xref: /linux/fs/btrfs/free-space-tree.c (revision a9aaf1ff88a8cb99a1335c9eb76de637f0cf8c10)
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 "messages.h"
9 #include "ctree.h"
10 #include "disk-io.h"
11 #include "locking.h"
12 #include "free-space-tree.h"
13 #include "transaction.h"
14 #include "block-group.h"
15 #include "fs.h"
16 #include "accessors.h"
17 #include "extent-tree.h"
18 #include "root-tree.h"
19 
20 static int __add_block_group_free_space(struct btrfs_trans_handle *trans,
21 					struct btrfs_block_group *block_group,
22 					struct btrfs_path *path);
23 
24 static struct btrfs_root *btrfs_free_space_root(
25 				struct btrfs_block_group *block_group)
26 {
27 	struct btrfs_key key = {
28 		.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID,
29 		.type = BTRFS_ROOT_ITEM_KEY,
30 		.offset = 0,
31 	};
32 
33 	if (btrfs_fs_incompat(block_group->fs_info, EXTENT_TREE_V2))
34 		key.offset = block_group->global_root_id;
35 	return btrfs_global_root(block_group->fs_info, &key);
36 }
37 
38 void set_free_space_tree_thresholds(struct btrfs_block_group *cache)
39 {
40 	u32 bitmap_range;
41 	size_t bitmap_size;
42 	u64 num_bitmaps, total_bitmap_size;
43 
44 	if (WARN_ON(cache->length == 0))
45 		btrfs_warn(cache->fs_info, "block group %llu length is zero",
46 			   cache->start);
47 
48 	/*
49 	 * We convert to bitmaps when the disk space required for using extents
50 	 * exceeds that required for using bitmaps.
51 	 */
52 	bitmap_range = cache->fs_info->sectorsize * BTRFS_FREE_SPACE_BITMAP_BITS;
53 	num_bitmaps = div_u64(cache->length + bitmap_range - 1, bitmap_range);
54 	bitmap_size = sizeof(struct btrfs_item) + BTRFS_FREE_SPACE_BITMAP_SIZE;
55 	total_bitmap_size = num_bitmaps * bitmap_size;
56 	cache->bitmap_high_thresh = div_u64(total_bitmap_size,
57 					    sizeof(struct btrfs_item));
58 
59 	/*
60 	 * We allow for a small buffer between the high threshold and low
61 	 * threshold to avoid thrashing back and forth between the two formats.
62 	 */
63 	if (cache->bitmap_high_thresh > 100)
64 		cache->bitmap_low_thresh = cache->bitmap_high_thresh - 100;
65 	else
66 		cache->bitmap_low_thresh = 0;
67 }
68 
69 static int add_new_free_space_info(struct btrfs_trans_handle *trans,
70 				   struct btrfs_block_group *block_group,
71 				   struct btrfs_path *path)
72 {
73 	struct btrfs_root *root = btrfs_free_space_root(block_group);
74 	struct btrfs_free_space_info *info;
75 	struct btrfs_key key;
76 	struct extent_buffer *leaf;
77 	int ret;
78 
79 	key.objectid = block_group->start;
80 	key.type = BTRFS_FREE_SPACE_INFO_KEY;
81 	key.offset = block_group->length;
82 
83 	ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*info));
84 	if (ret)
85 		goto out;
86 
87 	leaf = path->nodes[0];
88 	info = btrfs_item_ptr(leaf, path->slots[0],
89 			      struct btrfs_free_space_info);
90 	btrfs_set_free_space_extent_count(leaf, info, 0);
91 	btrfs_set_free_space_flags(leaf, info, 0);
92 	btrfs_mark_buffer_dirty(trans, leaf);
93 
94 	ret = 0;
95 out:
96 	btrfs_release_path(path);
97 	return ret;
98 }
99 
100 EXPORT_FOR_TESTS
101 struct btrfs_free_space_info *search_free_space_info(
102 		struct btrfs_trans_handle *trans,
103 		struct btrfs_block_group *block_group,
104 		struct btrfs_path *path, int cow)
105 {
106 	struct btrfs_fs_info *fs_info = block_group->fs_info;
107 	struct btrfs_root *root = btrfs_free_space_root(block_group);
108 	struct btrfs_key key;
109 	int ret;
110 
111 	key.objectid = block_group->start;
112 	key.type = BTRFS_FREE_SPACE_INFO_KEY;
113 	key.offset = block_group->length;
114 
115 	ret = btrfs_search_slot(trans, root, &key, path, 0, cow);
116 	if (ret < 0)
117 		return ERR_PTR(ret);
118 	if (ret != 0) {
119 		btrfs_warn(fs_info, "missing free space info for %llu",
120 			   block_group->start);
121 		ASSERT(0);
122 		return ERR_PTR(-ENOENT);
123 	}
124 
125 	return btrfs_item_ptr(path->nodes[0], path->slots[0],
126 			      struct btrfs_free_space_info);
127 }
128 
129 /*
130  * btrfs_search_slot() but we're looking for the greatest key less than the
131  * passed key.
132  */
133 static int btrfs_search_prev_slot(struct btrfs_trans_handle *trans,
134 				  struct btrfs_root *root,
135 				  struct btrfs_key *key, struct btrfs_path *p,
136 				  int ins_len, int cow)
137 {
138 	int ret;
139 
140 	ret = btrfs_search_slot(trans, root, key, p, ins_len, cow);
141 	if (ret < 0)
142 		return ret;
143 
144 	if (ret == 0) {
145 		ASSERT(0);
146 		return -EIO;
147 	}
148 
149 	if (p->slots[0] == 0) {
150 		ASSERT(0);
151 		return -EIO;
152 	}
153 	p->slots[0]--;
154 
155 	return 0;
156 }
157 
158 static inline u32 free_space_bitmap_size(const struct btrfs_fs_info *fs_info,
159 					 u64 size)
160 {
161 	return DIV_ROUND_UP(size >> fs_info->sectorsize_bits, BITS_PER_BYTE);
162 }
163 
164 static unsigned long *alloc_bitmap(u32 bitmap_size)
165 {
166 	unsigned long *ret;
167 	unsigned int nofs_flag;
168 	u32 bitmap_rounded_size = round_up(bitmap_size, sizeof(unsigned long));
169 
170 	/*
171 	 * GFP_NOFS doesn't work with kvmalloc(), but we really can't recurse
172 	 * into the filesystem as the free space bitmap can be modified in the
173 	 * critical section of a transaction commit.
174 	 *
175 	 * TODO: push the memalloc_nofs_{save,restore}() to the caller where we
176 	 * know that recursion is unsafe.
177 	 */
178 	nofs_flag = memalloc_nofs_save();
179 	ret = kvzalloc(bitmap_rounded_size, GFP_KERNEL);
180 	memalloc_nofs_restore(nofs_flag);
181 	return ret;
182 }
183 
184 static void le_bitmap_set(unsigned long *map, unsigned int start, int len)
185 {
186 	u8 *p = ((u8 *)map) + BIT_BYTE(start);
187 	const unsigned int size = start + len;
188 	int bits_to_set = BITS_PER_BYTE - (start % BITS_PER_BYTE);
189 	u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(start);
190 
191 	while (len - bits_to_set >= 0) {
192 		*p |= mask_to_set;
193 		len -= bits_to_set;
194 		bits_to_set = BITS_PER_BYTE;
195 		mask_to_set = ~0;
196 		p++;
197 	}
198 	if (len) {
199 		mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
200 		*p |= mask_to_set;
201 	}
202 }
203 
204 EXPORT_FOR_TESTS
205 int convert_free_space_to_bitmaps(struct btrfs_trans_handle *trans,
206 				  struct btrfs_block_group *block_group,
207 				  struct btrfs_path *path)
208 {
209 	struct btrfs_fs_info *fs_info = trans->fs_info;
210 	struct btrfs_root *root = btrfs_free_space_root(block_group);
211 	struct btrfs_free_space_info *info;
212 	struct btrfs_key key, found_key;
213 	struct extent_buffer *leaf;
214 	unsigned long *bitmap;
215 	char *bitmap_cursor;
216 	u64 start, end;
217 	u64 bitmap_range, i;
218 	u32 bitmap_size, flags, expected_extent_count;
219 	u32 extent_count = 0;
220 	int done = 0, nr;
221 	int ret;
222 
223 	bitmap_size = free_space_bitmap_size(fs_info, block_group->length);
224 	bitmap = alloc_bitmap(bitmap_size);
225 	if (!bitmap) {
226 		ret = -ENOMEM;
227 		goto out;
228 	}
229 
230 	start = block_group->start;
231 	end = block_group->start + block_group->length;
232 
233 	key.objectid = end - 1;
234 	key.type = (u8)-1;
235 	key.offset = (u64)-1;
236 
237 	while (!done) {
238 		ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
239 		if (ret)
240 			goto out;
241 
242 		leaf = path->nodes[0];
243 		nr = 0;
244 		path->slots[0]++;
245 		while (path->slots[0] > 0) {
246 			btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
247 
248 			if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
249 				ASSERT(found_key.objectid == block_group->start);
250 				ASSERT(found_key.offset == block_group->length);
251 				done = 1;
252 				break;
253 			} else if (found_key.type == BTRFS_FREE_SPACE_EXTENT_KEY) {
254 				u64 first, last;
255 
256 				ASSERT(found_key.objectid >= start);
257 				ASSERT(found_key.objectid < end);
258 				ASSERT(found_key.objectid + found_key.offset <= end);
259 
260 				first = div_u64(found_key.objectid - start,
261 						fs_info->sectorsize);
262 				last = div_u64(found_key.objectid + found_key.offset - start,
263 					       fs_info->sectorsize);
264 				le_bitmap_set(bitmap, first, last - first);
265 
266 				extent_count++;
267 				nr++;
268 				path->slots[0]--;
269 			} else {
270 				ASSERT(0);
271 			}
272 		}
273 
274 		ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
275 		if (ret)
276 			goto out;
277 		btrfs_release_path(path);
278 	}
279 
280 	info = search_free_space_info(trans, block_group, path, 1);
281 	if (IS_ERR(info)) {
282 		ret = PTR_ERR(info);
283 		goto out;
284 	}
285 	leaf = path->nodes[0];
286 	flags = btrfs_free_space_flags(leaf, info);
287 	flags |= BTRFS_FREE_SPACE_USING_BITMAPS;
288 	btrfs_set_free_space_flags(leaf, info, flags);
289 	expected_extent_count = btrfs_free_space_extent_count(leaf, info);
290 	btrfs_mark_buffer_dirty(trans, leaf);
291 	btrfs_release_path(path);
292 
293 	if (extent_count != expected_extent_count) {
294 		btrfs_err(fs_info,
295 			  "incorrect extent count for %llu; counted %u, expected %u",
296 			  block_group->start, extent_count,
297 			  expected_extent_count);
298 		ASSERT(0);
299 		ret = -EIO;
300 		goto out;
301 	}
302 
303 	bitmap_cursor = (char *)bitmap;
304 	bitmap_range = fs_info->sectorsize * BTRFS_FREE_SPACE_BITMAP_BITS;
305 	i = start;
306 	while (i < end) {
307 		unsigned long ptr;
308 		u64 extent_size;
309 		u32 data_size;
310 
311 		extent_size = min(end - i, bitmap_range);
312 		data_size = free_space_bitmap_size(fs_info, extent_size);
313 
314 		key.objectid = i;
315 		key.type = BTRFS_FREE_SPACE_BITMAP_KEY;
316 		key.offset = extent_size;
317 
318 		ret = btrfs_insert_empty_item(trans, root, path, &key,
319 					      data_size);
320 		if (ret)
321 			goto out;
322 
323 		leaf = path->nodes[0];
324 		ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
325 		write_extent_buffer(leaf, bitmap_cursor, ptr,
326 				    data_size);
327 		btrfs_mark_buffer_dirty(trans, leaf);
328 		btrfs_release_path(path);
329 
330 		i += extent_size;
331 		bitmap_cursor += data_size;
332 	}
333 
334 	ret = 0;
335 out:
336 	kvfree(bitmap);
337 	if (ret)
338 		btrfs_abort_transaction(trans, ret);
339 	return ret;
340 }
341 
342 EXPORT_FOR_TESTS
343 int convert_free_space_to_extents(struct btrfs_trans_handle *trans,
344 				  struct btrfs_block_group *block_group,
345 				  struct btrfs_path *path)
346 {
347 	struct btrfs_fs_info *fs_info = trans->fs_info;
348 	struct btrfs_root *root = btrfs_free_space_root(block_group);
349 	struct btrfs_free_space_info *info;
350 	struct btrfs_key key, found_key;
351 	struct extent_buffer *leaf;
352 	unsigned long *bitmap;
353 	u64 start, end;
354 	u32 bitmap_size, flags, expected_extent_count;
355 	unsigned long nrbits, start_bit, end_bit;
356 	u32 extent_count = 0;
357 	int done = 0, nr;
358 	int ret;
359 
360 	bitmap_size = free_space_bitmap_size(fs_info, block_group->length);
361 	bitmap = alloc_bitmap(bitmap_size);
362 	if (!bitmap) {
363 		ret = -ENOMEM;
364 		goto out;
365 	}
366 
367 	start = block_group->start;
368 	end = block_group->start + block_group->length;
369 
370 	key.objectid = end - 1;
371 	key.type = (u8)-1;
372 	key.offset = (u64)-1;
373 
374 	while (!done) {
375 		ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
376 		if (ret)
377 			goto out;
378 
379 		leaf = path->nodes[0];
380 		nr = 0;
381 		path->slots[0]++;
382 		while (path->slots[0] > 0) {
383 			btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
384 
385 			if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
386 				ASSERT(found_key.objectid == block_group->start);
387 				ASSERT(found_key.offset == block_group->length);
388 				done = 1;
389 				break;
390 			} else if (found_key.type == BTRFS_FREE_SPACE_BITMAP_KEY) {
391 				unsigned long ptr;
392 				char *bitmap_cursor;
393 				u32 bitmap_pos, data_size;
394 
395 				ASSERT(found_key.objectid >= start);
396 				ASSERT(found_key.objectid < end);
397 				ASSERT(found_key.objectid + found_key.offset <= end);
398 
399 				bitmap_pos = div_u64(found_key.objectid - start,
400 						     fs_info->sectorsize *
401 						     BITS_PER_BYTE);
402 				bitmap_cursor = ((char *)bitmap) + bitmap_pos;
403 				data_size = free_space_bitmap_size(fs_info,
404 								found_key.offset);
405 
406 				ptr = btrfs_item_ptr_offset(leaf, path->slots[0] - 1);
407 				read_extent_buffer(leaf, bitmap_cursor, ptr,
408 						   data_size);
409 
410 				nr++;
411 				path->slots[0]--;
412 			} else {
413 				ASSERT(0);
414 			}
415 		}
416 
417 		ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
418 		if (ret)
419 			goto out;
420 		btrfs_release_path(path);
421 	}
422 
423 	info = search_free_space_info(trans, block_group, path, 1);
424 	if (IS_ERR(info)) {
425 		ret = PTR_ERR(info);
426 		goto out;
427 	}
428 	leaf = path->nodes[0];
429 	flags = btrfs_free_space_flags(leaf, info);
430 	flags &= ~BTRFS_FREE_SPACE_USING_BITMAPS;
431 	btrfs_set_free_space_flags(leaf, info, flags);
432 	expected_extent_count = btrfs_free_space_extent_count(leaf, info);
433 	btrfs_mark_buffer_dirty(trans, leaf);
434 	btrfs_release_path(path);
435 
436 	nrbits = block_group->length >> block_group->fs_info->sectorsize_bits;
437 	start_bit = find_next_bit_le(bitmap, nrbits, 0);
438 
439 	while (start_bit < nrbits) {
440 		end_bit = find_next_zero_bit_le(bitmap, nrbits, start_bit);
441 		ASSERT(start_bit < end_bit);
442 
443 		key.objectid = start + start_bit * block_group->fs_info->sectorsize;
444 		key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
445 		key.offset = (end_bit - start_bit) * block_group->fs_info->sectorsize;
446 
447 		ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
448 		if (ret)
449 			goto out;
450 		btrfs_release_path(path);
451 
452 		extent_count++;
453 
454 		start_bit = find_next_bit_le(bitmap, nrbits, end_bit);
455 	}
456 
457 	if (extent_count != expected_extent_count) {
458 		btrfs_err(fs_info,
459 			  "incorrect extent count for %llu; counted %u, expected %u",
460 			  block_group->start, extent_count,
461 			  expected_extent_count);
462 		ASSERT(0);
463 		ret = -EIO;
464 		goto out;
465 	}
466 
467 	ret = 0;
468 out:
469 	kvfree(bitmap);
470 	if (ret)
471 		btrfs_abort_transaction(trans, ret);
472 	return ret;
473 }
474 
475 static int update_free_space_extent_count(struct btrfs_trans_handle *trans,
476 					  struct btrfs_block_group *block_group,
477 					  struct btrfs_path *path,
478 					  int new_extents)
479 {
480 	struct btrfs_free_space_info *info;
481 	u32 flags;
482 	u32 extent_count;
483 	int ret = 0;
484 
485 	if (new_extents == 0)
486 		return 0;
487 
488 	info = search_free_space_info(trans, block_group, path, 1);
489 	if (IS_ERR(info)) {
490 		ret = PTR_ERR(info);
491 		goto out;
492 	}
493 	flags = btrfs_free_space_flags(path->nodes[0], info);
494 	extent_count = btrfs_free_space_extent_count(path->nodes[0], info);
495 
496 	extent_count += new_extents;
497 	btrfs_set_free_space_extent_count(path->nodes[0], info, extent_count);
498 	btrfs_mark_buffer_dirty(trans, path->nodes[0]);
499 	btrfs_release_path(path);
500 
501 	if (!(flags & BTRFS_FREE_SPACE_USING_BITMAPS) &&
502 	    extent_count > block_group->bitmap_high_thresh) {
503 		ret = convert_free_space_to_bitmaps(trans, block_group, path);
504 	} else if ((flags & BTRFS_FREE_SPACE_USING_BITMAPS) &&
505 		   extent_count < block_group->bitmap_low_thresh) {
506 		ret = convert_free_space_to_extents(trans, block_group, path);
507 	}
508 
509 out:
510 	return ret;
511 }
512 
513 EXPORT_FOR_TESTS
514 int free_space_test_bit(struct btrfs_block_group *block_group,
515 			struct btrfs_path *path, u64 offset)
516 {
517 	struct extent_buffer *leaf;
518 	struct btrfs_key key;
519 	u64 found_start, found_end;
520 	unsigned long ptr, i;
521 
522 	leaf = path->nodes[0];
523 	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
524 	ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
525 
526 	found_start = key.objectid;
527 	found_end = key.objectid + key.offset;
528 	ASSERT(offset >= found_start && offset < found_end);
529 
530 	ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
531 	i = div_u64(offset - found_start,
532 		    block_group->fs_info->sectorsize);
533 	return !!extent_buffer_test_bit(leaf, ptr, i);
534 }
535 
536 static void free_space_set_bits(struct btrfs_trans_handle *trans,
537 				struct btrfs_block_group *block_group,
538 				struct btrfs_path *path, u64 *start, u64 *size,
539 				int bit)
540 {
541 	struct btrfs_fs_info *fs_info = block_group->fs_info;
542 	struct extent_buffer *leaf;
543 	struct btrfs_key key;
544 	u64 end = *start + *size;
545 	u64 found_start, found_end;
546 	unsigned long ptr, first, last;
547 
548 	leaf = path->nodes[0];
549 	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
550 	ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
551 
552 	found_start = key.objectid;
553 	found_end = key.objectid + key.offset;
554 	ASSERT(*start >= found_start && *start < found_end);
555 	ASSERT(end > found_start);
556 
557 	if (end > found_end)
558 		end = found_end;
559 
560 	ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
561 	first = (*start - found_start) >> fs_info->sectorsize_bits;
562 	last = (end - found_start) >> fs_info->sectorsize_bits;
563 	if (bit)
564 		extent_buffer_bitmap_set(leaf, ptr, first, last - first);
565 	else
566 		extent_buffer_bitmap_clear(leaf, ptr, first, last - first);
567 	btrfs_mark_buffer_dirty(trans, leaf);
568 
569 	*size -= end - *start;
570 	*start = end;
571 }
572 
573 /*
574  * We can't use btrfs_next_item() in modify_free_space_bitmap() because
575  * btrfs_next_leaf() doesn't get the path for writing. We can forgo the fancy
576  * tree walking in btrfs_next_leaf() anyways because we know exactly what we're
577  * looking for.
578  */
579 static int free_space_next_bitmap(struct btrfs_trans_handle *trans,
580 				  struct btrfs_root *root, struct btrfs_path *p)
581 {
582 	struct btrfs_key key;
583 
584 	if (p->slots[0] + 1 < btrfs_header_nritems(p->nodes[0])) {
585 		p->slots[0]++;
586 		return 0;
587 	}
588 
589 	btrfs_item_key_to_cpu(p->nodes[0], &key, p->slots[0]);
590 	btrfs_release_path(p);
591 
592 	key.objectid += key.offset;
593 	key.type = (u8)-1;
594 	key.offset = (u64)-1;
595 
596 	return btrfs_search_prev_slot(trans, root, &key, p, 0, 1);
597 }
598 
599 /*
600  * If remove is 1, then we are removing free space, thus clearing bits in the
601  * bitmap. If remove is 0, then we are adding free space, thus setting bits in
602  * the bitmap.
603  */
604 static int modify_free_space_bitmap(struct btrfs_trans_handle *trans,
605 				    struct btrfs_block_group *block_group,
606 				    struct btrfs_path *path,
607 				    u64 start, u64 size, int remove)
608 {
609 	struct btrfs_root *root = btrfs_free_space_root(block_group);
610 	struct btrfs_key key;
611 	u64 end = start + size;
612 	u64 cur_start, cur_size;
613 	int prev_bit, next_bit;
614 	int new_extents;
615 	int ret;
616 
617 	/*
618 	 * Read the bit for the block immediately before the extent of space if
619 	 * that block is within the block group.
620 	 */
621 	if (start > block_group->start) {
622 		u64 prev_block = start - block_group->fs_info->sectorsize;
623 
624 		key.objectid = prev_block;
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 = free_space_test_bit(block_group, path, prev_block);
633 
634 		/* The previous block may have been in the previous bitmap. */
635 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
636 		if (start >= key.objectid + key.offset) {
637 			ret = free_space_next_bitmap(trans, root, path);
638 			if (ret)
639 				goto out;
640 		}
641 	} else {
642 		key.objectid = start;
643 		key.type = (u8)-1;
644 		key.offset = (u64)-1;
645 
646 		ret = btrfs_search_prev_slot(trans, root, &key, path, 0, 1);
647 		if (ret)
648 			goto out;
649 
650 		prev_bit = -1;
651 	}
652 
653 	/*
654 	 * Iterate over all of the bitmaps overlapped by the extent of space,
655 	 * clearing/setting bits as required.
656 	 */
657 	cur_start = start;
658 	cur_size = size;
659 	while (1) {
660 		free_space_set_bits(trans, block_group, path, &cur_start, &cur_size,
661 				    !remove);
662 		if (cur_size == 0)
663 			break;
664 		ret = free_space_next_bitmap(trans, root, path);
665 		if (ret)
666 			goto out;
667 	}
668 
669 	/*
670 	 * Read the bit for the block immediately after the extent of space if
671 	 * that block is within the block group.
672 	 */
673 	if (end < block_group->start + block_group->length) {
674 		/* The next block may be in the next bitmap. */
675 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
676 		if (end >= key.objectid + key.offset) {
677 			ret = free_space_next_bitmap(trans, root, path);
678 			if (ret)
679 				goto out;
680 		}
681 
682 		next_bit = free_space_test_bit(block_group, path, end);
683 	} else {
684 		next_bit = -1;
685 	}
686 
687 	if (remove) {
688 		new_extents = -1;
689 		if (prev_bit == 1) {
690 			/* Leftover on the left. */
691 			new_extents++;
692 		}
693 		if (next_bit == 1) {
694 			/* Leftover on the right. */
695 			new_extents++;
696 		}
697 	} else {
698 		new_extents = 1;
699 		if (prev_bit == 1) {
700 			/* Merging with neighbor on the left. */
701 			new_extents--;
702 		}
703 		if (next_bit == 1) {
704 			/* Merging with neighbor on the right. */
705 			new_extents--;
706 		}
707 	}
708 
709 	btrfs_release_path(path);
710 	ret = update_free_space_extent_count(trans, block_group, path,
711 					     new_extents);
712 
713 out:
714 	return ret;
715 }
716 
717 static int remove_free_space_extent(struct btrfs_trans_handle *trans,
718 				    struct btrfs_block_group *block_group,
719 				    struct btrfs_path *path,
720 				    u64 start, u64 size)
721 {
722 	struct btrfs_root *root = btrfs_free_space_root(block_group);
723 	struct btrfs_key key;
724 	u64 found_start, found_end;
725 	u64 end = start + size;
726 	int new_extents = -1;
727 	int ret;
728 
729 	key.objectid = start;
730 	key.type = (u8)-1;
731 	key.offset = (u64)-1;
732 
733 	ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
734 	if (ret)
735 		goto out;
736 
737 	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
738 
739 	ASSERT(key.type == BTRFS_FREE_SPACE_EXTENT_KEY);
740 
741 	found_start = key.objectid;
742 	found_end = key.objectid + key.offset;
743 	ASSERT(start >= found_start && end <= found_end);
744 
745 	/*
746 	 * Okay, now that we've found the free space extent which contains the
747 	 * free space that we are removing, there are four cases:
748 	 *
749 	 * 1. We're using the whole extent: delete the key we found and
750 	 * decrement the free space extent count.
751 	 * 2. We are using part of the extent starting at the beginning: delete
752 	 * the key we found and insert a new key representing the leftover at
753 	 * the end. There is no net change in the number of extents.
754 	 * 3. We are using part of the extent ending at the end: delete the key
755 	 * we found and insert a new key representing the leftover at the
756 	 * beginning. There is no net change in the number of extents.
757 	 * 4. We are using part of the extent in the middle: delete the key we
758 	 * found and insert two new keys representing the leftovers on each
759 	 * side. Where we used to have one extent, we now have two, so increment
760 	 * the extent count. We may need to convert the block group to bitmaps
761 	 * as a result.
762 	 */
763 
764 	/* Delete the existing key (cases 1-4). */
765 	ret = btrfs_del_item(trans, root, path);
766 	if (ret)
767 		goto out;
768 
769 	/* Add a key for leftovers at the beginning (cases 3 and 4). */
770 	if (start > found_start) {
771 		key.objectid = found_start;
772 		key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
773 		key.offset = start - found_start;
774 
775 		btrfs_release_path(path);
776 		ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
777 		if (ret)
778 			goto out;
779 		new_extents++;
780 	}
781 
782 	/* Add a key for leftovers at the end (cases 2 and 4). */
783 	if (end < found_end) {
784 		key.objectid = end;
785 		key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
786 		key.offset = found_end - end;
787 
788 		btrfs_release_path(path);
789 		ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
790 		if (ret)
791 			goto out;
792 		new_extents++;
793 	}
794 
795 	btrfs_release_path(path);
796 	ret = update_free_space_extent_count(trans, block_group, path,
797 					     new_extents);
798 
799 out:
800 	return ret;
801 }
802 
803 EXPORT_FOR_TESTS
804 int __remove_from_free_space_tree(struct btrfs_trans_handle *trans,
805 				  struct btrfs_block_group *block_group,
806 				  struct btrfs_path *path, u64 start, u64 size)
807 {
808 	struct btrfs_free_space_info *info;
809 	u32 flags;
810 	int ret;
811 
812 	if (test_bit(BLOCK_GROUP_FLAG_NEEDS_FREE_SPACE, &block_group->runtime_flags)) {
813 		ret = __add_block_group_free_space(trans, block_group, path);
814 		if (ret)
815 			return ret;
816 	}
817 
818 	info = search_free_space_info(NULL, block_group, path, 0);
819 	if (IS_ERR(info))
820 		return PTR_ERR(info);
821 	flags = btrfs_free_space_flags(path->nodes[0], info);
822 	btrfs_release_path(path);
823 
824 	if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) {
825 		return modify_free_space_bitmap(trans, block_group, path,
826 						start, size, 1);
827 	} else {
828 		return remove_free_space_extent(trans, block_group, path,
829 						start, size);
830 	}
831 }
832 
833 int remove_from_free_space_tree(struct btrfs_trans_handle *trans,
834 				u64 start, u64 size)
835 {
836 	struct btrfs_block_group *block_group;
837 	struct btrfs_path *path;
838 	int ret;
839 
840 	if (!btrfs_fs_compat_ro(trans->fs_info, FREE_SPACE_TREE))
841 		return 0;
842 
843 	path = btrfs_alloc_path();
844 	if (!path) {
845 		ret = -ENOMEM;
846 		goto out;
847 	}
848 
849 	block_group = btrfs_lookup_block_group(trans->fs_info, start);
850 	if (!block_group) {
851 		ASSERT(0);
852 		ret = -ENOENT;
853 		goto out;
854 	}
855 
856 	mutex_lock(&block_group->free_space_lock);
857 	ret = __remove_from_free_space_tree(trans, block_group, path, start,
858 					    size);
859 	mutex_unlock(&block_group->free_space_lock);
860 
861 	btrfs_put_block_group(block_group);
862 out:
863 	btrfs_free_path(path);
864 	if (ret)
865 		btrfs_abort_transaction(trans, ret);
866 	return ret;
867 }
868 
869 static int add_free_space_extent(struct btrfs_trans_handle *trans,
870 				 struct btrfs_block_group *block_group,
871 				 struct btrfs_path *path,
872 				 u64 start, u64 size)
873 {
874 	struct btrfs_root *root = btrfs_free_space_root(block_group);
875 	struct btrfs_key key, new_key;
876 	u64 found_start, found_end;
877 	u64 end = start + size;
878 	int new_extents = 1;
879 	int ret;
880 
881 	/*
882 	 * We are adding a new extent of free space, but we need to merge
883 	 * extents. There are four cases here:
884 	 *
885 	 * 1. The new extent does not have any immediate neighbors to merge
886 	 * with: add the new key and increment the free space extent count. We
887 	 * may need to convert the block group to bitmaps as a result.
888 	 * 2. The new extent has an immediate neighbor before it: remove the
889 	 * previous key and insert a new key combining both of them. There is no
890 	 * net change in the number of extents.
891 	 * 3. The new extent has an immediate neighbor after it: remove the next
892 	 * key and insert a new key combining both of them. There is no net
893 	 * change in the number of extents.
894 	 * 4. The new extent has immediate neighbors on both sides: remove both
895 	 * of the keys and insert a new key combining all of them. Where we used
896 	 * to have two extents, we now have one, so decrement the extent count.
897 	 */
898 
899 	new_key.objectid = start;
900 	new_key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
901 	new_key.offset = size;
902 
903 	/* Search for a neighbor on the left. */
904 	if (start == block_group->start)
905 		goto right;
906 	key.objectid = start - 1;
907 	key.type = (u8)-1;
908 	key.offset = (u64)-1;
909 
910 	ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
911 	if (ret)
912 		goto out;
913 
914 	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
915 
916 	if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) {
917 		ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY);
918 		btrfs_release_path(path);
919 		goto right;
920 	}
921 
922 	found_start = key.objectid;
923 	found_end = key.objectid + key.offset;
924 	ASSERT(found_start >= block_group->start &&
925 	       found_end > block_group->start);
926 	ASSERT(found_start < start && found_end <= start);
927 
928 	/*
929 	 * Delete the neighbor on the left and absorb it into the new key (cases
930 	 * 2 and 4).
931 	 */
932 	if (found_end == start) {
933 		ret = btrfs_del_item(trans, root, path);
934 		if (ret)
935 			goto out;
936 		new_key.objectid = found_start;
937 		new_key.offset += key.offset;
938 		new_extents--;
939 	}
940 	btrfs_release_path(path);
941 
942 right:
943 	/* Search for a neighbor on the right. */
944 	if (end == block_group->start + block_group->length)
945 		goto insert;
946 	key.objectid = end;
947 	key.type = (u8)-1;
948 	key.offset = (u64)-1;
949 
950 	ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
951 	if (ret)
952 		goto out;
953 
954 	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
955 
956 	if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) {
957 		ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY);
958 		btrfs_release_path(path);
959 		goto insert;
960 	}
961 
962 	found_start = key.objectid;
963 	found_end = key.objectid + key.offset;
964 	ASSERT(found_start >= block_group->start &&
965 	       found_end > block_group->start);
966 	ASSERT((found_start < start && found_end <= start) ||
967 	       (found_start >= end && found_end > end));
968 
969 	/*
970 	 * Delete the neighbor on the right and absorb it into the new key
971 	 * (cases 3 and 4).
972 	 */
973 	if (found_start == end) {
974 		ret = btrfs_del_item(trans, root, path);
975 		if (ret)
976 			goto out;
977 		new_key.offset += key.offset;
978 		new_extents--;
979 	}
980 	btrfs_release_path(path);
981 
982 insert:
983 	/* Insert the new key (cases 1-4). */
984 	ret = btrfs_insert_empty_item(trans, root, path, &new_key, 0);
985 	if (ret)
986 		goto out;
987 
988 	btrfs_release_path(path);
989 	ret = update_free_space_extent_count(trans, block_group, path,
990 					     new_extents);
991 
992 out:
993 	return ret;
994 }
995 
996 EXPORT_FOR_TESTS
997 int __add_to_free_space_tree(struct btrfs_trans_handle *trans,
998 			     struct btrfs_block_group *block_group,
999 			     struct btrfs_path *path, u64 start, u64 size)
1000 {
1001 	struct btrfs_free_space_info *info;
1002 	u32 flags;
1003 	int ret;
1004 
1005 	if (test_bit(BLOCK_GROUP_FLAG_NEEDS_FREE_SPACE, &block_group->runtime_flags)) {
1006 		ret = __add_block_group_free_space(trans, block_group, path);
1007 		if (ret)
1008 			return ret;
1009 	}
1010 
1011 	info = search_free_space_info(NULL, block_group, path, 0);
1012 	if (IS_ERR(info))
1013 		return PTR_ERR(info);
1014 	flags = btrfs_free_space_flags(path->nodes[0], info);
1015 	btrfs_release_path(path);
1016 
1017 	if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) {
1018 		return modify_free_space_bitmap(trans, block_group, path,
1019 						start, size, 0);
1020 	} else {
1021 		return add_free_space_extent(trans, block_group, path, start,
1022 					     size);
1023 	}
1024 }
1025 
1026 int add_to_free_space_tree(struct btrfs_trans_handle *trans,
1027 			   u64 start, u64 size)
1028 {
1029 	struct btrfs_block_group *block_group;
1030 	struct btrfs_path *path;
1031 	int ret;
1032 
1033 	if (!btrfs_fs_compat_ro(trans->fs_info, FREE_SPACE_TREE))
1034 		return 0;
1035 
1036 	path = btrfs_alloc_path();
1037 	if (!path) {
1038 		ret = -ENOMEM;
1039 		goto out;
1040 	}
1041 
1042 	block_group = btrfs_lookup_block_group(trans->fs_info, start);
1043 	if (!block_group) {
1044 		ASSERT(0);
1045 		ret = -ENOENT;
1046 		goto out;
1047 	}
1048 
1049 	mutex_lock(&block_group->free_space_lock);
1050 	ret = __add_to_free_space_tree(trans, block_group, path, start, size);
1051 	mutex_unlock(&block_group->free_space_lock);
1052 
1053 	btrfs_put_block_group(block_group);
1054 out:
1055 	btrfs_free_path(path);
1056 	if (ret)
1057 		btrfs_abort_transaction(trans, ret);
1058 	return ret;
1059 }
1060 
1061 /*
1062  * Populate the free space tree by walking the extent tree. Operations on the
1063  * extent tree that happen as a result of writes to the free space tree will go
1064  * through the normal add/remove hooks.
1065  */
1066 static int populate_free_space_tree(struct btrfs_trans_handle *trans,
1067 				    struct btrfs_block_group *block_group)
1068 {
1069 	struct btrfs_root *extent_root;
1070 	struct btrfs_path *path, *path2;
1071 	struct btrfs_key key;
1072 	u64 start, end;
1073 	int ret;
1074 
1075 	path = btrfs_alloc_path();
1076 	if (!path)
1077 		return -ENOMEM;
1078 	path->reada = READA_FORWARD;
1079 
1080 	path2 = btrfs_alloc_path();
1081 	if (!path2) {
1082 		btrfs_free_path(path);
1083 		return -ENOMEM;
1084 	}
1085 
1086 	ret = add_new_free_space_info(trans, block_group, path2);
1087 	if (ret)
1088 		goto out;
1089 
1090 	mutex_lock(&block_group->free_space_lock);
1091 
1092 	/*
1093 	 * Iterate through all of the extent and metadata items in this block
1094 	 * group, adding the free space between them and the free space at the
1095 	 * end. Note that EXTENT_ITEM and METADATA_ITEM are less than
1096 	 * BLOCK_GROUP_ITEM, so an extent may precede the block group that it's
1097 	 * contained in.
1098 	 */
1099 	key.objectid = block_group->start;
1100 	key.type = BTRFS_EXTENT_ITEM_KEY;
1101 	key.offset = 0;
1102 
1103 	extent_root = btrfs_extent_root(trans->fs_info, key.objectid);
1104 	ret = btrfs_search_slot_for_read(extent_root, &key, path, 1, 0);
1105 	if (ret < 0)
1106 		goto out_locked;
1107 	ASSERT(ret == 0);
1108 
1109 	start = block_group->start;
1110 	end = block_group->start + block_group->length;
1111 	while (1) {
1112 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1113 
1114 		if (key.type == BTRFS_EXTENT_ITEM_KEY ||
1115 		    key.type == BTRFS_METADATA_ITEM_KEY) {
1116 			if (key.objectid >= end)
1117 				break;
1118 
1119 			if (start < key.objectid) {
1120 				ret = __add_to_free_space_tree(trans,
1121 							       block_group,
1122 							       path2, start,
1123 							       key.objectid -
1124 							       start);
1125 				if (ret)
1126 					goto out_locked;
1127 			}
1128 			start = key.objectid;
1129 			if (key.type == BTRFS_METADATA_ITEM_KEY)
1130 				start += trans->fs_info->nodesize;
1131 			else
1132 				start += key.offset;
1133 		} else if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
1134 			if (key.objectid != block_group->start)
1135 				break;
1136 		}
1137 
1138 		ret = btrfs_next_item(extent_root, path);
1139 		if (ret < 0)
1140 			goto out_locked;
1141 		if (ret)
1142 			break;
1143 	}
1144 	if (start < end) {
1145 		ret = __add_to_free_space_tree(trans, block_group, path2,
1146 					       start, end - start);
1147 		if (ret)
1148 			goto out_locked;
1149 	}
1150 
1151 	ret = 0;
1152 out_locked:
1153 	mutex_unlock(&block_group->free_space_lock);
1154 out:
1155 	btrfs_free_path(path2);
1156 	btrfs_free_path(path);
1157 	return ret;
1158 }
1159 
1160 int btrfs_create_free_space_tree(struct btrfs_fs_info *fs_info)
1161 {
1162 	struct btrfs_trans_handle *trans;
1163 	struct btrfs_root *tree_root = fs_info->tree_root;
1164 	struct btrfs_root *free_space_root;
1165 	struct btrfs_block_group *block_group;
1166 	struct rb_node *node;
1167 	int ret;
1168 
1169 	trans = btrfs_start_transaction(tree_root, 0);
1170 	if (IS_ERR(trans))
1171 		return PTR_ERR(trans);
1172 
1173 	set_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1174 	set_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags);
1175 	free_space_root = btrfs_create_tree(trans,
1176 					    BTRFS_FREE_SPACE_TREE_OBJECTID);
1177 	if (IS_ERR(free_space_root)) {
1178 		ret = PTR_ERR(free_space_root);
1179 		btrfs_abort_transaction(trans, ret);
1180 		btrfs_end_transaction(trans);
1181 		goto out_clear;
1182 	}
1183 	ret = btrfs_global_root_insert(free_space_root);
1184 	if (ret) {
1185 		btrfs_put_root(free_space_root);
1186 		btrfs_abort_transaction(trans, ret);
1187 		btrfs_end_transaction(trans);
1188 		goto out_clear;
1189 	}
1190 
1191 	node = rb_first_cached(&fs_info->block_group_cache_tree);
1192 	while (node) {
1193 		block_group = rb_entry(node, struct btrfs_block_group,
1194 				       cache_node);
1195 		ret = populate_free_space_tree(trans, block_group);
1196 		if (ret) {
1197 			btrfs_abort_transaction(trans, ret);
1198 			btrfs_end_transaction(trans);
1199 			goto out_clear;
1200 		}
1201 		node = rb_next(node);
1202 	}
1203 
1204 	btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE);
1205 	btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID);
1206 	clear_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1207 	ret = btrfs_commit_transaction(trans);
1208 
1209 	/*
1210 	 * Now that we've committed the transaction any reading of our commit
1211 	 * root will be safe, so we can cache from the free space tree now.
1212 	 */
1213 	clear_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags);
1214 	return ret;
1215 
1216 out_clear:
1217 	clear_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1218 	clear_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags);
1219 	return ret;
1220 }
1221 
1222 static int clear_free_space_tree(struct btrfs_trans_handle *trans,
1223 				 struct btrfs_root *root)
1224 {
1225 	struct btrfs_path *path;
1226 	struct btrfs_key key;
1227 	int nr;
1228 	int ret;
1229 
1230 	path = btrfs_alloc_path();
1231 	if (!path)
1232 		return -ENOMEM;
1233 
1234 	key.objectid = 0;
1235 	key.type = 0;
1236 	key.offset = 0;
1237 
1238 	while (1) {
1239 		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1240 		if (ret < 0)
1241 			goto out;
1242 
1243 		nr = btrfs_header_nritems(path->nodes[0]);
1244 		if (!nr)
1245 			break;
1246 
1247 		path->slots[0] = 0;
1248 		ret = btrfs_del_items(trans, root, path, 0, nr);
1249 		if (ret)
1250 			goto out;
1251 
1252 		btrfs_release_path(path);
1253 	}
1254 
1255 	ret = 0;
1256 out:
1257 	btrfs_free_path(path);
1258 	return ret;
1259 }
1260 
1261 int btrfs_delete_free_space_tree(struct btrfs_fs_info *fs_info)
1262 {
1263 	struct btrfs_trans_handle *trans;
1264 	struct btrfs_root *tree_root = fs_info->tree_root;
1265 	struct btrfs_key key = {
1266 		.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID,
1267 		.type = BTRFS_ROOT_ITEM_KEY,
1268 		.offset = 0,
1269 	};
1270 	struct btrfs_root *free_space_root = btrfs_global_root(fs_info, &key);
1271 	int ret;
1272 
1273 	trans = btrfs_start_transaction(tree_root, 0);
1274 	if (IS_ERR(trans))
1275 		return PTR_ERR(trans);
1276 
1277 	btrfs_clear_fs_compat_ro(fs_info, FREE_SPACE_TREE);
1278 	btrfs_clear_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID);
1279 
1280 	ret = clear_free_space_tree(trans, free_space_root);
1281 	if (ret) {
1282 		btrfs_abort_transaction(trans, ret);
1283 		btrfs_end_transaction(trans);
1284 		return ret;
1285 	}
1286 
1287 	ret = btrfs_del_root(trans, &free_space_root->root_key);
1288 	if (ret) {
1289 		btrfs_abort_transaction(trans, ret);
1290 		btrfs_end_transaction(trans);
1291 		return ret;
1292 	}
1293 
1294 	btrfs_global_root_delete(free_space_root);
1295 
1296 	spin_lock(&fs_info->trans_lock);
1297 	list_del(&free_space_root->dirty_list);
1298 	spin_unlock(&fs_info->trans_lock);
1299 
1300 	btrfs_tree_lock(free_space_root->node);
1301 	btrfs_clear_buffer_dirty(trans, free_space_root->node);
1302 	btrfs_tree_unlock(free_space_root->node);
1303 	ret = btrfs_free_tree_block(trans, btrfs_root_id(free_space_root),
1304 				    free_space_root->node, 0, 1);
1305 	btrfs_put_root(free_space_root);
1306 	if (ret < 0) {
1307 		btrfs_abort_transaction(trans, ret);
1308 		btrfs_end_transaction(trans);
1309 		return ret;
1310 	}
1311 
1312 	return btrfs_commit_transaction(trans);
1313 }
1314 
1315 int btrfs_rebuild_free_space_tree(struct btrfs_fs_info *fs_info)
1316 {
1317 	struct btrfs_trans_handle *trans;
1318 	struct btrfs_key key = {
1319 		.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID,
1320 		.type = BTRFS_ROOT_ITEM_KEY,
1321 		.offset = 0,
1322 	};
1323 	struct btrfs_root *free_space_root = btrfs_global_root(fs_info, &key);
1324 	struct rb_node *node;
1325 	int ret;
1326 
1327 	trans = btrfs_start_transaction(free_space_root, 1);
1328 	if (IS_ERR(trans))
1329 		return PTR_ERR(trans);
1330 
1331 	set_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1332 	set_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags);
1333 
1334 	ret = clear_free_space_tree(trans, free_space_root);
1335 	if (ret) {
1336 		btrfs_abort_transaction(trans, ret);
1337 		btrfs_end_transaction(trans);
1338 		return ret;
1339 	}
1340 
1341 	node = rb_first_cached(&fs_info->block_group_cache_tree);
1342 	while (node) {
1343 		struct btrfs_block_group *block_group;
1344 
1345 		block_group = rb_entry(node, struct btrfs_block_group,
1346 				       cache_node);
1347 		ret = populate_free_space_tree(trans, block_group);
1348 		if (ret) {
1349 			btrfs_abort_transaction(trans, ret);
1350 			btrfs_end_transaction(trans);
1351 			return ret;
1352 		}
1353 		node = rb_next(node);
1354 	}
1355 
1356 	btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE);
1357 	btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID);
1358 	clear_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1359 
1360 	ret = btrfs_commit_transaction(trans);
1361 	clear_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags);
1362 	return ret;
1363 }
1364 
1365 static int __add_block_group_free_space(struct btrfs_trans_handle *trans,
1366 					struct btrfs_block_group *block_group,
1367 					struct btrfs_path *path)
1368 {
1369 	int ret;
1370 
1371 	clear_bit(BLOCK_GROUP_FLAG_NEEDS_FREE_SPACE, &block_group->runtime_flags);
1372 
1373 	ret = add_new_free_space_info(trans, block_group, path);
1374 	if (ret)
1375 		return ret;
1376 
1377 	return __add_to_free_space_tree(trans, block_group, path,
1378 					block_group->start,
1379 					block_group->length);
1380 }
1381 
1382 int add_block_group_free_space(struct btrfs_trans_handle *trans,
1383 			       struct btrfs_block_group *block_group)
1384 {
1385 	struct btrfs_fs_info *fs_info = trans->fs_info;
1386 	struct btrfs_path *path = NULL;
1387 	int ret = 0;
1388 
1389 	if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
1390 		return 0;
1391 
1392 	mutex_lock(&block_group->free_space_lock);
1393 	if (!test_bit(BLOCK_GROUP_FLAG_NEEDS_FREE_SPACE, &block_group->runtime_flags))
1394 		goto out;
1395 
1396 	path = btrfs_alloc_path();
1397 	if (!path) {
1398 		ret = -ENOMEM;
1399 		goto out;
1400 	}
1401 
1402 	ret = __add_block_group_free_space(trans, block_group, path);
1403 
1404 out:
1405 	btrfs_free_path(path);
1406 	mutex_unlock(&block_group->free_space_lock);
1407 	if (ret)
1408 		btrfs_abort_transaction(trans, ret);
1409 	return ret;
1410 }
1411 
1412 int remove_block_group_free_space(struct btrfs_trans_handle *trans,
1413 				  struct btrfs_block_group *block_group)
1414 {
1415 	struct btrfs_root *root = btrfs_free_space_root(block_group);
1416 	struct btrfs_path *path;
1417 	struct btrfs_key key, found_key;
1418 	struct extent_buffer *leaf;
1419 	u64 start, end;
1420 	int done = 0, nr;
1421 	int ret;
1422 
1423 	if (!btrfs_fs_compat_ro(trans->fs_info, FREE_SPACE_TREE))
1424 		return 0;
1425 
1426 	if (test_bit(BLOCK_GROUP_FLAG_NEEDS_FREE_SPACE, &block_group->runtime_flags)) {
1427 		/* We never added this block group to the free space tree. */
1428 		return 0;
1429 	}
1430 
1431 	path = btrfs_alloc_path();
1432 	if (!path) {
1433 		ret = -ENOMEM;
1434 		goto out;
1435 	}
1436 
1437 	start = block_group->start;
1438 	end = block_group->start + block_group->length;
1439 
1440 	key.objectid = end - 1;
1441 	key.type = (u8)-1;
1442 	key.offset = (u64)-1;
1443 
1444 	while (!done) {
1445 		ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
1446 		if (ret)
1447 			goto out;
1448 
1449 		leaf = path->nodes[0];
1450 		nr = 0;
1451 		path->slots[0]++;
1452 		while (path->slots[0] > 0) {
1453 			btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
1454 
1455 			if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
1456 				ASSERT(found_key.objectid == block_group->start);
1457 				ASSERT(found_key.offset == block_group->length);
1458 				done = 1;
1459 				nr++;
1460 				path->slots[0]--;
1461 				break;
1462 			} else if (found_key.type == BTRFS_FREE_SPACE_EXTENT_KEY ||
1463 				   found_key.type == BTRFS_FREE_SPACE_BITMAP_KEY) {
1464 				ASSERT(found_key.objectid >= start);
1465 				ASSERT(found_key.objectid < end);
1466 				ASSERT(found_key.objectid + found_key.offset <= end);
1467 				nr++;
1468 				path->slots[0]--;
1469 			} else {
1470 				ASSERT(0);
1471 			}
1472 		}
1473 
1474 		ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
1475 		if (ret)
1476 			goto out;
1477 		btrfs_release_path(path);
1478 	}
1479 
1480 	ret = 0;
1481 out:
1482 	btrfs_free_path(path);
1483 	if (ret)
1484 		btrfs_abort_transaction(trans, ret);
1485 	return ret;
1486 }
1487 
1488 static int load_free_space_bitmaps(struct btrfs_caching_control *caching_ctl,
1489 				   struct btrfs_path *path,
1490 				   u32 expected_extent_count)
1491 {
1492 	struct btrfs_block_group *block_group;
1493 	struct btrfs_fs_info *fs_info;
1494 	struct btrfs_root *root;
1495 	struct btrfs_key key;
1496 	int prev_bit = 0, bit;
1497 	/* Initialize to silence GCC. */
1498 	u64 extent_start = 0;
1499 	u64 end, offset;
1500 	u64 total_found = 0;
1501 	u32 extent_count = 0;
1502 	int ret;
1503 
1504 	block_group = caching_ctl->block_group;
1505 	fs_info = block_group->fs_info;
1506 	root = btrfs_free_space_root(block_group);
1507 
1508 	end = block_group->start + block_group->length;
1509 
1510 	while (1) {
1511 		ret = btrfs_next_item(root, path);
1512 		if (ret < 0)
1513 			goto out;
1514 		if (ret)
1515 			break;
1516 
1517 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1518 
1519 		if (key.type == BTRFS_FREE_SPACE_INFO_KEY)
1520 			break;
1521 
1522 		ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
1523 		ASSERT(key.objectid < end && key.objectid + key.offset <= end);
1524 
1525 		offset = key.objectid;
1526 		while (offset < key.objectid + key.offset) {
1527 			bit = free_space_test_bit(block_group, path, offset);
1528 			if (prev_bit == 0 && bit == 1) {
1529 				extent_start = offset;
1530 			} else if (prev_bit == 1 && bit == 0) {
1531 				u64 space_added;
1532 
1533 				ret = btrfs_add_new_free_space(block_group,
1534 							       extent_start,
1535 							       offset,
1536 							       &space_added);
1537 				if (ret)
1538 					goto out;
1539 				total_found += space_added;
1540 				if (total_found > CACHING_CTL_WAKE_UP) {
1541 					total_found = 0;
1542 					wake_up(&caching_ctl->wait);
1543 				}
1544 				extent_count++;
1545 			}
1546 			prev_bit = bit;
1547 			offset += fs_info->sectorsize;
1548 		}
1549 	}
1550 	if (prev_bit == 1) {
1551 		ret = btrfs_add_new_free_space(block_group, extent_start, end, NULL);
1552 		if (ret)
1553 			goto out;
1554 		extent_count++;
1555 	}
1556 
1557 	if (extent_count != expected_extent_count) {
1558 		btrfs_err(fs_info,
1559 			  "incorrect extent count for %llu; counted %u, expected %u",
1560 			  block_group->start, extent_count,
1561 			  expected_extent_count);
1562 		ASSERT(0);
1563 		ret = -EIO;
1564 		goto out;
1565 	}
1566 
1567 	ret = 0;
1568 out:
1569 	return ret;
1570 }
1571 
1572 static int load_free_space_extents(struct btrfs_caching_control *caching_ctl,
1573 				   struct btrfs_path *path,
1574 				   u32 expected_extent_count)
1575 {
1576 	struct btrfs_block_group *block_group;
1577 	struct btrfs_fs_info *fs_info;
1578 	struct btrfs_root *root;
1579 	struct btrfs_key key;
1580 	u64 end;
1581 	u64 total_found = 0;
1582 	u32 extent_count = 0;
1583 	int ret;
1584 
1585 	block_group = caching_ctl->block_group;
1586 	fs_info = block_group->fs_info;
1587 	root = btrfs_free_space_root(block_group);
1588 
1589 	end = block_group->start + block_group->length;
1590 
1591 	while (1) {
1592 		u64 space_added;
1593 
1594 		ret = btrfs_next_item(root, path);
1595 		if (ret < 0)
1596 			goto out;
1597 		if (ret)
1598 			break;
1599 
1600 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1601 
1602 		if (key.type == BTRFS_FREE_SPACE_INFO_KEY)
1603 			break;
1604 
1605 		ASSERT(key.type == BTRFS_FREE_SPACE_EXTENT_KEY);
1606 		ASSERT(key.objectid < end && key.objectid + key.offset <= end);
1607 
1608 		ret = btrfs_add_new_free_space(block_group, key.objectid,
1609 					       key.objectid + key.offset,
1610 					       &space_added);
1611 		if (ret)
1612 			goto out;
1613 		total_found += space_added;
1614 		if (total_found > CACHING_CTL_WAKE_UP) {
1615 			total_found = 0;
1616 			wake_up(&caching_ctl->wait);
1617 		}
1618 		extent_count++;
1619 	}
1620 
1621 	if (extent_count != expected_extent_count) {
1622 		btrfs_err(fs_info,
1623 			  "incorrect extent count for %llu; counted %u, expected %u",
1624 			  block_group->start, extent_count,
1625 			  expected_extent_count);
1626 		ASSERT(0);
1627 		ret = -EIO;
1628 		goto out;
1629 	}
1630 
1631 	ret = 0;
1632 out:
1633 	return ret;
1634 }
1635 
1636 int load_free_space_tree(struct btrfs_caching_control *caching_ctl)
1637 {
1638 	struct btrfs_block_group *block_group;
1639 	struct btrfs_free_space_info *info;
1640 	struct btrfs_path *path;
1641 	u32 extent_count, flags;
1642 	int ret;
1643 
1644 	block_group = caching_ctl->block_group;
1645 
1646 	path = btrfs_alloc_path();
1647 	if (!path)
1648 		return -ENOMEM;
1649 
1650 	/*
1651 	 * Just like caching_thread() doesn't want to deadlock on the extent
1652 	 * tree, we don't want to deadlock on the free space tree.
1653 	 */
1654 	path->skip_locking = 1;
1655 	path->search_commit_root = 1;
1656 	path->reada = READA_FORWARD;
1657 
1658 	info = search_free_space_info(NULL, block_group, path, 0);
1659 	if (IS_ERR(info)) {
1660 		ret = PTR_ERR(info);
1661 		goto out;
1662 	}
1663 	extent_count = btrfs_free_space_extent_count(path->nodes[0], info);
1664 	flags = btrfs_free_space_flags(path->nodes[0], info);
1665 
1666 	/*
1667 	 * We left path pointing to the free space info item, so now
1668 	 * load_free_space_foo can just iterate through the free space tree from
1669 	 * there.
1670 	 */
1671 	if (flags & BTRFS_FREE_SPACE_USING_BITMAPS)
1672 		ret = load_free_space_bitmaps(caching_ctl, path, extent_count);
1673 	else
1674 		ret = load_free_space_extents(caching_ctl, path, extent_count);
1675 
1676 out:
1677 	btrfs_free_path(path);
1678 	return ret;
1679 }
1680