xref: /openbmc/linux/fs/btrfs/disk-io.c (revision 161c3549b45aeef05451b6822d8aaaf39c7bedce)
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/freezer.h>
29 #include <linux/slab.h>
30 #include <linux/migrate.h>
31 #include <linux/ratelimit.h>
32 #include <linux/uuid.h>
33 #include <linux/semaphore.h>
34 #include <asm/unaligned.h>
35 #include "ctree.h"
36 #include "disk-io.h"
37 #include "hash.h"
38 #include "transaction.h"
39 #include "btrfs_inode.h"
40 #include "volumes.h"
41 #include "print-tree.h"
42 #include "locking.h"
43 #include "tree-log.h"
44 #include "free-space-cache.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 
53 #ifdef CONFIG_X86
54 #include <asm/cpufeature.h>
55 #endif
56 
57 static const struct extent_io_ops btree_extent_io_ops;
58 static void end_workqueue_fn(struct btrfs_work *work);
59 static void free_fs_root(struct btrfs_root *root);
60 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
61 				    int read_only);
62 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
63 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
64 				      struct btrfs_root *root);
65 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
66 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
67 					struct extent_io_tree *dirty_pages,
68 					int mark);
69 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
70 				       struct extent_io_tree *pinned_extents);
71 static int btrfs_cleanup_transaction(struct btrfs_root *root);
72 static void btrfs_error_commit_super(struct btrfs_root *root);
73 
74 /*
75  * btrfs_end_io_wq structs are used to do processing in task context when an IO
76  * is complete.  This is used during reads to verify checksums, and it is used
77  * by writes to insert metadata for new file extents after IO is complete.
78  */
79 struct btrfs_end_io_wq {
80 	struct bio *bio;
81 	bio_end_io_t *end_io;
82 	void *private;
83 	struct btrfs_fs_info *info;
84 	int error;
85 	enum btrfs_wq_endio_type metadata;
86 	struct list_head list;
87 	struct btrfs_work work;
88 };
89 
90 static struct kmem_cache *btrfs_end_io_wq_cache;
91 
92 int __init btrfs_end_io_wq_init(void)
93 {
94 	btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
95 					sizeof(struct btrfs_end_io_wq),
96 					0,
97 					SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
98 					NULL);
99 	if (!btrfs_end_io_wq_cache)
100 		return -ENOMEM;
101 	return 0;
102 }
103 
104 void btrfs_end_io_wq_exit(void)
105 {
106 	if (btrfs_end_io_wq_cache)
107 		kmem_cache_destroy(btrfs_end_io_wq_cache);
108 }
109 
110 /*
111  * async submit bios are used to offload expensive checksumming
112  * onto the worker threads.  They checksum file and metadata bios
113  * just before they are sent down the IO stack.
114  */
115 struct async_submit_bio {
116 	struct inode *inode;
117 	struct bio *bio;
118 	struct list_head list;
119 	extent_submit_bio_hook_t *submit_bio_start;
120 	extent_submit_bio_hook_t *submit_bio_done;
121 	int rw;
122 	int mirror_num;
123 	unsigned long bio_flags;
124 	/*
125 	 * bio_offset is optional, can be used if the pages in the bio
126 	 * can't tell us where in the file the bio should go
127 	 */
128 	u64 bio_offset;
129 	struct btrfs_work work;
130 	int error;
131 };
132 
133 /*
134  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
135  * eb, the lockdep key is determined by the btrfs_root it belongs to and
136  * the level the eb occupies in the tree.
137  *
138  * Different roots are used for different purposes and may nest inside each
139  * other and they require separate keysets.  As lockdep keys should be
140  * static, assign keysets according to the purpose of the root as indicated
141  * by btrfs_root->objectid.  This ensures that all special purpose roots
142  * have separate keysets.
143  *
144  * Lock-nesting across peer nodes is always done with the immediate parent
145  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
146  * subclass to avoid triggering lockdep warning in such cases.
147  *
148  * The key is set by the readpage_end_io_hook after the buffer has passed
149  * csum validation but before the pages are unlocked.  It is also set by
150  * btrfs_init_new_buffer on freshly allocated blocks.
151  *
152  * We also add a check to make sure the highest level of the tree is the
153  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
154  * needs update as well.
155  */
156 #ifdef CONFIG_DEBUG_LOCK_ALLOC
157 # if BTRFS_MAX_LEVEL != 8
158 #  error
159 # endif
160 
161 static struct btrfs_lockdep_keyset {
162 	u64			id;		/* root objectid */
163 	const char		*name_stem;	/* lock name stem */
164 	char			names[BTRFS_MAX_LEVEL + 1][20];
165 	struct lock_class_key	keys[BTRFS_MAX_LEVEL + 1];
166 } btrfs_lockdep_keysets[] = {
167 	{ .id = BTRFS_ROOT_TREE_OBJECTID,	.name_stem = "root"	},
168 	{ .id = BTRFS_EXTENT_TREE_OBJECTID,	.name_stem = "extent"	},
169 	{ .id = BTRFS_CHUNK_TREE_OBJECTID,	.name_stem = "chunk"	},
170 	{ .id = BTRFS_DEV_TREE_OBJECTID,	.name_stem = "dev"	},
171 	{ .id = BTRFS_FS_TREE_OBJECTID,		.name_stem = "fs"	},
172 	{ .id = BTRFS_CSUM_TREE_OBJECTID,	.name_stem = "csum"	},
173 	{ .id = BTRFS_QUOTA_TREE_OBJECTID,	.name_stem = "quota"	},
174 	{ .id = BTRFS_TREE_LOG_OBJECTID,	.name_stem = "log"	},
175 	{ .id = BTRFS_TREE_RELOC_OBJECTID,	.name_stem = "treloc"	},
176 	{ .id = BTRFS_DATA_RELOC_TREE_OBJECTID,	.name_stem = "dreloc"	},
177 	{ .id = BTRFS_UUID_TREE_OBJECTID,	.name_stem = "uuid"	},
178 	{ .id = 0,				.name_stem = "tree"	},
179 };
180 
181 void __init btrfs_init_lockdep(void)
182 {
183 	int i, j;
184 
185 	/* initialize lockdep class names */
186 	for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
187 		struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
188 
189 		for (j = 0; j < ARRAY_SIZE(ks->names); j++)
190 			snprintf(ks->names[j], sizeof(ks->names[j]),
191 				 "btrfs-%s-%02d", ks->name_stem, j);
192 	}
193 }
194 
195 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
196 				    int level)
197 {
198 	struct btrfs_lockdep_keyset *ks;
199 
200 	BUG_ON(level >= ARRAY_SIZE(ks->keys));
201 
202 	/* find the matching keyset, id 0 is the default entry */
203 	for (ks = btrfs_lockdep_keysets; ks->id; ks++)
204 		if (ks->id == objectid)
205 			break;
206 
207 	lockdep_set_class_and_name(&eb->lock,
208 				   &ks->keys[level], ks->names[level]);
209 }
210 
211 #endif
212 
213 /*
214  * extents on the btree inode are pretty simple, there's one extent
215  * that covers the entire device
216  */
217 static struct extent_map *btree_get_extent(struct inode *inode,
218 		struct page *page, size_t pg_offset, u64 start, u64 len,
219 		int create)
220 {
221 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
222 	struct extent_map *em;
223 	int ret;
224 
225 	read_lock(&em_tree->lock);
226 	em = lookup_extent_mapping(em_tree, start, len);
227 	if (em) {
228 		em->bdev =
229 			BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
230 		read_unlock(&em_tree->lock);
231 		goto out;
232 	}
233 	read_unlock(&em_tree->lock);
234 
235 	em = alloc_extent_map();
236 	if (!em) {
237 		em = ERR_PTR(-ENOMEM);
238 		goto out;
239 	}
240 	em->start = 0;
241 	em->len = (u64)-1;
242 	em->block_len = (u64)-1;
243 	em->block_start = 0;
244 	em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
245 
246 	write_lock(&em_tree->lock);
247 	ret = add_extent_mapping(em_tree, em, 0);
248 	if (ret == -EEXIST) {
249 		free_extent_map(em);
250 		em = lookup_extent_mapping(em_tree, start, len);
251 		if (!em)
252 			em = ERR_PTR(-EIO);
253 	} else if (ret) {
254 		free_extent_map(em);
255 		em = ERR_PTR(ret);
256 	}
257 	write_unlock(&em_tree->lock);
258 
259 out:
260 	return em;
261 }
262 
263 u32 btrfs_csum_data(char *data, u32 seed, size_t len)
264 {
265 	return btrfs_crc32c(seed, data, len);
266 }
267 
268 void btrfs_csum_final(u32 crc, char *result)
269 {
270 	put_unaligned_le32(~crc, result);
271 }
272 
273 /*
274  * compute the csum for a btree block, and either verify it or write it
275  * into the csum field of the block.
276  */
277 static int csum_tree_block(struct btrfs_fs_info *fs_info,
278 			   struct extent_buffer *buf,
279 			   int verify)
280 {
281 	u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
282 	char *result = NULL;
283 	unsigned long len;
284 	unsigned long cur_len;
285 	unsigned long offset = BTRFS_CSUM_SIZE;
286 	char *kaddr;
287 	unsigned long map_start;
288 	unsigned long map_len;
289 	int err;
290 	u32 crc = ~(u32)0;
291 	unsigned long inline_result;
292 
293 	len = buf->len - offset;
294 	while (len > 0) {
295 		err = map_private_extent_buffer(buf, offset, 32,
296 					&kaddr, &map_start, &map_len);
297 		if (err)
298 			return 1;
299 		cur_len = min(len, map_len - (offset - map_start));
300 		crc = btrfs_csum_data(kaddr + offset - map_start,
301 				      crc, cur_len);
302 		len -= cur_len;
303 		offset += cur_len;
304 	}
305 	if (csum_size > sizeof(inline_result)) {
306 		result = kzalloc(csum_size, GFP_NOFS);
307 		if (!result)
308 			return 1;
309 	} else {
310 		result = (char *)&inline_result;
311 	}
312 
313 	btrfs_csum_final(crc, result);
314 
315 	if (verify) {
316 		if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
317 			u32 val;
318 			u32 found = 0;
319 			memcpy(&found, result, csum_size);
320 
321 			read_extent_buffer(buf, &val, 0, csum_size);
322 			btrfs_warn_rl(fs_info,
323 				"%s checksum verify failed on %llu wanted %X found %X "
324 				"level %d",
325 				fs_info->sb->s_id, buf->start,
326 				val, found, btrfs_header_level(buf));
327 			if (result != (char *)&inline_result)
328 				kfree(result);
329 			return 1;
330 		}
331 	} else {
332 		write_extent_buffer(buf, result, 0, csum_size);
333 	}
334 	if (result != (char *)&inline_result)
335 		kfree(result);
336 	return 0;
337 }
338 
339 /*
340  * we can't consider a given block up to date unless the transid of the
341  * block matches the transid in the parent node's pointer.  This is how we
342  * detect blocks that either didn't get written at all or got written
343  * in the wrong place.
344  */
345 static int verify_parent_transid(struct extent_io_tree *io_tree,
346 				 struct extent_buffer *eb, u64 parent_transid,
347 				 int atomic)
348 {
349 	struct extent_state *cached_state = NULL;
350 	int ret;
351 	bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
352 
353 	if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
354 		return 0;
355 
356 	if (atomic)
357 		return -EAGAIN;
358 
359 	if (need_lock) {
360 		btrfs_tree_read_lock(eb);
361 		btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
362 	}
363 
364 	lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
365 			 0, &cached_state);
366 	if (extent_buffer_uptodate(eb) &&
367 	    btrfs_header_generation(eb) == parent_transid) {
368 		ret = 0;
369 		goto out;
370 	}
371 	btrfs_err_rl(eb->fs_info,
372 		"parent transid verify failed on %llu wanted %llu found %llu",
373 			eb->start,
374 			parent_transid, btrfs_header_generation(eb));
375 	ret = 1;
376 
377 	/*
378 	 * Things reading via commit roots that don't have normal protection,
379 	 * like send, can have a really old block in cache that may point at a
380 	 * block that has been free'd and re-allocated.  So don't clear uptodate
381 	 * if we find an eb that is under IO (dirty/writeback) because we could
382 	 * end up reading in the stale data and then writing it back out and
383 	 * making everybody very sad.
384 	 */
385 	if (!extent_buffer_under_io(eb))
386 		clear_extent_buffer_uptodate(eb);
387 out:
388 	unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
389 			     &cached_state, GFP_NOFS);
390 	if (need_lock)
391 		btrfs_tree_read_unlock_blocking(eb);
392 	return ret;
393 }
394 
395 /*
396  * Return 0 if the superblock checksum type matches the checksum value of that
397  * algorithm. Pass the raw disk superblock data.
398  */
399 static int btrfs_check_super_csum(char *raw_disk_sb)
400 {
401 	struct btrfs_super_block *disk_sb =
402 		(struct btrfs_super_block *)raw_disk_sb;
403 	u16 csum_type = btrfs_super_csum_type(disk_sb);
404 	int ret = 0;
405 
406 	if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
407 		u32 crc = ~(u32)0;
408 		const int csum_size = sizeof(crc);
409 		char result[csum_size];
410 
411 		/*
412 		 * The super_block structure does not span the whole
413 		 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
414 		 * is filled with zeros and is included in the checkum.
415 		 */
416 		crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
417 				crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
418 		btrfs_csum_final(crc, result);
419 
420 		if (memcmp(raw_disk_sb, result, csum_size))
421 			ret = 1;
422 	}
423 
424 	if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
425 		printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n",
426 				csum_type);
427 		ret = 1;
428 	}
429 
430 	return ret;
431 }
432 
433 /*
434  * helper to read a given tree block, doing retries as required when
435  * the checksums don't match and we have alternate mirrors to try.
436  */
437 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
438 					  struct extent_buffer *eb,
439 					  u64 start, u64 parent_transid)
440 {
441 	struct extent_io_tree *io_tree;
442 	int failed = 0;
443 	int ret;
444 	int num_copies = 0;
445 	int mirror_num = 0;
446 	int failed_mirror = 0;
447 
448 	clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
449 	io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
450 	while (1) {
451 		ret = read_extent_buffer_pages(io_tree, eb, start,
452 					       WAIT_COMPLETE,
453 					       btree_get_extent, mirror_num);
454 		if (!ret) {
455 			if (!verify_parent_transid(io_tree, eb,
456 						   parent_transid, 0))
457 				break;
458 			else
459 				ret = -EIO;
460 		}
461 
462 		/*
463 		 * This buffer's crc is fine, but its contents are corrupted, so
464 		 * there is no reason to read the other copies, they won't be
465 		 * any less wrong.
466 		 */
467 		if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
468 			break;
469 
470 		num_copies = btrfs_num_copies(root->fs_info,
471 					      eb->start, eb->len);
472 		if (num_copies == 1)
473 			break;
474 
475 		if (!failed_mirror) {
476 			failed = 1;
477 			failed_mirror = eb->read_mirror;
478 		}
479 
480 		mirror_num++;
481 		if (mirror_num == failed_mirror)
482 			mirror_num++;
483 
484 		if (mirror_num > num_copies)
485 			break;
486 	}
487 
488 	if (failed && !ret && failed_mirror)
489 		repair_eb_io_failure(root, eb, failed_mirror);
490 
491 	return ret;
492 }
493 
494 /*
495  * checksum a dirty tree block before IO.  This has extra checks to make sure
496  * we only fill in the checksum field in the first page of a multi-page block
497  */
498 
499 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
500 {
501 	u64 start = page_offset(page);
502 	u64 found_start;
503 	struct extent_buffer *eb;
504 
505 	eb = (struct extent_buffer *)page->private;
506 	if (page != eb->pages[0])
507 		return 0;
508 	found_start = btrfs_header_bytenr(eb);
509 	if (WARN_ON(found_start != start || !PageUptodate(page)))
510 		return 0;
511 	csum_tree_block(fs_info, eb, 0);
512 	return 0;
513 }
514 
515 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
516 				 struct extent_buffer *eb)
517 {
518 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
519 	u8 fsid[BTRFS_UUID_SIZE];
520 	int ret = 1;
521 
522 	read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
523 	while (fs_devices) {
524 		if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
525 			ret = 0;
526 			break;
527 		}
528 		fs_devices = fs_devices->seed;
529 	}
530 	return ret;
531 }
532 
533 #define CORRUPT(reason, eb, root, slot)				\
534 	btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu,"	\
535 		   "root=%llu, slot=%d", reason,			\
536 	       btrfs_header_bytenr(eb),	root->objectid, slot)
537 
538 static noinline int check_leaf(struct btrfs_root *root,
539 			       struct extent_buffer *leaf)
540 {
541 	struct btrfs_key key;
542 	struct btrfs_key leaf_key;
543 	u32 nritems = btrfs_header_nritems(leaf);
544 	int slot;
545 
546 	if (nritems == 0)
547 		return 0;
548 
549 	/* Check the 0 item */
550 	if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
551 	    BTRFS_LEAF_DATA_SIZE(root)) {
552 		CORRUPT("invalid item offset size pair", leaf, root, 0);
553 		return -EIO;
554 	}
555 
556 	/*
557 	 * Check to make sure each items keys are in the correct order and their
558 	 * offsets make sense.  We only have to loop through nritems-1 because
559 	 * we check the current slot against the next slot, which verifies the
560 	 * next slot's offset+size makes sense and that the current's slot
561 	 * offset is correct.
562 	 */
563 	for (slot = 0; slot < nritems - 1; slot++) {
564 		btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
565 		btrfs_item_key_to_cpu(leaf, &key, slot + 1);
566 
567 		/* Make sure the keys are in the right order */
568 		if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
569 			CORRUPT("bad key order", leaf, root, slot);
570 			return -EIO;
571 		}
572 
573 		/*
574 		 * Make sure the offset and ends are right, remember that the
575 		 * item data starts at the end of the leaf and grows towards the
576 		 * front.
577 		 */
578 		if (btrfs_item_offset_nr(leaf, slot) !=
579 			btrfs_item_end_nr(leaf, slot + 1)) {
580 			CORRUPT("slot offset bad", leaf, root, slot);
581 			return -EIO;
582 		}
583 
584 		/*
585 		 * Check to make sure that we don't point outside of the leaf,
586 		 * just incase all the items are consistent to eachother, but
587 		 * all point outside of the leaf.
588 		 */
589 		if (btrfs_item_end_nr(leaf, slot) >
590 		    BTRFS_LEAF_DATA_SIZE(root)) {
591 			CORRUPT("slot end outside of leaf", leaf, root, slot);
592 			return -EIO;
593 		}
594 	}
595 
596 	return 0;
597 }
598 
599 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
600 				      u64 phy_offset, struct page *page,
601 				      u64 start, u64 end, int mirror)
602 {
603 	u64 found_start;
604 	int found_level;
605 	struct extent_buffer *eb;
606 	struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
607 	int ret = 0;
608 	int reads_done;
609 
610 	if (!page->private)
611 		goto out;
612 
613 	eb = (struct extent_buffer *)page->private;
614 
615 	/* the pending IO might have been the only thing that kept this buffer
616 	 * in memory.  Make sure we have a ref for all this other checks
617 	 */
618 	extent_buffer_get(eb);
619 
620 	reads_done = atomic_dec_and_test(&eb->io_pages);
621 	if (!reads_done)
622 		goto err;
623 
624 	eb->read_mirror = mirror;
625 	if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
626 		ret = -EIO;
627 		goto err;
628 	}
629 
630 	found_start = btrfs_header_bytenr(eb);
631 	if (found_start != eb->start) {
632 		btrfs_err_rl(eb->fs_info, "bad tree block start %llu %llu",
633 			       found_start, eb->start);
634 		ret = -EIO;
635 		goto err;
636 	}
637 	if (check_tree_block_fsid(root->fs_info, eb)) {
638 		btrfs_err_rl(eb->fs_info, "bad fsid on block %llu",
639 			       eb->start);
640 		ret = -EIO;
641 		goto err;
642 	}
643 	found_level = btrfs_header_level(eb);
644 	if (found_level >= BTRFS_MAX_LEVEL) {
645 		btrfs_err(root->fs_info, "bad tree block level %d",
646 			   (int)btrfs_header_level(eb));
647 		ret = -EIO;
648 		goto err;
649 	}
650 
651 	btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
652 				       eb, found_level);
653 
654 	ret = csum_tree_block(root->fs_info, eb, 1);
655 	if (ret) {
656 		ret = -EIO;
657 		goto err;
658 	}
659 
660 	/*
661 	 * If this is a leaf block and it is corrupt, set the corrupt bit so
662 	 * that we don't try and read the other copies of this block, just
663 	 * return -EIO.
664 	 */
665 	if (found_level == 0 && check_leaf(root, eb)) {
666 		set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
667 		ret = -EIO;
668 	}
669 
670 	if (!ret)
671 		set_extent_buffer_uptodate(eb);
672 err:
673 	if (reads_done &&
674 	    test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
675 		btree_readahead_hook(root, eb, eb->start, ret);
676 
677 	if (ret) {
678 		/*
679 		 * our io error hook is going to dec the io pages
680 		 * again, we have to make sure it has something
681 		 * to decrement
682 		 */
683 		atomic_inc(&eb->io_pages);
684 		clear_extent_buffer_uptodate(eb);
685 	}
686 	free_extent_buffer(eb);
687 out:
688 	return ret;
689 }
690 
691 static int btree_io_failed_hook(struct page *page, int failed_mirror)
692 {
693 	struct extent_buffer *eb;
694 	struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
695 
696 	eb = (struct extent_buffer *)page->private;
697 	set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
698 	eb->read_mirror = failed_mirror;
699 	atomic_dec(&eb->io_pages);
700 	if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
701 		btree_readahead_hook(root, eb, eb->start, -EIO);
702 	return -EIO;	/* we fixed nothing */
703 }
704 
705 static void end_workqueue_bio(struct bio *bio)
706 {
707 	struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
708 	struct btrfs_fs_info *fs_info;
709 	struct btrfs_workqueue *wq;
710 	btrfs_work_func_t func;
711 
712 	fs_info = end_io_wq->info;
713 	end_io_wq->error = bio->bi_error;
714 
715 	if (bio->bi_rw & REQ_WRITE) {
716 		if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
717 			wq = fs_info->endio_meta_write_workers;
718 			func = btrfs_endio_meta_write_helper;
719 		} else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
720 			wq = fs_info->endio_freespace_worker;
721 			func = btrfs_freespace_write_helper;
722 		} else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
723 			wq = fs_info->endio_raid56_workers;
724 			func = btrfs_endio_raid56_helper;
725 		} else {
726 			wq = fs_info->endio_write_workers;
727 			func = btrfs_endio_write_helper;
728 		}
729 	} else {
730 		if (unlikely(end_io_wq->metadata ==
731 			     BTRFS_WQ_ENDIO_DIO_REPAIR)) {
732 			wq = fs_info->endio_repair_workers;
733 			func = btrfs_endio_repair_helper;
734 		} else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
735 			wq = fs_info->endio_raid56_workers;
736 			func = btrfs_endio_raid56_helper;
737 		} else if (end_io_wq->metadata) {
738 			wq = fs_info->endio_meta_workers;
739 			func = btrfs_endio_meta_helper;
740 		} else {
741 			wq = fs_info->endio_workers;
742 			func = btrfs_endio_helper;
743 		}
744 	}
745 
746 	btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
747 	btrfs_queue_work(wq, &end_io_wq->work);
748 }
749 
750 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
751 			enum btrfs_wq_endio_type metadata)
752 {
753 	struct btrfs_end_io_wq *end_io_wq;
754 
755 	end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
756 	if (!end_io_wq)
757 		return -ENOMEM;
758 
759 	end_io_wq->private = bio->bi_private;
760 	end_io_wq->end_io = bio->bi_end_io;
761 	end_io_wq->info = info;
762 	end_io_wq->error = 0;
763 	end_io_wq->bio = bio;
764 	end_io_wq->metadata = metadata;
765 
766 	bio->bi_private = end_io_wq;
767 	bio->bi_end_io = end_workqueue_bio;
768 	return 0;
769 }
770 
771 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
772 {
773 	unsigned long limit = min_t(unsigned long,
774 				    info->thread_pool_size,
775 				    info->fs_devices->open_devices);
776 	return 256 * limit;
777 }
778 
779 static void run_one_async_start(struct btrfs_work *work)
780 {
781 	struct async_submit_bio *async;
782 	int ret;
783 
784 	async = container_of(work, struct  async_submit_bio, work);
785 	ret = async->submit_bio_start(async->inode, async->rw, async->bio,
786 				      async->mirror_num, async->bio_flags,
787 				      async->bio_offset);
788 	if (ret)
789 		async->error = ret;
790 }
791 
792 static void run_one_async_done(struct btrfs_work *work)
793 {
794 	struct btrfs_fs_info *fs_info;
795 	struct async_submit_bio *async;
796 	int limit;
797 
798 	async = container_of(work, struct  async_submit_bio, work);
799 	fs_info = BTRFS_I(async->inode)->root->fs_info;
800 
801 	limit = btrfs_async_submit_limit(fs_info);
802 	limit = limit * 2 / 3;
803 
804 	/*
805 	 * atomic_dec_return implies a barrier for waitqueue_active
806 	 */
807 	if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
808 	    waitqueue_active(&fs_info->async_submit_wait))
809 		wake_up(&fs_info->async_submit_wait);
810 
811 	/* If an error occured we just want to clean up the bio and move on */
812 	if (async->error) {
813 		async->bio->bi_error = async->error;
814 		bio_endio(async->bio);
815 		return;
816 	}
817 
818 	async->submit_bio_done(async->inode, async->rw, async->bio,
819 			       async->mirror_num, async->bio_flags,
820 			       async->bio_offset);
821 }
822 
823 static void run_one_async_free(struct btrfs_work *work)
824 {
825 	struct async_submit_bio *async;
826 
827 	async = container_of(work, struct  async_submit_bio, work);
828 	kfree(async);
829 }
830 
831 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
832 			int rw, struct bio *bio, int mirror_num,
833 			unsigned long bio_flags,
834 			u64 bio_offset,
835 			extent_submit_bio_hook_t *submit_bio_start,
836 			extent_submit_bio_hook_t *submit_bio_done)
837 {
838 	struct async_submit_bio *async;
839 
840 	async = kmalloc(sizeof(*async), GFP_NOFS);
841 	if (!async)
842 		return -ENOMEM;
843 
844 	async->inode = inode;
845 	async->rw = rw;
846 	async->bio = bio;
847 	async->mirror_num = mirror_num;
848 	async->submit_bio_start = submit_bio_start;
849 	async->submit_bio_done = submit_bio_done;
850 
851 	btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
852 			run_one_async_done, run_one_async_free);
853 
854 	async->bio_flags = bio_flags;
855 	async->bio_offset = bio_offset;
856 
857 	async->error = 0;
858 
859 	atomic_inc(&fs_info->nr_async_submits);
860 
861 	if (rw & REQ_SYNC)
862 		btrfs_set_work_high_priority(&async->work);
863 
864 	btrfs_queue_work(fs_info->workers, &async->work);
865 
866 	while (atomic_read(&fs_info->async_submit_draining) &&
867 	      atomic_read(&fs_info->nr_async_submits)) {
868 		wait_event(fs_info->async_submit_wait,
869 			   (atomic_read(&fs_info->nr_async_submits) == 0));
870 	}
871 
872 	return 0;
873 }
874 
875 static int btree_csum_one_bio(struct bio *bio)
876 {
877 	struct bio_vec *bvec;
878 	struct btrfs_root *root;
879 	int i, ret = 0;
880 
881 	bio_for_each_segment_all(bvec, bio, i) {
882 		root = BTRFS_I(bvec->bv_page->mapping->host)->root;
883 		ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
884 		if (ret)
885 			break;
886 	}
887 
888 	return ret;
889 }
890 
891 static int __btree_submit_bio_start(struct inode *inode, int rw,
892 				    struct bio *bio, int mirror_num,
893 				    unsigned long bio_flags,
894 				    u64 bio_offset)
895 {
896 	/*
897 	 * when we're called for a write, we're already in the async
898 	 * submission context.  Just jump into btrfs_map_bio
899 	 */
900 	return btree_csum_one_bio(bio);
901 }
902 
903 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
904 				 int mirror_num, unsigned long bio_flags,
905 				 u64 bio_offset)
906 {
907 	int ret;
908 
909 	/*
910 	 * when we're called for a write, we're already in the async
911 	 * submission context.  Just jump into btrfs_map_bio
912 	 */
913 	ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
914 	if (ret) {
915 		bio->bi_error = ret;
916 		bio_endio(bio);
917 	}
918 	return ret;
919 }
920 
921 static int check_async_write(struct inode *inode, unsigned long bio_flags)
922 {
923 	if (bio_flags & EXTENT_BIO_TREE_LOG)
924 		return 0;
925 #ifdef CONFIG_X86
926 	if (cpu_has_xmm4_2)
927 		return 0;
928 #endif
929 	return 1;
930 }
931 
932 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
933 				 int mirror_num, unsigned long bio_flags,
934 				 u64 bio_offset)
935 {
936 	int async = check_async_write(inode, bio_flags);
937 	int ret;
938 
939 	if (!(rw & REQ_WRITE)) {
940 		/*
941 		 * called for a read, do the setup so that checksum validation
942 		 * can happen in the async kernel threads
943 		 */
944 		ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
945 					  bio, BTRFS_WQ_ENDIO_METADATA);
946 		if (ret)
947 			goto out_w_error;
948 		ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
949 				    mirror_num, 0);
950 	} else if (!async) {
951 		ret = btree_csum_one_bio(bio);
952 		if (ret)
953 			goto out_w_error;
954 		ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
955 				    mirror_num, 0);
956 	} else {
957 		/*
958 		 * kthread helpers are used to submit writes so that
959 		 * checksumming can happen in parallel across all CPUs
960 		 */
961 		ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
962 					  inode, rw, bio, mirror_num, 0,
963 					  bio_offset,
964 					  __btree_submit_bio_start,
965 					  __btree_submit_bio_done);
966 	}
967 
968 	if (ret)
969 		goto out_w_error;
970 	return 0;
971 
972 out_w_error:
973 	bio->bi_error = ret;
974 	bio_endio(bio);
975 	return ret;
976 }
977 
978 #ifdef CONFIG_MIGRATION
979 static int btree_migratepage(struct address_space *mapping,
980 			struct page *newpage, struct page *page,
981 			enum migrate_mode mode)
982 {
983 	/*
984 	 * we can't safely write a btree page from here,
985 	 * we haven't done the locking hook
986 	 */
987 	if (PageDirty(page))
988 		return -EAGAIN;
989 	/*
990 	 * Buffers may be managed in a filesystem specific way.
991 	 * We must have no buffers or drop them.
992 	 */
993 	if (page_has_private(page) &&
994 	    !try_to_release_page(page, GFP_KERNEL))
995 		return -EAGAIN;
996 	return migrate_page(mapping, newpage, page, mode);
997 }
998 #endif
999 
1000 
1001 static int btree_writepages(struct address_space *mapping,
1002 			    struct writeback_control *wbc)
1003 {
1004 	struct btrfs_fs_info *fs_info;
1005 	int ret;
1006 
1007 	if (wbc->sync_mode == WB_SYNC_NONE) {
1008 
1009 		if (wbc->for_kupdate)
1010 			return 0;
1011 
1012 		fs_info = BTRFS_I(mapping->host)->root->fs_info;
1013 		/* this is a bit racy, but that's ok */
1014 		ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
1015 					     BTRFS_DIRTY_METADATA_THRESH);
1016 		if (ret < 0)
1017 			return 0;
1018 	}
1019 	return btree_write_cache_pages(mapping, wbc);
1020 }
1021 
1022 static int btree_readpage(struct file *file, struct page *page)
1023 {
1024 	struct extent_io_tree *tree;
1025 	tree = &BTRFS_I(page->mapping->host)->io_tree;
1026 	return extent_read_full_page(tree, page, btree_get_extent, 0);
1027 }
1028 
1029 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1030 {
1031 	if (PageWriteback(page) || PageDirty(page))
1032 		return 0;
1033 
1034 	return try_release_extent_buffer(page);
1035 }
1036 
1037 static void btree_invalidatepage(struct page *page, unsigned int offset,
1038 				 unsigned int length)
1039 {
1040 	struct extent_io_tree *tree;
1041 	tree = &BTRFS_I(page->mapping->host)->io_tree;
1042 	extent_invalidatepage(tree, page, offset);
1043 	btree_releasepage(page, GFP_NOFS);
1044 	if (PagePrivate(page)) {
1045 		btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1046 			   "page private not zero on page %llu",
1047 			   (unsigned long long)page_offset(page));
1048 		ClearPagePrivate(page);
1049 		set_page_private(page, 0);
1050 		page_cache_release(page);
1051 	}
1052 }
1053 
1054 static int btree_set_page_dirty(struct page *page)
1055 {
1056 #ifdef DEBUG
1057 	struct extent_buffer *eb;
1058 
1059 	BUG_ON(!PagePrivate(page));
1060 	eb = (struct extent_buffer *)page->private;
1061 	BUG_ON(!eb);
1062 	BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1063 	BUG_ON(!atomic_read(&eb->refs));
1064 	btrfs_assert_tree_locked(eb);
1065 #endif
1066 	return __set_page_dirty_nobuffers(page);
1067 }
1068 
1069 static const struct address_space_operations btree_aops = {
1070 	.readpage	= btree_readpage,
1071 	.writepages	= btree_writepages,
1072 	.releasepage	= btree_releasepage,
1073 	.invalidatepage = btree_invalidatepage,
1074 #ifdef CONFIG_MIGRATION
1075 	.migratepage	= btree_migratepage,
1076 #endif
1077 	.set_page_dirty = btree_set_page_dirty,
1078 };
1079 
1080 void readahead_tree_block(struct btrfs_root *root, u64 bytenr)
1081 {
1082 	struct extent_buffer *buf = NULL;
1083 	struct inode *btree_inode = root->fs_info->btree_inode;
1084 
1085 	buf = btrfs_find_create_tree_block(root, bytenr);
1086 	if (!buf)
1087 		return;
1088 	read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1089 				 buf, 0, WAIT_NONE, btree_get_extent, 0);
1090 	free_extent_buffer(buf);
1091 }
1092 
1093 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr,
1094 			 int mirror_num, struct extent_buffer **eb)
1095 {
1096 	struct extent_buffer *buf = NULL;
1097 	struct inode *btree_inode = root->fs_info->btree_inode;
1098 	struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1099 	int ret;
1100 
1101 	buf = btrfs_find_create_tree_block(root, bytenr);
1102 	if (!buf)
1103 		return 0;
1104 
1105 	set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1106 
1107 	ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1108 				       btree_get_extent, mirror_num);
1109 	if (ret) {
1110 		free_extent_buffer(buf);
1111 		return ret;
1112 	}
1113 
1114 	if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1115 		free_extent_buffer(buf);
1116 		return -EIO;
1117 	} else if (extent_buffer_uptodate(buf)) {
1118 		*eb = buf;
1119 	} else {
1120 		free_extent_buffer(buf);
1121 	}
1122 	return 0;
1123 }
1124 
1125 struct extent_buffer *btrfs_find_tree_block(struct btrfs_fs_info *fs_info,
1126 					    u64 bytenr)
1127 {
1128 	return find_extent_buffer(fs_info, bytenr);
1129 }
1130 
1131 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1132 						 u64 bytenr)
1133 {
1134 	if (btrfs_test_is_dummy_root(root))
1135 		return alloc_test_extent_buffer(root->fs_info, bytenr);
1136 	return alloc_extent_buffer(root->fs_info, bytenr);
1137 }
1138 
1139 
1140 int btrfs_write_tree_block(struct extent_buffer *buf)
1141 {
1142 	return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1143 					buf->start + buf->len - 1);
1144 }
1145 
1146 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1147 {
1148 	return filemap_fdatawait_range(buf->pages[0]->mapping,
1149 				       buf->start, buf->start + buf->len - 1);
1150 }
1151 
1152 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1153 				      u64 parent_transid)
1154 {
1155 	struct extent_buffer *buf = NULL;
1156 	int ret;
1157 
1158 	buf = btrfs_find_create_tree_block(root, bytenr);
1159 	if (!buf)
1160 		return ERR_PTR(-ENOMEM);
1161 
1162 	ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1163 	if (ret) {
1164 		free_extent_buffer(buf);
1165 		return ERR_PTR(ret);
1166 	}
1167 	return buf;
1168 
1169 }
1170 
1171 void clean_tree_block(struct btrfs_trans_handle *trans,
1172 		      struct btrfs_fs_info *fs_info,
1173 		      struct extent_buffer *buf)
1174 {
1175 	if (btrfs_header_generation(buf) ==
1176 	    fs_info->running_transaction->transid) {
1177 		btrfs_assert_tree_locked(buf);
1178 
1179 		if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1180 			__percpu_counter_add(&fs_info->dirty_metadata_bytes,
1181 					     -buf->len,
1182 					     fs_info->dirty_metadata_batch);
1183 			/* ugh, clear_extent_buffer_dirty needs to lock the page */
1184 			btrfs_set_lock_blocking(buf);
1185 			clear_extent_buffer_dirty(buf);
1186 		}
1187 	}
1188 }
1189 
1190 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1191 {
1192 	struct btrfs_subvolume_writers *writers;
1193 	int ret;
1194 
1195 	writers = kmalloc(sizeof(*writers), GFP_NOFS);
1196 	if (!writers)
1197 		return ERR_PTR(-ENOMEM);
1198 
1199 	ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL);
1200 	if (ret < 0) {
1201 		kfree(writers);
1202 		return ERR_PTR(ret);
1203 	}
1204 
1205 	init_waitqueue_head(&writers->wait);
1206 	return writers;
1207 }
1208 
1209 static void
1210 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1211 {
1212 	percpu_counter_destroy(&writers->counter);
1213 	kfree(writers);
1214 }
1215 
1216 static void __setup_root(u32 nodesize, u32 sectorsize, u32 stripesize,
1217 			 struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1218 			 u64 objectid)
1219 {
1220 	root->node = NULL;
1221 	root->commit_root = NULL;
1222 	root->sectorsize = sectorsize;
1223 	root->nodesize = nodesize;
1224 	root->stripesize = stripesize;
1225 	root->state = 0;
1226 	root->orphan_cleanup_state = 0;
1227 
1228 	root->objectid = objectid;
1229 	root->last_trans = 0;
1230 	root->highest_objectid = 0;
1231 	root->nr_delalloc_inodes = 0;
1232 	root->nr_ordered_extents = 0;
1233 	root->name = NULL;
1234 	root->inode_tree = RB_ROOT;
1235 	INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1236 	root->block_rsv = NULL;
1237 	root->orphan_block_rsv = NULL;
1238 
1239 	INIT_LIST_HEAD(&root->dirty_list);
1240 	INIT_LIST_HEAD(&root->root_list);
1241 	INIT_LIST_HEAD(&root->delalloc_inodes);
1242 	INIT_LIST_HEAD(&root->delalloc_root);
1243 	INIT_LIST_HEAD(&root->ordered_extents);
1244 	INIT_LIST_HEAD(&root->ordered_root);
1245 	INIT_LIST_HEAD(&root->logged_list[0]);
1246 	INIT_LIST_HEAD(&root->logged_list[1]);
1247 	spin_lock_init(&root->orphan_lock);
1248 	spin_lock_init(&root->inode_lock);
1249 	spin_lock_init(&root->delalloc_lock);
1250 	spin_lock_init(&root->ordered_extent_lock);
1251 	spin_lock_init(&root->accounting_lock);
1252 	spin_lock_init(&root->log_extents_lock[0]);
1253 	spin_lock_init(&root->log_extents_lock[1]);
1254 	mutex_init(&root->objectid_mutex);
1255 	mutex_init(&root->log_mutex);
1256 	mutex_init(&root->ordered_extent_mutex);
1257 	mutex_init(&root->delalloc_mutex);
1258 	init_waitqueue_head(&root->log_writer_wait);
1259 	init_waitqueue_head(&root->log_commit_wait[0]);
1260 	init_waitqueue_head(&root->log_commit_wait[1]);
1261 	INIT_LIST_HEAD(&root->log_ctxs[0]);
1262 	INIT_LIST_HEAD(&root->log_ctxs[1]);
1263 	atomic_set(&root->log_commit[0], 0);
1264 	atomic_set(&root->log_commit[1], 0);
1265 	atomic_set(&root->log_writers, 0);
1266 	atomic_set(&root->log_batch, 0);
1267 	atomic_set(&root->orphan_inodes, 0);
1268 	atomic_set(&root->refs, 1);
1269 	atomic_set(&root->will_be_snapshoted, 0);
1270 	root->log_transid = 0;
1271 	root->log_transid_committed = -1;
1272 	root->last_log_commit = 0;
1273 	if (fs_info)
1274 		extent_io_tree_init(&root->dirty_log_pages,
1275 				     fs_info->btree_inode->i_mapping);
1276 
1277 	memset(&root->root_key, 0, sizeof(root->root_key));
1278 	memset(&root->root_item, 0, sizeof(root->root_item));
1279 	memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1280 	if (fs_info)
1281 		root->defrag_trans_start = fs_info->generation;
1282 	else
1283 		root->defrag_trans_start = 0;
1284 	root->root_key.objectid = objectid;
1285 	root->anon_dev = 0;
1286 
1287 	spin_lock_init(&root->root_item_lock);
1288 }
1289 
1290 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1291 {
1292 	struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1293 	if (root)
1294 		root->fs_info = fs_info;
1295 	return root;
1296 }
1297 
1298 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1299 /* Should only be used by the testing infrastructure */
1300 struct btrfs_root *btrfs_alloc_dummy_root(void)
1301 {
1302 	struct btrfs_root *root;
1303 
1304 	root = btrfs_alloc_root(NULL);
1305 	if (!root)
1306 		return ERR_PTR(-ENOMEM);
1307 	__setup_root(4096, 4096, 4096, root, NULL, 1);
1308 	set_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state);
1309 	root->alloc_bytenr = 0;
1310 
1311 	return root;
1312 }
1313 #endif
1314 
1315 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1316 				     struct btrfs_fs_info *fs_info,
1317 				     u64 objectid)
1318 {
1319 	struct extent_buffer *leaf;
1320 	struct btrfs_root *tree_root = fs_info->tree_root;
1321 	struct btrfs_root *root;
1322 	struct btrfs_key key;
1323 	int ret = 0;
1324 	uuid_le uuid;
1325 
1326 	root = btrfs_alloc_root(fs_info);
1327 	if (!root)
1328 		return ERR_PTR(-ENOMEM);
1329 
1330 	__setup_root(tree_root->nodesize, tree_root->sectorsize,
1331 		tree_root->stripesize, root, fs_info, objectid);
1332 	root->root_key.objectid = objectid;
1333 	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1334 	root->root_key.offset = 0;
1335 
1336 	leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1337 	if (IS_ERR(leaf)) {
1338 		ret = PTR_ERR(leaf);
1339 		leaf = NULL;
1340 		goto fail;
1341 	}
1342 
1343 	memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1344 	btrfs_set_header_bytenr(leaf, leaf->start);
1345 	btrfs_set_header_generation(leaf, trans->transid);
1346 	btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1347 	btrfs_set_header_owner(leaf, objectid);
1348 	root->node = leaf;
1349 
1350 	write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1351 			    BTRFS_FSID_SIZE);
1352 	write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1353 			    btrfs_header_chunk_tree_uuid(leaf),
1354 			    BTRFS_UUID_SIZE);
1355 	btrfs_mark_buffer_dirty(leaf);
1356 
1357 	root->commit_root = btrfs_root_node(root);
1358 	set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1359 
1360 	root->root_item.flags = 0;
1361 	root->root_item.byte_limit = 0;
1362 	btrfs_set_root_bytenr(&root->root_item, leaf->start);
1363 	btrfs_set_root_generation(&root->root_item, trans->transid);
1364 	btrfs_set_root_level(&root->root_item, 0);
1365 	btrfs_set_root_refs(&root->root_item, 1);
1366 	btrfs_set_root_used(&root->root_item, leaf->len);
1367 	btrfs_set_root_last_snapshot(&root->root_item, 0);
1368 	btrfs_set_root_dirid(&root->root_item, 0);
1369 	uuid_le_gen(&uuid);
1370 	memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1371 	root->root_item.drop_level = 0;
1372 
1373 	key.objectid = objectid;
1374 	key.type = BTRFS_ROOT_ITEM_KEY;
1375 	key.offset = 0;
1376 	ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1377 	if (ret)
1378 		goto fail;
1379 
1380 	btrfs_tree_unlock(leaf);
1381 
1382 	return root;
1383 
1384 fail:
1385 	if (leaf) {
1386 		btrfs_tree_unlock(leaf);
1387 		free_extent_buffer(root->commit_root);
1388 		free_extent_buffer(leaf);
1389 	}
1390 	kfree(root);
1391 
1392 	return ERR_PTR(ret);
1393 }
1394 
1395 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1396 					 struct btrfs_fs_info *fs_info)
1397 {
1398 	struct btrfs_root *root;
1399 	struct btrfs_root *tree_root = fs_info->tree_root;
1400 	struct extent_buffer *leaf;
1401 
1402 	root = btrfs_alloc_root(fs_info);
1403 	if (!root)
1404 		return ERR_PTR(-ENOMEM);
1405 
1406 	__setup_root(tree_root->nodesize, tree_root->sectorsize,
1407 		     tree_root->stripesize, root, fs_info,
1408 		     BTRFS_TREE_LOG_OBJECTID);
1409 
1410 	root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1411 	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1412 	root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1413 
1414 	/*
1415 	 * DON'T set REF_COWS for log trees
1416 	 *
1417 	 * log trees do not get reference counted because they go away
1418 	 * before a real commit is actually done.  They do store pointers
1419 	 * to file data extents, and those reference counts still get
1420 	 * updated (along with back refs to the log tree).
1421 	 */
1422 
1423 	leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1424 			NULL, 0, 0, 0);
1425 	if (IS_ERR(leaf)) {
1426 		kfree(root);
1427 		return ERR_CAST(leaf);
1428 	}
1429 
1430 	memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1431 	btrfs_set_header_bytenr(leaf, leaf->start);
1432 	btrfs_set_header_generation(leaf, trans->transid);
1433 	btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1434 	btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1435 	root->node = leaf;
1436 
1437 	write_extent_buffer(root->node, root->fs_info->fsid,
1438 			    btrfs_header_fsid(), BTRFS_FSID_SIZE);
1439 	btrfs_mark_buffer_dirty(root->node);
1440 	btrfs_tree_unlock(root->node);
1441 	return root;
1442 }
1443 
1444 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1445 			     struct btrfs_fs_info *fs_info)
1446 {
1447 	struct btrfs_root *log_root;
1448 
1449 	log_root = alloc_log_tree(trans, fs_info);
1450 	if (IS_ERR(log_root))
1451 		return PTR_ERR(log_root);
1452 	WARN_ON(fs_info->log_root_tree);
1453 	fs_info->log_root_tree = log_root;
1454 	return 0;
1455 }
1456 
1457 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1458 		       struct btrfs_root *root)
1459 {
1460 	struct btrfs_root *log_root;
1461 	struct btrfs_inode_item *inode_item;
1462 
1463 	log_root = alloc_log_tree(trans, root->fs_info);
1464 	if (IS_ERR(log_root))
1465 		return PTR_ERR(log_root);
1466 
1467 	log_root->last_trans = trans->transid;
1468 	log_root->root_key.offset = root->root_key.objectid;
1469 
1470 	inode_item = &log_root->root_item.inode;
1471 	btrfs_set_stack_inode_generation(inode_item, 1);
1472 	btrfs_set_stack_inode_size(inode_item, 3);
1473 	btrfs_set_stack_inode_nlink(inode_item, 1);
1474 	btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
1475 	btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1476 
1477 	btrfs_set_root_node(&log_root->root_item, log_root->node);
1478 
1479 	WARN_ON(root->log_root);
1480 	root->log_root = log_root;
1481 	root->log_transid = 0;
1482 	root->log_transid_committed = -1;
1483 	root->last_log_commit = 0;
1484 	return 0;
1485 }
1486 
1487 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1488 					       struct btrfs_key *key)
1489 {
1490 	struct btrfs_root *root;
1491 	struct btrfs_fs_info *fs_info = tree_root->fs_info;
1492 	struct btrfs_path *path;
1493 	u64 generation;
1494 	int ret;
1495 
1496 	path = btrfs_alloc_path();
1497 	if (!path)
1498 		return ERR_PTR(-ENOMEM);
1499 
1500 	root = btrfs_alloc_root(fs_info);
1501 	if (!root) {
1502 		ret = -ENOMEM;
1503 		goto alloc_fail;
1504 	}
1505 
1506 	__setup_root(tree_root->nodesize, tree_root->sectorsize,
1507 		tree_root->stripesize, root, fs_info, key->objectid);
1508 
1509 	ret = btrfs_find_root(tree_root, key, path,
1510 			      &root->root_item, &root->root_key);
1511 	if (ret) {
1512 		if (ret > 0)
1513 			ret = -ENOENT;
1514 		goto find_fail;
1515 	}
1516 
1517 	generation = btrfs_root_generation(&root->root_item);
1518 	root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1519 				     generation);
1520 	if (IS_ERR(root->node)) {
1521 		ret = PTR_ERR(root->node);
1522 		goto find_fail;
1523 	} else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1524 		ret = -EIO;
1525 		free_extent_buffer(root->node);
1526 		goto find_fail;
1527 	}
1528 	root->commit_root = btrfs_root_node(root);
1529 out:
1530 	btrfs_free_path(path);
1531 	return root;
1532 
1533 find_fail:
1534 	kfree(root);
1535 alloc_fail:
1536 	root = ERR_PTR(ret);
1537 	goto out;
1538 }
1539 
1540 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1541 				      struct btrfs_key *location)
1542 {
1543 	struct btrfs_root *root;
1544 
1545 	root = btrfs_read_tree_root(tree_root, location);
1546 	if (IS_ERR(root))
1547 		return root;
1548 
1549 	if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1550 		set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1551 		btrfs_check_and_init_root_item(&root->root_item);
1552 	}
1553 
1554 	return root;
1555 }
1556 
1557 int btrfs_init_fs_root(struct btrfs_root *root)
1558 {
1559 	int ret;
1560 	struct btrfs_subvolume_writers *writers;
1561 
1562 	root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1563 	root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1564 					GFP_NOFS);
1565 	if (!root->free_ino_pinned || !root->free_ino_ctl) {
1566 		ret = -ENOMEM;
1567 		goto fail;
1568 	}
1569 
1570 	writers = btrfs_alloc_subvolume_writers();
1571 	if (IS_ERR(writers)) {
1572 		ret = PTR_ERR(writers);
1573 		goto fail;
1574 	}
1575 	root->subv_writers = writers;
1576 
1577 	btrfs_init_free_ino_ctl(root);
1578 	spin_lock_init(&root->ino_cache_lock);
1579 	init_waitqueue_head(&root->ino_cache_wait);
1580 
1581 	ret = get_anon_bdev(&root->anon_dev);
1582 	if (ret)
1583 		goto free_writers;
1584 	return 0;
1585 
1586 free_writers:
1587 	btrfs_free_subvolume_writers(root->subv_writers);
1588 fail:
1589 	kfree(root->free_ino_ctl);
1590 	kfree(root->free_ino_pinned);
1591 	return ret;
1592 }
1593 
1594 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1595 					       u64 root_id)
1596 {
1597 	struct btrfs_root *root;
1598 
1599 	spin_lock(&fs_info->fs_roots_radix_lock);
1600 	root = radix_tree_lookup(&fs_info->fs_roots_radix,
1601 				 (unsigned long)root_id);
1602 	spin_unlock(&fs_info->fs_roots_radix_lock);
1603 	return root;
1604 }
1605 
1606 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1607 			 struct btrfs_root *root)
1608 {
1609 	int ret;
1610 
1611 	ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1612 	if (ret)
1613 		return ret;
1614 
1615 	spin_lock(&fs_info->fs_roots_radix_lock);
1616 	ret = radix_tree_insert(&fs_info->fs_roots_radix,
1617 				(unsigned long)root->root_key.objectid,
1618 				root);
1619 	if (ret == 0)
1620 		set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1621 	spin_unlock(&fs_info->fs_roots_radix_lock);
1622 	radix_tree_preload_end();
1623 
1624 	return ret;
1625 }
1626 
1627 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1628 				     struct btrfs_key *location,
1629 				     bool check_ref)
1630 {
1631 	struct btrfs_root *root;
1632 	struct btrfs_path *path;
1633 	struct btrfs_key key;
1634 	int ret;
1635 
1636 	if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1637 		return fs_info->tree_root;
1638 	if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1639 		return fs_info->extent_root;
1640 	if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1641 		return fs_info->chunk_root;
1642 	if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1643 		return fs_info->dev_root;
1644 	if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1645 		return fs_info->csum_root;
1646 	if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1647 		return fs_info->quota_root ? fs_info->quota_root :
1648 					     ERR_PTR(-ENOENT);
1649 	if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1650 		return fs_info->uuid_root ? fs_info->uuid_root :
1651 					    ERR_PTR(-ENOENT);
1652 again:
1653 	root = btrfs_lookup_fs_root(fs_info, location->objectid);
1654 	if (root) {
1655 		if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1656 			return ERR_PTR(-ENOENT);
1657 		return root;
1658 	}
1659 
1660 	root = btrfs_read_fs_root(fs_info->tree_root, location);
1661 	if (IS_ERR(root))
1662 		return root;
1663 
1664 	if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1665 		ret = -ENOENT;
1666 		goto fail;
1667 	}
1668 
1669 	ret = btrfs_init_fs_root(root);
1670 	if (ret)
1671 		goto fail;
1672 
1673 	path = btrfs_alloc_path();
1674 	if (!path) {
1675 		ret = -ENOMEM;
1676 		goto fail;
1677 	}
1678 	key.objectid = BTRFS_ORPHAN_OBJECTID;
1679 	key.type = BTRFS_ORPHAN_ITEM_KEY;
1680 	key.offset = location->objectid;
1681 
1682 	ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1683 	btrfs_free_path(path);
1684 	if (ret < 0)
1685 		goto fail;
1686 	if (ret == 0)
1687 		set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1688 
1689 	ret = btrfs_insert_fs_root(fs_info, root);
1690 	if (ret) {
1691 		if (ret == -EEXIST) {
1692 			free_fs_root(root);
1693 			goto again;
1694 		}
1695 		goto fail;
1696 	}
1697 	return root;
1698 fail:
1699 	free_fs_root(root);
1700 	return ERR_PTR(ret);
1701 }
1702 
1703 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1704 {
1705 	struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1706 	int ret = 0;
1707 	struct btrfs_device *device;
1708 	struct backing_dev_info *bdi;
1709 
1710 	rcu_read_lock();
1711 	list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1712 		if (!device->bdev)
1713 			continue;
1714 		bdi = blk_get_backing_dev_info(device->bdev);
1715 		if (bdi_congested(bdi, bdi_bits)) {
1716 			ret = 1;
1717 			break;
1718 		}
1719 	}
1720 	rcu_read_unlock();
1721 	return ret;
1722 }
1723 
1724 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1725 {
1726 	int err;
1727 
1728 	err = bdi_setup_and_register(bdi, "btrfs");
1729 	if (err)
1730 		return err;
1731 
1732 	bdi->ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE;
1733 	bdi->congested_fn	= btrfs_congested_fn;
1734 	bdi->congested_data	= info;
1735 	bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
1736 	return 0;
1737 }
1738 
1739 /*
1740  * called by the kthread helper functions to finally call the bio end_io
1741  * functions.  This is where read checksum verification actually happens
1742  */
1743 static void end_workqueue_fn(struct btrfs_work *work)
1744 {
1745 	struct bio *bio;
1746 	struct btrfs_end_io_wq *end_io_wq;
1747 
1748 	end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1749 	bio = end_io_wq->bio;
1750 
1751 	bio->bi_error = end_io_wq->error;
1752 	bio->bi_private = end_io_wq->private;
1753 	bio->bi_end_io = end_io_wq->end_io;
1754 	kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1755 	bio_endio(bio);
1756 }
1757 
1758 static int cleaner_kthread(void *arg)
1759 {
1760 	struct btrfs_root *root = arg;
1761 	int again;
1762 	struct btrfs_trans_handle *trans;
1763 
1764 	do {
1765 		again = 0;
1766 
1767 		/* Make the cleaner go to sleep early. */
1768 		if (btrfs_need_cleaner_sleep(root))
1769 			goto sleep;
1770 
1771 		if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1772 			goto sleep;
1773 
1774 		/*
1775 		 * Avoid the problem that we change the status of the fs
1776 		 * during the above check and trylock.
1777 		 */
1778 		if (btrfs_need_cleaner_sleep(root)) {
1779 			mutex_unlock(&root->fs_info->cleaner_mutex);
1780 			goto sleep;
1781 		}
1782 
1783 		btrfs_run_delayed_iputs(root);
1784 		again = btrfs_clean_one_deleted_snapshot(root);
1785 		mutex_unlock(&root->fs_info->cleaner_mutex);
1786 
1787 		/*
1788 		 * The defragger has dealt with the R/O remount and umount,
1789 		 * needn't do anything special here.
1790 		 */
1791 		btrfs_run_defrag_inodes(root->fs_info);
1792 
1793 		/*
1794 		 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1795 		 * with relocation (btrfs_relocate_chunk) and relocation
1796 		 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1797 		 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1798 		 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1799 		 * unused block groups.
1800 		 */
1801 		btrfs_delete_unused_bgs(root->fs_info);
1802 sleep:
1803 		if (!try_to_freeze() && !again) {
1804 			set_current_state(TASK_INTERRUPTIBLE);
1805 			if (!kthread_should_stop())
1806 				schedule();
1807 			__set_current_state(TASK_RUNNING);
1808 		}
1809 	} while (!kthread_should_stop());
1810 
1811 	/*
1812 	 * Transaction kthread is stopped before us and wakes us up.
1813 	 * However we might have started a new transaction and COWed some
1814 	 * tree blocks when deleting unused block groups for example. So
1815 	 * make sure we commit the transaction we started to have a clean
1816 	 * shutdown when evicting the btree inode - if it has dirty pages
1817 	 * when we do the final iput() on it, eviction will trigger a
1818 	 * writeback for it which will fail with null pointer dereferences
1819 	 * since work queues and other resources were already released and
1820 	 * destroyed by the time the iput/eviction/writeback is made.
1821 	 */
1822 	trans = btrfs_attach_transaction(root);
1823 	if (IS_ERR(trans)) {
1824 		if (PTR_ERR(trans) != -ENOENT)
1825 			btrfs_err(root->fs_info,
1826 				  "cleaner transaction attach returned %ld",
1827 				  PTR_ERR(trans));
1828 	} else {
1829 		int ret;
1830 
1831 		ret = btrfs_commit_transaction(trans, root);
1832 		if (ret)
1833 			btrfs_err(root->fs_info,
1834 				  "cleaner open transaction commit returned %d",
1835 				  ret);
1836 	}
1837 
1838 	return 0;
1839 }
1840 
1841 static int transaction_kthread(void *arg)
1842 {
1843 	struct btrfs_root *root = arg;
1844 	struct btrfs_trans_handle *trans;
1845 	struct btrfs_transaction *cur;
1846 	u64 transid;
1847 	unsigned long now;
1848 	unsigned long delay;
1849 	bool cannot_commit;
1850 
1851 	do {
1852 		cannot_commit = false;
1853 		delay = HZ * root->fs_info->commit_interval;
1854 		mutex_lock(&root->fs_info->transaction_kthread_mutex);
1855 
1856 		spin_lock(&root->fs_info->trans_lock);
1857 		cur = root->fs_info->running_transaction;
1858 		if (!cur) {
1859 			spin_unlock(&root->fs_info->trans_lock);
1860 			goto sleep;
1861 		}
1862 
1863 		now = get_seconds();
1864 		if (cur->state < TRANS_STATE_BLOCKED &&
1865 		    (now < cur->start_time ||
1866 		     now - cur->start_time < root->fs_info->commit_interval)) {
1867 			spin_unlock(&root->fs_info->trans_lock);
1868 			delay = HZ * 5;
1869 			goto sleep;
1870 		}
1871 		transid = cur->transid;
1872 		spin_unlock(&root->fs_info->trans_lock);
1873 
1874 		/* If the file system is aborted, this will always fail. */
1875 		trans = btrfs_attach_transaction(root);
1876 		if (IS_ERR(trans)) {
1877 			if (PTR_ERR(trans) != -ENOENT)
1878 				cannot_commit = true;
1879 			goto sleep;
1880 		}
1881 		if (transid == trans->transid) {
1882 			btrfs_commit_transaction(trans, root);
1883 		} else {
1884 			btrfs_end_transaction(trans, root);
1885 		}
1886 sleep:
1887 		wake_up_process(root->fs_info->cleaner_kthread);
1888 		mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1889 
1890 		if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1891 				      &root->fs_info->fs_state)))
1892 			btrfs_cleanup_transaction(root);
1893 		if (!try_to_freeze()) {
1894 			set_current_state(TASK_INTERRUPTIBLE);
1895 			if (!kthread_should_stop() &&
1896 			    (!btrfs_transaction_blocked(root->fs_info) ||
1897 			     cannot_commit))
1898 				schedule_timeout(delay);
1899 			__set_current_state(TASK_RUNNING);
1900 		}
1901 	} while (!kthread_should_stop());
1902 	return 0;
1903 }
1904 
1905 /*
1906  * this will find the highest generation in the array of
1907  * root backups.  The index of the highest array is returned,
1908  * or -1 if we can't find anything.
1909  *
1910  * We check to make sure the array is valid by comparing the
1911  * generation of the latest  root in the array with the generation
1912  * in the super block.  If they don't match we pitch it.
1913  */
1914 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1915 {
1916 	u64 cur;
1917 	int newest_index = -1;
1918 	struct btrfs_root_backup *root_backup;
1919 	int i;
1920 
1921 	for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1922 		root_backup = info->super_copy->super_roots + i;
1923 		cur = btrfs_backup_tree_root_gen(root_backup);
1924 		if (cur == newest_gen)
1925 			newest_index = i;
1926 	}
1927 
1928 	/* check to see if we actually wrapped around */
1929 	if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1930 		root_backup = info->super_copy->super_roots;
1931 		cur = btrfs_backup_tree_root_gen(root_backup);
1932 		if (cur == newest_gen)
1933 			newest_index = 0;
1934 	}
1935 	return newest_index;
1936 }
1937 
1938 
1939 /*
1940  * find the oldest backup so we know where to store new entries
1941  * in the backup array.  This will set the backup_root_index
1942  * field in the fs_info struct
1943  */
1944 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1945 				     u64 newest_gen)
1946 {
1947 	int newest_index = -1;
1948 
1949 	newest_index = find_newest_super_backup(info, newest_gen);
1950 	/* if there was garbage in there, just move along */
1951 	if (newest_index == -1) {
1952 		info->backup_root_index = 0;
1953 	} else {
1954 		info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1955 	}
1956 }
1957 
1958 /*
1959  * copy all the root pointers into the super backup array.
1960  * this will bump the backup pointer by one when it is
1961  * done
1962  */
1963 static void backup_super_roots(struct btrfs_fs_info *info)
1964 {
1965 	int next_backup;
1966 	struct btrfs_root_backup *root_backup;
1967 	int last_backup;
1968 
1969 	next_backup = info->backup_root_index;
1970 	last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1971 		BTRFS_NUM_BACKUP_ROOTS;
1972 
1973 	/*
1974 	 * just overwrite the last backup if we're at the same generation
1975 	 * this happens only at umount
1976 	 */
1977 	root_backup = info->super_for_commit->super_roots + last_backup;
1978 	if (btrfs_backup_tree_root_gen(root_backup) ==
1979 	    btrfs_header_generation(info->tree_root->node))
1980 		next_backup = last_backup;
1981 
1982 	root_backup = info->super_for_commit->super_roots + next_backup;
1983 
1984 	/*
1985 	 * make sure all of our padding and empty slots get zero filled
1986 	 * regardless of which ones we use today
1987 	 */
1988 	memset(root_backup, 0, sizeof(*root_backup));
1989 
1990 	info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1991 
1992 	btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1993 	btrfs_set_backup_tree_root_gen(root_backup,
1994 			       btrfs_header_generation(info->tree_root->node));
1995 
1996 	btrfs_set_backup_tree_root_level(root_backup,
1997 			       btrfs_header_level(info->tree_root->node));
1998 
1999 	btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2000 	btrfs_set_backup_chunk_root_gen(root_backup,
2001 			       btrfs_header_generation(info->chunk_root->node));
2002 	btrfs_set_backup_chunk_root_level(root_backup,
2003 			       btrfs_header_level(info->chunk_root->node));
2004 
2005 	btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
2006 	btrfs_set_backup_extent_root_gen(root_backup,
2007 			       btrfs_header_generation(info->extent_root->node));
2008 	btrfs_set_backup_extent_root_level(root_backup,
2009 			       btrfs_header_level(info->extent_root->node));
2010 
2011 	/*
2012 	 * we might commit during log recovery, which happens before we set
2013 	 * the fs_root.  Make sure it is valid before we fill it in.
2014 	 */
2015 	if (info->fs_root && info->fs_root->node) {
2016 		btrfs_set_backup_fs_root(root_backup,
2017 					 info->fs_root->node->start);
2018 		btrfs_set_backup_fs_root_gen(root_backup,
2019 			       btrfs_header_generation(info->fs_root->node));
2020 		btrfs_set_backup_fs_root_level(root_backup,
2021 			       btrfs_header_level(info->fs_root->node));
2022 	}
2023 
2024 	btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2025 	btrfs_set_backup_dev_root_gen(root_backup,
2026 			       btrfs_header_generation(info->dev_root->node));
2027 	btrfs_set_backup_dev_root_level(root_backup,
2028 				       btrfs_header_level(info->dev_root->node));
2029 
2030 	btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
2031 	btrfs_set_backup_csum_root_gen(root_backup,
2032 			       btrfs_header_generation(info->csum_root->node));
2033 	btrfs_set_backup_csum_root_level(root_backup,
2034 			       btrfs_header_level(info->csum_root->node));
2035 
2036 	btrfs_set_backup_total_bytes(root_backup,
2037 			     btrfs_super_total_bytes(info->super_copy));
2038 	btrfs_set_backup_bytes_used(root_backup,
2039 			     btrfs_super_bytes_used(info->super_copy));
2040 	btrfs_set_backup_num_devices(root_backup,
2041 			     btrfs_super_num_devices(info->super_copy));
2042 
2043 	/*
2044 	 * if we don't copy this out to the super_copy, it won't get remembered
2045 	 * for the next commit
2046 	 */
2047 	memcpy(&info->super_copy->super_roots,
2048 	       &info->super_for_commit->super_roots,
2049 	       sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2050 }
2051 
2052 /*
2053  * this copies info out of the root backup array and back into
2054  * the in-memory super block.  It is meant to help iterate through
2055  * the array, so you send it the number of backups you've already
2056  * tried and the last backup index you used.
2057  *
2058  * this returns -1 when it has tried all the backups
2059  */
2060 static noinline int next_root_backup(struct btrfs_fs_info *info,
2061 				     struct btrfs_super_block *super,
2062 				     int *num_backups_tried, int *backup_index)
2063 {
2064 	struct btrfs_root_backup *root_backup;
2065 	int newest = *backup_index;
2066 
2067 	if (*num_backups_tried == 0) {
2068 		u64 gen = btrfs_super_generation(super);
2069 
2070 		newest = find_newest_super_backup(info, gen);
2071 		if (newest == -1)
2072 			return -1;
2073 
2074 		*backup_index = newest;
2075 		*num_backups_tried = 1;
2076 	} else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2077 		/* we've tried all the backups, all done */
2078 		return -1;
2079 	} else {
2080 		/* jump to the next oldest backup */
2081 		newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2082 			BTRFS_NUM_BACKUP_ROOTS;
2083 		*backup_index = newest;
2084 		*num_backups_tried += 1;
2085 	}
2086 	root_backup = super->super_roots + newest;
2087 
2088 	btrfs_set_super_generation(super,
2089 				   btrfs_backup_tree_root_gen(root_backup));
2090 	btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2091 	btrfs_set_super_root_level(super,
2092 				   btrfs_backup_tree_root_level(root_backup));
2093 	btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2094 
2095 	/*
2096 	 * fixme: the total bytes and num_devices need to match or we should
2097 	 * need a fsck
2098 	 */
2099 	btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2100 	btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2101 	return 0;
2102 }
2103 
2104 /* helper to cleanup workers */
2105 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2106 {
2107 	btrfs_destroy_workqueue(fs_info->fixup_workers);
2108 	btrfs_destroy_workqueue(fs_info->delalloc_workers);
2109 	btrfs_destroy_workqueue(fs_info->workers);
2110 	btrfs_destroy_workqueue(fs_info->endio_workers);
2111 	btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2112 	btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2113 	btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2114 	btrfs_destroy_workqueue(fs_info->rmw_workers);
2115 	btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2116 	btrfs_destroy_workqueue(fs_info->endio_write_workers);
2117 	btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2118 	btrfs_destroy_workqueue(fs_info->submit_workers);
2119 	btrfs_destroy_workqueue(fs_info->delayed_workers);
2120 	btrfs_destroy_workqueue(fs_info->caching_workers);
2121 	btrfs_destroy_workqueue(fs_info->readahead_workers);
2122 	btrfs_destroy_workqueue(fs_info->flush_workers);
2123 	btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2124 	btrfs_destroy_workqueue(fs_info->extent_workers);
2125 }
2126 
2127 static void free_root_extent_buffers(struct btrfs_root *root)
2128 {
2129 	if (root) {
2130 		free_extent_buffer(root->node);
2131 		free_extent_buffer(root->commit_root);
2132 		root->node = NULL;
2133 		root->commit_root = NULL;
2134 	}
2135 }
2136 
2137 /* helper to cleanup tree roots */
2138 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2139 {
2140 	free_root_extent_buffers(info->tree_root);
2141 
2142 	free_root_extent_buffers(info->dev_root);
2143 	free_root_extent_buffers(info->extent_root);
2144 	free_root_extent_buffers(info->csum_root);
2145 	free_root_extent_buffers(info->quota_root);
2146 	free_root_extent_buffers(info->uuid_root);
2147 	if (chunk_root)
2148 		free_root_extent_buffers(info->chunk_root);
2149 }
2150 
2151 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2152 {
2153 	int ret;
2154 	struct btrfs_root *gang[8];
2155 	int i;
2156 
2157 	while (!list_empty(&fs_info->dead_roots)) {
2158 		gang[0] = list_entry(fs_info->dead_roots.next,
2159 				     struct btrfs_root, root_list);
2160 		list_del(&gang[0]->root_list);
2161 
2162 		if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2163 			btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2164 		} else {
2165 			free_extent_buffer(gang[0]->node);
2166 			free_extent_buffer(gang[0]->commit_root);
2167 			btrfs_put_fs_root(gang[0]);
2168 		}
2169 	}
2170 
2171 	while (1) {
2172 		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2173 					     (void **)gang, 0,
2174 					     ARRAY_SIZE(gang));
2175 		if (!ret)
2176 			break;
2177 		for (i = 0; i < ret; i++)
2178 			btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2179 	}
2180 
2181 	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2182 		btrfs_free_log_root_tree(NULL, fs_info);
2183 		btrfs_destroy_pinned_extent(fs_info->tree_root,
2184 					    fs_info->pinned_extents);
2185 	}
2186 }
2187 
2188 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2189 {
2190 	mutex_init(&fs_info->scrub_lock);
2191 	atomic_set(&fs_info->scrubs_running, 0);
2192 	atomic_set(&fs_info->scrub_pause_req, 0);
2193 	atomic_set(&fs_info->scrubs_paused, 0);
2194 	atomic_set(&fs_info->scrub_cancel_req, 0);
2195 	init_waitqueue_head(&fs_info->scrub_pause_wait);
2196 	fs_info->scrub_workers_refcnt = 0;
2197 }
2198 
2199 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2200 {
2201 	spin_lock_init(&fs_info->balance_lock);
2202 	mutex_init(&fs_info->balance_mutex);
2203 	atomic_set(&fs_info->balance_running, 0);
2204 	atomic_set(&fs_info->balance_pause_req, 0);
2205 	atomic_set(&fs_info->balance_cancel_req, 0);
2206 	fs_info->balance_ctl = NULL;
2207 	init_waitqueue_head(&fs_info->balance_wait_q);
2208 }
2209 
2210 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info,
2211 				   struct btrfs_root *tree_root)
2212 {
2213 	fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2214 	set_nlink(fs_info->btree_inode, 1);
2215 	/*
2216 	 * we set the i_size on the btree inode to the max possible int.
2217 	 * the real end of the address space is determined by all of
2218 	 * the devices in the system
2219 	 */
2220 	fs_info->btree_inode->i_size = OFFSET_MAX;
2221 	fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2222 
2223 	RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2224 	extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2225 			     fs_info->btree_inode->i_mapping);
2226 	BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2227 	extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2228 
2229 	BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2230 
2231 	BTRFS_I(fs_info->btree_inode)->root = tree_root;
2232 	memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2233 	       sizeof(struct btrfs_key));
2234 	set_bit(BTRFS_INODE_DUMMY,
2235 		&BTRFS_I(fs_info->btree_inode)->runtime_flags);
2236 	btrfs_insert_inode_hash(fs_info->btree_inode);
2237 }
2238 
2239 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2240 {
2241 	fs_info->dev_replace.lock_owner = 0;
2242 	atomic_set(&fs_info->dev_replace.nesting_level, 0);
2243 	mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2244 	mutex_init(&fs_info->dev_replace.lock_management_lock);
2245 	mutex_init(&fs_info->dev_replace.lock);
2246 	init_waitqueue_head(&fs_info->replace_wait);
2247 }
2248 
2249 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2250 {
2251 	spin_lock_init(&fs_info->qgroup_lock);
2252 	mutex_init(&fs_info->qgroup_ioctl_lock);
2253 	fs_info->qgroup_tree = RB_ROOT;
2254 	fs_info->qgroup_op_tree = RB_ROOT;
2255 	INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2256 	fs_info->qgroup_seq = 1;
2257 	fs_info->quota_enabled = 0;
2258 	fs_info->pending_quota_state = 0;
2259 	fs_info->qgroup_ulist = NULL;
2260 	mutex_init(&fs_info->qgroup_rescan_lock);
2261 }
2262 
2263 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2264 		struct btrfs_fs_devices *fs_devices)
2265 {
2266 	int max_active = fs_info->thread_pool_size;
2267 	unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2268 
2269 	fs_info->workers =
2270 		btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2271 				      max_active, 16);
2272 
2273 	fs_info->delalloc_workers =
2274 		btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
2275 
2276 	fs_info->flush_workers =
2277 		btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0);
2278 
2279 	fs_info->caching_workers =
2280 		btrfs_alloc_workqueue("cache", flags, max_active, 0);
2281 
2282 	/*
2283 	 * a higher idle thresh on the submit workers makes it much more
2284 	 * likely that bios will be send down in a sane order to the
2285 	 * devices
2286 	 */
2287 	fs_info->submit_workers =
2288 		btrfs_alloc_workqueue("submit", flags,
2289 				      min_t(u64, fs_devices->num_devices,
2290 					    max_active), 64);
2291 
2292 	fs_info->fixup_workers =
2293 		btrfs_alloc_workqueue("fixup", flags, 1, 0);
2294 
2295 	/*
2296 	 * endios are largely parallel and should have a very
2297 	 * low idle thresh
2298 	 */
2299 	fs_info->endio_workers =
2300 		btrfs_alloc_workqueue("endio", flags, max_active, 4);
2301 	fs_info->endio_meta_workers =
2302 		btrfs_alloc_workqueue("endio-meta", flags, max_active, 4);
2303 	fs_info->endio_meta_write_workers =
2304 		btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2);
2305 	fs_info->endio_raid56_workers =
2306 		btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4);
2307 	fs_info->endio_repair_workers =
2308 		btrfs_alloc_workqueue("endio-repair", flags, 1, 0);
2309 	fs_info->rmw_workers =
2310 		btrfs_alloc_workqueue("rmw", flags, max_active, 2);
2311 	fs_info->endio_write_workers =
2312 		btrfs_alloc_workqueue("endio-write", flags, max_active, 2);
2313 	fs_info->endio_freespace_worker =
2314 		btrfs_alloc_workqueue("freespace-write", flags, max_active, 0);
2315 	fs_info->delayed_workers =
2316 		btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0);
2317 	fs_info->readahead_workers =
2318 		btrfs_alloc_workqueue("readahead", flags, max_active, 2);
2319 	fs_info->qgroup_rescan_workers =
2320 		btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0);
2321 	fs_info->extent_workers =
2322 		btrfs_alloc_workqueue("extent-refs", flags,
2323 				      min_t(u64, fs_devices->num_devices,
2324 					    max_active), 8);
2325 
2326 	if (!(fs_info->workers && fs_info->delalloc_workers &&
2327 	      fs_info->submit_workers && fs_info->flush_workers &&
2328 	      fs_info->endio_workers && fs_info->endio_meta_workers &&
2329 	      fs_info->endio_meta_write_workers &&
2330 	      fs_info->endio_repair_workers &&
2331 	      fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2332 	      fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2333 	      fs_info->caching_workers && fs_info->readahead_workers &&
2334 	      fs_info->fixup_workers && fs_info->delayed_workers &&
2335 	      fs_info->extent_workers &&
2336 	      fs_info->qgroup_rescan_workers)) {
2337 		return -ENOMEM;
2338 	}
2339 
2340 	return 0;
2341 }
2342 
2343 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2344 			    struct btrfs_fs_devices *fs_devices)
2345 {
2346 	int ret;
2347 	struct btrfs_root *tree_root = fs_info->tree_root;
2348 	struct btrfs_root *log_tree_root;
2349 	struct btrfs_super_block *disk_super = fs_info->super_copy;
2350 	u64 bytenr = btrfs_super_log_root(disk_super);
2351 
2352 	if (fs_devices->rw_devices == 0) {
2353 		btrfs_warn(fs_info, "log replay required on RO media");
2354 		return -EIO;
2355 	}
2356 
2357 	log_tree_root = btrfs_alloc_root(fs_info);
2358 	if (!log_tree_root)
2359 		return -ENOMEM;
2360 
2361 	__setup_root(tree_root->nodesize, tree_root->sectorsize,
2362 			tree_root->stripesize, log_tree_root, fs_info,
2363 			BTRFS_TREE_LOG_OBJECTID);
2364 
2365 	log_tree_root->node = read_tree_block(tree_root, bytenr,
2366 			fs_info->generation + 1);
2367 	if (IS_ERR(log_tree_root->node)) {
2368 		btrfs_warn(fs_info, "failed to read log tree");
2369 		ret = PTR_ERR(log_tree_root->node);
2370 		kfree(log_tree_root);
2371 		return ret;
2372 	} else if (!extent_buffer_uptodate(log_tree_root->node)) {
2373 		btrfs_err(fs_info, "failed to read log tree");
2374 		free_extent_buffer(log_tree_root->node);
2375 		kfree(log_tree_root);
2376 		return -EIO;
2377 	}
2378 	/* returns with log_tree_root freed on success */
2379 	ret = btrfs_recover_log_trees(log_tree_root);
2380 	if (ret) {
2381 		btrfs_std_error(tree_root->fs_info, ret,
2382 			    "Failed to recover log tree");
2383 		free_extent_buffer(log_tree_root->node);
2384 		kfree(log_tree_root);
2385 		return ret;
2386 	}
2387 
2388 	if (fs_info->sb->s_flags & MS_RDONLY) {
2389 		ret = btrfs_commit_super(tree_root);
2390 		if (ret)
2391 			return ret;
2392 	}
2393 
2394 	return 0;
2395 }
2396 
2397 static int btrfs_read_roots(struct btrfs_fs_info *fs_info,
2398 			    struct btrfs_root *tree_root)
2399 {
2400 	struct btrfs_root *root;
2401 	struct btrfs_key location;
2402 	int ret;
2403 
2404 	location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2405 	location.type = BTRFS_ROOT_ITEM_KEY;
2406 	location.offset = 0;
2407 
2408 	root = btrfs_read_tree_root(tree_root, &location);
2409 	if (IS_ERR(root))
2410 		return PTR_ERR(root);
2411 	set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2412 	fs_info->extent_root = root;
2413 
2414 	location.objectid = BTRFS_DEV_TREE_OBJECTID;
2415 	root = btrfs_read_tree_root(tree_root, &location);
2416 	if (IS_ERR(root))
2417 		return PTR_ERR(root);
2418 	set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2419 	fs_info->dev_root = root;
2420 	btrfs_init_devices_late(fs_info);
2421 
2422 	location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2423 	root = btrfs_read_tree_root(tree_root, &location);
2424 	if (IS_ERR(root))
2425 		return PTR_ERR(root);
2426 	set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2427 	fs_info->csum_root = root;
2428 
2429 	location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2430 	root = btrfs_read_tree_root(tree_root, &location);
2431 	if (!IS_ERR(root)) {
2432 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2433 		fs_info->quota_enabled = 1;
2434 		fs_info->pending_quota_state = 1;
2435 		fs_info->quota_root = root;
2436 	}
2437 
2438 	location.objectid = BTRFS_UUID_TREE_OBJECTID;
2439 	root = btrfs_read_tree_root(tree_root, &location);
2440 	if (IS_ERR(root)) {
2441 		ret = PTR_ERR(root);
2442 		if (ret != -ENOENT)
2443 			return ret;
2444 	} else {
2445 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2446 		fs_info->uuid_root = root;
2447 	}
2448 
2449 	return 0;
2450 }
2451 
2452 int open_ctree(struct super_block *sb,
2453 	       struct btrfs_fs_devices *fs_devices,
2454 	       char *options)
2455 {
2456 	u32 sectorsize;
2457 	u32 nodesize;
2458 	u32 stripesize;
2459 	u64 generation;
2460 	u64 features;
2461 	struct btrfs_key location;
2462 	struct buffer_head *bh;
2463 	struct btrfs_super_block *disk_super;
2464 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2465 	struct btrfs_root *tree_root;
2466 	struct btrfs_root *chunk_root;
2467 	int ret;
2468 	int err = -EINVAL;
2469 	int num_backups_tried = 0;
2470 	int backup_index = 0;
2471 	int max_active;
2472 
2473 	tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2474 	chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2475 	if (!tree_root || !chunk_root) {
2476 		err = -ENOMEM;
2477 		goto fail;
2478 	}
2479 
2480 	ret = init_srcu_struct(&fs_info->subvol_srcu);
2481 	if (ret) {
2482 		err = ret;
2483 		goto fail;
2484 	}
2485 
2486 	ret = setup_bdi(fs_info, &fs_info->bdi);
2487 	if (ret) {
2488 		err = ret;
2489 		goto fail_srcu;
2490 	}
2491 
2492 	ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2493 	if (ret) {
2494 		err = ret;
2495 		goto fail_bdi;
2496 	}
2497 	fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2498 					(1 + ilog2(nr_cpu_ids));
2499 
2500 	ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2501 	if (ret) {
2502 		err = ret;
2503 		goto fail_dirty_metadata_bytes;
2504 	}
2505 
2506 	ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2507 	if (ret) {
2508 		err = ret;
2509 		goto fail_delalloc_bytes;
2510 	}
2511 
2512 	fs_info->btree_inode = new_inode(sb);
2513 	if (!fs_info->btree_inode) {
2514 		err = -ENOMEM;
2515 		goto fail_bio_counter;
2516 	}
2517 
2518 	mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2519 
2520 	INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2521 	INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2522 	INIT_LIST_HEAD(&fs_info->trans_list);
2523 	INIT_LIST_HEAD(&fs_info->dead_roots);
2524 	INIT_LIST_HEAD(&fs_info->delayed_iputs);
2525 	INIT_LIST_HEAD(&fs_info->delalloc_roots);
2526 	INIT_LIST_HEAD(&fs_info->caching_block_groups);
2527 	spin_lock_init(&fs_info->delalloc_root_lock);
2528 	spin_lock_init(&fs_info->trans_lock);
2529 	spin_lock_init(&fs_info->fs_roots_radix_lock);
2530 	spin_lock_init(&fs_info->delayed_iput_lock);
2531 	spin_lock_init(&fs_info->defrag_inodes_lock);
2532 	spin_lock_init(&fs_info->free_chunk_lock);
2533 	spin_lock_init(&fs_info->tree_mod_seq_lock);
2534 	spin_lock_init(&fs_info->super_lock);
2535 	spin_lock_init(&fs_info->qgroup_op_lock);
2536 	spin_lock_init(&fs_info->buffer_lock);
2537 	spin_lock_init(&fs_info->unused_bgs_lock);
2538 	rwlock_init(&fs_info->tree_mod_log_lock);
2539 	mutex_init(&fs_info->unused_bg_unpin_mutex);
2540 	mutex_init(&fs_info->delete_unused_bgs_mutex);
2541 	mutex_init(&fs_info->reloc_mutex);
2542 	mutex_init(&fs_info->delalloc_root_mutex);
2543 	seqlock_init(&fs_info->profiles_lock);
2544 	init_rwsem(&fs_info->delayed_iput_sem);
2545 
2546 	INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2547 	INIT_LIST_HEAD(&fs_info->space_info);
2548 	INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2549 	INIT_LIST_HEAD(&fs_info->unused_bgs);
2550 	btrfs_mapping_init(&fs_info->mapping_tree);
2551 	btrfs_init_block_rsv(&fs_info->global_block_rsv,
2552 			     BTRFS_BLOCK_RSV_GLOBAL);
2553 	btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2554 			     BTRFS_BLOCK_RSV_DELALLOC);
2555 	btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2556 	btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2557 	btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2558 	btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2559 			     BTRFS_BLOCK_RSV_DELOPS);
2560 	atomic_set(&fs_info->nr_async_submits, 0);
2561 	atomic_set(&fs_info->async_delalloc_pages, 0);
2562 	atomic_set(&fs_info->async_submit_draining, 0);
2563 	atomic_set(&fs_info->nr_async_bios, 0);
2564 	atomic_set(&fs_info->defrag_running, 0);
2565 	atomic_set(&fs_info->qgroup_op_seq, 0);
2566 	atomic64_set(&fs_info->tree_mod_seq, 0);
2567 	fs_info->sb = sb;
2568 	fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2569 	fs_info->metadata_ratio = 0;
2570 	fs_info->defrag_inodes = RB_ROOT;
2571 	fs_info->free_chunk_space = 0;
2572 	fs_info->tree_mod_log = RB_ROOT;
2573 	fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2574 	fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2575 	/* readahead state */
2576 	INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2577 	spin_lock_init(&fs_info->reada_lock);
2578 
2579 	fs_info->thread_pool_size = min_t(unsigned long,
2580 					  num_online_cpus() + 2, 8);
2581 
2582 	INIT_LIST_HEAD(&fs_info->ordered_roots);
2583 	spin_lock_init(&fs_info->ordered_root_lock);
2584 	fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2585 					GFP_NOFS);
2586 	if (!fs_info->delayed_root) {
2587 		err = -ENOMEM;
2588 		goto fail_iput;
2589 	}
2590 	btrfs_init_delayed_root(fs_info->delayed_root);
2591 
2592 	btrfs_init_scrub(fs_info);
2593 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2594 	fs_info->check_integrity_print_mask = 0;
2595 #endif
2596 	btrfs_init_balance(fs_info);
2597 	btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2598 
2599 	sb->s_blocksize = 4096;
2600 	sb->s_blocksize_bits = blksize_bits(4096);
2601 	sb->s_bdi = &fs_info->bdi;
2602 
2603 	btrfs_init_btree_inode(fs_info, tree_root);
2604 
2605 	spin_lock_init(&fs_info->block_group_cache_lock);
2606 	fs_info->block_group_cache_tree = RB_ROOT;
2607 	fs_info->first_logical_byte = (u64)-1;
2608 
2609 	extent_io_tree_init(&fs_info->freed_extents[0],
2610 			     fs_info->btree_inode->i_mapping);
2611 	extent_io_tree_init(&fs_info->freed_extents[1],
2612 			     fs_info->btree_inode->i_mapping);
2613 	fs_info->pinned_extents = &fs_info->freed_extents[0];
2614 	fs_info->do_barriers = 1;
2615 
2616 
2617 	mutex_init(&fs_info->ordered_operations_mutex);
2618 	mutex_init(&fs_info->tree_log_mutex);
2619 	mutex_init(&fs_info->chunk_mutex);
2620 	mutex_init(&fs_info->transaction_kthread_mutex);
2621 	mutex_init(&fs_info->cleaner_mutex);
2622 	mutex_init(&fs_info->volume_mutex);
2623 	mutex_init(&fs_info->ro_block_group_mutex);
2624 	init_rwsem(&fs_info->commit_root_sem);
2625 	init_rwsem(&fs_info->cleanup_work_sem);
2626 	init_rwsem(&fs_info->subvol_sem);
2627 	sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2628 
2629 	btrfs_init_dev_replace_locks(fs_info);
2630 	btrfs_init_qgroup(fs_info);
2631 
2632 	btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2633 	btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2634 
2635 	init_waitqueue_head(&fs_info->transaction_throttle);
2636 	init_waitqueue_head(&fs_info->transaction_wait);
2637 	init_waitqueue_head(&fs_info->transaction_blocked_wait);
2638 	init_waitqueue_head(&fs_info->async_submit_wait);
2639 
2640 	INIT_LIST_HEAD(&fs_info->pinned_chunks);
2641 
2642 	ret = btrfs_alloc_stripe_hash_table(fs_info);
2643 	if (ret) {
2644 		err = ret;
2645 		goto fail_alloc;
2646 	}
2647 
2648 	__setup_root(4096, 4096, 4096, tree_root,
2649 		     fs_info, BTRFS_ROOT_TREE_OBJECTID);
2650 
2651 	invalidate_bdev(fs_devices->latest_bdev);
2652 
2653 	/*
2654 	 * Read super block and check the signature bytes only
2655 	 */
2656 	bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2657 	if (IS_ERR(bh)) {
2658 		err = PTR_ERR(bh);
2659 		goto fail_alloc;
2660 	}
2661 
2662 	/*
2663 	 * We want to check superblock checksum, the type is stored inside.
2664 	 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2665 	 */
2666 	if (btrfs_check_super_csum(bh->b_data)) {
2667 		printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
2668 		err = -EINVAL;
2669 		goto fail_alloc;
2670 	}
2671 
2672 	/*
2673 	 * super_copy is zeroed at allocation time and we never touch the
2674 	 * following bytes up to INFO_SIZE, the checksum is calculated from
2675 	 * the whole block of INFO_SIZE
2676 	 */
2677 	memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2678 	memcpy(fs_info->super_for_commit, fs_info->super_copy,
2679 	       sizeof(*fs_info->super_for_commit));
2680 	brelse(bh);
2681 
2682 	memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2683 
2684 	ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2685 	if (ret) {
2686 		printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
2687 		err = -EINVAL;
2688 		goto fail_alloc;
2689 	}
2690 
2691 	disk_super = fs_info->super_copy;
2692 	if (!btrfs_super_root(disk_super))
2693 		goto fail_alloc;
2694 
2695 	/* check FS state, whether FS is broken. */
2696 	if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2697 		set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2698 
2699 	/*
2700 	 * run through our array of backup supers and setup
2701 	 * our ring pointer to the oldest one
2702 	 */
2703 	generation = btrfs_super_generation(disk_super);
2704 	find_oldest_super_backup(fs_info, generation);
2705 
2706 	/*
2707 	 * In the long term, we'll store the compression type in the super
2708 	 * block, and it'll be used for per file compression control.
2709 	 */
2710 	fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2711 
2712 	ret = btrfs_parse_options(tree_root, options);
2713 	if (ret) {
2714 		err = ret;
2715 		goto fail_alloc;
2716 	}
2717 
2718 	features = btrfs_super_incompat_flags(disk_super) &
2719 		~BTRFS_FEATURE_INCOMPAT_SUPP;
2720 	if (features) {
2721 		printk(KERN_ERR "BTRFS: couldn't mount because of "
2722 		       "unsupported optional features (%Lx).\n",
2723 		       features);
2724 		err = -EINVAL;
2725 		goto fail_alloc;
2726 	}
2727 
2728 	/*
2729 	 * Leafsize and nodesize were always equal, this is only a sanity check.
2730 	 */
2731 	if (le32_to_cpu(disk_super->__unused_leafsize) !=
2732 	    btrfs_super_nodesize(disk_super)) {
2733 		printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2734 		       "blocksizes don't match.  node %d leaf %d\n",
2735 		       btrfs_super_nodesize(disk_super),
2736 		       le32_to_cpu(disk_super->__unused_leafsize));
2737 		err = -EINVAL;
2738 		goto fail_alloc;
2739 	}
2740 	if (btrfs_super_nodesize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2741 		printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2742 		       "blocksize (%d) was too large\n",
2743 		       btrfs_super_nodesize(disk_super));
2744 		err = -EINVAL;
2745 		goto fail_alloc;
2746 	}
2747 
2748 	features = btrfs_super_incompat_flags(disk_super);
2749 	features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2750 	if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2751 		features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2752 
2753 	if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2754 		printk(KERN_INFO "BTRFS: has skinny extents\n");
2755 
2756 	/*
2757 	 * flag our filesystem as having big metadata blocks if
2758 	 * they are bigger than the page size
2759 	 */
2760 	if (btrfs_super_nodesize(disk_super) > PAGE_CACHE_SIZE) {
2761 		if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2762 			printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n");
2763 		features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2764 	}
2765 
2766 	nodesize = btrfs_super_nodesize(disk_super);
2767 	sectorsize = btrfs_super_sectorsize(disk_super);
2768 	stripesize = btrfs_super_stripesize(disk_super);
2769 	fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2770 	fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2771 
2772 	/*
2773 	 * mixed block groups end up with duplicate but slightly offset
2774 	 * extent buffers for the same range.  It leads to corruptions
2775 	 */
2776 	if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2777 	    (sectorsize != nodesize)) {
2778 		printk(KERN_ERR "BTRFS: unequal leaf/node/sector sizes "
2779 				"are not allowed for mixed block groups on %s\n",
2780 				sb->s_id);
2781 		goto fail_alloc;
2782 	}
2783 
2784 	/*
2785 	 * Needn't use the lock because there is no other task which will
2786 	 * update the flag.
2787 	 */
2788 	btrfs_set_super_incompat_flags(disk_super, features);
2789 
2790 	features = btrfs_super_compat_ro_flags(disk_super) &
2791 		~BTRFS_FEATURE_COMPAT_RO_SUPP;
2792 	if (!(sb->s_flags & MS_RDONLY) && features) {
2793 		printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2794 		       "unsupported option features (%Lx).\n",
2795 		       features);
2796 		err = -EINVAL;
2797 		goto fail_alloc;
2798 	}
2799 
2800 	max_active = fs_info->thread_pool_size;
2801 
2802 	ret = btrfs_init_workqueues(fs_info, fs_devices);
2803 	if (ret) {
2804 		err = ret;
2805 		goto fail_sb_buffer;
2806 	}
2807 
2808 	fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2809 	fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2810 				    4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2811 
2812 	tree_root->nodesize = nodesize;
2813 	tree_root->sectorsize = sectorsize;
2814 	tree_root->stripesize = stripesize;
2815 
2816 	sb->s_blocksize = sectorsize;
2817 	sb->s_blocksize_bits = blksize_bits(sectorsize);
2818 
2819 	if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2820 		printk(KERN_ERR "BTRFS: valid FS not found on %s\n", sb->s_id);
2821 		goto fail_sb_buffer;
2822 	}
2823 
2824 	if (sectorsize != PAGE_SIZE) {
2825 		printk(KERN_ERR "BTRFS: incompatible sector size (%lu) "
2826 		       "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2827 		goto fail_sb_buffer;
2828 	}
2829 
2830 	mutex_lock(&fs_info->chunk_mutex);
2831 	ret = btrfs_read_sys_array(tree_root);
2832 	mutex_unlock(&fs_info->chunk_mutex);
2833 	if (ret) {
2834 		printk(KERN_ERR "BTRFS: failed to read the system "
2835 		       "array on %s\n", sb->s_id);
2836 		goto fail_sb_buffer;
2837 	}
2838 
2839 	generation = btrfs_super_chunk_root_generation(disk_super);
2840 
2841 	__setup_root(nodesize, sectorsize, stripesize, chunk_root,
2842 		     fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2843 
2844 	chunk_root->node = read_tree_block(chunk_root,
2845 					   btrfs_super_chunk_root(disk_super),
2846 					   generation);
2847 	if (IS_ERR(chunk_root->node) ||
2848 	    !extent_buffer_uptodate(chunk_root->node)) {
2849 		printk(KERN_ERR "BTRFS: failed to read chunk root on %s\n",
2850 		       sb->s_id);
2851 		if (!IS_ERR(chunk_root->node))
2852 			free_extent_buffer(chunk_root->node);
2853 		chunk_root->node = NULL;
2854 		goto fail_tree_roots;
2855 	}
2856 	btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2857 	chunk_root->commit_root = btrfs_root_node(chunk_root);
2858 
2859 	read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2860 	   btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2861 
2862 	ret = btrfs_read_chunk_tree(chunk_root);
2863 	if (ret) {
2864 		printk(KERN_ERR "BTRFS: failed to read chunk tree on %s\n",
2865 		       sb->s_id);
2866 		goto fail_tree_roots;
2867 	}
2868 
2869 	/*
2870 	 * keep the device that is marked to be the target device for the
2871 	 * dev_replace procedure
2872 	 */
2873 	btrfs_close_extra_devices(fs_devices, 0);
2874 
2875 	if (!fs_devices->latest_bdev) {
2876 		printk(KERN_ERR "BTRFS: failed to read devices on %s\n",
2877 		       sb->s_id);
2878 		goto fail_tree_roots;
2879 	}
2880 
2881 retry_root_backup:
2882 	generation = btrfs_super_generation(disk_super);
2883 
2884 	tree_root->node = read_tree_block(tree_root,
2885 					  btrfs_super_root(disk_super),
2886 					  generation);
2887 	if (IS_ERR(tree_root->node) ||
2888 	    !extent_buffer_uptodate(tree_root->node)) {
2889 		printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
2890 		       sb->s_id);
2891 		if (!IS_ERR(tree_root->node))
2892 			free_extent_buffer(tree_root->node);
2893 		tree_root->node = NULL;
2894 		goto recovery_tree_root;
2895 	}
2896 
2897 	btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2898 	tree_root->commit_root = btrfs_root_node(tree_root);
2899 	btrfs_set_root_refs(&tree_root->root_item, 1);
2900 
2901 	ret = btrfs_read_roots(fs_info, tree_root);
2902 	if (ret)
2903 		goto recovery_tree_root;
2904 
2905 	fs_info->generation = generation;
2906 	fs_info->last_trans_committed = generation;
2907 
2908 	ret = btrfs_recover_balance(fs_info);
2909 	if (ret) {
2910 		printk(KERN_ERR "BTRFS: failed to recover balance\n");
2911 		goto fail_block_groups;
2912 	}
2913 
2914 	ret = btrfs_init_dev_stats(fs_info);
2915 	if (ret) {
2916 		printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n",
2917 		       ret);
2918 		goto fail_block_groups;
2919 	}
2920 
2921 	ret = btrfs_init_dev_replace(fs_info);
2922 	if (ret) {
2923 		pr_err("BTRFS: failed to init dev_replace: %d\n", ret);
2924 		goto fail_block_groups;
2925 	}
2926 
2927 	btrfs_close_extra_devices(fs_devices, 1);
2928 
2929 	ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
2930 	if (ret) {
2931 		pr_err("BTRFS: failed to init sysfs fsid interface: %d\n", ret);
2932 		goto fail_block_groups;
2933 	}
2934 
2935 	ret = btrfs_sysfs_add_device(fs_devices);
2936 	if (ret) {
2937 		pr_err("BTRFS: failed to init sysfs device interface: %d\n", ret);
2938 		goto fail_fsdev_sysfs;
2939 	}
2940 
2941 	ret = btrfs_sysfs_add_mounted(fs_info);
2942 	if (ret) {
2943 		pr_err("BTRFS: failed to init sysfs interface: %d\n", ret);
2944 		goto fail_fsdev_sysfs;
2945 	}
2946 
2947 	ret = btrfs_init_space_info(fs_info);
2948 	if (ret) {
2949 		printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret);
2950 		goto fail_sysfs;
2951 	}
2952 
2953 	ret = btrfs_read_block_groups(fs_info->extent_root);
2954 	if (ret) {
2955 		printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret);
2956 		goto fail_sysfs;
2957 	}
2958 	fs_info->num_tolerated_disk_barrier_failures =
2959 		btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2960 	if (fs_info->fs_devices->missing_devices >
2961 	     fs_info->num_tolerated_disk_barrier_failures &&
2962 	    !(sb->s_flags & MS_RDONLY)) {
2963 		pr_warn("BTRFS: missing devices(%llu) exceeds the limit(%d), writeable mount is not allowed\n",
2964 			fs_info->fs_devices->missing_devices,
2965 			fs_info->num_tolerated_disk_barrier_failures);
2966 		goto fail_sysfs;
2967 	}
2968 
2969 	fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2970 					       "btrfs-cleaner");
2971 	if (IS_ERR(fs_info->cleaner_kthread))
2972 		goto fail_sysfs;
2973 
2974 	fs_info->transaction_kthread = kthread_run(transaction_kthread,
2975 						   tree_root,
2976 						   "btrfs-transaction");
2977 	if (IS_ERR(fs_info->transaction_kthread))
2978 		goto fail_cleaner;
2979 
2980 	if (!btrfs_test_opt(tree_root, SSD) &&
2981 	    !btrfs_test_opt(tree_root, NOSSD) &&
2982 	    !fs_info->fs_devices->rotating) {
2983 		printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD "
2984 		       "mode\n");
2985 		btrfs_set_opt(fs_info->mount_opt, SSD);
2986 	}
2987 
2988 	/*
2989 	 * Mount does not set all options immediatelly, we can do it now and do
2990 	 * not have to wait for transaction commit
2991 	 */
2992 	btrfs_apply_pending_changes(fs_info);
2993 
2994 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2995 	if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2996 		ret = btrfsic_mount(tree_root, fs_devices,
2997 				    btrfs_test_opt(tree_root,
2998 					CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2999 				    1 : 0,
3000 				    fs_info->check_integrity_print_mask);
3001 		if (ret)
3002 			printk(KERN_WARNING "BTRFS: failed to initialize"
3003 			       " integrity check module %s\n", sb->s_id);
3004 	}
3005 #endif
3006 	ret = btrfs_read_qgroup_config(fs_info);
3007 	if (ret)
3008 		goto fail_trans_kthread;
3009 
3010 	/* do not make disk changes in broken FS */
3011 	if (btrfs_super_log_root(disk_super) != 0) {
3012 		ret = btrfs_replay_log(fs_info, fs_devices);
3013 		if (ret) {
3014 			err = ret;
3015 			goto fail_qgroup;
3016 		}
3017 	}
3018 
3019 	ret = btrfs_find_orphan_roots(tree_root);
3020 	if (ret)
3021 		goto fail_qgroup;
3022 
3023 	if (!(sb->s_flags & MS_RDONLY)) {
3024 		ret = btrfs_cleanup_fs_roots(fs_info);
3025 		if (ret)
3026 			goto fail_qgroup;
3027 
3028 		mutex_lock(&fs_info->cleaner_mutex);
3029 		ret = btrfs_recover_relocation(tree_root);
3030 		mutex_unlock(&fs_info->cleaner_mutex);
3031 		if (ret < 0) {
3032 			printk(KERN_WARNING
3033 			       "BTRFS: failed to recover relocation\n");
3034 			err = -EINVAL;
3035 			goto fail_qgroup;
3036 		}
3037 	}
3038 
3039 	location.objectid = BTRFS_FS_TREE_OBJECTID;
3040 	location.type = BTRFS_ROOT_ITEM_KEY;
3041 	location.offset = 0;
3042 
3043 	fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3044 	if (IS_ERR(fs_info->fs_root)) {
3045 		err = PTR_ERR(fs_info->fs_root);
3046 		goto fail_qgroup;
3047 	}
3048 
3049 	if (sb->s_flags & MS_RDONLY)
3050 		return 0;
3051 
3052 	down_read(&fs_info->cleanup_work_sem);
3053 	if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3054 	    (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3055 		up_read(&fs_info->cleanup_work_sem);
3056 		close_ctree(tree_root);
3057 		return ret;
3058 	}
3059 	up_read(&fs_info->cleanup_work_sem);
3060 
3061 	ret = btrfs_resume_balance_async(fs_info);
3062 	if (ret) {
3063 		printk(KERN_WARNING "BTRFS: failed to resume balance\n");
3064 		close_ctree(tree_root);
3065 		return ret;
3066 	}
3067 
3068 	ret = btrfs_resume_dev_replace_async(fs_info);
3069 	if (ret) {
3070 		pr_warn("BTRFS: failed to resume dev_replace\n");
3071 		close_ctree(tree_root);
3072 		return ret;
3073 	}
3074 
3075 	btrfs_qgroup_rescan_resume(fs_info);
3076 
3077 	if (!fs_info->uuid_root) {
3078 		pr_info("BTRFS: creating UUID tree\n");
3079 		ret = btrfs_create_uuid_tree(fs_info);
3080 		if (ret) {
3081 			pr_warn("BTRFS: failed to create the UUID tree %d\n",
3082 				ret);
3083 			close_ctree(tree_root);
3084 			return ret;
3085 		}
3086 	} else if (btrfs_test_opt(tree_root, RESCAN_UUID_TREE) ||
3087 		   fs_info->generation !=
3088 				btrfs_super_uuid_tree_generation(disk_super)) {
3089 		pr_info("BTRFS: checking UUID tree\n");
3090 		ret = btrfs_check_uuid_tree(fs_info);
3091 		if (ret) {
3092 			pr_warn("BTRFS: failed to check the UUID tree %d\n",
3093 				ret);
3094 			close_ctree(tree_root);
3095 			return ret;
3096 		}
3097 	} else {
3098 		fs_info->update_uuid_tree_gen = 1;
3099 	}
3100 
3101 	fs_info->open = 1;
3102 
3103 	return 0;
3104 
3105 fail_qgroup:
3106 	btrfs_free_qgroup_config(fs_info);
3107 fail_trans_kthread:
3108 	kthread_stop(fs_info->transaction_kthread);
3109 	btrfs_cleanup_transaction(fs_info->tree_root);
3110 	btrfs_free_fs_roots(fs_info);
3111 fail_cleaner:
3112 	kthread_stop(fs_info->cleaner_kthread);
3113 
3114 	/*
3115 	 * make sure we're done with the btree inode before we stop our
3116 	 * kthreads
3117 	 */
3118 	filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3119 
3120 fail_sysfs:
3121 	btrfs_sysfs_remove_mounted(fs_info);
3122 
3123 fail_fsdev_sysfs:
3124 	btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3125 
3126 fail_block_groups:
3127 	btrfs_put_block_group_cache(fs_info);
3128 	btrfs_free_block_groups(fs_info);
3129 
3130 fail_tree_roots:
3131 	free_root_pointers(fs_info, 1);
3132 	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3133 
3134 fail_sb_buffer:
3135 	btrfs_stop_all_workers(fs_info);
3136 fail_alloc:
3137 fail_iput:
3138 	btrfs_mapping_tree_free(&fs_info->mapping_tree);
3139 
3140 	iput(fs_info->btree_inode);
3141 fail_bio_counter:
3142 	percpu_counter_destroy(&fs_info->bio_counter);
3143 fail_delalloc_bytes:
3144 	percpu_counter_destroy(&fs_info->delalloc_bytes);
3145 fail_dirty_metadata_bytes:
3146 	percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3147 fail_bdi:
3148 	bdi_destroy(&fs_info->bdi);
3149 fail_srcu:
3150 	cleanup_srcu_struct(&fs_info->subvol_srcu);
3151 fail:
3152 	btrfs_free_stripe_hash_table(fs_info);
3153 	btrfs_close_devices(fs_info->fs_devices);
3154 	return err;
3155 
3156 recovery_tree_root:
3157 	if (!btrfs_test_opt(tree_root, RECOVERY))
3158 		goto fail_tree_roots;
3159 
3160 	free_root_pointers(fs_info, 0);
3161 
3162 	/* don't use the log in recovery mode, it won't be valid */
3163 	btrfs_set_super_log_root(disk_super, 0);
3164 
3165 	/* we can't trust the free space cache either */
3166 	btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3167 
3168 	ret = next_root_backup(fs_info, fs_info->super_copy,
3169 			       &num_backups_tried, &backup_index);
3170 	if (ret == -1)
3171 		goto fail_block_groups;
3172 	goto retry_root_backup;
3173 }
3174 
3175 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3176 {
3177 	if (uptodate) {
3178 		set_buffer_uptodate(bh);
3179 	} else {
3180 		struct btrfs_device *device = (struct btrfs_device *)
3181 			bh->b_private;
3182 
3183 		btrfs_warn_rl_in_rcu(device->dev_root->fs_info,
3184 				"lost page write due to IO error on %s",
3185 					  rcu_str_deref(device->name));
3186 		/* note, we dont' set_buffer_write_io_error because we have
3187 		 * our own ways of dealing with the IO errors
3188 		 */
3189 		clear_buffer_uptodate(bh);
3190 		btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3191 	}
3192 	unlock_buffer(bh);
3193 	put_bh(bh);
3194 }
3195 
3196 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3197 			struct buffer_head **bh_ret)
3198 {
3199 	struct buffer_head *bh;
3200 	struct btrfs_super_block *super;
3201 	u64 bytenr;
3202 
3203 	bytenr = btrfs_sb_offset(copy_num);
3204 	if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3205 		return -EINVAL;
3206 
3207 	bh = __bread(bdev, bytenr / 4096, BTRFS_SUPER_INFO_SIZE);
3208 	/*
3209 	 * If we fail to read from the underlying devices, as of now
3210 	 * the best option we have is to mark it EIO.
3211 	 */
3212 	if (!bh)
3213 		return -EIO;
3214 
3215 	super = (struct btrfs_super_block *)bh->b_data;
3216 	if (btrfs_super_bytenr(super) != bytenr ||
3217 		    btrfs_super_magic(super) != BTRFS_MAGIC) {
3218 		brelse(bh);
3219 		return -EINVAL;
3220 	}
3221 
3222 	*bh_ret = bh;
3223 	return 0;
3224 }
3225 
3226 
3227 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3228 {
3229 	struct buffer_head *bh;
3230 	struct buffer_head *latest = NULL;
3231 	struct btrfs_super_block *super;
3232 	int i;
3233 	u64 transid = 0;
3234 	int ret = -EINVAL;
3235 
3236 	/* we would like to check all the supers, but that would make
3237 	 * a btrfs mount succeed after a mkfs from a different FS.
3238 	 * So, we need to add a special mount option to scan for
3239 	 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3240 	 */
3241 	for (i = 0; i < 1; i++) {
3242 		ret = btrfs_read_dev_one_super(bdev, i, &bh);
3243 		if (ret)
3244 			continue;
3245 
3246 		super = (struct btrfs_super_block *)bh->b_data;
3247 
3248 		if (!latest || btrfs_super_generation(super) > transid) {
3249 			brelse(latest);
3250 			latest = bh;
3251 			transid = btrfs_super_generation(super);
3252 		} else {
3253 			brelse(bh);
3254 		}
3255 	}
3256 
3257 	if (!latest)
3258 		return ERR_PTR(ret);
3259 
3260 	return latest;
3261 }
3262 
3263 /*
3264  * this should be called twice, once with wait == 0 and
3265  * once with wait == 1.  When wait == 0 is done, all the buffer heads
3266  * we write are pinned.
3267  *
3268  * They are released when wait == 1 is done.
3269  * max_mirrors must be the same for both runs, and it indicates how
3270  * many supers on this one device should be written.
3271  *
3272  * max_mirrors == 0 means to write them all.
3273  */
3274 static int write_dev_supers(struct btrfs_device *device,
3275 			    struct btrfs_super_block *sb,
3276 			    int do_barriers, int wait, int max_mirrors)
3277 {
3278 	struct buffer_head *bh;
3279 	int i;
3280 	int ret;
3281 	int errors = 0;
3282 	u32 crc;
3283 	u64 bytenr;
3284 
3285 	if (max_mirrors == 0)
3286 		max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3287 
3288 	for (i = 0; i < max_mirrors; i++) {
3289 		bytenr = btrfs_sb_offset(i);
3290 		if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3291 		    device->commit_total_bytes)
3292 			break;
3293 
3294 		if (wait) {
3295 			bh = __find_get_block(device->bdev, bytenr / 4096,
3296 					      BTRFS_SUPER_INFO_SIZE);
3297 			if (!bh) {
3298 				errors++;
3299 				continue;
3300 			}
3301 			wait_on_buffer(bh);
3302 			if (!buffer_uptodate(bh))
3303 				errors++;
3304 
3305 			/* drop our reference */
3306 			brelse(bh);
3307 
3308 			/* drop the reference from the wait == 0 run */
3309 			brelse(bh);
3310 			continue;
3311 		} else {
3312 			btrfs_set_super_bytenr(sb, bytenr);
3313 
3314 			crc = ~(u32)0;
3315 			crc = btrfs_csum_data((char *)sb +
3316 					      BTRFS_CSUM_SIZE, crc,
3317 					      BTRFS_SUPER_INFO_SIZE -
3318 					      BTRFS_CSUM_SIZE);
3319 			btrfs_csum_final(crc, sb->csum);
3320 
3321 			/*
3322 			 * one reference for us, and we leave it for the
3323 			 * caller
3324 			 */
3325 			bh = __getblk(device->bdev, bytenr / 4096,
3326 				      BTRFS_SUPER_INFO_SIZE);
3327 			if (!bh) {
3328 				btrfs_err(device->dev_root->fs_info,
3329 				    "couldn't get super buffer head for bytenr %llu",
3330 				    bytenr);
3331 				errors++;
3332 				continue;
3333 			}
3334 
3335 			memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3336 
3337 			/* one reference for submit_bh */
3338 			get_bh(bh);
3339 
3340 			set_buffer_uptodate(bh);
3341 			lock_buffer(bh);
3342 			bh->b_end_io = btrfs_end_buffer_write_sync;
3343 			bh->b_private = device;
3344 		}
3345 
3346 		/*
3347 		 * we fua the first super.  The others we allow
3348 		 * to go down lazy.
3349 		 */
3350 		if (i == 0)
3351 			ret = btrfsic_submit_bh(WRITE_FUA, bh);
3352 		else
3353 			ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3354 		if (ret)
3355 			errors++;
3356 	}
3357 	return errors < i ? 0 : -1;
3358 }
3359 
3360 /*
3361  * endio for the write_dev_flush, this will wake anyone waiting
3362  * for the barrier when it is done
3363  */
3364 static void btrfs_end_empty_barrier(struct bio *bio)
3365 {
3366 	if (bio->bi_private)
3367 		complete(bio->bi_private);
3368 	bio_put(bio);
3369 }
3370 
3371 /*
3372  * trigger flushes for one the devices.  If you pass wait == 0, the flushes are
3373  * sent down.  With wait == 1, it waits for the previous flush.
3374  *
3375  * any device where the flush fails with eopnotsupp are flagged as not-barrier
3376  * capable
3377  */
3378 static int write_dev_flush(struct btrfs_device *device, int wait)
3379 {
3380 	struct bio *bio;
3381 	int ret = 0;
3382 
3383 	if (device->nobarriers)
3384 		return 0;
3385 
3386 	if (wait) {
3387 		bio = device->flush_bio;
3388 		if (!bio)
3389 			return 0;
3390 
3391 		wait_for_completion(&device->flush_wait);
3392 
3393 		if (bio->bi_error) {
3394 			ret = bio->bi_error;
3395 			btrfs_dev_stat_inc_and_print(device,
3396 				BTRFS_DEV_STAT_FLUSH_ERRS);
3397 		}
3398 
3399 		/* drop the reference from the wait == 0 run */
3400 		bio_put(bio);
3401 		device->flush_bio = NULL;
3402 
3403 		return ret;
3404 	}
3405 
3406 	/*
3407 	 * one reference for us, and we leave it for the
3408 	 * caller
3409 	 */
3410 	device->flush_bio = NULL;
3411 	bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3412 	if (!bio)
3413 		return -ENOMEM;
3414 
3415 	bio->bi_end_io = btrfs_end_empty_barrier;
3416 	bio->bi_bdev = device->bdev;
3417 	init_completion(&device->flush_wait);
3418 	bio->bi_private = &device->flush_wait;
3419 	device->flush_bio = bio;
3420 
3421 	bio_get(bio);
3422 	btrfsic_submit_bio(WRITE_FLUSH, bio);
3423 
3424 	return 0;
3425 }
3426 
3427 /*
3428  * send an empty flush down to each device in parallel,
3429  * then wait for them
3430  */
3431 static int barrier_all_devices(struct btrfs_fs_info *info)
3432 {
3433 	struct list_head *head;
3434 	struct btrfs_device *dev;
3435 	int errors_send = 0;
3436 	int errors_wait = 0;
3437 	int ret;
3438 
3439 	/* send down all the barriers */
3440 	head = &info->fs_devices->devices;
3441 	list_for_each_entry_rcu(dev, head, dev_list) {
3442 		if (dev->missing)
3443 			continue;
3444 		if (!dev->bdev) {
3445 			errors_send++;
3446 			continue;
3447 		}
3448 		if (!dev->in_fs_metadata || !dev->writeable)
3449 			continue;
3450 
3451 		ret = write_dev_flush(dev, 0);
3452 		if (ret)
3453 			errors_send++;
3454 	}
3455 
3456 	/* wait for all the barriers */
3457 	list_for_each_entry_rcu(dev, head, dev_list) {
3458 		if (dev->missing)
3459 			continue;
3460 		if (!dev->bdev) {
3461 			errors_wait++;
3462 			continue;
3463 		}
3464 		if (!dev->in_fs_metadata || !dev->writeable)
3465 			continue;
3466 
3467 		ret = write_dev_flush(dev, 1);
3468 		if (ret)
3469 			errors_wait++;
3470 	}
3471 	if (errors_send > info->num_tolerated_disk_barrier_failures ||
3472 	    errors_wait > info->num_tolerated_disk_barrier_failures)
3473 		return -EIO;
3474 	return 0;
3475 }
3476 
3477 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3478 {
3479 	int raid_type;
3480 	int min_tolerated = INT_MAX;
3481 
3482 	if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3483 	    (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3484 		min_tolerated = min(min_tolerated,
3485 				    btrfs_raid_array[BTRFS_RAID_SINGLE].
3486 				    tolerated_failures);
3487 
3488 	for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3489 		if (raid_type == BTRFS_RAID_SINGLE)
3490 			continue;
3491 		if (!(flags & btrfs_raid_group[raid_type]))
3492 			continue;
3493 		min_tolerated = min(min_tolerated,
3494 				    btrfs_raid_array[raid_type].
3495 				    tolerated_failures);
3496 	}
3497 
3498 	if (min_tolerated == INT_MAX) {
3499 		pr_warn("BTRFS: unknown raid flag: %llu\n", flags);
3500 		min_tolerated = 0;
3501 	}
3502 
3503 	return min_tolerated;
3504 }
3505 
3506 int btrfs_calc_num_tolerated_disk_barrier_failures(
3507 	struct btrfs_fs_info *fs_info)
3508 {
3509 	struct btrfs_ioctl_space_info space;
3510 	struct btrfs_space_info *sinfo;
3511 	u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3512 		       BTRFS_BLOCK_GROUP_SYSTEM,
3513 		       BTRFS_BLOCK_GROUP_METADATA,
3514 		       BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3515 	int i;
3516 	int c;
3517 	int num_tolerated_disk_barrier_failures =
3518 		(int)fs_info->fs_devices->num_devices;
3519 
3520 	for (i = 0; i < ARRAY_SIZE(types); i++) {
3521 		struct btrfs_space_info *tmp;
3522 
3523 		sinfo = NULL;
3524 		rcu_read_lock();
3525 		list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3526 			if (tmp->flags == types[i]) {
3527 				sinfo = tmp;
3528 				break;
3529 			}
3530 		}
3531 		rcu_read_unlock();
3532 
3533 		if (!sinfo)
3534 			continue;
3535 
3536 		down_read(&sinfo->groups_sem);
3537 		for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3538 			u64 flags;
3539 
3540 			if (list_empty(&sinfo->block_groups[c]))
3541 				continue;
3542 
3543 			btrfs_get_block_group_info(&sinfo->block_groups[c],
3544 						   &space);
3545 			if (space.total_bytes == 0 || space.used_bytes == 0)
3546 				continue;
3547 			flags = space.flags;
3548 
3549 			num_tolerated_disk_barrier_failures = min(
3550 				num_tolerated_disk_barrier_failures,
3551 				btrfs_get_num_tolerated_disk_barrier_failures(
3552 					flags));
3553 		}
3554 		up_read(&sinfo->groups_sem);
3555 	}
3556 
3557 	return num_tolerated_disk_barrier_failures;
3558 }
3559 
3560 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3561 {
3562 	struct list_head *head;
3563 	struct btrfs_device *dev;
3564 	struct btrfs_super_block *sb;
3565 	struct btrfs_dev_item *dev_item;
3566 	int ret;
3567 	int do_barriers;
3568 	int max_errors;
3569 	int total_errors = 0;
3570 	u64 flags;
3571 
3572 	do_barriers = !btrfs_test_opt(root, NOBARRIER);
3573 	backup_super_roots(root->fs_info);
3574 
3575 	sb = root->fs_info->super_for_commit;
3576 	dev_item = &sb->dev_item;
3577 
3578 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3579 	head = &root->fs_info->fs_devices->devices;
3580 	max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3581 
3582 	if (do_barriers) {
3583 		ret = barrier_all_devices(root->fs_info);
3584 		if (ret) {
3585 			mutex_unlock(
3586 				&root->fs_info->fs_devices->device_list_mutex);
3587 			btrfs_std_error(root->fs_info, ret,
3588 				    "errors while submitting device barriers.");
3589 			return ret;
3590 		}
3591 	}
3592 
3593 	list_for_each_entry_rcu(dev, head, dev_list) {
3594 		if (!dev->bdev) {
3595 			total_errors++;
3596 			continue;
3597 		}
3598 		if (!dev->in_fs_metadata || !dev->writeable)
3599 			continue;
3600 
3601 		btrfs_set_stack_device_generation(dev_item, 0);
3602 		btrfs_set_stack_device_type(dev_item, dev->type);
3603 		btrfs_set_stack_device_id(dev_item, dev->devid);
3604 		btrfs_set_stack_device_total_bytes(dev_item,
3605 						   dev->commit_total_bytes);
3606 		btrfs_set_stack_device_bytes_used(dev_item,
3607 						  dev->commit_bytes_used);
3608 		btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3609 		btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3610 		btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3611 		memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3612 		memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3613 
3614 		flags = btrfs_super_flags(sb);
3615 		btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3616 
3617 		ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3618 		if (ret)
3619 			total_errors++;
3620 	}
3621 	if (total_errors > max_errors) {
3622 		btrfs_err(root->fs_info, "%d errors while writing supers",
3623 		       total_errors);
3624 		mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3625 
3626 		/* FUA is masked off if unsupported and can't be the reason */
3627 		btrfs_std_error(root->fs_info, -EIO,
3628 			    "%d errors while writing supers", total_errors);
3629 		return -EIO;
3630 	}
3631 
3632 	total_errors = 0;
3633 	list_for_each_entry_rcu(dev, head, dev_list) {
3634 		if (!dev->bdev)
3635 			continue;
3636 		if (!dev->in_fs_metadata || !dev->writeable)
3637 			continue;
3638 
3639 		ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3640 		if (ret)
3641 			total_errors++;
3642 	}
3643 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3644 	if (total_errors > max_errors) {
3645 		btrfs_std_error(root->fs_info, -EIO,
3646 			    "%d errors while writing supers", total_errors);
3647 		return -EIO;
3648 	}
3649 	return 0;
3650 }
3651 
3652 int write_ctree_super(struct btrfs_trans_handle *trans,
3653 		      struct btrfs_root *root, int max_mirrors)
3654 {
3655 	return write_all_supers(root, max_mirrors);
3656 }
3657 
3658 /* Drop a fs root from the radix tree and free it. */
3659 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3660 				  struct btrfs_root *root)
3661 {
3662 	spin_lock(&fs_info->fs_roots_radix_lock);
3663 	radix_tree_delete(&fs_info->fs_roots_radix,
3664 			  (unsigned long)root->root_key.objectid);
3665 	spin_unlock(&fs_info->fs_roots_radix_lock);
3666 
3667 	if (btrfs_root_refs(&root->root_item) == 0)
3668 		synchronize_srcu(&fs_info->subvol_srcu);
3669 
3670 	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3671 		btrfs_free_log(NULL, root);
3672 
3673 	if (root->free_ino_pinned)
3674 		__btrfs_remove_free_space_cache(root->free_ino_pinned);
3675 	if (root->free_ino_ctl)
3676 		__btrfs_remove_free_space_cache(root->free_ino_ctl);
3677 	free_fs_root(root);
3678 }
3679 
3680 static void free_fs_root(struct btrfs_root *root)
3681 {
3682 	iput(root->ino_cache_inode);
3683 	WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3684 	btrfs_free_block_rsv(root, root->orphan_block_rsv);
3685 	root->orphan_block_rsv = NULL;
3686 	if (root->anon_dev)
3687 		free_anon_bdev(root->anon_dev);
3688 	if (root->subv_writers)
3689 		btrfs_free_subvolume_writers(root->subv_writers);
3690 	free_extent_buffer(root->node);
3691 	free_extent_buffer(root->commit_root);
3692 	kfree(root->free_ino_ctl);
3693 	kfree(root->free_ino_pinned);
3694 	kfree(root->name);
3695 	btrfs_put_fs_root(root);
3696 }
3697 
3698 void btrfs_free_fs_root(struct btrfs_root *root)
3699 {
3700 	free_fs_root(root);
3701 }
3702 
3703 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3704 {
3705 	u64 root_objectid = 0;
3706 	struct btrfs_root *gang[8];
3707 	int i = 0;
3708 	int err = 0;
3709 	unsigned int ret = 0;
3710 	int index;
3711 
3712 	while (1) {
3713 		index = srcu_read_lock(&fs_info->subvol_srcu);
3714 		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3715 					     (void **)gang, root_objectid,
3716 					     ARRAY_SIZE(gang));
3717 		if (!ret) {
3718 			srcu_read_unlock(&fs_info->subvol_srcu, index);
3719 			break;
3720 		}
3721 		root_objectid = gang[ret - 1]->root_key.objectid + 1;
3722 
3723 		for (i = 0; i < ret; i++) {
3724 			/* Avoid to grab roots in dead_roots */
3725 			if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3726 				gang[i] = NULL;
3727 				continue;
3728 			}
3729 			/* grab all the search result for later use */
3730 			gang[i] = btrfs_grab_fs_root(gang[i]);
3731 		}
3732 		srcu_read_unlock(&fs_info->subvol_srcu, index);
3733 
3734 		for (i = 0; i < ret; i++) {
3735 			if (!gang[i])
3736 				continue;
3737 			root_objectid = gang[i]->root_key.objectid;
3738 			err = btrfs_orphan_cleanup(gang[i]);
3739 			if (err)
3740 				break;
3741 			btrfs_put_fs_root(gang[i]);
3742 		}
3743 		root_objectid++;
3744 	}
3745 
3746 	/* release the uncleaned roots due to error */
3747 	for (; i < ret; i++) {
3748 		if (gang[i])
3749 			btrfs_put_fs_root(gang[i]);
3750 	}
3751 	return err;
3752 }
3753 
3754 int btrfs_commit_super(struct btrfs_root *root)
3755 {
3756 	struct btrfs_trans_handle *trans;
3757 
3758 	mutex_lock(&root->fs_info->cleaner_mutex);
3759 	btrfs_run_delayed_iputs(root);
3760 	mutex_unlock(&root->fs_info->cleaner_mutex);
3761 	wake_up_process(root->fs_info->cleaner_kthread);
3762 
3763 	/* wait until ongoing cleanup work done */
3764 	down_write(&root->fs_info->cleanup_work_sem);
3765 	up_write(&root->fs_info->cleanup_work_sem);
3766 
3767 	trans = btrfs_join_transaction(root);
3768 	if (IS_ERR(trans))
3769 		return PTR_ERR(trans);
3770 	return btrfs_commit_transaction(trans, root);
3771 }
3772 
3773 void close_ctree(struct btrfs_root *root)
3774 {
3775 	struct btrfs_fs_info *fs_info = root->fs_info;
3776 	int ret;
3777 
3778 	fs_info->closing = 1;
3779 	smp_mb();
3780 
3781 	/* wait for the uuid_scan task to finish */
3782 	down(&fs_info->uuid_tree_rescan_sem);
3783 	/* avoid complains from lockdep et al., set sem back to initial state */
3784 	up(&fs_info->uuid_tree_rescan_sem);
3785 
3786 	/* pause restriper - we want to resume on mount */
3787 	btrfs_pause_balance(fs_info);
3788 
3789 	btrfs_dev_replace_suspend_for_unmount(fs_info);
3790 
3791 	btrfs_scrub_cancel(fs_info);
3792 
3793 	/* wait for any defraggers to finish */
3794 	wait_event(fs_info->transaction_wait,
3795 		   (atomic_read(&fs_info->defrag_running) == 0));
3796 
3797 	/* clear out the rbtree of defraggable inodes */
3798 	btrfs_cleanup_defrag_inodes(fs_info);
3799 
3800 	cancel_work_sync(&fs_info->async_reclaim_work);
3801 
3802 	if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3803 		/*
3804 		 * If the cleaner thread is stopped and there are
3805 		 * block groups queued for removal, the deletion will be
3806 		 * skipped when we quit the cleaner thread.
3807 		 */
3808 		btrfs_delete_unused_bgs(root->fs_info);
3809 
3810 		ret = btrfs_commit_super(root);
3811 		if (ret)
3812 			btrfs_err(fs_info, "commit super ret %d", ret);
3813 	}
3814 
3815 	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3816 		btrfs_error_commit_super(root);
3817 
3818 	kthread_stop(fs_info->transaction_kthread);
3819 	kthread_stop(fs_info->cleaner_kthread);
3820 
3821 	fs_info->closing = 2;
3822 	smp_mb();
3823 
3824 	btrfs_free_qgroup_config(fs_info);
3825 
3826 	if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3827 		btrfs_info(fs_info, "at unmount delalloc count %lld",
3828 		       percpu_counter_sum(&fs_info->delalloc_bytes));
3829 	}
3830 
3831 	btrfs_sysfs_remove_mounted(fs_info);
3832 	btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3833 
3834 	btrfs_free_fs_roots(fs_info);
3835 
3836 	btrfs_put_block_group_cache(fs_info);
3837 
3838 	btrfs_free_block_groups(fs_info);
3839 
3840 	/*
3841 	 * we must make sure there is not any read request to
3842 	 * submit after we stopping all workers.
3843 	 */
3844 	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3845 	btrfs_stop_all_workers(fs_info);
3846 
3847 	fs_info->open = 0;
3848 	free_root_pointers(fs_info, 1);
3849 
3850 	iput(fs_info->btree_inode);
3851 
3852 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3853 	if (btrfs_test_opt(root, CHECK_INTEGRITY))
3854 		btrfsic_unmount(root, fs_info->fs_devices);
3855 #endif
3856 
3857 	btrfs_close_devices(fs_info->fs_devices);
3858 	btrfs_mapping_tree_free(&fs_info->mapping_tree);
3859 
3860 	percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3861 	percpu_counter_destroy(&fs_info->delalloc_bytes);
3862 	percpu_counter_destroy(&fs_info->bio_counter);
3863 	bdi_destroy(&fs_info->bdi);
3864 	cleanup_srcu_struct(&fs_info->subvol_srcu);
3865 
3866 	btrfs_free_stripe_hash_table(fs_info);
3867 
3868 	__btrfs_free_block_rsv(root->orphan_block_rsv);
3869 	root->orphan_block_rsv = NULL;
3870 
3871 	lock_chunks(root);
3872 	while (!list_empty(&fs_info->pinned_chunks)) {
3873 		struct extent_map *em;
3874 
3875 		em = list_first_entry(&fs_info->pinned_chunks,
3876 				      struct extent_map, list);
3877 		list_del_init(&em->list);
3878 		free_extent_map(em);
3879 	}
3880 	unlock_chunks(root);
3881 }
3882 
3883 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3884 			  int atomic)
3885 {
3886 	int ret;
3887 	struct inode *btree_inode = buf->pages[0]->mapping->host;
3888 
3889 	ret = extent_buffer_uptodate(buf);
3890 	if (!ret)
3891 		return ret;
3892 
3893 	ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3894 				    parent_transid, atomic);
3895 	if (ret == -EAGAIN)
3896 		return ret;
3897 	return !ret;
3898 }
3899 
3900 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3901 {
3902 	return set_extent_buffer_uptodate(buf);
3903 }
3904 
3905 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3906 {
3907 	struct btrfs_root *root;
3908 	u64 transid = btrfs_header_generation(buf);
3909 	int was_dirty;
3910 
3911 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3912 	/*
3913 	 * This is a fast path so only do this check if we have sanity tests
3914 	 * enabled.  Normal people shouldn't be marking dummy buffers as dirty
3915 	 * outside of the sanity tests.
3916 	 */
3917 	if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3918 		return;
3919 #endif
3920 	root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3921 	btrfs_assert_tree_locked(buf);
3922 	if (transid != root->fs_info->generation)
3923 		WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3924 		       "found %llu running %llu\n",
3925 			buf->start, transid, root->fs_info->generation);
3926 	was_dirty = set_extent_buffer_dirty(buf);
3927 	if (!was_dirty)
3928 		__percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3929 				     buf->len,
3930 				     root->fs_info->dirty_metadata_batch);
3931 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3932 	if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
3933 		btrfs_print_leaf(root, buf);
3934 		ASSERT(0);
3935 	}
3936 #endif
3937 }
3938 
3939 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3940 					int flush_delayed)
3941 {
3942 	/*
3943 	 * looks as though older kernels can get into trouble with
3944 	 * this code, they end up stuck in balance_dirty_pages forever
3945 	 */
3946 	int ret;
3947 
3948 	if (current->flags & PF_MEMALLOC)
3949 		return;
3950 
3951 	if (flush_delayed)
3952 		btrfs_balance_delayed_items(root);
3953 
3954 	ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3955 				     BTRFS_DIRTY_METADATA_THRESH);
3956 	if (ret > 0) {
3957 		balance_dirty_pages_ratelimited(
3958 				   root->fs_info->btree_inode->i_mapping);
3959 	}
3960 	return;
3961 }
3962 
3963 void btrfs_btree_balance_dirty(struct btrfs_root *root)
3964 {
3965 	__btrfs_btree_balance_dirty(root, 1);
3966 }
3967 
3968 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3969 {
3970 	__btrfs_btree_balance_dirty(root, 0);
3971 }
3972 
3973 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3974 {
3975 	struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3976 	return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3977 }
3978 
3979 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3980 			      int read_only)
3981 {
3982 	struct btrfs_super_block *sb = fs_info->super_copy;
3983 	int ret = 0;
3984 
3985 	if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
3986 		printk(KERN_ERR "BTRFS: tree_root level too big: %d >= %d\n",
3987 				btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
3988 		ret = -EINVAL;
3989 	}
3990 	if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
3991 		printk(KERN_ERR "BTRFS: chunk_root level too big: %d >= %d\n",
3992 				btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
3993 		ret = -EINVAL;
3994 	}
3995 	if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
3996 		printk(KERN_ERR "BTRFS: log_root level too big: %d >= %d\n",
3997 				btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
3998 		ret = -EINVAL;
3999 	}
4000 
4001 	/*
4002 	 * The common minimum, we don't know if we can trust the nodesize/sectorsize
4003 	 * items yet, they'll be verified later. Issue just a warning.
4004 	 */
4005 	if (!IS_ALIGNED(btrfs_super_root(sb), 4096))
4006 		printk(KERN_WARNING "BTRFS: tree_root block unaligned: %llu\n",
4007 				btrfs_super_root(sb));
4008 	if (!IS_ALIGNED(btrfs_super_chunk_root(sb), 4096))
4009 		printk(KERN_WARNING "BTRFS: chunk_root block unaligned: %llu\n",
4010 				btrfs_super_chunk_root(sb));
4011 	if (!IS_ALIGNED(btrfs_super_log_root(sb), 4096))
4012 		printk(KERN_WARNING "BTRFS: log_root block unaligned: %llu\n",
4013 				btrfs_super_log_root(sb));
4014 
4015 	/*
4016 	 * Check the lower bound, the alignment and other constraints are
4017 	 * checked later.
4018 	 */
4019 	if (btrfs_super_nodesize(sb) < 4096) {
4020 		printk(KERN_ERR "BTRFS: nodesize too small: %u < 4096\n",
4021 				btrfs_super_nodesize(sb));
4022 		ret = -EINVAL;
4023 	}
4024 	if (btrfs_super_sectorsize(sb) < 4096) {
4025 		printk(KERN_ERR "BTRFS: sectorsize too small: %u < 4096\n",
4026 				btrfs_super_sectorsize(sb));
4027 		ret = -EINVAL;
4028 	}
4029 
4030 	if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) {
4031 		printk(KERN_ERR "BTRFS: dev_item UUID does not match fsid: %pU != %pU\n",
4032 				fs_info->fsid, sb->dev_item.fsid);
4033 		ret = -EINVAL;
4034 	}
4035 
4036 	/*
4037 	 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
4038 	 * done later
4039 	 */
4040 	if (btrfs_super_num_devices(sb) > (1UL << 31))
4041 		printk(KERN_WARNING "BTRFS: suspicious number of devices: %llu\n",
4042 				btrfs_super_num_devices(sb));
4043 	if (btrfs_super_num_devices(sb) == 0) {
4044 		printk(KERN_ERR "BTRFS: number of devices is 0\n");
4045 		ret = -EINVAL;
4046 	}
4047 
4048 	if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
4049 		printk(KERN_ERR "BTRFS: super offset mismatch %llu != %u\n",
4050 				btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
4051 		ret = -EINVAL;
4052 	}
4053 
4054 	/*
4055 	 * Obvious sys_chunk_array corruptions, it must hold at least one key
4056 	 * and one chunk
4057 	 */
4058 	if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4059 		printk(KERN_ERR "BTRFS: system chunk array too big %u > %u\n",
4060 				btrfs_super_sys_array_size(sb),
4061 				BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
4062 		ret = -EINVAL;
4063 	}
4064 	if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
4065 			+ sizeof(struct btrfs_chunk)) {
4066 		printk(KERN_ERR "BTRFS: system chunk array too small %u < %zu\n",
4067 				btrfs_super_sys_array_size(sb),
4068 				sizeof(struct btrfs_disk_key)
4069 				+ sizeof(struct btrfs_chunk));
4070 		ret = -EINVAL;
4071 	}
4072 
4073 	/*
4074 	 * The generation is a global counter, we'll trust it more than the others
4075 	 * but it's still possible that it's the one that's wrong.
4076 	 */
4077 	if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
4078 		printk(KERN_WARNING
4079 			"BTRFS: suspicious: generation < chunk_root_generation: %llu < %llu\n",
4080 			btrfs_super_generation(sb), btrfs_super_chunk_root_generation(sb));
4081 	if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
4082 	    && btrfs_super_cache_generation(sb) != (u64)-1)
4083 		printk(KERN_WARNING
4084 			"BTRFS: suspicious: generation < cache_generation: %llu < %llu\n",
4085 			btrfs_super_generation(sb), btrfs_super_cache_generation(sb));
4086 
4087 	return ret;
4088 }
4089 
4090 static void btrfs_error_commit_super(struct btrfs_root *root)
4091 {
4092 	mutex_lock(&root->fs_info->cleaner_mutex);
4093 	btrfs_run_delayed_iputs(root);
4094 	mutex_unlock(&root->fs_info->cleaner_mutex);
4095 
4096 	down_write(&root->fs_info->cleanup_work_sem);
4097 	up_write(&root->fs_info->cleanup_work_sem);
4098 
4099 	/* cleanup FS via transaction */
4100 	btrfs_cleanup_transaction(root);
4101 }
4102 
4103 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4104 {
4105 	struct btrfs_ordered_extent *ordered;
4106 
4107 	spin_lock(&root->ordered_extent_lock);
4108 	/*
4109 	 * This will just short circuit the ordered completion stuff which will
4110 	 * make sure the ordered extent gets properly cleaned up.
4111 	 */
4112 	list_for_each_entry(ordered, &root->ordered_extents,
4113 			    root_extent_list)
4114 		set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4115 	spin_unlock(&root->ordered_extent_lock);
4116 }
4117 
4118 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4119 {
4120 	struct btrfs_root *root;
4121 	struct list_head splice;
4122 
4123 	INIT_LIST_HEAD(&splice);
4124 
4125 	spin_lock(&fs_info->ordered_root_lock);
4126 	list_splice_init(&fs_info->ordered_roots, &splice);
4127 	while (!list_empty(&splice)) {
4128 		root = list_first_entry(&splice, struct btrfs_root,
4129 					ordered_root);
4130 		list_move_tail(&root->ordered_root,
4131 			       &fs_info->ordered_roots);
4132 
4133 		spin_unlock(&fs_info->ordered_root_lock);
4134 		btrfs_destroy_ordered_extents(root);
4135 
4136 		cond_resched();
4137 		spin_lock(&fs_info->ordered_root_lock);
4138 	}
4139 	spin_unlock(&fs_info->ordered_root_lock);
4140 }
4141 
4142 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4143 				      struct btrfs_root *root)
4144 {
4145 	struct rb_node *node;
4146 	struct btrfs_delayed_ref_root *delayed_refs;
4147 	struct btrfs_delayed_ref_node *ref;
4148 	int ret = 0;
4149 
4150 	delayed_refs = &trans->delayed_refs;
4151 
4152 	spin_lock(&delayed_refs->lock);
4153 	if (atomic_read(&delayed_refs->num_entries) == 0) {
4154 		spin_unlock(&delayed_refs->lock);
4155 		btrfs_info(root->fs_info, "delayed_refs has NO entry");
4156 		return ret;
4157 	}
4158 
4159 	while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4160 		struct btrfs_delayed_ref_head *head;
4161 		struct btrfs_delayed_ref_node *tmp;
4162 		bool pin_bytes = false;
4163 
4164 		head = rb_entry(node, struct btrfs_delayed_ref_head,
4165 				href_node);
4166 		if (!mutex_trylock(&head->mutex)) {
4167 			atomic_inc(&head->node.refs);
4168 			spin_unlock(&delayed_refs->lock);
4169 
4170 			mutex_lock(&head->mutex);
4171 			mutex_unlock(&head->mutex);
4172 			btrfs_put_delayed_ref(&head->node);
4173 			spin_lock(&delayed_refs->lock);
4174 			continue;
4175 		}
4176 		spin_lock(&head->lock);
4177 		list_for_each_entry_safe_reverse(ref, tmp, &head->ref_list,
4178 						 list) {
4179 			ref->in_tree = 0;
4180 			list_del(&ref->list);
4181 			atomic_dec(&delayed_refs->num_entries);
4182 			btrfs_put_delayed_ref(ref);
4183 		}
4184 		if (head->must_insert_reserved)
4185 			pin_bytes = true;
4186 		btrfs_free_delayed_extent_op(head->extent_op);
4187 		delayed_refs->num_heads--;
4188 		if (head->processing == 0)
4189 			delayed_refs->num_heads_ready--;
4190 		atomic_dec(&delayed_refs->num_entries);
4191 		head->node.in_tree = 0;
4192 		rb_erase(&head->href_node, &delayed_refs->href_root);
4193 		spin_unlock(&head->lock);
4194 		spin_unlock(&delayed_refs->lock);
4195 		mutex_unlock(&head->mutex);
4196 
4197 		if (pin_bytes)
4198 			btrfs_pin_extent(root, head->node.bytenr,
4199 					 head->node.num_bytes, 1);
4200 		btrfs_put_delayed_ref(&head->node);
4201 		cond_resched();
4202 		spin_lock(&delayed_refs->lock);
4203 	}
4204 
4205 	spin_unlock(&delayed_refs->lock);
4206 
4207 	return ret;
4208 }
4209 
4210 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4211 {
4212 	struct btrfs_inode *btrfs_inode;
4213 	struct list_head splice;
4214 
4215 	INIT_LIST_HEAD(&splice);
4216 
4217 	spin_lock(&root->delalloc_lock);
4218 	list_splice_init(&root->delalloc_inodes, &splice);
4219 
4220 	while (!list_empty(&splice)) {
4221 		btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4222 					       delalloc_inodes);
4223 
4224 		list_del_init(&btrfs_inode->delalloc_inodes);
4225 		clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4226 			  &btrfs_inode->runtime_flags);
4227 		spin_unlock(&root->delalloc_lock);
4228 
4229 		btrfs_invalidate_inodes(btrfs_inode->root);
4230 
4231 		spin_lock(&root->delalloc_lock);
4232 	}
4233 
4234 	spin_unlock(&root->delalloc_lock);
4235 }
4236 
4237 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4238 {
4239 	struct btrfs_root *root;
4240 	struct list_head splice;
4241 
4242 	INIT_LIST_HEAD(&splice);
4243 
4244 	spin_lock(&fs_info->delalloc_root_lock);
4245 	list_splice_init(&fs_info->delalloc_roots, &splice);
4246 	while (!list_empty(&splice)) {
4247 		root = list_first_entry(&splice, struct btrfs_root,
4248 					 delalloc_root);
4249 		list_del_init(&root->delalloc_root);
4250 		root = btrfs_grab_fs_root(root);
4251 		BUG_ON(!root);
4252 		spin_unlock(&fs_info->delalloc_root_lock);
4253 
4254 		btrfs_destroy_delalloc_inodes(root);
4255 		btrfs_put_fs_root(root);
4256 
4257 		spin_lock(&fs_info->delalloc_root_lock);
4258 	}
4259 	spin_unlock(&fs_info->delalloc_root_lock);
4260 }
4261 
4262 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
4263 					struct extent_io_tree *dirty_pages,
4264 					int mark)
4265 {
4266 	int ret;
4267 	struct extent_buffer *eb;
4268 	u64 start = 0;
4269 	u64 end;
4270 
4271 	while (1) {
4272 		ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4273 					    mark, NULL);
4274 		if (ret)
4275 			break;
4276 
4277 		clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
4278 		while (start <= end) {
4279 			eb = btrfs_find_tree_block(root->fs_info, start);
4280 			start += root->nodesize;
4281 			if (!eb)
4282 				continue;
4283 			wait_on_extent_buffer_writeback(eb);
4284 
4285 			if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4286 					       &eb->bflags))
4287 				clear_extent_buffer_dirty(eb);
4288 			free_extent_buffer_stale(eb);
4289 		}
4290 	}
4291 
4292 	return ret;
4293 }
4294 
4295 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4296 				       struct extent_io_tree *pinned_extents)
4297 {
4298 	struct extent_io_tree *unpin;
4299 	u64 start;
4300 	u64 end;
4301 	int ret;
4302 	bool loop = true;
4303 
4304 	unpin = pinned_extents;
4305 again:
4306 	while (1) {
4307 		ret = find_first_extent_bit(unpin, 0, &start, &end,
4308 					    EXTENT_DIRTY, NULL);
4309 		if (ret)
4310 			break;
4311 
4312 		clear_extent_dirty(unpin, start, end, GFP_NOFS);
4313 		btrfs_error_unpin_extent_range(root, start, end);
4314 		cond_resched();
4315 	}
4316 
4317 	if (loop) {
4318 		if (unpin == &root->fs_info->freed_extents[0])
4319 			unpin = &root->fs_info->freed_extents[1];
4320 		else
4321 			unpin = &root->fs_info->freed_extents[0];
4322 		loop = false;
4323 		goto again;
4324 	}
4325 
4326 	return 0;
4327 }
4328 
4329 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4330 				   struct btrfs_root *root)
4331 {
4332 	btrfs_destroy_delayed_refs(cur_trans, root);
4333 
4334 	cur_trans->state = TRANS_STATE_COMMIT_START;
4335 	wake_up(&root->fs_info->transaction_blocked_wait);
4336 
4337 	cur_trans->state = TRANS_STATE_UNBLOCKED;
4338 	wake_up(&root->fs_info->transaction_wait);
4339 
4340 	btrfs_destroy_delayed_inodes(root);
4341 	btrfs_assert_delayed_root_empty(root);
4342 
4343 	btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4344 				     EXTENT_DIRTY);
4345 	btrfs_destroy_pinned_extent(root,
4346 				    root->fs_info->pinned_extents);
4347 
4348 	cur_trans->state =TRANS_STATE_COMPLETED;
4349 	wake_up(&cur_trans->commit_wait);
4350 
4351 	/*
4352 	memset(cur_trans, 0, sizeof(*cur_trans));
4353 	kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4354 	*/
4355 }
4356 
4357 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4358 {
4359 	struct btrfs_transaction *t;
4360 
4361 	mutex_lock(&root->fs_info->transaction_kthread_mutex);
4362 
4363 	spin_lock(&root->fs_info->trans_lock);
4364 	while (!list_empty(&root->fs_info->trans_list)) {
4365 		t = list_first_entry(&root->fs_info->trans_list,
4366 				     struct btrfs_transaction, list);
4367 		if (t->state >= TRANS_STATE_COMMIT_START) {
4368 			atomic_inc(&t->use_count);
4369 			spin_unlock(&root->fs_info->trans_lock);
4370 			btrfs_wait_for_commit(root, t->transid);
4371 			btrfs_put_transaction(t);
4372 			spin_lock(&root->fs_info->trans_lock);
4373 			continue;
4374 		}
4375 		if (t == root->fs_info->running_transaction) {
4376 			t->state = TRANS_STATE_COMMIT_DOING;
4377 			spin_unlock(&root->fs_info->trans_lock);
4378 			/*
4379 			 * We wait for 0 num_writers since we don't hold a trans
4380 			 * handle open currently for this transaction.
4381 			 */
4382 			wait_event(t->writer_wait,
4383 				   atomic_read(&t->num_writers) == 0);
4384 		} else {
4385 			spin_unlock(&root->fs_info->trans_lock);
4386 		}
4387 		btrfs_cleanup_one_transaction(t, root);
4388 
4389 		spin_lock(&root->fs_info->trans_lock);
4390 		if (t == root->fs_info->running_transaction)
4391 			root->fs_info->running_transaction = NULL;
4392 		list_del_init(&t->list);
4393 		spin_unlock(&root->fs_info->trans_lock);
4394 
4395 		btrfs_put_transaction(t);
4396 		trace_btrfs_transaction_commit(root);
4397 		spin_lock(&root->fs_info->trans_lock);
4398 	}
4399 	spin_unlock(&root->fs_info->trans_lock);
4400 	btrfs_destroy_all_ordered_extents(root->fs_info);
4401 	btrfs_destroy_delayed_inodes(root);
4402 	btrfs_assert_delayed_root_empty(root);
4403 	btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4404 	btrfs_destroy_all_delalloc_inodes(root->fs_info);
4405 	mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4406 
4407 	return 0;
4408 }
4409 
4410 static const struct extent_io_ops btree_extent_io_ops = {
4411 	.readpage_end_io_hook = btree_readpage_end_io_hook,
4412 	.readpage_io_failed_hook = btree_io_failed_hook,
4413 	.submit_bio_hook = btree_submit_bio_hook,
4414 	/* note we're sharing with inode.c for the merge bio hook */
4415 	.merge_bio_hook = btrfs_merge_bio_hook,
4416 };
4417