xref: /openbmc/linux/fs/btrfs/disk-io.c (revision 5bd8e16d)
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_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1565 					      struct btrfs_key *location)
1566 {
1567 	struct btrfs_root *root;
1568 	int ret;
1569 
1570 	if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1571 		return fs_info->tree_root;
1572 	if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1573 		return fs_info->extent_root;
1574 	if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1575 		return fs_info->chunk_root;
1576 	if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1577 		return fs_info->dev_root;
1578 	if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1579 		return fs_info->csum_root;
1580 	if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1581 		return fs_info->quota_root ? fs_info->quota_root :
1582 					     ERR_PTR(-ENOENT);
1583 	if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1584 		return fs_info->uuid_root ? fs_info->uuid_root :
1585 					    ERR_PTR(-ENOENT);
1586 again:
1587 	root = btrfs_lookup_fs_root(fs_info, location->objectid);
1588 	if (root) {
1589 		if (btrfs_root_refs(&root->root_item) == 0)
1590 			return ERR_PTR(-ENOENT);
1591 		return root;
1592 	}
1593 
1594 	root = btrfs_read_fs_root(fs_info->tree_root, location);
1595 	if (IS_ERR(root))
1596 		return root;
1597 
1598 	if (btrfs_root_refs(&root->root_item) == 0) {
1599 		ret = -ENOENT;
1600 		goto fail;
1601 	}
1602 
1603 	ret = btrfs_init_fs_root(root);
1604 	if (ret)
1605 		goto fail;
1606 
1607 	ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1608 	if (ret < 0)
1609 		goto fail;
1610 	if (ret == 0)
1611 		root->orphan_item_inserted = 1;
1612 
1613 	ret = btrfs_insert_fs_root(fs_info, root);
1614 	if (ret) {
1615 		if (ret == -EEXIST) {
1616 			free_fs_root(root);
1617 			goto again;
1618 		}
1619 		goto fail;
1620 	}
1621 	return root;
1622 fail:
1623 	free_fs_root(root);
1624 	return ERR_PTR(ret);
1625 }
1626 
1627 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1628 {
1629 	struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1630 	int ret = 0;
1631 	struct btrfs_device *device;
1632 	struct backing_dev_info *bdi;
1633 
1634 	rcu_read_lock();
1635 	list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1636 		if (!device->bdev)
1637 			continue;
1638 		bdi = blk_get_backing_dev_info(device->bdev);
1639 		if (bdi && bdi_congested(bdi, bdi_bits)) {
1640 			ret = 1;
1641 			break;
1642 		}
1643 	}
1644 	rcu_read_unlock();
1645 	return ret;
1646 }
1647 
1648 /*
1649  * If this fails, caller must call bdi_destroy() to get rid of the
1650  * bdi again.
1651  */
1652 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1653 {
1654 	int err;
1655 
1656 	bdi->capabilities = BDI_CAP_MAP_COPY;
1657 	err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1658 	if (err)
1659 		return err;
1660 
1661 	bdi->ra_pages	= default_backing_dev_info.ra_pages;
1662 	bdi->congested_fn	= btrfs_congested_fn;
1663 	bdi->congested_data	= info;
1664 	return 0;
1665 }
1666 
1667 /*
1668  * called by the kthread helper functions to finally call the bio end_io
1669  * functions.  This is where read checksum verification actually happens
1670  */
1671 static void end_workqueue_fn(struct btrfs_work *work)
1672 {
1673 	struct bio *bio;
1674 	struct end_io_wq *end_io_wq;
1675 	struct btrfs_fs_info *fs_info;
1676 	int error;
1677 
1678 	end_io_wq = container_of(work, struct end_io_wq, work);
1679 	bio = end_io_wq->bio;
1680 	fs_info = end_io_wq->info;
1681 
1682 	error = end_io_wq->error;
1683 	bio->bi_private = end_io_wq->private;
1684 	bio->bi_end_io = end_io_wq->end_io;
1685 	kfree(end_io_wq);
1686 	bio_endio(bio, error);
1687 }
1688 
1689 static int cleaner_kthread(void *arg)
1690 {
1691 	struct btrfs_root *root = arg;
1692 	int again;
1693 
1694 	do {
1695 		again = 0;
1696 
1697 		/* Make the cleaner go to sleep early. */
1698 		if (btrfs_need_cleaner_sleep(root))
1699 			goto sleep;
1700 
1701 		if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1702 			goto sleep;
1703 
1704 		/*
1705 		 * Avoid the problem that we change the status of the fs
1706 		 * during the above check and trylock.
1707 		 */
1708 		if (btrfs_need_cleaner_sleep(root)) {
1709 			mutex_unlock(&root->fs_info->cleaner_mutex);
1710 			goto sleep;
1711 		}
1712 
1713 		btrfs_run_delayed_iputs(root);
1714 		again = btrfs_clean_one_deleted_snapshot(root);
1715 		mutex_unlock(&root->fs_info->cleaner_mutex);
1716 
1717 		/*
1718 		 * The defragger has dealt with the R/O remount and umount,
1719 		 * needn't do anything special here.
1720 		 */
1721 		btrfs_run_defrag_inodes(root->fs_info);
1722 sleep:
1723 		if (!try_to_freeze() && !again) {
1724 			set_current_state(TASK_INTERRUPTIBLE);
1725 			if (!kthread_should_stop())
1726 				schedule();
1727 			__set_current_state(TASK_RUNNING);
1728 		}
1729 	} while (!kthread_should_stop());
1730 	return 0;
1731 }
1732 
1733 static int transaction_kthread(void *arg)
1734 {
1735 	struct btrfs_root *root = arg;
1736 	struct btrfs_trans_handle *trans;
1737 	struct btrfs_transaction *cur;
1738 	u64 transid;
1739 	unsigned long now;
1740 	unsigned long delay;
1741 	bool cannot_commit;
1742 
1743 	do {
1744 		cannot_commit = false;
1745 		delay = HZ * root->fs_info->commit_interval;
1746 		mutex_lock(&root->fs_info->transaction_kthread_mutex);
1747 
1748 		spin_lock(&root->fs_info->trans_lock);
1749 		cur = root->fs_info->running_transaction;
1750 		if (!cur) {
1751 			spin_unlock(&root->fs_info->trans_lock);
1752 			goto sleep;
1753 		}
1754 
1755 		now = get_seconds();
1756 		if (cur->state < TRANS_STATE_BLOCKED &&
1757 		    (now < cur->start_time ||
1758 		     now - cur->start_time < root->fs_info->commit_interval)) {
1759 			spin_unlock(&root->fs_info->trans_lock);
1760 			delay = HZ * 5;
1761 			goto sleep;
1762 		}
1763 		transid = cur->transid;
1764 		spin_unlock(&root->fs_info->trans_lock);
1765 
1766 		/* If the file system is aborted, this will always fail. */
1767 		trans = btrfs_attach_transaction(root);
1768 		if (IS_ERR(trans)) {
1769 			if (PTR_ERR(trans) != -ENOENT)
1770 				cannot_commit = true;
1771 			goto sleep;
1772 		}
1773 		if (transid == trans->transid) {
1774 			btrfs_commit_transaction(trans, root);
1775 		} else {
1776 			btrfs_end_transaction(trans, root);
1777 		}
1778 sleep:
1779 		wake_up_process(root->fs_info->cleaner_kthread);
1780 		mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1781 
1782 		if (!try_to_freeze()) {
1783 			set_current_state(TASK_INTERRUPTIBLE);
1784 			if (!kthread_should_stop() &&
1785 			    (!btrfs_transaction_blocked(root->fs_info) ||
1786 			     cannot_commit))
1787 				schedule_timeout(delay);
1788 			__set_current_state(TASK_RUNNING);
1789 		}
1790 	} while (!kthread_should_stop());
1791 	return 0;
1792 }
1793 
1794 /*
1795  * this will find the highest generation in the array of
1796  * root backups.  The index of the highest array is returned,
1797  * or -1 if we can't find anything.
1798  *
1799  * We check to make sure the array is valid by comparing the
1800  * generation of the latest  root in the array with the generation
1801  * in the super block.  If they don't match we pitch it.
1802  */
1803 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1804 {
1805 	u64 cur;
1806 	int newest_index = -1;
1807 	struct btrfs_root_backup *root_backup;
1808 	int i;
1809 
1810 	for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1811 		root_backup = info->super_copy->super_roots + i;
1812 		cur = btrfs_backup_tree_root_gen(root_backup);
1813 		if (cur == newest_gen)
1814 			newest_index = i;
1815 	}
1816 
1817 	/* check to see if we actually wrapped around */
1818 	if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1819 		root_backup = info->super_copy->super_roots;
1820 		cur = btrfs_backup_tree_root_gen(root_backup);
1821 		if (cur == newest_gen)
1822 			newest_index = 0;
1823 	}
1824 	return newest_index;
1825 }
1826 
1827 
1828 /*
1829  * find the oldest backup so we know where to store new entries
1830  * in the backup array.  This will set the backup_root_index
1831  * field in the fs_info struct
1832  */
1833 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1834 				     u64 newest_gen)
1835 {
1836 	int newest_index = -1;
1837 
1838 	newest_index = find_newest_super_backup(info, newest_gen);
1839 	/* if there was garbage in there, just move along */
1840 	if (newest_index == -1) {
1841 		info->backup_root_index = 0;
1842 	} else {
1843 		info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1844 	}
1845 }
1846 
1847 /*
1848  * copy all the root pointers into the super backup array.
1849  * this will bump the backup pointer by one when it is
1850  * done
1851  */
1852 static void backup_super_roots(struct btrfs_fs_info *info)
1853 {
1854 	int next_backup;
1855 	struct btrfs_root_backup *root_backup;
1856 	int last_backup;
1857 
1858 	next_backup = info->backup_root_index;
1859 	last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1860 		BTRFS_NUM_BACKUP_ROOTS;
1861 
1862 	/*
1863 	 * just overwrite the last backup if we're at the same generation
1864 	 * this happens only at umount
1865 	 */
1866 	root_backup = info->super_for_commit->super_roots + last_backup;
1867 	if (btrfs_backup_tree_root_gen(root_backup) ==
1868 	    btrfs_header_generation(info->tree_root->node))
1869 		next_backup = last_backup;
1870 
1871 	root_backup = info->super_for_commit->super_roots + next_backup;
1872 
1873 	/*
1874 	 * make sure all of our padding and empty slots get zero filled
1875 	 * regardless of which ones we use today
1876 	 */
1877 	memset(root_backup, 0, sizeof(*root_backup));
1878 
1879 	info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1880 
1881 	btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1882 	btrfs_set_backup_tree_root_gen(root_backup,
1883 			       btrfs_header_generation(info->tree_root->node));
1884 
1885 	btrfs_set_backup_tree_root_level(root_backup,
1886 			       btrfs_header_level(info->tree_root->node));
1887 
1888 	btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1889 	btrfs_set_backup_chunk_root_gen(root_backup,
1890 			       btrfs_header_generation(info->chunk_root->node));
1891 	btrfs_set_backup_chunk_root_level(root_backup,
1892 			       btrfs_header_level(info->chunk_root->node));
1893 
1894 	btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1895 	btrfs_set_backup_extent_root_gen(root_backup,
1896 			       btrfs_header_generation(info->extent_root->node));
1897 	btrfs_set_backup_extent_root_level(root_backup,
1898 			       btrfs_header_level(info->extent_root->node));
1899 
1900 	/*
1901 	 * we might commit during log recovery, which happens before we set
1902 	 * the fs_root.  Make sure it is valid before we fill it in.
1903 	 */
1904 	if (info->fs_root && info->fs_root->node) {
1905 		btrfs_set_backup_fs_root(root_backup,
1906 					 info->fs_root->node->start);
1907 		btrfs_set_backup_fs_root_gen(root_backup,
1908 			       btrfs_header_generation(info->fs_root->node));
1909 		btrfs_set_backup_fs_root_level(root_backup,
1910 			       btrfs_header_level(info->fs_root->node));
1911 	}
1912 
1913 	btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1914 	btrfs_set_backup_dev_root_gen(root_backup,
1915 			       btrfs_header_generation(info->dev_root->node));
1916 	btrfs_set_backup_dev_root_level(root_backup,
1917 				       btrfs_header_level(info->dev_root->node));
1918 
1919 	btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1920 	btrfs_set_backup_csum_root_gen(root_backup,
1921 			       btrfs_header_generation(info->csum_root->node));
1922 	btrfs_set_backup_csum_root_level(root_backup,
1923 			       btrfs_header_level(info->csum_root->node));
1924 
1925 	btrfs_set_backup_total_bytes(root_backup,
1926 			     btrfs_super_total_bytes(info->super_copy));
1927 	btrfs_set_backup_bytes_used(root_backup,
1928 			     btrfs_super_bytes_used(info->super_copy));
1929 	btrfs_set_backup_num_devices(root_backup,
1930 			     btrfs_super_num_devices(info->super_copy));
1931 
1932 	/*
1933 	 * if we don't copy this out to the super_copy, it won't get remembered
1934 	 * for the next commit
1935 	 */
1936 	memcpy(&info->super_copy->super_roots,
1937 	       &info->super_for_commit->super_roots,
1938 	       sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1939 }
1940 
1941 /*
1942  * this copies info out of the root backup array and back into
1943  * the in-memory super block.  It is meant to help iterate through
1944  * the array, so you send it the number of backups you've already
1945  * tried and the last backup index you used.
1946  *
1947  * this returns -1 when it has tried all the backups
1948  */
1949 static noinline int next_root_backup(struct btrfs_fs_info *info,
1950 				     struct btrfs_super_block *super,
1951 				     int *num_backups_tried, int *backup_index)
1952 {
1953 	struct btrfs_root_backup *root_backup;
1954 	int newest = *backup_index;
1955 
1956 	if (*num_backups_tried == 0) {
1957 		u64 gen = btrfs_super_generation(super);
1958 
1959 		newest = find_newest_super_backup(info, gen);
1960 		if (newest == -1)
1961 			return -1;
1962 
1963 		*backup_index = newest;
1964 		*num_backups_tried = 1;
1965 	} else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1966 		/* we've tried all the backups, all done */
1967 		return -1;
1968 	} else {
1969 		/* jump to the next oldest backup */
1970 		newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1971 			BTRFS_NUM_BACKUP_ROOTS;
1972 		*backup_index = newest;
1973 		*num_backups_tried += 1;
1974 	}
1975 	root_backup = super->super_roots + newest;
1976 
1977 	btrfs_set_super_generation(super,
1978 				   btrfs_backup_tree_root_gen(root_backup));
1979 	btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1980 	btrfs_set_super_root_level(super,
1981 				   btrfs_backup_tree_root_level(root_backup));
1982 	btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1983 
1984 	/*
1985 	 * fixme: the total bytes and num_devices need to match or we should
1986 	 * need a fsck
1987 	 */
1988 	btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1989 	btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1990 	return 0;
1991 }
1992 
1993 /* helper to cleanup workers */
1994 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1995 {
1996 	btrfs_stop_workers(&fs_info->generic_worker);
1997 	btrfs_stop_workers(&fs_info->fixup_workers);
1998 	btrfs_stop_workers(&fs_info->delalloc_workers);
1999 	btrfs_stop_workers(&fs_info->workers);
2000 	btrfs_stop_workers(&fs_info->endio_workers);
2001 	btrfs_stop_workers(&fs_info->endio_meta_workers);
2002 	btrfs_stop_workers(&fs_info->endio_raid56_workers);
2003 	btrfs_stop_workers(&fs_info->rmw_workers);
2004 	btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2005 	btrfs_stop_workers(&fs_info->endio_write_workers);
2006 	btrfs_stop_workers(&fs_info->endio_freespace_worker);
2007 	btrfs_stop_workers(&fs_info->submit_workers);
2008 	btrfs_stop_workers(&fs_info->delayed_workers);
2009 	btrfs_stop_workers(&fs_info->caching_workers);
2010 	btrfs_stop_workers(&fs_info->readahead_workers);
2011 	btrfs_stop_workers(&fs_info->flush_workers);
2012 	btrfs_stop_workers(&fs_info->qgroup_rescan_workers);
2013 }
2014 
2015 /* helper to cleanup tree roots */
2016 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2017 {
2018 	free_extent_buffer(info->tree_root->node);
2019 	free_extent_buffer(info->tree_root->commit_root);
2020 	info->tree_root->node = NULL;
2021 	info->tree_root->commit_root = NULL;
2022 
2023 	if (info->dev_root) {
2024 		free_extent_buffer(info->dev_root->node);
2025 		free_extent_buffer(info->dev_root->commit_root);
2026 		info->dev_root->node = NULL;
2027 		info->dev_root->commit_root = NULL;
2028 	}
2029 	if (info->extent_root) {
2030 		free_extent_buffer(info->extent_root->node);
2031 		free_extent_buffer(info->extent_root->commit_root);
2032 		info->extent_root->node = NULL;
2033 		info->extent_root->commit_root = NULL;
2034 	}
2035 	if (info->csum_root) {
2036 		free_extent_buffer(info->csum_root->node);
2037 		free_extent_buffer(info->csum_root->commit_root);
2038 		info->csum_root->node = NULL;
2039 		info->csum_root->commit_root = NULL;
2040 	}
2041 	if (info->quota_root) {
2042 		free_extent_buffer(info->quota_root->node);
2043 		free_extent_buffer(info->quota_root->commit_root);
2044 		info->quota_root->node = NULL;
2045 		info->quota_root->commit_root = NULL;
2046 	}
2047 	if (info->uuid_root) {
2048 		free_extent_buffer(info->uuid_root->node);
2049 		free_extent_buffer(info->uuid_root->commit_root);
2050 		info->uuid_root->node = NULL;
2051 		info->uuid_root->commit_root = NULL;
2052 	}
2053 	if (chunk_root) {
2054 		free_extent_buffer(info->chunk_root->node);
2055 		free_extent_buffer(info->chunk_root->commit_root);
2056 		info->chunk_root->node = NULL;
2057 		info->chunk_root->commit_root = NULL;
2058 	}
2059 }
2060 
2061 static void del_fs_roots(struct btrfs_fs_info *fs_info)
2062 {
2063 	int ret;
2064 	struct btrfs_root *gang[8];
2065 	int i;
2066 
2067 	while (!list_empty(&fs_info->dead_roots)) {
2068 		gang[0] = list_entry(fs_info->dead_roots.next,
2069 				     struct btrfs_root, root_list);
2070 		list_del(&gang[0]->root_list);
2071 
2072 		if (gang[0]->in_radix) {
2073 			btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2074 		} else {
2075 			free_extent_buffer(gang[0]->node);
2076 			free_extent_buffer(gang[0]->commit_root);
2077 			btrfs_put_fs_root(gang[0]);
2078 		}
2079 	}
2080 
2081 	while (1) {
2082 		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2083 					     (void **)gang, 0,
2084 					     ARRAY_SIZE(gang));
2085 		if (!ret)
2086 			break;
2087 		for (i = 0; i < ret; i++)
2088 			btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2089 	}
2090 }
2091 
2092 int open_ctree(struct super_block *sb,
2093 	       struct btrfs_fs_devices *fs_devices,
2094 	       char *options)
2095 {
2096 	u32 sectorsize;
2097 	u32 nodesize;
2098 	u32 leafsize;
2099 	u32 blocksize;
2100 	u32 stripesize;
2101 	u64 generation;
2102 	u64 features;
2103 	struct btrfs_key location;
2104 	struct buffer_head *bh;
2105 	struct btrfs_super_block *disk_super;
2106 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2107 	struct btrfs_root *tree_root;
2108 	struct btrfs_root *extent_root;
2109 	struct btrfs_root *csum_root;
2110 	struct btrfs_root *chunk_root;
2111 	struct btrfs_root *dev_root;
2112 	struct btrfs_root *quota_root;
2113 	struct btrfs_root *uuid_root;
2114 	struct btrfs_root *log_tree_root;
2115 	int ret;
2116 	int err = -EINVAL;
2117 	int num_backups_tried = 0;
2118 	int backup_index = 0;
2119 	bool create_uuid_tree;
2120 	bool check_uuid_tree;
2121 
2122 	tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2123 	chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2124 	if (!tree_root || !chunk_root) {
2125 		err = -ENOMEM;
2126 		goto fail;
2127 	}
2128 
2129 	ret = init_srcu_struct(&fs_info->subvol_srcu);
2130 	if (ret) {
2131 		err = ret;
2132 		goto fail;
2133 	}
2134 
2135 	ret = setup_bdi(fs_info, &fs_info->bdi);
2136 	if (ret) {
2137 		err = ret;
2138 		goto fail_srcu;
2139 	}
2140 
2141 	ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2142 	if (ret) {
2143 		err = ret;
2144 		goto fail_bdi;
2145 	}
2146 	fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2147 					(1 + ilog2(nr_cpu_ids));
2148 
2149 	ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2150 	if (ret) {
2151 		err = ret;
2152 		goto fail_dirty_metadata_bytes;
2153 	}
2154 
2155 	fs_info->btree_inode = new_inode(sb);
2156 	if (!fs_info->btree_inode) {
2157 		err = -ENOMEM;
2158 		goto fail_delalloc_bytes;
2159 	}
2160 
2161 	mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2162 
2163 	INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2164 	INIT_LIST_HEAD(&fs_info->trans_list);
2165 	INIT_LIST_HEAD(&fs_info->dead_roots);
2166 	INIT_LIST_HEAD(&fs_info->delayed_iputs);
2167 	INIT_LIST_HEAD(&fs_info->delalloc_roots);
2168 	INIT_LIST_HEAD(&fs_info->caching_block_groups);
2169 	spin_lock_init(&fs_info->delalloc_root_lock);
2170 	spin_lock_init(&fs_info->trans_lock);
2171 	spin_lock_init(&fs_info->fs_roots_radix_lock);
2172 	spin_lock_init(&fs_info->delayed_iput_lock);
2173 	spin_lock_init(&fs_info->defrag_inodes_lock);
2174 	spin_lock_init(&fs_info->free_chunk_lock);
2175 	spin_lock_init(&fs_info->tree_mod_seq_lock);
2176 	spin_lock_init(&fs_info->super_lock);
2177 	rwlock_init(&fs_info->tree_mod_log_lock);
2178 	mutex_init(&fs_info->reloc_mutex);
2179 	seqlock_init(&fs_info->profiles_lock);
2180 
2181 	init_completion(&fs_info->kobj_unregister);
2182 	INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2183 	INIT_LIST_HEAD(&fs_info->space_info);
2184 	INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2185 	btrfs_mapping_init(&fs_info->mapping_tree);
2186 	btrfs_init_block_rsv(&fs_info->global_block_rsv,
2187 			     BTRFS_BLOCK_RSV_GLOBAL);
2188 	btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2189 			     BTRFS_BLOCK_RSV_DELALLOC);
2190 	btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2191 	btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2192 	btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2193 	btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2194 			     BTRFS_BLOCK_RSV_DELOPS);
2195 	atomic_set(&fs_info->nr_async_submits, 0);
2196 	atomic_set(&fs_info->async_delalloc_pages, 0);
2197 	atomic_set(&fs_info->async_submit_draining, 0);
2198 	atomic_set(&fs_info->nr_async_bios, 0);
2199 	atomic_set(&fs_info->defrag_running, 0);
2200 	atomic64_set(&fs_info->tree_mod_seq, 0);
2201 	fs_info->sb = sb;
2202 	fs_info->max_inline = 8192 * 1024;
2203 	fs_info->metadata_ratio = 0;
2204 	fs_info->defrag_inodes = RB_ROOT;
2205 	fs_info->free_chunk_space = 0;
2206 	fs_info->tree_mod_log = RB_ROOT;
2207 	fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2208 
2209 	/* readahead state */
2210 	INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2211 	spin_lock_init(&fs_info->reada_lock);
2212 
2213 	fs_info->thread_pool_size = min_t(unsigned long,
2214 					  num_online_cpus() + 2, 8);
2215 
2216 	INIT_LIST_HEAD(&fs_info->ordered_roots);
2217 	spin_lock_init(&fs_info->ordered_root_lock);
2218 	fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2219 					GFP_NOFS);
2220 	if (!fs_info->delayed_root) {
2221 		err = -ENOMEM;
2222 		goto fail_iput;
2223 	}
2224 	btrfs_init_delayed_root(fs_info->delayed_root);
2225 
2226 	mutex_init(&fs_info->scrub_lock);
2227 	atomic_set(&fs_info->scrubs_running, 0);
2228 	atomic_set(&fs_info->scrub_pause_req, 0);
2229 	atomic_set(&fs_info->scrubs_paused, 0);
2230 	atomic_set(&fs_info->scrub_cancel_req, 0);
2231 	init_waitqueue_head(&fs_info->scrub_pause_wait);
2232 	init_rwsem(&fs_info->scrub_super_lock);
2233 	fs_info->scrub_workers_refcnt = 0;
2234 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2235 	fs_info->check_integrity_print_mask = 0;
2236 #endif
2237 
2238 	spin_lock_init(&fs_info->balance_lock);
2239 	mutex_init(&fs_info->balance_mutex);
2240 	atomic_set(&fs_info->balance_running, 0);
2241 	atomic_set(&fs_info->balance_pause_req, 0);
2242 	atomic_set(&fs_info->balance_cancel_req, 0);
2243 	fs_info->balance_ctl = NULL;
2244 	init_waitqueue_head(&fs_info->balance_wait_q);
2245 
2246 	sb->s_blocksize = 4096;
2247 	sb->s_blocksize_bits = blksize_bits(4096);
2248 	sb->s_bdi = &fs_info->bdi;
2249 
2250 	fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2251 	set_nlink(fs_info->btree_inode, 1);
2252 	/*
2253 	 * we set the i_size on the btree inode to the max possible int.
2254 	 * the real end of the address space is determined by all of
2255 	 * the devices in the system
2256 	 */
2257 	fs_info->btree_inode->i_size = OFFSET_MAX;
2258 	fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2259 	fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2260 
2261 	RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2262 	extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2263 			     fs_info->btree_inode->i_mapping);
2264 	BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2265 	extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2266 
2267 	BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2268 
2269 	BTRFS_I(fs_info->btree_inode)->root = tree_root;
2270 	memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2271 	       sizeof(struct btrfs_key));
2272 	set_bit(BTRFS_INODE_DUMMY,
2273 		&BTRFS_I(fs_info->btree_inode)->runtime_flags);
2274 	insert_inode_hash(fs_info->btree_inode);
2275 
2276 	spin_lock_init(&fs_info->block_group_cache_lock);
2277 	fs_info->block_group_cache_tree = RB_ROOT;
2278 	fs_info->first_logical_byte = (u64)-1;
2279 
2280 	extent_io_tree_init(&fs_info->freed_extents[0],
2281 			     fs_info->btree_inode->i_mapping);
2282 	extent_io_tree_init(&fs_info->freed_extents[1],
2283 			     fs_info->btree_inode->i_mapping);
2284 	fs_info->pinned_extents = &fs_info->freed_extents[0];
2285 	fs_info->do_barriers = 1;
2286 
2287 
2288 	mutex_init(&fs_info->ordered_operations_mutex);
2289 	mutex_init(&fs_info->ordered_extent_flush_mutex);
2290 	mutex_init(&fs_info->tree_log_mutex);
2291 	mutex_init(&fs_info->chunk_mutex);
2292 	mutex_init(&fs_info->transaction_kthread_mutex);
2293 	mutex_init(&fs_info->cleaner_mutex);
2294 	mutex_init(&fs_info->volume_mutex);
2295 	init_rwsem(&fs_info->extent_commit_sem);
2296 	init_rwsem(&fs_info->cleanup_work_sem);
2297 	init_rwsem(&fs_info->subvol_sem);
2298 	sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2299 	fs_info->dev_replace.lock_owner = 0;
2300 	atomic_set(&fs_info->dev_replace.nesting_level, 0);
2301 	mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2302 	mutex_init(&fs_info->dev_replace.lock_management_lock);
2303 	mutex_init(&fs_info->dev_replace.lock);
2304 
2305 	spin_lock_init(&fs_info->qgroup_lock);
2306 	mutex_init(&fs_info->qgroup_ioctl_lock);
2307 	fs_info->qgroup_tree = RB_ROOT;
2308 	INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2309 	fs_info->qgroup_seq = 1;
2310 	fs_info->quota_enabled = 0;
2311 	fs_info->pending_quota_state = 0;
2312 	fs_info->qgroup_ulist = NULL;
2313 	mutex_init(&fs_info->qgroup_rescan_lock);
2314 
2315 	btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2316 	btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2317 
2318 	init_waitqueue_head(&fs_info->transaction_throttle);
2319 	init_waitqueue_head(&fs_info->transaction_wait);
2320 	init_waitqueue_head(&fs_info->transaction_blocked_wait);
2321 	init_waitqueue_head(&fs_info->async_submit_wait);
2322 
2323 	ret = btrfs_alloc_stripe_hash_table(fs_info);
2324 	if (ret) {
2325 		err = ret;
2326 		goto fail_alloc;
2327 	}
2328 
2329 	__setup_root(4096, 4096, 4096, 4096, tree_root,
2330 		     fs_info, BTRFS_ROOT_TREE_OBJECTID);
2331 
2332 	invalidate_bdev(fs_devices->latest_bdev);
2333 
2334 	/*
2335 	 * Read super block and check the signature bytes only
2336 	 */
2337 	bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2338 	if (!bh) {
2339 		err = -EINVAL;
2340 		goto fail_alloc;
2341 	}
2342 
2343 	/*
2344 	 * We want to check superblock checksum, the type is stored inside.
2345 	 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2346 	 */
2347 	if (btrfs_check_super_csum(bh->b_data)) {
2348 		printk(KERN_ERR "btrfs: superblock checksum mismatch\n");
2349 		err = -EINVAL;
2350 		goto fail_alloc;
2351 	}
2352 
2353 	/*
2354 	 * super_copy is zeroed at allocation time and we never touch the
2355 	 * following bytes up to INFO_SIZE, the checksum is calculated from
2356 	 * the whole block of INFO_SIZE
2357 	 */
2358 	memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2359 	memcpy(fs_info->super_for_commit, fs_info->super_copy,
2360 	       sizeof(*fs_info->super_for_commit));
2361 	brelse(bh);
2362 
2363 	memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2364 
2365 	ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2366 	if (ret) {
2367 		printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2368 		err = -EINVAL;
2369 		goto fail_alloc;
2370 	}
2371 
2372 	disk_super = fs_info->super_copy;
2373 	if (!btrfs_super_root(disk_super))
2374 		goto fail_alloc;
2375 
2376 	/* check FS state, whether FS is broken. */
2377 	if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2378 		set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2379 
2380 	/*
2381 	 * run through our array of backup supers and setup
2382 	 * our ring pointer to the oldest one
2383 	 */
2384 	generation = btrfs_super_generation(disk_super);
2385 	find_oldest_super_backup(fs_info, generation);
2386 
2387 	/*
2388 	 * In the long term, we'll store the compression type in the super
2389 	 * block, and it'll be used for per file compression control.
2390 	 */
2391 	fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2392 
2393 	ret = btrfs_parse_options(tree_root, options);
2394 	if (ret) {
2395 		err = ret;
2396 		goto fail_alloc;
2397 	}
2398 
2399 	features = btrfs_super_incompat_flags(disk_super) &
2400 		~BTRFS_FEATURE_INCOMPAT_SUPP;
2401 	if (features) {
2402 		printk(KERN_ERR "BTRFS: couldn't mount because of "
2403 		       "unsupported optional features (%Lx).\n",
2404 		       features);
2405 		err = -EINVAL;
2406 		goto fail_alloc;
2407 	}
2408 
2409 	if (btrfs_super_leafsize(disk_super) !=
2410 	    btrfs_super_nodesize(disk_super)) {
2411 		printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2412 		       "blocksizes don't match.  node %d leaf %d\n",
2413 		       btrfs_super_nodesize(disk_super),
2414 		       btrfs_super_leafsize(disk_super));
2415 		err = -EINVAL;
2416 		goto fail_alloc;
2417 	}
2418 	if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2419 		printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2420 		       "blocksize (%d) was too large\n",
2421 		       btrfs_super_leafsize(disk_super));
2422 		err = -EINVAL;
2423 		goto fail_alloc;
2424 	}
2425 
2426 	features = btrfs_super_incompat_flags(disk_super);
2427 	features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2428 	if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2429 		features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2430 
2431 	if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2432 		printk(KERN_ERR "btrfs: has skinny extents\n");
2433 
2434 	/*
2435 	 * flag our filesystem as having big metadata blocks if
2436 	 * they are bigger than the page size
2437 	 */
2438 	if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2439 		if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2440 			printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2441 		features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2442 	}
2443 
2444 	nodesize = btrfs_super_nodesize(disk_super);
2445 	leafsize = btrfs_super_leafsize(disk_super);
2446 	sectorsize = btrfs_super_sectorsize(disk_super);
2447 	stripesize = btrfs_super_stripesize(disk_super);
2448 	fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
2449 	fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2450 
2451 	/*
2452 	 * mixed block groups end up with duplicate but slightly offset
2453 	 * extent buffers for the same range.  It leads to corruptions
2454 	 */
2455 	if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2456 	    (sectorsize != leafsize)) {
2457 		printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2458 				"are not allowed for mixed block groups on %s\n",
2459 				sb->s_id);
2460 		goto fail_alloc;
2461 	}
2462 
2463 	/*
2464 	 * Needn't use the lock because there is no other task which will
2465 	 * update the flag.
2466 	 */
2467 	btrfs_set_super_incompat_flags(disk_super, features);
2468 
2469 	features = btrfs_super_compat_ro_flags(disk_super) &
2470 		~BTRFS_FEATURE_COMPAT_RO_SUPP;
2471 	if (!(sb->s_flags & MS_RDONLY) && features) {
2472 		printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2473 		       "unsupported option features (%Lx).\n",
2474 		       features);
2475 		err = -EINVAL;
2476 		goto fail_alloc;
2477 	}
2478 
2479 	btrfs_init_workers(&fs_info->generic_worker,
2480 			   "genwork", 1, NULL);
2481 
2482 	btrfs_init_workers(&fs_info->workers, "worker",
2483 			   fs_info->thread_pool_size,
2484 			   &fs_info->generic_worker);
2485 
2486 	btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2487 			   fs_info->thread_pool_size, NULL);
2488 
2489 	btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc",
2490 			   fs_info->thread_pool_size, NULL);
2491 
2492 	btrfs_init_workers(&fs_info->submit_workers, "submit",
2493 			   min_t(u64, fs_devices->num_devices,
2494 			   fs_info->thread_pool_size), NULL);
2495 
2496 	btrfs_init_workers(&fs_info->caching_workers, "cache",
2497 			   fs_info->thread_pool_size, NULL);
2498 
2499 	/* a higher idle thresh on the submit workers makes it much more
2500 	 * likely that bios will be send down in a sane order to the
2501 	 * devices
2502 	 */
2503 	fs_info->submit_workers.idle_thresh = 64;
2504 
2505 	fs_info->workers.idle_thresh = 16;
2506 	fs_info->workers.ordered = 1;
2507 
2508 	fs_info->delalloc_workers.idle_thresh = 2;
2509 	fs_info->delalloc_workers.ordered = 1;
2510 
2511 	btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2512 			   &fs_info->generic_worker);
2513 	btrfs_init_workers(&fs_info->endio_workers, "endio",
2514 			   fs_info->thread_pool_size,
2515 			   &fs_info->generic_worker);
2516 	btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2517 			   fs_info->thread_pool_size,
2518 			   &fs_info->generic_worker);
2519 	btrfs_init_workers(&fs_info->endio_meta_write_workers,
2520 			   "endio-meta-write", fs_info->thread_pool_size,
2521 			   &fs_info->generic_worker);
2522 	btrfs_init_workers(&fs_info->endio_raid56_workers,
2523 			   "endio-raid56", fs_info->thread_pool_size,
2524 			   &fs_info->generic_worker);
2525 	btrfs_init_workers(&fs_info->rmw_workers,
2526 			   "rmw", fs_info->thread_pool_size,
2527 			   &fs_info->generic_worker);
2528 	btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2529 			   fs_info->thread_pool_size,
2530 			   &fs_info->generic_worker);
2531 	btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2532 			   1, &fs_info->generic_worker);
2533 	btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2534 			   fs_info->thread_pool_size,
2535 			   &fs_info->generic_worker);
2536 	btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2537 			   fs_info->thread_pool_size,
2538 			   &fs_info->generic_worker);
2539 	btrfs_init_workers(&fs_info->qgroup_rescan_workers, "qgroup-rescan", 1,
2540 			   &fs_info->generic_worker);
2541 
2542 	/*
2543 	 * endios are largely parallel and should have a very
2544 	 * low idle thresh
2545 	 */
2546 	fs_info->endio_workers.idle_thresh = 4;
2547 	fs_info->endio_meta_workers.idle_thresh = 4;
2548 	fs_info->endio_raid56_workers.idle_thresh = 4;
2549 	fs_info->rmw_workers.idle_thresh = 2;
2550 
2551 	fs_info->endio_write_workers.idle_thresh = 2;
2552 	fs_info->endio_meta_write_workers.idle_thresh = 2;
2553 	fs_info->readahead_workers.idle_thresh = 2;
2554 
2555 	/*
2556 	 * btrfs_start_workers can really only fail because of ENOMEM so just
2557 	 * return -ENOMEM if any of these fail.
2558 	 */
2559 	ret = btrfs_start_workers(&fs_info->workers);
2560 	ret |= btrfs_start_workers(&fs_info->generic_worker);
2561 	ret |= btrfs_start_workers(&fs_info->submit_workers);
2562 	ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2563 	ret |= btrfs_start_workers(&fs_info->fixup_workers);
2564 	ret |= btrfs_start_workers(&fs_info->endio_workers);
2565 	ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2566 	ret |= btrfs_start_workers(&fs_info->rmw_workers);
2567 	ret |= btrfs_start_workers(&fs_info->endio_raid56_workers);
2568 	ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2569 	ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2570 	ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2571 	ret |= btrfs_start_workers(&fs_info->delayed_workers);
2572 	ret |= btrfs_start_workers(&fs_info->caching_workers);
2573 	ret |= btrfs_start_workers(&fs_info->readahead_workers);
2574 	ret |= btrfs_start_workers(&fs_info->flush_workers);
2575 	ret |= btrfs_start_workers(&fs_info->qgroup_rescan_workers);
2576 	if (ret) {
2577 		err = -ENOMEM;
2578 		goto fail_sb_buffer;
2579 	}
2580 
2581 	fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2582 	fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2583 				    4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2584 
2585 	tree_root->nodesize = nodesize;
2586 	tree_root->leafsize = leafsize;
2587 	tree_root->sectorsize = sectorsize;
2588 	tree_root->stripesize = stripesize;
2589 
2590 	sb->s_blocksize = sectorsize;
2591 	sb->s_blocksize_bits = blksize_bits(sectorsize);
2592 
2593 	if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2594 		printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2595 		goto fail_sb_buffer;
2596 	}
2597 
2598 	if (sectorsize != PAGE_SIZE) {
2599 		printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2600 		       "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2601 		goto fail_sb_buffer;
2602 	}
2603 
2604 	mutex_lock(&fs_info->chunk_mutex);
2605 	ret = btrfs_read_sys_array(tree_root);
2606 	mutex_unlock(&fs_info->chunk_mutex);
2607 	if (ret) {
2608 		printk(KERN_WARNING "btrfs: failed to read the system "
2609 		       "array on %s\n", sb->s_id);
2610 		goto fail_sb_buffer;
2611 	}
2612 
2613 	blocksize = btrfs_level_size(tree_root,
2614 				     btrfs_super_chunk_root_level(disk_super));
2615 	generation = btrfs_super_chunk_root_generation(disk_super);
2616 
2617 	__setup_root(nodesize, leafsize, sectorsize, stripesize,
2618 		     chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2619 
2620 	chunk_root->node = read_tree_block(chunk_root,
2621 					   btrfs_super_chunk_root(disk_super),
2622 					   blocksize, generation);
2623 	if (!chunk_root->node ||
2624 	    !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2625 		printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2626 		       sb->s_id);
2627 		goto fail_tree_roots;
2628 	}
2629 	btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2630 	chunk_root->commit_root = btrfs_root_node(chunk_root);
2631 
2632 	read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2633 	   btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2634 
2635 	ret = btrfs_read_chunk_tree(chunk_root);
2636 	if (ret) {
2637 		printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2638 		       sb->s_id);
2639 		goto fail_tree_roots;
2640 	}
2641 
2642 	/*
2643 	 * keep the device that is marked to be the target device for the
2644 	 * dev_replace procedure
2645 	 */
2646 	btrfs_close_extra_devices(fs_info, fs_devices, 0);
2647 
2648 	if (!fs_devices->latest_bdev) {
2649 		printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2650 		       sb->s_id);
2651 		goto fail_tree_roots;
2652 	}
2653 
2654 retry_root_backup:
2655 	blocksize = btrfs_level_size(tree_root,
2656 				     btrfs_super_root_level(disk_super));
2657 	generation = btrfs_super_generation(disk_super);
2658 
2659 	tree_root->node = read_tree_block(tree_root,
2660 					  btrfs_super_root(disk_super),
2661 					  blocksize, generation);
2662 	if (!tree_root->node ||
2663 	    !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2664 		printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2665 		       sb->s_id);
2666 
2667 		goto recovery_tree_root;
2668 	}
2669 
2670 	btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2671 	tree_root->commit_root = btrfs_root_node(tree_root);
2672 
2673 	location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2674 	location.type = BTRFS_ROOT_ITEM_KEY;
2675 	location.offset = 0;
2676 
2677 	extent_root = btrfs_read_tree_root(tree_root, &location);
2678 	if (IS_ERR(extent_root)) {
2679 		ret = PTR_ERR(extent_root);
2680 		goto recovery_tree_root;
2681 	}
2682 	extent_root->track_dirty = 1;
2683 	fs_info->extent_root = extent_root;
2684 
2685 	location.objectid = BTRFS_DEV_TREE_OBJECTID;
2686 	dev_root = btrfs_read_tree_root(tree_root, &location);
2687 	if (IS_ERR(dev_root)) {
2688 		ret = PTR_ERR(dev_root);
2689 		goto recovery_tree_root;
2690 	}
2691 	dev_root->track_dirty = 1;
2692 	fs_info->dev_root = dev_root;
2693 	btrfs_init_devices_late(fs_info);
2694 
2695 	location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2696 	csum_root = btrfs_read_tree_root(tree_root, &location);
2697 	if (IS_ERR(csum_root)) {
2698 		ret = PTR_ERR(csum_root);
2699 		goto recovery_tree_root;
2700 	}
2701 	csum_root->track_dirty = 1;
2702 	fs_info->csum_root = csum_root;
2703 
2704 	location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2705 	quota_root = btrfs_read_tree_root(tree_root, &location);
2706 	if (!IS_ERR(quota_root)) {
2707 		quota_root->track_dirty = 1;
2708 		fs_info->quota_enabled = 1;
2709 		fs_info->pending_quota_state = 1;
2710 		fs_info->quota_root = quota_root;
2711 	}
2712 
2713 	location.objectid = BTRFS_UUID_TREE_OBJECTID;
2714 	uuid_root = btrfs_read_tree_root(tree_root, &location);
2715 	if (IS_ERR(uuid_root)) {
2716 		ret = PTR_ERR(uuid_root);
2717 		if (ret != -ENOENT)
2718 			goto recovery_tree_root;
2719 		create_uuid_tree = true;
2720 		check_uuid_tree = false;
2721 	} else {
2722 		uuid_root->track_dirty = 1;
2723 		fs_info->uuid_root = uuid_root;
2724 		create_uuid_tree = false;
2725 		check_uuid_tree =
2726 		    generation != btrfs_super_uuid_tree_generation(disk_super);
2727 	}
2728 
2729 	fs_info->generation = generation;
2730 	fs_info->last_trans_committed = generation;
2731 
2732 	ret = btrfs_recover_balance(fs_info);
2733 	if (ret) {
2734 		printk(KERN_WARNING "btrfs: failed to recover balance\n");
2735 		goto fail_block_groups;
2736 	}
2737 
2738 	ret = btrfs_init_dev_stats(fs_info);
2739 	if (ret) {
2740 		printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
2741 		       ret);
2742 		goto fail_block_groups;
2743 	}
2744 
2745 	ret = btrfs_init_dev_replace(fs_info);
2746 	if (ret) {
2747 		pr_err("btrfs: failed to init dev_replace: %d\n", ret);
2748 		goto fail_block_groups;
2749 	}
2750 
2751 	btrfs_close_extra_devices(fs_info, fs_devices, 1);
2752 
2753 	ret = btrfs_init_space_info(fs_info);
2754 	if (ret) {
2755 		printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2756 		goto fail_block_groups;
2757 	}
2758 
2759 	ret = btrfs_read_block_groups(extent_root);
2760 	if (ret) {
2761 		printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2762 		goto fail_block_groups;
2763 	}
2764 	fs_info->num_tolerated_disk_barrier_failures =
2765 		btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2766 	if (fs_info->fs_devices->missing_devices >
2767 	     fs_info->num_tolerated_disk_barrier_failures &&
2768 	    !(sb->s_flags & MS_RDONLY)) {
2769 		printk(KERN_WARNING
2770 		       "Btrfs: too many missing devices, writeable mount is not allowed\n");
2771 		goto fail_block_groups;
2772 	}
2773 
2774 	fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2775 					       "btrfs-cleaner");
2776 	if (IS_ERR(fs_info->cleaner_kthread))
2777 		goto fail_block_groups;
2778 
2779 	fs_info->transaction_kthread = kthread_run(transaction_kthread,
2780 						   tree_root,
2781 						   "btrfs-transaction");
2782 	if (IS_ERR(fs_info->transaction_kthread))
2783 		goto fail_cleaner;
2784 
2785 	if (!btrfs_test_opt(tree_root, SSD) &&
2786 	    !btrfs_test_opt(tree_root, NOSSD) &&
2787 	    !fs_info->fs_devices->rotating) {
2788 		printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2789 		       "mode\n");
2790 		btrfs_set_opt(fs_info->mount_opt, SSD);
2791 	}
2792 
2793 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2794 	if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2795 		ret = btrfsic_mount(tree_root, fs_devices,
2796 				    btrfs_test_opt(tree_root,
2797 					CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2798 				    1 : 0,
2799 				    fs_info->check_integrity_print_mask);
2800 		if (ret)
2801 			printk(KERN_WARNING "btrfs: failed to initialize"
2802 			       " integrity check module %s\n", sb->s_id);
2803 	}
2804 #endif
2805 	ret = btrfs_read_qgroup_config(fs_info);
2806 	if (ret)
2807 		goto fail_trans_kthread;
2808 
2809 	/* do not make disk changes in broken FS */
2810 	if (btrfs_super_log_root(disk_super) != 0) {
2811 		u64 bytenr = btrfs_super_log_root(disk_super);
2812 
2813 		if (fs_devices->rw_devices == 0) {
2814 			printk(KERN_WARNING "Btrfs log replay required "
2815 			       "on RO media\n");
2816 			err = -EIO;
2817 			goto fail_qgroup;
2818 		}
2819 		blocksize =
2820 		     btrfs_level_size(tree_root,
2821 				      btrfs_super_log_root_level(disk_super));
2822 
2823 		log_tree_root = btrfs_alloc_root(fs_info);
2824 		if (!log_tree_root) {
2825 			err = -ENOMEM;
2826 			goto fail_qgroup;
2827 		}
2828 
2829 		__setup_root(nodesize, leafsize, sectorsize, stripesize,
2830 			     log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2831 
2832 		log_tree_root->node = read_tree_block(tree_root, bytenr,
2833 						      blocksize,
2834 						      generation + 1);
2835 		if (!log_tree_root->node ||
2836 		    !extent_buffer_uptodate(log_tree_root->node)) {
2837 			printk(KERN_ERR "btrfs: failed to read log tree\n");
2838 			free_extent_buffer(log_tree_root->node);
2839 			kfree(log_tree_root);
2840 			goto fail_trans_kthread;
2841 		}
2842 		/* returns with log_tree_root freed on success */
2843 		ret = btrfs_recover_log_trees(log_tree_root);
2844 		if (ret) {
2845 			btrfs_error(tree_root->fs_info, ret,
2846 				    "Failed to recover log tree");
2847 			free_extent_buffer(log_tree_root->node);
2848 			kfree(log_tree_root);
2849 			goto fail_trans_kthread;
2850 		}
2851 
2852 		if (sb->s_flags & MS_RDONLY) {
2853 			ret = btrfs_commit_super(tree_root);
2854 			if (ret)
2855 				goto fail_trans_kthread;
2856 		}
2857 	}
2858 
2859 	ret = btrfs_find_orphan_roots(tree_root);
2860 	if (ret)
2861 		goto fail_trans_kthread;
2862 
2863 	if (!(sb->s_flags & MS_RDONLY)) {
2864 		ret = btrfs_cleanup_fs_roots(fs_info);
2865 		if (ret)
2866 			goto fail_trans_kthread;
2867 
2868 		ret = btrfs_recover_relocation(tree_root);
2869 		if (ret < 0) {
2870 			printk(KERN_WARNING
2871 			       "btrfs: failed to recover relocation\n");
2872 			err = -EINVAL;
2873 			goto fail_qgroup;
2874 		}
2875 	}
2876 
2877 	location.objectid = BTRFS_FS_TREE_OBJECTID;
2878 	location.type = BTRFS_ROOT_ITEM_KEY;
2879 	location.offset = 0;
2880 
2881 	fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2882 	if (IS_ERR(fs_info->fs_root)) {
2883 		err = PTR_ERR(fs_info->fs_root);
2884 		goto fail_qgroup;
2885 	}
2886 
2887 	if (sb->s_flags & MS_RDONLY)
2888 		return 0;
2889 
2890 	down_read(&fs_info->cleanup_work_sem);
2891 	if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2892 	    (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2893 		up_read(&fs_info->cleanup_work_sem);
2894 		close_ctree(tree_root);
2895 		return ret;
2896 	}
2897 	up_read(&fs_info->cleanup_work_sem);
2898 
2899 	ret = btrfs_resume_balance_async(fs_info);
2900 	if (ret) {
2901 		printk(KERN_WARNING "btrfs: failed to resume balance\n");
2902 		close_ctree(tree_root);
2903 		return ret;
2904 	}
2905 
2906 	ret = btrfs_resume_dev_replace_async(fs_info);
2907 	if (ret) {
2908 		pr_warn("btrfs: failed to resume dev_replace\n");
2909 		close_ctree(tree_root);
2910 		return ret;
2911 	}
2912 
2913 	btrfs_qgroup_rescan_resume(fs_info);
2914 
2915 	if (create_uuid_tree) {
2916 		pr_info("btrfs: creating UUID tree\n");
2917 		ret = btrfs_create_uuid_tree(fs_info);
2918 		if (ret) {
2919 			pr_warn("btrfs: failed to create the UUID tree %d\n",
2920 				ret);
2921 			close_ctree(tree_root);
2922 			return ret;
2923 		}
2924 	} else if (check_uuid_tree ||
2925 		   btrfs_test_opt(tree_root, RESCAN_UUID_TREE)) {
2926 		pr_info("btrfs: checking UUID tree\n");
2927 		ret = btrfs_check_uuid_tree(fs_info);
2928 		if (ret) {
2929 			pr_warn("btrfs: failed to check the UUID tree %d\n",
2930 				ret);
2931 			close_ctree(tree_root);
2932 			return ret;
2933 		}
2934 	} else {
2935 		fs_info->update_uuid_tree_gen = 1;
2936 	}
2937 
2938 	return 0;
2939 
2940 fail_qgroup:
2941 	btrfs_free_qgroup_config(fs_info);
2942 fail_trans_kthread:
2943 	kthread_stop(fs_info->transaction_kthread);
2944 	btrfs_cleanup_transaction(fs_info->tree_root);
2945 	del_fs_roots(fs_info);
2946 fail_cleaner:
2947 	kthread_stop(fs_info->cleaner_kthread);
2948 
2949 	/*
2950 	 * make sure we're done with the btree inode before we stop our
2951 	 * kthreads
2952 	 */
2953 	filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2954 
2955 fail_block_groups:
2956 	btrfs_put_block_group_cache(fs_info);
2957 	btrfs_free_block_groups(fs_info);
2958 
2959 fail_tree_roots:
2960 	free_root_pointers(fs_info, 1);
2961 	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2962 
2963 fail_sb_buffer:
2964 	btrfs_stop_all_workers(fs_info);
2965 fail_alloc:
2966 fail_iput:
2967 	btrfs_mapping_tree_free(&fs_info->mapping_tree);
2968 
2969 	iput(fs_info->btree_inode);
2970 fail_delalloc_bytes:
2971 	percpu_counter_destroy(&fs_info->delalloc_bytes);
2972 fail_dirty_metadata_bytes:
2973 	percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
2974 fail_bdi:
2975 	bdi_destroy(&fs_info->bdi);
2976 fail_srcu:
2977 	cleanup_srcu_struct(&fs_info->subvol_srcu);
2978 fail:
2979 	btrfs_free_stripe_hash_table(fs_info);
2980 	btrfs_close_devices(fs_info->fs_devices);
2981 	return err;
2982 
2983 recovery_tree_root:
2984 	if (!btrfs_test_opt(tree_root, RECOVERY))
2985 		goto fail_tree_roots;
2986 
2987 	free_root_pointers(fs_info, 0);
2988 
2989 	/* don't use the log in recovery mode, it won't be valid */
2990 	btrfs_set_super_log_root(disk_super, 0);
2991 
2992 	/* we can't trust the free space cache either */
2993 	btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2994 
2995 	ret = next_root_backup(fs_info, fs_info->super_copy,
2996 			       &num_backups_tried, &backup_index);
2997 	if (ret == -1)
2998 		goto fail_block_groups;
2999 	goto retry_root_backup;
3000 }
3001 
3002 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3003 {
3004 	if (uptodate) {
3005 		set_buffer_uptodate(bh);
3006 	} else {
3007 		struct btrfs_device *device = (struct btrfs_device *)
3008 			bh->b_private;
3009 
3010 		printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
3011 					  "I/O error on %s\n",
3012 					  rcu_str_deref(device->name));
3013 		/* note, we dont' set_buffer_write_io_error because we have
3014 		 * our own ways of dealing with the IO errors
3015 		 */
3016 		clear_buffer_uptodate(bh);
3017 		btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3018 	}
3019 	unlock_buffer(bh);
3020 	put_bh(bh);
3021 }
3022 
3023 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3024 {
3025 	struct buffer_head *bh;
3026 	struct buffer_head *latest = NULL;
3027 	struct btrfs_super_block *super;
3028 	int i;
3029 	u64 transid = 0;
3030 	u64 bytenr;
3031 
3032 	/* we would like to check all the supers, but that would make
3033 	 * a btrfs mount succeed after a mkfs from a different FS.
3034 	 * So, we need to add a special mount option to scan for
3035 	 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3036 	 */
3037 	for (i = 0; i < 1; i++) {
3038 		bytenr = btrfs_sb_offset(i);
3039 		if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3040 					i_size_read(bdev->bd_inode))
3041 			break;
3042 		bh = __bread(bdev, bytenr / 4096,
3043 					BTRFS_SUPER_INFO_SIZE);
3044 		if (!bh)
3045 			continue;
3046 
3047 		super = (struct btrfs_super_block *)bh->b_data;
3048 		if (btrfs_super_bytenr(super) != bytenr ||
3049 		    btrfs_super_magic(super) != BTRFS_MAGIC) {
3050 			brelse(bh);
3051 			continue;
3052 		}
3053 
3054 		if (!latest || btrfs_super_generation(super) > transid) {
3055 			brelse(latest);
3056 			latest = bh;
3057 			transid = btrfs_super_generation(super);
3058 		} else {
3059 			brelse(bh);
3060 		}
3061 	}
3062 	return latest;
3063 }
3064 
3065 /*
3066  * this should be called twice, once with wait == 0 and
3067  * once with wait == 1.  When wait == 0 is done, all the buffer heads
3068  * we write are pinned.
3069  *
3070  * They are released when wait == 1 is done.
3071  * max_mirrors must be the same for both runs, and it indicates how
3072  * many supers on this one device should be written.
3073  *
3074  * max_mirrors == 0 means to write them all.
3075  */
3076 static int write_dev_supers(struct btrfs_device *device,
3077 			    struct btrfs_super_block *sb,
3078 			    int do_barriers, int wait, int max_mirrors)
3079 {
3080 	struct buffer_head *bh;
3081 	int i;
3082 	int ret;
3083 	int errors = 0;
3084 	u32 crc;
3085 	u64 bytenr;
3086 
3087 	if (max_mirrors == 0)
3088 		max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3089 
3090 	for (i = 0; i < max_mirrors; i++) {
3091 		bytenr = btrfs_sb_offset(i);
3092 		if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3093 			break;
3094 
3095 		if (wait) {
3096 			bh = __find_get_block(device->bdev, bytenr / 4096,
3097 					      BTRFS_SUPER_INFO_SIZE);
3098 			if (!bh) {
3099 				errors++;
3100 				continue;
3101 			}
3102 			wait_on_buffer(bh);
3103 			if (!buffer_uptodate(bh))
3104 				errors++;
3105 
3106 			/* drop our reference */
3107 			brelse(bh);
3108 
3109 			/* drop the reference from the wait == 0 run */
3110 			brelse(bh);
3111 			continue;
3112 		} else {
3113 			btrfs_set_super_bytenr(sb, bytenr);
3114 
3115 			crc = ~(u32)0;
3116 			crc = btrfs_csum_data((char *)sb +
3117 					      BTRFS_CSUM_SIZE, crc,
3118 					      BTRFS_SUPER_INFO_SIZE -
3119 					      BTRFS_CSUM_SIZE);
3120 			btrfs_csum_final(crc, sb->csum);
3121 
3122 			/*
3123 			 * one reference for us, and we leave it for the
3124 			 * caller
3125 			 */
3126 			bh = __getblk(device->bdev, bytenr / 4096,
3127 				      BTRFS_SUPER_INFO_SIZE);
3128 			if (!bh) {
3129 				printk(KERN_ERR "btrfs: couldn't get super "
3130 				       "buffer head for bytenr %Lu\n", bytenr);
3131 				errors++;
3132 				continue;
3133 			}
3134 
3135 			memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3136 
3137 			/* one reference for submit_bh */
3138 			get_bh(bh);
3139 
3140 			set_buffer_uptodate(bh);
3141 			lock_buffer(bh);
3142 			bh->b_end_io = btrfs_end_buffer_write_sync;
3143 			bh->b_private = device;
3144 		}
3145 
3146 		/*
3147 		 * we fua the first super.  The others we allow
3148 		 * to go down lazy.
3149 		 */
3150 		ret = btrfsic_submit_bh(WRITE_FUA, bh);
3151 		if (ret)
3152 			errors++;
3153 	}
3154 	return errors < i ? 0 : -1;
3155 }
3156 
3157 /*
3158  * endio for the write_dev_flush, this will wake anyone waiting
3159  * for the barrier when it is done
3160  */
3161 static void btrfs_end_empty_barrier(struct bio *bio, int err)
3162 {
3163 	if (err) {
3164 		if (err == -EOPNOTSUPP)
3165 			set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3166 		clear_bit(BIO_UPTODATE, &bio->bi_flags);
3167 	}
3168 	if (bio->bi_private)
3169 		complete(bio->bi_private);
3170 	bio_put(bio);
3171 }
3172 
3173 /*
3174  * trigger flushes for one the devices.  If you pass wait == 0, the flushes are
3175  * sent down.  With wait == 1, it waits for the previous flush.
3176  *
3177  * any device where the flush fails with eopnotsupp are flagged as not-barrier
3178  * capable
3179  */
3180 static int write_dev_flush(struct btrfs_device *device, int wait)
3181 {
3182 	struct bio *bio;
3183 	int ret = 0;
3184 
3185 	if (device->nobarriers)
3186 		return 0;
3187 
3188 	if (wait) {
3189 		bio = device->flush_bio;
3190 		if (!bio)
3191 			return 0;
3192 
3193 		wait_for_completion(&device->flush_wait);
3194 
3195 		if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3196 			printk_in_rcu("btrfs: disabling barriers on dev %s\n",
3197 				      rcu_str_deref(device->name));
3198 			device->nobarriers = 1;
3199 		} else if (!bio_flagged(bio, BIO_UPTODATE)) {
3200 			ret = -EIO;
3201 			btrfs_dev_stat_inc_and_print(device,
3202 				BTRFS_DEV_STAT_FLUSH_ERRS);
3203 		}
3204 
3205 		/* drop the reference from the wait == 0 run */
3206 		bio_put(bio);
3207 		device->flush_bio = NULL;
3208 
3209 		return ret;
3210 	}
3211 
3212 	/*
3213 	 * one reference for us, and we leave it for the
3214 	 * caller
3215 	 */
3216 	device->flush_bio = NULL;
3217 	bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3218 	if (!bio)
3219 		return -ENOMEM;
3220 
3221 	bio->bi_end_io = btrfs_end_empty_barrier;
3222 	bio->bi_bdev = device->bdev;
3223 	init_completion(&device->flush_wait);
3224 	bio->bi_private = &device->flush_wait;
3225 	device->flush_bio = bio;
3226 
3227 	bio_get(bio);
3228 	btrfsic_submit_bio(WRITE_FLUSH, bio);
3229 
3230 	return 0;
3231 }
3232 
3233 /*
3234  * send an empty flush down to each device in parallel,
3235  * then wait for them
3236  */
3237 static int barrier_all_devices(struct btrfs_fs_info *info)
3238 {
3239 	struct list_head *head;
3240 	struct btrfs_device *dev;
3241 	int errors_send = 0;
3242 	int errors_wait = 0;
3243 	int ret;
3244 
3245 	/* send down all the barriers */
3246 	head = &info->fs_devices->devices;
3247 	list_for_each_entry_rcu(dev, head, dev_list) {
3248 		if (!dev->bdev) {
3249 			errors_send++;
3250 			continue;
3251 		}
3252 		if (!dev->in_fs_metadata || !dev->writeable)
3253 			continue;
3254 
3255 		ret = write_dev_flush(dev, 0);
3256 		if (ret)
3257 			errors_send++;
3258 	}
3259 
3260 	/* wait for all the barriers */
3261 	list_for_each_entry_rcu(dev, head, dev_list) {
3262 		if (!dev->bdev) {
3263 			errors_wait++;
3264 			continue;
3265 		}
3266 		if (!dev->in_fs_metadata || !dev->writeable)
3267 			continue;
3268 
3269 		ret = write_dev_flush(dev, 1);
3270 		if (ret)
3271 			errors_wait++;
3272 	}
3273 	if (errors_send > info->num_tolerated_disk_barrier_failures ||
3274 	    errors_wait > info->num_tolerated_disk_barrier_failures)
3275 		return -EIO;
3276 	return 0;
3277 }
3278 
3279 int btrfs_calc_num_tolerated_disk_barrier_failures(
3280 	struct btrfs_fs_info *fs_info)
3281 {
3282 	struct btrfs_ioctl_space_info space;
3283 	struct btrfs_space_info *sinfo;
3284 	u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3285 		       BTRFS_BLOCK_GROUP_SYSTEM,
3286 		       BTRFS_BLOCK_GROUP_METADATA,
3287 		       BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3288 	int num_types = 4;
3289 	int i;
3290 	int c;
3291 	int num_tolerated_disk_barrier_failures =
3292 		(int)fs_info->fs_devices->num_devices;
3293 
3294 	for (i = 0; i < num_types; i++) {
3295 		struct btrfs_space_info *tmp;
3296 
3297 		sinfo = NULL;
3298 		rcu_read_lock();
3299 		list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3300 			if (tmp->flags == types[i]) {
3301 				sinfo = tmp;
3302 				break;
3303 			}
3304 		}
3305 		rcu_read_unlock();
3306 
3307 		if (!sinfo)
3308 			continue;
3309 
3310 		down_read(&sinfo->groups_sem);
3311 		for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3312 			if (!list_empty(&sinfo->block_groups[c])) {
3313 				u64 flags;
3314 
3315 				btrfs_get_block_group_info(
3316 					&sinfo->block_groups[c], &space);
3317 				if (space.total_bytes == 0 ||
3318 				    space.used_bytes == 0)
3319 					continue;
3320 				flags = space.flags;
3321 				/*
3322 				 * return
3323 				 * 0: if dup, single or RAID0 is configured for
3324 				 *    any of metadata, system or data, else
3325 				 * 1: if RAID5 is configured, or if RAID1 or
3326 				 *    RAID10 is configured and only two mirrors
3327 				 *    are used, else
3328 				 * 2: if RAID6 is configured, else
3329 				 * num_mirrors - 1: if RAID1 or RAID10 is
3330 				 *                  configured and more than
3331 				 *                  2 mirrors are used.
3332 				 */
3333 				if (num_tolerated_disk_barrier_failures > 0 &&
3334 				    ((flags & (BTRFS_BLOCK_GROUP_DUP |
3335 					       BTRFS_BLOCK_GROUP_RAID0)) ||
3336 				     ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3337 				      == 0)))
3338 					num_tolerated_disk_barrier_failures = 0;
3339 				else if (num_tolerated_disk_barrier_failures > 1) {
3340 					if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3341 					    BTRFS_BLOCK_GROUP_RAID5 |
3342 					    BTRFS_BLOCK_GROUP_RAID10)) {
3343 						num_tolerated_disk_barrier_failures = 1;
3344 					} else if (flags &
3345 						   BTRFS_BLOCK_GROUP_RAID6) {
3346 						num_tolerated_disk_barrier_failures = 2;
3347 					}
3348 				}
3349 			}
3350 		}
3351 		up_read(&sinfo->groups_sem);
3352 	}
3353 
3354 	return num_tolerated_disk_barrier_failures;
3355 }
3356 
3357 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3358 {
3359 	struct list_head *head;
3360 	struct btrfs_device *dev;
3361 	struct btrfs_super_block *sb;
3362 	struct btrfs_dev_item *dev_item;
3363 	int ret;
3364 	int do_barriers;
3365 	int max_errors;
3366 	int total_errors = 0;
3367 	u64 flags;
3368 
3369 	do_barriers = !btrfs_test_opt(root, NOBARRIER);
3370 	backup_super_roots(root->fs_info);
3371 
3372 	sb = root->fs_info->super_for_commit;
3373 	dev_item = &sb->dev_item;
3374 
3375 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3376 	head = &root->fs_info->fs_devices->devices;
3377 	max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3378 
3379 	if (do_barriers) {
3380 		ret = barrier_all_devices(root->fs_info);
3381 		if (ret) {
3382 			mutex_unlock(
3383 				&root->fs_info->fs_devices->device_list_mutex);
3384 			btrfs_error(root->fs_info, ret,
3385 				    "errors while submitting device barriers.");
3386 			return ret;
3387 		}
3388 	}
3389 
3390 	list_for_each_entry_rcu(dev, head, dev_list) {
3391 		if (!dev->bdev) {
3392 			total_errors++;
3393 			continue;
3394 		}
3395 		if (!dev->in_fs_metadata || !dev->writeable)
3396 			continue;
3397 
3398 		btrfs_set_stack_device_generation(dev_item, 0);
3399 		btrfs_set_stack_device_type(dev_item, dev->type);
3400 		btrfs_set_stack_device_id(dev_item, dev->devid);
3401 		btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3402 		btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3403 		btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3404 		btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3405 		btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3406 		memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3407 		memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3408 
3409 		flags = btrfs_super_flags(sb);
3410 		btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3411 
3412 		ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3413 		if (ret)
3414 			total_errors++;
3415 	}
3416 	if (total_errors > max_errors) {
3417 		printk(KERN_ERR "btrfs: %d errors while writing supers\n",
3418 		       total_errors);
3419 		mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3420 
3421 		/* FUA is masked off if unsupported and can't be the reason */
3422 		btrfs_error(root->fs_info, -EIO,
3423 			    "%d errors while writing supers", total_errors);
3424 		return -EIO;
3425 	}
3426 
3427 	total_errors = 0;
3428 	list_for_each_entry_rcu(dev, head, dev_list) {
3429 		if (!dev->bdev)
3430 			continue;
3431 		if (!dev->in_fs_metadata || !dev->writeable)
3432 			continue;
3433 
3434 		ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3435 		if (ret)
3436 			total_errors++;
3437 	}
3438 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3439 	if (total_errors > max_errors) {
3440 		btrfs_error(root->fs_info, -EIO,
3441 			    "%d errors while writing supers", total_errors);
3442 		return -EIO;
3443 	}
3444 	return 0;
3445 }
3446 
3447 int write_ctree_super(struct btrfs_trans_handle *trans,
3448 		      struct btrfs_root *root, int max_mirrors)
3449 {
3450 	int ret;
3451 
3452 	ret = write_all_supers(root, max_mirrors);
3453 	return ret;
3454 }
3455 
3456 /* Drop a fs root from the radix tree and free it. */
3457 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3458 				  struct btrfs_root *root)
3459 {
3460 	spin_lock(&fs_info->fs_roots_radix_lock);
3461 	radix_tree_delete(&fs_info->fs_roots_radix,
3462 			  (unsigned long)root->root_key.objectid);
3463 	spin_unlock(&fs_info->fs_roots_radix_lock);
3464 
3465 	if (btrfs_root_refs(&root->root_item) == 0)
3466 		synchronize_srcu(&fs_info->subvol_srcu);
3467 
3468 	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3469 		btrfs_free_log(NULL, root);
3470 		btrfs_free_log_root_tree(NULL, fs_info);
3471 	}
3472 
3473 	__btrfs_remove_free_space_cache(root->free_ino_pinned);
3474 	__btrfs_remove_free_space_cache(root->free_ino_ctl);
3475 	free_fs_root(root);
3476 }
3477 
3478 static void free_fs_root(struct btrfs_root *root)
3479 {
3480 	iput(root->cache_inode);
3481 	WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3482 	btrfs_free_block_rsv(root, root->orphan_block_rsv);
3483 	root->orphan_block_rsv = NULL;
3484 	if (root->anon_dev)
3485 		free_anon_bdev(root->anon_dev);
3486 	free_extent_buffer(root->node);
3487 	free_extent_buffer(root->commit_root);
3488 	kfree(root->free_ino_ctl);
3489 	kfree(root->free_ino_pinned);
3490 	kfree(root->name);
3491 	btrfs_put_fs_root(root);
3492 }
3493 
3494 void btrfs_free_fs_root(struct btrfs_root *root)
3495 {
3496 	free_fs_root(root);
3497 }
3498 
3499 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3500 {
3501 	u64 root_objectid = 0;
3502 	struct btrfs_root *gang[8];
3503 	int i;
3504 	int ret;
3505 
3506 	while (1) {
3507 		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3508 					     (void **)gang, root_objectid,
3509 					     ARRAY_SIZE(gang));
3510 		if (!ret)
3511 			break;
3512 
3513 		root_objectid = gang[ret - 1]->root_key.objectid + 1;
3514 		for (i = 0; i < ret; i++) {
3515 			int err;
3516 
3517 			root_objectid = gang[i]->root_key.objectid;
3518 			err = btrfs_orphan_cleanup(gang[i]);
3519 			if (err)
3520 				return err;
3521 		}
3522 		root_objectid++;
3523 	}
3524 	return 0;
3525 }
3526 
3527 int btrfs_commit_super(struct btrfs_root *root)
3528 {
3529 	struct btrfs_trans_handle *trans;
3530 	int ret;
3531 
3532 	mutex_lock(&root->fs_info->cleaner_mutex);
3533 	btrfs_run_delayed_iputs(root);
3534 	mutex_unlock(&root->fs_info->cleaner_mutex);
3535 	wake_up_process(root->fs_info->cleaner_kthread);
3536 
3537 	/* wait until ongoing cleanup work done */
3538 	down_write(&root->fs_info->cleanup_work_sem);
3539 	up_write(&root->fs_info->cleanup_work_sem);
3540 
3541 	trans = btrfs_join_transaction(root);
3542 	if (IS_ERR(trans))
3543 		return PTR_ERR(trans);
3544 	ret = btrfs_commit_transaction(trans, root);
3545 	if (ret)
3546 		return ret;
3547 	/* run commit again to drop the original snapshot */
3548 	trans = btrfs_join_transaction(root);
3549 	if (IS_ERR(trans))
3550 		return PTR_ERR(trans);
3551 	ret = btrfs_commit_transaction(trans, root);
3552 	if (ret)
3553 		return ret;
3554 	ret = btrfs_write_and_wait_transaction(NULL, root);
3555 	if (ret) {
3556 		btrfs_error(root->fs_info, ret,
3557 			    "Failed to sync btree inode to disk.");
3558 		return ret;
3559 	}
3560 
3561 	ret = write_ctree_super(NULL, root, 0);
3562 	return ret;
3563 }
3564 
3565 int close_ctree(struct btrfs_root *root)
3566 {
3567 	struct btrfs_fs_info *fs_info = root->fs_info;
3568 	int ret;
3569 
3570 	fs_info->closing = 1;
3571 	smp_mb();
3572 
3573 	/* wait for the uuid_scan task to finish */
3574 	down(&fs_info->uuid_tree_rescan_sem);
3575 	/* avoid complains from lockdep et al., set sem back to initial state */
3576 	up(&fs_info->uuid_tree_rescan_sem);
3577 
3578 	/* pause restriper - we want to resume on mount */
3579 	btrfs_pause_balance(fs_info);
3580 
3581 	btrfs_dev_replace_suspend_for_unmount(fs_info);
3582 
3583 	btrfs_scrub_cancel(fs_info);
3584 
3585 	/* wait for any defraggers to finish */
3586 	wait_event(fs_info->transaction_wait,
3587 		   (atomic_read(&fs_info->defrag_running) == 0));
3588 
3589 	/* clear out the rbtree of defraggable inodes */
3590 	btrfs_cleanup_defrag_inodes(fs_info);
3591 
3592 	if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3593 		ret = btrfs_commit_super(root);
3594 		if (ret)
3595 			printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3596 	}
3597 
3598 	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3599 		btrfs_error_commit_super(root);
3600 
3601 	btrfs_put_block_group_cache(fs_info);
3602 
3603 	kthread_stop(fs_info->transaction_kthread);
3604 	kthread_stop(fs_info->cleaner_kthread);
3605 
3606 	fs_info->closing = 2;
3607 	smp_mb();
3608 
3609 	btrfs_free_qgroup_config(root->fs_info);
3610 
3611 	if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3612 		printk(KERN_INFO "btrfs: at unmount delalloc count %lld\n",
3613 		       percpu_counter_sum(&fs_info->delalloc_bytes));
3614 	}
3615 
3616 	btrfs_free_block_groups(fs_info);
3617 
3618 	btrfs_stop_all_workers(fs_info);
3619 
3620 	del_fs_roots(fs_info);
3621 
3622 	free_root_pointers(fs_info, 1);
3623 
3624 	iput(fs_info->btree_inode);
3625 
3626 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3627 	if (btrfs_test_opt(root, CHECK_INTEGRITY))
3628 		btrfsic_unmount(root, fs_info->fs_devices);
3629 #endif
3630 
3631 	btrfs_close_devices(fs_info->fs_devices);
3632 	btrfs_mapping_tree_free(&fs_info->mapping_tree);
3633 
3634 	percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3635 	percpu_counter_destroy(&fs_info->delalloc_bytes);
3636 	bdi_destroy(&fs_info->bdi);
3637 	cleanup_srcu_struct(&fs_info->subvol_srcu);
3638 
3639 	btrfs_free_stripe_hash_table(fs_info);
3640 
3641 	btrfs_free_block_rsv(root, root->orphan_block_rsv);
3642 	root->orphan_block_rsv = NULL;
3643 
3644 	return 0;
3645 }
3646 
3647 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3648 			  int atomic)
3649 {
3650 	int ret;
3651 	struct inode *btree_inode = buf->pages[0]->mapping->host;
3652 
3653 	ret = extent_buffer_uptodate(buf);
3654 	if (!ret)
3655 		return ret;
3656 
3657 	ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3658 				    parent_transid, atomic);
3659 	if (ret == -EAGAIN)
3660 		return ret;
3661 	return !ret;
3662 }
3663 
3664 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3665 {
3666 	return set_extent_buffer_uptodate(buf);
3667 }
3668 
3669 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3670 {
3671 	struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3672 	u64 transid = btrfs_header_generation(buf);
3673 	int was_dirty;
3674 
3675 	btrfs_assert_tree_locked(buf);
3676 	if (transid != root->fs_info->generation)
3677 		WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3678 		       "found %llu running %llu\n",
3679 			buf->start, transid, root->fs_info->generation);
3680 	was_dirty = set_extent_buffer_dirty(buf);
3681 	if (!was_dirty)
3682 		__percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3683 				     buf->len,
3684 				     root->fs_info->dirty_metadata_batch);
3685 }
3686 
3687 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3688 					int flush_delayed)
3689 {
3690 	/*
3691 	 * looks as though older kernels can get into trouble with
3692 	 * this code, they end up stuck in balance_dirty_pages forever
3693 	 */
3694 	int ret;
3695 
3696 	if (current->flags & PF_MEMALLOC)
3697 		return;
3698 
3699 	if (flush_delayed)
3700 		btrfs_balance_delayed_items(root);
3701 
3702 	ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3703 				     BTRFS_DIRTY_METADATA_THRESH);
3704 	if (ret > 0) {
3705 		balance_dirty_pages_ratelimited(
3706 				   root->fs_info->btree_inode->i_mapping);
3707 	}
3708 	return;
3709 }
3710 
3711 void btrfs_btree_balance_dirty(struct btrfs_root *root)
3712 {
3713 	__btrfs_btree_balance_dirty(root, 1);
3714 }
3715 
3716 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3717 {
3718 	__btrfs_btree_balance_dirty(root, 0);
3719 }
3720 
3721 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3722 {
3723 	struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3724 	return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3725 }
3726 
3727 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3728 			      int read_only)
3729 {
3730 	/*
3731 	 * Placeholder for checks
3732 	 */
3733 	return 0;
3734 }
3735 
3736 static void btrfs_error_commit_super(struct btrfs_root *root)
3737 {
3738 	mutex_lock(&root->fs_info->cleaner_mutex);
3739 	btrfs_run_delayed_iputs(root);
3740 	mutex_unlock(&root->fs_info->cleaner_mutex);
3741 
3742 	down_write(&root->fs_info->cleanup_work_sem);
3743 	up_write(&root->fs_info->cleanup_work_sem);
3744 
3745 	/* cleanup FS via transaction */
3746 	btrfs_cleanup_transaction(root);
3747 }
3748 
3749 static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3750 					     struct btrfs_root *root)
3751 {
3752 	struct btrfs_inode *btrfs_inode;
3753 	struct list_head splice;
3754 
3755 	INIT_LIST_HEAD(&splice);
3756 
3757 	mutex_lock(&root->fs_info->ordered_operations_mutex);
3758 	spin_lock(&root->fs_info->ordered_root_lock);
3759 
3760 	list_splice_init(&t->ordered_operations, &splice);
3761 	while (!list_empty(&splice)) {
3762 		btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3763 					 ordered_operations);
3764 
3765 		list_del_init(&btrfs_inode->ordered_operations);
3766 		spin_unlock(&root->fs_info->ordered_root_lock);
3767 
3768 		btrfs_invalidate_inodes(btrfs_inode->root);
3769 
3770 		spin_lock(&root->fs_info->ordered_root_lock);
3771 	}
3772 
3773 	spin_unlock(&root->fs_info->ordered_root_lock);
3774 	mutex_unlock(&root->fs_info->ordered_operations_mutex);
3775 }
3776 
3777 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3778 {
3779 	struct btrfs_ordered_extent *ordered;
3780 
3781 	spin_lock(&root->ordered_extent_lock);
3782 	/*
3783 	 * This will just short circuit the ordered completion stuff which will
3784 	 * make sure the ordered extent gets properly cleaned up.
3785 	 */
3786 	list_for_each_entry(ordered, &root->ordered_extents,
3787 			    root_extent_list)
3788 		set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
3789 	spin_unlock(&root->ordered_extent_lock);
3790 }
3791 
3792 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3793 {
3794 	struct btrfs_root *root;
3795 	struct list_head splice;
3796 
3797 	INIT_LIST_HEAD(&splice);
3798 
3799 	spin_lock(&fs_info->ordered_root_lock);
3800 	list_splice_init(&fs_info->ordered_roots, &splice);
3801 	while (!list_empty(&splice)) {
3802 		root = list_first_entry(&splice, struct btrfs_root,
3803 					ordered_root);
3804 		list_del_init(&root->ordered_root);
3805 
3806 		btrfs_destroy_ordered_extents(root);
3807 
3808 		cond_resched_lock(&fs_info->ordered_root_lock);
3809 	}
3810 	spin_unlock(&fs_info->ordered_root_lock);
3811 }
3812 
3813 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3814 				      struct btrfs_root *root)
3815 {
3816 	struct rb_node *node;
3817 	struct btrfs_delayed_ref_root *delayed_refs;
3818 	struct btrfs_delayed_ref_node *ref;
3819 	int ret = 0;
3820 
3821 	delayed_refs = &trans->delayed_refs;
3822 
3823 	spin_lock(&delayed_refs->lock);
3824 	if (delayed_refs->num_entries == 0) {
3825 		spin_unlock(&delayed_refs->lock);
3826 		printk(KERN_INFO "delayed_refs has NO entry\n");
3827 		return ret;
3828 	}
3829 
3830 	while ((node = rb_first(&delayed_refs->root)) != NULL) {
3831 		struct btrfs_delayed_ref_head *head = NULL;
3832 		bool pin_bytes = false;
3833 
3834 		ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3835 		atomic_set(&ref->refs, 1);
3836 		if (btrfs_delayed_ref_is_head(ref)) {
3837 
3838 			head = btrfs_delayed_node_to_head(ref);
3839 			if (!mutex_trylock(&head->mutex)) {
3840 				atomic_inc(&ref->refs);
3841 				spin_unlock(&delayed_refs->lock);
3842 
3843 				/* Need to wait for the delayed ref to run */
3844 				mutex_lock(&head->mutex);
3845 				mutex_unlock(&head->mutex);
3846 				btrfs_put_delayed_ref(ref);
3847 
3848 				spin_lock(&delayed_refs->lock);
3849 				continue;
3850 			}
3851 
3852 			if (head->must_insert_reserved)
3853 				pin_bytes = true;
3854 			btrfs_free_delayed_extent_op(head->extent_op);
3855 			delayed_refs->num_heads--;
3856 			if (list_empty(&head->cluster))
3857 				delayed_refs->num_heads_ready--;
3858 			list_del_init(&head->cluster);
3859 		}
3860 
3861 		ref->in_tree = 0;
3862 		rb_erase(&ref->rb_node, &delayed_refs->root);
3863 		delayed_refs->num_entries--;
3864 		spin_unlock(&delayed_refs->lock);
3865 		if (head) {
3866 			if (pin_bytes)
3867 				btrfs_pin_extent(root, ref->bytenr,
3868 						 ref->num_bytes, 1);
3869 			mutex_unlock(&head->mutex);
3870 		}
3871 		btrfs_put_delayed_ref(ref);
3872 
3873 		cond_resched();
3874 		spin_lock(&delayed_refs->lock);
3875 	}
3876 
3877 	spin_unlock(&delayed_refs->lock);
3878 
3879 	return ret;
3880 }
3881 
3882 static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t)
3883 {
3884 	struct btrfs_pending_snapshot *snapshot;
3885 	struct list_head splice;
3886 
3887 	INIT_LIST_HEAD(&splice);
3888 
3889 	list_splice_init(&t->pending_snapshots, &splice);
3890 
3891 	while (!list_empty(&splice)) {
3892 		snapshot = list_entry(splice.next,
3893 				      struct btrfs_pending_snapshot,
3894 				      list);
3895 		snapshot->error = -ECANCELED;
3896 		list_del_init(&snapshot->list);
3897 	}
3898 }
3899 
3900 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3901 {
3902 	struct btrfs_inode *btrfs_inode;
3903 	struct list_head splice;
3904 
3905 	INIT_LIST_HEAD(&splice);
3906 
3907 	spin_lock(&root->delalloc_lock);
3908 	list_splice_init(&root->delalloc_inodes, &splice);
3909 
3910 	while (!list_empty(&splice)) {
3911 		btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3912 					       delalloc_inodes);
3913 
3914 		list_del_init(&btrfs_inode->delalloc_inodes);
3915 		clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3916 			  &btrfs_inode->runtime_flags);
3917 		spin_unlock(&root->delalloc_lock);
3918 
3919 		btrfs_invalidate_inodes(btrfs_inode->root);
3920 
3921 		spin_lock(&root->delalloc_lock);
3922 	}
3923 
3924 	spin_unlock(&root->delalloc_lock);
3925 }
3926 
3927 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3928 {
3929 	struct btrfs_root *root;
3930 	struct list_head splice;
3931 
3932 	INIT_LIST_HEAD(&splice);
3933 
3934 	spin_lock(&fs_info->delalloc_root_lock);
3935 	list_splice_init(&fs_info->delalloc_roots, &splice);
3936 	while (!list_empty(&splice)) {
3937 		root = list_first_entry(&splice, struct btrfs_root,
3938 					 delalloc_root);
3939 		list_del_init(&root->delalloc_root);
3940 		root = btrfs_grab_fs_root(root);
3941 		BUG_ON(!root);
3942 		spin_unlock(&fs_info->delalloc_root_lock);
3943 
3944 		btrfs_destroy_delalloc_inodes(root);
3945 		btrfs_put_fs_root(root);
3946 
3947 		spin_lock(&fs_info->delalloc_root_lock);
3948 	}
3949 	spin_unlock(&fs_info->delalloc_root_lock);
3950 }
3951 
3952 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3953 					struct extent_io_tree *dirty_pages,
3954 					int mark)
3955 {
3956 	int ret;
3957 	struct extent_buffer *eb;
3958 	u64 start = 0;
3959 	u64 end;
3960 
3961 	while (1) {
3962 		ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3963 					    mark, NULL);
3964 		if (ret)
3965 			break;
3966 
3967 		clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3968 		while (start <= end) {
3969 			eb = btrfs_find_tree_block(root, start,
3970 						   root->leafsize);
3971 			start += root->leafsize;
3972 			if (!eb)
3973 				continue;
3974 			wait_on_extent_buffer_writeback(eb);
3975 
3976 			if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3977 					       &eb->bflags))
3978 				clear_extent_buffer_dirty(eb);
3979 			free_extent_buffer_stale(eb);
3980 		}
3981 	}
3982 
3983 	return ret;
3984 }
3985 
3986 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3987 				       struct extent_io_tree *pinned_extents)
3988 {
3989 	struct extent_io_tree *unpin;
3990 	u64 start;
3991 	u64 end;
3992 	int ret;
3993 	bool loop = true;
3994 
3995 	unpin = pinned_extents;
3996 again:
3997 	while (1) {
3998 		ret = find_first_extent_bit(unpin, 0, &start, &end,
3999 					    EXTENT_DIRTY, NULL);
4000 		if (ret)
4001 			break;
4002 
4003 		/* opt_discard */
4004 		if (btrfs_test_opt(root, DISCARD))
4005 			ret = btrfs_error_discard_extent(root, start,
4006 							 end + 1 - start,
4007 							 NULL);
4008 
4009 		clear_extent_dirty(unpin, start, end, GFP_NOFS);
4010 		btrfs_error_unpin_extent_range(root, start, end);
4011 		cond_resched();
4012 	}
4013 
4014 	if (loop) {
4015 		if (unpin == &root->fs_info->freed_extents[0])
4016 			unpin = &root->fs_info->freed_extents[1];
4017 		else
4018 			unpin = &root->fs_info->freed_extents[0];
4019 		loop = false;
4020 		goto again;
4021 	}
4022 
4023 	return 0;
4024 }
4025 
4026 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4027 				   struct btrfs_root *root)
4028 {
4029 	btrfs_destroy_delayed_refs(cur_trans, root);
4030 	btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
4031 				cur_trans->dirty_pages.dirty_bytes);
4032 
4033 	cur_trans->state = TRANS_STATE_COMMIT_START;
4034 	wake_up(&root->fs_info->transaction_blocked_wait);
4035 
4036 	btrfs_evict_pending_snapshots(cur_trans);
4037 
4038 	cur_trans->state = TRANS_STATE_UNBLOCKED;
4039 	wake_up(&root->fs_info->transaction_wait);
4040 
4041 	btrfs_destroy_delayed_inodes(root);
4042 	btrfs_assert_delayed_root_empty(root);
4043 
4044 	btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4045 				     EXTENT_DIRTY);
4046 	btrfs_destroy_pinned_extent(root,
4047 				    root->fs_info->pinned_extents);
4048 
4049 	cur_trans->state =TRANS_STATE_COMPLETED;
4050 	wake_up(&cur_trans->commit_wait);
4051 
4052 	/*
4053 	memset(cur_trans, 0, sizeof(*cur_trans));
4054 	kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4055 	*/
4056 }
4057 
4058 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4059 {
4060 	struct btrfs_transaction *t;
4061 	LIST_HEAD(list);
4062 
4063 	mutex_lock(&root->fs_info->transaction_kthread_mutex);
4064 
4065 	spin_lock(&root->fs_info->trans_lock);
4066 	list_splice_init(&root->fs_info->trans_list, &list);
4067 	root->fs_info->running_transaction = NULL;
4068 	spin_unlock(&root->fs_info->trans_lock);
4069 
4070 	while (!list_empty(&list)) {
4071 		t = list_entry(list.next, struct btrfs_transaction, list);
4072 
4073 		btrfs_destroy_ordered_operations(t, root);
4074 
4075 		btrfs_destroy_all_ordered_extents(root->fs_info);
4076 
4077 		btrfs_destroy_delayed_refs(t, root);
4078 
4079 		/*
4080 		 *  FIXME: cleanup wait for commit
4081 		 *  We needn't acquire the lock here, because we are during
4082 		 *  the umount, there is no other task which will change it.
4083 		 */
4084 		t->state = TRANS_STATE_COMMIT_START;
4085 		smp_mb();
4086 		if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
4087 			wake_up(&root->fs_info->transaction_blocked_wait);
4088 
4089 		btrfs_evict_pending_snapshots(t);
4090 
4091 		t->state = TRANS_STATE_UNBLOCKED;
4092 		smp_mb();
4093 		if (waitqueue_active(&root->fs_info->transaction_wait))
4094 			wake_up(&root->fs_info->transaction_wait);
4095 
4096 		btrfs_destroy_delayed_inodes(root);
4097 		btrfs_assert_delayed_root_empty(root);
4098 
4099 		btrfs_destroy_all_delalloc_inodes(root->fs_info);
4100 
4101 		btrfs_destroy_marked_extents(root, &t->dirty_pages,
4102 					     EXTENT_DIRTY);
4103 
4104 		btrfs_destroy_pinned_extent(root,
4105 					    root->fs_info->pinned_extents);
4106 
4107 		t->state = TRANS_STATE_COMPLETED;
4108 		smp_mb();
4109 		if (waitqueue_active(&t->commit_wait))
4110 			wake_up(&t->commit_wait);
4111 
4112 		atomic_set(&t->use_count, 0);
4113 		list_del_init(&t->list);
4114 		memset(t, 0, sizeof(*t));
4115 		kmem_cache_free(btrfs_transaction_cachep, t);
4116 	}
4117 
4118 	mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4119 
4120 	return 0;
4121 }
4122 
4123 static struct extent_io_ops btree_extent_io_ops = {
4124 	.readpage_end_io_hook = btree_readpage_end_io_hook,
4125 	.readpage_io_failed_hook = btree_io_failed_hook,
4126 	.submit_bio_hook = btree_submit_bio_hook,
4127 	/* note we're sharing with inode.c for the merge bio hook */
4128 	.merge_bio_hook = btrfs_merge_bio_hook,
4129 };
4130