xref: /linux/fs/btrfs/free-space-tree.c (revision 4c06e63b92038fadb566b652ec3ec04e228931e8)
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 
btrfs_free_space_root(struct btrfs_block_group * block_group)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 
set_free_space_tree_thresholds(struct btrfs_block_group * cache)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 
add_new_free_space_info(struct btrfs_trans_handle * trans,struct btrfs_block_group * block_group,struct btrfs_path * path)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 
93 	ret = 0;
94 out:
95 	btrfs_release_path(path);
96 	return ret;
97 }
98 
99 EXPORT_FOR_TESTS
search_free_space_info(struct btrfs_trans_handle * trans,struct btrfs_block_group * block_group,struct btrfs_path * path,int cow)100 struct btrfs_free_space_info *search_free_space_info(
101 		struct btrfs_trans_handle *trans,
102 		struct btrfs_block_group *block_group,
103 		struct btrfs_path *path, int cow)
104 {
105 	struct btrfs_fs_info *fs_info = block_group->fs_info;
106 	struct btrfs_root *root = btrfs_free_space_root(block_group);
107 	struct btrfs_key key;
108 	int ret;
109 
110 	key.objectid = block_group->start;
111 	key.type = BTRFS_FREE_SPACE_INFO_KEY;
112 	key.offset = block_group->length;
113 
114 	ret = btrfs_search_slot(trans, root, &key, path, 0, cow);
115 	if (ret < 0)
116 		return ERR_PTR(ret);
117 	if (ret != 0) {
118 		btrfs_warn(fs_info, "missing free space info for %llu",
119 			   block_group->start);
120 		DEBUG_WARN();
121 		return ERR_PTR(-ENOENT);
122 	}
123 
124 	return btrfs_item_ptr(path->nodes[0], path->slots[0],
125 			      struct btrfs_free_space_info);
126 }
127 
128 /*
129  * btrfs_search_slot() but we're looking for the greatest key less than the
130  * passed key.
131  */
btrfs_search_prev_slot(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_key * key,struct btrfs_path * p,int ins_len,int cow)132 static int btrfs_search_prev_slot(struct btrfs_trans_handle *trans,
133 				  struct btrfs_root *root,
134 				  struct btrfs_key *key, struct btrfs_path *p,
135 				  int ins_len, int cow)
136 {
137 	int ret;
138 
139 	ret = btrfs_search_slot(trans, root, key, p, ins_len, cow);
140 	if (ret < 0)
141 		return ret;
142 
143 	if (ret == 0) {
144 		DEBUG_WARN();
145 		return -EIO;
146 	}
147 
148 	if (p->slots[0] == 0) {
149 		DEBUG_WARN("no previous slot found");
150 		return -EIO;
151 	}
152 	p->slots[0]--;
153 
154 	return 0;
155 }
156 
free_space_bitmap_size(const struct btrfs_fs_info * fs_info,u64 size)157 static inline u32 free_space_bitmap_size(const struct btrfs_fs_info *fs_info,
158 					 u64 size)
159 {
160 	return DIV_ROUND_UP(size >> fs_info->sectorsize_bits, BITS_PER_BYTE);
161 }
162 
alloc_bitmap(u32 bitmap_size)163 static unsigned long *alloc_bitmap(u32 bitmap_size)
164 {
165 	unsigned long *ret;
166 	unsigned int nofs_flag;
167 	u32 bitmap_rounded_size = round_up(bitmap_size, sizeof(unsigned long));
168 
169 	/*
170 	 * GFP_NOFS doesn't work with kvmalloc(), but we really can't recurse
171 	 * into the filesystem as the free space bitmap can be modified in the
172 	 * critical section of a transaction commit.
173 	 *
174 	 * TODO: push the memalloc_nofs_{save,restore}() to the caller where we
175 	 * know that recursion is unsafe.
176 	 */
177 	nofs_flag = memalloc_nofs_save();
178 	ret = kvzalloc(bitmap_rounded_size, GFP_KERNEL);
179 	memalloc_nofs_restore(nofs_flag);
180 	return ret;
181 }
182 
le_bitmap_set(unsigned long * map,unsigned int start,int len)183 static void le_bitmap_set(unsigned long *map, unsigned int start, int len)
184 {
185 	u8 *p = ((u8 *)map) + BIT_BYTE(start);
186 	const unsigned int size = start + len;
187 	int bits_to_set = BITS_PER_BYTE - (start % BITS_PER_BYTE);
188 	u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(start);
189 
190 	while (len - bits_to_set >= 0) {
191 		*p |= mask_to_set;
192 		len -= bits_to_set;
193 		bits_to_set = BITS_PER_BYTE;
194 		mask_to_set = ~0;
195 		p++;
196 	}
197 	if (len) {
198 		mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
199 		*p |= mask_to_set;
200 	}
201 }
202 
203 EXPORT_FOR_TESTS
convert_free_space_to_bitmaps(struct btrfs_trans_handle * trans,struct btrfs_block_group * block_group,struct btrfs_path * path)204 int convert_free_space_to_bitmaps(struct btrfs_trans_handle *trans,
205 				  struct btrfs_block_group *block_group,
206 				  struct btrfs_path *path)
207 {
208 	struct btrfs_fs_info *fs_info = trans->fs_info;
209 	struct btrfs_root *root = btrfs_free_space_root(block_group);
210 	struct btrfs_free_space_info *info;
211 	struct btrfs_key key, found_key;
212 	struct extent_buffer *leaf;
213 	unsigned long *bitmap;
214 	char *bitmap_cursor;
215 	u64 start, end;
216 	u64 bitmap_range, i;
217 	u32 bitmap_size, flags, expected_extent_count;
218 	u32 extent_count = 0;
219 	int done = 0, nr;
220 	int ret;
221 
222 	bitmap_size = free_space_bitmap_size(fs_info, block_group->length);
223 	bitmap = alloc_bitmap(bitmap_size);
224 	if (!bitmap) {
225 		ret = -ENOMEM;
226 		btrfs_abort_transaction(trans, ret);
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 			btrfs_abort_transaction(trans, ret);
241 			goto out;
242 		}
243 
244 		leaf = path->nodes[0];
245 		nr = 0;
246 		path->slots[0]++;
247 		while (path->slots[0] > 0) {
248 			btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
249 
250 			if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
251 				ASSERT(found_key.objectid == block_group->start);
252 				ASSERT(found_key.offset == block_group->length);
253 				done = 1;
254 				break;
255 			} else if (found_key.type == BTRFS_FREE_SPACE_EXTENT_KEY) {
256 				u64 first, last;
257 
258 				ASSERT(found_key.objectid >= start);
259 				ASSERT(found_key.objectid < end);
260 				ASSERT(found_key.objectid + found_key.offset <= end);
261 
262 				first = div_u64(found_key.objectid - start,
263 						fs_info->sectorsize);
264 				last = div_u64(found_key.objectid + found_key.offset - start,
265 					       fs_info->sectorsize);
266 				le_bitmap_set(bitmap, first, last - first);
267 
268 				extent_count++;
269 				nr++;
270 				path->slots[0]--;
271 			} else {
272 				ASSERT(0);
273 			}
274 		}
275 
276 		ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
277 		if (ret) {
278 			btrfs_abort_transaction(trans, ret);
279 			goto out;
280 		}
281 		btrfs_release_path(path);
282 	}
283 
284 	info = search_free_space_info(trans, block_group, path, 1);
285 	if (IS_ERR(info)) {
286 		ret = PTR_ERR(info);
287 		btrfs_abort_transaction(trans, ret);
288 		goto out;
289 	}
290 	leaf = path->nodes[0];
291 	flags = btrfs_free_space_flags(leaf, info);
292 	flags |= BTRFS_FREE_SPACE_USING_BITMAPS;
293 	btrfs_set_free_space_flags(leaf, info, flags);
294 	expected_extent_count = btrfs_free_space_extent_count(leaf, info);
295 	btrfs_release_path(path);
296 
297 	if (extent_count != expected_extent_count) {
298 		btrfs_err(fs_info,
299 			  "incorrect extent count for %llu; counted %u, expected %u",
300 			  block_group->start, extent_count,
301 			  expected_extent_count);
302 		ret = -EIO;
303 		btrfs_abort_transaction(trans, ret);
304 		goto out;
305 	}
306 
307 	bitmap_cursor = (char *)bitmap;
308 	bitmap_range = fs_info->sectorsize * BTRFS_FREE_SPACE_BITMAP_BITS;
309 	i = start;
310 	while (i < end) {
311 		unsigned long ptr;
312 		u64 extent_size;
313 		u32 data_size;
314 
315 		extent_size = min(end - i, bitmap_range);
316 		data_size = free_space_bitmap_size(fs_info, extent_size);
317 
318 		key.objectid = i;
319 		key.type = BTRFS_FREE_SPACE_BITMAP_KEY;
320 		key.offset = extent_size;
321 
322 		ret = btrfs_insert_empty_item(trans, root, path, &key,
323 					      data_size);
324 		if (ret) {
325 			btrfs_abort_transaction(trans, ret);
326 			goto out;
327 		}
328 
329 		leaf = path->nodes[0];
330 		ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
331 		write_extent_buffer(leaf, bitmap_cursor, ptr,
332 				    data_size);
333 		btrfs_release_path(path);
334 
335 		i += extent_size;
336 		bitmap_cursor += data_size;
337 	}
338 
339 	ret = 0;
340 out:
341 	kvfree(bitmap);
342 	return ret;
343 }
344 
345 EXPORT_FOR_TESTS
convert_free_space_to_extents(struct btrfs_trans_handle * trans,struct btrfs_block_group * block_group,struct btrfs_path * path)346 int convert_free_space_to_extents(struct btrfs_trans_handle *trans,
347 				  struct btrfs_block_group *block_group,
348 				  struct btrfs_path *path)
349 {
350 	struct btrfs_fs_info *fs_info = trans->fs_info;
351 	struct btrfs_root *root = btrfs_free_space_root(block_group);
352 	struct btrfs_free_space_info *info;
353 	struct btrfs_key key, found_key;
354 	struct extent_buffer *leaf;
355 	unsigned long *bitmap;
356 	u64 start, end;
357 	u32 bitmap_size, flags, expected_extent_count;
358 	unsigned long nrbits, start_bit, end_bit;
359 	u32 extent_count = 0;
360 	int done = 0, nr;
361 	int ret;
362 
363 	bitmap_size = free_space_bitmap_size(fs_info, block_group->length);
364 	bitmap = alloc_bitmap(bitmap_size);
365 	if (!bitmap) {
366 		ret = -ENOMEM;
367 		btrfs_abort_transaction(trans, ret);
368 		goto out;
369 	}
370 
371 	start = block_group->start;
372 	end = block_group->start + block_group->length;
373 
374 	key.objectid = end - 1;
375 	key.type = (u8)-1;
376 	key.offset = (u64)-1;
377 
378 	while (!done) {
379 		ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
380 		if (ret) {
381 			btrfs_abort_transaction(trans, ret);
382 			goto out;
383 		}
384 
385 		leaf = path->nodes[0];
386 		nr = 0;
387 		path->slots[0]++;
388 		while (path->slots[0] > 0) {
389 			btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
390 
391 			if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
392 				ASSERT(found_key.objectid == block_group->start);
393 				ASSERT(found_key.offset == block_group->length);
394 				done = 1;
395 				break;
396 			} else if (found_key.type == BTRFS_FREE_SPACE_BITMAP_KEY) {
397 				unsigned long ptr;
398 				char *bitmap_cursor;
399 				u32 bitmap_pos, data_size;
400 
401 				ASSERT(found_key.objectid >= start);
402 				ASSERT(found_key.objectid < end);
403 				ASSERT(found_key.objectid + found_key.offset <= end);
404 
405 				bitmap_pos = div_u64(found_key.objectid - start,
406 						     fs_info->sectorsize *
407 						     BITS_PER_BYTE);
408 				bitmap_cursor = ((char *)bitmap) + bitmap_pos;
409 				data_size = free_space_bitmap_size(fs_info,
410 								found_key.offset);
411 
412 				ptr = btrfs_item_ptr_offset(leaf, path->slots[0] - 1);
413 				read_extent_buffer(leaf, bitmap_cursor, ptr,
414 						   data_size);
415 
416 				nr++;
417 				path->slots[0]--;
418 			} else {
419 				ASSERT(0);
420 			}
421 		}
422 
423 		ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
424 		if (ret) {
425 			btrfs_abort_transaction(trans, ret);
426 			goto out;
427 		}
428 		btrfs_release_path(path);
429 	}
430 
431 	info = search_free_space_info(trans, block_group, path, 1);
432 	if (IS_ERR(info)) {
433 		ret = PTR_ERR(info);
434 		btrfs_abort_transaction(trans, ret);
435 		goto out;
436 	}
437 	leaf = path->nodes[0];
438 	flags = btrfs_free_space_flags(leaf, info);
439 	flags &= ~BTRFS_FREE_SPACE_USING_BITMAPS;
440 	btrfs_set_free_space_flags(leaf, info, flags);
441 	expected_extent_count = btrfs_free_space_extent_count(leaf, info);
442 	btrfs_release_path(path);
443 
444 	nrbits = block_group->length >> block_group->fs_info->sectorsize_bits;
445 	start_bit = find_next_bit_le(bitmap, nrbits, 0);
446 
447 	while (start_bit < nrbits) {
448 		end_bit = find_next_zero_bit_le(bitmap, nrbits, start_bit);
449 		ASSERT(start_bit < end_bit);
450 
451 		key.objectid = start + start_bit * block_group->fs_info->sectorsize;
452 		key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
453 		key.offset = (end_bit - start_bit) * block_group->fs_info->sectorsize;
454 
455 		ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
456 		if (ret) {
457 			btrfs_abort_transaction(trans, ret);
458 			goto out;
459 		}
460 		btrfs_release_path(path);
461 
462 		extent_count++;
463 
464 		start_bit = find_next_bit_le(bitmap, nrbits, end_bit);
465 	}
466 
467 	if (extent_count != expected_extent_count) {
468 		btrfs_err(fs_info,
469 			  "incorrect extent count for %llu; counted %u, expected %u",
470 			  block_group->start, extent_count,
471 			  expected_extent_count);
472 		ret = -EIO;
473 		btrfs_abort_transaction(trans, ret);
474 		goto out;
475 	}
476 
477 	ret = 0;
478 out:
479 	kvfree(bitmap);
480 	return ret;
481 }
482 
update_free_space_extent_count(struct btrfs_trans_handle * trans,struct btrfs_block_group * block_group,struct btrfs_path * path,int new_extents)483 static int update_free_space_extent_count(struct btrfs_trans_handle *trans,
484 					  struct btrfs_block_group *block_group,
485 					  struct btrfs_path *path,
486 					  int new_extents)
487 {
488 	struct btrfs_free_space_info *info;
489 	u32 flags;
490 	u32 extent_count;
491 	int ret = 0;
492 
493 	if (new_extents == 0)
494 		return 0;
495 
496 	info = search_free_space_info(trans, block_group, path, 1);
497 	if (IS_ERR(info)) {
498 		ret = PTR_ERR(info);
499 		goto out;
500 	}
501 	flags = btrfs_free_space_flags(path->nodes[0], info);
502 	extent_count = btrfs_free_space_extent_count(path->nodes[0], info);
503 
504 	extent_count += new_extents;
505 	btrfs_set_free_space_extent_count(path->nodes[0], info, extent_count);
506 	btrfs_release_path(path);
507 
508 	if (!(flags & BTRFS_FREE_SPACE_USING_BITMAPS) &&
509 	    extent_count > block_group->bitmap_high_thresh) {
510 		ret = convert_free_space_to_bitmaps(trans, block_group, path);
511 	} else if ((flags & BTRFS_FREE_SPACE_USING_BITMAPS) &&
512 		   extent_count < block_group->bitmap_low_thresh) {
513 		ret = convert_free_space_to_extents(trans, block_group, path);
514 	}
515 
516 out:
517 	return ret;
518 }
519 
520 EXPORT_FOR_TESTS
free_space_test_bit(struct btrfs_block_group * block_group,struct btrfs_path * path,u64 offset)521 int free_space_test_bit(struct btrfs_block_group *block_group,
522 			struct btrfs_path *path, u64 offset)
523 {
524 	struct extent_buffer *leaf;
525 	struct btrfs_key key;
526 	u64 found_start, found_end;
527 	unsigned long ptr, i;
528 
529 	leaf = path->nodes[0];
530 	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
531 	ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
532 
533 	found_start = key.objectid;
534 	found_end = key.objectid + key.offset;
535 	ASSERT(offset >= found_start && offset < found_end);
536 
537 	ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
538 	i = div_u64(offset - found_start,
539 		    block_group->fs_info->sectorsize);
540 	return !!extent_buffer_test_bit(leaf, ptr, i);
541 }
542 
free_space_set_bits(struct btrfs_trans_handle * trans,struct btrfs_block_group * block_group,struct btrfs_path * path,u64 * start,u64 * size,int bit)543 static void free_space_set_bits(struct btrfs_trans_handle *trans,
544 				struct btrfs_block_group *block_group,
545 				struct btrfs_path *path, u64 *start, u64 *size,
546 				int bit)
547 {
548 	struct btrfs_fs_info *fs_info = block_group->fs_info;
549 	struct extent_buffer *leaf;
550 	struct btrfs_key key;
551 	u64 end = *start + *size;
552 	u64 found_start, found_end;
553 	unsigned long ptr, first, last;
554 
555 	leaf = path->nodes[0];
556 	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
557 	ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
558 
559 	found_start = key.objectid;
560 	found_end = key.objectid + key.offset;
561 	ASSERT(*start >= found_start && *start < found_end);
562 	ASSERT(end > found_start);
563 
564 	if (end > found_end)
565 		end = found_end;
566 
567 	ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
568 	first = (*start - found_start) >> fs_info->sectorsize_bits;
569 	last = (end - found_start) >> fs_info->sectorsize_bits;
570 	if (bit)
571 		extent_buffer_bitmap_set(leaf, ptr, first, last - first);
572 	else
573 		extent_buffer_bitmap_clear(leaf, ptr, first, last - first);
574 	btrfs_mark_buffer_dirty(trans, leaf);
575 
576 	*size -= end - *start;
577 	*start = end;
578 }
579 
580 /*
581  * We can't use btrfs_next_item() in modify_free_space_bitmap() because
582  * btrfs_next_leaf() doesn't get the path for writing. We can forgo the fancy
583  * tree walking in btrfs_next_leaf() anyways because we know exactly what we're
584  * looking for.
585  */
free_space_next_bitmap(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * p)586 static int free_space_next_bitmap(struct btrfs_trans_handle *trans,
587 				  struct btrfs_root *root, struct btrfs_path *p)
588 {
589 	struct btrfs_key key;
590 
591 	if (p->slots[0] + 1 < btrfs_header_nritems(p->nodes[0])) {
592 		p->slots[0]++;
593 		return 0;
594 	}
595 
596 	btrfs_item_key_to_cpu(p->nodes[0], &key, p->slots[0]);
597 	btrfs_release_path(p);
598 
599 	key.objectid += key.offset;
600 	key.type = (u8)-1;
601 	key.offset = (u64)-1;
602 
603 	return btrfs_search_prev_slot(trans, root, &key, p, 0, 1);
604 }
605 
606 /*
607  * If remove is 1, then we are removing free space, thus clearing bits in the
608  * bitmap. If remove is 0, then we are adding free space, thus setting bits in
609  * the bitmap.
610  */
modify_free_space_bitmap(struct btrfs_trans_handle * trans,struct btrfs_block_group * block_group,struct btrfs_path * path,u64 start,u64 size,int remove)611 static int modify_free_space_bitmap(struct btrfs_trans_handle *trans,
612 				    struct btrfs_block_group *block_group,
613 				    struct btrfs_path *path,
614 				    u64 start, u64 size, int remove)
615 {
616 	struct btrfs_root *root = btrfs_free_space_root(block_group);
617 	struct btrfs_key key;
618 	u64 end = start + size;
619 	u64 cur_start, cur_size;
620 	int prev_bit, next_bit;
621 	int new_extents;
622 	int ret;
623 
624 	/*
625 	 * Read the bit for the block immediately before the extent of space if
626 	 * that block is within the block group.
627 	 */
628 	if (start > block_group->start) {
629 		u64 prev_block = start - block_group->fs_info->sectorsize;
630 
631 		key.objectid = prev_block;
632 		key.type = (u8)-1;
633 		key.offset = (u64)-1;
634 
635 		ret = btrfs_search_prev_slot(trans, root, &key, path, 0, 1);
636 		if (ret)
637 			goto out;
638 
639 		prev_bit = free_space_test_bit(block_group, path, prev_block);
640 
641 		/* The previous block may have been in the previous bitmap. */
642 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
643 		if (start >= key.objectid + key.offset) {
644 			ret = free_space_next_bitmap(trans, root, path);
645 			if (ret)
646 				goto out;
647 		}
648 	} else {
649 		key.objectid = start;
650 		key.type = (u8)-1;
651 		key.offset = (u64)-1;
652 
653 		ret = btrfs_search_prev_slot(trans, root, &key, path, 0, 1);
654 		if (ret)
655 			goto out;
656 
657 		prev_bit = -1;
658 	}
659 
660 	/*
661 	 * Iterate over all of the bitmaps overlapped by the extent of space,
662 	 * clearing/setting bits as required.
663 	 */
664 	cur_start = start;
665 	cur_size = size;
666 	while (1) {
667 		free_space_set_bits(trans, block_group, path, &cur_start, &cur_size,
668 				    !remove);
669 		if (cur_size == 0)
670 			break;
671 		ret = free_space_next_bitmap(trans, root, path);
672 		if (ret)
673 			goto out;
674 	}
675 
676 	/*
677 	 * Read the bit for the block immediately after the extent of space if
678 	 * that block is within the block group.
679 	 */
680 	if (end < block_group->start + block_group->length) {
681 		/* The next block may be in the next bitmap. */
682 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
683 		if (end >= key.objectid + key.offset) {
684 			ret = free_space_next_bitmap(trans, root, path);
685 			if (ret)
686 				goto out;
687 		}
688 
689 		next_bit = free_space_test_bit(block_group, path, end);
690 	} else {
691 		next_bit = -1;
692 	}
693 
694 	if (remove) {
695 		new_extents = -1;
696 		if (prev_bit == 1) {
697 			/* Leftover on the left. */
698 			new_extents++;
699 		}
700 		if (next_bit == 1) {
701 			/* Leftover on the right. */
702 			new_extents++;
703 		}
704 	} else {
705 		new_extents = 1;
706 		if (prev_bit == 1) {
707 			/* Merging with neighbor on the left. */
708 			new_extents--;
709 		}
710 		if (next_bit == 1) {
711 			/* Merging with neighbor on the right. */
712 			new_extents--;
713 		}
714 	}
715 
716 	btrfs_release_path(path);
717 	ret = update_free_space_extent_count(trans, block_group, path,
718 					     new_extents);
719 
720 out:
721 	return ret;
722 }
723 
remove_free_space_extent(struct btrfs_trans_handle * trans,struct btrfs_block_group * block_group,struct btrfs_path * path,u64 start,u64 size)724 static int remove_free_space_extent(struct btrfs_trans_handle *trans,
725 				    struct btrfs_block_group *block_group,
726 				    struct btrfs_path *path,
727 				    u64 start, u64 size)
728 {
729 	struct btrfs_root *root = btrfs_free_space_root(block_group);
730 	struct btrfs_key key;
731 	u64 found_start, found_end;
732 	u64 end = start + size;
733 	int new_extents = -1;
734 	int ret;
735 
736 	key.objectid = start;
737 	key.type = (u8)-1;
738 	key.offset = (u64)-1;
739 
740 	ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
741 	if (ret)
742 		goto out;
743 
744 	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
745 
746 	ASSERT(key.type == BTRFS_FREE_SPACE_EXTENT_KEY);
747 
748 	found_start = key.objectid;
749 	found_end = key.objectid + key.offset;
750 	ASSERT(start >= found_start && end <= found_end);
751 
752 	/*
753 	 * Okay, now that we've found the free space extent which contains the
754 	 * free space that we are removing, there are four cases:
755 	 *
756 	 * 1. We're using the whole extent: delete the key we found and
757 	 * decrement the free space extent count.
758 	 * 2. We are using part of the extent starting at the beginning: delete
759 	 * the key we found and insert a new key representing the leftover at
760 	 * the end. There is no net change in the number of extents.
761 	 * 3. We are using part of the extent ending at the end: delete the key
762 	 * we found and insert a new key representing the leftover at the
763 	 * beginning. There is no net change in the number of extents.
764 	 * 4. We are using part of the extent in the middle: delete the key we
765 	 * found and insert two new keys representing the leftovers on each
766 	 * side. Where we used to have one extent, we now have two, so increment
767 	 * the extent count. We may need to convert the block group to bitmaps
768 	 * as a result.
769 	 */
770 
771 	/* Delete the existing key (cases 1-4). */
772 	ret = btrfs_del_item(trans, root, path);
773 	if (ret)
774 		goto out;
775 
776 	/* Add a key for leftovers at the beginning (cases 3 and 4). */
777 	if (start > found_start) {
778 		key.objectid = found_start;
779 		key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
780 		key.offset = start - found_start;
781 
782 		btrfs_release_path(path);
783 		ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
784 		if (ret)
785 			goto out;
786 		new_extents++;
787 	}
788 
789 	/* Add a key for leftovers at the end (cases 2 and 4). */
790 	if (end < found_end) {
791 		key.objectid = end;
792 		key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
793 		key.offset = found_end - end;
794 
795 		btrfs_release_path(path);
796 		ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
797 		if (ret)
798 			goto out;
799 		new_extents++;
800 	}
801 
802 	btrfs_release_path(path);
803 	ret = update_free_space_extent_count(trans, block_group, path,
804 					     new_extents);
805 
806 out:
807 	return ret;
808 }
809 
810 EXPORT_FOR_TESTS
__remove_from_free_space_tree(struct btrfs_trans_handle * trans,struct btrfs_block_group * block_group,struct btrfs_path * path,u64 start,u64 size)811 int __remove_from_free_space_tree(struct btrfs_trans_handle *trans,
812 				  struct btrfs_block_group *block_group,
813 				  struct btrfs_path *path, u64 start, u64 size)
814 {
815 	struct btrfs_free_space_info *info;
816 	u32 flags;
817 	int ret;
818 
819 	if (test_bit(BLOCK_GROUP_FLAG_NEEDS_FREE_SPACE, &block_group->runtime_flags)) {
820 		ret = __add_block_group_free_space(trans, block_group, path);
821 		if (ret)
822 			return ret;
823 	}
824 
825 	info = search_free_space_info(NULL, block_group, path, 0);
826 	if (IS_ERR(info))
827 		return PTR_ERR(info);
828 	flags = btrfs_free_space_flags(path->nodes[0], info);
829 	btrfs_release_path(path);
830 
831 	if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) {
832 		return modify_free_space_bitmap(trans, block_group, path,
833 						start, size, 1);
834 	} else {
835 		return remove_free_space_extent(trans, block_group, path,
836 						start, size);
837 	}
838 }
839 
remove_from_free_space_tree(struct btrfs_trans_handle * trans,u64 start,u64 size)840 int remove_from_free_space_tree(struct btrfs_trans_handle *trans,
841 				u64 start, u64 size)
842 {
843 	struct btrfs_block_group *block_group;
844 	struct btrfs_path *path;
845 	int ret;
846 
847 	if (!btrfs_fs_compat_ro(trans->fs_info, FREE_SPACE_TREE))
848 		return 0;
849 
850 	path = btrfs_alloc_path();
851 	if (!path) {
852 		ret = -ENOMEM;
853 		btrfs_abort_transaction(trans, ret);
854 		goto out;
855 	}
856 
857 	block_group = btrfs_lookup_block_group(trans->fs_info, start);
858 	if (!block_group) {
859 		DEBUG_WARN("no block group found for start=%llu", start);
860 		ret = -ENOENT;
861 		btrfs_abort_transaction(trans, ret);
862 		goto out;
863 	}
864 
865 	mutex_lock(&block_group->free_space_lock);
866 	ret = __remove_from_free_space_tree(trans, block_group, path, start,
867 					    size);
868 	mutex_unlock(&block_group->free_space_lock);
869 	if (ret)
870 		btrfs_abort_transaction(trans, ret);
871 
872 	btrfs_put_block_group(block_group);
873 out:
874 	btrfs_free_path(path);
875 	return ret;
876 }
877 
add_free_space_extent(struct btrfs_trans_handle * trans,struct btrfs_block_group * block_group,struct btrfs_path * path,u64 start,u64 size)878 static int add_free_space_extent(struct btrfs_trans_handle *trans,
879 				 struct btrfs_block_group *block_group,
880 				 struct btrfs_path *path,
881 				 u64 start, u64 size)
882 {
883 	struct btrfs_root *root = btrfs_free_space_root(block_group);
884 	struct btrfs_key key, new_key;
885 	u64 found_start, found_end;
886 	u64 end = start + size;
887 	int new_extents = 1;
888 	int ret;
889 
890 	/*
891 	 * We are adding a new extent of free space, but we need to merge
892 	 * extents. There are four cases here:
893 	 *
894 	 * 1. The new extent does not have any immediate neighbors to merge
895 	 * with: add the new key and increment the free space extent count. We
896 	 * may need to convert the block group to bitmaps as a result.
897 	 * 2. The new extent has an immediate neighbor before it: remove the
898 	 * previous key and insert a new key combining both of them. There is no
899 	 * net change in the number of extents.
900 	 * 3. The new extent has an immediate neighbor after it: remove the next
901 	 * key and insert a new key combining both of them. There is no net
902 	 * change in the number of extents.
903 	 * 4. The new extent has immediate neighbors on both sides: remove both
904 	 * of the keys and insert a new key combining all of them. Where we used
905 	 * to have two extents, we now have one, so decrement the extent count.
906 	 */
907 
908 	new_key.objectid = start;
909 	new_key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
910 	new_key.offset = size;
911 
912 	/* Search for a neighbor on the left. */
913 	if (start == block_group->start)
914 		goto right;
915 	key.objectid = start - 1;
916 	key.type = (u8)-1;
917 	key.offset = (u64)-1;
918 
919 	ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
920 	if (ret)
921 		goto out;
922 
923 	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
924 
925 	if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) {
926 		ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY);
927 		btrfs_release_path(path);
928 		goto right;
929 	}
930 
931 	found_start = key.objectid;
932 	found_end = key.objectid + key.offset;
933 	ASSERT(found_start >= block_group->start &&
934 	       found_end > block_group->start);
935 	ASSERT(found_start < start && found_end <= start);
936 
937 	/*
938 	 * Delete the neighbor on the left and absorb it into the new key (cases
939 	 * 2 and 4).
940 	 */
941 	if (found_end == start) {
942 		ret = btrfs_del_item(trans, root, path);
943 		if (ret)
944 			goto out;
945 		new_key.objectid = found_start;
946 		new_key.offset += key.offset;
947 		new_extents--;
948 	}
949 	btrfs_release_path(path);
950 
951 right:
952 	/* Search for a neighbor on the right. */
953 	if (end == block_group->start + block_group->length)
954 		goto insert;
955 	key.objectid = end;
956 	key.type = (u8)-1;
957 	key.offset = (u64)-1;
958 
959 	ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
960 	if (ret)
961 		goto out;
962 
963 	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
964 
965 	if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) {
966 		ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY);
967 		btrfs_release_path(path);
968 		goto insert;
969 	}
970 
971 	found_start = key.objectid;
972 	found_end = key.objectid + key.offset;
973 	ASSERT(found_start >= block_group->start &&
974 	       found_end > block_group->start);
975 	ASSERT((found_start < start && found_end <= start) ||
976 	       (found_start >= end && found_end > end));
977 
978 	/*
979 	 * Delete the neighbor on the right and absorb it into the new key
980 	 * (cases 3 and 4).
981 	 */
982 	if (found_start == end) {
983 		ret = btrfs_del_item(trans, root, path);
984 		if (ret)
985 			goto out;
986 		new_key.offset += key.offset;
987 		new_extents--;
988 	}
989 	btrfs_release_path(path);
990 
991 insert:
992 	/* Insert the new key (cases 1-4). */
993 	ret = btrfs_insert_empty_item(trans, root, path, &new_key, 0);
994 	if (ret)
995 		goto out;
996 
997 	btrfs_release_path(path);
998 	ret = update_free_space_extent_count(trans, block_group, path,
999 					     new_extents);
1000 
1001 out:
1002 	return ret;
1003 }
1004 
1005 EXPORT_FOR_TESTS
__add_to_free_space_tree(struct btrfs_trans_handle * trans,struct btrfs_block_group * block_group,struct btrfs_path * path,u64 start,u64 size)1006 int __add_to_free_space_tree(struct btrfs_trans_handle *trans,
1007 			     struct btrfs_block_group *block_group,
1008 			     struct btrfs_path *path, u64 start, u64 size)
1009 {
1010 	struct btrfs_free_space_info *info;
1011 	u32 flags;
1012 	int ret;
1013 
1014 	if (test_bit(BLOCK_GROUP_FLAG_NEEDS_FREE_SPACE, &block_group->runtime_flags)) {
1015 		ret = __add_block_group_free_space(trans, block_group, path);
1016 		if (ret)
1017 			return ret;
1018 	}
1019 
1020 	info = search_free_space_info(NULL, block_group, path, 0);
1021 	if (IS_ERR(info))
1022 		return PTR_ERR(info);
1023 	flags = btrfs_free_space_flags(path->nodes[0], info);
1024 	btrfs_release_path(path);
1025 
1026 	if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) {
1027 		return modify_free_space_bitmap(trans, block_group, path,
1028 						start, size, 0);
1029 	} else {
1030 		return add_free_space_extent(trans, block_group, path, start,
1031 					     size);
1032 	}
1033 }
1034 
add_to_free_space_tree(struct btrfs_trans_handle * trans,u64 start,u64 size)1035 int add_to_free_space_tree(struct btrfs_trans_handle *trans,
1036 			   u64 start, u64 size)
1037 {
1038 	struct btrfs_block_group *block_group;
1039 	struct btrfs_path *path;
1040 	int ret;
1041 
1042 	if (!btrfs_fs_compat_ro(trans->fs_info, FREE_SPACE_TREE))
1043 		return 0;
1044 
1045 	path = btrfs_alloc_path();
1046 	if (!path) {
1047 		ret = -ENOMEM;
1048 		btrfs_abort_transaction(trans, ret);
1049 		goto out;
1050 	}
1051 
1052 	block_group = btrfs_lookup_block_group(trans->fs_info, start);
1053 	if (!block_group) {
1054 		DEBUG_WARN("no block group found for start=%llu", start);
1055 		ret = -ENOENT;
1056 		btrfs_abort_transaction(trans, ret);
1057 		goto out;
1058 	}
1059 
1060 	mutex_lock(&block_group->free_space_lock);
1061 	ret = __add_to_free_space_tree(trans, block_group, path, start, size);
1062 	mutex_unlock(&block_group->free_space_lock);
1063 	if (ret)
1064 		btrfs_abort_transaction(trans, ret);
1065 
1066 	btrfs_put_block_group(block_group);
1067 out:
1068 	btrfs_free_path(path);
1069 	return ret;
1070 }
1071 
1072 /*
1073  * Populate the free space tree by walking the extent tree. Operations on the
1074  * extent tree that happen as a result of writes to the free space tree will go
1075  * through the normal add/remove hooks.
1076  */
populate_free_space_tree(struct btrfs_trans_handle * trans,struct btrfs_block_group * block_group)1077 static int populate_free_space_tree(struct btrfs_trans_handle *trans,
1078 				    struct btrfs_block_group *block_group)
1079 {
1080 	struct btrfs_root *extent_root;
1081 	BTRFS_PATH_AUTO_FREE(path);
1082 	BTRFS_PATH_AUTO_FREE(path2);
1083 	struct btrfs_key key;
1084 	u64 start, end;
1085 	int ret;
1086 
1087 	path = btrfs_alloc_path();
1088 	if (!path)
1089 		return -ENOMEM;
1090 
1091 	path2 = btrfs_alloc_path();
1092 	if (!path2)
1093 		return -ENOMEM;
1094 
1095 	path->reada = READA_FORWARD;
1096 
1097 	ret = add_new_free_space_info(trans, block_group, path2);
1098 	if (ret)
1099 		return ret;
1100 
1101 	mutex_lock(&block_group->free_space_lock);
1102 
1103 	/*
1104 	 * Iterate through all of the extent and metadata items in this block
1105 	 * group, adding the free space between them and the free space at the
1106 	 * end. Note that EXTENT_ITEM and METADATA_ITEM are less than
1107 	 * BLOCK_GROUP_ITEM, so an extent may precede the block group that it's
1108 	 * contained in.
1109 	 */
1110 	key.objectid = block_group->start;
1111 	key.type = BTRFS_EXTENT_ITEM_KEY;
1112 	key.offset = 0;
1113 
1114 	extent_root = btrfs_extent_root(trans->fs_info, key.objectid);
1115 	ret = btrfs_search_slot_for_read(extent_root, &key, path, 1, 0);
1116 	if (ret < 0)
1117 		goto out_locked;
1118 	/*
1119 	 * If ret is 1 (no key found), it means this is an empty block group,
1120 	 * without any extents allocated from it and there's no block group
1121 	 * item (key BTRFS_BLOCK_GROUP_ITEM_KEY) located in the extent tree
1122 	 * because we are using the block group tree feature, so block group
1123 	 * items are stored in the block group tree. It also means there are no
1124 	 * extents allocated for block groups with a start offset beyond this
1125 	 * block group's end offset (this is the last, highest, block group).
1126 	 */
1127 	if (!btrfs_fs_compat_ro(trans->fs_info, BLOCK_GROUP_TREE))
1128 		ASSERT(ret == 0);
1129 
1130 	start = block_group->start;
1131 	end = block_group->start + block_group->length;
1132 	while (ret == 0) {
1133 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1134 
1135 		if (key.type == BTRFS_EXTENT_ITEM_KEY ||
1136 		    key.type == BTRFS_METADATA_ITEM_KEY) {
1137 			if (key.objectid >= end)
1138 				break;
1139 
1140 			if (start < key.objectid) {
1141 				ret = __add_to_free_space_tree(trans,
1142 							       block_group,
1143 							       path2, start,
1144 							       key.objectid -
1145 							       start);
1146 				if (ret)
1147 					goto out_locked;
1148 			}
1149 			start = key.objectid;
1150 			if (key.type == BTRFS_METADATA_ITEM_KEY)
1151 				start += trans->fs_info->nodesize;
1152 			else
1153 				start += key.offset;
1154 		} else if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
1155 			if (key.objectid != block_group->start)
1156 				break;
1157 		}
1158 
1159 		ret = btrfs_next_item(extent_root, path);
1160 		if (ret < 0)
1161 			goto out_locked;
1162 	}
1163 	if (start < end) {
1164 		ret = __add_to_free_space_tree(trans, block_group, path2,
1165 					       start, end - start);
1166 		if (ret)
1167 			goto out_locked;
1168 	}
1169 
1170 	ret = 0;
1171 out_locked:
1172 	mutex_unlock(&block_group->free_space_lock);
1173 
1174 	return ret;
1175 }
1176 
btrfs_create_free_space_tree(struct btrfs_fs_info * fs_info)1177 int btrfs_create_free_space_tree(struct btrfs_fs_info *fs_info)
1178 {
1179 	struct btrfs_trans_handle *trans;
1180 	struct btrfs_root *tree_root = fs_info->tree_root;
1181 	struct btrfs_root *free_space_root;
1182 	struct btrfs_block_group *block_group;
1183 	struct rb_node *node;
1184 	int ret;
1185 
1186 	trans = btrfs_start_transaction(tree_root, 0);
1187 	if (IS_ERR(trans))
1188 		return PTR_ERR(trans);
1189 
1190 	set_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1191 	set_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags);
1192 	free_space_root = btrfs_create_tree(trans,
1193 					    BTRFS_FREE_SPACE_TREE_OBJECTID);
1194 	if (IS_ERR(free_space_root)) {
1195 		ret = PTR_ERR(free_space_root);
1196 		btrfs_abort_transaction(trans, ret);
1197 		btrfs_end_transaction(trans);
1198 		goto out_clear;
1199 	}
1200 	ret = btrfs_global_root_insert(free_space_root);
1201 	if (ret) {
1202 		btrfs_put_root(free_space_root);
1203 		btrfs_abort_transaction(trans, ret);
1204 		btrfs_end_transaction(trans);
1205 		goto out_clear;
1206 	}
1207 
1208 	node = rb_first_cached(&fs_info->block_group_cache_tree);
1209 	while (node) {
1210 		block_group = rb_entry(node, struct btrfs_block_group,
1211 				       cache_node);
1212 		ret = populate_free_space_tree(trans, block_group);
1213 		if (ret) {
1214 			btrfs_abort_transaction(trans, ret);
1215 			btrfs_end_transaction(trans);
1216 			goto out_clear;
1217 		}
1218 		node = rb_next(node);
1219 	}
1220 
1221 	btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE);
1222 	btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID);
1223 	clear_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1224 	ret = btrfs_commit_transaction(trans);
1225 
1226 	/*
1227 	 * Now that we've committed the transaction any reading of our commit
1228 	 * root will be safe, so we can cache from the free space tree now.
1229 	 */
1230 	clear_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags);
1231 	return ret;
1232 
1233 out_clear:
1234 	clear_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1235 	clear_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags);
1236 	return ret;
1237 }
1238 
clear_free_space_tree(struct btrfs_trans_handle * trans,struct btrfs_root * root)1239 static int clear_free_space_tree(struct btrfs_trans_handle *trans,
1240 				 struct btrfs_root *root)
1241 {
1242 	BTRFS_PATH_AUTO_FREE(path);
1243 	struct btrfs_key key;
1244 	struct rb_node *node;
1245 	int nr;
1246 	int ret;
1247 
1248 	path = btrfs_alloc_path();
1249 	if (!path)
1250 		return -ENOMEM;
1251 
1252 	key.objectid = 0;
1253 	key.type = 0;
1254 	key.offset = 0;
1255 
1256 	while (1) {
1257 		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1258 		if (ret < 0)
1259 			return ret;
1260 
1261 		nr = btrfs_header_nritems(path->nodes[0]);
1262 		if (!nr)
1263 			break;
1264 
1265 		path->slots[0] = 0;
1266 		ret = btrfs_del_items(trans, root, path, 0, nr);
1267 		if (ret)
1268 			return ret;
1269 
1270 		btrfs_release_path(path);
1271 	}
1272 
1273 	node = rb_first_cached(&trans->fs_info->block_group_cache_tree);
1274 	while (node) {
1275 		struct btrfs_block_group *bg;
1276 
1277 		bg = rb_entry(node, struct btrfs_block_group, cache_node);
1278 		clear_bit(BLOCK_GROUP_FLAG_FREE_SPACE_ADDED, &bg->runtime_flags);
1279 		node = rb_next(node);
1280 		cond_resched();
1281 	}
1282 
1283 	return 0;
1284 }
1285 
btrfs_delete_free_space_tree(struct btrfs_fs_info * fs_info)1286 int btrfs_delete_free_space_tree(struct btrfs_fs_info *fs_info)
1287 {
1288 	struct btrfs_trans_handle *trans;
1289 	struct btrfs_root *tree_root = fs_info->tree_root;
1290 	struct btrfs_key key = {
1291 		.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID,
1292 		.type = BTRFS_ROOT_ITEM_KEY,
1293 		.offset = 0,
1294 	};
1295 	struct btrfs_root *free_space_root = btrfs_global_root(fs_info, &key);
1296 	int ret;
1297 
1298 	trans = btrfs_start_transaction(tree_root, 0);
1299 	if (IS_ERR(trans))
1300 		return PTR_ERR(trans);
1301 
1302 	btrfs_clear_fs_compat_ro(fs_info, FREE_SPACE_TREE);
1303 	btrfs_clear_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID);
1304 
1305 	ret = clear_free_space_tree(trans, free_space_root);
1306 	if (ret) {
1307 		btrfs_abort_transaction(trans, ret);
1308 		btrfs_end_transaction(trans);
1309 		return ret;
1310 	}
1311 
1312 	ret = btrfs_del_root(trans, &free_space_root->root_key);
1313 	if (ret) {
1314 		btrfs_abort_transaction(trans, ret);
1315 		btrfs_end_transaction(trans);
1316 		return ret;
1317 	}
1318 
1319 	btrfs_global_root_delete(free_space_root);
1320 
1321 	spin_lock(&fs_info->trans_lock);
1322 	list_del(&free_space_root->dirty_list);
1323 	spin_unlock(&fs_info->trans_lock);
1324 
1325 	btrfs_tree_lock(free_space_root->node);
1326 	btrfs_clear_buffer_dirty(trans, free_space_root->node);
1327 	btrfs_tree_unlock(free_space_root->node);
1328 	ret = btrfs_free_tree_block(trans, btrfs_root_id(free_space_root),
1329 				    free_space_root->node, 0, 1);
1330 	btrfs_put_root(free_space_root);
1331 	if (ret < 0) {
1332 		btrfs_abort_transaction(trans, ret);
1333 		btrfs_end_transaction(trans);
1334 		return ret;
1335 	}
1336 
1337 	return btrfs_commit_transaction(trans);
1338 }
1339 
btrfs_rebuild_free_space_tree(struct btrfs_fs_info * fs_info)1340 int btrfs_rebuild_free_space_tree(struct btrfs_fs_info *fs_info)
1341 {
1342 	struct btrfs_trans_handle *trans;
1343 	struct btrfs_key key = {
1344 		.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID,
1345 		.type = BTRFS_ROOT_ITEM_KEY,
1346 		.offset = 0,
1347 	};
1348 	struct btrfs_root *free_space_root = btrfs_global_root(fs_info, &key);
1349 	struct rb_node *node;
1350 	int ret;
1351 
1352 	trans = btrfs_start_transaction(free_space_root, 1);
1353 	if (IS_ERR(trans))
1354 		return PTR_ERR(trans);
1355 
1356 	set_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1357 	set_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags);
1358 
1359 	ret = clear_free_space_tree(trans, free_space_root);
1360 	if (ret) {
1361 		btrfs_abort_transaction(trans, ret);
1362 		btrfs_end_transaction(trans);
1363 		return ret;
1364 	}
1365 
1366 	node = rb_first_cached(&fs_info->block_group_cache_tree);
1367 	while (node) {
1368 		struct btrfs_block_group *block_group;
1369 
1370 		block_group = rb_entry(node, struct btrfs_block_group,
1371 				       cache_node);
1372 
1373 		if (test_bit(BLOCK_GROUP_FLAG_FREE_SPACE_ADDED,
1374 			     &block_group->runtime_flags))
1375 			goto next;
1376 
1377 		ret = populate_free_space_tree(trans, block_group);
1378 		if (ret) {
1379 			btrfs_abort_transaction(trans, ret);
1380 			btrfs_end_transaction(trans);
1381 			return ret;
1382 		}
1383 next:
1384 		if (btrfs_should_end_transaction(trans)) {
1385 			btrfs_end_transaction(trans);
1386 			trans = btrfs_start_transaction(free_space_root, 1);
1387 			if (IS_ERR(trans))
1388 				return PTR_ERR(trans);
1389 		}
1390 		node = rb_next(node);
1391 	}
1392 
1393 	btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE);
1394 	btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID);
1395 	clear_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1396 
1397 	ret = btrfs_commit_transaction(trans);
1398 	clear_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags);
1399 	return ret;
1400 }
1401 
__add_block_group_free_space(struct btrfs_trans_handle * trans,struct btrfs_block_group * block_group,struct btrfs_path * path)1402 static int __add_block_group_free_space(struct btrfs_trans_handle *trans,
1403 					struct btrfs_block_group *block_group,
1404 					struct btrfs_path *path)
1405 {
1406 	int ret;
1407 
1408 	clear_bit(BLOCK_GROUP_FLAG_NEEDS_FREE_SPACE, &block_group->runtime_flags);
1409 
1410 	/*
1411 	 * While rebuilding the free space tree we may allocate new metadata
1412 	 * block groups while modifying the free space tree.
1413 	 *
1414 	 * Because during the rebuild (at btrfs_rebuild_free_space_tree()) we
1415 	 * can use multiple transactions, every time btrfs_end_transaction() is
1416 	 * called at btrfs_rebuild_free_space_tree() we finish the creation of
1417 	 * new block groups by calling btrfs_create_pending_block_groups(), and
1418 	 * that in turn calls us, through add_block_group_free_space(), to add
1419 	 * a free space info item and a free space extent item for the block
1420 	 * group.
1421 	 *
1422 	 * Then later btrfs_rebuild_free_space_tree() may find such new block
1423 	 * groups and processes them with populate_free_space_tree(), which can
1424 	 * fail with EEXIST since there are already items for the block group in
1425 	 * the free space tree. Notice that we say "may find" because a new
1426 	 * block group may be added to the block groups rbtree in a node before
1427 	 * or after the block group currently being processed by the rebuild
1428 	 * process. So signal the rebuild process to skip such new block groups
1429 	 * if it finds them.
1430 	 */
1431 	set_bit(BLOCK_GROUP_FLAG_FREE_SPACE_ADDED, &block_group->runtime_flags);
1432 
1433 	ret = add_new_free_space_info(trans, block_group, path);
1434 	if (ret)
1435 		return ret;
1436 
1437 	return __add_to_free_space_tree(trans, block_group, path,
1438 					block_group->start,
1439 					block_group->length);
1440 }
1441 
add_block_group_free_space(struct btrfs_trans_handle * trans,struct btrfs_block_group * block_group)1442 int add_block_group_free_space(struct btrfs_trans_handle *trans,
1443 			       struct btrfs_block_group *block_group)
1444 {
1445 	struct btrfs_fs_info *fs_info = trans->fs_info;
1446 	struct btrfs_path *path = NULL;
1447 	int ret = 0;
1448 
1449 	if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
1450 		return 0;
1451 
1452 	mutex_lock(&block_group->free_space_lock);
1453 	if (!test_bit(BLOCK_GROUP_FLAG_NEEDS_FREE_SPACE, &block_group->runtime_flags))
1454 		goto out;
1455 
1456 	path = btrfs_alloc_path();
1457 	if (!path) {
1458 		ret = -ENOMEM;
1459 		goto out;
1460 	}
1461 
1462 	ret = __add_block_group_free_space(trans, block_group, path);
1463 
1464 out:
1465 	btrfs_free_path(path);
1466 	mutex_unlock(&block_group->free_space_lock);
1467 	if (ret)
1468 		btrfs_abort_transaction(trans, ret);
1469 	return ret;
1470 }
1471 
remove_block_group_free_space(struct btrfs_trans_handle * trans,struct btrfs_block_group * block_group)1472 int remove_block_group_free_space(struct btrfs_trans_handle *trans,
1473 				  struct btrfs_block_group *block_group)
1474 {
1475 	struct btrfs_root *root = btrfs_free_space_root(block_group);
1476 	struct btrfs_path *path;
1477 	struct btrfs_key key, found_key;
1478 	struct extent_buffer *leaf;
1479 	u64 start, end;
1480 	int done = 0, nr;
1481 	int ret;
1482 
1483 	if (!btrfs_fs_compat_ro(trans->fs_info, FREE_SPACE_TREE))
1484 		return 0;
1485 
1486 	if (test_bit(BLOCK_GROUP_FLAG_NEEDS_FREE_SPACE, &block_group->runtime_flags)) {
1487 		/* We never added this block group to the free space tree. */
1488 		return 0;
1489 	}
1490 
1491 	path = btrfs_alloc_path();
1492 	if (!path) {
1493 		ret = -ENOMEM;
1494 		goto out;
1495 	}
1496 
1497 	start = block_group->start;
1498 	end = block_group->start + block_group->length;
1499 
1500 	key.objectid = end - 1;
1501 	key.type = (u8)-1;
1502 	key.offset = (u64)-1;
1503 
1504 	while (!done) {
1505 		ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
1506 		if (ret)
1507 			goto out;
1508 
1509 		leaf = path->nodes[0];
1510 		nr = 0;
1511 		path->slots[0]++;
1512 		while (path->slots[0] > 0) {
1513 			btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
1514 
1515 			if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
1516 				ASSERT(found_key.objectid == block_group->start);
1517 				ASSERT(found_key.offset == block_group->length);
1518 				done = 1;
1519 				nr++;
1520 				path->slots[0]--;
1521 				break;
1522 			} else if (found_key.type == BTRFS_FREE_SPACE_EXTENT_KEY ||
1523 				   found_key.type == BTRFS_FREE_SPACE_BITMAP_KEY) {
1524 				ASSERT(found_key.objectid >= start);
1525 				ASSERT(found_key.objectid < end);
1526 				ASSERT(found_key.objectid + found_key.offset <= end);
1527 				nr++;
1528 				path->slots[0]--;
1529 			} else {
1530 				ASSERT(0);
1531 			}
1532 		}
1533 
1534 		ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
1535 		if (ret)
1536 			goto out;
1537 		btrfs_release_path(path);
1538 	}
1539 
1540 	ret = 0;
1541 out:
1542 	btrfs_free_path(path);
1543 	if (ret)
1544 		btrfs_abort_transaction(trans, ret);
1545 	return ret;
1546 }
1547 
load_free_space_bitmaps(struct btrfs_caching_control * caching_ctl,struct btrfs_path * path,u32 expected_extent_count)1548 static int load_free_space_bitmaps(struct btrfs_caching_control *caching_ctl,
1549 				   struct btrfs_path *path,
1550 				   u32 expected_extent_count)
1551 {
1552 	struct btrfs_block_group *block_group;
1553 	struct btrfs_fs_info *fs_info;
1554 	struct btrfs_root *root;
1555 	struct btrfs_key key;
1556 	int prev_bit = 0, bit;
1557 	/* Initialize to silence GCC. */
1558 	u64 extent_start = 0;
1559 	u64 end, offset;
1560 	u64 total_found = 0;
1561 	u32 extent_count = 0;
1562 	int ret;
1563 
1564 	block_group = caching_ctl->block_group;
1565 	fs_info = block_group->fs_info;
1566 	root = btrfs_free_space_root(block_group);
1567 
1568 	end = block_group->start + block_group->length;
1569 
1570 	while (1) {
1571 		ret = btrfs_next_item(root, path);
1572 		if (ret < 0)
1573 			goto out;
1574 		if (ret)
1575 			break;
1576 
1577 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1578 
1579 		if (key.type == BTRFS_FREE_SPACE_INFO_KEY)
1580 			break;
1581 
1582 		ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
1583 		ASSERT(key.objectid < end && key.objectid + key.offset <= end);
1584 
1585 		offset = key.objectid;
1586 		while (offset < key.objectid + key.offset) {
1587 			bit = free_space_test_bit(block_group, path, offset);
1588 			if (prev_bit == 0 && bit == 1) {
1589 				extent_start = offset;
1590 			} else if (prev_bit == 1 && bit == 0) {
1591 				u64 space_added;
1592 
1593 				ret = btrfs_add_new_free_space(block_group,
1594 							       extent_start,
1595 							       offset,
1596 							       &space_added);
1597 				if (ret)
1598 					goto out;
1599 				total_found += space_added;
1600 				if (total_found > CACHING_CTL_WAKE_UP) {
1601 					total_found = 0;
1602 					wake_up(&caching_ctl->wait);
1603 				}
1604 				extent_count++;
1605 			}
1606 			prev_bit = bit;
1607 			offset += fs_info->sectorsize;
1608 		}
1609 	}
1610 	if (prev_bit == 1) {
1611 		ret = btrfs_add_new_free_space(block_group, extent_start, end, NULL);
1612 		if (ret)
1613 			goto out;
1614 		extent_count++;
1615 	}
1616 
1617 	if (extent_count != expected_extent_count) {
1618 		btrfs_err(fs_info,
1619 			  "incorrect extent count for %llu; counted %u, expected %u",
1620 			  block_group->start, extent_count,
1621 			  expected_extent_count);
1622 		DEBUG_WARN();
1623 		ret = -EIO;
1624 		goto out;
1625 	}
1626 
1627 	ret = 0;
1628 out:
1629 	return ret;
1630 }
1631 
load_free_space_extents(struct btrfs_caching_control * caching_ctl,struct btrfs_path * path,u32 expected_extent_count)1632 static int load_free_space_extents(struct btrfs_caching_control *caching_ctl,
1633 				   struct btrfs_path *path,
1634 				   u32 expected_extent_count)
1635 {
1636 	struct btrfs_block_group *block_group;
1637 	struct btrfs_fs_info *fs_info;
1638 	struct btrfs_root *root;
1639 	struct btrfs_key key;
1640 	u64 end;
1641 	u64 total_found = 0;
1642 	u32 extent_count = 0;
1643 	int ret;
1644 
1645 	block_group = caching_ctl->block_group;
1646 	fs_info = block_group->fs_info;
1647 	root = btrfs_free_space_root(block_group);
1648 
1649 	end = block_group->start + block_group->length;
1650 
1651 	while (1) {
1652 		u64 space_added;
1653 
1654 		ret = btrfs_next_item(root, path);
1655 		if (ret < 0)
1656 			goto out;
1657 		if (ret)
1658 			break;
1659 
1660 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1661 
1662 		if (key.type == BTRFS_FREE_SPACE_INFO_KEY)
1663 			break;
1664 
1665 		ASSERT(key.type == BTRFS_FREE_SPACE_EXTENT_KEY);
1666 		ASSERT(key.objectid < end && key.objectid + key.offset <= end);
1667 
1668 		ret = btrfs_add_new_free_space(block_group, key.objectid,
1669 					       key.objectid + key.offset,
1670 					       &space_added);
1671 		if (ret)
1672 			goto out;
1673 		total_found += space_added;
1674 		if (total_found > CACHING_CTL_WAKE_UP) {
1675 			total_found = 0;
1676 			wake_up(&caching_ctl->wait);
1677 		}
1678 		extent_count++;
1679 	}
1680 
1681 	if (extent_count != expected_extent_count) {
1682 		btrfs_err(fs_info,
1683 			  "incorrect extent count for %llu; counted %u, expected %u",
1684 			  block_group->start, extent_count,
1685 			  expected_extent_count);
1686 		DEBUG_WARN();
1687 		ret = -EIO;
1688 		goto out;
1689 	}
1690 
1691 	ret = 0;
1692 out:
1693 	return ret;
1694 }
1695 
load_free_space_tree(struct btrfs_caching_control * caching_ctl)1696 int load_free_space_tree(struct btrfs_caching_control *caching_ctl)
1697 {
1698 	struct btrfs_block_group *block_group;
1699 	struct btrfs_free_space_info *info;
1700 	BTRFS_PATH_AUTO_FREE(path);
1701 	u32 extent_count, flags;
1702 
1703 	block_group = caching_ctl->block_group;
1704 
1705 	path = btrfs_alloc_path();
1706 	if (!path)
1707 		return -ENOMEM;
1708 
1709 	/*
1710 	 * Just like caching_thread() doesn't want to deadlock on the extent
1711 	 * tree, we don't want to deadlock on the free space tree.
1712 	 */
1713 	path->skip_locking = 1;
1714 	path->search_commit_root = 1;
1715 	path->reada = READA_FORWARD;
1716 
1717 	info = search_free_space_info(NULL, block_group, path, 0);
1718 	if (IS_ERR(info))
1719 		return PTR_ERR(info);
1720 
1721 	extent_count = btrfs_free_space_extent_count(path->nodes[0], info);
1722 	flags = btrfs_free_space_flags(path->nodes[0], info);
1723 
1724 	/*
1725 	 * We left path pointing to the free space info item, so now
1726 	 * load_free_space_foo can just iterate through the free space tree from
1727 	 * there.
1728 	 */
1729 	if (flags & BTRFS_FREE_SPACE_USING_BITMAPS)
1730 		return load_free_space_bitmaps(caching_ctl, path, extent_count);
1731 	else
1732 		return load_free_space_extents(caching_ctl, path, extent_count);
1733 }
1734