xref: /openbmc/linux/fs/btrfs/free-space-tree.c (revision 941518d6)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2015 Facebook.  All rights reserved.
4  */
5 
6 #include <linux/kernel.h>
7 #include <linux/sched/mm.h>
8 #include "ctree.h"
9 #include "disk-io.h"
10 #include "locking.h"
11 #include "free-space-tree.h"
12 #include "transaction.h"
13 #include "block-group.h"
14 
15 static int __add_block_group_free_space(struct btrfs_trans_handle *trans,
16 					struct btrfs_block_group *block_group,
17 					struct btrfs_path *path);
18 
19 static struct btrfs_root *btrfs_free_space_root(
20 				struct btrfs_block_group *block_group)
21 {
22 	struct btrfs_key key = {
23 		.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID,
24 		.type = BTRFS_ROOT_ITEM_KEY,
25 		.offset = 0,
26 	};
27 
28 	return btrfs_global_root(block_group->fs_info, &key);
29 }
30 
31 void set_free_space_tree_thresholds(struct btrfs_block_group *cache)
32 {
33 	u32 bitmap_range;
34 	size_t bitmap_size;
35 	u64 num_bitmaps, total_bitmap_size;
36 
37 	if (WARN_ON(cache->length == 0))
38 		btrfs_warn(cache->fs_info, "block group %llu length is zero",
39 			   cache->start);
40 
41 	/*
42 	 * We convert to bitmaps when the disk space required for using extents
43 	 * exceeds that required for using bitmaps.
44 	 */
45 	bitmap_range = cache->fs_info->sectorsize * BTRFS_FREE_SPACE_BITMAP_BITS;
46 	num_bitmaps = div_u64(cache->length + bitmap_range - 1, bitmap_range);
47 	bitmap_size = sizeof(struct btrfs_item) + BTRFS_FREE_SPACE_BITMAP_SIZE;
48 	total_bitmap_size = num_bitmaps * bitmap_size;
49 	cache->bitmap_high_thresh = div_u64(total_bitmap_size,
50 					    sizeof(struct btrfs_item));
51 
52 	/*
53 	 * We allow for a small buffer between the high threshold and low
54 	 * threshold to avoid thrashing back and forth between the two formats.
55 	 */
56 	if (cache->bitmap_high_thresh > 100)
57 		cache->bitmap_low_thresh = cache->bitmap_high_thresh - 100;
58 	else
59 		cache->bitmap_low_thresh = 0;
60 }
61 
62 static int add_new_free_space_info(struct btrfs_trans_handle *trans,
63 				   struct btrfs_block_group *block_group,
64 				   struct btrfs_path *path)
65 {
66 	struct btrfs_root *root = btrfs_free_space_root(block_group);
67 	struct btrfs_free_space_info *info;
68 	struct btrfs_key key;
69 	struct extent_buffer *leaf;
70 	int ret;
71 
72 	key.objectid = block_group->start;
73 	key.type = BTRFS_FREE_SPACE_INFO_KEY;
74 	key.offset = block_group->length;
75 
76 	ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*info));
77 	if (ret)
78 		goto out;
79 
80 	leaf = path->nodes[0];
81 	info = btrfs_item_ptr(leaf, path->slots[0],
82 			      struct btrfs_free_space_info);
83 	btrfs_set_free_space_extent_count(leaf, info, 0);
84 	btrfs_set_free_space_flags(leaf, info, 0);
85 	btrfs_mark_buffer_dirty(leaf);
86 
87 	ret = 0;
88 out:
89 	btrfs_release_path(path);
90 	return ret;
91 }
92 
93 EXPORT_FOR_TESTS
94 struct btrfs_free_space_info *search_free_space_info(
95 		struct btrfs_trans_handle *trans,
96 		struct btrfs_block_group *block_group,
97 		struct btrfs_path *path, int cow)
98 {
99 	struct btrfs_fs_info *fs_info = block_group->fs_info;
100 	struct btrfs_root *root = btrfs_free_space_root(block_group);
101 	struct btrfs_key key;
102 	int ret;
103 
104 	key.objectid = block_group->start;
105 	key.type = BTRFS_FREE_SPACE_INFO_KEY;
106 	key.offset = block_group->length;
107 
108 	ret = btrfs_search_slot(trans, root, &key, path, 0, cow);
109 	if (ret < 0)
110 		return ERR_PTR(ret);
111 	if (ret != 0) {
112 		btrfs_warn(fs_info, "missing free space info for %llu",
113 			   block_group->start);
114 		ASSERT(0);
115 		return ERR_PTR(-ENOENT);
116 	}
117 
118 	return btrfs_item_ptr(path->nodes[0], path->slots[0],
119 			      struct btrfs_free_space_info);
120 }
121 
122 /*
123  * btrfs_search_slot() but we're looking for the greatest key less than the
124  * passed key.
125  */
126 static int btrfs_search_prev_slot(struct btrfs_trans_handle *trans,
127 				  struct btrfs_root *root,
128 				  struct btrfs_key *key, struct btrfs_path *p,
129 				  int ins_len, int cow)
130 {
131 	int ret;
132 
133 	ret = btrfs_search_slot(trans, root, key, p, ins_len, cow);
134 	if (ret < 0)
135 		return ret;
136 
137 	if (ret == 0) {
138 		ASSERT(0);
139 		return -EIO;
140 	}
141 
142 	if (p->slots[0] == 0) {
143 		ASSERT(0);
144 		return -EIO;
145 	}
146 	p->slots[0]--;
147 
148 	return 0;
149 }
150 
151 static inline u32 free_space_bitmap_size(const struct btrfs_fs_info *fs_info,
152 					 u64 size)
153 {
154 	return DIV_ROUND_UP(size >> fs_info->sectorsize_bits, BITS_PER_BYTE);
155 }
156 
157 static unsigned long *alloc_bitmap(u32 bitmap_size)
158 {
159 	unsigned long *ret;
160 	unsigned int nofs_flag;
161 	u32 bitmap_rounded_size = round_up(bitmap_size, sizeof(unsigned long));
162 
163 	/*
164 	 * GFP_NOFS doesn't work with kvmalloc(), but we really can't recurse
165 	 * into the filesystem as the free space bitmap can be modified in the
166 	 * critical section of a transaction commit.
167 	 *
168 	 * TODO: push the memalloc_nofs_{save,restore}() to the caller where we
169 	 * know that recursion is unsafe.
170 	 */
171 	nofs_flag = memalloc_nofs_save();
172 	ret = kvzalloc(bitmap_rounded_size, GFP_KERNEL);
173 	memalloc_nofs_restore(nofs_flag);
174 	return ret;
175 }
176 
177 static void le_bitmap_set(unsigned long *map, unsigned int start, int len)
178 {
179 	u8 *p = ((u8 *)map) + BIT_BYTE(start);
180 	const unsigned int size = start + len;
181 	int bits_to_set = BITS_PER_BYTE - (start % BITS_PER_BYTE);
182 	u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(start);
183 
184 	while (len - bits_to_set >= 0) {
185 		*p |= mask_to_set;
186 		len -= bits_to_set;
187 		bits_to_set = BITS_PER_BYTE;
188 		mask_to_set = ~0;
189 		p++;
190 	}
191 	if (len) {
192 		mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
193 		*p |= mask_to_set;
194 	}
195 }
196 
197 EXPORT_FOR_TESTS
198 int convert_free_space_to_bitmaps(struct btrfs_trans_handle *trans,
199 				  struct btrfs_block_group *block_group,
200 				  struct btrfs_path *path)
201 {
202 	struct btrfs_fs_info *fs_info = trans->fs_info;
203 	struct btrfs_root *root = btrfs_free_space_root(block_group);
204 	struct btrfs_free_space_info *info;
205 	struct btrfs_key key, found_key;
206 	struct extent_buffer *leaf;
207 	unsigned long *bitmap;
208 	char *bitmap_cursor;
209 	u64 start, end;
210 	u64 bitmap_range, i;
211 	u32 bitmap_size, flags, expected_extent_count;
212 	u32 extent_count = 0;
213 	int done = 0, nr;
214 	int ret;
215 
216 	bitmap_size = free_space_bitmap_size(fs_info, block_group->length);
217 	bitmap = alloc_bitmap(bitmap_size);
218 	if (!bitmap) {
219 		ret = -ENOMEM;
220 		goto out;
221 	}
222 
223 	start = block_group->start;
224 	end = block_group->start + block_group->length;
225 
226 	key.objectid = end - 1;
227 	key.type = (u8)-1;
228 	key.offset = (u64)-1;
229 
230 	while (!done) {
231 		ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
232 		if (ret)
233 			goto out;
234 
235 		leaf = path->nodes[0];
236 		nr = 0;
237 		path->slots[0]++;
238 		while (path->slots[0] > 0) {
239 			btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
240 
241 			if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
242 				ASSERT(found_key.objectid == block_group->start);
243 				ASSERT(found_key.offset == block_group->length);
244 				done = 1;
245 				break;
246 			} else if (found_key.type == BTRFS_FREE_SPACE_EXTENT_KEY) {
247 				u64 first, last;
248 
249 				ASSERT(found_key.objectid >= start);
250 				ASSERT(found_key.objectid < end);
251 				ASSERT(found_key.objectid + found_key.offset <= end);
252 
253 				first = div_u64(found_key.objectid - start,
254 						fs_info->sectorsize);
255 				last = div_u64(found_key.objectid + found_key.offset - start,
256 					       fs_info->sectorsize);
257 				le_bitmap_set(bitmap, first, last - first);
258 
259 				extent_count++;
260 				nr++;
261 				path->slots[0]--;
262 			} else {
263 				ASSERT(0);
264 			}
265 		}
266 
267 		ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
268 		if (ret)
269 			goto out;
270 		btrfs_release_path(path);
271 	}
272 
273 	info = search_free_space_info(trans, block_group, path, 1);
274 	if (IS_ERR(info)) {
275 		ret = PTR_ERR(info);
276 		goto out;
277 	}
278 	leaf = path->nodes[0];
279 	flags = btrfs_free_space_flags(leaf, info);
280 	flags |= BTRFS_FREE_SPACE_USING_BITMAPS;
281 	btrfs_set_free_space_flags(leaf, info, flags);
282 	expected_extent_count = btrfs_free_space_extent_count(leaf, info);
283 	btrfs_mark_buffer_dirty(leaf);
284 	btrfs_release_path(path);
285 
286 	if (extent_count != expected_extent_count) {
287 		btrfs_err(fs_info,
288 			  "incorrect extent count for %llu; counted %u, expected %u",
289 			  block_group->start, extent_count,
290 			  expected_extent_count);
291 		ASSERT(0);
292 		ret = -EIO;
293 		goto out;
294 	}
295 
296 	bitmap_cursor = (char *)bitmap;
297 	bitmap_range = fs_info->sectorsize * BTRFS_FREE_SPACE_BITMAP_BITS;
298 	i = start;
299 	while (i < end) {
300 		unsigned long ptr;
301 		u64 extent_size;
302 		u32 data_size;
303 
304 		extent_size = min(end - i, bitmap_range);
305 		data_size = free_space_bitmap_size(fs_info, extent_size);
306 
307 		key.objectid = i;
308 		key.type = BTRFS_FREE_SPACE_BITMAP_KEY;
309 		key.offset = extent_size;
310 
311 		ret = btrfs_insert_empty_item(trans, root, path, &key,
312 					      data_size);
313 		if (ret)
314 			goto out;
315 
316 		leaf = path->nodes[0];
317 		ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
318 		write_extent_buffer(leaf, bitmap_cursor, ptr,
319 				    data_size);
320 		btrfs_mark_buffer_dirty(leaf);
321 		btrfs_release_path(path);
322 
323 		i += extent_size;
324 		bitmap_cursor += data_size;
325 	}
326 
327 	ret = 0;
328 out:
329 	kvfree(bitmap);
330 	if (ret)
331 		btrfs_abort_transaction(trans, ret);
332 	return ret;
333 }
334 
335 EXPORT_FOR_TESTS
336 int convert_free_space_to_extents(struct btrfs_trans_handle *trans,
337 				  struct btrfs_block_group *block_group,
338 				  struct btrfs_path *path)
339 {
340 	struct btrfs_fs_info *fs_info = trans->fs_info;
341 	struct btrfs_root *root = btrfs_free_space_root(block_group);
342 	struct btrfs_free_space_info *info;
343 	struct btrfs_key key, found_key;
344 	struct extent_buffer *leaf;
345 	unsigned long *bitmap;
346 	u64 start, end;
347 	u32 bitmap_size, flags, expected_extent_count;
348 	unsigned long nrbits, start_bit, end_bit;
349 	u32 extent_count = 0;
350 	int done = 0, nr;
351 	int ret;
352 
353 	bitmap_size = free_space_bitmap_size(fs_info, block_group->length);
354 	bitmap = alloc_bitmap(bitmap_size);
355 	if (!bitmap) {
356 		ret = -ENOMEM;
357 		goto out;
358 	}
359 
360 	start = block_group->start;
361 	end = block_group->start + block_group->length;
362 
363 	key.objectid = end - 1;
364 	key.type = (u8)-1;
365 	key.offset = (u64)-1;
366 
367 	while (!done) {
368 		ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
369 		if (ret)
370 			goto out;
371 
372 		leaf = path->nodes[0];
373 		nr = 0;
374 		path->slots[0]++;
375 		while (path->slots[0] > 0) {
376 			btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
377 
378 			if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
379 				ASSERT(found_key.objectid == block_group->start);
380 				ASSERT(found_key.offset == block_group->length);
381 				done = 1;
382 				break;
383 			} else if (found_key.type == BTRFS_FREE_SPACE_BITMAP_KEY) {
384 				unsigned long ptr;
385 				char *bitmap_cursor;
386 				u32 bitmap_pos, data_size;
387 
388 				ASSERT(found_key.objectid >= start);
389 				ASSERT(found_key.objectid < end);
390 				ASSERT(found_key.objectid + found_key.offset <= end);
391 
392 				bitmap_pos = div_u64(found_key.objectid - start,
393 						     fs_info->sectorsize *
394 						     BITS_PER_BYTE);
395 				bitmap_cursor = ((char *)bitmap) + bitmap_pos;
396 				data_size = free_space_bitmap_size(fs_info,
397 								found_key.offset);
398 
399 				ptr = btrfs_item_ptr_offset(leaf, path->slots[0] - 1);
400 				read_extent_buffer(leaf, bitmap_cursor, ptr,
401 						   data_size);
402 
403 				nr++;
404 				path->slots[0]--;
405 			} else {
406 				ASSERT(0);
407 			}
408 		}
409 
410 		ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
411 		if (ret)
412 			goto out;
413 		btrfs_release_path(path);
414 	}
415 
416 	info = search_free_space_info(trans, block_group, path, 1);
417 	if (IS_ERR(info)) {
418 		ret = PTR_ERR(info);
419 		goto out;
420 	}
421 	leaf = path->nodes[0];
422 	flags = btrfs_free_space_flags(leaf, info);
423 	flags &= ~BTRFS_FREE_SPACE_USING_BITMAPS;
424 	btrfs_set_free_space_flags(leaf, info, flags);
425 	expected_extent_count = btrfs_free_space_extent_count(leaf, info);
426 	btrfs_mark_buffer_dirty(leaf);
427 	btrfs_release_path(path);
428 
429 	nrbits = block_group->length >> block_group->fs_info->sectorsize_bits;
430 	start_bit = find_next_bit_le(bitmap, nrbits, 0);
431 
432 	while (start_bit < nrbits) {
433 		end_bit = find_next_zero_bit_le(bitmap, nrbits, start_bit);
434 		ASSERT(start_bit < end_bit);
435 
436 		key.objectid = start + start_bit * block_group->fs_info->sectorsize;
437 		key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
438 		key.offset = (end_bit - start_bit) * block_group->fs_info->sectorsize;
439 
440 		ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
441 		if (ret)
442 			goto out;
443 		btrfs_release_path(path);
444 
445 		extent_count++;
446 
447 		start_bit = find_next_bit_le(bitmap, nrbits, end_bit);
448 	}
449 
450 	if (extent_count != expected_extent_count) {
451 		btrfs_err(fs_info,
452 			  "incorrect extent count for %llu; counted %u, expected %u",
453 			  block_group->start, extent_count,
454 			  expected_extent_count);
455 		ASSERT(0);
456 		ret = -EIO;
457 		goto out;
458 	}
459 
460 	ret = 0;
461 out:
462 	kvfree(bitmap);
463 	if (ret)
464 		btrfs_abort_transaction(trans, ret);
465 	return ret;
466 }
467 
468 static int update_free_space_extent_count(struct btrfs_trans_handle *trans,
469 					  struct btrfs_block_group *block_group,
470 					  struct btrfs_path *path,
471 					  int new_extents)
472 {
473 	struct btrfs_free_space_info *info;
474 	u32 flags;
475 	u32 extent_count;
476 	int ret = 0;
477 
478 	if (new_extents == 0)
479 		return 0;
480 
481 	info = search_free_space_info(trans, block_group, path, 1);
482 	if (IS_ERR(info)) {
483 		ret = PTR_ERR(info);
484 		goto out;
485 	}
486 	flags = btrfs_free_space_flags(path->nodes[0], info);
487 	extent_count = btrfs_free_space_extent_count(path->nodes[0], info);
488 
489 	extent_count += new_extents;
490 	btrfs_set_free_space_extent_count(path->nodes[0], info, extent_count);
491 	btrfs_mark_buffer_dirty(path->nodes[0]);
492 	btrfs_release_path(path);
493 
494 	if (!(flags & BTRFS_FREE_SPACE_USING_BITMAPS) &&
495 	    extent_count > block_group->bitmap_high_thresh) {
496 		ret = convert_free_space_to_bitmaps(trans, block_group, path);
497 	} else if ((flags & BTRFS_FREE_SPACE_USING_BITMAPS) &&
498 		   extent_count < block_group->bitmap_low_thresh) {
499 		ret = convert_free_space_to_extents(trans, block_group, path);
500 	}
501 
502 out:
503 	return ret;
504 }
505 
506 EXPORT_FOR_TESTS
507 int free_space_test_bit(struct btrfs_block_group *block_group,
508 			struct btrfs_path *path, u64 offset)
509 {
510 	struct extent_buffer *leaf;
511 	struct btrfs_key key;
512 	u64 found_start, found_end;
513 	unsigned long ptr, i;
514 
515 	leaf = path->nodes[0];
516 	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
517 	ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
518 
519 	found_start = key.objectid;
520 	found_end = key.objectid + key.offset;
521 	ASSERT(offset >= found_start && offset < found_end);
522 
523 	ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
524 	i = div_u64(offset - found_start,
525 		    block_group->fs_info->sectorsize);
526 	return !!extent_buffer_test_bit(leaf, ptr, i);
527 }
528 
529 static void free_space_set_bits(struct btrfs_block_group *block_group,
530 				struct btrfs_path *path, u64 *start, u64 *size,
531 				int bit)
532 {
533 	struct btrfs_fs_info *fs_info = block_group->fs_info;
534 	struct extent_buffer *leaf;
535 	struct btrfs_key key;
536 	u64 end = *start + *size;
537 	u64 found_start, found_end;
538 	unsigned long ptr, first, last;
539 
540 	leaf = path->nodes[0];
541 	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
542 	ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
543 
544 	found_start = key.objectid;
545 	found_end = key.objectid + key.offset;
546 	ASSERT(*start >= found_start && *start < found_end);
547 	ASSERT(end > found_start);
548 
549 	if (end > found_end)
550 		end = found_end;
551 
552 	ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
553 	first = (*start - found_start) >> fs_info->sectorsize_bits;
554 	last = (end - found_start) >> fs_info->sectorsize_bits;
555 	if (bit)
556 		extent_buffer_bitmap_set(leaf, ptr, first, last - first);
557 	else
558 		extent_buffer_bitmap_clear(leaf, ptr, first, last - first);
559 	btrfs_mark_buffer_dirty(leaf);
560 
561 	*size -= end - *start;
562 	*start = end;
563 }
564 
565 /*
566  * We can't use btrfs_next_item() in modify_free_space_bitmap() because
567  * btrfs_next_leaf() doesn't get the path for writing. We can forgo the fancy
568  * tree walking in btrfs_next_leaf() anyways because we know exactly what we're
569  * looking for.
570  */
571 static int free_space_next_bitmap(struct btrfs_trans_handle *trans,
572 				  struct btrfs_root *root, struct btrfs_path *p)
573 {
574 	struct btrfs_key key;
575 
576 	if (p->slots[0] + 1 < btrfs_header_nritems(p->nodes[0])) {
577 		p->slots[0]++;
578 		return 0;
579 	}
580 
581 	btrfs_item_key_to_cpu(p->nodes[0], &key, p->slots[0]);
582 	btrfs_release_path(p);
583 
584 	key.objectid += key.offset;
585 	key.type = (u8)-1;
586 	key.offset = (u64)-1;
587 
588 	return btrfs_search_prev_slot(trans, root, &key, p, 0, 1);
589 }
590 
591 /*
592  * If remove is 1, then we are removing free space, thus clearing bits in the
593  * bitmap. If remove is 0, then we are adding free space, thus setting bits in
594  * the bitmap.
595  */
596 static int modify_free_space_bitmap(struct btrfs_trans_handle *trans,
597 				    struct btrfs_block_group *block_group,
598 				    struct btrfs_path *path,
599 				    u64 start, u64 size, int remove)
600 {
601 	struct btrfs_root *root = btrfs_free_space_root(block_group);
602 	struct btrfs_key key;
603 	u64 end = start + size;
604 	u64 cur_start, cur_size;
605 	int prev_bit, next_bit;
606 	int new_extents;
607 	int ret;
608 
609 	/*
610 	 * Read the bit for the block immediately before the extent of space if
611 	 * that block is within the block group.
612 	 */
613 	if (start > block_group->start) {
614 		u64 prev_block = start - block_group->fs_info->sectorsize;
615 
616 		key.objectid = prev_block;
617 		key.type = (u8)-1;
618 		key.offset = (u64)-1;
619 
620 		ret = btrfs_search_prev_slot(trans, root, &key, path, 0, 1);
621 		if (ret)
622 			goto out;
623 
624 		prev_bit = free_space_test_bit(block_group, path, prev_block);
625 
626 		/* The previous block may have been in the previous bitmap. */
627 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
628 		if (start >= key.objectid + key.offset) {
629 			ret = free_space_next_bitmap(trans, root, path);
630 			if (ret)
631 				goto out;
632 		}
633 	} else {
634 		key.objectid = start;
635 		key.type = (u8)-1;
636 		key.offset = (u64)-1;
637 
638 		ret = btrfs_search_prev_slot(trans, root, &key, path, 0, 1);
639 		if (ret)
640 			goto out;
641 
642 		prev_bit = -1;
643 	}
644 
645 	/*
646 	 * Iterate over all of the bitmaps overlapped by the extent of space,
647 	 * clearing/setting bits as required.
648 	 */
649 	cur_start = start;
650 	cur_size = size;
651 	while (1) {
652 		free_space_set_bits(block_group, path, &cur_start, &cur_size,
653 				    !remove);
654 		if (cur_size == 0)
655 			break;
656 		ret = free_space_next_bitmap(trans, root, path);
657 		if (ret)
658 			goto out;
659 	}
660 
661 	/*
662 	 * Read the bit for the block immediately after the extent of space if
663 	 * that block is within the block group.
664 	 */
665 	if (end < block_group->start + block_group->length) {
666 		/* The next block may be in the next bitmap. */
667 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
668 		if (end >= key.objectid + key.offset) {
669 			ret = free_space_next_bitmap(trans, root, path);
670 			if (ret)
671 				goto out;
672 		}
673 
674 		next_bit = free_space_test_bit(block_group, path, end);
675 	} else {
676 		next_bit = -1;
677 	}
678 
679 	if (remove) {
680 		new_extents = -1;
681 		if (prev_bit == 1) {
682 			/* Leftover on the left. */
683 			new_extents++;
684 		}
685 		if (next_bit == 1) {
686 			/* Leftover on the right. */
687 			new_extents++;
688 		}
689 	} else {
690 		new_extents = 1;
691 		if (prev_bit == 1) {
692 			/* Merging with neighbor on the left. */
693 			new_extents--;
694 		}
695 		if (next_bit == 1) {
696 			/* Merging with neighbor on the right. */
697 			new_extents--;
698 		}
699 	}
700 
701 	btrfs_release_path(path);
702 	ret = update_free_space_extent_count(trans, block_group, path,
703 					     new_extents);
704 
705 out:
706 	return ret;
707 }
708 
709 static int remove_free_space_extent(struct btrfs_trans_handle *trans,
710 				    struct btrfs_block_group *block_group,
711 				    struct btrfs_path *path,
712 				    u64 start, u64 size)
713 {
714 	struct btrfs_root *root = btrfs_free_space_root(block_group);
715 	struct btrfs_key key;
716 	u64 found_start, found_end;
717 	u64 end = start + size;
718 	int new_extents = -1;
719 	int ret;
720 
721 	key.objectid = start;
722 	key.type = (u8)-1;
723 	key.offset = (u64)-1;
724 
725 	ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
726 	if (ret)
727 		goto out;
728 
729 	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
730 
731 	ASSERT(key.type == BTRFS_FREE_SPACE_EXTENT_KEY);
732 
733 	found_start = key.objectid;
734 	found_end = key.objectid + key.offset;
735 	ASSERT(start >= found_start && end <= found_end);
736 
737 	/*
738 	 * Okay, now that we've found the free space extent which contains the
739 	 * free space that we are removing, there are four cases:
740 	 *
741 	 * 1. We're using the whole extent: delete the key we found and
742 	 * decrement the free space extent count.
743 	 * 2. We are using part of the extent starting at the beginning: delete
744 	 * the key we found and insert a new key representing the leftover at
745 	 * the end. There is no net change in the number of extents.
746 	 * 3. We are using part of the extent ending at the end: delete the key
747 	 * we found and insert a new key representing the leftover at the
748 	 * beginning. There is no net change in the number of extents.
749 	 * 4. We are using part of the extent in the middle: delete the key we
750 	 * found and insert two new keys representing the leftovers on each
751 	 * side. Where we used to have one extent, we now have two, so increment
752 	 * the extent count. We may need to convert the block group to bitmaps
753 	 * as a result.
754 	 */
755 
756 	/* Delete the existing key (cases 1-4). */
757 	ret = btrfs_del_item(trans, root, path);
758 	if (ret)
759 		goto out;
760 
761 	/* Add a key for leftovers at the beginning (cases 3 and 4). */
762 	if (start > found_start) {
763 		key.objectid = found_start;
764 		key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
765 		key.offset = start - found_start;
766 
767 		btrfs_release_path(path);
768 		ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
769 		if (ret)
770 			goto out;
771 		new_extents++;
772 	}
773 
774 	/* Add a key for leftovers at the end (cases 2 and 4). */
775 	if (end < found_end) {
776 		key.objectid = end;
777 		key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
778 		key.offset = found_end - end;
779 
780 		btrfs_release_path(path);
781 		ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
782 		if (ret)
783 			goto out;
784 		new_extents++;
785 	}
786 
787 	btrfs_release_path(path);
788 	ret = update_free_space_extent_count(trans, block_group, path,
789 					     new_extents);
790 
791 out:
792 	return ret;
793 }
794 
795 EXPORT_FOR_TESTS
796 int __remove_from_free_space_tree(struct btrfs_trans_handle *trans,
797 				  struct btrfs_block_group *block_group,
798 				  struct btrfs_path *path, u64 start, u64 size)
799 {
800 	struct btrfs_free_space_info *info;
801 	u32 flags;
802 	int ret;
803 
804 	if (block_group->needs_free_space) {
805 		ret = __add_block_group_free_space(trans, block_group, path);
806 		if (ret)
807 			return ret;
808 	}
809 
810 	info = search_free_space_info(NULL, block_group, path, 0);
811 	if (IS_ERR(info))
812 		return PTR_ERR(info);
813 	flags = btrfs_free_space_flags(path->nodes[0], info);
814 	btrfs_release_path(path);
815 
816 	if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) {
817 		return modify_free_space_bitmap(trans, block_group, path,
818 						start, size, 1);
819 	} else {
820 		return remove_free_space_extent(trans, block_group, path,
821 						start, size);
822 	}
823 }
824 
825 int remove_from_free_space_tree(struct btrfs_trans_handle *trans,
826 				u64 start, u64 size)
827 {
828 	struct btrfs_block_group *block_group;
829 	struct btrfs_path *path;
830 	int ret;
831 
832 	if (!btrfs_fs_compat_ro(trans->fs_info, FREE_SPACE_TREE))
833 		return 0;
834 
835 	path = btrfs_alloc_path();
836 	if (!path) {
837 		ret = -ENOMEM;
838 		goto out;
839 	}
840 
841 	block_group = btrfs_lookup_block_group(trans->fs_info, start);
842 	if (!block_group) {
843 		ASSERT(0);
844 		ret = -ENOENT;
845 		goto out;
846 	}
847 
848 	mutex_lock(&block_group->free_space_lock);
849 	ret = __remove_from_free_space_tree(trans, block_group, path, start,
850 					    size);
851 	mutex_unlock(&block_group->free_space_lock);
852 
853 	btrfs_put_block_group(block_group);
854 out:
855 	btrfs_free_path(path);
856 	if (ret)
857 		btrfs_abort_transaction(trans, ret);
858 	return ret;
859 }
860 
861 static int add_free_space_extent(struct btrfs_trans_handle *trans,
862 				 struct btrfs_block_group *block_group,
863 				 struct btrfs_path *path,
864 				 u64 start, u64 size)
865 {
866 	struct btrfs_root *root = btrfs_free_space_root(block_group);
867 	struct btrfs_key key, new_key;
868 	u64 found_start, found_end;
869 	u64 end = start + size;
870 	int new_extents = 1;
871 	int ret;
872 
873 	/*
874 	 * We are adding a new extent of free space, but we need to merge
875 	 * extents. There are four cases here:
876 	 *
877 	 * 1. The new extent does not have any immediate neighbors to merge
878 	 * with: add the new key and increment the free space extent count. We
879 	 * may need to convert the block group to bitmaps as a result.
880 	 * 2. The new extent has an immediate neighbor before it: remove the
881 	 * previous key and insert a new key combining both of them. There is no
882 	 * net change in the number of extents.
883 	 * 3. The new extent has an immediate neighbor after it: remove the next
884 	 * key and insert a new key combining both of them. There is no net
885 	 * change in the number of extents.
886 	 * 4. The new extent has immediate neighbors on both sides: remove both
887 	 * of the keys and insert a new key combining all of them. Where we used
888 	 * to have two extents, we now have one, so decrement the extent count.
889 	 */
890 
891 	new_key.objectid = start;
892 	new_key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
893 	new_key.offset = size;
894 
895 	/* Search for a neighbor on the left. */
896 	if (start == block_group->start)
897 		goto right;
898 	key.objectid = start - 1;
899 	key.type = (u8)-1;
900 	key.offset = (u64)-1;
901 
902 	ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
903 	if (ret)
904 		goto out;
905 
906 	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
907 
908 	if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) {
909 		ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY);
910 		btrfs_release_path(path);
911 		goto right;
912 	}
913 
914 	found_start = key.objectid;
915 	found_end = key.objectid + key.offset;
916 	ASSERT(found_start >= block_group->start &&
917 	       found_end > block_group->start);
918 	ASSERT(found_start < start && found_end <= start);
919 
920 	/*
921 	 * Delete the neighbor on the left and absorb it into the new key (cases
922 	 * 2 and 4).
923 	 */
924 	if (found_end == start) {
925 		ret = btrfs_del_item(trans, root, path);
926 		if (ret)
927 			goto out;
928 		new_key.objectid = found_start;
929 		new_key.offset += key.offset;
930 		new_extents--;
931 	}
932 	btrfs_release_path(path);
933 
934 right:
935 	/* Search for a neighbor on the right. */
936 	if (end == block_group->start + block_group->length)
937 		goto insert;
938 	key.objectid = end;
939 	key.type = (u8)-1;
940 	key.offset = (u64)-1;
941 
942 	ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
943 	if (ret)
944 		goto out;
945 
946 	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
947 
948 	if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) {
949 		ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY);
950 		btrfs_release_path(path);
951 		goto insert;
952 	}
953 
954 	found_start = key.objectid;
955 	found_end = key.objectid + key.offset;
956 	ASSERT(found_start >= block_group->start &&
957 	       found_end > block_group->start);
958 	ASSERT((found_start < start && found_end <= start) ||
959 	       (found_start >= end && found_end > end));
960 
961 	/*
962 	 * Delete the neighbor on the right and absorb it into the new key
963 	 * (cases 3 and 4).
964 	 */
965 	if (found_start == end) {
966 		ret = btrfs_del_item(trans, root, path);
967 		if (ret)
968 			goto out;
969 		new_key.offset += key.offset;
970 		new_extents--;
971 	}
972 	btrfs_release_path(path);
973 
974 insert:
975 	/* Insert the new key (cases 1-4). */
976 	ret = btrfs_insert_empty_item(trans, root, path, &new_key, 0);
977 	if (ret)
978 		goto out;
979 
980 	btrfs_release_path(path);
981 	ret = update_free_space_extent_count(trans, block_group, path,
982 					     new_extents);
983 
984 out:
985 	return ret;
986 }
987 
988 EXPORT_FOR_TESTS
989 int __add_to_free_space_tree(struct btrfs_trans_handle *trans,
990 			     struct btrfs_block_group *block_group,
991 			     struct btrfs_path *path, u64 start, u64 size)
992 {
993 	struct btrfs_free_space_info *info;
994 	u32 flags;
995 	int ret;
996 
997 	if (block_group->needs_free_space) {
998 		ret = __add_block_group_free_space(trans, block_group, path);
999 		if (ret)
1000 			return ret;
1001 	}
1002 
1003 	info = search_free_space_info(NULL, block_group, path, 0);
1004 	if (IS_ERR(info))
1005 		return PTR_ERR(info);
1006 	flags = btrfs_free_space_flags(path->nodes[0], info);
1007 	btrfs_release_path(path);
1008 
1009 	if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) {
1010 		return modify_free_space_bitmap(trans, block_group, path,
1011 						start, size, 0);
1012 	} else {
1013 		return add_free_space_extent(trans, block_group, path, start,
1014 					     size);
1015 	}
1016 }
1017 
1018 int add_to_free_space_tree(struct btrfs_trans_handle *trans,
1019 			   u64 start, u64 size)
1020 {
1021 	struct btrfs_block_group *block_group;
1022 	struct btrfs_path *path;
1023 	int ret;
1024 
1025 	if (!btrfs_fs_compat_ro(trans->fs_info, FREE_SPACE_TREE))
1026 		return 0;
1027 
1028 	path = btrfs_alloc_path();
1029 	if (!path) {
1030 		ret = -ENOMEM;
1031 		goto out;
1032 	}
1033 
1034 	block_group = btrfs_lookup_block_group(trans->fs_info, start);
1035 	if (!block_group) {
1036 		ASSERT(0);
1037 		ret = -ENOENT;
1038 		goto out;
1039 	}
1040 
1041 	mutex_lock(&block_group->free_space_lock);
1042 	ret = __add_to_free_space_tree(trans, block_group, path, start, size);
1043 	mutex_unlock(&block_group->free_space_lock);
1044 
1045 	btrfs_put_block_group(block_group);
1046 out:
1047 	btrfs_free_path(path);
1048 	if (ret)
1049 		btrfs_abort_transaction(trans, ret);
1050 	return ret;
1051 }
1052 
1053 /*
1054  * Populate the free space tree by walking the extent tree. Operations on the
1055  * extent tree that happen as a result of writes to the free space tree will go
1056  * through the normal add/remove hooks.
1057  */
1058 static int populate_free_space_tree(struct btrfs_trans_handle *trans,
1059 				    struct btrfs_block_group *block_group)
1060 {
1061 	struct btrfs_root *extent_root;
1062 	struct btrfs_path *path, *path2;
1063 	struct btrfs_key key;
1064 	u64 start, end;
1065 	int ret;
1066 
1067 	path = btrfs_alloc_path();
1068 	if (!path)
1069 		return -ENOMEM;
1070 	path->reada = READA_FORWARD;
1071 
1072 	path2 = btrfs_alloc_path();
1073 	if (!path2) {
1074 		btrfs_free_path(path);
1075 		return -ENOMEM;
1076 	}
1077 
1078 	ret = add_new_free_space_info(trans, block_group, path2);
1079 	if (ret)
1080 		goto out;
1081 
1082 	mutex_lock(&block_group->free_space_lock);
1083 
1084 	/*
1085 	 * Iterate through all of the extent and metadata items in this block
1086 	 * group, adding the free space between them and the free space at the
1087 	 * end. Note that EXTENT_ITEM and METADATA_ITEM are less than
1088 	 * BLOCK_GROUP_ITEM, so an extent may precede the block group that it's
1089 	 * contained in.
1090 	 */
1091 	key.objectid = block_group->start;
1092 	key.type = BTRFS_EXTENT_ITEM_KEY;
1093 	key.offset = 0;
1094 
1095 	extent_root = btrfs_extent_root(trans->fs_info, key.objectid);
1096 	ret = btrfs_search_slot_for_read(extent_root, &key, path, 1, 0);
1097 	if (ret < 0)
1098 		goto out_locked;
1099 	ASSERT(ret == 0);
1100 
1101 	start = block_group->start;
1102 	end = block_group->start + block_group->length;
1103 	while (1) {
1104 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1105 
1106 		if (key.type == BTRFS_EXTENT_ITEM_KEY ||
1107 		    key.type == BTRFS_METADATA_ITEM_KEY) {
1108 			if (key.objectid >= end)
1109 				break;
1110 
1111 			if (start < key.objectid) {
1112 				ret = __add_to_free_space_tree(trans,
1113 							       block_group,
1114 							       path2, start,
1115 							       key.objectid -
1116 							       start);
1117 				if (ret)
1118 					goto out_locked;
1119 			}
1120 			start = key.objectid;
1121 			if (key.type == BTRFS_METADATA_ITEM_KEY)
1122 				start += trans->fs_info->nodesize;
1123 			else
1124 				start += key.offset;
1125 		} else if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
1126 			if (key.objectid != block_group->start)
1127 				break;
1128 		}
1129 
1130 		ret = btrfs_next_item(extent_root, path);
1131 		if (ret < 0)
1132 			goto out_locked;
1133 		if (ret)
1134 			break;
1135 	}
1136 	if (start < end) {
1137 		ret = __add_to_free_space_tree(trans, block_group, path2,
1138 					       start, end - start);
1139 		if (ret)
1140 			goto out_locked;
1141 	}
1142 
1143 	ret = 0;
1144 out_locked:
1145 	mutex_unlock(&block_group->free_space_lock);
1146 out:
1147 	btrfs_free_path(path2);
1148 	btrfs_free_path(path);
1149 	return ret;
1150 }
1151 
1152 int btrfs_create_free_space_tree(struct btrfs_fs_info *fs_info)
1153 {
1154 	struct btrfs_trans_handle *trans;
1155 	struct btrfs_root *tree_root = fs_info->tree_root;
1156 	struct btrfs_root *free_space_root;
1157 	struct btrfs_block_group *block_group;
1158 	struct rb_node *node;
1159 	int ret;
1160 
1161 	trans = btrfs_start_transaction(tree_root, 0);
1162 	if (IS_ERR(trans))
1163 		return PTR_ERR(trans);
1164 
1165 	set_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1166 	set_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags);
1167 	free_space_root = btrfs_create_tree(trans,
1168 					    BTRFS_FREE_SPACE_TREE_OBJECTID);
1169 	if (IS_ERR(free_space_root)) {
1170 		ret = PTR_ERR(free_space_root);
1171 		goto abort;
1172 	}
1173 	ret = btrfs_global_root_insert(free_space_root);
1174 	if (ret) {
1175 		btrfs_put_root(free_space_root);
1176 		goto abort;
1177 	}
1178 
1179 	node = rb_first(&fs_info->block_group_cache_tree);
1180 	while (node) {
1181 		block_group = rb_entry(node, struct btrfs_block_group,
1182 				       cache_node);
1183 		ret = populate_free_space_tree(trans, block_group);
1184 		if (ret)
1185 			goto abort;
1186 		node = rb_next(node);
1187 	}
1188 
1189 	btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE);
1190 	btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID);
1191 	clear_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1192 	ret = btrfs_commit_transaction(trans);
1193 
1194 	/*
1195 	 * Now that we've committed the transaction any reading of our commit
1196 	 * root will be safe, so we can cache from the free space tree now.
1197 	 */
1198 	clear_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags);
1199 	return ret;
1200 
1201 abort:
1202 	clear_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1203 	clear_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags);
1204 	btrfs_abort_transaction(trans, ret);
1205 	btrfs_end_transaction(trans);
1206 	return ret;
1207 }
1208 
1209 static int clear_free_space_tree(struct btrfs_trans_handle *trans,
1210 				 struct btrfs_root *root)
1211 {
1212 	struct btrfs_path *path;
1213 	struct btrfs_key key;
1214 	int nr;
1215 	int ret;
1216 
1217 	path = btrfs_alloc_path();
1218 	if (!path)
1219 		return -ENOMEM;
1220 
1221 	key.objectid = 0;
1222 	key.type = 0;
1223 	key.offset = 0;
1224 
1225 	while (1) {
1226 		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1227 		if (ret < 0)
1228 			goto out;
1229 
1230 		nr = btrfs_header_nritems(path->nodes[0]);
1231 		if (!nr)
1232 			break;
1233 
1234 		path->slots[0] = 0;
1235 		ret = btrfs_del_items(trans, root, path, 0, nr);
1236 		if (ret)
1237 			goto out;
1238 
1239 		btrfs_release_path(path);
1240 	}
1241 
1242 	ret = 0;
1243 out:
1244 	btrfs_free_path(path);
1245 	return ret;
1246 }
1247 
1248 int btrfs_clear_free_space_tree(struct btrfs_fs_info *fs_info)
1249 {
1250 	struct btrfs_trans_handle *trans;
1251 	struct btrfs_root *tree_root = fs_info->tree_root;
1252 	struct btrfs_key key = {
1253 		.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID,
1254 		.type = BTRFS_ROOT_ITEM_KEY,
1255 		.offset = 0,
1256 	};
1257 	struct btrfs_root *free_space_root = btrfs_global_root(fs_info, &key);
1258 	int ret;
1259 
1260 	trans = btrfs_start_transaction(tree_root, 0);
1261 	if (IS_ERR(trans))
1262 		return PTR_ERR(trans);
1263 
1264 	btrfs_clear_fs_compat_ro(fs_info, FREE_SPACE_TREE);
1265 	btrfs_clear_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID);
1266 
1267 	ret = clear_free_space_tree(trans, free_space_root);
1268 	if (ret)
1269 		goto abort;
1270 
1271 	ret = btrfs_del_root(trans, &free_space_root->root_key);
1272 	if (ret)
1273 		goto abort;
1274 
1275 	btrfs_global_root_delete(free_space_root);
1276 	list_del(&free_space_root->dirty_list);
1277 
1278 	btrfs_tree_lock(free_space_root->node);
1279 	btrfs_clean_tree_block(free_space_root->node);
1280 	btrfs_tree_unlock(free_space_root->node);
1281 	btrfs_free_tree_block(trans, btrfs_root_id(free_space_root),
1282 			      free_space_root->node, 0, 1);
1283 
1284 	btrfs_put_root(free_space_root);
1285 
1286 	return btrfs_commit_transaction(trans);
1287 
1288 abort:
1289 	btrfs_abort_transaction(trans, ret);
1290 	btrfs_end_transaction(trans);
1291 	return ret;
1292 }
1293 
1294 static int __add_block_group_free_space(struct btrfs_trans_handle *trans,
1295 					struct btrfs_block_group *block_group,
1296 					struct btrfs_path *path)
1297 {
1298 	int ret;
1299 
1300 	block_group->needs_free_space = 0;
1301 
1302 	ret = add_new_free_space_info(trans, block_group, path);
1303 	if (ret)
1304 		return ret;
1305 
1306 	return __add_to_free_space_tree(trans, block_group, path,
1307 					block_group->start,
1308 					block_group->length);
1309 }
1310 
1311 int add_block_group_free_space(struct btrfs_trans_handle *trans,
1312 			       struct btrfs_block_group *block_group)
1313 {
1314 	struct btrfs_fs_info *fs_info = trans->fs_info;
1315 	struct btrfs_path *path = NULL;
1316 	int ret = 0;
1317 
1318 	if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
1319 		return 0;
1320 
1321 	mutex_lock(&block_group->free_space_lock);
1322 	if (!block_group->needs_free_space)
1323 		goto out;
1324 
1325 	path = btrfs_alloc_path();
1326 	if (!path) {
1327 		ret = -ENOMEM;
1328 		goto out;
1329 	}
1330 
1331 	ret = __add_block_group_free_space(trans, block_group, path);
1332 
1333 out:
1334 	btrfs_free_path(path);
1335 	mutex_unlock(&block_group->free_space_lock);
1336 	if (ret)
1337 		btrfs_abort_transaction(trans, ret);
1338 	return ret;
1339 }
1340 
1341 int remove_block_group_free_space(struct btrfs_trans_handle *trans,
1342 				  struct btrfs_block_group *block_group)
1343 {
1344 	struct btrfs_root *root = btrfs_free_space_root(block_group);
1345 	struct btrfs_path *path;
1346 	struct btrfs_key key, found_key;
1347 	struct extent_buffer *leaf;
1348 	u64 start, end;
1349 	int done = 0, nr;
1350 	int ret;
1351 
1352 	if (!btrfs_fs_compat_ro(trans->fs_info, FREE_SPACE_TREE))
1353 		return 0;
1354 
1355 	if (block_group->needs_free_space) {
1356 		/* We never added this block group to the free space tree. */
1357 		return 0;
1358 	}
1359 
1360 	path = btrfs_alloc_path();
1361 	if (!path) {
1362 		ret = -ENOMEM;
1363 		goto out;
1364 	}
1365 
1366 	start = block_group->start;
1367 	end = block_group->start + block_group->length;
1368 
1369 	key.objectid = end - 1;
1370 	key.type = (u8)-1;
1371 	key.offset = (u64)-1;
1372 
1373 	while (!done) {
1374 		ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
1375 		if (ret)
1376 			goto out;
1377 
1378 		leaf = path->nodes[0];
1379 		nr = 0;
1380 		path->slots[0]++;
1381 		while (path->slots[0] > 0) {
1382 			btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
1383 
1384 			if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
1385 				ASSERT(found_key.objectid == block_group->start);
1386 				ASSERT(found_key.offset == block_group->length);
1387 				done = 1;
1388 				nr++;
1389 				path->slots[0]--;
1390 				break;
1391 			} else if (found_key.type == BTRFS_FREE_SPACE_EXTENT_KEY ||
1392 				   found_key.type == BTRFS_FREE_SPACE_BITMAP_KEY) {
1393 				ASSERT(found_key.objectid >= start);
1394 				ASSERT(found_key.objectid < end);
1395 				ASSERT(found_key.objectid + found_key.offset <= end);
1396 				nr++;
1397 				path->slots[0]--;
1398 			} else {
1399 				ASSERT(0);
1400 			}
1401 		}
1402 
1403 		ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
1404 		if (ret)
1405 			goto out;
1406 		btrfs_release_path(path);
1407 	}
1408 
1409 	ret = 0;
1410 out:
1411 	btrfs_free_path(path);
1412 	if (ret)
1413 		btrfs_abort_transaction(trans, ret);
1414 	return ret;
1415 }
1416 
1417 static int load_free_space_bitmaps(struct btrfs_caching_control *caching_ctl,
1418 				   struct btrfs_path *path,
1419 				   u32 expected_extent_count)
1420 {
1421 	struct btrfs_block_group *block_group;
1422 	struct btrfs_fs_info *fs_info;
1423 	struct btrfs_root *root;
1424 	struct btrfs_key key;
1425 	int prev_bit = 0, bit;
1426 	/* Initialize to silence GCC. */
1427 	u64 extent_start = 0;
1428 	u64 end, offset;
1429 	u64 total_found = 0;
1430 	u32 extent_count = 0;
1431 	int ret;
1432 
1433 	block_group = caching_ctl->block_group;
1434 	fs_info = block_group->fs_info;
1435 	root = btrfs_free_space_root(block_group);
1436 
1437 	end = block_group->start + block_group->length;
1438 
1439 	while (1) {
1440 		ret = btrfs_next_item(root, path);
1441 		if (ret < 0)
1442 			goto out;
1443 		if (ret)
1444 			break;
1445 
1446 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1447 
1448 		if (key.type == BTRFS_FREE_SPACE_INFO_KEY)
1449 			break;
1450 
1451 		ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
1452 		ASSERT(key.objectid < end && key.objectid + key.offset <= end);
1453 
1454 		caching_ctl->progress = key.objectid;
1455 
1456 		offset = key.objectid;
1457 		while (offset < key.objectid + key.offset) {
1458 			bit = free_space_test_bit(block_group, path, offset);
1459 			if (prev_bit == 0 && bit == 1) {
1460 				extent_start = offset;
1461 			} else if (prev_bit == 1 && bit == 0) {
1462 				total_found += add_new_free_space(block_group,
1463 								  extent_start,
1464 								  offset);
1465 				if (total_found > CACHING_CTL_WAKE_UP) {
1466 					total_found = 0;
1467 					wake_up(&caching_ctl->wait);
1468 				}
1469 				extent_count++;
1470 			}
1471 			prev_bit = bit;
1472 			offset += fs_info->sectorsize;
1473 		}
1474 	}
1475 	if (prev_bit == 1) {
1476 		total_found += add_new_free_space(block_group, extent_start,
1477 						  end);
1478 		extent_count++;
1479 	}
1480 
1481 	if (extent_count != expected_extent_count) {
1482 		btrfs_err(fs_info,
1483 			  "incorrect extent count for %llu; counted %u, expected %u",
1484 			  block_group->start, extent_count,
1485 			  expected_extent_count);
1486 		ASSERT(0);
1487 		ret = -EIO;
1488 		goto out;
1489 	}
1490 
1491 	caching_ctl->progress = (u64)-1;
1492 
1493 	ret = 0;
1494 out:
1495 	return ret;
1496 }
1497 
1498 static int load_free_space_extents(struct btrfs_caching_control *caching_ctl,
1499 				   struct btrfs_path *path,
1500 				   u32 expected_extent_count)
1501 {
1502 	struct btrfs_block_group *block_group;
1503 	struct btrfs_fs_info *fs_info;
1504 	struct btrfs_root *root;
1505 	struct btrfs_key key;
1506 	u64 end;
1507 	u64 total_found = 0;
1508 	u32 extent_count = 0;
1509 	int ret;
1510 
1511 	block_group = caching_ctl->block_group;
1512 	fs_info = block_group->fs_info;
1513 	root = btrfs_free_space_root(block_group);
1514 
1515 	end = block_group->start + block_group->length;
1516 
1517 	while (1) {
1518 		ret = btrfs_next_item(root, path);
1519 		if (ret < 0)
1520 			goto out;
1521 		if (ret)
1522 			break;
1523 
1524 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1525 
1526 		if (key.type == BTRFS_FREE_SPACE_INFO_KEY)
1527 			break;
1528 
1529 		ASSERT(key.type == BTRFS_FREE_SPACE_EXTENT_KEY);
1530 		ASSERT(key.objectid < end && key.objectid + key.offset <= end);
1531 
1532 		caching_ctl->progress = key.objectid;
1533 
1534 		total_found += add_new_free_space(block_group, key.objectid,
1535 						  key.objectid + key.offset);
1536 		if (total_found > CACHING_CTL_WAKE_UP) {
1537 			total_found = 0;
1538 			wake_up(&caching_ctl->wait);
1539 		}
1540 		extent_count++;
1541 	}
1542 
1543 	if (extent_count != expected_extent_count) {
1544 		btrfs_err(fs_info,
1545 			  "incorrect extent count for %llu; counted %u, expected %u",
1546 			  block_group->start, extent_count,
1547 			  expected_extent_count);
1548 		ASSERT(0);
1549 		ret = -EIO;
1550 		goto out;
1551 	}
1552 
1553 	caching_ctl->progress = (u64)-1;
1554 
1555 	ret = 0;
1556 out:
1557 	return ret;
1558 }
1559 
1560 int load_free_space_tree(struct btrfs_caching_control *caching_ctl)
1561 {
1562 	struct btrfs_block_group *block_group;
1563 	struct btrfs_free_space_info *info;
1564 	struct btrfs_path *path;
1565 	u32 extent_count, flags;
1566 	int ret;
1567 
1568 	block_group = caching_ctl->block_group;
1569 
1570 	path = btrfs_alloc_path();
1571 	if (!path)
1572 		return -ENOMEM;
1573 
1574 	/*
1575 	 * Just like caching_thread() doesn't want to deadlock on the extent
1576 	 * tree, we don't want to deadlock on the free space tree.
1577 	 */
1578 	path->skip_locking = 1;
1579 	path->search_commit_root = 1;
1580 	path->reada = READA_FORWARD;
1581 
1582 	info = search_free_space_info(NULL, block_group, path, 0);
1583 	if (IS_ERR(info)) {
1584 		ret = PTR_ERR(info);
1585 		goto out;
1586 	}
1587 	extent_count = btrfs_free_space_extent_count(path->nodes[0], info);
1588 	flags = btrfs_free_space_flags(path->nodes[0], info);
1589 
1590 	/*
1591 	 * We left path pointing to the free space info item, so now
1592 	 * load_free_space_foo can just iterate through the free space tree from
1593 	 * there.
1594 	 */
1595 	if (flags & BTRFS_FREE_SPACE_USING_BITMAPS)
1596 		ret = load_free_space_bitmaps(caching_ctl, path, extent_count);
1597 	else
1598 		ret = load_free_space_extents(caching_ctl, path, extent_count);
1599 
1600 out:
1601 	btrfs_free_path(path);
1602 	return ret;
1603 }
1604