xref: /openbmc/linux/fs/btrfs/disk-io.c (revision 5104d265)
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_QUOTA_TREE_OBJECTID,	.name_stem = "quota"	},
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 int offset,
1017 				 unsigned int length)
1018 {
1019 	struct extent_io_tree *tree;
1020 	tree = &BTRFS_I(page->mapping->host)->io_tree;
1021 	extent_invalidatepage(tree, page, offset);
1022 	btree_releasepage(page, GFP_NOFS);
1023 	if (PagePrivate(page)) {
1024 		printk(KERN_WARNING "btrfs warning page private not zero "
1025 		       "on page %llu\n", (unsigned long long)page_offset(page));
1026 		ClearPagePrivate(page);
1027 		set_page_private(page, 0);
1028 		page_cache_release(page);
1029 	}
1030 }
1031 
1032 static int btree_set_page_dirty(struct page *page)
1033 {
1034 #ifdef DEBUG
1035 	struct extent_buffer *eb;
1036 
1037 	BUG_ON(!PagePrivate(page));
1038 	eb = (struct extent_buffer *)page->private;
1039 	BUG_ON(!eb);
1040 	BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1041 	BUG_ON(!atomic_read(&eb->refs));
1042 	btrfs_assert_tree_locked(eb);
1043 #endif
1044 	return __set_page_dirty_nobuffers(page);
1045 }
1046 
1047 static const struct address_space_operations btree_aops = {
1048 	.readpage	= btree_readpage,
1049 	.writepages	= btree_writepages,
1050 	.releasepage	= btree_releasepage,
1051 	.invalidatepage = btree_invalidatepage,
1052 #ifdef CONFIG_MIGRATION
1053 	.migratepage	= btree_migratepage,
1054 #endif
1055 	.set_page_dirty = btree_set_page_dirty,
1056 };
1057 
1058 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1059 			 u64 parent_transid)
1060 {
1061 	struct extent_buffer *buf = NULL;
1062 	struct inode *btree_inode = root->fs_info->btree_inode;
1063 	int ret = 0;
1064 
1065 	buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1066 	if (!buf)
1067 		return 0;
1068 	read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1069 				 buf, 0, WAIT_NONE, btree_get_extent, 0);
1070 	free_extent_buffer(buf);
1071 	return ret;
1072 }
1073 
1074 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1075 			 int mirror_num, struct extent_buffer **eb)
1076 {
1077 	struct extent_buffer *buf = NULL;
1078 	struct inode *btree_inode = root->fs_info->btree_inode;
1079 	struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1080 	int ret;
1081 
1082 	buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1083 	if (!buf)
1084 		return 0;
1085 
1086 	set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1087 
1088 	ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1089 				       btree_get_extent, mirror_num);
1090 	if (ret) {
1091 		free_extent_buffer(buf);
1092 		return ret;
1093 	}
1094 
1095 	if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1096 		free_extent_buffer(buf);
1097 		return -EIO;
1098 	} else if (extent_buffer_uptodate(buf)) {
1099 		*eb = buf;
1100 	} else {
1101 		free_extent_buffer(buf);
1102 	}
1103 	return 0;
1104 }
1105 
1106 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1107 					    u64 bytenr, u32 blocksize)
1108 {
1109 	struct inode *btree_inode = root->fs_info->btree_inode;
1110 	struct extent_buffer *eb;
1111 	eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1112 				bytenr, blocksize);
1113 	return eb;
1114 }
1115 
1116 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1117 						 u64 bytenr, u32 blocksize)
1118 {
1119 	struct inode *btree_inode = root->fs_info->btree_inode;
1120 	struct extent_buffer *eb;
1121 
1122 	eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1123 				 bytenr, blocksize);
1124 	return eb;
1125 }
1126 
1127 
1128 int btrfs_write_tree_block(struct extent_buffer *buf)
1129 {
1130 	return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1131 					buf->start + buf->len - 1);
1132 }
1133 
1134 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1135 {
1136 	return filemap_fdatawait_range(buf->pages[0]->mapping,
1137 				       buf->start, buf->start + buf->len - 1);
1138 }
1139 
1140 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1141 				      u32 blocksize, u64 parent_transid)
1142 {
1143 	struct extent_buffer *buf = NULL;
1144 	int ret;
1145 
1146 	buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1147 	if (!buf)
1148 		return NULL;
1149 
1150 	ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1151 	return buf;
1152 
1153 }
1154 
1155 void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1156 		      struct extent_buffer *buf)
1157 {
1158 	struct btrfs_fs_info *fs_info = root->fs_info;
1159 
1160 	if (btrfs_header_generation(buf) ==
1161 	    fs_info->running_transaction->transid) {
1162 		btrfs_assert_tree_locked(buf);
1163 
1164 		if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1165 			__percpu_counter_add(&fs_info->dirty_metadata_bytes,
1166 					     -buf->len,
1167 					     fs_info->dirty_metadata_batch);
1168 			/* ugh, clear_extent_buffer_dirty needs to lock the page */
1169 			btrfs_set_lock_blocking(buf);
1170 			clear_extent_buffer_dirty(buf);
1171 		}
1172 	}
1173 }
1174 
1175 static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1176 			 u32 stripesize, struct btrfs_root *root,
1177 			 struct btrfs_fs_info *fs_info,
1178 			 u64 objectid)
1179 {
1180 	root->node = NULL;
1181 	root->commit_root = NULL;
1182 	root->sectorsize = sectorsize;
1183 	root->nodesize = nodesize;
1184 	root->leafsize = leafsize;
1185 	root->stripesize = stripesize;
1186 	root->ref_cows = 0;
1187 	root->track_dirty = 0;
1188 	root->in_radix = 0;
1189 	root->orphan_item_inserted = 0;
1190 	root->orphan_cleanup_state = 0;
1191 
1192 	root->objectid = objectid;
1193 	root->last_trans = 0;
1194 	root->highest_objectid = 0;
1195 	root->nr_delalloc_inodes = 0;
1196 	root->nr_ordered_extents = 0;
1197 	root->name = NULL;
1198 	root->inode_tree = RB_ROOT;
1199 	INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1200 	root->block_rsv = NULL;
1201 	root->orphan_block_rsv = NULL;
1202 
1203 	INIT_LIST_HEAD(&root->dirty_list);
1204 	INIT_LIST_HEAD(&root->root_list);
1205 	INIT_LIST_HEAD(&root->delalloc_inodes);
1206 	INIT_LIST_HEAD(&root->delalloc_root);
1207 	INIT_LIST_HEAD(&root->ordered_extents);
1208 	INIT_LIST_HEAD(&root->ordered_root);
1209 	INIT_LIST_HEAD(&root->logged_list[0]);
1210 	INIT_LIST_HEAD(&root->logged_list[1]);
1211 	spin_lock_init(&root->orphan_lock);
1212 	spin_lock_init(&root->inode_lock);
1213 	spin_lock_init(&root->delalloc_lock);
1214 	spin_lock_init(&root->ordered_extent_lock);
1215 	spin_lock_init(&root->accounting_lock);
1216 	spin_lock_init(&root->log_extents_lock[0]);
1217 	spin_lock_init(&root->log_extents_lock[1]);
1218 	mutex_init(&root->objectid_mutex);
1219 	mutex_init(&root->log_mutex);
1220 	init_waitqueue_head(&root->log_writer_wait);
1221 	init_waitqueue_head(&root->log_commit_wait[0]);
1222 	init_waitqueue_head(&root->log_commit_wait[1]);
1223 	atomic_set(&root->log_commit[0], 0);
1224 	atomic_set(&root->log_commit[1], 0);
1225 	atomic_set(&root->log_writers, 0);
1226 	atomic_set(&root->log_batch, 0);
1227 	atomic_set(&root->orphan_inodes, 0);
1228 	atomic_set(&root->refs, 1);
1229 	root->log_transid = 0;
1230 	root->last_log_commit = 0;
1231 	extent_io_tree_init(&root->dirty_log_pages,
1232 			     fs_info->btree_inode->i_mapping);
1233 
1234 	memset(&root->root_key, 0, sizeof(root->root_key));
1235 	memset(&root->root_item, 0, sizeof(root->root_item));
1236 	memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1237 	memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1238 	root->defrag_trans_start = fs_info->generation;
1239 	init_completion(&root->kobj_unregister);
1240 	root->defrag_running = 0;
1241 	root->root_key.objectid = objectid;
1242 	root->anon_dev = 0;
1243 
1244 	spin_lock_init(&root->root_item_lock);
1245 }
1246 
1247 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1248 {
1249 	struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1250 	if (root)
1251 		root->fs_info = fs_info;
1252 	return root;
1253 }
1254 
1255 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1256 				     struct btrfs_fs_info *fs_info,
1257 				     u64 objectid)
1258 {
1259 	struct extent_buffer *leaf;
1260 	struct btrfs_root *tree_root = fs_info->tree_root;
1261 	struct btrfs_root *root;
1262 	struct btrfs_key key;
1263 	int ret = 0;
1264 	u64 bytenr;
1265 	uuid_le uuid;
1266 
1267 	root = btrfs_alloc_root(fs_info);
1268 	if (!root)
1269 		return ERR_PTR(-ENOMEM);
1270 
1271 	__setup_root(tree_root->nodesize, tree_root->leafsize,
1272 		     tree_root->sectorsize, tree_root->stripesize,
1273 		     root, fs_info, objectid);
1274 	root->root_key.objectid = objectid;
1275 	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1276 	root->root_key.offset = 0;
1277 
1278 	leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1279 				      0, objectid, NULL, 0, 0, 0);
1280 	if (IS_ERR(leaf)) {
1281 		ret = PTR_ERR(leaf);
1282 		leaf = NULL;
1283 		goto fail;
1284 	}
1285 
1286 	bytenr = leaf->start;
1287 	memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1288 	btrfs_set_header_bytenr(leaf, leaf->start);
1289 	btrfs_set_header_generation(leaf, trans->transid);
1290 	btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1291 	btrfs_set_header_owner(leaf, objectid);
1292 	root->node = leaf;
1293 
1294 	write_extent_buffer(leaf, fs_info->fsid,
1295 			    (unsigned long)btrfs_header_fsid(leaf),
1296 			    BTRFS_FSID_SIZE);
1297 	write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1298 			    (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
1299 			    BTRFS_UUID_SIZE);
1300 	btrfs_mark_buffer_dirty(leaf);
1301 
1302 	root->commit_root = btrfs_root_node(root);
1303 	root->track_dirty = 1;
1304 
1305 
1306 	root->root_item.flags = 0;
1307 	root->root_item.byte_limit = 0;
1308 	btrfs_set_root_bytenr(&root->root_item, leaf->start);
1309 	btrfs_set_root_generation(&root->root_item, trans->transid);
1310 	btrfs_set_root_level(&root->root_item, 0);
1311 	btrfs_set_root_refs(&root->root_item, 1);
1312 	btrfs_set_root_used(&root->root_item, leaf->len);
1313 	btrfs_set_root_last_snapshot(&root->root_item, 0);
1314 	btrfs_set_root_dirid(&root->root_item, 0);
1315 	uuid_le_gen(&uuid);
1316 	memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1317 	root->root_item.drop_level = 0;
1318 
1319 	key.objectid = objectid;
1320 	key.type = BTRFS_ROOT_ITEM_KEY;
1321 	key.offset = 0;
1322 	ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1323 	if (ret)
1324 		goto fail;
1325 
1326 	btrfs_tree_unlock(leaf);
1327 
1328 	return root;
1329 
1330 fail:
1331 	if (leaf) {
1332 		btrfs_tree_unlock(leaf);
1333 		free_extent_buffer(leaf);
1334 	}
1335 	kfree(root);
1336 
1337 	return ERR_PTR(ret);
1338 }
1339 
1340 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1341 					 struct btrfs_fs_info *fs_info)
1342 {
1343 	struct btrfs_root *root;
1344 	struct btrfs_root *tree_root = fs_info->tree_root;
1345 	struct extent_buffer *leaf;
1346 
1347 	root = btrfs_alloc_root(fs_info);
1348 	if (!root)
1349 		return ERR_PTR(-ENOMEM);
1350 
1351 	__setup_root(tree_root->nodesize, tree_root->leafsize,
1352 		     tree_root->sectorsize, tree_root->stripesize,
1353 		     root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1354 
1355 	root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1356 	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1357 	root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1358 	/*
1359 	 * log trees do not get reference counted because they go away
1360 	 * before a real commit is actually done.  They do store pointers
1361 	 * to file data extents, and those reference counts still get
1362 	 * updated (along with back refs to the log tree).
1363 	 */
1364 	root->ref_cows = 0;
1365 
1366 	leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1367 				      BTRFS_TREE_LOG_OBJECTID, NULL,
1368 				      0, 0, 0);
1369 	if (IS_ERR(leaf)) {
1370 		kfree(root);
1371 		return ERR_CAST(leaf);
1372 	}
1373 
1374 	memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1375 	btrfs_set_header_bytenr(leaf, leaf->start);
1376 	btrfs_set_header_generation(leaf, trans->transid);
1377 	btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1378 	btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1379 	root->node = leaf;
1380 
1381 	write_extent_buffer(root->node, root->fs_info->fsid,
1382 			    (unsigned long)btrfs_header_fsid(root->node),
1383 			    BTRFS_FSID_SIZE);
1384 	btrfs_mark_buffer_dirty(root->node);
1385 	btrfs_tree_unlock(root->node);
1386 	return root;
1387 }
1388 
1389 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1390 			     struct btrfs_fs_info *fs_info)
1391 {
1392 	struct btrfs_root *log_root;
1393 
1394 	log_root = alloc_log_tree(trans, fs_info);
1395 	if (IS_ERR(log_root))
1396 		return PTR_ERR(log_root);
1397 	WARN_ON(fs_info->log_root_tree);
1398 	fs_info->log_root_tree = log_root;
1399 	return 0;
1400 }
1401 
1402 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1403 		       struct btrfs_root *root)
1404 {
1405 	struct btrfs_root *log_root;
1406 	struct btrfs_inode_item *inode_item;
1407 
1408 	log_root = alloc_log_tree(trans, root->fs_info);
1409 	if (IS_ERR(log_root))
1410 		return PTR_ERR(log_root);
1411 
1412 	log_root->last_trans = trans->transid;
1413 	log_root->root_key.offset = root->root_key.objectid;
1414 
1415 	inode_item = &log_root->root_item.inode;
1416 	inode_item->generation = cpu_to_le64(1);
1417 	inode_item->size = cpu_to_le64(3);
1418 	inode_item->nlink = cpu_to_le32(1);
1419 	inode_item->nbytes = cpu_to_le64(root->leafsize);
1420 	inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1421 
1422 	btrfs_set_root_node(&log_root->root_item, log_root->node);
1423 
1424 	WARN_ON(root->log_root);
1425 	root->log_root = log_root;
1426 	root->log_transid = 0;
1427 	root->last_log_commit = 0;
1428 	return 0;
1429 }
1430 
1431 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1432 					struct btrfs_key *key)
1433 {
1434 	struct btrfs_root *root;
1435 	struct btrfs_fs_info *fs_info = tree_root->fs_info;
1436 	struct btrfs_path *path;
1437 	u64 generation;
1438 	u32 blocksize;
1439 	int ret;
1440 
1441 	path = btrfs_alloc_path();
1442 	if (!path)
1443 		return ERR_PTR(-ENOMEM);
1444 
1445 	root = btrfs_alloc_root(fs_info);
1446 	if (!root) {
1447 		ret = -ENOMEM;
1448 		goto alloc_fail;
1449 	}
1450 
1451 	__setup_root(tree_root->nodesize, tree_root->leafsize,
1452 		     tree_root->sectorsize, tree_root->stripesize,
1453 		     root, fs_info, key->objectid);
1454 
1455 	ret = btrfs_find_root(tree_root, key, path,
1456 			      &root->root_item, &root->root_key);
1457 	if (ret) {
1458 		if (ret > 0)
1459 			ret = -ENOENT;
1460 		goto find_fail;
1461 	}
1462 
1463 	generation = btrfs_root_generation(&root->root_item);
1464 	blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1465 	root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1466 				     blocksize, generation);
1467 	if (!root->node) {
1468 		ret = -ENOMEM;
1469 		goto find_fail;
1470 	} else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1471 		ret = -EIO;
1472 		goto read_fail;
1473 	}
1474 	root->commit_root = btrfs_root_node(root);
1475 out:
1476 	btrfs_free_path(path);
1477 	return root;
1478 
1479 read_fail:
1480 	free_extent_buffer(root->node);
1481 find_fail:
1482 	kfree(root);
1483 alloc_fail:
1484 	root = ERR_PTR(ret);
1485 	goto out;
1486 }
1487 
1488 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1489 				      struct btrfs_key *location)
1490 {
1491 	struct btrfs_root *root;
1492 
1493 	root = btrfs_read_tree_root(tree_root, location);
1494 	if (IS_ERR(root))
1495 		return root;
1496 
1497 	if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1498 		root->ref_cows = 1;
1499 		btrfs_check_and_init_root_item(&root->root_item);
1500 	}
1501 
1502 	return root;
1503 }
1504 
1505 int btrfs_init_fs_root(struct btrfs_root *root)
1506 {
1507 	int ret;
1508 
1509 	root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1510 	root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1511 					GFP_NOFS);
1512 	if (!root->free_ino_pinned || !root->free_ino_ctl) {
1513 		ret = -ENOMEM;
1514 		goto fail;
1515 	}
1516 
1517 	btrfs_init_free_ino_ctl(root);
1518 	mutex_init(&root->fs_commit_mutex);
1519 	spin_lock_init(&root->cache_lock);
1520 	init_waitqueue_head(&root->cache_wait);
1521 
1522 	ret = get_anon_bdev(&root->anon_dev);
1523 	if (ret)
1524 		goto fail;
1525 	return 0;
1526 fail:
1527 	kfree(root->free_ino_ctl);
1528 	kfree(root->free_ino_pinned);
1529 	return ret;
1530 }
1531 
1532 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1533 					u64 root_id)
1534 {
1535 	struct btrfs_root *root;
1536 
1537 	spin_lock(&fs_info->fs_roots_radix_lock);
1538 	root = radix_tree_lookup(&fs_info->fs_roots_radix,
1539 				 (unsigned long)root_id);
1540 	spin_unlock(&fs_info->fs_roots_radix_lock);
1541 	return root;
1542 }
1543 
1544 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1545 			 struct btrfs_root *root)
1546 {
1547 	int ret;
1548 
1549 	ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1550 	if (ret)
1551 		return ret;
1552 
1553 	spin_lock(&fs_info->fs_roots_radix_lock);
1554 	ret = radix_tree_insert(&fs_info->fs_roots_radix,
1555 				(unsigned long)root->root_key.objectid,
1556 				root);
1557 	if (ret == 0)
1558 		root->in_radix = 1;
1559 	spin_unlock(&fs_info->fs_roots_radix_lock);
1560 	radix_tree_preload_end();
1561 
1562 	return ret;
1563 }
1564 
1565 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1566 					      struct btrfs_key *location)
1567 {
1568 	struct btrfs_root *root;
1569 	int ret;
1570 
1571 	if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1572 		return fs_info->tree_root;
1573 	if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1574 		return fs_info->extent_root;
1575 	if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1576 		return fs_info->chunk_root;
1577 	if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1578 		return fs_info->dev_root;
1579 	if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1580 		return fs_info->csum_root;
1581 	if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1582 		return fs_info->quota_root ? fs_info->quota_root :
1583 					     ERR_PTR(-ENOENT);
1584 again:
1585 	root = btrfs_lookup_fs_root(fs_info, location->objectid);
1586 	if (root)
1587 		return root;
1588 
1589 	root = btrfs_read_fs_root(fs_info->tree_root, location);
1590 	if (IS_ERR(root))
1591 		return root;
1592 
1593 	if (btrfs_root_refs(&root->root_item) == 0) {
1594 		ret = -ENOENT;
1595 		goto fail;
1596 	}
1597 
1598 	ret = btrfs_init_fs_root(root);
1599 	if (ret)
1600 		goto fail;
1601 
1602 	ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1603 	if (ret < 0)
1604 		goto fail;
1605 	if (ret == 0)
1606 		root->orphan_item_inserted = 1;
1607 
1608 	ret = btrfs_insert_fs_root(fs_info, root);
1609 	if (ret) {
1610 		if (ret == -EEXIST) {
1611 			free_fs_root(root);
1612 			goto again;
1613 		}
1614 		goto fail;
1615 	}
1616 	return root;
1617 fail:
1618 	free_fs_root(root);
1619 	return ERR_PTR(ret);
1620 }
1621 
1622 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1623 {
1624 	struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1625 	int ret = 0;
1626 	struct btrfs_device *device;
1627 	struct backing_dev_info *bdi;
1628 
1629 	rcu_read_lock();
1630 	list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1631 		if (!device->bdev)
1632 			continue;
1633 		bdi = blk_get_backing_dev_info(device->bdev);
1634 		if (bdi && bdi_congested(bdi, bdi_bits)) {
1635 			ret = 1;
1636 			break;
1637 		}
1638 	}
1639 	rcu_read_unlock();
1640 	return ret;
1641 }
1642 
1643 /*
1644  * If this fails, caller must call bdi_destroy() to get rid of the
1645  * bdi again.
1646  */
1647 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1648 {
1649 	int err;
1650 
1651 	bdi->capabilities = BDI_CAP_MAP_COPY;
1652 	err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1653 	if (err)
1654 		return err;
1655 
1656 	bdi->ra_pages	= default_backing_dev_info.ra_pages;
1657 	bdi->congested_fn	= btrfs_congested_fn;
1658 	bdi->congested_data	= info;
1659 	return 0;
1660 }
1661 
1662 /*
1663  * called by the kthread helper functions to finally call the bio end_io
1664  * functions.  This is where read checksum verification actually happens
1665  */
1666 static void end_workqueue_fn(struct btrfs_work *work)
1667 {
1668 	struct bio *bio;
1669 	struct end_io_wq *end_io_wq;
1670 	struct btrfs_fs_info *fs_info;
1671 	int error;
1672 
1673 	end_io_wq = container_of(work, struct end_io_wq, work);
1674 	bio = end_io_wq->bio;
1675 	fs_info = end_io_wq->info;
1676 
1677 	error = end_io_wq->error;
1678 	bio->bi_private = end_io_wq->private;
1679 	bio->bi_end_io = end_io_wq->end_io;
1680 	kfree(end_io_wq);
1681 	bio_endio(bio, error);
1682 }
1683 
1684 static int cleaner_kthread(void *arg)
1685 {
1686 	struct btrfs_root *root = arg;
1687 	int again;
1688 
1689 	do {
1690 		again = 0;
1691 
1692 		/* Make the cleaner go to sleep early. */
1693 		if (btrfs_need_cleaner_sleep(root))
1694 			goto sleep;
1695 
1696 		if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1697 			goto sleep;
1698 
1699 		/*
1700 		 * Avoid the problem that we change the status of the fs
1701 		 * during the above check and trylock.
1702 		 */
1703 		if (btrfs_need_cleaner_sleep(root)) {
1704 			mutex_unlock(&root->fs_info->cleaner_mutex);
1705 			goto sleep;
1706 		}
1707 
1708 		btrfs_run_delayed_iputs(root);
1709 		again = btrfs_clean_one_deleted_snapshot(root);
1710 		mutex_unlock(&root->fs_info->cleaner_mutex);
1711 
1712 		/*
1713 		 * The defragger has dealt with the R/O remount and umount,
1714 		 * needn't do anything special here.
1715 		 */
1716 		btrfs_run_defrag_inodes(root->fs_info);
1717 sleep:
1718 		if (!try_to_freeze() && !again) {
1719 			set_current_state(TASK_INTERRUPTIBLE);
1720 			if (!kthread_should_stop())
1721 				schedule();
1722 			__set_current_state(TASK_RUNNING);
1723 		}
1724 	} while (!kthread_should_stop());
1725 	return 0;
1726 }
1727 
1728 static int transaction_kthread(void *arg)
1729 {
1730 	struct btrfs_root *root = arg;
1731 	struct btrfs_trans_handle *trans;
1732 	struct btrfs_transaction *cur;
1733 	u64 transid;
1734 	unsigned long now;
1735 	unsigned long delay;
1736 	bool cannot_commit;
1737 
1738 	do {
1739 		cannot_commit = false;
1740 		delay = HZ * 30;
1741 		mutex_lock(&root->fs_info->transaction_kthread_mutex);
1742 
1743 		spin_lock(&root->fs_info->trans_lock);
1744 		cur = root->fs_info->running_transaction;
1745 		if (!cur) {
1746 			spin_unlock(&root->fs_info->trans_lock);
1747 			goto sleep;
1748 		}
1749 
1750 		now = get_seconds();
1751 		if (cur->state < TRANS_STATE_BLOCKED &&
1752 		    (now < cur->start_time || now - cur->start_time < 30)) {
1753 			spin_unlock(&root->fs_info->trans_lock);
1754 			delay = HZ * 5;
1755 			goto sleep;
1756 		}
1757 		transid = cur->transid;
1758 		spin_unlock(&root->fs_info->trans_lock);
1759 
1760 		/* If the file system is aborted, this will always fail. */
1761 		trans = btrfs_attach_transaction(root);
1762 		if (IS_ERR(trans)) {
1763 			if (PTR_ERR(trans) != -ENOENT)
1764 				cannot_commit = true;
1765 			goto sleep;
1766 		}
1767 		if (transid == trans->transid) {
1768 			btrfs_commit_transaction(trans, root);
1769 		} else {
1770 			btrfs_end_transaction(trans, root);
1771 		}
1772 sleep:
1773 		wake_up_process(root->fs_info->cleaner_kthread);
1774 		mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1775 
1776 		if (!try_to_freeze()) {
1777 			set_current_state(TASK_INTERRUPTIBLE);
1778 			if (!kthread_should_stop() &&
1779 			    (!btrfs_transaction_blocked(root->fs_info) ||
1780 			     cannot_commit))
1781 				schedule_timeout(delay);
1782 			__set_current_state(TASK_RUNNING);
1783 		}
1784 	} while (!kthread_should_stop());
1785 	return 0;
1786 }
1787 
1788 /*
1789  * this will find the highest generation in the array of
1790  * root backups.  The index of the highest array is returned,
1791  * or -1 if we can't find anything.
1792  *
1793  * We check to make sure the array is valid by comparing the
1794  * generation of the latest  root in the array with the generation
1795  * in the super block.  If they don't match we pitch it.
1796  */
1797 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1798 {
1799 	u64 cur;
1800 	int newest_index = -1;
1801 	struct btrfs_root_backup *root_backup;
1802 	int i;
1803 
1804 	for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1805 		root_backup = info->super_copy->super_roots + i;
1806 		cur = btrfs_backup_tree_root_gen(root_backup);
1807 		if (cur == newest_gen)
1808 			newest_index = i;
1809 	}
1810 
1811 	/* check to see if we actually wrapped around */
1812 	if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1813 		root_backup = info->super_copy->super_roots;
1814 		cur = btrfs_backup_tree_root_gen(root_backup);
1815 		if (cur == newest_gen)
1816 			newest_index = 0;
1817 	}
1818 	return newest_index;
1819 }
1820 
1821 
1822 /*
1823  * find the oldest backup so we know where to store new entries
1824  * in the backup array.  This will set the backup_root_index
1825  * field in the fs_info struct
1826  */
1827 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1828 				     u64 newest_gen)
1829 {
1830 	int newest_index = -1;
1831 
1832 	newest_index = find_newest_super_backup(info, newest_gen);
1833 	/* if there was garbage in there, just move along */
1834 	if (newest_index == -1) {
1835 		info->backup_root_index = 0;
1836 	} else {
1837 		info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1838 	}
1839 }
1840 
1841 /*
1842  * copy all the root pointers into the super backup array.
1843  * this will bump the backup pointer by one when it is
1844  * done
1845  */
1846 static void backup_super_roots(struct btrfs_fs_info *info)
1847 {
1848 	int next_backup;
1849 	struct btrfs_root_backup *root_backup;
1850 	int last_backup;
1851 
1852 	next_backup = info->backup_root_index;
1853 	last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1854 		BTRFS_NUM_BACKUP_ROOTS;
1855 
1856 	/*
1857 	 * just overwrite the last backup if we're at the same generation
1858 	 * this happens only at umount
1859 	 */
1860 	root_backup = info->super_for_commit->super_roots + last_backup;
1861 	if (btrfs_backup_tree_root_gen(root_backup) ==
1862 	    btrfs_header_generation(info->tree_root->node))
1863 		next_backup = last_backup;
1864 
1865 	root_backup = info->super_for_commit->super_roots + next_backup;
1866 
1867 	/*
1868 	 * make sure all of our padding and empty slots get zero filled
1869 	 * regardless of which ones we use today
1870 	 */
1871 	memset(root_backup, 0, sizeof(*root_backup));
1872 
1873 	info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1874 
1875 	btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1876 	btrfs_set_backup_tree_root_gen(root_backup,
1877 			       btrfs_header_generation(info->tree_root->node));
1878 
1879 	btrfs_set_backup_tree_root_level(root_backup,
1880 			       btrfs_header_level(info->tree_root->node));
1881 
1882 	btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1883 	btrfs_set_backup_chunk_root_gen(root_backup,
1884 			       btrfs_header_generation(info->chunk_root->node));
1885 	btrfs_set_backup_chunk_root_level(root_backup,
1886 			       btrfs_header_level(info->chunk_root->node));
1887 
1888 	btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1889 	btrfs_set_backup_extent_root_gen(root_backup,
1890 			       btrfs_header_generation(info->extent_root->node));
1891 	btrfs_set_backup_extent_root_level(root_backup,
1892 			       btrfs_header_level(info->extent_root->node));
1893 
1894 	/*
1895 	 * we might commit during log recovery, which happens before we set
1896 	 * the fs_root.  Make sure it is valid before we fill it in.
1897 	 */
1898 	if (info->fs_root && info->fs_root->node) {
1899 		btrfs_set_backup_fs_root(root_backup,
1900 					 info->fs_root->node->start);
1901 		btrfs_set_backup_fs_root_gen(root_backup,
1902 			       btrfs_header_generation(info->fs_root->node));
1903 		btrfs_set_backup_fs_root_level(root_backup,
1904 			       btrfs_header_level(info->fs_root->node));
1905 	}
1906 
1907 	btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1908 	btrfs_set_backup_dev_root_gen(root_backup,
1909 			       btrfs_header_generation(info->dev_root->node));
1910 	btrfs_set_backup_dev_root_level(root_backup,
1911 				       btrfs_header_level(info->dev_root->node));
1912 
1913 	btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1914 	btrfs_set_backup_csum_root_gen(root_backup,
1915 			       btrfs_header_generation(info->csum_root->node));
1916 	btrfs_set_backup_csum_root_level(root_backup,
1917 			       btrfs_header_level(info->csum_root->node));
1918 
1919 	btrfs_set_backup_total_bytes(root_backup,
1920 			     btrfs_super_total_bytes(info->super_copy));
1921 	btrfs_set_backup_bytes_used(root_backup,
1922 			     btrfs_super_bytes_used(info->super_copy));
1923 	btrfs_set_backup_num_devices(root_backup,
1924 			     btrfs_super_num_devices(info->super_copy));
1925 
1926 	/*
1927 	 * if we don't copy this out to the super_copy, it won't get remembered
1928 	 * for the next commit
1929 	 */
1930 	memcpy(&info->super_copy->super_roots,
1931 	       &info->super_for_commit->super_roots,
1932 	       sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1933 }
1934 
1935 /*
1936  * this copies info out of the root backup array and back into
1937  * the in-memory super block.  It is meant to help iterate through
1938  * the array, so you send it the number of backups you've already
1939  * tried and the last backup index you used.
1940  *
1941  * this returns -1 when it has tried all the backups
1942  */
1943 static noinline int next_root_backup(struct btrfs_fs_info *info,
1944 				     struct btrfs_super_block *super,
1945 				     int *num_backups_tried, int *backup_index)
1946 {
1947 	struct btrfs_root_backup *root_backup;
1948 	int newest = *backup_index;
1949 
1950 	if (*num_backups_tried == 0) {
1951 		u64 gen = btrfs_super_generation(super);
1952 
1953 		newest = find_newest_super_backup(info, gen);
1954 		if (newest == -1)
1955 			return -1;
1956 
1957 		*backup_index = newest;
1958 		*num_backups_tried = 1;
1959 	} else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1960 		/* we've tried all the backups, all done */
1961 		return -1;
1962 	} else {
1963 		/* jump to the next oldest backup */
1964 		newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1965 			BTRFS_NUM_BACKUP_ROOTS;
1966 		*backup_index = newest;
1967 		*num_backups_tried += 1;
1968 	}
1969 	root_backup = super->super_roots + newest;
1970 
1971 	btrfs_set_super_generation(super,
1972 				   btrfs_backup_tree_root_gen(root_backup));
1973 	btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1974 	btrfs_set_super_root_level(super,
1975 				   btrfs_backup_tree_root_level(root_backup));
1976 	btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1977 
1978 	/*
1979 	 * fixme: the total bytes and num_devices need to match or we should
1980 	 * need a fsck
1981 	 */
1982 	btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1983 	btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1984 	return 0;
1985 }
1986 
1987 /* helper to cleanup workers */
1988 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1989 {
1990 	btrfs_stop_workers(&fs_info->generic_worker);
1991 	btrfs_stop_workers(&fs_info->fixup_workers);
1992 	btrfs_stop_workers(&fs_info->delalloc_workers);
1993 	btrfs_stop_workers(&fs_info->workers);
1994 	btrfs_stop_workers(&fs_info->endio_workers);
1995 	btrfs_stop_workers(&fs_info->endio_meta_workers);
1996 	btrfs_stop_workers(&fs_info->endio_raid56_workers);
1997 	btrfs_stop_workers(&fs_info->rmw_workers);
1998 	btrfs_stop_workers(&fs_info->endio_meta_write_workers);
1999 	btrfs_stop_workers(&fs_info->endio_write_workers);
2000 	btrfs_stop_workers(&fs_info->endio_freespace_worker);
2001 	btrfs_stop_workers(&fs_info->submit_workers);
2002 	btrfs_stop_workers(&fs_info->delayed_workers);
2003 	btrfs_stop_workers(&fs_info->caching_workers);
2004 	btrfs_stop_workers(&fs_info->readahead_workers);
2005 	btrfs_stop_workers(&fs_info->flush_workers);
2006 	btrfs_stop_workers(&fs_info->qgroup_rescan_workers);
2007 }
2008 
2009 /* helper to cleanup tree roots */
2010 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2011 {
2012 	free_extent_buffer(info->tree_root->node);
2013 	free_extent_buffer(info->tree_root->commit_root);
2014 	info->tree_root->node = NULL;
2015 	info->tree_root->commit_root = NULL;
2016 
2017 	if (info->dev_root) {
2018 		free_extent_buffer(info->dev_root->node);
2019 		free_extent_buffer(info->dev_root->commit_root);
2020 		info->dev_root->node = NULL;
2021 		info->dev_root->commit_root = NULL;
2022 	}
2023 	if (info->extent_root) {
2024 		free_extent_buffer(info->extent_root->node);
2025 		free_extent_buffer(info->extent_root->commit_root);
2026 		info->extent_root->node = NULL;
2027 		info->extent_root->commit_root = NULL;
2028 	}
2029 	if (info->csum_root) {
2030 		free_extent_buffer(info->csum_root->node);
2031 		free_extent_buffer(info->csum_root->commit_root);
2032 		info->csum_root->node = NULL;
2033 		info->csum_root->commit_root = NULL;
2034 	}
2035 	if (info->quota_root) {
2036 		free_extent_buffer(info->quota_root->node);
2037 		free_extent_buffer(info->quota_root->commit_root);
2038 		info->quota_root->node = NULL;
2039 		info->quota_root->commit_root = NULL;
2040 	}
2041 	if (chunk_root) {
2042 		free_extent_buffer(info->chunk_root->node);
2043 		free_extent_buffer(info->chunk_root->commit_root);
2044 		info->chunk_root->node = NULL;
2045 		info->chunk_root->commit_root = NULL;
2046 	}
2047 }
2048 
2049 static void del_fs_roots(struct btrfs_fs_info *fs_info)
2050 {
2051 	int ret;
2052 	struct btrfs_root *gang[8];
2053 	int i;
2054 
2055 	while (!list_empty(&fs_info->dead_roots)) {
2056 		gang[0] = list_entry(fs_info->dead_roots.next,
2057 				     struct btrfs_root, root_list);
2058 		list_del(&gang[0]->root_list);
2059 
2060 		if (gang[0]->in_radix) {
2061 			btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2062 		} else {
2063 			free_extent_buffer(gang[0]->node);
2064 			free_extent_buffer(gang[0]->commit_root);
2065 			btrfs_put_fs_root(gang[0]);
2066 		}
2067 	}
2068 
2069 	while (1) {
2070 		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2071 					     (void **)gang, 0,
2072 					     ARRAY_SIZE(gang));
2073 		if (!ret)
2074 			break;
2075 		for (i = 0; i < ret; i++)
2076 			btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2077 	}
2078 }
2079 
2080 int open_ctree(struct super_block *sb,
2081 	       struct btrfs_fs_devices *fs_devices,
2082 	       char *options)
2083 {
2084 	u32 sectorsize;
2085 	u32 nodesize;
2086 	u32 leafsize;
2087 	u32 blocksize;
2088 	u32 stripesize;
2089 	u64 generation;
2090 	u64 features;
2091 	struct btrfs_key location;
2092 	struct buffer_head *bh;
2093 	struct btrfs_super_block *disk_super;
2094 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2095 	struct btrfs_root *tree_root;
2096 	struct btrfs_root *extent_root;
2097 	struct btrfs_root *csum_root;
2098 	struct btrfs_root *chunk_root;
2099 	struct btrfs_root *dev_root;
2100 	struct btrfs_root *quota_root;
2101 	struct btrfs_root *log_tree_root;
2102 	int ret;
2103 	int err = -EINVAL;
2104 	int num_backups_tried = 0;
2105 	int backup_index = 0;
2106 
2107 	tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2108 	chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2109 	if (!tree_root || !chunk_root) {
2110 		err = -ENOMEM;
2111 		goto fail;
2112 	}
2113 
2114 	ret = init_srcu_struct(&fs_info->subvol_srcu);
2115 	if (ret) {
2116 		err = ret;
2117 		goto fail;
2118 	}
2119 
2120 	ret = setup_bdi(fs_info, &fs_info->bdi);
2121 	if (ret) {
2122 		err = ret;
2123 		goto fail_srcu;
2124 	}
2125 
2126 	ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2127 	if (ret) {
2128 		err = ret;
2129 		goto fail_bdi;
2130 	}
2131 	fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2132 					(1 + ilog2(nr_cpu_ids));
2133 
2134 	ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2135 	if (ret) {
2136 		err = ret;
2137 		goto fail_dirty_metadata_bytes;
2138 	}
2139 
2140 	fs_info->btree_inode = new_inode(sb);
2141 	if (!fs_info->btree_inode) {
2142 		err = -ENOMEM;
2143 		goto fail_delalloc_bytes;
2144 	}
2145 
2146 	mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2147 
2148 	INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2149 	INIT_LIST_HEAD(&fs_info->trans_list);
2150 	INIT_LIST_HEAD(&fs_info->dead_roots);
2151 	INIT_LIST_HEAD(&fs_info->delayed_iputs);
2152 	INIT_LIST_HEAD(&fs_info->delalloc_roots);
2153 	INIT_LIST_HEAD(&fs_info->caching_block_groups);
2154 	spin_lock_init(&fs_info->delalloc_root_lock);
2155 	spin_lock_init(&fs_info->trans_lock);
2156 	spin_lock_init(&fs_info->fs_roots_radix_lock);
2157 	spin_lock_init(&fs_info->delayed_iput_lock);
2158 	spin_lock_init(&fs_info->defrag_inodes_lock);
2159 	spin_lock_init(&fs_info->free_chunk_lock);
2160 	spin_lock_init(&fs_info->tree_mod_seq_lock);
2161 	spin_lock_init(&fs_info->super_lock);
2162 	rwlock_init(&fs_info->tree_mod_log_lock);
2163 	mutex_init(&fs_info->reloc_mutex);
2164 	seqlock_init(&fs_info->profiles_lock);
2165 
2166 	init_completion(&fs_info->kobj_unregister);
2167 	INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2168 	INIT_LIST_HEAD(&fs_info->space_info);
2169 	INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2170 	btrfs_mapping_init(&fs_info->mapping_tree);
2171 	btrfs_init_block_rsv(&fs_info->global_block_rsv,
2172 			     BTRFS_BLOCK_RSV_GLOBAL);
2173 	btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2174 			     BTRFS_BLOCK_RSV_DELALLOC);
2175 	btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2176 	btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2177 	btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2178 	btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2179 			     BTRFS_BLOCK_RSV_DELOPS);
2180 	atomic_set(&fs_info->nr_async_submits, 0);
2181 	atomic_set(&fs_info->async_delalloc_pages, 0);
2182 	atomic_set(&fs_info->async_submit_draining, 0);
2183 	atomic_set(&fs_info->nr_async_bios, 0);
2184 	atomic_set(&fs_info->defrag_running, 0);
2185 	atomic64_set(&fs_info->tree_mod_seq, 0);
2186 	fs_info->sb = sb;
2187 	fs_info->max_inline = 8192 * 1024;
2188 	fs_info->metadata_ratio = 0;
2189 	fs_info->defrag_inodes = RB_ROOT;
2190 	fs_info->free_chunk_space = 0;
2191 	fs_info->tree_mod_log = RB_ROOT;
2192 
2193 	/* readahead state */
2194 	INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2195 	spin_lock_init(&fs_info->reada_lock);
2196 
2197 	fs_info->thread_pool_size = min_t(unsigned long,
2198 					  num_online_cpus() + 2, 8);
2199 
2200 	INIT_LIST_HEAD(&fs_info->ordered_roots);
2201 	spin_lock_init(&fs_info->ordered_root_lock);
2202 	fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2203 					GFP_NOFS);
2204 	if (!fs_info->delayed_root) {
2205 		err = -ENOMEM;
2206 		goto fail_iput;
2207 	}
2208 	btrfs_init_delayed_root(fs_info->delayed_root);
2209 
2210 	mutex_init(&fs_info->scrub_lock);
2211 	atomic_set(&fs_info->scrubs_running, 0);
2212 	atomic_set(&fs_info->scrub_pause_req, 0);
2213 	atomic_set(&fs_info->scrubs_paused, 0);
2214 	atomic_set(&fs_info->scrub_cancel_req, 0);
2215 	init_waitqueue_head(&fs_info->scrub_pause_wait);
2216 	init_rwsem(&fs_info->scrub_super_lock);
2217 	fs_info->scrub_workers_refcnt = 0;
2218 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2219 	fs_info->check_integrity_print_mask = 0;
2220 #endif
2221 
2222 	spin_lock_init(&fs_info->balance_lock);
2223 	mutex_init(&fs_info->balance_mutex);
2224 	atomic_set(&fs_info->balance_running, 0);
2225 	atomic_set(&fs_info->balance_pause_req, 0);
2226 	atomic_set(&fs_info->balance_cancel_req, 0);
2227 	fs_info->balance_ctl = NULL;
2228 	init_waitqueue_head(&fs_info->balance_wait_q);
2229 
2230 	sb->s_blocksize = 4096;
2231 	sb->s_blocksize_bits = blksize_bits(4096);
2232 	sb->s_bdi = &fs_info->bdi;
2233 
2234 	fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2235 	set_nlink(fs_info->btree_inode, 1);
2236 	/*
2237 	 * we set the i_size on the btree inode to the max possible int.
2238 	 * the real end of the address space is determined by all of
2239 	 * the devices in the system
2240 	 */
2241 	fs_info->btree_inode->i_size = OFFSET_MAX;
2242 	fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2243 	fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2244 
2245 	RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2246 	extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2247 			     fs_info->btree_inode->i_mapping);
2248 	BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2249 	extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2250 
2251 	BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2252 
2253 	BTRFS_I(fs_info->btree_inode)->root = tree_root;
2254 	memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2255 	       sizeof(struct btrfs_key));
2256 	set_bit(BTRFS_INODE_DUMMY,
2257 		&BTRFS_I(fs_info->btree_inode)->runtime_flags);
2258 	insert_inode_hash(fs_info->btree_inode);
2259 
2260 	spin_lock_init(&fs_info->block_group_cache_lock);
2261 	fs_info->block_group_cache_tree = RB_ROOT;
2262 	fs_info->first_logical_byte = (u64)-1;
2263 
2264 	extent_io_tree_init(&fs_info->freed_extents[0],
2265 			     fs_info->btree_inode->i_mapping);
2266 	extent_io_tree_init(&fs_info->freed_extents[1],
2267 			     fs_info->btree_inode->i_mapping);
2268 	fs_info->pinned_extents = &fs_info->freed_extents[0];
2269 	fs_info->do_barriers = 1;
2270 
2271 
2272 	mutex_init(&fs_info->ordered_operations_mutex);
2273 	mutex_init(&fs_info->tree_log_mutex);
2274 	mutex_init(&fs_info->chunk_mutex);
2275 	mutex_init(&fs_info->transaction_kthread_mutex);
2276 	mutex_init(&fs_info->cleaner_mutex);
2277 	mutex_init(&fs_info->volume_mutex);
2278 	init_rwsem(&fs_info->extent_commit_sem);
2279 	init_rwsem(&fs_info->cleanup_work_sem);
2280 	init_rwsem(&fs_info->subvol_sem);
2281 	fs_info->dev_replace.lock_owner = 0;
2282 	atomic_set(&fs_info->dev_replace.nesting_level, 0);
2283 	mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2284 	mutex_init(&fs_info->dev_replace.lock_management_lock);
2285 	mutex_init(&fs_info->dev_replace.lock);
2286 
2287 	spin_lock_init(&fs_info->qgroup_lock);
2288 	mutex_init(&fs_info->qgroup_ioctl_lock);
2289 	fs_info->qgroup_tree = RB_ROOT;
2290 	INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2291 	fs_info->qgroup_seq = 1;
2292 	fs_info->quota_enabled = 0;
2293 	fs_info->pending_quota_state = 0;
2294 	fs_info->qgroup_ulist = NULL;
2295 	mutex_init(&fs_info->qgroup_rescan_lock);
2296 
2297 	btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2298 	btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2299 
2300 	init_waitqueue_head(&fs_info->transaction_throttle);
2301 	init_waitqueue_head(&fs_info->transaction_wait);
2302 	init_waitqueue_head(&fs_info->transaction_blocked_wait);
2303 	init_waitqueue_head(&fs_info->async_submit_wait);
2304 
2305 	ret = btrfs_alloc_stripe_hash_table(fs_info);
2306 	if (ret) {
2307 		err = ret;
2308 		goto fail_alloc;
2309 	}
2310 
2311 	__setup_root(4096, 4096, 4096, 4096, tree_root,
2312 		     fs_info, BTRFS_ROOT_TREE_OBJECTID);
2313 
2314 	invalidate_bdev(fs_devices->latest_bdev);
2315 
2316 	/*
2317 	 * Read super block and check the signature bytes only
2318 	 */
2319 	bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2320 	if (!bh) {
2321 		err = -EINVAL;
2322 		goto fail_alloc;
2323 	}
2324 
2325 	/*
2326 	 * We want to check superblock checksum, the type is stored inside.
2327 	 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2328 	 */
2329 	if (btrfs_check_super_csum(bh->b_data)) {
2330 		printk(KERN_ERR "btrfs: superblock checksum mismatch\n");
2331 		err = -EINVAL;
2332 		goto fail_alloc;
2333 	}
2334 
2335 	/*
2336 	 * super_copy is zeroed at allocation time and we never touch the
2337 	 * following bytes up to INFO_SIZE, the checksum is calculated from
2338 	 * the whole block of INFO_SIZE
2339 	 */
2340 	memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2341 	memcpy(fs_info->super_for_commit, fs_info->super_copy,
2342 	       sizeof(*fs_info->super_for_commit));
2343 	brelse(bh);
2344 
2345 	memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2346 
2347 	ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2348 	if (ret) {
2349 		printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2350 		err = -EINVAL;
2351 		goto fail_alloc;
2352 	}
2353 
2354 	disk_super = fs_info->super_copy;
2355 	if (!btrfs_super_root(disk_super))
2356 		goto fail_alloc;
2357 
2358 	/* check FS state, whether FS is broken. */
2359 	if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2360 		set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2361 
2362 	/*
2363 	 * run through our array of backup supers and setup
2364 	 * our ring pointer to the oldest one
2365 	 */
2366 	generation = btrfs_super_generation(disk_super);
2367 	find_oldest_super_backup(fs_info, generation);
2368 
2369 	/*
2370 	 * In the long term, we'll store the compression type in the super
2371 	 * block, and it'll be used for per file compression control.
2372 	 */
2373 	fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2374 
2375 	ret = btrfs_parse_options(tree_root, options);
2376 	if (ret) {
2377 		err = ret;
2378 		goto fail_alloc;
2379 	}
2380 
2381 	features = btrfs_super_incompat_flags(disk_super) &
2382 		~BTRFS_FEATURE_INCOMPAT_SUPP;
2383 	if (features) {
2384 		printk(KERN_ERR "BTRFS: couldn't mount because of "
2385 		       "unsupported optional features (%Lx).\n",
2386 		       (unsigned long long)features);
2387 		err = -EINVAL;
2388 		goto fail_alloc;
2389 	}
2390 
2391 	if (btrfs_super_leafsize(disk_super) !=
2392 	    btrfs_super_nodesize(disk_super)) {
2393 		printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2394 		       "blocksizes don't match.  node %d leaf %d\n",
2395 		       btrfs_super_nodesize(disk_super),
2396 		       btrfs_super_leafsize(disk_super));
2397 		err = -EINVAL;
2398 		goto fail_alloc;
2399 	}
2400 	if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2401 		printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2402 		       "blocksize (%d) was too large\n",
2403 		       btrfs_super_leafsize(disk_super));
2404 		err = -EINVAL;
2405 		goto fail_alloc;
2406 	}
2407 
2408 	features = btrfs_super_incompat_flags(disk_super);
2409 	features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2410 	if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2411 		features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2412 
2413 	if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2414 		printk(KERN_ERR "btrfs: has skinny extents\n");
2415 
2416 	/*
2417 	 * flag our filesystem as having big metadata blocks if
2418 	 * they are bigger than the page size
2419 	 */
2420 	if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2421 		if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2422 			printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2423 		features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2424 	}
2425 
2426 	nodesize = btrfs_super_nodesize(disk_super);
2427 	leafsize = btrfs_super_leafsize(disk_super);
2428 	sectorsize = btrfs_super_sectorsize(disk_super);
2429 	stripesize = btrfs_super_stripesize(disk_super);
2430 	fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
2431 	fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2432 
2433 	/*
2434 	 * mixed block groups end up with duplicate but slightly offset
2435 	 * extent buffers for the same range.  It leads to corruptions
2436 	 */
2437 	if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2438 	    (sectorsize != leafsize)) {
2439 		printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2440 				"are not allowed for mixed block groups on %s\n",
2441 				sb->s_id);
2442 		goto fail_alloc;
2443 	}
2444 
2445 	/*
2446 	 * Needn't use the lock because there is no other task which will
2447 	 * update the flag.
2448 	 */
2449 	btrfs_set_super_incompat_flags(disk_super, features);
2450 
2451 	features = btrfs_super_compat_ro_flags(disk_super) &
2452 		~BTRFS_FEATURE_COMPAT_RO_SUPP;
2453 	if (!(sb->s_flags & MS_RDONLY) && features) {
2454 		printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2455 		       "unsupported option features (%Lx).\n",
2456 		       (unsigned long long)features);
2457 		err = -EINVAL;
2458 		goto fail_alloc;
2459 	}
2460 
2461 	btrfs_init_workers(&fs_info->generic_worker,
2462 			   "genwork", 1, NULL);
2463 
2464 	btrfs_init_workers(&fs_info->workers, "worker",
2465 			   fs_info->thread_pool_size,
2466 			   &fs_info->generic_worker);
2467 
2468 	btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2469 			   fs_info->thread_pool_size,
2470 			   &fs_info->generic_worker);
2471 
2472 	btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc",
2473 			   fs_info->thread_pool_size,
2474 			   &fs_info->generic_worker);
2475 
2476 	btrfs_init_workers(&fs_info->submit_workers, "submit",
2477 			   min_t(u64, fs_devices->num_devices,
2478 			   fs_info->thread_pool_size),
2479 			   &fs_info->generic_worker);
2480 
2481 	btrfs_init_workers(&fs_info->caching_workers, "cache",
2482 			   2, &fs_info->generic_worker);
2483 
2484 	/* a higher idle thresh on the submit workers makes it much more
2485 	 * likely that bios will be send down in a sane order to the
2486 	 * devices
2487 	 */
2488 	fs_info->submit_workers.idle_thresh = 64;
2489 
2490 	fs_info->workers.idle_thresh = 16;
2491 	fs_info->workers.ordered = 1;
2492 
2493 	fs_info->delalloc_workers.idle_thresh = 2;
2494 	fs_info->delalloc_workers.ordered = 1;
2495 
2496 	btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2497 			   &fs_info->generic_worker);
2498 	btrfs_init_workers(&fs_info->endio_workers, "endio",
2499 			   fs_info->thread_pool_size,
2500 			   &fs_info->generic_worker);
2501 	btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2502 			   fs_info->thread_pool_size,
2503 			   &fs_info->generic_worker);
2504 	btrfs_init_workers(&fs_info->endio_meta_write_workers,
2505 			   "endio-meta-write", fs_info->thread_pool_size,
2506 			   &fs_info->generic_worker);
2507 	btrfs_init_workers(&fs_info->endio_raid56_workers,
2508 			   "endio-raid56", fs_info->thread_pool_size,
2509 			   &fs_info->generic_worker);
2510 	btrfs_init_workers(&fs_info->rmw_workers,
2511 			   "rmw", fs_info->thread_pool_size,
2512 			   &fs_info->generic_worker);
2513 	btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2514 			   fs_info->thread_pool_size,
2515 			   &fs_info->generic_worker);
2516 	btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2517 			   1, &fs_info->generic_worker);
2518 	btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2519 			   fs_info->thread_pool_size,
2520 			   &fs_info->generic_worker);
2521 	btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2522 			   fs_info->thread_pool_size,
2523 			   &fs_info->generic_worker);
2524 	btrfs_init_workers(&fs_info->qgroup_rescan_workers, "qgroup-rescan", 1,
2525 			   &fs_info->generic_worker);
2526 
2527 	/*
2528 	 * endios are largely parallel and should have a very
2529 	 * low idle thresh
2530 	 */
2531 	fs_info->endio_workers.idle_thresh = 4;
2532 	fs_info->endio_meta_workers.idle_thresh = 4;
2533 	fs_info->endio_raid56_workers.idle_thresh = 4;
2534 	fs_info->rmw_workers.idle_thresh = 2;
2535 
2536 	fs_info->endio_write_workers.idle_thresh = 2;
2537 	fs_info->endio_meta_write_workers.idle_thresh = 2;
2538 	fs_info->readahead_workers.idle_thresh = 2;
2539 
2540 	/*
2541 	 * btrfs_start_workers can really only fail because of ENOMEM so just
2542 	 * return -ENOMEM if any of these fail.
2543 	 */
2544 	ret = btrfs_start_workers(&fs_info->workers);
2545 	ret |= btrfs_start_workers(&fs_info->generic_worker);
2546 	ret |= btrfs_start_workers(&fs_info->submit_workers);
2547 	ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2548 	ret |= btrfs_start_workers(&fs_info->fixup_workers);
2549 	ret |= btrfs_start_workers(&fs_info->endio_workers);
2550 	ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2551 	ret |= btrfs_start_workers(&fs_info->rmw_workers);
2552 	ret |= btrfs_start_workers(&fs_info->endio_raid56_workers);
2553 	ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2554 	ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2555 	ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2556 	ret |= btrfs_start_workers(&fs_info->delayed_workers);
2557 	ret |= btrfs_start_workers(&fs_info->caching_workers);
2558 	ret |= btrfs_start_workers(&fs_info->readahead_workers);
2559 	ret |= btrfs_start_workers(&fs_info->flush_workers);
2560 	ret |= btrfs_start_workers(&fs_info->qgroup_rescan_workers);
2561 	if (ret) {
2562 		err = -ENOMEM;
2563 		goto fail_sb_buffer;
2564 	}
2565 
2566 	fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2567 	fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2568 				    4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2569 
2570 	tree_root->nodesize = nodesize;
2571 	tree_root->leafsize = leafsize;
2572 	tree_root->sectorsize = sectorsize;
2573 	tree_root->stripesize = stripesize;
2574 
2575 	sb->s_blocksize = sectorsize;
2576 	sb->s_blocksize_bits = blksize_bits(sectorsize);
2577 
2578 	if (disk_super->magic != cpu_to_le64(BTRFS_MAGIC)) {
2579 		printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2580 		goto fail_sb_buffer;
2581 	}
2582 
2583 	if (sectorsize != PAGE_SIZE) {
2584 		printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2585 		       "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2586 		goto fail_sb_buffer;
2587 	}
2588 
2589 	mutex_lock(&fs_info->chunk_mutex);
2590 	ret = btrfs_read_sys_array(tree_root);
2591 	mutex_unlock(&fs_info->chunk_mutex);
2592 	if (ret) {
2593 		printk(KERN_WARNING "btrfs: failed to read the system "
2594 		       "array on %s\n", sb->s_id);
2595 		goto fail_sb_buffer;
2596 	}
2597 
2598 	blocksize = btrfs_level_size(tree_root,
2599 				     btrfs_super_chunk_root_level(disk_super));
2600 	generation = btrfs_super_chunk_root_generation(disk_super);
2601 
2602 	__setup_root(nodesize, leafsize, sectorsize, stripesize,
2603 		     chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2604 
2605 	chunk_root->node = read_tree_block(chunk_root,
2606 					   btrfs_super_chunk_root(disk_super),
2607 					   blocksize, generation);
2608 	if (!chunk_root->node ||
2609 	    !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2610 		printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2611 		       sb->s_id);
2612 		goto fail_tree_roots;
2613 	}
2614 	btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2615 	chunk_root->commit_root = btrfs_root_node(chunk_root);
2616 
2617 	read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2618 	   (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2619 	   BTRFS_UUID_SIZE);
2620 
2621 	ret = btrfs_read_chunk_tree(chunk_root);
2622 	if (ret) {
2623 		printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2624 		       sb->s_id);
2625 		goto fail_tree_roots;
2626 	}
2627 
2628 	/*
2629 	 * keep the device that is marked to be the target device for the
2630 	 * dev_replace procedure
2631 	 */
2632 	btrfs_close_extra_devices(fs_info, fs_devices, 0);
2633 
2634 	if (!fs_devices->latest_bdev) {
2635 		printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2636 		       sb->s_id);
2637 		goto fail_tree_roots;
2638 	}
2639 
2640 retry_root_backup:
2641 	blocksize = btrfs_level_size(tree_root,
2642 				     btrfs_super_root_level(disk_super));
2643 	generation = btrfs_super_generation(disk_super);
2644 
2645 	tree_root->node = read_tree_block(tree_root,
2646 					  btrfs_super_root(disk_super),
2647 					  blocksize, generation);
2648 	if (!tree_root->node ||
2649 	    !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2650 		printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2651 		       sb->s_id);
2652 
2653 		goto recovery_tree_root;
2654 	}
2655 
2656 	btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2657 	tree_root->commit_root = btrfs_root_node(tree_root);
2658 
2659 	location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2660 	location.type = BTRFS_ROOT_ITEM_KEY;
2661 	location.offset = 0;
2662 
2663 	extent_root = btrfs_read_tree_root(tree_root, &location);
2664 	if (IS_ERR(extent_root)) {
2665 		ret = PTR_ERR(extent_root);
2666 		goto recovery_tree_root;
2667 	}
2668 	extent_root->track_dirty = 1;
2669 	fs_info->extent_root = extent_root;
2670 
2671 	location.objectid = BTRFS_DEV_TREE_OBJECTID;
2672 	dev_root = btrfs_read_tree_root(tree_root, &location);
2673 	if (IS_ERR(dev_root)) {
2674 		ret = PTR_ERR(dev_root);
2675 		goto recovery_tree_root;
2676 	}
2677 	dev_root->track_dirty = 1;
2678 	fs_info->dev_root = dev_root;
2679 	btrfs_init_devices_late(fs_info);
2680 
2681 	location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2682 	csum_root = btrfs_read_tree_root(tree_root, &location);
2683 	if (IS_ERR(csum_root)) {
2684 		ret = PTR_ERR(csum_root);
2685 		goto recovery_tree_root;
2686 	}
2687 	csum_root->track_dirty = 1;
2688 	fs_info->csum_root = csum_root;
2689 
2690 	location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2691 	quota_root = btrfs_read_tree_root(tree_root, &location);
2692 	if (!IS_ERR(quota_root)) {
2693 		quota_root->track_dirty = 1;
2694 		fs_info->quota_enabled = 1;
2695 		fs_info->pending_quota_state = 1;
2696 		fs_info->quota_root = quota_root;
2697 	}
2698 
2699 	fs_info->generation = generation;
2700 	fs_info->last_trans_committed = generation;
2701 
2702 	ret = btrfs_recover_balance(fs_info);
2703 	if (ret) {
2704 		printk(KERN_WARNING "btrfs: failed to recover balance\n");
2705 		goto fail_block_groups;
2706 	}
2707 
2708 	ret = btrfs_init_dev_stats(fs_info);
2709 	if (ret) {
2710 		printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
2711 		       ret);
2712 		goto fail_block_groups;
2713 	}
2714 
2715 	ret = btrfs_init_dev_replace(fs_info);
2716 	if (ret) {
2717 		pr_err("btrfs: failed to init dev_replace: %d\n", ret);
2718 		goto fail_block_groups;
2719 	}
2720 
2721 	btrfs_close_extra_devices(fs_info, fs_devices, 1);
2722 
2723 	ret = btrfs_init_space_info(fs_info);
2724 	if (ret) {
2725 		printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2726 		goto fail_block_groups;
2727 	}
2728 
2729 	ret = btrfs_read_block_groups(extent_root);
2730 	if (ret) {
2731 		printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2732 		goto fail_block_groups;
2733 	}
2734 	fs_info->num_tolerated_disk_barrier_failures =
2735 		btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2736 	if (fs_info->fs_devices->missing_devices >
2737 	     fs_info->num_tolerated_disk_barrier_failures &&
2738 	    !(sb->s_flags & MS_RDONLY)) {
2739 		printk(KERN_WARNING
2740 		       "Btrfs: too many missing devices, writeable mount is not allowed\n");
2741 		goto fail_block_groups;
2742 	}
2743 
2744 	fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2745 					       "btrfs-cleaner");
2746 	if (IS_ERR(fs_info->cleaner_kthread))
2747 		goto fail_block_groups;
2748 
2749 	fs_info->transaction_kthread = kthread_run(transaction_kthread,
2750 						   tree_root,
2751 						   "btrfs-transaction");
2752 	if (IS_ERR(fs_info->transaction_kthread))
2753 		goto fail_cleaner;
2754 
2755 	if (!btrfs_test_opt(tree_root, SSD) &&
2756 	    !btrfs_test_opt(tree_root, NOSSD) &&
2757 	    !fs_info->fs_devices->rotating) {
2758 		printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2759 		       "mode\n");
2760 		btrfs_set_opt(fs_info->mount_opt, SSD);
2761 	}
2762 
2763 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2764 	if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2765 		ret = btrfsic_mount(tree_root, fs_devices,
2766 				    btrfs_test_opt(tree_root,
2767 					CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2768 				    1 : 0,
2769 				    fs_info->check_integrity_print_mask);
2770 		if (ret)
2771 			printk(KERN_WARNING "btrfs: failed to initialize"
2772 			       " integrity check module %s\n", sb->s_id);
2773 	}
2774 #endif
2775 	ret = btrfs_read_qgroup_config(fs_info);
2776 	if (ret)
2777 		goto fail_trans_kthread;
2778 
2779 	/* do not make disk changes in broken FS */
2780 	if (btrfs_super_log_root(disk_super) != 0) {
2781 		u64 bytenr = btrfs_super_log_root(disk_super);
2782 
2783 		if (fs_devices->rw_devices == 0) {
2784 			printk(KERN_WARNING "Btrfs log replay required "
2785 			       "on RO media\n");
2786 			err = -EIO;
2787 			goto fail_qgroup;
2788 		}
2789 		blocksize =
2790 		     btrfs_level_size(tree_root,
2791 				      btrfs_super_log_root_level(disk_super));
2792 
2793 		log_tree_root = btrfs_alloc_root(fs_info);
2794 		if (!log_tree_root) {
2795 			err = -ENOMEM;
2796 			goto fail_qgroup;
2797 		}
2798 
2799 		__setup_root(nodesize, leafsize, sectorsize, stripesize,
2800 			     log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2801 
2802 		log_tree_root->node = read_tree_block(tree_root, bytenr,
2803 						      blocksize,
2804 						      generation + 1);
2805 		if (!log_tree_root->node ||
2806 		    !extent_buffer_uptodate(log_tree_root->node)) {
2807 			printk(KERN_ERR "btrfs: failed to read log tree\n");
2808 			free_extent_buffer(log_tree_root->node);
2809 			kfree(log_tree_root);
2810 			goto fail_trans_kthread;
2811 		}
2812 		/* returns with log_tree_root freed on success */
2813 		ret = btrfs_recover_log_trees(log_tree_root);
2814 		if (ret) {
2815 			btrfs_error(tree_root->fs_info, ret,
2816 				    "Failed to recover log tree");
2817 			free_extent_buffer(log_tree_root->node);
2818 			kfree(log_tree_root);
2819 			goto fail_trans_kthread;
2820 		}
2821 
2822 		if (sb->s_flags & MS_RDONLY) {
2823 			ret = btrfs_commit_super(tree_root);
2824 			if (ret)
2825 				goto fail_trans_kthread;
2826 		}
2827 	}
2828 
2829 	ret = btrfs_find_orphan_roots(tree_root);
2830 	if (ret)
2831 		goto fail_trans_kthread;
2832 
2833 	if (!(sb->s_flags & MS_RDONLY)) {
2834 		ret = btrfs_cleanup_fs_roots(fs_info);
2835 		if (ret)
2836 			goto fail_trans_kthread;
2837 
2838 		ret = btrfs_recover_relocation(tree_root);
2839 		if (ret < 0) {
2840 			printk(KERN_WARNING
2841 			       "btrfs: failed to recover relocation\n");
2842 			err = -EINVAL;
2843 			goto fail_qgroup;
2844 		}
2845 	}
2846 
2847 	location.objectid = BTRFS_FS_TREE_OBJECTID;
2848 	location.type = BTRFS_ROOT_ITEM_KEY;
2849 	location.offset = 0;
2850 
2851 	fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2852 	if (IS_ERR(fs_info->fs_root)) {
2853 		err = PTR_ERR(fs_info->fs_root);
2854 		goto fail_qgroup;
2855 	}
2856 
2857 	if (sb->s_flags & MS_RDONLY)
2858 		return 0;
2859 
2860 	down_read(&fs_info->cleanup_work_sem);
2861 	if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2862 	    (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2863 		up_read(&fs_info->cleanup_work_sem);
2864 		close_ctree(tree_root);
2865 		return ret;
2866 	}
2867 	up_read(&fs_info->cleanup_work_sem);
2868 
2869 	ret = btrfs_resume_balance_async(fs_info);
2870 	if (ret) {
2871 		printk(KERN_WARNING "btrfs: failed to resume balance\n");
2872 		close_ctree(tree_root);
2873 		return ret;
2874 	}
2875 
2876 	ret = btrfs_resume_dev_replace_async(fs_info);
2877 	if (ret) {
2878 		pr_warn("btrfs: failed to resume dev_replace\n");
2879 		close_ctree(tree_root);
2880 		return ret;
2881 	}
2882 
2883 	btrfs_qgroup_rescan_resume(fs_info);
2884 
2885 	return 0;
2886 
2887 fail_qgroup:
2888 	btrfs_free_qgroup_config(fs_info);
2889 fail_trans_kthread:
2890 	kthread_stop(fs_info->transaction_kthread);
2891 	btrfs_cleanup_transaction(fs_info->tree_root);
2892 	del_fs_roots(fs_info);
2893 fail_cleaner:
2894 	kthread_stop(fs_info->cleaner_kthread);
2895 
2896 	/*
2897 	 * make sure we're done with the btree inode before we stop our
2898 	 * kthreads
2899 	 */
2900 	filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2901 
2902 fail_block_groups:
2903 	btrfs_put_block_group_cache(fs_info);
2904 	btrfs_free_block_groups(fs_info);
2905 
2906 fail_tree_roots:
2907 	free_root_pointers(fs_info, 1);
2908 	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2909 
2910 fail_sb_buffer:
2911 	btrfs_stop_all_workers(fs_info);
2912 fail_alloc:
2913 fail_iput:
2914 	btrfs_mapping_tree_free(&fs_info->mapping_tree);
2915 
2916 	iput(fs_info->btree_inode);
2917 fail_delalloc_bytes:
2918 	percpu_counter_destroy(&fs_info->delalloc_bytes);
2919 fail_dirty_metadata_bytes:
2920 	percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
2921 fail_bdi:
2922 	bdi_destroy(&fs_info->bdi);
2923 fail_srcu:
2924 	cleanup_srcu_struct(&fs_info->subvol_srcu);
2925 fail:
2926 	btrfs_free_stripe_hash_table(fs_info);
2927 	btrfs_close_devices(fs_info->fs_devices);
2928 	return err;
2929 
2930 recovery_tree_root:
2931 	if (!btrfs_test_opt(tree_root, RECOVERY))
2932 		goto fail_tree_roots;
2933 
2934 	free_root_pointers(fs_info, 0);
2935 
2936 	/* don't use the log in recovery mode, it won't be valid */
2937 	btrfs_set_super_log_root(disk_super, 0);
2938 
2939 	/* we can't trust the free space cache either */
2940 	btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2941 
2942 	ret = next_root_backup(fs_info, fs_info->super_copy,
2943 			       &num_backups_tried, &backup_index);
2944 	if (ret == -1)
2945 		goto fail_block_groups;
2946 	goto retry_root_backup;
2947 }
2948 
2949 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2950 {
2951 	if (uptodate) {
2952 		set_buffer_uptodate(bh);
2953 	} else {
2954 		struct btrfs_device *device = (struct btrfs_device *)
2955 			bh->b_private;
2956 
2957 		printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
2958 					  "I/O error on %s\n",
2959 					  rcu_str_deref(device->name));
2960 		/* note, we dont' set_buffer_write_io_error because we have
2961 		 * our own ways of dealing with the IO errors
2962 		 */
2963 		clear_buffer_uptodate(bh);
2964 		btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
2965 	}
2966 	unlock_buffer(bh);
2967 	put_bh(bh);
2968 }
2969 
2970 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2971 {
2972 	struct buffer_head *bh;
2973 	struct buffer_head *latest = NULL;
2974 	struct btrfs_super_block *super;
2975 	int i;
2976 	u64 transid = 0;
2977 	u64 bytenr;
2978 
2979 	/* we would like to check all the supers, but that would make
2980 	 * a btrfs mount succeed after a mkfs from a different FS.
2981 	 * So, we need to add a special mount option to scan for
2982 	 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2983 	 */
2984 	for (i = 0; i < 1; i++) {
2985 		bytenr = btrfs_sb_offset(i);
2986 		if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2987 			break;
2988 		bh = __bread(bdev, bytenr / 4096, 4096);
2989 		if (!bh)
2990 			continue;
2991 
2992 		super = (struct btrfs_super_block *)bh->b_data;
2993 		if (btrfs_super_bytenr(super) != bytenr ||
2994 		    super->magic != cpu_to_le64(BTRFS_MAGIC)) {
2995 			brelse(bh);
2996 			continue;
2997 		}
2998 
2999 		if (!latest || btrfs_super_generation(super) > transid) {
3000 			brelse(latest);
3001 			latest = bh;
3002 			transid = btrfs_super_generation(super);
3003 		} else {
3004 			brelse(bh);
3005 		}
3006 	}
3007 	return latest;
3008 }
3009 
3010 /*
3011  * this should be called twice, once with wait == 0 and
3012  * once with wait == 1.  When wait == 0 is done, all the buffer heads
3013  * we write are pinned.
3014  *
3015  * They are released when wait == 1 is done.
3016  * max_mirrors must be the same for both runs, and it indicates how
3017  * many supers on this one device should be written.
3018  *
3019  * max_mirrors == 0 means to write them all.
3020  */
3021 static int write_dev_supers(struct btrfs_device *device,
3022 			    struct btrfs_super_block *sb,
3023 			    int do_barriers, int wait, int max_mirrors)
3024 {
3025 	struct buffer_head *bh;
3026 	int i;
3027 	int ret;
3028 	int errors = 0;
3029 	u32 crc;
3030 	u64 bytenr;
3031 
3032 	if (max_mirrors == 0)
3033 		max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3034 
3035 	for (i = 0; i < max_mirrors; i++) {
3036 		bytenr = btrfs_sb_offset(i);
3037 		if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3038 			break;
3039 
3040 		if (wait) {
3041 			bh = __find_get_block(device->bdev, bytenr / 4096,
3042 					      BTRFS_SUPER_INFO_SIZE);
3043 			if (!bh) {
3044 				errors++;
3045 				continue;
3046 			}
3047 			wait_on_buffer(bh);
3048 			if (!buffer_uptodate(bh))
3049 				errors++;
3050 
3051 			/* drop our reference */
3052 			brelse(bh);
3053 
3054 			/* drop the reference from the wait == 0 run */
3055 			brelse(bh);
3056 			continue;
3057 		} else {
3058 			btrfs_set_super_bytenr(sb, bytenr);
3059 
3060 			crc = ~(u32)0;
3061 			crc = btrfs_csum_data((char *)sb +
3062 					      BTRFS_CSUM_SIZE, crc,
3063 					      BTRFS_SUPER_INFO_SIZE -
3064 					      BTRFS_CSUM_SIZE);
3065 			btrfs_csum_final(crc, sb->csum);
3066 
3067 			/*
3068 			 * one reference for us, and we leave it for the
3069 			 * caller
3070 			 */
3071 			bh = __getblk(device->bdev, bytenr / 4096,
3072 				      BTRFS_SUPER_INFO_SIZE);
3073 			if (!bh) {
3074 				printk(KERN_ERR "btrfs: couldn't get super "
3075 				       "buffer head for bytenr %Lu\n", bytenr);
3076 				errors++;
3077 				continue;
3078 			}
3079 
3080 			memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3081 
3082 			/* one reference for submit_bh */
3083 			get_bh(bh);
3084 
3085 			set_buffer_uptodate(bh);
3086 			lock_buffer(bh);
3087 			bh->b_end_io = btrfs_end_buffer_write_sync;
3088 			bh->b_private = device;
3089 		}
3090 
3091 		/*
3092 		 * we fua the first super.  The others we allow
3093 		 * to go down lazy.
3094 		 */
3095 		ret = btrfsic_submit_bh(WRITE_FUA, bh);
3096 		if (ret)
3097 			errors++;
3098 	}
3099 	return errors < i ? 0 : -1;
3100 }
3101 
3102 /*
3103  * endio for the write_dev_flush, this will wake anyone waiting
3104  * for the barrier when it is done
3105  */
3106 static void btrfs_end_empty_barrier(struct bio *bio, int err)
3107 {
3108 	if (err) {
3109 		if (err == -EOPNOTSUPP)
3110 			set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3111 		clear_bit(BIO_UPTODATE, &bio->bi_flags);
3112 	}
3113 	if (bio->bi_private)
3114 		complete(bio->bi_private);
3115 	bio_put(bio);
3116 }
3117 
3118 /*
3119  * trigger flushes for one the devices.  If you pass wait == 0, the flushes are
3120  * sent down.  With wait == 1, it waits for the previous flush.
3121  *
3122  * any device where the flush fails with eopnotsupp are flagged as not-barrier
3123  * capable
3124  */
3125 static int write_dev_flush(struct btrfs_device *device, int wait)
3126 {
3127 	struct bio *bio;
3128 	int ret = 0;
3129 
3130 	if (device->nobarriers)
3131 		return 0;
3132 
3133 	if (wait) {
3134 		bio = device->flush_bio;
3135 		if (!bio)
3136 			return 0;
3137 
3138 		wait_for_completion(&device->flush_wait);
3139 
3140 		if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3141 			printk_in_rcu("btrfs: disabling barriers on dev %s\n",
3142 				      rcu_str_deref(device->name));
3143 			device->nobarriers = 1;
3144 		} else if (!bio_flagged(bio, BIO_UPTODATE)) {
3145 			ret = -EIO;
3146 			btrfs_dev_stat_inc_and_print(device,
3147 				BTRFS_DEV_STAT_FLUSH_ERRS);
3148 		}
3149 
3150 		/* drop the reference from the wait == 0 run */
3151 		bio_put(bio);
3152 		device->flush_bio = NULL;
3153 
3154 		return ret;
3155 	}
3156 
3157 	/*
3158 	 * one reference for us, and we leave it for the
3159 	 * caller
3160 	 */
3161 	device->flush_bio = NULL;
3162 	bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3163 	if (!bio)
3164 		return -ENOMEM;
3165 
3166 	bio->bi_end_io = btrfs_end_empty_barrier;
3167 	bio->bi_bdev = device->bdev;
3168 	init_completion(&device->flush_wait);
3169 	bio->bi_private = &device->flush_wait;
3170 	device->flush_bio = bio;
3171 
3172 	bio_get(bio);
3173 	btrfsic_submit_bio(WRITE_FLUSH, bio);
3174 
3175 	return 0;
3176 }
3177 
3178 /*
3179  * send an empty flush down to each device in parallel,
3180  * then wait for them
3181  */
3182 static int barrier_all_devices(struct btrfs_fs_info *info)
3183 {
3184 	struct list_head *head;
3185 	struct btrfs_device *dev;
3186 	int errors_send = 0;
3187 	int errors_wait = 0;
3188 	int ret;
3189 
3190 	/* send down all the barriers */
3191 	head = &info->fs_devices->devices;
3192 	list_for_each_entry_rcu(dev, head, dev_list) {
3193 		if (!dev->bdev) {
3194 			errors_send++;
3195 			continue;
3196 		}
3197 		if (!dev->in_fs_metadata || !dev->writeable)
3198 			continue;
3199 
3200 		ret = write_dev_flush(dev, 0);
3201 		if (ret)
3202 			errors_send++;
3203 	}
3204 
3205 	/* wait for all the barriers */
3206 	list_for_each_entry_rcu(dev, head, dev_list) {
3207 		if (!dev->bdev) {
3208 			errors_wait++;
3209 			continue;
3210 		}
3211 		if (!dev->in_fs_metadata || !dev->writeable)
3212 			continue;
3213 
3214 		ret = write_dev_flush(dev, 1);
3215 		if (ret)
3216 			errors_wait++;
3217 	}
3218 	if (errors_send > info->num_tolerated_disk_barrier_failures ||
3219 	    errors_wait > info->num_tolerated_disk_barrier_failures)
3220 		return -EIO;
3221 	return 0;
3222 }
3223 
3224 int btrfs_calc_num_tolerated_disk_barrier_failures(
3225 	struct btrfs_fs_info *fs_info)
3226 {
3227 	struct btrfs_ioctl_space_info space;
3228 	struct btrfs_space_info *sinfo;
3229 	u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3230 		       BTRFS_BLOCK_GROUP_SYSTEM,
3231 		       BTRFS_BLOCK_GROUP_METADATA,
3232 		       BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3233 	int num_types = 4;
3234 	int i;
3235 	int c;
3236 	int num_tolerated_disk_barrier_failures =
3237 		(int)fs_info->fs_devices->num_devices;
3238 
3239 	for (i = 0; i < num_types; i++) {
3240 		struct btrfs_space_info *tmp;
3241 
3242 		sinfo = NULL;
3243 		rcu_read_lock();
3244 		list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3245 			if (tmp->flags == types[i]) {
3246 				sinfo = tmp;
3247 				break;
3248 			}
3249 		}
3250 		rcu_read_unlock();
3251 
3252 		if (!sinfo)
3253 			continue;
3254 
3255 		down_read(&sinfo->groups_sem);
3256 		for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3257 			if (!list_empty(&sinfo->block_groups[c])) {
3258 				u64 flags;
3259 
3260 				btrfs_get_block_group_info(
3261 					&sinfo->block_groups[c], &space);
3262 				if (space.total_bytes == 0 ||
3263 				    space.used_bytes == 0)
3264 					continue;
3265 				flags = space.flags;
3266 				/*
3267 				 * return
3268 				 * 0: if dup, single or RAID0 is configured for
3269 				 *    any of metadata, system or data, else
3270 				 * 1: if RAID5 is configured, or if RAID1 or
3271 				 *    RAID10 is configured and only two mirrors
3272 				 *    are used, else
3273 				 * 2: if RAID6 is configured, else
3274 				 * num_mirrors - 1: if RAID1 or RAID10 is
3275 				 *                  configured and more than
3276 				 *                  2 mirrors are used.
3277 				 */
3278 				if (num_tolerated_disk_barrier_failures > 0 &&
3279 				    ((flags & (BTRFS_BLOCK_GROUP_DUP |
3280 					       BTRFS_BLOCK_GROUP_RAID0)) ||
3281 				     ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3282 				      == 0)))
3283 					num_tolerated_disk_barrier_failures = 0;
3284 				else if (num_tolerated_disk_barrier_failures > 1) {
3285 					if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3286 					    BTRFS_BLOCK_GROUP_RAID5 |
3287 					    BTRFS_BLOCK_GROUP_RAID10)) {
3288 						num_tolerated_disk_barrier_failures = 1;
3289 					} else if (flags &
3290 						   BTRFS_BLOCK_GROUP_RAID6) {
3291 						num_tolerated_disk_barrier_failures = 2;
3292 					}
3293 				}
3294 			}
3295 		}
3296 		up_read(&sinfo->groups_sem);
3297 	}
3298 
3299 	return num_tolerated_disk_barrier_failures;
3300 }
3301 
3302 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3303 {
3304 	struct list_head *head;
3305 	struct btrfs_device *dev;
3306 	struct btrfs_super_block *sb;
3307 	struct btrfs_dev_item *dev_item;
3308 	int ret;
3309 	int do_barriers;
3310 	int max_errors;
3311 	int total_errors = 0;
3312 	u64 flags;
3313 
3314 	max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3315 	do_barriers = !btrfs_test_opt(root, NOBARRIER);
3316 	backup_super_roots(root->fs_info);
3317 
3318 	sb = root->fs_info->super_for_commit;
3319 	dev_item = &sb->dev_item;
3320 
3321 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3322 	head = &root->fs_info->fs_devices->devices;
3323 
3324 	if (do_barriers) {
3325 		ret = barrier_all_devices(root->fs_info);
3326 		if (ret) {
3327 			mutex_unlock(
3328 				&root->fs_info->fs_devices->device_list_mutex);
3329 			btrfs_error(root->fs_info, ret,
3330 				    "errors while submitting device barriers.");
3331 			return ret;
3332 		}
3333 	}
3334 
3335 	list_for_each_entry_rcu(dev, head, dev_list) {
3336 		if (!dev->bdev) {
3337 			total_errors++;
3338 			continue;
3339 		}
3340 		if (!dev->in_fs_metadata || !dev->writeable)
3341 			continue;
3342 
3343 		btrfs_set_stack_device_generation(dev_item, 0);
3344 		btrfs_set_stack_device_type(dev_item, dev->type);
3345 		btrfs_set_stack_device_id(dev_item, dev->devid);
3346 		btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3347 		btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3348 		btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3349 		btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3350 		btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3351 		memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3352 		memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3353 
3354 		flags = btrfs_super_flags(sb);
3355 		btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3356 
3357 		ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3358 		if (ret)
3359 			total_errors++;
3360 	}
3361 	if (total_errors > max_errors) {
3362 		printk(KERN_ERR "btrfs: %d errors while writing supers\n",
3363 		       total_errors);
3364 
3365 		/* This shouldn't happen. FUA is masked off if unsupported */
3366 		BUG();
3367 	}
3368 
3369 	total_errors = 0;
3370 	list_for_each_entry_rcu(dev, head, dev_list) {
3371 		if (!dev->bdev)
3372 			continue;
3373 		if (!dev->in_fs_metadata || !dev->writeable)
3374 			continue;
3375 
3376 		ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3377 		if (ret)
3378 			total_errors++;
3379 	}
3380 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3381 	if (total_errors > max_errors) {
3382 		btrfs_error(root->fs_info, -EIO,
3383 			    "%d errors while writing supers", total_errors);
3384 		return -EIO;
3385 	}
3386 	return 0;
3387 }
3388 
3389 int write_ctree_super(struct btrfs_trans_handle *trans,
3390 		      struct btrfs_root *root, int max_mirrors)
3391 {
3392 	int ret;
3393 
3394 	ret = write_all_supers(root, max_mirrors);
3395 	return ret;
3396 }
3397 
3398 /* Drop a fs root from the radix tree and free it. */
3399 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3400 				  struct btrfs_root *root)
3401 {
3402 	spin_lock(&fs_info->fs_roots_radix_lock);
3403 	radix_tree_delete(&fs_info->fs_roots_radix,
3404 			  (unsigned long)root->root_key.objectid);
3405 	spin_unlock(&fs_info->fs_roots_radix_lock);
3406 
3407 	if (btrfs_root_refs(&root->root_item) == 0)
3408 		synchronize_srcu(&fs_info->subvol_srcu);
3409 
3410 	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3411 		btrfs_free_log(NULL, root);
3412 		btrfs_free_log_root_tree(NULL, fs_info);
3413 	}
3414 
3415 	__btrfs_remove_free_space_cache(root->free_ino_pinned);
3416 	__btrfs_remove_free_space_cache(root->free_ino_ctl);
3417 	free_fs_root(root);
3418 }
3419 
3420 static void free_fs_root(struct btrfs_root *root)
3421 {
3422 	iput(root->cache_inode);
3423 	WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3424 	if (root->anon_dev)
3425 		free_anon_bdev(root->anon_dev);
3426 	free_extent_buffer(root->node);
3427 	free_extent_buffer(root->commit_root);
3428 	kfree(root->free_ino_ctl);
3429 	kfree(root->free_ino_pinned);
3430 	kfree(root->name);
3431 	btrfs_put_fs_root(root);
3432 }
3433 
3434 void btrfs_free_fs_root(struct btrfs_root *root)
3435 {
3436 	free_fs_root(root);
3437 }
3438 
3439 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3440 {
3441 	u64 root_objectid = 0;
3442 	struct btrfs_root *gang[8];
3443 	int i;
3444 	int ret;
3445 
3446 	while (1) {
3447 		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3448 					     (void **)gang, root_objectid,
3449 					     ARRAY_SIZE(gang));
3450 		if (!ret)
3451 			break;
3452 
3453 		root_objectid = gang[ret - 1]->root_key.objectid + 1;
3454 		for (i = 0; i < ret; i++) {
3455 			int err;
3456 
3457 			root_objectid = gang[i]->root_key.objectid;
3458 			err = btrfs_orphan_cleanup(gang[i]);
3459 			if (err)
3460 				return err;
3461 		}
3462 		root_objectid++;
3463 	}
3464 	return 0;
3465 }
3466 
3467 int btrfs_commit_super(struct btrfs_root *root)
3468 {
3469 	struct btrfs_trans_handle *trans;
3470 	int ret;
3471 
3472 	mutex_lock(&root->fs_info->cleaner_mutex);
3473 	btrfs_run_delayed_iputs(root);
3474 	mutex_unlock(&root->fs_info->cleaner_mutex);
3475 	wake_up_process(root->fs_info->cleaner_kthread);
3476 
3477 	/* wait until ongoing cleanup work done */
3478 	down_write(&root->fs_info->cleanup_work_sem);
3479 	up_write(&root->fs_info->cleanup_work_sem);
3480 
3481 	trans = btrfs_join_transaction(root);
3482 	if (IS_ERR(trans))
3483 		return PTR_ERR(trans);
3484 	ret = btrfs_commit_transaction(trans, root);
3485 	if (ret)
3486 		return ret;
3487 	/* run commit again to drop the original snapshot */
3488 	trans = btrfs_join_transaction(root);
3489 	if (IS_ERR(trans))
3490 		return PTR_ERR(trans);
3491 	ret = btrfs_commit_transaction(trans, root);
3492 	if (ret)
3493 		return ret;
3494 	ret = btrfs_write_and_wait_transaction(NULL, root);
3495 	if (ret) {
3496 		btrfs_error(root->fs_info, ret,
3497 			    "Failed to sync btree inode to disk.");
3498 		return ret;
3499 	}
3500 
3501 	ret = write_ctree_super(NULL, root, 0);
3502 	return ret;
3503 }
3504 
3505 int close_ctree(struct btrfs_root *root)
3506 {
3507 	struct btrfs_fs_info *fs_info = root->fs_info;
3508 	int ret;
3509 
3510 	fs_info->closing = 1;
3511 	smp_mb();
3512 
3513 	/* pause restriper - we want to resume on mount */
3514 	btrfs_pause_balance(fs_info);
3515 
3516 	btrfs_dev_replace_suspend_for_unmount(fs_info);
3517 
3518 	btrfs_scrub_cancel(fs_info);
3519 
3520 	/* wait for any defraggers to finish */
3521 	wait_event(fs_info->transaction_wait,
3522 		   (atomic_read(&fs_info->defrag_running) == 0));
3523 
3524 	/* clear out the rbtree of defraggable inodes */
3525 	btrfs_cleanup_defrag_inodes(fs_info);
3526 
3527 	if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3528 		ret = btrfs_commit_super(root);
3529 		if (ret)
3530 			printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3531 	}
3532 
3533 	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3534 		btrfs_error_commit_super(root);
3535 
3536 	btrfs_put_block_group_cache(fs_info);
3537 
3538 	kthread_stop(fs_info->transaction_kthread);
3539 	kthread_stop(fs_info->cleaner_kthread);
3540 
3541 	fs_info->closing = 2;
3542 	smp_mb();
3543 
3544 	btrfs_free_qgroup_config(root->fs_info);
3545 
3546 	if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3547 		printk(KERN_INFO "btrfs: at unmount delalloc count %lld\n",
3548 		       percpu_counter_sum(&fs_info->delalloc_bytes));
3549 	}
3550 
3551 	btrfs_free_block_groups(fs_info);
3552 
3553 	btrfs_stop_all_workers(fs_info);
3554 
3555 	del_fs_roots(fs_info);
3556 
3557 	free_root_pointers(fs_info, 1);
3558 
3559 	iput(fs_info->btree_inode);
3560 
3561 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3562 	if (btrfs_test_opt(root, CHECK_INTEGRITY))
3563 		btrfsic_unmount(root, fs_info->fs_devices);
3564 #endif
3565 
3566 	btrfs_close_devices(fs_info->fs_devices);
3567 	btrfs_mapping_tree_free(&fs_info->mapping_tree);
3568 
3569 	percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3570 	percpu_counter_destroy(&fs_info->delalloc_bytes);
3571 	bdi_destroy(&fs_info->bdi);
3572 	cleanup_srcu_struct(&fs_info->subvol_srcu);
3573 
3574 	btrfs_free_stripe_hash_table(fs_info);
3575 
3576 	return 0;
3577 }
3578 
3579 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3580 			  int atomic)
3581 {
3582 	int ret;
3583 	struct inode *btree_inode = buf->pages[0]->mapping->host;
3584 
3585 	ret = extent_buffer_uptodate(buf);
3586 	if (!ret)
3587 		return ret;
3588 
3589 	ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3590 				    parent_transid, atomic);
3591 	if (ret == -EAGAIN)
3592 		return ret;
3593 	return !ret;
3594 }
3595 
3596 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3597 {
3598 	return set_extent_buffer_uptodate(buf);
3599 }
3600 
3601 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3602 {
3603 	struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3604 	u64 transid = btrfs_header_generation(buf);
3605 	int was_dirty;
3606 
3607 	btrfs_assert_tree_locked(buf);
3608 	if (transid != root->fs_info->generation)
3609 		WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3610 		       "found %llu running %llu\n",
3611 			(unsigned long long)buf->start,
3612 			(unsigned long long)transid,
3613 			(unsigned long long)root->fs_info->generation);
3614 	was_dirty = set_extent_buffer_dirty(buf);
3615 	if (!was_dirty)
3616 		__percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3617 				     buf->len,
3618 				     root->fs_info->dirty_metadata_batch);
3619 }
3620 
3621 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3622 					int flush_delayed)
3623 {
3624 	/*
3625 	 * looks as though older kernels can get into trouble with
3626 	 * this code, they end up stuck in balance_dirty_pages forever
3627 	 */
3628 	int ret;
3629 
3630 	if (current->flags & PF_MEMALLOC)
3631 		return;
3632 
3633 	if (flush_delayed)
3634 		btrfs_balance_delayed_items(root);
3635 
3636 	ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3637 				     BTRFS_DIRTY_METADATA_THRESH);
3638 	if (ret > 0) {
3639 		balance_dirty_pages_ratelimited(
3640 				   root->fs_info->btree_inode->i_mapping);
3641 	}
3642 	return;
3643 }
3644 
3645 void btrfs_btree_balance_dirty(struct btrfs_root *root)
3646 {
3647 	__btrfs_btree_balance_dirty(root, 1);
3648 }
3649 
3650 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3651 {
3652 	__btrfs_btree_balance_dirty(root, 0);
3653 }
3654 
3655 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3656 {
3657 	struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3658 	return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3659 }
3660 
3661 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3662 			      int read_only)
3663 {
3664 	/*
3665 	 * Placeholder for checks
3666 	 */
3667 	return 0;
3668 }
3669 
3670 static void btrfs_error_commit_super(struct btrfs_root *root)
3671 {
3672 	mutex_lock(&root->fs_info->cleaner_mutex);
3673 	btrfs_run_delayed_iputs(root);
3674 	mutex_unlock(&root->fs_info->cleaner_mutex);
3675 
3676 	down_write(&root->fs_info->cleanup_work_sem);
3677 	up_write(&root->fs_info->cleanup_work_sem);
3678 
3679 	/* cleanup FS via transaction */
3680 	btrfs_cleanup_transaction(root);
3681 }
3682 
3683 static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3684 					     struct btrfs_root *root)
3685 {
3686 	struct btrfs_inode *btrfs_inode;
3687 	struct list_head splice;
3688 
3689 	INIT_LIST_HEAD(&splice);
3690 
3691 	mutex_lock(&root->fs_info->ordered_operations_mutex);
3692 	spin_lock(&root->fs_info->ordered_root_lock);
3693 
3694 	list_splice_init(&t->ordered_operations, &splice);
3695 	while (!list_empty(&splice)) {
3696 		btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3697 					 ordered_operations);
3698 
3699 		list_del_init(&btrfs_inode->ordered_operations);
3700 		spin_unlock(&root->fs_info->ordered_root_lock);
3701 
3702 		btrfs_invalidate_inodes(btrfs_inode->root);
3703 
3704 		spin_lock(&root->fs_info->ordered_root_lock);
3705 	}
3706 
3707 	spin_unlock(&root->fs_info->ordered_root_lock);
3708 	mutex_unlock(&root->fs_info->ordered_operations_mutex);
3709 }
3710 
3711 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3712 {
3713 	struct btrfs_ordered_extent *ordered;
3714 
3715 	spin_lock(&root->ordered_extent_lock);
3716 	/*
3717 	 * This will just short circuit the ordered completion stuff which will
3718 	 * make sure the ordered extent gets properly cleaned up.
3719 	 */
3720 	list_for_each_entry(ordered, &root->ordered_extents,
3721 			    root_extent_list)
3722 		set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
3723 	spin_unlock(&root->ordered_extent_lock);
3724 }
3725 
3726 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3727 {
3728 	struct btrfs_root *root;
3729 	struct list_head splice;
3730 
3731 	INIT_LIST_HEAD(&splice);
3732 
3733 	spin_lock(&fs_info->ordered_root_lock);
3734 	list_splice_init(&fs_info->ordered_roots, &splice);
3735 	while (!list_empty(&splice)) {
3736 		root = list_first_entry(&splice, struct btrfs_root,
3737 					ordered_root);
3738 		list_del_init(&root->ordered_root);
3739 
3740 		btrfs_destroy_ordered_extents(root);
3741 
3742 		cond_resched_lock(&fs_info->ordered_root_lock);
3743 	}
3744 	spin_unlock(&fs_info->ordered_root_lock);
3745 }
3746 
3747 int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3748 			       struct btrfs_root *root)
3749 {
3750 	struct rb_node *node;
3751 	struct btrfs_delayed_ref_root *delayed_refs;
3752 	struct btrfs_delayed_ref_node *ref;
3753 	int ret = 0;
3754 
3755 	delayed_refs = &trans->delayed_refs;
3756 
3757 	spin_lock(&delayed_refs->lock);
3758 	if (delayed_refs->num_entries == 0) {
3759 		spin_unlock(&delayed_refs->lock);
3760 		printk(KERN_INFO "delayed_refs has NO entry\n");
3761 		return ret;
3762 	}
3763 
3764 	while ((node = rb_first(&delayed_refs->root)) != NULL) {
3765 		struct btrfs_delayed_ref_head *head = NULL;
3766 		bool pin_bytes = false;
3767 
3768 		ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3769 		atomic_set(&ref->refs, 1);
3770 		if (btrfs_delayed_ref_is_head(ref)) {
3771 
3772 			head = btrfs_delayed_node_to_head(ref);
3773 			if (!mutex_trylock(&head->mutex)) {
3774 				atomic_inc(&ref->refs);
3775 				spin_unlock(&delayed_refs->lock);
3776 
3777 				/* Need to wait for the delayed ref to run */
3778 				mutex_lock(&head->mutex);
3779 				mutex_unlock(&head->mutex);
3780 				btrfs_put_delayed_ref(ref);
3781 
3782 				spin_lock(&delayed_refs->lock);
3783 				continue;
3784 			}
3785 
3786 			if (head->must_insert_reserved)
3787 				pin_bytes = true;
3788 			btrfs_free_delayed_extent_op(head->extent_op);
3789 			delayed_refs->num_heads--;
3790 			if (list_empty(&head->cluster))
3791 				delayed_refs->num_heads_ready--;
3792 			list_del_init(&head->cluster);
3793 		}
3794 
3795 		ref->in_tree = 0;
3796 		rb_erase(&ref->rb_node, &delayed_refs->root);
3797 		delayed_refs->num_entries--;
3798 		spin_unlock(&delayed_refs->lock);
3799 		if (head) {
3800 			if (pin_bytes)
3801 				btrfs_pin_extent(root, ref->bytenr,
3802 						 ref->num_bytes, 1);
3803 			mutex_unlock(&head->mutex);
3804 		}
3805 		btrfs_put_delayed_ref(ref);
3806 
3807 		cond_resched();
3808 		spin_lock(&delayed_refs->lock);
3809 	}
3810 
3811 	spin_unlock(&delayed_refs->lock);
3812 
3813 	return ret;
3814 }
3815 
3816 static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t)
3817 {
3818 	struct btrfs_pending_snapshot *snapshot;
3819 	struct list_head splice;
3820 
3821 	INIT_LIST_HEAD(&splice);
3822 
3823 	list_splice_init(&t->pending_snapshots, &splice);
3824 
3825 	while (!list_empty(&splice)) {
3826 		snapshot = list_entry(splice.next,
3827 				      struct btrfs_pending_snapshot,
3828 				      list);
3829 		snapshot->error = -ECANCELED;
3830 		list_del_init(&snapshot->list);
3831 	}
3832 }
3833 
3834 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3835 {
3836 	struct btrfs_inode *btrfs_inode;
3837 	struct list_head splice;
3838 
3839 	INIT_LIST_HEAD(&splice);
3840 
3841 	spin_lock(&root->delalloc_lock);
3842 	list_splice_init(&root->delalloc_inodes, &splice);
3843 
3844 	while (!list_empty(&splice)) {
3845 		btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3846 					       delalloc_inodes);
3847 
3848 		list_del_init(&btrfs_inode->delalloc_inodes);
3849 		clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3850 			  &btrfs_inode->runtime_flags);
3851 		spin_unlock(&root->delalloc_lock);
3852 
3853 		btrfs_invalidate_inodes(btrfs_inode->root);
3854 
3855 		spin_lock(&root->delalloc_lock);
3856 	}
3857 
3858 	spin_unlock(&root->delalloc_lock);
3859 }
3860 
3861 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3862 {
3863 	struct btrfs_root *root;
3864 	struct list_head splice;
3865 
3866 	INIT_LIST_HEAD(&splice);
3867 
3868 	spin_lock(&fs_info->delalloc_root_lock);
3869 	list_splice_init(&fs_info->delalloc_roots, &splice);
3870 	while (!list_empty(&splice)) {
3871 		root = list_first_entry(&splice, struct btrfs_root,
3872 					 delalloc_root);
3873 		list_del_init(&root->delalloc_root);
3874 		root = btrfs_grab_fs_root(root);
3875 		BUG_ON(!root);
3876 		spin_unlock(&fs_info->delalloc_root_lock);
3877 
3878 		btrfs_destroy_delalloc_inodes(root);
3879 		btrfs_put_fs_root(root);
3880 
3881 		spin_lock(&fs_info->delalloc_root_lock);
3882 	}
3883 	spin_unlock(&fs_info->delalloc_root_lock);
3884 }
3885 
3886 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3887 					struct extent_io_tree *dirty_pages,
3888 					int mark)
3889 {
3890 	int ret;
3891 	struct extent_buffer *eb;
3892 	u64 start = 0;
3893 	u64 end;
3894 
3895 	while (1) {
3896 		ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3897 					    mark, NULL);
3898 		if (ret)
3899 			break;
3900 
3901 		clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3902 		while (start <= end) {
3903 			eb = btrfs_find_tree_block(root, start,
3904 						   root->leafsize);
3905 			start += root->leafsize;
3906 			if (!eb)
3907 				continue;
3908 			wait_on_extent_buffer_writeback(eb);
3909 
3910 			if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3911 					       &eb->bflags))
3912 				clear_extent_buffer_dirty(eb);
3913 			free_extent_buffer_stale(eb);
3914 		}
3915 	}
3916 
3917 	return ret;
3918 }
3919 
3920 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3921 				       struct extent_io_tree *pinned_extents)
3922 {
3923 	struct extent_io_tree *unpin;
3924 	u64 start;
3925 	u64 end;
3926 	int ret;
3927 	bool loop = true;
3928 
3929 	unpin = pinned_extents;
3930 again:
3931 	while (1) {
3932 		ret = find_first_extent_bit(unpin, 0, &start, &end,
3933 					    EXTENT_DIRTY, NULL);
3934 		if (ret)
3935 			break;
3936 
3937 		/* opt_discard */
3938 		if (btrfs_test_opt(root, DISCARD))
3939 			ret = btrfs_error_discard_extent(root, start,
3940 							 end + 1 - start,
3941 							 NULL);
3942 
3943 		clear_extent_dirty(unpin, start, end, GFP_NOFS);
3944 		btrfs_error_unpin_extent_range(root, start, end);
3945 		cond_resched();
3946 	}
3947 
3948 	if (loop) {
3949 		if (unpin == &root->fs_info->freed_extents[0])
3950 			unpin = &root->fs_info->freed_extents[1];
3951 		else
3952 			unpin = &root->fs_info->freed_extents[0];
3953 		loop = false;
3954 		goto again;
3955 	}
3956 
3957 	return 0;
3958 }
3959 
3960 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
3961 				   struct btrfs_root *root)
3962 {
3963 	btrfs_destroy_delayed_refs(cur_trans, root);
3964 	btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
3965 				cur_trans->dirty_pages.dirty_bytes);
3966 
3967 	cur_trans->state = TRANS_STATE_COMMIT_START;
3968 	wake_up(&root->fs_info->transaction_blocked_wait);
3969 
3970 	btrfs_evict_pending_snapshots(cur_trans);
3971 
3972 	cur_trans->state = TRANS_STATE_UNBLOCKED;
3973 	wake_up(&root->fs_info->transaction_wait);
3974 
3975 	btrfs_destroy_delayed_inodes(root);
3976 	btrfs_assert_delayed_root_empty(root);
3977 
3978 	btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
3979 				     EXTENT_DIRTY);
3980 	btrfs_destroy_pinned_extent(root,
3981 				    root->fs_info->pinned_extents);
3982 
3983 	cur_trans->state =TRANS_STATE_COMPLETED;
3984 	wake_up(&cur_trans->commit_wait);
3985 
3986 	/*
3987 	memset(cur_trans, 0, sizeof(*cur_trans));
3988 	kmem_cache_free(btrfs_transaction_cachep, cur_trans);
3989 	*/
3990 }
3991 
3992 static int btrfs_cleanup_transaction(struct btrfs_root *root)
3993 {
3994 	struct btrfs_transaction *t;
3995 	LIST_HEAD(list);
3996 
3997 	mutex_lock(&root->fs_info->transaction_kthread_mutex);
3998 
3999 	spin_lock(&root->fs_info->trans_lock);
4000 	list_splice_init(&root->fs_info->trans_list, &list);
4001 	root->fs_info->running_transaction = NULL;
4002 	spin_unlock(&root->fs_info->trans_lock);
4003 
4004 	while (!list_empty(&list)) {
4005 		t = list_entry(list.next, struct btrfs_transaction, list);
4006 
4007 		btrfs_destroy_ordered_operations(t, root);
4008 
4009 		btrfs_destroy_all_ordered_extents(root->fs_info);
4010 
4011 		btrfs_destroy_delayed_refs(t, root);
4012 
4013 		/*
4014 		 *  FIXME: cleanup wait for commit
4015 		 *  We needn't acquire the lock here, because we are during
4016 		 *  the umount, there is no other task which will change it.
4017 		 */
4018 		t->state = TRANS_STATE_COMMIT_START;
4019 		smp_mb();
4020 		if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
4021 			wake_up(&root->fs_info->transaction_blocked_wait);
4022 
4023 		btrfs_evict_pending_snapshots(t);
4024 
4025 		t->state = TRANS_STATE_UNBLOCKED;
4026 		smp_mb();
4027 		if (waitqueue_active(&root->fs_info->transaction_wait))
4028 			wake_up(&root->fs_info->transaction_wait);
4029 
4030 		btrfs_destroy_delayed_inodes(root);
4031 		btrfs_assert_delayed_root_empty(root);
4032 
4033 		btrfs_destroy_all_delalloc_inodes(root->fs_info);
4034 
4035 		btrfs_destroy_marked_extents(root, &t->dirty_pages,
4036 					     EXTENT_DIRTY);
4037 
4038 		btrfs_destroy_pinned_extent(root,
4039 					    root->fs_info->pinned_extents);
4040 
4041 		t->state = TRANS_STATE_COMPLETED;
4042 		smp_mb();
4043 		if (waitqueue_active(&t->commit_wait))
4044 			wake_up(&t->commit_wait);
4045 
4046 		atomic_set(&t->use_count, 0);
4047 		list_del_init(&t->list);
4048 		memset(t, 0, sizeof(*t));
4049 		kmem_cache_free(btrfs_transaction_cachep, t);
4050 	}
4051 
4052 	mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4053 
4054 	return 0;
4055 }
4056 
4057 static struct extent_io_ops btree_extent_io_ops = {
4058 	.readpage_end_io_hook = btree_readpage_end_io_hook,
4059 	.readpage_io_failed_hook = btree_io_failed_hook,
4060 	.submit_bio_hook = btree_submit_bio_hook,
4061 	/* note we're sharing with inode.c for the merge bio hook */
4062 	.merge_bio_hook = btrfs_merge_bio_hook,
4063 };
4064