xref: /openbmc/linux/fs/btrfs/disk-io.c (revision d1310b2e0cd98eb1348553e69b73827b436dca7b)
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/crc32c.h>
22 #include <linux/scatterlist.h>
23 #include <linux/swap.h>
24 #include <linux/radix-tree.h>
25 #include <linux/writeback.h>
26 #include <linux/buffer_head.h> // for block_sync_page
27 #include "ctree.h"
28 #include "disk-io.h"
29 #include "transaction.h"
30 #include "btrfs_inode.h"
31 #include "print-tree.h"
32 
33 #if 0
34 static int check_tree_block(struct btrfs_root *root, struct extent_buffer *buf)
35 {
36 	if (extent_buffer_blocknr(buf) != btrfs_header_blocknr(buf)) {
37 		printk(KERN_CRIT "buf blocknr(buf) is %llu, header is %llu\n",
38 		       (unsigned long long)extent_buffer_blocknr(buf),
39 		       (unsigned long long)btrfs_header_blocknr(buf));
40 		return 1;
41 	}
42 	return 0;
43 }
44 #endif
45 
46 static struct extent_io_ops btree_extent_io_ops;
47 
48 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
49 					    u64 bytenr, u32 blocksize)
50 {
51 	struct inode *btree_inode = root->fs_info->btree_inode;
52 	struct extent_buffer *eb;
53 	eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
54 				bytenr, blocksize, GFP_NOFS);
55 	return eb;
56 }
57 
58 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
59 						 u64 bytenr, u32 blocksize)
60 {
61 	struct inode *btree_inode = root->fs_info->btree_inode;
62 	struct extent_buffer *eb;
63 
64 	eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
65 				 bytenr, blocksize, NULL, GFP_NOFS);
66 	return eb;
67 }
68 
69 struct extent_map *btree_get_extent(struct inode *inode, struct page *page,
70 				    size_t page_offset, u64 start, u64 len,
71 				    int create)
72 {
73 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
74 	struct extent_map *em;
75 	int ret;
76 
77 again:
78 	spin_lock(&em_tree->lock);
79 	em = lookup_extent_mapping(em_tree, start, len);
80 	spin_unlock(&em_tree->lock);
81 	if (em) {
82 		goto out;
83 	}
84 	em = alloc_extent_map(GFP_NOFS);
85 	if (!em) {
86 		em = ERR_PTR(-ENOMEM);
87 		goto out;
88 	}
89 	em->start = 0;
90 	em->len = i_size_read(inode);
91 	em->block_start = 0;
92 	em->bdev = inode->i_sb->s_bdev;
93 
94 	spin_lock(&em_tree->lock);
95 	ret = add_extent_mapping(em_tree, em);
96 	spin_unlock(&em_tree->lock);
97 
98 	if (ret == -EEXIST) {
99 		free_extent_map(em);
100 		em = NULL;
101 		goto again;
102 	} else if (ret) {
103 		em = ERR_PTR(ret);
104 	}
105 out:
106 	return em;
107 }
108 
109 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
110 {
111 	return crc32c(seed, data, len);
112 }
113 
114 void btrfs_csum_final(u32 crc, char *result)
115 {
116 	*(__le32 *)result = ~cpu_to_le32(crc);
117 }
118 
119 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
120 			   int verify)
121 {
122 	char result[BTRFS_CRC32_SIZE];
123 	unsigned long len;
124 	unsigned long cur_len;
125 	unsigned long offset = BTRFS_CSUM_SIZE;
126 	char *map_token = NULL;
127 	char *kaddr;
128 	unsigned long map_start;
129 	unsigned long map_len;
130 	int err;
131 	u32 crc = ~(u32)0;
132 
133 	len = buf->len - offset;
134 	while(len > 0) {
135 		err = map_private_extent_buffer(buf, offset, 32,
136 					&map_token, &kaddr,
137 					&map_start, &map_len, KM_USER0);
138 		if (err) {
139 			printk("failed to map extent buffer! %lu\n",
140 			       offset);
141 			return 1;
142 		}
143 		cur_len = min(len, map_len - (offset - map_start));
144 		crc = btrfs_csum_data(root, kaddr + offset - map_start,
145 				      crc, cur_len);
146 		len -= cur_len;
147 		offset += cur_len;
148 		unmap_extent_buffer(buf, map_token, KM_USER0);
149 	}
150 	btrfs_csum_final(crc, result);
151 
152 	if (verify) {
153 		int from_this_trans = 0;
154 
155 		if (root->fs_info->running_transaction &&
156 		    btrfs_header_generation(buf) ==
157 		    root->fs_info->running_transaction->transid)
158 			from_this_trans = 1;
159 
160 		/* FIXME, this is not good */
161 		if (from_this_trans == 0 &&
162 		    memcmp_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE)) {
163 			u32 val;
164 			u32 found = 0;
165 			memcpy(&found, result, BTRFS_CRC32_SIZE);
166 
167 			read_extent_buffer(buf, &val, 0, BTRFS_CRC32_SIZE);
168 			printk("btrfs: %s checksum verify failed on %llu "
169 			       "wanted %X found %X from_this_trans %d\n",
170 			       root->fs_info->sb->s_id,
171 			       buf->start, val, found, from_this_trans);
172 			return 1;
173 		}
174 	} else {
175 		write_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE);
176 	}
177 	return 0;
178 }
179 
180 
181 int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
182 {
183 	struct extent_io_tree *tree;
184 	u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
185 	u64 found_start;
186 	int found_level;
187 	unsigned long len;
188 	struct extent_buffer *eb;
189 	tree = &BTRFS_I(page->mapping->host)->io_tree;
190 
191 	if (page->private == EXTENT_PAGE_PRIVATE)
192 		goto out;
193 	if (!page->private)
194 		goto out;
195 	len = page->private >> 2;
196 	if (len == 0) {
197 		WARN_ON(1);
198 	}
199 	eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
200 	read_extent_buffer_pages(tree, eb, start + PAGE_CACHE_SIZE, 1);
201 	btrfs_clear_buffer_defrag(eb);
202 	found_start = btrfs_header_bytenr(eb);
203 	if (found_start != start) {
204 		printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
205 		       start, found_start, len);
206 		WARN_ON(1);
207 		goto err;
208 	}
209 	if (eb->first_page != page) {
210 		printk("bad first page %lu %lu\n", eb->first_page->index,
211 		       page->index);
212 		WARN_ON(1);
213 		goto err;
214 	}
215 	if (!PageUptodate(page)) {
216 		printk("csum not up to date page %lu\n", page->index);
217 		WARN_ON(1);
218 		goto err;
219 	}
220 	found_level = btrfs_header_level(eb);
221 	csum_tree_block(root, eb, 0);
222 err:
223 	free_extent_buffer(eb);
224 out:
225 	return 0;
226 }
227 
228 static int btree_writepage_io_hook(struct page *page, u64 start, u64 end)
229 {
230 	struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
231 
232 	csum_dirty_buffer(root, page);
233 	return 0;
234 }
235 
236 static int btree_writepage(struct page *page, struct writeback_control *wbc)
237 {
238 	struct extent_io_tree *tree;
239 	tree = &BTRFS_I(page->mapping->host)->io_tree;
240 	return extent_write_full_page(tree, page, btree_get_extent, wbc);
241 }
242 
243 static int btree_writepages(struct address_space *mapping,
244 			    struct writeback_control *wbc)
245 {
246 	struct extent_io_tree *tree;
247 	tree = &BTRFS_I(mapping->host)->io_tree;
248 	if (wbc->sync_mode == WB_SYNC_NONE) {
249 		u64 num_dirty;
250 		u64 start = 0;
251 		unsigned long thresh = 96 * 1024 * 1024;
252 
253 		if (wbc->for_kupdate)
254 			return 0;
255 
256 		if (current_is_pdflush()) {
257 			thresh = 96 * 1024 * 1024;
258 		} else {
259 			thresh = 8 * 1024 * 1024;
260 		}
261 		num_dirty = count_range_bits(tree, &start, (u64)-1,
262 					     thresh, EXTENT_DIRTY);
263 		if (num_dirty < thresh) {
264 			return 0;
265 		}
266 	}
267 	return extent_writepages(tree, mapping, btree_get_extent, wbc);
268 }
269 
270 int btree_readpage(struct file *file, struct page *page)
271 {
272 	struct extent_io_tree *tree;
273 	tree = &BTRFS_I(page->mapping->host)->io_tree;
274 	return extent_read_full_page(tree, page, btree_get_extent);
275 }
276 
277 static int btree_releasepage(struct page *page, gfp_t unused_gfp_flags)
278 {
279 	struct extent_io_tree *tree;
280 	struct extent_map_tree *map;
281 	int ret;
282 
283 	tree = &BTRFS_I(page->mapping->host)->io_tree;
284 	map = &BTRFS_I(page->mapping->host)->extent_tree;
285 	ret = try_release_extent_mapping(map, tree, page);
286 	if (ret == 1) {
287 		ClearPagePrivate(page);
288 		set_page_private(page, 0);
289 		page_cache_release(page);
290 	}
291 	return ret;
292 }
293 
294 static void btree_invalidatepage(struct page *page, unsigned long offset)
295 {
296 	struct extent_io_tree *tree;
297 	tree = &BTRFS_I(page->mapping->host)->io_tree;
298 	extent_invalidatepage(tree, page, offset);
299 	btree_releasepage(page, GFP_NOFS);
300 }
301 
302 #if 0
303 static int btree_writepage(struct page *page, struct writeback_control *wbc)
304 {
305 	struct buffer_head *bh;
306 	struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
307 	struct buffer_head *head;
308 	if (!page_has_buffers(page)) {
309 		create_empty_buffers(page, root->fs_info->sb->s_blocksize,
310 					(1 << BH_Dirty)|(1 << BH_Uptodate));
311 	}
312 	head = page_buffers(page);
313 	bh = head;
314 	do {
315 		if (buffer_dirty(bh))
316 			csum_tree_block(root, bh, 0);
317 		bh = bh->b_this_page;
318 	} while (bh != head);
319 	return block_write_full_page(page, btree_get_block, wbc);
320 }
321 #endif
322 
323 static struct address_space_operations btree_aops = {
324 	.readpage	= btree_readpage,
325 	.writepage	= btree_writepage,
326 	.writepages	= btree_writepages,
327 	.releasepage	= btree_releasepage,
328 	.invalidatepage = btree_invalidatepage,
329 	.sync_page	= block_sync_page,
330 };
331 
332 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize)
333 {
334 	struct extent_buffer *buf = NULL;
335 	struct inode *btree_inode = root->fs_info->btree_inode;
336 	int ret = 0;
337 
338 	buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
339 	if (!buf)
340 		return 0;
341 	read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
342 				 buf, 0, 0);
343 	free_extent_buffer(buf);
344 	return ret;
345 }
346 
347 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
348 				      u32 blocksize)
349 {
350 	struct extent_buffer *buf = NULL;
351 	struct inode *btree_inode = root->fs_info->btree_inode;
352 	struct extent_io_tree *io_tree;
353 	u64 end;
354 	int ret;
355 
356 	io_tree = &BTRFS_I(btree_inode)->io_tree;
357 
358 	buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
359 	if (!buf)
360 		return NULL;
361 	read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree, buf, 0, 1);
362 
363 	if (buf->flags & EXTENT_CSUM)
364 		return buf;
365 
366 	end = buf->start + PAGE_CACHE_SIZE - 1;
367 	if (test_range_bit(io_tree, buf->start, end, EXTENT_CSUM, 1)) {
368 		buf->flags |= EXTENT_CSUM;
369 		return buf;
370 	}
371 
372 	lock_extent(io_tree, buf->start, end, GFP_NOFS);
373 
374 	if (test_range_bit(io_tree, buf->start, end, EXTENT_CSUM, 1)) {
375 		buf->flags |= EXTENT_CSUM;
376 		goto out_unlock;
377 	}
378 
379 	ret = csum_tree_block(root, buf, 1);
380 	set_extent_bits(io_tree, buf->start, end, EXTENT_CSUM, GFP_NOFS);
381 	buf->flags |= EXTENT_CSUM;
382 
383 out_unlock:
384 	unlock_extent(io_tree, buf->start, end, GFP_NOFS);
385 	return buf;
386 }
387 
388 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
389 		     struct extent_buffer *buf)
390 {
391 	struct inode *btree_inode = root->fs_info->btree_inode;
392 	if (btrfs_header_generation(buf) ==
393 	    root->fs_info->running_transaction->transid)
394 		clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
395 					  buf);
396 	return 0;
397 }
398 
399 int wait_on_tree_block_writeback(struct btrfs_root *root,
400 				 struct extent_buffer *buf)
401 {
402 	struct inode *btree_inode = root->fs_info->btree_inode;
403 	wait_on_extent_buffer_writeback(&BTRFS_I(btree_inode)->io_tree,
404 					buf);
405 	return 0;
406 }
407 
408 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
409 			u32 stripesize, struct btrfs_root *root,
410 			struct btrfs_fs_info *fs_info,
411 			u64 objectid)
412 {
413 	root->node = NULL;
414 	root->inode = NULL;
415 	root->commit_root = NULL;
416 	root->sectorsize = sectorsize;
417 	root->nodesize = nodesize;
418 	root->leafsize = leafsize;
419 	root->stripesize = stripesize;
420 	root->ref_cows = 0;
421 	root->fs_info = fs_info;
422 	root->objectid = objectid;
423 	root->last_trans = 0;
424 	root->highest_inode = 0;
425 	root->last_inode_alloc = 0;
426 	root->name = NULL;
427 	root->in_sysfs = 0;
428 	memset(&root->root_key, 0, sizeof(root->root_key));
429 	memset(&root->root_item, 0, sizeof(root->root_item));
430 	memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
431 	memset(&root->root_kobj, 0, sizeof(root->root_kobj));
432 	init_completion(&root->kobj_unregister);
433 	root->defrag_running = 0;
434 	root->defrag_level = 0;
435 	root->root_key.objectid = objectid;
436 	return 0;
437 }
438 
439 static int find_and_setup_root(struct btrfs_root *tree_root,
440 			       struct btrfs_fs_info *fs_info,
441 			       u64 objectid,
442 			       struct btrfs_root *root)
443 {
444 	int ret;
445 	u32 blocksize;
446 
447 	__setup_root(tree_root->nodesize, tree_root->leafsize,
448 		     tree_root->sectorsize, tree_root->stripesize,
449 		     root, fs_info, objectid);
450 	ret = btrfs_find_last_root(tree_root, objectid,
451 				   &root->root_item, &root->root_key);
452 	BUG_ON(ret);
453 
454 	blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
455 	root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
456 				     blocksize);
457 	BUG_ON(!root->node);
458 	return 0;
459 }
460 
461 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_fs_info *fs_info,
462 					       struct btrfs_key *location)
463 {
464 	struct btrfs_root *root;
465 	struct btrfs_root *tree_root = fs_info->tree_root;
466 	struct btrfs_path *path;
467 	struct extent_buffer *l;
468 	u64 highest_inode;
469 	u32 blocksize;
470 	int ret = 0;
471 
472 	root = kzalloc(sizeof(*root), GFP_NOFS);
473 	if (!root)
474 		return ERR_PTR(-ENOMEM);
475 	if (location->offset == (u64)-1) {
476 		ret = find_and_setup_root(tree_root, fs_info,
477 					  location->objectid, root);
478 		if (ret) {
479 			kfree(root);
480 			return ERR_PTR(ret);
481 		}
482 		goto insert;
483 	}
484 
485 	__setup_root(tree_root->nodesize, tree_root->leafsize,
486 		     tree_root->sectorsize, tree_root->stripesize,
487 		     root, fs_info, location->objectid);
488 
489 	path = btrfs_alloc_path();
490 	BUG_ON(!path);
491 	ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
492 	if (ret != 0) {
493 		if (ret > 0)
494 			ret = -ENOENT;
495 		goto out;
496 	}
497 	l = path->nodes[0];
498 	read_extent_buffer(l, &root->root_item,
499 	       btrfs_item_ptr_offset(l, path->slots[0]),
500 	       sizeof(root->root_item));
501 	memcpy(&root->root_key, location, sizeof(*location));
502 	ret = 0;
503 out:
504 	btrfs_release_path(root, path);
505 	btrfs_free_path(path);
506 	if (ret) {
507 		kfree(root);
508 		return ERR_PTR(ret);
509 	}
510 	blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
511 	root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
512 				     blocksize);
513 	BUG_ON(!root->node);
514 insert:
515 	root->ref_cows = 1;
516 	ret = btrfs_find_highest_inode(root, &highest_inode);
517 	if (ret == 0) {
518 		root->highest_inode = highest_inode;
519 		root->last_inode_alloc = highest_inode;
520 	}
521 	return root;
522 }
523 
524 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
525 					u64 root_objectid)
526 {
527 	struct btrfs_root *root;
528 
529 	if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
530 		return fs_info->tree_root;
531 	if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
532 		return fs_info->extent_root;
533 
534 	root = radix_tree_lookup(&fs_info->fs_roots_radix,
535 				 (unsigned long)root_objectid);
536 	return root;
537 }
538 
539 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
540 					      struct btrfs_key *location)
541 {
542 	struct btrfs_root *root;
543 	int ret;
544 
545 	if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
546 		return fs_info->tree_root;
547 	if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
548 		return fs_info->extent_root;
549 
550 	root = radix_tree_lookup(&fs_info->fs_roots_radix,
551 				 (unsigned long)location->objectid);
552 	if (root)
553 		return root;
554 
555 	root = btrfs_read_fs_root_no_radix(fs_info, location);
556 	if (IS_ERR(root))
557 		return root;
558 	ret = radix_tree_insert(&fs_info->fs_roots_radix,
559 				(unsigned long)root->root_key.objectid,
560 				root);
561 	if (ret) {
562 		free_extent_buffer(root->node);
563 		kfree(root);
564 		return ERR_PTR(ret);
565 	}
566 	ret = btrfs_find_dead_roots(fs_info->tree_root,
567 				    root->root_key.objectid, root);
568 	BUG_ON(ret);
569 
570 	return root;
571 }
572 
573 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
574 				      struct btrfs_key *location,
575 				      const char *name, int namelen)
576 {
577 	struct btrfs_root *root;
578 	int ret;
579 
580 	root = btrfs_read_fs_root_no_name(fs_info, location);
581 	if (!root)
582 		return NULL;
583 
584 	if (root->in_sysfs)
585 		return root;
586 
587 	ret = btrfs_set_root_name(root, name, namelen);
588 	if (ret) {
589 		free_extent_buffer(root->node);
590 		kfree(root);
591 		return ERR_PTR(ret);
592 	}
593 
594 	ret = btrfs_sysfs_add_root(root);
595 	if (ret) {
596 		free_extent_buffer(root->node);
597 		kfree(root->name);
598 		kfree(root);
599 		return ERR_PTR(ret);
600 	}
601 	root->in_sysfs = 1;
602 	return root;
603 }
604 #if 0
605 static int add_hasher(struct btrfs_fs_info *info, char *type) {
606 	struct btrfs_hasher *hasher;
607 
608 	hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
609 	if (!hasher)
610 		return -ENOMEM;
611 	hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
612 	if (!hasher->hash_tfm) {
613 		kfree(hasher);
614 		return -EINVAL;
615 	}
616 	spin_lock(&info->hash_lock);
617 	list_add(&hasher->list, &info->hashers);
618 	spin_unlock(&info->hash_lock);
619 	return 0;
620 }
621 #endif
622 struct btrfs_root *open_ctree(struct super_block *sb)
623 {
624 	u32 sectorsize;
625 	u32 nodesize;
626 	u32 leafsize;
627 	u32 blocksize;
628 	u32 stripesize;
629 	struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
630 						 GFP_NOFS);
631 	struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
632 					       GFP_NOFS);
633 	struct btrfs_fs_info *fs_info = kmalloc(sizeof(*fs_info),
634 						GFP_NOFS);
635 	int ret;
636 	int err = -EIO;
637 	struct btrfs_super_block *disk_super;
638 
639 	if (!extent_root || !tree_root || !fs_info) {
640 		err = -ENOMEM;
641 		goto fail;
642 	}
643 	INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
644 	INIT_LIST_HEAD(&fs_info->trans_list);
645 	INIT_LIST_HEAD(&fs_info->dead_roots);
646 	INIT_LIST_HEAD(&fs_info->hashers);
647 	spin_lock_init(&fs_info->hash_lock);
648 	spin_lock_init(&fs_info->delalloc_lock);
649 	spin_lock_init(&fs_info->new_trans_lock);
650 
651 	memset(&fs_info->super_kobj, 0, sizeof(fs_info->super_kobj));
652 	init_completion(&fs_info->kobj_unregister);
653 	sb_set_blocksize(sb, 4096);
654 	fs_info->running_transaction = NULL;
655 	fs_info->last_trans_committed = 0;
656 	fs_info->tree_root = tree_root;
657 	fs_info->extent_root = extent_root;
658 	fs_info->sb = sb;
659 	fs_info->throttles = 0;
660 	fs_info->mount_opt = 0;
661 	fs_info->max_extent = (u64)-1;
662 	fs_info->delalloc_bytes = 0;
663 	fs_info->btree_inode = new_inode(sb);
664 	fs_info->btree_inode->i_ino = 1;
665 	fs_info->btree_inode->i_nlink = 1;
666 	fs_info->btree_inode->i_size = sb->s_bdev->bd_inode->i_size;
667 	fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
668 	extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
669 			     fs_info->btree_inode->i_mapping,
670 			     GFP_NOFS);
671 	extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
672 			     GFP_NOFS);
673 
674 	BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
675 
676 	extent_io_tree_init(&fs_info->free_space_cache,
677 			     fs_info->btree_inode->i_mapping, GFP_NOFS);
678 	extent_io_tree_init(&fs_info->block_group_cache,
679 			     fs_info->btree_inode->i_mapping, GFP_NOFS);
680 	extent_io_tree_init(&fs_info->pinned_extents,
681 			     fs_info->btree_inode->i_mapping, GFP_NOFS);
682 	extent_io_tree_init(&fs_info->pending_del,
683 			     fs_info->btree_inode->i_mapping, GFP_NOFS);
684 	extent_io_tree_init(&fs_info->extent_ins,
685 			     fs_info->btree_inode->i_mapping, GFP_NOFS);
686 	fs_info->do_barriers = 1;
687 	fs_info->closing = 0;
688 	fs_info->total_pinned = 0;
689 	fs_info->last_alloc = 0;
690 
691 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
692 	INIT_WORK(&fs_info->trans_work, btrfs_transaction_cleaner, fs_info);
693 #else
694 	INIT_DELAYED_WORK(&fs_info->trans_work, btrfs_transaction_cleaner);
695 #endif
696 	BTRFS_I(fs_info->btree_inode)->root = tree_root;
697 	memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
698 	       sizeof(struct btrfs_key));
699 	insert_inode_hash(fs_info->btree_inode);
700 	mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
701 
702 	mutex_init(&fs_info->trans_mutex);
703 	mutex_init(&fs_info->fs_mutex);
704 
705 #if 0
706 	ret = add_hasher(fs_info, "crc32c");
707 	if (ret) {
708 		printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
709 		err = -ENOMEM;
710 		goto fail_iput;
711 	}
712 #endif
713 	__setup_root(512, 512, 512, 512, tree_root,
714 		     fs_info, BTRFS_ROOT_TREE_OBJECTID);
715 
716 	fs_info->sb_buffer = read_tree_block(tree_root,
717 					     BTRFS_SUPER_INFO_OFFSET,
718 					     512);
719 
720 	if (!fs_info->sb_buffer)
721 		goto fail_iput;
722 
723 	read_extent_buffer(fs_info->sb_buffer, &fs_info->super_copy, 0,
724 			   sizeof(fs_info->super_copy));
725 
726 	read_extent_buffer(fs_info->sb_buffer, fs_info->fsid,
727 			   (unsigned long)btrfs_super_fsid(fs_info->sb_buffer),
728 			   BTRFS_FSID_SIZE);
729 	disk_super = &fs_info->super_copy;
730 	if (!btrfs_super_root(disk_super))
731 		goto fail_sb_buffer;
732 
733 	nodesize = btrfs_super_nodesize(disk_super);
734 	leafsize = btrfs_super_leafsize(disk_super);
735 	sectorsize = btrfs_super_sectorsize(disk_super);
736 	stripesize = btrfs_super_stripesize(disk_super);
737 	tree_root->nodesize = nodesize;
738 	tree_root->leafsize = leafsize;
739 	tree_root->sectorsize = sectorsize;
740 	tree_root->stripesize = stripesize;
741 	sb_set_blocksize(sb, sectorsize);
742 
743 	i_size_write(fs_info->btree_inode,
744 		     btrfs_super_total_bytes(disk_super));
745 
746 	if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
747 		    sizeof(disk_super->magic))) {
748 		printk("btrfs: valid FS not found on %s\n", sb->s_id);
749 		goto fail_sb_buffer;
750 	}
751 
752 	blocksize = btrfs_level_size(tree_root,
753 				     btrfs_super_root_level(disk_super));
754 
755 	tree_root->node = read_tree_block(tree_root,
756 					  btrfs_super_root(disk_super),
757 					  blocksize);
758 	if (!tree_root->node)
759 		goto fail_sb_buffer;
760 
761 	mutex_lock(&fs_info->fs_mutex);
762 
763 	ret = find_and_setup_root(tree_root, fs_info,
764 				  BTRFS_EXTENT_TREE_OBJECTID, extent_root);
765 	if (ret) {
766 		mutex_unlock(&fs_info->fs_mutex);
767 		goto fail_tree_root;
768 	}
769 
770 	btrfs_read_block_groups(extent_root);
771 
772 	fs_info->generation = btrfs_super_generation(disk_super) + 1;
773 	mutex_unlock(&fs_info->fs_mutex);
774 	return tree_root;
775 
776 fail_tree_root:
777 	free_extent_buffer(tree_root->node);
778 fail_sb_buffer:
779 	free_extent_buffer(fs_info->sb_buffer);
780 fail_iput:
781 	iput(fs_info->btree_inode);
782 fail:
783 	kfree(extent_root);
784 	kfree(tree_root);
785 	kfree(fs_info);
786 	return ERR_PTR(err);
787 }
788 
789 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
790 		      *root)
791 {
792 	int ret;
793 	struct extent_buffer *super = root->fs_info->sb_buffer;
794 	struct inode *btree_inode = root->fs_info->btree_inode;
795 	struct super_block *sb = root->fs_info->sb;
796 
797 	if (!btrfs_test_opt(root, NOBARRIER))
798 		blkdev_issue_flush(sb->s_bdev, NULL);
799 	set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, super);
800 	ret = sync_page_range_nolock(btree_inode, btree_inode->i_mapping,
801 				     super->start, super->len);
802 	if (!btrfs_test_opt(root, NOBARRIER))
803 		blkdev_issue_flush(sb->s_bdev, NULL);
804 	return ret;
805 }
806 
807 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
808 {
809 	radix_tree_delete(&fs_info->fs_roots_radix,
810 			  (unsigned long)root->root_key.objectid);
811 	if (root->in_sysfs)
812 		btrfs_sysfs_del_root(root);
813 	if (root->inode)
814 		iput(root->inode);
815 	if (root->node)
816 		free_extent_buffer(root->node);
817 	if (root->commit_root)
818 		free_extent_buffer(root->commit_root);
819 	if (root->name)
820 		kfree(root->name);
821 	kfree(root);
822 	return 0;
823 }
824 
825 static int del_fs_roots(struct btrfs_fs_info *fs_info)
826 {
827 	int ret;
828 	struct btrfs_root *gang[8];
829 	int i;
830 
831 	while(1) {
832 		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
833 					     (void **)gang, 0,
834 					     ARRAY_SIZE(gang));
835 		if (!ret)
836 			break;
837 		for (i = 0; i < ret; i++)
838 			btrfs_free_fs_root(fs_info, gang[i]);
839 	}
840 	return 0;
841 }
842 
843 int close_ctree(struct btrfs_root *root)
844 {
845 	int ret;
846 	struct btrfs_trans_handle *trans;
847 	struct btrfs_fs_info *fs_info = root->fs_info;
848 
849 	fs_info->closing = 1;
850 	btrfs_transaction_flush_work(root);
851 	mutex_lock(&fs_info->fs_mutex);
852 	btrfs_defrag_dirty_roots(root->fs_info);
853 	trans = btrfs_start_transaction(root, 1);
854 	ret = btrfs_commit_transaction(trans, root);
855 	/* run commit again to  drop the original snapshot */
856 	trans = btrfs_start_transaction(root, 1);
857 	btrfs_commit_transaction(trans, root);
858 	ret = btrfs_write_and_wait_transaction(NULL, root);
859 	BUG_ON(ret);
860 	write_ctree_super(NULL, root);
861 	mutex_unlock(&fs_info->fs_mutex);
862 
863 	if (fs_info->extent_root->node)
864 		free_extent_buffer(fs_info->extent_root->node);
865 
866 	if (fs_info->tree_root->node)
867 		free_extent_buffer(fs_info->tree_root->node);
868 
869 	free_extent_buffer(fs_info->sb_buffer);
870 
871 	btrfs_free_block_groups(root->fs_info);
872 	del_fs_roots(fs_info);
873 
874 	filemap_write_and_wait(fs_info->btree_inode->i_mapping);
875 
876 	extent_io_tree_empty_lru(&fs_info->free_space_cache);
877 	extent_io_tree_empty_lru(&fs_info->block_group_cache);
878 	extent_io_tree_empty_lru(&fs_info->pinned_extents);
879 	extent_io_tree_empty_lru(&fs_info->pending_del);
880 	extent_io_tree_empty_lru(&fs_info->extent_ins);
881 	extent_io_tree_empty_lru(&BTRFS_I(fs_info->btree_inode)->io_tree);
882 
883 	truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
884 
885 	iput(fs_info->btree_inode);
886 #if 0
887 	while(!list_empty(&fs_info->hashers)) {
888 		struct btrfs_hasher *hasher;
889 		hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
890 				    hashers);
891 		list_del(&hasher->hashers);
892 		crypto_free_hash(&fs_info->hash_tfm);
893 		kfree(hasher);
894 	}
895 #endif
896 	kfree(fs_info->extent_root);
897 	kfree(fs_info->tree_root);
898 	return 0;
899 }
900 
901 int btrfs_buffer_uptodate(struct extent_buffer *buf)
902 {
903 	struct inode *btree_inode = buf->first_page->mapping->host;
904 	return extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
905 }
906 
907 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
908 {
909 	struct inode *btree_inode = buf->first_page->mapping->host;
910 	return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
911 					  buf);
912 }
913 
914 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
915 {
916 	struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
917 	u64 transid = btrfs_header_generation(buf);
918 	struct inode *btree_inode = root->fs_info->btree_inode;
919 
920 	if (transid != root->fs_info->generation) {
921 		printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
922 			(unsigned long long)buf->start,
923 			transid, root->fs_info->generation);
924 		WARN_ON(1);
925 	}
926 	set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
927 }
928 
929 void btrfs_throttle(struct btrfs_root *root)
930 {
931 	struct backing_dev_info *bdi;
932 
933 	bdi = root->fs_info->sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
934 	if (root->fs_info->throttles && bdi_write_congested(bdi)) {
935 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,18)
936 		congestion_wait(WRITE, HZ/20);
937 #else
938 		blk_congestion_wait(WRITE, HZ/20);
939 #endif
940 	}
941 }
942 
943 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
944 {
945 	balance_dirty_pages_ratelimited_nr(
946 			root->fs_info->btree_inode->i_mapping, 1);
947 }
948 
949 void btrfs_set_buffer_defrag(struct extent_buffer *buf)
950 {
951 	struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
952 	struct inode *btree_inode = root->fs_info->btree_inode;
953 	set_extent_bits(&BTRFS_I(btree_inode)->io_tree, buf->start,
954 			buf->start + buf->len - 1, EXTENT_DEFRAG, GFP_NOFS);
955 }
956 
957 void btrfs_set_buffer_defrag_done(struct extent_buffer *buf)
958 {
959 	struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
960 	struct inode *btree_inode = root->fs_info->btree_inode;
961 	set_extent_bits(&BTRFS_I(btree_inode)->io_tree, buf->start,
962 			buf->start + buf->len - 1, EXTENT_DEFRAG_DONE,
963 			GFP_NOFS);
964 }
965 
966 int btrfs_buffer_defrag(struct extent_buffer *buf)
967 {
968 	struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
969 	struct inode *btree_inode = root->fs_info->btree_inode;
970 	return test_range_bit(&BTRFS_I(btree_inode)->io_tree,
971 		     buf->start, buf->start + buf->len - 1, EXTENT_DEFRAG, 0);
972 }
973 
974 int btrfs_buffer_defrag_done(struct extent_buffer *buf)
975 {
976 	struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
977 	struct inode *btree_inode = root->fs_info->btree_inode;
978 	return test_range_bit(&BTRFS_I(btree_inode)->io_tree,
979 		     buf->start, buf->start + buf->len - 1,
980 		     EXTENT_DEFRAG_DONE, 0);
981 }
982 
983 int btrfs_clear_buffer_defrag_done(struct extent_buffer *buf)
984 {
985 	struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
986 	struct inode *btree_inode = root->fs_info->btree_inode;
987 	return clear_extent_bits(&BTRFS_I(btree_inode)->io_tree,
988 		     buf->start, buf->start + buf->len - 1,
989 		     EXTENT_DEFRAG_DONE, GFP_NOFS);
990 }
991 
992 int btrfs_clear_buffer_defrag(struct extent_buffer *buf)
993 {
994 	struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
995 	struct inode *btree_inode = root->fs_info->btree_inode;
996 	return clear_extent_bits(&BTRFS_I(btree_inode)->io_tree,
997 		     buf->start, buf->start + buf->len - 1,
998 		     EXTENT_DEFRAG, GFP_NOFS);
999 }
1000 
1001 int btrfs_read_buffer(struct extent_buffer *buf)
1002 {
1003 	struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1004 	struct inode *btree_inode = root->fs_info->btree_inode;
1005 	return read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1006 					buf, 0, 1);
1007 }
1008 
1009 static struct extent_io_ops btree_extent_io_ops = {
1010 	.writepage_io_hook = btree_writepage_io_hook,
1011 };
1012