xref: /openbmc/linux/fs/btrfs/disk-io.c (revision 161f4089)
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/crc32c.h>
30 #include <linux/slab.h>
31 #include <linux/migrate.h>
32 #include <linux/ratelimit.h>
33 #include <linux/uuid.h>
34 #include <linux/semaphore.h>
35 #include <asm/unaligned.h>
36 #include "compat.h"
37 #include "ctree.h"
38 #include "disk-io.h"
39 #include "transaction.h"
40 #include "btrfs_inode.h"
41 #include "volumes.h"
42 #include "print-tree.h"
43 #include "async-thread.h"
44 #include "locking.h"
45 #include "tree-log.h"
46 #include "free-space-cache.h"
47 #include "inode-map.h"
48 #include "check-integrity.h"
49 #include "rcu-string.h"
50 #include "dev-replace.h"
51 #include "raid56.h"
52 
53 #ifdef CONFIG_X86
54 #include <asm/cpufeature.h>
55 #endif
56 
57 static 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_operations(struct btrfs_transaction *t,
63 					     struct btrfs_root *root);
64 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
65 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
66 				      struct btrfs_root *root);
67 static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t);
68 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
69 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
70 					struct extent_io_tree *dirty_pages,
71 					int mark);
72 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
73 				       struct extent_io_tree *pinned_extents);
74 static int btrfs_cleanup_transaction(struct btrfs_root *root);
75 static void btrfs_error_commit_super(struct btrfs_root *root);
76 
77 /*
78  * end_io_wq structs are used to do processing in task context when an IO is
79  * complete.  This is used during reads to verify checksums, and it is used
80  * by writes to insert metadata for new file extents after IO is complete.
81  */
82 struct end_io_wq {
83 	struct bio *bio;
84 	bio_end_io_t *end_io;
85 	void *private;
86 	struct btrfs_fs_info *info;
87 	int error;
88 	int metadata;
89 	struct list_head list;
90 	struct btrfs_work work;
91 };
92 
93 /*
94  * async submit bios are used to offload expensive checksumming
95  * onto the worker threads.  They checksum file and metadata bios
96  * just before they are sent down the IO stack.
97  */
98 struct async_submit_bio {
99 	struct inode *inode;
100 	struct bio *bio;
101 	struct list_head list;
102 	extent_submit_bio_hook_t *submit_bio_start;
103 	extent_submit_bio_hook_t *submit_bio_done;
104 	int rw;
105 	int mirror_num;
106 	unsigned long bio_flags;
107 	/*
108 	 * bio_offset is optional, can be used if the pages in the bio
109 	 * can't tell us where in the file the bio should go
110 	 */
111 	u64 bio_offset;
112 	struct btrfs_work work;
113 	int error;
114 };
115 
116 /*
117  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
118  * eb, the lockdep key is determined by the btrfs_root it belongs to and
119  * the level the eb occupies in the tree.
120  *
121  * Different roots are used for different purposes and may nest inside each
122  * other and they require separate keysets.  As lockdep keys should be
123  * static, assign keysets according to the purpose of the root as indicated
124  * by btrfs_root->objectid.  This ensures that all special purpose roots
125  * have separate keysets.
126  *
127  * Lock-nesting across peer nodes is always done with the immediate parent
128  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
129  * subclass to avoid triggering lockdep warning in such cases.
130  *
131  * The key is set by the readpage_end_io_hook after the buffer has passed
132  * csum validation but before the pages are unlocked.  It is also set by
133  * btrfs_init_new_buffer on freshly allocated blocks.
134  *
135  * We also add a check to make sure the highest level of the tree is the
136  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
137  * needs update as well.
138  */
139 #ifdef CONFIG_DEBUG_LOCK_ALLOC
140 # if BTRFS_MAX_LEVEL != 8
141 #  error
142 # endif
143 
144 static struct btrfs_lockdep_keyset {
145 	u64			id;		/* root objectid */
146 	const char		*name_stem;	/* lock name stem */
147 	char			names[BTRFS_MAX_LEVEL + 1][20];
148 	struct lock_class_key	keys[BTRFS_MAX_LEVEL + 1];
149 } btrfs_lockdep_keysets[] = {
150 	{ .id = BTRFS_ROOT_TREE_OBJECTID,	.name_stem = "root"	},
151 	{ .id = BTRFS_EXTENT_TREE_OBJECTID,	.name_stem = "extent"	},
152 	{ .id = BTRFS_CHUNK_TREE_OBJECTID,	.name_stem = "chunk"	},
153 	{ .id = BTRFS_DEV_TREE_OBJECTID,	.name_stem = "dev"	},
154 	{ .id = BTRFS_FS_TREE_OBJECTID,		.name_stem = "fs"	},
155 	{ .id = BTRFS_CSUM_TREE_OBJECTID,	.name_stem = "csum"	},
156 	{ .id = BTRFS_QUOTA_TREE_OBJECTID,	.name_stem = "quota"	},
157 	{ .id = BTRFS_TREE_LOG_OBJECTID,	.name_stem = "log"	},
158 	{ .id = BTRFS_TREE_RELOC_OBJECTID,	.name_stem = "treloc"	},
159 	{ .id = BTRFS_DATA_RELOC_TREE_OBJECTID,	.name_stem = "dreloc"	},
160 	{ .id = BTRFS_UUID_TREE_OBJECTID,	.name_stem = "uuid"	},
161 	{ .id = 0,				.name_stem = "tree"	},
162 };
163 
164 void __init btrfs_init_lockdep(void)
165 {
166 	int i, j;
167 
168 	/* initialize lockdep class names */
169 	for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
170 		struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
171 
172 		for (j = 0; j < ARRAY_SIZE(ks->names); j++)
173 			snprintf(ks->names[j], sizeof(ks->names[j]),
174 				 "btrfs-%s-%02d", ks->name_stem, j);
175 	}
176 }
177 
178 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
179 				    int level)
180 {
181 	struct btrfs_lockdep_keyset *ks;
182 
183 	BUG_ON(level >= ARRAY_SIZE(ks->keys));
184 
185 	/* find the matching keyset, id 0 is the default entry */
186 	for (ks = btrfs_lockdep_keysets; ks->id; ks++)
187 		if (ks->id == objectid)
188 			break;
189 
190 	lockdep_set_class_and_name(&eb->lock,
191 				   &ks->keys[level], ks->names[level]);
192 }
193 
194 #endif
195 
196 /*
197  * extents on the btree inode are pretty simple, there's one extent
198  * that covers the entire device
199  */
200 static struct extent_map *btree_get_extent(struct inode *inode,
201 		struct page *page, size_t pg_offset, u64 start, u64 len,
202 		int create)
203 {
204 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
205 	struct extent_map *em;
206 	int ret;
207 
208 	read_lock(&em_tree->lock);
209 	em = lookup_extent_mapping(em_tree, start, len);
210 	if (em) {
211 		em->bdev =
212 			BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
213 		read_unlock(&em_tree->lock);
214 		goto out;
215 	}
216 	read_unlock(&em_tree->lock);
217 
218 	em = alloc_extent_map();
219 	if (!em) {
220 		em = ERR_PTR(-ENOMEM);
221 		goto out;
222 	}
223 	em->start = 0;
224 	em->len = (u64)-1;
225 	em->block_len = (u64)-1;
226 	em->block_start = 0;
227 	em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
228 
229 	write_lock(&em_tree->lock);
230 	ret = add_extent_mapping(em_tree, em, 0);
231 	if (ret == -EEXIST) {
232 		free_extent_map(em);
233 		em = lookup_extent_mapping(em_tree, start, len);
234 		if (!em)
235 			em = ERR_PTR(-EIO);
236 	} else if (ret) {
237 		free_extent_map(em);
238 		em = ERR_PTR(ret);
239 	}
240 	write_unlock(&em_tree->lock);
241 
242 out:
243 	return em;
244 }
245 
246 u32 btrfs_csum_data(char *data, u32 seed, size_t len)
247 {
248 	return crc32c(seed, data, len);
249 }
250 
251 void btrfs_csum_final(u32 crc, char *result)
252 {
253 	put_unaligned_le32(~crc, result);
254 }
255 
256 /*
257  * compute the csum for a btree block, and either verify it or write it
258  * into the csum field of the block.
259  */
260 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
261 			   int verify)
262 {
263 	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
264 	char *result = NULL;
265 	unsigned long len;
266 	unsigned long cur_len;
267 	unsigned long offset = BTRFS_CSUM_SIZE;
268 	char *kaddr;
269 	unsigned long map_start;
270 	unsigned long map_len;
271 	int err;
272 	u32 crc = ~(u32)0;
273 	unsigned long inline_result;
274 
275 	len = buf->len - offset;
276 	while (len > 0) {
277 		err = map_private_extent_buffer(buf, offset, 32,
278 					&kaddr, &map_start, &map_len);
279 		if (err)
280 			return 1;
281 		cur_len = min(len, map_len - (offset - map_start));
282 		crc = btrfs_csum_data(kaddr + offset - map_start,
283 				      crc, cur_len);
284 		len -= cur_len;
285 		offset += cur_len;
286 	}
287 	if (csum_size > sizeof(inline_result)) {
288 		result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
289 		if (!result)
290 			return 1;
291 	} else {
292 		result = (char *)&inline_result;
293 	}
294 
295 	btrfs_csum_final(crc, result);
296 
297 	if (verify) {
298 		if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
299 			u32 val;
300 			u32 found = 0;
301 			memcpy(&found, result, csum_size);
302 
303 			read_extent_buffer(buf, &val, 0, csum_size);
304 			printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
305 				       "failed on %llu wanted %X found %X "
306 				       "level %d\n",
307 				       root->fs_info->sb->s_id, buf->start,
308 				       val, found, btrfs_header_level(buf));
309 			if (result != (char *)&inline_result)
310 				kfree(result);
311 			return 1;
312 		}
313 	} else {
314 		write_extent_buffer(buf, result, 0, csum_size);
315 	}
316 	if (result != (char *)&inline_result)
317 		kfree(result);
318 	return 0;
319 }
320 
321 /*
322  * we can't consider a given block up to date unless the transid of the
323  * block matches the transid in the parent node's pointer.  This is how we
324  * detect blocks that either didn't get written at all or got written
325  * in the wrong place.
326  */
327 static int verify_parent_transid(struct extent_io_tree *io_tree,
328 				 struct extent_buffer *eb, u64 parent_transid,
329 				 int atomic)
330 {
331 	struct extent_state *cached_state = NULL;
332 	int ret;
333 
334 	if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
335 		return 0;
336 
337 	if (atomic)
338 		return -EAGAIN;
339 
340 	lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
341 			 0, &cached_state);
342 	if (extent_buffer_uptodate(eb) &&
343 	    btrfs_header_generation(eb) == parent_transid) {
344 		ret = 0;
345 		goto out;
346 	}
347 	printk_ratelimited("parent transid verify failed on %llu wanted %llu "
348 		       "found %llu\n",
349 		       eb->start, parent_transid, btrfs_header_generation(eb));
350 	ret = 1;
351 	clear_extent_buffer_uptodate(eb);
352 out:
353 	unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
354 			     &cached_state, GFP_NOFS);
355 	return ret;
356 }
357 
358 /*
359  * Return 0 if the superblock checksum type matches the checksum value of that
360  * algorithm. Pass the raw disk superblock data.
361  */
362 static int btrfs_check_super_csum(char *raw_disk_sb)
363 {
364 	struct btrfs_super_block *disk_sb =
365 		(struct btrfs_super_block *)raw_disk_sb;
366 	u16 csum_type = btrfs_super_csum_type(disk_sb);
367 	int ret = 0;
368 
369 	if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
370 		u32 crc = ~(u32)0;
371 		const int csum_size = sizeof(crc);
372 		char result[csum_size];
373 
374 		/*
375 		 * The super_block structure does not span the whole
376 		 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
377 		 * is filled with zeros and is included in the checkum.
378 		 */
379 		crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
380 				crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
381 		btrfs_csum_final(crc, result);
382 
383 		if (memcmp(raw_disk_sb, result, csum_size))
384 			ret = 1;
385 
386 		if (ret && btrfs_super_generation(disk_sb) < 10) {
387 			printk(KERN_WARNING "btrfs: super block crcs don't match, older mkfs detected\n");
388 			ret = 0;
389 		}
390 	}
391 
392 	if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
393 		printk(KERN_ERR "btrfs: unsupported checksum algorithm %u\n",
394 				csum_type);
395 		ret = 1;
396 	}
397 
398 	return ret;
399 }
400 
401 /*
402  * helper to read a given tree block, doing retries as required when
403  * the checksums don't match and we have alternate mirrors to try.
404  */
405 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
406 					  struct extent_buffer *eb,
407 					  u64 start, u64 parent_transid)
408 {
409 	struct extent_io_tree *io_tree;
410 	int failed = 0;
411 	int ret;
412 	int num_copies = 0;
413 	int mirror_num = 0;
414 	int failed_mirror = 0;
415 
416 	clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
417 	io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
418 	while (1) {
419 		ret = read_extent_buffer_pages(io_tree, eb, start,
420 					       WAIT_COMPLETE,
421 					       btree_get_extent, mirror_num);
422 		if (!ret) {
423 			if (!verify_parent_transid(io_tree, eb,
424 						   parent_transid, 0))
425 				break;
426 			else
427 				ret = -EIO;
428 		}
429 
430 		/*
431 		 * This buffer's crc is fine, but its contents are corrupted, so
432 		 * there is no reason to read the other copies, they won't be
433 		 * any less wrong.
434 		 */
435 		if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
436 			break;
437 
438 		num_copies = btrfs_num_copies(root->fs_info,
439 					      eb->start, eb->len);
440 		if (num_copies == 1)
441 			break;
442 
443 		if (!failed_mirror) {
444 			failed = 1;
445 			failed_mirror = eb->read_mirror;
446 		}
447 
448 		mirror_num++;
449 		if (mirror_num == failed_mirror)
450 			mirror_num++;
451 
452 		if (mirror_num > num_copies)
453 			break;
454 	}
455 
456 	if (failed && !ret && failed_mirror)
457 		repair_eb_io_failure(root, eb, failed_mirror);
458 
459 	return ret;
460 }
461 
462 /*
463  * checksum a dirty tree block before IO.  This has extra checks to make sure
464  * we only fill in the checksum field in the first page of a multi-page block
465  */
466 
467 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
468 {
469 	struct extent_io_tree *tree;
470 	u64 start = page_offset(page);
471 	u64 found_start;
472 	struct extent_buffer *eb;
473 
474 	tree = &BTRFS_I(page->mapping->host)->io_tree;
475 
476 	eb = (struct extent_buffer *)page->private;
477 	if (page != eb->pages[0])
478 		return 0;
479 	found_start = btrfs_header_bytenr(eb);
480 	if (found_start != start) {
481 		WARN_ON(1);
482 		return 0;
483 	}
484 	if (!PageUptodate(page)) {
485 		WARN_ON(1);
486 		return 0;
487 	}
488 	csum_tree_block(root, eb, 0);
489 	return 0;
490 }
491 
492 static int check_tree_block_fsid(struct btrfs_root *root,
493 				 struct extent_buffer *eb)
494 {
495 	struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
496 	u8 fsid[BTRFS_UUID_SIZE];
497 	int ret = 1;
498 
499 	read_extent_buffer(eb, fsid, btrfs_header_fsid(eb), BTRFS_FSID_SIZE);
500 	while (fs_devices) {
501 		if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
502 			ret = 0;
503 			break;
504 		}
505 		fs_devices = fs_devices->seed;
506 	}
507 	return ret;
508 }
509 
510 #define CORRUPT(reason, eb, root, slot)				\
511 	printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu,"	\
512 	       "root=%llu, slot=%d\n", reason,			\
513 	       btrfs_header_bytenr(eb),	root->objectid, slot)
514 
515 static noinline int check_leaf(struct btrfs_root *root,
516 			       struct extent_buffer *leaf)
517 {
518 	struct btrfs_key key;
519 	struct btrfs_key leaf_key;
520 	u32 nritems = btrfs_header_nritems(leaf);
521 	int slot;
522 
523 	if (nritems == 0)
524 		return 0;
525 
526 	/* Check the 0 item */
527 	if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
528 	    BTRFS_LEAF_DATA_SIZE(root)) {
529 		CORRUPT("invalid item offset size pair", leaf, root, 0);
530 		return -EIO;
531 	}
532 
533 	/*
534 	 * Check to make sure each items keys are in the correct order and their
535 	 * offsets make sense.  We only have to loop through nritems-1 because
536 	 * we check the current slot against the next slot, which verifies the
537 	 * next slot's offset+size makes sense and that the current's slot
538 	 * offset is correct.
539 	 */
540 	for (slot = 0; slot < nritems - 1; slot++) {
541 		btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
542 		btrfs_item_key_to_cpu(leaf, &key, slot + 1);
543 
544 		/* Make sure the keys are in the right order */
545 		if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
546 			CORRUPT("bad key order", leaf, root, slot);
547 			return -EIO;
548 		}
549 
550 		/*
551 		 * Make sure the offset and ends are right, remember that the
552 		 * item data starts at the end of the leaf and grows towards the
553 		 * front.
554 		 */
555 		if (btrfs_item_offset_nr(leaf, slot) !=
556 			btrfs_item_end_nr(leaf, slot + 1)) {
557 			CORRUPT("slot offset bad", leaf, root, slot);
558 			return -EIO;
559 		}
560 
561 		/*
562 		 * Check to make sure that we don't point outside of the leaf,
563 		 * just incase all the items are consistent to eachother, but
564 		 * all point outside of the leaf.
565 		 */
566 		if (btrfs_item_end_nr(leaf, slot) >
567 		    BTRFS_LEAF_DATA_SIZE(root)) {
568 			CORRUPT("slot end outside of leaf", leaf, root, slot);
569 			return -EIO;
570 		}
571 	}
572 
573 	return 0;
574 }
575 
576 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
577 				      u64 phy_offset, struct page *page,
578 				      u64 start, u64 end, int mirror)
579 {
580 	struct extent_io_tree *tree;
581 	u64 found_start;
582 	int found_level;
583 	struct extent_buffer *eb;
584 	struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
585 	int ret = 0;
586 	int reads_done;
587 
588 	if (!page->private)
589 		goto out;
590 
591 	tree = &BTRFS_I(page->mapping->host)->io_tree;
592 	eb = (struct extent_buffer *)page->private;
593 
594 	/* the pending IO might have been the only thing that kept this buffer
595 	 * in memory.  Make sure we have a ref for all this other checks
596 	 */
597 	extent_buffer_get(eb);
598 
599 	reads_done = atomic_dec_and_test(&eb->io_pages);
600 	if (!reads_done)
601 		goto err;
602 
603 	eb->read_mirror = mirror;
604 	if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
605 		ret = -EIO;
606 		goto err;
607 	}
608 
609 	found_start = btrfs_header_bytenr(eb);
610 	if (found_start != eb->start) {
611 		printk_ratelimited(KERN_INFO "btrfs bad tree block start "
612 			       "%llu %llu\n",
613 			       found_start, eb->start);
614 		ret = -EIO;
615 		goto err;
616 	}
617 	if (check_tree_block_fsid(root, eb)) {
618 		printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
619 			       eb->start);
620 		ret = -EIO;
621 		goto err;
622 	}
623 	found_level = btrfs_header_level(eb);
624 	if (found_level >= BTRFS_MAX_LEVEL) {
625 		btrfs_info(root->fs_info, "bad tree block level %d\n",
626 			   (int)btrfs_header_level(eb));
627 		ret = -EIO;
628 		goto err;
629 	}
630 
631 	btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
632 				       eb, found_level);
633 
634 	ret = csum_tree_block(root, eb, 1);
635 	if (ret) {
636 		ret = -EIO;
637 		goto err;
638 	}
639 
640 	/*
641 	 * If this is a leaf block and it is corrupt, set the corrupt bit so
642 	 * that we don't try and read the other copies of this block, just
643 	 * return -EIO.
644 	 */
645 	if (found_level == 0 && check_leaf(root, eb)) {
646 		set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
647 		ret = -EIO;
648 	}
649 
650 	if (!ret)
651 		set_extent_buffer_uptodate(eb);
652 err:
653 	if (reads_done &&
654 	    test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
655 		btree_readahead_hook(root, eb, eb->start, ret);
656 
657 	if (ret) {
658 		/*
659 		 * our io error hook is going to dec the io pages
660 		 * again, we have to make sure it has something
661 		 * to decrement
662 		 */
663 		atomic_inc(&eb->io_pages);
664 		clear_extent_buffer_uptodate(eb);
665 	}
666 	free_extent_buffer(eb);
667 out:
668 	return ret;
669 }
670 
671 static int btree_io_failed_hook(struct page *page, int failed_mirror)
672 {
673 	struct extent_buffer *eb;
674 	struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
675 
676 	eb = (struct extent_buffer *)page->private;
677 	set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
678 	eb->read_mirror = failed_mirror;
679 	atomic_dec(&eb->io_pages);
680 	if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
681 		btree_readahead_hook(root, eb, eb->start, -EIO);
682 	return -EIO;	/* we fixed nothing */
683 }
684 
685 static void end_workqueue_bio(struct bio *bio, int err)
686 {
687 	struct end_io_wq *end_io_wq = bio->bi_private;
688 	struct btrfs_fs_info *fs_info;
689 
690 	fs_info = end_io_wq->info;
691 	end_io_wq->error = err;
692 	end_io_wq->work.func = end_workqueue_fn;
693 	end_io_wq->work.flags = 0;
694 
695 	if (bio->bi_rw & REQ_WRITE) {
696 		if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
697 			btrfs_queue_worker(&fs_info->endio_meta_write_workers,
698 					   &end_io_wq->work);
699 		else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
700 			btrfs_queue_worker(&fs_info->endio_freespace_worker,
701 					   &end_io_wq->work);
702 		else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
703 			btrfs_queue_worker(&fs_info->endio_raid56_workers,
704 					   &end_io_wq->work);
705 		else
706 			btrfs_queue_worker(&fs_info->endio_write_workers,
707 					   &end_io_wq->work);
708 	} else {
709 		if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
710 			btrfs_queue_worker(&fs_info->endio_raid56_workers,
711 					   &end_io_wq->work);
712 		else if (end_io_wq->metadata)
713 			btrfs_queue_worker(&fs_info->endio_meta_workers,
714 					   &end_io_wq->work);
715 		else
716 			btrfs_queue_worker(&fs_info->endio_workers,
717 					   &end_io_wq->work);
718 	}
719 }
720 
721 /*
722  * For the metadata arg you want
723  *
724  * 0 - if data
725  * 1 - if normal metadta
726  * 2 - if writing to the free space cache area
727  * 3 - raid parity work
728  */
729 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
730 			int metadata)
731 {
732 	struct end_io_wq *end_io_wq;
733 	end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
734 	if (!end_io_wq)
735 		return -ENOMEM;
736 
737 	end_io_wq->private = bio->bi_private;
738 	end_io_wq->end_io = bio->bi_end_io;
739 	end_io_wq->info = info;
740 	end_io_wq->error = 0;
741 	end_io_wq->bio = bio;
742 	end_io_wq->metadata = metadata;
743 
744 	bio->bi_private = end_io_wq;
745 	bio->bi_end_io = end_workqueue_bio;
746 	return 0;
747 }
748 
749 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
750 {
751 	unsigned long limit = min_t(unsigned long,
752 				    info->workers.max_workers,
753 				    info->fs_devices->open_devices);
754 	return 256 * limit;
755 }
756 
757 static void run_one_async_start(struct btrfs_work *work)
758 {
759 	struct async_submit_bio *async;
760 	int ret;
761 
762 	async = container_of(work, struct  async_submit_bio, work);
763 	ret = async->submit_bio_start(async->inode, async->rw, async->bio,
764 				      async->mirror_num, async->bio_flags,
765 				      async->bio_offset);
766 	if (ret)
767 		async->error = ret;
768 }
769 
770 static void run_one_async_done(struct btrfs_work *work)
771 {
772 	struct btrfs_fs_info *fs_info;
773 	struct async_submit_bio *async;
774 	int limit;
775 
776 	async = container_of(work, struct  async_submit_bio, work);
777 	fs_info = BTRFS_I(async->inode)->root->fs_info;
778 
779 	limit = btrfs_async_submit_limit(fs_info);
780 	limit = limit * 2 / 3;
781 
782 	if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
783 	    waitqueue_active(&fs_info->async_submit_wait))
784 		wake_up(&fs_info->async_submit_wait);
785 
786 	/* If an error occured we just want to clean up the bio and move on */
787 	if (async->error) {
788 		bio_endio(async->bio, async->error);
789 		return;
790 	}
791 
792 	async->submit_bio_done(async->inode, async->rw, async->bio,
793 			       async->mirror_num, async->bio_flags,
794 			       async->bio_offset);
795 }
796 
797 static void run_one_async_free(struct btrfs_work *work)
798 {
799 	struct async_submit_bio *async;
800 
801 	async = container_of(work, struct  async_submit_bio, work);
802 	kfree(async);
803 }
804 
805 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
806 			int rw, struct bio *bio, int mirror_num,
807 			unsigned long bio_flags,
808 			u64 bio_offset,
809 			extent_submit_bio_hook_t *submit_bio_start,
810 			extent_submit_bio_hook_t *submit_bio_done)
811 {
812 	struct async_submit_bio *async;
813 
814 	async = kmalloc(sizeof(*async), GFP_NOFS);
815 	if (!async)
816 		return -ENOMEM;
817 
818 	async->inode = inode;
819 	async->rw = rw;
820 	async->bio = bio;
821 	async->mirror_num = mirror_num;
822 	async->submit_bio_start = submit_bio_start;
823 	async->submit_bio_done = submit_bio_done;
824 
825 	async->work.func = run_one_async_start;
826 	async->work.ordered_func = run_one_async_done;
827 	async->work.ordered_free = run_one_async_free;
828 
829 	async->work.flags = 0;
830 	async->bio_flags = bio_flags;
831 	async->bio_offset = bio_offset;
832 
833 	async->error = 0;
834 
835 	atomic_inc(&fs_info->nr_async_submits);
836 
837 	if (rw & REQ_SYNC)
838 		btrfs_set_work_high_prio(&async->work);
839 
840 	btrfs_queue_worker(&fs_info->workers, &async->work);
841 
842 	while (atomic_read(&fs_info->async_submit_draining) &&
843 	      atomic_read(&fs_info->nr_async_submits)) {
844 		wait_event(fs_info->async_submit_wait,
845 			   (atomic_read(&fs_info->nr_async_submits) == 0));
846 	}
847 
848 	return 0;
849 }
850 
851 static int btree_csum_one_bio(struct bio *bio)
852 {
853 	struct bio_vec *bvec = bio->bi_io_vec;
854 	int bio_index = 0;
855 	struct btrfs_root *root;
856 	int ret = 0;
857 
858 	WARN_ON(bio->bi_vcnt <= 0);
859 	while (bio_index < bio->bi_vcnt) {
860 		root = BTRFS_I(bvec->bv_page->mapping->host)->root;
861 		ret = csum_dirty_buffer(root, bvec->bv_page);
862 		if (ret)
863 			break;
864 		bio_index++;
865 		bvec++;
866 	}
867 	return ret;
868 }
869 
870 static int __btree_submit_bio_start(struct inode *inode, int rw,
871 				    struct bio *bio, int mirror_num,
872 				    unsigned long bio_flags,
873 				    u64 bio_offset)
874 {
875 	/*
876 	 * when we're called for a write, we're already in the async
877 	 * submission context.  Just jump into btrfs_map_bio
878 	 */
879 	return btree_csum_one_bio(bio);
880 }
881 
882 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
883 				 int mirror_num, unsigned long bio_flags,
884 				 u64 bio_offset)
885 {
886 	int ret;
887 
888 	/*
889 	 * when we're called for a write, we're already in the async
890 	 * submission context.  Just jump into btrfs_map_bio
891 	 */
892 	ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
893 	if (ret)
894 		bio_endio(bio, ret);
895 	return ret;
896 }
897 
898 static int check_async_write(struct inode *inode, unsigned long bio_flags)
899 {
900 	if (bio_flags & EXTENT_BIO_TREE_LOG)
901 		return 0;
902 #ifdef CONFIG_X86
903 	if (cpu_has_xmm4_2)
904 		return 0;
905 #endif
906 	return 1;
907 }
908 
909 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
910 				 int mirror_num, unsigned long bio_flags,
911 				 u64 bio_offset)
912 {
913 	int async = check_async_write(inode, bio_flags);
914 	int ret;
915 
916 	if (!(rw & REQ_WRITE)) {
917 		/*
918 		 * called for a read, do the setup so that checksum validation
919 		 * can happen in the async kernel threads
920 		 */
921 		ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
922 					  bio, 1);
923 		if (ret)
924 			goto out_w_error;
925 		ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
926 				    mirror_num, 0);
927 	} else if (!async) {
928 		ret = btree_csum_one_bio(bio);
929 		if (ret)
930 			goto out_w_error;
931 		ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
932 				    mirror_num, 0);
933 	} else {
934 		/*
935 		 * kthread helpers are used to submit writes so that
936 		 * checksumming can happen in parallel across all CPUs
937 		 */
938 		ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
939 					  inode, rw, bio, mirror_num, 0,
940 					  bio_offset,
941 					  __btree_submit_bio_start,
942 					  __btree_submit_bio_done);
943 	}
944 
945 	if (ret) {
946 out_w_error:
947 		bio_endio(bio, ret);
948 	}
949 	return ret;
950 }
951 
952 #ifdef CONFIG_MIGRATION
953 static int btree_migratepage(struct address_space *mapping,
954 			struct page *newpage, struct page *page,
955 			enum migrate_mode mode)
956 {
957 	/*
958 	 * we can't safely write a btree page from here,
959 	 * we haven't done the locking hook
960 	 */
961 	if (PageDirty(page))
962 		return -EAGAIN;
963 	/*
964 	 * Buffers may be managed in a filesystem specific way.
965 	 * We must have no buffers or drop them.
966 	 */
967 	if (page_has_private(page) &&
968 	    !try_to_release_page(page, GFP_KERNEL))
969 		return -EAGAIN;
970 	return migrate_page(mapping, newpage, page, mode);
971 }
972 #endif
973 
974 
975 static int btree_writepages(struct address_space *mapping,
976 			    struct writeback_control *wbc)
977 {
978 	struct extent_io_tree *tree;
979 	struct btrfs_fs_info *fs_info;
980 	int ret;
981 
982 	tree = &BTRFS_I(mapping->host)->io_tree;
983 	if (wbc->sync_mode == WB_SYNC_NONE) {
984 
985 		if (wbc->for_kupdate)
986 			return 0;
987 
988 		fs_info = BTRFS_I(mapping->host)->root->fs_info;
989 		/* this is a bit racy, but that's ok */
990 		ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
991 					     BTRFS_DIRTY_METADATA_THRESH);
992 		if (ret < 0)
993 			return 0;
994 	}
995 	return btree_write_cache_pages(mapping, wbc);
996 }
997 
998 static int btree_readpage(struct file *file, struct page *page)
999 {
1000 	struct extent_io_tree *tree;
1001 	tree = &BTRFS_I(page->mapping->host)->io_tree;
1002 	return extent_read_full_page(tree, page, btree_get_extent, 0);
1003 }
1004 
1005 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1006 {
1007 	if (PageWriteback(page) || PageDirty(page))
1008 		return 0;
1009 
1010 	return try_release_extent_buffer(page);
1011 }
1012 
1013 static void btree_invalidatepage(struct page *page, unsigned int offset,
1014 				 unsigned int length)
1015 {
1016 	struct extent_io_tree *tree;
1017 	tree = &BTRFS_I(page->mapping->host)->io_tree;
1018 	extent_invalidatepage(tree, page, offset);
1019 	btree_releasepage(page, GFP_NOFS);
1020 	if (PagePrivate(page)) {
1021 		printk(KERN_WARNING "btrfs warning page private not zero "
1022 		       "on page %llu\n", (unsigned long long)page_offset(page));
1023 		ClearPagePrivate(page);
1024 		set_page_private(page, 0);
1025 		page_cache_release(page);
1026 	}
1027 }
1028 
1029 static int btree_set_page_dirty(struct page *page)
1030 {
1031 #ifdef DEBUG
1032 	struct extent_buffer *eb;
1033 
1034 	BUG_ON(!PagePrivate(page));
1035 	eb = (struct extent_buffer *)page->private;
1036 	BUG_ON(!eb);
1037 	BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1038 	BUG_ON(!atomic_read(&eb->refs));
1039 	btrfs_assert_tree_locked(eb);
1040 #endif
1041 	return __set_page_dirty_nobuffers(page);
1042 }
1043 
1044 static const struct address_space_operations btree_aops = {
1045 	.readpage	= btree_readpage,
1046 	.writepages	= btree_writepages,
1047 	.releasepage	= btree_releasepage,
1048 	.invalidatepage = btree_invalidatepage,
1049 #ifdef CONFIG_MIGRATION
1050 	.migratepage	= btree_migratepage,
1051 #endif
1052 	.set_page_dirty = btree_set_page_dirty,
1053 };
1054 
1055 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1056 			 u64 parent_transid)
1057 {
1058 	struct extent_buffer *buf = NULL;
1059 	struct inode *btree_inode = root->fs_info->btree_inode;
1060 	int ret = 0;
1061 
1062 	buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1063 	if (!buf)
1064 		return 0;
1065 	read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1066 				 buf, 0, WAIT_NONE, btree_get_extent, 0);
1067 	free_extent_buffer(buf);
1068 	return ret;
1069 }
1070 
1071 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1072 			 int mirror_num, struct extent_buffer **eb)
1073 {
1074 	struct extent_buffer *buf = NULL;
1075 	struct inode *btree_inode = root->fs_info->btree_inode;
1076 	struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1077 	int ret;
1078 
1079 	buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1080 	if (!buf)
1081 		return 0;
1082 
1083 	set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1084 
1085 	ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1086 				       btree_get_extent, mirror_num);
1087 	if (ret) {
1088 		free_extent_buffer(buf);
1089 		return ret;
1090 	}
1091 
1092 	if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1093 		free_extent_buffer(buf);
1094 		return -EIO;
1095 	} else if (extent_buffer_uptodate(buf)) {
1096 		*eb = buf;
1097 	} else {
1098 		free_extent_buffer(buf);
1099 	}
1100 	return 0;
1101 }
1102 
1103 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1104 					    u64 bytenr, u32 blocksize)
1105 {
1106 	struct inode *btree_inode = root->fs_info->btree_inode;
1107 	struct extent_buffer *eb;
1108 	eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1109 				bytenr, blocksize);
1110 	return eb;
1111 }
1112 
1113 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1114 						 u64 bytenr, u32 blocksize)
1115 {
1116 	struct inode *btree_inode = root->fs_info->btree_inode;
1117 	struct extent_buffer *eb;
1118 
1119 	eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1120 				 bytenr, blocksize);
1121 	return eb;
1122 }
1123 
1124 
1125 int btrfs_write_tree_block(struct extent_buffer *buf)
1126 {
1127 	return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1128 					buf->start + buf->len - 1);
1129 }
1130 
1131 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1132 {
1133 	return filemap_fdatawait_range(buf->pages[0]->mapping,
1134 				       buf->start, buf->start + buf->len - 1);
1135 }
1136 
1137 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1138 				      u32 blocksize, u64 parent_transid)
1139 {
1140 	struct extent_buffer *buf = NULL;
1141 	int ret;
1142 
1143 	buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1144 	if (!buf)
1145 		return NULL;
1146 
1147 	ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1148 	if (ret) {
1149 		free_extent_buffer(buf);
1150 		return NULL;
1151 	}
1152 	return buf;
1153 
1154 }
1155 
1156 void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1157 		      struct extent_buffer *buf)
1158 {
1159 	struct btrfs_fs_info *fs_info = root->fs_info;
1160 
1161 	if (btrfs_header_generation(buf) ==
1162 	    fs_info->running_transaction->transid) {
1163 		btrfs_assert_tree_locked(buf);
1164 
1165 		if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1166 			__percpu_counter_add(&fs_info->dirty_metadata_bytes,
1167 					     -buf->len,
1168 					     fs_info->dirty_metadata_batch);
1169 			/* ugh, clear_extent_buffer_dirty needs to lock the page */
1170 			btrfs_set_lock_blocking(buf);
1171 			clear_extent_buffer_dirty(buf);
1172 		}
1173 	}
1174 }
1175 
1176 static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1177 			 u32 stripesize, struct btrfs_root *root,
1178 			 struct btrfs_fs_info *fs_info,
1179 			 u64 objectid)
1180 {
1181 	root->node = NULL;
1182 	root->commit_root = NULL;
1183 	root->sectorsize = sectorsize;
1184 	root->nodesize = nodesize;
1185 	root->leafsize = leafsize;
1186 	root->stripesize = stripesize;
1187 	root->ref_cows = 0;
1188 	root->track_dirty = 0;
1189 	root->in_radix = 0;
1190 	root->orphan_item_inserted = 0;
1191 	root->orphan_cleanup_state = 0;
1192 
1193 	root->objectid = objectid;
1194 	root->last_trans = 0;
1195 	root->highest_objectid = 0;
1196 	root->nr_delalloc_inodes = 0;
1197 	root->nr_ordered_extents = 0;
1198 	root->name = NULL;
1199 	root->inode_tree = RB_ROOT;
1200 	INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1201 	root->block_rsv = NULL;
1202 	root->orphan_block_rsv = NULL;
1203 
1204 	INIT_LIST_HEAD(&root->dirty_list);
1205 	INIT_LIST_HEAD(&root->root_list);
1206 	INIT_LIST_HEAD(&root->delalloc_inodes);
1207 	INIT_LIST_HEAD(&root->delalloc_root);
1208 	INIT_LIST_HEAD(&root->ordered_extents);
1209 	INIT_LIST_HEAD(&root->ordered_root);
1210 	INIT_LIST_HEAD(&root->logged_list[0]);
1211 	INIT_LIST_HEAD(&root->logged_list[1]);
1212 	spin_lock_init(&root->orphan_lock);
1213 	spin_lock_init(&root->inode_lock);
1214 	spin_lock_init(&root->delalloc_lock);
1215 	spin_lock_init(&root->ordered_extent_lock);
1216 	spin_lock_init(&root->accounting_lock);
1217 	spin_lock_init(&root->log_extents_lock[0]);
1218 	spin_lock_init(&root->log_extents_lock[1]);
1219 	mutex_init(&root->objectid_mutex);
1220 	mutex_init(&root->log_mutex);
1221 	init_waitqueue_head(&root->log_writer_wait);
1222 	init_waitqueue_head(&root->log_commit_wait[0]);
1223 	init_waitqueue_head(&root->log_commit_wait[1]);
1224 	atomic_set(&root->log_commit[0], 0);
1225 	atomic_set(&root->log_commit[1], 0);
1226 	atomic_set(&root->log_writers, 0);
1227 	atomic_set(&root->log_batch, 0);
1228 	atomic_set(&root->orphan_inodes, 0);
1229 	atomic_set(&root->refs, 1);
1230 	root->log_transid = 0;
1231 	root->last_log_commit = 0;
1232 	extent_io_tree_init(&root->dirty_log_pages,
1233 			     fs_info->btree_inode->i_mapping);
1234 
1235 	memset(&root->root_key, 0, sizeof(root->root_key));
1236 	memset(&root->root_item, 0, sizeof(root->root_item));
1237 	memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1238 	memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1239 	root->defrag_trans_start = fs_info->generation;
1240 	init_completion(&root->kobj_unregister);
1241 	root->defrag_running = 0;
1242 	root->root_key.objectid = objectid;
1243 	root->anon_dev = 0;
1244 
1245 	spin_lock_init(&root->root_item_lock);
1246 }
1247 
1248 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1249 {
1250 	struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1251 	if (root)
1252 		root->fs_info = fs_info;
1253 	return root;
1254 }
1255 
1256 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1257 				     struct btrfs_fs_info *fs_info,
1258 				     u64 objectid)
1259 {
1260 	struct extent_buffer *leaf;
1261 	struct btrfs_root *tree_root = fs_info->tree_root;
1262 	struct btrfs_root *root;
1263 	struct btrfs_key key;
1264 	int ret = 0;
1265 	u64 bytenr;
1266 	uuid_le uuid;
1267 
1268 	root = btrfs_alloc_root(fs_info);
1269 	if (!root)
1270 		return ERR_PTR(-ENOMEM);
1271 
1272 	__setup_root(tree_root->nodesize, tree_root->leafsize,
1273 		     tree_root->sectorsize, tree_root->stripesize,
1274 		     root, fs_info, objectid);
1275 	root->root_key.objectid = objectid;
1276 	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1277 	root->root_key.offset = 0;
1278 
1279 	leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1280 				      0, objectid, NULL, 0, 0, 0);
1281 	if (IS_ERR(leaf)) {
1282 		ret = PTR_ERR(leaf);
1283 		leaf = NULL;
1284 		goto fail;
1285 	}
1286 
1287 	bytenr = leaf->start;
1288 	memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1289 	btrfs_set_header_bytenr(leaf, leaf->start);
1290 	btrfs_set_header_generation(leaf, trans->transid);
1291 	btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1292 	btrfs_set_header_owner(leaf, objectid);
1293 	root->node = leaf;
1294 
1295 	write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(leaf),
1296 			    BTRFS_FSID_SIZE);
1297 	write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1298 			    btrfs_header_chunk_tree_uuid(leaf),
1299 			    BTRFS_UUID_SIZE);
1300 	btrfs_mark_buffer_dirty(leaf);
1301 
1302 	root->commit_root = btrfs_root_node(root);
1303 	root->track_dirty = 1;
1304 
1305 
1306 	root->root_item.flags = 0;
1307 	root->root_item.byte_limit = 0;
1308 	btrfs_set_root_bytenr(&root->root_item, leaf->start);
1309 	btrfs_set_root_generation(&root->root_item, trans->transid);
1310 	btrfs_set_root_level(&root->root_item, 0);
1311 	btrfs_set_root_refs(&root->root_item, 1);
1312 	btrfs_set_root_used(&root->root_item, leaf->len);
1313 	btrfs_set_root_last_snapshot(&root->root_item, 0);
1314 	btrfs_set_root_dirid(&root->root_item, 0);
1315 	uuid_le_gen(&uuid);
1316 	memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1317 	root->root_item.drop_level = 0;
1318 
1319 	key.objectid = objectid;
1320 	key.type = BTRFS_ROOT_ITEM_KEY;
1321 	key.offset = 0;
1322 	ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1323 	if (ret)
1324 		goto fail;
1325 
1326 	btrfs_tree_unlock(leaf);
1327 
1328 	return root;
1329 
1330 fail:
1331 	if (leaf) {
1332 		btrfs_tree_unlock(leaf);
1333 		free_extent_buffer(leaf);
1334 	}
1335 	kfree(root);
1336 
1337 	return ERR_PTR(ret);
1338 }
1339 
1340 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1341 					 struct btrfs_fs_info *fs_info)
1342 {
1343 	struct btrfs_root *root;
1344 	struct btrfs_root *tree_root = fs_info->tree_root;
1345 	struct extent_buffer *leaf;
1346 
1347 	root = btrfs_alloc_root(fs_info);
1348 	if (!root)
1349 		return ERR_PTR(-ENOMEM);
1350 
1351 	__setup_root(tree_root->nodesize, tree_root->leafsize,
1352 		     tree_root->sectorsize, tree_root->stripesize,
1353 		     root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1354 
1355 	root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1356 	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1357 	root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1358 	/*
1359 	 * log trees do not get reference counted because they go away
1360 	 * before a real commit is actually done.  They do store pointers
1361 	 * to file data extents, and those reference counts still get
1362 	 * updated (along with back refs to the log tree).
1363 	 */
1364 	root->ref_cows = 0;
1365 
1366 	leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1367 				      BTRFS_TREE_LOG_OBJECTID, NULL,
1368 				      0, 0, 0);
1369 	if (IS_ERR(leaf)) {
1370 		kfree(root);
1371 		return ERR_CAST(leaf);
1372 	}
1373 
1374 	memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1375 	btrfs_set_header_bytenr(leaf, leaf->start);
1376 	btrfs_set_header_generation(leaf, trans->transid);
1377 	btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1378 	btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1379 	root->node = leaf;
1380 
1381 	write_extent_buffer(root->node, root->fs_info->fsid,
1382 			    btrfs_header_fsid(root->node), BTRFS_FSID_SIZE);
1383 	btrfs_mark_buffer_dirty(root->node);
1384 	btrfs_tree_unlock(root->node);
1385 	return root;
1386 }
1387 
1388 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1389 			     struct btrfs_fs_info *fs_info)
1390 {
1391 	struct btrfs_root *log_root;
1392 
1393 	log_root = alloc_log_tree(trans, fs_info);
1394 	if (IS_ERR(log_root))
1395 		return PTR_ERR(log_root);
1396 	WARN_ON(fs_info->log_root_tree);
1397 	fs_info->log_root_tree = log_root;
1398 	return 0;
1399 }
1400 
1401 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1402 		       struct btrfs_root *root)
1403 {
1404 	struct btrfs_root *log_root;
1405 	struct btrfs_inode_item *inode_item;
1406 
1407 	log_root = alloc_log_tree(trans, root->fs_info);
1408 	if (IS_ERR(log_root))
1409 		return PTR_ERR(log_root);
1410 
1411 	log_root->last_trans = trans->transid;
1412 	log_root->root_key.offset = root->root_key.objectid;
1413 
1414 	inode_item = &log_root->root_item.inode;
1415 	btrfs_set_stack_inode_generation(inode_item, 1);
1416 	btrfs_set_stack_inode_size(inode_item, 3);
1417 	btrfs_set_stack_inode_nlink(inode_item, 1);
1418 	btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
1419 	btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1420 
1421 	btrfs_set_root_node(&log_root->root_item, log_root->node);
1422 
1423 	WARN_ON(root->log_root);
1424 	root->log_root = log_root;
1425 	root->log_transid = 0;
1426 	root->last_log_commit = 0;
1427 	return 0;
1428 }
1429 
1430 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1431 					       struct btrfs_key *key)
1432 {
1433 	struct btrfs_root *root;
1434 	struct btrfs_fs_info *fs_info = tree_root->fs_info;
1435 	struct btrfs_path *path;
1436 	u64 generation;
1437 	u32 blocksize;
1438 	int ret;
1439 
1440 	path = btrfs_alloc_path();
1441 	if (!path)
1442 		return ERR_PTR(-ENOMEM);
1443 
1444 	root = btrfs_alloc_root(fs_info);
1445 	if (!root) {
1446 		ret = -ENOMEM;
1447 		goto alloc_fail;
1448 	}
1449 
1450 	__setup_root(tree_root->nodesize, tree_root->leafsize,
1451 		     tree_root->sectorsize, tree_root->stripesize,
1452 		     root, fs_info, key->objectid);
1453 
1454 	ret = btrfs_find_root(tree_root, key, path,
1455 			      &root->root_item, &root->root_key);
1456 	if (ret) {
1457 		if (ret > 0)
1458 			ret = -ENOENT;
1459 		goto find_fail;
1460 	}
1461 
1462 	generation = btrfs_root_generation(&root->root_item);
1463 	blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1464 	root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1465 				     blocksize, generation);
1466 	if (!root->node) {
1467 		ret = -ENOMEM;
1468 		goto find_fail;
1469 	} else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1470 		ret = -EIO;
1471 		goto read_fail;
1472 	}
1473 	root->commit_root = btrfs_root_node(root);
1474 out:
1475 	btrfs_free_path(path);
1476 	return root;
1477 
1478 read_fail:
1479 	free_extent_buffer(root->node);
1480 find_fail:
1481 	kfree(root);
1482 alloc_fail:
1483 	root = ERR_PTR(ret);
1484 	goto out;
1485 }
1486 
1487 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1488 				      struct btrfs_key *location)
1489 {
1490 	struct btrfs_root *root;
1491 
1492 	root = btrfs_read_tree_root(tree_root, location);
1493 	if (IS_ERR(root))
1494 		return root;
1495 
1496 	if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1497 		root->ref_cows = 1;
1498 		btrfs_check_and_init_root_item(&root->root_item);
1499 	}
1500 
1501 	return root;
1502 }
1503 
1504 int btrfs_init_fs_root(struct btrfs_root *root)
1505 {
1506 	int ret;
1507 
1508 	root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1509 	root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1510 					GFP_NOFS);
1511 	if (!root->free_ino_pinned || !root->free_ino_ctl) {
1512 		ret = -ENOMEM;
1513 		goto fail;
1514 	}
1515 
1516 	btrfs_init_free_ino_ctl(root);
1517 	mutex_init(&root->fs_commit_mutex);
1518 	spin_lock_init(&root->cache_lock);
1519 	init_waitqueue_head(&root->cache_wait);
1520 
1521 	ret = get_anon_bdev(&root->anon_dev);
1522 	if (ret)
1523 		goto fail;
1524 	return 0;
1525 fail:
1526 	kfree(root->free_ino_ctl);
1527 	kfree(root->free_ino_pinned);
1528 	return ret;
1529 }
1530 
1531 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1532 					       u64 root_id)
1533 {
1534 	struct btrfs_root *root;
1535 
1536 	spin_lock(&fs_info->fs_roots_radix_lock);
1537 	root = radix_tree_lookup(&fs_info->fs_roots_radix,
1538 				 (unsigned long)root_id);
1539 	spin_unlock(&fs_info->fs_roots_radix_lock);
1540 	return root;
1541 }
1542 
1543 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1544 			 struct btrfs_root *root)
1545 {
1546 	int ret;
1547 
1548 	ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1549 	if (ret)
1550 		return ret;
1551 
1552 	spin_lock(&fs_info->fs_roots_radix_lock);
1553 	ret = radix_tree_insert(&fs_info->fs_roots_radix,
1554 				(unsigned long)root->root_key.objectid,
1555 				root);
1556 	if (ret == 0)
1557 		root->in_radix = 1;
1558 	spin_unlock(&fs_info->fs_roots_radix_lock);
1559 	radix_tree_preload_end();
1560 
1561 	return ret;
1562 }
1563 
1564 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1565 				     struct btrfs_key *location,
1566 				     bool check_ref)
1567 {
1568 	struct btrfs_root *root;
1569 	int ret;
1570 
1571 	if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1572 		return fs_info->tree_root;
1573 	if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1574 		return fs_info->extent_root;
1575 	if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1576 		return fs_info->chunk_root;
1577 	if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1578 		return fs_info->dev_root;
1579 	if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1580 		return fs_info->csum_root;
1581 	if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1582 		return fs_info->quota_root ? fs_info->quota_root :
1583 					     ERR_PTR(-ENOENT);
1584 	if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1585 		return fs_info->uuid_root ? fs_info->uuid_root :
1586 					    ERR_PTR(-ENOENT);
1587 again:
1588 	root = btrfs_lookup_fs_root(fs_info, location->objectid);
1589 	if (root) {
1590 		if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1591 			return ERR_PTR(-ENOENT);
1592 		return root;
1593 	}
1594 
1595 	root = btrfs_read_fs_root(fs_info->tree_root, location);
1596 	if (IS_ERR(root))
1597 		return root;
1598 
1599 	if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1600 		ret = -ENOENT;
1601 		goto fail;
1602 	}
1603 
1604 	ret = btrfs_init_fs_root(root);
1605 	if (ret)
1606 		goto fail;
1607 
1608 	ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1609 	if (ret < 0)
1610 		goto fail;
1611 	if (ret == 0)
1612 		root->orphan_item_inserted = 1;
1613 
1614 	ret = btrfs_insert_fs_root(fs_info, root);
1615 	if (ret) {
1616 		if (ret == -EEXIST) {
1617 			free_fs_root(root);
1618 			goto again;
1619 		}
1620 		goto fail;
1621 	}
1622 	return root;
1623 fail:
1624 	free_fs_root(root);
1625 	return ERR_PTR(ret);
1626 }
1627 
1628 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1629 {
1630 	struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1631 	int ret = 0;
1632 	struct btrfs_device *device;
1633 	struct backing_dev_info *bdi;
1634 
1635 	rcu_read_lock();
1636 	list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1637 		if (!device->bdev)
1638 			continue;
1639 		bdi = blk_get_backing_dev_info(device->bdev);
1640 		if (bdi && bdi_congested(bdi, bdi_bits)) {
1641 			ret = 1;
1642 			break;
1643 		}
1644 	}
1645 	rcu_read_unlock();
1646 	return ret;
1647 }
1648 
1649 /*
1650  * If this fails, caller must call bdi_destroy() to get rid of the
1651  * bdi again.
1652  */
1653 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1654 {
1655 	int err;
1656 
1657 	bdi->capabilities = BDI_CAP_MAP_COPY;
1658 	err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1659 	if (err)
1660 		return err;
1661 
1662 	bdi->ra_pages	= default_backing_dev_info.ra_pages;
1663 	bdi->congested_fn	= btrfs_congested_fn;
1664 	bdi->congested_data	= info;
1665 	return 0;
1666 }
1667 
1668 /*
1669  * called by the kthread helper functions to finally call the bio end_io
1670  * functions.  This is where read checksum verification actually happens
1671  */
1672 static void end_workqueue_fn(struct btrfs_work *work)
1673 {
1674 	struct bio *bio;
1675 	struct end_io_wq *end_io_wq;
1676 	struct btrfs_fs_info *fs_info;
1677 	int error;
1678 
1679 	end_io_wq = container_of(work, struct end_io_wq, work);
1680 	bio = end_io_wq->bio;
1681 	fs_info = end_io_wq->info;
1682 
1683 	error = end_io_wq->error;
1684 	bio->bi_private = end_io_wq->private;
1685 	bio->bi_end_io = end_io_wq->end_io;
1686 	kfree(end_io_wq);
1687 	bio_endio(bio, error);
1688 }
1689 
1690 static int cleaner_kthread(void *arg)
1691 {
1692 	struct btrfs_root *root = arg;
1693 	int again;
1694 
1695 	do {
1696 		again = 0;
1697 
1698 		/* Make the cleaner go to sleep early. */
1699 		if (btrfs_need_cleaner_sleep(root))
1700 			goto sleep;
1701 
1702 		if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1703 			goto sleep;
1704 
1705 		/*
1706 		 * Avoid the problem that we change the status of the fs
1707 		 * during the above check and trylock.
1708 		 */
1709 		if (btrfs_need_cleaner_sleep(root)) {
1710 			mutex_unlock(&root->fs_info->cleaner_mutex);
1711 			goto sleep;
1712 		}
1713 
1714 		btrfs_run_delayed_iputs(root);
1715 		again = btrfs_clean_one_deleted_snapshot(root);
1716 		mutex_unlock(&root->fs_info->cleaner_mutex);
1717 
1718 		/*
1719 		 * The defragger has dealt with the R/O remount and umount,
1720 		 * needn't do anything special here.
1721 		 */
1722 		btrfs_run_defrag_inodes(root->fs_info);
1723 sleep:
1724 		if (!try_to_freeze() && !again) {
1725 			set_current_state(TASK_INTERRUPTIBLE);
1726 			if (!kthread_should_stop())
1727 				schedule();
1728 			__set_current_state(TASK_RUNNING);
1729 		}
1730 	} while (!kthread_should_stop());
1731 	return 0;
1732 }
1733 
1734 static int transaction_kthread(void *arg)
1735 {
1736 	struct btrfs_root *root = arg;
1737 	struct btrfs_trans_handle *trans;
1738 	struct btrfs_transaction *cur;
1739 	u64 transid;
1740 	unsigned long now;
1741 	unsigned long delay;
1742 	bool cannot_commit;
1743 
1744 	do {
1745 		cannot_commit = false;
1746 		delay = HZ * root->fs_info->commit_interval;
1747 		mutex_lock(&root->fs_info->transaction_kthread_mutex);
1748 
1749 		spin_lock(&root->fs_info->trans_lock);
1750 		cur = root->fs_info->running_transaction;
1751 		if (!cur) {
1752 			spin_unlock(&root->fs_info->trans_lock);
1753 			goto sleep;
1754 		}
1755 
1756 		now = get_seconds();
1757 		if (cur->state < TRANS_STATE_BLOCKED &&
1758 		    (now < cur->start_time ||
1759 		     now - cur->start_time < root->fs_info->commit_interval)) {
1760 			spin_unlock(&root->fs_info->trans_lock);
1761 			delay = HZ * 5;
1762 			goto sleep;
1763 		}
1764 		transid = cur->transid;
1765 		spin_unlock(&root->fs_info->trans_lock);
1766 
1767 		/* If the file system is aborted, this will always fail. */
1768 		trans = btrfs_attach_transaction(root);
1769 		if (IS_ERR(trans)) {
1770 			if (PTR_ERR(trans) != -ENOENT)
1771 				cannot_commit = true;
1772 			goto sleep;
1773 		}
1774 		if (transid == trans->transid) {
1775 			btrfs_commit_transaction(trans, root);
1776 		} else {
1777 			btrfs_end_transaction(trans, root);
1778 		}
1779 sleep:
1780 		wake_up_process(root->fs_info->cleaner_kthread);
1781 		mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1782 
1783 		if (!try_to_freeze()) {
1784 			set_current_state(TASK_INTERRUPTIBLE);
1785 			if (!kthread_should_stop() &&
1786 			    (!btrfs_transaction_blocked(root->fs_info) ||
1787 			     cannot_commit))
1788 				schedule_timeout(delay);
1789 			__set_current_state(TASK_RUNNING);
1790 		}
1791 	} while (!kthread_should_stop());
1792 	return 0;
1793 }
1794 
1795 /*
1796  * this will find the highest generation in the array of
1797  * root backups.  The index of the highest array is returned,
1798  * or -1 if we can't find anything.
1799  *
1800  * We check to make sure the array is valid by comparing the
1801  * generation of the latest  root in the array with the generation
1802  * in the super block.  If they don't match we pitch it.
1803  */
1804 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1805 {
1806 	u64 cur;
1807 	int newest_index = -1;
1808 	struct btrfs_root_backup *root_backup;
1809 	int i;
1810 
1811 	for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1812 		root_backup = info->super_copy->super_roots + i;
1813 		cur = btrfs_backup_tree_root_gen(root_backup);
1814 		if (cur == newest_gen)
1815 			newest_index = i;
1816 	}
1817 
1818 	/* check to see if we actually wrapped around */
1819 	if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1820 		root_backup = info->super_copy->super_roots;
1821 		cur = btrfs_backup_tree_root_gen(root_backup);
1822 		if (cur == newest_gen)
1823 			newest_index = 0;
1824 	}
1825 	return newest_index;
1826 }
1827 
1828 
1829 /*
1830  * find the oldest backup so we know where to store new entries
1831  * in the backup array.  This will set the backup_root_index
1832  * field in the fs_info struct
1833  */
1834 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1835 				     u64 newest_gen)
1836 {
1837 	int newest_index = -1;
1838 
1839 	newest_index = find_newest_super_backup(info, newest_gen);
1840 	/* if there was garbage in there, just move along */
1841 	if (newest_index == -1) {
1842 		info->backup_root_index = 0;
1843 	} else {
1844 		info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1845 	}
1846 }
1847 
1848 /*
1849  * copy all the root pointers into the super backup array.
1850  * this will bump the backup pointer by one when it is
1851  * done
1852  */
1853 static void backup_super_roots(struct btrfs_fs_info *info)
1854 {
1855 	int next_backup;
1856 	struct btrfs_root_backup *root_backup;
1857 	int last_backup;
1858 
1859 	next_backup = info->backup_root_index;
1860 	last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1861 		BTRFS_NUM_BACKUP_ROOTS;
1862 
1863 	/*
1864 	 * just overwrite the last backup if we're at the same generation
1865 	 * this happens only at umount
1866 	 */
1867 	root_backup = info->super_for_commit->super_roots + last_backup;
1868 	if (btrfs_backup_tree_root_gen(root_backup) ==
1869 	    btrfs_header_generation(info->tree_root->node))
1870 		next_backup = last_backup;
1871 
1872 	root_backup = info->super_for_commit->super_roots + next_backup;
1873 
1874 	/*
1875 	 * make sure all of our padding and empty slots get zero filled
1876 	 * regardless of which ones we use today
1877 	 */
1878 	memset(root_backup, 0, sizeof(*root_backup));
1879 
1880 	info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1881 
1882 	btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1883 	btrfs_set_backup_tree_root_gen(root_backup,
1884 			       btrfs_header_generation(info->tree_root->node));
1885 
1886 	btrfs_set_backup_tree_root_level(root_backup,
1887 			       btrfs_header_level(info->tree_root->node));
1888 
1889 	btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1890 	btrfs_set_backup_chunk_root_gen(root_backup,
1891 			       btrfs_header_generation(info->chunk_root->node));
1892 	btrfs_set_backup_chunk_root_level(root_backup,
1893 			       btrfs_header_level(info->chunk_root->node));
1894 
1895 	btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1896 	btrfs_set_backup_extent_root_gen(root_backup,
1897 			       btrfs_header_generation(info->extent_root->node));
1898 	btrfs_set_backup_extent_root_level(root_backup,
1899 			       btrfs_header_level(info->extent_root->node));
1900 
1901 	/*
1902 	 * we might commit during log recovery, which happens before we set
1903 	 * the fs_root.  Make sure it is valid before we fill it in.
1904 	 */
1905 	if (info->fs_root && info->fs_root->node) {
1906 		btrfs_set_backup_fs_root(root_backup,
1907 					 info->fs_root->node->start);
1908 		btrfs_set_backup_fs_root_gen(root_backup,
1909 			       btrfs_header_generation(info->fs_root->node));
1910 		btrfs_set_backup_fs_root_level(root_backup,
1911 			       btrfs_header_level(info->fs_root->node));
1912 	}
1913 
1914 	btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1915 	btrfs_set_backup_dev_root_gen(root_backup,
1916 			       btrfs_header_generation(info->dev_root->node));
1917 	btrfs_set_backup_dev_root_level(root_backup,
1918 				       btrfs_header_level(info->dev_root->node));
1919 
1920 	btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1921 	btrfs_set_backup_csum_root_gen(root_backup,
1922 			       btrfs_header_generation(info->csum_root->node));
1923 	btrfs_set_backup_csum_root_level(root_backup,
1924 			       btrfs_header_level(info->csum_root->node));
1925 
1926 	btrfs_set_backup_total_bytes(root_backup,
1927 			     btrfs_super_total_bytes(info->super_copy));
1928 	btrfs_set_backup_bytes_used(root_backup,
1929 			     btrfs_super_bytes_used(info->super_copy));
1930 	btrfs_set_backup_num_devices(root_backup,
1931 			     btrfs_super_num_devices(info->super_copy));
1932 
1933 	/*
1934 	 * if we don't copy this out to the super_copy, it won't get remembered
1935 	 * for the next commit
1936 	 */
1937 	memcpy(&info->super_copy->super_roots,
1938 	       &info->super_for_commit->super_roots,
1939 	       sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1940 }
1941 
1942 /*
1943  * this copies info out of the root backup array and back into
1944  * the in-memory super block.  It is meant to help iterate through
1945  * the array, so you send it the number of backups you've already
1946  * tried and the last backup index you used.
1947  *
1948  * this returns -1 when it has tried all the backups
1949  */
1950 static noinline int next_root_backup(struct btrfs_fs_info *info,
1951 				     struct btrfs_super_block *super,
1952 				     int *num_backups_tried, int *backup_index)
1953 {
1954 	struct btrfs_root_backup *root_backup;
1955 	int newest = *backup_index;
1956 
1957 	if (*num_backups_tried == 0) {
1958 		u64 gen = btrfs_super_generation(super);
1959 
1960 		newest = find_newest_super_backup(info, gen);
1961 		if (newest == -1)
1962 			return -1;
1963 
1964 		*backup_index = newest;
1965 		*num_backups_tried = 1;
1966 	} else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1967 		/* we've tried all the backups, all done */
1968 		return -1;
1969 	} else {
1970 		/* jump to the next oldest backup */
1971 		newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1972 			BTRFS_NUM_BACKUP_ROOTS;
1973 		*backup_index = newest;
1974 		*num_backups_tried += 1;
1975 	}
1976 	root_backup = super->super_roots + newest;
1977 
1978 	btrfs_set_super_generation(super,
1979 				   btrfs_backup_tree_root_gen(root_backup));
1980 	btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1981 	btrfs_set_super_root_level(super,
1982 				   btrfs_backup_tree_root_level(root_backup));
1983 	btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1984 
1985 	/*
1986 	 * fixme: the total bytes and num_devices need to match or we should
1987 	 * need a fsck
1988 	 */
1989 	btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1990 	btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1991 	return 0;
1992 }
1993 
1994 /* helper to cleanup workers */
1995 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1996 {
1997 	btrfs_stop_workers(&fs_info->generic_worker);
1998 	btrfs_stop_workers(&fs_info->fixup_workers);
1999 	btrfs_stop_workers(&fs_info->delalloc_workers);
2000 	btrfs_stop_workers(&fs_info->workers);
2001 	btrfs_stop_workers(&fs_info->endio_workers);
2002 	btrfs_stop_workers(&fs_info->endio_meta_workers);
2003 	btrfs_stop_workers(&fs_info->endio_raid56_workers);
2004 	btrfs_stop_workers(&fs_info->rmw_workers);
2005 	btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2006 	btrfs_stop_workers(&fs_info->endio_write_workers);
2007 	btrfs_stop_workers(&fs_info->endio_freespace_worker);
2008 	btrfs_stop_workers(&fs_info->submit_workers);
2009 	btrfs_stop_workers(&fs_info->delayed_workers);
2010 	btrfs_stop_workers(&fs_info->caching_workers);
2011 	btrfs_stop_workers(&fs_info->readahead_workers);
2012 	btrfs_stop_workers(&fs_info->flush_workers);
2013 	btrfs_stop_workers(&fs_info->qgroup_rescan_workers);
2014 }
2015 
2016 /* helper to cleanup tree roots */
2017 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2018 {
2019 	free_extent_buffer(info->tree_root->node);
2020 	free_extent_buffer(info->tree_root->commit_root);
2021 	info->tree_root->node = NULL;
2022 	info->tree_root->commit_root = NULL;
2023 
2024 	if (info->dev_root) {
2025 		free_extent_buffer(info->dev_root->node);
2026 		free_extent_buffer(info->dev_root->commit_root);
2027 		info->dev_root->node = NULL;
2028 		info->dev_root->commit_root = NULL;
2029 	}
2030 	if (info->extent_root) {
2031 		free_extent_buffer(info->extent_root->node);
2032 		free_extent_buffer(info->extent_root->commit_root);
2033 		info->extent_root->node = NULL;
2034 		info->extent_root->commit_root = NULL;
2035 	}
2036 	if (info->csum_root) {
2037 		free_extent_buffer(info->csum_root->node);
2038 		free_extent_buffer(info->csum_root->commit_root);
2039 		info->csum_root->node = NULL;
2040 		info->csum_root->commit_root = NULL;
2041 	}
2042 	if (info->quota_root) {
2043 		free_extent_buffer(info->quota_root->node);
2044 		free_extent_buffer(info->quota_root->commit_root);
2045 		info->quota_root->node = NULL;
2046 		info->quota_root->commit_root = NULL;
2047 	}
2048 	if (info->uuid_root) {
2049 		free_extent_buffer(info->uuid_root->node);
2050 		free_extent_buffer(info->uuid_root->commit_root);
2051 		info->uuid_root->node = NULL;
2052 		info->uuid_root->commit_root = NULL;
2053 	}
2054 	if (chunk_root) {
2055 		free_extent_buffer(info->chunk_root->node);
2056 		free_extent_buffer(info->chunk_root->commit_root);
2057 		info->chunk_root->node = NULL;
2058 		info->chunk_root->commit_root = NULL;
2059 	}
2060 }
2061 
2062 static void del_fs_roots(struct btrfs_fs_info *fs_info)
2063 {
2064 	int ret;
2065 	struct btrfs_root *gang[8];
2066 	int i;
2067 
2068 	while (!list_empty(&fs_info->dead_roots)) {
2069 		gang[0] = list_entry(fs_info->dead_roots.next,
2070 				     struct btrfs_root, root_list);
2071 		list_del(&gang[0]->root_list);
2072 
2073 		if (gang[0]->in_radix) {
2074 			btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2075 		} else {
2076 			free_extent_buffer(gang[0]->node);
2077 			free_extent_buffer(gang[0]->commit_root);
2078 			btrfs_put_fs_root(gang[0]);
2079 		}
2080 	}
2081 
2082 	while (1) {
2083 		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2084 					     (void **)gang, 0,
2085 					     ARRAY_SIZE(gang));
2086 		if (!ret)
2087 			break;
2088 		for (i = 0; i < ret; i++)
2089 			btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2090 	}
2091 }
2092 
2093 int open_ctree(struct super_block *sb,
2094 	       struct btrfs_fs_devices *fs_devices,
2095 	       char *options)
2096 {
2097 	u32 sectorsize;
2098 	u32 nodesize;
2099 	u32 leafsize;
2100 	u32 blocksize;
2101 	u32 stripesize;
2102 	u64 generation;
2103 	u64 features;
2104 	struct btrfs_key location;
2105 	struct buffer_head *bh;
2106 	struct btrfs_super_block *disk_super;
2107 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2108 	struct btrfs_root *tree_root;
2109 	struct btrfs_root *extent_root;
2110 	struct btrfs_root *csum_root;
2111 	struct btrfs_root *chunk_root;
2112 	struct btrfs_root *dev_root;
2113 	struct btrfs_root *quota_root;
2114 	struct btrfs_root *uuid_root;
2115 	struct btrfs_root *log_tree_root;
2116 	int ret;
2117 	int err = -EINVAL;
2118 	int num_backups_tried = 0;
2119 	int backup_index = 0;
2120 	bool create_uuid_tree;
2121 	bool check_uuid_tree;
2122 
2123 	tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2124 	chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2125 	if (!tree_root || !chunk_root) {
2126 		err = -ENOMEM;
2127 		goto fail;
2128 	}
2129 
2130 	ret = init_srcu_struct(&fs_info->subvol_srcu);
2131 	if (ret) {
2132 		err = ret;
2133 		goto fail;
2134 	}
2135 
2136 	ret = setup_bdi(fs_info, &fs_info->bdi);
2137 	if (ret) {
2138 		err = ret;
2139 		goto fail_srcu;
2140 	}
2141 
2142 	ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2143 	if (ret) {
2144 		err = ret;
2145 		goto fail_bdi;
2146 	}
2147 	fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2148 					(1 + ilog2(nr_cpu_ids));
2149 
2150 	ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2151 	if (ret) {
2152 		err = ret;
2153 		goto fail_dirty_metadata_bytes;
2154 	}
2155 
2156 	fs_info->btree_inode = new_inode(sb);
2157 	if (!fs_info->btree_inode) {
2158 		err = -ENOMEM;
2159 		goto fail_delalloc_bytes;
2160 	}
2161 
2162 	mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2163 
2164 	INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2165 	INIT_LIST_HEAD(&fs_info->trans_list);
2166 	INIT_LIST_HEAD(&fs_info->dead_roots);
2167 	INIT_LIST_HEAD(&fs_info->delayed_iputs);
2168 	INIT_LIST_HEAD(&fs_info->delalloc_roots);
2169 	INIT_LIST_HEAD(&fs_info->caching_block_groups);
2170 	spin_lock_init(&fs_info->delalloc_root_lock);
2171 	spin_lock_init(&fs_info->trans_lock);
2172 	spin_lock_init(&fs_info->fs_roots_radix_lock);
2173 	spin_lock_init(&fs_info->delayed_iput_lock);
2174 	spin_lock_init(&fs_info->defrag_inodes_lock);
2175 	spin_lock_init(&fs_info->free_chunk_lock);
2176 	spin_lock_init(&fs_info->tree_mod_seq_lock);
2177 	spin_lock_init(&fs_info->super_lock);
2178 	rwlock_init(&fs_info->tree_mod_log_lock);
2179 	mutex_init(&fs_info->reloc_mutex);
2180 	seqlock_init(&fs_info->profiles_lock);
2181 
2182 	init_completion(&fs_info->kobj_unregister);
2183 	INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2184 	INIT_LIST_HEAD(&fs_info->space_info);
2185 	INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2186 	btrfs_mapping_init(&fs_info->mapping_tree);
2187 	btrfs_init_block_rsv(&fs_info->global_block_rsv,
2188 			     BTRFS_BLOCK_RSV_GLOBAL);
2189 	btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2190 			     BTRFS_BLOCK_RSV_DELALLOC);
2191 	btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2192 	btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2193 	btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2194 	btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2195 			     BTRFS_BLOCK_RSV_DELOPS);
2196 	atomic_set(&fs_info->nr_async_submits, 0);
2197 	atomic_set(&fs_info->async_delalloc_pages, 0);
2198 	atomic_set(&fs_info->async_submit_draining, 0);
2199 	atomic_set(&fs_info->nr_async_bios, 0);
2200 	atomic_set(&fs_info->defrag_running, 0);
2201 	atomic64_set(&fs_info->tree_mod_seq, 0);
2202 	fs_info->sb = sb;
2203 	fs_info->max_inline = 8192 * 1024;
2204 	fs_info->metadata_ratio = 0;
2205 	fs_info->defrag_inodes = RB_ROOT;
2206 	fs_info->free_chunk_space = 0;
2207 	fs_info->tree_mod_log = RB_ROOT;
2208 	fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2209 
2210 	/* readahead state */
2211 	INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2212 	spin_lock_init(&fs_info->reada_lock);
2213 
2214 	fs_info->thread_pool_size = min_t(unsigned long,
2215 					  num_online_cpus() + 2, 8);
2216 
2217 	INIT_LIST_HEAD(&fs_info->ordered_roots);
2218 	spin_lock_init(&fs_info->ordered_root_lock);
2219 	fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2220 					GFP_NOFS);
2221 	if (!fs_info->delayed_root) {
2222 		err = -ENOMEM;
2223 		goto fail_iput;
2224 	}
2225 	btrfs_init_delayed_root(fs_info->delayed_root);
2226 
2227 	mutex_init(&fs_info->scrub_lock);
2228 	atomic_set(&fs_info->scrubs_running, 0);
2229 	atomic_set(&fs_info->scrub_pause_req, 0);
2230 	atomic_set(&fs_info->scrubs_paused, 0);
2231 	atomic_set(&fs_info->scrub_cancel_req, 0);
2232 	init_waitqueue_head(&fs_info->scrub_pause_wait);
2233 	init_rwsem(&fs_info->scrub_super_lock);
2234 	fs_info->scrub_workers_refcnt = 0;
2235 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2236 	fs_info->check_integrity_print_mask = 0;
2237 #endif
2238 
2239 	spin_lock_init(&fs_info->balance_lock);
2240 	mutex_init(&fs_info->balance_mutex);
2241 	atomic_set(&fs_info->balance_running, 0);
2242 	atomic_set(&fs_info->balance_pause_req, 0);
2243 	atomic_set(&fs_info->balance_cancel_req, 0);
2244 	fs_info->balance_ctl = NULL;
2245 	init_waitqueue_head(&fs_info->balance_wait_q);
2246 
2247 	sb->s_blocksize = 4096;
2248 	sb->s_blocksize_bits = blksize_bits(4096);
2249 	sb->s_bdi = &fs_info->bdi;
2250 
2251 	fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2252 	set_nlink(fs_info->btree_inode, 1);
2253 	/*
2254 	 * we set the i_size on the btree inode to the max possible int.
2255 	 * the real end of the address space is determined by all of
2256 	 * the devices in the system
2257 	 */
2258 	fs_info->btree_inode->i_size = OFFSET_MAX;
2259 	fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2260 	fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2261 
2262 	RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2263 	extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2264 			     fs_info->btree_inode->i_mapping);
2265 	BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2266 	extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2267 
2268 	BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2269 
2270 	BTRFS_I(fs_info->btree_inode)->root = tree_root;
2271 	memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2272 	       sizeof(struct btrfs_key));
2273 	set_bit(BTRFS_INODE_DUMMY,
2274 		&BTRFS_I(fs_info->btree_inode)->runtime_flags);
2275 	insert_inode_hash(fs_info->btree_inode);
2276 
2277 	spin_lock_init(&fs_info->block_group_cache_lock);
2278 	fs_info->block_group_cache_tree = RB_ROOT;
2279 	fs_info->first_logical_byte = (u64)-1;
2280 
2281 	extent_io_tree_init(&fs_info->freed_extents[0],
2282 			     fs_info->btree_inode->i_mapping);
2283 	extent_io_tree_init(&fs_info->freed_extents[1],
2284 			     fs_info->btree_inode->i_mapping);
2285 	fs_info->pinned_extents = &fs_info->freed_extents[0];
2286 	fs_info->do_barriers = 1;
2287 
2288 
2289 	mutex_init(&fs_info->ordered_operations_mutex);
2290 	mutex_init(&fs_info->ordered_extent_flush_mutex);
2291 	mutex_init(&fs_info->tree_log_mutex);
2292 	mutex_init(&fs_info->chunk_mutex);
2293 	mutex_init(&fs_info->transaction_kthread_mutex);
2294 	mutex_init(&fs_info->cleaner_mutex);
2295 	mutex_init(&fs_info->volume_mutex);
2296 	init_rwsem(&fs_info->extent_commit_sem);
2297 	init_rwsem(&fs_info->cleanup_work_sem);
2298 	init_rwsem(&fs_info->subvol_sem);
2299 	sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2300 	fs_info->dev_replace.lock_owner = 0;
2301 	atomic_set(&fs_info->dev_replace.nesting_level, 0);
2302 	mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2303 	mutex_init(&fs_info->dev_replace.lock_management_lock);
2304 	mutex_init(&fs_info->dev_replace.lock);
2305 
2306 	spin_lock_init(&fs_info->qgroup_lock);
2307 	mutex_init(&fs_info->qgroup_ioctl_lock);
2308 	fs_info->qgroup_tree = RB_ROOT;
2309 	INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2310 	fs_info->qgroup_seq = 1;
2311 	fs_info->quota_enabled = 0;
2312 	fs_info->pending_quota_state = 0;
2313 	fs_info->qgroup_ulist = NULL;
2314 	mutex_init(&fs_info->qgroup_rescan_lock);
2315 
2316 	btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2317 	btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2318 
2319 	init_waitqueue_head(&fs_info->transaction_throttle);
2320 	init_waitqueue_head(&fs_info->transaction_wait);
2321 	init_waitqueue_head(&fs_info->transaction_blocked_wait);
2322 	init_waitqueue_head(&fs_info->async_submit_wait);
2323 
2324 	ret = btrfs_alloc_stripe_hash_table(fs_info);
2325 	if (ret) {
2326 		err = ret;
2327 		goto fail_alloc;
2328 	}
2329 
2330 	__setup_root(4096, 4096, 4096, 4096, tree_root,
2331 		     fs_info, BTRFS_ROOT_TREE_OBJECTID);
2332 
2333 	invalidate_bdev(fs_devices->latest_bdev);
2334 
2335 	/*
2336 	 * Read super block and check the signature bytes only
2337 	 */
2338 	bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2339 	if (!bh) {
2340 		err = -EINVAL;
2341 		goto fail_alloc;
2342 	}
2343 
2344 	/*
2345 	 * We want to check superblock checksum, the type is stored inside.
2346 	 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2347 	 */
2348 	if (btrfs_check_super_csum(bh->b_data)) {
2349 		printk(KERN_ERR "btrfs: superblock checksum mismatch\n");
2350 		err = -EINVAL;
2351 		goto fail_alloc;
2352 	}
2353 
2354 	/*
2355 	 * super_copy is zeroed at allocation time and we never touch the
2356 	 * following bytes up to INFO_SIZE, the checksum is calculated from
2357 	 * the whole block of INFO_SIZE
2358 	 */
2359 	memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2360 	memcpy(fs_info->super_for_commit, fs_info->super_copy,
2361 	       sizeof(*fs_info->super_for_commit));
2362 	brelse(bh);
2363 
2364 	memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2365 
2366 	ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2367 	if (ret) {
2368 		printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2369 		err = -EINVAL;
2370 		goto fail_alloc;
2371 	}
2372 
2373 	disk_super = fs_info->super_copy;
2374 	if (!btrfs_super_root(disk_super))
2375 		goto fail_alloc;
2376 
2377 	/* check FS state, whether FS is broken. */
2378 	if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2379 		set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2380 
2381 	/*
2382 	 * run through our array of backup supers and setup
2383 	 * our ring pointer to the oldest one
2384 	 */
2385 	generation = btrfs_super_generation(disk_super);
2386 	find_oldest_super_backup(fs_info, generation);
2387 
2388 	/*
2389 	 * In the long term, we'll store the compression type in the super
2390 	 * block, and it'll be used for per file compression control.
2391 	 */
2392 	fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2393 
2394 	ret = btrfs_parse_options(tree_root, options);
2395 	if (ret) {
2396 		err = ret;
2397 		goto fail_alloc;
2398 	}
2399 
2400 	features = btrfs_super_incompat_flags(disk_super) &
2401 		~BTRFS_FEATURE_INCOMPAT_SUPP;
2402 	if (features) {
2403 		printk(KERN_ERR "BTRFS: couldn't mount because of "
2404 		       "unsupported optional features (%Lx).\n",
2405 		       features);
2406 		err = -EINVAL;
2407 		goto fail_alloc;
2408 	}
2409 
2410 	if (btrfs_super_leafsize(disk_super) !=
2411 	    btrfs_super_nodesize(disk_super)) {
2412 		printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2413 		       "blocksizes don't match.  node %d leaf %d\n",
2414 		       btrfs_super_nodesize(disk_super),
2415 		       btrfs_super_leafsize(disk_super));
2416 		err = -EINVAL;
2417 		goto fail_alloc;
2418 	}
2419 	if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2420 		printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2421 		       "blocksize (%d) was too large\n",
2422 		       btrfs_super_leafsize(disk_super));
2423 		err = -EINVAL;
2424 		goto fail_alloc;
2425 	}
2426 
2427 	features = btrfs_super_incompat_flags(disk_super);
2428 	features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2429 	if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2430 		features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2431 
2432 	if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2433 		printk(KERN_ERR "btrfs: has skinny extents\n");
2434 
2435 	/*
2436 	 * flag our filesystem as having big metadata blocks if
2437 	 * they are bigger than the page size
2438 	 */
2439 	if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2440 		if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2441 			printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2442 		features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2443 	}
2444 
2445 	nodesize = btrfs_super_nodesize(disk_super);
2446 	leafsize = btrfs_super_leafsize(disk_super);
2447 	sectorsize = btrfs_super_sectorsize(disk_super);
2448 	stripesize = btrfs_super_stripesize(disk_super);
2449 	fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
2450 	fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2451 
2452 	/*
2453 	 * mixed block groups end up with duplicate but slightly offset
2454 	 * extent buffers for the same range.  It leads to corruptions
2455 	 */
2456 	if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2457 	    (sectorsize != leafsize)) {
2458 		printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2459 				"are not allowed for mixed block groups on %s\n",
2460 				sb->s_id);
2461 		goto fail_alloc;
2462 	}
2463 
2464 	/*
2465 	 * Needn't use the lock because there is no other task which will
2466 	 * update the flag.
2467 	 */
2468 	btrfs_set_super_incompat_flags(disk_super, features);
2469 
2470 	features = btrfs_super_compat_ro_flags(disk_super) &
2471 		~BTRFS_FEATURE_COMPAT_RO_SUPP;
2472 	if (!(sb->s_flags & MS_RDONLY) && features) {
2473 		printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2474 		       "unsupported option features (%Lx).\n",
2475 		       features);
2476 		err = -EINVAL;
2477 		goto fail_alloc;
2478 	}
2479 
2480 	btrfs_init_workers(&fs_info->generic_worker,
2481 			   "genwork", 1, NULL);
2482 
2483 	btrfs_init_workers(&fs_info->workers, "worker",
2484 			   fs_info->thread_pool_size,
2485 			   &fs_info->generic_worker);
2486 
2487 	btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2488 			   fs_info->thread_pool_size, NULL);
2489 
2490 	btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc",
2491 			   fs_info->thread_pool_size, NULL);
2492 
2493 	btrfs_init_workers(&fs_info->submit_workers, "submit",
2494 			   min_t(u64, fs_devices->num_devices,
2495 			   fs_info->thread_pool_size), NULL);
2496 
2497 	btrfs_init_workers(&fs_info->caching_workers, "cache",
2498 			   fs_info->thread_pool_size, NULL);
2499 
2500 	/* a higher idle thresh on the submit workers makes it much more
2501 	 * likely that bios will be send down in a sane order to the
2502 	 * devices
2503 	 */
2504 	fs_info->submit_workers.idle_thresh = 64;
2505 
2506 	fs_info->workers.idle_thresh = 16;
2507 	fs_info->workers.ordered = 1;
2508 
2509 	fs_info->delalloc_workers.idle_thresh = 2;
2510 	fs_info->delalloc_workers.ordered = 1;
2511 
2512 	btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2513 			   &fs_info->generic_worker);
2514 	btrfs_init_workers(&fs_info->endio_workers, "endio",
2515 			   fs_info->thread_pool_size,
2516 			   &fs_info->generic_worker);
2517 	btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2518 			   fs_info->thread_pool_size,
2519 			   &fs_info->generic_worker);
2520 	btrfs_init_workers(&fs_info->endio_meta_write_workers,
2521 			   "endio-meta-write", fs_info->thread_pool_size,
2522 			   &fs_info->generic_worker);
2523 	btrfs_init_workers(&fs_info->endio_raid56_workers,
2524 			   "endio-raid56", fs_info->thread_pool_size,
2525 			   &fs_info->generic_worker);
2526 	btrfs_init_workers(&fs_info->rmw_workers,
2527 			   "rmw", fs_info->thread_pool_size,
2528 			   &fs_info->generic_worker);
2529 	btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2530 			   fs_info->thread_pool_size,
2531 			   &fs_info->generic_worker);
2532 	btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2533 			   1, &fs_info->generic_worker);
2534 	btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2535 			   fs_info->thread_pool_size,
2536 			   &fs_info->generic_worker);
2537 	btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2538 			   fs_info->thread_pool_size,
2539 			   &fs_info->generic_worker);
2540 	btrfs_init_workers(&fs_info->qgroup_rescan_workers, "qgroup-rescan", 1,
2541 			   &fs_info->generic_worker);
2542 
2543 	/*
2544 	 * endios are largely parallel and should have a very
2545 	 * low idle thresh
2546 	 */
2547 	fs_info->endio_workers.idle_thresh = 4;
2548 	fs_info->endio_meta_workers.idle_thresh = 4;
2549 	fs_info->endio_raid56_workers.idle_thresh = 4;
2550 	fs_info->rmw_workers.idle_thresh = 2;
2551 
2552 	fs_info->endio_write_workers.idle_thresh = 2;
2553 	fs_info->endio_meta_write_workers.idle_thresh = 2;
2554 	fs_info->readahead_workers.idle_thresh = 2;
2555 
2556 	/*
2557 	 * btrfs_start_workers can really only fail because of ENOMEM so just
2558 	 * return -ENOMEM if any of these fail.
2559 	 */
2560 	ret = btrfs_start_workers(&fs_info->workers);
2561 	ret |= btrfs_start_workers(&fs_info->generic_worker);
2562 	ret |= btrfs_start_workers(&fs_info->submit_workers);
2563 	ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2564 	ret |= btrfs_start_workers(&fs_info->fixup_workers);
2565 	ret |= btrfs_start_workers(&fs_info->endio_workers);
2566 	ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2567 	ret |= btrfs_start_workers(&fs_info->rmw_workers);
2568 	ret |= btrfs_start_workers(&fs_info->endio_raid56_workers);
2569 	ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2570 	ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2571 	ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2572 	ret |= btrfs_start_workers(&fs_info->delayed_workers);
2573 	ret |= btrfs_start_workers(&fs_info->caching_workers);
2574 	ret |= btrfs_start_workers(&fs_info->readahead_workers);
2575 	ret |= btrfs_start_workers(&fs_info->flush_workers);
2576 	ret |= btrfs_start_workers(&fs_info->qgroup_rescan_workers);
2577 	if (ret) {
2578 		err = -ENOMEM;
2579 		goto fail_sb_buffer;
2580 	}
2581 
2582 	fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2583 	fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2584 				    4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2585 
2586 	tree_root->nodesize = nodesize;
2587 	tree_root->leafsize = leafsize;
2588 	tree_root->sectorsize = sectorsize;
2589 	tree_root->stripesize = stripesize;
2590 
2591 	sb->s_blocksize = sectorsize;
2592 	sb->s_blocksize_bits = blksize_bits(sectorsize);
2593 
2594 	if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2595 		printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2596 		goto fail_sb_buffer;
2597 	}
2598 
2599 	if (sectorsize != PAGE_SIZE) {
2600 		printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2601 		       "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2602 		goto fail_sb_buffer;
2603 	}
2604 
2605 	mutex_lock(&fs_info->chunk_mutex);
2606 	ret = btrfs_read_sys_array(tree_root);
2607 	mutex_unlock(&fs_info->chunk_mutex);
2608 	if (ret) {
2609 		printk(KERN_WARNING "btrfs: failed to read the system "
2610 		       "array on %s\n", sb->s_id);
2611 		goto fail_sb_buffer;
2612 	}
2613 
2614 	blocksize = btrfs_level_size(tree_root,
2615 				     btrfs_super_chunk_root_level(disk_super));
2616 	generation = btrfs_super_chunk_root_generation(disk_super);
2617 
2618 	__setup_root(nodesize, leafsize, sectorsize, stripesize,
2619 		     chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2620 
2621 	chunk_root->node = read_tree_block(chunk_root,
2622 					   btrfs_super_chunk_root(disk_super),
2623 					   blocksize, generation);
2624 	if (!chunk_root->node ||
2625 	    !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2626 		printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2627 		       sb->s_id);
2628 		goto fail_tree_roots;
2629 	}
2630 	btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2631 	chunk_root->commit_root = btrfs_root_node(chunk_root);
2632 
2633 	read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2634 	   btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2635 
2636 	ret = btrfs_read_chunk_tree(chunk_root);
2637 	if (ret) {
2638 		printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2639 		       sb->s_id);
2640 		goto fail_tree_roots;
2641 	}
2642 
2643 	/*
2644 	 * keep the device that is marked to be the target device for the
2645 	 * dev_replace procedure
2646 	 */
2647 	btrfs_close_extra_devices(fs_info, fs_devices, 0);
2648 
2649 	if (!fs_devices->latest_bdev) {
2650 		printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2651 		       sb->s_id);
2652 		goto fail_tree_roots;
2653 	}
2654 
2655 retry_root_backup:
2656 	blocksize = btrfs_level_size(tree_root,
2657 				     btrfs_super_root_level(disk_super));
2658 	generation = btrfs_super_generation(disk_super);
2659 
2660 	tree_root->node = read_tree_block(tree_root,
2661 					  btrfs_super_root(disk_super),
2662 					  blocksize, generation);
2663 	if (!tree_root->node ||
2664 	    !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2665 		printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2666 		       sb->s_id);
2667 
2668 		goto recovery_tree_root;
2669 	}
2670 
2671 	btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2672 	tree_root->commit_root = btrfs_root_node(tree_root);
2673 
2674 	location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2675 	location.type = BTRFS_ROOT_ITEM_KEY;
2676 	location.offset = 0;
2677 
2678 	extent_root = btrfs_read_tree_root(tree_root, &location);
2679 	if (IS_ERR(extent_root)) {
2680 		ret = PTR_ERR(extent_root);
2681 		goto recovery_tree_root;
2682 	}
2683 	extent_root->track_dirty = 1;
2684 	fs_info->extent_root = extent_root;
2685 
2686 	location.objectid = BTRFS_DEV_TREE_OBJECTID;
2687 	dev_root = btrfs_read_tree_root(tree_root, &location);
2688 	if (IS_ERR(dev_root)) {
2689 		ret = PTR_ERR(dev_root);
2690 		goto recovery_tree_root;
2691 	}
2692 	dev_root->track_dirty = 1;
2693 	fs_info->dev_root = dev_root;
2694 	btrfs_init_devices_late(fs_info);
2695 
2696 	location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2697 	csum_root = btrfs_read_tree_root(tree_root, &location);
2698 	if (IS_ERR(csum_root)) {
2699 		ret = PTR_ERR(csum_root);
2700 		goto recovery_tree_root;
2701 	}
2702 	csum_root->track_dirty = 1;
2703 	fs_info->csum_root = csum_root;
2704 
2705 	location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2706 	quota_root = btrfs_read_tree_root(tree_root, &location);
2707 	if (!IS_ERR(quota_root)) {
2708 		quota_root->track_dirty = 1;
2709 		fs_info->quota_enabled = 1;
2710 		fs_info->pending_quota_state = 1;
2711 		fs_info->quota_root = quota_root;
2712 	}
2713 
2714 	location.objectid = BTRFS_UUID_TREE_OBJECTID;
2715 	uuid_root = btrfs_read_tree_root(tree_root, &location);
2716 	if (IS_ERR(uuid_root)) {
2717 		ret = PTR_ERR(uuid_root);
2718 		if (ret != -ENOENT)
2719 			goto recovery_tree_root;
2720 		create_uuid_tree = true;
2721 		check_uuid_tree = false;
2722 	} else {
2723 		uuid_root->track_dirty = 1;
2724 		fs_info->uuid_root = uuid_root;
2725 		create_uuid_tree = false;
2726 		check_uuid_tree =
2727 		    generation != btrfs_super_uuid_tree_generation(disk_super);
2728 	}
2729 
2730 	fs_info->generation = generation;
2731 	fs_info->last_trans_committed = generation;
2732 
2733 	ret = btrfs_recover_balance(fs_info);
2734 	if (ret) {
2735 		printk(KERN_WARNING "btrfs: failed to recover balance\n");
2736 		goto fail_block_groups;
2737 	}
2738 
2739 	ret = btrfs_init_dev_stats(fs_info);
2740 	if (ret) {
2741 		printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
2742 		       ret);
2743 		goto fail_block_groups;
2744 	}
2745 
2746 	ret = btrfs_init_dev_replace(fs_info);
2747 	if (ret) {
2748 		pr_err("btrfs: failed to init dev_replace: %d\n", ret);
2749 		goto fail_block_groups;
2750 	}
2751 
2752 	btrfs_close_extra_devices(fs_info, fs_devices, 1);
2753 
2754 	ret = btrfs_init_space_info(fs_info);
2755 	if (ret) {
2756 		printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2757 		goto fail_block_groups;
2758 	}
2759 
2760 	ret = btrfs_read_block_groups(extent_root);
2761 	if (ret) {
2762 		printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2763 		goto fail_block_groups;
2764 	}
2765 	fs_info->num_tolerated_disk_barrier_failures =
2766 		btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2767 	if (fs_info->fs_devices->missing_devices >
2768 	     fs_info->num_tolerated_disk_barrier_failures &&
2769 	    !(sb->s_flags & MS_RDONLY)) {
2770 		printk(KERN_WARNING
2771 		       "Btrfs: too many missing devices, writeable mount is not allowed\n");
2772 		goto fail_block_groups;
2773 	}
2774 
2775 	fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2776 					       "btrfs-cleaner");
2777 	if (IS_ERR(fs_info->cleaner_kthread))
2778 		goto fail_block_groups;
2779 
2780 	fs_info->transaction_kthread = kthread_run(transaction_kthread,
2781 						   tree_root,
2782 						   "btrfs-transaction");
2783 	if (IS_ERR(fs_info->transaction_kthread))
2784 		goto fail_cleaner;
2785 
2786 	if (!btrfs_test_opt(tree_root, SSD) &&
2787 	    !btrfs_test_opt(tree_root, NOSSD) &&
2788 	    !fs_info->fs_devices->rotating) {
2789 		printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2790 		       "mode\n");
2791 		btrfs_set_opt(fs_info->mount_opt, SSD);
2792 	}
2793 
2794 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2795 	if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2796 		ret = btrfsic_mount(tree_root, fs_devices,
2797 				    btrfs_test_opt(tree_root,
2798 					CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2799 				    1 : 0,
2800 				    fs_info->check_integrity_print_mask);
2801 		if (ret)
2802 			printk(KERN_WARNING "btrfs: failed to initialize"
2803 			       " integrity check module %s\n", sb->s_id);
2804 	}
2805 #endif
2806 	ret = btrfs_read_qgroup_config(fs_info);
2807 	if (ret)
2808 		goto fail_trans_kthread;
2809 
2810 	/* do not make disk changes in broken FS */
2811 	if (btrfs_super_log_root(disk_super) != 0) {
2812 		u64 bytenr = btrfs_super_log_root(disk_super);
2813 
2814 		if (fs_devices->rw_devices == 0) {
2815 			printk(KERN_WARNING "Btrfs log replay required "
2816 			       "on RO media\n");
2817 			err = -EIO;
2818 			goto fail_qgroup;
2819 		}
2820 		blocksize =
2821 		     btrfs_level_size(tree_root,
2822 				      btrfs_super_log_root_level(disk_super));
2823 
2824 		log_tree_root = btrfs_alloc_root(fs_info);
2825 		if (!log_tree_root) {
2826 			err = -ENOMEM;
2827 			goto fail_qgroup;
2828 		}
2829 
2830 		__setup_root(nodesize, leafsize, sectorsize, stripesize,
2831 			     log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2832 
2833 		log_tree_root->node = read_tree_block(tree_root, bytenr,
2834 						      blocksize,
2835 						      generation + 1);
2836 		if (!log_tree_root->node ||
2837 		    !extent_buffer_uptodate(log_tree_root->node)) {
2838 			printk(KERN_ERR "btrfs: failed to read log tree\n");
2839 			free_extent_buffer(log_tree_root->node);
2840 			kfree(log_tree_root);
2841 			goto fail_trans_kthread;
2842 		}
2843 		/* returns with log_tree_root freed on success */
2844 		ret = btrfs_recover_log_trees(log_tree_root);
2845 		if (ret) {
2846 			btrfs_error(tree_root->fs_info, ret,
2847 				    "Failed to recover log tree");
2848 			free_extent_buffer(log_tree_root->node);
2849 			kfree(log_tree_root);
2850 			goto fail_trans_kthread;
2851 		}
2852 
2853 		if (sb->s_flags & MS_RDONLY) {
2854 			ret = btrfs_commit_super(tree_root);
2855 			if (ret)
2856 				goto fail_trans_kthread;
2857 		}
2858 	}
2859 
2860 	ret = btrfs_find_orphan_roots(tree_root);
2861 	if (ret)
2862 		goto fail_trans_kthread;
2863 
2864 	if (!(sb->s_flags & MS_RDONLY)) {
2865 		ret = btrfs_cleanup_fs_roots(fs_info);
2866 		if (ret)
2867 			goto fail_trans_kthread;
2868 
2869 		ret = btrfs_recover_relocation(tree_root);
2870 		if (ret < 0) {
2871 			printk(KERN_WARNING
2872 			       "btrfs: failed to recover relocation\n");
2873 			err = -EINVAL;
2874 			goto fail_qgroup;
2875 		}
2876 	}
2877 
2878 	location.objectid = BTRFS_FS_TREE_OBJECTID;
2879 	location.type = BTRFS_ROOT_ITEM_KEY;
2880 	location.offset = 0;
2881 
2882 	fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2883 	if (IS_ERR(fs_info->fs_root)) {
2884 		err = PTR_ERR(fs_info->fs_root);
2885 		goto fail_qgroup;
2886 	}
2887 
2888 	if (sb->s_flags & MS_RDONLY)
2889 		return 0;
2890 
2891 	down_read(&fs_info->cleanup_work_sem);
2892 	if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2893 	    (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2894 		up_read(&fs_info->cleanup_work_sem);
2895 		close_ctree(tree_root);
2896 		return ret;
2897 	}
2898 	up_read(&fs_info->cleanup_work_sem);
2899 
2900 	ret = btrfs_resume_balance_async(fs_info);
2901 	if (ret) {
2902 		printk(KERN_WARNING "btrfs: failed to resume balance\n");
2903 		close_ctree(tree_root);
2904 		return ret;
2905 	}
2906 
2907 	ret = btrfs_resume_dev_replace_async(fs_info);
2908 	if (ret) {
2909 		pr_warn("btrfs: failed to resume dev_replace\n");
2910 		close_ctree(tree_root);
2911 		return ret;
2912 	}
2913 
2914 	btrfs_qgroup_rescan_resume(fs_info);
2915 
2916 	if (create_uuid_tree) {
2917 		pr_info("btrfs: creating UUID tree\n");
2918 		ret = btrfs_create_uuid_tree(fs_info);
2919 		if (ret) {
2920 			pr_warn("btrfs: failed to create the UUID tree %d\n",
2921 				ret);
2922 			close_ctree(tree_root);
2923 			return ret;
2924 		}
2925 	} else if (check_uuid_tree ||
2926 		   btrfs_test_opt(tree_root, RESCAN_UUID_TREE)) {
2927 		pr_info("btrfs: checking UUID tree\n");
2928 		ret = btrfs_check_uuid_tree(fs_info);
2929 		if (ret) {
2930 			pr_warn("btrfs: failed to check the UUID tree %d\n",
2931 				ret);
2932 			close_ctree(tree_root);
2933 			return ret;
2934 		}
2935 	} else {
2936 		fs_info->update_uuid_tree_gen = 1;
2937 	}
2938 
2939 	return 0;
2940 
2941 fail_qgroup:
2942 	btrfs_free_qgroup_config(fs_info);
2943 fail_trans_kthread:
2944 	kthread_stop(fs_info->transaction_kthread);
2945 	btrfs_cleanup_transaction(fs_info->tree_root);
2946 	del_fs_roots(fs_info);
2947 fail_cleaner:
2948 	kthread_stop(fs_info->cleaner_kthread);
2949 
2950 	/*
2951 	 * make sure we're done with the btree inode before we stop our
2952 	 * kthreads
2953 	 */
2954 	filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2955 
2956 fail_block_groups:
2957 	btrfs_put_block_group_cache(fs_info);
2958 	btrfs_free_block_groups(fs_info);
2959 
2960 fail_tree_roots:
2961 	free_root_pointers(fs_info, 1);
2962 	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2963 
2964 fail_sb_buffer:
2965 	btrfs_stop_all_workers(fs_info);
2966 fail_alloc:
2967 fail_iput:
2968 	btrfs_mapping_tree_free(&fs_info->mapping_tree);
2969 
2970 	iput(fs_info->btree_inode);
2971 fail_delalloc_bytes:
2972 	percpu_counter_destroy(&fs_info->delalloc_bytes);
2973 fail_dirty_metadata_bytes:
2974 	percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
2975 fail_bdi:
2976 	bdi_destroy(&fs_info->bdi);
2977 fail_srcu:
2978 	cleanup_srcu_struct(&fs_info->subvol_srcu);
2979 fail:
2980 	btrfs_free_stripe_hash_table(fs_info);
2981 	btrfs_close_devices(fs_info->fs_devices);
2982 	return err;
2983 
2984 recovery_tree_root:
2985 	if (!btrfs_test_opt(tree_root, RECOVERY))
2986 		goto fail_tree_roots;
2987 
2988 	free_root_pointers(fs_info, 0);
2989 
2990 	/* don't use the log in recovery mode, it won't be valid */
2991 	btrfs_set_super_log_root(disk_super, 0);
2992 
2993 	/* we can't trust the free space cache either */
2994 	btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2995 
2996 	ret = next_root_backup(fs_info, fs_info->super_copy,
2997 			       &num_backups_tried, &backup_index);
2998 	if (ret == -1)
2999 		goto fail_block_groups;
3000 	goto retry_root_backup;
3001 }
3002 
3003 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3004 {
3005 	if (uptodate) {
3006 		set_buffer_uptodate(bh);
3007 	} else {
3008 		struct btrfs_device *device = (struct btrfs_device *)
3009 			bh->b_private;
3010 
3011 		printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
3012 					  "I/O error on %s\n",
3013 					  rcu_str_deref(device->name));
3014 		/* note, we dont' set_buffer_write_io_error because we have
3015 		 * our own ways of dealing with the IO errors
3016 		 */
3017 		clear_buffer_uptodate(bh);
3018 		btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3019 	}
3020 	unlock_buffer(bh);
3021 	put_bh(bh);
3022 }
3023 
3024 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3025 {
3026 	struct buffer_head *bh;
3027 	struct buffer_head *latest = NULL;
3028 	struct btrfs_super_block *super;
3029 	int i;
3030 	u64 transid = 0;
3031 	u64 bytenr;
3032 
3033 	/* we would like to check all the supers, but that would make
3034 	 * a btrfs mount succeed after a mkfs from a different FS.
3035 	 * So, we need to add a special mount option to scan for
3036 	 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3037 	 */
3038 	for (i = 0; i < 1; i++) {
3039 		bytenr = btrfs_sb_offset(i);
3040 		if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3041 					i_size_read(bdev->bd_inode))
3042 			break;
3043 		bh = __bread(bdev, bytenr / 4096,
3044 					BTRFS_SUPER_INFO_SIZE);
3045 		if (!bh)
3046 			continue;
3047 
3048 		super = (struct btrfs_super_block *)bh->b_data;
3049 		if (btrfs_super_bytenr(super) != bytenr ||
3050 		    btrfs_super_magic(super) != BTRFS_MAGIC) {
3051 			brelse(bh);
3052 			continue;
3053 		}
3054 
3055 		if (!latest || btrfs_super_generation(super) > transid) {
3056 			brelse(latest);
3057 			latest = bh;
3058 			transid = btrfs_super_generation(super);
3059 		} else {
3060 			brelse(bh);
3061 		}
3062 	}
3063 	return latest;
3064 }
3065 
3066 /*
3067  * this should be called twice, once with wait == 0 and
3068  * once with wait == 1.  When wait == 0 is done, all the buffer heads
3069  * we write are pinned.
3070  *
3071  * They are released when wait == 1 is done.
3072  * max_mirrors must be the same for both runs, and it indicates how
3073  * many supers on this one device should be written.
3074  *
3075  * max_mirrors == 0 means to write them all.
3076  */
3077 static int write_dev_supers(struct btrfs_device *device,
3078 			    struct btrfs_super_block *sb,
3079 			    int do_barriers, int wait, int max_mirrors)
3080 {
3081 	struct buffer_head *bh;
3082 	int i;
3083 	int ret;
3084 	int errors = 0;
3085 	u32 crc;
3086 	u64 bytenr;
3087 
3088 	if (max_mirrors == 0)
3089 		max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3090 
3091 	for (i = 0; i < max_mirrors; i++) {
3092 		bytenr = btrfs_sb_offset(i);
3093 		if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3094 			break;
3095 
3096 		if (wait) {
3097 			bh = __find_get_block(device->bdev, bytenr / 4096,
3098 					      BTRFS_SUPER_INFO_SIZE);
3099 			if (!bh) {
3100 				errors++;
3101 				continue;
3102 			}
3103 			wait_on_buffer(bh);
3104 			if (!buffer_uptodate(bh))
3105 				errors++;
3106 
3107 			/* drop our reference */
3108 			brelse(bh);
3109 
3110 			/* drop the reference from the wait == 0 run */
3111 			brelse(bh);
3112 			continue;
3113 		} else {
3114 			btrfs_set_super_bytenr(sb, bytenr);
3115 
3116 			crc = ~(u32)0;
3117 			crc = btrfs_csum_data((char *)sb +
3118 					      BTRFS_CSUM_SIZE, crc,
3119 					      BTRFS_SUPER_INFO_SIZE -
3120 					      BTRFS_CSUM_SIZE);
3121 			btrfs_csum_final(crc, sb->csum);
3122 
3123 			/*
3124 			 * one reference for us, and we leave it for the
3125 			 * caller
3126 			 */
3127 			bh = __getblk(device->bdev, bytenr / 4096,
3128 				      BTRFS_SUPER_INFO_SIZE);
3129 			if (!bh) {
3130 				printk(KERN_ERR "btrfs: couldn't get super "
3131 				       "buffer head for bytenr %Lu\n", bytenr);
3132 				errors++;
3133 				continue;
3134 			}
3135 
3136 			memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3137 
3138 			/* one reference for submit_bh */
3139 			get_bh(bh);
3140 
3141 			set_buffer_uptodate(bh);
3142 			lock_buffer(bh);
3143 			bh->b_end_io = btrfs_end_buffer_write_sync;
3144 			bh->b_private = device;
3145 		}
3146 
3147 		/*
3148 		 * we fua the first super.  The others we allow
3149 		 * to go down lazy.
3150 		 */
3151 		ret = btrfsic_submit_bh(WRITE_FUA, bh);
3152 		if (ret)
3153 			errors++;
3154 	}
3155 	return errors < i ? 0 : -1;
3156 }
3157 
3158 /*
3159  * endio for the write_dev_flush, this will wake anyone waiting
3160  * for the barrier when it is done
3161  */
3162 static void btrfs_end_empty_barrier(struct bio *bio, int err)
3163 {
3164 	if (err) {
3165 		if (err == -EOPNOTSUPP)
3166 			set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3167 		clear_bit(BIO_UPTODATE, &bio->bi_flags);
3168 	}
3169 	if (bio->bi_private)
3170 		complete(bio->bi_private);
3171 	bio_put(bio);
3172 }
3173 
3174 /*
3175  * trigger flushes for one the devices.  If you pass wait == 0, the flushes are
3176  * sent down.  With wait == 1, it waits for the previous flush.
3177  *
3178  * any device where the flush fails with eopnotsupp are flagged as not-barrier
3179  * capable
3180  */
3181 static int write_dev_flush(struct btrfs_device *device, int wait)
3182 {
3183 	struct bio *bio;
3184 	int ret = 0;
3185 
3186 	if (device->nobarriers)
3187 		return 0;
3188 
3189 	if (wait) {
3190 		bio = device->flush_bio;
3191 		if (!bio)
3192 			return 0;
3193 
3194 		wait_for_completion(&device->flush_wait);
3195 
3196 		if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3197 			printk_in_rcu("btrfs: disabling barriers on dev %s\n",
3198 				      rcu_str_deref(device->name));
3199 			device->nobarriers = 1;
3200 		} else if (!bio_flagged(bio, BIO_UPTODATE)) {
3201 			ret = -EIO;
3202 			btrfs_dev_stat_inc_and_print(device,
3203 				BTRFS_DEV_STAT_FLUSH_ERRS);
3204 		}
3205 
3206 		/* drop the reference from the wait == 0 run */
3207 		bio_put(bio);
3208 		device->flush_bio = NULL;
3209 
3210 		return ret;
3211 	}
3212 
3213 	/*
3214 	 * one reference for us, and we leave it for the
3215 	 * caller
3216 	 */
3217 	device->flush_bio = NULL;
3218 	bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3219 	if (!bio)
3220 		return -ENOMEM;
3221 
3222 	bio->bi_end_io = btrfs_end_empty_barrier;
3223 	bio->bi_bdev = device->bdev;
3224 	init_completion(&device->flush_wait);
3225 	bio->bi_private = &device->flush_wait;
3226 	device->flush_bio = bio;
3227 
3228 	bio_get(bio);
3229 	btrfsic_submit_bio(WRITE_FLUSH, bio);
3230 
3231 	return 0;
3232 }
3233 
3234 /*
3235  * send an empty flush down to each device in parallel,
3236  * then wait for them
3237  */
3238 static int barrier_all_devices(struct btrfs_fs_info *info)
3239 {
3240 	struct list_head *head;
3241 	struct btrfs_device *dev;
3242 	int errors_send = 0;
3243 	int errors_wait = 0;
3244 	int ret;
3245 
3246 	/* send down all the barriers */
3247 	head = &info->fs_devices->devices;
3248 	list_for_each_entry_rcu(dev, head, dev_list) {
3249 		if (!dev->bdev) {
3250 			errors_send++;
3251 			continue;
3252 		}
3253 		if (!dev->in_fs_metadata || !dev->writeable)
3254 			continue;
3255 
3256 		ret = write_dev_flush(dev, 0);
3257 		if (ret)
3258 			errors_send++;
3259 	}
3260 
3261 	/* wait for all the barriers */
3262 	list_for_each_entry_rcu(dev, head, dev_list) {
3263 		if (!dev->bdev) {
3264 			errors_wait++;
3265 			continue;
3266 		}
3267 		if (!dev->in_fs_metadata || !dev->writeable)
3268 			continue;
3269 
3270 		ret = write_dev_flush(dev, 1);
3271 		if (ret)
3272 			errors_wait++;
3273 	}
3274 	if (errors_send > info->num_tolerated_disk_barrier_failures ||
3275 	    errors_wait > info->num_tolerated_disk_barrier_failures)
3276 		return -EIO;
3277 	return 0;
3278 }
3279 
3280 int btrfs_calc_num_tolerated_disk_barrier_failures(
3281 	struct btrfs_fs_info *fs_info)
3282 {
3283 	struct btrfs_ioctl_space_info space;
3284 	struct btrfs_space_info *sinfo;
3285 	u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3286 		       BTRFS_BLOCK_GROUP_SYSTEM,
3287 		       BTRFS_BLOCK_GROUP_METADATA,
3288 		       BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3289 	int num_types = 4;
3290 	int i;
3291 	int c;
3292 	int num_tolerated_disk_barrier_failures =
3293 		(int)fs_info->fs_devices->num_devices;
3294 
3295 	for (i = 0; i < num_types; i++) {
3296 		struct btrfs_space_info *tmp;
3297 
3298 		sinfo = NULL;
3299 		rcu_read_lock();
3300 		list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3301 			if (tmp->flags == types[i]) {
3302 				sinfo = tmp;
3303 				break;
3304 			}
3305 		}
3306 		rcu_read_unlock();
3307 
3308 		if (!sinfo)
3309 			continue;
3310 
3311 		down_read(&sinfo->groups_sem);
3312 		for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3313 			if (!list_empty(&sinfo->block_groups[c])) {
3314 				u64 flags;
3315 
3316 				btrfs_get_block_group_info(
3317 					&sinfo->block_groups[c], &space);
3318 				if (space.total_bytes == 0 ||
3319 				    space.used_bytes == 0)
3320 					continue;
3321 				flags = space.flags;
3322 				/*
3323 				 * return
3324 				 * 0: if dup, single or RAID0 is configured for
3325 				 *    any of metadata, system or data, else
3326 				 * 1: if RAID5 is configured, or if RAID1 or
3327 				 *    RAID10 is configured and only two mirrors
3328 				 *    are used, else
3329 				 * 2: if RAID6 is configured, else
3330 				 * num_mirrors - 1: if RAID1 or RAID10 is
3331 				 *                  configured and more than
3332 				 *                  2 mirrors are used.
3333 				 */
3334 				if (num_tolerated_disk_barrier_failures > 0 &&
3335 				    ((flags & (BTRFS_BLOCK_GROUP_DUP |
3336 					       BTRFS_BLOCK_GROUP_RAID0)) ||
3337 				     ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3338 				      == 0)))
3339 					num_tolerated_disk_barrier_failures = 0;
3340 				else if (num_tolerated_disk_barrier_failures > 1) {
3341 					if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3342 					    BTRFS_BLOCK_GROUP_RAID5 |
3343 					    BTRFS_BLOCK_GROUP_RAID10)) {
3344 						num_tolerated_disk_barrier_failures = 1;
3345 					} else if (flags &
3346 						   BTRFS_BLOCK_GROUP_RAID6) {
3347 						num_tolerated_disk_barrier_failures = 2;
3348 					}
3349 				}
3350 			}
3351 		}
3352 		up_read(&sinfo->groups_sem);
3353 	}
3354 
3355 	return num_tolerated_disk_barrier_failures;
3356 }
3357 
3358 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3359 {
3360 	struct list_head *head;
3361 	struct btrfs_device *dev;
3362 	struct btrfs_super_block *sb;
3363 	struct btrfs_dev_item *dev_item;
3364 	int ret;
3365 	int do_barriers;
3366 	int max_errors;
3367 	int total_errors = 0;
3368 	u64 flags;
3369 
3370 	do_barriers = !btrfs_test_opt(root, NOBARRIER);
3371 	backup_super_roots(root->fs_info);
3372 
3373 	sb = root->fs_info->super_for_commit;
3374 	dev_item = &sb->dev_item;
3375 
3376 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3377 	head = &root->fs_info->fs_devices->devices;
3378 	max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3379 
3380 	if (do_barriers) {
3381 		ret = barrier_all_devices(root->fs_info);
3382 		if (ret) {
3383 			mutex_unlock(
3384 				&root->fs_info->fs_devices->device_list_mutex);
3385 			btrfs_error(root->fs_info, ret,
3386 				    "errors while submitting device barriers.");
3387 			return ret;
3388 		}
3389 	}
3390 
3391 	list_for_each_entry_rcu(dev, head, dev_list) {
3392 		if (!dev->bdev) {
3393 			total_errors++;
3394 			continue;
3395 		}
3396 		if (!dev->in_fs_metadata || !dev->writeable)
3397 			continue;
3398 
3399 		btrfs_set_stack_device_generation(dev_item, 0);
3400 		btrfs_set_stack_device_type(dev_item, dev->type);
3401 		btrfs_set_stack_device_id(dev_item, dev->devid);
3402 		btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3403 		btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3404 		btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3405 		btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3406 		btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3407 		memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3408 		memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3409 
3410 		flags = btrfs_super_flags(sb);
3411 		btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3412 
3413 		ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3414 		if (ret)
3415 			total_errors++;
3416 	}
3417 	if (total_errors > max_errors) {
3418 		printk(KERN_ERR "btrfs: %d errors while writing supers\n",
3419 		       total_errors);
3420 		mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3421 
3422 		/* FUA is masked off if unsupported and can't be the reason */
3423 		btrfs_error(root->fs_info, -EIO,
3424 			    "%d errors while writing supers", total_errors);
3425 		return -EIO;
3426 	}
3427 
3428 	total_errors = 0;
3429 	list_for_each_entry_rcu(dev, head, dev_list) {
3430 		if (!dev->bdev)
3431 			continue;
3432 		if (!dev->in_fs_metadata || !dev->writeable)
3433 			continue;
3434 
3435 		ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3436 		if (ret)
3437 			total_errors++;
3438 	}
3439 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3440 	if (total_errors > max_errors) {
3441 		btrfs_error(root->fs_info, -EIO,
3442 			    "%d errors while writing supers", total_errors);
3443 		return -EIO;
3444 	}
3445 	return 0;
3446 }
3447 
3448 int write_ctree_super(struct btrfs_trans_handle *trans,
3449 		      struct btrfs_root *root, int max_mirrors)
3450 {
3451 	int ret;
3452 
3453 	ret = write_all_supers(root, max_mirrors);
3454 	return ret;
3455 }
3456 
3457 /* Drop a fs root from the radix tree and free it. */
3458 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3459 				  struct btrfs_root *root)
3460 {
3461 	spin_lock(&fs_info->fs_roots_radix_lock);
3462 	radix_tree_delete(&fs_info->fs_roots_radix,
3463 			  (unsigned long)root->root_key.objectid);
3464 	spin_unlock(&fs_info->fs_roots_radix_lock);
3465 
3466 	if (btrfs_root_refs(&root->root_item) == 0)
3467 		synchronize_srcu(&fs_info->subvol_srcu);
3468 
3469 	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3470 		btrfs_free_log(NULL, root);
3471 		btrfs_free_log_root_tree(NULL, fs_info);
3472 	}
3473 
3474 	__btrfs_remove_free_space_cache(root->free_ino_pinned);
3475 	__btrfs_remove_free_space_cache(root->free_ino_ctl);
3476 	free_fs_root(root);
3477 }
3478 
3479 static void free_fs_root(struct btrfs_root *root)
3480 {
3481 	iput(root->cache_inode);
3482 	WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3483 	btrfs_free_block_rsv(root, root->orphan_block_rsv);
3484 	root->orphan_block_rsv = NULL;
3485 	if (root->anon_dev)
3486 		free_anon_bdev(root->anon_dev);
3487 	free_extent_buffer(root->node);
3488 	free_extent_buffer(root->commit_root);
3489 	kfree(root->free_ino_ctl);
3490 	kfree(root->free_ino_pinned);
3491 	kfree(root->name);
3492 	btrfs_put_fs_root(root);
3493 }
3494 
3495 void btrfs_free_fs_root(struct btrfs_root *root)
3496 {
3497 	free_fs_root(root);
3498 }
3499 
3500 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3501 {
3502 	u64 root_objectid = 0;
3503 	struct btrfs_root *gang[8];
3504 	int i;
3505 	int ret;
3506 
3507 	while (1) {
3508 		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3509 					     (void **)gang, root_objectid,
3510 					     ARRAY_SIZE(gang));
3511 		if (!ret)
3512 			break;
3513 
3514 		root_objectid = gang[ret - 1]->root_key.objectid + 1;
3515 		for (i = 0; i < ret; i++) {
3516 			int err;
3517 
3518 			root_objectid = gang[i]->root_key.objectid;
3519 			err = btrfs_orphan_cleanup(gang[i]);
3520 			if (err)
3521 				return err;
3522 		}
3523 		root_objectid++;
3524 	}
3525 	return 0;
3526 }
3527 
3528 int btrfs_commit_super(struct btrfs_root *root)
3529 {
3530 	struct btrfs_trans_handle *trans;
3531 	int ret;
3532 
3533 	mutex_lock(&root->fs_info->cleaner_mutex);
3534 	btrfs_run_delayed_iputs(root);
3535 	mutex_unlock(&root->fs_info->cleaner_mutex);
3536 	wake_up_process(root->fs_info->cleaner_kthread);
3537 
3538 	/* wait until ongoing cleanup work done */
3539 	down_write(&root->fs_info->cleanup_work_sem);
3540 	up_write(&root->fs_info->cleanup_work_sem);
3541 
3542 	trans = btrfs_join_transaction(root);
3543 	if (IS_ERR(trans))
3544 		return PTR_ERR(trans);
3545 	ret = btrfs_commit_transaction(trans, root);
3546 	if (ret)
3547 		return ret;
3548 	/* run commit again to drop the original snapshot */
3549 	trans = btrfs_join_transaction(root);
3550 	if (IS_ERR(trans))
3551 		return PTR_ERR(trans);
3552 	ret = btrfs_commit_transaction(trans, root);
3553 	if (ret)
3554 		return ret;
3555 	ret = btrfs_write_and_wait_transaction(NULL, root);
3556 	if (ret) {
3557 		btrfs_error(root->fs_info, ret,
3558 			    "Failed to sync btree inode to disk.");
3559 		return ret;
3560 	}
3561 
3562 	ret = write_ctree_super(NULL, root, 0);
3563 	return ret;
3564 }
3565 
3566 int close_ctree(struct btrfs_root *root)
3567 {
3568 	struct btrfs_fs_info *fs_info = root->fs_info;
3569 	int ret;
3570 
3571 	fs_info->closing = 1;
3572 	smp_mb();
3573 
3574 	/* wait for the uuid_scan task to finish */
3575 	down(&fs_info->uuid_tree_rescan_sem);
3576 	/* avoid complains from lockdep et al., set sem back to initial state */
3577 	up(&fs_info->uuid_tree_rescan_sem);
3578 
3579 	/* pause restriper - we want to resume on mount */
3580 	btrfs_pause_balance(fs_info);
3581 
3582 	btrfs_dev_replace_suspend_for_unmount(fs_info);
3583 
3584 	btrfs_scrub_cancel(fs_info);
3585 
3586 	/* wait for any defraggers to finish */
3587 	wait_event(fs_info->transaction_wait,
3588 		   (atomic_read(&fs_info->defrag_running) == 0));
3589 
3590 	/* clear out the rbtree of defraggable inodes */
3591 	btrfs_cleanup_defrag_inodes(fs_info);
3592 
3593 	if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3594 		ret = btrfs_commit_super(root);
3595 		if (ret)
3596 			printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3597 	}
3598 
3599 	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3600 		btrfs_error_commit_super(root);
3601 
3602 	btrfs_put_block_group_cache(fs_info);
3603 
3604 	kthread_stop(fs_info->transaction_kthread);
3605 	kthread_stop(fs_info->cleaner_kthread);
3606 
3607 	fs_info->closing = 2;
3608 	smp_mb();
3609 
3610 	btrfs_free_qgroup_config(root->fs_info);
3611 
3612 	if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3613 		printk(KERN_INFO "btrfs: at unmount delalloc count %lld\n",
3614 		       percpu_counter_sum(&fs_info->delalloc_bytes));
3615 	}
3616 
3617 	btrfs_free_block_groups(fs_info);
3618 
3619 	btrfs_stop_all_workers(fs_info);
3620 
3621 	del_fs_roots(fs_info);
3622 
3623 	free_root_pointers(fs_info, 1);
3624 
3625 	iput(fs_info->btree_inode);
3626 
3627 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3628 	if (btrfs_test_opt(root, CHECK_INTEGRITY))
3629 		btrfsic_unmount(root, fs_info->fs_devices);
3630 #endif
3631 
3632 	btrfs_close_devices(fs_info->fs_devices);
3633 	btrfs_mapping_tree_free(&fs_info->mapping_tree);
3634 
3635 	percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3636 	percpu_counter_destroy(&fs_info->delalloc_bytes);
3637 	bdi_destroy(&fs_info->bdi);
3638 	cleanup_srcu_struct(&fs_info->subvol_srcu);
3639 
3640 	btrfs_free_stripe_hash_table(fs_info);
3641 
3642 	btrfs_free_block_rsv(root, root->orphan_block_rsv);
3643 	root->orphan_block_rsv = NULL;
3644 
3645 	return 0;
3646 }
3647 
3648 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3649 			  int atomic)
3650 {
3651 	int ret;
3652 	struct inode *btree_inode = buf->pages[0]->mapping->host;
3653 
3654 	ret = extent_buffer_uptodate(buf);
3655 	if (!ret)
3656 		return ret;
3657 
3658 	ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3659 				    parent_transid, atomic);
3660 	if (ret == -EAGAIN)
3661 		return ret;
3662 	return !ret;
3663 }
3664 
3665 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3666 {
3667 	return set_extent_buffer_uptodate(buf);
3668 }
3669 
3670 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3671 {
3672 	struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3673 	u64 transid = btrfs_header_generation(buf);
3674 	int was_dirty;
3675 
3676 	btrfs_assert_tree_locked(buf);
3677 	if (transid != root->fs_info->generation)
3678 		WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3679 		       "found %llu running %llu\n",
3680 			buf->start, transid, root->fs_info->generation);
3681 	was_dirty = set_extent_buffer_dirty(buf);
3682 	if (!was_dirty)
3683 		__percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3684 				     buf->len,
3685 				     root->fs_info->dirty_metadata_batch);
3686 }
3687 
3688 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3689 					int flush_delayed)
3690 {
3691 	/*
3692 	 * looks as though older kernels can get into trouble with
3693 	 * this code, they end up stuck in balance_dirty_pages forever
3694 	 */
3695 	int ret;
3696 
3697 	if (current->flags & PF_MEMALLOC)
3698 		return;
3699 
3700 	if (flush_delayed)
3701 		btrfs_balance_delayed_items(root);
3702 
3703 	ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3704 				     BTRFS_DIRTY_METADATA_THRESH);
3705 	if (ret > 0) {
3706 		balance_dirty_pages_ratelimited(
3707 				   root->fs_info->btree_inode->i_mapping);
3708 	}
3709 	return;
3710 }
3711 
3712 void btrfs_btree_balance_dirty(struct btrfs_root *root)
3713 {
3714 	__btrfs_btree_balance_dirty(root, 1);
3715 }
3716 
3717 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3718 {
3719 	__btrfs_btree_balance_dirty(root, 0);
3720 }
3721 
3722 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3723 {
3724 	struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3725 	return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3726 }
3727 
3728 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3729 			      int read_only)
3730 {
3731 	/*
3732 	 * Placeholder for checks
3733 	 */
3734 	return 0;
3735 }
3736 
3737 static void btrfs_error_commit_super(struct btrfs_root *root)
3738 {
3739 	mutex_lock(&root->fs_info->cleaner_mutex);
3740 	btrfs_run_delayed_iputs(root);
3741 	mutex_unlock(&root->fs_info->cleaner_mutex);
3742 
3743 	down_write(&root->fs_info->cleanup_work_sem);
3744 	up_write(&root->fs_info->cleanup_work_sem);
3745 
3746 	/* cleanup FS via transaction */
3747 	btrfs_cleanup_transaction(root);
3748 }
3749 
3750 static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3751 					     struct btrfs_root *root)
3752 {
3753 	struct btrfs_inode *btrfs_inode;
3754 	struct list_head splice;
3755 
3756 	INIT_LIST_HEAD(&splice);
3757 
3758 	mutex_lock(&root->fs_info->ordered_operations_mutex);
3759 	spin_lock(&root->fs_info->ordered_root_lock);
3760 
3761 	list_splice_init(&t->ordered_operations, &splice);
3762 	while (!list_empty(&splice)) {
3763 		btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3764 					 ordered_operations);
3765 
3766 		list_del_init(&btrfs_inode->ordered_operations);
3767 		spin_unlock(&root->fs_info->ordered_root_lock);
3768 
3769 		btrfs_invalidate_inodes(btrfs_inode->root);
3770 
3771 		spin_lock(&root->fs_info->ordered_root_lock);
3772 	}
3773 
3774 	spin_unlock(&root->fs_info->ordered_root_lock);
3775 	mutex_unlock(&root->fs_info->ordered_operations_mutex);
3776 }
3777 
3778 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3779 {
3780 	struct btrfs_ordered_extent *ordered;
3781 
3782 	spin_lock(&root->ordered_extent_lock);
3783 	/*
3784 	 * This will just short circuit the ordered completion stuff which will
3785 	 * make sure the ordered extent gets properly cleaned up.
3786 	 */
3787 	list_for_each_entry(ordered, &root->ordered_extents,
3788 			    root_extent_list)
3789 		set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
3790 	spin_unlock(&root->ordered_extent_lock);
3791 }
3792 
3793 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3794 {
3795 	struct btrfs_root *root;
3796 	struct list_head splice;
3797 
3798 	INIT_LIST_HEAD(&splice);
3799 
3800 	spin_lock(&fs_info->ordered_root_lock);
3801 	list_splice_init(&fs_info->ordered_roots, &splice);
3802 	while (!list_empty(&splice)) {
3803 		root = list_first_entry(&splice, struct btrfs_root,
3804 					ordered_root);
3805 		list_del_init(&root->ordered_root);
3806 
3807 		btrfs_destroy_ordered_extents(root);
3808 
3809 		cond_resched_lock(&fs_info->ordered_root_lock);
3810 	}
3811 	spin_unlock(&fs_info->ordered_root_lock);
3812 }
3813 
3814 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3815 				      struct btrfs_root *root)
3816 {
3817 	struct rb_node *node;
3818 	struct btrfs_delayed_ref_root *delayed_refs;
3819 	struct btrfs_delayed_ref_node *ref;
3820 	int ret = 0;
3821 
3822 	delayed_refs = &trans->delayed_refs;
3823 
3824 	spin_lock(&delayed_refs->lock);
3825 	if (delayed_refs->num_entries == 0) {
3826 		spin_unlock(&delayed_refs->lock);
3827 		printk(KERN_INFO "delayed_refs has NO entry\n");
3828 		return ret;
3829 	}
3830 
3831 	while ((node = rb_first(&delayed_refs->root)) != NULL) {
3832 		struct btrfs_delayed_ref_head *head = NULL;
3833 		bool pin_bytes = false;
3834 
3835 		ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3836 		atomic_set(&ref->refs, 1);
3837 		if (btrfs_delayed_ref_is_head(ref)) {
3838 
3839 			head = btrfs_delayed_node_to_head(ref);
3840 			if (!mutex_trylock(&head->mutex)) {
3841 				atomic_inc(&ref->refs);
3842 				spin_unlock(&delayed_refs->lock);
3843 
3844 				/* Need to wait for the delayed ref to run */
3845 				mutex_lock(&head->mutex);
3846 				mutex_unlock(&head->mutex);
3847 				btrfs_put_delayed_ref(ref);
3848 
3849 				spin_lock(&delayed_refs->lock);
3850 				continue;
3851 			}
3852 
3853 			if (head->must_insert_reserved)
3854 				pin_bytes = true;
3855 			btrfs_free_delayed_extent_op(head->extent_op);
3856 			delayed_refs->num_heads--;
3857 			if (list_empty(&head->cluster))
3858 				delayed_refs->num_heads_ready--;
3859 			list_del_init(&head->cluster);
3860 		}
3861 
3862 		ref->in_tree = 0;
3863 		rb_erase(&ref->rb_node, &delayed_refs->root);
3864 		delayed_refs->num_entries--;
3865 		spin_unlock(&delayed_refs->lock);
3866 		if (head) {
3867 			if (pin_bytes)
3868 				btrfs_pin_extent(root, ref->bytenr,
3869 						 ref->num_bytes, 1);
3870 			mutex_unlock(&head->mutex);
3871 		}
3872 		btrfs_put_delayed_ref(ref);
3873 
3874 		cond_resched();
3875 		spin_lock(&delayed_refs->lock);
3876 	}
3877 
3878 	spin_unlock(&delayed_refs->lock);
3879 
3880 	return ret;
3881 }
3882 
3883 static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t)
3884 {
3885 	struct btrfs_pending_snapshot *snapshot;
3886 	struct list_head splice;
3887 
3888 	INIT_LIST_HEAD(&splice);
3889 
3890 	list_splice_init(&t->pending_snapshots, &splice);
3891 
3892 	while (!list_empty(&splice)) {
3893 		snapshot = list_entry(splice.next,
3894 				      struct btrfs_pending_snapshot,
3895 				      list);
3896 		snapshot->error = -ECANCELED;
3897 		list_del_init(&snapshot->list);
3898 	}
3899 }
3900 
3901 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3902 {
3903 	struct btrfs_inode *btrfs_inode;
3904 	struct list_head splice;
3905 
3906 	INIT_LIST_HEAD(&splice);
3907 
3908 	spin_lock(&root->delalloc_lock);
3909 	list_splice_init(&root->delalloc_inodes, &splice);
3910 
3911 	while (!list_empty(&splice)) {
3912 		btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3913 					       delalloc_inodes);
3914 
3915 		list_del_init(&btrfs_inode->delalloc_inodes);
3916 		clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3917 			  &btrfs_inode->runtime_flags);
3918 		spin_unlock(&root->delalloc_lock);
3919 
3920 		btrfs_invalidate_inodes(btrfs_inode->root);
3921 
3922 		spin_lock(&root->delalloc_lock);
3923 	}
3924 
3925 	spin_unlock(&root->delalloc_lock);
3926 }
3927 
3928 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3929 {
3930 	struct btrfs_root *root;
3931 	struct list_head splice;
3932 
3933 	INIT_LIST_HEAD(&splice);
3934 
3935 	spin_lock(&fs_info->delalloc_root_lock);
3936 	list_splice_init(&fs_info->delalloc_roots, &splice);
3937 	while (!list_empty(&splice)) {
3938 		root = list_first_entry(&splice, struct btrfs_root,
3939 					 delalloc_root);
3940 		list_del_init(&root->delalloc_root);
3941 		root = btrfs_grab_fs_root(root);
3942 		BUG_ON(!root);
3943 		spin_unlock(&fs_info->delalloc_root_lock);
3944 
3945 		btrfs_destroy_delalloc_inodes(root);
3946 		btrfs_put_fs_root(root);
3947 
3948 		spin_lock(&fs_info->delalloc_root_lock);
3949 	}
3950 	spin_unlock(&fs_info->delalloc_root_lock);
3951 }
3952 
3953 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3954 					struct extent_io_tree *dirty_pages,
3955 					int mark)
3956 {
3957 	int ret;
3958 	struct extent_buffer *eb;
3959 	u64 start = 0;
3960 	u64 end;
3961 
3962 	while (1) {
3963 		ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3964 					    mark, NULL);
3965 		if (ret)
3966 			break;
3967 
3968 		clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3969 		while (start <= end) {
3970 			eb = btrfs_find_tree_block(root, start,
3971 						   root->leafsize);
3972 			start += root->leafsize;
3973 			if (!eb)
3974 				continue;
3975 			wait_on_extent_buffer_writeback(eb);
3976 
3977 			if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3978 					       &eb->bflags))
3979 				clear_extent_buffer_dirty(eb);
3980 			free_extent_buffer_stale(eb);
3981 		}
3982 	}
3983 
3984 	return ret;
3985 }
3986 
3987 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3988 				       struct extent_io_tree *pinned_extents)
3989 {
3990 	struct extent_io_tree *unpin;
3991 	u64 start;
3992 	u64 end;
3993 	int ret;
3994 	bool loop = true;
3995 
3996 	unpin = pinned_extents;
3997 again:
3998 	while (1) {
3999 		ret = find_first_extent_bit(unpin, 0, &start, &end,
4000 					    EXTENT_DIRTY, NULL);
4001 		if (ret)
4002 			break;
4003 
4004 		/* opt_discard */
4005 		if (btrfs_test_opt(root, DISCARD))
4006 			ret = btrfs_error_discard_extent(root, start,
4007 							 end + 1 - start,
4008 							 NULL);
4009 
4010 		clear_extent_dirty(unpin, start, end, GFP_NOFS);
4011 		btrfs_error_unpin_extent_range(root, start, end);
4012 		cond_resched();
4013 	}
4014 
4015 	if (loop) {
4016 		if (unpin == &root->fs_info->freed_extents[0])
4017 			unpin = &root->fs_info->freed_extents[1];
4018 		else
4019 			unpin = &root->fs_info->freed_extents[0];
4020 		loop = false;
4021 		goto again;
4022 	}
4023 
4024 	return 0;
4025 }
4026 
4027 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4028 				   struct btrfs_root *root)
4029 {
4030 	btrfs_destroy_delayed_refs(cur_trans, root);
4031 	btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
4032 				cur_trans->dirty_pages.dirty_bytes);
4033 
4034 	cur_trans->state = TRANS_STATE_COMMIT_START;
4035 	wake_up(&root->fs_info->transaction_blocked_wait);
4036 
4037 	btrfs_evict_pending_snapshots(cur_trans);
4038 
4039 	cur_trans->state = TRANS_STATE_UNBLOCKED;
4040 	wake_up(&root->fs_info->transaction_wait);
4041 
4042 	btrfs_destroy_delayed_inodes(root);
4043 	btrfs_assert_delayed_root_empty(root);
4044 
4045 	btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4046 				     EXTENT_DIRTY);
4047 	btrfs_destroy_pinned_extent(root,
4048 				    root->fs_info->pinned_extents);
4049 
4050 	cur_trans->state =TRANS_STATE_COMPLETED;
4051 	wake_up(&cur_trans->commit_wait);
4052 
4053 	/*
4054 	memset(cur_trans, 0, sizeof(*cur_trans));
4055 	kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4056 	*/
4057 }
4058 
4059 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4060 {
4061 	struct btrfs_transaction *t;
4062 	LIST_HEAD(list);
4063 
4064 	mutex_lock(&root->fs_info->transaction_kthread_mutex);
4065 
4066 	spin_lock(&root->fs_info->trans_lock);
4067 	list_splice_init(&root->fs_info->trans_list, &list);
4068 	root->fs_info->running_transaction = NULL;
4069 	spin_unlock(&root->fs_info->trans_lock);
4070 
4071 	while (!list_empty(&list)) {
4072 		t = list_entry(list.next, struct btrfs_transaction, list);
4073 
4074 		btrfs_destroy_ordered_operations(t, root);
4075 
4076 		btrfs_destroy_all_ordered_extents(root->fs_info);
4077 
4078 		btrfs_destroy_delayed_refs(t, root);
4079 
4080 		/*
4081 		 *  FIXME: cleanup wait for commit
4082 		 *  We needn't acquire the lock here, because we are during
4083 		 *  the umount, there is no other task which will change it.
4084 		 */
4085 		t->state = TRANS_STATE_COMMIT_START;
4086 		smp_mb();
4087 		if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
4088 			wake_up(&root->fs_info->transaction_blocked_wait);
4089 
4090 		btrfs_evict_pending_snapshots(t);
4091 
4092 		t->state = TRANS_STATE_UNBLOCKED;
4093 		smp_mb();
4094 		if (waitqueue_active(&root->fs_info->transaction_wait))
4095 			wake_up(&root->fs_info->transaction_wait);
4096 
4097 		btrfs_destroy_delayed_inodes(root);
4098 		btrfs_assert_delayed_root_empty(root);
4099 
4100 		btrfs_destroy_all_delalloc_inodes(root->fs_info);
4101 
4102 		btrfs_destroy_marked_extents(root, &t->dirty_pages,
4103 					     EXTENT_DIRTY);
4104 
4105 		btrfs_destroy_pinned_extent(root,
4106 					    root->fs_info->pinned_extents);
4107 
4108 		t->state = TRANS_STATE_COMPLETED;
4109 		smp_mb();
4110 		if (waitqueue_active(&t->commit_wait))
4111 			wake_up(&t->commit_wait);
4112 
4113 		atomic_set(&t->use_count, 0);
4114 		list_del_init(&t->list);
4115 		memset(t, 0, sizeof(*t));
4116 		kmem_cache_free(btrfs_transaction_cachep, t);
4117 	}
4118 
4119 	mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4120 
4121 	return 0;
4122 }
4123 
4124 static struct extent_io_ops btree_extent_io_ops = {
4125 	.readpage_end_io_hook = btree_readpage_end_io_hook,
4126 	.readpage_io_failed_hook = btree_io_failed_hook,
4127 	.submit_bio_hook = btree_submit_bio_hook,
4128 	/* note we're sharing with inode.c for the merge bio hook */
4129 	.merge_bio_hook = btrfs_merge_bio_hook,
4130 };
4131