xref: /openbmc/linux/fs/btrfs/free-space-tree.c (revision 15e3ae36)
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 void set_free_space_tree_thresholds(struct btrfs_block_group *cache)
20 {
21 	u32 bitmap_range;
22 	size_t bitmap_size;
23 	u64 num_bitmaps, total_bitmap_size;
24 
25 	/*
26 	 * We convert to bitmaps when the disk space required for using extents
27 	 * exceeds that required for using bitmaps.
28 	 */
29 	bitmap_range = cache->fs_info->sectorsize * BTRFS_FREE_SPACE_BITMAP_BITS;
30 	num_bitmaps = div_u64(cache->length + bitmap_range - 1, bitmap_range);
31 	bitmap_size = sizeof(struct btrfs_item) + BTRFS_FREE_SPACE_BITMAP_SIZE;
32 	total_bitmap_size = num_bitmaps * bitmap_size;
33 	cache->bitmap_high_thresh = div_u64(total_bitmap_size,
34 					    sizeof(struct btrfs_item));
35 
36 	/*
37 	 * We allow for a small buffer between the high threshold and low
38 	 * threshold to avoid thrashing back and forth between the two formats.
39 	 */
40 	if (cache->bitmap_high_thresh > 100)
41 		cache->bitmap_low_thresh = cache->bitmap_high_thresh - 100;
42 	else
43 		cache->bitmap_low_thresh = 0;
44 }
45 
46 static int add_new_free_space_info(struct btrfs_trans_handle *trans,
47 				   struct btrfs_block_group *block_group,
48 				   struct btrfs_path *path)
49 {
50 	struct btrfs_root *root = trans->fs_info->free_space_root;
51 	struct btrfs_free_space_info *info;
52 	struct btrfs_key key;
53 	struct extent_buffer *leaf;
54 	int ret;
55 
56 	key.objectid = block_group->start;
57 	key.type = BTRFS_FREE_SPACE_INFO_KEY;
58 	key.offset = block_group->length;
59 
60 	ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*info));
61 	if (ret)
62 		goto out;
63 
64 	leaf = path->nodes[0];
65 	info = btrfs_item_ptr(leaf, path->slots[0],
66 			      struct btrfs_free_space_info);
67 	btrfs_set_free_space_extent_count(leaf, info, 0);
68 	btrfs_set_free_space_flags(leaf, info, 0);
69 	btrfs_mark_buffer_dirty(leaf);
70 
71 	ret = 0;
72 out:
73 	btrfs_release_path(path);
74 	return ret;
75 }
76 
77 EXPORT_FOR_TESTS
78 struct btrfs_free_space_info *search_free_space_info(
79 		struct btrfs_trans_handle *trans,
80 		struct btrfs_block_group *block_group,
81 		struct btrfs_path *path, int cow)
82 {
83 	struct btrfs_fs_info *fs_info = block_group->fs_info;
84 	struct btrfs_root *root = fs_info->free_space_root;
85 	struct btrfs_key key;
86 	int ret;
87 
88 	key.objectid = block_group->start;
89 	key.type = BTRFS_FREE_SPACE_INFO_KEY;
90 	key.offset = block_group->length;
91 
92 	ret = btrfs_search_slot(trans, root, &key, path, 0, cow);
93 	if (ret < 0)
94 		return ERR_PTR(ret);
95 	if (ret != 0) {
96 		btrfs_warn(fs_info, "missing free space info for %llu",
97 			   block_group->start);
98 		ASSERT(0);
99 		return ERR_PTR(-ENOENT);
100 	}
101 
102 	return btrfs_item_ptr(path->nodes[0], path->slots[0],
103 			      struct btrfs_free_space_info);
104 }
105 
106 /*
107  * btrfs_search_slot() but we're looking for the greatest key less than the
108  * passed key.
109  */
110 static int btrfs_search_prev_slot(struct btrfs_trans_handle *trans,
111 				  struct btrfs_root *root,
112 				  struct btrfs_key *key, struct btrfs_path *p,
113 				  int ins_len, int cow)
114 {
115 	int ret;
116 
117 	ret = btrfs_search_slot(trans, root, key, p, ins_len, cow);
118 	if (ret < 0)
119 		return ret;
120 
121 	if (ret == 0) {
122 		ASSERT(0);
123 		return -EIO;
124 	}
125 
126 	if (p->slots[0] == 0) {
127 		ASSERT(0);
128 		return -EIO;
129 	}
130 	p->slots[0]--;
131 
132 	return 0;
133 }
134 
135 static inline u32 free_space_bitmap_size(u64 size, u32 sectorsize)
136 {
137 	return DIV_ROUND_UP((u32)div_u64(size, sectorsize), BITS_PER_BYTE);
138 }
139 
140 static unsigned long *alloc_bitmap(u32 bitmap_size)
141 {
142 	unsigned long *ret;
143 	unsigned int nofs_flag;
144 	u32 bitmap_rounded_size = round_up(bitmap_size, sizeof(unsigned long));
145 
146 	/*
147 	 * GFP_NOFS doesn't work with kvmalloc(), but we really can't recurse
148 	 * into the filesystem as the free space bitmap can be modified in the
149 	 * critical section of a transaction commit.
150 	 *
151 	 * TODO: push the memalloc_nofs_{save,restore}() to the caller where we
152 	 * know that recursion is unsafe.
153 	 */
154 	nofs_flag = memalloc_nofs_save();
155 	ret = kvzalloc(bitmap_rounded_size, GFP_KERNEL);
156 	memalloc_nofs_restore(nofs_flag);
157 	return ret;
158 }
159 
160 static void le_bitmap_set(unsigned long *map, unsigned int start, int len)
161 {
162 	u8 *p = ((u8 *)map) + BIT_BYTE(start);
163 	const unsigned int size = start + len;
164 	int bits_to_set = BITS_PER_BYTE - (start % BITS_PER_BYTE);
165 	u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(start);
166 
167 	while (len - bits_to_set >= 0) {
168 		*p |= mask_to_set;
169 		len -= bits_to_set;
170 		bits_to_set = BITS_PER_BYTE;
171 		mask_to_set = ~0;
172 		p++;
173 	}
174 	if (len) {
175 		mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
176 		*p |= mask_to_set;
177 	}
178 }
179 
180 EXPORT_FOR_TESTS
181 int convert_free_space_to_bitmaps(struct btrfs_trans_handle *trans,
182 				  struct btrfs_block_group *block_group,
183 				  struct btrfs_path *path)
184 {
185 	struct btrfs_fs_info *fs_info = trans->fs_info;
186 	struct btrfs_root *root = fs_info->free_space_root;
187 	struct btrfs_free_space_info *info;
188 	struct btrfs_key key, found_key;
189 	struct extent_buffer *leaf;
190 	unsigned long *bitmap;
191 	char *bitmap_cursor;
192 	u64 start, end;
193 	u64 bitmap_range, i;
194 	u32 bitmap_size, flags, expected_extent_count;
195 	u32 extent_count = 0;
196 	int done = 0, nr;
197 	int ret;
198 
199 	bitmap_size = free_space_bitmap_size(block_group->length,
200 					     fs_info->sectorsize);
201 	bitmap = alloc_bitmap(bitmap_size);
202 	if (!bitmap) {
203 		ret = -ENOMEM;
204 		goto out;
205 	}
206 
207 	start = block_group->start;
208 	end = block_group->start + block_group->length;
209 
210 	key.objectid = end - 1;
211 	key.type = (u8)-1;
212 	key.offset = (u64)-1;
213 
214 	while (!done) {
215 		ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
216 		if (ret)
217 			goto out;
218 
219 		leaf = path->nodes[0];
220 		nr = 0;
221 		path->slots[0]++;
222 		while (path->slots[0] > 0) {
223 			btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
224 
225 			if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
226 				ASSERT(found_key.objectid == block_group->start);
227 				ASSERT(found_key.offset == block_group->length);
228 				done = 1;
229 				break;
230 			} else if (found_key.type == BTRFS_FREE_SPACE_EXTENT_KEY) {
231 				u64 first, last;
232 
233 				ASSERT(found_key.objectid >= start);
234 				ASSERT(found_key.objectid < end);
235 				ASSERT(found_key.objectid + found_key.offset <= end);
236 
237 				first = div_u64(found_key.objectid - start,
238 						fs_info->sectorsize);
239 				last = div_u64(found_key.objectid + found_key.offset - start,
240 					       fs_info->sectorsize);
241 				le_bitmap_set(bitmap, first, last - first);
242 
243 				extent_count++;
244 				nr++;
245 				path->slots[0]--;
246 			} else {
247 				ASSERT(0);
248 			}
249 		}
250 
251 		ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
252 		if (ret)
253 			goto out;
254 		btrfs_release_path(path);
255 	}
256 
257 	info = search_free_space_info(trans, block_group, path, 1);
258 	if (IS_ERR(info)) {
259 		ret = PTR_ERR(info);
260 		goto out;
261 	}
262 	leaf = path->nodes[0];
263 	flags = btrfs_free_space_flags(leaf, info);
264 	flags |= BTRFS_FREE_SPACE_USING_BITMAPS;
265 	btrfs_set_free_space_flags(leaf, info, flags);
266 	expected_extent_count = btrfs_free_space_extent_count(leaf, info);
267 	btrfs_mark_buffer_dirty(leaf);
268 	btrfs_release_path(path);
269 
270 	if (extent_count != expected_extent_count) {
271 		btrfs_err(fs_info,
272 			  "incorrect extent count for %llu; counted %u, expected %u",
273 			  block_group->start, extent_count,
274 			  expected_extent_count);
275 		ASSERT(0);
276 		ret = -EIO;
277 		goto out;
278 	}
279 
280 	bitmap_cursor = (char *)bitmap;
281 	bitmap_range = fs_info->sectorsize * BTRFS_FREE_SPACE_BITMAP_BITS;
282 	i = start;
283 	while (i < end) {
284 		unsigned long ptr;
285 		u64 extent_size;
286 		u32 data_size;
287 
288 		extent_size = min(end - i, bitmap_range);
289 		data_size = free_space_bitmap_size(extent_size,
290 						   fs_info->sectorsize);
291 
292 		key.objectid = i;
293 		key.type = BTRFS_FREE_SPACE_BITMAP_KEY;
294 		key.offset = extent_size;
295 
296 		ret = btrfs_insert_empty_item(trans, root, path, &key,
297 					      data_size);
298 		if (ret)
299 			goto out;
300 
301 		leaf = path->nodes[0];
302 		ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
303 		write_extent_buffer(leaf, bitmap_cursor, ptr,
304 				    data_size);
305 		btrfs_mark_buffer_dirty(leaf);
306 		btrfs_release_path(path);
307 
308 		i += extent_size;
309 		bitmap_cursor += data_size;
310 	}
311 
312 	ret = 0;
313 out:
314 	kvfree(bitmap);
315 	if (ret)
316 		btrfs_abort_transaction(trans, ret);
317 	return ret;
318 }
319 
320 EXPORT_FOR_TESTS
321 int convert_free_space_to_extents(struct btrfs_trans_handle *trans,
322 				  struct btrfs_block_group *block_group,
323 				  struct btrfs_path *path)
324 {
325 	struct btrfs_fs_info *fs_info = trans->fs_info;
326 	struct btrfs_root *root = fs_info->free_space_root;
327 	struct btrfs_free_space_info *info;
328 	struct btrfs_key key, found_key;
329 	struct extent_buffer *leaf;
330 	unsigned long *bitmap;
331 	u64 start, end;
332 	u32 bitmap_size, flags, expected_extent_count;
333 	unsigned long nrbits, start_bit, end_bit;
334 	u32 extent_count = 0;
335 	int done = 0, nr;
336 	int ret;
337 
338 	bitmap_size = free_space_bitmap_size(block_group->length,
339 					     fs_info->sectorsize);
340 	bitmap = alloc_bitmap(bitmap_size);
341 	if (!bitmap) {
342 		ret = -ENOMEM;
343 		goto out;
344 	}
345 
346 	start = block_group->start;
347 	end = block_group->start + block_group->length;
348 
349 	key.objectid = end - 1;
350 	key.type = (u8)-1;
351 	key.offset = (u64)-1;
352 
353 	while (!done) {
354 		ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
355 		if (ret)
356 			goto out;
357 
358 		leaf = path->nodes[0];
359 		nr = 0;
360 		path->slots[0]++;
361 		while (path->slots[0] > 0) {
362 			btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
363 
364 			if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
365 				ASSERT(found_key.objectid == block_group->start);
366 				ASSERT(found_key.offset == block_group->length);
367 				done = 1;
368 				break;
369 			} else if (found_key.type == BTRFS_FREE_SPACE_BITMAP_KEY) {
370 				unsigned long ptr;
371 				char *bitmap_cursor;
372 				u32 bitmap_pos, data_size;
373 
374 				ASSERT(found_key.objectid >= start);
375 				ASSERT(found_key.objectid < end);
376 				ASSERT(found_key.objectid + found_key.offset <= end);
377 
378 				bitmap_pos = div_u64(found_key.objectid - start,
379 						     fs_info->sectorsize *
380 						     BITS_PER_BYTE);
381 				bitmap_cursor = ((char *)bitmap) + bitmap_pos;
382 				data_size = free_space_bitmap_size(found_key.offset,
383 								   fs_info->sectorsize);
384 
385 				ptr = btrfs_item_ptr_offset(leaf, path->slots[0] - 1);
386 				read_extent_buffer(leaf, bitmap_cursor, ptr,
387 						   data_size);
388 
389 				nr++;
390 				path->slots[0]--;
391 			} else {
392 				ASSERT(0);
393 			}
394 		}
395 
396 		ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
397 		if (ret)
398 			goto out;
399 		btrfs_release_path(path);
400 	}
401 
402 	info = search_free_space_info(trans, block_group, path, 1);
403 	if (IS_ERR(info)) {
404 		ret = PTR_ERR(info);
405 		goto out;
406 	}
407 	leaf = path->nodes[0];
408 	flags = btrfs_free_space_flags(leaf, info);
409 	flags &= ~BTRFS_FREE_SPACE_USING_BITMAPS;
410 	btrfs_set_free_space_flags(leaf, info, flags);
411 	expected_extent_count = btrfs_free_space_extent_count(leaf, info);
412 	btrfs_mark_buffer_dirty(leaf);
413 	btrfs_release_path(path);
414 
415 	nrbits = div_u64(block_group->length, block_group->fs_info->sectorsize);
416 	start_bit = find_next_bit_le(bitmap, nrbits, 0);
417 
418 	while (start_bit < nrbits) {
419 		end_bit = find_next_zero_bit_le(bitmap, nrbits, start_bit);
420 		ASSERT(start_bit < end_bit);
421 
422 		key.objectid = start + start_bit * block_group->fs_info->sectorsize;
423 		key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
424 		key.offset = (end_bit - start_bit) * block_group->fs_info->sectorsize;
425 
426 		ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
427 		if (ret)
428 			goto out;
429 		btrfs_release_path(path);
430 
431 		extent_count++;
432 
433 		start_bit = find_next_bit_le(bitmap, nrbits, end_bit);
434 	}
435 
436 	if (extent_count != expected_extent_count) {
437 		btrfs_err(fs_info,
438 			  "incorrect extent count for %llu; counted %u, expected %u",
439 			  block_group->start, extent_count,
440 			  expected_extent_count);
441 		ASSERT(0);
442 		ret = -EIO;
443 		goto out;
444 	}
445 
446 	ret = 0;
447 out:
448 	kvfree(bitmap);
449 	if (ret)
450 		btrfs_abort_transaction(trans, ret);
451 	return ret;
452 }
453 
454 static int update_free_space_extent_count(struct btrfs_trans_handle *trans,
455 					  struct btrfs_block_group *block_group,
456 					  struct btrfs_path *path,
457 					  int new_extents)
458 {
459 	struct btrfs_free_space_info *info;
460 	u32 flags;
461 	u32 extent_count;
462 	int ret = 0;
463 
464 	if (new_extents == 0)
465 		return 0;
466 
467 	info = search_free_space_info(trans, block_group, path, 1);
468 	if (IS_ERR(info)) {
469 		ret = PTR_ERR(info);
470 		goto out;
471 	}
472 	flags = btrfs_free_space_flags(path->nodes[0], info);
473 	extent_count = btrfs_free_space_extent_count(path->nodes[0], info);
474 
475 	extent_count += new_extents;
476 	btrfs_set_free_space_extent_count(path->nodes[0], info, extent_count);
477 	btrfs_mark_buffer_dirty(path->nodes[0]);
478 	btrfs_release_path(path);
479 
480 	if (!(flags & BTRFS_FREE_SPACE_USING_BITMAPS) &&
481 	    extent_count > block_group->bitmap_high_thresh) {
482 		ret = convert_free_space_to_bitmaps(trans, block_group, path);
483 	} else if ((flags & BTRFS_FREE_SPACE_USING_BITMAPS) &&
484 		   extent_count < block_group->bitmap_low_thresh) {
485 		ret = convert_free_space_to_extents(trans, block_group, path);
486 	}
487 
488 out:
489 	return ret;
490 }
491 
492 EXPORT_FOR_TESTS
493 int free_space_test_bit(struct btrfs_block_group *block_group,
494 			struct btrfs_path *path, u64 offset)
495 {
496 	struct extent_buffer *leaf;
497 	struct btrfs_key key;
498 	u64 found_start, found_end;
499 	unsigned long ptr, i;
500 
501 	leaf = path->nodes[0];
502 	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
503 	ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
504 
505 	found_start = key.objectid;
506 	found_end = key.objectid + key.offset;
507 	ASSERT(offset >= found_start && offset < found_end);
508 
509 	ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
510 	i = div_u64(offset - found_start,
511 		    block_group->fs_info->sectorsize);
512 	return !!extent_buffer_test_bit(leaf, ptr, i);
513 }
514 
515 static void free_space_set_bits(struct btrfs_block_group *block_group,
516 				struct btrfs_path *path, u64 *start, u64 *size,
517 				int bit)
518 {
519 	struct btrfs_fs_info *fs_info = block_group->fs_info;
520 	struct extent_buffer *leaf;
521 	struct btrfs_key key;
522 	u64 end = *start + *size;
523 	u64 found_start, found_end;
524 	unsigned long ptr, first, last;
525 
526 	leaf = path->nodes[0];
527 	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
528 	ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
529 
530 	found_start = key.objectid;
531 	found_end = key.objectid + key.offset;
532 	ASSERT(*start >= found_start && *start < found_end);
533 	ASSERT(end > found_start);
534 
535 	if (end > found_end)
536 		end = found_end;
537 
538 	ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
539 	first = div_u64(*start - found_start, fs_info->sectorsize);
540 	last = div_u64(end - found_start, fs_info->sectorsize);
541 	if (bit)
542 		extent_buffer_bitmap_set(leaf, ptr, first, last - first);
543 	else
544 		extent_buffer_bitmap_clear(leaf, ptr, first, last - first);
545 	btrfs_mark_buffer_dirty(leaf);
546 
547 	*size -= end - *start;
548 	*start = end;
549 }
550 
551 /*
552  * We can't use btrfs_next_item() in modify_free_space_bitmap() because
553  * btrfs_next_leaf() doesn't get the path for writing. We can forgo the fancy
554  * tree walking in btrfs_next_leaf() anyways because we know exactly what we're
555  * looking for.
556  */
557 static int free_space_next_bitmap(struct btrfs_trans_handle *trans,
558 				  struct btrfs_root *root, struct btrfs_path *p)
559 {
560 	struct btrfs_key key;
561 
562 	if (p->slots[0] + 1 < btrfs_header_nritems(p->nodes[0])) {
563 		p->slots[0]++;
564 		return 0;
565 	}
566 
567 	btrfs_item_key_to_cpu(p->nodes[0], &key, p->slots[0]);
568 	btrfs_release_path(p);
569 
570 	key.objectid += key.offset;
571 	key.type = (u8)-1;
572 	key.offset = (u64)-1;
573 
574 	return btrfs_search_prev_slot(trans, root, &key, p, 0, 1);
575 }
576 
577 /*
578  * If remove is 1, then we are removing free space, thus clearing bits in the
579  * bitmap. If remove is 0, then we are adding free space, thus setting bits in
580  * the bitmap.
581  */
582 static int modify_free_space_bitmap(struct btrfs_trans_handle *trans,
583 				    struct btrfs_block_group *block_group,
584 				    struct btrfs_path *path,
585 				    u64 start, u64 size, int remove)
586 {
587 	struct btrfs_root *root = block_group->fs_info->free_space_root;
588 	struct btrfs_key key;
589 	u64 end = start + size;
590 	u64 cur_start, cur_size;
591 	int prev_bit, next_bit;
592 	int new_extents;
593 	int ret;
594 
595 	/*
596 	 * Read the bit for the block immediately before the extent of space if
597 	 * that block is within the block group.
598 	 */
599 	if (start > block_group->start) {
600 		u64 prev_block = start - block_group->fs_info->sectorsize;
601 
602 		key.objectid = prev_block;
603 		key.type = (u8)-1;
604 		key.offset = (u64)-1;
605 
606 		ret = btrfs_search_prev_slot(trans, root, &key, path, 0, 1);
607 		if (ret)
608 			goto out;
609 
610 		prev_bit = free_space_test_bit(block_group, path, prev_block);
611 
612 		/* The previous block may have been in the previous bitmap. */
613 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
614 		if (start >= key.objectid + key.offset) {
615 			ret = free_space_next_bitmap(trans, root, path);
616 			if (ret)
617 				goto out;
618 		}
619 	} else {
620 		key.objectid = start;
621 		key.type = (u8)-1;
622 		key.offset = (u64)-1;
623 
624 		ret = btrfs_search_prev_slot(trans, root, &key, path, 0, 1);
625 		if (ret)
626 			goto out;
627 
628 		prev_bit = -1;
629 	}
630 
631 	/*
632 	 * Iterate over all of the bitmaps overlapped by the extent of space,
633 	 * clearing/setting bits as required.
634 	 */
635 	cur_start = start;
636 	cur_size = size;
637 	while (1) {
638 		free_space_set_bits(block_group, path, &cur_start, &cur_size,
639 				    !remove);
640 		if (cur_size == 0)
641 			break;
642 		ret = free_space_next_bitmap(trans, root, path);
643 		if (ret)
644 			goto out;
645 	}
646 
647 	/*
648 	 * Read the bit for the block immediately after the extent of space if
649 	 * that block is within the block group.
650 	 */
651 	if (end < block_group->start + block_group->length) {
652 		/* The next block may be in the next bitmap. */
653 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
654 		if (end >= key.objectid + key.offset) {
655 			ret = free_space_next_bitmap(trans, root, path);
656 			if (ret)
657 				goto out;
658 		}
659 
660 		next_bit = free_space_test_bit(block_group, path, end);
661 	} else {
662 		next_bit = -1;
663 	}
664 
665 	if (remove) {
666 		new_extents = -1;
667 		if (prev_bit == 1) {
668 			/* Leftover on the left. */
669 			new_extents++;
670 		}
671 		if (next_bit == 1) {
672 			/* Leftover on the right. */
673 			new_extents++;
674 		}
675 	} else {
676 		new_extents = 1;
677 		if (prev_bit == 1) {
678 			/* Merging with neighbor on the left. */
679 			new_extents--;
680 		}
681 		if (next_bit == 1) {
682 			/* Merging with neighbor on the right. */
683 			new_extents--;
684 		}
685 	}
686 
687 	btrfs_release_path(path);
688 	ret = update_free_space_extent_count(trans, block_group, path,
689 					     new_extents);
690 
691 out:
692 	return ret;
693 }
694 
695 static int remove_free_space_extent(struct btrfs_trans_handle *trans,
696 				    struct btrfs_block_group *block_group,
697 				    struct btrfs_path *path,
698 				    u64 start, u64 size)
699 {
700 	struct btrfs_root *root = trans->fs_info->free_space_root;
701 	struct btrfs_key key;
702 	u64 found_start, found_end;
703 	u64 end = start + size;
704 	int new_extents = -1;
705 	int ret;
706 
707 	key.objectid = start;
708 	key.type = (u8)-1;
709 	key.offset = (u64)-1;
710 
711 	ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
712 	if (ret)
713 		goto out;
714 
715 	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
716 
717 	ASSERT(key.type == BTRFS_FREE_SPACE_EXTENT_KEY);
718 
719 	found_start = key.objectid;
720 	found_end = key.objectid + key.offset;
721 	ASSERT(start >= found_start && end <= found_end);
722 
723 	/*
724 	 * Okay, now that we've found the free space extent which contains the
725 	 * free space that we are removing, there are four cases:
726 	 *
727 	 * 1. We're using the whole extent: delete the key we found and
728 	 * decrement the free space extent count.
729 	 * 2. We are using part of the extent starting at the beginning: delete
730 	 * the key we found and insert a new key representing the leftover at
731 	 * the end. There is no net change in the number of extents.
732 	 * 3. We are using part of the extent ending at the end: delete the key
733 	 * we found and insert a new key representing the leftover at the
734 	 * beginning. There is no net change in the number of extents.
735 	 * 4. We are using part of the extent in the middle: delete the key we
736 	 * found and insert two new keys representing the leftovers on each
737 	 * side. Where we used to have one extent, we now have two, so increment
738 	 * the extent count. We may need to convert the block group to bitmaps
739 	 * as a result.
740 	 */
741 
742 	/* Delete the existing key (cases 1-4). */
743 	ret = btrfs_del_item(trans, root, path);
744 	if (ret)
745 		goto out;
746 
747 	/* Add a key for leftovers at the beginning (cases 3 and 4). */
748 	if (start > found_start) {
749 		key.objectid = found_start;
750 		key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
751 		key.offset = start - found_start;
752 
753 		btrfs_release_path(path);
754 		ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
755 		if (ret)
756 			goto out;
757 		new_extents++;
758 	}
759 
760 	/* Add a key for leftovers at the end (cases 2 and 4). */
761 	if (end < found_end) {
762 		key.objectid = end;
763 		key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
764 		key.offset = found_end - end;
765 
766 		btrfs_release_path(path);
767 		ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
768 		if (ret)
769 			goto out;
770 		new_extents++;
771 	}
772 
773 	btrfs_release_path(path);
774 	ret = update_free_space_extent_count(trans, block_group, path,
775 					     new_extents);
776 
777 out:
778 	return ret;
779 }
780 
781 EXPORT_FOR_TESTS
782 int __remove_from_free_space_tree(struct btrfs_trans_handle *trans,
783 				  struct btrfs_block_group *block_group,
784 				  struct btrfs_path *path, u64 start, u64 size)
785 {
786 	struct btrfs_free_space_info *info;
787 	u32 flags;
788 	int ret;
789 
790 	if (block_group->needs_free_space) {
791 		ret = __add_block_group_free_space(trans, block_group, path);
792 		if (ret)
793 			return ret;
794 	}
795 
796 	info = search_free_space_info(NULL, block_group, path, 0);
797 	if (IS_ERR(info))
798 		return PTR_ERR(info);
799 	flags = btrfs_free_space_flags(path->nodes[0], info);
800 	btrfs_release_path(path);
801 
802 	if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) {
803 		return modify_free_space_bitmap(trans, block_group, path,
804 						start, size, 1);
805 	} else {
806 		return remove_free_space_extent(trans, block_group, path,
807 						start, size);
808 	}
809 }
810 
811 int remove_from_free_space_tree(struct btrfs_trans_handle *trans,
812 				u64 start, u64 size)
813 {
814 	struct btrfs_block_group *block_group;
815 	struct btrfs_path *path;
816 	int ret;
817 
818 	if (!btrfs_fs_compat_ro(trans->fs_info, FREE_SPACE_TREE))
819 		return 0;
820 
821 	path = btrfs_alloc_path();
822 	if (!path) {
823 		ret = -ENOMEM;
824 		goto out;
825 	}
826 
827 	block_group = btrfs_lookup_block_group(trans->fs_info, start);
828 	if (!block_group) {
829 		ASSERT(0);
830 		ret = -ENOENT;
831 		goto out;
832 	}
833 
834 	mutex_lock(&block_group->free_space_lock);
835 	ret = __remove_from_free_space_tree(trans, block_group, path, start,
836 					    size);
837 	mutex_unlock(&block_group->free_space_lock);
838 
839 	btrfs_put_block_group(block_group);
840 out:
841 	btrfs_free_path(path);
842 	if (ret)
843 		btrfs_abort_transaction(trans, ret);
844 	return ret;
845 }
846 
847 static int add_free_space_extent(struct btrfs_trans_handle *trans,
848 				 struct btrfs_block_group *block_group,
849 				 struct btrfs_path *path,
850 				 u64 start, u64 size)
851 {
852 	struct btrfs_root *root = trans->fs_info->free_space_root;
853 	struct btrfs_key key, new_key;
854 	u64 found_start, found_end;
855 	u64 end = start + size;
856 	int new_extents = 1;
857 	int ret;
858 
859 	/*
860 	 * We are adding a new extent of free space, but we need to merge
861 	 * extents. There are four cases here:
862 	 *
863 	 * 1. The new extent does not have any immediate neighbors to merge
864 	 * with: add the new key and increment the free space extent count. We
865 	 * may need to convert the block group to bitmaps as a result.
866 	 * 2. The new extent has an immediate neighbor before it: remove the
867 	 * previous key and insert a new key combining both of them. There is no
868 	 * net change in the number of extents.
869 	 * 3. The new extent has an immediate neighbor after it: remove the next
870 	 * key and insert a new key combining both of them. There is no net
871 	 * change in the number of extents.
872 	 * 4. The new extent has immediate neighbors on both sides: remove both
873 	 * of the keys and insert a new key combining all of them. Where we used
874 	 * to have two extents, we now have one, so decrement the extent count.
875 	 */
876 
877 	new_key.objectid = start;
878 	new_key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
879 	new_key.offset = size;
880 
881 	/* Search for a neighbor on the left. */
882 	if (start == block_group->start)
883 		goto right;
884 	key.objectid = start - 1;
885 	key.type = (u8)-1;
886 	key.offset = (u64)-1;
887 
888 	ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
889 	if (ret)
890 		goto out;
891 
892 	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
893 
894 	if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) {
895 		ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY);
896 		btrfs_release_path(path);
897 		goto right;
898 	}
899 
900 	found_start = key.objectid;
901 	found_end = key.objectid + key.offset;
902 	ASSERT(found_start >= block_group->start &&
903 	       found_end > block_group->start);
904 	ASSERT(found_start < start && found_end <= start);
905 
906 	/*
907 	 * Delete the neighbor on the left and absorb it into the new key (cases
908 	 * 2 and 4).
909 	 */
910 	if (found_end == start) {
911 		ret = btrfs_del_item(trans, root, path);
912 		if (ret)
913 			goto out;
914 		new_key.objectid = found_start;
915 		new_key.offset += key.offset;
916 		new_extents--;
917 	}
918 	btrfs_release_path(path);
919 
920 right:
921 	/* Search for a neighbor on the right. */
922 	if (end == block_group->start + block_group->length)
923 		goto insert;
924 	key.objectid = end;
925 	key.type = (u8)-1;
926 	key.offset = (u64)-1;
927 
928 	ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
929 	if (ret)
930 		goto out;
931 
932 	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
933 
934 	if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) {
935 		ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY);
936 		btrfs_release_path(path);
937 		goto insert;
938 	}
939 
940 	found_start = key.objectid;
941 	found_end = key.objectid + key.offset;
942 	ASSERT(found_start >= block_group->start &&
943 	       found_end > block_group->start);
944 	ASSERT((found_start < start && found_end <= start) ||
945 	       (found_start >= end && found_end > end));
946 
947 	/*
948 	 * Delete the neighbor on the right and absorb it into the new key
949 	 * (cases 3 and 4).
950 	 */
951 	if (found_start == end) {
952 		ret = btrfs_del_item(trans, root, path);
953 		if (ret)
954 			goto out;
955 		new_key.offset += key.offset;
956 		new_extents--;
957 	}
958 	btrfs_release_path(path);
959 
960 insert:
961 	/* Insert the new key (cases 1-4). */
962 	ret = btrfs_insert_empty_item(trans, root, path, &new_key, 0);
963 	if (ret)
964 		goto out;
965 
966 	btrfs_release_path(path);
967 	ret = update_free_space_extent_count(trans, block_group, path,
968 					     new_extents);
969 
970 out:
971 	return ret;
972 }
973 
974 EXPORT_FOR_TESTS
975 int __add_to_free_space_tree(struct btrfs_trans_handle *trans,
976 			     struct btrfs_block_group *block_group,
977 			     struct btrfs_path *path, u64 start, u64 size)
978 {
979 	struct btrfs_free_space_info *info;
980 	u32 flags;
981 	int ret;
982 
983 	if (block_group->needs_free_space) {
984 		ret = __add_block_group_free_space(trans, block_group, path);
985 		if (ret)
986 			return ret;
987 	}
988 
989 	info = search_free_space_info(NULL, block_group, path, 0);
990 	if (IS_ERR(info))
991 		return PTR_ERR(info);
992 	flags = btrfs_free_space_flags(path->nodes[0], info);
993 	btrfs_release_path(path);
994 
995 	if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) {
996 		return modify_free_space_bitmap(trans, block_group, path,
997 						start, size, 0);
998 	} else {
999 		return add_free_space_extent(trans, block_group, path, start,
1000 					     size);
1001 	}
1002 }
1003 
1004 int add_to_free_space_tree(struct btrfs_trans_handle *trans,
1005 			   u64 start, u64 size)
1006 {
1007 	struct btrfs_block_group *block_group;
1008 	struct btrfs_path *path;
1009 	int ret;
1010 
1011 	if (!btrfs_fs_compat_ro(trans->fs_info, FREE_SPACE_TREE))
1012 		return 0;
1013 
1014 	path = btrfs_alloc_path();
1015 	if (!path) {
1016 		ret = -ENOMEM;
1017 		goto out;
1018 	}
1019 
1020 	block_group = btrfs_lookup_block_group(trans->fs_info, start);
1021 	if (!block_group) {
1022 		ASSERT(0);
1023 		ret = -ENOENT;
1024 		goto out;
1025 	}
1026 
1027 	mutex_lock(&block_group->free_space_lock);
1028 	ret = __add_to_free_space_tree(trans, block_group, path, start, size);
1029 	mutex_unlock(&block_group->free_space_lock);
1030 
1031 	btrfs_put_block_group(block_group);
1032 out:
1033 	btrfs_free_path(path);
1034 	if (ret)
1035 		btrfs_abort_transaction(trans, ret);
1036 	return ret;
1037 }
1038 
1039 /*
1040  * Populate the free space tree by walking the extent tree. Operations on the
1041  * extent tree that happen as a result of writes to the free space tree will go
1042  * through the normal add/remove hooks.
1043  */
1044 static int populate_free_space_tree(struct btrfs_trans_handle *trans,
1045 				    struct btrfs_block_group *block_group)
1046 {
1047 	struct btrfs_root *extent_root = trans->fs_info->extent_root;
1048 	struct btrfs_path *path, *path2;
1049 	struct btrfs_key key;
1050 	u64 start, end;
1051 	int ret;
1052 
1053 	path = btrfs_alloc_path();
1054 	if (!path)
1055 		return -ENOMEM;
1056 	path->reada = READA_FORWARD;
1057 
1058 	path2 = btrfs_alloc_path();
1059 	if (!path2) {
1060 		btrfs_free_path(path);
1061 		return -ENOMEM;
1062 	}
1063 
1064 	ret = add_new_free_space_info(trans, block_group, path2);
1065 	if (ret)
1066 		goto out;
1067 
1068 	mutex_lock(&block_group->free_space_lock);
1069 
1070 	/*
1071 	 * Iterate through all of the extent and metadata items in this block
1072 	 * group, adding the free space between them and the free space at the
1073 	 * end. Note that EXTENT_ITEM and METADATA_ITEM are less than
1074 	 * BLOCK_GROUP_ITEM, so an extent may precede the block group that it's
1075 	 * contained in.
1076 	 */
1077 	key.objectid = block_group->start;
1078 	key.type = BTRFS_EXTENT_ITEM_KEY;
1079 	key.offset = 0;
1080 
1081 	ret = btrfs_search_slot_for_read(extent_root, &key, path, 1, 0);
1082 	if (ret < 0)
1083 		goto out_locked;
1084 	ASSERT(ret == 0);
1085 
1086 	start = block_group->start;
1087 	end = block_group->start + block_group->length;
1088 	while (1) {
1089 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1090 
1091 		if (key.type == BTRFS_EXTENT_ITEM_KEY ||
1092 		    key.type == BTRFS_METADATA_ITEM_KEY) {
1093 			if (key.objectid >= end)
1094 				break;
1095 
1096 			if (start < key.objectid) {
1097 				ret = __add_to_free_space_tree(trans,
1098 							       block_group,
1099 							       path2, start,
1100 							       key.objectid -
1101 							       start);
1102 				if (ret)
1103 					goto out_locked;
1104 			}
1105 			start = key.objectid;
1106 			if (key.type == BTRFS_METADATA_ITEM_KEY)
1107 				start += trans->fs_info->nodesize;
1108 			else
1109 				start += key.offset;
1110 		} else if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
1111 			if (key.objectid != block_group->start)
1112 				break;
1113 		}
1114 
1115 		ret = btrfs_next_item(extent_root, path);
1116 		if (ret < 0)
1117 			goto out_locked;
1118 		if (ret)
1119 			break;
1120 	}
1121 	if (start < end) {
1122 		ret = __add_to_free_space_tree(trans, block_group, path2,
1123 					       start, end - start);
1124 		if (ret)
1125 			goto out_locked;
1126 	}
1127 
1128 	ret = 0;
1129 out_locked:
1130 	mutex_unlock(&block_group->free_space_lock);
1131 out:
1132 	btrfs_free_path(path2);
1133 	btrfs_free_path(path);
1134 	return ret;
1135 }
1136 
1137 int btrfs_create_free_space_tree(struct btrfs_fs_info *fs_info)
1138 {
1139 	struct btrfs_trans_handle *trans;
1140 	struct btrfs_root *tree_root = fs_info->tree_root;
1141 	struct btrfs_root *free_space_root;
1142 	struct btrfs_block_group *block_group;
1143 	struct rb_node *node;
1144 	int ret;
1145 
1146 	trans = btrfs_start_transaction(tree_root, 0);
1147 	if (IS_ERR(trans))
1148 		return PTR_ERR(trans);
1149 
1150 	set_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1151 	free_space_root = btrfs_create_tree(trans,
1152 					    BTRFS_FREE_SPACE_TREE_OBJECTID);
1153 	if (IS_ERR(free_space_root)) {
1154 		ret = PTR_ERR(free_space_root);
1155 		goto abort;
1156 	}
1157 	fs_info->free_space_root = free_space_root;
1158 
1159 	node = rb_first(&fs_info->block_group_cache_tree);
1160 	while (node) {
1161 		block_group = rb_entry(node, struct btrfs_block_group,
1162 				       cache_node);
1163 		ret = populate_free_space_tree(trans, block_group);
1164 		if (ret)
1165 			goto abort;
1166 		node = rb_next(node);
1167 	}
1168 
1169 	btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE);
1170 	btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID);
1171 	clear_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1172 
1173 	return btrfs_commit_transaction(trans);
1174 
1175 abort:
1176 	clear_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1177 	btrfs_abort_transaction(trans, ret);
1178 	btrfs_end_transaction(trans);
1179 	return ret;
1180 }
1181 
1182 static int clear_free_space_tree(struct btrfs_trans_handle *trans,
1183 				 struct btrfs_root *root)
1184 {
1185 	struct btrfs_path *path;
1186 	struct btrfs_key key;
1187 	int nr;
1188 	int ret;
1189 
1190 	path = btrfs_alloc_path();
1191 	if (!path)
1192 		return -ENOMEM;
1193 
1194 	path->leave_spinning = 1;
1195 
1196 	key.objectid = 0;
1197 	key.type = 0;
1198 	key.offset = 0;
1199 
1200 	while (1) {
1201 		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1202 		if (ret < 0)
1203 			goto out;
1204 
1205 		nr = btrfs_header_nritems(path->nodes[0]);
1206 		if (!nr)
1207 			break;
1208 
1209 		path->slots[0] = 0;
1210 		ret = btrfs_del_items(trans, root, path, 0, nr);
1211 		if (ret)
1212 			goto out;
1213 
1214 		btrfs_release_path(path);
1215 	}
1216 
1217 	ret = 0;
1218 out:
1219 	btrfs_free_path(path);
1220 	return ret;
1221 }
1222 
1223 int btrfs_clear_free_space_tree(struct btrfs_fs_info *fs_info)
1224 {
1225 	struct btrfs_trans_handle *trans;
1226 	struct btrfs_root *tree_root = fs_info->tree_root;
1227 	struct btrfs_root *free_space_root = fs_info->free_space_root;
1228 	int ret;
1229 
1230 	trans = btrfs_start_transaction(tree_root, 0);
1231 	if (IS_ERR(trans))
1232 		return PTR_ERR(trans);
1233 
1234 	btrfs_clear_fs_compat_ro(fs_info, FREE_SPACE_TREE);
1235 	btrfs_clear_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID);
1236 	fs_info->free_space_root = NULL;
1237 
1238 	ret = clear_free_space_tree(trans, free_space_root);
1239 	if (ret)
1240 		goto abort;
1241 
1242 	ret = btrfs_del_root(trans, &free_space_root->root_key);
1243 	if (ret)
1244 		goto abort;
1245 
1246 	list_del(&free_space_root->dirty_list);
1247 
1248 	btrfs_tree_lock(free_space_root->node);
1249 	btrfs_clean_tree_block(free_space_root->node);
1250 	btrfs_tree_unlock(free_space_root->node);
1251 	btrfs_free_tree_block(trans, free_space_root, free_space_root->node,
1252 			      0, 1);
1253 
1254 	btrfs_put_root(free_space_root);
1255 
1256 	return btrfs_commit_transaction(trans);
1257 
1258 abort:
1259 	btrfs_abort_transaction(trans, ret);
1260 	btrfs_end_transaction(trans);
1261 	return ret;
1262 }
1263 
1264 static int __add_block_group_free_space(struct btrfs_trans_handle *trans,
1265 					struct btrfs_block_group *block_group,
1266 					struct btrfs_path *path)
1267 {
1268 	int ret;
1269 
1270 	block_group->needs_free_space = 0;
1271 
1272 	ret = add_new_free_space_info(trans, block_group, path);
1273 	if (ret)
1274 		return ret;
1275 
1276 	return __add_to_free_space_tree(trans, block_group, path,
1277 					block_group->start,
1278 					block_group->length);
1279 }
1280 
1281 int add_block_group_free_space(struct btrfs_trans_handle *trans,
1282 			       struct btrfs_block_group *block_group)
1283 {
1284 	struct btrfs_fs_info *fs_info = trans->fs_info;
1285 	struct btrfs_path *path = NULL;
1286 	int ret = 0;
1287 
1288 	if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
1289 		return 0;
1290 
1291 	mutex_lock(&block_group->free_space_lock);
1292 	if (!block_group->needs_free_space)
1293 		goto out;
1294 
1295 	path = btrfs_alloc_path();
1296 	if (!path) {
1297 		ret = -ENOMEM;
1298 		goto out;
1299 	}
1300 
1301 	ret = __add_block_group_free_space(trans, block_group, path);
1302 
1303 out:
1304 	btrfs_free_path(path);
1305 	mutex_unlock(&block_group->free_space_lock);
1306 	if (ret)
1307 		btrfs_abort_transaction(trans, ret);
1308 	return ret;
1309 }
1310 
1311 int remove_block_group_free_space(struct btrfs_trans_handle *trans,
1312 				  struct btrfs_block_group *block_group)
1313 {
1314 	struct btrfs_root *root = trans->fs_info->free_space_root;
1315 	struct btrfs_path *path;
1316 	struct btrfs_key key, found_key;
1317 	struct extent_buffer *leaf;
1318 	u64 start, end;
1319 	int done = 0, nr;
1320 	int ret;
1321 
1322 	if (!btrfs_fs_compat_ro(trans->fs_info, FREE_SPACE_TREE))
1323 		return 0;
1324 
1325 	if (block_group->needs_free_space) {
1326 		/* We never added this block group to the free space tree. */
1327 		return 0;
1328 	}
1329 
1330 	path = btrfs_alloc_path();
1331 	if (!path) {
1332 		ret = -ENOMEM;
1333 		goto out;
1334 	}
1335 
1336 	start = block_group->start;
1337 	end = block_group->start + block_group->length;
1338 
1339 	key.objectid = end - 1;
1340 	key.type = (u8)-1;
1341 	key.offset = (u64)-1;
1342 
1343 	while (!done) {
1344 		ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
1345 		if (ret)
1346 			goto out;
1347 
1348 		leaf = path->nodes[0];
1349 		nr = 0;
1350 		path->slots[0]++;
1351 		while (path->slots[0] > 0) {
1352 			btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
1353 
1354 			if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
1355 				ASSERT(found_key.objectid == block_group->start);
1356 				ASSERT(found_key.offset == block_group->length);
1357 				done = 1;
1358 				nr++;
1359 				path->slots[0]--;
1360 				break;
1361 			} else if (found_key.type == BTRFS_FREE_SPACE_EXTENT_KEY ||
1362 				   found_key.type == BTRFS_FREE_SPACE_BITMAP_KEY) {
1363 				ASSERT(found_key.objectid >= start);
1364 				ASSERT(found_key.objectid < end);
1365 				ASSERT(found_key.objectid + found_key.offset <= end);
1366 				nr++;
1367 				path->slots[0]--;
1368 			} else {
1369 				ASSERT(0);
1370 			}
1371 		}
1372 
1373 		ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
1374 		if (ret)
1375 			goto out;
1376 		btrfs_release_path(path);
1377 	}
1378 
1379 	ret = 0;
1380 out:
1381 	btrfs_free_path(path);
1382 	if (ret)
1383 		btrfs_abort_transaction(trans, ret);
1384 	return ret;
1385 }
1386 
1387 static int load_free_space_bitmaps(struct btrfs_caching_control *caching_ctl,
1388 				   struct btrfs_path *path,
1389 				   u32 expected_extent_count)
1390 {
1391 	struct btrfs_block_group *block_group;
1392 	struct btrfs_fs_info *fs_info;
1393 	struct btrfs_root *root;
1394 	struct btrfs_key key;
1395 	int prev_bit = 0, bit;
1396 	/* Initialize to silence GCC. */
1397 	u64 extent_start = 0;
1398 	u64 end, offset;
1399 	u64 total_found = 0;
1400 	u32 extent_count = 0;
1401 	int ret;
1402 
1403 	block_group = caching_ctl->block_group;
1404 	fs_info = block_group->fs_info;
1405 	root = fs_info->free_space_root;
1406 
1407 	end = block_group->start + block_group->length;
1408 
1409 	while (1) {
1410 		ret = btrfs_next_item(root, path);
1411 		if (ret < 0)
1412 			goto out;
1413 		if (ret)
1414 			break;
1415 
1416 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1417 
1418 		if (key.type == BTRFS_FREE_SPACE_INFO_KEY)
1419 			break;
1420 
1421 		ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
1422 		ASSERT(key.objectid < end && key.objectid + key.offset <= end);
1423 
1424 		caching_ctl->progress = key.objectid;
1425 
1426 		offset = key.objectid;
1427 		while (offset < key.objectid + key.offset) {
1428 			bit = free_space_test_bit(block_group, path, offset);
1429 			if (prev_bit == 0 && bit == 1) {
1430 				extent_start = offset;
1431 			} else if (prev_bit == 1 && bit == 0) {
1432 				total_found += add_new_free_space(block_group,
1433 								  extent_start,
1434 								  offset);
1435 				if (total_found > CACHING_CTL_WAKE_UP) {
1436 					total_found = 0;
1437 					wake_up(&caching_ctl->wait);
1438 				}
1439 				extent_count++;
1440 			}
1441 			prev_bit = bit;
1442 			offset += fs_info->sectorsize;
1443 		}
1444 	}
1445 	if (prev_bit == 1) {
1446 		total_found += add_new_free_space(block_group, extent_start,
1447 						  end);
1448 		extent_count++;
1449 	}
1450 
1451 	if (extent_count != expected_extent_count) {
1452 		btrfs_err(fs_info,
1453 			  "incorrect extent count for %llu; counted %u, expected %u",
1454 			  block_group->start, extent_count,
1455 			  expected_extent_count);
1456 		ASSERT(0);
1457 		ret = -EIO;
1458 		goto out;
1459 	}
1460 
1461 	caching_ctl->progress = (u64)-1;
1462 
1463 	ret = 0;
1464 out:
1465 	return ret;
1466 }
1467 
1468 static int load_free_space_extents(struct btrfs_caching_control *caching_ctl,
1469 				   struct btrfs_path *path,
1470 				   u32 expected_extent_count)
1471 {
1472 	struct btrfs_block_group *block_group;
1473 	struct btrfs_fs_info *fs_info;
1474 	struct btrfs_root *root;
1475 	struct btrfs_key key;
1476 	u64 end;
1477 	u64 total_found = 0;
1478 	u32 extent_count = 0;
1479 	int ret;
1480 
1481 	block_group = caching_ctl->block_group;
1482 	fs_info = block_group->fs_info;
1483 	root = fs_info->free_space_root;
1484 
1485 	end = block_group->start + block_group->length;
1486 
1487 	while (1) {
1488 		ret = btrfs_next_item(root, path);
1489 		if (ret < 0)
1490 			goto out;
1491 		if (ret)
1492 			break;
1493 
1494 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1495 
1496 		if (key.type == BTRFS_FREE_SPACE_INFO_KEY)
1497 			break;
1498 
1499 		ASSERT(key.type == BTRFS_FREE_SPACE_EXTENT_KEY);
1500 		ASSERT(key.objectid < end && key.objectid + key.offset <= end);
1501 
1502 		caching_ctl->progress = key.objectid;
1503 
1504 		total_found += add_new_free_space(block_group, key.objectid,
1505 						  key.objectid + key.offset);
1506 		if (total_found > CACHING_CTL_WAKE_UP) {
1507 			total_found = 0;
1508 			wake_up(&caching_ctl->wait);
1509 		}
1510 		extent_count++;
1511 	}
1512 
1513 	if (extent_count != expected_extent_count) {
1514 		btrfs_err(fs_info,
1515 			  "incorrect extent count for %llu; counted %u, expected %u",
1516 			  block_group->start, extent_count,
1517 			  expected_extent_count);
1518 		ASSERT(0);
1519 		ret = -EIO;
1520 		goto out;
1521 	}
1522 
1523 	caching_ctl->progress = (u64)-1;
1524 
1525 	ret = 0;
1526 out:
1527 	return ret;
1528 }
1529 
1530 int load_free_space_tree(struct btrfs_caching_control *caching_ctl)
1531 {
1532 	struct btrfs_block_group *block_group;
1533 	struct btrfs_free_space_info *info;
1534 	struct btrfs_path *path;
1535 	u32 extent_count, flags;
1536 	int ret;
1537 
1538 	block_group = caching_ctl->block_group;
1539 
1540 	path = btrfs_alloc_path();
1541 	if (!path)
1542 		return -ENOMEM;
1543 
1544 	/*
1545 	 * Just like caching_thread() doesn't want to deadlock on the extent
1546 	 * tree, we don't want to deadlock on the free space tree.
1547 	 */
1548 	path->skip_locking = 1;
1549 	path->search_commit_root = 1;
1550 	path->reada = READA_FORWARD;
1551 
1552 	info = search_free_space_info(NULL, block_group, path, 0);
1553 	if (IS_ERR(info)) {
1554 		ret = PTR_ERR(info);
1555 		goto out;
1556 	}
1557 	extent_count = btrfs_free_space_extent_count(path->nodes[0], info);
1558 	flags = btrfs_free_space_flags(path->nodes[0], info);
1559 
1560 	/*
1561 	 * We left path pointing to the free space info item, so now
1562 	 * load_free_space_foo can just iterate through the free space tree from
1563 	 * there.
1564 	 */
1565 	if (flags & BTRFS_FREE_SPACE_USING_BITMAPS)
1566 		ret = load_free_space_bitmaps(caching_ctl, path, extent_count);
1567 	else
1568 		ret = load_free_space_extents(caching_ctl, path, extent_count);
1569 
1570 out:
1571 	btrfs_free_path(path);
1572 	return ret;
1573 }
1574