xref: /linux/fs/btrfs/disk-io.c (revision f79e4d5f92a129a1159c973735007d4ddc8541f3)
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/scatterlist.h>
9 #include <linux/swap.h>
10 #include <linux/radix-tree.h>
11 #include <linux/writeback.h>
12 #include <linux/buffer_head.h>
13 #include <linux/workqueue.h>
14 #include <linux/kthread.h>
15 #include <linux/slab.h>
16 #include <linux/migrate.h>
17 #include <linux/ratelimit.h>
18 #include <linux/uuid.h>
19 #include <linux/semaphore.h>
20 #include <linux/error-injection.h>
21 #include <linux/crc32c.h>
22 #include <asm/unaligned.h>
23 #include "ctree.h"
24 #include "disk-io.h"
25 #include "transaction.h"
26 #include "btrfs_inode.h"
27 #include "volumes.h"
28 #include "print-tree.h"
29 #include "locking.h"
30 #include "tree-log.h"
31 #include "free-space-cache.h"
32 #include "free-space-tree.h"
33 #include "inode-map.h"
34 #include "check-integrity.h"
35 #include "rcu-string.h"
36 #include "dev-replace.h"
37 #include "raid56.h"
38 #include "sysfs.h"
39 #include "qgroup.h"
40 #include "compression.h"
41 #include "tree-checker.h"
42 #include "ref-verify.h"
43 
44 #ifdef CONFIG_X86
45 #include <asm/cpufeature.h>
46 #endif
47 
48 #define BTRFS_SUPER_FLAG_SUPP	(BTRFS_HEADER_FLAG_WRITTEN |\
49 				 BTRFS_HEADER_FLAG_RELOC |\
50 				 BTRFS_SUPER_FLAG_ERROR |\
51 				 BTRFS_SUPER_FLAG_SEEDING |\
52 				 BTRFS_SUPER_FLAG_METADUMP |\
53 				 BTRFS_SUPER_FLAG_METADUMP_V2)
54 
55 static const struct extent_io_ops btree_extent_io_ops;
56 static void end_workqueue_fn(struct btrfs_work *work);
57 static void free_fs_root(struct btrfs_root *root);
58 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
59 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
60 				      struct btrfs_fs_info *fs_info);
61 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
62 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
63 					struct extent_io_tree *dirty_pages,
64 					int mark);
65 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
66 				       struct extent_io_tree *pinned_extents);
67 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
68 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
69 
70 /*
71  * btrfs_end_io_wq structs are used to do processing in task context when an IO
72  * is complete.  This is used during reads to verify checksums, and it is used
73  * by writes to insert metadata for new file extents after IO is complete.
74  */
75 struct btrfs_end_io_wq {
76 	struct bio *bio;
77 	bio_end_io_t *end_io;
78 	void *private;
79 	struct btrfs_fs_info *info;
80 	blk_status_t status;
81 	enum btrfs_wq_endio_type metadata;
82 	struct btrfs_work work;
83 };
84 
85 static struct kmem_cache *btrfs_end_io_wq_cache;
86 
87 int __init btrfs_end_io_wq_init(void)
88 {
89 	btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
90 					sizeof(struct btrfs_end_io_wq),
91 					0,
92 					SLAB_MEM_SPREAD,
93 					NULL);
94 	if (!btrfs_end_io_wq_cache)
95 		return -ENOMEM;
96 	return 0;
97 }
98 
99 void __cold btrfs_end_io_wq_exit(void)
100 {
101 	kmem_cache_destroy(btrfs_end_io_wq_cache);
102 }
103 
104 /*
105  * async submit bios are used to offload expensive checksumming
106  * onto the worker threads.  They checksum file and metadata bios
107  * just before they are sent down the IO stack.
108  */
109 struct async_submit_bio {
110 	void *private_data;
111 	struct btrfs_fs_info *fs_info;
112 	struct bio *bio;
113 	extent_submit_bio_start_t *submit_bio_start;
114 	extent_submit_bio_done_t *submit_bio_done;
115 	int mirror_num;
116 	unsigned long bio_flags;
117 	/*
118 	 * bio_offset is optional, can be used if the pages in the bio
119 	 * can't tell us where in the file the bio should go
120 	 */
121 	u64 bio_offset;
122 	struct btrfs_work work;
123 	blk_status_t status;
124 };
125 
126 /*
127  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
128  * eb, the lockdep key is determined by the btrfs_root it belongs to and
129  * the level the eb occupies in the tree.
130  *
131  * Different roots are used for different purposes and may nest inside each
132  * other and they require separate keysets.  As lockdep keys should be
133  * static, assign keysets according to the purpose of the root as indicated
134  * by btrfs_root->objectid.  This ensures that all special purpose roots
135  * have separate keysets.
136  *
137  * Lock-nesting across peer nodes is always done with the immediate parent
138  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
139  * subclass to avoid triggering lockdep warning in such cases.
140  *
141  * The key is set by the readpage_end_io_hook after the buffer has passed
142  * csum validation but before the pages are unlocked.  It is also set by
143  * btrfs_init_new_buffer on freshly allocated blocks.
144  *
145  * We also add a check to make sure the highest level of the tree is the
146  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
147  * needs update as well.
148  */
149 #ifdef CONFIG_DEBUG_LOCK_ALLOC
150 # if BTRFS_MAX_LEVEL != 8
151 #  error
152 # endif
153 
154 static struct btrfs_lockdep_keyset {
155 	u64			id;		/* root objectid */
156 	const char		*name_stem;	/* lock name stem */
157 	char			names[BTRFS_MAX_LEVEL + 1][20];
158 	struct lock_class_key	keys[BTRFS_MAX_LEVEL + 1];
159 } btrfs_lockdep_keysets[] = {
160 	{ .id = BTRFS_ROOT_TREE_OBJECTID,	.name_stem = "root"	},
161 	{ .id = BTRFS_EXTENT_TREE_OBJECTID,	.name_stem = "extent"	},
162 	{ .id = BTRFS_CHUNK_TREE_OBJECTID,	.name_stem = "chunk"	},
163 	{ .id = BTRFS_DEV_TREE_OBJECTID,	.name_stem = "dev"	},
164 	{ .id = BTRFS_FS_TREE_OBJECTID,		.name_stem = "fs"	},
165 	{ .id = BTRFS_CSUM_TREE_OBJECTID,	.name_stem = "csum"	},
166 	{ .id = BTRFS_QUOTA_TREE_OBJECTID,	.name_stem = "quota"	},
167 	{ .id = BTRFS_TREE_LOG_OBJECTID,	.name_stem = "log"	},
168 	{ .id = BTRFS_TREE_RELOC_OBJECTID,	.name_stem = "treloc"	},
169 	{ .id = BTRFS_DATA_RELOC_TREE_OBJECTID,	.name_stem = "dreloc"	},
170 	{ .id = BTRFS_UUID_TREE_OBJECTID,	.name_stem = "uuid"	},
171 	{ .id = BTRFS_FREE_SPACE_TREE_OBJECTID,	.name_stem = "free-space" },
172 	{ .id = 0,				.name_stem = "tree"	},
173 };
174 
175 void __init btrfs_init_lockdep(void)
176 {
177 	int i, j;
178 
179 	/* initialize lockdep class names */
180 	for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
181 		struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
182 
183 		for (j = 0; j < ARRAY_SIZE(ks->names); j++)
184 			snprintf(ks->names[j], sizeof(ks->names[j]),
185 				 "btrfs-%s-%02d", ks->name_stem, j);
186 	}
187 }
188 
189 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
190 				    int level)
191 {
192 	struct btrfs_lockdep_keyset *ks;
193 
194 	BUG_ON(level >= ARRAY_SIZE(ks->keys));
195 
196 	/* find the matching keyset, id 0 is the default entry */
197 	for (ks = btrfs_lockdep_keysets; ks->id; ks++)
198 		if (ks->id == objectid)
199 			break;
200 
201 	lockdep_set_class_and_name(&eb->lock,
202 				   &ks->keys[level], ks->names[level]);
203 }
204 
205 #endif
206 
207 /*
208  * extents on the btree inode are pretty simple, there's one extent
209  * that covers the entire device
210  */
211 struct extent_map *btree_get_extent(struct btrfs_inode *inode,
212 		struct page *page, size_t pg_offset, u64 start, u64 len,
213 		int create)
214 {
215 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
216 	struct extent_map_tree *em_tree = &inode->extent_tree;
217 	struct extent_map *em;
218 	int ret;
219 
220 	read_lock(&em_tree->lock);
221 	em = lookup_extent_mapping(em_tree, start, len);
222 	if (em) {
223 		em->bdev = fs_info->fs_devices->latest_bdev;
224 		read_unlock(&em_tree->lock);
225 		goto out;
226 	}
227 	read_unlock(&em_tree->lock);
228 
229 	em = alloc_extent_map();
230 	if (!em) {
231 		em = ERR_PTR(-ENOMEM);
232 		goto out;
233 	}
234 	em->start = 0;
235 	em->len = (u64)-1;
236 	em->block_len = (u64)-1;
237 	em->block_start = 0;
238 	em->bdev = fs_info->fs_devices->latest_bdev;
239 
240 	write_lock(&em_tree->lock);
241 	ret = add_extent_mapping(em_tree, em, 0);
242 	if (ret == -EEXIST) {
243 		free_extent_map(em);
244 		em = lookup_extent_mapping(em_tree, start, len);
245 		if (!em)
246 			em = ERR_PTR(-EIO);
247 	} else if (ret) {
248 		free_extent_map(em);
249 		em = ERR_PTR(ret);
250 	}
251 	write_unlock(&em_tree->lock);
252 
253 out:
254 	return em;
255 }
256 
257 u32 btrfs_csum_data(const char *data, u32 seed, size_t len)
258 {
259 	return crc32c(seed, data, len);
260 }
261 
262 void btrfs_csum_final(u32 crc, u8 *result)
263 {
264 	put_unaligned_le32(~crc, result);
265 }
266 
267 /*
268  * compute the csum for a btree block, and either verify it or write it
269  * into the csum field of the block.
270  */
271 static int csum_tree_block(struct btrfs_fs_info *fs_info,
272 			   struct extent_buffer *buf,
273 			   int verify)
274 {
275 	u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
276 	char result[BTRFS_CSUM_SIZE];
277 	unsigned long len;
278 	unsigned long cur_len;
279 	unsigned long offset = BTRFS_CSUM_SIZE;
280 	char *kaddr;
281 	unsigned long map_start;
282 	unsigned long map_len;
283 	int err;
284 	u32 crc = ~(u32)0;
285 
286 	len = buf->len - offset;
287 	while (len > 0) {
288 		err = map_private_extent_buffer(buf, offset, 32,
289 					&kaddr, &map_start, &map_len);
290 		if (err)
291 			return err;
292 		cur_len = min(len, map_len - (offset - map_start));
293 		crc = btrfs_csum_data(kaddr + offset - map_start,
294 				      crc, cur_len);
295 		len -= cur_len;
296 		offset += cur_len;
297 	}
298 	memset(result, 0, BTRFS_CSUM_SIZE);
299 
300 	btrfs_csum_final(crc, result);
301 
302 	if (verify) {
303 		if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
304 			u32 val;
305 			u32 found = 0;
306 			memcpy(&found, result, csum_size);
307 
308 			read_extent_buffer(buf, &val, 0, csum_size);
309 			btrfs_warn_rl(fs_info,
310 				"%s checksum verify failed on %llu wanted %X found %X level %d",
311 				fs_info->sb->s_id, buf->start,
312 				val, found, btrfs_header_level(buf));
313 			return -EUCLEAN;
314 		}
315 	} else {
316 		write_extent_buffer(buf, result, 0, csum_size);
317 	}
318 
319 	return 0;
320 }
321 
322 /*
323  * we can't consider a given block up to date unless the transid of the
324  * block matches the transid in the parent node's pointer.  This is how we
325  * detect blocks that either didn't get written at all or got written
326  * in the wrong place.
327  */
328 static int verify_parent_transid(struct extent_io_tree *io_tree,
329 				 struct extent_buffer *eb, u64 parent_transid,
330 				 int atomic)
331 {
332 	struct extent_state *cached_state = NULL;
333 	int ret;
334 	bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
335 
336 	if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
337 		return 0;
338 
339 	if (atomic)
340 		return -EAGAIN;
341 
342 	if (need_lock) {
343 		btrfs_tree_read_lock(eb);
344 		btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
345 	}
346 
347 	lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
348 			 &cached_state);
349 	if (extent_buffer_uptodate(eb) &&
350 	    btrfs_header_generation(eb) == parent_transid) {
351 		ret = 0;
352 		goto out;
353 	}
354 	btrfs_err_rl(eb->fs_info,
355 		"parent transid verify failed on %llu wanted %llu found %llu",
356 			eb->start,
357 			parent_transid, btrfs_header_generation(eb));
358 	ret = 1;
359 
360 	/*
361 	 * Things reading via commit roots that don't have normal protection,
362 	 * like send, can have a really old block in cache that may point at a
363 	 * block that has been freed and re-allocated.  So don't clear uptodate
364 	 * if we find an eb that is under IO (dirty/writeback) because we could
365 	 * end up reading in the stale data and then writing it back out and
366 	 * making everybody very sad.
367 	 */
368 	if (!extent_buffer_under_io(eb))
369 		clear_extent_buffer_uptodate(eb);
370 out:
371 	unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
372 			     &cached_state);
373 	if (need_lock)
374 		btrfs_tree_read_unlock_blocking(eb);
375 	return ret;
376 }
377 
378 /*
379  * Return 0 if the superblock checksum type matches the checksum value of that
380  * algorithm. Pass the raw disk superblock data.
381  */
382 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
383 				  char *raw_disk_sb)
384 {
385 	struct btrfs_super_block *disk_sb =
386 		(struct btrfs_super_block *)raw_disk_sb;
387 	u16 csum_type = btrfs_super_csum_type(disk_sb);
388 	int ret = 0;
389 
390 	if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
391 		u32 crc = ~(u32)0;
392 		char result[sizeof(crc)];
393 
394 		/*
395 		 * The super_block structure does not span the whole
396 		 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
397 		 * is filled with zeros and is included in the checksum.
398 		 */
399 		crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
400 				crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
401 		btrfs_csum_final(crc, result);
402 
403 		if (memcmp(raw_disk_sb, result, sizeof(result)))
404 			ret = 1;
405 	}
406 
407 	if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
408 		btrfs_err(fs_info, "unsupported checksum algorithm %u",
409 				csum_type);
410 		ret = 1;
411 	}
412 
413 	return ret;
414 }
415 
416 static int verify_level_key(struct btrfs_fs_info *fs_info,
417 			    struct extent_buffer *eb, int level,
418 			    struct btrfs_key *first_key, u64 parent_transid)
419 {
420 	int found_level;
421 	struct btrfs_key found_key;
422 	int ret;
423 
424 	found_level = btrfs_header_level(eb);
425 	if (found_level != level) {
426 #ifdef CONFIG_BTRFS_DEBUG
427 		WARN_ON(1);
428 		btrfs_err(fs_info,
429 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
430 			  eb->start, level, found_level);
431 #endif
432 		return -EIO;
433 	}
434 
435 	if (!first_key)
436 		return 0;
437 
438 	/*
439 	 * For live tree block (new tree blocks in current transaction),
440 	 * we need proper lock context to avoid race, which is impossible here.
441 	 * So we only checks tree blocks which is read from disk, whose
442 	 * generation <= fs_info->last_trans_committed.
443 	 */
444 	if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
445 		return 0;
446 	if (found_level)
447 		btrfs_node_key_to_cpu(eb, &found_key, 0);
448 	else
449 		btrfs_item_key_to_cpu(eb, &found_key, 0);
450 	ret = btrfs_comp_cpu_keys(first_key, &found_key);
451 
452 #ifdef CONFIG_BTRFS_DEBUG
453 	if (ret) {
454 		WARN_ON(1);
455 		btrfs_err(fs_info,
456 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
457 			  eb->start, parent_transid, first_key->objectid,
458 			  first_key->type, first_key->offset,
459 			  found_key.objectid, found_key.type,
460 			  found_key.offset);
461 	}
462 #endif
463 	return ret;
464 }
465 
466 /*
467  * helper to read a given tree block, doing retries as required when
468  * the checksums don't match and we have alternate mirrors to try.
469  *
470  * @parent_transid:	expected transid, skip check if 0
471  * @level:		expected level, mandatory check
472  * @first_key:		expected key of first slot, skip check if NULL
473  */
474 static int btree_read_extent_buffer_pages(struct btrfs_fs_info *fs_info,
475 					  struct extent_buffer *eb,
476 					  u64 parent_transid, int level,
477 					  struct btrfs_key *first_key)
478 {
479 	struct extent_io_tree *io_tree;
480 	int failed = 0;
481 	int ret;
482 	int num_copies = 0;
483 	int mirror_num = 0;
484 	int failed_mirror = 0;
485 
486 	clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
487 	io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
488 	while (1) {
489 		ret = read_extent_buffer_pages(io_tree, eb, WAIT_COMPLETE,
490 					       mirror_num);
491 		if (!ret) {
492 			if (verify_parent_transid(io_tree, eb,
493 						   parent_transid, 0))
494 				ret = -EIO;
495 			else if (verify_level_key(fs_info, eb, level,
496 						  first_key, parent_transid))
497 				ret = -EUCLEAN;
498 			else
499 				break;
500 		}
501 
502 		/*
503 		 * This buffer's crc is fine, but its contents are corrupted, so
504 		 * there is no reason to read the other copies, they won't be
505 		 * any less wrong.
506 		 */
507 		if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags) ||
508 		    ret == -EUCLEAN)
509 			break;
510 
511 		num_copies = btrfs_num_copies(fs_info,
512 					      eb->start, eb->len);
513 		if (num_copies == 1)
514 			break;
515 
516 		if (!failed_mirror) {
517 			failed = 1;
518 			failed_mirror = eb->read_mirror;
519 		}
520 
521 		mirror_num++;
522 		if (mirror_num == failed_mirror)
523 			mirror_num++;
524 
525 		if (mirror_num > num_copies)
526 			break;
527 	}
528 
529 	if (failed && !ret && failed_mirror)
530 		repair_eb_io_failure(fs_info, eb, failed_mirror);
531 
532 	return ret;
533 }
534 
535 /*
536  * checksum a dirty tree block before IO.  This has extra checks to make sure
537  * we only fill in the checksum field in the first page of a multi-page block
538  */
539 
540 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
541 {
542 	u64 start = page_offset(page);
543 	u64 found_start;
544 	struct extent_buffer *eb;
545 
546 	eb = (struct extent_buffer *)page->private;
547 	if (page != eb->pages[0])
548 		return 0;
549 
550 	found_start = btrfs_header_bytenr(eb);
551 	/*
552 	 * Please do not consolidate these warnings into a single if.
553 	 * It is useful to know what went wrong.
554 	 */
555 	if (WARN_ON(found_start != start))
556 		return -EUCLEAN;
557 	if (WARN_ON(!PageUptodate(page)))
558 		return -EUCLEAN;
559 
560 	ASSERT(memcmp_extent_buffer(eb, fs_info->fsid,
561 			btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
562 
563 	return csum_tree_block(fs_info, eb, 0);
564 }
565 
566 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
567 				 struct extent_buffer *eb)
568 {
569 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
570 	u8 fsid[BTRFS_FSID_SIZE];
571 	int ret = 1;
572 
573 	read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
574 	while (fs_devices) {
575 		if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
576 			ret = 0;
577 			break;
578 		}
579 		fs_devices = fs_devices->seed;
580 	}
581 	return ret;
582 }
583 
584 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
585 				      u64 phy_offset, struct page *page,
586 				      u64 start, u64 end, int mirror)
587 {
588 	u64 found_start;
589 	int found_level;
590 	struct extent_buffer *eb;
591 	struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
592 	struct btrfs_fs_info *fs_info = root->fs_info;
593 	int ret = 0;
594 	int reads_done;
595 
596 	if (!page->private)
597 		goto out;
598 
599 	eb = (struct extent_buffer *)page->private;
600 
601 	/* the pending IO might have been the only thing that kept this buffer
602 	 * in memory.  Make sure we have a ref for all this other checks
603 	 */
604 	extent_buffer_get(eb);
605 
606 	reads_done = atomic_dec_and_test(&eb->io_pages);
607 	if (!reads_done)
608 		goto err;
609 
610 	eb->read_mirror = mirror;
611 	if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
612 		ret = -EIO;
613 		goto err;
614 	}
615 
616 	found_start = btrfs_header_bytenr(eb);
617 	if (found_start != eb->start) {
618 		btrfs_err_rl(fs_info, "bad tree block start %llu %llu",
619 			     found_start, eb->start);
620 		ret = -EIO;
621 		goto err;
622 	}
623 	if (check_tree_block_fsid(fs_info, eb)) {
624 		btrfs_err_rl(fs_info, "bad fsid on block %llu",
625 			     eb->start);
626 		ret = -EIO;
627 		goto err;
628 	}
629 	found_level = btrfs_header_level(eb);
630 	if (found_level >= BTRFS_MAX_LEVEL) {
631 		btrfs_err(fs_info, "bad tree block level %d",
632 			  (int)btrfs_header_level(eb));
633 		ret = -EIO;
634 		goto err;
635 	}
636 
637 	btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
638 				       eb, found_level);
639 
640 	ret = csum_tree_block(fs_info, eb, 1);
641 	if (ret)
642 		goto err;
643 
644 	/*
645 	 * If this is a leaf block and it is corrupt, set the corrupt bit so
646 	 * that we don't try and read the other copies of this block, just
647 	 * return -EIO.
648 	 */
649 	if (found_level == 0 && btrfs_check_leaf_full(fs_info, eb)) {
650 		set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
651 		ret = -EIO;
652 	}
653 
654 	if (found_level > 0 && btrfs_check_node(fs_info, eb))
655 		ret = -EIO;
656 
657 	if (!ret)
658 		set_extent_buffer_uptodate(eb);
659 err:
660 	if (reads_done &&
661 	    test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
662 		btree_readahead_hook(eb, ret);
663 
664 	if (ret) {
665 		/*
666 		 * our io error hook is going to dec the io pages
667 		 * again, we have to make sure it has something
668 		 * to decrement
669 		 */
670 		atomic_inc(&eb->io_pages);
671 		clear_extent_buffer_uptodate(eb);
672 	}
673 	free_extent_buffer(eb);
674 out:
675 	return ret;
676 }
677 
678 static int btree_io_failed_hook(struct page *page, int failed_mirror)
679 {
680 	struct extent_buffer *eb;
681 
682 	eb = (struct extent_buffer *)page->private;
683 	set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
684 	eb->read_mirror = failed_mirror;
685 	atomic_dec(&eb->io_pages);
686 	if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
687 		btree_readahead_hook(eb, -EIO);
688 	return -EIO;	/* we fixed nothing */
689 }
690 
691 static void end_workqueue_bio(struct bio *bio)
692 {
693 	struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
694 	struct btrfs_fs_info *fs_info;
695 	struct btrfs_workqueue *wq;
696 	btrfs_work_func_t func;
697 
698 	fs_info = end_io_wq->info;
699 	end_io_wq->status = bio->bi_status;
700 
701 	if (bio_op(bio) == REQ_OP_WRITE) {
702 		if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
703 			wq = fs_info->endio_meta_write_workers;
704 			func = btrfs_endio_meta_write_helper;
705 		} else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
706 			wq = fs_info->endio_freespace_worker;
707 			func = btrfs_freespace_write_helper;
708 		} else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
709 			wq = fs_info->endio_raid56_workers;
710 			func = btrfs_endio_raid56_helper;
711 		} else {
712 			wq = fs_info->endio_write_workers;
713 			func = btrfs_endio_write_helper;
714 		}
715 	} else {
716 		if (unlikely(end_io_wq->metadata ==
717 			     BTRFS_WQ_ENDIO_DIO_REPAIR)) {
718 			wq = fs_info->endio_repair_workers;
719 			func = btrfs_endio_repair_helper;
720 		} else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
721 			wq = fs_info->endio_raid56_workers;
722 			func = btrfs_endio_raid56_helper;
723 		} else if (end_io_wq->metadata) {
724 			wq = fs_info->endio_meta_workers;
725 			func = btrfs_endio_meta_helper;
726 		} else {
727 			wq = fs_info->endio_workers;
728 			func = btrfs_endio_helper;
729 		}
730 	}
731 
732 	btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
733 	btrfs_queue_work(wq, &end_io_wq->work);
734 }
735 
736 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
737 			enum btrfs_wq_endio_type metadata)
738 {
739 	struct btrfs_end_io_wq *end_io_wq;
740 
741 	end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
742 	if (!end_io_wq)
743 		return BLK_STS_RESOURCE;
744 
745 	end_io_wq->private = bio->bi_private;
746 	end_io_wq->end_io = bio->bi_end_io;
747 	end_io_wq->info = info;
748 	end_io_wq->status = 0;
749 	end_io_wq->bio = bio;
750 	end_io_wq->metadata = metadata;
751 
752 	bio->bi_private = end_io_wq;
753 	bio->bi_end_io = end_workqueue_bio;
754 	return 0;
755 }
756 
757 static void run_one_async_start(struct btrfs_work *work)
758 {
759 	struct async_submit_bio *async;
760 	blk_status_t ret;
761 
762 	async = container_of(work, struct  async_submit_bio, work);
763 	ret = async->submit_bio_start(async->private_data, async->bio,
764 				      async->bio_offset);
765 	if (ret)
766 		async->status = ret;
767 }
768 
769 static void run_one_async_done(struct btrfs_work *work)
770 {
771 	struct async_submit_bio *async;
772 
773 	async = container_of(work, struct  async_submit_bio, work);
774 
775 	/* If an error occurred we just want to clean up the bio and move on */
776 	if (async->status) {
777 		async->bio->bi_status = async->status;
778 		bio_endio(async->bio);
779 		return;
780 	}
781 
782 	async->submit_bio_done(async->private_data, async->bio, async->mirror_num);
783 }
784 
785 static void run_one_async_free(struct btrfs_work *work)
786 {
787 	struct async_submit_bio *async;
788 
789 	async = container_of(work, struct  async_submit_bio, work);
790 	kfree(async);
791 }
792 
793 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
794 				 int mirror_num, unsigned long bio_flags,
795 				 u64 bio_offset, void *private_data,
796 				 extent_submit_bio_start_t *submit_bio_start,
797 				 extent_submit_bio_done_t *submit_bio_done)
798 {
799 	struct async_submit_bio *async;
800 
801 	async = kmalloc(sizeof(*async), GFP_NOFS);
802 	if (!async)
803 		return BLK_STS_RESOURCE;
804 
805 	async->private_data = private_data;
806 	async->fs_info = fs_info;
807 	async->bio = bio;
808 	async->mirror_num = mirror_num;
809 	async->submit_bio_start = submit_bio_start;
810 	async->submit_bio_done = submit_bio_done;
811 
812 	btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
813 			run_one_async_done, run_one_async_free);
814 
815 	async->bio_flags = bio_flags;
816 	async->bio_offset = bio_offset;
817 
818 	async->status = 0;
819 
820 	if (op_is_sync(bio->bi_opf))
821 		btrfs_set_work_high_priority(&async->work);
822 
823 	btrfs_queue_work(fs_info->workers, &async->work);
824 	return 0;
825 }
826 
827 static blk_status_t btree_csum_one_bio(struct bio *bio)
828 {
829 	struct bio_vec *bvec;
830 	struct btrfs_root *root;
831 	int i, ret = 0;
832 
833 	ASSERT(!bio_flagged(bio, BIO_CLONED));
834 	bio_for_each_segment_all(bvec, bio, i) {
835 		root = BTRFS_I(bvec->bv_page->mapping->host)->root;
836 		ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
837 		if (ret)
838 			break;
839 	}
840 
841 	return errno_to_blk_status(ret);
842 }
843 
844 static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
845 					     u64 bio_offset)
846 {
847 	/*
848 	 * when we're called for a write, we're already in the async
849 	 * submission context.  Just jump into btrfs_map_bio
850 	 */
851 	return btree_csum_one_bio(bio);
852 }
853 
854 static blk_status_t btree_submit_bio_done(void *private_data, struct bio *bio,
855 					    int mirror_num)
856 {
857 	struct inode *inode = private_data;
858 	blk_status_t ret;
859 
860 	/*
861 	 * when we're called for a write, we're already in the async
862 	 * submission context.  Just jump into btrfs_map_bio
863 	 */
864 	ret = btrfs_map_bio(btrfs_sb(inode->i_sb), bio, mirror_num, 1);
865 	if (ret) {
866 		bio->bi_status = ret;
867 		bio_endio(bio);
868 	}
869 	return ret;
870 }
871 
872 static int check_async_write(struct btrfs_inode *bi)
873 {
874 	if (atomic_read(&bi->sync_writers))
875 		return 0;
876 #ifdef CONFIG_X86
877 	if (static_cpu_has(X86_FEATURE_XMM4_2))
878 		return 0;
879 #endif
880 	return 1;
881 }
882 
883 static blk_status_t btree_submit_bio_hook(void *private_data, struct bio *bio,
884 					  int mirror_num, unsigned long bio_flags,
885 					  u64 bio_offset)
886 {
887 	struct inode *inode = private_data;
888 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
889 	int async = check_async_write(BTRFS_I(inode));
890 	blk_status_t ret;
891 
892 	if (bio_op(bio) != REQ_OP_WRITE) {
893 		/*
894 		 * called for a read, do the setup so that checksum validation
895 		 * can happen in the async kernel threads
896 		 */
897 		ret = btrfs_bio_wq_end_io(fs_info, bio,
898 					  BTRFS_WQ_ENDIO_METADATA);
899 		if (ret)
900 			goto out_w_error;
901 		ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
902 	} else if (!async) {
903 		ret = btree_csum_one_bio(bio);
904 		if (ret)
905 			goto out_w_error;
906 		ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
907 	} else {
908 		/*
909 		 * kthread helpers are used to submit writes so that
910 		 * checksumming can happen in parallel across all CPUs
911 		 */
912 		ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
913 					  bio_offset, private_data,
914 					  btree_submit_bio_start,
915 					  btree_submit_bio_done);
916 	}
917 
918 	if (ret)
919 		goto out_w_error;
920 	return 0;
921 
922 out_w_error:
923 	bio->bi_status = ret;
924 	bio_endio(bio);
925 	return ret;
926 }
927 
928 #ifdef CONFIG_MIGRATION
929 static int btree_migratepage(struct address_space *mapping,
930 			struct page *newpage, struct page *page,
931 			enum migrate_mode mode)
932 {
933 	/*
934 	 * we can't safely write a btree page from here,
935 	 * we haven't done the locking hook
936 	 */
937 	if (PageDirty(page))
938 		return -EAGAIN;
939 	/*
940 	 * Buffers may be managed in a filesystem specific way.
941 	 * We must have no buffers or drop them.
942 	 */
943 	if (page_has_private(page) &&
944 	    !try_to_release_page(page, GFP_KERNEL))
945 		return -EAGAIN;
946 	return migrate_page(mapping, newpage, page, mode);
947 }
948 #endif
949 
950 
951 static int btree_writepages(struct address_space *mapping,
952 			    struct writeback_control *wbc)
953 {
954 	struct btrfs_fs_info *fs_info;
955 	int ret;
956 
957 	if (wbc->sync_mode == WB_SYNC_NONE) {
958 
959 		if (wbc->for_kupdate)
960 			return 0;
961 
962 		fs_info = BTRFS_I(mapping->host)->root->fs_info;
963 		/* this is a bit racy, but that's ok */
964 		ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
965 					     BTRFS_DIRTY_METADATA_THRESH);
966 		if (ret < 0)
967 			return 0;
968 	}
969 	return btree_write_cache_pages(mapping, wbc);
970 }
971 
972 static int btree_readpage(struct file *file, struct page *page)
973 {
974 	struct extent_io_tree *tree;
975 	tree = &BTRFS_I(page->mapping->host)->io_tree;
976 	return extent_read_full_page(tree, page, btree_get_extent, 0);
977 }
978 
979 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
980 {
981 	if (PageWriteback(page) || PageDirty(page))
982 		return 0;
983 
984 	return try_release_extent_buffer(page);
985 }
986 
987 static void btree_invalidatepage(struct page *page, unsigned int offset,
988 				 unsigned int length)
989 {
990 	struct extent_io_tree *tree;
991 	tree = &BTRFS_I(page->mapping->host)->io_tree;
992 	extent_invalidatepage(tree, page, offset);
993 	btree_releasepage(page, GFP_NOFS);
994 	if (PagePrivate(page)) {
995 		btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
996 			   "page private not zero on page %llu",
997 			   (unsigned long long)page_offset(page));
998 		ClearPagePrivate(page);
999 		set_page_private(page, 0);
1000 		put_page(page);
1001 	}
1002 }
1003 
1004 static int btree_set_page_dirty(struct page *page)
1005 {
1006 #ifdef DEBUG
1007 	struct extent_buffer *eb;
1008 
1009 	BUG_ON(!PagePrivate(page));
1010 	eb = (struct extent_buffer *)page->private;
1011 	BUG_ON(!eb);
1012 	BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1013 	BUG_ON(!atomic_read(&eb->refs));
1014 	btrfs_assert_tree_locked(eb);
1015 #endif
1016 	return __set_page_dirty_nobuffers(page);
1017 }
1018 
1019 static const struct address_space_operations btree_aops = {
1020 	.readpage	= btree_readpage,
1021 	.writepages	= btree_writepages,
1022 	.releasepage	= btree_releasepage,
1023 	.invalidatepage = btree_invalidatepage,
1024 #ifdef CONFIG_MIGRATION
1025 	.migratepage	= btree_migratepage,
1026 #endif
1027 	.set_page_dirty = btree_set_page_dirty,
1028 };
1029 
1030 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1031 {
1032 	struct extent_buffer *buf = NULL;
1033 	struct inode *btree_inode = fs_info->btree_inode;
1034 
1035 	buf = btrfs_find_create_tree_block(fs_info, bytenr);
1036 	if (IS_ERR(buf))
1037 		return;
1038 	read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1039 				 buf, WAIT_NONE, 0);
1040 	free_extent_buffer(buf);
1041 }
1042 
1043 int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr,
1044 			 int mirror_num, struct extent_buffer **eb)
1045 {
1046 	struct extent_buffer *buf = NULL;
1047 	struct inode *btree_inode = fs_info->btree_inode;
1048 	struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1049 	int ret;
1050 
1051 	buf = btrfs_find_create_tree_block(fs_info, bytenr);
1052 	if (IS_ERR(buf))
1053 		return 0;
1054 
1055 	set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1056 
1057 	ret = read_extent_buffer_pages(io_tree, buf, WAIT_PAGE_LOCK,
1058 				       mirror_num);
1059 	if (ret) {
1060 		free_extent_buffer(buf);
1061 		return ret;
1062 	}
1063 
1064 	if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1065 		free_extent_buffer(buf);
1066 		return -EIO;
1067 	} else if (extent_buffer_uptodate(buf)) {
1068 		*eb = buf;
1069 	} else {
1070 		free_extent_buffer(buf);
1071 	}
1072 	return 0;
1073 }
1074 
1075 struct extent_buffer *btrfs_find_create_tree_block(
1076 						struct btrfs_fs_info *fs_info,
1077 						u64 bytenr)
1078 {
1079 	if (btrfs_is_testing(fs_info))
1080 		return alloc_test_extent_buffer(fs_info, bytenr);
1081 	return alloc_extent_buffer(fs_info, bytenr);
1082 }
1083 
1084 
1085 int btrfs_write_tree_block(struct extent_buffer *buf)
1086 {
1087 	return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1088 					buf->start + buf->len - 1);
1089 }
1090 
1091 void btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1092 {
1093 	filemap_fdatawait_range(buf->pages[0]->mapping,
1094 			        buf->start, buf->start + buf->len - 1);
1095 }
1096 
1097 /*
1098  * Read tree block at logical address @bytenr and do variant basic but critical
1099  * verification.
1100  *
1101  * @parent_transid:	expected transid of this tree block, skip check if 0
1102  * @level:		expected level, mandatory check
1103  * @first_key:		expected key in slot 0, skip check if NULL
1104  */
1105 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1106 				      u64 parent_transid, int level,
1107 				      struct btrfs_key *first_key)
1108 {
1109 	struct extent_buffer *buf = NULL;
1110 	int ret;
1111 
1112 	buf = btrfs_find_create_tree_block(fs_info, bytenr);
1113 	if (IS_ERR(buf))
1114 		return buf;
1115 
1116 	ret = btree_read_extent_buffer_pages(fs_info, buf, parent_transid,
1117 					     level, first_key);
1118 	if (ret) {
1119 		free_extent_buffer(buf);
1120 		return ERR_PTR(ret);
1121 	}
1122 	return buf;
1123 
1124 }
1125 
1126 void clean_tree_block(struct btrfs_fs_info *fs_info,
1127 		      struct extent_buffer *buf)
1128 {
1129 	if (btrfs_header_generation(buf) ==
1130 	    fs_info->running_transaction->transid) {
1131 		btrfs_assert_tree_locked(buf);
1132 
1133 		if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1134 			percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1135 						 -buf->len,
1136 						 fs_info->dirty_metadata_batch);
1137 			/* ugh, clear_extent_buffer_dirty needs to lock the page */
1138 			btrfs_set_lock_blocking(buf);
1139 			clear_extent_buffer_dirty(buf);
1140 		}
1141 	}
1142 }
1143 
1144 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1145 {
1146 	struct btrfs_subvolume_writers *writers;
1147 	int ret;
1148 
1149 	writers = kmalloc(sizeof(*writers), GFP_NOFS);
1150 	if (!writers)
1151 		return ERR_PTR(-ENOMEM);
1152 
1153 	ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS);
1154 	if (ret < 0) {
1155 		kfree(writers);
1156 		return ERR_PTR(ret);
1157 	}
1158 
1159 	init_waitqueue_head(&writers->wait);
1160 	return writers;
1161 }
1162 
1163 static void
1164 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1165 {
1166 	percpu_counter_destroy(&writers->counter);
1167 	kfree(writers);
1168 }
1169 
1170 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1171 			 u64 objectid)
1172 {
1173 	bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1174 	root->node = NULL;
1175 	root->commit_root = NULL;
1176 	root->state = 0;
1177 	root->orphan_cleanup_state = 0;
1178 
1179 	root->objectid = objectid;
1180 	root->last_trans = 0;
1181 	root->highest_objectid = 0;
1182 	root->nr_delalloc_inodes = 0;
1183 	root->nr_ordered_extents = 0;
1184 	root->name = NULL;
1185 	root->inode_tree = RB_ROOT;
1186 	INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1187 	root->block_rsv = NULL;
1188 
1189 	INIT_LIST_HEAD(&root->dirty_list);
1190 	INIT_LIST_HEAD(&root->root_list);
1191 	INIT_LIST_HEAD(&root->delalloc_inodes);
1192 	INIT_LIST_HEAD(&root->delalloc_root);
1193 	INIT_LIST_HEAD(&root->ordered_extents);
1194 	INIT_LIST_HEAD(&root->ordered_root);
1195 	INIT_LIST_HEAD(&root->logged_list[0]);
1196 	INIT_LIST_HEAD(&root->logged_list[1]);
1197 	spin_lock_init(&root->inode_lock);
1198 	spin_lock_init(&root->delalloc_lock);
1199 	spin_lock_init(&root->ordered_extent_lock);
1200 	spin_lock_init(&root->accounting_lock);
1201 	spin_lock_init(&root->log_extents_lock[0]);
1202 	spin_lock_init(&root->log_extents_lock[1]);
1203 	spin_lock_init(&root->qgroup_meta_rsv_lock);
1204 	mutex_init(&root->objectid_mutex);
1205 	mutex_init(&root->log_mutex);
1206 	mutex_init(&root->ordered_extent_mutex);
1207 	mutex_init(&root->delalloc_mutex);
1208 	init_waitqueue_head(&root->log_writer_wait);
1209 	init_waitqueue_head(&root->log_commit_wait[0]);
1210 	init_waitqueue_head(&root->log_commit_wait[1]);
1211 	INIT_LIST_HEAD(&root->log_ctxs[0]);
1212 	INIT_LIST_HEAD(&root->log_ctxs[1]);
1213 	atomic_set(&root->log_commit[0], 0);
1214 	atomic_set(&root->log_commit[1], 0);
1215 	atomic_set(&root->log_writers, 0);
1216 	atomic_set(&root->log_batch, 0);
1217 	refcount_set(&root->refs, 1);
1218 	atomic_set(&root->will_be_snapshotted, 0);
1219 	root->log_transid = 0;
1220 	root->log_transid_committed = -1;
1221 	root->last_log_commit = 0;
1222 	if (!dummy)
1223 		extent_io_tree_init(&root->dirty_log_pages, NULL);
1224 
1225 	memset(&root->root_key, 0, sizeof(root->root_key));
1226 	memset(&root->root_item, 0, sizeof(root->root_item));
1227 	memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1228 	if (!dummy)
1229 		root->defrag_trans_start = fs_info->generation;
1230 	else
1231 		root->defrag_trans_start = 0;
1232 	root->root_key.objectid = objectid;
1233 	root->anon_dev = 0;
1234 
1235 	spin_lock_init(&root->root_item_lock);
1236 }
1237 
1238 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1239 		gfp_t flags)
1240 {
1241 	struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1242 	if (root)
1243 		root->fs_info = fs_info;
1244 	return root;
1245 }
1246 
1247 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1248 /* Should only be used by the testing infrastructure */
1249 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1250 {
1251 	struct btrfs_root *root;
1252 
1253 	if (!fs_info)
1254 		return ERR_PTR(-EINVAL);
1255 
1256 	root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1257 	if (!root)
1258 		return ERR_PTR(-ENOMEM);
1259 
1260 	/* We don't use the stripesize in selftest, set it as sectorsize */
1261 	__setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1262 	root->alloc_bytenr = 0;
1263 
1264 	return root;
1265 }
1266 #endif
1267 
1268 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1269 				     struct btrfs_fs_info *fs_info,
1270 				     u64 objectid)
1271 {
1272 	struct extent_buffer *leaf;
1273 	struct btrfs_root *tree_root = fs_info->tree_root;
1274 	struct btrfs_root *root;
1275 	struct btrfs_key key;
1276 	int ret = 0;
1277 	uuid_le uuid = NULL_UUID_LE;
1278 
1279 	root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1280 	if (!root)
1281 		return ERR_PTR(-ENOMEM);
1282 
1283 	__setup_root(root, fs_info, objectid);
1284 	root->root_key.objectid = objectid;
1285 	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1286 	root->root_key.offset = 0;
1287 
1288 	leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1289 	if (IS_ERR(leaf)) {
1290 		ret = PTR_ERR(leaf);
1291 		leaf = NULL;
1292 		goto fail;
1293 	}
1294 
1295 	memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1296 	btrfs_set_header_bytenr(leaf, leaf->start);
1297 	btrfs_set_header_generation(leaf, trans->transid);
1298 	btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1299 	btrfs_set_header_owner(leaf, objectid);
1300 	root->node = leaf;
1301 
1302 	write_extent_buffer_fsid(leaf, fs_info->fsid);
1303 	write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
1304 	btrfs_mark_buffer_dirty(leaf);
1305 
1306 	root->commit_root = btrfs_root_node(root);
1307 	set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1308 
1309 	root->root_item.flags = 0;
1310 	root->root_item.byte_limit = 0;
1311 	btrfs_set_root_bytenr(&root->root_item, leaf->start);
1312 	btrfs_set_root_generation(&root->root_item, trans->transid);
1313 	btrfs_set_root_level(&root->root_item, 0);
1314 	btrfs_set_root_refs(&root->root_item, 1);
1315 	btrfs_set_root_used(&root->root_item, leaf->len);
1316 	btrfs_set_root_last_snapshot(&root->root_item, 0);
1317 	btrfs_set_root_dirid(&root->root_item, 0);
1318 	if (is_fstree(objectid))
1319 		uuid_le_gen(&uuid);
1320 	memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1321 	root->root_item.drop_level = 0;
1322 
1323 	key.objectid = objectid;
1324 	key.type = BTRFS_ROOT_ITEM_KEY;
1325 	key.offset = 0;
1326 	ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1327 	if (ret)
1328 		goto fail;
1329 
1330 	btrfs_tree_unlock(leaf);
1331 
1332 	return root;
1333 
1334 fail:
1335 	if (leaf) {
1336 		btrfs_tree_unlock(leaf);
1337 		free_extent_buffer(root->commit_root);
1338 		free_extent_buffer(leaf);
1339 	}
1340 	kfree(root);
1341 
1342 	return ERR_PTR(ret);
1343 }
1344 
1345 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1346 					 struct btrfs_fs_info *fs_info)
1347 {
1348 	struct btrfs_root *root;
1349 	struct extent_buffer *leaf;
1350 
1351 	root = btrfs_alloc_root(fs_info, GFP_NOFS);
1352 	if (!root)
1353 		return ERR_PTR(-ENOMEM);
1354 
1355 	__setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1356 
1357 	root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1358 	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1359 	root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1360 
1361 	/*
1362 	 * DON'T set REF_COWS for log trees
1363 	 *
1364 	 * log trees do not get reference counted because they go away
1365 	 * before a real commit is actually done.  They do store pointers
1366 	 * to file data extents, and those reference counts still get
1367 	 * updated (along with back refs to the log tree).
1368 	 */
1369 
1370 	leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1371 			NULL, 0, 0, 0);
1372 	if (IS_ERR(leaf)) {
1373 		kfree(root);
1374 		return ERR_CAST(leaf);
1375 	}
1376 
1377 	memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1378 	btrfs_set_header_bytenr(leaf, leaf->start);
1379 	btrfs_set_header_generation(leaf, trans->transid);
1380 	btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1381 	btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1382 	root->node = leaf;
1383 
1384 	write_extent_buffer_fsid(root->node, fs_info->fsid);
1385 	btrfs_mark_buffer_dirty(root->node);
1386 	btrfs_tree_unlock(root->node);
1387 	return root;
1388 }
1389 
1390 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1391 			     struct btrfs_fs_info *fs_info)
1392 {
1393 	struct btrfs_root *log_root;
1394 
1395 	log_root = alloc_log_tree(trans, fs_info);
1396 	if (IS_ERR(log_root))
1397 		return PTR_ERR(log_root);
1398 	WARN_ON(fs_info->log_root_tree);
1399 	fs_info->log_root_tree = log_root;
1400 	return 0;
1401 }
1402 
1403 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1404 		       struct btrfs_root *root)
1405 {
1406 	struct btrfs_fs_info *fs_info = root->fs_info;
1407 	struct btrfs_root *log_root;
1408 	struct btrfs_inode_item *inode_item;
1409 
1410 	log_root = alloc_log_tree(trans, fs_info);
1411 	if (IS_ERR(log_root))
1412 		return PTR_ERR(log_root);
1413 
1414 	log_root->last_trans = trans->transid;
1415 	log_root->root_key.offset = root->root_key.objectid;
1416 
1417 	inode_item = &log_root->root_item.inode;
1418 	btrfs_set_stack_inode_generation(inode_item, 1);
1419 	btrfs_set_stack_inode_size(inode_item, 3);
1420 	btrfs_set_stack_inode_nlink(inode_item, 1);
1421 	btrfs_set_stack_inode_nbytes(inode_item,
1422 				     fs_info->nodesize);
1423 	btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1424 
1425 	btrfs_set_root_node(&log_root->root_item, log_root->node);
1426 
1427 	WARN_ON(root->log_root);
1428 	root->log_root = log_root;
1429 	root->log_transid = 0;
1430 	root->log_transid_committed = -1;
1431 	root->last_log_commit = 0;
1432 	return 0;
1433 }
1434 
1435 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1436 					       struct btrfs_key *key)
1437 {
1438 	struct btrfs_root *root;
1439 	struct btrfs_fs_info *fs_info = tree_root->fs_info;
1440 	struct btrfs_path *path;
1441 	u64 generation;
1442 	int ret;
1443 	int level;
1444 
1445 	path = btrfs_alloc_path();
1446 	if (!path)
1447 		return ERR_PTR(-ENOMEM);
1448 
1449 	root = btrfs_alloc_root(fs_info, GFP_NOFS);
1450 	if (!root) {
1451 		ret = -ENOMEM;
1452 		goto alloc_fail;
1453 	}
1454 
1455 	__setup_root(root, fs_info, key->objectid);
1456 
1457 	ret = btrfs_find_root(tree_root, key, path,
1458 			      &root->root_item, &root->root_key);
1459 	if (ret) {
1460 		if (ret > 0)
1461 			ret = -ENOENT;
1462 		goto find_fail;
1463 	}
1464 
1465 	generation = btrfs_root_generation(&root->root_item);
1466 	level = btrfs_root_level(&root->root_item);
1467 	root->node = read_tree_block(fs_info,
1468 				     btrfs_root_bytenr(&root->root_item),
1469 				     generation, level, NULL);
1470 	if (IS_ERR(root->node)) {
1471 		ret = PTR_ERR(root->node);
1472 		goto find_fail;
1473 	} else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1474 		ret = -EIO;
1475 		free_extent_buffer(root->node);
1476 		goto find_fail;
1477 	}
1478 	root->commit_root = btrfs_root_node(root);
1479 out:
1480 	btrfs_free_path(path);
1481 	return root;
1482 
1483 find_fail:
1484 	kfree(root);
1485 alloc_fail:
1486 	root = ERR_PTR(ret);
1487 	goto out;
1488 }
1489 
1490 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1491 				      struct btrfs_key *location)
1492 {
1493 	struct btrfs_root *root;
1494 
1495 	root = btrfs_read_tree_root(tree_root, location);
1496 	if (IS_ERR(root))
1497 		return root;
1498 
1499 	if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1500 		set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1501 		btrfs_check_and_init_root_item(&root->root_item);
1502 	}
1503 
1504 	return root;
1505 }
1506 
1507 int btrfs_init_fs_root(struct btrfs_root *root)
1508 {
1509 	int ret;
1510 	struct btrfs_subvolume_writers *writers;
1511 
1512 	root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1513 	root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1514 					GFP_NOFS);
1515 	if (!root->free_ino_pinned || !root->free_ino_ctl) {
1516 		ret = -ENOMEM;
1517 		goto fail;
1518 	}
1519 
1520 	writers = btrfs_alloc_subvolume_writers();
1521 	if (IS_ERR(writers)) {
1522 		ret = PTR_ERR(writers);
1523 		goto fail;
1524 	}
1525 	root->subv_writers = writers;
1526 
1527 	btrfs_init_free_ino_ctl(root);
1528 	spin_lock_init(&root->ino_cache_lock);
1529 	init_waitqueue_head(&root->ino_cache_wait);
1530 
1531 	ret = get_anon_bdev(&root->anon_dev);
1532 	if (ret)
1533 		goto fail;
1534 
1535 	mutex_lock(&root->objectid_mutex);
1536 	ret = btrfs_find_highest_objectid(root,
1537 					&root->highest_objectid);
1538 	if (ret) {
1539 		mutex_unlock(&root->objectid_mutex);
1540 		goto fail;
1541 	}
1542 
1543 	ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1544 
1545 	mutex_unlock(&root->objectid_mutex);
1546 
1547 	return 0;
1548 fail:
1549 	/* the caller is responsible to call free_fs_root */
1550 	return ret;
1551 }
1552 
1553 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1554 					u64 root_id)
1555 {
1556 	struct btrfs_root *root;
1557 
1558 	spin_lock(&fs_info->fs_roots_radix_lock);
1559 	root = radix_tree_lookup(&fs_info->fs_roots_radix,
1560 				 (unsigned long)root_id);
1561 	spin_unlock(&fs_info->fs_roots_radix_lock);
1562 	return root;
1563 }
1564 
1565 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1566 			 struct btrfs_root *root)
1567 {
1568 	int ret;
1569 
1570 	ret = radix_tree_preload(GFP_NOFS);
1571 	if (ret)
1572 		return ret;
1573 
1574 	spin_lock(&fs_info->fs_roots_radix_lock);
1575 	ret = radix_tree_insert(&fs_info->fs_roots_radix,
1576 				(unsigned long)root->root_key.objectid,
1577 				root);
1578 	if (ret == 0)
1579 		set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1580 	spin_unlock(&fs_info->fs_roots_radix_lock);
1581 	radix_tree_preload_end();
1582 
1583 	return ret;
1584 }
1585 
1586 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1587 				     struct btrfs_key *location,
1588 				     bool check_ref)
1589 {
1590 	struct btrfs_root *root;
1591 	struct btrfs_path *path;
1592 	struct btrfs_key key;
1593 	int ret;
1594 
1595 	if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1596 		return fs_info->tree_root;
1597 	if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1598 		return fs_info->extent_root;
1599 	if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1600 		return fs_info->chunk_root;
1601 	if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1602 		return fs_info->dev_root;
1603 	if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1604 		return fs_info->csum_root;
1605 	if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1606 		return fs_info->quota_root ? fs_info->quota_root :
1607 					     ERR_PTR(-ENOENT);
1608 	if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1609 		return fs_info->uuid_root ? fs_info->uuid_root :
1610 					    ERR_PTR(-ENOENT);
1611 	if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1612 		return fs_info->free_space_root ? fs_info->free_space_root :
1613 						  ERR_PTR(-ENOENT);
1614 again:
1615 	root = btrfs_lookup_fs_root(fs_info, location->objectid);
1616 	if (root) {
1617 		if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1618 			return ERR_PTR(-ENOENT);
1619 		return root;
1620 	}
1621 
1622 	root = btrfs_read_fs_root(fs_info->tree_root, location);
1623 	if (IS_ERR(root))
1624 		return root;
1625 
1626 	if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1627 		ret = -ENOENT;
1628 		goto fail;
1629 	}
1630 
1631 	ret = btrfs_init_fs_root(root);
1632 	if (ret)
1633 		goto fail;
1634 
1635 	path = btrfs_alloc_path();
1636 	if (!path) {
1637 		ret = -ENOMEM;
1638 		goto fail;
1639 	}
1640 	key.objectid = BTRFS_ORPHAN_OBJECTID;
1641 	key.type = BTRFS_ORPHAN_ITEM_KEY;
1642 	key.offset = location->objectid;
1643 
1644 	ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1645 	btrfs_free_path(path);
1646 	if (ret < 0)
1647 		goto fail;
1648 	if (ret == 0)
1649 		set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1650 
1651 	ret = btrfs_insert_fs_root(fs_info, root);
1652 	if (ret) {
1653 		if (ret == -EEXIST) {
1654 			free_fs_root(root);
1655 			goto again;
1656 		}
1657 		goto fail;
1658 	}
1659 	return root;
1660 fail:
1661 	free_fs_root(root);
1662 	return ERR_PTR(ret);
1663 }
1664 
1665 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1666 {
1667 	struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1668 	int ret = 0;
1669 	struct btrfs_device *device;
1670 	struct backing_dev_info *bdi;
1671 
1672 	rcu_read_lock();
1673 	list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1674 		if (!device->bdev)
1675 			continue;
1676 		bdi = device->bdev->bd_bdi;
1677 		if (bdi_congested(bdi, bdi_bits)) {
1678 			ret = 1;
1679 			break;
1680 		}
1681 	}
1682 	rcu_read_unlock();
1683 	return ret;
1684 }
1685 
1686 /*
1687  * called by the kthread helper functions to finally call the bio end_io
1688  * functions.  This is where read checksum verification actually happens
1689  */
1690 static void end_workqueue_fn(struct btrfs_work *work)
1691 {
1692 	struct bio *bio;
1693 	struct btrfs_end_io_wq *end_io_wq;
1694 
1695 	end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1696 	bio = end_io_wq->bio;
1697 
1698 	bio->bi_status = end_io_wq->status;
1699 	bio->bi_private = end_io_wq->private;
1700 	bio->bi_end_io = end_io_wq->end_io;
1701 	kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1702 	bio_endio(bio);
1703 }
1704 
1705 static int cleaner_kthread(void *arg)
1706 {
1707 	struct btrfs_root *root = arg;
1708 	struct btrfs_fs_info *fs_info = root->fs_info;
1709 	int again;
1710 	struct btrfs_trans_handle *trans;
1711 
1712 	do {
1713 		again = 0;
1714 
1715 		/* Make the cleaner go to sleep early. */
1716 		if (btrfs_need_cleaner_sleep(fs_info))
1717 			goto sleep;
1718 
1719 		/*
1720 		 * Do not do anything if we might cause open_ctree() to block
1721 		 * before we have finished mounting the filesystem.
1722 		 */
1723 		if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1724 			goto sleep;
1725 
1726 		if (!mutex_trylock(&fs_info->cleaner_mutex))
1727 			goto sleep;
1728 
1729 		/*
1730 		 * Avoid the problem that we change the status of the fs
1731 		 * during the above check and trylock.
1732 		 */
1733 		if (btrfs_need_cleaner_sleep(fs_info)) {
1734 			mutex_unlock(&fs_info->cleaner_mutex);
1735 			goto sleep;
1736 		}
1737 
1738 		mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
1739 		btrfs_run_delayed_iputs(fs_info);
1740 		mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
1741 
1742 		again = btrfs_clean_one_deleted_snapshot(root);
1743 		mutex_unlock(&fs_info->cleaner_mutex);
1744 
1745 		/*
1746 		 * The defragger has dealt with the R/O remount and umount,
1747 		 * needn't do anything special here.
1748 		 */
1749 		btrfs_run_defrag_inodes(fs_info);
1750 
1751 		/*
1752 		 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1753 		 * with relocation (btrfs_relocate_chunk) and relocation
1754 		 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1755 		 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1756 		 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1757 		 * unused block groups.
1758 		 */
1759 		btrfs_delete_unused_bgs(fs_info);
1760 sleep:
1761 		if (!again) {
1762 			set_current_state(TASK_INTERRUPTIBLE);
1763 			if (!kthread_should_stop())
1764 				schedule();
1765 			__set_current_state(TASK_RUNNING);
1766 		}
1767 	} while (!kthread_should_stop());
1768 
1769 	/*
1770 	 * Transaction kthread is stopped before us and wakes us up.
1771 	 * However we might have started a new transaction and COWed some
1772 	 * tree blocks when deleting unused block groups for example. So
1773 	 * make sure we commit the transaction we started to have a clean
1774 	 * shutdown when evicting the btree inode - if it has dirty pages
1775 	 * when we do the final iput() on it, eviction will trigger a
1776 	 * writeback for it which will fail with null pointer dereferences
1777 	 * since work queues and other resources were already released and
1778 	 * destroyed by the time the iput/eviction/writeback is made.
1779 	 */
1780 	trans = btrfs_attach_transaction(root);
1781 	if (IS_ERR(trans)) {
1782 		if (PTR_ERR(trans) != -ENOENT)
1783 			btrfs_err(fs_info,
1784 				  "cleaner transaction attach returned %ld",
1785 				  PTR_ERR(trans));
1786 	} else {
1787 		int ret;
1788 
1789 		ret = btrfs_commit_transaction(trans);
1790 		if (ret)
1791 			btrfs_err(fs_info,
1792 				  "cleaner open transaction commit returned %d",
1793 				  ret);
1794 	}
1795 
1796 	return 0;
1797 }
1798 
1799 static int transaction_kthread(void *arg)
1800 {
1801 	struct btrfs_root *root = arg;
1802 	struct btrfs_fs_info *fs_info = root->fs_info;
1803 	struct btrfs_trans_handle *trans;
1804 	struct btrfs_transaction *cur;
1805 	u64 transid;
1806 	unsigned long now;
1807 	unsigned long delay;
1808 	bool cannot_commit;
1809 
1810 	do {
1811 		cannot_commit = false;
1812 		delay = HZ * fs_info->commit_interval;
1813 		mutex_lock(&fs_info->transaction_kthread_mutex);
1814 
1815 		spin_lock(&fs_info->trans_lock);
1816 		cur = fs_info->running_transaction;
1817 		if (!cur) {
1818 			spin_unlock(&fs_info->trans_lock);
1819 			goto sleep;
1820 		}
1821 
1822 		now = get_seconds();
1823 		if (cur->state < TRANS_STATE_BLOCKED &&
1824 		    !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) &&
1825 		    (now < cur->start_time ||
1826 		     now - cur->start_time < fs_info->commit_interval)) {
1827 			spin_unlock(&fs_info->trans_lock);
1828 			delay = HZ * 5;
1829 			goto sleep;
1830 		}
1831 		transid = cur->transid;
1832 		spin_unlock(&fs_info->trans_lock);
1833 
1834 		/* If the file system is aborted, this will always fail. */
1835 		trans = btrfs_attach_transaction(root);
1836 		if (IS_ERR(trans)) {
1837 			if (PTR_ERR(trans) != -ENOENT)
1838 				cannot_commit = true;
1839 			goto sleep;
1840 		}
1841 		if (transid == trans->transid) {
1842 			btrfs_commit_transaction(trans);
1843 		} else {
1844 			btrfs_end_transaction(trans);
1845 		}
1846 sleep:
1847 		wake_up_process(fs_info->cleaner_kthread);
1848 		mutex_unlock(&fs_info->transaction_kthread_mutex);
1849 
1850 		if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1851 				      &fs_info->fs_state)))
1852 			btrfs_cleanup_transaction(fs_info);
1853 		if (!kthread_should_stop() &&
1854 				(!btrfs_transaction_blocked(fs_info) ||
1855 				 cannot_commit))
1856 			schedule_timeout_interruptible(delay);
1857 	} while (!kthread_should_stop());
1858 	return 0;
1859 }
1860 
1861 /*
1862  * this will find the highest generation in the array of
1863  * root backups.  The index of the highest array is returned,
1864  * or -1 if we can't find anything.
1865  *
1866  * We check to make sure the array is valid by comparing the
1867  * generation of the latest  root in the array with the generation
1868  * in the super block.  If they don't match we pitch it.
1869  */
1870 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1871 {
1872 	u64 cur;
1873 	int newest_index = -1;
1874 	struct btrfs_root_backup *root_backup;
1875 	int i;
1876 
1877 	for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1878 		root_backup = info->super_copy->super_roots + i;
1879 		cur = btrfs_backup_tree_root_gen(root_backup);
1880 		if (cur == newest_gen)
1881 			newest_index = i;
1882 	}
1883 
1884 	/* check to see if we actually wrapped around */
1885 	if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1886 		root_backup = info->super_copy->super_roots;
1887 		cur = btrfs_backup_tree_root_gen(root_backup);
1888 		if (cur == newest_gen)
1889 			newest_index = 0;
1890 	}
1891 	return newest_index;
1892 }
1893 
1894 
1895 /*
1896  * find the oldest backup so we know where to store new entries
1897  * in the backup array.  This will set the backup_root_index
1898  * field in the fs_info struct
1899  */
1900 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1901 				     u64 newest_gen)
1902 {
1903 	int newest_index = -1;
1904 
1905 	newest_index = find_newest_super_backup(info, newest_gen);
1906 	/* if there was garbage in there, just move along */
1907 	if (newest_index == -1) {
1908 		info->backup_root_index = 0;
1909 	} else {
1910 		info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1911 	}
1912 }
1913 
1914 /*
1915  * copy all the root pointers into the super backup array.
1916  * this will bump the backup pointer by one when it is
1917  * done
1918  */
1919 static void backup_super_roots(struct btrfs_fs_info *info)
1920 {
1921 	int next_backup;
1922 	struct btrfs_root_backup *root_backup;
1923 	int last_backup;
1924 
1925 	next_backup = info->backup_root_index;
1926 	last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1927 		BTRFS_NUM_BACKUP_ROOTS;
1928 
1929 	/*
1930 	 * just overwrite the last backup if we're at the same generation
1931 	 * this happens only at umount
1932 	 */
1933 	root_backup = info->super_for_commit->super_roots + last_backup;
1934 	if (btrfs_backup_tree_root_gen(root_backup) ==
1935 	    btrfs_header_generation(info->tree_root->node))
1936 		next_backup = last_backup;
1937 
1938 	root_backup = info->super_for_commit->super_roots + next_backup;
1939 
1940 	/*
1941 	 * make sure all of our padding and empty slots get zero filled
1942 	 * regardless of which ones we use today
1943 	 */
1944 	memset(root_backup, 0, sizeof(*root_backup));
1945 
1946 	info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1947 
1948 	btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1949 	btrfs_set_backup_tree_root_gen(root_backup,
1950 			       btrfs_header_generation(info->tree_root->node));
1951 
1952 	btrfs_set_backup_tree_root_level(root_backup,
1953 			       btrfs_header_level(info->tree_root->node));
1954 
1955 	btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1956 	btrfs_set_backup_chunk_root_gen(root_backup,
1957 			       btrfs_header_generation(info->chunk_root->node));
1958 	btrfs_set_backup_chunk_root_level(root_backup,
1959 			       btrfs_header_level(info->chunk_root->node));
1960 
1961 	btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1962 	btrfs_set_backup_extent_root_gen(root_backup,
1963 			       btrfs_header_generation(info->extent_root->node));
1964 	btrfs_set_backup_extent_root_level(root_backup,
1965 			       btrfs_header_level(info->extent_root->node));
1966 
1967 	/*
1968 	 * we might commit during log recovery, which happens before we set
1969 	 * the fs_root.  Make sure it is valid before we fill it in.
1970 	 */
1971 	if (info->fs_root && info->fs_root->node) {
1972 		btrfs_set_backup_fs_root(root_backup,
1973 					 info->fs_root->node->start);
1974 		btrfs_set_backup_fs_root_gen(root_backup,
1975 			       btrfs_header_generation(info->fs_root->node));
1976 		btrfs_set_backup_fs_root_level(root_backup,
1977 			       btrfs_header_level(info->fs_root->node));
1978 	}
1979 
1980 	btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1981 	btrfs_set_backup_dev_root_gen(root_backup,
1982 			       btrfs_header_generation(info->dev_root->node));
1983 	btrfs_set_backup_dev_root_level(root_backup,
1984 				       btrfs_header_level(info->dev_root->node));
1985 
1986 	btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1987 	btrfs_set_backup_csum_root_gen(root_backup,
1988 			       btrfs_header_generation(info->csum_root->node));
1989 	btrfs_set_backup_csum_root_level(root_backup,
1990 			       btrfs_header_level(info->csum_root->node));
1991 
1992 	btrfs_set_backup_total_bytes(root_backup,
1993 			     btrfs_super_total_bytes(info->super_copy));
1994 	btrfs_set_backup_bytes_used(root_backup,
1995 			     btrfs_super_bytes_used(info->super_copy));
1996 	btrfs_set_backup_num_devices(root_backup,
1997 			     btrfs_super_num_devices(info->super_copy));
1998 
1999 	/*
2000 	 * if we don't copy this out to the super_copy, it won't get remembered
2001 	 * for the next commit
2002 	 */
2003 	memcpy(&info->super_copy->super_roots,
2004 	       &info->super_for_commit->super_roots,
2005 	       sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2006 }
2007 
2008 /*
2009  * this copies info out of the root backup array and back into
2010  * the in-memory super block.  It is meant to help iterate through
2011  * the array, so you send it the number of backups you've already
2012  * tried and the last backup index you used.
2013  *
2014  * this returns -1 when it has tried all the backups
2015  */
2016 static noinline int next_root_backup(struct btrfs_fs_info *info,
2017 				     struct btrfs_super_block *super,
2018 				     int *num_backups_tried, int *backup_index)
2019 {
2020 	struct btrfs_root_backup *root_backup;
2021 	int newest = *backup_index;
2022 
2023 	if (*num_backups_tried == 0) {
2024 		u64 gen = btrfs_super_generation(super);
2025 
2026 		newest = find_newest_super_backup(info, gen);
2027 		if (newest == -1)
2028 			return -1;
2029 
2030 		*backup_index = newest;
2031 		*num_backups_tried = 1;
2032 	} else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2033 		/* we've tried all the backups, all done */
2034 		return -1;
2035 	} else {
2036 		/* jump to the next oldest backup */
2037 		newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2038 			BTRFS_NUM_BACKUP_ROOTS;
2039 		*backup_index = newest;
2040 		*num_backups_tried += 1;
2041 	}
2042 	root_backup = super->super_roots + newest;
2043 
2044 	btrfs_set_super_generation(super,
2045 				   btrfs_backup_tree_root_gen(root_backup));
2046 	btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2047 	btrfs_set_super_root_level(super,
2048 				   btrfs_backup_tree_root_level(root_backup));
2049 	btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2050 
2051 	/*
2052 	 * fixme: the total bytes and num_devices need to match or we should
2053 	 * need a fsck
2054 	 */
2055 	btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2056 	btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2057 	return 0;
2058 }
2059 
2060 /* helper to cleanup workers */
2061 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2062 {
2063 	btrfs_destroy_workqueue(fs_info->fixup_workers);
2064 	btrfs_destroy_workqueue(fs_info->delalloc_workers);
2065 	btrfs_destroy_workqueue(fs_info->workers);
2066 	btrfs_destroy_workqueue(fs_info->endio_workers);
2067 	btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2068 	btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2069 	btrfs_destroy_workqueue(fs_info->rmw_workers);
2070 	btrfs_destroy_workqueue(fs_info->endio_write_workers);
2071 	btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2072 	btrfs_destroy_workqueue(fs_info->submit_workers);
2073 	btrfs_destroy_workqueue(fs_info->delayed_workers);
2074 	btrfs_destroy_workqueue(fs_info->caching_workers);
2075 	btrfs_destroy_workqueue(fs_info->readahead_workers);
2076 	btrfs_destroy_workqueue(fs_info->flush_workers);
2077 	btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2078 	btrfs_destroy_workqueue(fs_info->extent_workers);
2079 	/*
2080 	 * Now that all other work queues are destroyed, we can safely destroy
2081 	 * the queues used for metadata I/O, since tasks from those other work
2082 	 * queues can do metadata I/O operations.
2083 	 */
2084 	btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2085 	btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2086 }
2087 
2088 static void free_root_extent_buffers(struct btrfs_root *root)
2089 {
2090 	if (root) {
2091 		free_extent_buffer(root->node);
2092 		free_extent_buffer(root->commit_root);
2093 		root->node = NULL;
2094 		root->commit_root = NULL;
2095 	}
2096 }
2097 
2098 /* helper to cleanup tree roots */
2099 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2100 {
2101 	free_root_extent_buffers(info->tree_root);
2102 
2103 	free_root_extent_buffers(info->dev_root);
2104 	free_root_extent_buffers(info->extent_root);
2105 	free_root_extent_buffers(info->csum_root);
2106 	free_root_extent_buffers(info->quota_root);
2107 	free_root_extent_buffers(info->uuid_root);
2108 	if (chunk_root)
2109 		free_root_extent_buffers(info->chunk_root);
2110 	free_root_extent_buffers(info->free_space_root);
2111 }
2112 
2113 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2114 {
2115 	int ret;
2116 	struct btrfs_root *gang[8];
2117 	int i;
2118 
2119 	while (!list_empty(&fs_info->dead_roots)) {
2120 		gang[0] = list_entry(fs_info->dead_roots.next,
2121 				     struct btrfs_root, root_list);
2122 		list_del(&gang[0]->root_list);
2123 
2124 		if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2125 			btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2126 		} else {
2127 			free_extent_buffer(gang[0]->node);
2128 			free_extent_buffer(gang[0]->commit_root);
2129 			btrfs_put_fs_root(gang[0]);
2130 		}
2131 	}
2132 
2133 	while (1) {
2134 		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2135 					     (void **)gang, 0,
2136 					     ARRAY_SIZE(gang));
2137 		if (!ret)
2138 			break;
2139 		for (i = 0; i < ret; i++)
2140 			btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2141 	}
2142 
2143 	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2144 		btrfs_free_log_root_tree(NULL, fs_info);
2145 		btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2146 	}
2147 }
2148 
2149 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2150 {
2151 	mutex_init(&fs_info->scrub_lock);
2152 	atomic_set(&fs_info->scrubs_running, 0);
2153 	atomic_set(&fs_info->scrub_pause_req, 0);
2154 	atomic_set(&fs_info->scrubs_paused, 0);
2155 	atomic_set(&fs_info->scrub_cancel_req, 0);
2156 	init_waitqueue_head(&fs_info->scrub_pause_wait);
2157 	fs_info->scrub_workers_refcnt = 0;
2158 }
2159 
2160 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2161 {
2162 	spin_lock_init(&fs_info->balance_lock);
2163 	mutex_init(&fs_info->balance_mutex);
2164 	atomic_set(&fs_info->balance_pause_req, 0);
2165 	atomic_set(&fs_info->balance_cancel_req, 0);
2166 	fs_info->balance_ctl = NULL;
2167 	init_waitqueue_head(&fs_info->balance_wait_q);
2168 }
2169 
2170 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2171 {
2172 	struct inode *inode = fs_info->btree_inode;
2173 
2174 	inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2175 	set_nlink(inode, 1);
2176 	/*
2177 	 * we set the i_size on the btree inode to the max possible int.
2178 	 * the real end of the address space is determined by all of
2179 	 * the devices in the system
2180 	 */
2181 	inode->i_size = OFFSET_MAX;
2182 	inode->i_mapping->a_ops = &btree_aops;
2183 
2184 	RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2185 	extent_io_tree_init(&BTRFS_I(inode)->io_tree, inode);
2186 	BTRFS_I(inode)->io_tree.track_uptodate = 0;
2187 	extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2188 
2189 	BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2190 
2191 	BTRFS_I(inode)->root = fs_info->tree_root;
2192 	memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2193 	set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2194 	btrfs_insert_inode_hash(inode);
2195 }
2196 
2197 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2198 {
2199 	fs_info->dev_replace.lock_owner = 0;
2200 	atomic_set(&fs_info->dev_replace.nesting_level, 0);
2201 	mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2202 	rwlock_init(&fs_info->dev_replace.lock);
2203 	atomic_set(&fs_info->dev_replace.read_locks, 0);
2204 	atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2205 	init_waitqueue_head(&fs_info->replace_wait);
2206 	init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2207 }
2208 
2209 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2210 {
2211 	spin_lock_init(&fs_info->qgroup_lock);
2212 	mutex_init(&fs_info->qgroup_ioctl_lock);
2213 	fs_info->qgroup_tree = RB_ROOT;
2214 	fs_info->qgroup_op_tree = RB_ROOT;
2215 	INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2216 	fs_info->qgroup_seq = 1;
2217 	fs_info->qgroup_ulist = NULL;
2218 	fs_info->qgroup_rescan_running = false;
2219 	mutex_init(&fs_info->qgroup_rescan_lock);
2220 }
2221 
2222 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2223 		struct btrfs_fs_devices *fs_devices)
2224 {
2225 	u32 max_active = fs_info->thread_pool_size;
2226 	unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2227 
2228 	fs_info->workers =
2229 		btrfs_alloc_workqueue(fs_info, "worker",
2230 				      flags | WQ_HIGHPRI, max_active, 16);
2231 
2232 	fs_info->delalloc_workers =
2233 		btrfs_alloc_workqueue(fs_info, "delalloc",
2234 				      flags, max_active, 2);
2235 
2236 	fs_info->flush_workers =
2237 		btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2238 				      flags, max_active, 0);
2239 
2240 	fs_info->caching_workers =
2241 		btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2242 
2243 	/*
2244 	 * a higher idle thresh on the submit workers makes it much more
2245 	 * likely that bios will be send down in a sane order to the
2246 	 * devices
2247 	 */
2248 	fs_info->submit_workers =
2249 		btrfs_alloc_workqueue(fs_info, "submit", flags,
2250 				      min_t(u64, fs_devices->num_devices,
2251 					    max_active), 64);
2252 
2253 	fs_info->fixup_workers =
2254 		btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2255 
2256 	/*
2257 	 * endios are largely parallel and should have a very
2258 	 * low idle thresh
2259 	 */
2260 	fs_info->endio_workers =
2261 		btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2262 	fs_info->endio_meta_workers =
2263 		btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2264 				      max_active, 4);
2265 	fs_info->endio_meta_write_workers =
2266 		btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2267 				      max_active, 2);
2268 	fs_info->endio_raid56_workers =
2269 		btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2270 				      max_active, 4);
2271 	fs_info->endio_repair_workers =
2272 		btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2273 	fs_info->rmw_workers =
2274 		btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2275 	fs_info->endio_write_workers =
2276 		btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2277 				      max_active, 2);
2278 	fs_info->endio_freespace_worker =
2279 		btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2280 				      max_active, 0);
2281 	fs_info->delayed_workers =
2282 		btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2283 				      max_active, 0);
2284 	fs_info->readahead_workers =
2285 		btrfs_alloc_workqueue(fs_info, "readahead", flags,
2286 				      max_active, 2);
2287 	fs_info->qgroup_rescan_workers =
2288 		btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2289 	fs_info->extent_workers =
2290 		btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2291 				      min_t(u64, fs_devices->num_devices,
2292 					    max_active), 8);
2293 
2294 	if (!(fs_info->workers && fs_info->delalloc_workers &&
2295 	      fs_info->submit_workers && fs_info->flush_workers &&
2296 	      fs_info->endio_workers && fs_info->endio_meta_workers &&
2297 	      fs_info->endio_meta_write_workers &&
2298 	      fs_info->endio_repair_workers &&
2299 	      fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2300 	      fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2301 	      fs_info->caching_workers && fs_info->readahead_workers &&
2302 	      fs_info->fixup_workers && fs_info->delayed_workers &&
2303 	      fs_info->extent_workers &&
2304 	      fs_info->qgroup_rescan_workers)) {
2305 		return -ENOMEM;
2306 	}
2307 
2308 	return 0;
2309 }
2310 
2311 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2312 			    struct btrfs_fs_devices *fs_devices)
2313 {
2314 	int ret;
2315 	struct btrfs_root *log_tree_root;
2316 	struct btrfs_super_block *disk_super = fs_info->super_copy;
2317 	u64 bytenr = btrfs_super_log_root(disk_super);
2318 	int level = btrfs_super_log_root_level(disk_super);
2319 
2320 	if (fs_devices->rw_devices == 0) {
2321 		btrfs_warn(fs_info, "log replay required on RO media");
2322 		return -EIO;
2323 	}
2324 
2325 	log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2326 	if (!log_tree_root)
2327 		return -ENOMEM;
2328 
2329 	__setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2330 
2331 	log_tree_root->node = read_tree_block(fs_info, bytenr,
2332 					      fs_info->generation + 1,
2333 					      level, NULL);
2334 	if (IS_ERR(log_tree_root->node)) {
2335 		btrfs_warn(fs_info, "failed to read log tree");
2336 		ret = PTR_ERR(log_tree_root->node);
2337 		kfree(log_tree_root);
2338 		return ret;
2339 	} else if (!extent_buffer_uptodate(log_tree_root->node)) {
2340 		btrfs_err(fs_info, "failed to read log tree");
2341 		free_extent_buffer(log_tree_root->node);
2342 		kfree(log_tree_root);
2343 		return -EIO;
2344 	}
2345 	/* returns with log_tree_root freed on success */
2346 	ret = btrfs_recover_log_trees(log_tree_root);
2347 	if (ret) {
2348 		btrfs_handle_fs_error(fs_info, ret,
2349 				      "Failed to recover log tree");
2350 		free_extent_buffer(log_tree_root->node);
2351 		kfree(log_tree_root);
2352 		return ret;
2353 	}
2354 
2355 	if (sb_rdonly(fs_info->sb)) {
2356 		ret = btrfs_commit_super(fs_info);
2357 		if (ret)
2358 			return ret;
2359 	}
2360 
2361 	return 0;
2362 }
2363 
2364 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2365 {
2366 	struct btrfs_root *tree_root = fs_info->tree_root;
2367 	struct btrfs_root *root;
2368 	struct btrfs_key location;
2369 	int ret;
2370 
2371 	BUG_ON(!fs_info->tree_root);
2372 
2373 	location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2374 	location.type = BTRFS_ROOT_ITEM_KEY;
2375 	location.offset = 0;
2376 
2377 	root = btrfs_read_tree_root(tree_root, &location);
2378 	if (IS_ERR(root)) {
2379 		ret = PTR_ERR(root);
2380 		goto out;
2381 	}
2382 	set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2383 	fs_info->extent_root = root;
2384 
2385 	location.objectid = BTRFS_DEV_TREE_OBJECTID;
2386 	root = btrfs_read_tree_root(tree_root, &location);
2387 	if (IS_ERR(root)) {
2388 		ret = PTR_ERR(root);
2389 		goto out;
2390 	}
2391 	set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2392 	fs_info->dev_root = root;
2393 	btrfs_init_devices_late(fs_info);
2394 
2395 	location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2396 	root = btrfs_read_tree_root(tree_root, &location);
2397 	if (IS_ERR(root)) {
2398 		ret = PTR_ERR(root);
2399 		goto out;
2400 	}
2401 	set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2402 	fs_info->csum_root = root;
2403 
2404 	location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2405 	root = btrfs_read_tree_root(tree_root, &location);
2406 	if (!IS_ERR(root)) {
2407 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2408 		set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2409 		fs_info->quota_root = root;
2410 	}
2411 
2412 	location.objectid = BTRFS_UUID_TREE_OBJECTID;
2413 	root = btrfs_read_tree_root(tree_root, &location);
2414 	if (IS_ERR(root)) {
2415 		ret = PTR_ERR(root);
2416 		if (ret != -ENOENT)
2417 			goto out;
2418 	} else {
2419 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2420 		fs_info->uuid_root = root;
2421 	}
2422 
2423 	if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2424 		location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2425 		root = btrfs_read_tree_root(tree_root, &location);
2426 		if (IS_ERR(root)) {
2427 			ret = PTR_ERR(root);
2428 			goto out;
2429 		}
2430 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2431 		fs_info->free_space_root = root;
2432 	}
2433 
2434 	return 0;
2435 out:
2436 	btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2437 		   location.objectid, ret);
2438 	return ret;
2439 }
2440 
2441 /*
2442  * Real super block validation
2443  * NOTE: super csum type and incompat features will not be checked here.
2444  *
2445  * @sb:		super block to check
2446  * @mirror_num:	the super block number to check its bytenr:
2447  * 		0	the primary (1st) sb
2448  * 		1, 2	2nd and 3rd backup copy
2449  * 	       -1	skip bytenr check
2450  */
2451 static int validate_super(struct btrfs_fs_info *fs_info,
2452 			    struct btrfs_super_block *sb, int mirror_num)
2453 {
2454 	u64 nodesize = btrfs_super_nodesize(sb);
2455 	u64 sectorsize = btrfs_super_sectorsize(sb);
2456 	int ret = 0;
2457 
2458 	if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2459 		btrfs_err(fs_info, "no valid FS found");
2460 		ret = -EINVAL;
2461 	}
2462 	if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2463 		btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2464 				btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2465 		ret = -EINVAL;
2466 	}
2467 	if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2468 		btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2469 				btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2470 		ret = -EINVAL;
2471 	}
2472 	if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2473 		btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2474 				btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2475 		ret = -EINVAL;
2476 	}
2477 	if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2478 		btrfs_err(fs_info, "log_root level too big: %d >= %d",
2479 				btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2480 		ret = -EINVAL;
2481 	}
2482 
2483 	/*
2484 	 * Check sectorsize and nodesize first, other check will need it.
2485 	 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2486 	 */
2487 	if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2488 	    sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2489 		btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2490 		ret = -EINVAL;
2491 	}
2492 	/* Only PAGE SIZE is supported yet */
2493 	if (sectorsize != PAGE_SIZE) {
2494 		btrfs_err(fs_info,
2495 			"sectorsize %llu not supported yet, only support %lu",
2496 			sectorsize, PAGE_SIZE);
2497 		ret = -EINVAL;
2498 	}
2499 	if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2500 	    nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2501 		btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2502 		ret = -EINVAL;
2503 	}
2504 	if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2505 		btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2506 			  le32_to_cpu(sb->__unused_leafsize), nodesize);
2507 		ret = -EINVAL;
2508 	}
2509 
2510 	/* Root alignment check */
2511 	if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2512 		btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2513 			   btrfs_super_root(sb));
2514 		ret = -EINVAL;
2515 	}
2516 	if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2517 		btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2518 			   btrfs_super_chunk_root(sb));
2519 		ret = -EINVAL;
2520 	}
2521 	if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2522 		btrfs_warn(fs_info, "log_root block unaligned: %llu",
2523 			   btrfs_super_log_root(sb));
2524 		ret = -EINVAL;
2525 	}
2526 
2527 	if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_FSID_SIZE) != 0) {
2528 		btrfs_err(fs_info,
2529 			   "dev_item UUID does not match fsid: %pU != %pU",
2530 			   fs_info->fsid, sb->dev_item.fsid);
2531 		ret = -EINVAL;
2532 	}
2533 
2534 	/*
2535 	 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2536 	 * done later
2537 	 */
2538 	if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2539 		btrfs_err(fs_info, "bytes_used is too small %llu",
2540 			  btrfs_super_bytes_used(sb));
2541 		ret = -EINVAL;
2542 	}
2543 	if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2544 		btrfs_err(fs_info, "invalid stripesize %u",
2545 			  btrfs_super_stripesize(sb));
2546 		ret = -EINVAL;
2547 	}
2548 	if (btrfs_super_num_devices(sb) > (1UL << 31))
2549 		btrfs_warn(fs_info, "suspicious number of devices: %llu",
2550 			   btrfs_super_num_devices(sb));
2551 	if (btrfs_super_num_devices(sb) == 0) {
2552 		btrfs_err(fs_info, "number of devices is 0");
2553 		ret = -EINVAL;
2554 	}
2555 
2556 	if (mirror_num >= 0 &&
2557 	    btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2558 		btrfs_err(fs_info, "super offset mismatch %llu != %u",
2559 			  btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2560 		ret = -EINVAL;
2561 	}
2562 
2563 	/*
2564 	 * Obvious sys_chunk_array corruptions, it must hold at least one key
2565 	 * and one chunk
2566 	 */
2567 	if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2568 		btrfs_err(fs_info, "system chunk array too big %u > %u",
2569 			  btrfs_super_sys_array_size(sb),
2570 			  BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2571 		ret = -EINVAL;
2572 	}
2573 	if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2574 			+ sizeof(struct btrfs_chunk)) {
2575 		btrfs_err(fs_info, "system chunk array too small %u < %zu",
2576 			  btrfs_super_sys_array_size(sb),
2577 			  sizeof(struct btrfs_disk_key)
2578 			  + sizeof(struct btrfs_chunk));
2579 		ret = -EINVAL;
2580 	}
2581 
2582 	/*
2583 	 * The generation is a global counter, we'll trust it more than the others
2584 	 * but it's still possible that it's the one that's wrong.
2585 	 */
2586 	if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2587 		btrfs_warn(fs_info,
2588 			"suspicious: generation < chunk_root_generation: %llu < %llu",
2589 			btrfs_super_generation(sb),
2590 			btrfs_super_chunk_root_generation(sb));
2591 	if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2592 	    && btrfs_super_cache_generation(sb) != (u64)-1)
2593 		btrfs_warn(fs_info,
2594 			"suspicious: generation < cache_generation: %llu < %llu",
2595 			btrfs_super_generation(sb),
2596 			btrfs_super_cache_generation(sb));
2597 
2598 	return ret;
2599 }
2600 
2601 /*
2602  * Validation of super block at mount time.
2603  * Some checks already done early at mount time, like csum type and incompat
2604  * flags will be skipped.
2605  */
2606 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2607 {
2608 	return validate_super(fs_info, fs_info->super_copy, 0);
2609 }
2610 
2611 /*
2612  * Validation of super block at write time.
2613  * Some checks like bytenr check will be skipped as their values will be
2614  * overwritten soon.
2615  * Extra checks like csum type and incompat flags will be done here.
2616  */
2617 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2618 				      struct btrfs_super_block *sb)
2619 {
2620 	int ret;
2621 
2622 	ret = validate_super(fs_info, sb, -1);
2623 	if (ret < 0)
2624 		goto out;
2625 	if (btrfs_super_csum_type(sb) != BTRFS_CSUM_TYPE_CRC32) {
2626 		ret = -EUCLEAN;
2627 		btrfs_err(fs_info, "invalid csum type, has %u want %u",
2628 			  btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2629 		goto out;
2630 	}
2631 	if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2632 		ret = -EUCLEAN;
2633 		btrfs_err(fs_info,
2634 		"invalid incompat flags, has 0x%llx valid mask 0x%llx",
2635 			  btrfs_super_incompat_flags(sb),
2636 			  (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2637 		goto out;
2638 	}
2639 out:
2640 	if (ret < 0)
2641 		btrfs_err(fs_info,
2642 		"super block corruption detected before writing it to disk");
2643 	return ret;
2644 }
2645 
2646 int open_ctree(struct super_block *sb,
2647 	       struct btrfs_fs_devices *fs_devices,
2648 	       char *options)
2649 {
2650 	u32 sectorsize;
2651 	u32 nodesize;
2652 	u32 stripesize;
2653 	u64 generation;
2654 	u64 features;
2655 	struct btrfs_key location;
2656 	struct buffer_head *bh;
2657 	struct btrfs_super_block *disk_super;
2658 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2659 	struct btrfs_root *tree_root;
2660 	struct btrfs_root *chunk_root;
2661 	int ret;
2662 	int err = -EINVAL;
2663 	int num_backups_tried = 0;
2664 	int backup_index = 0;
2665 	int clear_free_space_tree = 0;
2666 	int level;
2667 
2668 	tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2669 	chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2670 	if (!tree_root || !chunk_root) {
2671 		err = -ENOMEM;
2672 		goto fail;
2673 	}
2674 
2675 	ret = init_srcu_struct(&fs_info->subvol_srcu);
2676 	if (ret) {
2677 		err = ret;
2678 		goto fail;
2679 	}
2680 
2681 	ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2682 	if (ret) {
2683 		err = ret;
2684 		goto fail_srcu;
2685 	}
2686 	fs_info->dirty_metadata_batch = PAGE_SIZE *
2687 					(1 + ilog2(nr_cpu_ids));
2688 
2689 	ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2690 	if (ret) {
2691 		err = ret;
2692 		goto fail_dirty_metadata_bytes;
2693 	}
2694 
2695 	ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2696 	if (ret) {
2697 		err = ret;
2698 		goto fail_delalloc_bytes;
2699 	}
2700 
2701 	INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2702 	INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2703 	INIT_LIST_HEAD(&fs_info->trans_list);
2704 	INIT_LIST_HEAD(&fs_info->dead_roots);
2705 	INIT_LIST_HEAD(&fs_info->delayed_iputs);
2706 	INIT_LIST_HEAD(&fs_info->delalloc_roots);
2707 	INIT_LIST_HEAD(&fs_info->caching_block_groups);
2708 	INIT_LIST_HEAD(&fs_info->pending_raid_kobjs);
2709 	spin_lock_init(&fs_info->pending_raid_kobjs_lock);
2710 	spin_lock_init(&fs_info->delalloc_root_lock);
2711 	spin_lock_init(&fs_info->trans_lock);
2712 	spin_lock_init(&fs_info->fs_roots_radix_lock);
2713 	spin_lock_init(&fs_info->delayed_iput_lock);
2714 	spin_lock_init(&fs_info->defrag_inodes_lock);
2715 	spin_lock_init(&fs_info->tree_mod_seq_lock);
2716 	spin_lock_init(&fs_info->super_lock);
2717 	spin_lock_init(&fs_info->qgroup_op_lock);
2718 	spin_lock_init(&fs_info->buffer_lock);
2719 	spin_lock_init(&fs_info->unused_bgs_lock);
2720 	rwlock_init(&fs_info->tree_mod_log_lock);
2721 	mutex_init(&fs_info->unused_bg_unpin_mutex);
2722 	mutex_init(&fs_info->delete_unused_bgs_mutex);
2723 	mutex_init(&fs_info->reloc_mutex);
2724 	mutex_init(&fs_info->delalloc_root_mutex);
2725 	mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2726 	seqlock_init(&fs_info->profiles_lock);
2727 
2728 	INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2729 	INIT_LIST_HEAD(&fs_info->space_info);
2730 	INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2731 	INIT_LIST_HEAD(&fs_info->unused_bgs);
2732 	btrfs_mapping_init(&fs_info->mapping_tree);
2733 	btrfs_init_block_rsv(&fs_info->global_block_rsv,
2734 			     BTRFS_BLOCK_RSV_GLOBAL);
2735 	btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2736 	btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2737 	btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2738 	btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2739 			     BTRFS_BLOCK_RSV_DELOPS);
2740 	atomic_set(&fs_info->async_delalloc_pages, 0);
2741 	atomic_set(&fs_info->defrag_running, 0);
2742 	atomic_set(&fs_info->qgroup_op_seq, 0);
2743 	atomic_set(&fs_info->reada_works_cnt, 0);
2744 	atomic64_set(&fs_info->tree_mod_seq, 0);
2745 	fs_info->sb = sb;
2746 	fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2747 	fs_info->metadata_ratio = 0;
2748 	fs_info->defrag_inodes = RB_ROOT;
2749 	atomic64_set(&fs_info->free_chunk_space, 0);
2750 	fs_info->tree_mod_log = RB_ROOT;
2751 	fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2752 	fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2753 	/* readahead state */
2754 	INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2755 	spin_lock_init(&fs_info->reada_lock);
2756 	btrfs_init_ref_verify(fs_info);
2757 
2758 	fs_info->thread_pool_size = min_t(unsigned long,
2759 					  num_online_cpus() + 2, 8);
2760 
2761 	INIT_LIST_HEAD(&fs_info->ordered_roots);
2762 	spin_lock_init(&fs_info->ordered_root_lock);
2763 
2764 	fs_info->btree_inode = new_inode(sb);
2765 	if (!fs_info->btree_inode) {
2766 		err = -ENOMEM;
2767 		goto fail_bio_counter;
2768 	}
2769 	mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2770 
2771 	fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2772 					GFP_KERNEL);
2773 	if (!fs_info->delayed_root) {
2774 		err = -ENOMEM;
2775 		goto fail_iput;
2776 	}
2777 	btrfs_init_delayed_root(fs_info->delayed_root);
2778 
2779 	btrfs_init_scrub(fs_info);
2780 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2781 	fs_info->check_integrity_print_mask = 0;
2782 #endif
2783 	btrfs_init_balance(fs_info);
2784 	btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2785 
2786 	sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2787 	sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2788 
2789 	btrfs_init_btree_inode(fs_info);
2790 
2791 	spin_lock_init(&fs_info->block_group_cache_lock);
2792 	fs_info->block_group_cache_tree = RB_ROOT;
2793 	fs_info->first_logical_byte = (u64)-1;
2794 
2795 	extent_io_tree_init(&fs_info->freed_extents[0], NULL);
2796 	extent_io_tree_init(&fs_info->freed_extents[1], NULL);
2797 	fs_info->pinned_extents = &fs_info->freed_extents[0];
2798 	set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2799 
2800 	mutex_init(&fs_info->ordered_operations_mutex);
2801 	mutex_init(&fs_info->tree_log_mutex);
2802 	mutex_init(&fs_info->chunk_mutex);
2803 	mutex_init(&fs_info->transaction_kthread_mutex);
2804 	mutex_init(&fs_info->cleaner_mutex);
2805 	mutex_init(&fs_info->ro_block_group_mutex);
2806 	init_rwsem(&fs_info->commit_root_sem);
2807 	init_rwsem(&fs_info->cleanup_work_sem);
2808 	init_rwsem(&fs_info->subvol_sem);
2809 	sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2810 
2811 	btrfs_init_dev_replace_locks(fs_info);
2812 	btrfs_init_qgroup(fs_info);
2813 
2814 	btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2815 	btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2816 
2817 	init_waitqueue_head(&fs_info->transaction_throttle);
2818 	init_waitqueue_head(&fs_info->transaction_wait);
2819 	init_waitqueue_head(&fs_info->transaction_blocked_wait);
2820 	init_waitqueue_head(&fs_info->async_submit_wait);
2821 
2822 	INIT_LIST_HEAD(&fs_info->pinned_chunks);
2823 
2824 	/* Usable values until the real ones are cached from the superblock */
2825 	fs_info->nodesize = 4096;
2826 	fs_info->sectorsize = 4096;
2827 	fs_info->stripesize = 4096;
2828 
2829 	ret = btrfs_alloc_stripe_hash_table(fs_info);
2830 	if (ret) {
2831 		err = ret;
2832 		goto fail_alloc;
2833 	}
2834 
2835 	__setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2836 
2837 	invalidate_bdev(fs_devices->latest_bdev);
2838 
2839 	/*
2840 	 * Read super block and check the signature bytes only
2841 	 */
2842 	bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2843 	if (IS_ERR(bh)) {
2844 		err = PTR_ERR(bh);
2845 		goto fail_alloc;
2846 	}
2847 
2848 	/*
2849 	 * We want to check superblock checksum, the type is stored inside.
2850 	 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2851 	 */
2852 	if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2853 		btrfs_err(fs_info, "superblock checksum mismatch");
2854 		err = -EINVAL;
2855 		brelse(bh);
2856 		goto fail_alloc;
2857 	}
2858 
2859 	/*
2860 	 * super_copy is zeroed at allocation time and we never touch the
2861 	 * following bytes up to INFO_SIZE, the checksum is calculated from
2862 	 * the whole block of INFO_SIZE
2863 	 */
2864 	memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2865 	memcpy(fs_info->super_for_commit, fs_info->super_copy,
2866 	       sizeof(*fs_info->super_for_commit));
2867 	brelse(bh);
2868 
2869 	memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2870 
2871 	ret = btrfs_validate_mount_super(fs_info);
2872 	if (ret) {
2873 		btrfs_err(fs_info, "superblock contains fatal errors");
2874 		err = -EINVAL;
2875 		goto fail_alloc;
2876 	}
2877 
2878 	disk_super = fs_info->super_copy;
2879 	if (!btrfs_super_root(disk_super))
2880 		goto fail_alloc;
2881 
2882 	/* check FS state, whether FS is broken. */
2883 	if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2884 		set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2885 
2886 	/*
2887 	 * run through our array of backup supers and setup
2888 	 * our ring pointer to the oldest one
2889 	 */
2890 	generation = btrfs_super_generation(disk_super);
2891 	find_oldest_super_backup(fs_info, generation);
2892 
2893 	/*
2894 	 * In the long term, we'll store the compression type in the super
2895 	 * block, and it'll be used for per file compression control.
2896 	 */
2897 	fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2898 
2899 	ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2900 	if (ret) {
2901 		err = ret;
2902 		goto fail_alloc;
2903 	}
2904 
2905 	features = btrfs_super_incompat_flags(disk_super) &
2906 		~BTRFS_FEATURE_INCOMPAT_SUPP;
2907 	if (features) {
2908 		btrfs_err(fs_info,
2909 		    "cannot mount because of unsupported optional features (%llx)",
2910 		    features);
2911 		err = -EINVAL;
2912 		goto fail_alloc;
2913 	}
2914 
2915 	features = btrfs_super_incompat_flags(disk_super);
2916 	features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2917 	if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2918 		features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2919 	else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2920 		features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2921 
2922 	if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2923 		btrfs_info(fs_info, "has skinny extents");
2924 
2925 	/*
2926 	 * flag our filesystem as having big metadata blocks if
2927 	 * they are bigger than the page size
2928 	 */
2929 	if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2930 		if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2931 			btrfs_info(fs_info,
2932 				"flagging fs with big metadata feature");
2933 		features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2934 	}
2935 
2936 	nodesize = btrfs_super_nodesize(disk_super);
2937 	sectorsize = btrfs_super_sectorsize(disk_super);
2938 	stripesize = sectorsize;
2939 	fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2940 	fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2941 
2942 	/* Cache block sizes */
2943 	fs_info->nodesize = nodesize;
2944 	fs_info->sectorsize = sectorsize;
2945 	fs_info->stripesize = stripesize;
2946 
2947 	/*
2948 	 * mixed block groups end up with duplicate but slightly offset
2949 	 * extent buffers for the same range.  It leads to corruptions
2950 	 */
2951 	if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2952 	    (sectorsize != nodesize)) {
2953 		btrfs_err(fs_info,
2954 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2955 			nodesize, sectorsize);
2956 		goto fail_alloc;
2957 	}
2958 
2959 	/*
2960 	 * Needn't use the lock because there is no other task which will
2961 	 * update the flag.
2962 	 */
2963 	btrfs_set_super_incompat_flags(disk_super, features);
2964 
2965 	features = btrfs_super_compat_ro_flags(disk_super) &
2966 		~BTRFS_FEATURE_COMPAT_RO_SUPP;
2967 	if (!sb_rdonly(sb) && features) {
2968 		btrfs_err(fs_info,
2969 	"cannot mount read-write because of unsupported optional features (%llx)",
2970 		       features);
2971 		err = -EINVAL;
2972 		goto fail_alloc;
2973 	}
2974 
2975 	ret = btrfs_init_workqueues(fs_info, fs_devices);
2976 	if (ret) {
2977 		err = ret;
2978 		goto fail_sb_buffer;
2979 	}
2980 
2981 	sb->s_bdi->congested_fn = btrfs_congested_fn;
2982 	sb->s_bdi->congested_data = fs_info;
2983 	sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2984 	sb->s_bdi->ra_pages = VM_MAX_READAHEAD * SZ_1K / PAGE_SIZE;
2985 	sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2986 	sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2987 
2988 	sb->s_blocksize = sectorsize;
2989 	sb->s_blocksize_bits = blksize_bits(sectorsize);
2990 	memcpy(&sb->s_uuid, fs_info->fsid, BTRFS_FSID_SIZE);
2991 
2992 	mutex_lock(&fs_info->chunk_mutex);
2993 	ret = btrfs_read_sys_array(fs_info);
2994 	mutex_unlock(&fs_info->chunk_mutex);
2995 	if (ret) {
2996 		btrfs_err(fs_info, "failed to read the system array: %d", ret);
2997 		goto fail_sb_buffer;
2998 	}
2999 
3000 	generation = btrfs_super_chunk_root_generation(disk_super);
3001 	level = btrfs_super_chunk_root_level(disk_super);
3002 
3003 	__setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
3004 
3005 	chunk_root->node = read_tree_block(fs_info,
3006 					   btrfs_super_chunk_root(disk_super),
3007 					   generation, level, NULL);
3008 	if (IS_ERR(chunk_root->node) ||
3009 	    !extent_buffer_uptodate(chunk_root->node)) {
3010 		btrfs_err(fs_info, "failed to read chunk root");
3011 		if (!IS_ERR(chunk_root->node))
3012 			free_extent_buffer(chunk_root->node);
3013 		chunk_root->node = NULL;
3014 		goto fail_tree_roots;
3015 	}
3016 	btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
3017 	chunk_root->commit_root = btrfs_root_node(chunk_root);
3018 
3019 	read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3020 	   btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
3021 
3022 	ret = btrfs_read_chunk_tree(fs_info);
3023 	if (ret) {
3024 		btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3025 		goto fail_tree_roots;
3026 	}
3027 
3028 	/*
3029 	 * Keep the devid that is marked to be the target device for the
3030 	 * device replace procedure
3031 	 */
3032 	btrfs_free_extra_devids(fs_devices, 0);
3033 
3034 	if (!fs_devices->latest_bdev) {
3035 		btrfs_err(fs_info, "failed to read devices");
3036 		goto fail_tree_roots;
3037 	}
3038 
3039 retry_root_backup:
3040 	generation = btrfs_super_generation(disk_super);
3041 	level = btrfs_super_root_level(disk_super);
3042 
3043 	tree_root->node = read_tree_block(fs_info,
3044 					  btrfs_super_root(disk_super),
3045 					  generation, level, NULL);
3046 	if (IS_ERR(tree_root->node) ||
3047 	    !extent_buffer_uptodate(tree_root->node)) {
3048 		btrfs_warn(fs_info, "failed to read tree root");
3049 		if (!IS_ERR(tree_root->node))
3050 			free_extent_buffer(tree_root->node);
3051 		tree_root->node = NULL;
3052 		goto recovery_tree_root;
3053 	}
3054 
3055 	btrfs_set_root_node(&tree_root->root_item, tree_root->node);
3056 	tree_root->commit_root = btrfs_root_node(tree_root);
3057 	btrfs_set_root_refs(&tree_root->root_item, 1);
3058 
3059 	mutex_lock(&tree_root->objectid_mutex);
3060 	ret = btrfs_find_highest_objectid(tree_root,
3061 					&tree_root->highest_objectid);
3062 	if (ret) {
3063 		mutex_unlock(&tree_root->objectid_mutex);
3064 		goto recovery_tree_root;
3065 	}
3066 
3067 	ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
3068 
3069 	mutex_unlock(&tree_root->objectid_mutex);
3070 
3071 	ret = btrfs_read_roots(fs_info);
3072 	if (ret)
3073 		goto recovery_tree_root;
3074 
3075 	fs_info->generation = generation;
3076 	fs_info->last_trans_committed = generation;
3077 
3078 	ret = btrfs_recover_balance(fs_info);
3079 	if (ret) {
3080 		btrfs_err(fs_info, "failed to recover balance: %d", ret);
3081 		goto fail_block_groups;
3082 	}
3083 
3084 	ret = btrfs_init_dev_stats(fs_info);
3085 	if (ret) {
3086 		btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3087 		goto fail_block_groups;
3088 	}
3089 
3090 	ret = btrfs_init_dev_replace(fs_info);
3091 	if (ret) {
3092 		btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3093 		goto fail_block_groups;
3094 	}
3095 
3096 	btrfs_free_extra_devids(fs_devices, 1);
3097 
3098 	ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
3099 	if (ret) {
3100 		btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3101 				ret);
3102 		goto fail_block_groups;
3103 	}
3104 
3105 	ret = btrfs_sysfs_add_device(fs_devices);
3106 	if (ret) {
3107 		btrfs_err(fs_info, "failed to init sysfs device interface: %d",
3108 				ret);
3109 		goto fail_fsdev_sysfs;
3110 	}
3111 
3112 	ret = btrfs_sysfs_add_mounted(fs_info);
3113 	if (ret) {
3114 		btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3115 		goto fail_fsdev_sysfs;
3116 	}
3117 
3118 	ret = btrfs_init_space_info(fs_info);
3119 	if (ret) {
3120 		btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3121 		goto fail_sysfs;
3122 	}
3123 
3124 	ret = btrfs_read_block_groups(fs_info);
3125 	if (ret) {
3126 		btrfs_err(fs_info, "failed to read block groups: %d", ret);
3127 		goto fail_sysfs;
3128 	}
3129 
3130 	if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3131 		btrfs_warn(fs_info,
3132 		"writeable mount is not allowed due to too many missing devices");
3133 		goto fail_sysfs;
3134 	}
3135 
3136 	fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3137 					       "btrfs-cleaner");
3138 	if (IS_ERR(fs_info->cleaner_kthread))
3139 		goto fail_sysfs;
3140 
3141 	fs_info->transaction_kthread = kthread_run(transaction_kthread,
3142 						   tree_root,
3143 						   "btrfs-transaction");
3144 	if (IS_ERR(fs_info->transaction_kthread))
3145 		goto fail_cleaner;
3146 
3147 	if (!btrfs_test_opt(fs_info, NOSSD) &&
3148 	    !fs_info->fs_devices->rotating) {
3149 		btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3150 	}
3151 
3152 	/*
3153 	 * Mount does not set all options immediately, we can do it now and do
3154 	 * not have to wait for transaction commit
3155 	 */
3156 	btrfs_apply_pending_changes(fs_info);
3157 
3158 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3159 	if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3160 		ret = btrfsic_mount(fs_info, fs_devices,
3161 				    btrfs_test_opt(fs_info,
3162 					CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3163 				    1 : 0,
3164 				    fs_info->check_integrity_print_mask);
3165 		if (ret)
3166 			btrfs_warn(fs_info,
3167 				"failed to initialize integrity check module: %d",
3168 				ret);
3169 	}
3170 #endif
3171 	ret = btrfs_read_qgroup_config(fs_info);
3172 	if (ret)
3173 		goto fail_trans_kthread;
3174 
3175 	if (btrfs_build_ref_tree(fs_info))
3176 		btrfs_err(fs_info, "couldn't build ref tree");
3177 
3178 	/* do not make disk changes in broken FS or nologreplay is given */
3179 	if (btrfs_super_log_root(disk_super) != 0 &&
3180 	    !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3181 		ret = btrfs_replay_log(fs_info, fs_devices);
3182 		if (ret) {
3183 			err = ret;
3184 			goto fail_qgroup;
3185 		}
3186 	}
3187 
3188 	ret = btrfs_find_orphan_roots(fs_info);
3189 	if (ret)
3190 		goto fail_qgroup;
3191 
3192 	if (!sb_rdonly(sb)) {
3193 		ret = btrfs_cleanup_fs_roots(fs_info);
3194 		if (ret)
3195 			goto fail_qgroup;
3196 
3197 		mutex_lock(&fs_info->cleaner_mutex);
3198 		ret = btrfs_recover_relocation(tree_root);
3199 		mutex_unlock(&fs_info->cleaner_mutex);
3200 		if (ret < 0) {
3201 			btrfs_warn(fs_info, "failed to recover relocation: %d",
3202 					ret);
3203 			err = -EINVAL;
3204 			goto fail_qgroup;
3205 		}
3206 	}
3207 
3208 	location.objectid = BTRFS_FS_TREE_OBJECTID;
3209 	location.type = BTRFS_ROOT_ITEM_KEY;
3210 	location.offset = 0;
3211 
3212 	fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3213 	if (IS_ERR(fs_info->fs_root)) {
3214 		err = PTR_ERR(fs_info->fs_root);
3215 		btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3216 		goto fail_qgroup;
3217 	}
3218 
3219 	if (sb_rdonly(sb))
3220 		return 0;
3221 
3222 	if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3223 	    btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3224 		clear_free_space_tree = 1;
3225 	} else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3226 		   !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3227 		btrfs_warn(fs_info, "free space tree is invalid");
3228 		clear_free_space_tree = 1;
3229 	}
3230 
3231 	if (clear_free_space_tree) {
3232 		btrfs_info(fs_info, "clearing free space tree");
3233 		ret = btrfs_clear_free_space_tree(fs_info);
3234 		if (ret) {
3235 			btrfs_warn(fs_info,
3236 				   "failed to clear free space tree: %d", ret);
3237 			close_ctree(fs_info);
3238 			return ret;
3239 		}
3240 	}
3241 
3242 	if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3243 	    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3244 		btrfs_info(fs_info, "creating free space tree");
3245 		ret = btrfs_create_free_space_tree(fs_info);
3246 		if (ret) {
3247 			btrfs_warn(fs_info,
3248 				"failed to create free space tree: %d", ret);
3249 			close_ctree(fs_info);
3250 			return ret;
3251 		}
3252 	}
3253 
3254 	down_read(&fs_info->cleanup_work_sem);
3255 	if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3256 	    (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3257 		up_read(&fs_info->cleanup_work_sem);
3258 		close_ctree(fs_info);
3259 		return ret;
3260 	}
3261 	up_read(&fs_info->cleanup_work_sem);
3262 
3263 	ret = btrfs_resume_balance_async(fs_info);
3264 	if (ret) {
3265 		btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3266 		close_ctree(fs_info);
3267 		return ret;
3268 	}
3269 
3270 	ret = btrfs_resume_dev_replace_async(fs_info);
3271 	if (ret) {
3272 		btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3273 		close_ctree(fs_info);
3274 		return ret;
3275 	}
3276 
3277 	btrfs_qgroup_rescan_resume(fs_info);
3278 
3279 	if (!fs_info->uuid_root) {
3280 		btrfs_info(fs_info, "creating UUID tree");
3281 		ret = btrfs_create_uuid_tree(fs_info);
3282 		if (ret) {
3283 			btrfs_warn(fs_info,
3284 				"failed to create the UUID tree: %d", ret);
3285 			close_ctree(fs_info);
3286 			return ret;
3287 		}
3288 	} else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3289 		   fs_info->generation !=
3290 				btrfs_super_uuid_tree_generation(disk_super)) {
3291 		btrfs_info(fs_info, "checking UUID tree");
3292 		ret = btrfs_check_uuid_tree(fs_info);
3293 		if (ret) {
3294 			btrfs_warn(fs_info,
3295 				"failed to check the UUID tree: %d", ret);
3296 			close_ctree(fs_info);
3297 			return ret;
3298 		}
3299 	} else {
3300 		set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3301 	}
3302 	set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3303 
3304 	/*
3305 	 * backuproot only affect mount behavior, and if open_ctree succeeded,
3306 	 * no need to keep the flag
3307 	 */
3308 	btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3309 
3310 	return 0;
3311 
3312 fail_qgroup:
3313 	btrfs_free_qgroup_config(fs_info);
3314 fail_trans_kthread:
3315 	kthread_stop(fs_info->transaction_kthread);
3316 	btrfs_cleanup_transaction(fs_info);
3317 	btrfs_free_fs_roots(fs_info);
3318 fail_cleaner:
3319 	kthread_stop(fs_info->cleaner_kthread);
3320 
3321 	/*
3322 	 * make sure we're done with the btree inode before we stop our
3323 	 * kthreads
3324 	 */
3325 	filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3326 
3327 fail_sysfs:
3328 	btrfs_sysfs_remove_mounted(fs_info);
3329 
3330 fail_fsdev_sysfs:
3331 	btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3332 
3333 fail_block_groups:
3334 	btrfs_put_block_group_cache(fs_info);
3335 
3336 fail_tree_roots:
3337 	free_root_pointers(fs_info, 1);
3338 	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3339 
3340 fail_sb_buffer:
3341 	btrfs_stop_all_workers(fs_info);
3342 	btrfs_free_block_groups(fs_info);
3343 fail_alloc:
3344 fail_iput:
3345 	btrfs_mapping_tree_free(&fs_info->mapping_tree);
3346 
3347 	iput(fs_info->btree_inode);
3348 fail_bio_counter:
3349 	percpu_counter_destroy(&fs_info->bio_counter);
3350 fail_delalloc_bytes:
3351 	percpu_counter_destroy(&fs_info->delalloc_bytes);
3352 fail_dirty_metadata_bytes:
3353 	percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3354 fail_srcu:
3355 	cleanup_srcu_struct(&fs_info->subvol_srcu);
3356 fail:
3357 	btrfs_free_stripe_hash_table(fs_info);
3358 	btrfs_close_devices(fs_info->fs_devices);
3359 	return err;
3360 
3361 recovery_tree_root:
3362 	if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3363 		goto fail_tree_roots;
3364 
3365 	free_root_pointers(fs_info, 0);
3366 
3367 	/* don't use the log in recovery mode, it won't be valid */
3368 	btrfs_set_super_log_root(disk_super, 0);
3369 
3370 	/* we can't trust the free space cache either */
3371 	btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3372 
3373 	ret = next_root_backup(fs_info, fs_info->super_copy,
3374 			       &num_backups_tried, &backup_index);
3375 	if (ret == -1)
3376 		goto fail_block_groups;
3377 	goto retry_root_backup;
3378 }
3379 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3380 
3381 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3382 {
3383 	if (uptodate) {
3384 		set_buffer_uptodate(bh);
3385 	} else {
3386 		struct btrfs_device *device = (struct btrfs_device *)
3387 			bh->b_private;
3388 
3389 		btrfs_warn_rl_in_rcu(device->fs_info,
3390 				"lost page write due to IO error on %s",
3391 					  rcu_str_deref(device->name));
3392 		/* note, we don't set_buffer_write_io_error because we have
3393 		 * our own ways of dealing with the IO errors
3394 		 */
3395 		clear_buffer_uptodate(bh);
3396 		btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3397 	}
3398 	unlock_buffer(bh);
3399 	put_bh(bh);
3400 }
3401 
3402 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3403 			struct buffer_head **bh_ret)
3404 {
3405 	struct buffer_head *bh;
3406 	struct btrfs_super_block *super;
3407 	u64 bytenr;
3408 
3409 	bytenr = btrfs_sb_offset(copy_num);
3410 	if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3411 		return -EINVAL;
3412 
3413 	bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE);
3414 	/*
3415 	 * If we fail to read from the underlying devices, as of now
3416 	 * the best option we have is to mark it EIO.
3417 	 */
3418 	if (!bh)
3419 		return -EIO;
3420 
3421 	super = (struct btrfs_super_block *)bh->b_data;
3422 	if (btrfs_super_bytenr(super) != bytenr ||
3423 		    btrfs_super_magic(super) != BTRFS_MAGIC) {
3424 		brelse(bh);
3425 		return -EINVAL;
3426 	}
3427 
3428 	*bh_ret = bh;
3429 	return 0;
3430 }
3431 
3432 
3433 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3434 {
3435 	struct buffer_head *bh;
3436 	struct buffer_head *latest = NULL;
3437 	struct btrfs_super_block *super;
3438 	int i;
3439 	u64 transid = 0;
3440 	int ret = -EINVAL;
3441 
3442 	/* we would like to check all the supers, but that would make
3443 	 * a btrfs mount succeed after a mkfs from a different FS.
3444 	 * So, we need to add a special mount option to scan for
3445 	 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3446 	 */
3447 	for (i = 0; i < 1; i++) {
3448 		ret = btrfs_read_dev_one_super(bdev, i, &bh);
3449 		if (ret)
3450 			continue;
3451 
3452 		super = (struct btrfs_super_block *)bh->b_data;
3453 
3454 		if (!latest || btrfs_super_generation(super) > transid) {
3455 			brelse(latest);
3456 			latest = bh;
3457 			transid = btrfs_super_generation(super);
3458 		} else {
3459 			brelse(bh);
3460 		}
3461 	}
3462 
3463 	if (!latest)
3464 		return ERR_PTR(ret);
3465 
3466 	return latest;
3467 }
3468 
3469 /*
3470  * Write superblock @sb to the @device. Do not wait for completion, all the
3471  * buffer heads we write are pinned.
3472  *
3473  * Write @max_mirrors copies of the superblock, where 0 means default that fit
3474  * the expected device size at commit time. Note that max_mirrors must be
3475  * same for write and wait phases.
3476  *
3477  * Return number of errors when buffer head is not found or submission fails.
3478  */
3479 static int write_dev_supers(struct btrfs_device *device,
3480 			    struct btrfs_super_block *sb, int max_mirrors)
3481 {
3482 	struct buffer_head *bh;
3483 	int i;
3484 	int ret;
3485 	int errors = 0;
3486 	u32 crc;
3487 	u64 bytenr;
3488 	int op_flags;
3489 
3490 	if (max_mirrors == 0)
3491 		max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3492 
3493 	for (i = 0; i < max_mirrors; i++) {
3494 		bytenr = btrfs_sb_offset(i);
3495 		if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3496 		    device->commit_total_bytes)
3497 			break;
3498 
3499 		btrfs_set_super_bytenr(sb, bytenr);
3500 
3501 		crc = ~(u32)0;
3502 		crc = btrfs_csum_data((const char *)sb + BTRFS_CSUM_SIZE, crc,
3503 				      BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3504 		btrfs_csum_final(crc, sb->csum);
3505 
3506 		/* One reference for us, and we leave it for the caller */
3507 		bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE,
3508 			      BTRFS_SUPER_INFO_SIZE);
3509 		if (!bh) {
3510 			btrfs_err(device->fs_info,
3511 			    "couldn't get super buffer head for bytenr %llu",
3512 			    bytenr);
3513 			errors++;
3514 			continue;
3515 		}
3516 
3517 		memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3518 
3519 		/* one reference for submit_bh */
3520 		get_bh(bh);
3521 
3522 		set_buffer_uptodate(bh);
3523 		lock_buffer(bh);
3524 		bh->b_end_io = btrfs_end_buffer_write_sync;
3525 		bh->b_private = device;
3526 
3527 		/*
3528 		 * we fua the first super.  The others we allow
3529 		 * to go down lazy.
3530 		 */
3531 		op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
3532 		if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3533 			op_flags |= REQ_FUA;
3534 		ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh);
3535 		if (ret)
3536 			errors++;
3537 	}
3538 	return errors < i ? 0 : -1;
3539 }
3540 
3541 /*
3542  * Wait for write completion of superblocks done by write_dev_supers,
3543  * @max_mirrors same for write and wait phases.
3544  *
3545  * Return number of errors when buffer head is not found or not marked up to
3546  * date.
3547  */
3548 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3549 {
3550 	struct buffer_head *bh;
3551 	int i;
3552 	int errors = 0;
3553 	bool primary_failed = false;
3554 	u64 bytenr;
3555 
3556 	if (max_mirrors == 0)
3557 		max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3558 
3559 	for (i = 0; i < max_mirrors; i++) {
3560 		bytenr = btrfs_sb_offset(i);
3561 		if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3562 		    device->commit_total_bytes)
3563 			break;
3564 
3565 		bh = __find_get_block(device->bdev,
3566 				      bytenr / BTRFS_BDEV_BLOCKSIZE,
3567 				      BTRFS_SUPER_INFO_SIZE);
3568 		if (!bh) {
3569 			errors++;
3570 			if (i == 0)
3571 				primary_failed = true;
3572 			continue;
3573 		}
3574 		wait_on_buffer(bh);
3575 		if (!buffer_uptodate(bh)) {
3576 			errors++;
3577 			if (i == 0)
3578 				primary_failed = true;
3579 		}
3580 
3581 		/* drop our reference */
3582 		brelse(bh);
3583 
3584 		/* drop the reference from the writing run */
3585 		brelse(bh);
3586 	}
3587 
3588 	/* log error, force error return */
3589 	if (primary_failed) {
3590 		btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3591 			  device->devid);
3592 		return -1;
3593 	}
3594 
3595 	return errors < i ? 0 : -1;
3596 }
3597 
3598 /*
3599  * endio for the write_dev_flush, this will wake anyone waiting
3600  * for the barrier when it is done
3601  */
3602 static void btrfs_end_empty_barrier(struct bio *bio)
3603 {
3604 	complete(bio->bi_private);
3605 }
3606 
3607 /*
3608  * Submit a flush request to the device if it supports it. Error handling is
3609  * done in the waiting counterpart.
3610  */
3611 static void write_dev_flush(struct btrfs_device *device)
3612 {
3613 	struct request_queue *q = bdev_get_queue(device->bdev);
3614 	struct bio *bio = device->flush_bio;
3615 
3616 	if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3617 		return;
3618 
3619 	bio_reset(bio);
3620 	bio->bi_end_io = btrfs_end_empty_barrier;
3621 	bio_set_dev(bio, device->bdev);
3622 	bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3623 	init_completion(&device->flush_wait);
3624 	bio->bi_private = &device->flush_wait;
3625 
3626 	btrfsic_submit_bio(bio);
3627 	set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3628 }
3629 
3630 /*
3631  * If the flush bio has been submitted by write_dev_flush, wait for it.
3632  */
3633 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3634 {
3635 	struct bio *bio = device->flush_bio;
3636 
3637 	if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3638 		return BLK_STS_OK;
3639 
3640 	clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3641 	wait_for_completion_io(&device->flush_wait);
3642 
3643 	return bio->bi_status;
3644 }
3645 
3646 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3647 {
3648 	if (!btrfs_check_rw_degradable(fs_info, NULL))
3649 		return -EIO;
3650 	return 0;
3651 }
3652 
3653 /*
3654  * send an empty flush down to each device in parallel,
3655  * then wait for them
3656  */
3657 static int barrier_all_devices(struct btrfs_fs_info *info)
3658 {
3659 	struct list_head *head;
3660 	struct btrfs_device *dev;
3661 	int errors_wait = 0;
3662 	blk_status_t ret;
3663 
3664 	lockdep_assert_held(&info->fs_devices->device_list_mutex);
3665 	/* send down all the barriers */
3666 	head = &info->fs_devices->devices;
3667 	list_for_each_entry(dev, head, dev_list) {
3668 		if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3669 			continue;
3670 		if (!dev->bdev)
3671 			continue;
3672 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3673 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3674 			continue;
3675 
3676 		write_dev_flush(dev);
3677 		dev->last_flush_error = BLK_STS_OK;
3678 	}
3679 
3680 	/* wait for all the barriers */
3681 	list_for_each_entry(dev, head, dev_list) {
3682 		if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3683 			continue;
3684 		if (!dev->bdev) {
3685 			errors_wait++;
3686 			continue;
3687 		}
3688 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3689 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3690 			continue;
3691 
3692 		ret = wait_dev_flush(dev);
3693 		if (ret) {
3694 			dev->last_flush_error = ret;
3695 			btrfs_dev_stat_inc_and_print(dev,
3696 					BTRFS_DEV_STAT_FLUSH_ERRS);
3697 			errors_wait++;
3698 		}
3699 	}
3700 
3701 	if (errors_wait) {
3702 		/*
3703 		 * At some point we need the status of all disks
3704 		 * to arrive at the volume status. So error checking
3705 		 * is being pushed to a separate loop.
3706 		 */
3707 		return check_barrier_error(info);
3708 	}
3709 	return 0;
3710 }
3711 
3712 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3713 {
3714 	int raid_type;
3715 	int min_tolerated = INT_MAX;
3716 
3717 	if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3718 	    (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3719 		min_tolerated = min(min_tolerated,
3720 				    btrfs_raid_array[BTRFS_RAID_SINGLE].
3721 				    tolerated_failures);
3722 
3723 	for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3724 		if (raid_type == BTRFS_RAID_SINGLE)
3725 			continue;
3726 		if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3727 			continue;
3728 		min_tolerated = min(min_tolerated,
3729 				    btrfs_raid_array[raid_type].
3730 				    tolerated_failures);
3731 	}
3732 
3733 	if (min_tolerated == INT_MAX) {
3734 		pr_warn("BTRFS: unknown raid flag: %llu", flags);
3735 		min_tolerated = 0;
3736 	}
3737 
3738 	return min_tolerated;
3739 }
3740 
3741 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3742 {
3743 	struct list_head *head;
3744 	struct btrfs_device *dev;
3745 	struct btrfs_super_block *sb;
3746 	struct btrfs_dev_item *dev_item;
3747 	int ret;
3748 	int do_barriers;
3749 	int max_errors;
3750 	int total_errors = 0;
3751 	u64 flags;
3752 
3753 	do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3754 
3755 	/*
3756 	 * max_mirrors == 0 indicates we're from commit_transaction,
3757 	 * not from fsync where the tree roots in fs_info have not
3758 	 * been consistent on disk.
3759 	 */
3760 	if (max_mirrors == 0)
3761 		backup_super_roots(fs_info);
3762 
3763 	sb = fs_info->super_for_commit;
3764 	dev_item = &sb->dev_item;
3765 
3766 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
3767 	head = &fs_info->fs_devices->devices;
3768 	max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3769 
3770 	if (do_barriers) {
3771 		ret = barrier_all_devices(fs_info);
3772 		if (ret) {
3773 			mutex_unlock(
3774 				&fs_info->fs_devices->device_list_mutex);
3775 			btrfs_handle_fs_error(fs_info, ret,
3776 					      "errors while submitting device barriers.");
3777 			return ret;
3778 		}
3779 	}
3780 
3781 	list_for_each_entry(dev, head, dev_list) {
3782 		if (!dev->bdev) {
3783 			total_errors++;
3784 			continue;
3785 		}
3786 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3787 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3788 			continue;
3789 
3790 		btrfs_set_stack_device_generation(dev_item, 0);
3791 		btrfs_set_stack_device_type(dev_item, dev->type);
3792 		btrfs_set_stack_device_id(dev_item, dev->devid);
3793 		btrfs_set_stack_device_total_bytes(dev_item,
3794 						   dev->commit_total_bytes);
3795 		btrfs_set_stack_device_bytes_used(dev_item,
3796 						  dev->commit_bytes_used);
3797 		btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3798 		btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3799 		btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3800 		memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3801 		memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_FSID_SIZE);
3802 
3803 		flags = btrfs_super_flags(sb);
3804 		btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3805 
3806 		ret = btrfs_validate_write_super(fs_info, sb);
3807 		if (ret < 0) {
3808 			mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3809 			btrfs_handle_fs_error(fs_info, -EUCLEAN,
3810 				"unexpected superblock corruption detected");
3811 			return -EUCLEAN;
3812 		}
3813 
3814 		ret = write_dev_supers(dev, sb, max_mirrors);
3815 		if (ret)
3816 			total_errors++;
3817 	}
3818 	if (total_errors > max_errors) {
3819 		btrfs_err(fs_info, "%d errors while writing supers",
3820 			  total_errors);
3821 		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3822 
3823 		/* FUA is masked off if unsupported and can't be the reason */
3824 		btrfs_handle_fs_error(fs_info, -EIO,
3825 				      "%d errors while writing supers",
3826 				      total_errors);
3827 		return -EIO;
3828 	}
3829 
3830 	total_errors = 0;
3831 	list_for_each_entry(dev, head, dev_list) {
3832 		if (!dev->bdev)
3833 			continue;
3834 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3835 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3836 			continue;
3837 
3838 		ret = wait_dev_supers(dev, max_mirrors);
3839 		if (ret)
3840 			total_errors++;
3841 	}
3842 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3843 	if (total_errors > max_errors) {
3844 		btrfs_handle_fs_error(fs_info, -EIO,
3845 				      "%d errors while writing supers",
3846 				      total_errors);
3847 		return -EIO;
3848 	}
3849 	return 0;
3850 }
3851 
3852 /* Drop a fs root from the radix tree and free it. */
3853 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3854 				  struct btrfs_root *root)
3855 {
3856 	spin_lock(&fs_info->fs_roots_radix_lock);
3857 	radix_tree_delete(&fs_info->fs_roots_radix,
3858 			  (unsigned long)root->root_key.objectid);
3859 	spin_unlock(&fs_info->fs_roots_radix_lock);
3860 
3861 	if (btrfs_root_refs(&root->root_item) == 0)
3862 		synchronize_srcu(&fs_info->subvol_srcu);
3863 
3864 	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3865 		btrfs_free_log(NULL, root);
3866 		if (root->reloc_root) {
3867 			free_extent_buffer(root->reloc_root->node);
3868 			free_extent_buffer(root->reloc_root->commit_root);
3869 			btrfs_put_fs_root(root->reloc_root);
3870 			root->reloc_root = NULL;
3871 		}
3872 	}
3873 
3874 	if (root->free_ino_pinned)
3875 		__btrfs_remove_free_space_cache(root->free_ino_pinned);
3876 	if (root->free_ino_ctl)
3877 		__btrfs_remove_free_space_cache(root->free_ino_ctl);
3878 	free_fs_root(root);
3879 }
3880 
3881 static void free_fs_root(struct btrfs_root *root)
3882 {
3883 	iput(root->ino_cache_inode);
3884 	WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3885 	if (root->anon_dev)
3886 		free_anon_bdev(root->anon_dev);
3887 	if (root->subv_writers)
3888 		btrfs_free_subvolume_writers(root->subv_writers);
3889 	free_extent_buffer(root->node);
3890 	free_extent_buffer(root->commit_root);
3891 	kfree(root->free_ino_ctl);
3892 	kfree(root->free_ino_pinned);
3893 	kfree(root->name);
3894 	btrfs_put_fs_root(root);
3895 }
3896 
3897 void btrfs_free_fs_root(struct btrfs_root *root)
3898 {
3899 	free_fs_root(root);
3900 }
3901 
3902 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3903 {
3904 	u64 root_objectid = 0;
3905 	struct btrfs_root *gang[8];
3906 	int i = 0;
3907 	int err = 0;
3908 	unsigned int ret = 0;
3909 	int index;
3910 
3911 	while (1) {
3912 		index = srcu_read_lock(&fs_info->subvol_srcu);
3913 		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3914 					     (void **)gang, root_objectid,
3915 					     ARRAY_SIZE(gang));
3916 		if (!ret) {
3917 			srcu_read_unlock(&fs_info->subvol_srcu, index);
3918 			break;
3919 		}
3920 		root_objectid = gang[ret - 1]->root_key.objectid + 1;
3921 
3922 		for (i = 0; i < ret; i++) {
3923 			/* Avoid to grab roots in dead_roots */
3924 			if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3925 				gang[i] = NULL;
3926 				continue;
3927 			}
3928 			/* grab all the search result for later use */
3929 			gang[i] = btrfs_grab_fs_root(gang[i]);
3930 		}
3931 		srcu_read_unlock(&fs_info->subvol_srcu, index);
3932 
3933 		for (i = 0; i < ret; i++) {
3934 			if (!gang[i])
3935 				continue;
3936 			root_objectid = gang[i]->root_key.objectid;
3937 			err = btrfs_orphan_cleanup(gang[i]);
3938 			if (err)
3939 				break;
3940 			btrfs_put_fs_root(gang[i]);
3941 		}
3942 		root_objectid++;
3943 	}
3944 
3945 	/* release the uncleaned roots due to error */
3946 	for (; i < ret; i++) {
3947 		if (gang[i])
3948 			btrfs_put_fs_root(gang[i]);
3949 	}
3950 	return err;
3951 }
3952 
3953 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3954 {
3955 	struct btrfs_root *root = fs_info->tree_root;
3956 	struct btrfs_trans_handle *trans;
3957 
3958 	mutex_lock(&fs_info->cleaner_mutex);
3959 	btrfs_run_delayed_iputs(fs_info);
3960 	mutex_unlock(&fs_info->cleaner_mutex);
3961 	wake_up_process(fs_info->cleaner_kthread);
3962 
3963 	/* wait until ongoing cleanup work done */
3964 	down_write(&fs_info->cleanup_work_sem);
3965 	up_write(&fs_info->cleanup_work_sem);
3966 
3967 	trans = btrfs_join_transaction(root);
3968 	if (IS_ERR(trans))
3969 		return PTR_ERR(trans);
3970 	return btrfs_commit_transaction(trans);
3971 }
3972 
3973 void close_ctree(struct btrfs_fs_info *fs_info)
3974 {
3975 	int ret;
3976 
3977 	set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
3978 
3979 	/* wait for the qgroup rescan worker to stop */
3980 	btrfs_qgroup_wait_for_completion(fs_info, false);
3981 
3982 	/* wait for the uuid_scan task to finish */
3983 	down(&fs_info->uuid_tree_rescan_sem);
3984 	/* avoid complains from lockdep et al., set sem back to initial state */
3985 	up(&fs_info->uuid_tree_rescan_sem);
3986 
3987 	/* pause restriper - we want to resume on mount */
3988 	btrfs_pause_balance(fs_info);
3989 
3990 	btrfs_dev_replace_suspend_for_unmount(fs_info);
3991 
3992 	btrfs_scrub_cancel(fs_info);
3993 
3994 	/* wait for any defraggers to finish */
3995 	wait_event(fs_info->transaction_wait,
3996 		   (atomic_read(&fs_info->defrag_running) == 0));
3997 
3998 	/* clear out the rbtree of defraggable inodes */
3999 	btrfs_cleanup_defrag_inodes(fs_info);
4000 
4001 	cancel_work_sync(&fs_info->async_reclaim_work);
4002 
4003 	if (!sb_rdonly(fs_info->sb)) {
4004 		/*
4005 		 * If the cleaner thread is stopped and there are
4006 		 * block groups queued for removal, the deletion will be
4007 		 * skipped when we quit the cleaner thread.
4008 		 */
4009 		btrfs_delete_unused_bgs(fs_info);
4010 
4011 		ret = btrfs_commit_super(fs_info);
4012 		if (ret)
4013 			btrfs_err(fs_info, "commit super ret %d", ret);
4014 	}
4015 
4016 	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
4017 	    test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
4018 		btrfs_error_commit_super(fs_info);
4019 
4020 	kthread_stop(fs_info->transaction_kthread);
4021 	kthread_stop(fs_info->cleaner_kthread);
4022 
4023 	set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4024 
4025 	btrfs_free_qgroup_config(fs_info);
4026 	ASSERT(list_empty(&fs_info->delalloc_roots));
4027 
4028 	if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4029 		btrfs_info(fs_info, "at unmount delalloc count %lld",
4030 		       percpu_counter_sum(&fs_info->delalloc_bytes));
4031 	}
4032 
4033 	btrfs_sysfs_remove_mounted(fs_info);
4034 	btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4035 
4036 	btrfs_free_fs_roots(fs_info);
4037 
4038 	btrfs_put_block_group_cache(fs_info);
4039 
4040 	/*
4041 	 * we must make sure there is not any read request to
4042 	 * submit after we stopping all workers.
4043 	 */
4044 	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4045 	btrfs_stop_all_workers(fs_info);
4046 
4047 	btrfs_free_block_groups(fs_info);
4048 
4049 	clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4050 	free_root_pointers(fs_info, 1);
4051 
4052 	iput(fs_info->btree_inode);
4053 
4054 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4055 	if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4056 		btrfsic_unmount(fs_info->fs_devices);
4057 #endif
4058 
4059 	btrfs_close_devices(fs_info->fs_devices);
4060 	btrfs_mapping_tree_free(&fs_info->mapping_tree);
4061 
4062 	percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
4063 	percpu_counter_destroy(&fs_info->delalloc_bytes);
4064 	percpu_counter_destroy(&fs_info->bio_counter);
4065 	cleanup_srcu_struct(&fs_info->subvol_srcu);
4066 
4067 	btrfs_free_stripe_hash_table(fs_info);
4068 	btrfs_free_ref_cache(fs_info);
4069 
4070 	while (!list_empty(&fs_info->pinned_chunks)) {
4071 		struct extent_map *em;
4072 
4073 		em = list_first_entry(&fs_info->pinned_chunks,
4074 				      struct extent_map, list);
4075 		list_del_init(&em->list);
4076 		free_extent_map(em);
4077 	}
4078 }
4079 
4080 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4081 			  int atomic)
4082 {
4083 	int ret;
4084 	struct inode *btree_inode = buf->pages[0]->mapping->host;
4085 
4086 	ret = extent_buffer_uptodate(buf);
4087 	if (!ret)
4088 		return ret;
4089 
4090 	ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4091 				    parent_transid, atomic);
4092 	if (ret == -EAGAIN)
4093 		return ret;
4094 	return !ret;
4095 }
4096 
4097 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4098 {
4099 	struct btrfs_fs_info *fs_info;
4100 	struct btrfs_root *root;
4101 	u64 transid = btrfs_header_generation(buf);
4102 	int was_dirty;
4103 
4104 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4105 	/*
4106 	 * This is a fast path so only do this check if we have sanity tests
4107 	 * enabled.  Normal people shouldn't be marking dummy buffers as dirty
4108 	 * outside of the sanity tests.
4109 	 */
4110 	if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
4111 		return;
4112 #endif
4113 	root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4114 	fs_info = root->fs_info;
4115 	btrfs_assert_tree_locked(buf);
4116 	if (transid != fs_info->generation)
4117 		WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4118 			buf->start, transid, fs_info->generation);
4119 	was_dirty = set_extent_buffer_dirty(buf);
4120 	if (!was_dirty)
4121 		percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4122 					 buf->len,
4123 					 fs_info->dirty_metadata_batch);
4124 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4125 	/*
4126 	 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4127 	 * but item data not updated.
4128 	 * So here we should only check item pointers, not item data.
4129 	 */
4130 	if (btrfs_header_level(buf) == 0 &&
4131 	    btrfs_check_leaf_relaxed(fs_info, buf)) {
4132 		btrfs_print_leaf(buf);
4133 		ASSERT(0);
4134 	}
4135 #endif
4136 }
4137 
4138 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4139 					int flush_delayed)
4140 {
4141 	/*
4142 	 * looks as though older kernels can get into trouble with
4143 	 * this code, they end up stuck in balance_dirty_pages forever
4144 	 */
4145 	int ret;
4146 
4147 	if (current->flags & PF_MEMALLOC)
4148 		return;
4149 
4150 	if (flush_delayed)
4151 		btrfs_balance_delayed_items(fs_info);
4152 
4153 	ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4154 				     BTRFS_DIRTY_METADATA_THRESH);
4155 	if (ret > 0) {
4156 		balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4157 	}
4158 }
4159 
4160 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4161 {
4162 	__btrfs_btree_balance_dirty(fs_info, 1);
4163 }
4164 
4165 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4166 {
4167 	__btrfs_btree_balance_dirty(fs_info, 0);
4168 }
4169 
4170 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4171 		      struct btrfs_key *first_key)
4172 {
4173 	struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4174 	struct btrfs_fs_info *fs_info = root->fs_info;
4175 
4176 	return btree_read_extent_buffer_pages(fs_info, buf, parent_transid,
4177 					      level, first_key);
4178 }
4179 
4180 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4181 {
4182 	/* cleanup FS via transaction */
4183 	btrfs_cleanup_transaction(fs_info);
4184 
4185 	mutex_lock(&fs_info->cleaner_mutex);
4186 	btrfs_run_delayed_iputs(fs_info);
4187 	mutex_unlock(&fs_info->cleaner_mutex);
4188 
4189 	down_write(&fs_info->cleanup_work_sem);
4190 	up_write(&fs_info->cleanup_work_sem);
4191 }
4192 
4193 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4194 {
4195 	struct btrfs_ordered_extent *ordered;
4196 
4197 	spin_lock(&root->ordered_extent_lock);
4198 	/*
4199 	 * This will just short circuit the ordered completion stuff which will
4200 	 * make sure the ordered extent gets properly cleaned up.
4201 	 */
4202 	list_for_each_entry(ordered, &root->ordered_extents,
4203 			    root_extent_list)
4204 		set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4205 	spin_unlock(&root->ordered_extent_lock);
4206 }
4207 
4208 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4209 {
4210 	struct btrfs_root *root;
4211 	struct list_head splice;
4212 
4213 	INIT_LIST_HEAD(&splice);
4214 
4215 	spin_lock(&fs_info->ordered_root_lock);
4216 	list_splice_init(&fs_info->ordered_roots, &splice);
4217 	while (!list_empty(&splice)) {
4218 		root = list_first_entry(&splice, struct btrfs_root,
4219 					ordered_root);
4220 		list_move_tail(&root->ordered_root,
4221 			       &fs_info->ordered_roots);
4222 
4223 		spin_unlock(&fs_info->ordered_root_lock);
4224 		btrfs_destroy_ordered_extents(root);
4225 
4226 		cond_resched();
4227 		spin_lock(&fs_info->ordered_root_lock);
4228 	}
4229 	spin_unlock(&fs_info->ordered_root_lock);
4230 }
4231 
4232 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4233 				      struct btrfs_fs_info *fs_info)
4234 {
4235 	struct rb_node *node;
4236 	struct btrfs_delayed_ref_root *delayed_refs;
4237 	struct btrfs_delayed_ref_node *ref;
4238 	int ret = 0;
4239 
4240 	delayed_refs = &trans->delayed_refs;
4241 
4242 	spin_lock(&delayed_refs->lock);
4243 	if (atomic_read(&delayed_refs->num_entries) == 0) {
4244 		spin_unlock(&delayed_refs->lock);
4245 		btrfs_info(fs_info, "delayed_refs has NO entry");
4246 		return ret;
4247 	}
4248 
4249 	while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4250 		struct btrfs_delayed_ref_head *head;
4251 		struct rb_node *n;
4252 		bool pin_bytes = false;
4253 
4254 		head = rb_entry(node, struct btrfs_delayed_ref_head,
4255 				href_node);
4256 		if (!mutex_trylock(&head->mutex)) {
4257 			refcount_inc(&head->refs);
4258 			spin_unlock(&delayed_refs->lock);
4259 
4260 			mutex_lock(&head->mutex);
4261 			mutex_unlock(&head->mutex);
4262 			btrfs_put_delayed_ref_head(head);
4263 			spin_lock(&delayed_refs->lock);
4264 			continue;
4265 		}
4266 		spin_lock(&head->lock);
4267 		while ((n = rb_first(&head->ref_tree)) != NULL) {
4268 			ref = rb_entry(n, struct btrfs_delayed_ref_node,
4269 				       ref_node);
4270 			ref->in_tree = 0;
4271 			rb_erase(&ref->ref_node, &head->ref_tree);
4272 			RB_CLEAR_NODE(&ref->ref_node);
4273 			if (!list_empty(&ref->add_list))
4274 				list_del(&ref->add_list);
4275 			atomic_dec(&delayed_refs->num_entries);
4276 			btrfs_put_delayed_ref(ref);
4277 		}
4278 		if (head->must_insert_reserved)
4279 			pin_bytes = true;
4280 		btrfs_free_delayed_extent_op(head->extent_op);
4281 		delayed_refs->num_heads--;
4282 		if (head->processing == 0)
4283 			delayed_refs->num_heads_ready--;
4284 		atomic_dec(&delayed_refs->num_entries);
4285 		rb_erase(&head->href_node, &delayed_refs->href_root);
4286 		RB_CLEAR_NODE(&head->href_node);
4287 		spin_unlock(&head->lock);
4288 		spin_unlock(&delayed_refs->lock);
4289 		mutex_unlock(&head->mutex);
4290 
4291 		if (pin_bytes)
4292 			btrfs_pin_extent(fs_info, head->bytenr,
4293 					 head->num_bytes, 1);
4294 		btrfs_put_delayed_ref_head(head);
4295 		cond_resched();
4296 		spin_lock(&delayed_refs->lock);
4297 	}
4298 
4299 	spin_unlock(&delayed_refs->lock);
4300 
4301 	return ret;
4302 }
4303 
4304 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4305 {
4306 	struct btrfs_inode *btrfs_inode;
4307 	struct list_head splice;
4308 
4309 	INIT_LIST_HEAD(&splice);
4310 
4311 	spin_lock(&root->delalloc_lock);
4312 	list_splice_init(&root->delalloc_inodes, &splice);
4313 
4314 	while (!list_empty(&splice)) {
4315 		struct inode *inode = NULL;
4316 		btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4317 					       delalloc_inodes);
4318 		__btrfs_del_delalloc_inode(root, btrfs_inode);
4319 		spin_unlock(&root->delalloc_lock);
4320 
4321 		/*
4322 		 * Make sure we get a live inode and that it'll not disappear
4323 		 * meanwhile.
4324 		 */
4325 		inode = igrab(&btrfs_inode->vfs_inode);
4326 		if (inode) {
4327 			invalidate_inode_pages2(inode->i_mapping);
4328 			iput(inode);
4329 		}
4330 		spin_lock(&root->delalloc_lock);
4331 	}
4332 	spin_unlock(&root->delalloc_lock);
4333 }
4334 
4335 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4336 {
4337 	struct btrfs_root *root;
4338 	struct list_head splice;
4339 
4340 	INIT_LIST_HEAD(&splice);
4341 
4342 	spin_lock(&fs_info->delalloc_root_lock);
4343 	list_splice_init(&fs_info->delalloc_roots, &splice);
4344 	while (!list_empty(&splice)) {
4345 		root = list_first_entry(&splice, struct btrfs_root,
4346 					 delalloc_root);
4347 		root = btrfs_grab_fs_root(root);
4348 		BUG_ON(!root);
4349 		spin_unlock(&fs_info->delalloc_root_lock);
4350 
4351 		btrfs_destroy_delalloc_inodes(root);
4352 		btrfs_put_fs_root(root);
4353 
4354 		spin_lock(&fs_info->delalloc_root_lock);
4355 	}
4356 	spin_unlock(&fs_info->delalloc_root_lock);
4357 }
4358 
4359 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4360 					struct extent_io_tree *dirty_pages,
4361 					int mark)
4362 {
4363 	int ret;
4364 	struct extent_buffer *eb;
4365 	u64 start = 0;
4366 	u64 end;
4367 
4368 	while (1) {
4369 		ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4370 					    mark, NULL);
4371 		if (ret)
4372 			break;
4373 
4374 		clear_extent_bits(dirty_pages, start, end, mark);
4375 		while (start <= end) {
4376 			eb = find_extent_buffer(fs_info, start);
4377 			start += fs_info->nodesize;
4378 			if (!eb)
4379 				continue;
4380 			wait_on_extent_buffer_writeback(eb);
4381 
4382 			if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4383 					       &eb->bflags))
4384 				clear_extent_buffer_dirty(eb);
4385 			free_extent_buffer_stale(eb);
4386 		}
4387 	}
4388 
4389 	return ret;
4390 }
4391 
4392 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4393 				       struct extent_io_tree *pinned_extents)
4394 {
4395 	struct extent_io_tree *unpin;
4396 	u64 start;
4397 	u64 end;
4398 	int ret;
4399 	bool loop = true;
4400 
4401 	unpin = pinned_extents;
4402 again:
4403 	while (1) {
4404 		ret = find_first_extent_bit(unpin, 0, &start, &end,
4405 					    EXTENT_DIRTY, NULL);
4406 		if (ret)
4407 			break;
4408 
4409 		clear_extent_dirty(unpin, start, end);
4410 		btrfs_error_unpin_extent_range(fs_info, start, end);
4411 		cond_resched();
4412 	}
4413 
4414 	if (loop) {
4415 		if (unpin == &fs_info->freed_extents[0])
4416 			unpin = &fs_info->freed_extents[1];
4417 		else
4418 			unpin = &fs_info->freed_extents[0];
4419 		loop = false;
4420 		goto again;
4421 	}
4422 
4423 	return 0;
4424 }
4425 
4426 static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache)
4427 {
4428 	struct inode *inode;
4429 
4430 	inode = cache->io_ctl.inode;
4431 	if (inode) {
4432 		invalidate_inode_pages2(inode->i_mapping);
4433 		BTRFS_I(inode)->generation = 0;
4434 		cache->io_ctl.inode = NULL;
4435 		iput(inode);
4436 	}
4437 	btrfs_put_block_group(cache);
4438 }
4439 
4440 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4441 			     struct btrfs_fs_info *fs_info)
4442 {
4443 	struct btrfs_block_group_cache *cache;
4444 
4445 	spin_lock(&cur_trans->dirty_bgs_lock);
4446 	while (!list_empty(&cur_trans->dirty_bgs)) {
4447 		cache = list_first_entry(&cur_trans->dirty_bgs,
4448 					 struct btrfs_block_group_cache,
4449 					 dirty_list);
4450 
4451 		if (!list_empty(&cache->io_list)) {
4452 			spin_unlock(&cur_trans->dirty_bgs_lock);
4453 			list_del_init(&cache->io_list);
4454 			btrfs_cleanup_bg_io(cache);
4455 			spin_lock(&cur_trans->dirty_bgs_lock);
4456 		}
4457 
4458 		list_del_init(&cache->dirty_list);
4459 		spin_lock(&cache->lock);
4460 		cache->disk_cache_state = BTRFS_DC_ERROR;
4461 		spin_unlock(&cache->lock);
4462 
4463 		spin_unlock(&cur_trans->dirty_bgs_lock);
4464 		btrfs_put_block_group(cache);
4465 		spin_lock(&cur_trans->dirty_bgs_lock);
4466 	}
4467 	spin_unlock(&cur_trans->dirty_bgs_lock);
4468 
4469 	/*
4470 	 * Refer to the definition of io_bgs member for details why it's safe
4471 	 * to use it without any locking
4472 	 */
4473 	while (!list_empty(&cur_trans->io_bgs)) {
4474 		cache = list_first_entry(&cur_trans->io_bgs,
4475 					 struct btrfs_block_group_cache,
4476 					 io_list);
4477 
4478 		list_del_init(&cache->io_list);
4479 		spin_lock(&cache->lock);
4480 		cache->disk_cache_state = BTRFS_DC_ERROR;
4481 		spin_unlock(&cache->lock);
4482 		btrfs_cleanup_bg_io(cache);
4483 	}
4484 }
4485 
4486 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4487 				   struct btrfs_fs_info *fs_info)
4488 {
4489 	btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4490 	ASSERT(list_empty(&cur_trans->dirty_bgs));
4491 	ASSERT(list_empty(&cur_trans->io_bgs));
4492 
4493 	btrfs_destroy_delayed_refs(cur_trans, fs_info);
4494 
4495 	cur_trans->state = TRANS_STATE_COMMIT_START;
4496 	wake_up(&fs_info->transaction_blocked_wait);
4497 
4498 	cur_trans->state = TRANS_STATE_UNBLOCKED;
4499 	wake_up(&fs_info->transaction_wait);
4500 
4501 	btrfs_destroy_delayed_inodes(fs_info);
4502 	btrfs_assert_delayed_root_empty(fs_info);
4503 
4504 	btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4505 				     EXTENT_DIRTY);
4506 	btrfs_destroy_pinned_extent(fs_info,
4507 				    fs_info->pinned_extents);
4508 
4509 	cur_trans->state =TRANS_STATE_COMPLETED;
4510 	wake_up(&cur_trans->commit_wait);
4511 }
4512 
4513 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4514 {
4515 	struct btrfs_transaction *t;
4516 
4517 	mutex_lock(&fs_info->transaction_kthread_mutex);
4518 
4519 	spin_lock(&fs_info->trans_lock);
4520 	while (!list_empty(&fs_info->trans_list)) {
4521 		t = list_first_entry(&fs_info->trans_list,
4522 				     struct btrfs_transaction, list);
4523 		if (t->state >= TRANS_STATE_COMMIT_START) {
4524 			refcount_inc(&t->use_count);
4525 			spin_unlock(&fs_info->trans_lock);
4526 			btrfs_wait_for_commit(fs_info, t->transid);
4527 			btrfs_put_transaction(t);
4528 			spin_lock(&fs_info->trans_lock);
4529 			continue;
4530 		}
4531 		if (t == fs_info->running_transaction) {
4532 			t->state = TRANS_STATE_COMMIT_DOING;
4533 			spin_unlock(&fs_info->trans_lock);
4534 			/*
4535 			 * We wait for 0 num_writers since we don't hold a trans
4536 			 * handle open currently for this transaction.
4537 			 */
4538 			wait_event(t->writer_wait,
4539 				   atomic_read(&t->num_writers) == 0);
4540 		} else {
4541 			spin_unlock(&fs_info->trans_lock);
4542 		}
4543 		btrfs_cleanup_one_transaction(t, fs_info);
4544 
4545 		spin_lock(&fs_info->trans_lock);
4546 		if (t == fs_info->running_transaction)
4547 			fs_info->running_transaction = NULL;
4548 		list_del_init(&t->list);
4549 		spin_unlock(&fs_info->trans_lock);
4550 
4551 		btrfs_put_transaction(t);
4552 		trace_btrfs_transaction_commit(fs_info->tree_root);
4553 		spin_lock(&fs_info->trans_lock);
4554 	}
4555 	spin_unlock(&fs_info->trans_lock);
4556 	btrfs_destroy_all_ordered_extents(fs_info);
4557 	btrfs_destroy_delayed_inodes(fs_info);
4558 	btrfs_assert_delayed_root_empty(fs_info);
4559 	btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
4560 	btrfs_destroy_all_delalloc_inodes(fs_info);
4561 	mutex_unlock(&fs_info->transaction_kthread_mutex);
4562 
4563 	return 0;
4564 }
4565 
4566 static struct btrfs_fs_info *btree_fs_info(void *private_data)
4567 {
4568 	struct inode *inode = private_data;
4569 	return btrfs_sb(inode->i_sb);
4570 }
4571 
4572 static const struct extent_io_ops btree_extent_io_ops = {
4573 	/* mandatory callbacks */
4574 	.submit_bio_hook = btree_submit_bio_hook,
4575 	.readpage_end_io_hook = btree_readpage_end_io_hook,
4576 	/* note we're sharing with inode.c for the merge bio hook */
4577 	.merge_bio_hook = btrfs_merge_bio_hook,
4578 	.readpage_io_failed_hook = btree_io_failed_hook,
4579 	.set_range_writeback = btrfs_set_range_writeback,
4580 	.tree_fs_info = btree_fs_info,
4581 
4582 	/* optional callbacks */
4583 };
4584