xref: /linux/fs/btrfs/disk-io.c (revision a67ff6a54095e27093ea501fb143fefe51a536c2)
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 <asm/unaligned.h>
34 #include "compat.h"
35 #include "ctree.h"
36 #include "disk-io.h"
37 #include "transaction.h"
38 #include "btrfs_inode.h"
39 #include "volumes.h"
40 #include "print-tree.h"
41 #include "async-thread.h"
42 #include "locking.h"
43 #include "tree-log.h"
44 #include "free-space-cache.h"
45 #include "inode-map.h"
46 
47 static struct extent_io_ops btree_extent_io_ops;
48 static void end_workqueue_fn(struct btrfs_work *work);
49 static void free_fs_root(struct btrfs_root *root);
50 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
51 				    int read_only);
52 static int btrfs_destroy_ordered_operations(struct btrfs_root *root);
53 static int btrfs_destroy_ordered_extents(struct btrfs_root *root);
54 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
55 				      struct btrfs_root *root);
56 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
57 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
58 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
59 					struct extent_io_tree *dirty_pages,
60 					int mark);
61 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
62 				       struct extent_io_tree *pinned_extents);
63 static int btrfs_cleanup_transaction(struct btrfs_root *root);
64 
65 /*
66  * end_io_wq structs are used to do processing in task context when an IO is
67  * complete.  This is used during reads to verify checksums, and it is used
68  * by writes to insert metadata for new file extents after IO is complete.
69  */
70 struct end_io_wq {
71 	struct bio *bio;
72 	bio_end_io_t *end_io;
73 	void *private;
74 	struct btrfs_fs_info *info;
75 	int error;
76 	int metadata;
77 	struct list_head list;
78 	struct btrfs_work work;
79 };
80 
81 /*
82  * async submit bios are used to offload expensive checksumming
83  * onto the worker threads.  They checksum file and metadata bios
84  * just before they are sent down the IO stack.
85  */
86 struct async_submit_bio {
87 	struct inode *inode;
88 	struct bio *bio;
89 	struct list_head list;
90 	extent_submit_bio_hook_t *submit_bio_start;
91 	extent_submit_bio_hook_t *submit_bio_done;
92 	int rw;
93 	int mirror_num;
94 	unsigned long bio_flags;
95 	/*
96 	 * bio_offset is optional, can be used if the pages in the bio
97 	 * can't tell us where in the file the bio should go
98 	 */
99 	u64 bio_offset;
100 	struct btrfs_work work;
101 };
102 
103 /*
104  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
105  * eb, the lockdep key is determined by the btrfs_root it belongs to and
106  * the level the eb occupies in the tree.
107  *
108  * Different roots are used for different purposes and may nest inside each
109  * other and they require separate keysets.  As lockdep keys should be
110  * static, assign keysets according to the purpose of the root as indicated
111  * by btrfs_root->objectid.  This ensures that all special purpose roots
112  * have separate keysets.
113  *
114  * Lock-nesting across peer nodes is always done with the immediate parent
115  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
116  * subclass to avoid triggering lockdep warning in such cases.
117  *
118  * The key is set by the readpage_end_io_hook after the buffer has passed
119  * csum validation but before the pages are unlocked.  It is also set by
120  * btrfs_init_new_buffer on freshly allocated blocks.
121  *
122  * We also add a check to make sure the highest level of the tree is the
123  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
124  * needs update as well.
125  */
126 #ifdef CONFIG_DEBUG_LOCK_ALLOC
127 # if BTRFS_MAX_LEVEL != 8
128 #  error
129 # endif
130 
131 static struct btrfs_lockdep_keyset {
132 	u64			id;		/* root objectid */
133 	const char		*name_stem;	/* lock name stem */
134 	char			names[BTRFS_MAX_LEVEL + 1][20];
135 	struct lock_class_key	keys[BTRFS_MAX_LEVEL + 1];
136 } btrfs_lockdep_keysets[] = {
137 	{ .id = BTRFS_ROOT_TREE_OBJECTID,	.name_stem = "root"	},
138 	{ .id = BTRFS_EXTENT_TREE_OBJECTID,	.name_stem = "extent"	},
139 	{ .id = BTRFS_CHUNK_TREE_OBJECTID,	.name_stem = "chunk"	},
140 	{ .id = BTRFS_DEV_TREE_OBJECTID,	.name_stem = "dev"	},
141 	{ .id = BTRFS_FS_TREE_OBJECTID,		.name_stem = "fs"	},
142 	{ .id = BTRFS_CSUM_TREE_OBJECTID,	.name_stem = "csum"	},
143 	{ .id = BTRFS_ORPHAN_OBJECTID,		.name_stem = "orphan"	},
144 	{ .id = BTRFS_TREE_LOG_OBJECTID,	.name_stem = "log"	},
145 	{ .id = BTRFS_TREE_RELOC_OBJECTID,	.name_stem = "treloc"	},
146 	{ .id = BTRFS_DATA_RELOC_TREE_OBJECTID,	.name_stem = "dreloc"	},
147 	{ .id = 0,				.name_stem = "tree"	},
148 };
149 
150 void __init btrfs_init_lockdep(void)
151 {
152 	int i, j;
153 
154 	/* initialize lockdep class names */
155 	for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
156 		struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
157 
158 		for (j = 0; j < ARRAY_SIZE(ks->names); j++)
159 			snprintf(ks->names[j], sizeof(ks->names[j]),
160 				 "btrfs-%s-%02d", ks->name_stem, j);
161 	}
162 }
163 
164 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
165 				    int level)
166 {
167 	struct btrfs_lockdep_keyset *ks;
168 
169 	BUG_ON(level >= ARRAY_SIZE(ks->keys));
170 
171 	/* find the matching keyset, id 0 is the default entry */
172 	for (ks = btrfs_lockdep_keysets; ks->id; ks++)
173 		if (ks->id == objectid)
174 			break;
175 
176 	lockdep_set_class_and_name(&eb->lock,
177 				   &ks->keys[level], ks->names[level]);
178 }
179 
180 #endif
181 
182 /*
183  * extents on the btree inode are pretty simple, there's one extent
184  * that covers the entire device
185  */
186 static struct extent_map *btree_get_extent(struct inode *inode,
187 		struct page *page, size_t pg_offset, u64 start, u64 len,
188 		int create)
189 {
190 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
191 	struct extent_map *em;
192 	int ret;
193 
194 	read_lock(&em_tree->lock);
195 	em = lookup_extent_mapping(em_tree, start, len);
196 	if (em) {
197 		em->bdev =
198 			BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
199 		read_unlock(&em_tree->lock);
200 		goto out;
201 	}
202 	read_unlock(&em_tree->lock);
203 
204 	em = alloc_extent_map();
205 	if (!em) {
206 		em = ERR_PTR(-ENOMEM);
207 		goto out;
208 	}
209 	em->start = 0;
210 	em->len = (u64)-1;
211 	em->block_len = (u64)-1;
212 	em->block_start = 0;
213 	em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
214 
215 	write_lock(&em_tree->lock);
216 	ret = add_extent_mapping(em_tree, em);
217 	if (ret == -EEXIST) {
218 		u64 failed_start = em->start;
219 		u64 failed_len = em->len;
220 
221 		free_extent_map(em);
222 		em = lookup_extent_mapping(em_tree, start, len);
223 		if (em) {
224 			ret = 0;
225 		} else {
226 			em = lookup_extent_mapping(em_tree, failed_start,
227 						   failed_len);
228 			ret = -EIO;
229 		}
230 	} else if (ret) {
231 		free_extent_map(em);
232 		em = NULL;
233 	}
234 	write_unlock(&em_tree->lock);
235 
236 	if (ret)
237 		em = ERR_PTR(ret);
238 out:
239 	return em;
240 }
241 
242 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
243 {
244 	return crc32c(seed, data, len);
245 }
246 
247 void btrfs_csum_final(u32 crc, char *result)
248 {
249 	put_unaligned_le32(~crc, result);
250 }
251 
252 /*
253  * compute the csum for a btree block, and either verify it or write it
254  * into the csum field of the block.
255  */
256 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
257 			   int verify)
258 {
259 	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
260 	char *result = NULL;
261 	unsigned long len;
262 	unsigned long cur_len;
263 	unsigned long offset = BTRFS_CSUM_SIZE;
264 	char *kaddr;
265 	unsigned long map_start;
266 	unsigned long map_len;
267 	int err;
268 	u32 crc = ~(u32)0;
269 	unsigned long inline_result;
270 
271 	len = buf->len - offset;
272 	while (len > 0) {
273 		err = map_private_extent_buffer(buf, offset, 32,
274 					&kaddr, &map_start, &map_len);
275 		if (err)
276 			return 1;
277 		cur_len = min(len, map_len - (offset - map_start));
278 		crc = btrfs_csum_data(root, kaddr + offset - map_start,
279 				      crc, cur_len);
280 		len -= cur_len;
281 		offset += cur_len;
282 	}
283 	if (csum_size > sizeof(inline_result)) {
284 		result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
285 		if (!result)
286 			return 1;
287 	} else {
288 		result = (char *)&inline_result;
289 	}
290 
291 	btrfs_csum_final(crc, result);
292 
293 	if (verify) {
294 		if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
295 			u32 val;
296 			u32 found = 0;
297 			memcpy(&found, result, csum_size);
298 
299 			read_extent_buffer(buf, &val, 0, csum_size);
300 			printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
301 				       "failed on %llu wanted %X found %X "
302 				       "level %d\n",
303 				       root->fs_info->sb->s_id,
304 				       (unsigned long long)buf->start, val, found,
305 				       btrfs_header_level(buf));
306 			if (result != (char *)&inline_result)
307 				kfree(result);
308 			return 1;
309 		}
310 	} else {
311 		write_extent_buffer(buf, result, 0, csum_size);
312 	}
313 	if (result != (char *)&inline_result)
314 		kfree(result);
315 	return 0;
316 }
317 
318 /*
319  * we can't consider a given block up to date unless the transid of the
320  * block matches the transid in the parent node's pointer.  This is how we
321  * detect blocks that either didn't get written at all or got written
322  * in the wrong place.
323  */
324 static int verify_parent_transid(struct extent_io_tree *io_tree,
325 				 struct extent_buffer *eb, u64 parent_transid)
326 {
327 	struct extent_state *cached_state = NULL;
328 	int ret;
329 
330 	if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
331 		return 0;
332 
333 	lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
334 			 0, &cached_state, GFP_NOFS);
335 	if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
336 	    btrfs_header_generation(eb) == parent_transid) {
337 		ret = 0;
338 		goto out;
339 	}
340 	printk_ratelimited("parent transid verify failed on %llu wanted %llu "
341 		       "found %llu\n",
342 		       (unsigned long long)eb->start,
343 		       (unsigned long long)parent_transid,
344 		       (unsigned long long)btrfs_header_generation(eb));
345 	ret = 1;
346 	clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
347 out:
348 	unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
349 			     &cached_state, GFP_NOFS);
350 	return ret;
351 }
352 
353 /*
354  * helper to read a given tree block, doing retries as required when
355  * the checksums don't match and we have alternate mirrors to try.
356  */
357 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
358 					  struct extent_buffer *eb,
359 					  u64 start, u64 parent_transid)
360 {
361 	struct extent_io_tree *io_tree;
362 	int ret;
363 	int num_copies = 0;
364 	int mirror_num = 0;
365 
366 	clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
367 	io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
368 	while (1) {
369 		ret = read_extent_buffer_pages(io_tree, eb, start,
370 					       WAIT_COMPLETE,
371 					       btree_get_extent, mirror_num);
372 		if (!ret &&
373 		    !verify_parent_transid(io_tree, eb, parent_transid))
374 			return ret;
375 
376 		/*
377 		 * This buffer's crc is fine, but its contents are corrupted, so
378 		 * there is no reason to read the other copies, they won't be
379 		 * any less wrong.
380 		 */
381 		if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
382 			return ret;
383 
384 		num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
385 					      eb->start, eb->len);
386 		if (num_copies == 1)
387 			return ret;
388 
389 		mirror_num++;
390 		if (mirror_num > num_copies)
391 			return ret;
392 	}
393 	return -EIO;
394 }
395 
396 /*
397  * checksum a dirty tree block before IO.  This has extra checks to make sure
398  * we only fill in the checksum field in the first page of a multi-page block
399  */
400 
401 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
402 {
403 	struct extent_io_tree *tree;
404 	u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
405 	u64 found_start;
406 	unsigned long len;
407 	struct extent_buffer *eb;
408 	int ret;
409 
410 	tree = &BTRFS_I(page->mapping->host)->io_tree;
411 
412 	if (page->private == EXTENT_PAGE_PRIVATE) {
413 		WARN_ON(1);
414 		goto out;
415 	}
416 	if (!page->private) {
417 		WARN_ON(1);
418 		goto out;
419 	}
420 	len = page->private >> 2;
421 	WARN_ON(len == 0);
422 
423 	eb = alloc_extent_buffer(tree, start, len, page);
424 	if (eb == NULL) {
425 		WARN_ON(1);
426 		goto out;
427 	}
428 	ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
429 					     btrfs_header_generation(eb));
430 	BUG_ON(ret);
431 	WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
432 
433 	found_start = btrfs_header_bytenr(eb);
434 	if (found_start != start) {
435 		WARN_ON(1);
436 		goto err;
437 	}
438 	if (eb->first_page != page) {
439 		WARN_ON(1);
440 		goto err;
441 	}
442 	if (!PageUptodate(page)) {
443 		WARN_ON(1);
444 		goto err;
445 	}
446 	csum_tree_block(root, eb, 0);
447 err:
448 	free_extent_buffer(eb);
449 out:
450 	return 0;
451 }
452 
453 static int check_tree_block_fsid(struct btrfs_root *root,
454 				 struct extent_buffer *eb)
455 {
456 	struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
457 	u8 fsid[BTRFS_UUID_SIZE];
458 	int ret = 1;
459 
460 	read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
461 			   BTRFS_FSID_SIZE);
462 	while (fs_devices) {
463 		if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
464 			ret = 0;
465 			break;
466 		}
467 		fs_devices = fs_devices->seed;
468 	}
469 	return ret;
470 }
471 
472 #define CORRUPT(reason, eb, root, slot)				\
473 	printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu,"	\
474 	       "root=%llu, slot=%d\n", reason,			\
475 	       (unsigned long long)btrfs_header_bytenr(eb),	\
476 	       (unsigned long long)root->objectid, slot)
477 
478 static noinline int check_leaf(struct btrfs_root *root,
479 			       struct extent_buffer *leaf)
480 {
481 	struct btrfs_key key;
482 	struct btrfs_key leaf_key;
483 	u32 nritems = btrfs_header_nritems(leaf);
484 	int slot;
485 
486 	if (nritems == 0)
487 		return 0;
488 
489 	/* Check the 0 item */
490 	if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
491 	    BTRFS_LEAF_DATA_SIZE(root)) {
492 		CORRUPT("invalid item offset size pair", leaf, root, 0);
493 		return -EIO;
494 	}
495 
496 	/*
497 	 * Check to make sure each items keys are in the correct order and their
498 	 * offsets make sense.  We only have to loop through nritems-1 because
499 	 * we check the current slot against the next slot, which verifies the
500 	 * next slot's offset+size makes sense and that the current's slot
501 	 * offset is correct.
502 	 */
503 	for (slot = 0; slot < nritems - 1; slot++) {
504 		btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
505 		btrfs_item_key_to_cpu(leaf, &key, slot + 1);
506 
507 		/* Make sure the keys are in the right order */
508 		if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
509 			CORRUPT("bad key order", leaf, root, slot);
510 			return -EIO;
511 		}
512 
513 		/*
514 		 * Make sure the offset and ends are right, remember that the
515 		 * item data starts at the end of the leaf and grows towards the
516 		 * front.
517 		 */
518 		if (btrfs_item_offset_nr(leaf, slot) !=
519 			btrfs_item_end_nr(leaf, slot + 1)) {
520 			CORRUPT("slot offset bad", leaf, root, slot);
521 			return -EIO;
522 		}
523 
524 		/*
525 		 * Check to make sure that we don't point outside of the leaf,
526 		 * just incase all the items are consistent to eachother, but
527 		 * all point outside of the leaf.
528 		 */
529 		if (btrfs_item_end_nr(leaf, slot) >
530 		    BTRFS_LEAF_DATA_SIZE(root)) {
531 			CORRUPT("slot end outside of leaf", leaf, root, slot);
532 			return -EIO;
533 		}
534 	}
535 
536 	return 0;
537 }
538 
539 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
540 			       struct extent_state *state)
541 {
542 	struct extent_io_tree *tree;
543 	u64 found_start;
544 	int found_level;
545 	unsigned long len;
546 	struct extent_buffer *eb;
547 	struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
548 	int ret = 0;
549 
550 	tree = &BTRFS_I(page->mapping->host)->io_tree;
551 	if (page->private == EXTENT_PAGE_PRIVATE)
552 		goto out;
553 	if (!page->private)
554 		goto out;
555 
556 	len = page->private >> 2;
557 	WARN_ON(len == 0);
558 
559 	eb = alloc_extent_buffer(tree, start, len, page);
560 	if (eb == NULL) {
561 		ret = -EIO;
562 		goto out;
563 	}
564 
565 	found_start = btrfs_header_bytenr(eb);
566 	if (found_start != start) {
567 		printk_ratelimited(KERN_INFO "btrfs bad tree block start "
568 			       "%llu %llu\n",
569 			       (unsigned long long)found_start,
570 			       (unsigned long long)eb->start);
571 		ret = -EIO;
572 		goto err;
573 	}
574 	if (eb->first_page != page) {
575 		printk(KERN_INFO "btrfs bad first page %lu %lu\n",
576 		       eb->first_page->index, page->index);
577 		WARN_ON(1);
578 		ret = -EIO;
579 		goto err;
580 	}
581 	if (check_tree_block_fsid(root, eb)) {
582 		printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
583 			       (unsigned long long)eb->start);
584 		ret = -EIO;
585 		goto err;
586 	}
587 	found_level = btrfs_header_level(eb);
588 
589 	btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
590 				       eb, found_level);
591 
592 	ret = csum_tree_block(root, eb, 1);
593 	if (ret) {
594 		ret = -EIO;
595 		goto err;
596 	}
597 
598 	/*
599 	 * If this is a leaf block and it is corrupt, set the corrupt bit so
600 	 * that we don't try and read the other copies of this block, just
601 	 * return -EIO.
602 	 */
603 	if (found_level == 0 && check_leaf(root, eb)) {
604 		set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
605 		ret = -EIO;
606 	}
607 
608 	end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
609 	end = eb->start + end - 1;
610 err:
611 	if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) {
612 		clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags);
613 		btree_readahead_hook(root, eb, eb->start, ret);
614 	}
615 
616 	free_extent_buffer(eb);
617 out:
618 	return ret;
619 }
620 
621 static int btree_io_failed_hook(struct bio *failed_bio,
622 			 struct page *page, u64 start, u64 end,
623 			 u64 mirror_num, struct extent_state *state)
624 {
625 	struct extent_io_tree *tree;
626 	unsigned long len;
627 	struct extent_buffer *eb;
628 	struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
629 
630 	tree = &BTRFS_I(page->mapping->host)->io_tree;
631 	if (page->private == EXTENT_PAGE_PRIVATE)
632 		goto out;
633 	if (!page->private)
634 		goto out;
635 
636 	len = page->private >> 2;
637 	WARN_ON(len == 0);
638 
639 	eb = alloc_extent_buffer(tree, start, len, page);
640 	if (eb == NULL)
641 		goto out;
642 
643 	if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) {
644 		clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags);
645 		btree_readahead_hook(root, eb, eb->start, -EIO);
646 	}
647 	free_extent_buffer(eb);
648 
649 out:
650 	return -EIO;	/* we fixed nothing */
651 }
652 
653 static void end_workqueue_bio(struct bio *bio, int err)
654 {
655 	struct end_io_wq *end_io_wq = bio->bi_private;
656 	struct btrfs_fs_info *fs_info;
657 
658 	fs_info = end_io_wq->info;
659 	end_io_wq->error = err;
660 	end_io_wq->work.func = end_workqueue_fn;
661 	end_io_wq->work.flags = 0;
662 
663 	if (bio->bi_rw & REQ_WRITE) {
664 		if (end_io_wq->metadata == 1)
665 			btrfs_queue_worker(&fs_info->endio_meta_write_workers,
666 					   &end_io_wq->work);
667 		else if (end_io_wq->metadata == 2)
668 			btrfs_queue_worker(&fs_info->endio_freespace_worker,
669 					   &end_io_wq->work);
670 		else
671 			btrfs_queue_worker(&fs_info->endio_write_workers,
672 					   &end_io_wq->work);
673 	} else {
674 		if (end_io_wq->metadata)
675 			btrfs_queue_worker(&fs_info->endio_meta_workers,
676 					   &end_io_wq->work);
677 		else
678 			btrfs_queue_worker(&fs_info->endio_workers,
679 					   &end_io_wq->work);
680 	}
681 }
682 
683 /*
684  * For the metadata arg you want
685  *
686  * 0 - if data
687  * 1 - if normal metadta
688  * 2 - if writing to the free space cache area
689  */
690 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
691 			int metadata)
692 {
693 	struct end_io_wq *end_io_wq;
694 	end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
695 	if (!end_io_wq)
696 		return -ENOMEM;
697 
698 	end_io_wq->private = bio->bi_private;
699 	end_io_wq->end_io = bio->bi_end_io;
700 	end_io_wq->info = info;
701 	end_io_wq->error = 0;
702 	end_io_wq->bio = bio;
703 	end_io_wq->metadata = metadata;
704 
705 	bio->bi_private = end_io_wq;
706 	bio->bi_end_io = end_workqueue_bio;
707 	return 0;
708 }
709 
710 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
711 {
712 	unsigned long limit = min_t(unsigned long,
713 				    info->workers.max_workers,
714 				    info->fs_devices->open_devices);
715 	return 256 * limit;
716 }
717 
718 static void run_one_async_start(struct btrfs_work *work)
719 {
720 	struct async_submit_bio *async;
721 
722 	async = container_of(work, struct  async_submit_bio, work);
723 	async->submit_bio_start(async->inode, async->rw, async->bio,
724 			       async->mirror_num, async->bio_flags,
725 			       async->bio_offset);
726 }
727 
728 static void run_one_async_done(struct btrfs_work *work)
729 {
730 	struct btrfs_fs_info *fs_info;
731 	struct async_submit_bio *async;
732 	int limit;
733 
734 	async = container_of(work, struct  async_submit_bio, work);
735 	fs_info = BTRFS_I(async->inode)->root->fs_info;
736 
737 	limit = btrfs_async_submit_limit(fs_info);
738 	limit = limit * 2 / 3;
739 
740 	atomic_dec(&fs_info->nr_async_submits);
741 
742 	if (atomic_read(&fs_info->nr_async_submits) < limit &&
743 	    waitqueue_active(&fs_info->async_submit_wait))
744 		wake_up(&fs_info->async_submit_wait);
745 
746 	async->submit_bio_done(async->inode, async->rw, async->bio,
747 			       async->mirror_num, async->bio_flags,
748 			       async->bio_offset);
749 }
750 
751 static void run_one_async_free(struct btrfs_work *work)
752 {
753 	struct async_submit_bio *async;
754 
755 	async = container_of(work, struct  async_submit_bio, work);
756 	kfree(async);
757 }
758 
759 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
760 			int rw, struct bio *bio, int mirror_num,
761 			unsigned long bio_flags,
762 			u64 bio_offset,
763 			extent_submit_bio_hook_t *submit_bio_start,
764 			extent_submit_bio_hook_t *submit_bio_done)
765 {
766 	struct async_submit_bio *async;
767 
768 	async = kmalloc(sizeof(*async), GFP_NOFS);
769 	if (!async)
770 		return -ENOMEM;
771 
772 	async->inode = inode;
773 	async->rw = rw;
774 	async->bio = bio;
775 	async->mirror_num = mirror_num;
776 	async->submit_bio_start = submit_bio_start;
777 	async->submit_bio_done = submit_bio_done;
778 
779 	async->work.func = run_one_async_start;
780 	async->work.ordered_func = run_one_async_done;
781 	async->work.ordered_free = run_one_async_free;
782 
783 	async->work.flags = 0;
784 	async->bio_flags = bio_flags;
785 	async->bio_offset = bio_offset;
786 
787 	atomic_inc(&fs_info->nr_async_submits);
788 
789 	if (rw & REQ_SYNC)
790 		btrfs_set_work_high_prio(&async->work);
791 
792 	btrfs_queue_worker(&fs_info->workers, &async->work);
793 
794 	while (atomic_read(&fs_info->async_submit_draining) &&
795 	      atomic_read(&fs_info->nr_async_submits)) {
796 		wait_event(fs_info->async_submit_wait,
797 			   (atomic_read(&fs_info->nr_async_submits) == 0));
798 	}
799 
800 	return 0;
801 }
802 
803 static int btree_csum_one_bio(struct bio *bio)
804 {
805 	struct bio_vec *bvec = bio->bi_io_vec;
806 	int bio_index = 0;
807 	struct btrfs_root *root;
808 
809 	WARN_ON(bio->bi_vcnt <= 0);
810 	while (bio_index < bio->bi_vcnt) {
811 		root = BTRFS_I(bvec->bv_page->mapping->host)->root;
812 		csum_dirty_buffer(root, bvec->bv_page);
813 		bio_index++;
814 		bvec++;
815 	}
816 	return 0;
817 }
818 
819 static int __btree_submit_bio_start(struct inode *inode, int rw,
820 				    struct bio *bio, int mirror_num,
821 				    unsigned long bio_flags,
822 				    u64 bio_offset)
823 {
824 	/*
825 	 * when we're called for a write, we're already in the async
826 	 * submission context.  Just jump into btrfs_map_bio
827 	 */
828 	btree_csum_one_bio(bio);
829 	return 0;
830 }
831 
832 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
833 				 int mirror_num, unsigned long bio_flags,
834 				 u64 bio_offset)
835 {
836 	/*
837 	 * when we're called for a write, we're already in the async
838 	 * submission context.  Just jump into btrfs_map_bio
839 	 */
840 	return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
841 }
842 
843 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
844 				 int mirror_num, unsigned long bio_flags,
845 				 u64 bio_offset)
846 {
847 	int ret;
848 
849 	ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
850 					  bio, 1);
851 	BUG_ON(ret);
852 
853 	if (!(rw & REQ_WRITE)) {
854 		/*
855 		 * called for a read, do the setup so that checksum validation
856 		 * can happen in the async kernel threads
857 		 */
858 		return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
859 				     mirror_num, 0);
860 	}
861 
862 	/*
863 	 * kthread helpers are used to submit writes so that checksumming
864 	 * can happen in parallel across all CPUs
865 	 */
866 	return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
867 				   inode, rw, bio, mirror_num, 0,
868 				   bio_offset,
869 				   __btree_submit_bio_start,
870 				   __btree_submit_bio_done);
871 }
872 
873 #ifdef CONFIG_MIGRATION
874 static int btree_migratepage(struct address_space *mapping,
875 			struct page *newpage, struct page *page)
876 {
877 	/*
878 	 * we can't safely write a btree page from here,
879 	 * we haven't done the locking hook
880 	 */
881 	if (PageDirty(page))
882 		return -EAGAIN;
883 	/*
884 	 * Buffers may be managed in a filesystem specific way.
885 	 * We must have no buffers or drop them.
886 	 */
887 	if (page_has_private(page) &&
888 	    !try_to_release_page(page, GFP_KERNEL))
889 		return -EAGAIN;
890 	return migrate_page(mapping, newpage, page);
891 }
892 #endif
893 
894 static int btree_writepage(struct page *page, struct writeback_control *wbc)
895 {
896 	struct extent_io_tree *tree;
897 	struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
898 	struct extent_buffer *eb;
899 	int was_dirty;
900 
901 	tree = &BTRFS_I(page->mapping->host)->io_tree;
902 	if (!(current->flags & PF_MEMALLOC)) {
903 		return extent_write_full_page(tree, page,
904 					      btree_get_extent, wbc);
905 	}
906 
907 	redirty_page_for_writepage(wbc, page);
908 	eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
909 	WARN_ON(!eb);
910 
911 	was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
912 	if (!was_dirty) {
913 		spin_lock(&root->fs_info->delalloc_lock);
914 		root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
915 		spin_unlock(&root->fs_info->delalloc_lock);
916 	}
917 	free_extent_buffer(eb);
918 
919 	unlock_page(page);
920 	return 0;
921 }
922 
923 static int btree_writepages(struct address_space *mapping,
924 			    struct writeback_control *wbc)
925 {
926 	struct extent_io_tree *tree;
927 	tree = &BTRFS_I(mapping->host)->io_tree;
928 	if (wbc->sync_mode == WB_SYNC_NONE) {
929 		struct btrfs_root *root = BTRFS_I(mapping->host)->root;
930 		u64 num_dirty;
931 		unsigned long thresh = 32 * 1024 * 1024;
932 
933 		if (wbc->for_kupdate)
934 			return 0;
935 
936 		/* this is a bit racy, but that's ok */
937 		num_dirty = root->fs_info->dirty_metadata_bytes;
938 		if (num_dirty < thresh)
939 			return 0;
940 	}
941 	return extent_writepages(tree, mapping, btree_get_extent, wbc);
942 }
943 
944 static int btree_readpage(struct file *file, struct page *page)
945 {
946 	struct extent_io_tree *tree;
947 	tree = &BTRFS_I(page->mapping->host)->io_tree;
948 	return extent_read_full_page(tree, page, btree_get_extent, 0);
949 }
950 
951 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
952 {
953 	struct extent_io_tree *tree;
954 	struct extent_map_tree *map;
955 	int ret;
956 
957 	if (PageWriteback(page) || PageDirty(page))
958 		return 0;
959 
960 	tree = &BTRFS_I(page->mapping->host)->io_tree;
961 	map = &BTRFS_I(page->mapping->host)->extent_tree;
962 
963 	ret = try_release_extent_state(map, tree, page, gfp_flags);
964 	if (!ret)
965 		return 0;
966 
967 	ret = try_release_extent_buffer(tree, page);
968 	if (ret == 1) {
969 		ClearPagePrivate(page);
970 		set_page_private(page, 0);
971 		page_cache_release(page);
972 	}
973 
974 	return ret;
975 }
976 
977 static void btree_invalidatepage(struct page *page, unsigned long offset)
978 {
979 	struct extent_io_tree *tree;
980 	tree = &BTRFS_I(page->mapping->host)->io_tree;
981 	extent_invalidatepage(tree, page, offset);
982 	btree_releasepage(page, GFP_NOFS);
983 	if (PagePrivate(page)) {
984 		printk(KERN_WARNING "btrfs warning page private not zero "
985 		       "on page %llu\n", (unsigned long long)page_offset(page));
986 		ClearPagePrivate(page);
987 		set_page_private(page, 0);
988 		page_cache_release(page);
989 	}
990 }
991 
992 static const struct address_space_operations btree_aops = {
993 	.readpage	= btree_readpage,
994 	.writepage	= btree_writepage,
995 	.writepages	= btree_writepages,
996 	.releasepage	= btree_releasepage,
997 	.invalidatepage = btree_invalidatepage,
998 #ifdef CONFIG_MIGRATION
999 	.migratepage	= btree_migratepage,
1000 #endif
1001 };
1002 
1003 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1004 			 u64 parent_transid)
1005 {
1006 	struct extent_buffer *buf = NULL;
1007 	struct inode *btree_inode = root->fs_info->btree_inode;
1008 	int ret = 0;
1009 
1010 	buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1011 	if (!buf)
1012 		return 0;
1013 	read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1014 				 buf, 0, WAIT_NONE, btree_get_extent, 0);
1015 	free_extent_buffer(buf);
1016 	return ret;
1017 }
1018 
1019 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1020 			 int mirror_num, struct extent_buffer **eb)
1021 {
1022 	struct extent_buffer *buf = NULL;
1023 	struct inode *btree_inode = root->fs_info->btree_inode;
1024 	struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1025 	int ret;
1026 
1027 	buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1028 	if (!buf)
1029 		return 0;
1030 
1031 	set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1032 
1033 	ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1034 				       btree_get_extent, mirror_num);
1035 	if (ret) {
1036 		free_extent_buffer(buf);
1037 		return ret;
1038 	}
1039 
1040 	if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1041 		free_extent_buffer(buf);
1042 		return -EIO;
1043 	} else if (extent_buffer_uptodate(io_tree, buf, NULL)) {
1044 		*eb = buf;
1045 	} else {
1046 		free_extent_buffer(buf);
1047 	}
1048 	return 0;
1049 }
1050 
1051 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1052 					    u64 bytenr, u32 blocksize)
1053 {
1054 	struct inode *btree_inode = root->fs_info->btree_inode;
1055 	struct extent_buffer *eb;
1056 	eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1057 				bytenr, blocksize);
1058 	return eb;
1059 }
1060 
1061 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1062 						 u64 bytenr, u32 blocksize)
1063 {
1064 	struct inode *btree_inode = root->fs_info->btree_inode;
1065 	struct extent_buffer *eb;
1066 
1067 	eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1068 				 bytenr, blocksize, NULL);
1069 	return eb;
1070 }
1071 
1072 
1073 int btrfs_write_tree_block(struct extent_buffer *buf)
1074 {
1075 	return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
1076 					buf->start + buf->len - 1);
1077 }
1078 
1079 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1080 {
1081 	return filemap_fdatawait_range(buf->first_page->mapping,
1082 				       buf->start, buf->start + buf->len - 1);
1083 }
1084 
1085 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1086 				      u32 blocksize, u64 parent_transid)
1087 {
1088 	struct extent_buffer *buf = NULL;
1089 	int ret;
1090 
1091 	buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1092 	if (!buf)
1093 		return NULL;
1094 
1095 	ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1096 
1097 	if (ret == 0)
1098 		set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
1099 	return buf;
1100 
1101 }
1102 
1103 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1104 		     struct extent_buffer *buf)
1105 {
1106 	struct inode *btree_inode = root->fs_info->btree_inode;
1107 	if (btrfs_header_generation(buf) ==
1108 	    root->fs_info->running_transaction->transid) {
1109 		btrfs_assert_tree_locked(buf);
1110 
1111 		if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1112 			spin_lock(&root->fs_info->delalloc_lock);
1113 			if (root->fs_info->dirty_metadata_bytes >= buf->len)
1114 				root->fs_info->dirty_metadata_bytes -= buf->len;
1115 			else
1116 				WARN_ON(1);
1117 			spin_unlock(&root->fs_info->delalloc_lock);
1118 		}
1119 
1120 		/* ugh, clear_extent_buffer_dirty needs to lock the page */
1121 		btrfs_set_lock_blocking(buf);
1122 		clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
1123 					  buf);
1124 	}
1125 	return 0;
1126 }
1127 
1128 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1129 			u32 stripesize, struct btrfs_root *root,
1130 			struct btrfs_fs_info *fs_info,
1131 			u64 objectid)
1132 {
1133 	root->node = NULL;
1134 	root->commit_root = NULL;
1135 	root->sectorsize = sectorsize;
1136 	root->nodesize = nodesize;
1137 	root->leafsize = leafsize;
1138 	root->stripesize = stripesize;
1139 	root->ref_cows = 0;
1140 	root->track_dirty = 0;
1141 	root->in_radix = 0;
1142 	root->orphan_item_inserted = 0;
1143 	root->orphan_cleanup_state = 0;
1144 
1145 	root->fs_info = fs_info;
1146 	root->objectid = objectid;
1147 	root->last_trans = 0;
1148 	root->highest_objectid = 0;
1149 	root->name = NULL;
1150 	root->inode_tree = RB_ROOT;
1151 	INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1152 	root->block_rsv = NULL;
1153 	root->orphan_block_rsv = NULL;
1154 
1155 	INIT_LIST_HEAD(&root->dirty_list);
1156 	INIT_LIST_HEAD(&root->orphan_list);
1157 	INIT_LIST_HEAD(&root->root_list);
1158 	spin_lock_init(&root->orphan_lock);
1159 	spin_lock_init(&root->inode_lock);
1160 	spin_lock_init(&root->accounting_lock);
1161 	mutex_init(&root->objectid_mutex);
1162 	mutex_init(&root->log_mutex);
1163 	init_waitqueue_head(&root->log_writer_wait);
1164 	init_waitqueue_head(&root->log_commit_wait[0]);
1165 	init_waitqueue_head(&root->log_commit_wait[1]);
1166 	atomic_set(&root->log_commit[0], 0);
1167 	atomic_set(&root->log_commit[1], 0);
1168 	atomic_set(&root->log_writers, 0);
1169 	root->log_batch = 0;
1170 	root->log_transid = 0;
1171 	root->last_log_commit = 0;
1172 	extent_io_tree_init(&root->dirty_log_pages,
1173 			     fs_info->btree_inode->i_mapping);
1174 
1175 	memset(&root->root_key, 0, sizeof(root->root_key));
1176 	memset(&root->root_item, 0, sizeof(root->root_item));
1177 	memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1178 	memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1179 	root->defrag_trans_start = fs_info->generation;
1180 	init_completion(&root->kobj_unregister);
1181 	root->defrag_running = 0;
1182 	root->root_key.objectid = objectid;
1183 	root->anon_dev = 0;
1184 	return 0;
1185 }
1186 
1187 static int find_and_setup_root(struct btrfs_root *tree_root,
1188 			       struct btrfs_fs_info *fs_info,
1189 			       u64 objectid,
1190 			       struct btrfs_root *root)
1191 {
1192 	int ret;
1193 	u32 blocksize;
1194 	u64 generation;
1195 
1196 	__setup_root(tree_root->nodesize, tree_root->leafsize,
1197 		     tree_root->sectorsize, tree_root->stripesize,
1198 		     root, fs_info, objectid);
1199 	ret = btrfs_find_last_root(tree_root, objectid,
1200 				   &root->root_item, &root->root_key);
1201 	if (ret > 0)
1202 		return -ENOENT;
1203 	BUG_ON(ret);
1204 
1205 	generation = btrfs_root_generation(&root->root_item);
1206 	blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1207 	root->commit_root = NULL;
1208 	root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1209 				     blocksize, generation);
1210 	if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) {
1211 		free_extent_buffer(root->node);
1212 		root->node = NULL;
1213 		return -EIO;
1214 	}
1215 	root->commit_root = btrfs_root_node(root);
1216 	return 0;
1217 }
1218 
1219 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1220 					 struct btrfs_fs_info *fs_info)
1221 {
1222 	struct btrfs_root *root;
1223 	struct btrfs_root *tree_root = fs_info->tree_root;
1224 	struct extent_buffer *leaf;
1225 
1226 	root = kzalloc(sizeof(*root), GFP_NOFS);
1227 	if (!root)
1228 		return ERR_PTR(-ENOMEM);
1229 
1230 	__setup_root(tree_root->nodesize, tree_root->leafsize,
1231 		     tree_root->sectorsize, tree_root->stripesize,
1232 		     root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1233 
1234 	root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1235 	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1236 	root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1237 	/*
1238 	 * log trees do not get reference counted because they go away
1239 	 * before a real commit is actually done.  They do store pointers
1240 	 * to file data extents, and those reference counts still get
1241 	 * updated (along with back refs to the log tree).
1242 	 */
1243 	root->ref_cows = 0;
1244 
1245 	leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1246 				      BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
1247 	if (IS_ERR(leaf)) {
1248 		kfree(root);
1249 		return ERR_CAST(leaf);
1250 	}
1251 
1252 	memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1253 	btrfs_set_header_bytenr(leaf, leaf->start);
1254 	btrfs_set_header_generation(leaf, trans->transid);
1255 	btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1256 	btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1257 	root->node = leaf;
1258 
1259 	write_extent_buffer(root->node, root->fs_info->fsid,
1260 			    (unsigned long)btrfs_header_fsid(root->node),
1261 			    BTRFS_FSID_SIZE);
1262 	btrfs_mark_buffer_dirty(root->node);
1263 	btrfs_tree_unlock(root->node);
1264 	return root;
1265 }
1266 
1267 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1268 			     struct btrfs_fs_info *fs_info)
1269 {
1270 	struct btrfs_root *log_root;
1271 
1272 	log_root = alloc_log_tree(trans, fs_info);
1273 	if (IS_ERR(log_root))
1274 		return PTR_ERR(log_root);
1275 	WARN_ON(fs_info->log_root_tree);
1276 	fs_info->log_root_tree = log_root;
1277 	return 0;
1278 }
1279 
1280 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1281 		       struct btrfs_root *root)
1282 {
1283 	struct btrfs_root *log_root;
1284 	struct btrfs_inode_item *inode_item;
1285 
1286 	log_root = alloc_log_tree(trans, root->fs_info);
1287 	if (IS_ERR(log_root))
1288 		return PTR_ERR(log_root);
1289 
1290 	log_root->last_trans = trans->transid;
1291 	log_root->root_key.offset = root->root_key.objectid;
1292 
1293 	inode_item = &log_root->root_item.inode;
1294 	inode_item->generation = cpu_to_le64(1);
1295 	inode_item->size = cpu_to_le64(3);
1296 	inode_item->nlink = cpu_to_le32(1);
1297 	inode_item->nbytes = cpu_to_le64(root->leafsize);
1298 	inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1299 
1300 	btrfs_set_root_node(&log_root->root_item, log_root->node);
1301 
1302 	WARN_ON(root->log_root);
1303 	root->log_root = log_root;
1304 	root->log_transid = 0;
1305 	root->last_log_commit = 0;
1306 	return 0;
1307 }
1308 
1309 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1310 					       struct btrfs_key *location)
1311 {
1312 	struct btrfs_root *root;
1313 	struct btrfs_fs_info *fs_info = tree_root->fs_info;
1314 	struct btrfs_path *path;
1315 	struct extent_buffer *l;
1316 	u64 generation;
1317 	u32 blocksize;
1318 	int ret = 0;
1319 
1320 	root = kzalloc(sizeof(*root), GFP_NOFS);
1321 	if (!root)
1322 		return ERR_PTR(-ENOMEM);
1323 	if (location->offset == (u64)-1) {
1324 		ret = find_and_setup_root(tree_root, fs_info,
1325 					  location->objectid, root);
1326 		if (ret) {
1327 			kfree(root);
1328 			return ERR_PTR(ret);
1329 		}
1330 		goto out;
1331 	}
1332 
1333 	__setup_root(tree_root->nodesize, tree_root->leafsize,
1334 		     tree_root->sectorsize, tree_root->stripesize,
1335 		     root, fs_info, location->objectid);
1336 
1337 	path = btrfs_alloc_path();
1338 	if (!path) {
1339 		kfree(root);
1340 		return ERR_PTR(-ENOMEM);
1341 	}
1342 	ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1343 	if (ret == 0) {
1344 		l = path->nodes[0];
1345 		read_extent_buffer(l, &root->root_item,
1346 				btrfs_item_ptr_offset(l, path->slots[0]),
1347 				sizeof(root->root_item));
1348 		memcpy(&root->root_key, location, sizeof(*location));
1349 	}
1350 	btrfs_free_path(path);
1351 	if (ret) {
1352 		kfree(root);
1353 		if (ret > 0)
1354 			ret = -ENOENT;
1355 		return ERR_PTR(ret);
1356 	}
1357 
1358 	generation = btrfs_root_generation(&root->root_item);
1359 	blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1360 	root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1361 				     blocksize, generation);
1362 	root->commit_root = btrfs_root_node(root);
1363 	BUG_ON(!root->node);
1364 out:
1365 	if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1366 		root->ref_cows = 1;
1367 		btrfs_check_and_init_root_item(&root->root_item);
1368 	}
1369 
1370 	return root;
1371 }
1372 
1373 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1374 					      struct btrfs_key *location)
1375 {
1376 	struct btrfs_root *root;
1377 	int ret;
1378 
1379 	if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1380 		return fs_info->tree_root;
1381 	if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1382 		return fs_info->extent_root;
1383 	if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1384 		return fs_info->chunk_root;
1385 	if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1386 		return fs_info->dev_root;
1387 	if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1388 		return fs_info->csum_root;
1389 again:
1390 	spin_lock(&fs_info->fs_roots_radix_lock);
1391 	root = radix_tree_lookup(&fs_info->fs_roots_radix,
1392 				 (unsigned long)location->objectid);
1393 	spin_unlock(&fs_info->fs_roots_radix_lock);
1394 	if (root)
1395 		return root;
1396 
1397 	root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1398 	if (IS_ERR(root))
1399 		return root;
1400 
1401 	root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1402 	root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1403 					GFP_NOFS);
1404 	if (!root->free_ino_pinned || !root->free_ino_ctl) {
1405 		ret = -ENOMEM;
1406 		goto fail;
1407 	}
1408 
1409 	btrfs_init_free_ino_ctl(root);
1410 	mutex_init(&root->fs_commit_mutex);
1411 	spin_lock_init(&root->cache_lock);
1412 	init_waitqueue_head(&root->cache_wait);
1413 
1414 	ret = get_anon_bdev(&root->anon_dev);
1415 	if (ret)
1416 		goto fail;
1417 
1418 	if (btrfs_root_refs(&root->root_item) == 0) {
1419 		ret = -ENOENT;
1420 		goto fail;
1421 	}
1422 
1423 	ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1424 	if (ret < 0)
1425 		goto fail;
1426 	if (ret == 0)
1427 		root->orphan_item_inserted = 1;
1428 
1429 	ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1430 	if (ret)
1431 		goto fail;
1432 
1433 	spin_lock(&fs_info->fs_roots_radix_lock);
1434 	ret = radix_tree_insert(&fs_info->fs_roots_radix,
1435 				(unsigned long)root->root_key.objectid,
1436 				root);
1437 	if (ret == 0)
1438 		root->in_radix = 1;
1439 
1440 	spin_unlock(&fs_info->fs_roots_radix_lock);
1441 	radix_tree_preload_end();
1442 	if (ret) {
1443 		if (ret == -EEXIST) {
1444 			free_fs_root(root);
1445 			goto again;
1446 		}
1447 		goto fail;
1448 	}
1449 
1450 	ret = btrfs_find_dead_roots(fs_info->tree_root,
1451 				    root->root_key.objectid);
1452 	WARN_ON(ret);
1453 	return root;
1454 fail:
1455 	free_fs_root(root);
1456 	return ERR_PTR(ret);
1457 }
1458 
1459 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1460 {
1461 	struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1462 	int ret = 0;
1463 	struct btrfs_device *device;
1464 	struct backing_dev_info *bdi;
1465 
1466 	rcu_read_lock();
1467 	list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1468 		if (!device->bdev)
1469 			continue;
1470 		bdi = blk_get_backing_dev_info(device->bdev);
1471 		if (bdi && bdi_congested(bdi, bdi_bits)) {
1472 			ret = 1;
1473 			break;
1474 		}
1475 	}
1476 	rcu_read_unlock();
1477 	return ret;
1478 }
1479 
1480 /*
1481  * If this fails, caller must call bdi_destroy() to get rid of the
1482  * bdi again.
1483  */
1484 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1485 {
1486 	int err;
1487 
1488 	bdi->capabilities = BDI_CAP_MAP_COPY;
1489 	err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1490 	if (err)
1491 		return err;
1492 
1493 	bdi->ra_pages	= default_backing_dev_info.ra_pages;
1494 	bdi->congested_fn	= btrfs_congested_fn;
1495 	bdi->congested_data	= info;
1496 	return 0;
1497 }
1498 
1499 static int bio_ready_for_csum(struct bio *bio)
1500 {
1501 	u64 length = 0;
1502 	u64 buf_len = 0;
1503 	u64 start = 0;
1504 	struct page *page;
1505 	struct extent_io_tree *io_tree = NULL;
1506 	struct bio_vec *bvec;
1507 	int i;
1508 	int ret;
1509 
1510 	bio_for_each_segment(bvec, bio, i) {
1511 		page = bvec->bv_page;
1512 		if (page->private == EXTENT_PAGE_PRIVATE) {
1513 			length += bvec->bv_len;
1514 			continue;
1515 		}
1516 		if (!page->private) {
1517 			length += bvec->bv_len;
1518 			continue;
1519 		}
1520 		length = bvec->bv_len;
1521 		buf_len = page->private >> 2;
1522 		start = page_offset(page) + bvec->bv_offset;
1523 		io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1524 	}
1525 	/* are we fully contained in this bio? */
1526 	if (buf_len <= length)
1527 		return 1;
1528 
1529 	ret = extent_range_uptodate(io_tree, start + length,
1530 				    start + buf_len - 1);
1531 	return ret;
1532 }
1533 
1534 /*
1535  * called by the kthread helper functions to finally call the bio end_io
1536  * functions.  This is where read checksum verification actually happens
1537  */
1538 static void end_workqueue_fn(struct btrfs_work *work)
1539 {
1540 	struct bio *bio;
1541 	struct end_io_wq *end_io_wq;
1542 	struct btrfs_fs_info *fs_info;
1543 	int error;
1544 
1545 	end_io_wq = container_of(work, struct end_io_wq, work);
1546 	bio = end_io_wq->bio;
1547 	fs_info = end_io_wq->info;
1548 
1549 	/* metadata bio reads are special because the whole tree block must
1550 	 * be checksummed at once.  This makes sure the entire block is in
1551 	 * ram and up to date before trying to verify things.  For
1552 	 * blocksize <= pagesize, it is basically a noop
1553 	 */
1554 	if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1555 	    !bio_ready_for_csum(bio)) {
1556 		btrfs_queue_worker(&fs_info->endio_meta_workers,
1557 				   &end_io_wq->work);
1558 		return;
1559 	}
1560 	error = end_io_wq->error;
1561 	bio->bi_private = end_io_wq->private;
1562 	bio->bi_end_io = end_io_wq->end_io;
1563 	kfree(end_io_wq);
1564 	bio_endio(bio, error);
1565 }
1566 
1567 static int cleaner_kthread(void *arg)
1568 {
1569 	struct btrfs_root *root = arg;
1570 
1571 	do {
1572 		vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1573 
1574 		if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1575 		    mutex_trylock(&root->fs_info->cleaner_mutex)) {
1576 			btrfs_run_delayed_iputs(root);
1577 			btrfs_clean_old_snapshots(root);
1578 			mutex_unlock(&root->fs_info->cleaner_mutex);
1579 			btrfs_run_defrag_inodes(root->fs_info);
1580 		}
1581 
1582 		if (freezing(current)) {
1583 			refrigerator();
1584 		} else {
1585 			set_current_state(TASK_INTERRUPTIBLE);
1586 			if (!kthread_should_stop())
1587 				schedule();
1588 			__set_current_state(TASK_RUNNING);
1589 		}
1590 	} while (!kthread_should_stop());
1591 	return 0;
1592 }
1593 
1594 static int transaction_kthread(void *arg)
1595 {
1596 	struct btrfs_root *root = arg;
1597 	struct btrfs_trans_handle *trans;
1598 	struct btrfs_transaction *cur;
1599 	u64 transid;
1600 	unsigned long now;
1601 	unsigned long delay;
1602 	int ret;
1603 
1604 	do {
1605 		delay = HZ * 30;
1606 		vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1607 		mutex_lock(&root->fs_info->transaction_kthread_mutex);
1608 
1609 		spin_lock(&root->fs_info->trans_lock);
1610 		cur = root->fs_info->running_transaction;
1611 		if (!cur) {
1612 			spin_unlock(&root->fs_info->trans_lock);
1613 			goto sleep;
1614 		}
1615 
1616 		now = get_seconds();
1617 		if (!cur->blocked &&
1618 		    (now < cur->start_time || now - cur->start_time < 30)) {
1619 			spin_unlock(&root->fs_info->trans_lock);
1620 			delay = HZ * 5;
1621 			goto sleep;
1622 		}
1623 		transid = cur->transid;
1624 		spin_unlock(&root->fs_info->trans_lock);
1625 
1626 		trans = btrfs_join_transaction(root);
1627 		BUG_ON(IS_ERR(trans));
1628 		if (transid == trans->transid) {
1629 			ret = btrfs_commit_transaction(trans, root);
1630 			BUG_ON(ret);
1631 		} else {
1632 			btrfs_end_transaction(trans, root);
1633 		}
1634 sleep:
1635 		wake_up_process(root->fs_info->cleaner_kthread);
1636 		mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1637 
1638 		if (freezing(current)) {
1639 			refrigerator();
1640 		} else {
1641 			set_current_state(TASK_INTERRUPTIBLE);
1642 			if (!kthread_should_stop() &&
1643 			    !btrfs_transaction_blocked(root->fs_info))
1644 				schedule_timeout(delay);
1645 			__set_current_state(TASK_RUNNING);
1646 		}
1647 	} while (!kthread_should_stop());
1648 	return 0;
1649 }
1650 
1651 /*
1652  * this will find the highest generation in the array of
1653  * root backups.  The index of the highest array is returned,
1654  * or -1 if we can't find anything.
1655  *
1656  * We check to make sure the array is valid by comparing the
1657  * generation of the latest  root in the array with the generation
1658  * in the super block.  If they don't match we pitch it.
1659  */
1660 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1661 {
1662 	u64 cur;
1663 	int newest_index = -1;
1664 	struct btrfs_root_backup *root_backup;
1665 	int i;
1666 
1667 	for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1668 		root_backup = info->super_copy->super_roots + i;
1669 		cur = btrfs_backup_tree_root_gen(root_backup);
1670 		if (cur == newest_gen)
1671 			newest_index = i;
1672 	}
1673 
1674 	/* check to see if we actually wrapped around */
1675 	if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1676 		root_backup = info->super_copy->super_roots;
1677 		cur = btrfs_backup_tree_root_gen(root_backup);
1678 		if (cur == newest_gen)
1679 			newest_index = 0;
1680 	}
1681 	return newest_index;
1682 }
1683 
1684 
1685 /*
1686  * find the oldest backup so we know where to store new entries
1687  * in the backup array.  This will set the backup_root_index
1688  * field in the fs_info struct
1689  */
1690 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1691 				     u64 newest_gen)
1692 {
1693 	int newest_index = -1;
1694 
1695 	newest_index = find_newest_super_backup(info, newest_gen);
1696 	/* if there was garbage in there, just move along */
1697 	if (newest_index == -1) {
1698 		info->backup_root_index = 0;
1699 	} else {
1700 		info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1701 	}
1702 }
1703 
1704 /*
1705  * copy all the root pointers into the super backup array.
1706  * this will bump the backup pointer by one when it is
1707  * done
1708  */
1709 static void backup_super_roots(struct btrfs_fs_info *info)
1710 {
1711 	int next_backup;
1712 	struct btrfs_root_backup *root_backup;
1713 	int last_backup;
1714 
1715 	next_backup = info->backup_root_index;
1716 	last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1717 		BTRFS_NUM_BACKUP_ROOTS;
1718 
1719 	/*
1720 	 * just overwrite the last backup if we're at the same generation
1721 	 * this happens only at umount
1722 	 */
1723 	root_backup = info->super_for_commit->super_roots + last_backup;
1724 	if (btrfs_backup_tree_root_gen(root_backup) ==
1725 	    btrfs_header_generation(info->tree_root->node))
1726 		next_backup = last_backup;
1727 
1728 	root_backup = info->super_for_commit->super_roots + next_backup;
1729 
1730 	/*
1731 	 * make sure all of our padding and empty slots get zero filled
1732 	 * regardless of which ones we use today
1733 	 */
1734 	memset(root_backup, 0, sizeof(*root_backup));
1735 
1736 	info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1737 
1738 	btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1739 	btrfs_set_backup_tree_root_gen(root_backup,
1740 			       btrfs_header_generation(info->tree_root->node));
1741 
1742 	btrfs_set_backup_tree_root_level(root_backup,
1743 			       btrfs_header_level(info->tree_root->node));
1744 
1745 	btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1746 	btrfs_set_backup_chunk_root_gen(root_backup,
1747 			       btrfs_header_generation(info->chunk_root->node));
1748 	btrfs_set_backup_chunk_root_level(root_backup,
1749 			       btrfs_header_level(info->chunk_root->node));
1750 
1751 	btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1752 	btrfs_set_backup_extent_root_gen(root_backup,
1753 			       btrfs_header_generation(info->extent_root->node));
1754 	btrfs_set_backup_extent_root_level(root_backup,
1755 			       btrfs_header_level(info->extent_root->node));
1756 
1757 	/*
1758 	 * we might commit during log recovery, which happens before we set
1759 	 * the fs_root.  Make sure it is valid before we fill it in.
1760 	 */
1761 	if (info->fs_root && info->fs_root->node) {
1762 		btrfs_set_backup_fs_root(root_backup,
1763 					 info->fs_root->node->start);
1764 		btrfs_set_backup_fs_root_gen(root_backup,
1765 			       btrfs_header_generation(info->fs_root->node));
1766 		btrfs_set_backup_fs_root_level(root_backup,
1767 			       btrfs_header_level(info->fs_root->node));
1768 	}
1769 
1770 	btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1771 	btrfs_set_backup_dev_root_gen(root_backup,
1772 			       btrfs_header_generation(info->dev_root->node));
1773 	btrfs_set_backup_dev_root_level(root_backup,
1774 				       btrfs_header_level(info->dev_root->node));
1775 
1776 	btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1777 	btrfs_set_backup_csum_root_gen(root_backup,
1778 			       btrfs_header_generation(info->csum_root->node));
1779 	btrfs_set_backup_csum_root_level(root_backup,
1780 			       btrfs_header_level(info->csum_root->node));
1781 
1782 	btrfs_set_backup_total_bytes(root_backup,
1783 			     btrfs_super_total_bytes(info->super_copy));
1784 	btrfs_set_backup_bytes_used(root_backup,
1785 			     btrfs_super_bytes_used(info->super_copy));
1786 	btrfs_set_backup_num_devices(root_backup,
1787 			     btrfs_super_num_devices(info->super_copy));
1788 
1789 	/*
1790 	 * if we don't copy this out to the super_copy, it won't get remembered
1791 	 * for the next commit
1792 	 */
1793 	memcpy(&info->super_copy->super_roots,
1794 	       &info->super_for_commit->super_roots,
1795 	       sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1796 }
1797 
1798 /*
1799  * this copies info out of the root backup array and back into
1800  * the in-memory super block.  It is meant to help iterate through
1801  * the array, so you send it the number of backups you've already
1802  * tried and the last backup index you used.
1803  *
1804  * this returns -1 when it has tried all the backups
1805  */
1806 static noinline int next_root_backup(struct btrfs_fs_info *info,
1807 				     struct btrfs_super_block *super,
1808 				     int *num_backups_tried, int *backup_index)
1809 {
1810 	struct btrfs_root_backup *root_backup;
1811 	int newest = *backup_index;
1812 
1813 	if (*num_backups_tried == 0) {
1814 		u64 gen = btrfs_super_generation(super);
1815 
1816 		newest = find_newest_super_backup(info, gen);
1817 		if (newest == -1)
1818 			return -1;
1819 
1820 		*backup_index = newest;
1821 		*num_backups_tried = 1;
1822 	} else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1823 		/* we've tried all the backups, all done */
1824 		return -1;
1825 	} else {
1826 		/* jump to the next oldest backup */
1827 		newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1828 			BTRFS_NUM_BACKUP_ROOTS;
1829 		*backup_index = newest;
1830 		*num_backups_tried += 1;
1831 	}
1832 	root_backup = super->super_roots + newest;
1833 
1834 	btrfs_set_super_generation(super,
1835 				   btrfs_backup_tree_root_gen(root_backup));
1836 	btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1837 	btrfs_set_super_root_level(super,
1838 				   btrfs_backup_tree_root_level(root_backup));
1839 	btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1840 
1841 	/*
1842 	 * fixme: the total bytes and num_devices need to match or we should
1843 	 * need a fsck
1844 	 */
1845 	btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1846 	btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1847 	return 0;
1848 }
1849 
1850 /* helper to cleanup tree roots */
1851 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
1852 {
1853 	free_extent_buffer(info->tree_root->node);
1854 	free_extent_buffer(info->tree_root->commit_root);
1855 	free_extent_buffer(info->dev_root->node);
1856 	free_extent_buffer(info->dev_root->commit_root);
1857 	free_extent_buffer(info->extent_root->node);
1858 	free_extent_buffer(info->extent_root->commit_root);
1859 	free_extent_buffer(info->csum_root->node);
1860 	free_extent_buffer(info->csum_root->commit_root);
1861 
1862 	info->tree_root->node = NULL;
1863 	info->tree_root->commit_root = NULL;
1864 	info->dev_root->node = NULL;
1865 	info->dev_root->commit_root = NULL;
1866 	info->extent_root->node = NULL;
1867 	info->extent_root->commit_root = NULL;
1868 	info->csum_root->node = NULL;
1869 	info->csum_root->commit_root = NULL;
1870 
1871 	if (chunk_root) {
1872 		free_extent_buffer(info->chunk_root->node);
1873 		free_extent_buffer(info->chunk_root->commit_root);
1874 		info->chunk_root->node = NULL;
1875 		info->chunk_root->commit_root = NULL;
1876 	}
1877 }
1878 
1879 
1880 struct btrfs_root *open_ctree(struct super_block *sb,
1881 			      struct btrfs_fs_devices *fs_devices,
1882 			      char *options)
1883 {
1884 	u32 sectorsize;
1885 	u32 nodesize;
1886 	u32 leafsize;
1887 	u32 blocksize;
1888 	u32 stripesize;
1889 	u64 generation;
1890 	u64 features;
1891 	struct btrfs_key location;
1892 	struct buffer_head *bh;
1893 	struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1894 						 GFP_NOFS);
1895 	struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1896 						 GFP_NOFS);
1897 	struct btrfs_root *tree_root = btrfs_sb(sb);
1898 	struct btrfs_fs_info *fs_info = NULL;
1899 	struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1900 						GFP_NOFS);
1901 	struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1902 					      GFP_NOFS);
1903 	struct btrfs_root *log_tree_root;
1904 
1905 	int ret;
1906 	int err = -EINVAL;
1907 	int num_backups_tried = 0;
1908 	int backup_index = 0;
1909 
1910 	struct btrfs_super_block *disk_super;
1911 
1912 	if (!extent_root || !tree_root || !tree_root->fs_info ||
1913 	    !chunk_root || !dev_root || !csum_root) {
1914 		err = -ENOMEM;
1915 		goto fail;
1916 	}
1917 	fs_info = tree_root->fs_info;
1918 
1919 	ret = init_srcu_struct(&fs_info->subvol_srcu);
1920 	if (ret) {
1921 		err = ret;
1922 		goto fail;
1923 	}
1924 
1925 	ret = setup_bdi(fs_info, &fs_info->bdi);
1926 	if (ret) {
1927 		err = ret;
1928 		goto fail_srcu;
1929 	}
1930 
1931 	fs_info->btree_inode = new_inode(sb);
1932 	if (!fs_info->btree_inode) {
1933 		err = -ENOMEM;
1934 		goto fail_bdi;
1935 	}
1936 
1937 	mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1938 
1939 	INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1940 	INIT_LIST_HEAD(&fs_info->trans_list);
1941 	INIT_LIST_HEAD(&fs_info->dead_roots);
1942 	INIT_LIST_HEAD(&fs_info->delayed_iputs);
1943 	INIT_LIST_HEAD(&fs_info->hashers);
1944 	INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1945 	INIT_LIST_HEAD(&fs_info->ordered_operations);
1946 	INIT_LIST_HEAD(&fs_info->caching_block_groups);
1947 	spin_lock_init(&fs_info->delalloc_lock);
1948 	spin_lock_init(&fs_info->trans_lock);
1949 	spin_lock_init(&fs_info->ref_cache_lock);
1950 	spin_lock_init(&fs_info->fs_roots_radix_lock);
1951 	spin_lock_init(&fs_info->delayed_iput_lock);
1952 	spin_lock_init(&fs_info->defrag_inodes_lock);
1953 	spin_lock_init(&fs_info->free_chunk_lock);
1954 	mutex_init(&fs_info->reloc_mutex);
1955 
1956 	init_completion(&fs_info->kobj_unregister);
1957 	fs_info->tree_root = tree_root;
1958 	fs_info->extent_root = extent_root;
1959 	fs_info->csum_root = csum_root;
1960 	fs_info->chunk_root = chunk_root;
1961 	fs_info->dev_root = dev_root;
1962 	fs_info->fs_devices = fs_devices;
1963 	INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1964 	INIT_LIST_HEAD(&fs_info->space_info);
1965 	btrfs_mapping_init(&fs_info->mapping_tree);
1966 	btrfs_init_block_rsv(&fs_info->global_block_rsv);
1967 	btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1968 	btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1969 	btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1970 	btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1971 	btrfs_init_block_rsv(&fs_info->delayed_block_rsv);
1972 	atomic_set(&fs_info->nr_async_submits, 0);
1973 	atomic_set(&fs_info->async_delalloc_pages, 0);
1974 	atomic_set(&fs_info->async_submit_draining, 0);
1975 	atomic_set(&fs_info->nr_async_bios, 0);
1976 	atomic_set(&fs_info->defrag_running, 0);
1977 	fs_info->sb = sb;
1978 	fs_info->max_inline = 8192 * 1024;
1979 	fs_info->metadata_ratio = 0;
1980 	fs_info->defrag_inodes = RB_ROOT;
1981 	fs_info->trans_no_join = 0;
1982 	fs_info->free_chunk_space = 0;
1983 
1984 	/* readahead state */
1985 	INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
1986 	spin_lock_init(&fs_info->reada_lock);
1987 
1988 	fs_info->thread_pool_size = min_t(unsigned long,
1989 					  num_online_cpus() + 2, 8);
1990 
1991 	INIT_LIST_HEAD(&fs_info->ordered_extents);
1992 	spin_lock_init(&fs_info->ordered_extent_lock);
1993 	fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
1994 					GFP_NOFS);
1995 	if (!fs_info->delayed_root) {
1996 		err = -ENOMEM;
1997 		goto fail_iput;
1998 	}
1999 	btrfs_init_delayed_root(fs_info->delayed_root);
2000 
2001 	mutex_init(&fs_info->scrub_lock);
2002 	atomic_set(&fs_info->scrubs_running, 0);
2003 	atomic_set(&fs_info->scrub_pause_req, 0);
2004 	atomic_set(&fs_info->scrubs_paused, 0);
2005 	atomic_set(&fs_info->scrub_cancel_req, 0);
2006 	init_waitqueue_head(&fs_info->scrub_pause_wait);
2007 	init_rwsem(&fs_info->scrub_super_lock);
2008 	fs_info->scrub_workers_refcnt = 0;
2009 
2010 	sb->s_blocksize = 4096;
2011 	sb->s_blocksize_bits = blksize_bits(4096);
2012 	sb->s_bdi = &fs_info->bdi;
2013 
2014 	fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2015 	set_nlink(fs_info->btree_inode, 1);
2016 	/*
2017 	 * we set the i_size on the btree inode to the max possible int.
2018 	 * the real end of the address space is determined by all of
2019 	 * the devices in the system
2020 	 */
2021 	fs_info->btree_inode->i_size = OFFSET_MAX;
2022 	fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2023 	fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2024 
2025 	RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2026 	extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2027 			     fs_info->btree_inode->i_mapping);
2028 	extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2029 
2030 	BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2031 
2032 	BTRFS_I(fs_info->btree_inode)->root = tree_root;
2033 	memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2034 	       sizeof(struct btrfs_key));
2035 	BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
2036 	insert_inode_hash(fs_info->btree_inode);
2037 
2038 	spin_lock_init(&fs_info->block_group_cache_lock);
2039 	fs_info->block_group_cache_tree = RB_ROOT;
2040 
2041 	extent_io_tree_init(&fs_info->freed_extents[0],
2042 			     fs_info->btree_inode->i_mapping);
2043 	extent_io_tree_init(&fs_info->freed_extents[1],
2044 			     fs_info->btree_inode->i_mapping);
2045 	fs_info->pinned_extents = &fs_info->freed_extents[0];
2046 	fs_info->do_barriers = 1;
2047 
2048 
2049 	mutex_init(&fs_info->ordered_operations_mutex);
2050 	mutex_init(&fs_info->tree_log_mutex);
2051 	mutex_init(&fs_info->chunk_mutex);
2052 	mutex_init(&fs_info->transaction_kthread_mutex);
2053 	mutex_init(&fs_info->cleaner_mutex);
2054 	mutex_init(&fs_info->volume_mutex);
2055 	init_rwsem(&fs_info->extent_commit_sem);
2056 	init_rwsem(&fs_info->cleanup_work_sem);
2057 	init_rwsem(&fs_info->subvol_sem);
2058 
2059 	btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2060 	btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2061 
2062 	init_waitqueue_head(&fs_info->transaction_throttle);
2063 	init_waitqueue_head(&fs_info->transaction_wait);
2064 	init_waitqueue_head(&fs_info->transaction_blocked_wait);
2065 	init_waitqueue_head(&fs_info->async_submit_wait);
2066 
2067 	__setup_root(4096, 4096, 4096, 4096, tree_root,
2068 		     fs_info, BTRFS_ROOT_TREE_OBJECTID);
2069 
2070 	bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2071 	if (!bh) {
2072 		err = -EINVAL;
2073 		goto fail_alloc;
2074 	}
2075 
2076 	memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2077 	memcpy(fs_info->super_for_commit, fs_info->super_copy,
2078 	       sizeof(*fs_info->super_for_commit));
2079 	brelse(bh);
2080 
2081 	memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2082 
2083 	disk_super = fs_info->super_copy;
2084 	if (!btrfs_super_root(disk_super))
2085 		goto fail_alloc;
2086 
2087 	/* check FS state, whether FS is broken. */
2088 	fs_info->fs_state |= btrfs_super_flags(disk_super);
2089 
2090 	btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2091 
2092 	/*
2093 	 * run through our array of backup supers and setup
2094 	 * our ring pointer to the oldest one
2095 	 */
2096 	generation = btrfs_super_generation(disk_super);
2097 	find_oldest_super_backup(fs_info, generation);
2098 
2099 	/*
2100 	 * In the long term, we'll store the compression type in the super
2101 	 * block, and it'll be used for per file compression control.
2102 	 */
2103 	fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2104 
2105 	ret = btrfs_parse_options(tree_root, options);
2106 	if (ret) {
2107 		err = ret;
2108 		goto fail_alloc;
2109 	}
2110 
2111 	features = btrfs_super_incompat_flags(disk_super) &
2112 		~BTRFS_FEATURE_INCOMPAT_SUPP;
2113 	if (features) {
2114 		printk(KERN_ERR "BTRFS: couldn't mount because of "
2115 		       "unsupported optional features (%Lx).\n",
2116 		       (unsigned long long)features);
2117 		err = -EINVAL;
2118 		goto fail_alloc;
2119 	}
2120 
2121 	features = btrfs_super_incompat_flags(disk_super);
2122 	features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2123 	if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
2124 		features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2125 	btrfs_set_super_incompat_flags(disk_super, features);
2126 
2127 	features = btrfs_super_compat_ro_flags(disk_super) &
2128 		~BTRFS_FEATURE_COMPAT_RO_SUPP;
2129 	if (!(sb->s_flags & MS_RDONLY) && features) {
2130 		printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2131 		       "unsupported option features (%Lx).\n",
2132 		       (unsigned long long)features);
2133 		err = -EINVAL;
2134 		goto fail_alloc;
2135 	}
2136 
2137 	btrfs_init_workers(&fs_info->generic_worker,
2138 			   "genwork", 1, NULL);
2139 
2140 	btrfs_init_workers(&fs_info->workers, "worker",
2141 			   fs_info->thread_pool_size,
2142 			   &fs_info->generic_worker);
2143 
2144 	btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2145 			   fs_info->thread_pool_size,
2146 			   &fs_info->generic_worker);
2147 
2148 	btrfs_init_workers(&fs_info->submit_workers, "submit",
2149 			   min_t(u64, fs_devices->num_devices,
2150 			   fs_info->thread_pool_size),
2151 			   &fs_info->generic_worker);
2152 
2153 	btrfs_init_workers(&fs_info->caching_workers, "cache",
2154 			   2, &fs_info->generic_worker);
2155 
2156 	/* a higher idle thresh on the submit workers makes it much more
2157 	 * likely that bios will be send down in a sane order to the
2158 	 * devices
2159 	 */
2160 	fs_info->submit_workers.idle_thresh = 64;
2161 
2162 	fs_info->workers.idle_thresh = 16;
2163 	fs_info->workers.ordered = 1;
2164 
2165 	fs_info->delalloc_workers.idle_thresh = 2;
2166 	fs_info->delalloc_workers.ordered = 1;
2167 
2168 	btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2169 			   &fs_info->generic_worker);
2170 	btrfs_init_workers(&fs_info->endio_workers, "endio",
2171 			   fs_info->thread_pool_size,
2172 			   &fs_info->generic_worker);
2173 	btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2174 			   fs_info->thread_pool_size,
2175 			   &fs_info->generic_worker);
2176 	btrfs_init_workers(&fs_info->endio_meta_write_workers,
2177 			   "endio-meta-write", fs_info->thread_pool_size,
2178 			   &fs_info->generic_worker);
2179 	btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2180 			   fs_info->thread_pool_size,
2181 			   &fs_info->generic_worker);
2182 	btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2183 			   1, &fs_info->generic_worker);
2184 	btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2185 			   fs_info->thread_pool_size,
2186 			   &fs_info->generic_worker);
2187 	btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2188 			   fs_info->thread_pool_size,
2189 			   &fs_info->generic_worker);
2190 
2191 	/*
2192 	 * endios are largely parallel and should have a very
2193 	 * low idle thresh
2194 	 */
2195 	fs_info->endio_workers.idle_thresh = 4;
2196 	fs_info->endio_meta_workers.idle_thresh = 4;
2197 
2198 	fs_info->endio_write_workers.idle_thresh = 2;
2199 	fs_info->endio_meta_write_workers.idle_thresh = 2;
2200 	fs_info->readahead_workers.idle_thresh = 2;
2201 
2202 	btrfs_start_workers(&fs_info->workers, 1);
2203 	btrfs_start_workers(&fs_info->generic_worker, 1);
2204 	btrfs_start_workers(&fs_info->submit_workers, 1);
2205 	btrfs_start_workers(&fs_info->delalloc_workers, 1);
2206 	btrfs_start_workers(&fs_info->fixup_workers, 1);
2207 	btrfs_start_workers(&fs_info->endio_workers, 1);
2208 	btrfs_start_workers(&fs_info->endio_meta_workers, 1);
2209 	btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
2210 	btrfs_start_workers(&fs_info->endio_write_workers, 1);
2211 	btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
2212 	btrfs_start_workers(&fs_info->delayed_workers, 1);
2213 	btrfs_start_workers(&fs_info->caching_workers, 1);
2214 	btrfs_start_workers(&fs_info->readahead_workers, 1);
2215 
2216 	fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2217 	fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2218 				    4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2219 
2220 	nodesize = btrfs_super_nodesize(disk_super);
2221 	leafsize = btrfs_super_leafsize(disk_super);
2222 	sectorsize = btrfs_super_sectorsize(disk_super);
2223 	stripesize = btrfs_super_stripesize(disk_super);
2224 	tree_root->nodesize = nodesize;
2225 	tree_root->leafsize = leafsize;
2226 	tree_root->sectorsize = sectorsize;
2227 	tree_root->stripesize = stripesize;
2228 
2229 	sb->s_blocksize = sectorsize;
2230 	sb->s_blocksize_bits = blksize_bits(sectorsize);
2231 
2232 	if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
2233 		    sizeof(disk_super->magic))) {
2234 		printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2235 		goto fail_sb_buffer;
2236 	}
2237 
2238 	mutex_lock(&fs_info->chunk_mutex);
2239 	ret = btrfs_read_sys_array(tree_root);
2240 	mutex_unlock(&fs_info->chunk_mutex);
2241 	if (ret) {
2242 		printk(KERN_WARNING "btrfs: failed to read the system "
2243 		       "array on %s\n", sb->s_id);
2244 		goto fail_sb_buffer;
2245 	}
2246 
2247 	blocksize = btrfs_level_size(tree_root,
2248 				     btrfs_super_chunk_root_level(disk_super));
2249 	generation = btrfs_super_chunk_root_generation(disk_super);
2250 
2251 	__setup_root(nodesize, leafsize, sectorsize, stripesize,
2252 		     chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2253 
2254 	chunk_root->node = read_tree_block(chunk_root,
2255 					   btrfs_super_chunk_root(disk_super),
2256 					   blocksize, generation);
2257 	BUG_ON(!chunk_root->node);
2258 	if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2259 		printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2260 		       sb->s_id);
2261 		goto fail_tree_roots;
2262 	}
2263 	btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2264 	chunk_root->commit_root = btrfs_root_node(chunk_root);
2265 
2266 	read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2267 	   (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2268 	   BTRFS_UUID_SIZE);
2269 
2270 	mutex_lock(&fs_info->chunk_mutex);
2271 	ret = btrfs_read_chunk_tree(chunk_root);
2272 	mutex_unlock(&fs_info->chunk_mutex);
2273 	if (ret) {
2274 		printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2275 		       sb->s_id);
2276 		goto fail_tree_roots;
2277 	}
2278 
2279 	btrfs_close_extra_devices(fs_devices);
2280 
2281 retry_root_backup:
2282 	blocksize = btrfs_level_size(tree_root,
2283 				     btrfs_super_root_level(disk_super));
2284 	generation = btrfs_super_generation(disk_super);
2285 
2286 	tree_root->node = read_tree_block(tree_root,
2287 					  btrfs_super_root(disk_super),
2288 					  blocksize, generation);
2289 	if (!tree_root->node ||
2290 	    !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2291 		printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2292 		       sb->s_id);
2293 
2294 		goto recovery_tree_root;
2295 	}
2296 
2297 	btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2298 	tree_root->commit_root = btrfs_root_node(tree_root);
2299 
2300 	ret = find_and_setup_root(tree_root, fs_info,
2301 				  BTRFS_EXTENT_TREE_OBJECTID, extent_root);
2302 	if (ret)
2303 		goto recovery_tree_root;
2304 	extent_root->track_dirty = 1;
2305 
2306 	ret = find_and_setup_root(tree_root, fs_info,
2307 				  BTRFS_DEV_TREE_OBJECTID, dev_root);
2308 	if (ret)
2309 		goto recovery_tree_root;
2310 	dev_root->track_dirty = 1;
2311 
2312 	ret = find_and_setup_root(tree_root, fs_info,
2313 				  BTRFS_CSUM_TREE_OBJECTID, csum_root);
2314 	if (ret)
2315 		goto recovery_tree_root;
2316 
2317 	csum_root->track_dirty = 1;
2318 
2319 	fs_info->generation = generation;
2320 	fs_info->last_trans_committed = generation;
2321 	fs_info->data_alloc_profile = (u64)-1;
2322 	fs_info->metadata_alloc_profile = (u64)-1;
2323 	fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
2324 
2325 	ret = btrfs_init_space_info(fs_info);
2326 	if (ret) {
2327 		printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2328 		goto fail_block_groups;
2329 	}
2330 
2331 	ret = btrfs_read_block_groups(extent_root);
2332 	if (ret) {
2333 		printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2334 		goto fail_block_groups;
2335 	}
2336 
2337 	fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2338 					       "btrfs-cleaner");
2339 	if (IS_ERR(fs_info->cleaner_kthread))
2340 		goto fail_block_groups;
2341 
2342 	fs_info->transaction_kthread = kthread_run(transaction_kthread,
2343 						   tree_root,
2344 						   "btrfs-transaction");
2345 	if (IS_ERR(fs_info->transaction_kthread))
2346 		goto fail_cleaner;
2347 
2348 	if (!btrfs_test_opt(tree_root, SSD) &&
2349 	    !btrfs_test_opt(tree_root, NOSSD) &&
2350 	    !fs_info->fs_devices->rotating) {
2351 		printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2352 		       "mode\n");
2353 		btrfs_set_opt(fs_info->mount_opt, SSD);
2354 	}
2355 
2356 	/* do not make disk changes in broken FS */
2357 	if (btrfs_super_log_root(disk_super) != 0 &&
2358 	    !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
2359 		u64 bytenr = btrfs_super_log_root(disk_super);
2360 
2361 		if (fs_devices->rw_devices == 0) {
2362 			printk(KERN_WARNING "Btrfs log replay required "
2363 			       "on RO media\n");
2364 			err = -EIO;
2365 			goto fail_trans_kthread;
2366 		}
2367 		blocksize =
2368 		     btrfs_level_size(tree_root,
2369 				      btrfs_super_log_root_level(disk_super));
2370 
2371 		log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
2372 		if (!log_tree_root) {
2373 			err = -ENOMEM;
2374 			goto fail_trans_kthread;
2375 		}
2376 
2377 		__setup_root(nodesize, leafsize, sectorsize, stripesize,
2378 			     log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2379 
2380 		log_tree_root->node = read_tree_block(tree_root, bytenr,
2381 						      blocksize,
2382 						      generation + 1);
2383 		ret = btrfs_recover_log_trees(log_tree_root);
2384 		BUG_ON(ret);
2385 
2386 		if (sb->s_flags & MS_RDONLY) {
2387 			ret =  btrfs_commit_super(tree_root);
2388 			BUG_ON(ret);
2389 		}
2390 	}
2391 
2392 	ret = btrfs_find_orphan_roots(tree_root);
2393 	BUG_ON(ret);
2394 
2395 	if (!(sb->s_flags & MS_RDONLY)) {
2396 		ret = btrfs_cleanup_fs_roots(fs_info);
2397 		BUG_ON(ret);
2398 
2399 		ret = btrfs_recover_relocation(tree_root);
2400 		if (ret < 0) {
2401 			printk(KERN_WARNING
2402 			       "btrfs: failed to recover relocation\n");
2403 			err = -EINVAL;
2404 			goto fail_trans_kthread;
2405 		}
2406 	}
2407 
2408 	location.objectid = BTRFS_FS_TREE_OBJECTID;
2409 	location.type = BTRFS_ROOT_ITEM_KEY;
2410 	location.offset = (u64)-1;
2411 
2412 	fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2413 	if (!fs_info->fs_root)
2414 		goto fail_trans_kthread;
2415 	if (IS_ERR(fs_info->fs_root)) {
2416 		err = PTR_ERR(fs_info->fs_root);
2417 		goto fail_trans_kthread;
2418 	}
2419 
2420 	if (!(sb->s_flags & MS_RDONLY)) {
2421 		down_read(&fs_info->cleanup_work_sem);
2422 		err = btrfs_orphan_cleanup(fs_info->fs_root);
2423 		if (!err)
2424 			err = btrfs_orphan_cleanup(fs_info->tree_root);
2425 		up_read(&fs_info->cleanup_work_sem);
2426 		if (err) {
2427 			close_ctree(tree_root);
2428 			return ERR_PTR(err);
2429 		}
2430 	}
2431 
2432 	return tree_root;
2433 
2434 fail_trans_kthread:
2435 	kthread_stop(fs_info->transaction_kthread);
2436 fail_cleaner:
2437 	kthread_stop(fs_info->cleaner_kthread);
2438 
2439 	/*
2440 	 * make sure we're done with the btree inode before we stop our
2441 	 * kthreads
2442 	 */
2443 	filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2444 	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2445 
2446 fail_block_groups:
2447 	btrfs_free_block_groups(fs_info);
2448 
2449 fail_tree_roots:
2450 	free_root_pointers(fs_info, 1);
2451 
2452 fail_sb_buffer:
2453 	btrfs_stop_workers(&fs_info->generic_worker);
2454 	btrfs_stop_workers(&fs_info->readahead_workers);
2455 	btrfs_stop_workers(&fs_info->fixup_workers);
2456 	btrfs_stop_workers(&fs_info->delalloc_workers);
2457 	btrfs_stop_workers(&fs_info->workers);
2458 	btrfs_stop_workers(&fs_info->endio_workers);
2459 	btrfs_stop_workers(&fs_info->endio_meta_workers);
2460 	btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2461 	btrfs_stop_workers(&fs_info->endio_write_workers);
2462 	btrfs_stop_workers(&fs_info->endio_freespace_worker);
2463 	btrfs_stop_workers(&fs_info->submit_workers);
2464 	btrfs_stop_workers(&fs_info->delayed_workers);
2465 	btrfs_stop_workers(&fs_info->caching_workers);
2466 fail_alloc:
2467 fail_iput:
2468 	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2469 	iput(fs_info->btree_inode);
2470 
2471 	btrfs_close_devices(fs_info->fs_devices);
2472 	btrfs_mapping_tree_free(&fs_info->mapping_tree);
2473 fail_bdi:
2474 	bdi_destroy(&fs_info->bdi);
2475 fail_srcu:
2476 	cleanup_srcu_struct(&fs_info->subvol_srcu);
2477 fail:
2478 	free_fs_info(fs_info);
2479 	return ERR_PTR(err);
2480 
2481 recovery_tree_root:
2482 
2483 	if (!btrfs_test_opt(tree_root, RECOVERY))
2484 		goto fail_tree_roots;
2485 
2486 	free_root_pointers(fs_info, 0);
2487 
2488 	/* don't use the log in recovery mode, it won't be valid */
2489 	btrfs_set_super_log_root(disk_super, 0);
2490 
2491 	/* we can't trust the free space cache either */
2492 	btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2493 
2494 	ret = next_root_backup(fs_info, fs_info->super_copy,
2495 			       &num_backups_tried, &backup_index);
2496 	if (ret == -1)
2497 		goto fail_block_groups;
2498 	goto retry_root_backup;
2499 }
2500 
2501 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2502 {
2503 	char b[BDEVNAME_SIZE];
2504 
2505 	if (uptodate) {
2506 		set_buffer_uptodate(bh);
2507 	} else {
2508 		printk_ratelimited(KERN_WARNING "lost page write due to "
2509 					"I/O error on %s\n",
2510 				       bdevname(bh->b_bdev, b));
2511 		/* note, we dont' set_buffer_write_io_error because we have
2512 		 * our own ways of dealing with the IO errors
2513 		 */
2514 		clear_buffer_uptodate(bh);
2515 	}
2516 	unlock_buffer(bh);
2517 	put_bh(bh);
2518 }
2519 
2520 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2521 {
2522 	struct buffer_head *bh;
2523 	struct buffer_head *latest = NULL;
2524 	struct btrfs_super_block *super;
2525 	int i;
2526 	u64 transid = 0;
2527 	u64 bytenr;
2528 
2529 	/* we would like to check all the supers, but that would make
2530 	 * a btrfs mount succeed after a mkfs from a different FS.
2531 	 * So, we need to add a special mount option to scan for
2532 	 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2533 	 */
2534 	for (i = 0; i < 1; i++) {
2535 		bytenr = btrfs_sb_offset(i);
2536 		if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2537 			break;
2538 		bh = __bread(bdev, bytenr / 4096, 4096);
2539 		if (!bh)
2540 			continue;
2541 
2542 		super = (struct btrfs_super_block *)bh->b_data;
2543 		if (btrfs_super_bytenr(super) != bytenr ||
2544 		    strncmp((char *)(&super->magic), BTRFS_MAGIC,
2545 			    sizeof(super->magic))) {
2546 			brelse(bh);
2547 			continue;
2548 		}
2549 
2550 		if (!latest || btrfs_super_generation(super) > transid) {
2551 			brelse(latest);
2552 			latest = bh;
2553 			transid = btrfs_super_generation(super);
2554 		} else {
2555 			brelse(bh);
2556 		}
2557 	}
2558 	return latest;
2559 }
2560 
2561 /*
2562  * this should be called twice, once with wait == 0 and
2563  * once with wait == 1.  When wait == 0 is done, all the buffer heads
2564  * we write are pinned.
2565  *
2566  * They are released when wait == 1 is done.
2567  * max_mirrors must be the same for both runs, and it indicates how
2568  * many supers on this one device should be written.
2569  *
2570  * max_mirrors == 0 means to write them all.
2571  */
2572 static int write_dev_supers(struct btrfs_device *device,
2573 			    struct btrfs_super_block *sb,
2574 			    int do_barriers, int wait, int max_mirrors)
2575 {
2576 	struct buffer_head *bh;
2577 	int i;
2578 	int ret;
2579 	int errors = 0;
2580 	u32 crc;
2581 	u64 bytenr;
2582 	int last_barrier = 0;
2583 
2584 	if (max_mirrors == 0)
2585 		max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2586 
2587 	/* make sure only the last submit_bh does a barrier */
2588 	if (do_barriers) {
2589 		for (i = 0; i < max_mirrors; i++) {
2590 			bytenr = btrfs_sb_offset(i);
2591 			if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2592 			    device->total_bytes)
2593 				break;
2594 			last_barrier = i;
2595 		}
2596 	}
2597 
2598 	for (i = 0; i < max_mirrors; i++) {
2599 		bytenr = btrfs_sb_offset(i);
2600 		if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2601 			break;
2602 
2603 		if (wait) {
2604 			bh = __find_get_block(device->bdev, bytenr / 4096,
2605 					      BTRFS_SUPER_INFO_SIZE);
2606 			BUG_ON(!bh);
2607 			wait_on_buffer(bh);
2608 			if (!buffer_uptodate(bh))
2609 				errors++;
2610 
2611 			/* drop our reference */
2612 			brelse(bh);
2613 
2614 			/* drop the reference from the wait == 0 run */
2615 			brelse(bh);
2616 			continue;
2617 		} else {
2618 			btrfs_set_super_bytenr(sb, bytenr);
2619 
2620 			crc = ~(u32)0;
2621 			crc = btrfs_csum_data(NULL, (char *)sb +
2622 					      BTRFS_CSUM_SIZE, crc,
2623 					      BTRFS_SUPER_INFO_SIZE -
2624 					      BTRFS_CSUM_SIZE);
2625 			btrfs_csum_final(crc, sb->csum);
2626 
2627 			/*
2628 			 * one reference for us, and we leave it for the
2629 			 * caller
2630 			 */
2631 			bh = __getblk(device->bdev, bytenr / 4096,
2632 				      BTRFS_SUPER_INFO_SIZE);
2633 			memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2634 
2635 			/* one reference for submit_bh */
2636 			get_bh(bh);
2637 
2638 			set_buffer_uptodate(bh);
2639 			lock_buffer(bh);
2640 			bh->b_end_io = btrfs_end_buffer_write_sync;
2641 		}
2642 
2643 		if (i == last_barrier && do_barriers)
2644 			ret = submit_bh(WRITE_FLUSH_FUA, bh);
2645 		else
2646 			ret = submit_bh(WRITE_SYNC, bh);
2647 
2648 		if (ret)
2649 			errors++;
2650 	}
2651 	return errors < i ? 0 : -1;
2652 }
2653 
2654 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2655 {
2656 	struct list_head *head;
2657 	struct btrfs_device *dev;
2658 	struct btrfs_super_block *sb;
2659 	struct btrfs_dev_item *dev_item;
2660 	int ret;
2661 	int do_barriers;
2662 	int max_errors;
2663 	int total_errors = 0;
2664 	u64 flags;
2665 
2666 	max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
2667 	do_barriers = !btrfs_test_opt(root, NOBARRIER);
2668 	backup_super_roots(root->fs_info);
2669 
2670 	sb = root->fs_info->super_for_commit;
2671 	dev_item = &sb->dev_item;
2672 
2673 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2674 	head = &root->fs_info->fs_devices->devices;
2675 	list_for_each_entry_rcu(dev, head, dev_list) {
2676 		if (!dev->bdev) {
2677 			total_errors++;
2678 			continue;
2679 		}
2680 		if (!dev->in_fs_metadata || !dev->writeable)
2681 			continue;
2682 
2683 		btrfs_set_stack_device_generation(dev_item, 0);
2684 		btrfs_set_stack_device_type(dev_item, dev->type);
2685 		btrfs_set_stack_device_id(dev_item, dev->devid);
2686 		btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2687 		btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2688 		btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2689 		btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2690 		btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2691 		memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2692 		memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2693 
2694 		flags = btrfs_super_flags(sb);
2695 		btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2696 
2697 		ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2698 		if (ret)
2699 			total_errors++;
2700 	}
2701 	if (total_errors > max_errors) {
2702 		printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2703 		       total_errors);
2704 		BUG();
2705 	}
2706 
2707 	total_errors = 0;
2708 	list_for_each_entry_rcu(dev, head, dev_list) {
2709 		if (!dev->bdev)
2710 			continue;
2711 		if (!dev->in_fs_metadata || !dev->writeable)
2712 			continue;
2713 
2714 		ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2715 		if (ret)
2716 			total_errors++;
2717 	}
2718 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2719 	if (total_errors > max_errors) {
2720 		printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2721 		       total_errors);
2722 		BUG();
2723 	}
2724 	return 0;
2725 }
2726 
2727 int write_ctree_super(struct btrfs_trans_handle *trans,
2728 		      struct btrfs_root *root, int max_mirrors)
2729 {
2730 	int ret;
2731 
2732 	ret = write_all_supers(root, max_mirrors);
2733 	return ret;
2734 }
2735 
2736 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2737 {
2738 	spin_lock(&fs_info->fs_roots_radix_lock);
2739 	radix_tree_delete(&fs_info->fs_roots_radix,
2740 			  (unsigned long)root->root_key.objectid);
2741 	spin_unlock(&fs_info->fs_roots_radix_lock);
2742 
2743 	if (btrfs_root_refs(&root->root_item) == 0)
2744 		synchronize_srcu(&fs_info->subvol_srcu);
2745 
2746 	__btrfs_remove_free_space_cache(root->free_ino_pinned);
2747 	__btrfs_remove_free_space_cache(root->free_ino_ctl);
2748 	free_fs_root(root);
2749 	return 0;
2750 }
2751 
2752 static void free_fs_root(struct btrfs_root *root)
2753 {
2754 	iput(root->cache_inode);
2755 	WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2756 	if (root->anon_dev)
2757 		free_anon_bdev(root->anon_dev);
2758 	free_extent_buffer(root->node);
2759 	free_extent_buffer(root->commit_root);
2760 	kfree(root->free_ino_ctl);
2761 	kfree(root->free_ino_pinned);
2762 	kfree(root->name);
2763 	kfree(root);
2764 }
2765 
2766 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2767 {
2768 	int ret;
2769 	struct btrfs_root *gang[8];
2770 	int i;
2771 
2772 	while (!list_empty(&fs_info->dead_roots)) {
2773 		gang[0] = list_entry(fs_info->dead_roots.next,
2774 				     struct btrfs_root, root_list);
2775 		list_del(&gang[0]->root_list);
2776 
2777 		if (gang[0]->in_radix) {
2778 			btrfs_free_fs_root(fs_info, gang[0]);
2779 		} else {
2780 			free_extent_buffer(gang[0]->node);
2781 			free_extent_buffer(gang[0]->commit_root);
2782 			kfree(gang[0]);
2783 		}
2784 	}
2785 
2786 	while (1) {
2787 		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2788 					     (void **)gang, 0,
2789 					     ARRAY_SIZE(gang));
2790 		if (!ret)
2791 			break;
2792 		for (i = 0; i < ret; i++)
2793 			btrfs_free_fs_root(fs_info, gang[i]);
2794 	}
2795 	return 0;
2796 }
2797 
2798 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2799 {
2800 	u64 root_objectid = 0;
2801 	struct btrfs_root *gang[8];
2802 	int i;
2803 	int ret;
2804 
2805 	while (1) {
2806 		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2807 					     (void **)gang, root_objectid,
2808 					     ARRAY_SIZE(gang));
2809 		if (!ret)
2810 			break;
2811 
2812 		root_objectid = gang[ret - 1]->root_key.objectid + 1;
2813 		for (i = 0; i < ret; i++) {
2814 			int err;
2815 
2816 			root_objectid = gang[i]->root_key.objectid;
2817 			err = btrfs_orphan_cleanup(gang[i]);
2818 			if (err)
2819 				return err;
2820 		}
2821 		root_objectid++;
2822 	}
2823 	return 0;
2824 }
2825 
2826 int btrfs_commit_super(struct btrfs_root *root)
2827 {
2828 	struct btrfs_trans_handle *trans;
2829 	int ret;
2830 
2831 	mutex_lock(&root->fs_info->cleaner_mutex);
2832 	btrfs_run_delayed_iputs(root);
2833 	btrfs_clean_old_snapshots(root);
2834 	mutex_unlock(&root->fs_info->cleaner_mutex);
2835 
2836 	/* wait until ongoing cleanup work done */
2837 	down_write(&root->fs_info->cleanup_work_sem);
2838 	up_write(&root->fs_info->cleanup_work_sem);
2839 
2840 	trans = btrfs_join_transaction(root);
2841 	if (IS_ERR(trans))
2842 		return PTR_ERR(trans);
2843 	ret = btrfs_commit_transaction(trans, root);
2844 	BUG_ON(ret);
2845 	/* run commit again to drop the original snapshot */
2846 	trans = btrfs_join_transaction(root);
2847 	if (IS_ERR(trans))
2848 		return PTR_ERR(trans);
2849 	btrfs_commit_transaction(trans, root);
2850 	ret = btrfs_write_and_wait_transaction(NULL, root);
2851 	BUG_ON(ret);
2852 
2853 	ret = write_ctree_super(NULL, root, 0);
2854 	return ret;
2855 }
2856 
2857 int close_ctree(struct btrfs_root *root)
2858 {
2859 	struct btrfs_fs_info *fs_info = root->fs_info;
2860 	int ret;
2861 
2862 	fs_info->closing = 1;
2863 	smp_mb();
2864 
2865 	btrfs_scrub_cancel(root);
2866 
2867 	/* wait for any defraggers to finish */
2868 	wait_event(fs_info->transaction_wait,
2869 		   (atomic_read(&fs_info->defrag_running) == 0));
2870 
2871 	/* clear out the rbtree of defraggable inodes */
2872 	btrfs_run_defrag_inodes(root->fs_info);
2873 
2874 	/*
2875 	 * Here come 2 situations when btrfs is broken to flip readonly:
2876 	 *
2877 	 * 1. when btrfs flips readonly somewhere else before
2878 	 * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
2879 	 * and btrfs will skip to write sb directly to keep
2880 	 * ERROR state on disk.
2881 	 *
2882 	 * 2. when btrfs flips readonly just in btrfs_commit_super,
2883 	 * and in such case, btrfs cannot write sb via btrfs_commit_super,
2884 	 * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
2885 	 * btrfs will cleanup all FS resources first and write sb then.
2886 	 */
2887 	if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2888 		ret = btrfs_commit_super(root);
2889 		if (ret)
2890 			printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2891 	}
2892 
2893 	if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
2894 		ret = btrfs_error_commit_super(root);
2895 		if (ret)
2896 			printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2897 	}
2898 
2899 	btrfs_put_block_group_cache(fs_info);
2900 
2901 	kthread_stop(root->fs_info->transaction_kthread);
2902 	kthread_stop(root->fs_info->cleaner_kthread);
2903 
2904 	fs_info->closing = 2;
2905 	smp_mb();
2906 
2907 	if (fs_info->delalloc_bytes) {
2908 		printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2909 		       (unsigned long long)fs_info->delalloc_bytes);
2910 	}
2911 	if (fs_info->total_ref_cache_size) {
2912 		printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2913 		       (unsigned long long)fs_info->total_ref_cache_size);
2914 	}
2915 
2916 	free_extent_buffer(fs_info->extent_root->node);
2917 	free_extent_buffer(fs_info->extent_root->commit_root);
2918 	free_extent_buffer(fs_info->tree_root->node);
2919 	free_extent_buffer(fs_info->tree_root->commit_root);
2920 	free_extent_buffer(root->fs_info->chunk_root->node);
2921 	free_extent_buffer(root->fs_info->chunk_root->commit_root);
2922 	free_extent_buffer(root->fs_info->dev_root->node);
2923 	free_extent_buffer(root->fs_info->dev_root->commit_root);
2924 	free_extent_buffer(root->fs_info->csum_root->node);
2925 	free_extent_buffer(root->fs_info->csum_root->commit_root);
2926 
2927 	btrfs_free_block_groups(root->fs_info);
2928 
2929 	del_fs_roots(fs_info);
2930 
2931 	iput(fs_info->btree_inode);
2932 
2933 	btrfs_stop_workers(&fs_info->generic_worker);
2934 	btrfs_stop_workers(&fs_info->fixup_workers);
2935 	btrfs_stop_workers(&fs_info->delalloc_workers);
2936 	btrfs_stop_workers(&fs_info->workers);
2937 	btrfs_stop_workers(&fs_info->endio_workers);
2938 	btrfs_stop_workers(&fs_info->endio_meta_workers);
2939 	btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2940 	btrfs_stop_workers(&fs_info->endio_write_workers);
2941 	btrfs_stop_workers(&fs_info->endio_freespace_worker);
2942 	btrfs_stop_workers(&fs_info->submit_workers);
2943 	btrfs_stop_workers(&fs_info->delayed_workers);
2944 	btrfs_stop_workers(&fs_info->caching_workers);
2945 	btrfs_stop_workers(&fs_info->readahead_workers);
2946 
2947 	btrfs_close_devices(fs_info->fs_devices);
2948 	btrfs_mapping_tree_free(&fs_info->mapping_tree);
2949 
2950 	bdi_destroy(&fs_info->bdi);
2951 	cleanup_srcu_struct(&fs_info->subvol_srcu);
2952 
2953 	free_fs_info(fs_info);
2954 
2955 	return 0;
2956 }
2957 
2958 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2959 {
2960 	int ret;
2961 	struct inode *btree_inode = buf->first_page->mapping->host;
2962 
2963 	ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2964 				     NULL);
2965 	if (!ret)
2966 		return ret;
2967 
2968 	ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2969 				    parent_transid);
2970 	return !ret;
2971 }
2972 
2973 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2974 {
2975 	struct inode *btree_inode = buf->first_page->mapping->host;
2976 	return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2977 					  buf);
2978 }
2979 
2980 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2981 {
2982 	struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2983 	u64 transid = btrfs_header_generation(buf);
2984 	struct inode *btree_inode = root->fs_info->btree_inode;
2985 	int was_dirty;
2986 
2987 	btrfs_assert_tree_locked(buf);
2988 	if (transid != root->fs_info->generation) {
2989 		printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2990 		       "found %llu running %llu\n",
2991 			(unsigned long long)buf->start,
2992 			(unsigned long long)transid,
2993 			(unsigned long long)root->fs_info->generation);
2994 		WARN_ON(1);
2995 	}
2996 	was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2997 					    buf);
2998 	if (!was_dirty) {
2999 		spin_lock(&root->fs_info->delalloc_lock);
3000 		root->fs_info->dirty_metadata_bytes += buf->len;
3001 		spin_unlock(&root->fs_info->delalloc_lock);
3002 	}
3003 }
3004 
3005 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
3006 {
3007 	/*
3008 	 * looks as though older kernels can get into trouble with
3009 	 * this code, they end up stuck in balance_dirty_pages forever
3010 	 */
3011 	u64 num_dirty;
3012 	unsigned long thresh = 32 * 1024 * 1024;
3013 
3014 	if (current->flags & PF_MEMALLOC)
3015 		return;
3016 
3017 	btrfs_balance_delayed_items(root);
3018 
3019 	num_dirty = root->fs_info->dirty_metadata_bytes;
3020 
3021 	if (num_dirty > thresh) {
3022 		balance_dirty_pages_ratelimited_nr(
3023 				   root->fs_info->btree_inode->i_mapping, 1);
3024 	}
3025 	return;
3026 }
3027 
3028 void __btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
3029 {
3030 	/*
3031 	 * looks as though older kernels can get into trouble with
3032 	 * this code, they end up stuck in balance_dirty_pages forever
3033 	 */
3034 	u64 num_dirty;
3035 	unsigned long thresh = 32 * 1024 * 1024;
3036 
3037 	if (current->flags & PF_MEMALLOC)
3038 		return;
3039 
3040 	num_dirty = root->fs_info->dirty_metadata_bytes;
3041 
3042 	if (num_dirty > thresh) {
3043 		balance_dirty_pages_ratelimited_nr(
3044 				   root->fs_info->btree_inode->i_mapping, 1);
3045 	}
3046 	return;
3047 }
3048 
3049 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3050 {
3051 	struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
3052 	int ret;
3053 	ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3054 	if (ret == 0)
3055 		set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
3056 	return ret;
3057 }
3058 
3059 static int btree_lock_page_hook(struct page *page, void *data,
3060 				void (*flush_fn)(void *))
3061 {
3062 	struct inode *inode = page->mapping->host;
3063 	struct btrfs_root *root = BTRFS_I(inode)->root;
3064 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3065 	struct extent_buffer *eb;
3066 	unsigned long len;
3067 	u64 bytenr = page_offset(page);
3068 
3069 	if (page->private == EXTENT_PAGE_PRIVATE)
3070 		goto out;
3071 
3072 	len = page->private >> 2;
3073 	eb = find_extent_buffer(io_tree, bytenr, len);
3074 	if (!eb)
3075 		goto out;
3076 
3077 	if (!btrfs_try_tree_write_lock(eb)) {
3078 		flush_fn(data);
3079 		btrfs_tree_lock(eb);
3080 	}
3081 	btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3082 
3083 	if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3084 		spin_lock(&root->fs_info->delalloc_lock);
3085 		if (root->fs_info->dirty_metadata_bytes >= eb->len)
3086 			root->fs_info->dirty_metadata_bytes -= eb->len;
3087 		else
3088 			WARN_ON(1);
3089 		spin_unlock(&root->fs_info->delalloc_lock);
3090 	}
3091 
3092 	btrfs_tree_unlock(eb);
3093 	free_extent_buffer(eb);
3094 out:
3095 	if (!trylock_page(page)) {
3096 		flush_fn(data);
3097 		lock_page(page);
3098 	}
3099 	return 0;
3100 }
3101 
3102 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3103 			      int read_only)
3104 {
3105 	if (read_only)
3106 		return;
3107 
3108 	if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
3109 		printk(KERN_WARNING "warning: mount fs with errors, "
3110 		       "running btrfsck is recommended\n");
3111 }
3112 
3113 int btrfs_error_commit_super(struct btrfs_root *root)
3114 {
3115 	int ret;
3116 
3117 	mutex_lock(&root->fs_info->cleaner_mutex);
3118 	btrfs_run_delayed_iputs(root);
3119 	mutex_unlock(&root->fs_info->cleaner_mutex);
3120 
3121 	down_write(&root->fs_info->cleanup_work_sem);
3122 	up_write(&root->fs_info->cleanup_work_sem);
3123 
3124 	/* cleanup FS via transaction */
3125 	btrfs_cleanup_transaction(root);
3126 
3127 	ret = write_ctree_super(NULL, root, 0);
3128 
3129 	return ret;
3130 }
3131 
3132 static int btrfs_destroy_ordered_operations(struct btrfs_root *root)
3133 {
3134 	struct btrfs_inode *btrfs_inode;
3135 	struct list_head splice;
3136 
3137 	INIT_LIST_HEAD(&splice);
3138 
3139 	mutex_lock(&root->fs_info->ordered_operations_mutex);
3140 	spin_lock(&root->fs_info->ordered_extent_lock);
3141 
3142 	list_splice_init(&root->fs_info->ordered_operations, &splice);
3143 	while (!list_empty(&splice)) {
3144 		btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3145 					 ordered_operations);
3146 
3147 		list_del_init(&btrfs_inode->ordered_operations);
3148 
3149 		btrfs_invalidate_inodes(btrfs_inode->root);
3150 	}
3151 
3152 	spin_unlock(&root->fs_info->ordered_extent_lock);
3153 	mutex_unlock(&root->fs_info->ordered_operations_mutex);
3154 
3155 	return 0;
3156 }
3157 
3158 static int btrfs_destroy_ordered_extents(struct btrfs_root *root)
3159 {
3160 	struct list_head splice;
3161 	struct btrfs_ordered_extent *ordered;
3162 	struct inode *inode;
3163 
3164 	INIT_LIST_HEAD(&splice);
3165 
3166 	spin_lock(&root->fs_info->ordered_extent_lock);
3167 
3168 	list_splice_init(&root->fs_info->ordered_extents, &splice);
3169 	while (!list_empty(&splice)) {
3170 		ordered = list_entry(splice.next, struct btrfs_ordered_extent,
3171 				     root_extent_list);
3172 
3173 		list_del_init(&ordered->root_extent_list);
3174 		atomic_inc(&ordered->refs);
3175 
3176 		/* the inode may be getting freed (in sys_unlink path). */
3177 		inode = igrab(ordered->inode);
3178 
3179 		spin_unlock(&root->fs_info->ordered_extent_lock);
3180 		if (inode)
3181 			iput(inode);
3182 
3183 		atomic_set(&ordered->refs, 1);
3184 		btrfs_put_ordered_extent(ordered);
3185 
3186 		spin_lock(&root->fs_info->ordered_extent_lock);
3187 	}
3188 
3189 	spin_unlock(&root->fs_info->ordered_extent_lock);
3190 
3191 	return 0;
3192 }
3193 
3194 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3195 				      struct btrfs_root *root)
3196 {
3197 	struct rb_node *node;
3198 	struct btrfs_delayed_ref_root *delayed_refs;
3199 	struct btrfs_delayed_ref_node *ref;
3200 	int ret = 0;
3201 
3202 	delayed_refs = &trans->delayed_refs;
3203 
3204 	spin_lock(&delayed_refs->lock);
3205 	if (delayed_refs->num_entries == 0) {
3206 		spin_unlock(&delayed_refs->lock);
3207 		printk(KERN_INFO "delayed_refs has NO entry\n");
3208 		return ret;
3209 	}
3210 
3211 	node = rb_first(&delayed_refs->root);
3212 	while (node) {
3213 		ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3214 		node = rb_next(node);
3215 
3216 		ref->in_tree = 0;
3217 		rb_erase(&ref->rb_node, &delayed_refs->root);
3218 		delayed_refs->num_entries--;
3219 
3220 		atomic_set(&ref->refs, 1);
3221 		if (btrfs_delayed_ref_is_head(ref)) {
3222 			struct btrfs_delayed_ref_head *head;
3223 
3224 			head = btrfs_delayed_node_to_head(ref);
3225 			mutex_lock(&head->mutex);
3226 			kfree(head->extent_op);
3227 			delayed_refs->num_heads--;
3228 			if (list_empty(&head->cluster))
3229 				delayed_refs->num_heads_ready--;
3230 			list_del_init(&head->cluster);
3231 			mutex_unlock(&head->mutex);
3232 		}
3233 
3234 		spin_unlock(&delayed_refs->lock);
3235 		btrfs_put_delayed_ref(ref);
3236 
3237 		cond_resched();
3238 		spin_lock(&delayed_refs->lock);
3239 	}
3240 
3241 	spin_unlock(&delayed_refs->lock);
3242 
3243 	return ret;
3244 }
3245 
3246 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
3247 {
3248 	struct btrfs_pending_snapshot *snapshot;
3249 	struct list_head splice;
3250 
3251 	INIT_LIST_HEAD(&splice);
3252 
3253 	list_splice_init(&t->pending_snapshots, &splice);
3254 
3255 	while (!list_empty(&splice)) {
3256 		snapshot = list_entry(splice.next,
3257 				      struct btrfs_pending_snapshot,
3258 				      list);
3259 
3260 		list_del_init(&snapshot->list);
3261 
3262 		kfree(snapshot);
3263 	}
3264 
3265 	return 0;
3266 }
3267 
3268 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3269 {
3270 	struct btrfs_inode *btrfs_inode;
3271 	struct list_head splice;
3272 
3273 	INIT_LIST_HEAD(&splice);
3274 
3275 	spin_lock(&root->fs_info->delalloc_lock);
3276 	list_splice_init(&root->fs_info->delalloc_inodes, &splice);
3277 
3278 	while (!list_empty(&splice)) {
3279 		btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3280 				    delalloc_inodes);
3281 
3282 		list_del_init(&btrfs_inode->delalloc_inodes);
3283 
3284 		btrfs_invalidate_inodes(btrfs_inode->root);
3285 	}
3286 
3287 	spin_unlock(&root->fs_info->delalloc_lock);
3288 
3289 	return 0;
3290 }
3291 
3292 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3293 					struct extent_io_tree *dirty_pages,
3294 					int mark)
3295 {
3296 	int ret;
3297 	struct page *page;
3298 	struct inode *btree_inode = root->fs_info->btree_inode;
3299 	struct extent_buffer *eb;
3300 	u64 start = 0;
3301 	u64 end;
3302 	u64 offset;
3303 	unsigned long index;
3304 
3305 	while (1) {
3306 		ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3307 					    mark);
3308 		if (ret)
3309 			break;
3310 
3311 		clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3312 		while (start <= end) {
3313 			index = start >> PAGE_CACHE_SHIFT;
3314 			start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
3315 			page = find_get_page(btree_inode->i_mapping, index);
3316 			if (!page)
3317 				continue;
3318 			offset = page_offset(page);
3319 
3320 			spin_lock(&dirty_pages->buffer_lock);
3321 			eb = radix_tree_lookup(
3322 			     &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
3323 					       offset >> PAGE_CACHE_SHIFT);
3324 			spin_unlock(&dirty_pages->buffer_lock);
3325 			if (eb) {
3326 				ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3327 							 &eb->bflags);
3328 				atomic_set(&eb->refs, 1);
3329 			}
3330 			if (PageWriteback(page))
3331 				end_page_writeback(page);
3332 
3333 			lock_page(page);
3334 			if (PageDirty(page)) {
3335 				clear_page_dirty_for_io(page);
3336 				spin_lock_irq(&page->mapping->tree_lock);
3337 				radix_tree_tag_clear(&page->mapping->page_tree,
3338 							page_index(page),
3339 							PAGECACHE_TAG_DIRTY);
3340 				spin_unlock_irq(&page->mapping->tree_lock);
3341 			}
3342 
3343 			page->mapping->a_ops->invalidatepage(page, 0);
3344 			unlock_page(page);
3345 		}
3346 	}
3347 
3348 	return ret;
3349 }
3350 
3351 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3352 				       struct extent_io_tree *pinned_extents)
3353 {
3354 	struct extent_io_tree *unpin;
3355 	u64 start;
3356 	u64 end;
3357 	int ret;
3358 
3359 	unpin = pinned_extents;
3360 	while (1) {
3361 		ret = find_first_extent_bit(unpin, 0, &start, &end,
3362 					    EXTENT_DIRTY);
3363 		if (ret)
3364 			break;
3365 
3366 		/* opt_discard */
3367 		if (btrfs_test_opt(root, DISCARD))
3368 			ret = btrfs_error_discard_extent(root, start,
3369 							 end + 1 - start,
3370 							 NULL);
3371 
3372 		clear_extent_dirty(unpin, start, end, GFP_NOFS);
3373 		btrfs_error_unpin_extent_range(root, start, end);
3374 		cond_resched();
3375 	}
3376 
3377 	return 0;
3378 }
3379 
3380 static int btrfs_cleanup_transaction(struct btrfs_root *root)
3381 {
3382 	struct btrfs_transaction *t;
3383 	LIST_HEAD(list);
3384 
3385 	WARN_ON(1);
3386 
3387 	mutex_lock(&root->fs_info->transaction_kthread_mutex);
3388 
3389 	spin_lock(&root->fs_info->trans_lock);
3390 	list_splice_init(&root->fs_info->trans_list, &list);
3391 	root->fs_info->trans_no_join = 1;
3392 	spin_unlock(&root->fs_info->trans_lock);
3393 
3394 	while (!list_empty(&list)) {
3395 		t = list_entry(list.next, struct btrfs_transaction, list);
3396 		if (!t)
3397 			break;
3398 
3399 		btrfs_destroy_ordered_operations(root);
3400 
3401 		btrfs_destroy_ordered_extents(root);
3402 
3403 		btrfs_destroy_delayed_refs(t, root);
3404 
3405 		btrfs_block_rsv_release(root,
3406 					&root->fs_info->trans_block_rsv,
3407 					t->dirty_pages.dirty_bytes);
3408 
3409 		/* FIXME: cleanup wait for commit */
3410 		t->in_commit = 1;
3411 		t->blocked = 1;
3412 		if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3413 			wake_up(&root->fs_info->transaction_blocked_wait);
3414 
3415 		t->blocked = 0;
3416 		if (waitqueue_active(&root->fs_info->transaction_wait))
3417 			wake_up(&root->fs_info->transaction_wait);
3418 
3419 		t->commit_done = 1;
3420 		if (waitqueue_active(&t->commit_wait))
3421 			wake_up(&t->commit_wait);
3422 
3423 		btrfs_destroy_pending_snapshots(t);
3424 
3425 		btrfs_destroy_delalloc_inodes(root);
3426 
3427 		spin_lock(&root->fs_info->trans_lock);
3428 		root->fs_info->running_transaction = NULL;
3429 		spin_unlock(&root->fs_info->trans_lock);
3430 
3431 		btrfs_destroy_marked_extents(root, &t->dirty_pages,
3432 					     EXTENT_DIRTY);
3433 
3434 		btrfs_destroy_pinned_extent(root,
3435 					    root->fs_info->pinned_extents);
3436 
3437 		atomic_set(&t->use_count, 0);
3438 		list_del_init(&t->list);
3439 		memset(t, 0, sizeof(*t));
3440 		kmem_cache_free(btrfs_transaction_cachep, t);
3441 	}
3442 
3443 	spin_lock(&root->fs_info->trans_lock);
3444 	root->fs_info->trans_no_join = 0;
3445 	spin_unlock(&root->fs_info->trans_lock);
3446 	mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3447 
3448 	return 0;
3449 }
3450 
3451 static struct extent_io_ops btree_extent_io_ops = {
3452 	.write_cache_pages_lock_hook = btree_lock_page_hook,
3453 	.readpage_end_io_hook = btree_readpage_end_io_hook,
3454 	.readpage_io_failed_hook = btree_io_failed_hook,
3455 	.submit_bio_hook = btree_submit_bio_hook,
3456 	/* note we're sharing with inode.c for the merge bio hook */
3457 	.merge_bio_hook = btrfs_merge_bio_hook,
3458 };
3459