xref: /openbmc/linux/fs/btrfs/free-space-tree.c (revision c5c87812)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2015 Facebook.  All rights reserved.
4  */
5 
6 #include <linux/kernel.h>
7 #include <linux/sched/mm.h>
8 #include "ctree.h"
9 #include "disk-io.h"
10 #include "locking.h"
11 #include "free-space-tree.h"
12 #include "transaction.h"
13 #include "block-group.h"
14 
15 static int __add_block_group_free_space(struct btrfs_trans_handle *trans,
16 					struct btrfs_block_group *block_group,
17 					struct btrfs_path *path);
18 
19 void set_free_space_tree_thresholds(struct btrfs_block_group *cache)
20 {
21 	u32 bitmap_range;
22 	size_t bitmap_size;
23 	u64 num_bitmaps, total_bitmap_size;
24 
25 	if (WARN_ON(cache->length == 0))
26 		btrfs_warn(cache->fs_info, "block group %llu length is zero",
27 			   cache->start);
28 
29 	/*
30 	 * We convert to bitmaps when the disk space required for using extents
31 	 * exceeds that required for using bitmaps.
32 	 */
33 	bitmap_range = cache->fs_info->sectorsize * BTRFS_FREE_SPACE_BITMAP_BITS;
34 	num_bitmaps = div_u64(cache->length + bitmap_range - 1, bitmap_range);
35 	bitmap_size = sizeof(struct btrfs_item) + BTRFS_FREE_SPACE_BITMAP_SIZE;
36 	total_bitmap_size = num_bitmaps * bitmap_size;
37 	cache->bitmap_high_thresh = div_u64(total_bitmap_size,
38 					    sizeof(struct btrfs_item));
39 
40 	/*
41 	 * We allow for a small buffer between the high threshold and low
42 	 * threshold to avoid thrashing back and forth between the two formats.
43 	 */
44 	if (cache->bitmap_high_thresh > 100)
45 		cache->bitmap_low_thresh = cache->bitmap_high_thresh - 100;
46 	else
47 		cache->bitmap_low_thresh = 0;
48 }
49 
50 static int add_new_free_space_info(struct btrfs_trans_handle *trans,
51 				   struct btrfs_block_group *block_group,
52 				   struct btrfs_path *path)
53 {
54 	struct btrfs_root *root = trans->fs_info->free_space_root;
55 	struct btrfs_free_space_info *info;
56 	struct btrfs_key key;
57 	struct extent_buffer *leaf;
58 	int ret;
59 
60 	key.objectid = block_group->start;
61 	key.type = BTRFS_FREE_SPACE_INFO_KEY;
62 	key.offset = block_group->length;
63 
64 	ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*info));
65 	if (ret)
66 		goto out;
67 
68 	leaf = path->nodes[0];
69 	info = btrfs_item_ptr(leaf, path->slots[0],
70 			      struct btrfs_free_space_info);
71 	btrfs_set_free_space_extent_count(leaf, info, 0);
72 	btrfs_set_free_space_flags(leaf, info, 0);
73 	btrfs_mark_buffer_dirty(leaf);
74 
75 	ret = 0;
76 out:
77 	btrfs_release_path(path);
78 	return ret;
79 }
80 
81 EXPORT_FOR_TESTS
82 struct btrfs_free_space_info *search_free_space_info(
83 		struct btrfs_trans_handle *trans,
84 		struct btrfs_block_group *block_group,
85 		struct btrfs_path *path, int cow)
86 {
87 	struct btrfs_fs_info *fs_info = block_group->fs_info;
88 	struct btrfs_root *root = fs_info->free_space_root;
89 	struct btrfs_key key;
90 	int ret;
91 
92 	key.objectid = block_group->start;
93 	key.type = BTRFS_FREE_SPACE_INFO_KEY;
94 	key.offset = block_group->length;
95 
96 	ret = btrfs_search_slot(trans, root, &key, path, 0, cow);
97 	if (ret < 0)
98 		return ERR_PTR(ret);
99 	if (ret != 0) {
100 		btrfs_warn(fs_info, "missing free space info for %llu",
101 			   block_group->start);
102 		ASSERT(0);
103 		return ERR_PTR(-ENOENT);
104 	}
105 
106 	return btrfs_item_ptr(path->nodes[0], path->slots[0],
107 			      struct btrfs_free_space_info);
108 }
109 
110 /*
111  * btrfs_search_slot() but we're looking for the greatest key less than the
112  * passed key.
113  */
114 static int btrfs_search_prev_slot(struct btrfs_trans_handle *trans,
115 				  struct btrfs_root *root,
116 				  struct btrfs_key *key, struct btrfs_path *p,
117 				  int ins_len, int cow)
118 {
119 	int ret;
120 
121 	ret = btrfs_search_slot(trans, root, key, p, ins_len, cow);
122 	if (ret < 0)
123 		return ret;
124 
125 	if (ret == 0) {
126 		ASSERT(0);
127 		return -EIO;
128 	}
129 
130 	if (p->slots[0] == 0) {
131 		ASSERT(0);
132 		return -EIO;
133 	}
134 	p->slots[0]--;
135 
136 	return 0;
137 }
138 
139 static inline u32 free_space_bitmap_size(u64 size, u32 sectorsize)
140 {
141 	return DIV_ROUND_UP((u32)div_u64(size, sectorsize), BITS_PER_BYTE);
142 }
143 
144 static unsigned long *alloc_bitmap(u32 bitmap_size)
145 {
146 	unsigned long *ret;
147 	unsigned int nofs_flag;
148 	u32 bitmap_rounded_size = round_up(bitmap_size, sizeof(unsigned long));
149 
150 	/*
151 	 * GFP_NOFS doesn't work with kvmalloc(), but we really can't recurse
152 	 * into the filesystem as the free space bitmap can be modified in the
153 	 * critical section of a transaction commit.
154 	 *
155 	 * TODO: push the memalloc_nofs_{save,restore}() to the caller where we
156 	 * know that recursion is unsafe.
157 	 */
158 	nofs_flag = memalloc_nofs_save();
159 	ret = kvzalloc(bitmap_rounded_size, GFP_KERNEL);
160 	memalloc_nofs_restore(nofs_flag);
161 	return ret;
162 }
163 
164 static void le_bitmap_set(unsigned long *map, unsigned int start, int len)
165 {
166 	u8 *p = ((u8 *)map) + BIT_BYTE(start);
167 	const unsigned int size = start + len;
168 	int bits_to_set = BITS_PER_BYTE - (start % BITS_PER_BYTE);
169 	u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(start);
170 
171 	while (len - bits_to_set >= 0) {
172 		*p |= mask_to_set;
173 		len -= bits_to_set;
174 		bits_to_set = BITS_PER_BYTE;
175 		mask_to_set = ~0;
176 		p++;
177 	}
178 	if (len) {
179 		mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
180 		*p |= mask_to_set;
181 	}
182 }
183 
184 EXPORT_FOR_TESTS
185 int convert_free_space_to_bitmaps(struct btrfs_trans_handle *trans,
186 				  struct btrfs_block_group *block_group,
187 				  struct btrfs_path *path)
188 {
189 	struct btrfs_fs_info *fs_info = trans->fs_info;
190 	struct btrfs_root *root = fs_info->free_space_root;
191 	struct btrfs_free_space_info *info;
192 	struct btrfs_key key, found_key;
193 	struct extent_buffer *leaf;
194 	unsigned long *bitmap;
195 	char *bitmap_cursor;
196 	u64 start, end;
197 	u64 bitmap_range, i;
198 	u32 bitmap_size, flags, expected_extent_count;
199 	u32 extent_count = 0;
200 	int done = 0, nr;
201 	int ret;
202 
203 	bitmap_size = free_space_bitmap_size(block_group->length,
204 					     fs_info->sectorsize);
205 	bitmap = alloc_bitmap(bitmap_size);
206 	if (!bitmap) {
207 		ret = -ENOMEM;
208 		goto out;
209 	}
210 
211 	start = block_group->start;
212 	end = block_group->start + block_group->length;
213 
214 	key.objectid = end - 1;
215 	key.type = (u8)-1;
216 	key.offset = (u64)-1;
217 
218 	while (!done) {
219 		ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
220 		if (ret)
221 			goto out;
222 
223 		leaf = path->nodes[0];
224 		nr = 0;
225 		path->slots[0]++;
226 		while (path->slots[0] > 0) {
227 			btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
228 
229 			if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
230 				ASSERT(found_key.objectid == block_group->start);
231 				ASSERT(found_key.offset == block_group->length);
232 				done = 1;
233 				break;
234 			} else if (found_key.type == BTRFS_FREE_SPACE_EXTENT_KEY) {
235 				u64 first, last;
236 
237 				ASSERT(found_key.objectid >= start);
238 				ASSERT(found_key.objectid < end);
239 				ASSERT(found_key.objectid + found_key.offset <= end);
240 
241 				first = div_u64(found_key.objectid - start,
242 						fs_info->sectorsize);
243 				last = div_u64(found_key.objectid + found_key.offset - start,
244 					       fs_info->sectorsize);
245 				le_bitmap_set(bitmap, first, last - first);
246 
247 				extent_count++;
248 				nr++;
249 				path->slots[0]--;
250 			} else {
251 				ASSERT(0);
252 			}
253 		}
254 
255 		ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
256 		if (ret)
257 			goto out;
258 		btrfs_release_path(path);
259 	}
260 
261 	info = search_free_space_info(trans, block_group, path, 1);
262 	if (IS_ERR(info)) {
263 		ret = PTR_ERR(info);
264 		goto out;
265 	}
266 	leaf = path->nodes[0];
267 	flags = btrfs_free_space_flags(leaf, info);
268 	flags |= BTRFS_FREE_SPACE_USING_BITMAPS;
269 	btrfs_set_free_space_flags(leaf, info, flags);
270 	expected_extent_count = btrfs_free_space_extent_count(leaf, info);
271 	btrfs_mark_buffer_dirty(leaf);
272 	btrfs_release_path(path);
273 
274 	if (extent_count != expected_extent_count) {
275 		btrfs_err(fs_info,
276 			  "incorrect extent count for %llu; counted %u, expected %u",
277 			  block_group->start, extent_count,
278 			  expected_extent_count);
279 		ASSERT(0);
280 		ret = -EIO;
281 		goto out;
282 	}
283 
284 	bitmap_cursor = (char *)bitmap;
285 	bitmap_range = fs_info->sectorsize * BTRFS_FREE_SPACE_BITMAP_BITS;
286 	i = start;
287 	while (i < end) {
288 		unsigned long ptr;
289 		u64 extent_size;
290 		u32 data_size;
291 
292 		extent_size = min(end - i, bitmap_range);
293 		data_size = free_space_bitmap_size(extent_size,
294 						   fs_info->sectorsize);
295 
296 		key.objectid = i;
297 		key.type = BTRFS_FREE_SPACE_BITMAP_KEY;
298 		key.offset = extent_size;
299 
300 		ret = btrfs_insert_empty_item(trans, root, path, &key,
301 					      data_size);
302 		if (ret)
303 			goto out;
304 
305 		leaf = path->nodes[0];
306 		ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
307 		write_extent_buffer(leaf, bitmap_cursor, ptr,
308 				    data_size);
309 		btrfs_mark_buffer_dirty(leaf);
310 		btrfs_release_path(path);
311 
312 		i += extent_size;
313 		bitmap_cursor += data_size;
314 	}
315 
316 	ret = 0;
317 out:
318 	kvfree(bitmap);
319 	if (ret)
320 		btrfs_abort_transaction(trans, ret);
321 	return ret;
322 }
323 
324 EXPORT_FOR_TESTS
325 int convert_free_space_to_extents(struct btrfs_trans_handle *trans,
326 				  struct btrfs_block_group *block_group,
327 				  struct btrfs_path *path)
328 {
329 	struct btrfs_fs_info *fs_info = trans->fs_info;
330 	struct btrfs_root *root = fs_info->free_space_root;
331 	struct btrfs_free_space_info *info;
332 	struct btrfs_key key, found_key;
333 	struct extent_buffer *leaf;
334 	unsigned long *bitmap;
335 	u64 start, end;
336 	u32 bitmap_size, flags, expected_extent_count;
337 	unsigned long nrbits, start_bit, end_bit;
338 	u32 extent_count = 0;
339 	int done = 0, nr;
340 	int ret;
341 
342 	bitmap_size = free_space_bitmap_size(block_group->length,
343 					     fs_info->sectorsize);
344 	bitmap = alloc_bitmap(bitmap_size);
345 	if (!bitmap) {
346 		ret = -ENOMEM;
347 		goto out;
348 	}
349 
350 	start = block_group->start;
351 	end = block_group->start + block_group->length;
352 
353 	key.objectid = end - 1;
354 	key.type = (u8)-1;
355 	key.offset = (u64)-1;
356 
357 	while (!done) {
358 		ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
359 		if (ret)
360 			goto out;
361 
362 		leaf = path->nodes[0];
363 		nr = 0;
364 		path->slots[0]++;
365 		while (path->slots[0] > 0) {
366 			btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
367 
368 			if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
369 				ASSERT(found_key.objectid == block_group->start);
370 				ASSERT(found_key.offset == block_group->length);
371 				done = 1;
372 				break;
373 			} else if (found_key.type == BTRFS_FREE_SPACE_BITMAP_KEY) {
374 				unsigned long ptr;
375 				char *bitmap_cursor;
376 				u32 bitmap_pos, data_size;
377 
378 				ASSERT(found_key.objectid >= start);
379 				ASSERT(found_key.objectid < end);
380 				ASSERT(found_key.objectid + found_key.offset <= end);
381 
382 				bitmap_pos = div_u64(found_key.objectid - start,
383 						     fs_info->sectorsize *
384 						     BITS_PER_BYTE);
385 				bitmap_cursor = ((char *)bitmap) + bitmap_pos;
386 				data_size = free_space_bitmap_size(found_key.offset,
387 								   fs_info->sectorsize);
388 
389 				ptr = btrfs_item_ptr_offset(leaf, path->slots[0] - 1);
390 				read_extent_buffer(leaf, bitmap_cursor, ptr,
391 						   data_size);
392 
393 				nr++;
394 				path->slots[0]--;
395 			} else {
396 				ASSERT(0);
397 			}
398 		}
399 
400 		ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
401 		if (ret)
402 			goto out;
403 		btrfs_release_path(path);
404 	}
405 
406 	info = search_free_space_info(trans, block_group, path, 1);
407 	if (IS_ERR(info)) {
408 		ret = PTR_ERR(info);
409 		goto out;
410 	}
411 	leaf = path->nodes[0];
412 	flags = btrfs_free_space_flags(leaf, info);
413 	flags &= ~BTRFS_FREE_SPACE_USING_BITMAPS;
414 	btrfs_set_free_space_flags(leaf, info, flags);
415 	expected_extent_count = btrfs_free_space_extent_count(leaf, info);
416 	btrfs_mark_buffer_dirty(leaf);
417 	btrfs_release_path(path);
418 
419 	nrbits = div_u64(block_group->length, block_group->fs_info->sectorsize);
420 	start_bit = find_next_bit_le(bitmap, nrbits, 0);
421 
422 	while (start_bit < nrbits) {
423 		end_bit = find_next_zero_bit_le(bitmap, nrbits, start_bit);
424 		ASSERT(start_bit < end_bit);
425 
426 		key.objectid = start + start_bit * block_group->fs_info->sectorsize;
427 		key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
428 		key.offset = (end_bit - start_bit) * block_group->fs_info->sectorsize;
429 
430 		ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
431 		if (ret)
432 			goto out;
433 		btrfs_release_path(path);
434 
435 		extent_count++;
436 
437 		start_bit = find_next_bit_le(bitmap, nrbits, end_bit);
438 	}
439 
440 	if (extent_count != expected_extent_count) {
441 		btrfs_err(fs_info,
442 			  "incorrect extent count for %llu; counted %u, expected %u",
443 			  block_group->start, extent_count,
444 			  expected_extent_count);
445 		ASSERT(0);
446 		ret = -EIO;
447 		goto out;
448 	}
449 
450 	ret = 0;
451 out:
452 	kvfree(bitmap);
453 	if (ret)
454 		btrfs_abort_transaction(trans, ret);
455 	return ret;
456 }
457 
458 static int update_free_space_extent_count(struct btrfs_trans_handle *trans,
459 					  struct btrfs_block_group *block_group,
460 					  struct btrfs_path *path,
461 					  int new_extents)
462 {
463 	struct btrfs_free_space_info *info;
464 	u32 flags;
465 	u32 extent_count;
466 	int ret = 0;
467 
468 	if (new_extents == 0)
469 		return 0;
470 
471 	info = search_free_space_info(trans, block_group, path, 1);
472 	if (IS_ERR(info)) {
473 		ret = PTR_ERR(info);
474 		goto out;
475 	}
476 	flags = btrfs_free_space_flags(path->nodes[0], info);
477 	extent_count = btrfs_free_space_extent_count(path->nodes[0], info);
478 
479 	extent_count += new_extents;
480 	btrfs_set_free_space_extent_count(path->nodes[0], info, extent_count);
481 	btrfs_mark_buffer_dirty(path->nodes[0]);
482 	btrfs_release_path(path);
483 
484 	if (!(flags & BTRFS_FREE_SPACE_USING_BITMAPS) &&
485 	    extent_count > block_group->bitmap_high_thresh) {
486 		ret = convert_free_space_to_bitmaps(trans, block_group, path);
487 	} else if ((flags & BTRFS_FREE_SPACE_USING_BITMAPS) &&
488 		   extent_count < block_group->bitmap_low_thresh) {
489 		ret = convert_free_space_to_extents(trans, block_group, path);
490 	}
491 
492 out:
493 	return ret;
494 }
495 
496 EXPORT_FOR_TESTS
497 int free_space_test_bit(struct btrfs_block_group *block_group,
498 			struct btrfs_path *path, u64 offset)
499 {
500 	struct extent_buffer *leaf;
501 	struct btrfs_key key;
502 	u64 found_start, found_end;
503 	unsigned long ptr, i;
504 
505 	leaf = path->nodes[0];
506 	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
507 	ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
508 
509 	found_start = key.objectid;
510 	found_end = key.objectid + key.offset;
511 	ASSERT(offset >= found_start && offset < found_end);
512 
513 	ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
514 	i = div_u64(offset - found_start,
515 		    block_group->fs_info->sectorsize);
516 	return !!extent_buffer_test_bit(leaf, ptr, i);
517 }
518 
519 static void free_space_set_bits(struct btrfs_block_group *block_group,
520 				struct btrfs_path *path, u64 *start, u64 *size,
521 				int bit)
522 {
523 	struct btrfs_fs_info *fs_info = block_group->fs_info;
524 	struct extent_buffer *leaf;
525 	struct btrfs_key key;
526 	u64 end = *start + *size;
527 	u64 found_start, found_end;
528 	unsigned long ptr, first, last;
529 
530 	leaf = path->nodes[0];
531 	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
532 	ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
533 
534 	found_start = key.objectid;
535 	found_end = key.objectid + key.offset;
536 	ASSERT(*start >= found_start && *start < found_end);
537 	ASSERT(end > found_start);
538 
539 	if (end > found_end)
540 		end = found_end;
541 
542 	ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
543 	first = div_u64(*start - found_start, fs_info->sectorsize);
544 	last = div_u64(end - found_start, fs_info->sectorsize);
545 	if (bit)
546 		extent_buffer_bitmap_set(leaf, ptr, first, last - first);
547 	else
548 		extent_buffer_bitmap_clear(leaf, ptr, first, last - first);
549 	btrfs_mark_buffer_dirty(leaf);
550 
551 	*size -= end - *start;
552 	*start = end;
553 }
554 
555 /*
556  * We can't use btrfs_next_item() in modify_free_space_bitmap() because
557  * btrfs_next_leaf() doesn't get the path for writing. We can forgo the fancy
558  * tree walking in btrfs_next_leaf() anyways because we know exactly what we're
559  * looking for.
560  */
561 static int free_space_next_bitmap(struct btrfs_trans_handle *trans,
562 				  struct btrfs_root *root, struct btrfs_path *p)
563 {
564 	struct btrfs_key key;
565 
566 	if (p->slots[0] + 1 < btrfs_header_nritems(p->nodes[0])) {
567 		p->slots[0]++;
568 		return 0;
569 	}
570 
571 	btrfs_item_key_to_cpu(p->nodes[0], &key, p->slots[0]);
572 	btrfs_release_path(p);
573 
574 	key.objectid += key.offset;
575 	key.type = (u8)-1;
576 	key.offset = (u64)-1;
577 
578 	return btrfs_search_prev_slot(trans, root, &key, p, 0, 1);
579 }
580 
581 /*
582  * If remove is 1, then we are removing free space, thus clearing bits in the
583  * bitmap. If remove is 0, then we are adding free space, thus setting bits in
584  * the bitmap.
585  */
586 static int modify_free_space_bitmap(struct btrfs_trans_handle *trans,
587 				    struct btrfs_block_group *block_group,
588 				    struct btrfs_path *path,
589 				    u64 start, u64 size, int remove)
590 {
591 	struct btrfs_root *root = block_group->fs_info->free_space_root;
592 	struct btrfs_key key;
593 	u64 end = start + size;
594 	u64 cur_start, cur_size;
595 	int prev_bit, next_bit;
596 	int new_extents;
597 	int ret;
598 
599 	/*
600 	 * Read the bit for the block immediately before the extent of space if
601 	 * that block is within the block group.
602 	 */
603 	if (start > block_group->start) {
604 		u64 prev_block = start - block_group->fs_info->sectorsize;
605 
606 		key.objectid = prev_block;
607 		key.type = (u8)-1;
608 		key.offset = (u64)-1;
609 
610 		ret = btrfs_search_prev_slot(trans, root, &key, path, 0, 1);
611 		if (ret)
612 			goto out;
613 
614 		prev_bit = free_space_test_bit(block_group, path, prev_block);
615 
616 		/* The previous block may have been in the previous bitmap. */
617 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
618 		if (start >= key.objectid + key.offset) {
619 			ret = free_space_next_bitmap(trans, root, path);
620 			if (ret)
621 				goto out;
622 		}
623 	} else {
624 		key.objectid = start;
625 		key.type = (u8)-1;
626 		key.offset = (u64)-1;
627 
628 		ret = btrfs_search_prev_slot(trans, root, &key, path, 0, 1);
629 		if (ret)
630 			goto out;
631 
632 		prev_bit = -1;
633 	}
634 
635 	/*
636 	 * Iterate over all of the bitmaps overlapped by the extent of space,
637 	 * clearing/setting bits as required.
638 	 */
639 	cur_start = start;
640 	cur_size = size;
641 	while (1) {
642 		free_space_set_bits(block_group, path, &cur_start, &cur_size,
643 				    !remove);
644 		if (cur_size == 0)
645 			break;
646 		ret = free_space_next_bitmap(trans, root, path);
647 		if (ret)
648 			goto out;
649 	}
650 
651 	/*
652 	 * Read the bit for the block immediately after the extent of space if
653 	 * that block is within the block group.
654 	 */
655 	if (end < block_group->start + block_group->length) {
656 		/* The next block may be in the next bitmap. */
657 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
658 		if (end >= key.objectid + key.offset) {
659 			ret = free_space_next_bitmap(trans, root, path);
660 			if (ret)
661 				goto out;
662 		}
663 
664 		next_bit = free_space_test_bit(block_group, path, end);
665 	} else {
666 		next_bit = -1;
667 	}
668 
669 	if (remove) {
670 		new_extents = -1;
671 		if (prev_bit == 1) {
672 			/* Leftover on the left. */
673 			new_extents++;
674 		}
675 		if (next_bit == 1) {
676 			/* Leftover on the right. */
677 			new_extents++;
678 		}
679 	} else {
680 		new_extents = 1;
681 		if (prev_bit == 1) {
682 			/* Merging with neighbor on the left. */
683 			new_extents--;
684 		}
685 		if (next_bit == 1) {
686 			/* Merging with neighbor on the right. */
687 			new_extents--;
688 		}
689 	}
690 
691 	btrfs_release_path(path);
692 	ret = update_free_space_extent_count(trans, block_group, path,
693 					     new_extents);
694 
695 out:
696 	return ret;
697 }
698 
699 static int remove_free_space_extent(struct btrfs_trans_handle *trans,
700 				    struct btrfs_block_group *block_group,
701 				    struct btrfs_path *path,
702 				    u64 start, u64 size)
703 {
704 	struct btrfs_root *root = trans->fs_info->free_space_root;
705 	struct btrfs_key key;
706 	u64 found_start, found_end;
707 	u64 end = start + size;
708 	int new_extents = -1;
709 	int ret;
710 
711 	key.objectid = start;
712 	key.type = (u8)-1;
713 	key.offset = (u64)-1;
714 
715 	ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
716 	if (ret)
717 		goto out;
718 
719 	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
720 
721 	ASSERT(key.type == BTRFS_FREE_SPACE_EXTENT_KEY);
722 
723 	found_start = key.objectid;
724 	found_end = key.objectid + key.offset;
725 	ASSERT(start >= found_start && end <= found_end);
726 
727 	/*
728 	 * Okay, now that we've found the free space extent which contains the
729 	 * free space that we are removing, there are four cases:
730 	 *
731 	 * 1. We're using the whole extent: delete the key we found and
732 	 * decrement the free space extent count.
733 	 * 2. We are using part of the extent starting at the beginning: delete
734 	 * the key we found and insert a new key representing the leftover at
735 	 * the end. There is no net change in the number of extents.
736 	 * 3. We are using part of the extent ending at the end: delete the key
737 	 * we found and insert a new key representing the leftover at the
738 	 * beginning. There is no net change in the number of extents.
739 	 * 4. We are using part of the extent in the middle: delete the key we
740 	 * found and insert two new keys representing the leftovers on each
741 	 * side. Where we used to have one extent, we now have two, so increment
742 	 * the extent count. We may need to convert the block group to bitmaps
743 	 * as a result.
744 	 */
745 
746 	/* Delete the existing key (cases 1-4). */
747 	ret = btrfs_del_item(trans, root, path);
748 	if (ret)
749 		goto out;
750 
751 	/* Add a key for leftovers at the beginning (cases 3 and 4). */
752 	if (start > found_start) {
753 		key.objectid = found_start;
754 		key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
755 		key.offset = start - found_start;
756 
757 		btrfs_release_path(path);
758 		ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
759 		if (ret)
760 			goto out;
761 		new_extents++;
762 	}
763 
764 	/* Add a key for leftovers at the end (cases 2 and 4). */
765 	if (end < found_end) {
766 		key.objectid = end;
767 		key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
768 		key.offset = found_end - end;
769 
770 		btrfs_release_path(path);
771 		ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
772 		if (ret)
773 			goto out;
774 		new_extents++;
775 	}
776 
777 	btrfs_release_path(path);
778 	ret = update_free_space_extent_count(trans, block_group, path,
779 					     new_extents);
780 
781 out:
782 	return ret;
783 }
784 
785 EXPORT_FOR_TESTS
786 int __remove_from_free_space_tree(struct btrfs_trans_handle *trans,
787 				  struct btrfs_block_group *block_group,
788 				  struct btrfs_path *path, u64 start, u64 size)
789 {
790 	struct btrfs_free_space_info *info;
791 	u32 flags;
792 	int ret;
793 
794 	if (block_group->needs_free_space) {
795 		ret = __add_block_group_free_space(trans, block_group, path);
796 		if (ret)
797 			return ret;
798 	}
799 
800 	info = search_free_space_info(NULL, block_group, path, 0);
801 	if (IS_ERR(info))
802 		return PTR_ERR(info);
803 	flags = btrfs_free_space_flags(path->nodes[0], info);
804 	btrfs_release_path(path);
805 
806 	if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) {
807 		return modify_free_space_bitmap(trans, block_group, path,
808 						start, size, 1);
809 	} else {
810 		return remove_free_space_extent(trans, block_group, path,
811 						start, size);
812 	}
813 }
814 
815 int remove_from_free_space_tree(struct btrfs_trans_handle *trans,
816 				u64 start, u64 size)
817 {
818 	struct btrfs_block_group *block_group;
819 	struct btrfs_path *path;
820 	int ret;
821 
822 	if (!btrfs_fs_compat_ro(trans->fs_info, FREE_SPACE_TREE))
823 		return 0;
824 
825 	path = btrfs_alloc_path();
826 	if (!path) {
827 		ret = -ENOMEM;
828 		goto out;
829 	}
830 
831 	block_group = btrfs_lookup_block_group(trans->fs_info, start);
832 	if (!block_group) {
833 		ASSERT(0);
834 		ret = -ENOENT;
835 		goto out;
836 	}
837 
838 	mutex_lock(&block_group->free_space_lock);
839 	ret = __remove_from_free_space_tree(trans, block_group, path, start,
840 					    size);
841 	mutex_unlock(&block_group->free_space_lock);
842 
843 	btrfs_put_block_group(block_group);
844 out:
845 	btrfs_free_path(path);
846 	if (ret)
847 		btrfs_abort_transaction(trans, ret);
848 	return ret;
849 }
850 
851 static int add_free_space_extent(struct btrfs_trans_handle *trans,
852 				 struct btrfs_block_group *block_group,
853 				 struct btrfs_path *path,
854 				 u64 start, u64 size)
855 {
856 	struct btrfs_root *root = trans->fs_info->free_space_root;
857 	struct btrfs_key key, new_key;
858 	u64 found_start, found_end;
859 	u64 end = start + size;
860 	int new_extents = 1;
861 	int ret;
862 
863 	/*
864 	 * We are adding a new extent of free space, but we need to merge
865 	 * extents. There are four cases here:
866 	 *
867 	 * 1. The new extent does not have any immediate neighbors to merge
868 	 * with: add the new key and increment the free space extent count. We
869 	 * may need to convert the block group to bitmaps as a result.
870 	 * 2. The new extent has an immediate neighbor before it: remove the
871 	 * previous key and insert a new key combining both of them. There is no
872 	 * net change in the number of extents.
873 	 * 3. The new extent has an immediate neighbor after it: remove the next
874 	 * key and insert a new key combining both of them. There is no net
875 	 * change in the number of extents.
876 	 * 4. The new extent has immediate neighbors on both sides: remove both
877 	 * of the keys and insert a new key combining all of them. Where we used
878 	 * to have two extents, we now have one, so decrement the extent count.
879 	 */
880 
881 	new_key.objectid = start;
882 	new_key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
883 	new_key.offset = size;
884 
885 	/* Search for a neighbor on the left. */
886 	if (start == block_group->start)
887 		goto right;
888 	key.objectid = start - 1;
889 	key.type = (u8)-1;
890 	key.offset = (u64)-1;
891 
892 	ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
893 	if (ret)
894 		goto out;
895 
896 	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
897 
898 	if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) {
899 		ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY);
900 		btrfs_release_path(path);
901 		goto right;
902 	}
903 
904 	found_start = key.objectid;
905 	found_end = key.objectid + key.offset;
906 	ASSERT(found_start >= block_group->start &&
907 	       found_end > block_group->start);
908 	ASSERT(found_start < start && found_end <= start);
909 
910 	/*
911 	 * Delete the neighbor on the left and absorb it into the new key (cases
912 	 * 2 and 4).
913 	 */
914 	if (found_end == start) {
915 		ret = btrfs_del_item(trans, root, path);
916 		if (ret)
917 			goto out;
918 		new_key.objectid = found_start;
919 		new_key.offset += key.offset;
920 		new_extents--;
921 	}
922 	btrfs_release_path(path);
923 
924 right:
925 	/* Search for a neighbor on the right. */
926 	if (end == block_group->start + block_group->length)
927 		goto insert;
928 	key.objectid = end;
929 	key.type = (u8)-1;
930 	key.offset = (u64)-1;
931 
932 	ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
933 	if (ret)
934 		goto out;
935 
936 	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
937 
938 	if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) {
939 		ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY);
940 		btrfs_release_path(path);
941 		goto insert;
942 	}
943 
944 	found_start = key.objectid;
945 	found_end = key.objectid + key.offset;
946 	ASSERT(found_start >= block_group->start &&
947 	       found_end > block_group->start);
948 	ASSERT((found_start < start && found_end <= start) ||
949 	       (found_start >= end && found_end > end));
950 
951 	/*
952 	 * Delete the neighbor on the right and absorb it into the new key
953 	 * (cases 3 and 4).
954 	 */
955 	if (found_start == end) {
956 		ret = btrfs_del_item(trans, root, path);
957 		if (ret)
958 			goto out;
959 		new_key.offset += key.offset;
960 		new_extents--;
961 	}
962 	btrfs_release_path(path);
963 
964 insert:
965 	/* Insert the new key (cases 1-4). */
966 	ret = btrfs_insert_empty_item(trans, root, path, &new_key, 0);
967 	if (ret)
968 		goto out;
969 
970 	btrfs_release_path(path);
971 	ret = update_free_space_extent_count(trans, block_group, path,
972 					     new_extents);
973 
974 out:
975 	return ret;
976 }
977 
978 EXPORT_FOR_TESTS
979 int __add_to_free_space_tree(struct btrfs_trans_handle *trans,
980 			     struct btrfs_block_group *block_group,
981 			     struct btrfs_path *path, u64 start, u64 size)
982 {
983 	struct btrfs_free_space_info *info;
984 	u32 flags;
985 	int ret;
986 
987 	if (block_group->needs_free_space) {
988 		ret = __add_block_group_free_space(trans, block_group, path);
989 		if (ret)
990 			return ret;
991 	}
992 
993 	info = search_free_space_info(NULL, block_group, path, 0);
994 	if (IS_ERR(info))
995 		return PTR_ERR(info);
996 	flags = btrfs_free_space_flags(path->nodes[0], info);
997 	btrfs_release_path(path);
998 
999 	if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) {
1000 		return modify_free_space_bitmap(trans, block_group, path,
1001 						start, size, 0);
1002 	} else {
1003 		return add_free_space_extent(trans, block_group, path, start,
1004 					     size);
1005 	}
1006 }
1007 
1008 int add_to_free_space_tree(struct btrfs_trans_handle *trans,
1009 			   u64 start, u64 size)
1010 {
1011 	struct btrfs_block_group *block_group;
1012 	struct btrfs_path *path;
1013 	int ret;
1014 
1015 	if (!btrfs_fs_compat_ro(trans->fs_info, FREE_SPACE_TREE))
1016 		return 0;
1017 
1018 	path = btrfs_alloc_path();
1019 	if (!path) {
1020 		ret = -ENOMEM;
1021 		goto out;
1022 	}
1023 
1024 	block_group = btrfs_lookup_block_group(trans->fs_info, start);
1025 	if (!block_group) {
1026 		ASSERT(0);
1027 		ret = -ENOENT;
1028 		goto out;
1029 	}
1030 
1031 	mutex_lock(&block_group->free_space_lock);
1032 	ret = __add_to_free_space_tree(trans, block_group, path, start, size);
1033 	mutex_unlock(&block_group->free_space_lock);
1034 
1035 	btrfs_put_block_group(block_group);
1036 out:
1037 	btrfs_free_path(path);
1038 	if (ret)
1039 		btrfs_abort_transaction(trans, ret);
1040 	return ret;
1041 }
1042 
1043 /*
1044  * Populate the free space tree by walking the extent tree. Operations on the
1045  * extent tree that happen as a result of writes to the free space tree will go
1046  * through the normal add/remove hooks.
1047  */
1048 static int populate_free_space_tree(struct btrfs_trans_handle *trans,
1049 				    struct btrfs_block_group *block_group)
1050 {
1051 	struct btrfs_root *extent_root = trans->fs_info->extent_root;
1052 	struct btrfs_path *path, *path2;
1053 	struct btrfs_key key;
1054 	u64 start, end;
1055 	int ret;
1056 
1057 	path = btrfs_alloc_path();
1058 	if (!path)
1059 		return -ENOMEM;
1060 	path->reada = READA_FORWARD;
1061 
1062 	path2 = btrfs_alloc_path();
1063 	if (!path2) {
1064 		btrfs_free_path(path);
1065 		return -ENOMEM;
1066 	}
1067 
1068 	ret = add_new_free_space_info(trans, block_group, path2);
1069 	if (ret)
1070 		goto out;
1071 
1072 	mutex_lock(&block_group->free_space_lock);
1073 
1074 	/*
1075 	 * Iterate through all of the extent and metadata items in this block
1076 	 * group, adding the free space between them and the free space at the
1077 	 * end. Note that EXTENT_ITEM and METADATA_ITEM are less than
1078 	 * BLOCK_GROUP_ITEM, so an extent may precede the block group that it's
1079 	 * contained in.
1080 	 */
1081 	key.objectid = block_group->start;
1082 	key.type = BTRFS_EXTENT_ITEM_KEY;
1083 	key.offset = 0;
1084 
1085 	ret = btrfs_search_slot_for_read(extent_root, &key, path, 1, 0);
1086 	if (ret < 0)
1087 		goto out_locked;
1088 	ASSERT(ret == 0);
1089 
1090 	start = block_group->start;
1091 	end = block_group->start + block_group->length;
1092 	while (1) {
1093 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1094 
1095 		if (key.type == BTRFS_EXTENT_ITEM_KEY ||
1096 		    key.type == BTRFS_METADATA_ITEM_KEY) {
1097 			if (key.objectid >= end)
1098 				break;
1099 
1100 			if (start < key.objectid) {
1101 				ret = __add_to_free_space_tree(trans,
1102 							       block_group,
1103 							       path2, start,
1104 							       key.objectid -
1105 							       start);
1106 				if (ret)
1107 					goto out_locked;
1108 			}
1109 			start = key.objectid;
1110 			if (key.type == BTRFS_METADATA_ITEM_KEY)
1111 				start += trans->fs_info->nodesize;
1112 			else
1113 				start += key.offset;
1114 		} else if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
1115 			if (key.objectid != block_group->start)
1116 				break;
1117 		}
1118 
1119 		ret = btrfs_next_item(extent_root, path);
1120 		if (ret < 0)
1121 			goto out_locked;
1122 		if (ret)
1123 			break;
1124 	}
1125 	if (start < end) {
1126 		ret = __add_to_free_space_tree(trans, block_group, path2,
1127 					       start, end - start);
1128 		if (ret)
1129 			goto out_locked;
1130 	}
1131 
1132 	ret = 0;
1133 out_locked:
1134 	mutex_unlock(&block_group->free_space_lock);
1135 out:
1136 	btrfs_free_path(path2);
1137 	btrfs_free_path(path);
1138 	return ret;
1139 }
1140 
1141 int btrfs_create_free_space_tree(struct btrfs_fs_info *fs_info)
1142 {
1143 	struct btrfs_trans_handle *trans;
1144 	struct btrfs_root *tree_root = fs_info->tree_root;
1145 	struct btrfs_root *free_space_root;
1146 	struct btrfs_block_group *block_group;
1147 	struct rb_node *node;
1148 	int ret;
1149 
1150 	trans = btrfs_start_transaction(tree_root, 0);
1151 	if (IS_ERR(trans))
1152 		return PTR_ERR(trans);
1153 
1154 	set_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1155 	free_space_root = btrfs_create_tree(trans,
1156 					    BTRFS_FREE_SPACE_TREE_OBJECTID);
1157 	if (IS_ERR(free_space_root)) {
1158 		ret = PTR_ERR(free_space_root);
1159 		goto abort;
1160 	}
1161 	fs_info->free_space_root = free_space_root;
1162 
1163 	node = rb_first(&fs_info->block_group_cache_tree);
1164 	while (node) {
1165 		block_group = rb_entry(node, struct btrfs_block_group,
1166 				       cache_node);
1167 		ret = populate_free_space_tree(trans, block_group);
1168 		if (ret)
1169 			goto abort;
1170 		node = rb_next(node);
1171 	}
1172 
1173 	btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE);
1174 	btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID);
1175 	clear_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1176 
1177 	return btrfs_commit_transaction(trans);
1178 
1179 abort:
1180 	clear_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1181 	btrfs_abort_transaction(trans, ret);
1182 	btrfs_end_transaction(trans);
1183 	return ret;
1184 }
1185 
1186 static int clear_free_space_tree(struct btrfs_trans_handle *trans,
1187 				 struct btrfs_root *root)
1188 {
1189 	struct btrfs_path *path;
1190 	struct btrfs_key key;
1191 	int nr;
1192 	int ret;
1193 
1194 	path = btrfs_alloc_path();
1195 	if (!path)
1196 		return -ENOMEM;
1197 
1198 	path->leave_spinning = 1;
1199 
1200 	key.objectid = 0;
1201 	key.type = 0;
1202 	key.offset = 0;
1203 
1204 	while (1) {
1205 		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1206 		if (ret < 0)
1207 			goto out;
1208 
1209 		nr = btrfs_header_nritems(path->nodes[0]);
1210 		if (!nr)
1211 			break;
1212 
1213 		path->slots[0] = 0;
1214 		ret = btrfs_del_items(trans, root, path, 0, nr);
1215 		if (ret)
1216 			goto out;
1217 
1218 		btrfs_release_path(path);
1219 	}
1220 
1221 	ret = 0;
1222 out:
1223 	btrfs_free_path(path);
1224 	return ret;
1225 }
1226 
1227 int btrfs_clear_free_space_tree(struct btrfs_fs_info *fs_info)
1228 {
1229 	struct btrfs_trans_handle *trans;
1230 	struct btrfs_root *tree_root = fs_info->tree_root;
1231 	struct btrfs_root *free_space_root = fs_info->free_space_root;
1232 	int ret;
1233 
1234 	trans = btrfs_start_transaction(tree_root, 0);
1235 	if (IS_ERR(trans))
1236 		return PTR_ERR(trans);
1237 
1238 	btrfs_clear_fs_compat_ro(fs_info, FREE_SPACE_TREE);
1239 	btrfs_clear_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID);
1240 	fs_info->free_space_root = NULL;
1241 
1242 	ret = clear_free_space_tree(trans, free_space_root);
1243 	if (ret)
1244 		goto abort;
1245 
1246 	ret = btrfs_del_root(trans, &free_space_root->root_key);
1247 	if (ret)
1248 		goto abort;
1249 
1250 	list_del(&free_space_root->dirty_list);
1251 
1252 	btrfs_tree_lock(free_space_root->node);
1253 	btrfs_clean_tree_block(free_space_root->node);
1254 	btrfs_tree_unlock(free_space_root->node);
1255 	btrfs_free_tree_block(trans, free_space_root, free_space_root->node,
1256 			      0, 1);
1257 
1258 	btrfs_put_root(free_space_root);
1259 
1260 	return btrfs_commit_transaction(trans);
1261 
1262 abort:
1263 	btrfs_abort_transaction(trans, ret);
1264 	btrfs_end_transaction(trans);
1265 	return ret;
1266 }
1267 
1268 static int __add_block_group_free_space(struct btrfs_trans_handle *trans,
1269 					struct btrfs_block_group *block_group,
1270 					struct btrfs_path *path)
1271 {
1272 	int ret;
1273 
1274 	block_group->needs_free_space = 0;
1275 
1276 	ret = add_new_free_space_info(trans, block_group, path);
1277 	if (ret)
1278 		return ret;
1279 
1280 	return __add_to_free_space_tree(trans, block_group, path,
1281 					block_group->start,
1282 					block_group->length);
1283 }
1284 
1285 int add_block_group_free_space(struct btrfs_trans_handle *trans,
1286 			       struct btrfs_block_group *block_group)
1287 {
1288 	struct btrfs_fs_info *fs_info = trans->fs_info;
1289 	struct btrfs_path *path = NULL;
1290 	int ret = 0;
1291 
1292 	if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
1293 		return 0;
1294 
1295 	mutex_lock(&block_group->free_space_lock);
1296 	if (!block_group->needs_free_space)
1297 		goto out;
1298 
1299 	path = btrfs_alloc_path();
1300 	if (!path) {
1301 		ret = -ENOMEM;
1302 		goto out;
1303 	}
1304 
1305 	ret = __add_block_group_free_space(trans, block_group, path);
1306 
1307 out:
1308 	btrfs_free_path(path);
1309 	mutex_unlock(&block_group->free_space_lock);
1310 	if (ret)
1311 		btrfs_abort_transaction(trans, ret);
1312 	return ret;
1313 }
1314 
1315 int remove_block_group_free_space(struct btrfs_trans_handle *trans,
1316 				  struct btrfs_block_group *block_group)
1317 {
1318 	struct btrfs_root *root = trans->fs_info->free_space_root;
1319 	struct btrfs_path *path;
1320 	struct btrfs_key key, found_key;
1321 	struct extent_buffer *leaf;
1322 	u64 start, end;
1323 	int done = 0, nr;
1324 	int ret;
1325 
1326 	if (!btrfs_fs_compat_ro(trans->fs_info, FREE_SPACE_TREE))
1327 		return 0;
1328 
1329 	if (block_group->needs_free_space) {
1330 		/* We never added this block group to the free space tree. */
1331 		return 0;
1332 	}
1333 
1334 	path = btrfs_alloc_path();
1335 	if (!path) {
1336 		ret = -ENOMEM;
1337 		goto out;
1338 	}
1339 
1340 	start = block_group->start;
1341 	end = block_group->start + block_group->length;
1342 
1343 	key.objectid = end - 1;
1344 	key.type = (u8)-1;
1345 	key.offset = (u64)-1;
1346 
1347 	while (!done) {
1348 		ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
1349 		if (ret)
1350 			goto out;
1351 
1352 		leaf = path->nodes[0];
1353 		nr = 0;
1354 		path->slots[0]++;
1355 		while (path->slots[0] > 0) {
1356 			btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
1357 
1358 			if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
1359 				ASSERT(found_key.objectid == block_group->start);
1360 				ASSERT(found_key.offset == block_group->length);
1361 				done = 1;
1362 				nr++;
1363 				path->slots[0]--;
1364 				break;
1365 			} else if (found_key.type == BTRFS_FREE_SPACE_EXTENT_KEY ||
1366 				   found_key.type == BTRFS_FREE_SPACE_BITMAP_KEY) {
1367 				ASSERT(found_key.objectid >= start);
1368 				ASSERT(found_key.objectid < end);
1369 				ASSERT(found_key.objectid + found_key.offset <= end);
1370 				nr++;
1371 				path->slots[0]--;
1372 			} else {
1373 				ASSERT(0);
1374 			}
1375 		}
1376 
1377 		ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
1378 		if (ret)
1379 			goto out;
1380 		btrfs_release_path(path);
1381 	}
1382 
1383 	ret = 0;
1384 out:
1385 	btrfs_free_path(path);
1386 	if (ret)
1387 		btrfs_abort_transaction(trans, ret);
1388 	return ret;
1389 }
1390 
1391 static int load_free_space_bitmaps(struct btrfs_caching_control *caching_ctl,
1392 				   struct btrfs_path *path,
1393 				   u32 expected_extent_count)
1394 {
1395 	struct btrfs_block_group *block_group;
1396 	struct btrfs_fs_info *fs_info;
1397 	struct btrfs_root *root;
1398 	struct btrfs_key key;
1399 	int prev_bit = 0, bit;
1400 	/* Initialize to silence GCC. */
1401 	u64 extent_start = 0;
1402 	u64 end, offset;
1403 	u64 total_found = 0;
1404 	u32 extent_count = 0;
1405 	int ret;
1406 
1407 	block_group = caching_ctl->block_group;
1408 	fs_info = block_group->fs_info;
1409 	root = fs_info->free_space_root;
1410 
1411 	end = block_group->start + block_group->length;
1412 
1413 	while (1) {
1414 		ret = btrfs_next_item(root, path);
1415 		if (ret < 0)
1416 			goto out;
1417 		if (ret)
1418 			break;
1419 
1420 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1421 
1422 		if (key.type == BTRFS_FREE_SPACE_INFO_KEY)
1423 			break;
1424 
1425 		ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
1426 		ASSERT(key.objectid < end && key.objectid + key.offset <= end);
1427 
1428 		caching_ctl->progress = key.objectid;
1429 
1430 		offset = key.objectid;
1431 		while (offset < key.objectid + key.offset) {
1432 			bit = free_space_test_bit(block_group, path, offset);
1433 			if (prev_bit == 0 && bit == 1) {
1434 				extent_start = offset;
1435 			} else if (prev_bit == 1 && bit == 0) {
1436 				total_found += add_new_free_space(block_group,
1437 								  extent_start,
1438 								  offset);
1439 				if (total_found > CACHING_CTL_WAKE_UP) {
1440 					total_found = 0;
1441 					wake_up(&caching_ctl->wait);
1442 				}
1443 				extent_count++;
1444 			}
1445 			prev_bit = bit;
1446 			offset += fs_info->sectorsize;
1447 		}
1448 	}
1449 	if (prev_bit == 1) {
1450 		total_found += add_new_free_space(block_group, extent_start,
1451 						  end);
1452 		extent_count++;
1453 	}
1454 
1455 	if (extent_count != expected_extent_count) {
1456 		btrfs_err(fs_info,
1457 			  "incorrect extent count for %llu; counted %u, expected %u",
1458 			  block_group->start, extent_count,
1459 			  expected_extent_count);
1460 		ASSERT(0);
1461 		ret = -EIO;
1462 		goto out;
1463 	}
1464 
1465 	caching_ctl->progress = (u64)-1;
1466 
1467 	ret = 0;
1468 out:
1469 	return ret;
1470 }
1471 
1472 static int load_free_space_extents(struct btrfs_caching_control *caching_ctl,
1473 				   struct btrfs_path *path,
1474 				   u32 expected_extent_count)
1475 {
1476 	struct btrfs_block_group *block_group;
1477 	struct btrfs_fs_info *fs_info;
1478 	struct btrfs_root *root;
1479 	struct btrfs_key key;
1480 	u64 end;
1481 	u64 total_found = 0;
1482 	u32 extent_count = 0;
1483 	int ret;
1484 
1485 	block_group = caching_ctl->block_group;
1486 	fs_info = block_group->fs_info;
1487 	root = fs_info->free_space_root;
1488 
1489 	end = block_group->start + block_group->length;
1490 
1491 	while (1) {
1492 		ret = btrfs_next_item(root, path);
1493 		if (ret < 0)
1494 			goto out;
1495 		if (ret)
1496 			break;
1497 
1498 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1499 
1500 		if (key.type == BTRFS_FREE_SPACE_INFO_KEY)
1501 			break;
1502 
1503 		ASSERT(key.type == BTRFS_FREE_SPACE_EXTENT_KEY);
1504 		ASSERT(key.objectid < end && key.objectid + key.offset <= end);
1505 
1506 		caching_ctl->progress = key.objectid;
1507 
1508 		total_found += add_new_free_space(block_group, key.objectid,
1509 						  key.objectid + key.offset);
1510 		if (total_found > CACHING_CTL_WAKE_UP) {
1511 			total_found = 0;
1512 			wake_up(&caching_ctl->wait);
1513 		}
1514 		extent_count++;
1515 	}
1516 
1517 	if (extent_count != expected_extent_count) {
1518 		btrfs_err(fs_info,
1519 			  "incorrect extent count for %llu; counted %u, expected %u",
1520 			  block_group->start, extent_count,
1521 			  expected_extent_count);
1522 		ASSERT(0);
1523 		ret = -EIO;
1524 		goto out;
1525 	}
1526 
1527 	caching_ctl->progress = (u64)-1;
1528 
1529 	ret = 0;
1530 out:
1531 	return ret;
1532 }
1533 
1534 int load_free_space_tree(struct btrfs_caching_control *caching_ctl)
1535 {
1536 	struct btrfs_block_group *block_group;
1537 	struct btrfs_free_space_info *info;
1538 	struct btrfs_path *path;
1539 	u32 extent_count, flags;
1540 	int ret;
1541 
1542 	block_group = caching_ctl->block_group;
1543 
1544 	path = btrfs_alloc_path();
1545 	if (!path)
1546 		return -ENOMEM;
1547 
1548 	/*
1549 	 * Just like caching_thread() doesn't want to deadlock on the extent
1550 	 * tree, we don't want to deadlock on the free space tree.
1551 	 */
1552 	path->skip_locking = 1;
1553 	path->search_commit_root = 1;
1554 	path->reada = READA_FORWARD;
1555 
1556 	info = search_free_space_info(NULL, block_group, path, 0);
1557 	if (IS_ERR(info)) {
1558 		ret = PTR_ERR(info);
1559 		goto out;
1560 	}
1561 	extent_count = btrfs_free_space_extent_count(path->nodes[0], info);
1562 	flags = btrfs_free_space_flags(path->nodes[0], info);
1563 
1564 	/*
1565 	 * We left path pointing to the free space info item, so now
1566 	 * load_free_space_foo can just iterate through the free space tree from
1567 	 * there.
1568 	 */
1569 	if (flags & BTRFS_FREE_SPACE_USING_BITMAPS)
1570 		ret = load_free_space_bitmaps(caching_ctl, path, extent_count);
1571 	else
1572 		ret = load_free_space_extents(caching_ctl, path, extent_count);
1573 
1574 out:
1575 	btrfs_free_path(path);
1576 	return ret;
1577 }
1578