xref: /linux/fs/btrfs/free-space-tree.c (revision 24bce201d79807b668bf9d9e0aca801c5c0d5f78)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2015 Facebook.  All rights reserved.
4  */
5 
6 #include <linux/kernel.h>
7 #include <linux/sched/mm.h>
8 #include "ctree.h"
9 #include "disk-io.h"
10 #include "locking.h"
11 #include "free-space-tree.h"
12 #include "transaction.h"
13 #include "block-group.h"
14 
15 static int __add_block_group_free_space(struct btrfs_trans_handle *trans,
16 					struct btrfs_block_group *block_group,
17 					struct btrfs_path *path);
18 
19 static struct btrfs_root *btrfs_free_space_root(
20 				struct btrfs_block_group *block_group)
21 {
22 	struct btrfs_key key = {
23 		.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID,
24 		.type = BTRFS_ROOT_ITEM_KEY,
25 		.offset = 0,
26 	};
27 
28 	if (btrfs_fs_incompat(block_group->fs_info, EXTENT_TREE_V2))
29 		key.offset = block_group->global_root_id;
30 	return btrfs_global_root(block_group->fs_info, &key);
31 }
32 
33 void set_free_space_tree_thresholds(struct btrfs_block_group *cache)
34 {
35 	u32 bitmap_range;
36 	size_t bitmap_size;
37 	u64 num_bitmaps, total_bitmap_size;
38 
39 	if (WARN_ON(cache->length == 0))
40 		btrfs_warn(cache->fs_info, "block group %llu length is zero",
41 			   cache->start);
42 
43 	/*
44 	 * We convert to bitmaps when the disk space required for using extents
45 	 * exceeds that required for using bitmaps.
46 	 */
47 	bitmap_range = cache->fs_info->sectorsize * BTRFS_FREE_SPACE_BITMAP_BITS;
48 	num_bitmaps = div_u64(cache->length + bitmap_range - 1, bitmap_range);
49 	bitmap_size = sizeof(struct btrfs_item) + BTRFS_FREE_SPACE_BITMAP_SIZE;
50 	total_bitmap_size = num_bitmaps * bitmap_size;
51 	cache->bitmap_high_thresh = div_u64(total_bitmap_size,
52 					    sizeof(struct btrfs_item));
53 
54 	/*
55 	 * We allow for a small buffer between the high threshold and low
56 	 * threshold to avoid thrashing back and forth between the two formats.
57 	 */
58 	if (cache->bitmap_high_thresh > 100)
59 		cache->bitmap_low_thresh = cache->bitmap_high_thresh - 100;
60 	else
61 		cache->bitmap_low_thresh = 0;
62 }
63 
64 static int add_new_free_space_info(struct btrfs_trans_handle *trans,
65 				   struct btrfs_block_group *block_group,
66 				   struct btrfs_path *path)
67 {
68 	struct btrfs_root *root = btrfs_free_space_root(block_group);
69 	struct btrfs_free_space_info *info;
70 	struct btrfs_key key;
71 	struct extent_buffer *leaf;
72 	int ret;
73 
74 	key.objectid = block_group->start;
75 	key.type = BTRFS_FREE_SPACE_INFO_KEY;
76 	key.offset = block_group->length;
77 
78 	ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*info));
79 	if (ret)
80 		goto out;
81 
82 	leaf = path->nodes[0];
83 	info = btrfs_item_ptr(leaf, path->slots[0],
84 			      struct btrfs_free_space_info);
85 	btrfs_set_free_space_extent_count(leaf, info, 0);
86 	btrfs_set_free_space_flags(leaf, info, 0);
87 	btrfs_mark_buffer_dirty(leaf);
88 
89 	ret = 0;
90 out:
91 	btrfs_release_path(path);
92 	return ret;
93 }
94 
95 EXPORT_FOR_TESTS
96 struct btrfs_free_space_info *search_free_space_info(
97 		struct btrfs_trans_handle *trans,
98 		struct btrfs_block_group *block_group,
99 		struct btrfs_path *path, int cow)
100 {
101 	struct btrfs_fs_info *fs_info = block_group->fs_info;
102 	struct btrfs_root *root = btrfs_free_space_root(block_group);
103 	struct btrfs_key key;
104 	int ret;
105 
106 	key.objectid = block_group->start;
107 	key.type = BTRFS_FREE_SPACE_INFO_KEY;
108 	key.offset = block_group->length;
109 
110 	ret = btrfs_search_slot(trans, root, &key, path, 0, cow);
111 	if (ret < 0)
112 		return ERR_PTR(ret);
113 	if (ret != 0) {
114 		btrfs_warn(fs_info, "missing free space info for %llu",
115 			   block_group->start);
116 		ASSERT(0);
117 		return ERR_PTR(-ENOENT);
118 	}
119 
120 	return btrfs_item_ptr(path->nodes[0], path->slots[0],
121 			      struct btrfs_free_space_info);
122 }
123 
124 /*
125  * btrfs_search_slot() but we're looking for the greatest key less than the
126  * passed key.
127  */
128 static int btrfs_search_prev_slot(struct btrfs_trans_handle *trans,
129 				  struct btrfs_root *root,
130 				  struct btrfs_key *key, struct btrfs_path *p,
131 				  int ins_len, int cow)
132 {
133 	int ret;
134 
135 	ret = btrfs_search_slot(trans, root, key, p, ins_len, cow);
136 	if (ret < 0)
137 		return ret;
138 
139 	if (ret == 0) {
140 		ASSERT(0);
141 		return -EIO;
142 	}
143 
144 	if (p->slots[0] == 0) {
145 		ASSERT(0);
146 		return -EIO;
147 	}
148 	p->slots[0]--;
149 
150 	return 0;
151 }
152 
153 static inline u32 free_space_bitmap_size(const struct btrfs_fs_info *fs_info,
154 					 u64 size)
155 {
156 	return DIV_ROUND_UP(size >> fs_info->sectorsize_bits, BITS_PER_BYTE);
157 }
158 
159 static unsigned long *alloc_bitmap(u32 bitmap_size)
160 {
161 	unsigned long *ret;
162 	unsigned int nofs_flag;
163 	u32 bitmap_rounded_size = round_up(bitmap_size, sizeof(unsigned long));
164 
165 	/*
166 	 * GFP_NOFS doesn't work with kvmalloc(), but we really can't recurse
167 	 * into the filesystem as the free space bitmap can be modified in the
168 	 * critical section of a transaction commit.
169 	 *
170 	 * TODO: push the memalloc_nofs_{save,restore}() to the caller where we
171 	 * know that recursion is unsafe.
172 	 */
173 	nofs_flag = memalloc_nofs_save();
174 	ret = kvzalloc(bitmap_rounded_size, GFP_KERNEL);
175 	memalloc_nofs_restore(nofs_flag);
176 	return ret;
177 }
178 
179 static void le_bitmap_set(unsigned long *map, unsigned int start, int len)
180 {
181 	u8 *p = ((u8 *)map) + BIT_BYTE(start);
182 	const unsigned int size = start + len;
183 	int bits_to_set = BITS_PER_BYTE - (start % BITS_PER_BYTE);
184 	u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(start);
185 
186 	while (len - bits_to_set >= 0) {
187 		*p |= mask_to_set;
188 		len -= bits_to_set;
189 		bits_to_set = BITS_PER_BYTE;
190 		mask_to_set = ~0;
191 		p++;
192 	}
193 	if (len) {
194 		mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
195 		*p |= mask_to_set;
196 	}
197 }
198 
199 EXPORT_FOR_TESTS
200 int convert_free_space_to_bitmaps(struct btrfs_trans_handle *trans,
201 				  struct btrfs_block_group *block_group,
202 				  struct btrfs_path *path)
203 {
204 	struct btrfs_fs_info *fs_info = trans->fs_info;
205 	struct btrfs_root *root = btrfs_free_space_root(block_group);
206 	struct btrfs_free_space_info *info;
207 	struct btrfs_key key, found_key;
208 	struct extent_buffer *leaf;
209 	unsigned long *bitmap;
210 	char *bitmap_cursor;
211 	u64 start, end;
212 	u64 bitmap_range, i;
213 	u32 bitmap_size, flags, expected_extent_count;
214 	u32 extent_count = 0;
215 	int done = 0, nr;
216 	int ret;
217 
218 	bitmap_size = free_space_bitmap_size(fs_info, block_group->length);
219 	bitmap = alloc_bitmap(bitmap_size);
220 	if (!bitmap) {
221 		ret = -ENOMEM;
222 		goto out;
223 	}
224 
225 	start = block_group->start;
226 	end = block_group->start + block_group->length;
227 
228 	key.objectid = end - 1;
229 	key.type = (u8)-1;
230 	key.offset = (u64)-1;
231 
232 	while (!done) {
233 		ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
234 		if (ret)
235 			goto out;
236 
237 		leaf = path->nodes[0];
238 		nr = 0;
239 		path->slots[0]++;
240 		while (path->slots[0] > 0) {
241 			btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
242 
243 			if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
244 				ASSERT(found_key.objectid == block_group->start);
245 				ASSERT(found_key.offset == block_group->length);
246 				done = 1;
247 				break;
248 			} else if (found_key.type == BTRFS_FREE_SPACE_EXTENT_KEY) {
249 				u64 first, last;
250 
251 				ASSERT(found_key.objectid >= start);
252 				ASSERT(found_key.objectid < end);
253 				ASSERT(found_key.objectid + found_key.offset <= end);
254 
255 				first = div_u64(found_key.objectid - start,
256 						fs_info->sectorsize);
257 				last = div_u64(found_key.objectid + found_key.offset - start,
258 					       fs_info->sectorsize);
259 				le_bitmap_set(bitmap, first, last - first);
260 
261 				extent_count++;
262 				nr++;
263 				path->slots[0]--;
264 			} else {
265 				ASSERT(0);
266 			}
267 		}
268 
269 		ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
270 		if (ret)
271 			goto out;
272 		btrfs_release_path(path);
273 	}
274 
275 	info = search_free_space_info(trans, block_group, path, 1);
276 	if (IS_ERR(info)) {
277 		ret = PTR_ERR(info);
278 		goto out;
279 	}
280 	leaf = path->nodes[0];
281 	flags = btrfs_free_space_flags(leaf, info);
282 	flags |= BTRFS_FREE_SPACE_USING_BITMAPS;
283 	btrfs_set_free_space_flags(leaf, info, flags);
284 	expected_extent_count = btrfs_free_space_extent_count(leaf, info);
285 	btrfs_mark_buffer_dirty(leaf);
286 	btrfs_release_path(path);
287 
288 	if (extent_count != expected_extent_count) {
289 		btrfs_err(fs_info,
290 			  "incorrect extent count for %llu; counted %u, expected %u",
291 			  block_group->start, extent_count,
292 			  expected_extent_count);
293 		ASSERT(0);
294 		ret = -EIO;
295 		goto out;
296 	}
297 
298 	bitmap_cursor = (char *)bitmap;
299 	bitmap_range = fs_info->sectorsize * BTRFS_FREE_SPACE_BITMAP_BITS;
300 	i = start;
301 	while (i < end) {
302 		unsigned long ptr;
303 		u64 extent_size;
304 		u32 data_size;
305 
306 		extent_size = min(end - i, bitmap_range);
307 		data_size = free_space_bitmap_size(fs_info, extent_size);
308 
309 		key.objectid = i;
310 		key.type = BTRFS_FREE_SPACE_BITMAP_KEY;
311 		key.offset = extent_size;
312 
313 		ret = btrfs_insert_empty_item(trans, root, path, &key,
314 					      data_size);
315 		if (ret)
316 			goto out;
317 
318 		leaf = path->nodes[0];
319 		ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
320 		write_extent_buffer(leaf, bitmap_cursor, ptr,
321 				    data_size);
322 		btrfs_mark_buffer_dirty(leaf);
323 		btrfs_release_path(path);
324 
325 		i += extent_size;
326 		bitmap_cursor += data_size;
327 	}
328 
329 	ret = 0;
330 out:
331 	kvfree(bitmap);
332 	if (ret)
333 		btrfs_abort_transaction(trans, ret);
334 	return ret;
335 }
336 
337 EXPORT_FOR_TESTS
338 int convert_free_space_to_extents(struct btrfs_trans_handle *trans,
339 				  struct btrfs_block_group *block_group,
340 				  struct btrfs_path *path)
341 {
342 	struct btrfs_fs_info *fs_info = trans->fs_info;
343 	struct btrfs_root *root = btrfs_free_space_root(block_group);
344 	struct btrfs_free_space_info *info;
345 	struct btrfs_key key, found_key;
346 	struct extent_buffer *leaf;
347 	unsigned long *bitmap;
348 	u64 start, end;
349 	u32 bitmap_size, flags, expected_extent_count;
350 	unsigned long nrbits, start_bit, end_bit;
351 	u32 extent_count = 0;
352 	int done = 0, nr;
353 	int ret;
354 
355 	bitmap_size = free_space_bitmap_size(fs_info, block_group->length);
356 	bitmap = alloc_bitmap(bitmap_size);
357 	if (!bitmap) {
358 		ret = -ENOMEM;
359 		goto out;
360 	}
361 
362 	start = block_group->start;
363 	end = block_group->start + block_group->length;
364 
365 	key.objectid = end - 1;
366 	key.type = (u8)-1;
367 	key.offset = (u64)-1;
368 
369 	while (!done) {
370 		ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
371 		if (ret)
372 			goto out;
373 
374 		leaf = path->nodes[0];
375 		nr = 0;
376 		path->slots[0]++;
377 		while (path->slots[0] > 0) {
378 			btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
379 
380 			if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
381 				ASSERT(found_key.objectid == block_group->start);
382 				ASSERT(found_key.offset == block_group->length);
383 				done = 1;
384 				break;
385 			} else if (found_key.type == BTRFS_FREE_SPACE_BITMAP_KEY) {
386 				unsigned long ptr;
387 				char *bitmap_cursor;
388 				u32 bitmap_pos, data_size;
389 
390 				ASSERT(found_key.objectid >= start);
391 				ASSERT(found_key.objectid < end);
392 				ASSERT(found_key.objectid + found_key.offset <= end);
393 
394 				bitmap_pos = div_u64(found_key.objectid - start,
395 						     fs_info->sectorsize *
396 						     BITS_PER_BYTE);
397 				bitmap_cursor = ((char *)bitmap) + bitmap_pos;
398 				data_size = free_space_bitmap_size(fs_info,
399 								found_key.offset);
400 
401 				ptr = btrfs_item_ptr_offset(leaf, path->slots[0] - 1);
402 				read_extent_buffer(leaf, bitmap_cursor, ptr,
403 						   data_size);
404 
405 				nr++;
406 				path->slots[0]--;
407 			} else {
408 				ASSERT(0);
409 			}
410 		}
411 
412 		ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
413 		if (ret)
414 			goto out;
415 		btrfs_release_path(path);
416 	}
417 
418 	info = search_free_space_info(trans, block_group, path, 1);
419 	if (IS_ERR(info)) {
420 		ret = PTR_ERR(info);
421 		goto out;
422 	}
423 	leaf = path->nodes[0];
424 	flags = btrfs_free_space_flags(leaf, info);
425 	flags &= ~BTRFS_FREE_SPACE_USING_BITMAPS;
426 	btrfs_set_free_space_flags(leaf, info, flags);
427 	expected_extent_count = btrfs_free_space_extent_count(leaf, info);
428 	btrfs_mark_buffer_dirty(leaf);
429 	btrfs_release_path(path);
430 
431 	nrbits = block_group->length >> block_group->fs_info->sectorsize_bits;
432 	start_bit = find_next_bit_le(bitmap, nrbits, 0);
433 
434 	while (start_bit < nrbits) {
435 		end_bit = find_next_zero_bit_le(bitmap, nrbits, start_bit);
436 		ASSERT(start_bit < end_bit);
437 
438 		key.objectid = start + start_bit * block_group->fs_info->sectorsize;
439 		key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
440 		key.offset = (end_bit - start_bit) * block_group->fs_info->sectorsize;
441 
442 		ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
443 		if (ret)
444 			goto out;
445 		btrfs_release_path(path);
446 
447 		extent_count++;
448 
449 		start_bit = find_next_bit_le(bitmap, nrbits, end_bit);
450 	}
451 
452 	if (extent_count != expected_extent_count) {
453 		btrfs_err(fs_info,
454 			  "incorrect extent count for %llu; counted %u, expected %u",
455 			  block_group->start, extent_count,
456 			  expected_extent_count);
457 		ASSERT(0);
458 		ret = -EIO;
459 		goto out;
460 	}
461 
462 	ret = 0;
463 out:
464 	kvfree(bitmap);
465 	if (ret)
466 		btrfs_abort_transaction(trans, ret);
467 	return ret;
468 }
469 
470 static int update_free_space_extent_count(struct btrfs_trans_handle *trans,
471 					  struct btrfs_block_group *block_group,
472 					  struct btrfs_path *path,
473 					  int new_extents)
474 {
475 	struct btrfs_free_space_info *info;
476 	u32 flags;
477 	u32 extent_count;
478 	int ret = 0;
479 
480 	if (new_extents == 0)
481 		return 0;
482 
483 	info = search_free_space_info(trans, block_group, path, 1);
484 	if (IS_ERR(info)) {
485 		ret = PTR_ERR(info);
486 		goto out;
487 	}
488 	flags = btrfs_free_space_flags(path->nodes[0], info);
489 	extent_count = btrfs_free_space_extent_count(path->nodes[0], info);
490 
491 	extent_count += new_extents;
492 	btrfs_set_free_space_extent_count(path->nodes[0], info, extent_count);
493 	btrfs_mark_buffer_dirty(path->nodes[0]);
494 	btrfs_release_path(path);
495 
496 	if (!(flags & BTRFS_FREE_SPACE_USING_BITMAPS) &&
497 	    extent_count > block_group->bitmap_high_thresh) {
498 		ret = convert_free_space_to_bitmaps(trans, block_group, path);
499 	} else if ((flags & BTRFS_FREE_SPACE_USING_BITMAPS) &&
500 		   extent_count < block_group->bitmap_low_thresh) {
501 		ret = convert_free_space_to_extents(trans, block_group, path);
502 	}
503 
504 out:
505 	return ret;
506 }
507 
508 EXPORT_FOR_TESTS
509 int free_space_test_bit(struct btrfs_block_group *block_group,
510 			struct btrfs_path *path, u64 offset)
511 {
512 	struct extent_buffer *leaf;
513 	struct btrfs_key key;
514 	u64 found_start, found_end;
515 	unsigned long ptr, i;
516 
517 	leaf = path->nodes[0];
518 	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
519 	ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
520 
521 	found_start = key.objectid;
522 	found_end = key.objectid + key.offset;
523 	ASSERT(offset >= found_start && offset < found_end);
524 
525 	ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
526 	i = div_u64(offset - found_start,
527 		    block_group->fs_info->sectorsize);
528 	return !!extent_buffer_test_bit(leaf, ptr, i);
529 }
530 
531 static void free_space_set_bits(struct btrfs_block_group *block_group,
532 				struct btrfs_path *path, u64 *start, u64 *size,
533 				int bit)
534 {
535 	struct btrfs_fs_info *fs_info = block_group->fs_info;
536 	struct extent_buffer *leaf;
537 	struct btrfs_key key;
538 	u64 end = *start + *size;
539 	u64 found_start, found_end;
540 	unsigned long ptr, first, last;
541 
542 	leaf = path->nodes[0];
543 	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
544 	ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
545 
546 	found_start = key.objectid;
547 	found_end = key.objectid + key.offset;
548 	ASSERT(*start >= found_start && *start < found_end);
549 	ASSERT(end > found_start);
550 
551 	if (end > found_end)
552 		end = found_end;
553 
554 	ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
555 	first = (*start - found_start) >> fs_info->sectorsize_bits;
556 	last = (end - found_start) >> fs_info->sectorsize_bits;
557 	if (bit)
558 		extent_buffer_bitmap_set(leaf, ptr, first, last - first);
559 	else
560 		extent_buffer_bitmap_clear(leaf, ptr, first, last - first);
561 	btrfs_mark_buffer_dirty(leaf);
562 
563 	*size -= end - *start;
564 	*start = end;
565 }
566 
567 /*
568  * We can't use btrfs_next_item() in modify_free_space_bitmap() because
569  * btrfs_next_leaf() doesn't get the path for writing. We can forgo the fancy
570  * tree walking in btrfs_next_leaf() anyways because we know exactly what we're
571  * looking for.
572  */
573 static int free_space_next_bitmap(struct btrfs_trans_handle *trans,
574 				  struct btrfs_root *root, struct btrfs_path *p)
575 {
576 	struct btrfs_key key;
577 
578 	if (p->slots[0] + 1 < btrfs_header_nritems(p->nodes[0])) {
579 		p->slots[0]++;
580 		return 0;
581 	}
582 
583 	btrfs_item_key_to_cpu(p->nodes[0], &key, p->slots[0]);
584 	btrfs_release_path(p);
585 
586 	key.objectid += key.offset;
587 	key.type = (u8)-1;
588 	key.offset = (u64)-1;
589 
590 	return btrfs_search_prev_slot(trans, root, &key, p, 0, 1);
591 }
592 
593 /*
594  * If remove is 1, then we are removing free space, thus clearing bits in the
595  * bitmap. If remove is 0, then we are adding free space, thus setting bits in
596  * the bitmap.
597  */
598 static int modify_free_space_bitmap(struct btrfs_trans_handle *trans,
599 				    struct btrfs_block_group *block_group,
600 				    struct btrfs_path *path,
601 				    u64 start, u64 size, int remove)
602 {
603 	struct btrfs_root *root = btrfs_free_space_root(block_group);
604 	struct btrfs_key key;
605 	u64 end = start + size;
606 	u64 cur_start, cur_size;
607 	int prev_bit, next_bit;
608 	int new_extents;
609 	int ret;
610 
611 	/*
612 	 * Read the bit for the block immediately before the extent of space if
613 	 * that block is within the block group.
614 	 */
615 	if (start > block_group->start) {
616 		u64 prev_block = start - block_group->fs_info->sectorsize;
617 
618 		key.objectid = prev_block;
619 		key.type = (u8)-1;
620 		key.offset = (u64)-1;
621 
622 		ret = btrfs_search_prev_slot(trans, root, &key, path, 0, 1);
623 		if (ret)
624 			goto out;
625 
626 		prev_bit = free_space_test_bit(block_group, path, prev_block);
627 
628 		/* The previous block may have been in the previous bitmap. */
629 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
630 		if (start >= key.objectid + key.offset) {
631 			ret = free_space_next_bitmap(trans, root, path);
632 			if (ret)
633 				goto out;
634 		}
635 	} else {
636 		key.objectid = start;
637 		key.type = (u8)-1;
638 		key.offset = (u64)-1;
639 
640 		ret = btrfs_search_prev_slot(trans, root, &key, path, 0, 1);
641 		if (ret)
642 			goto out;
643 
644 		prev_bit = -1;
645 	}
646 
647 	/*
648 	 * Iterate over all of the bitmaps overlapped by the extent of space,
649 	 * clearing/setting bits as required.
650 	 */
651 	cur_start = start;
652 	cur_size = size;
653 	while (1) {
654 		free_space_set_bits(block_group, path, &cur_start, &cur_size,
655 				    !remove);
656 		if (cur_size == 0)
657 			break;
658 		ret = free_space_next_bitmap(trans, root, path);
659 		if (ret)
660 			goto out;
661 	}
662 
663 	/*
664 	 * Read the bit for the block immediately after the extent of space if
665 	 * that block is within the block group.
666 	 */
667 	if (end < block_group->start + block_group->length) {
668 		/* The next block may be in the next bitmap. */
669 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
670 		if (end >= key.objectid + key.offset) {
671 			ret = free_space_next_bitmap(trans, root, path);
672 			if (ret)
673 				goto out;
674 		}
675 
676 		next_bit = free_space_test_bit(block_group, path, end);
677 	} else {
678 		next_bit = -1;
679 	}
680 
681 	if (remove) {
682 		new_extents = -1;
683 		if (prev_bit == 1) {
684 			/* Leftover on the left. */
685 			new_extents++;
686 		}
687 		if (next_bit == 1) {
688 			/* Leftover on the right. */
689 			new_extents++;
690 		}
691 	} else {
692 		new_extents = 1;
693 		if (prev_bit == 1) {
694 			/* Merging with neighbor on the left. */
695 			new_extents--;
696 		}
697 		if (next_bit == 1) {
698 			/* Merging with neighbor on the right. */
699 			new_extents--;
700 		}
701 	}
702 
703 	btrfs_release_path(path);
704 	ret = update_free_space_extent_count(trans, block_group, path,
705 					     new_extents);
706 
707 out:
708 	return ret;
709 }
710 
711 static int remove_free_space_extent(struct btrfs_trans_handle *trans,
712 				    struct btrfs_block_group *block_group,
713 				    struct btrfs_path *path,
714 				    u64 start, u64 size)
715 {
716 	struct btrfs_root *root = btrfs_free_space_root(block_group);
717 	struct btrfs_key key;
718 	u64 found_start, found_end;
719 	u64 end = start + size;
720 	int new_extents = -1;
721 	int ret;
722 
723 	key.objectid = start;
724 	key.type = (u8)-1;
725 	key.offset = (u64)-1;
726 
727 	ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
728 	if (ret)
729 		goto out;
730 
731 	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
732 
733 	ASSERT(key.type == BTRFS_FREE_SPACE_EXTENT_KEY);
734 
735 	found_start = key.objectid;
736 	found_end = key.objectid + key.offset;
737 	ASSERT(start >= found_start && end <= found_end);
738 
739 	/*
740 	 * Okay, now that we've found the free space extent which contains the
741 	 * free space that we are removing, there are four cases:
742 	 *
743 	 * 1. We're using the whole extent: delete the key we found and
744 	 * decrement the free space extent count.
745 	 * 2. We are using part of the extent starting at the beginning: delete
746 	 * the key we found and insert a new key representing the leftover at
747 	 * the end. There is no net change in the number of extents.
748 	 * 3. We are using part of the extent ending at the end: delete the key
749 	 * we found and insert a new key representing the leftover at the
750 	 * beginning. There is no net change in the number of extents.
751 	 * 4. We are using part of the extent in the middle: delete the key we
752 	 * found and insert two new keys representing the leftovers on each
753 	 * side. Where we used to have one extent, we now have two, so increment
754 	 * the extent count. We may need to convert the block group to bitmaps
755 	 * as a result.
756 	 */
757 
758 	/* Delete the existing key (cases 1-4). */
759 	ret = btrfs_del_item(trans, root, path);
760 	if (ret)
761 		goto out;
762 
763 	/* Add a key for leftovers at the beginning (cases 3 and 4). */
764 	if (start > found_start) {
765 		key.objectid = found_start;
766 		key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
767 		key.offset = start - found_start;
768 
769 		btrfs_release_path(path);
770 		ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
771 		if (ret)
772 			goto out;
773 		new_extents++;
774 	}
775 
776 	/* Add a key for leftovers at the end (cases 2 and 4). */
777 	if (end < found_end) {
778 		key.objectid = end;
779 		key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
780 		key.offset = found_end - end;
781 
782 		btrfs_release_path(path);
783 		ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
784 		if (ret)
785 			goto out;
786 		new_extents++;
787 	}
788 
789 	btrfs_release_path(path);
790 	ret = update_free_space_extent_count(trans, block_group, path,
791 					     new_extents);
792 
793 out:
794 	return ret;
795 }
796 
797 EXPORT_FOR_TESTS
798 int __remove_from_free_space_tree(struct btrfs_trans_handle *trans,
799 				  struct btrfs_block_group *block_group,
800 				  struct btrfs_path *path, u64 start, u64 size)
801 {
802 	struct btrfs_free_space_info *info;
803 	u32 flags;
804 	int ret;
805 
806 	if (block_group->needs_free_space) {
807 		ret = __add_block_group_free_space(trans, block_group, path);
808 		if (ret)
809 			return ret;
810 	}
811 
812 	info = search_free_space_info(NULL, block_group, path, 0);
813 	if (IS_ERR(info))
814 		return PTR_ERR(info);
815 	flags = btrfs_free_space_flags(path->nodes[0], info);
816 	btrfs_release_path(path);
817 
818 	if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) {
819 		return modify_free_space_bitmap(trans, block_group, path,
820 						start, size, 1);
821 	} else {
822 		return remove_free_space_extent(trans, block_group, path,
823 						start, size);
824 	}
825 }
826 
827 int remove_from_free_space_tree(struct btrfs_trans_handle *trans,
828 				u64 start, u64 size)
829 {
830 	struct btrfs_block_group *block_group;
831 	struct btrfs_path *path;
832 	int ret;
833 
834 	if (!btrfs_fs_compat_ro(trans->fs_info, FREE_SPACE_TREE))
835 		return 0;
836 
837 	path = btrfs_alloc_path();
838 	if (!path) {
839 		ret = -ENOMEM;
840 		goto out;
841 	}
842 
843 	block_group = btrfs_lookup_block_group(trans->fs_info, start);
844 	if (!block_group) {
845 		ASSERT(0);
846 		ret = -ENOENT;
847 		goto out;
848 	}
849 
850 	mutex_lock(&block_group->free_space_lock);
851 	ret = __remove_from_free_space_tree(trans, block_group, path, start,
852 					    size);
853 	mutex_unlock(&block_group->free_space_lock);
854 
855 	btrfs_put_block_group(block_group);
856 out:
857 	btrfs_free_path(path);
858 	if (ret)
859 		btrfs_abort_transaction(trans, ret);
860 	return ret;
861 }
862 
863 static int add_free_space_extent(struct btrfs_trans_handle *trans,
864 				 struct btrfs_block_group *block_group,
865 				 struct btrfs_path *path,
866 				 u64 start, u64 size)
867 {
868 	struct btrfs_root *root = btrfs_free_space_root(block_group);
869 	struct btrfs_key key, new_key;
870 	u64 found_start, found_end;
871 	u64 end = start + size;
872 	int new_extents = 1;
873 	int ret;
874 
875 	/*
876 	 * We are adding a new extent of free space, but we need to merge
877 	 * extents. There are four cases here:
878 	 *
879 	 * 1. The new extent does not have any immediate neighbors to merge
880 	 * with: add the new key and increment the free space extent count. We
881 	 * may need to convert the block group to bitmaps as a result.
882 	 * 2. The new extent has an immediate neighbor before it: remove the
883 	 * previous key and insert a new key combining both of them. There is no
884 	 * net change in the number of extents.
885 	 * 3. The new extent has an immediate neighbor after it: remove the next
886 	 * key and insert a new key combining both of them. There is no net
887 	 * change in the number of extents.
888 	 * 4. The new extent has immediate neighbors on both sides: remove both
889 	 * of the keys and insert a new key combining all of them. Where we used
890 	 * to have two extents, we now have one, so decrement the extent count.
891 	 */
892 
893 	new_key.objectid = start;
894 	new_key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
895 	new_key.offset = size;
896 
897 	/* Search for a neighbor on the left. */
898 	if (start == block_group->start)
899 		goto right;
900 	key.objectid = start - 1;
901 	key.type = (u8)-1;
902 	key.offset = (u64)-1;
903 
904 	ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
905 	if (ret)
906 		goto out;
907 
908 	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
909 
910 	if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) {
911 		ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY);
912 		btrfs_release_path(path);
913 		goto right;
914 	}
915 
916 	found_start = key.objectid;
917 	found_end = key.objectid + key.offset;
918 	ASSERT(found_start >= block_group->start &&
919 	       found_end > block_group->start);
920 	ASSERT(found_start < start && found_end <= start);
921 
922 	/*
923 	 * Delete the neighbor on the left and absorb it into the new key (cases
924 	 * 2 and 4).
925 	 */
926 	if (found_end == start) {
927 		ret = btrfs_del_item(trans, root, path);
928 		if (ret)
929 			goto out;
930 		new_key.objectid = found_start;
931 		new_key.offset += key.offset;
932 		new_extents--;
933 	}
934 	btrfs_release_path(path);
935 
936 right:
937 	/* Search for a neighbor on the right. */
938 	if (end == block_group->start + block_group->length)
939 		goto insert;
940 	key.objectid = end;
941 	key.type = (u8)-1;
942 	key.offset = (u64)-1;
943 
944 	ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
945 	if (ret)
946 		goto out;
947 
948 	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
949 
950 	if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) {
951 		ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY);
952 		btrfs_release_path(path);
953 		goto insert;
954 	}
955 
956 	found_start = key.objectid;
957 	found_end = key.objectid + key.offset;
958 	ASSERT(found_start >= block_group->start &&
959 	       found_end > block_group->start);
960 	ASSERT((found_start < start && found_end <= start) ||
961 	       (found_start >= end && found_end > end));
962 
963 	/*
964 	 * Delete the neighbor on the right and absorb it into the new key
965 	 * (cases 3 and 4).
966 	 */
967 	if (found_start == end) {
968 		ret = btrfs_del_item(trans, root, path);
969 		if (ret)
970 			goto out;
971 		new_key.offset += key.offset;
972 		new_extents--;
973 	}
974 	btrfs_release_path(path);
975 
976 insert:
977 	/* Insert the new key (cases 1-4). */
978 	ret = btrfs_insert_empty_item(trans, root, path, &new_key, 0);
979 	if (ret)
980 		goto out;
981 
982 	btrfs_release_path(path);
983 	ret = update_free_space_extent_count(trans, block_group, path,
984 					     new_extents);
985 
986 out:
987 	return ret;
988 }
989 
990 EXPORT_FOR_TESTS
991 int __add_to_free_space_tree(struct btrfs_trans_handle *trans,
992 			     struct btrfs_block_group *block_group,
993 			     struct btrfs_path *path, u64 start, u64 size)
994 {
995 	struct btrfs_free_space_info *info;
996 	u32 flags;
997 	int ret;
998 
999 	if (block_group->needs_free_space) {
1000 		ret = __add_block_group_free_space(trans, block_group, path);
1001 		if (ret)
1002 			return ret;
1003 	}
1004 
1005 	info = search_free_space_info(NULL, block_group, path, 0);
1006 	if (IS_ERR(info))
1007 		return PTR_ERR(info);
1008 	flags = btrfs_free_space_flags(path->nodes[0], info);
1009 	btrfs_release_path(path);
1010 
1011 	if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) {
1012 		return modify_free_space_bitmap(trans, block_group, path,
1013 						start, size, 0);
1014 	} else {
1015 		return add_free_space_extent(trans, block_group, path, start,
1016 					     size);
1017 	}
1018 }
1019 
1020 int add_to_free_space_tree(struct btrfs_trans_handle *trans,
1021 			   u64 start, u64 size)
1022 {
1023 	struct btrfs_block_group *block_group;
1024 	struct btrfs_path *path;
1025 	int ret;
1026 
1027 	if (!btrfs_fs_compat_ro(trans->fs_info, FREE_SPACE_TREE))
1028 		return 0;
1029 
1030 	path = btrfs_alloc_path();
1031 	if (!path) {
1032 		ret = -ENOMEM;
1033 		goto out;
1034 	}
1035 
1036 	block_group = btrfs_lookup_block_group(trans->fs_info, start);
1037 	if (!block_group) {
1038 		ASSERT(0);
1039 		ret = -ENOENT;
1040 		goto out;
1041 	}
1042 
1043 	mutex_lock(&block_group->free_space_lock);
1044 	ret = __add_to_free_space_tree(trans, block_group, path, start, size);
1045 	mutex_unlock(&block_group->free_space_lock);
1046 
1047 	btrfs_put_block_group(block_group);
1048 out:
1049 	btrfs_free_path(path);
1050 	if (ret)
1051 		btrfs_abort_transaction(trans, ret);
1052 	return ret;
1053 }
1054 
1055 /*
1056  * Populate the free space tree by walking the extent tree. Operations on the
1057  * extent tree that happen as a result of writes to the free space tree will go
1058  * through the normal add/remove hooks.
1059  */
1060 static int populate_free_space_tree(struct btrfs_trans_handle *trans,
1061 				    struct btrfs_block_group *block_group)
1062 {
1063 	struct btrfs_root *extent_root;
1064 	struct btrfs_path *path, *path2;
1065 	struct btrfs_key key;
1066 	u64 start, end;
1067 	int ret;
1068 
1069 	path = btrfs_alloc_path();
1070 	if (!path)
1071 		return -ENOMEM;
1072 	path->reada = READA_FORWARD;
1073 
1074 	path2 = btrfs_alloc_path();
1075 	if (!path2) {
1076 		btrfs_free_path(path);
1077 		return -ENOMEM;
1078 	}
1079 
1080 	ret = add_new_free_space_info(trans, block_group, path2);
1081 	if (ret)
1082 		goto out;
1083 
1084 	mutex_lock(&block_group->free_space_lock);
1085 
1086 	/*
1087 	 * Iterate through all of the extent and metadata items in this block
1088 	 * group, adding the free space between them and the free space at the
1089 	 * end. Note that EXTENT_ITEM and METADATA_ITEM are less than
1090 	 * BLOCK_GROUP_ITEM, so an extent may precede the block group that it's
1091 	 * contained in.
1092 	 */
1093 	key.objectid = block_group->start;
1094 	key.type = BTRFS_EXTENT_ITEM_KEY;
1095 	key.offset = 0;
1096 
1097 	extent_root = btrfs_extent_root(trans->fs_info, key.objectid);
1098 	ret = btrfs_search_slot_for_read(extent_root, &key, path, 1, 0);
1099 	if (ret < 0)
1100 		goto out_locked;
1101 	ASSERT(ret == 0);
1102 
1103 	start = block_group->start;
1104 	end = block_group->start + block_group->length;
1105 	while (1) {
1106 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1107 
1108 		if (key.type == BTRFS_EXTENT_ITEM_KEY ||
1109 		    key.type == BTRFS_METADATA_ITEM_KEY) {
1110 			if (key.objectid >= end)
1111 				break;
1112 
1113 			if (start < key.objectid) {
1114 				ret = __add_to_free_space_tree(trans,
1115 							       block_group,
1116 							       path2, start,
1117 							       key.objectid -
1118 							       start);
1119 				if (ret)
1120 					goto out_locked;
1121 			}
1122 			start = key.objectid;
1123 			if (key.type == BTRFS_METADATA_ITEM_KEY)
1124 				start += trans->fs_info->nodesize;
1125 			else
1126 				start += key.offset;
1127 		} else if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
1128 			if (key.objectid != block_group->start)
1129 				break;
1130 		}
1131 
1132 		ret = btrfs_next_item(extent_root, path);
1133 		if (ret < 0)
1134 			goto out_locked;
1135 		if (ret)
1136 			break;
1137 	}
1138 	if (start < end) {
1139 		ret = __add_to_free_space_tree(trans, block_group, path2,
1140 					       start, end - start);
1141 		if (ret)
1142 			goto out_locked;
1143 	}
1144 
1145 	ret = 0;
1146 out_locked:
1147 	mutex_unlock(&block_group->free_space_lock);
1148 out:
1149 	btrfs_free_path(path2);
1150 	btrfs_free_path(path);
1151 	return ret;
1152 }
1153 
1154 int btrfs_create_free_space_tree(struct btrfs_fs_info *fs_info)
1155 {
1156 	struct btrfs_trans_handle *trans;
1157 	struct btrfs_root *tree_root = fs_info->tree_root;
1158 	struct btrfs_root *free_space_root;
1159 	struct btrfs_block_group *block_group;
1160 	struct rb_node *node;
1161 	int ret;
1162 
1163 	trans = btrfs_start_transaction(tree_root, 0);
1164 	if (IS_ERR(trans))
1165 		return PTR_ERR(trans);
1166 
1167 	set_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1168 	set_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags);
1169 	free_space_root = btrfs_create_tree(trans,
1170 					    BTRFS_FREE_SPACE_TREE_OBJECTID);
1171 	if (IS_ERR(free_space_root)) {
1172 		ret = PTR_ERR(free_space_root);
1173 		goto abort;
1174 	}
1175 	ret = btrfs_global_root_insert(free_space_root);
1176 	if (ret) {
1177 		btrfs_put_root(free_space_root);
1178 		goto abort;
1179 	}
1180 
1181 	node = rb_first_cached(&fs_info->block_group_cache_tree);
1182 	while (node) {
1183 		block_group = rb_entry(node, struct btrfs_block_group,
1184 				       cache_node);
1185 		ret = populate_free_space_tree(trans, block_group);
1186 		if (ret)
1187 			goto abort;
1188 		node = rb_next(node);
1189 	}
1190 
1191 	btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE);
1192 	btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID);
1193 	clear_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1194 	ret = btrfs_commit_transaction(trans);
1195 
1196 	/*
1197 	 * Now that we've committed the transaction any reading of our commit
1198 	 * root will be safe, so we can cache from the free space tree now.
1199 	 */
1200 	clear_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags);
1201 	return ret;
1202 
1203 abort:
1204 	clear_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1205 	clear_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags);
1206 	btrfs_abort_transaction(trans, ret);
1207 	btrfs_end_transaction(trans);
1208 	return ret;
1209 }
1210 
1211 static int clear_free_space_tree(struct btrfs_trans_handle *trans,
1212 				 struct btrfs_root *root)
1213 {
1214 	struct btrfs_path *path;
1215 	struct btrfs_key key;
1216 	int nr;
1217 	int ret;
1218 
1219 	path = btrfs_alloc_path();
1220 	if (!path)
1221 		return -ENOMEM;
1222 
1223 	key.objectid = 0;
1224 	key.type = 0;
1225 	key.offset = 0;
1226 
1227 	while (1) {
1228 		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1229 		if (ret < 0)
1230 			goto out;
1231 
1232 		nr = btrfs_header_nritems(path->nodes[0]);
1233 		if (!nr)
1234 			break;
1235 
1236 		path->slots[0] = 0;
1237 		ret = btrfs_del_items(trans, root, path, 0, nr);
1238 		if (ret)
1239 			goto out;
1240 
1241 		btrfs_release_path(path);
1242 	}
1243 
1244 	ret = 0;
1245 out:
1246 	btrfs_free_path(path);
1247 	return ret;
1248 }
1249 
1250 int btrfs_clear_free_space_tree(struct btrfs_fs_info *fs_info)
1251 {
1252 	struct btrfs_trans_handle *trans;
1253 	struct btrfs_root *tree_root = fs_info->tree_root;
1254 	struct btrfs_key key = {
1255 		.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID,
1256 		.type = BTRFS_ROOT_ITEM_KEY,
1257 		.offset = 0,
1258 	};
1259 	struct btrfs_root *free_space_root = btrfs_global_root(fs_info, &key);
1260 	int ret;
1261 
1262 	trans = btrfs_start_transaction(tree_root, 0);
1263 	if (IS_ERR(trans))
1264 		return PTR_ERR(trans);
1265 
1266 	btrfs_clear_fs_compat_ro(fs_info, FREE_SPACE_TREE);
1267 	btrfs_clear_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID);
1268 
1269 	ret = clear_free_space_tree(trans, free_space_root);
1270 	if (ret)
1271 		goto abort;
1272 
1273 	ret = btrfs_del_root(trans, &free_space_root->root_key);
1274 	if (ret)
1275 		goto abort;
1276 
1277 	btrfs_global_root_delete(free_space_root);
1278 	list_del(&free_space_root->dirty_list);
1279 
1280 	btrfs_tree_lock(free_space_root->node);
1281 	btrfs_clean_tree_block(free_space_root->node);
1282 	btrfs_tree_unlock(free_space_root->node);
1283 	btrfs_free_tree_block(trans, btrfs_root_id(free_space_root),
1284 			      free_space_root->node, 0, 1);
1285 
1286 	btrfs_put_root(free_space_root);
1287 
1288 	return btrfs_commit_transaction(trans);
1289 
1290 abort:
1291 	btrfs_abort_transaction(trans, ret);
1292 	btrfs_end_transaction(trans);
1293 	return ret;
1294 }
1295 
1296 static int __add_block_group_free_space(struct btrfs_trans_handle *trans,
1297 					struct btrfs_block_group *block_group,
1298 					struct btrfs_path *path)
1299 {
1300 	int ret;
1301 
1302 	block_group->needs_free_space = 0;
1303 
1304 	ret = add_new_free_space_info(trans, block_group, path);
1305 	if (ret)
1306 		return ret;
1307 
1308 	return __add_to_free_space_tree(trans, block_group, path,
1309 					block_group->start,
1310 					block_group->length);
1311 }
1312 
1313 int add_block_group_free_space(struct btrfs_trans_handle *trans,
1314 			       struct btrfs_block_group *block_group)
1315 {
1316 	struct btrfs_fs_info *fs_info = trans->fs_info;
1317 	struct btrfs_path *path = NULL;
1318 	int ret = 0;
1319 
1320 	if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
1321 		return 0;
1322 
1323 	mutex_lock(&block_group->free_space_lock);
1324 	if (!block_group->needs_free_space)
1325 		goto out;
1326 
1327 	path = btrfs_alloc_path();
1328 	if (!path) {
1329 		ret = -ENOMEM;
1330 		goto out;
1331 	}
1332 
1333 	ret = __add_block_group_free_space(trans, block_group, path);
1334 
1335 out:
1336 	btrfs_free_path(path);
1337 	mutex_unlock(&block_group->free_space_lock);
1338 	if (ret)
1339 		btrfs_abort_transaction(trans, ret);
1340 	return ret;
1341 }
1342 
1343 int remove_block_group_free_space(struct btrfs_trans_handle *trans,
1344 				  struct btrfs_block_group *block_group)
1345 {
1346 	struct btrfs_root *root = btrfs_free_space_root(block_group);
1347 	struct btrfs_path *path;
1348 	struct btrfs_key key, found_key;
1349 	struct extent_buffer *leaf;
1350 	u64 start, end;
1351 	int done = 0, nr;
1352 	int ret;
1353 
1354 	if (!btrfs_fs_compat_ro(trans->fs_info, FREE_SPACE_TREE))
1355 		return 0;
1356 
1357 	if (block_group->needs_free_space) {
1358 		/* We never added this block group to the free space tree. */
1359 		return 0;
1360 	}
1361 
1362 	path = btrfs_alloc_path();
1363 	if (!path) {
1364 		ret = -ENOMEM;
1365 		goto out;
1366 	}
1367 
1368 	start = block_group->start;
1369 	end = block_group->start + block_group->length;
1370 
1371 	key.objectid = end - 1;
1372 	key.type = (u8)-1;
1373 	key.offset = (u64)-1;
1374 
1375 	while (!done) {
1376 		ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
1377 		if (ret)
1378 			goto out;
1379 
1380 		leaf = path->nodes[0];
1381 		nr = 0;
1382 		path->slots[0]++;
1383 		while (path->slots[0] > 0) {
1384 			btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
1385 
1386 			if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
1387 				ASSERT(found_key.objectid == block_group->start);
1388 				ASSERT(found_key.offset == block_group->length);
1389 				done = 1;
1390 				nr++;
1391 				path->slots[0]--;
1392 				break;
1393 			} else if (found_key.type == BTRFS_FREE_SPACE_EXTENT_KEY ||
1394 				   found_key.type == BTRFS_FREE_SPACE_BITMAP_KEY) {
1395 				ASSERT(found_key.objectid >= start);
1396 				ASSERT(found_key.objectid < end);
1397 				ASSERT(found_key.objectid + found_key.offset <= end);
1398 				nr++;
1399 				path->slots[0]--;
1400 			} else {
1401 				ASSERT(0);
1402 			}
1403 		}
1404 
1405 		ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
1406 		if (ret)
1407 			goto out;
1408 		btrfs_release_path(path);
1409 	}
1410 
1411 	ret = 0;
1412 out:
1413 	btrfs_free_path(path);
1414 	if (ret)
1415 		btrfs_abort_transaction(trans, ret);
1416 	return ret;
1417 }
1418 
1419 static int load_free_space_bitmaps(struct btrfs_caching_control *caching_ctl,
1420 				   struct btrfs_path *path,
1421 				   u32 expected_extent_count)
1422 {
1423 	struct btrfs_block_group *block_group;
1424 	struct btrfs_fs_info *fs_info;
1425 	struct btrfs_root *root;
1426 	struct btrfs_key key;
1427 	int prev_bit = 0, bit;
1428 	/* Initialize to silence GCC. */
1429 	u64 extent_start = 0;
1430 	u64 end, offset;
1431 	u64 total_found = 0;
1432 	u32 extent_count = 0;
1433 	int ret;
1434 
1435 	block_group = caching_ctl->block_group;
1436 	fs_info = block_group->fs_info;
1437 	root = btrfs_free_space_root(block_group);
1438 
1439 	end = block_group->start + block_group->length;
1440 
1441 	while (1) {
1442 		ret = btrfs_next_item(root, path);
1443 		if (ret < 0)
1444 			goto out;
1445 		if (ret)
1446 			break;
1447 
1448 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1449 
1450 		if (key.type == BTRFS_FREE_SPACE_INFO_KEY)
1451 			break;
1452 
1453 		ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
1454 		ASSERT(key.objectid < end && key.objectid + key.offset <= end);
1455 
1456 		caching_ctl->progress = key.objectid;
1457 
1458 		offset = key.objectid;
1459 		while (offset < key.objectid + key.offset) {
1460 			bit = free_space_test_bit(block_group, path, offset);
1461 			if (prev_bit == 0 && bit == 1) {
1462 				extent_start = offset;
1463 			} else if (prev_bit == 1 && bit == 0) {
1464 				total_found += add_new_free_space(block_group,
1465 								  extent_start,
1466 								  offset);
1467 				if (total_found > CACHING_CTL_WAKE_UP) {
1468 					total_found = 0;
1469 					wake_up(&caching_ctl->wait);
1470 				}
1471 				extent_count++;
1472 			}
1473 			prev_bit = bit;
1474 			offset += fs_info->sectorsize;
1475 		}
1476 	}
1477 	if (prev_bit == 1) {
1478 		total_found += add_new_free_space(block_group, extent_start,
1479 						  end);
1480 		extent_count++;
1481 	}
1482 
1483 	if (extent_count != expected_extent_count) {
1484 		btrfs_err(fs_info,
1485 			  "incorrect extent count for %llu; counted %u, expected %u",
1486 			  block_group->start, extent_count,
1487 			  expected_extent_count);
1488 		ASSERT(0);
1489 		ret = -EIO;
1490 		goto out;
1491 	}
1492 
1493 	caching_ctl->progress = (u64)-1;
1494 
1495 	ret = 0;
1496 out:
1497 	return ret;
1498 }
1499 
1500 static int load_free_space_extents(struct btrfs_caching_control *caching_ctl,
1501 				   struct btrfs_path *path,
1502 				   u32 expected_extent_count)
1503 {
1504 	struct btrfs_block_group *block_group;
1505 	struct btrfs_fs_info *fs_info;
1506 	struct btrfs_root *root;
1507 	struct btrfs_key key;
1508 	u64 end;
1509 	u64 total_found = 0;
1510 	u32 extent_count = 0;
1511 	int ret;
1512 
1513 	block_group = caching_ctl->block_group;
1514 	fs_info = block_group->fs_info;
1515 	root = btrfs_free_space_root(block_group);
1516 
1517 	end = block_group->start + block_group->length;
1518 
1519 	while (1) {
1520 		ret = btrfs_next_item(root, path);
1521 		if (ret < 0)
1522 			goto out;
1523 		if (ret)
1524 			break;
1525 
1526 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1527 
1528 		if (key.type == BTRFS_FREE_SPACE_INFO_KEY)
1529 			break;
1530 
1531 		ASSERT(key.type == BTRFS_FREE_SPACE_EXTENT_KEY);
1532 		ASSERT(key.objectid < end && key.objectid + key.offset <= end);
1533 
1534 		caching_ctl->progress = key.objectid;
1535 
1536 		total_found += add_new_free_space(block_group, key.objectid,
1537 						  key.objectid + key.offset);
1538 		if (total_found > CACHING_CTL_WAKE_UP) {
1539 			total_found = 0;
1540 			wake_up(&caching_ctl->wait);
1541 		}
1542 		extent_count++;
1543 	}
1544 
1545 	if (extent_count != expected_extent_count) {
1546 		btrfs_err(fs_info,
1547 			  "incorrect extent count for %llu; counted %u, expected %u",
1548 			  block_group->start, extent_count,
1549 			  expected_extent_count);
1550 		ASSERT(0);
1551 		ret = -EIO;
1552 		goto out;
1553 	}
1554 
1555 	caching_ctl->progress = (u64)-1;
1556 
1557 	ret = 0;
1558 out:
1559 	return ret;
1560 }
1561 
1562 int load_free_space_tree(struct btrfs_caching_control *caching_ctl)
1563 {
1564 	struct btrfs_block_group *block_group;
1565 	struct btrfs_free_space_info *info;
1566 	struct btrfs_path *path;
1567 	u32 extent_count, flags;
1568 	int ret;
1569 
1570 	block_group = caching_ctl->block_group;
1571 
1572 	path = btrfs_alloc_path();
1573 	if (!path)
1574 		return -ENOMEM;
1575 
1576 	/*
1577 	 * Just like caching_thread() doesn't want to deadlock on the extent
1578 	 * tree, we don't want to deadlock on the free space tree.
1579 	 */
1580 	path->skip_locking = 1;
1581 	path->search_commit_root = 1;
1582 	path->reada = READA_FORWARD;
1583 
1584 	info = search_free_space_info(NULL, block_group, path, 0);
1585 	if (IS_ERR(info)) {
1586 		ret = PTR_ERR(info);
1587 		goto out;
1588 	}
1589 	extent_count = btrfs_free_space_extent_count(path->nodes[0], info);
1590 	flags = btrfs_free_space_flags(path->nodes[0], info);
1591 
1592 	/*
1593 	 * We left path pointing to the free space info item, so now
1594 	 * load_free_space_foo can just iterate through the free space tree from
1595 	 * there.
1596 	 */
1597 	if (flags & BTRFS_FREE_SPACE_USING_BITMAPS)
1598 		ret = load_free_space_bitmaps(caching_ctl, path, extent_count);
1599 	else
1600 		ret = load_free_space_extents(caching_ctl, path, extent_count);
1601 
1602 out:
1603 	btrfs_free_path(path);
1604 	return ret;
1605 }
1606