xref: /openbmc/linux/fs/btrfs/free-space-tree.c (revision de2bdb3d)
1 /*
2  * Copyright (C) 2015 Facebook.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18 
19 #include <linux/kernel.h>
20 #include <linux/vmalloc.h>
21 #include "ctree.h"
22 #include "disk-io.h"
23 #include "locking.h"
24 #include "free-space-tree.h"
25 #include "transaction.h"
26 
27 static int __add_block_group_free_space(struct btrfs_trans_handle *trans,
28 					struct btrfs_fs_info *fs_info,
29 					struct btrfs_block_group_cache *block_group,
30 					struct btrfs_path *path);
31 
32 void set_free_space_tree_thresholds(struct btrfs_block_group_cache *cache)
33 {
34 	u32 bitmap_range;
35 	size_t bitmap_size;
36 	u64 num_bitmaps, total_bitmap_size;
37 
38 	/*
39 	 * We convert to bitmaps when the disk space required for using extents
40 	 * exceeds that required for using bitmaps.
41 	 */
42 	bitmap_range = cache->sectorsize * BTRFS_FREE_SPACE_BITMAP_BITS;
43 	num_bitmaps = div_u64(cache->key.offset + bitmap_range - 1,
44 			      bitmap_range);
45 	bitmap_size = sizeof(struct btrfs_item) + BTRFS_FREE_SPACE_BITMAP_SIZE;
46 	total_bitmap_size = num_bitmaps * bitmap_size;
47 	cache->bitmap_high_thresh = div_u64(total_bitmap_size,
48 					    sizeof(struct btrfs_item));
49 
50 	/*
51 	 * We allow for a small buffer between the high threshold and low
52 	 * threshold to avoid thrashing back and forth between the two formats.
53 	 */
54 	if (cache->bitmap_high_thresh > 100)
55 		cache->bitmap_low_thresh = cache->bitmap_high_thresh - 100;
56 	else
57 		cache->bitmap_low_thresh = 0;
58 }
59 
60 static int add_new_free_space_info(struct btrfs_trans_handle *trans,
61 				   struct btrfs_fs_info *fs_info,
62 				   struct btrfs_block_group_cache *block_group,
63 				   struct btrfs_path *path)
64 {
65 	struct btrfs_root *root = fs_info->free_space_root;
66 	struct btrfs_free_space_info *info;
67 	struct btrfs_key key;
68 	struct extent_buffer *leaf;
69 	int ret;
70 
71 	key.objectid = block_group->key.objectid;
72 	key.type = BTRFS_FREE_SPACE_INFO_KEY;
73 	key.offset = block_group->key.offset;
74 
75 	ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*info));
76 	if (ret)
77 		goto out;
78 
79 	leaf = path->nodes[0];
80 	info = btrfs_item_ptr(leaf, path->slots[0],
81 			      struct btrfs_free_space_info);
82 	btrfs_set_free_space_extent_count(leaf, info, 0);
83 	btrfs_set_free_space_flags(leaf, info, 0);
84 	btrfs_mark_buffer_dirty(leaf);
85 
86 	ret = 0;
87 out:
88 	btrfs_release_path(path);
89 	return ret;
90 }
91 
92 struct btrfs_free_space_info *
93 search_free_space_info(struct btrfs_trans_handle *trans,
94 		       struct btrfs_fs_info *fs_info,
95 		       struct btrfs_block_group_cache *block_group,
96 		       struct btrfs_path *path, int cow)
97 {
98 	struct btrfs_root *root = fs_info->free_space_root;
99 	struct btrfs_key key;
100 	int ret;
101 
102 	key.objectid = block_group->key.objectid;
103 	key.type = BTRFS_FREE_SPACE_INFO_KEY;
104 	key.offset = block_group->key.offset;
105 
106 	ret = btrfs_search_slot(trans, root, &key, path, 0, cow);
107 	if (ret < 0)
108 		return ERR_PTR(ret);
109 	if (ret != 0) {
110 		btrfs_warn(fs_info, "missing free space info for %llu",
111 			   block_group->key.objectid);
112 		ASSERT(0);
113 		return ERR_PTR(-ENOENT);
114 	}
115 
116 	return btrfs_item_ptr(path->nodes[0], path->slots[0],
117 			      struct btrfs_free_space_info);
118 }
119 
120 /*
121  * btrfs_search_slot() but we're looking for the greatest key less than the
122  * passed key.
123  */
124 static int btrfs_search_prev_slot(struct btrfs_trans_handle *trans,
125 				  struct btrfs_root *root,
126 				  struct btrfs_key *key, struct btrfs_path *p,
127 				  int ins_len, int cow)
128 {
129 	int ret;
130 
131 	ret = btrfs_search_slot(trans, root, key, p, ins_len, cow);
132 	if (ret < 0)
133 		return ret;
134 
135 	if (ret == 0) {
136 		ASSERT(0);
137 		return -EIO;
138 	}
139 
140 	if (p->slots[0] == 0) {
141 		ASSERT(0);
142 		return -EIO;
143 	}
144 	p->slots[0]--;
145 
146 	return 0;
147 }
148 
149 static inline u32 free_space_bitmap_size(u64 size, u32 sectorsize)
150 {
151 	return DIV_ROUND_UP((u32)div_u64(size, sectorsize), BITS_PER_BYTE);
152 }
153 
154 static u8 *alloc_bitmap(u32 bitmap_size)
155 {
156 	void *mem;
157 
158 	/*
159 	 * The allocation size varies, observed numbers were < 4K up to 16K.
160 	 * Using vmalloc unconditionally would be too heavy, we'll try
161 	 * contiguous allocations first.
162 	 */
163 	if  (bitmap_size <= PAGE_SIZE)
164 		return kzalloc(bitmap_size, GFP_NOFS);
165 
166 	mem = kzalloc(bitmap_size, GFP_NOFS | __GFP_NOWARN);
167 	if (mem)
168 		return mem;
169 
170 	return __vmalloc(bitmap_size, GFP_NOFS | __GFP_HIGHMEM | __GFP_ZERO,
171 			 PAGE_KERNEL);
172 }
173 
174 int convert_free_space_to_bitmaps(struct btrfs_trans_handle *trans,
175 				  struct btrfs_fs_info *fs_info,
176 				  struct btrfs_block_group_cache *block_group,
177 				  struct btrfs_path *path)
178 {
179 	struct btrfs_root *root = fs_info->free_space_root;
180 	struct btrfs_free_space_info *info;
181 	struct btrfs_key key, found_key;
182 	struct extent_buffer *leaf;
183 	u8 *bitmap, *bitmap_cursor;
184 	u64 start, end;
185 	u64 bitmap_range, i;
186 	u32 bitmap_size, flags, expected_extent_count;
187 	u32 extent_count = 0;
188 	int done = 0, nr;
189 	int ret;
190 
191 	bitmap_size = free_space_bitmap_size(block_group->key.offset,
192 					     block_group->sectorsize);
193 	bitmap = alloc_bitmap(bitmap_size);
194 	if (!bitmap) {
195 		ret = -ENOMEM;
196 		goto out;
197 	}
198 
199 	start = block_group->key.objectid;
200 	end = block_group->key.objectid + block_group->key.offset;
201 
202 	key.objectid = end - 1;
203 	key.type = (u8)-1;
204 	key.offset = (u64)-1;
205 
206 	while (!done) {
207 		ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
208 		if (ret)
209 			goto out;
210 
211 		leaf = path->nodes[0];
212 		nr = 0;
213 		path->slots[0]++;
214 		while (path->slots[0] > 0) {
215 			btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
216 
217 			if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
218 				ASSERT(found_key.objectid == block_group->key.objectid);
219 				ASSERT(found_key.offset == block_group->key.offset);
220 				done = 1;
221 				break;
222 			} else if (found_key.type == BTRFS_FREE_SPACE_EXTENT_KEY) {
223 				u64 first, last;
224 
225 				ASSERT(found_key.objectid >= start);
226 				ASSERT(found_key.objectid < end);
227 				ASSERT(found_key.objectid + found_key.offset <= end);
228 
229 				first = div_u64(found_key.objectid - start,
230 						block_group->sectorsize);
231 				last = div_u64(found_key.objectid + found_key.offset - start,
232 					       block_group->sectorsize);
233 				le_bitmap_set(bitmap, first, last - first);
234 
235 				extent_count++;
236 				nr++;
237 				path->slots[0]--;
238 			} else {
239 				ASSERT(0);
240 			}
241 		}
242 
243 		ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
244 		if (ret)
245 			goto out;
246 		btrfs_release_path(path);
247 	}
248 
249 	info = search_free_space_info(trans, fs_info, block_group, path, 1);
250 	if (IS_ERR(info)) {
251 		ret = PTR_ERR(info);
252 		goto out;
253 	}
254 	leaf = path->nodes[0];
255 	flags = btrfs_free_space_flags(leaf, info);
256 	flags |= BTRFS_FREE_SPACE_USING_BITMAPS;
257 	btrfs_set_free_space_flags(leaf, info, flags);
258 	expected_extent_count = btrfs_free_space_extent_count(leaf, info);
259 	btrfs_mark_buffer_dirty(leaf);
260 	btrfs_release_path(path);
261 
262 	if (extent_count != expected_extent_count) {
263 		btrfs_err(fs_info,
264 			  "incorrect extent count for %llu; counted %u, expected %u",
265 			  block_group->key.objectid, extent_count,
266 			  expected_extent_count);
267 		ASSERT(0);
268 		ret = -EIO;
269 		goto out;
270 	}
271 
272 	bitmap_cursor = bitmap;
273 	bitmap_range = block_group->sectorsize * BTRFS_FREE_SPACE_BITMAP_BITS;
274 	i = start;
275 	while (i < end) {
276 		unsigned long ptr;
277 		u64 extent_size;
278 		u32 data_size;
279 
280 		extent_size = min(end - i, bitmap_range);
281 		data_size = free_space_bitmap_size(extent_size,
282 						   block_group->sectorsize);
283 
284 		key.objectid = i;
285 		key.type = BTRFS_FREE_SPACE_BITMAP_KEY;
286 		key.offset = extent_size;
287 
288 		ret = btrfs_insert_empty_item(trans, root, path, &key,
289 					      data_size);
290 		if (ret)
291 			goto out;
292 
293 		leaf = path->nodes[0];
294 		ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
295 		write_extent_buffer(leaf, bitmap_cursor, ptr,
296 				    data_size);
297 		btrfs_mark_buffer_dirty(leaf);
298 		btrfs_release_path(path);
299 
300 		i += extent_size;
301 		bitmap_cursor += data_size;
302 	}
303 
304 	ret = 0;
305 out:
306 	kvfree(bitmap);
307 	if (ret)
308 		btrfs_abort_transaction(trans, ret);
309 	return ret;
310 }
311 
312 int convert_free_space_to_extents(struct btrfs_trans_handle *trans,
313 				  struct btrfs_fs_info *fs_info,
314 				  struct btrfs_block_group_cache *block_group,
315 				  struct btrfs_path *path)
316 {
317 	struct btrfs_root *root = fs_info->free_space_root;
318 	struct btrfs_free_space_info *info;
319 	struct btrfs_key key, found_key;
320 	struct extent_buffer *leaf;
321 	u8 *bitmap;
322 	u64 start, end;
323 	/* Initialize to silence GCC. */
324 	u64 extent_start = 0;
325 	u64 offset;
326 	u32 bitmap_size, flags, expected_extent_count;
327 	int prev_bit = 0, bit, bitnr;
328 	u32 extent_count = 0;
329 	int done = 0, nr;
330 	int ret;
331 
332 	bitmap_size = free_space_bitmap_size(block_group->key.offset,
333 					     block_group->sectorsize);
334 	bitmap = alloc_bitmap(bitmap_size);
335 	if (!bitmap) {
336 		ret = -ENOMEM;
337 		goto out;
338 	}
339 
340 	start = block_group->key.objectid;
341 	end = block_group->key.objectid + block_group->key.offset;
342 
343 	key.objectid = end - 1;
344 	key.type = (u8)-1;
345 	key.offset = (u64)-1;
346 
347 	while (!done) {
348 		ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
349 		if (ret)
350 			goto out;
351 
352 		leaf = path->nodes[0];
353 		nr = 0;
354 		path->slots[0]++;
355 		while (path->slots[0] > 0) {
356 			btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
357 
358 			if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
359 				ASSERT(found_key.objectid == block_group->key.objectid);
360 				ASSERT(found_key.offset == block_group->key.offset);
361 				done = 1;
362 				break;
363 			} else if (found_key.type == BTRFS_FREE_SPACE_BITMAP_KEY) {
364 				unsigned long ptr;
365 				u8 *bitmap_cursor;
366 				u32 bitmap_pos, data_size;
367 
368 				ASSERT(found_key.objectid >= start);
369 				ASSERT(found_key.objectid < end);
370 				ASSERT(found_key.objectid + found_key.offset <= end);
371 
372 				bitmap_pos = div_u64(found_key.objectid - start,
373 						     block_group->sectorsize *
374 						     BITS_PER_BYTE);
375 				bitmap_cursor = bitmap + bitmap_pos;
376 				data_size = free_space_bitmap_size(found_key.offset,
377 								   block_group->sectorsize);
378 
379 				ptr = btrfs_item_ptr_offset(leaf, path->slots[0] - 1);
380 				read_extent_buffer(leaf, bitmap_cursor, ptr,
381 						   data_size);
382 
383 				nr++;
384 				path->slots[0]--;
385 			} else {
386 				ASSERT(0);
387 			}
388 		}
389 
390 		ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
391 		if (ret)
392 			goto out;
393 		btrfs_release_path(path);
394 	}
395 
396 	info = search_free_space_info(trans, fs_info, block_group, path, 1);
397 	if (IS_ERR(info)) {
398 		ret = PTR_ERR(info);
399 		goto out;
400 	}
401 	leaf = path->nodes[0];
402 	flags = btrfs_free_space_flags(leaf, info);
403 	flags &= ~BTRFS_FREE_SPACE_USING_BITMAPS;
404 	btrfs_set_free_space_flags(leaf, info, flags);
405 	expected_extent_count = btrfs_free_space_extent_count(leaf, info);
406 	btrfs_mark_buffer_dirty(leaf);
407 	btrfs_release_path(path);
408 
409 	offset = start;
410 	bitnr = 0;
411 	while (offset < end) {
412 		bit = !!le_test_bit(bitnr, bitmap);
413 		if (prev_bit == 0 && bit == 1) {
414 			extent_start = offset;
415 		} else if (prev_bit == 1 && bit == 0) {
416 			key.objectid = extent_start;
417 			key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
418 			key.offset = offset - extent_start;
419 
420 			ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
421 			if (ret)
422 				goto out;
423 			btrfs_release_path(path);
424 
425 			extent_count++;
426 		}
427 		prev_bit = bit;
428 		offset += block_group->sectorsize;
429 		bitnr++;
430 	}
431 	if (prev_bit == 1) {
432 		key.objectid = extent_start;
433 		key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
434 		key.offset = end - extent_start;
435 
436 		ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
437 		if (ret)
438 			goto out;
439 		btrfs_release_path(path);
440 
441 		extent_count++;
442 	}
443 
444 	if (extent_count != expected_extent_count) {
445 		btrfs_err(fs_info,
446 			  "incorrect extent count for %llu; counted %u, expected %u",
447 			  block_group->key.objectid, extent_count,
448 			  expected_extent_count);
449 		ASSERT(0);
450 		ret = -EIO;
451 		goto out;
452 	}
453 
454 	ret = 0;
455 out:
456 	kvfree(bitmap);
457 	if (ret)
458 		btrfs_abort_transaction(trans, ret);
459 	return ret;
460 }
461 
462 static int update_free_space_extent_count(struct btrfs_trans_handle *trans,
463 					  struct btrfs_fs_info *fs_info,
464 					  struct btrfs_block_group_cache *block_group,
465 					  struct btrfs_path *path,
466 					  int new_extents)
467 {
468 	struct btrfs_free_space_info *info;
469 	u32 flags;
470 	u32 extent_count;
471 	int ret = 0;
472 
473 	if (new_extents == 0)
474 		return 0;
475 
476 	info = search_free_space_info(trans, fs_info, block_group, path, 1);
477 	if (IS_ERR(info)) {
478 		ret = PTR_ERR(info);
479 		goto out;
480 	}
481 	flags = btrfs_free_space_flags(path->nodes[0], info);
482 	extent_count = btrfs_free_space_extent_count(path->nodes[0], info);
483 
484 	extent_count += new_extents;
485 	btrfs_set_free_space_extent_count(path->nodes[0], info, extent_count);
486 	btrfs_mark_buffer_dirty(path->nodes[0]);
487 	btrfs_release_path(path);
488 
489 	if (!(flags & BTRFS_FREE_SPACE_USING_BITMAPS) &&
490 	    extent_count > block_group->bitmap_high_thresh) {
491 		ret = convert_free_space_to_bitmaps(trans, fs_info, block_group,
492 						    path);
493 	} else if ((flags & BTRFS_FREE_SPACE_USING_BITMAPS) &&
494 		   extent_count < block_group->bitmap_low_thresh) {
495 		ret = convert_free_space_to_extents(trans, fs_info, block_group,
496 						    path);
497 	}
498 
499 out:
500 	return ret;
501 }
502 
503 int free_space_test_bit(struct btrfs_block_group_cache *block_group,
504 			struct btrfs_path *path, u64 offset)
505 {
506 	struct extent_buffer *leaf;
507 	struct btrfs_key key;
508 	u64 found_start, found_end;
509 	unsigned long ptr, i;
510 
511 	leaf = path->nodes[0];
512 	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
513 	ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
514 
515 	found_start = key.objectid;
516 	found_end = key.objectid + key.offset;
517 	ASSERT(offset >= found_start && offset < found_end);
518 
519 	ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
520 	i = div_u64(offset - found_start, block_group->sectorsize);
521 	return !!extent_buffer_test_bit(leaf, ptr, i);
522 }
523 
524 static void free_space_set_bits(struct btrfs_block_group_cache *block_group,
525 				struct btrfs_path *path, u64 *start, u64 *size,
526 				int bit)
527 {
528 	struct extent_buffer *leaf;
529 	struct btrfs_key key;
530 	u64 end = *start + *size;
531 	u64 found_start, found_end;
532 	unsigned long ptr, first, last;
533 
534 	leaf = path->nodes[0];
535 	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
536 	ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
537 
538 	found_start = key.objectid;
539 	found_end = key.objectid + key.offset;
540 	ASSERT(*start >= found_start && *start < found_end);
541 	ASSERT(end > found_start);
542 
543 	if (end > found_end)
544 		end = found_end;
545 
546 	ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
547 	first = div_u64(*start - found_start, block_group->sectorsize);
548 	last = div_u64(end - found_start, block_group->sectorsize);
549 	if (bit)
550 		extent_buffer_bitmap_set(leaf, ptr, first, last - first);
551 	else
552 		extent_buffer_bitmap_clear(leaf, ptr, first, last - first);
553 	btrfs_mark_buffer_dirty(leaf);
554 
555 	*size -= end - *start;
556 	*start = end;
557 }
558 
559 /*
560  * We can't use btrfs_next_item() in modify_free_space_bitmap() because
561  * btrfs_next_leaf() doesn't get the path for writing. We can forgo the fancy
562  * tree walking in btrfs_next_leaf() anyways because we know exactly what we're
563  * looking for.
564  */
565 static int free_space_next_bitmap(struct btrfs_trans_handle *trans,
566 				  struct btrfs_root *root, struct btrfs_path *p)
567 {
568 	struct btrfs_key key;
569 
570 	if (p->slots[0] + 1 < btrfs_header_nritems(p->nodes[0])) {
571 		p->slots[0]++;
572 		return 0;
573 	}
574 
575 	btrfs_item_key_to_cpu(p->nodes[0], &key, p->slots[0]);
576 	btrfs_release_path(p);
577 
578 	key.objectid += key.offset;
579 	key.type = (u8)-1;
580 	key.offset = (u64)-1;
581 
582 	return btrfs_search_prev_slot(trans, root, &key, p, 0, 1);
583 }
584 
585 /*
586  * If remove is 1, then we are removing free space, thus clearing bits in the
587  * bitmap. If remove is 0, then we are adding free space, thus setting bits in
588  * the bitmap.
589  */
590 static int modify_free_space_bitmap(struct btrfs_trans_handle *trans,
591 				    struct btrfs_fs_info *fs_info,
592 				    struct btrfs_block_group_cache *block_group,
593 				    struct btrfs_path *path,
594 				    u64 start, u64 size, int remove)
595 {
596 	struct btrfs_root *root = fs_info->free_space_root;
597 	struct btrfs_key key;
598 	u64 end = start + size;
599 	u64 cur_start, cur_size;
600 	int prev_bit, next_bit;
601 	int new_extents;
602 	int ret;
603 
604 	/*
605 	 * Read the bit for the block immediately before the extent of space if
606 	 * that block is within the block group.
607 	 */
608 	if (start > block_group->key.objectid) {
609 		u64 prev_block = start - block_group->sectorsize;
610 
611 		key.objectid = prev_block;
612 		key.type = (u8)-1;
613 		key.offset = (u64)-1;
614 
615 		ret = btrfs_search_prev_slot(trans, root, &key, path, 0, 1);
616 		if (ret)
617 			goto out;
618 
619 		prev_bit = free_space_test_bit(block_group, path, prev_block);
620 
621 		/* The previous block may have been in the previous bitmap. */
622 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
623 		if (start >= key.objectid + key.offset) {
624 			ret = free_space_next_bitmap(trans, root, path);
625 			if (ret)
626 				goto out;
627 		}
628 	} else {
629 		key.objectid = start;
630 		key.type = (u8)-1;
631 		key.offset = (u64)-1;
632 
633 		ret = btrfs_search_prev_slot(trans, root, &key, path, 0, 1);
634 		if (ret)
635 			goto out;
636 
637 		prev_bit = -1;
638 	}
639 
640 	/*
641 	 * Iterate over all of the bitmaps overlapped by the extent of space,
642 	 * clearing/setting bits as required.
643 	 */
644 	cur_start = start;
645 	cur_size = size;
646 	while (1) {
647 		free_space_set_bits(block_group, path, &cur_start, &cur_size,
648 				    !remove);
649 		if (cur_size == 0)
650 			break;
651 		ret = free_space_next_bitmap(trans, root, path);
652 		if (ret)
653 			goto out;
654 	}
655 
656 	/*
657 	 * Read the bit for the block immediately after the extent of space if
658 	 * that block is within the block group.
659 	 */
660 	if (end < block_group->key.objectid + block_group->key.offset) {
661 		/* The next block may be in the next bitmap. */
662 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
663 		if (end >= key.objectid + key.offset) {
664 			ret = free_space_next_bitmap(trans, root, path);
665 			if (ret)
666 				goto out;
667 		}
668 
669 		next_bit = free_space_test_bit(block_group, path, end);
670 	} else {
671 		next_bit = -1;
672 	}
673 
674 	if (remove) {
675 		new_extents = -1;
676 		if (prev_bit == 1) {
677 			/* Leftover on the left. */
678 			new_extents++;
679 		}
680 		if (next_bit == 1) {
681 			/* Leftover on the right. */
682 			new_extents++;
683 		}
684 	} else {
685 		new_extents = 1;
686 		if (prev_bit == 1) {
687 			/* Merging with neighbor on the left. */
688 			new_extents--;
689 		}
690 		if (next_bit == 1) {
691 			/* Merging with neighbor on the right. */
692 			new_extents--;
693 		}
694 	}
695 
696 	btrfs_release_path(path);
697 	ret = update_free_space_extent_count(trans, fs_info, block_group, path,
698 					     new_extents);
699 
700 out:
701 	return ret;
702 }
703 
704 static int remove_free_space_extent(struct btrfs_trans_handle *trans,
705 				    struct btrfs_fs_info *fs_info,
706 				    struct btrfs_block_group_cache *block_group,
707 				    struct btrfs_path *path,
708 				    u64 start, u64 size)
709 {
710 	struct btrfs_root *root = fs_info->free_space_root;
711 	struct btrfs_key key;
712 	u64 found_start, found_end;
713 	u64 end = start + size;
714 	int new_extents = -1;
715 	int ret;
716 
717 	key.objectid = start;
718 	key.type = (u8)-1;
719 	key.offset = (u64)-1;
720 
721 	ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
722 	if (ret)
723 		goto out;
724 
725 	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
726 
727 	ASSERT(key.type == BTRFS_FREE_SPACE_EXTENT_KEY);
728 
729 	found_start = key.objectid;
730 	found_end = key.objectid + key.offset;
731 	ASSERT(start >= found_start && end <= found_end);
732 
733 	/*
734 	 * Okay, now that we've found the free space extent which contains the
735 	 * free space that we are removing, there are four cases:
736 	 *
737 	 * 1. We're using the whole extent: delete the key we found and
738 	 * decrement the free space extent count.
739 	 * 2. We are using part of the extent starting at the beginning: delete
740 	 * the key we found and insert a new key representing the leftover at
741 	 * the end. There is no net change in the number of extents.
742 	 * 3. We are using part of the extent ending at the end: delete the key
743 	 * we found and insert a new key representing the leftover at the
744 	 * beginning. There is no net change in the number of extents.
745 	 * 4. We are using part of the extent in the middle: delete the key we
746 	 * found and insert two new keys representing the leftovers on each
747 	 * side. Where we used to have one extent, we now have two, so increment
748 	 * the extent count. We may need to convert the block group to bitmaps
749 	 * as a result.
750 	 */
751 
752 	/* Delete the existing key (cases 1-4). */
753 	ret = btrfs_del_item(trans, root, path);
754 	if (ret)
755 		goto out;
756 
757 	/* Add a key for leftovers at the beginning (cases 3 and 4). */
758 	if (start > found_start) {
759 		key.objectid = found_start;
760 		key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
761 		key.offset = start - found_start;
762 
763 		btrfs_release_path(path);
764 		ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
765 		if (ret)
766 			goto out;
767 		new_extents++;
768 	}
769 
770 	/* Add a key for leftovers at the end (cases 2 and 4). */
771 	if (end < found_end) {
772 		key.objectid = end;
773 		key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
774 		key.offset = found_end - end;
775 
776 		btrfs_release_path(path);
777 		ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
778 		if (ret)
779 			goto out;
780 		new_extents++;
781 	}
782 
783 	btrfs_release_path(path);
784 	ret = update_free_space_extent_count(trans, fs_info, block_group, path,
785 					     new_extents);
786 
787 out:
788 	return ret;
789 }
790 
791 int __remove_from_free_space_tree(struct btrfs_trans_handle *trans,
792 				  struct btrfs_fs_info *fs_info,
793 				  struct btrfs_block_group_cache *block_group,
794 				  struct btrfs_path *path, u64 start, u64 size)
795 {
796 	struct btrfs_free_space_info *info;
797 	u32 flags;
798 	int ret;
799 
800 	if (block_group->needs_free_space) {
801 		ret = __add_block_group_free_space(trans, fs_info, block_group,
802 						   path);
803 		if (ret)
804 			return ret;
805 	}
806 
807 	info = search_free_space_info(NULL, fs_info, block_group, path, 0);
808 	if (IS_ERR(info))
809 		return PTR_ERR(info);
810 	flags = btrfs_free_space_flags(path->nodes[0], info);
811 	btrfs_release_path(path);
812 
813 	if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) {
814 		return modify_free_space_bitmap(trans, fs_info, block_group,
815 						path, start, size, 1);
816 	} else {
817 		return remove_free_space_extent(trans, fs_info, block_group,
818 						path, start, size);
819 	}
820 }
821 
822 int remove_from_free_space_tree(struct btrfs_trans_handle *trans,
823 				struct btrfs_fs_info *fs_info,
824 				u64 start, u64 size)
825 {
826 	struct btrfs_block_group_cache *block_group;
827 	struct btrfs_path *path;
828 	int ret;
829 
830 	if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
831 		return 0;
832 
833 	path = btrfs_alloc_path();
834 	if (!path) {
835 		ret = -ENOMEM;
836 		goto out;
837 	}
838 
839 	block_group = btrfs_lookup_block_group(fs_info, start);
840 	if (!block_group) {
841 		ASSERT(0);
842 		ret = -ENOENT;
843 		goto out;
844 	}
845 
846 	mutex_lock(&block_group->free_space_lock);
847 	ret = __remove_from_free_space_tree(trans, fs_info, block_group, path,
848 					    start, size);
849 	mutex_unlock(&block_group->free_space_lock);
850 
851 	btrfs_put_block_group(block_group);
852 out:
853 	btrfs_free_path(path);
854 	if (ret)
855 		btrfs_abort_transaction(trans, ret);
856 	return ret;
857 }
858 
859 static int add_free_space_extent(struct btrfs_trans_handle *trans,
860 				 struct btrfs_fs_info *fs_info,
861 				 struct btrfs_block_group_cache *block_group,
862 				 struct btrfs_path *path,
863 				 u64 start, u64 size)
864 {
865 	struct btrfs_root *root = fs_info->free_space_root;
866 	struct btrfs_key key, new_key;
867 	u64 found_start, found_end;
868 	u64 end = start + size;
869 	int new_extents = 1;
870 	int ret;
871 
872 	/*
873 	 * We are adding a new extent of free space, but we need to merge
874 	 * extents. There are four cases here:
875 	 *
876 	 * 1. The new extent does not have any immediate neighbors to merge
877 	 * with: add the new key and increment the free space extent count. We
878 	 * may need to convert the block group to bitmaps as a result.
879 	 * 2. The new extent has an immediate neighbor before it: remove the
880 	 * previous key and insert a new key combining both of them. There is no
881 	 * net change in the number of extents.
882 	 * 3. The new extent has an immediate neighbor after it: remove the next
883 	 * key and insert a new key combining both of them. There is no net
884 	 * change in the number of extents.
885 	 * 4. The new extent has immediate neighbors on both sides: remove both
886 	 * of the keys and insert a new key combining all of them. Where we used
887 	 * to have two extents, we now have one, so decrement the extent count.
888 	 */
889 
890 	new_key.objectid = start;
891 	new_key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
892 	new_key.offset = size;
893 
894 	/* Search for a neighbor on the left. */
895 	if (start == block_group->key.objectid)
896 		goto right;
897 	key.objectid = start - 1;
898 	key.type = (u8)-1;
899 	key.offset = (u64)-1;
900 
901 	ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
902 	if (ret)
903 		goto out;
904 
905 	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
906 
907 	if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) {
908 		ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY);
909 		btrfs_release_path(path);
910 		goto right;
911 	}
912 
913 	found_start = key.objectid;
914 	found_end = key.objectid + key.offset;
915 	ASSERT(found_start >= block_group->key.objectid &&
916 	       found_end > block_group->key.objectid);
917 	ASSERT(found_start < start && found_end <= start);
918 
919 	/*
920 	 * Delete the neighbor on the left and absorb it into the new key (cases
921 	 * 2 and 4).
922 	 */
923 	if (found_end == start) {
924 		ret = btrfs_del_item(trans, root, path);
925 		if (ret)
926 			goto out;
927 		new_key.objectid = found_start;
928 		new_key.offset += key.offset;
929 		new_extents--;
930 	}
931 	btrfs_release_path(path);
932 
933 right:
934 	/* Search for a neighbor on the right. */
935 	if (end == block_group->key.objectid + block_group->key.offset)
936 		goto insert;
937 	key.objectid = end;
938 	key.type = (u8)-1;
939 	key.offset = (u64)-1;
940 
941 	ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
942 	if (ret)
943 		goto out;
944 
945 	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
946 
947 	if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) {
948 		ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY);
949 		btrfs_release_path(path);
950 		goto insert;
951 	}
952 
953 	found_start = key.objectid;
954 	found_end = key.objectid + key.offset;
955 	ASSERT(found_start >= block_group->key.objectid &&
956 	       found_end > block_group->key.objectid);
957 	ASSERT((found_start < start && found_end <= start) ||
958 	       (found_start >= end && found_end > end));
959 
960 	/*
961 	 * Delete the neighbor on the right and absorb it into the new key
962 	 * (cases 3 and 4).
963 	 */
964 	if (found_start == end) {
965 		ret = btrfs_del_item(trans, root, path);
966 		if (ret)
967 			goto out;
968 		new_key.offset += key.offset;
969 		new_extents--;
970 	}
971 	btrfs_release_path(path);
972 
973 insert:
974 	/* Insert the new key (cases 1-4). */
975 	ret = btrfs_insert_empty_item(trans, root, path, &new_key, 0);
976 	if (ret)
977 		goto out;
978 
979 	btrfs_release_path(path);
980 	ret = update_free_space_extent_count(trans, fs_info, block_group, path,
981 					     new_extents);
982 
983 out:
984 	return ret;
985 }
986 
987 int __add_to_free_space_tree(struct btrfs_trans_handle *trans,
988 			     struct btrfs_fs_info *fs_info,
989 			     struct btrfs_block_group_cache *block_group,
990 			     struct btrfs_path *path, u64 start, u64 size)
991 {
992 	struct btrfs_free_space_info *info;
993 	u32 flags;
994 	int ret;
995 
996 	if (block_group->needs_free_space) {
997 		ret = __add_block_group_free_space(trans, fs_info, block_group,
998 						   path);
999 		if (ret)
1000 			return ret;
1001 	}
1002 
1003 	info = search_free_space_info(NULL, fs_info, block_group, path, 0);
1004 	if (IS_ERR(info))
1005 		return PTR_ERR(info);
1006 	flags = btrfs_free_space_flags(path->nodes[0], info);
1007 	btrfs_release_path(path);
1008 
1009 	if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) {
1010 		return modify_free_space_bitmap(trans, fs_info, block_group,
1011 						path, start, size, 0);
1012 	} else {
1013 		return add_free_space_extent(trans, fs_info, block_group, path,
1014 					     start, size);
1015 	}
1016 }
1017 
1018 int add_to_free_space_tree(struct btrfs_trans_handle *trans,
1019 			   struct btrfs_fs_info *fs_info,
1020 			   u64 start, u64 size)
1021 {
1022 	struct btrfs_block_group_cache *block_group;
1023 	struct btrfs_path *path;
1024 	int ret;
1025 
1026 	if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
1027 		return 0;
1028 
1029 	path = btrfs_alloc_path();
1030 	if (!path) {
1031 		ret = -ENOMEM;
1032 		goto out;
1033 	}
1034 
1035 	block_group = btrfs_lookup_block_group(fs_info, start);
1036 	if (!block_group) {
1037 		ASSERT(0);
1038 		ret = -ENOENT;
1039 		goto out;
1040 	}
1041 
1042 	mutex_lock(&block_group->free_space_lock);
1043 	ret = __add_to_free_space_tree(trans, fs_info, block_group, path, start,
1044 				       size);
1045 	mutex_unlock(&block_group->free_space_lock);
1046 
1047 	btrfs_put_block_group(block_group);
1048 out:
1049 	btrfs_free_path(path);
1050 	if (ret)
1051 		btrfs_abort_transaction(trans, ret);
1052 	return ret;
1053 }
1054 
1055 /*
1056  * Populate the free space tree by walking the extent tree. Operations on the
1057  * extent tree that happen as a result of writes to the free space tree will go
1058  * through the normal add/remove hooks.
1059  */
1060 static int populate_free_space_tree(struct btrfs_trans_handle *trans,
1061 				    struct btrfs_fs_info *fs_info,
1062 				    struct btrfs_block_group_cache *block_group)
1063 {
1064 	struct btrfs_root *extent_root = fs_info->extent_root;
1065 	struct btrfs_path *path, *path2;
1066 	struct btrfs_key key;
1067 	u64 start, end;
1068 	int ret;
1069 
1070 	path = btrfs_alloc_path();
1071 	if (!path)
1072 		return -ENOMEM;
1073 	path->reada = 1;
1074 
1075 	path2 = btrfs_alloc_path();
1076 	if (!path2) {
1077 		btrfs_free_path(path);
1078 		return -ENOMEM;
1079 	}
1080 
1081 	ret = add_new_free_space_info(trans, fs_info, block_group, path2);
1082 	if (ret)
1083 		goto out;
1084 
1085 	mutex_lock(&block_group->free_space_lock);
1086 
1087 	/*
1088 	 * Iterate through all of the extent and metadata items in this block
1089 	 * group, adding the free space between them and the free space at the
1090 	 * end. Note that EXTENT_ITEM and METADATA_ITEM are less than
1091 	 * BLOCK_GROUP_ITEM, so an extent may precede the block group that it's
1092 	 * contained in.
1093 	 */
1094 	key.objectid = block_group->key.objectid;
1095 	key.type = BTRFS_EXTENT_ITEM_KEY;
1096 	key.offset = 0;
1097 
1098 	ret = btrfs_search_slot_for_read(extent_root, &key, path, 1, 0);
1099 	if (ret < 0)
1100 		goto out_locked;
1101 	ASSERT(ret == 0);
1102 
1103 	start = block_group->key.objectid;
1104 	end = block_group->key.objectid + block_group->key.offset;
1105 	while (1) {
1106 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1107 
1108 		if (key.type == BTRFS_EXTENT_ITEM_KEY ||
1109 		    key.type == BTRFS_METADATA_ITEM_KEY) {
1110 			if (key.objectid >= end)
1111 				break;
1112 
1113 			if (start < key.objectid) {
1114 				ret = __add_to_free_space_tree(trans, fs_info,
1115 							       block_group,
1116 							       path2, start,
1117 							       key.objectid -
1118 							       start);
1119 				if (ret)
1120 					goto out_locked;
1121 			}
1122 			start = key.objectid;
1123 			if (key.type == BTRFS_METADATA_ITEM_KEY)
1124 				start += fs_info->tree_root->nodesize;
1125 			else
1126 				start += key.offset;
1127 		} else if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
1128 			if (key.objectid != block_group->key.objectid)
1129 				break;
1130 		}
1131 
1132 		ret = btrfs_next_item(extent_root, path);
1133 		if (ret < 0)
1134 			goto out_locked;
1135 		if (ret)
1136 			break;
1137 	}
1138 	if (start < end) {
1139 		ret = __add_to_free_space_tree(trans, fs_info, block_group,
1140 					       path2, start, end - start);
1141 		if (ret)
1142 			goto out_locked;
1143 	}
1144 
1145 	ret = 0;
1146 out_locked:
1147 	mutex_unlock(&block_group->free_space_lock);
1148 out:
1149 	btrfs_free_path(path2);
1150 	btrfs_free_path(path);
1151 	return ret;
1152 }
1153 
1154 int btrfs_create_free_space_tree(struct btrfs_fs_info *fs_info)
1155 {
1156 	struct btrfs_trans_handle *trans;
1157 	struct btrfs_root *tree_root = fs_info->tree_root;
1158 	struct btrfs_root *free_space_root;
1159 	struct btrfs_block_group_cache *block_group;
1160 	struct rb_node *node;
1161 	int ret;
1162 
1163 	trans = btrfs_start_transaction(tree_root, 0);
1164 	if (IS_ERR(trans))
1165 		return PTR_ERR(trans);
1166 
1167 	set_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1168 	free_space_root = btrfs_create_tree(trans, fs_info,
1169 					    BTRFS_FREE_SPACE_TREE_OBJECTID);
1170 	if (IS_ERR(free_space_root)) {
1171 		ret = PTR_ERR(free_space_root);
1172 		goto abort;
1173 	}
1174 	fs_info->free_space_root = free_space_root;
1175 
1176 	node = rb_first(&fs_info->block_group_cache_tree);
1177 	while (node) {
1178 		block_group = rb_entry(node, struct btrfs_block_group_cache,
1179 				       cache_node);
1180 		ret = populate_free_space_tree(trans, fs_info, block_group);
1181 		if (ret)
1182 			goto abort;
1183 		node = rb_next(node);
1184 	}
1185 
1186 	btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE);
1187 	btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID);
1188 	clear_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1189 
1190 	ret = btrfs_commit_transaction(trans, tree_root);
1191 	if (ret)
1192 		return ret;
1193 
1194 	return 0;
1195 
1196 abort:
1197 	clear_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1198 	btrfs_abort_transaction(trans, ret);
1199 	btrfs_end_transaction(trans, tree_root);
1200 	return ret;
1201 }
1202 
1203 static int clear_free_space_tree(struct btrfs_trans_handle *trans,
1204 				 struct btrfs_root *root)
1205 {
1206 	struct btrfs_path *path;
1207 	struct btrfs_key key;
1208 	int nr;
1209 	int ret;
1210 
1211 	path = btrfs_alloc_path();
1212 	if (!path)
1213 		return -ENOMEM;
1214 
1215 	path->leave_spinning = 1;
1216 
1217 	key.objectid = 0;
1218 	key.type = 0;
1219 	key.offset = 0;
1220 
1221 	while (1) {
1222 		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1223 		if (ret < 0)
1224 			goto out;
1225 
1226 		nr = btrfs_header_nritems(path->nodes[0]);
1227 		if (!nr)
1228 			break;
1229 
1230 		path->slots[0] = 0;
1231 		ret = btrfs_del_items(trans, root, path, 0, nr);
1232 		if (ret)
1233 			goto out;
1234 
1235 		btrfs_release_path(path);
1236 	}
1237 
1238 	ret = 0;
1239 out:
1240 	btrfs_free_path(path);
1241 	return ret;
1242 }
1243 
1244 int btrfs_clear_free_space_tree(struct btrfs_fs_info *fs_info)
1245 {
1246 	struct btrfs_trans_handle *trans;
1247 	struct btrfs_root *tree_root = fs_info->tree_root;
1248 	struct btrfs_root *free_space_root = fs_info->free_space_root;
1249 	int ret;
1250 
1251 	trans = btrfs_start_transaction(tree_root, 0);
1252 	if (IS_ERR(trans))
1253 		return PTR_ERR(trans);
1254 
1255 	btrfs_clear_fs_compat_ro(fs_info, FREE_SPACE_TREE);
1256 	btrfs_clear_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID);
1257 	fs_info->free_space_root = NULL;
1258 
1259 	ret = clear_free_space_tree(trans, free_space_root);
1260 	if (ret)
1261 		goto abort;
1262 
1263 	ret = btrfs_del_root(trans, tree_root, &free_space_root->root_key);
1264 	if (ret)
1265 		goto abort;
1266 
1267 	list_del(&free_space_root->dirty_list);
1268 
1269 	btrfs_tree_lock(free_space_root->node);
1270 	clean_tree_block(trans, tree_root->fs_info, free_space_root->node);
1271 	btrfs_tree_unlock(free_space_root->node);
1272 	btrfs_free_tree_block(trans, free_space_root, free_space_root->node,
1273 			      0, 1);
1274 
1275 	free_extent_buffer(free_space_root->node);
1276 	free_extent_buffer(free_space_root->commit_root);
1277 	kfree(free_space_root);
1278 
1279 	ret = btrfs_commit_transaction(trans, tree_root);
1280 	if (ret)
1281 		return ret;
1282 
1283 	return 0;
1284 
1285 abort:
1286 	btrfs_abort_transaction(trans, ret);
1287 	btrfs_end_transaction(trans, tree_root);
1288 	return ret;
1289 }
1290 
1291 static int __add_block_group_free_space(struct btrfs_trans_handle *trans,
1292 					struct btrfs_fs_info *fs_info,
1293 					struct btrfs_block_group_cache *block_group,
1294 					struct btrfs_path *path)
1295 {
1296 	u64 start, end;
1297 	int ret;
1298 
1299 	start = block_group->key.objectid;
1300 	end = block_group->key.objectid + block_group->key.offset;
1301 
1302 	block_group->needs_free_space = 0;
1303 
1304 	ret = add_new_free_space_info(trans, fs_info, block_group, path);
1305 	if (ret)
1306 		return ret;
1307 
1308 	return __add_to_free_space_tree(trans, fs_info, block_group, path,
1309 					block_group->key.objectid,
1310 					block_group->key.offset);
1311 }
1312 
1313 int add_block_group_free_space(struct btrfs_trans_handle *trans,
1314 			       struct btrfs_fs_info *fs_info,
1315 			       struct btrfs_block_group_cache *block_group)
1316 {
1317 	struct btrfs_path *path = NULL;
1318 	int ret = 0;
1319 
1320 	if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
1321 		return 0;
1322 
1323 	mutex_lock(&block_group->free_space_lock);
1324 	if (!block_group->needs_free_space)
1325 		goto out;
1326 
1327 	path = btrfs_alloc_path();
1328 	if (!path) {
1329 		ret = -ENOMEM;
1330 		goto out;
1331 	}
1332 
1333 	ret = __add_block_group_free_space(trans, fs_info, block_group, path);
1334 
1335 out:
1336 	btrfs_free_path(path);
1337 	mutex_unlock(&block_group->free_space_lock);
1338 	if (ret)
1339 		btrfs_abort_transaction(trans, ret);
1340 	return ret;
1341 }
1342 
1343 int remove_block_group_free_space(struct btrfs_trans_handle *trans,
1344 				  struct btrfs_fs_info *fs_info,
1345 				  struct btrfs_block_group_cache *block_group)
1346 {
1347 	struct btrfs_root *root = fs_info->free_space_root;
1348 	struct btrfs_path *path;
1349 	struct btrfs_key key, found_key;
1350 	struct extent_buffer *leaf;
1351 	u64 start, end;
1352 	int done = 0, nr;
1353 	int ret;
1354 
1355 	if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
1356 		return 0;
1357 
1358 	if (block_group->needs_free_space) {
1359 		/* We never added this block group to the free space tree. */
1360 		return 0;
1361 	}
1362 
1363 	path = btrfs_alloc_path();
1364 	if (!path) {
1365 		ret = -ENOMEM;
1366 		goto out;
1367 	}
1368 
1369 	start = block_group->key.objectid;
1370 	end = block_group->key.objectid + block_group->key.offset;
1371 
1372 	key.objectid = end - 1;
1373 	key.type = (u8)-1;
1374 	key.offset = (u64)-1;
1375 
1376 	while (!done) {
1377 		ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
1378 		if (ret)
1379 			goto out;
1380 
1381 		leaf = path->nodes[0];
1382 		nr = 0;
1383 		path->slots[0]++;
1384 		while (path->slots[0] > 0) {
1385 			btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
1386 
1387 			if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
1388 				ASSERT(found_key.objectid == block_group->key.objectid);
1389 				ASSERT(found_key.offset == block_group->key.offset);
1390 				done = 1;
1391 				nr++;
1392 				path->slots[0]--;
1393 				break;
1394 			} else if (found_key.type == BTRFS_FREE_SPACE_EXTENT_KEY ||
1395 				   found_key.type == BTRFS_FREE_SPACE_BITMAP_KEY) {
1396 				ASSERT(found_key.objectid >= start);
1397 				ASSERT(found_key.objectid < end);
1398 				ASSERT(found_key.objectid + found_key.offset <= end);
1399 				nr++;
1400 				path->slots[0]--;
1401 			} else {
1402 				ASSERT(0);
1403 			}
1404 		}
1405 
1406 		ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
1407 		if (ret)
1408 			goto out;
1409 		btrfs_release_path(path);
1410 	}
1411 
1412 	ret = 0;
1413 out:
1414 	btrfs_free_path(path);
1415 	if (ret)
1416 		btrfs_abort_transaction(trans, ret);
1417 	return ret;
1418 }
1419 
1420 static int load_free_space_bitmaps(struct btrfs_caching_control *caching_ctl,
1421 				   struct btrfs_path *path,
1422 				   u32 expected_extent_count)
1423 {
1424 	struct btrfs_block_group_cache *block_group;
1425 	struct btrfs_fs_info *fs_info;
1426 	struct btrfs_root *root;
1427 	struct btrfs_key key;
1428 	int prev_bit = 0, bit;
1429 	/* Initialize to silence GCC. */
1430 	u64 extent_start = 0;
1431 	u64 end, offset;
1432 	u64 total_found = 0;
1433 	u32 extent_count = 0;
1434 	int ret;
1435 
1436 	block_group = caching_ctl->block_group;
1437 	fs_info = block_group->fs_info;
1438 	root = fs_info->free_space_root;
1439 
1440 	end = block_group->key.objectid + block_group->key.offset;
1441 
1442 	while (1) {
1443 		ret = btrfs_next_item(root, path);
1444 		if (ret < 0)
1445 			goto out;
1446 		if (ret)
1447 			break;
1448 
1449 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1450 
1451 		if (key.type == BTRFS_FREE_SPACE_INFO_KEY)
1452 			break;
1453 
1454 		ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
1455 		ASSERT(key.objectid < end && key.objectid + key.offset <= end);
1456 
1457 		caching_ctl->progress = key.objectid;
1458 
1459 		offset = key.objectid;
1460 		while (offset < key.objectid + key.offset) {
1461 			bit = free_space_test_bit(block_group, path, offset);
1462 			if (prev_bit == 0 && bit == 1) {
1463 				extent_start = offset;
1464 			} else if (prev_bit == 1 && bit == 0) {
1465 				total_found += add_new_free_space(block_group,
1466 								  fs_info,
1467 								  extent_start,
1468 								  offset);
1469 				if (total_found > CACHING_CTL_WAKE_UP) {
1470 					total_found = 0;
1471 					wake_up(&caching_ctl->wait);
1472 				}
1473 				extent_count++;
1474 			}
1475 			prev_bit = bit;
1476 			offset += block_group->sectorsize;
1477 		}
1478 	}
1479 	if (prev_bit == 1) {
1480 		total_found += add_new_free_space(block_group, fs_info,
1481 						  extent_start, end);
1482 		extent_count++;
1483 	}
1484 
1485 	if (extent_count != expected_extent_count) {
1486 		btrfs_err(fs_info,
1487 			  "incorrect extent count for %llu; counted %u, expected %u",
1488 			  block_group->key.objectid, extent_count,
1489 			  expected_extent_count);
1490 		ASSERT(0);
1491 		ret = -EIO;
1492 		goto out;
1493 	}
1494 
1495 	caching_ctl->progress = (u64)-1;
1496 
1497 	ret = 0;
1498 out:
1499 	return ret;
1500 }
1501 
1502 static int load_free_space_extents(struct btrfs_caching_control *caching_ctl,
1503 				   struct btrfs_path *path,
1504 				   u32 expected_extent_count)
1505 {
1506 	struct btrfs_block_group_cache *block_group;
1507 	struct btrfs_fs_info *fs_info;
1508 	struct btrfs_root *root;
1509 	struct btrfs_key key;
1510 	u64 end;
1511 	u64 total_found = 0;
1512 	u32 extent_count = 0;
1513 	int ret;
1514 
1515 	block_group = caching_ctl->block_group;
1516 	fs_info = block_group->fs_info;
1517 	root = fs_info->free_space_root;
1518 
1519 	end = block_group->key.objectid + block_group->key.offset;
1520 
1521 	while (1) {
1522 		ret = btrfs_next_item(root, path);
1523 		if (ret < 0)
1524 			goto out;
1525 		if (ret)
1526 			break;
1527 
1528 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1529 
1530 		if (key.type == BTRFS_FREE_SPACE_INFO_KEY)
1531 			break;
1532 
1533 		ASSERT(key.type == BTRFS_FREE_SPACE_EXTENT_KEY);
1534 		ASSERT(key.objectid < end && key.objectid + key.offset <= end);
1535 
1536 		caching_ctl->progress = key.objectid;
1537 
1538 		total_found += add_new_free_space(block_group, fs_info,
1539 						  key.objectid,
1540 						  key.objectid + key.offset);
1541 		if (total_found > CACHING_CTL_WAKE_UP) {
1542 			total_found = 0;
1543 			wake_up(&caching_ctl->wait);
1544 		}
1545 		extent_count++;
1546 	}
1547 
1548 	if (extent_count != expected_extent_count) {
1549 		btrfs_err(fs_info,
1550 			  "incorrect extent count for %llu; counted %u, expected %u",
1551 			  block_group->key.objectid, extent_count,
1552 			  expected_extent_count);
1553 		ASSERT(0);
1554 		ret = -EIO;
1555 		goto out;
1556 	}
1557 
1558 	caching_ctl->progress = (u64)-1;
1559 
1560 	ret = 0;
1561 out:
1562 	return ret;
1563 }
1564 
1565 int load_free_space_tree(struct btrfs_caching_control *caching_ctl)
1566 {
1567 	struct btrfs_block_group_cache *block_group;
1568 	struct btrfs_fs_info *fs_info;
1569 	struct btrfs_free_space_info *info;
1570 	struct btrfs_path *path;
1571 	u32 extent_count, flags;
1572 	int ret;
1573 
1574 	block_group = caching_ctl->block_group;
1575 	fs_info = block_group->fs_info;
1576 
1577 	path = btrfs_alloc_path();
1578 	if (!path)
1579 		return -ENOMEM;
1580 
1581 	/*
1582 	 * Just like caching_thread() doesn't want to deadlock on the extent
1583 	 * tree, we don't want to deadlock on the free space tree.
1584 	 */
1585 	path->skip_locking = 1;
1586 	path->search_commit_root = 1;
1587 	path->reada = 1;
1588 
1589 	info = search_free_space_info(NULL, fs_info, block_group, path, 0);
1590 	if (IS_ERR(info)) {
1591 		ret = PTR_ERR(info);
1592 		goto out;
1593 	}
1594 	extent_count = btrfs_free_space_extent_count(path->nodes[0], info);
1595 	flags = btrfs_free_space_flags(path->nodes[0], info);
1596 
1597 	/*
1598 	 * We left path pointing to the free space info item, so now
1599 	 * load_free_space_foo can just iterate through the free space tree from
1600 	 * there.
1601 	 */
1602 	if (flags & BTRFS_FREE_SPACE_USING_BITMAPS)
1603 		ret = load_free_space_bitmaps(caching_ctl, path, extent_count);
1604 	else
1605 		ret = load_free_space_extents(caching_ctl, path, extent_count);
1606 
1607 out:
1608 	btrfs_free_path(path);
1609 	return ret;
1610 }
1611