xref: /linux/fs/btrfs/disk-io.c (revision 6eb2fb3170549737207974c2c6ad34bcc2f3025e)
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 <asm/unaligned.h>
35 #include "compat.h"
36 #include "ctree.h"
37 #include "disk-io.h"
38 #include "transaction.h"
39 #include "btrfs_inode.h"
40 #include "volumes.h"
41 #include "print-tree.h"
42 #include "async-thread.h"
43 #include "locking.h"
44 #include "tree-log.h"
45 #include "free-space-cache.h"
46 #include "inode-map.h"
47 #include "check-integrity.h"
48 #include "rcu-string.h"
49 #include "dev-replace.h"
50 #include "raid56.h"
51 
52 #ifdef CONFIG_X86
53 #include <asm/cpufeature.h>
54 #endif
55 
56 static struct extent_io_ops btree_extent_io_ops;
57 static void end_workqueue_fn(struct btrfs_work *work);
58 static void free_fs_root(struct btrfs_root *root);
59 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
60 				    int read_only);
61 static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
62 					     struct btrfs_root *root);
63 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
64 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
65 				      struct btrfs_root *root);
66 static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t);
67 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
68 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
69 					struct extent_io_tree *dirty_pages,
70 					int mark);
71 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
72 				       struct extent_io_tree *pinned_extents);
73 static int btrfs_cleanup_transaction(struct btrfs_root *root);
74 static void btrfs_error_commit_super(struct btrfs_root *root);
75 
76 /*
77  * end_io_wq structs are used to do processing in task context when an IO is
78  * complete.  This is used during reads to verify checksums, and it is used
79  * by writes to insert metadata for new file extents after IO is complete.
80  */
81 struct end_io_wq {
82 	struct bio *bio;
83 	bio_end_io_t *end_io;
84 	void *private;
85 	struct btrfs_fs_info *info;
86 	int error;
87 	int metadata;
88 	struct list_head list;
89 	struct btrfs_work work;
90 };
91 
92 /*
93  * async submit bios are used to offload expensive checksumming
94  * onto the worker threads.  They checksum file and metadata bios
95  * just before they are sent down the IO stack.
96  */
97 struct async_submit_bio {
98 	struct inode *inode;
99 	struct bio *bio;
100 	struct list_head list;
101 	extent_submit_bio_hook_t *submit_bio_start;
102 	extent_submit_bio_hook_t *submit_bio_done;
103 	int rw;
104 	int mirror_num;
105 	unsigned long bio_flags;
106 	/*
107 	 * bio_offset is optional, can be used if the pages in the bio
108 	 * can't tell us where in the file the bio should go
109 	 */
110 	u64 bio_offset;
111 	struct btrfs_work work;
112 	int error;
113 };
114 
115 /*
116  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
117  * eb, the lockdep key is determined by the btrfs_root it belongs to and
118  * the level the eb occupies in the tree.
119  *
120  * Different roots are used for different purposes and may nest inside each
121  * other and they require separate keysets.  As lockdep keys should be
122  * static, assign keysets according to the purpose of the root as indicated
123  * by btrfs_root->objectid.  This ensures that all special purpose roots
124  * have separate keysets.
125  *
126  * Lock-nesting across peer nodes is always done with the immediate parent
127  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
128  * subclass to avoid triggering lockdep warning in such cases.
129  *
130  * The key is set by the readpage_end_io_hook after the buffer has passed
131  * csum validation but before the pages are unlocked.  It is also set by
132  * btrfs_init_new_buffer on freshly allocated blocks.
133  *
134  * We also add a check to make sure the highest level of the tree is the
135  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
136  * needs update as well.
137  */
138 #ifdef CONFIG_DEBUG_LOCK_ALLOC
139 # if BTRFS_MAX_LEVEL != 8
140 #  error
141 # endif
142 
143 static struct btrfs_lockdep_keyset {
144 	u64			id;		/* root objectid */
145 	const char		*name_stem;	/* lock name stem */
146 	char			names[BTRFS_MAX_LEVEL + 1][20];
147 	struct lock_class_key	keys[BTRFS_MAX_LEVEL + 1];
148 } btrfs_lockdep_keysets[] = {
149 	{ .id = BTRFS_ROOT_TREE_OBJECTID,	.name_stem = "root"	},
150 	{ .id = BTRFS_EXTENT_TREE_OBJECTID,	.name_stem = "extent"	},
151 	{ .id = BTRFS_CHUNK_TREE_OBJECTID,	.name_stem = "chunk"	},
152 	{ .id = BTRFS_DEV_TREE_OBJECTID,	.name_stem = "dev"	},
153 	{ .id = BTRFS_FS_TREE_OBJECTID,		.name_stem = "fs"	},
154 	{ .id = BTRFS_CSUM_TREE_OBJECTID,	.name_stem = "csum"	},
155 	{ .id = BTRFS_ORPHAN_OBJECTID,		.name_stem = "orphan"	},
156 	{ .id = BTRFS_TREE_LOG_OBJECTID,	.name_stem = "log"	},
157 	{ .id = BTRFS_TREE_RELOC_OBJECTID,	.name_stem = "treloc"	},
158 	{ .id = BTRFS_DATA_RELOC_TREE_OBJECTID,	.name_stem = "dreloc"	},
159 	{ .id = 0,				.name_stem = "tree"	},
160 };
161 
162 void __init btrfs_init_lockdep(void)
163 {
164 	int i, j;
165 
166 	/* initialize lockdep class names */
167 	for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
168 		struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
169 
170 		for (j = 0; j < ARRAY_SIZE(ks->names); j++)
171 			snprintf(ks->names[j], sizeof(ks->names[j]),
172 				 "btrfs-%s-%02d", ks->name_stem, j);
173 	}
174 }
175 
176 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
177 				    int level)
178 {
179 	struct btrfs_lockdep_keyset *ks;
180 
181 	BUG_ON(level >= ARRAY_SIZE(ks->keys));
182 
183 	/* find the matching keyset, id 0 is the default entry */
184 	for (ks = btrfs_lockdep_keysets; ks->id; ks++)
185 		if (ks->id == objectid)
186 			break;
187 
188 	lockdep_set_class_and_name(&eb->lock,
189 				   &ks->keys[level], ks->names[level]);
190 }
191 
192 #endif
193 
194 /*
195  * extents on the btree inode are pretty simple, there's one extent
196  * that covers the entire device
197  */
198 static struct extent_map *btree_get_extent(struct inode *inode,
199 		struct page *page, size_t pg_offset, u64 start, u64 len,
200 		int create)
201 {
202 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
203 	struct extent_map *em;
204 	int ret;
205 
206 	read_lock(&em_tree->lock);
207 	em = lookup_extent_mapping(em_tree, start, len);
208 	if (em) {
209 		em->bdev =
210 			BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
211 		read_unlock(&em_tree->lock);
212 		goto out;
213 	}
214 	read_unlock(&em_tree->lock);
215 
216 	em = alloc_extent_map();
217 	if (!em) {
218 		em = ERR_PTR(-ENOMEM);
219 		goto out;
220 	}
221 	em->start = 0;
222 	em->len = (u64)-1;
223 	em->block_len = (u64)-1;
224 	em->block_start = 0;
225 	em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
226 
227 	write_lock(&em_tree->lock);
228 	ret = add_extent_mapping(em_tree, em, 0);
229 	if (ret == -EEXIST) {
230 		free_extent_map(em);
231 		em = lookup_extent_mapping(em_tree, start, len);
232 		if (!em)
233 			em = ERR_PTR(-EIO);
234 	} else if (ret) {
235 		free_extent_map(em);
236 		em = ERR_PTR(ret);
237 	}
238 	write_unlock(&em_tree->lock);
239 
240 out:
241 	return em;
242 }
243 
244 u32 btrfs_csum_data(char *data, u32 seed, size_t len)
245 {
246 	return crc32c(seed, data, len);
247 }
248 
249 void btrfs_csum_final(u32 crc, char *result)
250 {
251 	put_unaligned_le32(~crc, result);
252 }
253 
254 /*
255  * compute the csum for a btree block, and either verify it or write it
256  * into the csum field of the block.
257  */
258 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
259 			   int verify)
260 {
261 	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
262 	char *result = NULL;
263 	unsigned long len;
264 	unsigned long cur_len;
265 	unsigned long offset = BTRFS_CSUM_SIZE;
266 	char *kaddr;
267 	unsigned long map_start;
268 	unsigned long map_len;
269 	int err;
270 	u32 crc = ~(u32)0;
271 	unsigned long inline_result;
272 
273 	len = buf->len - offset;
274 	while (len > 0) {
275 		err = map_private_extent_buffer(buf, offset, 32,
276 					&kaddr, &map_start, &map_len);
277 		if (err)
278 			return 1;
279 		cur_len = min(len, map_len - (offset - map_start));
280 		crc = btrfs_csum_data(kaddr + offset - map_start,
281 				      crc, cur_len);
282 		len -= cur_len;
283 		offset += cur_len;
284 	}
285 	if (csum_size > sizeof(inline_result)) {
286 		result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
287 		if (!result)
288 			return 1;
289 	} else {
290 		result = (char *)&inline_result;
291 	}
292 
293 	btrfs_csum_final(crc, result);
294 
295 	if (verify) {
296 		if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
297 			u32 val;
298 			u32 found = 0;
299 			memcpy(&found, result, csum_size);
300 
301 			read_extent_buffer(buf, &val, 0, csum_size);
302 			printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
303 				       "failed on %llu wanted %X found %X "
304 				       "level %d\n",
305 				       root->fs_info->sb->s_id,
306 				       (unsigned long long)buf->start, val, found,
307 				       btrfs_header_level(buf));
308 			if (result != (char *)&inline_result)
309 				kfree(result);
310 			return 1;
311 		}
312 	} else {
313 		write_extent_buffer(buf, result, 0, csum_size);
314 	}
315 	if (result != (char *)&inline_result)
316 		kfree(result);
317 	return 0;
318 }
319 
320 /*
321  * we can't consider a given block up to date unless the transid of the
322  * block matches the transid in the parent node's pointer.  This is how we
323  * detect blocks that either didn't get written at all or got written
324  * in the wrong place.
325  */
326 static int verify_parent_transid(struct extent_io_tree *io_tree,
327 				 struct extent_buffer *eb, u64 parent_transid,
328 				 int atomic)
329 {
330 	struct extent_state *cached_state = NULL;
331 	int ret;
332 
333 	if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
334 		return 0;
335 
336 	if (atomic)
337 		return -EAGAIN;
338 
339 	lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
340 			 0, &cached_state);
341 	if (extent_buffer_uptodate(eb) &&
342 	    btrfs_header_generation(eb) == parent_transid) {
343 		ret = 0;
344 		goto out;
345 	}
346 	printk_ratelimited("parent transid verify failed on %llu wanted %llu "
347 		       "found %llu\n",
348 		       (unsigned long long)eb->start,
349 		       (unsigned long long)parent_transid,
350 		       (unsigned long long)btrfs_header_generation(eb));
351 	ret = 1;
352 	clear_extent_buffer_uptodate(eb);
353 out:
354 	unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
355 			     &cached_state, GFP_NOFS);
356 	return ret;
357 }
358 
359 /*
360  * Return 0 if the superblock checksum type matches the checksum value of that
361  * algorithm. Pass the raw disk superblock data.
362  */
363 static int btrfs_check_super_csum(char *raw_disk_sb)
364 {
365 	struct btrfs_super_block *disk_sb =
366 		(struct btrfs_super_block *)raw_disk_sb;
367 	u16 csum_type = btrfs_super_csum_type(disk_sb);
368 	int ret = 0;
369 
370 	if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
371 		u32 crc = ~(u32)0;
372 		const int csum_size = sizeof(crc);
373 		char result[csum_size];
374 
375 		/*
376 		 * The super_block structure does not span the whole
377 		 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
378 		 * is filled with zeros and is included in the checkum.
379 		 */
380 		crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
381 				crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
382 		btrfs_csum_final(crc, result);
383 
384 		if (memcmp(raw_disk_sb, result, csum_size))
385 			ret = 1;
386 
387 		if (ret && btrfs_super_generation(disk_sb) < 10) {
388 			printk(KERN_WARNING "btrfs: super block crcs don't match, older mkfs detected\n");
389 			ret = 0;
390 		}
391 	}
392 
393 	if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
394 		printk(KERN_ERR "btrfs: unsupported checksum algorithm %u\n",
395 				csum_type);
396 		ret = 1;
397 	}
398 
399 	return ret;
400 }
401 
402 /*
403  * helper to read a given tree block, doing retries as required when
404  * the checksums don't match and we have alternate mirrors to try.
405  */
406 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
407 					  struct extent_buffer *eb,
408 					  u64 start, u64 parent_transid)
409 {
410 	struct extent_io_tree *io_tree;
411 	int failed = 0;
412 	int ret;
413 	int num_copies = 0;
414 	int mirror_num = 0;
415 	int failed_mirror = 0;
416 
417 	clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
418 	io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
419 	while (1) {
420 		ret = read_extent_buffer_pages(io_tree, eb, start,
421 					       WAIT_COMPLETE,
422 					       btree_get_extent, mirror_num);
423 		if (!ret) {
424 			if (!verify_parent_transid(io_tree, eb,
425 						   parent_transid, 0))
426 				break;
427 			else
428 				ret = -EIO;
429 		}
430 
431 		/*
432 		 * This buffer's crc is fine, but its contents are corrupted, so
433 		 * there is no reason to read the other copies, they won't be
434 		 * any less wrong.
435 		 */
436 		if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
437 			break;
438 
439 		num_copies = btrfs_num_copies(root->fs_info,
440 					      eb->start, eb->len);
441 		if (num_copies == 1)
442 			break;
443 
444 		if (!failed_mirror) {
445 			failed = 1;
446 			failed_mirror = eb->read_mirror;
447 		}
448 
449 		mirror_num++;
450 		if (mirror_num == failed_mirror)
451 			mirror_num++;
452 
453 		if (mirror_num > num_copies)
454 			break;
455 	}
456 
457 	if (failed && !ret && failed_mirror)
458 		repair_eb_io_failure(root, eb, failed_mirror);
459 
460 	return ret;
461 }
462 
463 /*
464  * checksum a dirty tree block before IO.  This has extra checks to make sure
465  * we only fill in the checksum field in the first page of a multi-page block
466  */
467 
468 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
469 {
470 	struct extent_io_tree *tree;
471 	u64 start = page_offset(page);
472 	u64 found_start;
473 	struct extent_buffer *eb;
474 
475 	tree = &BTRFS_I(page->mapping->host)->io_tree;
476 
477 	eb = (struct extent_buffer *)page->private;
478 	if (page != eb->pages[0])
479 		return 0;
480 	found_start = btrfs_header_bytenr(eb);
481 	if (found_start != start) {
482 		WARN_ON(1);
483 		return 0;
484 	}
485 	if (!PageUptodate(page)) {
486 		WARN_ON(1);
487 		return 0;
488 	}
489 	csum_tree_block(root, eb, 0);
490 	return 0;
491 }
492 
493 static int check_tree_block_fsid(struct btrfs_root *root,
494 				 struct extent_buffer *eb)
495 {
496 	struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
497 	u8 fsid[BTRFS_UUID_SIZE];
498 	int ret = 1;
499 
500 	read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
501 			   BTRFS_FSID_SIZE);
502 	while (fs_devices) {
503 		if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
504 			ret = 0;
505 			break;
506 		}
507 		fs_devices = fs_devices->seed;
508 	}
509 	return ret;
510 }
511 
512 #define CORRUPT(reason, eb, root, slot)				\
513 	printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu,"	\
514 	       "root=%llu, slot=%d\n", reason,			\
515 	       (unsigned long long)btrfs_header_bytenr(eb),	\
516 	       (unsigned long long)root->objectid, slot)
517 
518 static noinline int check_leaf(struct btrfs_root *root,
519 			       struct extent_buffer *leaf)
520 {
521 	struct btrfs_key key;
522 	struct btrfs_key leaf_key;
523 	u32 nritems = btrfs_header_nritems(leaf);
524 	int slot;
525 
526 	if (nritems == 0)
527 		return 0;
528 
529 	/* Check the 0 item */
530 	if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
531 	    BTRFS_LEAF_DATA_SIZE(root)) {
532 		CORRUPT("invalid item offset size pair", leaf, root, 0);
533 		return -EIO;
534 	}
535 
536 	/*
537 	 * Check to make sure each items keys are in the correct order and their
538 	 * offsets make sense.  We only have to loop through nritems-1 because
539 	 * we check the current slot against the next slot, which verifies the
540 	 * next slot's offset+size makes sense and that the current's slot
541 	 * offset is correct.
542 	 */
543 	for (slot = 0; slot < nritems - 1; slot++) {
544 		btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
545 		btrfs_item_key_to_cpu(leaf, &key, slot + 1);
546 
547 		/* Make sure the keys are in the right order */
548 		if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
549 			CORRUPT("bad key order", leaf, root, slot);
550 			return -EIO;
551 		}
552 
553 		/*
554 		 * Make sure the offset and ends are right, remember that the
555 		 * item data starts at the end of the leaf and grows towards the
556 		 * front.
557 		 */
558 		if (btrfs_item_offset_nr(leaf, slot) !=
559 			btrfs_item_end_nr(leaf, slot + 1)) {
560 			CORRUPT("slot offset bad", leaf, root, slot);
561 			return -EIO;
562 		}
563 
564 		/*
565 		 * Check to make sure that we don't point outside of the leaf,
566 		 * just incase all the items are consistent to eachother, but
567 		 * all point outside of the leaf.
568 		 */
569 		if (btrfs_item_end_nr(leaf, slot) >
570 		    BTRFS_LEAF_DATA_SIZE(root)) {
571 			CORRUPT("slot end outside of leaf", leaf, root, slot);
572 			return -EIO;
573 		}
574 	}
575 
576 	return 0;
577 }
578 
579 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
580 			       struct extent_state *state, int mirror)
581 {
582 	struct extent_io_tree *tree;
583 	u64 found_start;
584 	int found_level;
585 	struct extent_buffer *eb;
586 	struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
587 	int ret = 0;
588 	int reads_done;
589 
590 	if (!page->private)
591 		goto out;
592 
593 	tree = &BTRFS_I(page->mapping->host)->io_tree;
594 	eb = (struct extent_buffer *)page->private;
595 
596 	/* the pending IO might have been the only thing that kept this buffer
597 	 * in memory.  Make sure we have a ref for all this other checks
598 	 */
599 	extent_buffer_get(eb);
600 
601 	reads_done = atomic_dec_and_test(&eb->io_pages);
602 	if (!reads_done)
603 		goto err;
604 
605 	eb->read_mirror = mirror;
606 	if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
607 		ret = -EIO;
608 		goto err;
609 	}
610 
611 	found_start = btrfs_header_bytenr(eb);
612 	if (found_start != eb->start) {
613 		printk_ratelimited(KERN_INFO "btrfs bad tree block start "
614 			       "%llu %llu\n",
615 			       (unsigned long long)found_start,
616 			       (unsigned long long)eb->start);
617 		ret = -EIO;
618 		goto err;
619 	}
620 	if (check_tree_block_fsid(root, eb)) {
621 		printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
622 			       (unsigned long long)eb->start);
623 		ret = -EIO;
624 		goto err;
625 	}
626 	found_level = btrfs_header_level(eb);
627 	if (found_level >= BTRFS_MAX_LEVEL) {
628 		btrfs_info(root->fs_info, "bad tree block level %d\n",
629 			   (int)btrfs_header_level(eb));
630 		ret = -EIO;
631 		goto err;
632 	}
633 
634 	btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
635 				       eb, found_level);
636 
637 	ret = csum_tree_block(root, eb, 1);
638 	if (ret) {
639 		ret = -EIO;
640 		goto err;
641 	}
642 
643 	/*
644 	 * If this is a leaf block and it is corrupt, set the corrupt bit so
645 	 * that we don't try and read the other copies of this block, just
646 	 * return -EIO.
647 	 */
648 	if (found_level == 0 && check_leaf(root, eb)) {
649 		set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
650 		ret = -EIO;
651 	}
652 
653 	if (!ret)
654 		set_extent_buffer_uptodate(eb);
655 err:
656 	if (reads_done &&
657 	    test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
658 		btree_readahead_hook(root, eb, eb->start, ret);
659 
660 	if (ret) {
661 		/*
662 		 * our io error hook is going to dec the io pages
663 		 * again, we have to make sure it has something
664 		 * to decrement
665 		 */
666 		atomic_inc(&eb->io_pages);
667 		clear_extent_buffer_uptodate(eb);
668 	}
669 	free_extent_buffer(eb);
670 out:
671 	return ret;
672 }
673 
674 static int btree_io_failed_hook(struct page *page, int failed_mirror)
675 {
676 	struct extent_buffer *eb;
677 	struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
678 
679 	eb = (struct extent_buffer *)page->private;
680 	set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
681 	eb->read_mirror = failed_mirror;
682 	atomic_dec(&eb->io_pages);
683 	if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
684 		btree_readahead_hook(root, eb, eb->start, -EIO);
685 	return -EIO;	/* we fixed nothing */
686 }
687 
688 static void end_workqueue_bio(struct bio *bio, int err)
689 {
690 	struct end_io_wq *end_io_wq = bio->bi_private;
691 	struct btrfs_fs_info *fs_info;
692 
693 	fs_info = end_io_wq->info;
694 	end_io_wq->error = err;
695 	end_io_wq->work.func = end_workqueue_fn;
696 	end_io_wq->work.flags = 0;
697 
698 	if (bio->bi_rw & REQ_WRITE) {
699 		if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
700 			btrfs_queue_worker(&fs_info->endio_meta_write_workers,
701 					   &end_io_wq->work);
702 		else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
703 			btrfs_queue_worker(&fs_info->endio_freespace_worker,
704 					   &end_io_wq->work);
705 		else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
706 			btrfs_queue_worker(&fs_info->endio_raid56_workers,
707 					   &end_io_wq->work);
708 		else
709 			btrfs_queue_worker(&fs_info->endio_write_workers,
710 					   &end_io_wq->work);
711 	} else {
712 		if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
713 			btrfs_queue_worker(&fs_info->endio_raid56_workers,
714 					   &end_io_wq->work);
715 		else if (end_io_wq->metadata)
716 			btrfs_queue_worker(&fs_info->endio_meta_workers,
717 					   &end_io_wq->work);
718 		else
719 			btrfs_queue_worker(&fs_info->endio_workers,
720 					   &end_io_wq->work);
721 	}
722 }
723 
724 /*
725  * For the metadata arg you want
726  *
727  * 0 - if data
728  * 1 - if normal metadta
729  * 2 - if writing to the free space cache area
730  * 3 - raid parity work
731  */
732 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
733 			int metadata)
734 {
735 	struct end_io_wq *end_io_wq;
736 	end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
737 	if (!end_io_wq)
738 		return -ENOMEM;
739 
740 	end_io_wq->private = bio->bi_private;
741 	end_io_wq->end_io = bio->bi_end_io;
742 	end_io_wq->info = info;
743 	end_io_wq->error = 0;
744 	end_io_wq->bio = bio;
745 	end_io_wq->metadata = metadata;
746 
747 	bio->bi_private = end_io_wq;
748 	bio->bi_end_io = end_workqueue_bio;
749 	return 0;
750 }
751 
752 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
753 {
754 	unsigned long limit = min_t(unsigned long,
755 				    info->workers.max_workers,
756 				    info->fs_devices->open_devices);
757 	return 256 * limit;
758 }
759 
760 static void run_one_async_start(struct btrfs_work *work)
761 {
762 	struct async_submit_bio *async;
763 	int ret;
764 
765 	async = container_of(work, struct  async_submit_bio, work);
766 	ret = async->submit_bio_start(async->inode, async->rw, async->bio,
767 				      async->mirror_num, async->bio_flags,
768 				      async->bio_offset);
769 	if (ret)
770 		async->error = ret;
771 }
772 
773 static void run_one_async_done(struct btrfs_work *work)
774 {
775 	struct btrfs_fs_info *fs_info;
776 	struct async_submit_bio *async;
777 	int limit;
778 
779 	async = container_of(work, struct  async_submit_bio, work);
780 	fs_info = BTRFS_I(async->inode)->root->fs_info;
781 
782 	limit = btrfs_async_submit_limit(fs_info);
783 	limit = limit * 2 / 3;
784 
785 	if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
786 	    waitqueue_active(&fs_info->async_submit_wait))
787 		wake_up(&fs_info->async_submit_wait);
788 
789 	/* If an error occured we just want to clean up the bio and move on */
790 	if (async->error) {
791 		bio_endio(async->bio, async->error);
792 		return;
793 	}
794 
795 	async->submit_bio_done(async->inode, async->rw, async->bio,
796 			       async->mirror_num, async->bio_flags,
797 			       async->bio_offset);
798 }
799 
800 static void run_one_async_free(struct btrfs_work *work)
801 {
802 	struct async_submit_bio *async;
803 
804 	async = container_of(work, struct  async_submit_bio, work);
805 	kfree(async);
806 }
807 
808 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
809 			int rw, struct bio *bio, int mirror_num,
810 			unsigned long bio_flags,
811 			u64 bio_offset,
812 			extent_submit_bio_hook_t *submit_bio_start,
813 			extent_submit_bio_hook_t *submit_bio_done)
814 {
815 	struct async_submit_bio *async;
816 
817 	async = kmalloc(sizeof(*async), GFP_NOFS);
818 	if (!async)
819 		return -ENOMEM;
820 
821 	async->inode = inode;
822 	async->rw = rw;
823 	async->bio = bio;
824 	async->mirror_num = mirror_num;
825 	async->submit_bio_start = submit_bio_start;
826 	async->submit_bio_done = submit_bio_done;
827 
828 	async->work.func = run_one_async_start;
829 	async->work.ordered_func = run_one_async_done;
830 	async->work.ordered_free = run_one_async_free;
831 
832 	async->work.flags = 0;
833 	async->bio_flags = bio_flags;
834 	async->bio_offset = bio_offset;
835 
836 	async->error = 0;
837 
838 	atomic_inc(&fs_info->nr_async_submits);
839 
840 	if (rw & REQ_SYNC)
841 		btrfs_set_work_high_prio(&async->work);
842 
843 	btrfs_queue_worker(&fs_info->workers, &async->work);
844 
845 	while (atomic_read(&fs_info->async_submit_draining) &&
846 	      atomic_read(&fs_info->nr_async_submits)) {
847 		wait_event(fs_info->async_submit_wait,
848 			   (atomic_read(&fs_info->nr_async_submits) == 0));
849 	}
850 
851 	return 0;
852 }
853 
854 static int btree_csum_one_bio(struct bio *bio)
855 {
856 	struct bio_vec *bvec = bio->bi_io_vec;
857 	int bio_index = 0;
858 	struct btrfs_root *root;
859 	int ret = 0;
860 
861 	WARN_ON(bio->bi_vcnt <= 0);
862 	while (bio_index < bio->bi_vcnt) {
863 		root = BTRFS_I(bvec->bv_page->mapping->host)->root;
864 		ret = csum_dirty_buffer(root, bvec->bv_page);
865 		if (ret)
866 			break;
867 		bio_index++;
868 		bvec++;
869 	}
870 	return ret;
871 }
872 
873 static int __btree_submit_bio_start(struct inode *inode, int rw,
874 				    struct bio *bio, int mirror_num,
875 				    unsigned long bio_flags,
876 				    u64 bio_offset)
877 {
878 	/*
879 	 * when we're called for a write, we're already in the async
880 	 * submission context.  Just jump into btrfs_map_bio
881 	 */
882 	return btree_csum_one_bio(bio);
883 }
884 
885 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
886 				 int mirror_num, unsigned long bio_flags,
887 				 u64 bio_offset)
888 {
889 	int ret;
890 
891 	/*
892 	 * when we're called for a write, we're already in the async
893 	 * submission context.  Just jump into btrfs_map_bio
894 	 */
895 	ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
896 	if (ret)
897 		bio_endio(bio, ret);
898 	return ret;
899 }
900 
901 static int check_async_write(struct inode *inode, unsigned long bio_flags)
902 {
903 	if (bio_flags & EXTENT_BIO_TREE_LOG)
904 		return 0;
905 #ifdef CONFIG_X86
906 	if (cpu_has_xmm4_2)
907 		return 0;
908 #endif
909 	return 1;
910 }
911 
912 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
913 				 int mirror_num, unsigned long bio_flags,
914 				 u64 bio_offset)
915 {
916 	int async = check_async_write(inode, bio_flags);
917 	int ret;
918 
919 	if (!(rw & REQ_WRITE)) {
920 		/*
921 		 * called for a read, do the setup so that checksum validation
922 		 * can happen in the async kernel threads
923 		 */
924 		ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
925 					  bio, 1);
926 		if (ret)
927 			goto out_w_error;
928 		ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
929 				    mirror_num, 0);
930 	} else if (!async) {
931 		ret = btree_csum_one_bio(bio);
932 		if (ret)
933 			goto out_w_error;
934 		ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
935 				    mirror_num, 0);
936 	} else {
937 		/*
938 		 * kthread helpers are used to submit writes so that
939 		 * checksumming can happen in parallel across all CPUs
940 		 */
941 		ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
942 					  inode, rw, bio, mirror_num, 0,
943 					  bio_offset,
944 					  __btree_submit_bio_start,
945 					  __btree_submit_bio_done);
946 	}
947 
948 	if (ret) {
949 out_w_error:
950 		bio_endio(bio, ret);
951 	}
952 	return ret;
953 }
954 
955 #ifdef CONFIG_MIGRATION
956 static int btree_migratepage(struct address_space *mapping,
957 			struct page *newpage, struct page *page,
958 			enum migrate_mode mode)
959 {
960 	/*
961 	 * we can't safely write a btree page from here,
962 	 * we haven't done the locking hook
963 	 */
964 	if (PageDirty(page))
965 		return -EAGAIN;
966 	/*
967 	 * Buffers may be managed in a filesystem specific way.
968 	 * We must have no buffers or drop them.
969 	 */
970 	if (page_has_private(page) &&
971 	    !try_to_release_page(page, GFP_KERNEL))
972 		return -EAGAIN;
973 	return migrate_page(mapping, newpage, page, mode);
974 }
975 #endif
976 
977 
978 static int btree_writepages(struct address_space *mapping,
979 			    struct writeback_control *wbc)
980 {
981 	struct extent_io_tree *tree;
982 	struct btrfs_fs_info *fs_info;
983 	int ret;
984 
985 	tree = &BTRFS_I(mapping->host)->io_tree;
986 	if (wbc->sync_mode == WB_SYNC_NONE) {
987 
988 		if (wbc->for_kupdate)
989 			return 0;
990 
991 		fs_info = BTRFS_I(mapping->host)->root->fs_info;
992 		/* this is a bit racy, but that's ok */
993 		ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
994 					     BTRFS_DIRTY_METADATA_THRESH);
995 		if (ret < 0)
996 			return 0;
997 	}
998 	return btree_write_cache_pages(mapping, wbc);
999 }
1000 
1001 static int btree_readpage(struct file *file, struct page *page)
1002 {
1003 	struct extent_io_tree *tree;
1004 	tree = &BTRFS_I(page->mapping->host)->io_tree;
1005 	return extent_read_full_page(tree, page, btree_get_extent, 0);
1006 }
1007 
1008 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1009 {
1010 	if (PageWriteback(page) || PageDirty(page))
1011 		return 0;
1012 
1013 	return try_release_extent_buffer(page);
1014 }
1015 
1016 static void btree_invalidatepage(struct page *page, unsigned long offset)
1017 {
1018 	struct extent_io_tree *tree;
1019 	tree = &BTRFS_I(page->mapping->host)->io_tree;
1020 	extent_invalidatepage(tree, page, offset);
1021 	btree_releasepage(page, GFP_NOFS);
1022 	if (PagePrivate(page)) {
1023 		printk(KERN_WARNING "btrfs warning page private not zero "
1024 		       "on page %llu\n", (unsigned long long)page_offset(page));
1025 		ClearPagePrivate(page);
1026 		set_page_private(page, 0);
1027 		page_cache_release(page);
1028 	}
1029 }
1030 
1031 static int btree_set_page_dirty(struct page *page)
1032 {
1033 #ifdef DEBUG
1034 	struct extent_buffer *eb;
1035 
1036 	BUG_ON(!PagePrivate(page));
1037 	eb = (struct extent_buffer *)page->private;
1038 	BUG_ON(!eb);
1039 	BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1040 	BUG_ON(!atomic_read(&eb->refs));
1041 	btrfs_assert_tree_locked(eb);
1042 #endif
1043 	return __set_page_dirty_nobuffers(page);
1044 }
1045 
1046 static const struct address_space_operations btree_aops = {
1047 	.readpage	= btree_readpage,
1048 	.writepages	= btree_writepages,
1049 	.releasepage	= btree_releasepage,
1050 	.invalidatepage = btree_invalidatepage,
1051 #ifdef CONFIG_MIGRATION
1052 	.migratepage	= btree_migratepage,
1053 #endif
1054 	.set_page_dirty = btree_set_page_dirty,
1055 };
1056 
1057 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1058 			 u64 parent_transid)
1059 {
1060 	struct extent_buffer *buf = NULL;
1061 	struct inode *btree_inode = root->fs_info->btree_inode;
1062 	int ret = 0;
1063 
1064 	buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1065 	if (!buf)
1066 		return 0;
1067 	read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1068 				 buf, 0, WAIT_NONE, btree_get_extent, 0);
1069 	free_extent_buffer(buf);
1070 	return ret;
1071 }
1072 
1073 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1074 			 int mirror_num, struct extent_buffer **eb)
1075 {
1076 	struct extent_buffer *buf = NULL;
1077 	struct inode *btree_inode = root->fs_info->btree_inode;
1078 	struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1079 	int ret;
1080 
1081 	buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1082 	if (!buf)
1083 		return 0;
1084 
1085 	set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1086 
1087 	ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1088 				       btree_get_extent, mirror_num);
1089 	if (ret) {
1090 		free_extent_buffer(buf);
1091 		return ret;
1092 	}
1093 
1094 	if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1095 		free_extent_buffer(buf);
1096 		return -EIO;
1097 	} else if (extent_buffer_uptodate(buf)) {
1098 		*eb = buf;
1099 	} else {
1100 		free_extent_buffer(buf);
1101 	}
1102 	return 0;
1103 }
1104 
1105 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1106 					    u64 bytenr, u32 blocksize)
1107 {
1108 	struct inode *btree_inode = root->fs_info->btree_inode;
1109 	struct extent_buffer *eb;
1110 	eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1111 				bytenr, blocksize);
1112 	return eb;
1113 }
1114 
1115 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1116 						 u64 bytenr, u32 blocksize)
1117 {
1118 	struct inode *btree_inode = root->fs_info->btree_inode;
1119 	struct extent_buffer *eb;
1120 
1121 	eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1122 				 bytenr, blocksize);
1123 	return eb;
1124 }
1125 
1126 
1127 int btrfs_write_tree_block(struct extent_buffer *buf)
1128 {
1129 	return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1130 					buf->start + buf->len - 1);
1131 }
1132 
1133 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1134 {
1135 	return filemap_fdatawait_range(buf->pages[0]->mapping,
1136 				       buf->start, buf->start + buf->len - 1);
1137 }
1138 
1139 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1140 				      u32 blocksize, u64 parent_transid)
1141 {
1142 	struct extent_buffer *buf = NULL;
1143 	int ret;
1144 
1145 	buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1146 	if (!buf)
1147 		return NULL;
1148 
1149 	ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1150 	return buf;
1151 
1152 }
1153 
1154 void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1155 		      struct extent_buffer *buf)
1156 {
1157 	struct btrfs_fs_info *fs_info = root->fs_info;
1158 
1159 	if (btrfs_header_generation(buf) ==
1160 	    fs_info->running_transaction->transid) {
1161 		btrfs_assert_tree_locked(buf);
1162 
1163 		if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1164 			__percpu_counter_add(&fs_info->dirty_metadata_bytes,
1165 					     -buf->len,
1166 					     fs_info->dirty_metadata_batch);
1167 			/* ugh, clear_extent_buffer_dirty needs to lock the page */
1168 			btrfs_set_lock_blocking(buf);
1169 			clear_extent_buffer_dirty(buf);
1170 		}
1171 	}
1172 }
1173 
1174 static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1175 			 u32 stripesize, struct btrfs_root *root,
1176 			 struct btrfs_fs_info *fs_info,
1177 			 u64 objectid)
1178 {
1179 	root->node = NULL;
1180 	root->commit_root = NULL;
1181 	root->sectorsize = sectorsize;
1182 	root->nodesize = nodesize;
1183 	root->leafsize = leafsize;
1184 	root->stripesize = stripesize;
1185 	root->ref_cows = 0;
1186 	root->track_dirty = 0;
1187 	root->in_radix = 0;
1188 	root->orphan_item_inserted = 0;
1189 	root->orphan_cleanup_state = 0;
1190 
1191 	root->objectid = objectid;
1192 	root->last_trans = 0;
1193 	root->highest_objectid = 0;
1194 	root->name = NULL;
1195 	root->inode_tree = RB_ROOT;
1196 	INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1197 	root->block_rsv = NULL;
1198 	root->orphan_block_rsv = NULL;
1199 
1200 	INIT_LIST_HEAD(&root->dirty_list);
1201 	INIT_LIST_HEAD(&root->root_list);
1202 	INIT_LIST_HEAD(&root->logged_list[0]);
1203 	INIT_LIST_HEAD(&root->logged_list[1]);
1204 	spin_lock_init(&root->orphan_lock);
1205 	spin_lock_init(&root->inode_lock);
1206 	spin_lock_init(&root->accounting_lock);
1207 	spin_lock_init(&root->log_extents_lock[0]);
1208 	spin_lock_init(&root->log_extents_lock[1]);
1209 	mutex_init(&root->objectid_mutex);
1210 	mutex_init(&root->log_mutex);
1211 	init_waitqueue_head(&root->log_writer_wait);
1212 	init_waitqueue_head(&root->log_commit_wait[0]);
1213 	init_waitqueue_head(&root->log_commit_wait[1]);
1214 	atomic_set(&root->log_commit[0], 0);
1215 	atomic_set(&root->log_commit[1], 0);
1216 	atomic_set(&root->log_writers, 0);
1217 	atomic_set(&root->log_batch, 0);
1218 	atomic_set(&root->orphan_inodes, 0);
1219 	root->log_transid = 0;
1220 	root->last_log_commit = 0;
1221 	extent_io_tree_init(&root->dirty_log_pages,
1222 			     fs_info->btree_inode->i_mapping);
1223 
1224 	memset(&root->root_key, 0, sizeof(root->root_key));
1225 	memset(&root->root_item, 0, sizeof(root->root_item));
1226 	memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1227 	memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1228 	root->defrag_trans_start = fs_info->generation;
1229 	init_completion(&root->kobj_unregister);
1230 	root->defrag_running = 0;
1231 	root->root_key.objectid = objectid;
1232 	root->anon_dev = 0;
1233 
1234 	spin_lock_init(&root->root_item_lock);
1235 }
1236 
1237 static int __must_check find_and_setup_root(struct btrfs_root *tree_root,
1238 					    struct btrfs_fs_info *fs_info,
1239 					    u64 objectid,
1240 					    struct btrfs_root *root)
1241 {
1242 	int ret;
1243 	u32 blocksize;
1244 	u64 generation;
1245 
1246 	__setup_root(tree_root->nodesize, tree_root->leafsize,
1247 		     tree_root->sectorsize, tree_root->stripesize,
1248 		     root, fs_info, objectid);
1249 	ret = btrfs_find_last_root(tree_root, objectid,
1250 				   &root->root_item, &root->root_key);
1251 	if (ret > 0)
1252 		return -ENOENT;
1253 	else if (ret < 0)
1254 		return ret;
1255 
1256 	generation = btrfs_root_generation(&root->root_item);
1257 	blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1258 	root->commit_root = NULL;
1259 	root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1260 				     blocksize, generation);
1261 	if (!root->node || !btrfs_buffer_uptodate(root->node, generation, 0)) {
1262 		free_extent_buffer(root->node);
1263 		root->node = NULL;
1264 		return -EIO;
1265 	}
1266 	root->commit_root = btrfs_root_node(root);
1267 	return 0;
1268 }
1269 
1270 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1271 {
1272 	struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1273 	if (root)
1274 		root->fs_info = fs_info;
1275 	return root;
1276 }
1277 
1278 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1279 				     struct btrfs_fs_info *fs_info,
1280 				     u64 objectid)
1281 {
1282 	struct extent_buffer *leaf;
1283 	struct btrfs_root *tree_root = fs_info->tree_root;
1284 	struct btrfs_root *root;
1285 	struct btrfs_key key;
1286 	int ret = 0;
1287 	u64 bytenr;
1288 	uuid_le uuid;
1289 
1290 	root = btrfs_alloc_root(fs_info);
1291 	if (!root)
1292 		return ERR_PTR(-ENOMEM);
1293 
1294 	__setup_root(tree_root->nodesize, tree_root->leafsize,
1295 		     tree_root->sectorsize, tree_root->stripesize,
1296 		     root, fs_info, objectid);
1297 	root->root_key.objectid = objectid;
1298 	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1299 	root->root_key.offset = 0;
1300 
1301 	leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1302 				      0, objectid, NULL, 0, 0, 0);
1303 	if (IS_ERR(leaf)) {
1304 		ret = PTR_ERR(leaf);
1305 		leaf = NULL;
1306 		goto fail;
1307 	}
1308 
1309 	bytenr = leaf->start;
1310 	memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1311 	btrfs_set_header_bytenr(leaf, leaf->start);
1312 	btrfs_set_header_generation(leaf, trans->transid);
1313 	btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1314 	btrfs_set_header_owner(leaf, objectid);
1315 	root->node = leaf;
1316 
1317 	write_extent_buffer(leaf, fs_info->fsid,
1318 			    (unsigned long)btrfs_header_fsid(leaf),
1319 			    BTRFS_FSID_SIZE);
1320 	write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1321 			    (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
1322 			    BTRFS_UUID_SIZE);
1323 	btrfs_mark_buffer_dirty(leaf);
1324 
1325 	root->commit_root = btrfs_root_node(root);
1326 	root->track_dirty = 1;
1327 
1328 
1329 	root->root_item.flags = 0;
1330 	root->root_item.byte_limit = 0;
1331 	btrfs_set_root_bytenr(&root->root_item, leaf->start);
1332 	btrfs_set_root_generation(&root->root_item, trans->transid);
1333 	btrfs_set_root_level(&root->root_item, 0);
1334 	btrfs_set_root_refs(&root->root_item, 1);
1335 	btrfs_set_root_used(&root->root_item, leaf->len);
1336 	btrfs_set_root_last_snapshot(&root->root_item, 0);
1337 	btrfs_set_root_dirid(&root->root_item, 0);
1338 	uuid_le_gen(&uuid);
1339 	memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1340 	root->root_item.drop_level = 0;
1341 
1342 	key.objectid = objectid;
1343 	key.type = BTRFS_ROOT_ITEM_KEY;
1344 	key.offset = 0;
1345 	ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1346 	if (ret)
1347 		goto fail;
1348 
1349 	btrfs_tree_unlock(leaf);
1350 
1351 	return root;
1352 
1353 fail:
1354 	if (leaf) {
1355 		btrfs_tree_unlock(leaf);
1356 		free_extent_buffer(leaf);
1357 	}
1358 	kfree(root);
1359 
1360 	return ERR_PTR(ret);
1361 }
1362 
1363 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1364 					 struct btrfs_fs_info *fs_info)
1365 {
1366 	struct btrfs_root *root;
1367 	struct btrfs_root *tree_root = fs_info->tree_root;
1368 	struct extent_buffer *leaf;
1369 
1370 	root = btrfs_alloc_root(fs_info);
1371 	if (!root)
1372 		return ERR_PTR(-ENOMEM);
1373 
1374 	__setup_root(tree_root->nodesize, tree_root->leafsize,
1375 		     tree_root->sectorsize, tree_root->stripesize,
1376 		     root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1377 
1378 	root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1379 	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1380 	root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1381 	/*
1382 	 * log trees do not get reference counted because they go away
1383 	 * before a real commit is actually done.  They do store pointers
1384 	 * to file data extents, and those reference counts still get
1385 	 * updated (along with back refs to the log tree).
1386 	 */
1387 	root->ref_cows = 0;
1388 
1389 	leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1390 				      BTRFS_TREE_LOG_OBJECTID, NULL,
1391 				      0, 0, 0);
1392 	if (IS_ERR(leaf)) {
1393 		kfree(root);
1394 		return ERR_CAST(leaf);
1395 	}
1396 
1397 	memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1398 	btrfs_set_header_bytenr(leaf, leaf->start);
1399 	btrfs_set_header_generation(leaf, trans->transid);
1400 	btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1401 	btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1402 	root->node = leaf;
1403 
1404 	write_extent_buffer(root->node, root->fs_info->fsid,
1405 			    (unsigned long)btrfs_header_fsid(root->node),
1406 			    BTRFS_FSID_SIZE);
1407 	btrfs_mark_buffer_dirty(root->node);
1408 	btrfs_tree_unlock(root->node);
1409 	return root;
1410 }
1411 
1412 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1413 			     struct btrfs_fs_info *fs_info)
1414 {
1415 	struct btrfs_root *log_root;
1416 
1417 	log_root = alloc_log_tree(trans, fs_info);
1418 	if (IS_ERR(log_root))
1419 		return PTR_ERR(log_root);
1420 	WARN_ON(fs_info->log_root_tree);
1421 	fs_info->log_root_tree = log_root;
1422 	return 0;
1423 }
1424 
1425 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1426 		       struct btrfs_root *root)
1427 {
1428 	struct btrfs_root *log_root;
1429 	struct btrfs_inode_item *inode_item;
1430 
1431 	log_root = alloc_log_tree(trans, root->fs_info);
1432 	if (IS_ERR(log_root))
1433 		return PTR_ERR(log_root);
1434 
1435 	log_root->last_trans = trans->transid;
1436 	log_root->root_key.offset = root->root_key.objectid;
1437 
1438 	inode_item = &log_root->root_item.inode;
1439 	inode_item->generation = cpu_to_le64(1);
1440 	inode_item->size = cpu_to_le64(3);
1441 	inode_item->nlink = cpu_to_le32(1);
1442 	inode_item->nbytes = cpu_to_le64(root->leafsize);
1443 	inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1444 
1445 	btrfs_set_root_node(&log_root->root_item, log_root->node);
1446 
1447 	WARN_ON(root->log_root);
1448 	root->log_root = log_root;
1449 	root->log_transid = 0;
1450 	root->last_log_commit = 0;
1451 	return 0;
1452 }
1453 
1454 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1455 					       struct btrfs_key *location)
1456 {
1457 	struct btrfs_root *root;
1458 	struct btrfs_fs_info *fs_info = tree_root->fs_info;
1459 	struct btrfs_path *path;
1460 	struct extent_buffer *l;
1461 	u64 generation;
1462 	u32 blocksize;
1463 	int ret = 0;
1464 	int slot;
1465 
1466 	root = btrfs_alloc_root(fs_info);
1467 	if (!root)
1468 		return ERR_PTR(-ENOMEM);
1469 	if (location->offset == (u64)-1) {
1470 		ret = find_and_setup_root(tree_root, fs_info,
1471 					  location->objectid, root);
1472 		if (ret) {
1473 			kfree(root);
1474 			return ERR_PTR(ret);
1475 		}
1476 		goto out;
1477 	}
1478 
1479 	__setup_root(tree_root->nodesize, tree_root->leafsize,
1480 		     tree_root->sectorsize, tree_root->stripesize,
1481 		     root, fs_info, location->objectid);
1482 
1483 	path = btrfs_alloc_path();
1484 	if (!path) {
1485 		kfree(root);
1486 		return ERR_PTR(-ENOMEM);
1487 	}
1488 	ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1489 	if (ret == 0) {
1490 		l = path->nodes[0];
1491 		slot = path->slots[0];
1492 		btrfs_read_root_item(l, slot, &root->root_item);
1493 		memcpy(&root->root_key, location, sizeof(*location));
1494 	}
1495 	btrfs_free_path(path);
1496 	if (ret) {
1497 		kfree(root);
1498 		if (ret > 0)
1499 			ret = -ENOENT;
1500 		return ERR_PTR(ret);
1501 	}
1502 
1503 	generation = btrfs_root_generation(&root->root_item);
1504 	blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1505 	root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1506 				     blocksize, generation);
1507 	if (!root->node || !extent_buffer_uptodate(root->node)) {
1508 		ret = (!root->node) ? -ENOMEM : -EIO;
1509 
1510 		free_extent_buffer(root->node);
1511 		kfree(root);
1512 		return ERR_PTR(ret);
1513 	}
1514 
1515 	root->commit_root = btrfs_root_node(root);
1516 	BUG_ON(!root->node); /* -ENOMEM */
1517 out:
1518 	if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1519 		root->ref_cows = 1;
1520 		btrfs_check_and_init_root_item(&root->root_item);
1521 	}
1522 
1523 	return root;
1524 }
1525 
1526 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1527 					      struct btrfs_key *location)
1528 {
1529 	struct btrfs_root *root;
1530 	int ret;
1531 
1532 	if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1533 		return fs_info->tree_root;
1534 	if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1535 		return fs_info->extent_root;
1536 	if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1537 		return fs_info->chunk_root;
1538 	if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1539 		return fs_info->dev_root;
1540 	if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1541 		return fs_info->csum_root;
1542 	if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1543 		return fs_info->quota_root ? fs_info->quota_root :
1544 					     ERR_PTR(-ENOENT);
1545 again:
1546 	spin_lock(&fs_info->fs_roots_radix_lock);
1547 	root = radix_tree_lookup(&fs_info->fs_roots_radix,
1548 				 (unsigned long)location->objectid);
1549 	spin_unlock(&fs_info->fs_roots_radix_lock);
1550 	if (root)
1551 		return root;
1552 
1553 	root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1554 	if (IS_ERR(root))
1555 		return root;
1556 
1557 	root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1558 	root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1559 					GFP_NOFS);
1560 	if (!root->free_ino_pinned || !root->free_ino_ctl) {
1561 		ret = -ENOMEM;
1562 		goto fail;
1563 	}
1564 
1565 	btrfs_init_free_ino_ctl(root);
1566 	mutex_init(&root->fs_commit_mutex);
1567 	spin_lock_init(&root->cache_lock);
1568 	init_waitqueue_head(&root->cache_wait);
1569 
1570 	ret = get_anon_bdev(&root->anon_dev);
1571 	if (ret)
1572 		goto fail;
1573 
1574 	if (btrfs_root_refs(&root->root_item) == 0) {
1575 		ret = -ENOENT;
1576 		goto fail;
1577 	}
1578 
1579 	ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1580 	if (ret < 0)
1581 		goto fail;
1582 	if (ret == 0)
1583 		root->orphan_item_inserted = 1;
1584 
1585 	ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1586 	if (ret)
1587 		goto fail;
1588 
1589 	spin_lock(&fs_info->fs_roots_radix_lock);
1590 	ret = radix_tree_insert(&fs_info->fs_roots_radix,
1591 				(unsigned long)root->root_key.objectid,
1592 				root);
1593 	if (ret == 0)
1594 		root->in_radix = 1;
1595 
1596 	spin_unlock(&fs_info->fs_roots_radix_lock);
1597 	radix_tree_preload_end();
1598 	if (ret) {
1599 		if (ret == -EEXIST) {
1600 			free_fs_root(root);
1601 			goto again;
1602 		}
1603 		goto fail;
1604 	}
1605 
1606 	ret = btrfs_find_dead_roots(fs_info->tree_root,
1607 				    root->root_key.objectid);
1608 	WARN_ON(ret);
1609 	return root;
1610 fail:
1611 	free_fs_root(root);
1612 	return ERR_PTR(ret);
1613 }
1614 
1615 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1616 {
1617 	struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1618 	int ret = 0;
1619 	struct btrfs_device *device;
1620 	struct backing_dev_info *bdi;
1621 
1622 	rcu_read_lock();
1623 	list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1624 		if (!device->bdev)
1625 			continue;
1626 		bdi = blk_get_backing_dev_info(device->bdev);
1627 		if (bdi && bdi_congested(bdi, bdi_bits)) {
1628 			ret = 1;
1629 			break;
1630 		}
1631 	}
1632 	rcu_read_unlock();
1633 	return ret;
1634 }
1635 
1636 /*
1637  * If this fails, caller must call bdi_destroy() to get rid of the
1638  * bdi again.
1639  */
1640 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1641 {
1642 	int err;
1643 
1644 	bdi->capabilities = BDI_CAP_MAP_COPY;
1645 	err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1646 	if (err)
1647 		return err;
1648 
1649 	bdi->ra_pages	= default_backing_dev_info.ra_pages;
1650 	bdi->congested_fn	= btrfs_congested_fn;
1651 	bdi->congested_data	= info;
1652 	return 0;
1653 }
1654 
1655 /*
1656  * called by the kthread helper functions to finally call the bio end_io
1657  * functions.  This is where read checksum verification actually happens
1658  */
1659 static void end_workqueue_fn(struct btrfs_work *work)
1660 {
1661 	struct bio *bio;
1662 	struct end_io_wq *end_io_wq;
1663 	struct btrfs_fs_info *fs_info;
1664 	int error;
1665 
1666 	end_io_wq = container_of(work, struct end_io_wq, work);
1667 	bio = end_io_wq->bio;
1668 	fs_info = end_io_wq->info;
1669 
1670 	error = end_io_wq->error;
1671 	bio->bi_private = end_io_wq->private;
1672 	bio->bi_end_io = end_io_wq->end_io;
1673 	kfree(end_io_wq);
1674 	bio_endio(bio, error);
1675 }
1676 
1677 static int cleaner_kthread(void *arg)
1678 {
1679 	struct btrfs_root *root = arg;
1680 
1681 	do {
1682 		int again = 0;
1683 
1684 		if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1685 		    down_read_trylock(&root->fs_info->sb->s_umount)) {
1686 			if (mutex_trylock(&root->fs_info->cleaner_mutex)) {
1687 				btrfs_run_delayed_iputs(root);
1688 				again = btrfs_clean_one_deleted_snapshot(root);
1689 				mutex_unlock(&root->fs_info->cleaner_mutex);
1690 			}
1691 			btrfs_run_defrag_inodes(root->fs_info);
1692 			up_read(&root->fs_info->sb->s_umount);
1693 		}
1694 
1695 		if (!try_to_freeze() && !again) {
1696 			set_current_state(TASK_INTERRUPTIBLE);
1697 			if (!kthread_should_stop())
1698 				schedule();
1699 			__set_current_state(TASK_RUNNING);
1700 		}
1701 	} while (!kthread_should_stop());
1702 	return 0;
1703 }
1704 
1705 static int transaction_kthread(void *arg)
1706 {
1707 	struct btrfs_root *root = arg;
1708 	struct btrfs_trans_handle *trans;
1709 	struct btrfs_transaction *cur;
1710 	u64 transid;
1711 	unsigned long now;
1712 	unsigned long delay;
1713 	bool cannot_commit;
1714 
1715 	do {
1716 		cannot_commit = false;
1717 		delay = HZ * 30;
1718 		mutex_lock(&root->fs_info->transaction_kthread_mutex);
1719 
1720 		spin_lock(&root->fs_info->trans_lock);
1721 		cur = root->fs_info->running_transaction;
1722 		if (!cur) {
1723 			spin_unlock(&root->fs_info->trans_lock);
1724 			goto sleep;
1725 		}
1726 
1727 		now = get_seconds();
1728 		if (!cur->blocked &&
1729 		    (now < cur->start_time || now - cur->start_time < 30)) {
1730 			spin_unlock(&root->fs_info->trans_lock);
1731 			delay = HZ * 5;
1732 			goto sleep;
1733 		}
1734 		transid = cur->transid;
1735 		spin_unlock(&root->fs_info->trans_lock);
1736 
1737 		/* If the file system is aborted, this will always fail. */
1738 		trans = btrfs_attach_transaction(root);
1739 		if (IS_ERR(trans)) {
1740 			if (PTR_ERR(trans) != -ENOENT)
1741 				cannot_commit = true;
1742 			goto sleep;
1743 		}
1744 		if (transid == trans->transid) {
1745 			btrfs_commit_transaction(trans, root);
1746 		} else {
1747 			btrfs_end_transaction(trans, root);
1748 		}
1749 sleep:
1750 		wake_up_process(root->fs_info->cleaner_kthread);
1751 		mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1752 
1753 		if (!try_to_freeze()) {
1754 			set_current_state(TASK_INTERRUPTIBLE);
1755 			if (!kthread_should_stop() &&
1756 			    (!btrfs_transaction_blocked(root->fs_info) ||
1757 			     cannot_commit))
1758 				schedule_timeout(delay);
1759 			__set_current_state(TASK_RUNNING);
1760 		}
1761 	} while (!kthread_should_stop());
1762 	return 0;
1763 }
1764 
1765 /*
1766  * this will find the highest generation in the array of
1767  * root backups.  The index of the highest array is returned,
1768  * or -1 if we can't find anything.
1769  *
1770  * We check to make sure the array is valid by comparing the
1771  * generation of the latest  root in the array with the generation
1772  * in the super block.  If they don't match we pitch it.
1773  */
1774 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1775 {
1776 	u64 cur;
1777 	int newest_index = -1;
1778 	struct btrfs_root_backup *root_backup;
1779 	int i;
1780 
1781 	for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1782 		root_backup = info->super_copy->super_roots + i;
1783 		cur = btrfs_backup_tree_root_gen(root_backup);
1784 		if (cur == newest_gen)
1785 			newest_index = i;
1786 	}
1787 
1788 	/* check to see if we actually wrapped around */
1789 	if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1790 		root_backup = info->super_copy->super_roots;
1791 		cur = btrfs_backup_tree_root_gen(root_backup);
1792 		if (cur == newest_gen)
1793 			newest_index = 0;
1794 	}
1795 	return newest_index;
1796 }
1797 
1798 
1799 /*
1800  * find the oldest backup so we know where to store new entries
1801  * in the backup array.  This will set the backup_root_index
1802  * field in the fs_info struct
1803  */
1804 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1805 				     u64 newest_gen)
1806 {
1807 	int newest_index = -1;
1808 
1809 	newest_index = find_newest_super_backup(info, newest_gen);
1810 	/* if there was garbage in there, just move along */
1811 	if (newest_index == -1) {
1812 		info->backup_root_index = 0;
1813 	} else {
1814 		info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1815 	}
1816 }
1817 
1818 /*
1819  * copy all the root pointers into the super backup array.
1820  * this will bump the backup pointer by one when it is
1821  * done
1822  */
1823 static void backup_super_roots(struct btrfs_fs_info *info)
1824 {
1825 	int next_backup;
1826 	struct btrfs_root_backup *root_backup;
1827 	int last_backup;
1828 
1829 	next_backup = info->backup_root_index;
1830 	last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1831 		BTRFS_NUM_BACKUP_ROOTS;
1832 
1833 	/*
1834 	 * just overwrite the last backup if we're at the same generation
1835 	 * this happens only at umount
1836 	 */
1837 	root_backup = info->super_for_commit->super_roots + last_backup;
1838 	if (btrfs_backup_tree_root_gen(root_backup) ==
1839 	    btrfs_header_generation(info->tree_root->node))
1840 		next_backup = last_backup;
1841 
1842 	root_backup = info->super_for_commit->super_roots + next_backup;
1843 
1844 	/*
1845 	 * make sure all of our padding and empty slots get zero filled
1846 	 * regardless of which ones we use today
1847 	 */
1848 	memset(root_backup, 0, sizeof(*root_backup));
1849 
1850 	info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1851 
1852 	btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1853 	btrfs_set_backup_tree_root_gen(root_backup,
1854 			       btrfs_header_generation(info->tree_root->node));
1855 
1856 	btrfs_set_backup_tree_root_level(root_backup,
1857 			       btrfs_header_level(info->tree_root->node));
1858 
1859 	btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1860 	btrfs_set_backup_chunk_root_gen(root_backup,
1861 			       btrfs_header_generation(info->chunk_root->node));
1862 	btrfs_set_backup_chunk_root_level(root_backup,
1863 			       btrfs_header_level(info->chunk_root->node));
1864 
1865 	btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1866 	btrfs_set_backup_extent_root_gen(root_backup,
1867 			       btrfs_header_generation(info->extent_root->node));
1868 	btrfs_set_backup_extent_root_level(root_backup,
1869 			       btrfs_header_level(info->extent_root->node));
1870 
1871 	/*
1872 	 * we might commit during log recovery, which happens before we set
1873 	 * the fs_root.  Make sure it is valid before we fill it in.
1874 	 */
1875 	if (info->fs_root && info->fs_root->node) {
1876 		btrfs_set_backup_fs_root(root_backup,
1877 					 info->fs_root->node->start);
1878 		btrfs_set_backup_fs_root_gen(root_backup,
1879 			       btrfs_header_generation(info->fs_root->node));
1880 		btrfs_set_backup_fs_root_level(root_backup,
1881 			       btrfs_header_level(info->fs_root->node));
1882 	}
1883 
1884 	btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1885 	btrfs_set_backup_dev_root_gen(root_backup,
1886 			       btrfs_header_generation(info->dev_root->node));
1887 	btrfs_set_backup_dev_root_level(root_backup,
1888 				       btrfs_header_level(info->dev_root->node));
1889 
1890 	btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1891 	btrfs_set_backup_csum_root_gen(root_backup,
1892 			       btrfs_header_generation(info->csum_root->node));
1893 	btrfs_set_backup_csum_root_level(root_backup,
1894 			       btrfs_header_level(info->csum_root->node));
1895 
1896 	btrfs_set_backup_total_bytes(root_backup,
1897 			     btrfs_super_total_bytes(info->super_copy));
1898 	btrfs_set_backup_bytes_used(root_backup,
1899 			     btrfs_super_bytes_used(info->super_copy));
1900 	btrfs_set_backup_num_devices(root_backup,
1901 			     btrfs_super_num_devices(info->super_copy));
1902 
1903 	/*
1904 	 * if we don't copy this out to the super_copy, it won't get remembered
1905 	 * for the next commit
1906 	 */
1907 	memcpy(&info->super_copy->super_roots,
1908 	       &info->super_for_commit->super_roots,
1909 	       sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1910 }
1911 
1912 /*
1913  * this copies info out of the root backup array and back into
1914  * the in-memory super block.  It is meant to help iterate through
1915  * the array, so you send it the number of backups you've already
1916  * tried and the last backup index you used.
1917  *
1918  * this returns -1 when it has tried all the backups
1919  */
1920 static noinline int next_root_backup(struct btrfs_fs_info *info,
1921 				     struct btrfs_super_block *super,
1922 				     int *num_backups_tried, int *backup_index)
1923 {
1924 	struct btrfs_root_backup *root_backup;
1925 	int newest = *backup_index;
1926 
1927 	if (*num_backups_tried == 0) {
1928 		u64 gen = btrfs_super_generation(super);
1929 
1930 		newest = find_newest_super_backup(info, gen);
1931 		if (newest == -1)
1932 			return -1;
1933 
1934 		*backup_index = newest;
1935 		*num_backups_tried = 1;
1936 	} else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1937 		/* we've tried all the backups, all done */
1938 		return -1;
1939 	} else {
1940 		/* jump to the next oldest backup */
1941 		newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1942 			BTRFS_NUM_BACKUP_ROOTS;
1943 		*backup_index = newest;
1944 		*num_backups_tried += 1;
1945 	}
1946 	root_backup = super->super_roots + newest;
1947 
1948 	btrfs_set_super_generation(super,
1949 				   btrfs_backup_tree_root_gen(root_backup));
1950 	btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1951 	btrfs_set_super_root_level(super,
1952 				   btrfs_backup_tree_root_level(root_backup));
1953 	btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1954 
1955 	/*
1956 	 * fixme: the total bytes and num_devices need to match or we should
1957 	 * need a fsck
1958 	 */
1959 	btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1960 	btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1961 	return 0;
1962 }
1963 
1964 /* helper to cleanup workers */
1965 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1966 {
1967 	btrfs_stop_workers(&fs_info->generic_worker);
1968 	btrfs_stop_workers(&fs_info->fixup_workers);
1969 	btrfs_stop_workers(&fs_info->delalloc_workers);
1970 	btrfs_stop_workers(&fs_info->workers);
1971 	btrfs_stop_workers(&fs_info->endio_workers);
1972 	btrfs_stop_workers(&fs_info->endio_meta_workers);
1973 	btrfs_stop_workers(&fs_info->endio_raid56_workers);
1974 	btrfs_stop_workers(&fs_info->rmw_workers);
1975 	btrfs_stop_workers(&fs_info->endio_meta_write_workers);
1976 	btrfs_stop_workers(&fs_info->endio_write_workers);
1977 	btrfs_stop_workers(&fs_info->endio_freespace_worker);
1978 	btrfs_stop_workers(&fs_info->submit_workers);
1979 	btrfs_stop_workers(&fs_info->delayed_workers);
1980 	btrfs_stop_workers(&fs_info->caching_workers);
1981 	btrfs_stop_workers(&fs_info->readahead_workers);
1982 	btrfs_stop_workers(&fs_info->flush_workers);
1983 	btrfs_stop_workers(&fs_info->qgroup_rescan_workers);
1984 }
1985 
1986 /* helper to cleanup tree roots */
1987 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
1988 {
1989 	free_extent_buffer(info->tree_root->node);
1990 	free_extent_buffer(info->tree_root->commit_root);
1991 	free_extent_buffer(info->dev_root->node);
1992 	free_extent_buffer(info->dev_root->commit_root);
1993 	free_extent_buffer(info->extent_root->node);
1994 	free_extent_buffer(info->extent_root->commit_root);
1995 	free_extent_buffer(info->csum_root->node);
1996 	free_extent_buffer(info->csum_root->commit_root);
1997 	if (info->quota_root) {
1998 		free_extent_buffer(info->quota_root->node);
1999 		free_extent_buffer(info->quota_root->commit_root);
2000 	}
2001 
2002 	info->tree_root->node = NULL;
2003 	info->tree_root->commit_root = NULL;
2004 	info->dev_root->node = NULL;
2005 	info->dev_root->commit_root = NULL;
2006 	info->extent_root->node = NULL;
2007 	info->extent_root->commit_root = NULL;
2008 	info->csum_root->node = NULL;
2009 	info->csum_root->commit_root = NULL;
2010 	if (info->quota_root) {
2011 		info->quota_root->node = NULL;
2012 		info->quota_root->commit_root = NULL;
2013 	}
2014 
2015 	if (chunk_root) {
2016 		free_extent_buffer(info->chunk_root->node);
2017 		free_extent_buffer(info->chunk_root->commit_root);
2018 		info->chunk_root->node = NULL;
2019 		info->chunk_root->commit_root = NULL;
2020 	}
2021 }
2022 
2023 static void del_fs_roots(struct btrfs_fs_info *fs_info)
2024 {
2025 	int ret;
2026 	struct btrfs_root *gang[8];
2027 	int i;
2028 
2029 	while (!list_empty(&fs_info->dead_roots)) {
2030 		gang[0] = list_entry(fs_info->dead_roots.next,
2031 				     struct btrfs_root, root_list);
2032 		list_del(&gang[0]->root_list);
2033 
2034 		if (gang[0]->in_radix) {
2035 			btrfs_free_fs_root(fs_info, gang[0]);
2036 		} else {
2037 			free_extent_buffer(gang[0]->node);
2038 			free_extent_buffer(gang[0]->commit_root);
2039 			kfree(gang[0]);
2040 		}
2041 	}
2042 
2043 	while (1) {
2044 		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2045 					     (void **)gang, 0,
2046 					     ARRAY_SIZE(gang));
2047 		if (!ret)
2048 			break;
2049 		for (i = 0; i < ret; i++)
2050 			btrfs_free_fs_root(fs_info, gang[i]);
2051 	}
2052 }
2053 
2054 int open_ctree(struct super_block *sb,
2055 	       struct btrfs_fs_devices *fs_devices,
2056 	       char *options)
2057 {
2058 	u32 sectorsize;
2059 	u32 nodesize;
2060 	u32 leafsize;
2061 	u32 blocksize;
2062 	u32 stripesize;
2063 	u64 generation;
2064 	u64 features;
2065 	struct btrfs_key location;
2066 	struct buffer_head *bh;
2067 	struct btrfs_super_block *disk_super;
2068 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2069 	struct btrfs_root *tree_root;
2070 	struct btrfs_root *extent_root;
2071 	struct btrfs_root *csum_root;
2072 	struct btrfs_root *chunk_root;
2073 	struct btrfs_root *dev_root;
2074 	struct btrfs_root *quota_root;
2075 	struct btrfs_root *log_tree_root;
2076 	int ret;
2077 	int err = -EINVAL;
2078 	int num_backups_tried = 0;
2079 	int backup_index = 0;
2080 
2081 	tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2082 	extent_root = fs_info->extent_root = btrfs_alloc_root(fs_info);
2083 	csum_root = fs_info->csum_root = btrfs_alloc_root(fs_info);
2084 	chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2085 	dev_root = fs_info->dev_root = btrfs_alloc_root(fs_info);
2086 	quota_root = fs_info->quota_root = btrfs_alloc_root(fs_info);
2087 
2088 	if (!tree_root || !extent_root || !csum_root ||
2089 	    !chunk_root || !dev_root || !quota_root) {
2090 		err = -ENOMEM;
2091 		goto fail;
2092 	}
2093 
2094 	ret = init_srcu_struct(&fs_info->subvol_srcu);
2095 	if (ret) {
2096 		err = ret;
2097 		goto fail;
2098 	}
2099 
2100 	ret = setup_bdi(fs_info, &fs_info->bdi);
2101 	if (ret) {
2102 		err = ret;
2103 		goto fail_srcu;
2104 	}
2105 
2106 	ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2107 	if (ret) {
2108 		err = ret;
2109 		goto fail_bdi;
2110 	}
2111 	fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2112 					(1 + ilog2(nr_cpu_ids));
2113 
2114 	ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2115 	if (ret) {
2116 		err = ret;
2117 		goto fail_dirty_metadata_bytes;
2118 	}
2119 
2120 	fs_info->btree_inode = new_inode(sb);
2121 	if (!fs_info->btree_inode) {
2122 		err = -ENOMEM;
2123 		goto fail_delalloc_bytes;
2124 	}
2125 
2126 	mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2127 
2128 	INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2129 	INIT_LIST_HEAD(&fs_info->trans_list);
2130 	INIT_LIST_HEAD(&fs_info->dead_roots);
2131 	INIT_LIST_HEAD(&fs_info->delayed_iputs);
2132 	INIT_LIST_HEAD(&fs_info->delalloc_inodes);
2133 	INIT_LIST_HEAD(&fs_info->caching_block_groups);
2134 	spin_lock_init(&fs_info->delalloc_lock);
2135 	spin_lock_init(&fs_info->trans_lock);
2136 	spin_lock_init(&fs_info->fs_roots_radix_lock);
2137 	spin_lock_init(&fs_info->delayed_iput_lock);
2138 	spin_lock_init(&fs_info->defrag_inodes_lock);
2139 	spin_lock_init(&fs_info->free_chunk_lock);
2140 	spin_lock_init(&fs_info->tree_mod_seq_lock);
2141 	spin_lock_init(&fs_info->super_lock);
2142 	rwlock_init(&fs_info->tree_mod_log_lock);
2143 	mutex_init(&fs_info->reloc_mutex);
2144 	seqlock_init(&fs_info->profiles_lock);
2145 
2146 	init_completion(&fs_info->kobj_unregister);
2147 	INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2148 	INIT_LIST_HEAD(&fs_info->space_info);
2149 	INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2150 	btrfs_mapping_init(&fs_info->mapping_tree);
2151 	btrfs_init_block_rsv(&fs_info->global_block_rsv,
2152 			     BTRFS_BLOCK_RSV_GLOBAL);
2153 	btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2154 			     BTRFS_BLOCK_RSV_DELALLOC);
2155 	btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2156 	btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2157 	btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2158 	btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2159 			     BTRFS_BLOCK_RSV_DELOPS);
2160 	atomic_set(&fs_info->nr_async_submits, 0);
2161 	atomic_set(&fs_info->async_delalloc_pages, 0);
2162 	atomic_set(&fs_info->async_submit_draining, 0);
2163 	atomic_set(&fs_info->nr_async_bios, 0);
2164 	atomic_set(&fs_info->defrag_running, 0);
2165 	atomic64_set(&fs_info->tree_mod_seq, 0);
2166 	fs_info->sb = sb;
2167 	fs_info->max_inline = 8192 * 1024;
2168 	fs_info->metadata_ratio = 0;
2169 	fs_info->defrag_inodes = RB_ROOT;
2170 	fs_info->trans_no_join = 0;
2171 	fs_info->free_chunk_space = 0;
2172 	fs_info->tree_mod_log = RB_ROOT;
2173 
2174 	/* readahead state */
2175 	INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2176 	spin_lock_init(&fs_info->reada_lock);
2177 
2178 	fs_info->thread_pool_size = min_t(unsigned long,
2179 					  num_online_cpus() + 2, 8);
2180 
2181 	INIT_LIST_HEAD(&fs_info->ordered_extents);
2182 	spin_lock_init(&fs_info->ordered_extent_lock);
2183 	fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2184 					GFP_NOFS);
2185 	if (!fs_info->delayed_root) {
2186 		err = -ENOMEM;
2187 		goto fail_iput;
2188 	}
2189 	btrfs_init_delayed_root(fs_info->delayed_root);
2190 
2191 	mutex_init(&fs_info->scrub_lock);
2192 	atomic_set(&fs_info->scrubs_running, 0);
2193 	atomic_set(&fs_info->scrub_pause_req, 0);
2194 	atomic_set(&fs_info->scrubs_paused, 0);
2195 	atomic_set(&fs_info->scrub_cancel_req, 0);
2196 	init_waitqueue_head(&fs_info->scrub_pause_wait);
2197 	init_rwsem(&fs_info->scrub_super_lock);
2198 	fs_info->scrub_workers_refcnt = 0;
2199 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2200 	fs_info->check_integrity_print_mask = 0;
2201 #endif
2202 
2203 	spin_lock_init(&fs_info->balance_lock);
2204 	mutex_init(&fs_info->balance_mutex);
2205 	atomic_set(&fs_info->balance_running, 0);
2206 	atomic_set(&fs_info->balance_pause_req, 0);
2207 	atomic_set(&fs_info->balance_cancel_req, 0);
2208 	fs_info->balance_ctl = NULL;
2209 	init_waitqueue_head(&fs_info->balance_wait_q);
2210 
2211 	sb->s_blocksize = 4096;
2212 	sb->s_blocksize_bits = blksize_bits(4096);
2213 	sb->s_bdi = &fs_info->bdi;
2214 
2215 	fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2216 	set_nlink(fs_info->btree_inode, 1);
2217 	/*
2218 	 * we set the i_size on the btree inode to the max possible int.
2219 	 * the real end of the address space is determined by all of
2220 	 * the devices in the system
2221 	 */
2222 	fs_info->btree_inode->i_size = OFFSET_MAX;
2223 	fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2224 	fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2225 
2226 	RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2227 	extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2228 			     fs_info->btree_inode->i_mapping);
2229 	BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2230 	extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2231 
2232 	BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2233 
2234 	BTRFS_I(fs_info->btree_inode)->root = tree_root;
2235 	memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2236 	       sizeof(struct btrfs_key));
2237 	set_bit(BTRFS_INODE_DUMMY,
2238 		&BTRFS_I(fs_info->btree_inode)->runtime_flags);
2239 	insert_inode_hash(fs_info->btree_inode);
2240 
2241 	spin_lock_init(&fs_info->block_group_cache_lock);
2242 	fs_info->block_group_cache_tree = RB_ROOT;
2243 	fs_info->first_logical_byte = (u64)-1;
2244 
2245 	extent_io_tree_init(&fs_info->freed_extents[0],
2246 			     fs_info->btree_inode->i_mapping);
2247 	extent_io_tree_init(&fs_info->freed_extents[1],
2248 			     fs_info->btree_inode->i_mapping);
2249 	fs_info->pinned_extents = &fs_info->freed_extents[0];
2250 	fs_info->do_barriers = 1;
2251 
2252 
2253 	mutex_init(&fs_info->ordered_operations_mutex);
2254 	mutex_init(&fs_info->tree_log_mutex);
2255 	mutex_init(&fs_info->chunk_mutex);
2256 	mutex_init(&fs_info->transaction_kthread_mutex);
2257 	mutex_init(&fs_info->cleaner_mutex);
2258 	mutex_init(&fs_info->volume_mutex);
2259 	init_rwsem(&fs_info->extent_commit_sem);
2260 	init_rwsem(&fs_info->cleanup_work_sem);
2261 	init_rwsem(&fs_info->subvol_sem);
2262 	fs_info->dev_replace.lock_owner = 0;
2263 	atomic_set(&fs_info->dev_replace.nesting_level, 0);
2264 	mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2265 	mutex_init(&fs_info->dev_replace.lock_management_lock);
2266 	mutex_init(&fs_info->dev_replace.lock);
2267 
2268 	spin_lock_init(&fs_info->qgroup_lock);
2269 	mutex_init(&fs_info->qgroup_ioctl_lock);
2270 	fs_info->qgroup_tree = RB_ROOT;
2271 	INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2272 	fs_info->qgroup_seq = 1;
2273 	fs_info->quota_enabled = 0;
2274 	fs_info->pending_quota_state = 0;
2275 	mutex_init(&fs_info->qgroup_rescan_lock);
2276 
2277 	btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2278 	btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2279 
2280 	init_waitqueue_head(&fs_info->transaction_throttle);
2281 	init_waitqueue_head(&fs_info->transaction_wait);
2282 	init_waitqueue_head(&fs_info->transaction_blocked_wait);
2283 	init_waitqueue_head(&fs_info->async_submit_wait);
2284 
2285 	ret = btrfs_alloc_stripe_hash_table(fs_info);
2286 	if (ret) {
2287 		err = ret;
2288 		goto fail_alloc;
2289 	}
2290 
2291 	__setup_root(4096, 4096, 4096, 4096, tree_root,
2292 		     fs_info, BTRFS_ROOT_TREE_OBJECTID);
2293 
2294 	invalidate_bdev(fs_devices->latest_bdev);
2295 
2296 	/*
2297 	 * Read super block and check the signature bytes only
2298 	 */
2299 	bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2300 	if (!bh) {
2301 		err = -EINVAL;
2302 		goto fail_alloc;
2303 	}
2304 
2305 	/*
2306 	 * We want to check superblock checksum, the type is stored inside.
2307 	 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2308 	 */
2309 	if (btrfs_check_super_csum(bh->b_data)) {
2310 		printk(KERN_ERR "btrfs: superblock checksum mismatch\n");
2311 		err = -EINVAL;
2312 		goto fail_alloc;
2313 	}
2314 
2315 	/*
2316 	 * super_copy is zeroed at allocation time and we never touch the
2317 	 * following bytes up to INFO_SIZE, the checksum is calculated from
2318 	 * the whole block of INFO_SIZE
2319 	 */
2320 	memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2321 	memcpy(fs_info->super_for_commit, fs_info->super_copy,
2322 	       sizeof(*fs_info->super_for_commit));
2323 	brelse(bh);
2324 
2325 	memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2326 
2327 	ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2328 	if (ret) {
2329 		printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2330 		err = -EINVAL;
2331 		goto fail_alloc;
2332 	}
2333 
2334 	disk_super = fs_info->super_copy;
2335 	if (!btrfs_super_root(disk_super))
2336 		goto fail_alloc;
2337 
2338 	/* check FS state, whether FS is broken. */
2339 	if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2340 		set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2341 
2342 	/*
2343 	 * run through our array of backup supers and setup
2344 	 * our ring pointer to the oldest one
2345 	 */
2346 	generation = btrfs_super_generation(disk_super);
2347 	find_oldest_super_backup(fs_info, generation);
2348 
2349 	/*
2350 	 * In the long term, we'll store the compression type in the super
2351 	 * block, and it'll be used for per file compression control.
2352 	 */
2353 	fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2354 
2355 	ret = btrfs_parse_options(tree_root, options);
2356 	if (ret) {
2357 		err = ret;
2358 		goto fail_alloc;
2359 	}
2360 
2361 	features = btrfs_super_incompat_flags(disk_super) &
2362 		~BTRFS_FEATURE_INCOMPAT_SUPP;
2363 	if (features) {
2364 		printk(KERN_ERR "BTRFS: couldn't mount because of "
2365 		       "unsupported optional features (%Lx).\n",
2366 		       (unsigned long long)features);
2367 		err = -EINVAL;
2368 		goto fail_alloc;
2369 	}
2370 
2371 	if (btrfs_super_leafsize(disk_super) !=
2372 	    btrfs_super_nodesize(disk_super)) {
2373 		printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2374 		       "blocksizes don't match.  node %d leaf %d\n",
2375 		       btrfs_super_nodesize(disk_super),
2376 		       btrfs_super_leafsize(disk_super));
2377 		err = -EINVAL;
2378 		goto fail_alloc;
2379 	}
2380 	if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2381 		printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2382 		       "blocksize (%d) was too large\n",
2383 		       btrfs_super_leafsize(disk_super));
2384 		err = -EINVAL;
2385 		goto fail_alloc;
2386 	}
2387 
2388 	features = btrfs_super_incompat_flags(disk_super);
2389 	features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2390 	if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2391 		features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2392 
2393 	if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2394 		printk(KERN_ERR "btrfs: has skinny extents\n");
2395 
2396 	/*
2397 	 * flag our filesystem as having big metadata blocks if
2398 	 * they are bigger than the page size
2399 	 */
2400 	if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2401 		if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2402 			printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2403 		features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2404 	}
2405 
2406 	nodesize = btrfs_super_nodesize(disk_super);
2407 	leafsize = btrfs_super_leafsize(disk_super);
2408 	sectorsize = btrfs_super_sectorsize(disk_super);
2409 	stripesize = btrfs_super_stripesize(disk_super);
2410 	fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
2411 	fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2412 
2413 	/*
2414 	 * mixed block groups end up with duplicate but slightly offset
2415 	 * extent buffers for the same range.  It leads to corruptions
2416 	 */
2417 	if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2418 	    (sectorsize != leafsize)) {
2419 		printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2420 				"are not allowed for mixed block groups on %s\n",
2421 				sb->s_id);
2422 		goto fail_alloc;
2423 	}
2424 
2425 	/*
2426 	 * Needn't use the lock because there is no other task which will
2427 	 * update the flag.
2428 	 */
2429 	btrfs_set_super_incompat_flags(disk_super, features);
2430 
2431 	features = btrfs_super_compat_ro_flags(disk_super) &
2432 		~BTRFS_FEATURE_COMPAT_RO_SUPP;
2433 	if (!(sb->s_flags & MS_RDONLY) && features) {
2434 		printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2435 		       "unsupported option features (%Lx).\n",
2436 		       (unsigned long long)features);
2437 		err = -EINVAL;
2438 		goto fail_alloc;
2439 	}
2440 
2441 	btrfs_init_workers(&fs_info->generic_worker,
2442 			   "genwork", 1, NULL);
2443 
2444 	btrfs_init_workers(&fs_info->workers, "worker",
2445 			   fs_info->thread_pool_size,
2446 			   &fs_info->generic_worker);
2447 
2448 	btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2449 			   fs_info->thread_pool_size,
2450 			   &fs_info->generic_worker);
2451 
2452 	btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc",
2453 			   fs_info->thread_pool_size,
2454 			   &fs_info->generic_worker);
2455 
2456 	btrfs_init_workers(&fs_info->submit_workers, "submit",
2457 			   min_t(u64, fs_devices->num_devices,
2458 			   fs_info->thread_pool_size),
2459 			   &fs_info->generic_worker);
2460 
2461 	btrfs_init_workers(&fs_info->caching_workers, "cache",
2462 			   2, &fs_info->generic_worker);
2463 
2464 	/* a higher idle thresh on the submit workers makes it much more
2465 	 * likely that bios will be send down in a sane order to the
2466 	 * devices
2467 	 */
2468 	fs_info->submit_workers.idle_thresh = 64;
2469 
2470 	fs_info->workers.idle_thresh = 16;
2471 	fs_info->workers.ordered = 1;
2472 
2473 	fs_info->delalloc_workers.idle_thresh = 2;
2474 	fs_info->delalloc_workers.ordered = 1;
2475 
2476 	btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2477 			   &fs_info->generic_worker);
2478 	btrfs_init_workers(&fs_info->endio_workers, "endio",
2479 			   fs_info->thread_pool_size,
2480 			   &fs_info->generic_worker);
2481 	btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2482 			   fs_info->thread_pool_size,
2483 			   &fs_info->generic_worker);
2484 	btrfs_init_workers(&fs_info->endio_meta_write_workers,
2485 			   "endio-meta-write", fs_info->thread_pool_size,
2486 			   &fs_info->generic_worker);
2487 	btrfs_init_workers(&fs_info->endio_raid56_workers,
2488 			   "endio-raid56", fs_info->thread_pool_size,
2489 			   &fs_info->generic_worker);
2490 	btrfs_init_workers(&fs_info->rmw_workers,
2491 			   "rmw", fs_info->thread_pool_size,
2492 			   &fs_info->generic_worker);
2493 	btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2494 			   fs_info->thread_pool_size,
2495 			   &fs_info->generic_worker);
2496 	btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2497 			   1, &fs_info->generic_worker);
2498 	btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2499 			   fs_info->thread_pool_size,
2500 			   &fs_info->generic_worker);
2501 	btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2502 			   fs_info->thread_pool_size,
2503 			   &fs_info->generic_worker);
2504 	btrfs_init_workers(&fs_info->qgroup_rescan_workers, "qgroup-rescan", 1,
2505 			   &fs_info->generic_worker);
2506 
2507 	/*
2508 	 * endios are largely parallel and should have a very
2509 	 * low idle thresh
2510 	 */
2511 	fs_info->endio_workers.idle_thresh = 4;
2512 	fs_info->endio_meta_workers.idle_thresh = 4;
2513 	fs_info->endio_raid56_workers.idle_thresh = 4;
2514 	fs_info->rmw_workers.idle_thresh = 2;
2515 
2516 	fs_info->endio_write_workers.idle_thresh = 2;
2517 	fs_info->endio_meta_write_workers.idle_thresh = 2;
2518 	fs_info->readahead_workers.idle_thresh = 2;
2519 
2520 	/*
2521 	 * btrfs_start_workers can really only fail because of ENOMEM so just
2522 	 * return -ENOMEM if any of these fail.
2523 	 */
2524 	ret = btrfs_start_workers(&fs_info->workers);
2525 	ret |= btrfs_start_workers(&fs_info->generic_worker);
2526 	ret |= btrfs_start_workers(&fs_info->submit_workers);
2527 	ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2528 	ret |= btrfs_start_workers(&fs_info->fixup_workers);
2529 	ret |= btrfs_start_workers(&fs_info->endio_workers);
2530 	ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2531 	ret |= btrfs_start_workers(&fs_info->rmw_workers);
2532 	ret |= btrfs_start_workers(&fs_info->endio_raid56_workers);
2533 	ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2534 	ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2535 	ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2536 	ret |= btrfs_start_workers(&fs_info->delayed_workers);
2537 	ret |= btrfs_start_workers(&fs_info->caching_workers);
2538 	ret |= btrfs_start_workers(&fs_info->readahead_workers);
2539 	ret |= btrfs_start_workers(&fs_info->flush_workers);
2540 	ret |= btrfs_start_workers(&fs_info->qgroup_rescan_workers);
2541 	if (ret) {
2542 		err = -ENOMEM;
2543 		goto fail_sb_buffer;
2544 	}
2545 
2546 	fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2547 	fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2548 				    4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2549 
2550 	tree_root->nodesize = nodesize;
2551 	tree_root->leafsize = leafsize;
2552 	tree_root->sectorsize = sectorsize;
2553 	tree_root->stripesize = stripesize;
2554 
2555 	sb->s_blocksize = sectorsize;
2556 	sb->s_blocksize_bits = blksize_bits(sectorsize);
2557 
2558 	if (disk_super->magic != cpu_to_le64(BTRFS_MAGIC)) {
2559 		printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2560 		goto fail_sb_buffer;
2561 	}
2562 
2563 	if (sectorsize != PAGE_SIZE) {
2564 		printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2565 		       "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2566 		goto fail_sb_buffer;
2567 	}
2568 
2569 	mutex_lock(&fs_info->chunk_mutex);
2570 	ret = btrfs_read_sys_array(tree_root);
2571 	mutex_unlock(&fs_info->chunk_mutex);
2572 	if (ret) {
2573 		printk(KERN_WARNING "btrfs: failed to read the system "
2574 		       "array on %s\n", sb->s_id);
2575 		goto fail_sb_buffer;
2576 	}
2577 
2578 	blocksize = btrfs_level_size(tree_root,
2579 				     btrfs_super_chunk_root_level(disk_super));
2580 	generation = btrfs_super_chunk_root_generation(disk_super);
2581 
2582 	__setup_root(nodesize, leafsize, sectorsize, stripesize,
2583 		     chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2584 
2585 	chunk_root->node = read_tree_block(chunk_root,
2586 					   btrfs_super_chunk_root(disk_super),
2587 					   blocksize, generation);
2588 	if (!chunk_root->node ||
2589 	    !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2590 		printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2591 		       sb->s_id);
2592 		goto fail_tree_roots;
2593 	}
2594 	btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2595 	chunk_root->commit_root = btrfs_root_node(chunk_root);
2596 
2597 	read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2598 	   (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2599 	   BTRFS_UUID_SIZE);
2600 
2601 	ret = btrfs_read_chunk_tree(chunk_root);
2602 	if (ret) {
2603 		printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2604 		       sb->s_id);
2605 		goto fail_tree_roots;
2606 	}
2607 
2608 	/*
2609 	 * keep the device that is marked to be the target device for the
2610 	 * dev_replace procedure
2611 	 */
2612 	btrfs_close_extra_devices(fs_info, fs_devices, 0);
2613 
2614 	if (!fs_devices->latest_bdev) {
2615 		printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2616 		       sb->s_id);
2617 		goto fail_tree_roots;
2618 	}
2619 
2620 retry_root_backup:
2621 	blocksize = btrfs_level_size(tree_root,
2622 				     btrfs_super_root_level(disk_super));
2623 	generation = btrfs_super_generation(disk_super);
2624 
2625 	tree_root->node = read_tree_block(tree_root,
2626 					  btrfs_super_root(disk_super),
2627 					  blocksize, generation);
2628 	if (!tree_root->node ||
2629 	    !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2630 		printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2631 		       sb->s_id);
2632 
2633 		goto recovery_tree_root;
2634 	}
2635 
2636 	btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2637 	tree_root->commit_root = btrfs_root_node(tree_root);
2638 
2639 	ret = find_and_setup_root(tree_root, fs_info,
2640 				  BTRFS_EXTENT_TREE_OBJECTID, extent_root);
2641 	if (ret)
2642 		goto recovery_tree_root;
2643 	extent_root->track_dirty = 1;
2644 
2645 	ret = find_and_setup_root(tree_root, fs_info,
2646 				  BTRFS_DEV_TREE_OBJECTID, dev_root);
2647 	if (ret)
2648 		goto recovery_tree_root;
2649 	dev_root->track_dirty = 1;
2650 
2651 	ret = find_and_setup_root(tree_root, fs_info,
2652 				  BTRFS_CSUM_TREE_OBJECTID, csum_root);
2653 	if (ret)
2654 		goto recovery_tree_root;
2655 	csum_root->track_dirty = 1;
2656 
2657 	ret = find_and_setup_root(tree_root, fs_info,
2658 				  BTRFS_QUOTA_TREE_OBJECTID, quota_root);
2659 	if (ret) {
2660 		kfree(quota_root);
2661 		quota_root = fs_info->quota_root = NULL;
2662 	} else {
2663 		quota_root->track_dirty = 1;
2664 		fs_info->quota_enabled = 1;
2665 		fs_info->pending_quota_state = 1;
2666 	}
2667 
2668 	fs_info->generation = generation;
2669 	fs_info->last_trans_committed = generation;
2670 
2671 	ret = btrfs_recover_balance(fs_info);
2672 	if (ret) {
2673 		printk(KERN_WARNING "btrfs: failed to recover balance\n");
2674 		goto fail_block_groups;
2675 	}
2676 
2677 	ret = btrfs_init_dev_stats(fs_info);
2678 	if (ret) {
2679 		printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
2680 		       ret);
2681 		goto fail_block_groups;
2682 	}
2683 
2684 	ret = btrfs_init_dev_replace(fs_info);
2685 	if (ret) {
2686 		pr_err("btrfs: failed to init dev_replace: %d\n", ret);
2687 		goto fail_block_groups;
2688 	}
2689 
2690 	btrfs_close_extra_devices(fs_info, fs_devices, 1);
2691 
2692 	ret = btrfs_init_space_info(fs_info);
2693 	if (ret) {
2694 		printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2695 		goto fail_block_groups;
2696 	}
2697 
2698 	ret = btrfs_read_block_groups(extent_root);
2699 	if (ret) {
2700 		printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2701 		goto fail_block_groups;
2702 	}
2703 	fs_info->num_tolerated_disk_barrier_failures =
2704 		btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2705 	if (fs_info->fs_devices->missing_devices >
2706 	     fs_info->num_tolerated_disk_barrier_failures &&
2707 	    !(sb->s_flags & MS_RDONLY)) {
2708 		printk(KERN_WARNING
2709 		       "Btrfs: too many missing devices, writeable mount is not allowed\n");
2710 		goto fail_block_groups;
2711 	}
2712 
2713 	fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2714 					       "btrfs-cleaner");
2715 	if (IS_ERR(fs_info->cleaner_kthread))
2716 		goto fail_block_groups;
2717 
2718 	fs_info->transaction_kthread = kthread_run(transaction_kthread,
2719 						   tree_root,
2720 						   "btrfs-transaction");
2721 	if (IS_ERR(fs_info->transaction_kthread))
2722 		goto fail_cleaner;
2723 
2724 	if (!btrfs_test_opt(tree_root, SSD) &&
2725 	    !btrfs_test_opt(tree_root, NOSSD) &&
2726 	    !fs_info->fs_devices->rotating) {
2727 		printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2728 		       "mode\n");
2729 		btrfs_set_opt(fs_info->mount_opt, SSD);
2730 	}
2731 
2732 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2733 	if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2734 		ret = btrfsic_mount(tree_root, fs_devices,
2735 				    btrfs_test_opt(tree_root,
2736 					CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2737 				    1 : 0,
2738 				    fs_info->check_integrity_print_mask);
2739 		if (ret)
2740 			printk(KERN_WARNING "btrfs: failed to initialize"
2741 			       " integrity check module %s\n", sb->s_id);
2742 	}
2743 #endif
2744 	ret = btrfs_read_qgroup_config(fs_info);
2745 	if (ret)
2746 		goto fail_trans_kthread;
2747 
2748 	/* do not make disk changes in broken FS */
2749 	if (btrfs_super_log_root(disk_super) != 0) {
2750 		u64 bytenr = btrfs_super_log_root(disk_super);
2751 
2752 		if (fs_devices->rw_devices == 0) {
2753 			printk(KERN_WARNING "Btrfs log replay required "
2754 			       "on RO media\n");
2755 			err = -EIO;
2756 			goto fail_qgroup;
2757 		}
2758 		blocksize =
2759 		     btrfs_level_size(tree_root,
2760 				      btrfs_super_log_root_level(disk_super));
2761 
2762 		log_tree_root = btrfs_alloc_root(fs_info);
2763 		if (!log_tree_root) {
2764 			err = -ENOMEM;
2765 			goto fail_qgroup;
2766 		}
2767 
2768 		__setup_root(nodesize, leafsize, sectorsize, stripesize,
2769 			     log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2770 
2771 		log_tree_root->node = read_tree_block(tree_root, bytenr,
2772 						      blocksize,
2773 						      generation + 1);
2774 		if (!log_tree_root->node ||
2775 		    !extent_buffer_uptodate(log_tree_root->node)) {
2776 			printk(KERN_ERR "btrfs: failed to read log tree\n");
2777 			free_extent_buffer(log_tree_root->node);
2778 			kfree(log_tree_root);
2779 			goto fail_trans_kthread;
2780 		}
2781 		/* returns with log_tree_root freed on success */
2782 		ret = btrfs_recover_log_trees(log_tree_root);
2783 		if (ret) {
2784 			btrfs_error(tree_root->fs_info, ret,
2785 				    "Failed to recover log tree");
2786 			free_extent_buffer(log_tree_root->node);
2787 			kfree(log_tree_root);
2788 			goto fail_trans_kthread;
2789 		}
2790 
2791 		if (sb->s_flags & MS_RDONLY) {
2792 			ret = btrfs_commit_super(tree_root);
2793 			if (ret)
2794 				goto fail_trans_kthread;
2795 		}
2796 	}
2797 
2798 	ret = btrfs_find_orphan_roots(tree_root);
2799 	if (ret)
2800 		goto fail_trans_kthread;
2801 
2802 	if (!(sb->s_flags & MS_RDONLY)) {
2803 		ret = btrfs_cleanup_fs_roots(fs_info);
2804 		if (ret)
2805 			goto fail_trans_kthread;
2806 
2807 		ret = btrfs_recover_relocation(tree_root);
2808 		if (ret < 0) {
2809 			printk(KERN_WARNING
2810 			       "btrfs: failed to recover relocation\n");
2811 			err = -EINVAL;
2812 			goto fail_qgroup;
2813 		}
2814 	}
2815 
2816 	location.objectid = BTRFS_FS_TREE_OBJECTID;
2817 	location.type = BTRFS_ROOT_ITEM_KEY;
2818 	location.offset = (u64)-1;
2819 
2820 	fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2821 	if (!fs_info->fs_root)
2822 		goto fail_qgroup;
2823 	if (IS_ERR(fs_info->fs_root)) {
2824 		err = PTR_ERR(fs_info->fs_root);
2825 		goto fail_qgroup;
2826 	}
2827 
2828 	if (sb->s_flags & MS_RDONLY)
2829 		return 0;
2830 
2831 	down_read(&fs_info->cleanup_work_sem);
2832 	if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2833 	    (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2834 		up_read(&fs_info->cleanup_work_sem);
2835 		close_ctree(tree_root);
2836 		return ret;
2837 	}
2838 	up_read(&fs_info->cleanup_work_sem);
2839 
2840 	ret = btrfs_resume_balance_async(fs_info);
2841 	if (ret) {
2842 		printk(KERN_WARNING "btrfs: failed to resume balance\n");
2843 		close_ctree(tree_root);
2844 		return ret;
2845 	}
2846 
2847 	ret = btrfs_resume_dev_replace_async(fs_info);
2848 	if (ret) {
2849 		pr_warn("btrfs: failed to resume dev_replace\n");
2850 		close_ctree(tree_root);
2851 		return ret;
2852 	}
2853 
2854 	return 0;
2855 
2856 fail_qgroup:
2857 	btrfs_free_qgroup_config(fs_info);
2858 fail_trans_kthread:
2859 	kthread_stop(fs_info->transaction_kthread);
2860 	del_fs_roots(fs_info);
2861 	btrfs_cleanup_transaction(fs_info->tree_root);
2862 fail_cleaner:
2863 	kthread_stop(fs_info->cleaner_kthread);
2864 
2865 	/*
2866 	 * make sure we're done with the btree inode before we stop our
2867 	 * kthreads
2868 	 */
2869 	filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2870 
2871 fail_block_groups:
2872 	btrfs_put_block_group_cache(fs_info);
2873 	btrfs_free_block_groups(fs_info);
2874 
2875 fail_tree_roots:
2876 	free_root_pointers(fs_info, 1);
2877 	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2878 
2879 fail_sb_buffer:
2880 	btrfs_stop_all_workers(fs_info);
2881 fail_alloc:
2882 fail_iput:
2883 	btrfs_mapping_tree_free(&fs_info->mapping_tree);
2884 
2885 	iput(fs_info->btree_inode);
2886 fail_delalloc_bytes:
2887 	percpu_counter_destroy(&fs_info->delalloc_bytes);
2888 fail_dirty_metadata_bytes:
2889 	percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
2890 fail_bdi:
2891 	bdi_destroy(&fs_info->bdi);
2892 fail_srcu:
2893 	cleanup_srcu_struct(&fs_info->subvol_srcu);
2894 fail:
2895 	btrfs_free_stripe_hash_table(fs_info);
2896 	btrfs_close_devices(fs_info->fs_devices);
2897 	return err;
2898 
2899 recovery_tree_root:
2900 	if (!btrfs_test_opt(tree_root, RECOVERY))
2901 		goto fail_tree_roots;
2902 
2903 	free_root_pointers(fs_info, 0);
2904 
2905 	/* don't use the log in recovery mode, it won't be valid */
2906 	btrfs_set_super_log_root(disk_super, 0);
2907 
2908 	/* we can't trust the free space cache either */
2909 	btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2910 
2911 	ret = next_root_backup(fs_info, fs_info->super_copy,
2912 			       &num_backups_tried, &backup_index);
2913 	if (ret == -1)
2914 		goto fail_block_groups;
2915 	goto retry_root_backup;
2916 }
2917 
2918 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2919 {
2920 	if (uptodate) {
2921 		set_buffer_uptodate(bh);
2922 	} else {
2923 		struct btrfs_device *device = (struct btrfs_device *)
2924 			bh->b_private;
2925 
2926 		printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
2927 					  "I/O error on %s\n",
2928 					  rcu_str_deref(device->name));
2929 		/* note, we dont' set_buffer_write_io_error because we have
2930 		 * our own ways of dealing with the IO errors
2931 		 */
2932 		clear_buffer_uptodate(bh);
2933 		btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
2934 	}
2935 	unlock_buffer(bh);
2936 	put_bh(bh);
2937 }
2938 
2939 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2940 {
2941 	struct buffer_head *bh;
2942 	struct buffer_head *latest = NULL;
2943 	struct btrfs_super_block *super;
2944 	int i;
2945 	u64 transid = 0;
2946 	u64 bytenr;
2947 
2948 	/* we would like to check all the supers, but that would make
2949 	 * a btrfs mount succeed after a mkfs from a different FS.
2950 	 * So, we need to add a special mount option to scan for
2951 	 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2952 	 */
2953 	for (i = 0; i < 1; i++) {
2954 		bytenr = btrfs_sb_offset(i);
2955 		if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2956 			break;
2957 		bh = __bread(bdev, bytenr / 4096, 4096);
2958 		if (!bh)
2959 			continue;
2960 
2961 		super = (struct btrfs_super_block *)bh->b_data;
2962 		if (btrfs_super_bytenr(super) != bytenr ||
2963 		    super->magic != cpu_to_le64(BTRFS_MAGIC)) {
2964 			brelse(bh);
2965 			continue;
2966 		}
2967 
2968 		if (!latest || btrfs_super_generation(super) > transid) {
2969 			brelse(latest);
2970 			latest = bh;
2971 			transid = btrfs_super_generation(super);
2972 		} else {
2973 			brelse(bh);
2974 		}
2975 	}
2976 	return latest;
2977 }
2978 
2979 /*
2980  * this should be called twice, once with wait == 0 and
2981  * once with wait == 1.  When wait == 0 is done, all the buffer heads
2982  * we write are pinned.
2983  *
2984  * They are released when wait == 1 is done.
2985  * max_mirrors must be the same for both runs, and it indicates how
2986  * many supers on this one device should be written.
2987  *
2988  * max_mirrors == 0 means to write them all.
2989  */
2990 static int write_dev_supers(struct btrfs_device *device,
2991 			    struct btrfs_super_block *sb,
2992 			    int do_barriers, int wait, int max_mirrors)
2993 {
2994 	struct buffer_head *bh;
2995 	int i;
2996 	int ret;
2997 	int errors = 0;
2998 	u32 crc;
2999 	u64 bytenr;
3000 
3001 	if (max_mirrors == 0)
3002 		max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3003 
3004 	for (i = 0; i < max_mirrors; i++) {
3005 		bytenr = btrfs_sb_offset(i);
3006 		if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3007 			break;
3008 
3009 		if (wait) {
3010 			bh = __find_get_block(device->bdev, bytenr / 4096,
3011 					      BTRFS_SUPER_INFO_SIZE);
3012 			if (!bh) {
3013 				errors++;
3014 				continue;
3015 			}
3016 			wait_on_buffer(bh);
3017 			if (!buffer_uptodate(bh))
3018 				errors++;
3019 
3020 			/* drop our reference */
3021 			brelse(bh);
3022 
3023 			/* drop the reference from the wait == 0 run */
3024 			brelse(bh);
3025 			continue;
3026 		} else {
3027 			btrfs_set_super_bytenr(sb, bytenr);
3028 
3029 			crc = ~(u32)0;
3030 			crc = btrfs_csum_data((char *)sb +
3031 					      BTRFS_CSUM_SIZE, crc,
3032 					      BTRFS_SUPER_INFO_SIZE -
3033 					      BTRFS_CSUM_SIZE);
3034 			btrfs_csum_final(crc, sb->csum);
3035 
3036 			/*
3037 			 * one reference for us, and we leave it for the
3038 			 * caller
3039 			 */
3040 			bh = __getblk(device->bdev, bytenr / 4096,
3041 				      BTRFS_SUPER_INFO_SIZE);
3042 			if (!bh) {
3043 				printk(KERN_ERR "btrfs: couldn't get super "
3044 				       "buffer head for bytenr %Lu\n", bytenr);
3045 				errors++;
3046 				continue;
3047 			}
3048 
3049 			memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3050 
3051 			/* one reference for submit_bh */
3052 			get_bh(bh);
3053 
3054 			set_buffer_uptodate(bh);
3055 			lock_buffer(bh);
3056 			bh->b_end_io = btrfs_end_buffer_write_sync;
3057 			bh->b_private = device;
3058 		}
3059 
3060 		/*
3061 		 * we fua the first super.  The others we allow
3062 		 * to go down lazy.
3063 		 */
3064 		ret = btrfsic_submit_bh(WRITE_FUA, bh);
3065 		if (ret)
3066 			errors++;
3067 	}
3068 	return errors < i ? 0 : -1;
3069 }
3070 
3071 /*
3072  * endio for the write_dev_flush, this will wake anyone waiting
3073  * for the barrier when it is done
3074  */
3075 static void btrfs_end_empty_barrier(struct bio *bio, int err)
3076 {
3077 	if (err) {
3078 		if (err == -EOPNOTSUPP)
3079 			set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3080 		clear_bit(BIO_UPTODATE, &bio->bi_flags);
3081 	}
3082 	if (bio->bi_private)
3083 		complete(bio->bi_private);
3084 	bio_put(bio);
3085 }
3086 
3087 /*
3088  * trigger flushes for one the devices.  If you pass wait == 0, the flushes are
3089  * sent down.  With wait == 1, it waits for the previous flush.
3090  *
3091  * any device where the flush fails with eopnotsupp are flagged as not-barrier
3092  * capable
3093  */
3094 static int write_dev_flush(struct btrfs_device *device, int wait)
3095 {
3096 	struct bio *bio;
3097 	int ret = 0;
3098 
3099 	if (device->nobarriers)
3100 		return 0;
3101 
3102 	if (wait) {
3103 		bio = device->flush_bio;
3104 		if (!bio)
3105 			return 0;
3106 
3107 		wait_for_completion(&device->flush_wait);
3108 
3109 		if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3110 			printk_in_rcu("btrfs: disabling barriers on dev %s\n",
3111 				      rcu_str_deref(device->name));
3112 			device->nobarriers = 1;
3113 		} else if (!bio_flagged(bio, BIO_UPTODATE)) {
3114 			ret = -EIO;
3115 			btrfs_dev_stat_inc_and_print(device,
3116 				BTRFS_DEV_STAT_FLUSH_ERRS);
3117 		}
3118 
3119 		/* drop the reference from the wait == 0 run */
3120 		bio_put(bio);
3121 		device->flush_bio = NULL;
3122 
3123 		return ret;
3124 	}
3125 
3126 	/*
3127 	 * one reference for us, and we leave it for the
3128 	 * caller
3129 	 */
3130 	device->flush_bio = NULL;
3131 	bio = bio_alloc(GFP_NOFS, 0);
3132 	if (!bio)
3133 		return -ENOMEM;
3134 
3135 	bio->bi_end_io = btrfs_end_empty_barrier;
3136 	bio->bi_bdev = device->bdev;
3137 	init_completion(&device->flush_wait);
3138 	bio->bi_private = &device->flush_wait;
3139 	device->flush_bio = bio;
3140 
3141 	bio_get(bio);
3142 	btrfsic_submit_bio(WRITE_FLUSH, bio);
3143 
3144 	return 0;
3145 }
3146 
3147 /*
3148  * send an empty flush down to each device in parallel,
3149  * then wait for them
3150  */
3151 static int barrier_all_devices(struct btrfs_fs_info *info)
3152 {
3153 	struct list_head *head;
3154 	struct btrfs_device *dev;
3155 	int errors_send = 0;
3156 	int errors_wait = 0;
3157 	int ret;
3158 
3159 	/* send down all the barriers */
3160 	head = &info->fs_devices->devices;
3161 	list_for_each_entry_rcu(dev, head, dev_list) {
3162 		if (!dev->bdev) {
3163 			errors_send++;
3164 			continue;
3165 		}
3166 		if (!dev->in_fs_metadata || !dev->writeable)
3167 			continue;
3168 
3169 		ret = write_dev_flush(dev, 0);
3170 		if (ret)
3171 			errors_send++;
3172 	}
3173 
3174 	/* wait for all the barriers */
3175 	list_for_each_entry_rcu(dev, head, dev_list) {
3176 		if (!dev->bdev) {
3177 			errors_wait++;
3178 			continue;
3179 		}
3180 		if (!dev->in_fs_metadata || !dev->writeable)
3181 			continue;
3182 
3183 		ret = write_dev_flush(dev, 1);
3184 		if (ret)
3185 			errors_wait++;
3186 	}
3187 	if (errors_send > info->num_tolerated_disk_barrier_failures ||
3188 	    errors_wait > info->num_tolerated_disk_barrier_failures)
3189 		return -EIO;
3190 	return 0;
3191 }
3192 
3193 int btrfs_calc_num_tolerated_disk_barrier_failures(
3194 	struct btrfs_fs_info *fs_info)
3195 {
3196 	struct btrfs_ioctl_space_info space;
3197 	struct btrfs_space_info *sinfo;
3198 	u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3199 		       BTRFS_BLOCK_GROUP_SYSTEM,
3200 		       BTRFS_BLOCK_GROUP_METADATA,
3201 		       BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3202 	int num_types = 4;
3203 	int i;
3204 	int c;
3205 	int num_tolerated_disk_barrier_failures =
3206 		(int)fs_info->fs_devices->num_devices;
3207 
3208 	for (i = 0; i < num_types; i++) {
3209 		struct btrfs_space_info *tmp;
3210 
3211 		sinfo = NULL;
3212 		rcu_read_lock();
3213 		list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3214 			if (tmp->flags == types[i]) {
3215 				sinfo = tmp;
3216 				break;
3217 			}
3218 		}
3219 		rcu_read_unlock();
3220 
3221 		if (!sinfo)
3222 			continue;
3223 
3224 		down_read(&sinfo->groups_sem);
3225 		for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3226 			if (!list_empty(&sinfo->block_groups[c])) {
3227 				u64 flags;
3228 
3229 				btrfs_get_block_group_info(
3230 					&sinfo->block_groups[c], &space);
3231 				if (space.total_bytes == 0 ||
3232 				    space.used_bytes == 0)
3233 					continue;
3234 				flags = space.flags;
3235 				/*
3236 				 * return
3237 				 * 0: if dup, single or RAID0 is configured for
3238 				 *    any of metadata, system or data, else
3239 				 * 1: if RAID5 is configured, or if RAID1 or
3240 				 *    RAID10 is configured and only two mirrors
3241 				 *    are used, else
3242 				 * 2: if RAID6 is configured, else
3243 				 * num_mirrors - 1: if RAID1 or RAID10 is
3244 				 *                  configured and more than
3245 				 *                  2 mirrors are used.
3246 				 */
3247 				if (num_tolerated_disk_barrier_failures > 0 &&
3248 				    ((flags & (BTRFS_BLOCK_GROUP_DUP |
3249 					       BTRFS_BLOCK_GROUP_RAID0)) ||
3250 				     ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3251 				      == 0)))
3252 					num_tolerated_disk_barrier_failures = 0;
3253 				else if (num_tolerated_disk_barrier_failures > 1) {
3254 					if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3255 					    BTRFS_BLOCK_GROUP_RAID5 |
3256 					    BTRFS_BLOCK_GROUP_RAID10)) {
3257 						num_tolerated_disk_barrier_failures = 1;
3258 					} else if (flags &
3259 						   BTRFS_BLOCK_GROUP_RAID5) {
3260 						num_tolerated_disk_barrier_failures = 2;
3261 					}
3262 				}
3263 			}
3264 		}
3265 		up_read(&sinfo->groups_sem);
3266 	}
3267 
3268 	return num_tolerated_disk_barrier_failures;
3269 }
3270 
3271 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3272 {
3273 	struct list_head *head;
3274 	struct btrfs_device *dev;
3275 	struct btrfs_super_block *sb;
3276 	struct btrfs_dev_item *dev_item;
3277 	int ret;
3278 	int do_barriers;
3279 	int max_errors;
3280 	int total_errors = 0;
3281 	u64 flags;
3282 
3283 	max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3284 	do_barriers = !btrfs_test_opt(root, NOBARRIER);
3285 	backup_super_roots(root->fs_info);
3286 
3287 	sb = root->fs_info->super_for_commit;
3288 	dev_item = &sb->dev_item;
3289 
3290 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3291 	head = &root->fs_info->fs_devices->devices;
3292 
3293 	if (do_barriers) {
3294 		ret = barrier_all_devices(root->fs_info);
3295 		if (ret) {
3296 			mutex_unlock(
3297 				&root->fs_info->fs_devices->device_list_mutex);
3298 			btrfs_error(root->fs_info, ret,
3299 				    "errors while submitting device barriers.");
3300 			return ret;
3301 		}
3302 	}
3303 
3304 	list_for_each_entry_rcu(dev, head, dev_list) {
3305 		if (!dev->bdev) {
3306 			total_errors++;
3307 			continue;
3308 		}
3309 		if (!dev->in_fs_metadata || !dev->writeable)
3310 			continue;
3311 
3312 		btrfs_set_stack_device_generation(dev_item, 0);
3313 		btrfs_set_stack_device_type(dev_item, dev->type);
3314 		btrfs_set_stack_device_id(dev_item, dev->devid);
3315 		btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3316 		btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3317 		btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3318 		btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3319 		btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3320 		memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3321 		memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3322 
3323 		flags = btrfs_super_flags(sb);
3324 		btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3325 
3326 		ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3327 		if (ret)
3328 			total_errors++;
3329 	}
3330 	if (total_errors > max_errors) {
3331 		printk(KERN_ERR "btrfs: %d errors while writing supers\n",
3332 		       total_errors);
3333 
3334 		/* This shouldn't happen. FUA is masked off if unsupported */
3335 		BUG();
3336 	}
3337 
3338 	total_errors = 0;
3339 	list_for_each_entry_rcu(dev, head, dev_list) {
3340 		if (!dev->bdev)
3341 			continue;
3342 		if (!dev->in_fs_metadata || !dev->writeable)
3343 			continue;
3344 
3345 		ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3346 		if (ret)
3347 			total_errors++;
3348 	}
3349 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3350 	if (total_errors > max_errors) {
3351 		btrfs_error(root->fs_info, -EIO,
3352 			    "%d errors while writing supers", total_errors);
3353 		return -EIO;
3354 	}
3355 	return 0;
3356 }
3357 
3358 int write_ctree_super(struct btrfs_trans_handle *trans,
3359 		      struct btrfs_root *root, int max_mirrors)
3360 {
3361 	int ret;
3362 
3363 	ret = write_all_supers(root, max_mirrors);
3364 	return ret;
3365 }
3366 
3367 void btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
3368 {
3369 	spin_lock(&fs_info->fs_roots_radix_lock);
3370 	radix_tree_delete(&fs_info->fs_roots_radix,
3371 			  (unsigned long)root->root_key.objectid);
3372 	spin_unlock(&fs_info->fs_roots_radix_lock);
3373 
3374 	if (btrfs_root_refs(&root->root_item) == 0)
3375 		synchronize_srcu(&fs_info->subvol_srcu);
3376 
3377 	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3378 		btrfs_free_log(NULL, root);
3379 		btrfs_free_log_root_tree(NULL, fs_info);
3380 	}
3381 
3382 	__btrfs_remove_free_space_cache(root->free_ino_pinned);
3383 	__btrfs_remove_free_space_cache(root->free_ino_ctl);
3384 	free_fs_root(root);
3385 }
3386 
3387 static void free_fs_root(struct btrfs_root *root)
3388 {
3389 	iput(root->cache_inode);
3390 	WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3391 	if (root->anon_dev)
3392 		free_anon_bdev(root->anon_dev);
3393 	free_extent_buffer(root->node);
3394 	free_extent_buffer(root->commit_root);
3395 	kfree(root->free_ino_ctl);
3396 	kfree(root->free_ino_pinned);
3397 	kfree(root->name);
3398 	kfree(root);
3399 }
3400 
3401 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3402 {
3403 	u64 root_objectid = 0;
3404 	struct btrfs_root *gang[8];
3405 	int i;
3406 	int ret;
3407 
3408 	while (1) {
3409 		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3410 					     (void **)gang, root_objectid,
3411 					     ARRAY_SIZE(gang));
3412 		if (!ret)
3413 			break;
3414 
3415 		root_objectid = gang[ret - 1]->root_key.objectid + 1;
3416 		for (i = 0; i < ret; i++) {
3417 			int err;
3418 
3419 			root_objectid = gang[i]->root_key.objectid;
3420 			err = btrfs_orphan_cleanup(gang[i]);
3421 			if (err)
3422 				return err;
3423 		}
3424 		root_objectid++;
3425 	}
3426 	return 0;
3427 }
3428 
3429 int btrfs_commit_super(struct btrfs_root *root)
3430 {
3431 	struct btrfs_trans_handle *trans;
3432 	int ret;
3433 
3434 	mutex_lock(&root->fs_info->cleaner_mutex);
3435 	btrfs_run_delayed_iputs(root);
3436 	mutex_unlock(&root->fs_info->cleaner_mutex);
3437 	wake_up_process(root->fs_info->cleaner_kthread);
3438 
3439 	/* wait until ongoing cleanup work done */
3440 	down_write(&root->fs_info->cleanup_work_sem);
3441 	up_write(&root->fs_info->cleanup_work_sem);
3442 
3443 	trans = btrfs_join_transaction(root);
3444 	if (IS_ERR(trans))
3445 		return PTR_ERR(trans);
3446 	ret = btrfs_commit_transaction(trans, root);
3447 	if (ret)
3448 		return ret;
3449 	/* run commit again to drop the original snapshot */
3450 	trans = btrfs_join_transaction(root);
3451 	if (IS_ERR(trans))
3452 		return PTR_ERR(trans);
3453 	ret = btrfs_commit_transaction(trans, root);
3454 	if (ret)
3455 		return ret;
3456 	ret = btrfs_write_and_wait_transaction(NULL, root);
3457 	if (ret) {
3458 		btrfs_error(root->fs_info, ret,
3459 			    "Failed to sync btree inode to disk.");
3460 		return ret;
3461 	}
3462 
3463 	ret = write_ctree_super(NULL, root, 0);
3464 	return ret;
3465 }
3466 
3467 int close_ctree(struct btrfs_root *root)
3468 {
3469 	struct btrfs_fs_info *fs_info = root->fs_info;
3470 	int ret;
3471 
3472 	fs_info->closing = 1;
3473 	smp_mb();
3474 
3475 	/* pause restriper - we want to resume on mount */
3476 	btrfs_pause_balance(fs_info);
3477 
3478 	btrfs_dev_replace_suspend_for_unmount(fs_info);
3479 
3480 	btrfs_scrub_cancel(fs_info);
3481 
3482 	/* wait for any defraggers to finish */
3483 	wait_event(fs_info->transaction_wait,
3484 		   (atomic_read(&fs_info->defrag_running) == 0));
3485 
3486 	/* clear out the rbtree of defraggable inodes */
3487 	btrfs_cleanup_defrag_inodes(fs_info);
3488 
3489 	if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3490 		ret = btrfs_commit_super(root);
3491 		if (ret)
3492 			printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3493 	}
3494 
3495 	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3496 		btrfs_error_commit_super(root);
3497 
3498 	btrfs_put_block_group_cache(fs_info);
3499 
3500 	kthread_stop(fs_info->transaction_kthread);
3501 	kthread_stop(fs_info->cleaner_kthread);
3502 
3503 	fs_info->closing = 2;
3504 	smp_mb();
3505 
3506 	btrfs_free_qgroup_config(root->fs_info);
3507 
3508 	if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3509 		printk(KERN_INFO "btrfs: at unmount delalloc count %lld\n",
3510 		       percpu_counter_sum(&fs_info->delalloc_bytes));
3511 	}
3512 
3513 	free_root_pointers(fs_info, 1);
3514 
3515 	btrfs_free_block_groups(fs_info);
3516 
3517 	del_fs_roots(fs_info);
3518 
3519 	iput(fs_info->btree_inode);
3520 
3521 	btrfs_stop_all_workers(fs_info);
3522 
3523 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3524 	if (btrfs_test_opt(root, CHECK_INTEGRITY))
3525 		btrfsic_unmount(root, fs_info->fs_devices);
3526 #endif
3527 
3528 	btrfs_close_devices(fs_info->fs_devices);
3529 	btrfs_mapping_tree_free(&fs_info->mapping_tree);
3530 
3531 	percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3532 	percpu_counter_destroy(&fs_info->delalloc_bytes);
3533 	bdi_destroy(&fs_info->bdi);
3534 	cleanup_srcu_struct(&fs_info->subvol_srcu);
3535 
3536 	btrfs_free_stripe_hash_table(fs_info);
3537 
3538 	return 0;
3539 }
3540 
3541 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3542 			  int atomic)
3543 {
3544 	int ret;
3545 	struct inode *btree_inode = buf->pages[0]->mapping->host;
3546 
3547 	ret = extent_buffer_uptodate(buf);
3548 	if (!ret)
3549 		return ret;
3550 
3551 	ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3552 				    parent_transid, atomic);
3553 	if (ret == -EAGAIN)
3554 		return ret;
3555 	return !ret;
3556 }
3557 
3558 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3559 {
3560 	return set_extent_buffer_uptodate(buf);
3561 }
3562 
3563 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3564 {
3565 	struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3566 	u64 transid = btrfs_header_generation(buf);
3567 	int was_dirty;
3568 
3569 	btrfs_assert_tree_locked(buf);
3570 	if (transid != root->fs_info->generation)
3571 		WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3572 		       "found %llu running %llu\n",
3573 			(unsigned long long)buf->start,
3574 			(unsigned long long)transid,
3575 			(unsigned long long)root->fs_info->generation);
3576 	was_dirty = set_extent_buffer_dirty(buf);
3577 	if (!was_dirty)
3578 		__percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3579 				     buf->len,
3580 				     root->fs_info->dirty_metadata_batch);
3581 }
3582 
3583 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3584 					int flush_delayed)
3585 {
3586 	/*
3587 	 * looks as though older kernels can get into trouble with
3588 	 * this code, they end up stuck in balance_dirty_pages forever
3589 	 */
3590 	int ret;
3591 
3592 	if (current->flags & PF_MEMALLOC)
3593 		return;
3594 
3595 	if (flush_delayed)
3596 		btrfs_balance_delayed_items(root);
3597 
3598 	ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3599 				     BTRFS_DIRTY_METADATA_THRESH);
3600 	if (ret > 0) {
3601 		balance_dirty_pages_ratelimited(
3602 				   root->fs_info->btree_inode->i_mapping);
3603 	}
3604 	return;
3605 }
3606 
3607 void btrfs_btree_balance_dirty(struct btrfs_root *root)
3608 {
3609 	__btrfs_btree_balance_dirty(root, 1);
3610 }
3611 
3612 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3613 {
3614 	__btrfs_btree_balance_dirty(root, 0);
3615 }
3616 
3617 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3618 {
3619 	struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3620 	return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3621 }
3622 
3623 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3624 			      int read_only)
3625 {
3626 	/*
3627 	 * Placeholder for checks
3628 	 */
3629 	return 0;
3630 }
3631 
3632 static void btrfs_error_commit_super(struct btrfs_root *root)
3633 {
3634 	mutex_lock(&root->fs_info->cleaner_mutex);
3635 	btrfs_run_delayed_iputs(root);
3636 	mutex_unlock(&root->fs_info->cleaner_mutex);
3637 
3638 	down_write(&root->fs_info->cleanup_work_sem);
3639 	up_write(&root->fs_info->cleanup_work_sem);
3640 
3641 	/* cleanup FS via transaction */
3642 	btrfs_cleanup_transaction(root);
3643 }
3644 
3645 static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3646 					     struct btrfs_root *root)
3647 {
3648 	struct btrfs_inode *btrfs_inode;
3649 	struct list_head splice;
3650 
3651 	INIT_LIST_HEAD(&splice);
3652 
3653 	mutex_lock(&root->fs_info->ordered_operations_mutex);
3654 	spin_lock(&root->fs_info->ordered_extent_lock);
3655 
3656 	list_splice_init(&t->ordered_operations, &splice);
3657 	while (!list_empty(&splice)) {
3658 		btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3659 					 ordered_operations);
3660 
3661 		list_del_init(&btrfs_inode->ordered_operations);
3662 
3663 		btrfs_invalidate_inodes(btrfs_inode->root);
3664 	}
3665 
3666 	spin_unlock(&root->fs_info->ordered_extent_lock);
3667 	mutex_unlock(&root->fs_info->ordered_operations_mutex);
3668 }
3669 
3670 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3671 {
3672 	struct btrfs_ordered_extent *ordered;
3673 
3674 	spin_lock(&root->fs_info->ordered_extent_lock);
3675 	/*
3676 	 * This will just short circuit the ordered completion stuff which will
3677 	 * make sure the ordered extent gets properly cleaned up.
3678 	 */
3679 	list_for_each_entry(ordered, &root->fs_info->ordered_extents,
3680 			    root_extent_list)
3681 		set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
3682 	spin_unlock(&root->fs_info->ordered_extent_lock);
3683 }
3684 
3685 int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3686 			       struct btrfs_root *root)
3687 {
3688 	struct rb_node *node;
3689 	struct btrfs_delayed_ref_root *delayed_refs;
3690 	struct btrfs_delayed_ref_node *ref;
3691 	int ret = 0;
3692 
3693 	delayed_refs = &trans->delayed_refs;
3694 
3695 	spin_lock(&delayed_refs->lock);
3696 	if (delayed_refs->num_entries == 0) {
3697 		spin_unlock(&delayed_refs->lock);
3698 		printk(KERN_INFO "delayed_refs has NO entry\n");
3699 		return ret;
3700 	}
3701 
3702 	while ((node = rb_first(&delayed_refs->root)) != NULL) {
3703 		struct btrfs_delayed_ref_head *head = NULL;
3704 
3705 		ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3706 		atomic_set(&ref->refs, 1);
3707 		if (btrfs_delayed_ref_is_head(ref)) {
3708 
3709 			head = btrfs_delayed_node_to_head(ref);
3710 			if (!mutex_trylock(&head->mutex)) {
3711 				atomic_inc(&ref->refs);
3712 				spin_unlock(&delayed_refs->lock);
3713 
3714 				/* Need to wait for the delayed ref to run */
3715 				mutex_lock(&head->mutex);
3716 				mutex_unlock(&head->mutex);
3717 				btrfs_put_delayed_ref(ref);
3718 
3719 				spin_lock(&delayed_refs->lock);
3720 				continue;
3721 			}
3722 
3723 			if (head->must_insert_reserved)
3724 				btrfs_pin_extent(root, ref->bytenr,
3725 						 ref->num_bytes, 1);
3726 			btrfs_free_delayed_extent_op(head->extent_op);
3727 			delayed_refs->num_heads--;
3728 			if (list_empty(&head->cluster))
3729 				delayed_refs->num_heads_ready--;
3730 			list_del_init(&head->cluster);
3731 		}
3732 
3733 		ref->in_tree = 0;
3734 		rb_erase(&ref->rb_node, &delayed_refs->root);
3735 		delayed_refs->num_entries--;
3736 		if (head)
3737 			mutex_unlock(&head->mutex);
3738 		spin_unlock(&delayed_refs->lock);
3739 		btrfs_put_delayed_ref(ref);
3740 
3741 		cond_resched();
3742 		spin_lock(&delayed_refs->lock);
3743 	}
3744 
3745 	spin_unlock(&delayed_refs->lock);
3746 
3747 	return ret;
3748 }
3749 
3750 static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t)
3751 {
3752 	struct btrfs_pending_snapshot *snapshot;
3753 	struct list_head splice;
3754 
3755 	INIT_LIST_HEAD(&splice);
3756 
3757 	list_splice_init(&t->pending_snapshots, &splice);
3758 
3759 	while (!list_empty(&splice)) {
3760 		snapshot = list_entry(splice.next,
3761 				      struct btrfs_pending_snapshot,
3762 				      list);
3763 		snapshot->error = -ECANCELED;
3764 		list_del_init(&snapshot->list);
3765 	}
3766 }
3767 
3768 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3769 {
3770 	struct btrfs_inode *btrfs_inode;
3771 	struct list_head splice;
3772 
3773 	INIT_LIST_HEAD(&splice);
3774 
3775 	spin_lock(&root->fs_info->delalloc_lock);
3776 	list_splice_init(&root->fs_info->delalloc_inodes, &splice);
3777 
3778 	while (!list_empty(&splice)) {
3779 		btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3780 				    delalloc_inodes);
3781 
3782 		list_del_init(&btrfs_inode->delalloc_inodes);
3783 		clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3784 			  &btrfs_inode->runtime_flags);
3785 
3786 		btrfs_invalidate_inodes(btrfs_inode->root);
3787 	}
3788 
3789 	spin_unlock(&root->fs_info->delalloc_lock);
3790 }
3791 
3792 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3793 					struct extent_io_tree *dirty_pages,
3794 					int mark)
3795 {
3796 	int ret;
3797 	struct extent_buffer *eb;
3798 	u64 start = 0;
3799 	u64 end;
3800 
3801 	while (1) {
3802 		ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3803 					    mark, NULL);
3804 		if (ret)
3805 			break;
3806 
3807 		clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3808 		while (start <= end) {
3809 			eb = btrfs_find_tree_block(root, start,
3810 						   root->leafsize);
3811 			start += eb->len;
3812 			if (!eb)
3813 				continue;
3814 			wait_on_extent_buffer_writeback(eb);
3815 
3816 			if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3817 					       &eb->bflags))
3818 				clear_extent_buffer_dirty(eb);
3819 			free_extent_buffer_stale(eb);
3820 		}
3821 	}
3822 
3823 	return ret;
3824 }
3825 
3826 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3827 				       struct extent_io_tree *pinned_extents)
3828 {
3829 	struct extent_io_tree *unpin;
3830 	u64 start;
3831 	u64 end;
3832 	int ret;
3833 	bool loop = true;
3834 
3835 	unpin = pinned_extents;
3836 again:
3837 	while (1) {
3838 		ret = find_first_extent_bit(unpin, 0, &start, &end,
3839 					    EXTENT_DIRTY, NULL);
3840 		if (ret)
3841 			break;
3842 
3843 		/* opt_discard */
3844 		if (btrfs_test_opt(root, DISCARD))
3845 			ret = btrfs_error_discard_extent(root, start,
3846 							 end + 1 - start,
3847 							 NULL);
3848 
3849 		clear_extent_dirty(unpin, start, end, GFP_NOFS);
3850 		btrfs_error_unpin_extent_range(root, start, end);
3851 		cond_resched();
3852 	}
3853 
3854 	if (loop) {
3855 		if (unpin == &root->fs_info->freed_extents[0])
3856 			unpin = &root->fs_info->freed_extents[1];
3857 		else
3858 			unpin = &root->fs_info->freed_extents[0];
3859 		loop = false;
3860 		goto again;
3861 	}
3862 
3863 	return 0;
3864 }
3865 
3866 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
3867 				   struct btrfs_root *root)
3868 {
3869 	btrfs_destroy_delayed_refs(cur_trans, root);
3870 	btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
3871 				cur_trans->dirty_pages.dirty_bytes);
3872 
3873 	/* FIXME: cleanup wait for commit */
3874 	cur_trans->in_commit = 1;
3875 	cur_trans->blocked = 1;
3876 	wake_up(&root->fs_info->transaction_blocked_wait);
3877 
3878 	btrfs_evict_pending_snapshots(cur_trans);
3879 
3880 	cur_trans->blocked = 0;
3881 	wake_up(&root->fs_info->transaction_wait);
3882 
3883 	cur_trans->commit_done = 1;
3884 	wake_up(&cur_trans->commit_wait);
3885 
3886 	btrfs_destroy_delayed_inodes(root);
3887 	btrfs_assert_delayed_root_empty(root);
3888 
3889 	btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
3890 				     EXTENT_DIRTY);
3891 	btrfs_destroy_pinned_extent(root,
3892 				    root->fs_info->pinned_extents);
3893 
3894 	/*
3895 	memset(cur_trans, 0, sizeof(*cur_trans));
3896 	kmem_cache_free(btrfs_transaction_cachep, cur_trans);
3897 	*/
3898 }
3899 
3900 static int btrfs_cleanup_transaction(struct btrfs_root *root)
3901 {
3902 	struct btrfs_transaction *t;
3903 	LIST_HEAD(list);
3904 
3905 	mutex_lock(&root->fs_info->transaction_kthread_mutex);
3906 
3907 	spin_lock(&root->fs_info->trans_lock);
3908 	list_splice_init(&root->fs_info->trans_list, &list);
3909 	root->fs_info->trans_no_join = 1;
3910 	spin_unlock(&root->fs_info->trans_lock);
3911 
3912 	while (!list_empty(&list)) {
3913 		t = list_entry(list.next, struct btrfs_transaction, list);
3914 
3915 		btrfs_destroy_ordered_operations(t, root);
3916 
3917 		btrfs_destroy_ordered_extents(root);
3918 
3919 		btrfs_destroy_delayed_refs(t, root);
3920 
3921 		/* FIXME: cleanup wait for commit */
3922 		t->in_commit = 1;
3923 		t->blocked = 1;
3924 		smp_mb();
3925 		if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3926 			wake_up(&root->fs_info->transaction_blocked_wait);
3927 
3928 		btrfs_evict_pending_snapshots(t);
3929 
3930 		t->blocked = 0;
3931 		smp_mb();
3932 		if (waitqueue_active(&root->fs_info->transaction_wait))
3933 			wake_up(&root->fs_info->transaction_wait);
3934 
3935 		t->commit_done = 1;
3936 		smp_mb();
3937 		if (waitqueue_active(&t->commit_wait))
3938 			wake_up(&t->commit_wait);
3939 
3940 		btrfs_destroy_delayed_inodes(root);
3941 		btrfs_assert_delayed_root_empty(root);
3942 
3943 		btrfs_destroy_delalloc_inodes(root);
3944 
3945 		spin_lock(&root->fs_info->trans_lock);
3946 		root->fs_info->running_transaction = NULL;
3947 		spin_unlock(&root->fs_info->trans_lock);
3948 
3949 		btrfs_destroy_marked_extents(root, &t->dirty_pages,
3950 					     EXTENT_DIRTY);
3951 
3952 		btrfs_destroy_pinned_extent(root,
3953 					    root->fs_info->pinned_extents);
3954 
3955 		atomic_set(&t->use_count, 0);
3956 		list_del_init(&t->list);
3957 		memset(t, 0, sizeof(*t));
3958 		kmem_cache_free(btrfs_transaction_cachep, t);
3959 	}
3960 
3961 	spin_lock(&root->fs_info->trans_lock);
3962 	root->fs_info->trans_no_join = 0;
3963 	spin_unlock(&root->fs_info->trans_lock);
3964 	mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3965 
3966 	return 0;
3967 }
3968 
3969 static struct extent_io_ops btree_extent_io_ops = {
3970 	.readpage_end_io_hook = btree_readpage_end_io_hook,
3971 	.readpage_io_failed_hook = btree_io_failed_hook,
3972 	.submit_bio_hook = btree_submit_bio_hook,
3973 	/* note we're sharing with inode.c for the merge bio hook */
3974 	.merge_bio_hook = btrfs_merge_bio_hook,
3975 };
3976