xref: /linux/fs/btrfs/disk-io.c (revision e5c86679d5e864947a52fb31e45a425dea3e7fa9)
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18 
19 #include <linux/fs.h>
20 #include <linux/blkdev.h>
21 #include <linux/scatterlist.h>
22 #include <linux/swap.h>
23 #include <linux/radix-tree.h>
24 #include <linux/writeback.h>
25 #include <linux/buffer_head.h>
26 #include <linux/workqueue.h>
27 #include <linux/kthread.h>
28 #include <linux/slab.h>
29 #include <linux/migrate.h>
30 #include <linux/ratelimit.h>
31 #include <linux/uuid.h>
32 #include <linux/semaphore.h>
33 #include <asm/unaligned.h>
34 #include "ctree.h"
35 #include "disk-io.h"
36 #include "hash.h"
37 #include "transaction.h"
38 #include "btrfs_inode.h"
39 #include "volumes.h"
40 #include "print-tree.h"
41 #include "locking.h"
42 #include "tree-log.h"
43 #include "free-space-cache.h"
44 #include "free-space-tree.h"
45 #include "inode-map.h"
46 #include "check-integrity.h"
47 #include "rcu-string.h"
48 #include "dev-replace.h"
49 #include "raid56.h"
50 #include "sysfs.h"
51 #include "qgroup.h"
52 #include "compression.h"
53 
54 #ifdef CONFIG_X86
55 #include <asm/cpufeature.h>
56 #endif
57 
58 #define BTRFS_SUPER_FLAG_SUPP	(BTRFS_HEADER_FLAG_WRITTEN |\
59 				 BTRFS_HEADER_FLAG_RELOC |\
60 				 BTRFS_SUPER_FLAG_ERROR |\
61 				 BTRFS_SUPER_FLAG_SEEDING |\
62 				 BTRFS_SUPER_FLAG_METADUMP)
63 
64 static const struct extent_io_ops btree_extent_io_ops;
65 static void end_workqueue_fn(struct btrfs_work *work);
66 static void free_fs_root(struct btrfs_root *root);
67 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info);
68 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
69 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
70 				      struct btrfs_fs_info *fs_info);
71 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
72 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
73 					struct extent_io_tree *dirty_pages,
74 					int mark);
75 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
76 				       struct extent_io_tree *pinned_extents);
77 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
78 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
79 
80 /*
81  * btrfs_end_io_wq structs are used to do processing in task context when an IO
82  * is complete.  This is used during reads to verify checksums, and it is used
83  * by writes to insert metadata for new file extents after IO is complete.
84  */
85 struct btrfs_end_io_wq {
86 	struct bio *bio;
87 	bio_end_io_t *end_io;
88 	void *private;
89 	struct btrfs_fs_info *info;
90 	int error;
91 	enum btrfs_wq_endio_type metadata;
92 	struct list_head list;
93 	struct btrfs_work work;
94 };
95 
96 static struct kmem_cache *btrfs_end_io_wq_cache;
97 
98 int __init btrfs_end_io_wq_init(void)
99 {
100 	btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
101 					sizeof(struct btrfs_end_io_wq),
102 					0,
103 					SLAB_MEM_SPREAD,
104 					NULL);
105 	if (!btrfs_end_io_wq_cache)
106 		return -ENOMEM;
107 	return 0;
108 }
109 
110 void btrfs_end_io_wq_exit(void)
111 {
112 	kmem_cache_destroy(btrfs_end_io_wq_cache);
113 }
114 
115 /*
116  * async submit bios are used to offload expensive checksumming
117  * onto the worker threads.  They checksum file and metadata bios
118  * just before they are sent down the IO stack.
119  */
120 struct async_submit_bio {
121 	struct inode *inode;
122 	struct bio *bio;
123 	struct list_head list;
124 	extent_submit_bio_hook_t *submit_bio_start;
125 	extent_submit_bio_hook_t *submit_bio_done;
126 	int mirror_num;
127 	unsigned long bio_flags;
128 	/*
129 	 * bio_offset is optional, can be used if the pages in the bio
130 	 * can't tell us where in the file the bio should go
131 	 */
132 	u64 bio_offset;
133 	struct btrfs_work work;
134 	int error;
135 };
136 
137 /*
138  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
139  * eb, the lockdep key is determined by the btrfs_root it belongs to and
140  * the level the eb occupies in the tree.
141  *
142  * Different roots are used for different purposes and may nest inside each
143  * other and they require separate keysets.  As lockdep keys should be
144  * static, assign keysets according to the purpose of the root as indicated
145  * by btrfs_root->objectid.  This ensures that all special purpose roots
146  * have separate keysets.
147  *
148  * Lock-nesting across peer nodes is always done with the immediate parent
149  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
150  * subclass to avoid triggering lockdep warning in such cases.
151  *
152  * The key is set by the readpage_end_io_hook after the buffer has passed
153  * csum validation but before the pages are unlocked.  It is also set by
154  * btrfs_init_new_buffer on freshly allocated blocks.
155  *
156  * We also add a check to make sure the highest level of the tree is the
157  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
158  * needs update as well.
159  */
160 #ifdef CONFIG_DEBUG_LOCK_ALLOC
161 # if BTRFS_MAX_LEVEL != 8
162 #  error
163 # endif
164 
165 static struct btrfs_lockdep_keyset {
166 	u64			id;		/* root objectid */
167 	const char		*name_stem;	/* lock name stem */
168 	char			names[BTRFS_MAX_LEVEL + 1][20];
169 	struct lock_class_key	keys[BTRFS_MAX_LEVEL + 1];
170 } btrfs_lockdep_keysets[] = {
171 	{ .id = BTRFS_ROOT_TREE_OBJECTID,	.name_stem = "root"	},
172 	{ .id = BTRFS_EXTENT_TREE_OBJECTID,	.name_stem = "extent"	},
173 	{ .id = BTRFS_CHUNK_TREE_OBJECTID,	.name_stem = "chunk"	},
174 	{ .id = BTRFS_DEV_TREE_OBJECTID,	.name_stem = "dev"	},
175 	{ .id = BTRFS_FS_TREE_OBJECTID,		.name_stem = "fs"	},
176 	{ .id = BTRFS_CSUM_TREE_OBJECTID,	.name_stem = "csum"	},
177 	{ .id = BTRFS_QUOTA_TREE_OBJECTID,	.name_stem = "quota"	},
178 	{ .id = BTRFS_TREE_LOG_OBJECTID,	.name_stem = "log"	},
179 	{ .id = BTRFS_TREE_RELOC_OBJECTID,	.name_stem = "treloc"	},
180 	{ .id = BTRFS_DATA_RELOC_TREE_OBJECTID,	.name_stem = "dreloc"	},
181 	{ .id = BTRFS_UUID_TREE_OBJECTID,	.name_stem = "uuid"	},
182 	{ .id = BTRFS_FREE_SPACE_TREE_OBJECTID,	.name_stem = "free-space" },
183 	{ .id = 0,				.name_stem = "tree"	},
184 };
185 
186 void __init btrfs_init_lockdep(void)
187 {
188 	int i, j;
189 
190 	/* initialize lockdep class names */
191 	for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
192 		struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
193 
194 		for (j = 0; j < ARRAY_SIZE(ks->names); j++)
195 			snprintf(ks->names[j], sizeof(ks->names[j]),
196 				 "btrfs-%s-%02d", ks->name_stem, j);
197 	}
198 }
199 
200 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
201 				    int level)
202 {
203 	struct btrfs_lockdep_keyset *ks;
204 
205 	BUG_ON(level >= ARRAY_SIZE(ks->keys));
206 
207 	/* find the matching keyset, id 0 is the default entry */
208 	for (ks = btrfs_lockdep_keysets; ks->id; ks++)
209 		if (ks->id == objectid)
210 			break;
211 
212 	lockdep_set_class_and_name(&eb->lock,
213 				   &ks->keys[level], ks->names[level]);
214 }
215 
216 #endif
217 
218 /*
219  * extents on the btree inode are pretty simple, there's one extent
220  * that covers the entire device
221  */
222 static struct extent_map *btree_get_extent(struct btrfs_inode *inode,
223 		struct page *page, size_t pg_offset, u64 start, u64 len,
224 		int create)
225 {
226 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
227 	struct extent_map_tree *em_tree = &inode->extent_tree;
228 	struct extent_map *em;
229 	int ret;
230 
231 	read_lock(&em_tree->lock);
232 	em = lookup_extent_mapping(em_tree, start, len);
233 	if (em) {
234 		em->bdev = fs_info->fs_devices->latest_bdev;
235 		read_unlock(&em_tree->lock);
236 		goto out;
237 	}
238 	read_unlock(&em_tree->lock);
239 
240 	em = alloc_extent_map();
241 	if (!em) {
242 		em = ERR_PTR(-ENOMEM);
243 		goto out;
244 	}
245 	em->start = 0;
246 	em->len = (u64)-1;
247 	em->block_len = (u64)-1;
248 	em->block_start = 0;
249 	em->bdev = fs_info->fs_devices->latest_bdev;
250 
251 	write_lock(&em_tree->lock);
252 	ret = add_extent_mapping(em_tree, em, 0);
253 	if (ret == -EEXIST) {
254 		free_extent_map(em);
255 		em = lookup_extent_mapping(em_tree, start, len);
256 		if (!em)
257 			em = ERR_PTR(-EIO);
258 	} else if (ret) {
259 		free_extent_map(em);
260 		em = ERR_PTR(ret);
261 	}
262 	write_unlock(&em_tree->lock);
263 
264 out:
265 	return em;
266 }
267 
268 u32 btrfs_csum_data(const char *data, u32 seed, size_t len)
269 {
270 	return btrfs_crc32c(seed, data, len);
271 }
272 
273 void btrfs_csum_final(u32 crc, u8 *result)
274 {
275 	put_unaligned_le32(~crc, result);
276 }
277 
278 /*
279  * compute the csum for a btree block, and either verify it or write it
280  * into the csum field of the block.
281  */
282 static int csum_tree_block(struct btrfs_fs_info *fs_info,
283 			   struct extent_buffer *buf,
284 			   int verify)
285 {
286 	u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
287 	char *result = NULL;
288 	unsigned long len;
289 	unsigned long cur_len;
290 	unsigned long offset = BTRFS_CSUM_SIZE;
291 	char *kaddr;
292 	unsigned long map_start;
293 	unsigned long map_len;
294 	int err;
295 	u32 crc = ~(u32)0;
296 	unsigned long inline_result;
297 
298 	len = buf->len - offset;
299 	while (len > 0) {
300 		err = map_private_extent_buffer(buf, offset, 32,
301 					&kaddr, &map_start, &map_len);
302 		if (err)
303 			return err;
304 		cur_len = min(len, map_len - (offset - map_start));
305 		crc = btrfs_csum_data(kaddr + offset - map_start,
306 				      crc, cur_len);
307 		len -= cur_len;
308 		offset += cur_len;
309 	}
310 	if (csum_size > sizeof(inline_result)) {
311 		result = kzalloc(csum_size, GFP_NOFS);
312 		if (!result)
313 			return -ENOMEM;
314 	} else {
315 		result = (char *)&inline_result;
316 	}
317 
318 	btrfs_csum_final(crc, result);
319 
320 	if (verify) {
321 		if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
322 			u32 val;
323 			u32 found = 0;
324 			memcpy(&found, result, csum_size);
325 
326 			read_extent_buffer(buf, &val, 0, csum_size);
327 			btrfs_warn_rl(fs_info,
328 				"%s checksum verify failed on %llu wanted %X found %X level %d",
329 				fs_info->sb->s_id, buf->start,
330 				val, found, btrfs_header_level(buf));
331 			if (result != (char *)&inline_result)
332 				kfree(result);
333 			return -EUCLEAN;
334 		}
335 	} else {
336 		write_extent_buffer(buf, result, 0, csum_size);
337 	}
338 	if (result != (char *)&inline_result)
339 		kfree(result);
340 	return 0;
341 }
342 
343 /*
344  * we can't consider a given block up to date unless the transid of the
345  * block matches the transid in the parent node's pointer.  This is how we
346  * detect blocks that either didn't get written at all or got written
347  * in the wrong place.
348  */
349 static int verify_parent_transid(struct extent_io_tree *io_tree,
350 				 struct extent_buffer *eb, u64 parent_transid,
351 				 int atomic)
352 {
353 	struct extent_state *cached_state = NULL;
354 	int ret;
355 	bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
356 
357 	if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
358 		return 0;
359 
360 	if (atomic)
361 		return -EAGAIN;
362 
363 	if (need_lock) {
364 		btrfs_tree_read_lock(eb);
365 		btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
366 	}
367 
368 	lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
369 			 &cached_state);
370 	if (extent_buffer_uptodate(eb) &&
371 	    btrfs_header_generation(eb) == parent_transid) {
372 		ret = 0;
373 		goto out;
374 	}
375 	btrfs_err_rl(eb->fs_info,
376 		"parent transid verify failed on %llu wanted %llu found %llu",
377 			eb->start,
378 			parent_transid, btrfs_header_generation(eb));
379 	ret = 1;
380 
381 	/*
382 	 * Things reading via commit roots that don't have normal protection,
383 	 * like send, can have a really old block in cache that may point at a
384 	 * block that has been freed and re-allocated.  So don't clear uptodate
385 	 * if we find an eb that is under IO (dirty/writeback) because we could
386 	 * end up reading in the stale data and then writing it back out and
387 	 * making everybody very sad.
388 	 */
389 	if (!extent_buffer_under_io(eb))
390 		clear_extent_buffer_uptodate(eb);
391 out:
392 	unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
393 			     &cached_state, GFP_NOFS);
394 	if (need_lock)
395 		btrfs_tree_read_unlock_blocking(eb);
396 	return ret;
397 }
398 
399 /*
400  * Return 0 if the superblock checksum type matches the checksum value of that
401  * algorithm. Pass the raw disk superblock data.
402  */
403 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
404 				  char *raw_disk_sb)
405 {
406 	struct btrfs_super_block *disk_sb =
407 		(struct btrfs_super_block *)raw_disk_sb;
408 	u16 csum_type = btrfs_super_csum_type(disk_sb);
409 	int ret = 0;
410 
411 	if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
412 		u32 crc = ~(u32)0;
413 		const int csum_size = sizeof(crc);
414 		char result[csum_size];
415 
416 		/*
417 		 * The super_block structure does not span the whole
418 		 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
419 		 * is filled with zeros and is included in the checksum.
420 		 */
421 		crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
422 				crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
423 		btrfs_csum_final(crc, result);
424 
425 		if (memcmp(raw_disk_sb, result, csum_size))
426 			ret = 1;
427 	}
428 
429 	if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
430 		btrfs_err(fs_info, "unsupported checksum algorithm %u",
431 				csum_type);
432 		ret = 1;
433 	}
434 
435 	return ret;
436 }
437 
438 /*
439  * helper to read a given tree block, doing retries as required when
440  * the checksums don't match and we have alternate mirrors to try.
441  */
442 static int btree_read_extent_buffer_pages(struct btrfs_fs_info *fs_info,
443 					  struct extent_buffer *eb,
444 					  u64 parent_transid)
445 {
446 	struct extent_io_tree *io_tree;
447 	int failed = 0;
448 	int ret;
449 	int num_copies = 0;
450 	int mirror_num = 0;
451 	int failed_mirror = 0;
452 
453 	clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
454 	io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
455 	while (1) {
456 		ret = read_extent_buffer_pages(io_tree, eb, WAIT_COMPLETE,
457 					       btree_get_extent, mirror_num);
458 		if (!ret) {
459 			if (!verify_parent_transid(io_tree, eb,
460 						   parent_transid, 0))
461 				break;
462 			else
463 				ret = -EIO;
464 		}
465 
466 		/*
467 		 * This buffer's crc is fine, but its contents are corrupted, so
468 		 * there is no reason to read the other copies, they won't be
469 		 * any less wrong.
470 		 */
471 		if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
472 			break;
473 
474 		num_copies = btrfs_num_copies(fs_info,
475 					      eb->start, eb->len);
476 		if (num_copies == 1)
477 			break;
478 
479 		if (!failed_mirror) {
480 			failed = 1;
481 			failed_mirror = eb->read_mirror;
482 		}
483 
484 		mirror_num++;
485 		if (mirror_num == failed_mirror)
486 			mirror_num++;
487 
488 		if (mirror_num > num_copies)
489 			break;
490 	}
491 
492 	if (failed && !ret && failed_mirror)
493 		repair_eb_io_failure(fs_info, eb, failed_mirror);
494 
495 	return ret;
496 }
497 
498 /*
499  * checksum a dirty tree block before IO.  This has extra checks to make sure
500  * we only fill in the checksum field in the first page of a multi-page block
501  */
502 
503 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
504 {
505 	u64 start = page_offset(page);
506 	u64 found_start;
507 	struct extent_buffer *eb;
508 
509 	eb = (struct extent_buffer *)page->private;
510 	if (page != eb->pages[0])
511 		return 0;
512 
513 	found_start = btrfs_header_bytenr(eb);
514 	/*
515 	 * Please do not consolidate these warnings into a single if.
516 	 * It is useful to know what went wrong.
517 	 */
518 	if (WARN_ON(found_start != start))
519 		return -EUCLEAN;
520 	if (WARN_ON(!PageUptodate(page)))
521 		return -EUCLEAN;
522 
523 	ASSERT(memcmp_extent_buffer(eb, fs_info->fsid,
524 			btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
525 
526 	return csum_tree_block(fs_info, eb, 0);
527 }
528 
529 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
530 				 struct extent_buffer *eb)
531 {
532 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
533 	u8 fsid[BTRFS_UUID_SIZE];
534 	int ret = 1;
535 
536 	read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
537 	while (fs_devices) {
538 		if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
539 			ret = 0;
540 			break;
541 		}
542 		fs_devices = fs_devices->seed;
543 	}
544 	return ret;
545 }
546 
547 #define CORRUPT(reason, eb, root, slot)					\
548 	btrfs_crit(root->fs_info,					\
549 		   "corrupt %s, %s: block=%llu, root=%llu, slot=%d",	\
550 		   btrfs_header_level(eb) == 0 ? "leaf" : "node",	\
551 		   reason, btrfs_header_bytenr(eb), root->objectid, slot)
552 
553 static noinline int check_leaf(struct btrfs_root *root,
554 			       struct extent_buffer *leaf)
555 {
556 	struct btrfs_fs_info *fs_info = root->fs_info;
557 	struct btrfs_key key;
558 	struct btrfs_key leaf_key;
559 	u32 nritems = btrfs_header_nritems(leaf);
560 	int slot;
561 
562 	/*
563 	 * Extent buffers from a relocation tree have a owner field that
564 	 * corresponds to the subvolume tree they are based on. So just from an
565 	 * extent buffer alone we can not find out what is the id of the
566 	 * corresponding subvolume tree, so we can not figure out if the extent
567 	 * buffer corresponds to the root of the relocation tree or not. So skip
568 	 * this check for relocation trees.
569 	 */
570 	if (nritems == 0 && !btrfs_header_flag(leaf, BTRFS_HEADER_FLAG_RELOC)) {
571 		struct btrfs_root *check_root;
572 
573 		key.objectid = btrfs_header_owner(leaf);
574 		key.type = BTRFS_ROOT_ITEM_KEY;
575 		key.offset = (u64)-1;
576 
577 		check_root = btrfs_get_fs_root(fs_info, &key, false);
578 		/*
579 		 * The only reason we also check NULL here is that during
580 		 * open_ctree() some roots has not yet been set up.
581 		 */
582 		if (!IS_ERR_OR_NULL(check_root)) {
583 			struct extent_buffer *eb;
584 
585 			eb = btrfs_root_node(check_root);
586 			/* if leaf is the root, then it's fine */
587 			if (leaf != eb) {
588 				CORRUPT("non-root leaf's nritems is 0",
589 					leaf, check_root, 0);
590 				free_extent_buffer(eb);
591 				return -EIO;
592 			}
593 			free_extent_buffer(eb);
594 		}
595 		return 0;
596 	}
597 
598 	if (nritems == 0)
599 		return 0;
600 
601 	/* Check the 0 item */
602 	if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
603 	    BTRFS_LEAF_DATA_SIZE(fs_info)) {
604 		CORRUPT("invalid item offset size pair", leaf, root, 0);
605 		return -EIO;
606 	}
607 
608 	/*
609 	 * Check to make sure each items keys are in the correct order and their
610 	 * offsets make sense.  We only have to loop through nritems-1 because
611 	 * we check the current slot against the next slot, which verifies the
612 	 * next slot's offset+size makes sense and that the current's slot
613 	 * offset is correct.
614 	 */
615 	for (slot = 0; slot < nritems - 1; slot++) {
616 		btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
617 		btrfs_item_key_to_cpu(leaf, &key, slot + 1);
618 
619 		/* Make sure the keys are in the right order */
620 		if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
621 			CORRUPT("bad key order", leaf, root, slot);
622 			return -EIO;
623 		}
624 
625 		/*
626 		 * Make sure the offset and ends are right, remember that the
627 		 * item data starts at the end of the leaf and grows towards the
628 		 * front.
629 		 */
630 		if (btrfs_item_offset_nr(leaf, slot) !=
631 			btrfs_item_end_nr(leaf, slot + 1)) {
632 			CORRUPT("slot offset bad", leaf, root, slot);
633 			return -EIO;
634 		}
635 
636 		/*
637 		 * Check to make sure that we don't point outside of the leaf,
638 		 * just in case all the items are consistent to each other, but
639 		 * all point outside of the leaf.
640 		 */
641 		if (btrfs_item_end_nr(leaf, slot) >
642 		    BTRFS_LEAF_DATA_SIZE(fs_info)) {
643 			CORRUPT("slot end outside of leaf", leaf, root, slot);
644 			return -EIO;
645 		}
646 	}
647 
648 	return 0;
649 }
650 
651 static int check_node(struct btrfs_root *root, struct extent_buffer *node)
652 {
653 	unsigned long nr = btrfs_header_nritems(node);
654 	struct btrfs_key key, next_key;
655 	int slot;
656 	u64 bytenr;
657 	int ret = 0;
658 
659 	if (nr == 0 || nr > BTRFS_NODEPTRS_PER_BLOCK(root->fs_info)) {
660 		btrfs_crit(root->fs_info,
661 			   "corrupt node: block %llu root %llu nritems %lu",
662 			   node->start, root->objectid, nr);
663 		return -EIO;
664 	}
665 
666 	for (slot = 0; slot < nr - 1; slot++) {
667 		bytenr = btrfs_node_blockptr(node, slot);
668 		btrfs_node_key_to_cpu(node, &key, slot);
669 		btrfs_node_key_to_cpu(node, &next_key, slot + 1);
670 
671 		if (!bytenr) {
672 			CORRUPT("invalid item slot", node, root, slot);
673 			ret = -EIO;
674 			goto out;
675 		}
676 
677 		if (btrfs_comp_cpu_keys(&key, &next_key) >= 0) {
678 			CORRUPT("bad key order", node, root, slot);
679 			ret = -EIO;
680 			goto out;
681 		}
682 	}
683 out:
684 	return ret;
685 }
686 
687 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
688 				      u64 phy_offset, struct page *page,
689 				      u64 start, u64 end, int mirror)
690 {
691 	u64 found_start;
692 	int found_level;
693 	struct extent_buffer *eb;
694 	struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
695 	struct btrfs_fs_info *fs_info = root->fs_info;
696 	int ret = 0;
697 	int reads_done;
698 
699 	if (!page->private)
700 		goto out;
701 
702 	eb = (struct extent_buffer *)page->private;
703 
704 	/* the pending IO might have been the only thing that kept this buffer
705 	 * in memory.  Make sure we have a ref for all this other checks
706 	 */
707 	extent_buffer_get(eb);
708 
709 	reads_done = atomic_dec_and_test(&eb->io_pages);
710 	if (!reads_done)
711 		goto err;
712 
713 	eb->read_mirror = mirror;
714 	if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
715 		ret = -EIO;
716 		goto err;
717 	}
718 
719 	found_start = btrfs_header_bytenr(eb);
720 	if (found_start != eb->start) {
721 		btrfs_err_rl(fs_info, "bad tree block start %llu %llu",
722 			     found_start, eb->start);
723 		ret = -EIO;
724 		goto err;
725 	}
726 	if (check_tree_block_fsid(fs_info, eb)) {
727 		btrfs_err_rl(fs_info, "bad fsid on block %llu",
728 			     eb->start);
729 		ret = -EIO;
730 		goto err;
731 	}
732 	found_level = btrfs_header_level(eb);
733 	if (found_level >= BTRFS_MAX_LEVEL) {
734 		btrfs_err(fs_info, "bad tree block level %d",
735 			  (int)btrfs_header_level(eb));
736 		ret = -EIO;
737 		goto err;
738 	}
739 
740 	btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
741 				       eb, found_level);
742 
743 	ret = csum_tree_block(fs_info, eb, 1);
744 	if (ret)
745 		goto err;
746 
747 	/*
748 	 * If this is a leaf block and it is corrupt, set the corrupt bit so
749 	 * that we don't try and read the other copies of this block, just
750 	 * return -EIO.
751 	 */
752 	if (found_level == 0 && check_leaf(root, eb)) {
753 		set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
754 		ret = -EIO;
755 	}
756 
757 	if (found_level > 0 && check_node(root, eb))
758 		ret = -EIO;
759 
760 	if (!ret)
761 		set_extent_buffer_uptodate(eb);
762 err:
763 	if (reads_done &&
764 	    test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
765 		btree_readahead_hook(fs_info, eb, ret);
766 
767 	if (ret) {
768 		/*
769 		 * our io error hook is going to dec the io pages
770 		 * again, we have to make sure it has something
771 		 * to decrement
772 		 */
773 		atomic_inc(&eb->io_pages);
774 		clear_extent_buffer_uptodate(eb);
775 	}
776 	free_extent_buffer(eb);
777 out:
778 	return ret;
779 }
780 
781 static int btree_io_failed_hook(struct page *page, int failed_mirror)
782 {
783 	struct extent_buffer *eb;
784 
785 	eb = (struct extent_buffer *)page->private;
786 	set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
787 	eb->read_mirror = failed_mirror;
788 	atomic_dec(&eb->io_pages);
789 	if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
790 		btree_readahead_hook(eb->fs_info, eb, -EIO);
791 	return -EIO;	/* we fixed nothing */
792 }
793 
794 static void end_workqueue_bio(struct bio *bio)
795 {
796 	struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
797 	struct btrfs_fs_info *fs_info;
798 	struct btrfs_workqueue *wq;
799 	btrfs_work_func_t func;
800 
801 	fs_info = end_io_wq->info;
802 	end_io_wq->error = bio->bi_error;
803 
804 	if (bio_op(bio) == REQ_OP_WRITE) {
805 		if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
806 			wq = fs_info->endio_meta_write_workers;
807 			func = btrfs_endio_meta_write_helper;
808 		} else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
809 			wq = fs_info->endio_freespace_worker;
810 			func = btrfs_freespace_write_helper;
811 		} else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
812 			wq = fs_info->endio_raid56_workers;
813 			func = btrfs_endio_raid56_helper;
814 		} else {
815 			wq = fs_info->endio_write_workers;
816 			func = btrfs_endio_write_helper;
817 		}
818 	} else {
819 		if (unlikely(end_io_wq->metadata ==
820 			     BTRFS_WQ_ENDIO_DIO_REPAIR)) {
821 			wq = fs_info->endio_repair_workers;
822 			func = btrfs_endio_repair_helper;
823 		} else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
824 			wq = fs_info->endio_raid56_workers;
825 			func = btrfs_endio_raid56_helper;
826 		} else if (end_io_wq->metadata) {
827 			wq = fs_info->endio_meta_workers;
828 			func = btrfs_endio_meta_helper;
829 		} else {
830 			wq = fs_info->endio_workers;
831 			func = btrfs_endio_helper;
832 		}
833 	}
834 
835 	btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
836 	btrfs_queue_work(wq, &end_io_wq->work);
837 }
838 
839 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
840 			enum btrfs_wq_endio_type metadata)
841 {
842 	struct btrfs_end_io_wq *end_io_wq;
843 
844 	end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
845 	if (!end_io_wq)
846 		return -ENOMEM;
847 
848 	end_io_wq->private = bio->bi_private;
849 	end_io_wq->end_io = bio->bi_end_io;
850 	end_io_wq->info = info;
851 	end_io_wq->error = 0;
852 	end_io_wq->bio = bio;
853 	end_io_wq->metadata = metadata;
854 
855 	bio->bi_private = end_io_wq;
856 	bio->bi_end_io = end_workqueue_bio;
857 	return 0;
858 }
859 
860 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
861 {
862 	unsigned long limit = min_t(unsigned long,
863 				    info->thread_pool_size,
864 				    info->fs_devices->open_devices);
865 	return 256 * limit;
866 }
867 
868 static void run_one_async_start(struct btrfs_work *work)
869 {
870 	struct async_submit_bio *async;
871 	int ret;
872 
873 	async = container_of(work, struct  async_submit_bio, work);
874 	ret = async->submit_bio_start(async->inode, async->bio,
875 				      async->mirror_num, async->bio_flags,
876 				      async->bio_offset);
877 	if (ret)
878 		async->error = ret;
879 }
880 
881 static void run_one_async_done(struct btrfs_work *work)
882 {
883 	struct btrfs_fs_info *fs_info;
884 	struct async_submit_bio *async;
885 	int limit;
886 
887 	async = container_of(work, struct  async_submit_bio, work);
888 	fs_info = BTRFS_I(async->inode)->root->fs_info;
889 
890 	limit = btrfs_async_submit_limit(fs_info);
891 	limit = limit * 2 / 3;
892 
893 	/*
894 	 * atomic_dec_return implies a barrier for waitqueue_active
895 	 */
896 	if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
897 	    waitqueue_active(&fs_info->async_submit_wait))
898 		wake_up(&fs_info->async_submit_wait);
899 
900 	/* If an error occurred we just want to clean up the bio and move on */
901 	if (async->error) {
902 		async->bio->bi_error = async->error;
903 		bio_endio(async->bio);
904 		return;
905 	}
906 
907 	async->submit_bio_done(async->inode, async->bio, async->mirror_num,
908 			       async->bio_flags, async->bio_offset);
909 }
910 
911 static void run_one_async_free(struct btrfs_work *work)
912 {
913 	struct async_submit_bio *async;
914 
915 	async = container_of(work, struct  async_submit_bio, work);
916 	kfree(async);
917 }
918 
919 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
920 			struct bio *bio, int mirror_num,
921 			unsigned long bio_flags,
922 			u64 bio_offset,
923 			extent_submit_bio_hook_t *submit_bio_start,
924 			extent_submit_bio_hook_t *submit_bio_done)
925 {
926 	struct async_submit_bio *async;
927 
928 	async = kmalloc(sizeof(*async), GFP_NOFS);
929 	if (!async)
930 		return -ENOMEM;
931 
932 	async->inode = inode;
933 	async->bio = bio;
934 	async->mirror_num = mirror_num;
935 	async->submit_bio_start = submit_bio_start;
936 	async->submit_bio_done = submit_bio_done;
937 
938 	btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
939 			run_one_async_done, run_one_async_free);
940 
941 	async->bio_flags = bio_flags;
942 	async->bio_offset = bio_offset;
943 
944 	async->error = 0;
945 
946 	atomic_inc(&fs_info->nr_async_submits);
947 
948 	if (op_is_sync(bio->bi_opf))
949 		btrfs_set_work_high_priority(&async->work);
950 
951 	btrfs_queue_work(fs_info->workers, &async->work);
952 
953 	while (atomic_read(&fs_info->async_submit_draining) &&
954 	      atomic_read(&fs_info->nr_async_submits)) {
955 		wait_event(fs_info->async_submit_wait,
956 			   (atomic_read(&fs_info->nr_async_submits) == 0));
957 	}
958 
959 	return 0;
960 }
961 
962 static int btree_csum_one_bio(struct bio *bio)
963 {
964 	struct bio_vec *bvec;
965 	struct btrfs_root *root;
966 	int i, ret = 0;
967 
968 	bio_for_each_segment_all(bvec, bio, i) {
969 		root = BTRFS_I(bvec->bv_page->mapping->host)->root;
970 		ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
971 		if (ret)
972 			break;
973 	}
974 
975 	return ret;
976 }
977 
978 static int __btree_submit_bio_start(struct inode *inode, struct bio *bio,
979 				    int mirror_num, unsigned long bio_flags,
980 				    u64 bio_offset)
981 {
982 	/*
983 	 * when we're called for a write, we're already in the async
984 	 * submission context.  Just jump into btrfs_map_bio
985 	 */
986 	return btree_csum_one_bio(bio);
987 }
988 
989 static int __btree_submit_bio_done(struct inode *inode, struct bio *bio,
990 				 int mirror_num, unsigned long bio_flags,
991 				 u64 bio_offset)
992 {
993 	int ret;
994 
995 	/*
996 	 * when we're called for a write, we're already in the async
997 	 * submission context.  Just jump into btrfs_map_bio
998 	 */
999 	ret = btrfs_map_bio(btrfs_sb(inode->i_sb), bio, mirror_num, 1);
1000 	if (ret) {
1001 		bio->bi_error = ret;
1002 		bio_endio(bio);
1003 	}
1004 	return ret;
1005 }
1006 
1007 static int check_async_write(unsigned long bio_flags)
1008 {
1009 	if (bio_flags & EXTENT_BIO_TREE_LOG)
1010 		return 0;
1011 #ifdef CONFIG_X86
1012 	if (static_cpu_has(X86_FEATURE_XMM4_2))
1013 		return 0;
1014 #endif
1015 	return 1;
1016 }
1017 
1018 static int btree_submit_bio_hook(struct inode *inode, struct bio *bio,
1019 				 int mirror_num, unsigned long bio_flags,
1020 				 u64 bio_offset)
1021 {
1022 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1023 	int async = check_async_write(bio_flags);
1024 	int ret;
1025 
1026 	if (bio_op(bio) != REQ_OP_WRITE) {
1027 		/*
1028 		 * called for a read, do the setup so that checksum validation
1029 		 * can happen in the async kernel threads
1030 		 */
1031 		ret = btrfs_bio_wq_end_io(fs_info, bio,
1032 					  BTRFS_WQ_ENDIO_METADATA);
1033 		if (ret)
1034 			goto out_w_error;
1035 		ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
1036 	} else if (!async) {
1037 		ret = btree_csum_one_bio(bio);
1038 		if (ret)
1039 			goto out_w_error;
1040 		ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
1041 	} else {
1042 		/*
1043 		 * kthread helpers are used to submit writes so that
1044 		 * checksumming can happen in parallel across all CPUs
1045 		 */
1046 		ret = btrfs_wq_submit_bio(fs_info, inode, bio, mirror_num, 0,
1047 					  bio_offset,
1048 					  __btree_submit_bio_start,
1049 					  __btree_submit_bio_done);
1050 	}
1051 
1052 	if (ret)
1053 		goto out_w_error;
1054 	return 0;
1055 
1056 out_w_error:
1057 	bio->bi_error = ret;
1058 	bio_endio(bio);
1059 	return ret;
1060 }
1061 
1062 #ifdef CONFIG_MIGRATION
1063 static int btree_migratepage(struct address_space *mapping,
1064 			struct page *newpage, struct page *page,
1065 			enum migrate_mode mode)
1066 {
1067 	/*
1068 	 * we can't safely write a btree page from here,
1069 	 * we haven't done the locking hook
1070 	 */
1071 	if (PageDirty(page))
1072 		return -EAGAIN;
1073 	/*
1074 	 * Buffers may be managed in a filesystem specific way.
1075 	 * We must have no buffers or drop them.
1076 	 */
1077 	if (page_has_private(page) &&
1078 	    !try_to_release_page(page, GFP_KERNEL))
1079 		return -EAGAIN;
1080 	return migrate_page(mapping, newpage, page, mode);
1081 }
1082 #endif
1083 
1084 
1085 static int btree_writepages(struct address_space *mapping,
1086 			    struct writeback_control *wbc)
1087 {
1088 	struct btrfs_fs_info *fs_info;
1089 	int ret;
1090 
1091 	if (wbc->sync_mode == WB_SYNC_NONE) {
1092 
1093 		if (wbc->for_kupdate)
1094 			return 0;
1095 
1096 		fs_info = BTRFS_I(mapping->host)->root->fs_info;
1097 		/* this is a bit racy, but that's ok */
1098 		ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
1099 					     BTRFS_DIRTY_METADATA_THRESH);
1100 		if (ret < 0)
1101 			return 0;
1102 	}
1103 	return btree_write_cache_pages(mapping, wbc);
1104 }
1105 
1106 static int btree_readpage(struct file *file, struct page *page)
1107 {
1108 	struct extent_io_tree *tree;
1109 	tree = &BTRFS_I(page->mapping->host)->io_tree;
1110 	return extent_read_full_page(tree, page, btree_get_extent, 0);
1111 }
1112 
1113 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1114 {
1115 	if (PageWriteback(page) || PageDirty(page))
1116 		return 0;
1117 
1118 	return try_release_extent_buffer(page);
1119 }
1120 
1121 static void btree_invalidatepage(struct page *page, unsigned int offset,
1122 				 unsigned int length)
1123 {
1124 	struct extent_io_tree *tree;
1125 	tree = &BTRFS_I(page->mapping->host)->io_tree;
1126 	extent_invalidatepage(tree, page, offset);
1127 	btree_releasepage(page, GFP_NOFS);
1128 	if (PagePrivate(page)) {
1129 		btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1130 			   "page private not zero on page %llu",
1131 			   (unsigned long long)page_offset(page));
1132 		ClearPagePrivate(page);
1133 		set_page_private(page, 0);
1134 		put_page(page);
1135 	}
1136 }
1137 
1138 static int btree_set_page_dirty(struct page *page)
1139 {
1140 #ifdef DEBUG
1141 	struct extent_buffer *eb;
1142 
1143 	BUG_ON(!PagePrivate(page));
1144 	eb = (struct extent_buffer *)page->private;
1145 	BUG_ON(!eb);
1146 	BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1147 	BUG_ON(!atomic_read(&eb->refs));
1148 	btrfs_assert_tree_locked(eb);
1149 #endif
1150 	return __set_page_dirty_nobuffers(page);
1151 }
1152 
1153 static const struct address_space_operations btree_aops = {
1154 	.readpage	= btree_readpage,
1155 	.writepages	= btree_writepages,
1156 	.releasepage	= btree_releasepage,
1157 	.invalidatepage = btree_invalidatepage,
1158 #ifdef CONFIG_MIGRATION
1159 	.migratepage	= btree_migratepage,
1160 #endif
1161 	.set_page_dirty = btree_set_page_dirty,
1162 };
1163 
1164 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1165 {
1166 	struct extent_buffer *buf = NULL;
1167 	struct inode *btree_inode = fs_info->btree_inode;
1168 
1169 	buf = btrfs_find_create_tree_block(fs_info, bytenr);
1170 	if (IS_ERR(buf))
1171 		return;
1172 	read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1173 				 buf, WAIT_NONE, btree_get_extent, 0);
1174 	free_extent_buffer(buf);
1175 }
1176 
1177 int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr,
1178 			 int mirror_num, struct extent_buffer **eb)
1179 {
1180 	struct extent_buffer *buf = NULL;
1181 	struct inode *btree_inode = fs_info->btree_inode;
1182 	struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1183 	int ret;
1184 
1185 	buf = btrfs_find_create_tree_block(fs_info, bytenr);
1186 	if (IS_ERR(buf))
1187 		return 0;
1188 
1189 	set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1190 
1191 	ret = read_extent_buffer_pages(io_tree, buf, WAIT_PAGE_LOCK,
1192 				       btree_get_extent, mirror_num);
1193 	if (ret) {
1194 		free_extent_buffer(buf);
1195 		return ret;
1196 	}
1197 
1198 	if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1199 		free_extent_buffer(buf);
1200 		return -EIO;
1201 	} else if (extent_buffer_uptodate(buf)) {
1202 		*eb = buf;
1203 	} else {
1204 		free_extent_buffer(buf);
1205 	}
1206 	return 0;
1207 }
1208 
1209 struct extent_buffer *btrfs_find_create_tree_block(
1210 						struct btrfs_fs_info *fs_info,
1211 						u64 bytenr)
1212 {
1213 	if (btrfs_is_testing(fs_info))
1214 		return alloc_test_extent_buffer(fs_info, bytenr);
1215 	return alloc_extent_buffer(fs_info, bytenr);
1216 }
1217 
1218 
1219 int btrfs_write_tree_block(struct extent_buffer *buf)
1220 {
1221 	return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1222 					buf->start + buf->len - 1);
1223 }
1224 
1225 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1226 {
1227 	return filemap_fdatawait_range(buf->pages[0]->mapping,
1228 				       buf->start, buf->start + buf->len - 1);
1229 }
1230 
1231 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1232 				      u64 parent_transid)
1233 {
1234 	struct extent_buffer *buf = NULL;
1235 	int ret;
1236 
1237 	buf = btrfs_find_create_tree_block(fs_info, bytenr);
1238 	if (IS_ERR(buf))
1239 		return buf;
1240 
1241 	ret = btree_read_extent_buffer_pages(fs_info, buf, parent_transid);
1242 	if (ret) {
1243 		free_extent_buffer(buf);
1244 		return ERR_PTR(ret);
1245 	}
1246 	return buf;
1247 
1248 }
1249 
1250 void clean_tree_block(struct btrfs_fs_info *fs_info,
1251 		      struct extent_buffer *buf)
1252 {
1253 	if (btrfs_header_generation(buf) ==
1254 	    fs_info->running_transaction->transid) {
1255 		btrfs_assert_tree_locked(buf);
1256 
1257 		if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1258 			__percpu_counter_add(&fs_info->dirty_metadata_bytes,
1259 					     -buf->len,
1260 					     fs_info->dirty_metadata_batch);
1261 			/* ugh, clear_extent_buffer_dirty needs to lock the page */
1262 			btrfs_set_lock_blocking(buf);
1263 			clear_extent_buffer_dirty(buf);
1264 		}
1265 	}
1266 }
1267 
1268 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1269 {
1270 	struct btrfs_subvolume_writers *writers;
1271 	int ret;
1272 
1273 	writers = kmalloc(sizeof(*writers), GFP_NOFS);
1274 	if (!writers)
1275 		return ERR_PTR(-ENOMEM);
1276 
1277 	ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL);
1278 	if (ret < 0) {
1279 		kfree(writers);
1280 		return ERR_PTR(ret);
1281 	}
1282 
1283 	init_waitqueue_head(&writers->wait);
1284 	return writers;
1285 }
1286 
1287 static void
1288 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1289 {
1290 	percpu_counter_destroy(&writers->counter);
1291 	kfree(writers);
1292 }
1293 
1294 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1295 			 u64 objectid)
1296 {
1297 	bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1298 	root->node = NULL;
1299 	root->commit_root = NULL;
1300 	root->state = 0;
1301 	root->orphan_cleanup_state = 0;
1302 
1303 	root->objectid = objectid;
1304 	root->last_trans = 0;
1305 	root->highest_objectid = 0;
1306 	root->nr_delalloc_inodes = 0;
1307 	root->nr_ordered_extents = 0;
1308 	root->name = NULL;
1309 	root->inode_tree = RB_ROOT;
1310 	INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1311 	root->block_rsv = NULL;
1312 	root->orphan_block_rsv = NULL;
1313 
1314 	INIT_LIST_HEAD(&root->dirty_list);
1315 	INIT_LIST_HEAD(&root->root_list);
1316 	INIT_LIST_HEAD(&root->delalloc_inodes);
1317 	INIT_LIST_HEAD(&root->delalloc_root);
1318 	INIT_LIST_HEAD(&root->ordered_extents);
1319 	INIT_LIST_HEAD(&root->ordered_root);
1320 	INIT_LIST_HEAD(&root->logged_list[0]);
1321 	INIT_LIST_HEAD(&root->logged_list[1]);
1322 	spin_lock_init(&root->orphan_lock);
1323 	spin_lock_init(&root->inode_lock);
1324 	spin_lock_init(&root->delalloc_lock);
1325 	spin_lock_init(&root->ordered_extent_lock);
1326 	spin_lock_init(&root->accounting_lock);
1327 	spin_lock_init(&root->log_extents_lock[0]);
1328 	spin_lock_init(&root->log_extents_lock[1]);
1329 	mutex_init(&root->objectid_mutex);
1330 	mutex_init(&root->log_mutex);
1331 	mutex_init(&root->ordered_extent_mutex);
1332 	mutex_init(&root->delalloc_mutex);
1333 	init_waitqueue_head(&root->log_writer_wait);
1334 	init_waitqueue_head(&root->log_commit_wait[0]);
1335 	init_waitqueue_head(&root->log_commit_wait[1]);
1336 	INIT_LIST_HEAD(&root->log_ctxs[0]);
1337 	INIT_LIST_HEAD(&root->log_ctxs[1]);
1338 	atomic_set(&root->log_commit[0], 0);
1339 	atomic_set(&root->log_commit[1], 0);
1340 	atomic_set(&root->log_writers, 0);
1341 	atomic_set(&root->log_batch, 0);
1342 	atomic_set(&root->orphan_inodes, 0);
1343 	atomic_set(&root->refs, 1);
1344 	atomic_set(&root->will_be_snapshoted, 0);
1345 	atomic64_set(&root->qgroup_meta_rsv, 0);
1346 	root->log_transid = 0;
1347 	root->log_transid_committed = -1;
1348 	root->last_log_commit = 0;
1349 	if (!dummy)
1350 		extent_io_tree_init(&root->dirty_log_pages,
1351 				     fs_info->btree_inode->i_mapping);
1352 
1353 	memset(&root->root_key, 0, sizeof(root->root_key));
1354 	memset(&root->root_item, 0, sizeof(root->root_item));
1355 	memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1356 	if (!dummy)
1357 		root->defrag_trans_start = fs_info->generation;
1358 	else
1359 		root->defrag_trans_start = 0;
1360 	root->root_key.objectid = objectid;
1361 	root->anon_dev = 0;
1362 
1363 	spin_lock_init(&root->root_item_lock);
1364 }
1365 
1366 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1367 		gfp_t flags)
1368 {
1369 	struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1370 	if (root)
1371 		root->fs_info = fs_info;
1372 	return root;
1373 }
1374 
1375 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1376 /* Should only be used by the testing infrastructure */
1377 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1378 {
1379 	struct btrfs_root *root;
1380 
1381 	if (!fs_info)
1382 		return ERR_PTR(-EINVAL);
1383 
1384 	root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1385 	if (!root)
1386 		return ERR_PTR(-ENOMEM);
1387 
1388 	/* We don't use the stripesize in selftest, set it as sectorsize */
1389 	__setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1390 	root->alloc_bytenr = 0;
1391 
1392 	return root;
1393 }
1394 #endif
1395 
1396 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1397 				     struct btrfs_fs_info *fs_info,
1398 				     u64 objectid)
1399 {
1400 	struct extent_buffer *leaf;
1401 	struct btrfs_root *tree_root = fs_info->tree_root;
1402 	struct btrfs_root *root;
1403 	struct btrfs_key key;
1404 	int ret = 0;
1405 	uuid_le uuid;
1406 
1407 	root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1408 	if (!root)
1409 		return ERR_PTR(-ENOMEM);
1410 
1411 	__setup_root(root, fs_info, objectid);
1412 	root->root_key.objectid = objectid;
1413 	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1414 	root->root_key.offset = 0;
1415 
1416 	leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1417 	if (IS_ERR(leaf)) {
1418 		ret = PTR_ERR(leaf);
1419 		leaf = NULL;
1420 		goto fail;
1421 	}
1422 
1423 	memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1424 	btrfs_set_header_bytenr(leaf, leaf->start);
1425 	btrfs_set_header_generation(leaf, trans->transid);
1426 	btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1427 	btrfs_set_header_owner(leaf, objectid);
1428 	root->node = leaf;
1429 
1430 	write_extent_buffer_fsid(leaf, fs_info->fsid);
1431 	write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
1432 	btrfs_mark_buffer_dirty(leaf);
1433 
1434 	root->commit_root = btrfs_root_node(root);
1435 	set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1436 
1437 	root->root_item.flags = 0;
1438 	root->root_item.byte_limit = 0;
1439 	btrfs_set_root_bytenr(&root->root_item, leaf->start);
1440 	btrfs_set_root_generation(&root->root_item, trans->transid);
1441 	btrfs_set_root_level(&root->root_item, 0);
1442 	btrfs_set_root_refs(&root->root_item, 1);
1443 	btrfs_set_root_used(&root->root_item, leaf->len);
1444 	btrfs_set_root_last_snapshot(&root->root_item, 0);
1445 	btrfs_set_root_dirid(&root->root_item, 0);
1446 	uuid_le_gen(&uuid);
1447 	memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1448 	root->root_item.drop_level = 0;
1449 
1450 	key.objectid = objectid;
1451 	key.type = BTRFS_ROOT_ITEM_KEY;
1452 	key.offset = 0;
1453 	ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1454 	if (ret)
1455 		goto fail;
1456 
1457 	btrfs_tree_unlock(leaf);
1458 
1459 	return root;
1460 
1461 fail:
1462 	if (leaf) {
1463 		btrfs_tree_unlock(leaf);
1464 		free_extent_buffer(root->commit_root);
1465 		free_extent_buffer(leaf);
1466 	}
1467 	kfree(root);
1468 
1469 	return ERR_PTR(ret);
1470 }
1471 
1472 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1473 					 struct btrfs_fs_info *fs_info)
1474 {
1475 	struct btrfs_root *root;
1476 	struct extent_buffer *leaf;
1477 
1478 	root = btrfs_alloc_root(fs_info, GFP_NOFS);
1479 	if (!root)
1480 		return ERR_PTR(-ENOMEM);
1481 
1482 	__setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1483 
1484 	root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1485 	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1486 	root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1487 
1488 	/*
1489 	 * DON'T set REF_COWS for log trees
1490 	 *
1491 	 * log trees do not get reference counted because they go away
1492 	 * before a real commit is actually done.  They do store pointers
1493 	 * to file data extents, and those reference counts still get
1494 	 * updated (along with back refs to the log tree).
1495 	 */
1496 
1497 	leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1498 			NULL, 0, 0, 0);
1499 	if (IS_ERR(leaf)) {
1500 		kfree(root);
1501 		return ERR_CAST(leaf);
1502 	}
1503 
1504 	memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1505 	btrfs_set_header_bytenr(leaf, leaf->start);
1506 	btrfs_set_header_generation(leaf, trans->transid);
1507 	btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1508 	btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1509 	root->node = leaf;
1510 
1511 	write_extent_buffer_fsid(root->node, fs_info->fsid);
1512 	btrfs_mark_buffer_dirty(root->node);
1513 	btrfs_tree_unlock(root->node);
1514 	return root;
1515 }
1516 
1517 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1518 			     struct btrfs_fs_info *fs_info)
1519 {
1520 	struct btrfs_root *log_root;
1521 
1522 	log_root = alloc_log_tree(trans, fs_info);
1523 	if (IS_ERR(log_root))
1524 		return PTR_ERR(log_root);
1525 	WARN_ON(fs_info->log_root_tree);
1526 	fs_info->log_root_tree = log_root;
1527 	return 0;
1528 }
1529 
1530 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1531 		       struct btrfs_root *root)
1532 {
1533 	struct btrfs_fs_info *fs_info = root->fs_info;
1534 	struct btrfs_root *log_root;
1535 	struct btrfs_inode_item *inode_item;
1536 
1537 	log_root = alloc_log_tree(trans, fs_info);
1538 	if (IS_ERR(log_root))
1539 		return PTR_ERR(log_root);
1540 
1541 	log_root->last_trans = trans->transid;
1542 	log_root->root_key.offset = root->root_key.objectid;
1543 
1544 	inode_item = &log_root->root_item.inode;
1545 	btrfs_set_stack_inode_generation(inode_item, 1);
1546 	btrfs_set_stack_inode_size(inode_item, 3);
1547 	btrfs_set_stack_inode_nlink(inode_item, 1);
1548 	btrfs_set_stack_inode_nbytes(inode_item,
1549 				     fs_info->nodesize);
1550 	btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1551 
1552 	btrfs_set_root_node(&log_root->root_item, log_root->node);
1553 
1554 	WARN_ON(root->log_root);
1555 	root->log_root = log_root;
1556 	root->log_transid = 0;
1557 	root->log_transid_committed = -1;
1558 	root->last_log_commit = 0;
1559 	return 0;
1560 }
1561 
1562 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1563 					       struct btrfs_key *key)
1564 {
1565 	struct btrfs_root *root;
1566 	struct btrfs_fs_info *fs_info = tree_root->fs_info;
1567 	struct btrfs_path *path;
1568 	u64 generation;
1569 	int ret;
1570 
1571 	path = btrfs_alloc_path();
1572 	if (!path)
1573 		return ERR_PTR(-ENOMEM);
1574 
1575 	root = btrfs_alloc_root(fs_info, GFP_NOFS);
1576 	if (!root) {
1577 		ret = -ENOMEM;
1578 		goto alloc_fail;
1579 	}
1580 
1581 	__setup_root(root, fs_info, key->objectid);
1582 
1583 	ret = btrfs_find_root(tree_root, key, path,
1584 			      &root->root_item, &root->root_key);
1585 	if (ret) {
1586 		if (ret > 0)
1587 			ret = -ENOENT;
1588 		goto find_fail;
1589 	}
1590 
1591 	generation = btrfs_root_generation(&root->root_item);
1592 	root->node = read_tree_block(fs_info,
1593 				     btrfs_root_bytenr(&root->root_item),
1594 				     generation);
1595 	if (IS_ERR(root->node)) {
1596 		ret = PTR_ERR(root->node);
1597 		goto find_fail;
1598 	} else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1599 		ret = -EIO;
1600 		free_extent_buffer(root->node);
1601 		goto find_fail;
1602 	}
1603 	root->commit_root = btrfs_root_node(root);
1604 out:
1605 	btrfs_free_path(path);
1606 	return root;
1607 
1608 find_fail:
1609 	kfree(root);
1610 alloc_fail:
1611 	root = ERR_PTR(ret);
1612 	goto out;
1613 }
1614 
1615 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1616 				      struct btrfs_key *location)
1617 {
1618 	struct btrfs_root *root;
1619 
1620 	root = btrfs_read_tree_root(tree_root, location);
1621 	if (IS_ERR(root))
1622 		return root;
1623 
1624 	if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1625 		set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1626 		btrfs_check_and_init_root_item(&root->root_item);
1627 	}
1628 
1629 	return root;
1630 }
1631 
1632 int btrfs_init_fs_root(struct btrfs_root *root)
1633 {
1634 	int ret;
1635 	struct btrfs_subvolume_writers *writers;
1636 
1637 	root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1638 	root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1639 					GFP_NOFS);
1640 	if (!root->free_ino_pinned || !root->free_ino_ctl) {
1641 		ret = -ENOMEM;
1642 		goto fail;
1643 	}
1644 
1645 	writers = btrfs_alloc_subvolume_writers();
1646 	if (IS_ERR(writers)) {
1647 		ret = PTR_ERR(writers);
1648 		goto fail;
1649 	}
1650 	root->subv_writers = writers;
1651 
1652 	btrfs_init_free_ino_ctl(root);
1653 	spin_lock_init(&root->ino_cache_lock);
1654 	init_waitqueue_head(&root->ino_cache_wait);
1655 
1656 	ret = get_anon_bdev(&root->anon_dev);
1657 	if (ret)
1658 		goto fail;
1659 
1660 	mutex_lock(&root->objectid_mutex);
1661 	ret = btrfs_find_highest_objectid(root,
1662 					&root->highest_objectid);
1663 	if (ret) {
1664 		mutex_unlock(&root->objectid_mutex);
1665 		goto fail;
1666 	}
1667 
1668 	ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1669 
1670 	mutex_unlock(&root->objectid_mutex);
1671 
1672 	return 0;
1673 fail:
1674 	/* the caller is responsible to call free_fs_root */
1675 	return ret;
1676 }
1677 
1678 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1679 					u64 root_id)
1680 {
1681 	struct btrfs_root *root;
1682 
1683 	spin_lock(&fs_info->fs_roots_radix_lock);
1684 	root = radix_tree_lookup(&fs_info->fs_roots_radix,
1685 				 (unsigned long)root_id);
1686 	spin_unlock(&fs_info->fs_roots_radix_lock);
1687 	return root;
1688 }
1689 
1690 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1691 			 struct btrfs_root *root)
1692 {
1693 	int ret;
1694 
1695 	ret = radix_tree_preload(GFP_NOFS);
1696 	if (ret)
1697 		return ret;
1698 
1699 	spin_lock(&fs_info->fs_roots_radix_lock);
1700 	ret = radix_tree_insert(&fs_info->fs_roots_radix,
1701 				(unsigned long)root->root_key.objectid,
1702 				root);
1703 	if (ret == 0)
1704 		set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1705 	spin_unlock(&fs_info->fs_roots_radix_lock);
1706 	radix_tree_preload_end();
1707 
1708 	return ret;
1709 }
1710 
1711 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1712 				     struct btrfs_key *location,
1713 				     bool check_ref)
1714 {
1715 	struct btrfs_root *root;
1716 	struct btrfs_path *path;
1717 	struct btrfs_key key;
1718 	int ret;
1719 
1720 	if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1721 		return fs_info->tree_root;
1722 	if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1723 		return fs_info->extent_root;
1724 	if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1725 		return fs_info->chunk_root;
1726 	if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1727 		return fs_info->dev_root;
1728 	if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1729 		return fs_info->csum_root;
1730 	if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1731 		return fs_info->quota_root ? fs_info->quota_root :
1732 					     ERR_PTR(-ENOENT);
1733 	if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1734 		return fs_info->uuid_root ? fs_info->uuid_root :
1735 					    ERR_PTR(-ENOENT);
1736 	if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1737 		return fs_info->free_space_root ? fs_info->free_space_root :
1738 						  ERR_PTR(-ENOENT);
1739 again:
1740 	root = btrfs_lookup_fs_root(fs_info, location->objectid);
1741 	if (root) {
1742 		if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1743 			return ERR_PTR(-ENOENT);
1744 		return root;
1745 	}
1746 
1747 	root = btrfs_read_fs_root(fs_info->tree_root, location);
1748 	if (IS_ERR(root))
1749 		return root;
1750 
1751 	if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1752 		ret = -ENOENT;
1753 		goto fail;
1754 	}
1755 
1756 	ret = btrfs_init_fs_root(root);
1757 	if (ret)
1758 		goto fail;
1759 
1760 	path = btrfs_alloc_path();
1761 	if (!path) {
1762 		ret = -ENOMEM;
1763 		goto fail;
1764 	}
1765 	key.objectid = BTRFS_ORPHAN_OBJECTID;
1766 	key.type = BTRFS_ORPHAN_ITEM_KEY;
1767 	key.offset = location->objectid;
1768 
1769 	ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1770 	btrfs_free_path(path);
1771 	if (ret < 0)
1772 		goto fail;
1773 	if (ret == 0)
1774 		set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1775 
1776 	ret = btrfs_insert_fs_root(fs_info, root);
1777 	if (ret) {
1778 		if (ret == -EEXIST) {
1779 			free_fs_root(root);
1780 			goto again;
1781 		}
1782 		goto fail;
1783 	}
1784 	return root;
1785 fail:
1786 	free_fs_root(root);
1787 	return ERR_PTR(ret);
1788 }
1789 
1790 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1791 {
1792 	struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1793 	int ret = 0;
1794 	struct btrfs_device *device;
1795 	struct backing_dev_info *bdi;
1796 
1797 	rcu_read_lock();
1798 	list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1799 		if (!device->bdev)
1800 			continue;
1801 		bdi = device->bdev->bd_bdi;
1802 		if (bdi_congested(bdi, bdi_bits)) {
1803 			ret = 1;
1804 			break;
1805 		}
1806 	}
1807 	rcu_read_unlock();
1808 	return ret;
1809 }
1810 
1811 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1812 {
1813 	int err;
1814 
1815 	err = bdi_setup_and_register(bdi, "btrfs");
1816 	if (err)
1817 		return err;
1818 
1819 	bdi->ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_SIZE;
1820 	bdi->congested_fn	= btrfs_congested_fn;
1821 	bdi->congested_data	= info;
1822 	bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
1823 	return 0;
1824 }
1825 
1826 /*
1827  * called by the kthread helper functions to finally call the bio end_io
1828  * functions.  This is where read checksum verification actually happens
1829  */
1830 static void end_workqueue_fn(struct btrfs_work *work)
1831 {
1832 	struct bio *bio;
1833 	struct btrfs_end_io_wq *end_io_wq;
1834 
1835 	end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1836 	bio = end_io_wq->bio;
1837 
1838 	bio->bi_error = end_io_wq->error;
1839 	bio->bi_private = end_io_wq->private;
1840 	bio->bi_end_io = end_io_wq->end_io;
1841 	kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1842 	bio_endio(bio);
1843 }
1844 
1845 static int cleaner_kthread(void *arg)
1846 {
1847 	struct btrfs_root *root = arg;
1848 	struct btrfs_fs_info *fs_info = root->fs_info;
1849 	int again;
1850 	struct btrfs_trans_handle *trans;
1851 
1852 	do {
1853 		again = 0;
1854 
1855 		/* Make the cleaner go to sleep early. */
1856 		if (btrfs_need_cleaner_sleep(fs_info))
1857 			goto sleep;
1858 
1859 		/*
1860 		 * Do not do anything if we might cause open_ctree() to block
1861 		 * before we have finished mounting the filesystem.
1862 		 */
1863 		if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1864 			goto sleep;
1865 
1866 		if (!mutex_trylock(&fs_info->cleaner_mutex))
1867 			goto sleep;
1868 
1869 		/*
1870 		 * Avoid the problem that we change the status of the fs
1871 		 * during the above check and trylock.
1872 		 */
1873 		if (btrfs_need_cleaner_sleep(fs_info)) {
1874 			mutex_unlock(&fs_info->cleaner_mutex);
1875 			goto sleep;
1876 		}
1877 
1878 		mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
1879 		btrfs_run_delayed_iputs(fs_info);
1880 		mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
1881 
1882 		again = btrfs_clean_one_deleted_snapshot(root);
1883 		mutex_unlock(&fs_info->cleaner_mutex);
1884 
1885 		/*
1886 		 * The defragger has dealt with the R/O remount and umount,
1887 		 * needn't do anything special here.
1888 		 */
1889 		btrfs_run_defrag_inodes(fs_info);
1890 
1891 		/*
1892 		 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1893 		 * with relocation (btrfs_relocate_chunk) and relocation
1894 		 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1895 		 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1896 		 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1897 		 * unused block groups.
1898 		 */
1899 		btrfs_delete_unused_bgs(fs_info);
1900 sleep:
1901 		if (!again) {
1902 			set_current_state(TASK_INTERRUPTIBLE);
1903 			if (!kthread_should_stop())
1904 				schedule();
1905 			__set_current_state(TASK_RUNNING);
1906 		}
1907 	} while (!kthread_should_stop());
1908 
1909 	/*
1910 	 * Transaction kthread is stopped before us and wakes us up.
1911 	 * However we might have started a new transaction and COWed some
1912 	 * tree blocks when deleting unused block groups for example. So
1913 	 * make sure we commit the transaction we started to have a clean
1914 	 * shutdown when evicting the btree inode - if it has dirty pages
1915 	 * when we do the final iput() on it, eviction will trigger a
1916 	 * writeback for it which will fail with null pointer dereferences
1917 	 * since work queues and other resources were already released and
1918 	 * destroyed by the time the iput/eviction/writeback is made.
1919 	 */
1920 	trans = btrfs_attach_transaction(root);
1921 	if (IS_ERR(trans)) {
1922 		if (PTR_ERR(trans) != -ENOENT)
1923 			btrfs_err(fs_info,
1924 				  "cleaner transaction attach returned %ld",
1925 				  PTR_ERR(trans));
1926 	} else {
1927 		int ret;
1928 
1929 		ret = btrfs_commit_transaction(trans);
1930 		if (ret)
1931 			btrfs_err(fs_info,
1932 				  "cleaner open transaction commit returned %d",
1933 				  ret);
1934 	}
1935 
1936 	return 0;
1937 }
1938 
1939 static int transaction_kthread(void *arg)
1940 {
1941 	struct btrfs_root *root = arg;
1942 	struct btrfs_fs_info *fs_info = root->fs_info;
1943 	struct btrfs_trans_handle *trans;
1944 	struct btrfs_transaction *cur;
1945 	u64 transid;
1946 	unsigned long now;
1947 	unsigned long delay;
1948 	bool cannot_commit;
1949 
1950 	do {
1951 		cannot_commit = false;
1952 		delay = HZ * fs_info->commit_interval;
1953 		mutex_lock(&fs_info->transaction_kthread_mutex);
1954 
1955 		spin_lock(&fs_info->trans_lock);
1956 		cur = fs_info->running_transaction;
1957 		if (!cur) {
1958 			spin_unlock(&fs_info->trans_lock);
1959 			goto sleep;
1960 		}
1961 
1962 		now = get_seconds();
1963 		if (cur->state < TRANS_STATE_BLOCKED &&
1964 		    (now < cur->start_time ||
1965 		     now - cur->start_time < fs_info->commit_interval)) {
1966 			spin_unlock(&fs_info->trans_lock);
1967 			delay = HZ * 5;
1968 			goto sleep;
1969 		}
1970 		transid = cur->transid;
1971 		spin_unlock(&fs_info->trans_lock);
1972 
1973 		/* If the file system is aborted, this will always fail. */
1974 		trans = btrfs_attach_transaction(root);
1975 		if (IS_ERR(trans)) {
1976 			if (PTR_ERR(trans) != -ENOENT)
1977 				cannot_commit = true;
1978 			goto sleep;
1979 		}
1980 		if (transid == trans->transid) {
1981 			btrfs_commit_transaction(trans);
1982 		} else {
1983 			btrfs_end_transaction(trans);
1984 		}
1985 sleep:
1986 		wake_up_process(fs_info->cleaner_kthread);
1987 		mutex_unlock(&fs_info->transaction_kthread_mutex);
1988 
1989 		if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1990 				      &fs_info->fs_state)))
1991 			btrfs_cleanup_transaction(fs_info);
1992 		set_current_state(TASK_INTERRUPTIBLE);
1993 		if (!kthread_should_stop() &&
1994 				(!btrfs_transaction_blocked(fs_info) ||
1995 				 cannot_commit))
1996 			schedule_timeout(delay);
1997 		__set_current_state(TASK_RUNNING);
1998 	} while (!kthread_should_stop());
1999 	return 0;
2000 }
2001 
2002 /*
2003  * this will find the highest generation in the array of
2004  * root backups.  The index of the highest array is returned,
2005  * or -1 if we can't find anything.
2006  *
2007  * We check to make sure the array is valid by comparing the
2008  * generation of the latest  root in the array with the generation
2009  * in the super block.  If they don't match we pitch it.
2010  */
2011 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
2012 {
2013 	u64 cur;
2014 	int newest_index = -1;
2015 	struct btrfs_root_backup *root_backup;
2016 	int i;
2017 
2018 	for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2019 		root_backup = info->super_copy->super_roots + i;
2020 		cur = btrfs_backup_tree_root_gen(root_backup);
2021 		if (cur == newest_gen)
2022 			newest_index = i;
2023 	}
2024 
2025 	/* check to see if we actually wrapped around */
2026 	if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
2027 		root_backup = info->super_copy->super_roots;
2028 		cur = btrfs_backup_tree_root_gen(root_backup);
2029 		if (cur == newest_gen)
2030 			newest_index = 0;
2031 	}
2032 	return newest_index;
2033 }
2034 
2035 
2036 /*
2037  * find the oldest backup so we know where to store new entries
2038  * in the backup array.  This will set the backup_root_index
2039  * field in the fs_info struct
2040  */
2041 static void find_oldest_super_backup(struct btrfs_fs_info *info,
2042 				     u64 newest_gen)
2043 {
2044 	int newest_index = -1;
2045 
2046 	newest_index = find_newest_super_backup(info, newest_gen);
2047 	/* if there was garbage in there, just move along */
2048 	if (newest_index == -1) {
2049 		info->backup_root_index = 0;
2050 	} else {
2051 		info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
2052 	}
2053 }
2054 
2055 /*
2056  * copy all the root pointers into the super backup array.
2057  * this will bump the backup pointer by one when it is
2058  * done
2059  */
2060 static void backup_super_roots(struct btrfs_fs_info *info)
2061 {
2062 	int next_backup;
2063 	struct btrfs_root_backup *root_backup;
2064 	int last_backup;
2065 
2066 	next_backup = info->backup_root_index;
2067 	last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
2068 		BTRFS_NUM_BACKUP_ROOTS;
2069 
2070 	/*
2071 	 * just overwrite the last backup if we're at the same generation
2072 	 * this happens only at umount
2073 	 */
2074 	root_backup = info->super_for_commit->super_roots + last_backup;
2075 	if (btrfs_backup_tree_root_gen(root_backup) ==
2076 	    btrfs_header_generation(info->tree_root->node))
2077 		next_backup = last_backup;
2078 
2079 	root_backup = info->super_for_commit->super_roots + next_backup;
2080 
2081 	/*
2082 	 * make sure all of our padding and empty slots get zero filled
2083 	 * regardless of which ones we use today
2084 	 */
2085 	memset(root_backup, 0, sizeof(*root_backup));
2086 
2087 	info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
2088 
2089 	btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
2090 	btrfs_set_backup_tree_root_gen(root_backup,
2091 			       btrfs_header_generation(info->tree_root->node));
2092 
2093 	btrfs_set_backup_tree_root_level(root_backup,
2094 			       btrfs_header_level(info->tree_root->node));
2095 
2096 	btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2097 	btrfs_set_backup_chunk_root_gen(root_backup,
2098 			       btrfs_header_generation(info->chunk_root->node));
2099 	btrfs_set_backup_chunk_root_level(root_backup,
2100 			       btrfs_header_level(info->chunk_root->node));
2101 
2102 	btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
2103 	btrfs_set_backup_extent_root_gen(root_backup,
2104 			       btrfs_header_generation(info->extent_root->node));
2105 	btrfs_set_backup_extent_root_level(root_backup,
2106 			       btrfs_header_level(info->extent_root->node));
2107 
2108 	/*
2109 	 * we might commit during log recovery, which happens before we set
2110 	 * the fs_root.  Make sure it is valid before we fill it in.
2111 	 */
2112 	if (info->fs_root && info->fs_root->node) {
2113 		btrfs_set_backup_fs_root(root_backup,
2114 					 info->fs_root->node->start);
2115 		btrfs_set_backup_fs_root_gen(root_backup,
2116 			       btrfs_header_generation(info->fs_root->node));
2117 		btrfs_set_backup_fs_root_level(root_backup,
2118 			       btrfs_header_level(info->fs_root->node));
2119 	}
2120 
2121 	btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2122 	btrfs_set_backup_dev_root_gen(root_backup,
2123 			       btrfs_header_generation(info->dev_root->node));
2124 	btrfs_set_backup_dev_root_level(root_backup,
2125 				       btrfs_header_level(info->dev_root->node));
2126 
2127 	btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
2128 	btrfs_set_backup_csum_root_gen(root_backup,
2129 			       btrfs_header_generation(info->csum_root->node));
2130 	btrfs_set_backup_csum_root_level(root_backup,
2131 			       btrfs_header_level(info->csum_root->node));
2132 
2133 	btrfs_set_backup_total_bytes(root_backup,
2134 			     btrfs_super_total_bytes(info->super_copy));
2135 	btrfs_set_backup_bytes_used(root_backup,
2136 			     btrfs_super_bytes_used(info->super_copy));
2137 	btrfs_set_backup_num_devices(root_backup,
2138 			     btrfs_super_num_devices(info->super_copy));
2139 
2140 	/*
2141 	 * if we don't copy this out to the super_copy, it won't get remembered
2142 	 * for the next commit
2143 	 */
2144 	memcpy(&info->super_copy->super_roots,
2145 	       &info->super_for_commit->super_roots,
2146 	       sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2147 }
2148 
2149 /*
2150  * this copies info out of the root backup array and back into
2151  * the in-memory super block.  It is meant to help iterate through
2152  * the array, so you send it the number of backups you've already
2153  * tried and the last backup index you used.
2154  *
2155  * this returns -1 when it has tried all the backups
2156  */
2157 static noinline int next_root_backup(struct btrfs_fs_info *info,
2158 				     struct btrfs_super_block *super,
2159 				     int *num_backups_tried, int *backup_index)
2160 {
2161 	struct btrfs_root_backup *root_backup;
2162 	int newest = *backup_index;
2163 
2164 	if (*num_backups_tried == 0) {
2165 		u64 gen = btrfs_super_generation(super);
2166 
2167 		newest = find_newest_super_backup(info, gen);
2168 		if (newest == -1)
2169 			return -1;
2170 
2171 		*backup_index = newest;
2172 		*num_backups_tried = 1;
2173 	} else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2174 		/* we've tried all the backups, all done */
2175 		return -1;
2176 	} else {
2177 		/* jump to the next oldest backup */
2178 		newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2179 			BTRFS_NUM_BACKUP_ROOTS;
2180 		*backup_index = newest;
2181 		*num_backups_tried += 1;
2182 	}
2183 	root_backup = super->super_roots + newest;
2184 
2185 	btrfs_set_super_generation(super,
2186 				   btrfs_backup_tree_root_gen(root_backup));
2187 	btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2188 	btrfs_set_super_root_level(super,
2189 				   btrfs_backup_tree_root_level(root_backup));
2190 	btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2191 
2192 	/*
2193 	 * fixme: the total bytes and num_devices need to match or we should
2194 	 * need a fsck
2195 	 */
2196 	btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2197 	btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2198 	return 0;
2199 }
2200 
2201 /* helper to cleanup workers */
2202 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2203 {
2204 	btrfs_destroy_workqueue(fs_info->fixup_workers);
2205 	btrfs_destroy_workqueue(fs_info->delalloc_workers);
2206 	btrfs_destroy_workqueue(fs_info->workers);
2207 	btrfs_destroy_workqueue(fs_info->endio_workers);
2208 	btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2209 	btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2210 	btrfs_destroy_workqueue(fs_info->rmw_workers);
2211 	btrfs_destroy_workqueue(fs_info->endio_write_workers);
2212 	btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2213 	btrfs_destroy_workqueue(fs_info->submit_workers);
2214 	btrfs_destroy_workqueue(fs_info->delayed_workers);
2215 	btrfs_destroy_workqueue(fs_info->caching_workers);
2216 	btrfs_destroy_workqueue(fs_info->readahead_workers);
2217 	btrfs_destroy_workqueue(fs_info->flush_workers);
2218 	btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2219 	btrfs_destroy_workqueue(fs_info->extent_workers);
2220 	/*
2221 	 * Now that all other work queues are destroyed, we can safely destroy
2222 	 * the queues used for metadata I/O, since tasks from those other work
2223 	 * queues can do metadata I/O operations.
2224 	 */
2225 	btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2226 	btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2227 }
2228 
2229 static void free_root_extent_buffers(struct btrfs_root *root)
2230 {
2231 	if (root) {
2232 		free_extent_buffer(root->node);
2233 		free_extent_buffer(root->commit_root);
2234 		root->node = NULL;
2235 		root->commit_root = NULL;
2236 	}
2237 }
2238 
2239 /* helper to cleanup tree roots */
2240 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2241 {
2242 	free_root_extent_buffers(info->tree_root);
2243 
2244 	free_root_extent_buffers(info->dev_root);
2245 	free_root_extent_buffers(info->extent_root);
2246 	free_root_extent_buffers(info->csum_root);
2247 	free_root_extent_buffers(info->quota_root);
2248 	free_root_extent_buffers(info->uuid_root);
2249 	if (chunk_root)
2250 		free_root_extent_buffers(info->chunk_root);
2251 	free_root_extent_buffers(info->free_space_root);
2252 }
2253 
2254 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2255 {
2256 	int ret;
2257 	struct btrfs_root *gang[8];
2258 	int i;
2259 
2260 	while (!list_empty(&fs_info->dead_roots)) {
2261 		gang[0] = list_entry(fs_info->dead_roots.next,
2262 				     struct btrfs_root, root_list);
2263 		list_del(&gang[0]->root_list);
2264 
2265 		if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2266 			btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2267 		} else {
2268 			free_extent_buffer(gang[0]->node);
2269 			free_extent_buffer(gang[0]->commit_root);
2270 			btrfs_put_fs_root(gang[0]);
2271 		}
2272 	}
2273 
2274 	while (1) {
2275 		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2276 					     (void **)gang, 0,
2277 					     ARRAY_SIZE(gang));
2278 		if (!ret)
2279 			break;
2280 		for (i = 0; i < ret; i++)
2281 			btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2282 	}
2283 
2284 	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2285 		btrfs_free_log_root_tree(NULL, fs_info);
2286 		btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2287 	}
2288 }
2289 
2290 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2291 {
2292 	mutex_init(&fs_info->scrub_lock);
2293 	atomic_set(&fs_info->scrubs_running, 0);
2294 	atomic_set(&fs_info->scrub_pause_req, 0);
2295 	atomic_set(&fs_info->scrubs_paused, 0);
2296 	atomic_set(&fs_info->scrub_cancel_req, 0);
2297 	init_waitqueue_head(&fs_info->scrub_pause_wait);
2298 	fs_info->scrub_workers_refcnt = 0;
2299 }
2300 
2301 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2302 {
2303 	spin_lock_init(&fs_info->balance_lock);
2304 	mutex_init(&fs_info->balance_mutex);
2305 	atomic_set(&fs_info->balance_running, 0);
2306 	atomic_set(&fs_info->balance_pause_req, 0);
2307 	atomic_set(&fs_info->balance_cancel_req, 0);
2308 	fs_info->balance_ctl = NULL;
2309 	init_waitqueue_head(&fs_info->balance_wait_q);
2310 }
2311 
2312 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2313 {
2314 	struct inode *inode = fs_info->btree_inode;
2315 
2316 	inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2317 	set_nlink(inode, 1);
2318 	/*
2319 	 * we set the i_size on the btree inode to the max possible int.
2320 	 * the real end of the address space is determined by all of
2321 	 * the devices in the system
2322 	 */
2323 	inode->i_size = OFFSET_MAX;
2324 	inode->i_mapping->a_ops = &btree_aops;
2325 
2326 	RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2327 	extent_io_tree_init(&BTRFS_I(inode)->io_tree, inode->i_mapping);
2328 	BTRFS_I(inode)->io_tree.track_uptodate = 0;
2329 	extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2330 
2331 	BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2332 
2333 	BTRFS_I(inode)->root = fs_info->tree_root;
2334 	memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2335 	set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2336 	btrfs_insert_inode_hash(inode);
2337 }
2338 
2339 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2340 {
2341 	fs_info->dev_replace.lock_owner = 0;
2342 	atomic_set(&fs_info->dev_replace.nesting_level, 0);
2343 	mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2344 	rwlock_init(&fs_info->dev_replace.lock);
2345 	atomic_set(&fs_info->dev_replace.read_locks, 0);
2346 	atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2347 	init_waitqueue_head(&fs_info->replace_wait);
2348 	init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2349 }
2350 
2351 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2352 {
2353 	spin_lock_init(&fs_info->qgroup_lock);
2354 	mutex_init(&fs_info->qgroup_ioctl_lock);
2355 	fs_info->qgroup_tree = RB_ROOT;
2356 	fs_info->qgroup_op_tree = RB_ROOT;
2357 	INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2358 	fs_info->qgroup_seq = 1;
2359 	fs_info->qgroup_ulist = NULL;
2360 	fs_info->qgroup_rescan_running = false;
2361 	mutex_init(&fs_info->qgroup_rescan_lock);
2362 }
2363 
2364 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2365 		struct btrfs_fs_devices *fs_devices)
2366 {
2367 	int max_active = fs_info->thread_pool_size;
2368 	unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2369 
2370 	fs_info->workers =
2371 		btrfs_alloc_workqueue(fs_info, "worker",
2372 				      flags | WQ_HIGHPRI, max_active, 16);
2373 
2374 	fs_info->delalloc_workers =
2375 		btrfs_alloc_workqueue(fs_info, "delalloc",
2376 				      flags, max_active, 2);
2377 
2378 	fs_info->flush_workers =
2379 		btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2380 				      flags, max_active, 0);
2381 
2382 	fs_info->caching_workers =
2383 		btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2384 
2385 	/*
2386 	 * a higher idle thresh on the submit workers makes it much more
2387 	 * likely that bios will be send down in a sane order to the
2388 	 * devices
2389 	 */
2390 	fs_info->submit_workers =
2391 		btrfs_alloc_workqueue(fs_info, "submit", flags,
2392 				      min_t(u64, fs_devices->num_devices,
2393 					    max_active), 64);
2394 
2395 	fs_info->fixup_workers =
2396 		btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2397 
2398 	/*
2399 	 * endios are largely parallel and should have a very
2400 	 * low idle thresh
2401 	 */
2402 	fs_info->endio_workers =
2403 		btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2404 	fs_info->endio_meta_workers =
2405 		btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2406 				      max_active, 4);
2407 	fs_info->endio_meta_write_workers =
2408 		btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2409 				      max_active, 2);
2410 	fs_info->endio_raid56_workers =
2411 		btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2412 				      max_active, 4);
2413 	fs_info->endio_repair_workers =
2414 		btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2415 	fs_info->rmw_workers =
2416 		btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2417 	fs_info->endio_write_workers =
2418 		btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2419 				      max_active, 2);
2420 	fs_info->endio_freespace_worker =
2421 		btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2422 				      max_active, 0);
2423 	fs_info->delayed_workers =
2424 		btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2425 				      max_active, 0);
2426 	fs_info->readahead_workers =
2427 		btrfs_alloc_workqueue(fs_info, "readahead", flags,
2428 				      max_active, 2);
2429 	fs_info->qgroup_rescan_workers =
2430 		btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2431 	fs_info->extent_workers =
2432 		btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2433 				      min_t(u64, fs_devices->num_devices,
2434 					    max_active), 8);
2435 
2436 	if (!(fs_info->workers && fs_info->delalloc_workers &&
2437 	      fs_info->submit_workers && fs_info->flush_workers &&
2438 	      fs_info->endio_workers && fs_info->endio_meta_workers &&
2439 	      fs_info->endio_meta_write_workers &&
2440 	      fs_info->endio_repair_workers &&
2441 	      fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2442 	      fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2443 	      fs_info->caching_workers && fs_info->readahead_workers &&
2444 	      fs_info->fixup_workers && fs_info->delayed_workers &&
2445 	      fs_info->extent_workers &&
2446 	      fs_info->qgroup_rescan_workers)) {
2447 		return -ENOMEM;
2448 	}
2449 
2450 	return 0;
2451 }
2452 
2453 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2454 			    struct btrfs_fs_devices *fs_devices)
2455 {
2456 	int ret;
2457 	struct btrfs_root *log_tree_root;
2458 	struct btrfs_super_block *disk_super = fs_info->super_copy;
2459 	u64 bytenr = btrfs_super_log_root(disk_super);
2460 
2461 	if (fs_devices->rw_devices == 0) {
2462 		btrfs_warn(fs_info, "log replay required on RO media");
2463 		return -EIO;
2464 	}
2465 
2466 	log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2467 	if (!log_tree_root)
2468 		return -ENOMEM;
2469 
2470 	__setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2471 
2472 	log_tree_root->node = read_tree_block(fs_info, bytenr,
2473 					      fs_info->generation + 1);
2474 	if (IS_ERR(log_tree_root->node)) {
2475 		btrfs_warn(fs_info, "failed to read log tree");
2476 		ret = PTR_ERR(log_tree_root->node);
2477 		kfree(log_tree_root);
2478 		return ret;
2479 	} else if (!extent_buffer_uptodate(log_tree_root->node)) {
2480 		btrfs_err(fs_info, "failed to read log tree");
2481 		free_extent_buffer(log_tree_root->node);
2482 		kfree(log_tree_root);
2483 		return -EIO;
2484 	}
2485 	/* returns with log_tree_root freed on success */
2486 	ret = btrfs_recover_log_trees(log_tree_root);
2487 	if (ret) {
2488 		btrfs_handle_fs_error(fs_info, ret,
2489 				      "Failed to recover log tree");
2490 		free_extent_buffer(log_tree_root->node);
2491 		kfree(log_tree_root);
2492 		return ret;
2493 	}
2494 
2495 	if (fs_info->sb->s_flags & MS_RDONLY) {
2496 		ret = btrfs_commit_super(fs_info);
2497 		if (ret)
2498 			return ret;
2499 	}
2500 
2501 	return 0;
2502 }
2503 
2504 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2505 {
2506 	struct btrfs_root *tree_root = fs_info->tree_root;
2507 	struct btrfs_root *root;
2508 	struct btrfs_key location;
2509 	int ret;
2510 
2511 	BUG_ON(!fs_info->tree_root);
2512 
2513 	location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2514 	location.type = BTRFS_ROOT_ITEM_KEY;
2515 	location.offset = 0;
2516 
2517 	root = btrfs_read_tree_root(tree_root, &location);
2518 	if (IS_ERR(root))
2519 		return PTR_ERR(root);
2520 	set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2521 	fs_info->extent_root = root;
2522 
2523 	location.objectid = BTRFS_DEV_TREE_OBJECTID;
2524 	root = btrfs_read_tree_root(tree_root, &location);
2525 	if (IS_ERR(root))
2526 		return PTR_ERR(root);
2527 	set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2528 	fs_info->dev_root = root;
2529 	btrfs_init_devices_late(fs_info);
2530 
2531 	location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2532 	root = btrfs_read_tree_root(tree_root, &location);
2533 	if (IS_ERR(root))
2534 		return PTR_ERR(root);
2535 	set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2536 	fs_info->csum_root = root;
2537 
2538 	location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2539 	root = btrfs_read_tree_root(tree_root, &location);
2540 	if (!IS_ERR(root)) {
2541 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2542 		set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2543 		fs_info->quota_root = root;
2544 	}
2545 
2546 	location.objectid = BTRFS_UUID_TREE_OBJECTID;
2547 	root = btrfs_read_tree_root(tree_root, &location);
2548 	if (IS_ERR(root)) {
2549 		ret = PTR_ERR(root);
2550 		if (ret != -ENOENT)
2551 			return ret;
2552 	} else {
2553 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2554 		fs_info->uuid_root = root;
2555 	}
2556 
2557 	if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2558 		location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2559 		root = btrfs_read_tree_root(tree_root, &location);
2560 		if (IS_ERR(root))
2561 			return PTR_ERR(root);
2562 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2563 		fs_info->free_space_root = root;
2564 	}
2565 
2566 	return 0;
2567 }
2568 
2569 int open_ctree(struct super_block *sb,
2570 	       struct btrfs_fs_devices *fs_devices,
2571 	       char *options)
2572 {
2573 	u32 sectorsize;
2574 	u32 nodesize;
2575 	u32 stripesize;
2576 	u64 generation;
2577 	u64 features;
2578 	struct btrfs_key location;
2579 	struct buffer_head *bh;
2580 	struct btrfs_super_block *disk_super;
2581 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2582 	struct btrfs_root *tree_root;
2583 	struct btrfs_root *chunk_root;
2584 	int ret;
2585 	int err = -EINVAL;
2586 	int num_backups_tried = 0;
2587 	int backup_index = 0;
2588 	int max_active;
2589 	int clear_free_space_tree = 0;
2590 
2591 	tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2592 	chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2593 	if (!tree_root || !chunk_root) {
2594 		err = -ENOMEM;
2595 		goto fail;
2596 	}
2597 
2598 	ret = init_srcu_struct(&fs_info->subvol_srcu);
2599 	if (ret) {
2600 		err = ret;
2601 		goto fail;
2602 	}
2603 
2604 	ret = setup_bdi(fs_info, &fs_info->bdi);
2605 	if (ret) {
2606 		err = ret;
2607 		goto fail_srcu;
2608 	}
2609 
2610 	ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2611 	if (ret) {
2612 		err = ret;
2613 		goto fail_bdi;
2614 	}
2615 	fs_info->dirty_metadata_batch = PAGE_SIZE *
2616 					(1 + ilog2(nr_cpu_ids));
2617 
2618 	ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2619 	if (ret) {
2620 		err = ret;
2621 		goto fail_dirty_metadata_bytes;
2622 	}
2623 
2624 	ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2625 	if (ret) {
2626 		err = ret;
2627 		goto fail_delalloc_bytes;
2628 	}
2629 
2630 	fs_info->btree_inode = new_inode(sb);
2631 	if (!fs_info->btree_inode) {
2632 		err = -ENOMEM;
2633 		goto fail_bio_counter;
2634 	}
2635 
2636 	mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2637 
2638 	INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2639 	INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2640 	INIT_LIST_HEAD(&fs_info->trans_list);
2641 	INIT_LIST_HEAD(&fs_info->dead_roots);
2642 	INIT_LIST_HEAD(&fs_info->delayed_iputs);
2643 	INIT_LIST_HEAD(&fs_info->delalloc_roots);
2644 	INIT_LIST_HEAD(&fs_info->caching_block_groups);
2645 	spin_lock_init(&fs_info->delalloc_root_lock);
2646 	spin_lock_init(&fs_info->trans_lock);
2647 	spin_lock_init(&fs_info->fs_roots_radix_lock);
2648 	spin_lock_init(&fs_info->delayed_iput_lock);
2649 	spin_lock_init(&fs_info->defrag_inodes_lock);
2650 	spin_lock_init(&fs_info->free_chunk_lock);
2651 	spin_lock_init(&fs_info->tree_mod_seq_lock);
2652 	spin_lock_init(&fs_info->super_lock);
2653 	spin_lock_init(&fs_info->qgroup_op_lock);
2654 	spin_lock_init(&fs_info->buffer_lock);
2655 	spin_lock_init(&fs_info->unused_bgs_lock);
2656 	rwlock_init(&fs_info->tree_mod_log_lock);
2657 	mutex_init(&fs_info->unused_bg_unpin_mutex);
2658 	mutex_init(&fs_info->delete_unused_bgs_mutex);
2659 	mutex_init(&fs_info->reloc_mutex);
2660 	mutex_init(&fs_info->delalloc_root_mutex);
2661 	mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2662 	seqlock_init(&fs_info->profiles_lock);
2663 
2664 	INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2665 	INIT_LIST_HEAD(&fs_info->space_info);
2666 	INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2667 	INIT_LIST_HEAD(&fs_info->unused_bgs);
2668 	btrfs_mapping_init(&fs_info->mapping_tree);
2669 	btrfs_init_block_rsv(&fs_info->global_block_rsv,
2670 			     BTRFS_BLOCK_RSV_GLOBAL);
2671 	btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2672 			     BTRFS_BLOCK_RSV_DELALLOC);
2673 	btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2674 	btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2675 	btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2676 	btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2677 			     BTRFS_BLOCK_RSV_DELOPS);
2678 	atomic_set(&fs_info->nr_async_submits, 0);
2679 	atomic_set(&fs_info->async_delalloc_pages, 0);
2680 	atomic_set(&fs_info->async_submit_draining, 0);
2681 	atomic_set(&fs_info->nr_async_bios, 0);
2682 	atomic_set(&fs_info->defrag_running, 0);
2683 	atomic_set(&fs_info->qgroup_op_seq, 0);
2684 	atomic_set(&fs_info->reada_works_cnt, 0);
2685 	atomic64_set(&fs_info->tree_mod_seq, 0);
2686 	fs_info->fs_frozen = 0;
2687 	fs_info->sb = sb;
2688 	fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2689 	fs_info->metadata_ratio = 0;
2690 	fs_info->defrag_inodes = RB_ROOT;
2691 	fs_info->free_chunk_space = 0;
2692 	fs_info->tree_mod_log = RB_ROOT;
2693 	fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2694 	fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2695 	/* readahead state */
2696 	INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2697 	spin_lock_init(&fs_info->reada_lock);
2698 
2699 	fs_info->thread_pool_size = min_t(unsigned long,
2700 					  num_online_cpus() + 2, 8);
2701 
2702 	INIT_LIST_HEAD(&fs_info->ordered_roots);
2703 	spin_lock_init(&fs_info->ordered_root_lock);
2704 	fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2705 					GFP_KERNEL);
2706 	if (!fs_info->delayed_root) {
2707 		err = -ENOMEM;
2708 		goto fail_iput;
2709 	}
2710 	btrfs_init_delayed_root(fs_info->delayed_root);
2711 
2712 	btrfs_init_scrub(fs_info);
2713 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2714 	fs_info->check_integrity_print_mask = 0;
2715 #endif
2716 	btrfs_init_balance(fs_info);
2717 	btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2718 
2719 	sb->s_blocksize = 4096;
2720 	sb->s_blocksize_bits = blksize_bits(4096);
2721 	sb->s_bdi = &fs_info->bdi;
2722 
2723 	btrfs_init_btree_inode(fs_info);
2724 
2725 	spin_lock_init(&fs_info->block_group_cache_lock);
2726 	fs_info->block_group_cache_tree = RB_ROOT;
2727 	fs_info->first_logical_byte = (u64)-1;
2728 
2729 	extent_io_tree_init(&fs_info->freed_extents[0],
2730 			     fs_info->btree_inode->i_mapping);
2731 	extent_io_tree_init(&fs_info->freed_extents[1],
2732 			     fs_info->btree_inode->i_mapping);
2733 	fs_info->pinned_extents = &fs_info->freed_extents[0];
2734 	set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2735 
2736 	mutex_init(&fs_info->ordered_operations_mutex);
2737 	mutex_init(&fs_info->tree_log_mutex);
2738 	mutex_init(&fs_info->chunk_mutex);
2739 	mutex_init(&fs_info->transaction_kthread_mutex);
2740 	mutex_init(&fs_info->cleaner_mutex);
2741 	mutex_init(&fs_info->volume_mutex);
2742 	mutex_init(&fs_info->ro_block_group_mutex);
2743 	init_rwsem(&fs_info->commit_root_sem);
2744 	init_rwsem(&fs_info->cleanup_work_sem);
2745 	init_rwsem(&fs_info->subvol_sem);
2746 	sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2747 
2748 	btrfs_init_dev_replace_locks(fs_info);
2749 	btrfs_init_qgroup(fs_info);
2750 
2751 	btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2752 	btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2753 
2754 	init_waitqueue_head(&fs_info->transaction_throttle);
2755 	init_waitqueue_head(&fs_info->transaction_wait);
2756 	init_waitqueue_head(&fs_info->transaction_blocked_wait);
2757 	init_waitqueue_head(&fs_info->async_submit_wait);
2758 
2759 	INIT_LIST_HEAD(&fs_info->pinned_chunks);
2760 
2761 	/* Usable values until the real ones are cached from the superblock */
2762 	fs_info->nodesize = 4096;
2763 	fs_info->sectorsize = 4096;
2764 	fs_info->stripesize = 4096;
2765 
2766 	ret = btrfs_alloc_stripe_hash_table(fs_info);
2767 	if (ret) {
2768 		err = ret;
2769 		goto fail_alloc;
2770 	}
2771 
2772 	__setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2773 
2774 	invalidate_bdev(fs_devices->latest_bdev);
2775 
2776 	/*
2777 	 * Read super block and check the signature bytes only
2778 	 */
2779 	bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2780 	if (IS_ERR(bh)) {
2781 		err = PTR_ERR(bh);
2782 		goto fail_alloc;
2783 	}
2784 
2785 	/*
2786 	 * We want to check superblock checksum, the type is stored inside.
2787 	 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2788 	 */
2789 	if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2790 		btrfs_err(fs_info, "superblock checksum mismatch");
2791 		err = -EINVAL;
2792 		brelse(bh);
2793 		goto fail_alloc;
2794 	}
2795 
2796 	/*
2797 	 * super_copy is zeroed at allocation time and we never touch the
2798 	 * following bytes up to INFO_SIZE, the checksum is calculated from
2799 	 * the whole block of INFO_SIZE
2800 	 */
2801 	memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2802 	memcpy(fs_info->super_for_commit, fs_info->super_copy,
2803 	       sizeof(*fs_info->super_for_commit));
2804 	brelse(bh);
2805 
2806 	memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2807 
2808 	ret = btrfs_check_super_valid(fs_info);
2809 	if (ret) {
2810 		btrfs_err(fs_info, "superblock contains fatal errors");
2811 		err = -EINVAL;
2812 		goto fail_alloc;
2813 	}
2814 
2815 	disk_super = fs_info->super_copy;
2816 	if (!btrfs_super_root(disk_super))
2817 		goto fail_alloc;
2818 
2819 	/* check FS state, whether FS is broken. */
2820 	if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2821 		set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2822 
2823 	/*
2824 	 * run through our array of backup supers and setup
2825 	 * our ring pointer to the oldest one
2826 	 */
2827 	generation = btrfs_super_generation(disk_super);
2828 	find_oldest_super_backup(fs_info, generation);
2829 
2830 	/*
2831 	 * In the long term, we'll store the compression type in the super
2832 	 * block, and it'll be used for per file compression control.
2833 	 */
2834 	fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2835 
2836 	ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2837 	if (ret) {
2838 		err = ret;
2839 		goto fail_alloc;
2840 	}
2841 
2842 	features = btrfs_super_incompat_flags(disk_super) &
2843 		~BTRFS_FEATURE_INCOMPAT_SUPP;
2844 	if (features) {
2845 		btrfs_err(fs_info,
2846 		    "cannot mount because of unsupported optional features (%llx)",
2847 		    features);
2848 		err = -EINVAL;
2849 		goto fail_alloc;
2850 	}
2851 
2852 	features = btrfs_super_incompat_flags(disk_super);
2853 	features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2854 	if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2855 		features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2856 
2857 	if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2858 		btrfs_info(fs_info, "has skinny extents");
2859 
2860 	/*
2861 	 * flag our filesystem as having big metadata blocks if
2862 	 * they are bigger than the page size
2863 	 */
2864 	if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2865 		if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2866 			btrfs_info(fs_info,
2867 				"flagging fs with big metadata feature");
2868 		features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2869 	}
2870 
2871 	nodesize = btrfs_super_nodesize(disk_super);
2872 	sectorsize = btrfs_super_sectorsize(disk_super);
2873 	stripesize = sectorsize;
2874 	fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2875 	fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2876 
2877 	/* Cache block sizes */
2878 	fs_info->nodesize = nodesize;
2879 	fs_info->sectorsize = sectorsize;
2880 	fs_info->stripesize = stripesize;
2881 
2882 	/*
2883 	 * mixed block groups end up with duplicate but slightly offset
2884 	 * extent buffers for the same range.  It leads to corruptions
2885 	 */
2886 	if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2887 	    (sectorsize != nodesize)) {
2888 		btrfs_err(fs_info,
2889 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2890 			nodesize, sectorsize);
2891 		goto fail_alloc;
2892 	}
2893 
2894 	/*
2895 	 * Needn't use the lock because there is no other task which will
2896 	 * update the flag.
2897 	 */
2898 	btrfs_set_super_incompat_flags(disk_super, features);
2899 
2900 	features = btrfs_super_compat_ro_flags(disk_super) &
2901 		~BTRFS_FEATURE_COMPAT_RO_SUPP;
2902 	if (!(sb->s_flags & MS_RDONLY) && features) {
2903 		btrfs_err(fs_info,
2904 	"cannot mount read-write because of unsupported optional features (%llx)",
2905 		       features);
2906 		err = -EINVAL;
2907 		goto fail_alloc;
2908 	}
2909 
2910 	max_active = fs_info->thread_pool_size;
2911 
2912 	ret = btrfs_init_workqueues(fs_info, fs_devices);
2913 	if (ret) {
2914 		err = ret;
2915 		goto fail_sb_buffer;
2916 	}
2917 
2918 	fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2919 	fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2920 				    SZ_4M / PAGE_SIZE);
2921 
2922 	sb->s_blocksize = sectorsize;
2923 	sb->s_blocksize_bits = blksize_bits(sectorsize);
2924 
2925 	mutex_lock(&fs_info->chunk_mutex);
2926 	ret = btrfs_read_sys_array(fs_info);
2927 	mutex_unlock(&fs_info->chunk_mutex);
2928 	if (ret) {
2929 		btrfs_err(fs_info, "failed to read the system array: %d", ret);
2930 		goto fail_sb_buffer;
2931 	}
2932 
2933 	generation = btrfs_super_chunk_root_generation(disk_super);
2934 
2935 	__setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2936 
2937 	chunk_root->node = read_tree_block(fs_info,
2938 					   btrfs_super_chunk_root(disk_super),
2939 					   generation);
2940 	if (IS_ERR(chunk_root->node) ||
2941 	    !extent_buffer_uptodate(chunk_root->node)) {
2942 		btrfs_err(fs_info, "failed to read chunk root");
2943 		if (!IS_ERR(chunk_root->node))
2944 			free_extent_buffer(chunk_root->node);
2945 		chunk_root->node = NULL;
2946 		goto fail_tree_roots;
2947 	}
2948 	btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2949 	chunk_root->commit_root = btrfs_root_node(chunk_root);
2950 
2951 	read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2952 	   btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2953 
2954 	ret = btrfs_read_chunk_tree(fs_info);
2955 	if (ret) {
2956 		btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
2957 		goto fail_tree_roots;
2958 	}
2959 
2960 	/*
2961 	 * keep the device that is marked to be the target device for the
2962 	 * dev_replace procedure
2963 	 */
2964 	btrfs_close_extra_devices(fs_devices, 0);
2965 
2966 	if (!fs_devices->latest_bdev) {
2967 		btrfs_err(fs_info, "failed to read devices");
2968 		goto fail_tree_roots;
2969 	}
2970 
2971 retry_root_backup:
2972 	generation = btrfs_super_generation(disk_super);
2973 
2974 	tree_root->node = read_tree_block(fs_info,
2975 					  btrfs_super_root(disk_super),
2976 					  generation);
2977 	if (IS_ERR(tree_root->node) ||
2978 	    !extent_buffer_uptodate(tree_root->node)) {
2979 		btrfs_warn(fs_info, "failed to read tree root");
2980 		if (!IS_ERR(tree_root->node))
2981 			free_extent_buffer(tree_root->node);
2982 		tree_root->node = NULL;
2983 		goto recovery_tree_root;
2984 	}
2985 
2986 	btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2987 	tree_root->commit_root = btrfs_root_node(tree_root);
2988 	btrfs_set_root_refs(&tree_root->root_item, 1);
2989 
2990 	mutex_lock(&tree_root->objectid_mutex);
2991 	ret = btrfs_find_highest_objectid(tree_root,
2992 					&tree_root->highest_objectid);
2993 	if (ret) {
2994 		mutex_unlock(&tree_root->objectid_mutex);
2995 		goto recovery_tree_root;
2996 	}
2997 
2998 	ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2999 
3000 	mutex_unlock(&tree_root->objectid_mutex);
3001 
3002 	ret = btrfs_read_roots(fs_info);
3003 	if (ret)
3004 		goto recovery_tree_root;
3005 
3006 	fs_info->generation = generation;
3007 	fs_info->last_trans_committed = generation;
3008 
3009 	ret = btrfs_recover_balance(fs_info);
3010 	if (ret) {
3011 		btrfs_err(fs_info, "failed to recover balance: %d", ret);
3012 		goto fail_block_groups;
3013 	}
3014 
3015 	ret = btrfs_init_dev_stats(fs_info);
3016 	if (ret) {
3017 		btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3018 		goto fail_block_groups;
3019 	}
3020 
3021 	ret = btrfs_init_dev_replace(fs_info);
3022 	if (ret) {
3023 		btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3024 		goto fail_block_groups;
3025 	}
3026 
3027 	btrfs_close_extra_devices(fs_devices, 1);
3028 
3029 	ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
3030 	if (ret) {
3031 		btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3032 				ret);
3033 		goto fail_block_groups;
3034 	}
3035 
3036 	ret = btrfs_sysfs_add_device(fs_devices);
3037 	if (ret) {
3038 		btrfs_err(fs_info, "failed to init sysfs device interface: %d",
3039 				ret);
3040 		goto fail_fsdev_sysfs;
3041 	}
3042 
3043 	ret = btrfs_sysfs_add_mounted(fs_info);
3044 	if (ret) {
3045 		btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3046 		goto fail_fsdev_sysfs;
3047 	}
3048 
3049 	ret = btrfs_init_space_info(fs_info);
3050 	if (ret) {
3051 		btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3052 		goto fail_sysfs;
3053 	}
3054 
3055 	ret = btrfs_read_block_groups(fs_info);
3056 	if (ret) {
3057 		btrfs_err(fs_info, "failed to read block groups: %d", ret);
3058 		goto fail_sysfs;
3059 	}
3060 	fs_info->num_tolerated_disk_barrier_failures =
3061 		btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3062 	if (fs_info->fs_devices->missing_devices >
3063 	     fs_info->num_tolerated_disk_barrier_failures &&
3064 	    !(sb->s_flags & MS_RDONLY)) {
3065 		btrfs_warn(fs_info,
3066 "missing devices (%llu) exceeds the limit (%d), writeable mount is not allowed",
3067 			fs_info->fs_devices->missing_devices,
3068 			fs_info->num_tolerated_disk_barrier_failures);
3069 		goto fail_sysfs;
3070 	}
3071 
3072 	fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3073 					       "btrfs-cleaner");
3074 	if (IS_ERR(fs_info->cleaner_kthread))
3075 		goto fail_sysfs;
3076 
3077 	fs_info->transaction_kthread = kthread_run(transaction_kthread,
3078 						   tree_root,
3079 						   "btrfs-transaction");
3080 	if (IS_ERR(fs_info->transaction_kthread))
3081 		goto fail_cleaner;
3082 
3083 	if (!btrfs_test_opt(fs_info, SSD) &&
3084 	    !btrfs_test_opt(fs_info, NOSSD) &&
3085 	    !fs_info->fs_devices->rotating) {
3086 		btrfs_info(fs_info, "detected SSD devices, enabling SSD mode");
3087 		btrfs_set_opt(fs_info->mount_opt, SSD);
3088 	}
3089 
3090 	/*
3091 	 * Mount does not set all options immediately, we can do it now and do
3092 	 * not have to wait for transaction commit
3093 	 */
3094 	btrfs_apply_pending_changes(fs_info);
3095 
3096 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3097 	if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3098 		ret = btrfsic_mount(fs_info, fs_devices,
3099 				    btrfs_test_opt(fs_info,
3100 					CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3101 				    1 : 0,
3102 				    fs_info->check_integrity_print_mask);
3103 		if (ret)
3104 			btrfs_warn(fs_info,
3105 				"failed to initialize integrity check module: %d",
3106 				ret);
3107 	}
3108 #endif
3109 	ret = btrfs_read_qgroup_config(fs_info);
3110 	if (ret)
3111 		goto fail_trans_kthread;
3112 
3113 	/* do not make disk changes in broken FS or nologreplay is given */
3114 	if (btrfs_super_log_root(disk_super) != 0 &&
3115 	    !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3116 		ret = btrfs_replay_log(fs_info, fs_devices);
3117 		if (ret) {
3118 			err = ret;
3119 			goto fail_qgroup;
3120 		}
3121 	}
3122 
3123 	ret = btrfs_find_orphan_roots(fs_info);
3124 	if (ret)
3125 		goto fail_qgroup;
3126 
3127 	if (!(sb->s_flags & MS_RDONLY)) {
3128 		ret = btrfs_cleanup_fs_roots(fs_info);
3129 		if (ret)
3130 			goto fail_qgroup;
3131 
3132 		mutex_lock(&fs_info->cleaner_mutex);
3133 		ret = btrfs_recover_relocation(tree_root);
3134 		mutex_unlock(&fs_info->cleaner_mutex);
3135 		if (ret < 0) {
3136 			btrfs_warn(fs_info, "failed to recover relocation: %d",
3137 					ret);
3138 			err = -EINVAL;
3139 			goto fail_qgroup;
3140 		}
3141 	}
3142 
3143 	location.objectid = BTRFS_FS_TREE_OBJECTID;
3144 	location.type = BTRFS_ROOT_ITEM_KEY;
3145 	location.offset = 0;
3146 
3147 	fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3148 	if (IS_ERR(fs_info->fs_root)) {
3149 		err = PTR_ERR(fs_info->fs_root);
3150 		goto fail_qgroup;
3151 	}
3152 
3153 	if (sb->s_flags & MS_RDONLY)
3154 		return 0;
3155 
3156 	if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3157 	    btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3158 		clear_free_space_tree = 1;
3159 	} else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3160 		   !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3161 		btrfs_warn(fs_info, "free space tree is invalid");
3162 		clear_free_space_tree = 1;
3163 	}
3164 
3165 	if (clear_free_space_tree) {
3166 		btrfs_info(fs_info, "clearing free space tree");
3167 		ret = btrfs_clear_free_space_tree(fs_info);
3168 		if (ret) {
3169 			btrfs_warn(fs_info,
3170 				   "failed to clear free space tree: %d", ret);
3171 			close_ctree(fs_info);
3172 			return ret;
3173 		}
3174 	}
3175 
3176 	if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3177 	    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3178 		btrfs_info(fs_info, "creating free space tree");
3179 		ret = btrfs_create_free_space_tree(fs_info);
3180 		if (ret) {
3181 			btrfs_warn(fs_info,
3182 				"failed to create free space tree: %d", ret);
3183 			close_ctree(fs_info);
3184 			return ret;
3185 		}
3186 	}
3187 
3188 	down_read(&fs_info->cleanup_work_sem);
3189 	if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3190 	    (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3191 		up_read(&fs_info->cleanup_work_sem);
3192 		close_ctree(fs_info);
3193 		return ret;
3194 	}
3195 	up_read(&fs_info->cleanup_work_sem);
3196 
3197 	ret = btrfs_resume_balance_async(fs_info);
3198 	if (ret) {
3199 		btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3200 		close_ctree(fs_info);
3201 		return ret;
3202 	}
3203 
3204 	ret = btrfs_resume_dev_replace_async(fs_info);
3205 	if (ret) {
3206 		btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3207 		close_ctree(fs_info);
3208 		return ret;
3209 	}
3210 
3211 	btrfs_qgroup_rescan_resume(fs_info);
3212 
3213 	if (!fs_info->uuid_root) {
3214 		btrfs_info(fs_info, "creating UUID tree");
3215 		ret = btrfs_create_uuid_tree(fs_info);
3216 		if (ret) {
3217 			btrfs_warn(fs_info,
3218 				"failed to create the UUID tree: %d", ret);
3219 			close_ctree(fs_info);
3220 			return ret;
3221 		}
3222 	} else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3223 		   fs_info->generation !=
3224 				btrfs_super_uuid_tree_generation(disk_super)) {
3225 		btrfs_info(fs_info, "checking UUID tree");
3226 		ret = btrfs_check_uuid_tree(fs_info);
3227 		if (ret) {
3228 			btrfs_warn(fs_info,
3229 				"failed to check the UUID tree: %d", ret);
3230 			close_ctree(fs_info);
3231 			return ret;
3232 		}
3233 	} else {
3234 		set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3235 	}
3236 	set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3237 
3238 	/*
3239 	 * backuproot only affect mount behavior, and if open_ctree succeeded,
3240 	 * no need to keep the flag
3241 	 */
3242 	btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3243 
3244 	return 0;
3245 
3246 fail_qgroup:
3247 	btrfs_free_qgroup_config(fs_info);
3248 fail_trans_kthread:
3249 	kthread_stop(fs_info->transaction_kthread);
3250 	btrfs_cleanup_transaction(fs_info);
3251 	btrfs_free_fs_roots(fs_info);
3252 fail_cleaner:
3253 	kthread_stop(fs_info->cleaner_kthread);
3254 
3255 	/*
3256 	 * make sure we're done with the btree inode before we stop our
3257 	 * kthreads
3258 	 */
3259 	filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3260 
3261 fail_sysfs:
3262 	btrfs_sysfs_remove_mounted(fs_info);
3263 
3264 fail_fsdev_sysfs:
3265 	btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3266 
3267 fail_block_groups:
3268 	btrfs_put_block_group_cache(fs_info);
3269 
3270 fail_tree_roots:
3271 	free_root_pointers(fs_info, 1);
3272 	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3273 
3274 fail_sb_buffer:
3275 	btrfs_stop_all_workers(fs_info);
3276 	btrfs_free_block_groups(fs_info);
3277 fail_alloc:
3278 fail_iput:
3279 	btrfs_mapping_tree_free(&fs_info->mapping_tree);
3280 
3281 	iput(fs_info->btree_inode);
3282 fail_bio_counter:
3283 	percpu_counter_destroy(&fs_info->bio_counter);
3284 fail_delalloc_bytes:
3285 	percpu_counter_destroy(&fs_info->delalloc_bytes);
3286 fail_dirty_metadata_bytes:
3287 	percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3288 fail_bdi:
3289 	bdi_destroy(&fs_info->bdi);
3290 fail_srcu:
3291 	cleanup_srcu_struct(&fs_info->subvol_srcu);
3292 fail:
3293 	btrfs_free_stripe_hash_table(fs_info);
3294 	btrfs_close_devices(fs_info->fs_devices);
3295 	return err;
3296 
3297 recovery_tree_root:
3298 	if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3299 		goto fail_tree_roots;
3300 
3301 	free_root_pointers(fs_info, 0);
3302 
3303 	/* don't use the log in recovery mode, it won't be valid */
3304 	btrfs_set_super_log_root(disk_super, 0);
3305 
3306 	/* we can't trust the free space cache either */
3307 	btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3308 
3309 	ret = next_root_backup(fs_info, fs_info->super_copy,
3310 			       &num_backups_tried, &backup_index);
3311 	if (ret == -1)
3312 		goto fail_block_groups;
3313 	goto retry_root_backup;
3314 }
3315 
3316 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3317 {
3318 	if (uptodate) {
3319 		set_buffer_uptodate(bh);
3320 	} else {
3321 		struct btrfs_device *device = (struct btrfs_device *)
3322 			bh->b_private;
3323 
3324 		btrfs_warn_rl_in_rcu(device->fs_info,
3325 				"lost page write due to IO error on %s",
3326 					  rcu_str_deref(device->name));
3327 		/* note, we don't set_buffer_write_io_error because we have
3328 		 * our own ways of dealing with the IO errors
3329 		 */
3330 		clear_buffer_uptodate(bh);
3331 		btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3332 	}
3333 	unlock_buffer(bh);
3334 	put_bh(bh);
3335 }
3336 
3337 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3338 			struct buffer_head **bh_ret)
3339 {
3340 	struct buffer_head *bh;
3341 	struct btrfs_super_block *super;
3342 	u64 bytenr;
3343 
3344 	bytenr = btrfs_sb_offset(copy_num);
3345 	if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3346 		return -EINVAL;
3347 
3348 	bh = __bread(bdev, bytenr / 4096, BTRFS_SUPER_INFO_SIZE);
3349 	/*
3350 	 * If we fail to read from the underlying devices, as of now
3351 	 * the best option we have is to mark it EIO.
3352 	 */
3353 	if (!bh)
3354 		return -EIO;
3355 
3356 	super = (struct btrfs_super_block *)bh->b_data;
3357 	if (btrfs_super_bytenr(super) != bytenr ||
3358 		    btrfs_super_magic(super) != BTRFS_MAGIC) {
3359 		brelse(bh);
3360 		return -EINVAL;
3361 	}
3362 
3363 	*bh_ret = bh;
3364 	return 0;
3365 }
3366 
3367 
3368 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3369 {
3370 	struct buffer_head *bh;
3371 	struct buffer_head *latest = NULL;
3372 	struct btrfs_super_block *super;
3373 	int i;
3374 	u64 transid = 0;
3375 	int ret = -EINVAL;
3376 
3377 	/* we would like to check all the supers, but that would make
3378 	 * a btrfs mount succeed after a mkfs from a different FS.
3379 	 * So, we need to add a special mount option to scan for
3380 	 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3381 	 */
3382 	for (i = 0; i < 1; i++) {
3383 		ret = btrfs_read_dev_one_super(bdev, i, &bh);
3384 		if (ret)
3385 			continue;
3386 
3387 		super = (struct btrfs_super_block *)bh->b_data;
3388 
3389 		if (!latest || btrfs_super_generation(super) > transid) {
3390 			brelse(latest);
3391 			latest = bh;
3392 			transid = btrfs_super_generation(super);
3393 		} else {
3394 			brelse(bh);
3395 		}
3396 	}
3397 
3398 	if (!latest)
3399 		return ERR_PTR(ret);
3400 
3401 	return latest;
3402 }
3403 
3404 /*
3405  * this should be called twice, once with wait == 0 and
3406  * once with wait == 1.  When wait == 0 is done, all the buffer heads
3407  * we write are pinned.
3408  *
3409  * They are released when wait == 1 is done.
3410  * max_mirrors must be the same for both runs, and it indicates how
3411  * many supers on this one device should be written.
3412  *
3413  * max_mirrors == 0 means to write them all.
3414  */
3415 static int write_dev_supers(struct btrfs_device *device,
3416 			    struct btrfs_super_block *sb,
3417 			    int wait, int max_mirrors)
3418 {
3419 	struct buffer_head *bh;
3420 	int i;
3421 	int ret;
3422 	int errors = 0;
3423 	u32 crc;
3424 	u64 bytenr;
3425 
3426 	if (max_mirrors == 0)
3427 		max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3428 
3429 	for (i = 0; i < max_mirrors; i++) {
3430 		bytenr = btrfs_sb_offset(i);
3431 		if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3432 		    device->commit_total_bytes)
3433 			break;
3434 
3435 		if (wait) {
3436 			bh = __find_get_block(device->bdev, bytenr / 4096,
3437 					      BTRFS_SUPER_INFO_SIZE);
3438 			if (!bh) {
3439 				errors++;
3440 				continue;
3441 			}
3442 			wait_on_buffer(bh);
3443 			if (!buffer_uptodate(bh))
3444 				errors++;
3445 
3446 			/* drop our reference */
3447 			brelse(bh);
3448 
3449 			/* drop the reference from the wait == 0 run */
3450 			brelse(bh);
3451 			continue;
3452 		} else {
3453 			btrfs_set_super_bytenr(sb, bytenr);
3454 
3455 			crc = ~(u32)0;
3456 			crc = btrfs_csum_data((const char *)sb +
3457 					      BTRFS_CSUM_SIZE, crc,
3458 					      BTRFS_SUPER_INFO_SIZE -
3459 					      BTRFS_CSUM_SIZE);
3460 			btrfs_csum_final(crc, sb->csum);
3461 
3462 			/*
3463 			 * one reference for us, and we leave it for the
3464 			 * caller
3465 			 */
3466 			bh = __getblk(device->bdev, bytenr / 4096,
3467 				      BTRFS_SUPER_INFO_SIZE);
3468 			if (!bh) {
3469 				btrfs_err(device->fs_info,
3470 				    "couldn't get super buffer head for bytenr %llu",
3471 				    bytenr);
3472 				errors++;
3473 				continue;
3474 			}
3475 
3476 			memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3477 
3478 			/* one reference for submit_bh */
3479 			get_bh(bh);
3480 
3481 			set_buffer_uptodate(bh);
3482 			lock_buffer(bh);
3483 			bh->b_end_io = btrfs_end_buffer_write_sync;
3484 			bh->b_private = device;
3485 		}
3486 
3487 		/*
3488 		 * we fua the first super.  The others we allow
3489 		 * to go down lazy.
3490 		 */
3491 		if (i == 0)
3492 			ret = btrfsic_submit_bh(REQ_OP_WRITE, REQ_FUA, bh);
3493 		else
3494 			ret = btrfsic_submit_bh(REQ_OP_WRITE, REQ_SYNC, bh);
3495 		if (ret)
3496 			errors++;
3497 	}
3498 	return errors < i ? 0 : -1;
3499 }
3500 
3501 /*
3502  * endio for the write_dev_flush, this will wake anyone waiting
3503  * for the barrier when it is done
3504  */
3505 static void btrfs_end_empty_barrier(struct bio *bio)
3506 {
3507 	if (bio->bi_private)
3508 		complete(bio->bi_private);
3509 	bio_put(bio);
3510 }
3511 
3512 /*
3513  * trigger flushes for one the devices.  If you pass wait == 0, the flushes are
3514  * sent down.  With wait == 1, it waits for the previous flush.
3515  *
3516  * any device where the flush fails with eopnotsupp are flagged as not-barrier
3517  * capable
3518  */
3519 static int write_dev_flush(struct btrfs_device *device, int wait)
3520 {
3521 	struct bio *bio;
3522 	int ret = 0;
3523 
3524 	if (device->nobarriers)
3525 		return 0;
3526 
3527 	if (wait) {
3528 		bio = device->flush_bio;
3529 		if (!bio)
3530 			return 0;
3531 
3532 		wait_for_completion(&device->flush_wait);
3533 
3534 		if (bio->bi_error) {
3535 			ret = bio->bi_error;
3536 			btrfs_dev_stat_inc_and_print(device,
3537 				BTRFS_DEV_STAT_FLUSH_ERRS);
3538 		}
3539 
3540 		/* drop the reference from the wait == 0 run */
3541 		bio_put(bio);
3542 		device->flush_bio = NULL;
3543 
3544 		return ret;
3545 	}
3546 
3547 	/*
3548 	 * one reference for us, and we leave it for the
3549 	 * caller
3550 	 */
3551 	device->flush_bio = NULL;
3552 	bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3553 	if (!bio)
3554 		return -ENOMEM;
3555 
3556 	bio->bi_end_io = btrfs_end_empty_barrier;
3557 	bio->bi_bdev = device->bdev;
3558 	bio->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
3559 	init_completion(&device->flush_wait);
3560 	bio->bi_private = &device->flush_wait;
3561 	device->flush_bio = bio;
3562 
3563 	bio_get(bio);
3564 	btrfsic_submit_bio(bio);
3565 
3566 	return 0;
3567 }
3568 
3569 /*
3570  * send an empty flush down to each device in parallel,
3571  * then wait for them
3572  */
3573 static int barrier_all_devices(struct btrfs_fs_info *info)
3574 {
3575 	struct list_head *head;
3576 	struct btrfs_device *dev;
3577 	int errors_send = 0;
3578 	int errors_wait = 0;
3579 	int ret;
3580 
3581 	/* send down all the barriers */
3582 	head = &info->fs_devices->devices;
3583 	list_for_each_entry_rcu(dev, head, dev_list) {
3584 		if (dev->missing)
3585 			continue;
3586 		if (!dev->bdev) {
3587 			errors_send++;
3588 			continue;
3589 		}
3590 		if (!dev->in_fs_metadata || !dev->writeable)
3591 			continue;
3592 
3593 		ret = write_dev_flush(dev, 0);
3594 		if (ret)
3595 			errors_send++;
3596 	}
3597 
3598 	/* wait for all the barriers */
3599 	list_for_each_entry_rcu(dev, head, dev_list) {
3600 		if (dev->missing)
3601 			continue;
3602 		if (!dev->bdev) {
3603 			errors_wait++;
3604 			continue;
3605 		}
3606 		if (!dev->in_fs_metadata || !dev->writeable)
3607 			continue;
3608 
3609 		ret = write_dev_flush(dev, 1);
3610 		if (ret)
3611 			errors_wait++;
3612 	}
3613 	if (errors_send > info->num_tolerated_disk_barrier_failures ||
3614 	    errors_wait > info->num_tolerated_disk_barrier_failures)
3615 		return -EIO;
3616 	return 0;
3617 }
3618 
3619 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3620 {
3621 	int raid_type;
3622 	int min_tolerated = INT_MAX;
3623 
3624 	if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3625 	    (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3626 		min_tolerated = min(min_tolerated,
3627 				    btrfs_raid_array[BTRFS_RAID_SINGLE].
3628 				    tolerated_failures);
3629 
3630 	for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3631 		if (raid_type == BTRFS_RAID_SINGLE)
3632 			continue;
3633 		if (!(flags & btrfs_raid_group[raid_type]))
3634 			continue;
3635 		min_tolerated = min(min_tolerated,
3636 				    btrfs_raid_array[raid_type].
3637 				    tolerated_failures);
3638 	}
3639 
3640 	if (min_tolerated == INT_MAX) {
3641 		pr_warn("BTRFS: unknown raid flag: %llu", flags);
3642 		min_tolerated = 0;
3643 	}
3644 
3645 	return min_tolerated;
3646 }
3647 
3648 int btrfs_calc_num_tolerated_disk_barrier_failures(
3649 	struct btrfs_fs_info *fs_info)
3650 {
3651 	struct btrfs_ioctl_space_info space;
3652 	struct btrfs_space_info *sinfo;
3653 	u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3654 		       BTRFS_BLOCK_GROUP_SYSTEM,
3655 		       BTRFS_BLOCK_GROUP_METADATA,
3656 		       BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3657 	int i;
3658 	int c;
3659 	int num_tolerated_disk_barrier_failures =
3660 		(int)fs_info->fs_devices->num_devices;
3661 
3662 	for (i = 0; i < ARRAY_SIZE(types); i++) {
3663 		struct btrfs_space_info *tmp;
3664 
3665 		sinfo = NULL;
3666 		rcu_read_lock();
3667 		list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3668 			if (tmp->flags == types[i]) {
3669 				sinfo = tmp;
3670 				break;
3671 			}
3672 		}
3673 		rcu_read_unlock();
3674 
3675 		if (!sinfo)
3676 			continue;
3677 
3678 		down_read(&sinfo->groups_sem);
3679 		for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3680 			u64 flags;
3681 
3682 			if (list_empty(&sinfo->block_groups[c]))
3683 				continue;
3684 
3685 			btrfs_get_block_group_info(&sinfo->block_groups[c],
3686 						   &space);
3687 			if (space.total_bytes == 0 || space.used_bytes == 0)
3688 				continue;
3689 			flags = space.flags;
3690 
3691 			num_tolerated_disk_barrier_failures = min(
3692 				num_tolerated_disk_barrier_failures,
3693 				btrfs_get_num_tolerated_disk_barrier_failures(
3694 					flags));
3695 		}
3696 		up_read(&sinfo->groups_sem);
3697 	}
3698 
3699 	return num_tolerated_disk_barrier_failures;
3700 }
3701 
3702 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3703 {
3704 	struct list_head *head;
3705 	struct btrfs_device *dev;
3706 	struct btrfs_super_block *sb;
3707 	struct btrfs_dev_item *dev_item;
3708 	int ret;
3709 	int do_barriers;
3710 	int max_errors;
3711 	int total_errors = 0;
3712 	u64 flags;
3713 
3714 	do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3715 	backup_super_roots(fs_info);
3716 
3717 	sb = fs_info->super_for_commit;
3718 	dev_item = &sb->dev_item;
3719 
3720 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
3721 	head = &fs_info->fs_devices->devices;
3722 	max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3723 
3724 	if (do_barriers) {
3725 		ret = barrier_all_devices(fs_info);
3726 		if (ret) {
3727 			mutex_unlock(
3728 				&fs_info->fs_devices->device_list_mutex);
3729 			btrfs_handle_fs_error(fs_info, ret,
3730 					      "errors while submitting device barriers.");
3731 			return ret;
3732 		}
3733 	}
3734 
3735 	list_for_each_entry_rcu(dev, head, dev_list) {
3736 		if (!dev->bdev) {
3737 			total_errors++;
3738 			continue;
3739 		}
3740 		if (!dev->in_fs_metadata || !dev->writeable)
3741 			continue;
3742 
3743 		btrfs_set_stack_device_generation(dev_item, 0);
3744 		btrfs_set_stack_device_type(dev_item, dev->type);
3745 		btrfs_set_stack_device_id(dev_item, dev->devid);
3746 		btrfs_set_stack_device_total_bytes(dev_item,
3747 						   dev->commit_total_bytes);
3748 		btrfs_set_stack_device_bytes_used(dev_item,
3749 						  dev->commit_bytes_used);
3750 		btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3751 		btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3752 		btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3753 		memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3754 		memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3755 
3756 		flags = btrfs_super_flags(sb);
3757 		btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3758 
3759 		ret = write_dev_supers(dev, sb, 0, max_mirrors);
3760 		if (ret)
3761 			total_errors++;
3762 	}
3763 	if (total_errors > max_errors) {
3764 		btrfs_err(fs_info, "%d errors while writing supers",
3765 			  total_errors);
3766 		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3767 
3768 		/* FUA is masked off if unsupported and can't be the reason */
3769 		btrfs_handle_fs_error(fs_info, -EIO,
3770 				      "%d errors while writing supers",
3771 				      total_errors);
3772 		return -EIO;
3773 	}
3774 
3775 	total_errors = 0;
3776 	list_for_each_entry_rcu(dev, head, dev_list) {
3777 		if (!dev->bdev)
3778 			continue;
3779 		if (!dev->in_fs_metadata || !dev->writeable)
3780 			continue;
3781 
3782 		ret = write_dev_supers(dev, sb, 1, max_mirrors);
3783 		if (ret)
3784 			total_errors++;
3785 	}
3786 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3787 	if (total_errors > max_errors) {
3788 		btrfs_handle_fs_error(fs_info, -EIO,
3789 				      "%d errors while writing supers",
3790 				      total_errors);
3791 		return -EIO;
3792 	}
3793 	return 0;
3794 }
3795 
3796 /* Drop a fs root from the radix tree and free it. */
3797 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3798 				  struct btrfs_root *root)
3799 {
3800 	spin_lock(&fs_info->fs_roots_radix_lock);
3801 	radix_tree_delete(&fs_info->fs_roots_radix,
3802 			  (unsigned long)root->root_key.objectid);
3803 	spin_unlock(&fs_info->fs_roots_radix_lock);
3804 
3805 	if (btrfs_root_refs(&root->root_item) == 0)
3806 		synchronize_srcu(&fs_info->subvol_srcu);
3807 
3808 	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3809 		btrfs_free_log(NULL, root);
3810 		if (root->reloc_root) {
3811 			free_extent_buffer(root->reloc_root->node);
3812 			free_extent_buffer(root->reloc_root->commit_root);
3813 			btrfs_put_fs_root(root->reloc_root);
3814 			root->reloc_root = NULL;
3815 		}
3816 	}
3817 
3818 	if (root->free_ino_pinned)
3819 		__btrfs_remove_free_space_cache(root->free_ino_pinned);
3820 	if (root->free_ino_ctl)
3821 		__btrfs_remove_free_space_cache(root->free_ino_ctl);
3822 	free_fs_root(root);
3823 }
3824 
3825 static void free_fs_root(struct btrfs_root *root)
3826 {
3827 	iput(root->ino_cache_inode);
3828 	WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3829 	btrfs_free_block_rsv(root->fs_info, root->orphan_block_rsv);
3830 	root->orphan_block_rsv = NULL;
3831 	if (root->anon_dev)
3832 		free_anon_bdev(root->anon_dev);
3833 	if (root->subv_writers)
3834 		btrfs_free_subvolume_writers(root->subv_writers);
3835 	free_extent_buffer(root->node);
3836 	free_extent_buffer(root->commit_root);
3837 	kfree(root->free_ino_ctl);
3838 	kfree(root->free_ino_pinned);
3839 	kfree(root->name);
3840 	btrfs_put_fs_root(root);
3841 }
3842 
3843 void btrfs_free_fs_root(struct btrfs_root *root)
3844 {
3845 	free_fs_root(root);
3846 }
3847 
3848 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3849 {
3850 	u64 root_objectid = 0;
3851 	struct btrfs_root *gang[8];
3852 	int i = 0;
3853 	int err = 0;
3854 	unsigned int ret = 0;
3855 	int index;
3856 
3857 	while (1) {
3858 		index = srcu_read_lock(&fs_info->subvol_srcu);
3859 		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3860 					     (void **)gang, root_objectid,
3861 					     ARRAY_SIZE(gang));
3862 		if (!ret) {
3863 			srcu_read_unlock(&fs_info->subvol_srcu, index);
3864 			break;
3865 		}
3866 		root_objectid = gang[ret - 1]->root_key.objectid + 1;
3867 
3868 		for (i = 0; i < ret; i++) {
3869 			/* Avoid to grab roots in dead_roots */
3870 			if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3871 				gang[i] = NULL;
3872 				continue;
3873 			}
3874 			/* grab all the search result for later use */
3875 			gang[i] = btrfs_grab_fs_root(gang[i]);
3876 		}
3877 		srcu_read_unlock(&fs_info->subvol_srcu, index);
3878 
3879 		for (i = 0; i < ret; i++) {
3880 			if (!gang[i])
3881 				continue;
3882 			root_objectid = gang[i]->root_key.objectid;
3883 			err = btrfs_orphan_cleanup(gang[i]);
3884 			if (err)
3885 				break;
3886 			btrfs_put_fs_root(gang[i]);
3887 		}
3888 		root_objectid++;
3889 	}
3890 
3891 	/* release the uncleaned roots due to error */
3892 	for (; i < ret; i++) {
3893 		if (gang[i])
3894 			btrfs_put_fs_root(gang[i]);
3895 	}
3896 	return err;
3897 }
3898 
3899 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3900 {
3901 	struct btrfs_root *root = fs_info->tree_root;
3902 	struct btrfs_trans_handle *trans;
3903 
3904 	mutex_lock(&fs_info->cleaner_mutex);
3905 	btrfs_run_delayed_iputs(fs_info);
3906 	mutex_unlock(&fs_info->cleaner_mutex);
3907 	wake_up_process(fs_info->cleaner_kthread);
3908 
3909 	/* wait until ongoing cleanup work done */
3910 	down_write(&fs_info->cleanup_work_sem);
3911 	up_write(&fs_info->cleanup_work_sem);
3912 
3913 	trans = btrfs_join_transaction(root);
3914 	if (IS_ERR(trans))
3915 		return PTR_ERR(trans);
3916 	return btrfs_commit_transaction(trans);
3917 }
3918 
3919 void close_ctree(struct btrfs_fs_info *fs_info)
3920 {
3921 	struct btrfs_root *root = fs_info->tree_root;
3922 	int ret;
3923 
3924 	set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
3925 
3926 	/* wait for the qgroup rescan worker to stop */
3927 	btrfs_qgroup_wait_for_completion(fs_info, false);
3928 
3929 	/* wait for the uuid_scan task to finish */
3930 	down(&fs_info->uuid_tree_rescan_sem);
3931 	/* avoid complains from lockdep et al., set sem back to initial state */
3932 	up(&fs_info->uuid_tree_rescan_sem);
3933 
3934 	/* pause restriper - we want to resume on mount */
3935 	btrfs_pause_balance(fs_info);
3936 
3937 	btrfs_dev_replace_suspend_for_unmount(fs_info);
3938 
3939 	btrfs_scrub_cancel(fs_info);
3940 
3941 	/* wait for any defraggers to finish */
3942 	wait_event(fs_info->transaction_wait,
3943 		   (atomic_read(&fs_info->defrag_running) == 0));
3944 
3945 	/* clear out the rbtree of defraggable inodes */
3946 	btrfs_cleanup_defrag_inodes(fs_info);
3947 
3948 	cancel_work_sync(&fs_info->async_reclaim_work);
3949 
3950 	if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3951 		/*
3952 		 * If the cleaner thread is stopped and there are
3953 		 * block groups queued for removal, the deletion will be
3954 		 * skipped when we quit the cleaner thread.
3955 		 */
3956 		btrfs_delete_unused_bgs(fs_info);
3957 
3958 		ret = btrfs_commit_super(fs_info);
3959 		if (ret)
3960 			btrfs_err(fs_info, "commit super ret %d", ret);
3961 	}
3962 
3963 	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3964 		btrfs_error_commit_super(fs_info);
3965 
3966 	kthread_stop(fs_info->transaction_kthread);
3967 	kthread_stop(fs_info->cleaner_kthread);
3968 
3969 	set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
3970 
3971 	btrfs_free_qgroup_config(fs_info);
3972 
3973 	if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3974 		btrfs_info(fs_info, "at unmount delalloc count %lld",
3975 		       percpu_counter_sum(&fs_info->delalloc_bytes));
3976 	}
3977 
3978 	btrfs_sysfs_remove_mounted(fs_info);
3979 	btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3980 
3981 	btrfs_free_fs_roots(fs_info);
3982 
3983 	btrfs_put_block_group_cache(fs_info);
3984 
3985 	/*
3986 	 * we must make sure there is not any read request to
3987 	 * submit after we stopping all workers.
3988 	 */
3989 	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3990 	btrfs_stop_all_workers(fs_info);
3991 
3992 	btrfs_free_block_groups(fs_info);
3993 
3994 	clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
3995 	free_root_pointers(fs_info, 1);
3996 
3997 	iput(fs_info->btree_inode);
3998 
3999 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4000 	if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4001 		btrfsic_unmount(fs_info->fs_devices);
4002 #endif
4003 
4004 	btrfs_close_devices(fs_info->fs_devices);
4005 	btrfs_mapping_tree_free(&fs_info->mapping_tree);
4006 
4007 	percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
4008 	percpu_counter_destroy(&fs_info->delalloc_bytes);
4009 	percpu_counter_destroy(&fs_info->bio_counter);
4010 	bdi_destroy(&fs_info->bdi);
4011 	cleanup_srcu_struct(&fs_info->subvol_srcu);
4012 
4013 	btrfs_free_stripe_hash_table(fs_info);
4014 
4015 	__btrfs_free_block_rsv(root->orphan_block_rsv);
4016 	root->orphan_block_rsv = NULL;
4017 
4018 	mutex_lock(&fs_info->chunk_mutex);
4019 	while (!list_empty(&fs_info->pinned_chunks)) {
4020 		struct extent_map *em;
4021 
4022 		em = list_first_entry(&fs_info->pinned_chunks,
4023 				      struct extent_map, list);
4024 		list_del_init(&em->list);
4025 		free_extent_map(em);
4026 	}
4027 	mutex_unlock(&fs_info->chunk_mutex);
4028 }
4029 
4030 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4031 			  int atomic)
4032 {
4033 	int ret;
4034 	struct inode *btree_inode = buf->pages[0]->mapping->host;
4035 
4036 	ret = extent_buffer_uptodate(buf);
4037 	if (!ret)
4038 		return ret;
4039 
4040 	ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4041 				    parent_transid, atomic);
4042 	if (ret == -EAGAIN)
4043 		return ret;
4044 	return !ret;
4045 }
4046 
4047 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4048 {
4049 	struct btrfs_fs_info *fs_info;
4050 	struct btrfs_root *root;
4051 	u64 transid = btrfs_header_generation(buf);
4052 	int was_dirty;
4053 
4054 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4055 	/*
4056 	 * This is a fast path so only do this check if we have sanity tests
4057 	 * enabled.  Normal people shouldn't be marking dummy buffers as dirty
4058 	 * outside of the sanity tests.
4059 	 */
4060 	if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
4061 		return;
4062 #endif
4063 	root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4064 	fs_info = root->fs_info;
4065 	btrfs_assert_tree_locked(buf);
4066 	if (transid != fs_info->generation)
4067 		WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4068 			buf->start, transid, fs_info->generation);
4069 	was_dirty = set_extent_buffer_dirty(buf);
4070 	if (!was_dirty)
4071 		__percpu_counter_add(&fs_info->dirty_metadata_bytes,
4072 				     buf->len,
4073 				     fs_info->dirty_metadata_batch);
4074 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4075 	if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
4076 		btrfs_print_leaf(fs_info, buf);
4077 		ASSERT(0);
4078 	}
4079 #endif
4080 }
4081 
4082 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4083 					int flush_delayed)
4084 {
4085 	/*
4086 	 * looks as though older kernels can get into trouble with
4087 	 * this code, they end up stuck in balance_dirty_pages forever
4088 	 */
4089 	int ret;
4090 
4091 	if (current->flags & PF_MEMALLOC)
4092 		return;
4093 
4094 	if (flush_delayed)
4095 		btrfs_balance_delayed_items(fs_info);
4096 
4097 	ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4098 				     BTRFS_DIRTY_METADATA_THRESH);
4099 	if (ret > 0) {
4100 		balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4101 	}
4102 }
4103 
4104 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4105 {
4106 	__btrfs_btree_balance_dirty(fs_info, 1);
4107 }
4108 
4109 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4110 {
4111 	__btrfs_btree_balance_dirty(fs_info, 0);
4112 }
4113 
4114 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
4115 {
4116 	struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4117 	struct btrfs_fs_info *fs_info = root->fs_info;
4118 
4119 	return btree_read_extent_buffer_pages(fs_info, buf, parent_transid);
4120 }
4121 
4122 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info)
4123 {
4124 	struct btrfs_super_block *sb = fs_info->super_copy;
4125 	u64 nodesize = btrfs_super_nodesize(sb);
4126 	u64 sectorsize = btrfs_super_sectorsize(sb);
4127 	int ret = 0;
4128 
4129 	if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
4130 		btrfs_err(fs_info, "no valid FS found");
4131 		ret = -EINVAL;
4132 	}
4133 	if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP)
4134 		btrfs_warn(fs_info, "unrecognized super flag: %llu",
4135 				btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
4136 	if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
4137 		btrfs_err(fs_info, "tree_root level too big: %d >= %d",
4138 				btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
4139 		ret = -EINVAL;
4140 	}
4141 	if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
4142 		btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
4143 				btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
4144 		ret = -EINVAL;
4145 	}
4146 	if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
4147 		btrfs_err(fs_info, "log_root level too big: %d >= %d",
4148 				btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
4149 		ret = -EINVAL;
4150 	}
4151 
4152 	/*
4153 	 * Check sectorsize and nodesize first, other check will need it.
4154 	 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
4155 	 */
4156 	if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
4157 	    sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4158 		btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
4159 		ret = -EINVAL;
4160 	}
4161 	/* Only PAGE SIZE is supported yet */
4162 	if (sectorsize != PAGE_SIZE) {
4163 		btrfs_err(fs_info,
4164 			"sectorsize %llu not supported yet, only support %lu",
4165 			sectorsize, PAGE_SIZE);
4166 		ret = -EINVAL;
4167 	}
4168 	if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
4169 	    nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4170 		btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
4171 		ret = -EINVAL;
4172 	}
4173 	if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
4174 		btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
4175 			  le32_to_cpu(sb->__unused_leafsize), nodesize);
4176 		ret = -EINVAL;
4177 	}
4178 
4179 	/* Root alignment check */
4180 	if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
4181 		btrfs_warn(fs_info, "tree_root block unaligned: %llu",
4182 			   btrfs_super_root(sb));
4183 		ret = -EINVAL;
4184 	}
4185 	if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
4186 		btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
4187 			   btrfs_super_chunk_root(sb));
4188 		ret = -EINVAL;
4189 	}
4190 	if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
4191 		btrfs_warn(fs_info, "log_root block unaligned: %llu",
4192 			   btrfs_super_log_root(sb));
4193 		ret = -EINVAL;
4194 	}
4195 
4196 	if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) {
4197 		btrfs_err(fs_info,
4198 			   "dev_item UUID does not match fsid: %pU != %pU",
4199 			   fs_info->fsid, sb->dev_item.fsid);
4200 		ret = -EINVAL;
4201 	}
4202 
4203 	/*
4204 	 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
4205 	 * done later
4206 	 */
4207 	if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
4208 		btrfs_err(fs_info, "bytes_used is too small %llu",
4209 			  btrfs_super_bytes_used(sb));
4210 		ret = -EINVAL;
4211 	}
4212 	if (!is_power_of_2(btrfs_super_stripesize(sb))) {
4213 		btrfs_err(fs_info, "invalid stripesize %u",
4214 			  btrfs_super_stripesize(sb));
4215 		ret = -EINVAL;
4216 	}
4217 	if (btrfs_super_num_devices(sb) > (1UL << 31))
4218 		btrfs_warn(fs_info, "suspicious number of devices: %llu",
4219 			   btrfs_super_num_devices(sb));
4220 	if (btrfs_super_num_devices(sb) == 0) {
4221 		btrfs_err(fs_info, "number of devices is 0");
4222 		ret = -EINVAL;
4223 	}
4224 
4225 	if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
4226 		btrfs_err(fs_info, "super offset mismatch %llu != %u",
4227 			  btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
4228 		ret = -EINVAL;
4229 	}
4230 
4231 	/*
4232 	 * Obvious sys_chunk_array corruptions, it must hold at least one key
4233 	 * and one chunk
4234 	 */
4235 	if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4236 		btrfs_err(fs_info, "system chunk array too big %u > %u",
4237 			  btrfs_super_sys_array_size(sb),
4238 			  BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
4239 		ret = -EINVAL;
4240 	}
4241 	if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
4242 			+ sizeof(struct btrfs_chunk)) {
4243 		btrfs_err(fs_info, "system chunk array too small %u < %zu",
4244 			  btrfs_super_sys_array_size(sb),
4245 			  sizeof(struct btrfs_disk_key)
4246 			  + sizeof(struct btrfs_chunk));
4247 		ret = -EINVAL;
4248 	}
4249 
4250 	/*
4251 	 * The generation is a global counter, we'll trust it more than the others
4252 	 * but it's still possible that it's the one that's wrong.
4253 	 */
4254 	if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
4255 		btrfs_warn(fs_info,
4256 			"suspicious: generation < chunk_root_generation: %llu < %llu",
4257 			btrfs_super_generation(sb),
4258 			btrfs_super_chunk_root_generation(sb));
4259 	if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
4260 	    && btrfs_super_cache_generation(sb) != (u64)-1)
4261 		btrfs_warn(fs_info,
4262 			"suspicious: generation < cache_generation: %llu < %llu",
4263 			btrfs_super_generation(sb),
4264 			btrfs_super_cache_generation(sb));
4265 
4266 	return ret;
4267 }
4268 
4269 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4270 {
4271 	mutex_lock(&fs_info->cleaner_mutex);
4272 	btrfs_run_delayed_iputs(fs_info);
4273 	mutex_unlock(&fs_info->cleaner_mutex);
4274 
4275 	down_write(&fs_info->cleanup_work_sem);
4276 	up_write(&fs_info->cleanup_work_sem);
4277 
4278 	/* cleanup FS via transaction */
4279 	btrfs_cleanup_transaction(fs_info);
4280 }
4281 
4282 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4283 {
4284 	struct btrfs_ordered_extent *ordered;
4285 
4286 	spin_lock(&root->ordered_extent_lock);
4287 	/*
4288 	 * This will just short circuit the ordered completion stuff which will
4289 	 * make sure the ordered extent gets properly cleaned up.
4290 	 */
4291 	list_for_each_entry(ordered, &root->ordered_extents,
4292 			    root_extent_list)
4293 		set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4294 	spin_unlock(&root->ordered_extent_lock);
4295 }
4296 
4297 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4298 {
4299 	struct btrfs_root *root;
4300 	struct list_head splice;
4301 
4302 	INIT_LIST_HEAD(&splice);
4303 
4304 	spin_lock(&fs_info->ordered_root_lock);
4305 	list_splice_init(&fs_info->ordered_roots, &splice);
4306 	while (!list_empty(&splice)) {
4307 		root = list_first_entry(&splice, struct btrfs_root,
4308 					ordered_root);
4309 		list_move_tail(&root->ordered_root,
4310 			       &fs_info->ordered_roots);
4311 
4312 		spin_unlock(&fs_info->ordered_root_lock);
4313 		btrfs_destroy_ordered_extents(root);
4314 
4315 		cond_resched();
4316 		spin_lock(&fs_info->ordered_root_lock);
4317 	}
4318 	spin_unlock(&fs_info->ordered_root_lock);
4319 }
4320 
4321 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4322 				      struct btrfs_fs_info *fs_info)
4323 {
4324 	struct rb_node *node;
4325 	struct btrfs_delayed_ref_root *delayed_refs;
4326 	struct btrfs_delayed_ref_node *ref;
4327 	int ret = 0;
4328 
4329 	delayed_refs = &trans->delayed_refs;
4330 
4331 	spin_lock(&delayed_refs->lock);
4332 	if (atomic_read(&delayed_refs->num_entries) == 0) {
4333 		spin_unlock(&delayed_refs->lock);
4334 		btrfs_info(fs_info, "delayed_refs has NO entry");
4335 		return ret;
4336 	}
4337 
4338 	while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4339 		struct btrfs_delayed_ref_head *head;
4340 		struct btrfs_delayed_ref_node *tmp;
4341 		bool pin_bytes = false;
4342 
4343 		head = rb_entry(node, struct btrfs_delayed_ref_head,
4344 				href_node);
4345 		if (!mutex_trylock(&head->mutex)) {
4346 			atomic_inc(&head->node.refs);
4347 			spin_unlock(&delayed_refs->lock);
4348 
4349 			mutex_lock(&head->mutex);
4350 			mutex_unlock(&head->mutex);
4351 			btrfs_put_delayed_ref(&head->node);
4352 			spin_lock(&delayed_refs->lock);
4353 			continue;
4354 		}
4355 		spin_lock(&head->lock);
4356 		list_for_each_entry_safe_reverse(ref, tmp, &head->ref_list,
4357 						 list) {
4358 			ref->in_tree = 0;
4359 			list_del(&ref->list);
4360 			if (!list_empty(&ref->add_list))
4361 				list_del(&ref->add_list);
4362 			atomic_dec(&delayed_refs->num_entries);
4363 			btrfs_put_delayed_ref(ref);
4364 		}
4365 		if (head->must_insert_reserved)
4366 			pin_bytes = true;
4367 		btrfs_free_delayed_extent_op(head->extent_op);
4368 		delayed_refs->num_heads--;
4369 		if (head->processing == 0)
4370 			delayed_refs->num_heads_ready--;
4371 		atomic_dec(&delayed_refs->num_entries);
4372 		head->node.in_tree = 0;
4373 		rb_erase(&head->href_node, &delayed_refs->href_root);
4374 		spin_unlock(&head->lock);
4375 		spin_unlock(&delayed_refs->lock);
4376 		mutex_unlock(&head->mutex);
4377 
4378 		if (pin_bytes)
4379 			btrfs_pin_extent(fs_info, head->node.bytenr,
4380 					 head->node.num_bytes, 1);
4381 		btrfs_put_delayed_ref(&head->node);
4382 		cond_resched();
4383 		spin_lock(&delayed_refs->lock);
4384 	}
4385 
4386 	spin_unlock(&delayed_refs->lock);
4387 
4388 	return ret;
4389 }
4390 
4391 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4392 {
4393 	struct btrfs_inode *btrfs_inode;
4394 	struct list_head splice;
4395 
4396 	INIT_LIST_HEAD(&splice);
4397 
4398 	spin_lock(&root->delalloc_lock);
4399 	list_splice_init(&root->delalloc_inodes, &splice);
4400 
4401 	while (!list_empty(&splice)) {
4402 		btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4403 					       delalloc_inodes);
4404 
4405 		list_del_init(&btrfs_inode->delalloc_inodes);
4406 		clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4407 			  &btrfs_inode->runtime_flags);
4408 		spin_unlock(&root->delalloc_lock);
4409 
4410 		btrfs_invalidate_inodes(btrfs_inode->root);
4411 
4412 		spin_lock(&root->delalloc_lock);
4413 	}
4414 
4415 	spin_unlock(&root->delalloc_lock);
4416 }
4417 
4418 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4419 {
4420 	struct btrfs_root *root;
4421 	struct list_head splice;
4422 
4423 	INIT_LIST_HEAD(&splice);
4424 
4425 	spin_lock(&fs_info->delalloc_root_lock);
4426 	list_splice_init(&fs_info->delalloc_roots, &splice);
4427 	while (!list_empty(&splice)) {
4428 		root = list_first_entry(&splice, struct btrfs_root,
4429 					 delalloc_root);
4430 		list_del_init(&root->delalloc_root);
4431 		root = btrfs_grab_fs_root(root);
4432 		BUG_ON(!root);
4433 		spin_unlock(&fs_info->delalloc_root_lock);
4434 
4435 		btrfs_destroy_delalloc_inodes(root);
4436 		btrfs_put_fs_root(root);
4437 
4438 		spin_lock(&fs_info->delalloc_root_lock);
4439 	}
4440 	spin_unlock(&fs_info->delalloc_root_lock);
4441 }
4442 
4443 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4444 					struct extent_io_tree *dirty_pages,
4445 					int mark)
4446 {
4447 	int ret;
4448 	struct extent_buffer *eb;
4449 	u64 start = 0;
4450 	u64 end;
4451 
4452 	while (1) {
4453 		ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4454 					    mark, NULL);
4455 		if (ret)
4456 			break;
4457 
4458 		clear_extent_bits(dirty_pages, start, end, mark);
4459 		while (start <= end) {
4460 			eb = find_extent_buffer(fs_info, start);
4461 			start += fs_info->nodesize;
4462 			if (!eb)
4463 				continue;
4464 			wait_on_extent_buffer_writeback(eb);
4465 
4466 			if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4467 					       &eb->bflags))
4468 				clear_extent_buffer_dirty(eb);
4469 			free_extent_buffer_stale(eb);
4470 		}
4471 	}
4472 
4473 	return ret;
4474 }
4475 
4476 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4477 				       struct extent_io_tree *pinned_extents)
4478 {
4479 	struct extent_io_tree *unpin;
4480 	u64 start;
4481 	u64 end;
4482 	int ret;
4483 	bool loop = true;
4484 
4485 	unpin = pinned_extents;
4486 again:
4487 	while (1) {
4488 		ret = find_first_extent_bit(unpin, 0, &start, &end,
4489 					    EXTENT_DIRTY, NULL);
4490 		if (ret)
4491 			break;
4492 
4493 		clear_extent_dirty(unpin, start, end);
4494 		btrfs_error_unpin_extent_range(fs_info, start, end);
4495 		cond_resched();
4496 	}
4497 
4498 	if (loop) {
4499 		if (unpin == &fs_info->freed_extents[0])
4500 			unpin = &fs_info->freed_extents[1];
4501 		else
4502 			unpin = &fs_info->freed_extents[0];
4503 		loop = false;
4504 		goto again;
4505 	}
4506 
4507 	return 0;
4508 }
4509 
4510 static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache)
4511 {
4512 	struct inode *inode;
4513 
4514 	inode = cache->io_ctl.inode;
4515 	if (inode) {
4516 		invalidate_inode_pages2(inode->i_mapping);
4517 		BTRFS_I(inode)->generation = 0;
4518 		cache->io_ctl.inode = NULL;
4519 		iput(inode);
4520 	}
4521 	btrfs_put_block_group(cache);
4522 }
4523 
4524 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4525 			     struct btrfs_fs_info *fs_info)
4526 {
4527 	struct btrfs_block_group_cache *cache;
4528 
4529 	spin_lock(&cur_trans->dirty_bgs_lock);
4530 	while (!list_empty(&cur_trans->dirty_bgs)) {
4531 		cache = list_first_entry(&cur_trans->dirty_bgs,
4532 					 struct btrfs_block_group_cache,
4533 					 dirty_list);
4534 		if (!cache) {
4535 			btrfs_err(fs_info, "orphan block group dirty_bgs list");
4536 			spin_unlock(&cur_trans->dirty_bgs_lock);
4537 			return;
4538 		}
4539 
4540 		if (!list_empty(&cache->io_list)) {
4541 			spin_unlock(&cur_trans->dirty_bgs_lock);
4542 			list_del_init(&cache->io_list);
4543 			btrfs_cleanup_bg_io(cache);
4544 			spin_lock(&cur_trans->dirty_bgs_lock);
4545 		}
4546 
4547 		list_del_init(&cache->dirty_list);
4548 		spin_lock(&cache->lock);
4549 		cache->disk_cache_state = BTRFS_DC_ERROR;
4550 		spin_unlock(&cache->lock);
4551 
4552 		spin_unlock(&cur_trans->dirty_bgs_lock);
4553 		btrfs_put_block_group(cache);
4554 		spin_lock(&cur_trans->dirty_bgs_lock);
4555 	}
4556 	spin_unlock(&cur_trans->dirty_bgs_lock);
4557 
4558 	while (!list_empty(&cur_trans->io_bgs)) {
4559 		cache = list_first_entry(&cur_trans->io_bgs,
4560 					 struct btrfs_block_group_cache,
4561 					 io_list);
4562 		if (!cache) {
4563 			btrfs_err(fs_info, "orphan block group on io_bgs list");
4564 			return;
4565 		}
4566 
4567 		list_del_init(&cache->io_list);
4568 		spin_lock(&cache->lock);
4569 		cache->disk_cache_state = BTRFS_DC_ERROR;
4570 		spin_unlock(&cache->lock);
4571 		btrfs_cleanup_bg_io(cache);
4572 	}
4573 }
4574 
4575 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4576 				   struct btrfs_fs_info *fs_info)
4577 {
4578 	btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4579 	ASSERT(list_empty(&cur_trans->dirty_bgs));
4580 	ASSERT(list_empty(&cur_trans->io_bgs));
4581 
4582 	btrfs_destroy_delayed_refs(cur_trans, fs_info);
4583 
4584 	cur_trans->state = TRANS_STATE_COMMIT_START;
4585 	wake_up(&fs_info->transaction_blocked_wait);
4586 
4587 	cur_trans->state = TRANS_STATE_UNBLOCKED;
4588 	wake_up(&fs_info->transaction_wait);
4589 
4590 	btrfs_destroy_delayed_inodes(fs_info);
4591 	btrfs_assert_delayed_root_empty(fs_info);
4592 
4593 	btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4594 				     EXTENT_DIRTY);
4595 	btrfs_destroy_pinned_extent(fs_info,
4596 				    fs_info->pinned_extents);
4597 
4598 	cur_trans->state =TRANS_STATE_COMPLETED;
4599 	wake_up(&cur_trans->commit_wait);
4600 
4601 	/*
4602 	memset(cur_trans, 0, sizeof(*cur_trans));
4603 	kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4604 	*/
4605 }
4606 
4607 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4608 {
4609 	struct btrfs_transaction *t;
4610 
4611 	mutex_lock(&fs_info->transaction_kthread_mutex);
4612 
4613 	spin_lock(&fs_info->trans_lock);
4614 	while (!list_empty(&fs_info->trans_list)) {
4615 		t = list_first_entry(&fs_info->trans_list,
4616 				     struct btrfs_transaction, list);
4617 		if (t->state >= TRANS_STATE_COMMIT_START) {
4618 			atomic_inc(&t->use_count);
4619 			spin_unlock(&fs_info->trans_lock);
4620 			btrfs_wait_for_commit(fs_info, t->transid);
4621 			btrfs_put_transaction(t);
4622 			spin_lock(&fs_info->trans_lock);
4623 			continue;
4624 		}
4625 		if (t == fs_info->running_transaction) {
4626 			t->state = TRANS_STATE_COMMIT_DOING;
4627 			spin_unlock(&fs_info->trans_lock);
4628 			/*
4629 			 * We wait for 0 num_writers since we don't hold a trans
4630 			 * handle open currently for this transaction.
4631 			 */
4632 			wait_event(t->writer_wait,
4633 				   atomic_read(&t->num_writers) == 0);
4634 		} else {
4635 			spin_unlock(&fs_info->trans_lock);
4636 		}
4637 		btrfs_cleanup_one_transaction(t, fs_info);
4638 
4639 		spin_lock(&fs_info->trans_lock);
4640 		if (t == fs_info->running_transaction)
4641 			fs_info->running_transaction = NULL;
4642 		list_del_init(&t->list);
4643 		spin_unlock(&fs_info->trans_lock);
4644 
4645 		btrfs_put_transaction(t);
4646 		trace_btrfs_transaction_commit(fs_info->tree_root);
4647 		spin_lock(&fs_info->trans_lock);
4648 	}
4649 	spin_unlock(&fs_info->trans_lock);
4650 	btrfs_destroy_all_ordered_extents(fs_info);
4651 	btrfs_destroy_delayed_inodes(fs_info);
4652 	btrfs_assert_delayed_root_empty(fs_info);
4653 	btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
4654 	btrfs_destroy_all_delalloc_inodes(fs_info);
4655 	mutex_unlock(&fs_info->transaction_kthread_mutex);
4656 
4657 	return 0;
4658 }
4659 
4660 static const struct extent_io_ops btree_extent_io_ops = {
4661 	/* mandatory callbacks */
4662 	.submit_bio_hook = btree_submit_bio_hook,
4663 	.readpage_end_io_hook = btree_readpage_end_io_hook,
4664 	/* note we're sharing with inode.c for the merge bio hook */
4665 	.merge_bio_hook = btrfs_merge_bio_hook,
4666 	.readpage_io_failed_hook = btree_io_failed_hook,
4667 
4668 	/* optional callbacks */
4669 };
4670