Merge tag 'for-4.18-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux
[muen/linux.git] / fs / btrfs / free-space-tree.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2015 Facebook.  All rights reserved.
4  */
5
6 #include <linux/kernel.h>
7 #include <linux/sched/mm.h>
8 #include "ctree.h"
9 #include "disk-io.h"
10 #include "locking.h"
11 #include "free-space-tree.h"
12 #include "transaction.h"
13
14 static int __add_block_group_free_space(struct btrfs_trans_handle *trans,
15                                         struct btrfs_block_group_cache *block_group,
16                                         struct btrfs_path *path);
17
18 void set_free_space_tree_thresholds(struct btrfs_block_group_cache *cache)
19 {
20         u32 bitmap_range;
21         size_t bitmap_size;
22         u64 num_bitmaps, total_bitmap_size;
23
24         /*
25          * We convert to bitmaps when the disk space required for using extents
26          * exceeds that required for using bitmaps.
27          */
28         bitmap_range = cache->fs_info->sectorsize * BTRFS_FREE_SPACE_BITMAP_BITS;
29         num_bitmaps = div_u64(cache->key.offset + bitmap_range - 1,
30                               bitmap_range);
31         bitmap_size = sizeof(struct btrfs_item) + BTRFS_FREE_SPACE_BITMAP_SIZE;
32         total_bitmap_size = num_bitmaps * bitmap_size;
33         cache->bitmap_high_thresh = div_u64(total_bitmap_size,
34                                             sizeof(struct btrfs_item));
35
36         /*
37          * We allow for a small buffer between the high threshold and low
38          * threshold to avoid thrashing back and forth between the two formats.
39          */
40         if (cache->bitmap_high_thresh > 100)
41                 cache->bitmap_low_thresh = cache->bitmap_high_thresh - 100;
42         else
43                 cache->bitmap_low_thresh = 0;
44 }
45
46 static int add_new_free_space_info(struct btrfs_trans_handle *trans,
47                                    struct btrfs_block_group_cache *block_group,
48                                    struct btrfs_path *path)
49 {
50         struct btrfs_root *root = trans->fs_info->free_space_root;
51         struct btrfs_free_space_info *info;
52         struct btrfs_key key;
53         struct extent_buffer *leaf;
54         int ret;
55
56         key.objectid = block_group->key.objectid;
57         key.type = BTRFS_FREE_SPACE_INFO_KEY;
58         key.offset = block_group->key.offset;
59
60         ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*info));
61         if (ret)
62                 goto out;
63
64         leaf = path->nodes[0];
65         info = btrfs_item_ptr(leaf, path->slots[0],
66                               struct btrfs_free_space_info);
67         btrfs_set_free_space_extent_count(leaf, info, 0);
68         btrfs_set_free_space_flags(leaf, info, 0);
69         btrfs_mark_buffer_dirty(leaf);
70
71         ret = 0;
72 out:
73         btrfs_release_path(path);
74         return ret;
75 }
76
77 struct btrfs_free_space_info *
78 search_free_space_info(struct btrfs_trans_handle *trans,
79                        struct btrfs_fs_info *fs_info,
80                        struct btrfs_block_group_cache *block_group,
81                        struct btrfs_path *path, int cow)
82 {
83         struct btrfs_root *root = fs_info->free_space_root;
84         struct btrfs_key key;
85         int ret;
86
87         key.objectid = block_group->key.objectid;
88         key.type = BTRFS_FREE_SPACE_INFO_KEY;
89         key.offset = block_group->key.offset;
90
91         ret = btrfs_search_slot(trans, root, &key, path, 0, cow);
92         if (ret < 0)
93                 return ERR_PTR(ret);
94         if (ret != 0) {
95                 btrfs_warn(fs_info, "missing free space info for %llu",
96                            block_group->key.objectid);
97                 ASSERT(0);
98                 return ERR_PTR(-ENOENT);
99         }
100
101         return btrfs_item_ptr(path->nodes[0], path->slots[0],
102                               struct btrfs_free_space_info);
103 }
104
105 /*
106  * btrfs_search_slot() but we're looking for the greatest key less than the
107  * passed key.
108  */
109 static int btrfs_search_prev_slot(struct btrfs_trans_handle *trans,
110                                   struct btrfs_root *root,
111                                   struct btrfs_key *key, struct btrfs_path *p,
112                                   int ins_len, int cow)
113 {
114         int ret;
115
116         ret = btrfs_search_slot(trans, root, key, p, ins_len, cow);
117         if (ret < 0)
118                 return ret;
119
120         if (ret == 0) {
121                 ASSERT(0);
122                 return -EIO;
123         }
124
125         if (p->slots[0] == 0) {
126                 ASSERT(0);
127                 return -EIO;
128         }
129         p->slots[0]--;
130
131         return 0;
132 }
133
134 static inline u32 free_space_bitmap_size(u64 size, u32 sectorsize)
135 {
136         return DIV_ROUND_UP((u32)div_u64(size, sectorsize), BITS_PER_BYTE);
137 }
138
139 static unsigned long *alloc_bitmap(u32 bitmap_size)
140 {
141         unsigned long *ret;
142         unsigned int nofs_flag;
143         u32 bitmap_rounded_size = round_up(bitmap_size, sizeof(unsigned long));
144
145         /*
146          * GFP_NOFS doesn't work with kvmalloc(), but we really can't recurse
147          * into the filesystem as the free space bitmap can be modified in the
148          * critical section of a transaction commit.
149          *
150          * TODO: push the memalloc_nofs_{save,restore}() to the caller where we
151          * know that recursion is unsafe.
152          */
153         nofs_flag = memalloc_nofs_save();
154         ret = kvzalloc(bitmap_rounded_size, GFP_KERNEL);
155         memalloc_nofs_restore(nofs_flag);
156         return ret;
157 }
158
159 static void le_bitmap_set(unsigned long *map, unsigned int start, int len)
160 {
161         u8 *p = ((u8 *)map) + BIT_BYTE(start);
162         const unsigned int size = start + len;
163         int bits_to_set = BITS_PER_BYTE - (start % BITS_PER_BYTE);
164         u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(start);
165
166         while (len - bits_to_set >= 0) {
167                 *p |= mask_to_set;
168                 len -= bits_to_set;
169                 bits_to_set = BITS_PER_BYTE;
170                 mask_to_set = ~0;
171                 p++;
172         }
173         if (len) {
174                 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
175                 *p |= mask_to_set;
176         }
177 }
178
179 int convert_free_space_to_bitmaps(struct btrfs_trans_handle *trans,
180                                   struct btrfs_block_group_cache *block_group,
181                                   struct btrfs_path *path)
182 {
183         struct btrfs_fs_info *fs_info = trans->fs_info;
184         struct btrfs_root *root = fs_info->free_space_root;
185         struct btrfs_free_space_info *info;
186         struct btrfs_key key, found_key;
187         struct extent_buffer *leaf;
188         unsigned long *bitmap;
189         char *bitmap_cursor;
190         u64 start, end;
191         u64 bitmap_range, i;
192         u32 bitmap_size, flags, expected_extent_count;
193         u32 extent_count = 0;
194         int done = 0, nr;
195         int ret;
196
197         bitmap_size = free_space_bitmap_size(block_group->key.offset,
198                                              fs_info->sectorsize);
199         bitmap = alloc_bitmap(bitmap_size);
200         if (!bitmap) {
201                 ret = -ENOMEM;
202                 goto out;
203         }
204
205         start = block_group->key.objectid;
206         end = block_group->key.objectid + block_group->key.offset;
207
208         key.objectid = end - 1;
209         key.type = (u8)-1;
210         key.offset = (u64)-1;
211
212         while (!done) {
213                 ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
214                 if (ret)
215                         goto out;
216
217                 leaf = path->nodes[0];
218                 nr = 0;
219                 path->slots[0]++;
220                 while (path->slots[0] > 0) {
221                         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
222
223                         if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
224                                 ASSERT(found_key.objectid == block_group->key.objectid);
225                                 ASSERT(found_key.offset == block_group->key.offset);
226                                 done = 1;
227                                 break;
228                         } else if (found_key.type == BTRFS_FREE_SPACE_EXTENT_KEY) {
229                                 u64 first, last;
230
231                                 ASSERT(found_key.objectid >= start);
232                                 ASSERT(found_key.objectid < end);
233                                 ASSERT(found_key.objectid + found_key.offset <= end);
234
235                                 first = div_u64(found_key.objectid - start,
236                                                 fs_info->sectorsize);
237                                 last = div_u64(found_key.objectid + found_key.offset - start,
238                                                fs_info->sectorsize);
239                                 le_bitmap_set(bitmap, first, last - first);
240
241                                 extent_count++;
242                                 nr++;
243                                 path->slots[0]--;
244                         } else {
245                                 ASSERT(0);
246                         }
247                 }
248
249                 ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
250                 if (ret)
251                         goto out;
252                 btrfs_release_path(path);
253         }
254
255         info = search_free_space_info(trans, fs_info, block_group, path, 1);
256         if (IS_ERR(info)) {
257                 ret = PTR_ERR(info);
258                 goto out;
259         }
260         leaf = path->nodes[0];
261         flags = btrfs_free_space_flags(leaf, info);
262         flags |= BTRFS_FREE_SPACE_USING_BITMAPS;
263         btrfs_set_free_space_flags(leaf, info, flags);
264         expected_extent_count = btrfs_free_space_extent_count(leaf, info);
265         btrfs_mark_buffer_dirty(leaf);
266         btrfs_release_path(path);
267
268         if (extent_count != expected_extent_count) {
269                 btrfs_err(fs_info,
270                           "incorrect extent count for %llu; counted %u, expected %u",
271                           block_group->key.objectid, extent_count,
272                           expected_extent_count);
273                 ASSERT(0);
274                 ret = -EIO;
275                 goto out;
276         }
277
278         bitmap_cursor = (char *)bitmap;
279         bitmap_range = fs_info->sectorsize * BTRFS_FREE_SPACE_BITMAP_BITS;
280         i = start;
281         while (i < end) {
282                 unsigned long ptr;
283                 u64 extent_size;
284                 u32 data_size;
285
286                 extent_size = min(end - i, bitmap_range);
287                 data_size = free_space_bitmap_size(extent_size,
288                                                    fs_info->sectorsize);
289
290                 key.objectid = i;
291                 key.type = BTRFS_FREE_SPACE_BITMAP_KEY;
292                 key.offset = extent_size;
293
294                 ret = btrfs_insert_empty_item(trans, root, path, &key,
295                                               data_size);
296                 if (ret)
297                         goto out;
298
299                 leaf = path->nodes[0];
300                 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
301                 write_extent_buffer(leaf, bitmap_cursor, ptr,
302                                     data_size);
303                 btrfs_mark_buffer_dirty(leaf);
304                 btrfs_release_path(path);
305
306                 i += extent_size;
307                 bitmap_cursor += data_size;
308         }
309
310         ret = 0;
311 out:
312         kvfree(bitmap);
313         if (ret)
314                 btrfs_abort_transaction(trans, ret);
315         return ret;
316 }
317
318 int convert_free_space_to_extents(struct btrfs_trans_handle *trans,
319                                   struct btrfs_block_group_cache *block_group,
320                                   struct btrfs_path *path)
321 {
322         struct btrfs_fs_info *fs_info = trans->fs_info;
323         struct btrfs_root *root = fs_info->free_space_root;
324         struct btrfs_free_space_info *info;
325         struct btrfs_key key, found_key;
326         struct extent_buffer *leaf;
327         unsigned long *bitmap;
328         u64 start, end;
329         u32 bitmap_size, flags, expected_extent_count;
330         unsigned long nrbits, start_bit, end_bit;
331         u32 extent_count = 0;
332         int done = 0, nr;
333         int ret;
334
335         bitmap_size = free_space_bitmap_size(block_group->key.offset,
336                                              fs_info->sectorsize);
337         bitmap = alloc_bitmap(bitmap_size);
338         if (!bitmap) {
339                 ret = -ENOMEM;
340                 goto out;
341         }
342
343         start = block_group->key.objectid;
344         end = block_group->key.objectid + block_group->key.offset;
345
346         key.objectid = end - 1;
347         key.type = (u8)-1;
348         key.offset = (u64)-1;
349
350         while (!done) {
351                 ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
352                 if (ret)
353                         goto out;
354
355                 leaf = path->nodes[0];
356                 nr = 0;
357                 path->slots[0]++;
358                 while (path->slots[0] > 0) {
359                         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
360
361                         if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
362                                 ASSERT(found_key.objectid == block_group->key.objectid);
363                                 ASSERT(found_key.offset == block_group->key.offset);
364                                 done = 1;
365                                 break;
366                         } else if (found_key.type == BTRFS_FREE_SPACE_BITMAP_KEY) {
367                                 unsigned long ptr;
368                                 char *bitmap_cursor;
369                                 u32 bitmap_pos, data_size;
370
371                                 ASSERT(found_key.objectid >= start);
372                                 ASSERT(found_key.objectid < end);
373                                 ASSERT(found_key.objectid + found_key.offset <= end);
374
375                                 bitmap_pos = div_u64(found_key.objectid - start,
376                                                      fs_info->sectorsize *
377                                                      BITS_PER_BYTE);
378                                 bitmap_cursor = ((char *)bitmap) + bitmap_pos;
379                                 data_size = free_space_bitmap_size(found_key.offset,
380                                                                    fs_info->sectorsize);
381
382                                 ptr = btrfs_item_ptr_offset(leaf, path->slots[0] - 1);
383                                 read_extent_buffer(leaf, bitmap_cursor, ptr,
384                                                    data_size);
385
386                                 nr++;
387                                 path->slots[0]--;
388                         } else {
389                                 ASSERT(0);
390                         }
391                 }
392
393                 ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
394                 if (ret)
395                         goto out;
396                 btrfs_release_path(path);
397         }
398
399         info = search_free_space_info(trans, fs_info, block_group, path, 1);
400         if (IS_ERR(info)) {
401                 ret = PTR_ERR(info);
402                 goto out;
403         }
404         leaf = path->nodes[0];
405         flags = btrfs_free_space_flags(leaf, info);
406         flags &= ~BTRFS_FREE_SPACE_USING_BITMAPS;
407         btrfs_set_free_space_flags(leaf, info, flags);
408         expected_extent_count = btrfs_free_space_extent_count(leaf, info);
409         btrfs_mark_buffer_dirty(leaf);
410         btrfs_release_path(path);
411
412         nrbits = div_u64(block_group->key.offset, block_group->fs_info->sectorsize);
413         start_bit = find_next_bit_le(bitmap, nrbits, 0);
414
415         while (start_bit < nrbits) {
416                 end_bit = find_next_zero_bit_le(bitmap, nrbits, start_bit);
417                 ASSERT(start_bit < end_bit);
418
419                 key.objectid = start + start_bit * block_group->fs_info->sectorsize;
420                 key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
421                 key.offset = (end_bit - start_bit) * block_group->fs_info->sectorsize;
422
423                 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
424                 if (ret)
425                         goto out;
426                 btrfs_release_path(path);
427
428                 extent_count++;
429
430                 start_bit = find_next_bit_le(bitmap, nrbits, end_bit);
431         }
432
433         if (extent_count != expected_extent_count) {
434                 btrfs_err(fs_info,
435                           "incorrect extent count for %llu; counted %u, expected %u",
436                           block_group->key.objectid, extent_count,
437                           expected_extent_count);
438                 ASSERT(0);
439                 ret = -EIO;
440                 goto out;
441         }
442
443         ret = 0;
444 out:
445         kvfree(bitmap);
446         if (ret)
447                 btrfs_abort_transaction(trans, ret);
448         return ret;
449 }
450
451 static int update_free_space_extent_count(struct btrfs_trans_handle *trans,
452                                           struct btrfs_block_group_cache *block_group,
453                                           struct btrfs_path *path,
454                                           int new_extents)
455 {
456         struct btrfs_free_space_info *info;
457         u32 flags;
458         u32 extent_count;
459         int ret = 0;
460
461         if (new_extents == 0)
462                 return 0;
463
464         info = search_free_space_info(trans, trans->fs_info, block_group, path,
465                                       1);
466         if (IS_ERR(info)) {
467                 ret = PTR_ERR(info);
468                 goto out;
469         }
470         flags = btrfs_free_space_flags(path->nodes[0], info);
471         extent_count = btrfs_free_space_extent_count(path->nodes[0], info);
472
473         extent_count += new_extents;
474         btrfs_set_free_space_extent_count(path->nodes[0], info, extent_count);
475         btrfs_mark_buffer_dirty(path->nodes[0]);
476         btrfs_release_path(path);
477
478         if (!(flags & BTRFS_FREE_SPACE_USING_BITMAPS) &&
479             extent_count > block_group->bitmap_high_thresh) {
480                 ret = convert_free_space_to_bitmaps(trans, block_group, path);
481         } else if ((flags & BTRFS_FREE_SPACE_USING_BITMAPS) &&
482                    extent_count < block_group->bitmap_low_thresh) {
483                 ret = convert_free_space_to_extents(trans, block_group, path);
484         }
485
486 out:
487         return ret;
488 }
489
490 int free_space_test_bit(struct btrfs_block_group_cache *block_group,
491                         struct btrfs_path *path, u64 offset)
492 {
493         struct extent_buffer *leaf;
494         struct btrfs_key key;
495         u64 found_start, found_end;
496         unsigned long ptr, i;
497
498         leaf = path->nodes[0];
499         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
500         ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
501
502         found_start = key.objectid;
503         found_end = key.objectid + key.offset;
504         ASSERT(offset >= found_start && offset < found_end);
505
506         ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
507         i = div_u64(offset - found_start,
508                     block_group->fs_info->sectorsize);
509         return !!extent_buffer_test_bit(leaf, ptr, i);
510 }
511
512 static void free_space_set_bits(struct btrfs_block_group_cache *block_group,
513                                 struct btrfs_path *path, u64 *start, u64 *size,
514                                 int bit)
515 {
516         struct btrfs_fs_info *fs_info = block_group->fs_info;
517         struct extent_buffer *leaf;
518         struct btrfs_key key;
519         u64 end = *start + *size;
520         u64 found_start, found_end;
521         unsigned long ptr, first, last;
522
523         leaf = path->nodes[0];
524         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
525         ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
526
527         found_start = key.objectid;
528         found_end = key.objectid + key.offset;
529         ASSERT(*start >= found_start && *start < found_end);
530         ASSERT(end > found_start);
531
532         if (end > found_end)
533                 end = found_end;
534
535         ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
536         first = div_u64(*start - found_start, fs_info->sectorsize);
537         last = div_u64(end - found_start, fs_info->sectorsize);
538         if (bit)
539                 extent_buffer_bitmap_set(leaf, ptr, first, last - first);
540         else
541                 extent_buffer_bitmap_clear(leaf, ptr, first, last - first);
542         btrfs_mark_buffer_dirty(leaf);
543
544         *size -= end - *start;
545         *start = end;
546 }
547
548 /*
549  * We can't use btrfs_next_item() in modify_free_space_bitmap() because
550  * btrfs_next_leaf() doesn't get the path for writing. We can forgo the fancy
551  * tree walking in btrfs_next_leaf() anyways because we know exactly what we're
552  * looking for.
553  */
554 static int free_space_next_bitmap(struct btrfs_trans_handle *trans,
555                                   struct btrfs_root *root, struct btrfs_path *p)
556 {
557         struct btrfs_key key;
558
559         if (p->slots[0] + 1 < btrfs_header_nritems(p->nodes[0])) {
560                 p->slots[0]++;
561                 return 0;
562         }
563
564         btrfs_item_key_to_cpu(p->nodes[0], &key, p->slots[0]);
565         btrfs_release_path(p);
566
567         key.objectid += key.offset;
568         key.type = (u8)-1;
569         key.offset = (u64)-1;
570
571         return btrfs_search_prev_slot(trans, root, &key, p, 0, 1);
572 }
573
574 /*
575  * If remove is 1, then we are removing free space, thus clearing bits in the
576  * bitmap. If remove is 0, then we are adding free space, thus setting bits in
577  * the bitmap.
578  */
579 static int modify_free_space_bitmap(struct btrfs_trans_handle *trans,
580                                     struct btrfs_block_group_cache *block_group,
581                                     struct btrfs_path *path,
582                                     u64 start, u64 size, int remove)
583 {
584         struct btrfs_root *root = block_group->fs_info->free_space_root;
585         struct btrfs_key key;
586         u64 end = start + size;
587         u64 cur_start, cur_size;
588         int prev_bit, next_bit;
589         int new_extents;
590         int ret;
591
592         /*
593          * Read the bit for the block immediately before the extent of space if
594          * that block is within the block group.
595          */
596         if (start > block_group->key.objectid) {
597                 u64 prev_block = start - block_group->fs_info->sectorsize;
598
599                 key.objectid = prev_block;
600                 key.type = (u8)-1;
601                 key.offset = (u64)-1;
602
603                 ret = btrfs_search_prev_slot(trans, root, &key, path, 0, 1);
604                 if (ret)
605                         goto out;
606
607                 prev_bit = free_space_test_bit(block_group, path, prev_block);
608
609                 /* The previous block may have been in the previous bitmap. */
610                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
611                 if (start >= key.objectid + key.offset) {
612                         ret = free_space_next_bitmap(trans, root, path);
613                         if (ret)
614                                 goto out;
615                 }
616         } else {
617                 key.objectid = start;
618                 key.type = (u8)-1;
619                 key.offset = (u64)-1;
620
621                 ret = btrfs_search_prev_slot(trans, root, &key, path, 0, 1);
622                 if (ret)
623                         goto out;
624
625                 prev_bit = -1;
626         }
627
628         /*
629          * Iterate over all of the bitmaps overlapped by the extent of space,
630          * clearing/setting bits as required.
631          */
632         cur_start = start;
633         cur_size = size;
634         while (1) {
635                 free_space_set_bits(block_group, path, &cur_start, &cur_size,
636                                     !remove);
637                 if (cur_size == 0)
638                         break;
639                 ret = free_space_next_bitmap(trans, root, path);
640                 if (ret)
641                         goto out;
642         }
643
644         /*
645          * Read the bit for the block immediately after the extent of space if
646          * that block is within the block group.
647          */
648         if (end < block_group->key.objectid + block_group->key.offset) {
649                 /* The next block may be in the next bitmap. */
650                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
651                 if (end >= key.objectid + key.offset) {
652                         ret = free_space_next_bitmap(trans, root, path);
653                         if (ret)
654                                 goto out;
655                 }
656
657                 next_bit = free_space_test_bit(block_group, path, end);
658         } else {
659                 next_bit = -1;
660         }
661
662         if (remove) {
663                 new_extents = -1;
664                 if (prev_bit == 1) {
665                         /* Leftover on the left. */
666                         new_extents++;
667                 }
668                 if (next_bit == 1) {
669                         /* Leftover on the right. */
670                         new_extents++;
671                 }
672         } else {
673                 new_extents = 1;
674                 if (prev_bit == 1) {
675                         /* Merging with neighbor on the left. */
676                         new_extents--;
677                 }
678                 if (next_bit == 1) {
679                         /* Merging with neighbor on the right. */
680                         new_extents--;
681                 }
682         }
683
684         btrfs_release_path(path);
685         ret = update_free_space_extent_count(trans, block_group, path,
686                                              new_extents);
687
688 out:
689         return ret;
690 }
691
692 static int remove_free_space_extent(struct btrfs_trans_handle *trans,
693                                     struct btrfs_block_group_cache *block_group,
694                                     struct btrfs_path *path,
695                                     u64 start, u64 size)
696 {
697         struct btrfs_root *root = trans->fs_info->free_space_root;
698         struct btrfs_key key;
699         u64 found_start, found_end;
700         u64 end = start + size;
701         int new_extents = -1;
702         int ret;
703
704         key.objectid = start;
705         key.type = (u8)-1;
706         key.offset = (u64)-1;
707
708         ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
709         if (ret)
710                 goto out;
711
712         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
713
714         ASSERT(key.type == BTRFS_FREE_SPACE_EXTENT_KEY);
715
716         found_start = key.objectid;
717         found_end = key.objectid + key.offset;
718         ASSERT(start >= found_start && end <= found_end);
719
720         /*
721          * Okay, now that we've found the free space extent which contains the
722          * free space that we are removing, there are four cases:
723          *
724          * 1. We're using the whole extent: delete the key we found and
725          * decrement the free space extent count.
726          * 2. We are using part of the extent starting at the beginning: delete
727          * the key we found and insert a new key representing the leftover at
728          * the end. There is no net change in the number of extents.
729          * 3. We are using part of the extent ending at the end: delete the key
730          * we found and insert a new key representing the leftover at the
731          * beginning. There is no net change in the number of extents.
732          * 4. We are using part of the extent in the middle: delete the key we
733          * found and insert two new keys representing the leftovers on each
734          * side. Where we used to have one extent, we now have two, so increment
735          * the extent count. We may need to convert the block group to bitmaps
736          * as a result.
737          */
738
739         /* Delete the existing key (cases 1-4). */
740         ret = btrfs_del_item(trans, root, path);
741         if (ret)
742                 goto out;
743
744         /* Add a key for leftovers at the beginning (cases 3 and 4). */
745         if (start > found_start) {
746                 key.objectid = found_start;
747                 key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
748                 key.offset = start - found_start;
749
750                 btrfs_release_path(path);
751                 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
752                 if (ret)
753                         goto out;
754                 new_extents++;
755         }
756
757         /* Add a key for leftovers at the end (cases 2 and 4). */
758         if (end < found_end) {
759                 key.objectid = end;
760                 key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
761                 key.offset = found_end - end;
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         btrfs_release_path(path);
771         ret = update_free_space_extent_count(trans, block_group, path,
772                                              new_extents);
773
774 out:
775         return ret;
776 }
777
778 int __remove_from_free_space_tree(struct btrfs_trans_handle *trans,
779                                   struct btrfs_block_group_cache *block_group,
780                                   struct btrfs_path *path, u64 start, u64 size)
781 {
782         struct btrfs_free_space_info *info;
783         u32 flags;
784         int ret;
785
786         if (block_group->needs_free_space) {
787                 ret = __add_block_group_free_space(trans, block_group, path);
788                 if (ret)
789                         return ret;
790         }
791
792         info = search_free_space_info(NULL, trans->fs_info, block_group, path,
793                                       0);
794         if (IS_ERR(info))
795                 return PTR_ERR(info);
796         flags = btrfs_free_space_flags(path->nodes[0], info);
797         btrfs_release_path(path);
798
799         if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) {
800                 return modify_free_space_bitmap(trans, block_group, path,
801                                                 start, size, 1);
802         } else {
803                 return remove_free_space_extent(trans, block_group, path,
804                                                 start, size);
805         }
806 }
807
808 int remove_from_free_space_tree(struct btrfs_trans_handle *trans,
809                                 u64 start, u64 size)
810 {
811         struct btrfs_block_group_cache *block_group;
812         struct btrfs_path *path;
813         int ret;
814
815         if (!btrfs_fs_compat_ro(trans->fs_info, FREE_SPACE_TREE))
816                 return 0;
817
818         path = btrfs_alloc_path();
819         if (!path) {
820                 ret = -ENOMEM;
821                 goto out;
822         }
823
824         block_group = btrfs_lookup_block_group(trans->fs_info, start);
825         if (!block_group) {
826                 ASSERT(0);
827                 ret = -ENOENT;
828                 goto out;
829         }
830
831         mutex_lock(&block_group->free_space_lock);
832         ret = __remove_from_free_space_tree(trans, block_group, path, start,
833                                             size);
834         mutex_unlock(&block_group->free_space_lock);
835
836         btrfs_put_block_group(block_group);
837 out:
838         btrfs_free_path(path);
839         if (ret)
840                 btrfs_abort_transaction(trans, ret);
841         return ret;
842 }
843
844 static int add_free_space_extent(struct btrfs_trans_handle *trans,
845                                  struct btrfs_block_group_cache *block_group,
846                                  struct btrfs_path *path,
847                                  u64 start, u64 size)
848 {
849         struct btrfs_root *root = trans->fs_info->free_space_root;
850         struct btrfs_key key, new_key;
851         u64 found_start, found_end;
852         u64 end = start + size;
853         int new_extents = 1;
854         int ret;
855
856         /*
857          * We are adding a new extent of free space, but we need to merge
858          * extents. There are four cases here:
859          *
860          * 1. The new extent does not have any immediate neighbors to merge
861          * with: add the new key and increment the free space extent count. We
862          * may need to convert the block group to bitmaps as a result.
863          * 2. The new extent has an immediate neighbor before it: remove the
864          * previous key and insert a new key combining both of them. There is no
865          * net change in the number of extents.
866          * 3. The new extent has an immediate neighbor after it: remove the next
867          * key and insert a new key combining both of them. There is no net
868          * change in the number of extents.
869          * 4. The new extent has immediate neighbors on both sides: remove both
870          * of the keys and insert a new key combining all of them. Where we used
871          * to have two extents, we now have one, so decrement the extent count.
872          */
873
874         new_key.objectid = start;
875         new_key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
876         new_key.offset = size;
877
878         /* Search for a neighbor on the left. */
879         if (start == block_group->key.objectid)
880                 goto right;
881         key.objectid = start - 1;
882         key.type = (u8)-1;
883         key.offset = (u64)-1;
884
885         ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
886         if (ret)
887                 goto out;
888
889         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
890
891         if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) {
892                 ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY);
893                 btrfs_release_path(path);
894                 goto right;
895         }
896
897         found_start = key.objectid;
898         found_end = key.objectid + key.offset;
899         ASSERT(found_start >= block_group->key.objectid &&
900                found_end > block_group->key.objectid);
901         ASSERT(found_start < start && found_end <= start);
902
903         /*
904          * Delete the neighbor on the left and absorb it into the new key (cases
905          * 2 and 4).
906          */
907         if (found_end == start) {
908                 ret = btrfs_del_item(trans, root, path);
909                 if (ret)
910                         goto out;
911                 new_key.objectid = found_start;
912                 new_key.offset += key.offset;
913                 new_extents--;
914         }
915         btrfs_release_path(path);
916
917 right:
918         /* Search for a neighbor on the right. */
919         if (end == block_group->key.objectid + block_group->key.offset)
920                 goto insert;
921         key.objectid = end;
922         key.type = (u8)-1;
923         key.offset = (u64)-1;
924
925         ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
926         if (ret)
927                 goto out;
928
929         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
930
931         if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) {
932                 ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY);
933                 btrfs_release_path(path);
934                 goto insert;
935         }
936
937         found_start = key.objectid;
938         found_end = key.objectid + key.offset;
939         ASSERT(found_start >= block_group->key.objectid &&
940                found_end > block_group->key.objectid);
941         ASSERT((found_start < start && found_end <= start) ||
942                (found_start >= end && found_end > end));
943
944         /*
945          * Delete the neighbor on the right and absorb it into the new key
946          * (cases 3 and 4).
947          */
948         if (found_start == end) {
949                 ret = btrfs_del_item(trans, root, path);
950                 if (ret)
951                         goto out;
952                 new_key.offset += key.offset;
953                 new_extents--;
954         }
955         btrfs_release_path(path);
956
957 insert:
958         /* Insert the new key (cases 1-4). */
959         ret = btrfs_insert_empty_item(trans, root, path, &new_key, 0);
960         if (ret)
961                 goto out;
962
963         btrfs_release_path(path);
964         ret = update_free_space_extent_count(trans, block_group, path,
965                                              new_extents);
966
967 out:
968         return ret;
969 }
970
971 int __add_to_free_space_tree(struct btrfs_trans_handle *trans,
972                              struct btrfs_block_group_cache *block_group,
973                              struct btrfs_path *path, u64 start, u64 size)
974 {
975         struct btrfs_fs_info *fs_info = trans->fs_info;
976         struct btrfs_free_space_info *info;
977         u32 flags;
978         int ret;
979
980         if (block_group->needs_free_space) {
981                 ret = __add_block_group_free_space(trans, block_group, path);
982                 if (ret)
983                         return ret;
984         }
985
986         info = search_free_space_info(NULL, fs_info, block_group, path, 0);
987         if (IS_ERR(info))
988                 return PTR_ERR(info);
989         flags = btrfs_free_space_flags(path->nodes[0], info);
990         btrfs_release_path(path);
991
992         if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) {
993                 return modify_free_space_bitmap(trans, block_group, path,
994                                                 start, size, 0);
995         } else {
996                 return add_free_space_extent(trans, block_group, path, start,
997                                              size);
998         }
999 }
1000
1001 int add_to_free_space_tree(struct btrfs_trans_handle *trans,
1002                            u64 start, u64 size)
1003 {
1004         struct btrfs_block_group_cache *block_group;
1005         struct btrfs_path *path;
1006         int ret;
1007
1008         if (!btrfs_fs_compat_ro(trans->fs_info, FREE_SPACE_TREE))
1009                 return 0;
1010
1011         path = btrfs_alloc_path();
1012         if (!path) {
1013                 ret = -ENOMEM;
1014                 goto out;
1015         }
1016
1017         block_group = btrfs_lookup_block_group(trans->fs_info, start);
1018         if (!block_group) {
1019                 ASSERT(0);
1020                 ret = -ENOENT;
1021                 goto out;
1022         }
1023
1024         mutex_lock(&block_group->free_space_lock);
1025         ret = __add_to_free_space_tree(trans, block_group, path, start, size);
1026         mutex_unlock(&block_group->free_space_lock);
1027
1028         btrfs_put_block_group(block_group);
1029 out:
1030         btrfs_free_path(path);
1031         if (ret)
1032                 btrfs_abort_transaction(trans, ret);
1033         return ret;
1034 }
1035
1036 /*
1037  * Populate the free space tree by walking the extent tree. Operations on the
1038  * extent tree that happen as a result of writes to the free space tree will go
1039  * through the normal add/remove hooks.
1040  */
1041 static int populate_free_space_tree(struct btrfs_trans_handle *trans,
1042                                     struct btrfs_block_group_cache *block_group)
1043 {
1044         struct btrfs_root *extent_root = trans->fs_info->extent_root;
1045         struct btrfs_path *path, *path2;
1046         struct btrfs_key key;
1047         u64 start, end;
1048         int ret;
1049
1050         path = btrfs_alloc_path();
1051         if (!path)
1052                 return -ENOMEM;
1053         path->reada = READA_FORWARD;
1054
1055         path2 = btrfs_alloc_path();
1056         if (!path2) {
1057                 btrfs_free_path(path);
1058                 return -ENOMEM;
1059         }
1060
1061         ret = add_new_free_space_info(trans, block_group, path2);
1062         if (ret)
1063                 goto out;
1064
1065         mutex_lock(&block_group->free_space_lock);
1066
1067         /*
1068          * Iterate through all of the extent and metadata items in this block
1069          * group, adding the free space between them and the free space at the
1070          * end. Note that EXTENT_ITEM and METADATA_ITEM are less than
1071          * BLOCK_GROUP_ITEM, so an extent may precede the block group that it's
1072          * contained in.
1073          */
1074         key.objectid = block_group->key.objectid;
1075         key.type = BTRFS_EXTENT_ITEM_KEY;
1076         key.offset = 0;
1077
1078         ret = btrfs_search_slot_for_read(extent_root, &key, path, 1, 0);
1079         if (ret < 0)
1080                 goto out_locked;
1081         ASSERT(ret == 0);
1082
1083         start = block_group->key.objectid;
1084         end = block_group->key.objectid + block_group->key.offset;
1085         while (1) {
1086                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1087
1088                 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
1089                     key.type == BTRFS_METADATA_ITEM_KEY) {
1090                         if (key.objectid >= end)
1091                                 break;
1092
1093                         if (start < key.objectid) {
1094                                 ret = __add_to_free_space_tree(trans,
1095                                                                block_group,
1096                                                                path2, start,
1097                                                                key.objectid -
1098                                                                start);
1099                                 if (ret)
1100                                         goto out_locked;
1101                         }
1102                         start = key.objectid;
1103                         if (key.type == BTRFS_METADATA_ITEM_KEY)
1104                                 start += trans->fs_info->nodesize;
1105                         else
1106                                 start += key.offset;
1107                 } else if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
1108                         if (key.objectid != block_group->key.objectid)
1109                                 break;
1110                 }
1111
1112                 ret = btrfs_next_item(extent_root, path);
1113                 if (ret < 0)
1114                         goto out_locked;
1115                 if (ret)
1116                         break;
1117         }
1118         if (start < end) {
1119                 ret = __add_to_free_space_tree(trans, block_group, path2,
1120                                                start, end - start);
1121                 if (ret)
1122                         goto out_locked;
1123         }
1124
1125         ret = 0;
1126 out_locked:
1127         mutex_unlock(&block_group->free_space_lock);
1128 out:
1129         btrfs_free_path(path2);
1130         btrfs_free_path(path);
1131         return ret;
1132 }
1133
1134 int btrfs_create_free_space_tree(struct btrfs_fs_info *fs_info)
1135 {
1136         struct btrfs_trans_handle *trans;
1137         struct btrfs_root *tree_root = fs_info->tree_root;
1138         struct btrfs_root *free_space_root;
1139         struct btrfs_block_group_cache *block_group;
1140         struct rb_node *node;
1141         int ret;
1142
1143         trans = btrfs_start_transaction(tree_root, 0);
1144         if (IS_ERR(trans))
1145                 return PTR_ERR(trans);
1146
1147         set_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1148         free_space_root = btrfs_create_tree(trans, fs_info,
1149                                             BTRFS_FREE_SPACE_TREE_OBJECTID);
1150         if (IS_ERR(free_space_root)) {
1151                 ret = PTR_ERR(free_space_root);
1152                 goto abort;
1153         }
1154         fs_info->free_space_root = free_space_root;
1155
1156         node = rb_first(&fs_info->block_group_cache_tree);
1157         while (node) {
1158                 block_group = rb_entry(node, struct btrfs_block_group_cache,
1159                                        cache_node);
1160                 ret = populate_free_space_tree(trans, block_group);
1161                 if (ret)
1162                         goto abort;
1163                 node = rb_next(node);
1164         }
1165
1166         btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE);
1167         btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID);
1168         clear_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1169
1170         return btrfs_commit_transaction(trans);
1171
1172 abort:
1173         clear_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1174         btrfs_abort_transaction(trans, ret);
1175         btrfs_end_transaction(trans);
1176         return ret;
1177 }
1178
1179 static int clear_free_space_tree(struct btrfs_trans_handle *trans,
1180                                  struct btrfs_root *root)
1181 {
1182         struct btrfs_path *path;
1183         struct btrfs_key key;
1184         int nr;
1185         int ret;
1186
1187         path = btrfs_alloc_path();
1188         if (!path)
1189                 return -ENOMEM;
1190
1191         path->leave_spinning = 1;
1192
1193         key.objectid = 0;
1194         key.type = 0;
1195         key.offset = 0;
1196
1197         while (1) {
1198                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1199                 if (ret < 0)
1200                         goto out;
1201
1202                 nr = btrfs_header_nritems(path->nodes[0]);
1203                 if (!nr)
1204                         break;
1205
1206                 path->slots[0] = 0;
1207                 ret = btrfs_del_items(trans, root, path, 0, nr);
1208                 if (ret)
1209                         goto out;
1210
1211                 btrfs_release_path(path);
1212         }
1213
1214         ret = 0;
1215 out:
1216         btrfs_free_path(path);
1217         return ret;
1218 }
1219
1220 int btrfs_clear_free_space_tree(struct btrfs_fs_info *fs_info)
1221 {
1222         struct btrfs_trans_handle *trans;
1223         struct btrfs_root *tree_root = fs_info->tree_root;
1224         struct btrfs_root *free_space_root = fs_info->free_space_root;
1225         int ret;
1226
1227         trans = btrfs_start_transaction(tree_root, 0);
1228         if (IS_ERR(trans))
1229                 return PTR_ERR(trans);
1230
1231         btrfs_clear_fs_compat_ro(fs_info, FREE_SPACE_TREE);
1232         btrfs_clear_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID);
1233         fs_info->free_space_root = NULL;
1234
1235         ret = clear_free_space_tree(trans, free_space_root);
1236         if (ret)
1237                 goto abort;
1238
1239         ret = btrfs_del_root(trans, fs_info, &free_space_root->root_key);
1240         if (ret)
1241                 goto abort;
1242
1243         list_del(&free_space_root->dirty_list);
1244
1245         btrfs_tree_lock(free_space_root->node);
1246         clean_tree_block(fs_info, free_space_root->node);
1247         btrfs_tree_unlock(free_space_root->node);
1248         btrfs_free_tree_block(trans, free_space_root, free_space_root->node,
1249                               0, 1);
1250
1251         free_extent_buffer(free_space_root->node);
1252         free_extent_buffer(free_space_root->commit_root);
1253         kfree(free_space_root);
1254
1255         return btrfs_commit_transaction(trans);
1256
1257 abort:
1258         btrfs_abort_transaction(trans, ret);
1259         btrfs_end_transaction(trans);
1260         return ret;
1261 }
1262
1263 static int __add_block_group_free_space(struct btrfs_trans_handle *trans,
1264                                         struct btrfs_block_group_cache *block_group,
1265                                         struct btrfs_path *path)
1266 {
1267         int ret;
1268
1269         block_group->needs_free_space = 0;
1270
1271         ret = add_new_free_space_info(trans, block_group, path);
1272         if (ret)
1273                 return ret;
1274
1275         return __add_to_free_space_tree(trans, block_group, path,
1276                                         block_group->key.objectid,
1277                                         block_group->key.offset);
1278 }
1279
1280 int add_block_group_free_space(struct btrfs_trans_handle *trans,
1281                                struct btrfs_block_group_cache *block_group)
1282 {
1283         struct btrfs_fs_info *fs_info = trans->fs_info;
1284         struct btrfs_path *path = NULL;
1285         int ret = 0;
1286
1287         if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
1288                 return 0;
1289
1290         mutex_lock(&block_group->free_space_lock);
1291         if (!block_group->needs_free_space)
1292                 goto out;
1293
1294         path = btrfs_alloc_path();
1295         if (!path) {
1296                 ret = -ENOMEM;
1297                 goto out;
1298         }
1299
1300         ret = __add_block_group_free_space(trans, block_group, path);
1301
1302 out:
1303         btrfs_free_path(path);
1304         mutex_unlock(&block_group->free_space_lock);
1305         if (ret)
1306                 btrfs_abort_transaction(trans, ret);
1307         return ret;
1308 }
1309
1310 int remove_block_group_free_space(struct btrfs_trans_handle *trans,
1311                                   struct btrfs_block_group_cache *block_group)
1312 {
1313         struct btrfs_root *root = trans->fs_info->free_space_root;
1314         struct btrfs_path *path;
1315         struct btrfs_key key, found_key;
1316         struct extent_buffer *leaf;
1317         u64 start, end;
1318         int done = 0, nr;
1319         int ret;
1320
1321         if (!btrfs_fs_compat_ro(trans->fs_info, FREE_SPACE_TREE))
1322                 return 0;
1323
1324         if (block_group->needs_free_space) {
1325                 /* We never added this block group to the free space tree. */
1326                 return 0;
1327         }
1328
1329         path = btrfs_alloc_path();
1330         if (!path) {
1331                 ret = -ENOMEM;
1332                 goto out;
1333         }
1334
1335         start = block_group->key.objectid;
1336         end = block_group->key.objectid + block_group->key.offset;
1337
1338         key.objectid = end - 1;
1339         key.type = (u8)-1;
1340         key.offset = (u64)-1;
1341
1342         while (!done) {
1343                 ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
1344                 if (ret)
1345                         goto out;
1346
1347                 leaf = path->nodes[0];
1348                 nr = 0;
1349                 path->slots[0]++;
1350                 while (path->slots[0] > 0) {
1351                         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
1352
1353                         if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
1354                                 ASSERT(found_key.objectid == block_group->key.objectid);
1355                                 ASSERT(found_key.offset == block_group->key.offset);
1356                                 done = 1;
1357                                 nr++;
1358                                 path->slots[0]--;
1359                                 break;
1360                         } else if (found_key.type == BTRFS_FREE_SPACE_EXTENT_KEY ||
1361                                    found_key.type == BTRFS_FREE_SPACE_BITMAP_KEY) {
1362                                 ASSERT(found_key.objectid >= start);
1363                                 ASSERT(found_key.objectid < end);
1364                                 ASSERT(found_key.objectid + found_key.offset <= end);
1365                                 nr++;
1366                                 path->slots[0]--;
1367                         } else {
1368                                 ASSERT(0);
1369                         }
1370                 }
1371
1372                 ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
1373                 if (ret)
1374                         goto out;
1375                 btrfs_release_path(path);
1376         }
1377
1378         ret = 0;
1379 out:
1380         btrfs_free_path(path);
1381         if (ret)
1382                 btrfs_abort_transaction(trans, ret);
1383         return ret;
1384 }
1385
1386 static int load_free_space_bitmaps(struct btrfs_caching_control *caching_ctl,
1387                                    struct btrfs_path *path,
1388                                    u32 expected_extent_count)
1389 {
1390         struct btrfs_block_group_cache *block_group;
1391         struct btrfs_fs_info *fs_info;
1392         struct btrfs_root *root;
1393         struct btrfs_key key;
1394         int prev_bit = 0, bit;
1395         /* Initialize to silence GCC. */
1396         u64 extent_start = 0;
1397         u64 end, offset;
1398         u64 total_found = 0;
1399         u32 extent_count = 0;
1400         int ret;
1401
1402         block_group = caching_ctl->block_group;
1403         fs_info = block_group->fs_info;
1404         root = fs_info->free_space_root;
1405
1406         end = block_group->key.objectid + block_group->key.offset;
1407
1408         while (1) {
1409                 ret = btrfs_next_item(root, path);
1410                 if (ret < 0)
1411                         goto out;
1412                 if (ret)
1413                         break;
1414
1415                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1416
1417                 if (key.type == BTRFS_FREE_SPACE_INFO_KEY)
1418                         break;
1419
1420                 ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
1421                 ASSERT(key.objectid < end && key.objectid + key.offset <= end);
1422
1423                 caching_ctl->progress = key.objectid;
1424
1425                 offset = key.objectid;
1426                 while (offset < key.objectid + key.offset) {
1427                         bit = free_space_test_bit(block_group, path, offset);
1428                         if (prev_bit == 0 && bit == 1) {
1429                                 extent_start = offset;
1430                         } else if (prev_bit == 1 && bit == 0) {
1431                                 total_found += add_new_free_space(block_group,
1432                                                                   extent_start,
1433                                                                   offset);
1434                                 if (total_found > CACHING_CTL_WAKE_UP) {
1435                                         total_found = 0;
1436                                         wake_up(&caching_ctl->wait);
1437                                 }
1438                                 extent_count++;
1439                         }
1440                         prev_bit = bit;
1441                         offset += fs_info->sectorsize;
1442                 }
1443         }
1444         if (prev_bit == 1) {
1445                 total_found += add_new_free_space(block_group, extent_start,
1446                                                   end);
1447                 extent_count++;
1448         }
1449
1450         if (extent_count != expected_extent_count) {
1451                 btrfs_err(fs_info,
1452                           "incorrect extent count for %llu; counted %u, expected %u",
1453                           block_group->key.objectid, extent_count,
1454                           expected_extent_count);
1455                 ASSERT(0);
1456                 ret = -EIO;
1457                 goto out;
1458         }
1459
1460         caching_ctl->progress = (u64)-1;
1461
1462         ret = 0;
1463 out:
1464         return ret;
1465 }
1466
1467 static int load_free_space_extents(struct btrfs_caching_control *caching_ctl,
1468                                    struct btrfs_path *path,
1469                                    u32 expected_extent_count)
1470 {
1471         struct btrfs_block_group_cache *block_group;
1472         struct btrfs_fs_info *fs_info;
1473         struct btrfs_root *root;
1474         struct btrfs_key key;
1475         u64 end;
1476         u64 total_found = 0;
1477         u32 extent_count = 0;
1478         int ret;
1479
1480         block_group = caching_ctl->block_group;
1481         fs_info = block_group->fs_info;
1482         root = fs_info->free_space_root;
1483
1484         end = block_group->key.objectid + block_group->key.offset;
1485
1486         while (1) {
1487                 ret = btrfs_next_item(root, path);
1488                 if (ret < 0)
1489                         goto out;
1490                 if (ret)
1491                         break;
1492
1493                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1494
1495                 if (key.type == BTRFS_FREE_SPACE_INFO_KEY)
1496                         break;
1497
1498                 ASSERT(key.type == BTRFS_FREE_SPACE_EXTENT_KEY);
1499                 ASSERT(key.objectid < end && key.objectid + key.offset <= end);
1500
1501                 caching_ctl->progress = key.objectid;
1502
1503                 total_found += add_new_free_space(block_group, key.objectid,
1504                                                   key.objectid + key.offset);
1505                 if (total_found > CACHING_CTL_WAKE_UP) {
1506                         total_found = 0;
1507                         wake_up(&caching_ctl->wait);
1508                 }
1509                 extent_count++;
1510         }
1511
1512         if (extent_count != expected_extent_count) {
1513                 btrfs_err(fs_info,
1514                           "incorrect extent count for %llu; counted %u, expected %u",
1515                           block_group->key.objectid, extent_count,
1516                           expected_extent_count);
1517                 ASSERT(0);
1518                 ret = -EIO;
1519                 goto out;
1520         }
1521
1522         caching_ctl->progress = (u64)-1;
1523
1524         ret = 0;
1525 out:
1526         return ret;
1527 }
1528
1529 int load_free_space_tree(struct btrfs_caching_control *caching_ctl)
1530 {
1531         struct btrfs_block_group_cache *block_group;
1532         struct btrfs_fs_info *fs_info;
1533         struct btrfs_free_space_info *info;
1534         struct btrfs_path *path;
1535         u32 extent_count, flags;
1536         int ret;
1537
1538         block_group = caching_ctl->block_group;
1539         fs_info = block_group->fs_info;
1540
1541         path = btrfs_alloc_path();
1542         if (!path)
1543                 return -ENOMEM;
1544
1545         /*
1546          * Just like caching_thread() doesn't want to deadlock on the extent
1547          * tree, we don't want to deadlock on the free space tree.
1548          */
1549         path->skip_locking = 1;
1550         path->search_commit_root = 1;
1551         path->reada = READA_FORWARD;
1552
1553         info = search_free_space_info(NULL, fs_info, block_group, path, 0);
1554         if (IS_ERR(info)) {
1555                 ret = PTR_ERR(info);
1556                 goto out;
1557         }
1558         extent_count = btrfs_free_space_extent_count(path->nodes[0], info);
1559         flags = btrfs_free_space_flags(path->nodes[0], info);
1560
1561         /*
1562          * We left path pointing to the free space info item, so now
1563          * load_free_space_foo can just iterate through the free space tree from
1564          * there.
1565          */
1566         if (flags & BTRFS_FREE_SPACE_USING_BITMAPS)
1567                 ret = load_free_space_bitmaps(caching_ctl, path, extent_count);
1568         else
1569                 ret = load_free_space_extents(caching_ctl, path, extent_count);
1570
1571 out:
1572         btrfs_free_path(path);
1573         return ret;
1574 }