xref: /openbmc/linux/fs/btrfs/free-space-tree.c (revision e1e38ea1)
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 
14 static int __add_block_group_free_space(struct btrfs_trans_handle *trans,
15 					struct btrfs_block_group_cache *block_group,
16 					struct btrfs_path *path);
17 
18 void set_free_space_tree_thresholds(struct btrfs_block_group_cache *cache)
19 {
20 	u32 bitmap_range;
21 	size_t bitmap_size;
22 	u64 num_bitmaps, total_bitmap_size;
23 
24 	/*
25 	 * We convert to bitmaps when the disk space required for using extents
26 	 * exceeds that required for using bitmaps.
27 	 */
28 	bitmap_range = cache->fs_info->sectorsize * BTRFS_FREE_SPACE_BITMAP_BITS;
29 	num_bitmaps = div_u64(cache->key.offset + bitmap_range - 1,
30 			      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_cache *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->key.objectid;
57 	key.type = BTRFS_FREE_SPACE_INFO_KEY;
58 	key.offset = block_group->key.offset;
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 struct btrfs_free_space_info *
78 search_free_space_info(struct btrfs_trans_handle *trans,
79 		       struct btrfs_fs_info *fs_info,
80 		       struct btrfs_block_group_cache *block_group,
81 		       struct btrfs_path *path, int cow)
82 {
83 	struct btrfs_root *root = fs_info->free_space_root;
84 	struct btrfs_key key;
85 	int ret;
86 
87 	key.objectid = block_group->key.objectid;
88 	key.type = BTRFS_FREE_SPACE_INFO_KEY;
89 	key.offset = block_group->key.offset;
90 
91 	ret = btrfs_search_slot(trans, root, &key, path, 0, cow);
92 	if (ret < 0)
93 		return ERR_PTR(ret);
94 	if (ret != 0) {
95 		btrfs_warn(fs_info, "missing free space info for %llu",
96 			   block_group->key.objectid);
97 		ASSERT(0);
98 		return ERR_PTR(-ENOENT);
99 	}
100 
101 	return btrfs_item_ptr(path->nodes[0], path->slots[0],
102 			      struct btrfs_free_space_info);
103 }
104 
105 /*
106  * btrfs_search_slot() but we're looking for the greatest key less than the
107  * passed key.
108  */
109 static int btrfs_search_prev_slot(struct btrfs_trans_handle *trans,
110 				  struct btrfs_root *root,
111 				  struct btrfs_key *key, struct btrfs_path *p,
112 				  int ins_len, int cow)
113 {
114 	int ret;
115 
116 	ret = btrfs_search_slot(trans, root, key, p, ins_len, cow);
117 	if (ret < 0)
118 		return ret;
119 
120 	if (ret == 0) {
121 		ASSERT(0);
122 		return -EIO;
123 	}
124 
125 	if (p->slots[0] == 0) {
126 		ASSERT(0);
127 		return -EIO;
128 	}
129 	p->slots[0]--;
130 
131 	return 0;
132 }
133 
134 static inline u32 free_space_bitmap_size(u64 size, u32 sectorsize)
135 {
136 	return DIV_ROUND_UP((u32)div_u64(size, sectorsize), BITS_PER_BYTE);
137 }
138 
139 static unsigned long *alloc_bitmap(u32 bitmap_size)
140 {
141 	unsigned long *ret;
142 	unsigned int nofs_flag;
143 	u32 bitmap_rounded_size = round_up(bitmap_size, sizeof(unsigned long));
144 
145 	/*
146 	 * GFP_NOFS doesn't work with kvmalloc(), but we really can't recurse
147 	 * into the filesystem as the free space bitmap can be modified in the
148 	 * critical section of a transaction commit.
149 	 *
150 	 * TODO: push the memalloc_nofs_{save,restore}() to the caller where we
151 	 * know that recursion is unsafe.
152 	 */
153 	nofs_flag = memalloc_nofs_save();
154 	ret = kvzalloc(bitmap_rounded_size, GFP_KERNEL);
155 	memalloc_nofs_restore(nofs_flag);
156 	return ret;
157 }
158 
159 static void le_bitmap_set(unsigned long *map, unsigned int start, int len)
160 {
161 	u8 *p = ((u8 *)map) + BIT_BYTE(start);
162 	const unsigned int size = start + len;
163 	int bits_to_set = BITS_PER_BYTE - (start % BITS_PER_BYTE);
164 	u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(start);
165 
166 	while (len - bits_to_set >= 0) {
167 		*p |= mask_to_set;
168 		len -= bits_to_set;
169 		bits_to_set = BITS_PER_BYTE;
170 		mask_to_set = ~0;
171 		p++;
172 	}
173 	if (len) {
174 		mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
175 		*p |= mask_to_set;
176 	}
177 }
178 
179 int convert_free_space_to_bitmaps(struct btrfs_trans_handle *trans,
180 				  struct btrfs_block_group_cache *block_group,
181 				  struct btrfs_path *path)
182 {
183 	struct btrfs_fs_info *fs_info = trans->fs_info;
184 	struct btrfs_root *root = fs_info->free_space_root;
185 	struct btrfs_free_space_info *info;
186 	struct btrfs_key key, found_key;
187 	struct extent_buffer *leaf;
188 	unsigned long *bitmap;
189 	char *bitmap_cursor;
190 	u64 start, end;
191 	u64 bitmap_range, i;
192 	u32 bitmap_size, flags, expected_extent_count;
193 	u32 extent_count = 0;
194 	int done = 0, nr;
195 	int ret;
196 
197 	bitmap_size = free_space_bitmap_size(block_group->key.offset,
198 					     fs_info->sectorsize);
199 	bitmap = alloc_bitmap(bitmap_size);
200 	if (!bitmap) {
201 		ret = -ENOMEM;
202 		goto out;
203 	}
204 
205 	start = block_group->key.objectid;
206 	end = block_group->key.objectid + block_group->key.offset;
207 
208 	key.objectid = end - 1;
209 	key.type = (u8)-1;
210 	key.offset = (u64)-1;
211 
212 	while (!done) {
213 		ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
214 		if (ret)
215 			goto out;
216 
217 		leaf = path->nodes[0];
218 		nr = 0;
219 		path->slots[0]++;
220 		while (path->slots[0] > 0) {
221 			btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
222 
223 			if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
224 				ASSERT(found_key.objectid == block_group->key.objectid);
225 				ASSERT(found_key.offset == block_group->key.offset);
226 				done = 1;
227 				break;
228 			} else if (found_key.type == BTRFS_FREE_SPACE_EXTENT_KEY) {
229 				u64 first, last;
230 
231 				ASSERT(found_key.objectid >= start);
232 				ASSERT(found_key.objectid < end);
233 				ASSERT(found_key.objectid + found_key.offset <= end);
234 
235 				first = div_u64(found_key.objectid - start,
236 						fs_info->sectorsize);
237 				last = div_u64(found_key.objectid + found_key.offset - start,
238 					       fs_info->sectorsize);
239 				le_bitmap_set(bitmap, first, last - first);
240 
241 				extent_count++;
242 				nr++;
243 				path->slots[0]--;
244 			} else {
245 				ASSERT(0);
246 			}
247 		}
248 
249 		ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
250 		if (ret)
251 			goto out;
252 		btrfs_release_path(path);
253 	}
254 
255 	info = search_free_space_info(trans, fs_info, block_group, path, 1);
256 	if (IS_ERR(info)) {
257 		ret = PTR_ERR(info);
258 		goto out;
259 	}
260 	leaf = path->nodes[0];
261 	flags = btrfs_free_space_flags(leaf, info);
262 	flags |= BTRFS_FREE_SPACE_USING_BITMAPS;
263 	btrfs_set_free_space_flags(leaf, info, flags);
264 	expected_extent_count = btrfs_free_space_extent_count(leaf, info);
265 	btrfs_mark_buffer_dirty(leaf);
266 	btrfs_release_path(path);
267 
268 	if (extent_count != expected_extent_count) {
269 		btrfs_err(fs_info,
270 			  "incorrect extent count for %llu; counted %u, expected %u",
271 			  block_group->key.objectid, extent_count,
272 			  expected_extent_count);
273 		ASSERT(0);
274 		ret = -EIO;
275 		goto out;
276 	}
277 
278 	bitmap_cursor = (char *)bitmap;
279 	bitmap_range = fs_info->sectorsize * BTRFS_FREE_SPACE_BITMAP_BITS;
280 	i = start;
281 	while (i < end) {
282 		unsigned long ptr;
283 		u64 extent_size;
284 		u32 data_size;
285 
286 		extent_size = min(end - i, bitmap_range);
287 		data_size = free_space_bitmap_size(extent_size,
288 						   fs_info->sectorsize);
289 
290 		key.objectid = i;
291 		key.type = BTRFS_FREE_SPACE_BITMAP_KEY;
292 		key.offset = extent_size;
293 
294 		ret = btrfs_insert_empty_item(trans, root, path, &key,
295 					      data_size);
296 		if (ret)
297 			goto out;
298 
299 		leaf = path->nodes[0];
300 		ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
301 		write_extent_buffer(leaf, bitmap_cursor, ptr,
302 				    data_size);
303 		btrfs_mark_buffer_dirty(leaf);
304 		btrfs_release_path(path);
305 
306 		i += extent_size;
307 		bitmap_cursor += data_size;
308 	}
309 
310 	ret = 0;
311 out:
312 	kvfree(bitmap);
313 	if (ret)
314 		btrfs_abort_transaction(trans, ret);
315 	return ret;
316 }
317 
318 int convert_free_space_to_extents(struct btrfs_trans_handle *trans,
319 				  struct btrfs_block_group_cache *block_group,
320 				  struct btrfs_path *path)
321 {
322 	struct btrfs_fs_info *fs_info = trans->fs_info;
323 	struct btrfs_root *root = fs_info->free_space_root;
324 	struct btrfs_free_space_info *info;
325 	struct btrfs_key key, found_key;
326 	struct extent_buffer *leaf;
327 	unsigned long *bitmap;
328 	u64 start, end;
329 	u32 bitmap_size, flags, expected_extent_count;
330 	unsigned long nrbits, start_bit, end_bit;
331 	u32 extent_count = 0;
332 	int done = 0, nr;
333 	int ret;
334 
335 	bitmap_size = free_space_bitmap_size(block_group->key.offset,
336 					     fs_info->sectorsize);
337 	bitmap = alloc_bitmap(bitmap_size);
338 	if (!bitmap) {
339 		ret = -ENOMEM;
340 		goto out;
341 	}
342 
343 	start = block_group->key.objectid;
344 	end = block_group->key.objectid + block_group->key.offset;
345 
346 	key.objectid = end - 1;
347 	key.type = (u8)-1;
348 	key.offset = (u64)-1;
349 
350 	while (!done) {
351 		ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
352 		if (ret)
353 			goto out;
354 
355 		leaf = path->nodes[0];
356 		nr = 0;
357 		path->slots[0]++;
358 		while (path->slots[0] > 0) {
359 			btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
360 
361 			if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
362 				ASSERT(found_key.objectid == block_group->key.objectid);
363 				ASSERT(found_key.offset == block_group->key.offset);
364 				done = 1;
365 				break;
366 			} else if (found_key.type == BTRFS_FREE_SPACE_BITMAP_KEY) {
367 				unsigned long ptr;
368 				char *bitmap_cursor;
369 				u32 bitmap_pos, data_size;
370 
371 				ASSERT(found_key.objectid >= start);
372 				ASSERT(found_key.objectid < end);
373 				ASSERT(found_key.objectid + found_key.offset <= end);
374 
375 				bitmap_pos = div_u64(found_key.objectid - start,
376 						     fs_info->sectorsize *
377 						     BITS_PER_BYTE);
378 				bitmap_cursor = ((char *)bitmap) + bitmap_pos;
379 				data_size = free_space_bitmap_size(found_key.offset,
380 								   fs_info->sectorsize);
381 
382 				ptr = btrfs_item_ptr_offset(leaf, path->slots[0] - 1);
383 				read_extent_buffer(leaf, bitmap_cursor, ptr,
384 						   data_size);
385 
386 				nr++;
387 				path->slots[0]--;
388 			} else {
389 				ASSERT(0);
390 			}
391 		}
392 
393 		ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
394 		if (ret)
395 			goto out;
396 		btrfs_release_path(path);
397 	}
398 
399 	info = search_free_space_info(trans, fs_info, block_group, path, 1);
400 	if (IS_ERR(info)) {
401 		ret = PTR_ERR(info);
402 		goto out;
403 	}
404 	leaf = path->nodes[0];
405 	flags = btrfs_free_space_flags(leaf, info);
406 	flags &= ~BTRFS_FREE_SPACE_USING_BITMAPS;
407 	btrfs_set_free_space_flags(leaf, info, flags);
408 	expected_extent_count = btrfs_free_space_extent_count(leaf, info);
409 	btrfs_mark_buffer_dirty(leaf);
410 	btrfs_release_path(path);
411 
412 	nrbits = div_u64(block_group->key.offset, block_group->fs_info->sectorsize);
413 	start_bit = find_next_bit_le(bitmap, nrbits, 0);
414 
415 	while (start_bit < nrbits) {
416 		end_bit = find_next_zero_bit_le(bitmap, nrbits, start_bit);
417 		ASSERT(start_bit < end_bit);
418 
419 		key.objectid = start + start_bit * block_group->fs_info->sectorsize;
420 		key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
421 		key.offset = (end_bit - start_bit) * block_group->fs_info->sectorsize;
422 
423 		ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
424 		if (ret)
425 			goto out;
426 		btrfs_release_path(path);
427 
428 		extent_count++;
429 
430 		start_bit = find_next_bit_le(bitmap, nrbits, end_bit);
431 	}
432 
433 	if (extent_count != expected_extent_count) {
434 		btrfs_err(fs_info,
435 			  "incorrect extent count for %llu; counted %u, expected %u",
436 			  block_group->key.objectid, extent_count,
437 			  expected_extent_count);
438 		ASSERT(0);
439 		ret = -EIO;
440 		goto out;
441 	}
442 
443 	ret = 0;
444 out:
445 	kvfree(bitmap);
446 	if (ret)
447 		btrfs_abort_transaction(trans, ret);
448 	return ret;
449 }
450 
451 static int update_free_space_extent_count(struct btrfs_trans_handle *trans,
452 					  struct btrfs_block_group_cache *block_group,
453 					  struct btrfs_path *path,
454 					  int new_extents)
455 {
456 	struct btrfs_free_space_info *info;
457 	u32 flags;
458 	u32 extent_count;
459 	int ret = 0;
460 
461 	if (new_extents == 0)
462 		return 0;
463 
464 	info = search_free_space_info(trans, trans->fs_info, block_group, path,
465 				      1);
466 	if (IS_ERR(info)) {
467 		ret = PTR_ERR(info);
468 		goto out;
469 	}
470 	flags = btrfs_free_space_flags(path->nodes[0], info);
471 	extent_count = btrfs_free_space_extent_count(path->nodes[0], info);
472 
473 	extent_count += new_extents;
474 	btrfs_set_free_space_extent_count(path->nodes[0], info, extent_count);
475 	btrfs_mark_buffer_dirty(path->nodes[0]);
476 	btrfs_release_path(path);
477 
478 	if (!(flags & BTRFS_FREE_SPACE_USING_BITMAPS) &&
479 	    extent_count > block_group->bitmap_high_thresh) {
480 		ret = convert_free_space_to_bitmaps(trans, block_group, path);
481 	} else if ((flags & BTRFS_FREE_SPACE_USING_BITMAPS) &&
482 		   extent_count < block_group->bitmap_low_thresh) {
483 		ret = convert_free_space_to_extents(trans, block_group, path);
484 	}
485 
486 out:
487 	return ret;
488 }
489 
490 int free_space_test_bit(struct btrfs_block_group_cache *block_group,
491 			struct btrfs_path *path, u64 offset)
492 {
493 	struct extent_buffer *leaf;
494 	struct btrfs_key key;
495 	u64 found_start, found_end;
496 	unsigned long ptr, i;
497 
498 	leaf = path->nodes[0];
499 	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
500 	ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
501 
502 	found_start = key.objectid;
503 	found_end = key.objectid + key.offset;
504 	ASSERT(offset >= found_start && offset < found_end);
505 
506 	ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
507 	i = div_u64(offset - found_start,
508 		    block_group->fs_info->sectorsize);
509 	return !!extent_buffer_test_bit(leaf, ptr, i);
510 }
511 
512 static void free_space_set_bits(struct btrfs_block_group_cache *block_group,
513 				struct btrfs_path *path, u64 *start, u64 *size,
514 				int bit)
515 {
516 	struct btrfs_fs_info *fs_info = block_group->fs_info;
517 	struct extent_buffer *leaf;
518 	struct btrfs_key key;
519 	u64 end = *start + *size;
520 	u64 found_start, found_end;
521 	unsigned long ptr, first, last;
522 
523 	leaf = path->nodes[0];
524 	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
525 	ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
526 
527 	found_start = key.objectid;
528 	found_end = key.objectid + key.offset;
529 	ASSERT(*start >= found_start && *start < found_end);
530 	ASSERT(end > found_start);
531 
532 	if (end > found_end)
533 		end = found_end;
534 
535 	ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
536 	first = div_u64(*start - found_start, fs_info->sectorsize);
537 	last = div_u64(end - found_start, fs_info->sectorsize);
538 	if (bit)
539 		extent_buffer_bitmap_set(leaf, ptr, first, last - first);
540 	else
541 		extent_buffer_bitmap_clear(leaf, ptr, first, last - first);
542 	btrfs_mark_buffer_dirty(leaf);
543 
544 	*size -= end - *start;
545 	*start = end;
546 }
547 
548 /*
549  * We can't use btrfs_next_item() in modify_free_space_bitmap() because
550  * btrfs_next_leaf() doesn't get the path for writing. We can forgo the fancy
551  * tree walking in btrfs_next_leaf() anyways because we know exactly what we're
552  * looking for.
553  */
554 static int free_space_next_bitmap(struct btrfs_trans_handle *trans,
555 				  struct btrfs_root *root, struct btrfs_path *p)
556 {
557 	struct btrfs_key key;
558 
559 	if (p->slots[0] + 1 < btrfs_header_nritems(p->nodes[0])) {
560 		p->slots[0]++;
561 		return 0;
562 	}
563 
564 	btrfs_item_key_to_cpu(p->nodes[0], &key, p->slots[0]);
565 	btrfs_release_path(p);
566 
567 	key.objectid += key.offset;
568 	key.type = (u8)-1;
569 	key.offset = (u64)-1;
570 
571 	return btrfs_search_prev_slot(trans, root, &key, p, 0, 1);
572 }
573 
574 /*
575  * If remove is 1, then we are removing free space, thus clearing bits in the
576  * bitmap. If remove is 0, then we are adding free space, thus setting bits in
577  * the bitmap.
578  */
579 static int modify_free_space_bitmap(struct btrfs_trans_handle *trans,
580 				    struct btrfs_block_group_cache *block_group,
581 				    struct btrfs_path *path,
582 				    u64 start, u64 size, int remove)
583 {
584 	struct btrfs_root *root = block_group->fs_info->free_space_root;
585 	struct btrfs_key key;
586 	u64 end = start + size;
587 	u64 cur_start, cur_size;
588 	int prev_bit, next_bit;
589 	int new_extents;
590 	int ret;
591 
592 	/*
593 	 * Read the bit for the block immediately before the extent of space if
594 	 * that block is within the block group.
595 	 */
596 	if (start > block_group->key.objectid) {
597 		u64 prev_block = start - block_group->fs_info->sectorsize;
598 
599 		key.objectid = prev_block;
600 		key.type = (u8)-1;
601 		key.offset = (u64)-1;
602 
603 		ret = btrfs_search_prev_slot(trans, root, &key, path, 0, 1);
604 		if (ret)
605 			goto out;
606 
607 		prev_bit = free_space_test_bit(block_group, path, prev_block);
608 
609 		/* The previous block may have been in the previous bitmap. */
610 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
611 		if (start >= key.objectid + key.offset) {
612 			ret = free_space_next_bitmap(trans, root, path);
613 			if (ret)
614 				goto out;
615 		}
616 	} else {
617 		key.objectid = start;
618 		key.type = (u8)-1;
619 		key.offset = (u64)-1;
620 
621 		ret = btrfs_search_prev_slot(trans, root, &key, path, 0, 1);
622 		if (ret)
623 			goto out;
624 
625 		prev_bit = -1;
626 	}
627 
628 	/*
629 	 * Iterate over all of the bitmaps overlapped by the extent of space,
630 	 * clearing/setting bits as required.
631 	 */
632 	cur_start = start;
633 	cur_size = size;
634 	while (1) {
635 		free_space_set_bits(block_group, path, &cur_start, &cur_size,
636 				    !remove);
637 		if (cur_size == 0)
638 			break;
639 		ret = free_space_next_bitmap(trans, root, path);
640 		if (ret)
641 			goto out;
642 	}
643 
644 	/*
645 	 * Read the bit for the block immediately after the extent of space if
646 	 * that block is within the block group.
647 	 */
648 	if (end < block_group->key.objectid + block_group->key.offset) {
649 		/* The next block may be in the next bitmap. */
650 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
651 		if (end >= key.objectid + key.offset) {
652 			ret = free_space_next_bitmap(trans, root, path);
653 			if (ret)
654 				goto out;
655 		}
656 
657 		next_bit = free_space_test_bit(block_group, path, end);
658 	} else {
659 		next_bit = -1;
660 	}
661 
662 	if (remove) {
663 		new_extents = -1;
664 		if (prev_bit == 1) {
665 			/* Leftover on the left. */
666 			new_extents++;
667 		}
668 		if (next_bit == 1) {
669 			/* Leftover on the right. */
670 			new_extents++;
671 		}
672 	} else {
673 		new_extents = 1;
674 		if (prev_bit == 1) {
675 			/* Merging with neighbor on the left. */
676 			new_extents--;
677 		}
678 		if (next_bit == 1) {
679 			/* Merging with neighbor on the right. */
680 			new_extents--;
681 		}
682 	}
683 
684 	btrfs_release_path(path);
685 	ret = update_free_space_extent_count(trans, block_group, path,
686 					     new_extents);
687 
688 out:
689 	return ret;
690 }
691 
692 static int remove_free_space_extent(struct btrfs_trans_handle *trans,
693 				    struct btrfs_block_group_cache *block_group,
694 				    struct btrfs_path *path,
695 				    u64 start, u64 size)
696 {
697 	struct btrfs_root *root = trans->fs_info->free_space_root;
698 	struct btrfs_key key;
699 	u64 found_start, found_end;
700 	u64 end = start + size;
701 	int new_extents = -1;
702 	int ret;
703 
704 	key.objectid = start;
705 	key.type = (u8)-1;
706 	key.offset = (u64)-1;
707 
708 	ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
709 	if (ret)
710 		goto out;
711 
712 	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
713 
714 	ASSERT(key.type == BTRFS_FREE_SPACE_EXTENT_KEY);
715 
716 	found_start = key.objectid;
717 	found_end = key.objectid + key.offset;
718 	ASSERT(start >= found_start && end <= found_end);
719 
720 	/*
721 	 * Okay, now that we've found the free space extent which contains the
722 	 * free space that we are removing, there are four cases:
723 	 *
724 	 * 1. We're using the whole extent: delete the key we found and
725 	 * decrement the free space extent count.
726 	 * 2. We are using part of the extent starting at the beginning: delete
727 	 * the key we found and insert a new key representing the leftover at
728 	 * the end. There is no net change in the number of extents.
729 	 * 3. We are using part of the extent ending at the end: delete the key
730 	 * we found and insert a new key representing the leftover at the
731 	 * beginning. There is no net change in the number of extents.
732 	 * 4. We are using part of the extent in the middle: delete the key we
733 	 * found and insert two new keys representing the leftovers on each
734 	 * side. Where we used to have one extent, we now have two, so increment
735 	 * the extent count. We may need to convert the block group to bitmaps
736 	 * as a result.
737 	 */
738 
739 	/* Delete the existing key (cases 1-4). */
740 	ret = btrfs_del_item(trans, root, path);
741 	if (ret)
742 		goto out;
743 
744 	/* Add a key for leftovers at the beginning (cases 3 and 4). */
745 	if (start > found_start) {
746 		key.objectid = found_start;
747 		key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
748 		key.offset = start - found_start;
749 
750 		btrfs_release_path(path);
751 		ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
752 		if (ret)
753 			goto out;
754 		new_extents++;
755 	}
756 
757 	/* Add a key for leftovers at the end (cases 2 and 4). */
758 	if (end < found_end) {
759 		key.objectid = end;
760 		key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
761 		key.offset = found_end - end;
762 
763 		btrfs_release_path(path);
764 		ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
765 		if (ret)
766 			goto out;
767 		new_extents++;
768 	}
769 
770 	btrfs_release_path(path);
771 	ret = update_free_space_extent_count(trans, block_group, path,
772 					     new_extents);
773 
774 out:
775 	return ret;
776 }
777 
778 int __remove_from_free_space_tree(struct btrfs_trans_handle *trans,
779 				  struct btrfs_block_group_cache *block_group,
780 				  struct btrfs_path *path, u64 start, u64 size)
781 {
782 	struct btrfs_free_space_info *info;
783 	u32 flags;
784 	int ret;
785 
786 	if (block_group->needs_free_space) {
787 		ret = __add_block_group_free_space(trans, block_group, path);
788 		if (ret)
789 			return ret;
790 	}
791 
792 	info = search_free_space_info(NULL, trans->fs_info, block_group, path,
793 				      0);
794 	if (IS_ERR(info))
795 		return PTR_ERR(info);
796 	flags = btrfs_free_space_flags(path->nodes[0], info);
797 	btrfs_release_path(path);
798 
799 	if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) {
800 		return modify_free_space_bitmap(trans, block_group, path,
801 						start, size, 1);
802 	} else {
803 		return remove_free_space_extent(trans, block_group, path,
804 						start, size);
805 	}
806 }
807 
808 int remove_from_free_space_tree(struct btrfs_trans_handle *trans,
809 				u64 start, u64 size)
810 {
811 	struct btrfs_block_group_cache *block_group;
812 	struct btrfs_path *path;
813 	int ret;
814 
815 	if (!btrfs_fs_compat_ro(trans->fs_info, FREE_SPACE_TREE))
816 		return 0;
817 
818 	path = btrfs_alloc_path();
819 	if (!path) {
820 		ret = -ENOMEM;
821 		goto out;
822 	}
823 
824 	block_group = btrfs_lookup_block_group(trans->fs_info, start);
825 	if (!block_group) {
826 		ASSERT(0);
827 		ret = -ENOENT;
828 		goto out;
829 	}
830 
831 	mutex_lock(&block_group->free_space_lock);
832 	ret = __remove_from_free_space_tree(trans, block_group, path, start,
833 					    size);
834 	mutex_unlock(&block_group->free_space_lock);
835 
836 	btrfs_put_block_group(block_group);
837 out:
838 	btrfs_free_path(path);
839 	if (ret)
840 		btrfs_abort_transaction(trans, ret);
841 	return ret;
842 }
843 
844 static int add_free_space_extent(struct btrfs_trans_handle *trans,
845 				 struct btrfs_block_group_cache *block_group,
846 				 struct btrfs_path *path,
847 				 u64 start, u64 size)
848 {
849 	struct btrfs_root *root = trans->fs_info->free_space_root;
850 	struct btrfs_key key, new_key;
851 	u64 found_start, found_end;
852 	u64 end = start + size;
853 	int new_extents = 1;
854 	int ret;
855 
856 	/*
857 	 * We are adding a new extent of free space, but we need to merge
858 	 * extents. There are four cases here:
859 	 *
860 	 * 1. The new extent does not have any immediate neighbors to merge
861 	 * with: add the new key and increment the free space extent count. We
862 	 * may need to convert the block group to bitmaps as a result.
863 	 * 2. The new extent has an immediate neighbor before it: remove the
864 	 * previous key and insert a new key combining both of them. There is no
865 	 * net change in the number of extents.
866 	 * 3. The new extent has an immediate neighbor after it: remove the next
867 	 * key and insert a new key combining both of them. There is no net
868 	 * change in the number of extents.
869 	 * 4. The new extent has immediate neighbors on both sides: remove both
870 	 * of the keys and insert a new key combining all of them. Where we used
871 	 * to have two extents, we now have one, so decrement the extent count.
872 	 */
873 
874 	new_key.objectid = start;
875 	new_key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
876 	new_key.offset = size;
877 
878 	/* Search for a neighbor on the left. */
879 	if (start == block_group->key.objectid)
880 		goto right;
881 	key.objectid = start - 1;
882 	key.type = (u8)-1;
883 	key.offset = (u64)-1;
884 
885 	ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
886 	if (ret)
887 		goto out;
888 
889 	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
890 
891 	if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) {
892 		ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY);
893 		btrfs_release_path(path);
894 		goto right;
895 	}
896 
897 	found_start = key.objectid;
898 	found_end = key.objectid + key.offset;
899 	ASSERT(found_start >= block_group->key.objectid &&
900 	       found_end > block_group->key.objectid);
901 	ASSERT(found_start < start && found_end <= start);
902 
903 	/*
904 	 * Delete the neighbor on the left and absorb it into the new key (cases
905 	 * 2 and 4).
906 	 */
907 	if (found_end == start) {
908 		ret = btrfs_del_item(trans, root, path);
909 		if (ret)
910 			goto out;
911 		new_key.objectid = found_start;
912 		new_key.offset += key.offset;
913 		new_extents--;
914 	}
915 	btrfs_release_path(path);
916 
917 right:
918 	/* Search for a neighbor on the right. */
919 	if (end == block_group->key.objectid + block_group->key.offset)
920 		goto insert;
921 	key.objectid = end;
922 	key.type = (u8)-1;
923 	key.offset = (u64)-1;
924 
925 	ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
926 	if (ret)
927 		goto out;
928 
929 	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
930 
931 	if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) {
932 		ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY);
933 		btrfs_release_path(path);
934 		goto insert;
935 	}
936 
937 	found_start = key.objectid;
938 	found_end = key.objectid + key.offset;
939 	ASSERT(found_start >= block_group->key.objectid &&
940 	       found_end > block_group->key.objectid);
941 	ASSERT((found_start < start && found_end <= start) ||
942 	       (found_start >= end && found_end > end));
943 
944 	/*
945 	 * Delete the neighbor on the right and absorb it into the new key
946 	 * (cases 3 and 4).
947 	 */
948 	if (found_start == end) {
949 		ret = btrfs_del_item(trans, root, path);
950 		if (ret)
951 			goto out;
952 		new_key.offset += key.offset;
953 		new_extents--;
954 	}
955 	btrfs_release_path(path);
956 
957 insert:
958 	/* Insert the new key (cases 1-4). */
959 	ret = btrfs_insert_empty_item(trans, root, path, &new_key, 0);
960 	if (ret)
961 		goto out;
962 
963 	btrfs_release_path(path);
964 	ret = update_free_space_extent_count(trans, block_group, path,
965 					     new_extents);
966 
967 out:
968 	return ret;
969 }
970 
971 int __add_to_free_space_tree(struct btrfs_trans_handle *trans,
972 			     struct btrfs_block_group_cache *block_group,
973 			     struct btrfs_path *path, u64 start, u64 size)
974 {
975 	struct btrfs_fs_info *fs_info = trans->fs_info;
976 	struct btrfs_free_space_info *info;
977 	u32 flags;
978 	int ret;
979 
980 	if (block_group->needs_free_space) {
981 		ret = __add_block_group_free_space(trans, block_group, path);
982 		if (ret)
983 			return ret;
984 	}
985 
986 	info = search_free_space_info(NULL, fs_info, block_group, path, 0);
987 	if (IS_ERR(info))
988 		return PTR_ERR(info);
989 	flags = btrfs_free_space_flags(path->nodes[0], info);
990 	btrfs_release_path(path);
991 
992 	if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) {
993 		return modify_free_space_bitmap(trans, block_group, path,
994 						start, size, 0);
995 	} else {
996 		return add_free_space_extent(trans, block_group, path, start,
997 					     size);
998 	}
999 }
1000 
1001 int add_to_free_space_tree(struct btrfs_trans_handle *trans,
1002 			   u64 start, u64 size)
1003 {
1004 	struct btrfs_block_group_cache *block_group;
1005 	struct btrfs_path *path;
1006 	int ret;
1007 
1008 	if (!btrfs_fs_compat_ro(trans->fs_info, FREE_SPACE_TREE))
1009 		return 0;
1010 
1011 	path = btrfs_alloc_path();
1012 	if (!path) {
1013 		ret = -ENOMEM;
1014 		goto out;
1015 	}
1016 
1017 	block_group = btrfs_lookup_block_group(trans->fs_info, start);
1018 	if (!block_group) {
1019 		ASSERT(0);
1020 		ret = -ENOENT;
1021 		goto out;
1022 	}
1023 
1024 	mutex_lock(&block_group->free_space_lock);
1025 	ret = __add_to_free_space_tree(trans, block_group, path, start, size);
1026 	mutex_unlock(&block_group->free_space_lock);
1027 
1028 	btrfs_put_block_group(block_group);
1029 out:
1030 	btrfs_free_path(path);
1031 	if (ret)
1032 		btrfs_abort_transaction(trans, ret);
1033 	return ret;
1034 }
1035 
1036 /*
1037  * Populate the free space tree by walking the extent tree. Operations on the
1038  * extent tree that happen as a result of writes to the free space tree will go
1039  * through the normal add/remove hooks.
1040  */
1041 static int populate_free_space_tree(struct btrfs_trans_handle *trans,
1042 				    struct btrfs_block_group_cache *block_group)
1043 {
1044 	struct btrfs_root *extent_root = trans->fs_info->extent_root;
1045 	struct btrfs_path *path, *path2;
1046 	struct btrfs_key key;
1047 	u64 start, end;
1048 	int ret;
1049 
1050 	path = btrfs_alloc_path();
1051 	if (!path)
1052 		return -ENOMEM;
1053 	path->reada = READA_FORWARD;
1054 
1055 	path2 = btrfs_alloc_path();
1056 	if (!path2) {
1057 		btrfs_free_path(path);
1058 		return -ENOMEM;
1059 	}
1060 
1061 	ret = add_new_free_space_info(trans, block_group, path2);
1062 	if (ret)
1063 		goto out;
1064 
1065 	mutex_lock(&block_group->free_space_lock);
1066 
1067 	/*
1068 	 * Iterate through all of the extent and metadata items in this block
1069 	 * group, adding the free space between them and the free space at the
1070 	 * end. Note that EXTENT_ITEM and METADATA_ITEM are less than
1071 	 * BLOCK_GROUP_ITEM, so an extent may precede the block group that it's
1072 	 * contained in.
1073 	 */
1074 	key.objectid = block_group->key.objectid;
1075 	key.type = BTRFS_EXTENT_ITEM_KEY;
1076 	key.offset = 0;
1077 
1078 	ret = btrfs_search_slot_for_read(extent_root, &key, path, 1, 0);
1079 	if (ret < 0)
1080 		goto out_locked;
1081 	ASSERT(ret == 0);
1082 
1083 	start = block_group->key.objectid;
1084 	end = block_group->key.objectid + block_group->key.offset;
1085 	while (1) {
1086 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1087 
1088 		if (key.type == BTRFS_EXTENT_ITEM_KEY ||
1089 		    key.type == BTRFS_METADATA_ITEM_KEY) {
1090 			if (key.objectid >= end)
1091 				break;
1092 
1093 			if (start < key.objectid) {
1094 				ret = __add_to_free_space_tree(trans,
1095 							       block_group,
1096 							       path2, start,
1097 							       key.objectid -
1098 							       start);
1099 				if (ret)
1100 					goto out_locked;
1101 			}
1102 			start = key.objectid;
1103 			if (key.type == BTRFS_METADATA_ITEM_KEY)
1104 				start += trans->fs_info->nodesize;
1105 			else
1106 				start += key.offset;
1107 		} else if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
1108 			if (key.objectid != block_group->key.objectid)
1109 				break;
1110 		}
1111 
1112 		ret = btrfs_next_item(extent_root, path);
1113 		if (ret < 0)
1114 			goto out_locked;
1115 		if (ret)
1116 			break;
1117 	}
1118 	if (start < end) {
1119 		ret = __add_to_free_space_tree(trans, block_group, path2,
1120 					       start, end - start);
1121 		if (ret)
1122 			goto out_locked;
1123 	}
1124 
1125 	ret = 0;
1126 out_locked:
1127 	mutex_unlock(&block_group->free_space_lock);
1128 out:
1129 	btrfs_free_path(path2);
1130 	btrfs_free_path(path);
1131 	return ret;
1132 }
1133 
1134 int btrfs_create_free_space_tree(struct btrfs_fs_info *fs_info)
1135 {
1136 	struct btrfs_trans_handle *trans;
1137 	struct btrfs_root *tree_root = fs_info->tree_root;
1138 	struct btrfs_root *free_space_root;
1139 	struct btrfs_block_group_cache *block_group;
1140 	struct rb_node *node;
1141 	int ret;
1142 
1143 	trans = btrfs_start_transaction(tree_root, 0);
1144 	if (IS_ERR(trans))
1145 		return PTR_ERR(trans);
1146 
1147 	set_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1148 	free_space_root = btrfs_create_tree(trans, fs_info,
1149 					    BTRFS_FREE_SPACE_TREE_OBJECTID);
1150 	if (IS_ERR(free_space_root)) {
1151 		ret = PTR_ERR(free_space_root);
1152 		goto abort;
1153 	}
1154 	fs_info->free_space_root = free_space_root;
1155 
1156 	node = rb_first(&fs_info->block_group_cache_tree);
1157 	while (node) {
1158 		block_group = rb_entry(node, struct btrfs_block_group_cache,
1159 				       cache_node);
1160 		ret = populate_free_space_tree(trans, block_group);
1161 		if (ret)
1162 			goto abort;
1163 		node = rb_next(node);
1164 	}
1165 
1166 	btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE);
1167 	btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID);
1168 	clear_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1169 
1170 	return btrfs_commit_transaction(trans);
1171 
1172 abort:
1173 	clear_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1174 	btrfs_abort_transaction(trans, ret);
1175 	btrfs_end_transaction(trans);
1176 	return ret;
1177 }
1178 
1179 static int clear_free_space_tree(struct btrfs_trans_handle *trans,
1180 				 struct btrfs_root *root)
1181 {
1182 	struct btrfs_path *path;
1183 	struct btrfs_key key;
1184 	int nr;
1185 	int ret;
1186 
1187 	path = btrfs_alloc_path();
1188 	if (!path)
1189 		return -ENOMEM;
1190 
1191 	path->leave_spinning = 1;
1192 
1193 	key.objectid = 0;
1194 	key.type = 0;
1195 	key.offset = 0;
1196 
1197 	while (1) {
1198 		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1199 		if (ret < 0)
1200 			goto out;
1201 
1202 		nr = btrfs_header_nritems(path->nodes[0]);
1203 		if (!nr)
1204 			break;
1205 
1206 		path->slots[0] = 0;
1207 		ret = btrfs_del_items(trans, root, path, 0, nr);
1208 		if (ret)
1209 			goto out;
1210 
1211 		btrfs_release_path(path);
1212 	}
1213 
1214 	ret = 0;
1215 out:
1216 	btrfs_free_path(path);
1217 	return ret;
1218 }
1219 
1220 int btrfs_clear_free_space_tree(struct btrfs_fs_info *fs_info)
1221 {
1222 	struct btrfs_trans_handle *trans;
1223 	struct btrfs_root *tree_root = fs_info->tree_root;
1224 	struct btrfs_root *free_space_root = fs_info->free_space_root;
1225 	int ret;
1226 
1227 	trans = btrfs_start_transaction(tree_root, 0);
1228 	if (IS_ERR(trans))
1229 		return PTR_ERR(trans);
1230 
1231 	btrfs_clear_fs_compat_ro(fs_info, FREE_SPACE_TREE);
1232 	btrfs_clear_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID);
1233 	fs_info->free_space_root = NULL;
1234 
1235 	ret = clear_free_space_tree(trans, free_space_root);
1236 	if (ret)
1237 		goto abort;
1238 
1239 	ret = btrfs_del_root(trans, &free_space_root->root_key);
1240 	if (ret)
1241 		goto abort;
1242 
1243 	list_del(&free_space_root->dirty_list);
1244 
1245 	btrfs_tree_lock(free_space_root->node);
1246 	clean_tree_block(fs_info, free_space_root->node);
1247 	btrfs_tree_unlock(free_space_root->node);
1248 	btrfs_free_tree_block(trans, free_space_root, free_space_root->node,
1249 			      0, 1);
1250 
1251 	free_extent_buffer(free_space_root->node);
1252 	free_extent_buffer(free_space_root->commit_root);
1253 	kfree(free_space_root);
1254 
1255 	return btrfs_commit_transaction(trans);
1256 
1257 abort:
1258 	btrfs_abort_transaction(trans, ret);
1259 	btrfs_end_transaction(trans);
1260 	return ret;
1261 }
1262 
1263 static int __add_block_group_free_space(struct btrfs_trans_handle *trans,
1264 					struct btrfs_block_group_cache *block_group,
1265 					struct btrfs_path *path)
1266 {
1267 	int ret;
1268 
1269 	block_group->needs_free_space = 0;
1270 
1271 	ret = add_new_free_space_info(trans, block_group, path);
1272 	if (ret)
1273 		return ret;
1274 
1275 	return __add_to_free_space_tree(trans, block_group, path,
1276 					block_group->key.objectid,
1277 					block_group->key.offset);
1278 }
1279 
1280 int add_block_group_free_space(struct btrfs_trans_handle *trans,
1281 			       struct btrfs_block_group_cache *block_group)
1282 {
1283 	struct btrfs_fs_info *fs_info = trans->fs_info;
1284 	struct btrfs_path *path = NULL;
1285 	int ret = 0;
1286 
1287 	if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
1288 		return 0;
1289 
1290 	mutex_lock(&block_group->free_space_lock);
1291 	if (!block_group->needs_free_space)
1292 		goto out;
1293 
1294 	path = btrfs_alloc_path();
1295 	if (!path) {
1296 		ret = -ENOMEM;
1297 		goto out;
1298 	}
1299 
1300 	ret = __add_block_group_free_space(trans, block_group, path);
1301 
1302 out:
1303 	btrfs_free_path(path);
1304 	mutex_unlock(&block_group->free_space_lock);
1305 	if (ret)
1306 		btrfs_abort_transaction(trans, ret);
1307 	return ret;
1308 }
1309 
1310 int remove_block_group_free_space(struct btrfs_trans_handle *trans,
1311 				  struct btrfs_block_group_cache *block_group)
1312 {
1313 	struct btrfs_root *root = trans->fs_info->free_space_root;
1314 	struct btrfs_path *path;
1315 	struct btrfs_key key, found_key;
1316 	struct extent_buffer *leaf;
1317 	u64 start, end;
1318 	int done = 0, nr;
1319 	int ret;
1320 
1321 	if (!btrfs_fs_compat_ro(trans->fs_info, FREE_SPACE_TREE))
1322 		return 0;
1323 
1324 	if (block_group->needs_free_space) {
1325 		/* We never added this block group to the free space tree. */
1326 		return 0;
1327 	}
1328 
1329 	path = btrfs_alloc_path();
1330 	if (!path) {
1331 		ret = -ENOMEM;
1332 		goto out;
1333 	}
1334 
1335 	start = block_group->key.objectid;
1336 	end = block_group->key.objectid + block_group->key.offset;
1337 
1338 	key.objectid = end - 1;
1339 	key.type = (u8)-1;
1340 	key.offset = (u64)-1;
1341 
1342 	while (!done) {
1343 		ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
1344 		if (ret)
1345 			goto out;
1346 
1347 		leaf = path->nodes[0];
1348 		nr = 0;
1349 		path->slots[0]++;
1350 		while (path->slots[0] > 0) {
1351 			btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
1352 
1353 			if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
1354 				ASSERT(found_key.objectid == block_group->key.objectid);
1355 				ASSERT(found_key.offset == block_group->key.offset);
1356 				done = 1;
1357 				nr++;
1358 				path->slots[0]--;
1359 				break;
1360 			} else if (found_key.type == BTRFS_FREE_SPACE_EXTENT_KEY ||
1361 				   found_key.type == BTRFS_FREE_SPACE_BITMAP_KEY) {
1362 				ASSERT(found_key.objectid >= start);
1363 				ASSERT(found_key.objectid < end);
1364 				ASSERT(found_key.objectid + found_key.offset <= end);
1365 				nr++;
1366 				path->slots[0]--;
1367 			} else {
1368 				ASSERT(0);
1369 			}
1370 		}
1371 
1372 		ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
1373 		if (ret)
1374 			goto out;
1375 		btrfs_release_path(path);
1376 	}
1377 
1378 	ret = 0;
1379 out:
1380 	btrfs_free_path(path);
1381 	if (ret)
1382 		btrfs_abort_transaction(trans, ret);
1383 	return ret;
1384 }
1385 
1386 static int load_free_space_bitmaps(struct btrfs_caching_control *caching_ctl,
1387 				   struct btrfs_path *path,
1388 				   u32 expected_extent_count)
1389 {
1390 	struct btrfs_block_group_cache *block_group;
1391 	struct btrfs_fs_info *fs_info;
1392 	struct btrfs_root *root;
1393 	struct btrfs_key key;
1394 	int prev_bit = 0, bit;
1395 	/* Initialize to silence GCC. */
1396 	u64 extent_start = 0;
1397 	u64 end, offset;
1398 	u64 total_found = 0;
1399 	u32 extent_count = 0;
1400 	int ret;
1401 
1402 	block_group = caching_ctl->block_group;
1403 	fs_info = block_group->fs_info;
1404 	root = fs_info->free_space_root;
1405 
1406 	end = block_group->key.objectid + block_group->key.offset;
1407 
1408 	while (1) {
1409 		ret = btrfs_next_item(root, path);
1410 		if (ret < 0)
1411 			goto out;
1412 		if (ret)
1413 			break;
1414 
1415 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1416 
1417 		if (key.type == BTRFS_FREE_SPACE_INFO_KEY)
1418 			break;
1419 
1420 		ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
1421 		ASSERT(key.objectid < end && key.objectid + key.offset <= end);
1422 
1423 		caching_ctl->progress = key.objectid;
1424 
1425 		offset = key.objectid;
1426 		while (offset < key.objectid + key.offset) {
1427 			bit = free_space_test_bit(block_group, path, offset);
1428 			if (prev_bit == 0 && bit == 1) {
1429 				extent_start = offset;
1430 			} else if (prev_bit == 1 && bit == 0) {
1431 				total_found += add_new_free_space(block_group,
1432 								  extent_start,
1433 								  offset);
1434 				if (total_found > CACHING_CTL_WAKE_UP) {
1435 					total_found = 0;
1436 					wake_up(&caching_ctl->wait);
1437 				}
1438 				extent_count++;
1439 			}
1440 			prev_bit = bit;
1441 			offset += fs_info->sectorsize;
1442 		}
1443 	}
1444 	if (prev_bit == 1) {
1445 		total_found += add_new_free_space(block_group, extent_start,
1446 						  end);
1447 		extent_count++;
1448 	}
1449 
1450 	if (extent_count != expected_extent_count) {
1451 		btrfs_err(fs_info,
1452 			  "incorrect extent count for %llu; counted %u, expected %u",
1453 			  block_group->key.objectid, extent_count,
1454 			  expected_extent_count);
1455 		ASSERT(0);
1456 		ret = -EIO;
1457 		goto out;
1458 	}
1459 
1460 	caching_ctl->progress = (u64)-1;
1461 
1462 	ret = 0;
1463 out:
1464 	return ret;
1465 }
1466 
1467 static int load_free_space_extents(struct btrfs_caching_control *caching_ctl,
1468 				   struct btrfs_path *path,
1469 				   u32 expected_extent_count)
1470 {
1471 	struct btrfs_block_group_cache *block_group;
1472 	struct btrfs_fs_info *fs_info;
1473 	struct btrfs_root *root;
1474 	struct btrfs_key key;
1475 	u64 end;
1476 	u64 total_found = 0;
1477 	u32 extent_count = 0;
1478 	int ret;
1479 
1480 	block_group = caching_ctl->block_group;
1481 	fs_info = block_group->fs_info;
1482 	root = fs_info->free_space_root;
1483 
1484 	end = block_group->key.objectid + block_group->key.offset;
1485 
1486 	while (1) {
1487 		ret = btrfs_next_item(root, path);
1488 		if (ret < 0)
1489 			goto out;
1490 		if (ret)
1491 			break;
1492 
1493 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1494 
1495 		if (key.type == BTRFS_FREE_SPACE_INFO_KEY)
1496 			break;
1497 
1498 		ASSERT(key.type == BTRFS_FREE_SPACE_EXTENT_KEY);
1499 		ASSERT(key.objectid < end && key.objectid + key.offset <= end);
1500 
1501 		caching_ctl->progress = key.objectid;
1502 
1503 		total_found += add_new_free_space(block_group, key.objectid,
1504 						  key.objectid + key.offset);
1505 		if (total_found > CACHING_CTL_WAKE_UP) {
1506 			total_found = 0;
1507 			wake_up(&caching_ctl->wait);
1508 		}
1509 		extent_count++;
1510 	}
1511 
1512 	if (extent_count != expected_extent_count) {
1513 		btrfs_err(fs_info,
1514 			  "incorrect extent count for %llu; counted %u, expected %u",
1515 			  block_group->key.objectid, extent_count,
1516 			  expected_extent_count);
1517 		ASSERT(0);
1518 		ret = -EIO;
1519 		goto out;
1520 	}
1521 
1522 	caching_ctl->progress = (u64)-1;
1523 
1524 	ret = 0;
1525 out:
1526 	return ret;
1527 }
1528 
1529 int load_free_space_tree(struct btrfs_caching_control *caching_ctl)
1530 {
1531 	struct btrfs_block_group_cache *block_group;
1532 	struct btrfs_fs_info *fs_info;
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 	fs_info = block_group->fs_info;
1540 
1541 	path = btrfs_alloc_path();
1542 	if (!path)
1543 		return -ENOMEM;
1544 
1545 	/*
1546 	 * Just like caching_thread() doesn't want to deadlock on the extent
1547 	 * tree, we don't want to deadlock on the free space tree.
1548 	 */
1549 	path->skip_locking = 1;
1550 	path->search_commit_root = 1;
1551 	path->reada = READA_FORWARD;
1552 
1553 	info = search_free_space_info(NULL, fs_info, block_group, path, 0);
1554 	if (IS_ERR(info)) {
1555 		ret = PTR_ERR(info);
1556 		goto out;
1557 	}
1558 	extent_count = btrfs_free_space_extent_count(path->nodes[0], info);
1559 	flags = btrfs_free_space_flags(path->nodes[0], info);
1560 
1561 	/*
1562 	 * We left path pointing to the free space info item, so now
1563 	 * load_free_space_foo can just iterate through the free space tree from
1564 	 * there.
1565 	 */
1566 	if (flags & BTRFS_FREE_SPACE_USING_BITMAPS)
1567 		ret = load_free_space_bitmaps(caching_ctl, path, extent_count);
1568 	else
1569 		ret = load_free_space_extents(caching_ctl, path, extent_count);
1570 
1571 out:
1572 	btrfs_free_path(path);
1573 	return ret;
1574 }
1575