Merge patch series "Add non-coherent DMA support for AX45MP"
[linux-block.git] / fs / btrfs / free-space-tree.c
CommitLineData
c1d7c514 1// SPDX-License-Identifier: GPL-2.0
a5ed9182
OS
2/*
3 * Copyright (C) 2015 Facebook. All rights reserved.
a5ed9182
OS
4 */
5
6#include <linux/kernel.h>
25ff17e8 7#include <linux/sched/mm.h>
9b569ea0 8#include "messages.h"
a5ed9182
OS
9#include "ctree.h"
10#include "disk-io.h"
11#include "locking.h"
12#include "free-space-tree.h"
13#include "transaction.h"
aac0023c 14#include "block-group.h"
c7f13d42 15#include "fs.h"
07e81dc9 16#include "accessors.h"
a0231804 17#include "extent-tree.h"
45c40c8f 18#include "root-tree.h"
a5ed9182
OS
19
20static int __add_block_group_free_space(struct btrfs_trans_handle *trans,
32da5386 21 struct btrfs_block_group *block_group,
a5ed9182
OS
22 struct btrfs_path *path);
23
7939dd9f
JB
24static struct btrfs_root *btrfs_free_space_root(
25 struct btrfs_block_group *block_group)
26{
abed4aaa
JB
27 struct btrfs_key key = {
28 .objectid = BTRFS_FREE_SPACE_TREE_OBJECTID,
29 .type = BTRFS_ROOT_ITEM_KEY,
30 .offset = 0,
31 };
32
f7238e50
JB
33 if (btrfs_fs_incompat(block_group->fs_info, EXTENT_TREE_V2))
34 key.offset = block_group->global_root_id;
abed4aaa 35 return btrfs_global_root(block_group->fs_info, &key);
7939dd9f
JB
36}
37
32da5386 38void set_free_space_tree_thresholds(struct btrfs_block_group *cache)
a5ed9182
OS
39{
40 u32 bitmap_range;
41 size_t bitmap_size;
42 u64 num_bitmaps, total_bitmap_size;
43
e3e39c72
MPS
44 if (WARN_ON(cache->length == 0))
45 btrfs_warn(cache->fs_info, "block group %llu length is zero",
46 cache->start);
47
a5ed9182
OS
48 /*
49 * We convert to bitmaps when the disk space required for using extents
50 * exceeds that required for using bitmaps.
51 */
da17066c 52 bitmap_range = cache->fs_info->sectorsize * BTRFS_FREE_SPACE_BITMAP_BITS;
b3470b5d 53 num_bitmaps = div_u64(cache->length + bitmap_range - 1, bitmap_range);
a5ed9182
OS
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
69static int add_new_free_space_info(struct btrfs_trans_handle *trans,
32da5386 70 struct btrfs_block_group *block_group,
a5ed9182
OS
71 struct btrfs_path *path)
72{
7939dd9f 73 struct btrfs_root *root = btrfs_free_space_root(block_group);
a5ed9182
OS
74 struct btrfs_free_space_info *info;
75 struct btrfs_key key;
76 struct extent_buffer *leaf;
77 int ret;
78
b3470b5d 79 key.objectid = block_group->start;
a5ed9182 80 key.type = BTRFS_FREE_SPACE_INFO_KEY;
b3470b5d 81 key.offset = block_group->length;
a5ed9182
OS
82
83 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*info));
84 if (ret)
85 goto out;
86
87 leaf = path->nodes[0];
88 info = btrfs_item_ptr(leaf, path->slots[0],
89 struct btrfs_free_space_info);
90 btrfs_set_free_space_extent_count(leaf, info, 0);
91 btrfs_set_free_space_flags(leaf, info, 0);
92 btrfs_mark_buffer_dirty(leaf);
93
94 ret = 0;
95out:
96 btrfs_release_path(path);
97 return ret;
98}
99
ce9f967f
JT
100EXPORT_FOR_TESTS
101struct btrfs_free_space_info *search_free_space_info(
2ccf545e 102 struct btrfs_trans_handle *trans,
32da5386 103 struct btrfs_block_group *block_group,
ce9f967f 104 struct btrfs_path *path, int cow)
a5ed9182 105{
2ccf545e 106 struct btrfs_fs_info *fs_info = block_group->fs_info;
7939dd9f 107 struct btrfs_root *root = btrfs_free_space_root(block_group);
a5ed9182
OS
108 struct btrfs_key key;
109 int ret;
110
b3470b5d 111 key.objectid = block_group->start;
a5ed9182 112 key.type = BTRFS_FREE_SPACE_INFO_KEY;
b3470b5d 113 key.offset = block_group->length;
a5ed9182
OS
114
115 ret = btrfs_search_slot(trans, root, &key, path, 0, cow);
116 if (ret < 0)
117 return ERR_PTR(ret);
118 if (ret != 0) {
ab8d0fc4 119 btrfs_warn(fs_info, "missing free space info for %llu",
b3470b5d 120 block_group->start);
a5ed9182
OS
121 ASSERT(0);
122 return ERR_PTR(-ENOENT);
123 }
124
125 return btrfs_item_ptr(path->nodes[0], path->slots[0],
126 struct btrfs_free_space_info);
127}
128
129/*
130 * btrfs_search_slot() but we're looking for the greatest key less than the
131 * passed key.
132 */
133static int btrfs_search_prev_slot(struct btrfs_trans_handle *trans,
134 struct btrfs_root *root,
135 struct btrfs_key *key, struct btrfs_path *p,
136 int ins_len, int cow)
137{
138 int ret;
139
140 ret = btrfs_search_slot(trans, root, key, p, ins_len, cow);
141 if (ret < 0)
142 return ret;
143
144 if (ret == 0) {
145 ASSERT(0);
146 return -EIO;
147 }
148
149 if (p->slots[0] == 0) {
150 ASSERT(0);
151 return -EIO;
152 }
153 p->slots[0]--;
154
155 return 0;
156}
157
098e6308
DS
158static inline u32 free_space_bitmap_size(const struct btrfs_fs_info *fs_info,
159 u64 size)
a5ed9182 160{
098e6308 161 return DIV_ROUND_UP(size >> fs_info->sectorsize_bits, BITS_PER_BYTE);
a5ed9182
OS
162}
163
a565971f 164static unsigned long *alloc_bitmap(u32 bitmap_size)
a5ed9182 165{
a565971f 166 unsigned long *ret;
25ff17e8 167 unsigned int nofs_flag;
a565971f 168 u32 bitmap_rounded_size = round_up(bitmap_size, sizeof(unsigned long));
79b134a2
DS
169
170 /*
25ff17e8
OS
171 * GFP_NOFS doesn't work with kvmalloc(), but we really can't recurse
172 * into the filesystem as the free space bitmap can be modified in the
173 * critical section of a transaction commit.
174 *
175 * TODO: push the memalloc_nofs_{save,restore}() to the caller where we
176 * know that recursion is unsafe.
79b134a2 177 */
25ff17e8 178 nofs_flag = memalloc_nofs_save();
a565971f 179 ret = kvzalloc(bitmap_rounded_size, GFP_KERNEL);
25ff17e8
OS
180 memalloc_nofs_restore(nofs_flag);
181 return ret;
a5ed9182
OS
182}
183
a565971f 184static void le_bitmap_set(unsigned long *map, unsigned int start, int len)
6faa8f47 185{
a565971f 186 u8 *p = ((u8 *)map) + BIT_BYTE(start);
6faa8f47
HM
187 const unsigned int size = start + len;
188 int bits_to_set = BITS_PER_BYTE - (start % BITS_PER_BYTE);
189 u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(start);
190
191 while (len - bits_to_set >= 0) {
192 *p |= mask_to_set;
193 len -= bits_to_set;
194 bits_to_set = BITS_PER_BYTE;
195 mask_to_set = ~0;
196 p++;
197 }
198 if (len) {
199 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
200 *p |= mask_to_set;
201 }
202}
203
ce9f967f 204EXPORT_FOR_TESTS
a5ed9182 205int convert_free_space_to_bitmaps(struct btrfs_trans_handle *trans,
32da5386 206 struct btrfs_block_group *block_group,
a5ed9182
OS
207 struct btrfs_path *path)
208{
719fb4de 209 struct btrfs_fs_info *fs_info = trans->fs_info;
7939dd9f 210 struct btrfs_root *root = btrfs_free_space_root(block_group);
a5ed9182
OS
211 struct btrfs_free_space_info *info;
212 struct btrfs_key key, found_key;
213 struct extent_buffer *leaf;
a565971f
HM
214 unsigned long *bitmap;
215 char *bitmap_cursor;
a5ed9182
OS
216 u64 start, end;
217 u64 bitmap_range, i;
218 u32 bitmap_size, flags, expected_extent_count;
219 u32 extent_count = 0;
220 int done = 0, nr;
221 int ret;
222
098e6308 223 bitmap_size = free_space_bitmap_size(fs_info, block_group->length);
a5ed9182
OS
224 bitmap = alloc_bitmap(bitmap_size);
225 if (!bitmap) {
226 ret = -ENOMEM;
227 goto out;
228 }
229
b3470b5d
DS
230 start = block_group->start;
231 end = block_group->start + block_group->length;
a5ed9182
OS
232
233 key.objectid = end - 1;
234 key.type = (u8)-1;
235 key.offset = (u64)-1;
236
237 while (!done) {
238 ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
239 if (ret)
240 goto out;
241
242 leaf = path->nodes[0];
243 nr = 0;
244 path->slots[0]++;
245 while (path->slots[0] > 0) {
246 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
247
248 if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
b3470b5d
DS
249 ASSERT(found_key.objectid == block_group->start);
250 ASSERT(found_key.offset == block_group->length);
a5ed9182
OS
251 done = 1;
252 break;
253 } else if (found_key.type == BTRFS_FREE_SPACE_EXTENT_KEY) {
254 u64 first, last;
255
256 ASSERT(found_key.objectid >= start);
257 ASSERT(found_key.objectid < end);
258 ASSERT(found_key.objectid + found_key.offset <= end);
259
260 first = div_u64(found_key.objectid - start,
0b246afa 261 fs_info->sectorsize);
a5ed9182 262 last = div_u64(found_key.objectid + found_key.offset - start,
0b246afa 263 fs_info->sectorsize);
2fe1d551 264 le_bitmap_set(bitmap, first, last - first);
a5ed9182
OS
265
266 extent_count++;
267 nr++;
268 path->slots[0]--;
269 } else {
270 ASSERT(0);
271 }
272 }
273
274 ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
275 if (ret)
276 goto out;
277 btrfs_release_path(path);
278 }
279
2ccf545e 280 info = search_free_space_info(trans, block_group, path, 1);
a5ed9182
OS
281 if (IS_ERR(info)) {
282 ret = PTR_ERR(info);
283 goto out;
284 }
285 leaf = path->nodes[0];
286 flags = btrfs_free_space_flags(leaf, info);
287 flags |= BTRFS_FREE_SPACE_USING_BITMAPS;
288 btrfs_set_free_space_flags(leaf, info, flags);
289 expected_extent_count = btrfs_free_space_extent_count(leaf, info);
290 btrfs_mark_buffer_dirty(leaf);
291 btrfs_release_path(path);
292
293 if (extent_count != expected_extent_count) {
5d163e0e
JM
294 btrfs_err(fs_info,
295 "incorrect extent count for %llu; counted %u, expected %u",
b3470b5d 296 block_group->start, extent_count,
a5ed9182
OS
297 expected_extent_count);
298 ASSERT(0);
299 ret = -EIO;
300 goto out;
301 }
302
a565971f 303 bitmap_cursor = (char *)bitmap;
0b246afa 304 bitmap_range = fs_info->sectorsize * BTRFS_FREE_SPACE_BITMAP_BITS;
a5ed9182
OS
305 i = start;
306 while (i < end) {
307 unsigned long ptr;
308 u64 extent_size;
309 u32 data_size;
310
311 extent_size = min(end - i, bitmap_range);
098e6308 312 data_size = free_space_bitmap_size(fs_info, extent_size);
a5ed9182
OS
313
314 key.objectid = i;
315 key.type = BTRFS_FREE_SPACE_BITMAP_KEY;
316 key.offset = extent_size;
317
318 ret = btrfs_insert_empty_item(trans, root, path, &key,
319 data_size);
320 if (ret)
321 goto out;
322
323 leaf = path->nodes[0];
324 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
325 write_extent_buffer(leaf, bitmap_cursor, ptr,
326 data_size);
327 btrfs_mark_buffer_dirty(leaf);
328 btrfs_release_path(path);
329
330 i += extent_size;
331 bitmap_cursor += data_size;
332 }
333
334 ret = 0;
335out:
79b134a2 336 kvfree(bitmap);
a5ed9182 337 if (ret)
66642832 338 btrfs_abort_transaction(trans, ret);
a5ed9182
OS
339 return ret;
340}
341
ce9f967f 342EXPORT_FOR_TESTS
a5ed9182 343int convert_free_space_to_extents(struct btrfs_trans_handle *trans,
32da5386 344 struct btrfs_block_group *block_group,
a5ed9182
OS
345 struct btrfs_path *path)
346{
5296c2bf 347 struct btrfs_fs_info *fs_info = trans->fs_info;
7939dd9f 348 struct btrfs_root *root = btrfs_free_space_root(block_group);
a5ed9182
OS
349 struct btrfs_free_space_info *info;
350 struct btrfs_key key, found_key;
351 struct extent_buffer *leaf;
a565971f 352 unsigned long *bitmap;
a5ed9182 353 u64 start, end;
a5ed9182 354 u32 bitmap_size, flags, expected_extent_count;
a565971f 355 unsigned long nrbits, start_bit, end_bit;
a5ed9182
OS
356 u32 extent_count = 0;
357 int done = 0, nr;
358 int ret;
359
098e6308 360 bitmap_size = free_space_bitmap_size(fs_info, block_group->length);
a5ed9182
OS
361 bitmap = alloc_bitmap(bitmap_size);
362 if (!bitmap) {
363 ret = -ENOMEM;
364 goto out;
365 }
366
b3470b5d
DS
367 start = block_group->start;
368 end = block_group->start + block_group->length;
a5ed9182
OS
369
370 key.objectid = end - 1;
371 key.type = (u8)-1;
372 key.offset = (u64)-1;
373
374 while (!done) {
375 ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
376 if (ret)
377 goto out;
378
379 leaf = path->nodes[0];
380 nr = 0;
381 path->slots[0]++;
382 while (path->slots[0] > 0) {
383 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
384
385 if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
b3470b5d
DS
386 ASSERT(found_key.objectid == block_group->start);
387 ASSERT(found_key.offset == block_group->length);
a5ed9182
OS
388 done = 1;
389 break;
390 } else if (found_key.type == BTRFS_FREE_SPACE_BITMAP_KEY) {
391 unsigned long ptr;
a565971f 392 char *bitmap_cursor;
a5ed9182
OS
393 u32 bitmap_pos, data_size;
394
395 ASSERT(found_key.objectid >= start);
396 ASSERT(found_key.objectid < end);
397 ASSERT(found_key.objectid + found_key.offset <= end);
398
399 bitmap_pos = div_u64(found_key.objectid - start,
0b246afa 400 fs_info->sectorsize *
a5ed9182 401 BITS_PER_BYTE);
a565971f 402 bitmap_cursor = ((char *)bitmap) + bitmap_pos;
098e6308
DS
403 data_size = free_space_bitmap_size(fs_info,
404 found_key.offset);
a5ed9182
OS
405
406 ptr = btrfs_item_ptr_offset(leaf, path->slots[0] - 1);
407 read_extent_buffer(leaf, bitmap_cursor, ptr,
408 data_size);
409
410 nr++;
411 path->slots[0]--;
412 } else {
413 ASSERT(0);
414 }
415 }
416
417 ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
418 if (ret)
419 goto out;
420 btrfs_release_path(path);
421 }
422
2ccf545e 423 info = search_free_space_info(trans, block_group, path, 1);
a5ed9182
OS
424 if (IS_ERR(info)) {
425 ret = PTR_ERR(info);
426 goto out;
427 }
428 leaf = path->nodes[0];
429 flags = btrfs_free_space_flags(leaf, info);
430 flags &= ~BTRFS_FREE_SPACE_USING_BITMAPS;
431 btrfs_set_free_space_flags(leaf, info, flags);
432 expected_extent_count = btrfs_free_space_extent_count(leaf, info);
433 btrfs_mark_buffer_dirty(leaf);
434 btrfs_release_path(path);
435
ab108d99 436 nrbits = block_group->length >> block_group->fs_info->sectorsize_bits;
a565971f
HM
437 start_bit = find_next_bit_le(bitmap, nrbits, 0);
438
439 while (start_bit < nrbits) {
440 end_bit = find_next_zero_bit_le(bitmap, nrbits, start_bit);
441 ASSERT(start_bit < end_bit);
442
443 key.objectid = start + start_bit * block_group->fs_info->sectorsize;
a5ed9182 444 key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
a565971f 445 key.offset = (end_bit - start_bit) * block_group->fs_info->sectorsize;
a5ed9182
OS
446
447 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
448 if (ret)
449 goto out;
450 btrfs_release_path(path);
451
452 extent_count++;
a565971f
HM
453
454 start_bit = find_next_bit_le(bitmap, nrbits, end_bit);
a5ed9182
OS
455 }
456
457 if (extent_count != expected_extent_count) {
5d163e0e
JM
458 btrfs_err(fs_info,
459 "incorrect extent count for %llu; counted %u, expected %u",
b3470b5d 460 block_group->start, extent_count,
a5ed9182
OS
461 expected_extent_count);
462 ASSERT(0);
463 ret = -EIO;
464 goto out;
465 }
466
467 ret = 0;
468out:
79b134a2 469 kvfree(bitmap);
a5ed9182 470 if (ret)
66642832 471 btrfs_abort_transaction(trans, ret);
a5ed9182
OS
472 return ret;
473}
474
475static int update_free_space_extent_count(struct btrfs_trans_handle *trans,
32da5386 476 struct btrfs_block_group *block_group,
a5ed9182
OS
477 struct btrfs_path *path,
478 int new_extents)
479{
480 struct btrfs_free_space_info *info;
481 u32 flags;
482 u32 extent_count;
483 int ret = 0;
484
485 if (new_extents == 0)
486 return 0;
487
2ccf545e 488 info = search_free_space_info(trans, block_group, path, 1);
a5ed9182
OS
489 if (IS_ERR(info)) {
490 ret = PTR_ERR(info);
491 goto out;
492 }
493 flags = btrfs_free_space_flags(path->nodes[0], info);
494 extent_count = btrfs_free_space_extent_count(path->nodes[0], info);
495
496 extent_count += new_extents;
497 btrfs_set_free_space_extent_count(path->nodes[0], info, extent_count);
498 btrfs_mark_buffer_dirty(path->nodes[0]);
499 btrfs_release_path(path);
500
501 if (!(flags & BTRFS_FREE_SPACE_USING_BITMAPS) &&
502 extent_count > block_group->bitmap_high_thresh) {
719fb4de 503 ret = convert_free_space_to_bitmaps(trans, block_group, path);
a5ed9182
OS
504 } else if ((flags & BTRFS_FREE_SPACE_USING_BITMAPS) &&
505 extent_count < block_group->bitmap_low_thresh) {
5296c2bf 506 ret = convert_free_space_to_extents(trans, block_group, path);
a5ed9182
OS
507 }
508
509out:
510 return ret;
511}
512
ce9f967f 513EXPORT_FOR_TESTS
32da5386 514int free_space_test_bit(struct btrfs_block_group *block_group,
a5ed9182
OS
515 struct btrfs_path *path, u64 offset)
516{
517 struct extent_buffer *leaf;
518 struct btrfs_key key;
519 u64 found_start, found_end;
520 unsigned long ptr, i;
521
522 leaf = path->nodes[0];
523 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
524 ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
525
526 found_start = key.objectid;
527 found_end = key.objectid + key.offset;
528 ASSERT(offset >= found_start && offset < found_end);
529
530 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
da17066c
JM
531 i = div_u64(offset - found_start,
532 block_group->fs_info->sectorsize);
a5ed9182
OS
533 return !!extent_buffer_test_bit(leaf, ptr, i);
534}
535
32da5386 536static void free_space_set_bits(struct btrfs_block_group *block_group,
a5ed9182
OS
537 struct btrfs_path *path, u64 *start, u64 *size,
538 int bit)
539{
0b246afa 540 struct btrfs_fs_info *fs_info = block_group->fs_info;
a5ed9182
OS
541 struct extent_buffer *leaf;
542 struct btrfs_key key;
543 u64 end = *start + *size;
544 u64 found_start, found_end;
545 unsigned long ptr, first, last;
546
547 leaf = path->nodes[0];
548 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
549 ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
550
551 found_start = key.objectid;
552 found_end = key.objectid + key.offset;
553 ASSERT(*start >= found_start && *start < found_end);
554 ASSERT(end > found_start);
555
556 if (end > found_end)
557 end = found_end;
558
559 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
ab108d99
DS
560 first = (*start - found_start) >> fs_info->sectorsize_bits;
561 last = (end - found_start) >> fs_info->sectorsize_bits;
a5ed9182
OS
562 if (bit)
563 extent_buffer_bitmap_set(leaf, ptr, first, last - first);
564 else
565 extent_buffer_bitmap_clear(leaf, ptr, first, last - first);
566 btrfs_mark_buffer_dirty(leaf);
567
568 *size -= end - *start;
569 *start = end;
570}
571
572/*
573 * We can't use btrfs_next_item() in modify_free_space_bitmap() because
574 * btrfs_next_leaf() doesn't get the path for writing. We can forgo the fancy
575 * tree walking in btrfs_next_leaf() anyways because we know exactly what we're
576 * looking for.
577 */
578static int free_space_next_bitmap(struct btrfs_trans_handle *trans,
579 struct btrfs_root *root, struct btrfs_path *p)
580{
581 struct btrfs_key key;
582
583 if (p->slots[0] + 1 < btrfs_header_nritems(p->nodes[0])) {
584 p->slots[0]++;
585 return 0;
586 }
587
588 btrfs_item_key_to_cpu(p->nodes[0], &key, p->slots[0]);
589 btrfs_release_path(p);
590
591 key.objectid += key.offset;
592 key.type = (u8)-1;
593 key.offset = (u64)-1;
594
595 return btrfs_search_prev_slot(trans, root, &key, p, 0, 1);
596}
597
598/*
599 * If remove is 1, then we are removing free space, thus clearing bits in the
600 * bitmap. If remove is 0, then we are adding free space, thus setting bits in
601 * the bitmap.
602 */
603static int modify_free_space_bitmap(struct btrfs_trans_handle *trans,
32da5386 604 struct btrfs_block_group *block_group,
a5ed9182
OS
605 struct btrfs_path *path,
606 u64 start, u64 size, int remove)
607{
7939dd9f 608 struct btrfs_root *root = btrfs_free_space_root(block_group);
a5ed9182
OS
609 struct btrfs_key key;
610 u64 end = start + size;
611 u64 cur_start, cur_size;
612 int prev_bit, next_bit;
613 int new_extents;
614 int ret;
615
616 /*
617 * Read the bit for the block immediately before the extent of space if
618 * that block is within the block group.
619 */
b3470b5d 620 if (start > block_group->start) {
da17066c 621 u64 prev_block = start - block_group->fs_info->sectorsize;
a5ed9182
OS
622
623 key.objectid = prev_block;
624 key.type = (u8)-1;
625 key.offset = (u64)-1;
626
627 ret = btrfs_search_prev_slot(trans, root, &key, path, 0, 1);
628 if (ret)
629 goto out;
630
631 prev_bit = free_space_test_bit(block_group, path, prev_block);
632
633 /* The previous block may have been in the previous bitmap. */
634 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
635 if (start >= key.objectid + key.offset) {
636 ret = free_space_next_bitmap(trans, root, path);
637 if (ret)
638 goto out;
639 }
640 } else {
641 key.objectid = start;
642 key.type = (u8)-1;
643 key.offset = (u64)-1;
644
645 ret = btrfs_search_prev_slot(trans, root, &key, path, 0, 1);
646 if (ret)
647 goto out;
648
649 prev_bit = -1;
650 }
651
652 /*
653 * Iterate over all of the bitmaps overlapped by the extent of space,
654 * clearing/setting bits as required.
655 */
656 cur_start = start;
657 cur_size = size;
658 while (1) {
659 free_space_set_bits(block_group, path, &cur_start, &cur_size,
660 !remove);
661 if (cur_size == 0)
662 break;
663 ret = free_space_next_bitmap(trans, root, path);
664 if (ret)
665 goto out;
666 }
667
668 /*
669 * Read the bit for the block immediately after the extent of space if
670 * that block is within the block group.
671 */
b3470b5d 672 if (end < block_group->start + block_group->length) {
a5ed9182
OS
673 /* The next block may be in the next bitmap. */
674 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
675 if (end >= key.objectid + key.offset) {
676 ret = free_space_next_bitmap(trans, root, path);
677 if (ret)
678 goto out;
679 }
680
681 next_bit = free_space_test_bit(block_group, path, end);
682 } else {
683 next_bit = -1;
684 }
685
686 if (remove) {
687 new_extents = -1;
688 if (prev_bit == 1) {
689 /* Leftover on the left. */
690 new_extents++;
691 }
692 if (next_bit == 1) {
693 /* Leftover on the right. */
694 new_extents++;
695 }
696 } else {
697 new_extents = 1;
698 if (prev_bit == 1) {
699 /* Merging with neighbor on the left. */
700 new_extents--;
701 }
702 if (next_bit == 1) {
703 /* Merging with neighbor on the right. */
704 new_extents--;
705 }
706 }
707
708 btrfs_release_path(path);
690d7682 709 ret = update_free_space_extent_count(trans, block_group, path,
a5ed9182
OS
710 new_extents);
711
712out:
713 return ret;
714}
715
716static int remove_free_space_extent(struct btrfs_trans_handle *trans,
32da5386 717 struct btrfs_block_group *block_group,
a5ed9182
OS
718 struct btrfs_path *path,
719 u64 start, u64 size)
720{
7939dd9f 721 struct btrfs_root *root = btrfs_free_space_root(block_group);
a5ed9182
OS
722 struct btrfs_key key;
723 u64 found_start, found_end;
724 u64 end = start + size;
725 int new_extents = -1;
726 int ret;
727
728 key.objectid = start;
729 key.type = (u8)-1;
730 key.offset = (u64)-1;
731
732 ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
733 if (ret)
734 goto out;
735
736 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
737
738 ASSERT(key.type == BTRFS_FREE_SPACE_EXTENT_KEY);
739
740 found_start = key.objectid;
741 found_end = key.objectid + key.offset;
742 ASSERT(start >= found_start && end <= found_end);
743
744 /*
745 * Okay, now that we've found the free space extent which contains the
746 * free space that we are removing, there are four cases:
747 *
748 * 1. We're using the whole extent: delete the key we found and
749 * decrement the free space extent count.
750 * 2. We are using part of the extent starting at the beginning: delete
751 * the key we found and insert a new key representing the leftover at
752 * the end. There is no net change in the number of extents.
753 * 3. We are using part of the extent ending at the end: delete the key
754 * we found and insert a new key representing the leftover at the
755 * beginning. There is no net change in the number of extents.
756 * 4. We are using part of the extent in the middle: delete the key we
757 * found and insert two new keys representing the leftovers on each
758 * side. Where we used to have one extent, we now have two, so increment
759 * the extent count. We may need to convert the block group to bitmaps
760 * as a result.
761 */
762
763 /* Delete the existing key (cases 1-4). */
764 ret = btrfs_del_item(trans, root, path);
765 if (ret)
766 goto out;
767
768 /* Add a key for leftovers at the beginning (cases 3 and 4). */
769 if (start > found_start) {
770 key.objectid = found_start;
771 key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
772 key.offset = start - found_start;
773
774 btrfs_release_path(path);
775 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
776 if (ret)
777 goto out;
778 new_extents++;
779 }
780
781 /* Add a key for leftovers at the end (cases 2 and 4). */
782 if (end < found_end) {
783 key.objectid = end;
784 key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
785 key.offset = found_end - end;
786
787 btrfs_release_path(path);
788 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
789 if (ret)
790 goto out;
791 new_extents++;
792 }
793
794 btrfs_release_path(path);
690d7682 795 ret = update_free_space_extent_count(trans, block_group, path,
a5ed9182
OS
796 new_extents);
797
798out:
799 return ret;
800}
801
ce9f967f 802EXPORT_FOR_TESTS
a5ed9182 803int __remove_from_free_space_tree(struct btrfs_trans_handle *trans,
32da5386 804 struct btrfs_block_group *block_group,
a5ed9182
OS
805 struct btrfs_path *path, u64 start, u64 size)
806{
807 struct btrfs_free_space_info *info;
808 u32 flags;
809 int ret;
810
0d7764ff 811 if (test_bit(BLOCK_GROUP_FLAG_NEEDS_FREE_SPACE, &block_group->runtime_flags)) {
9a7e0f92 812 ret = __add_block_group_free_space(trans, block_group, path);
a5ed9182
OS
813 if (ret)
814 return ret;
815 }
816
2ccf545e 817 info = search_free_space_info(NULL, block_group, path, 0);
a5ed9182
OS
818 if (IS_ERR(info))
819 return PTR_ERR(info);
820 flags = btrfs_free_space_flags(path->nodes[0], info);
821 btrfs_release_path(path);
822
823 if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) {
85a7ef13
NB
824 return modify_free_space_bitmap(trans, block_group, path,
825 start, size, 1);
a5ed9182 826 } else {
e581168d
NB
827 return remove_free_space_extent(trans, block_group, path,
828 start, size);
a5ed9182
OS
829 }
830}
831
832int remove_from_free_space_tree(struct btrfs_trans_handle *trans,
a5ed9182
OS
833 u64 start, u64 size)
834{
32da5386 835 struct btrfs_block_group *block_group;
a5ed9182
OS
836 struct btrfs_path *path;
837 int ret;
838
25a356d3 839 if (!btrfs_fs_compat_ro(trans->fs_info, FREE_SPACE_TREE))
a5ed9182
OS
840 return 0;
841
842 path = btrfs_alloc_path();
843 if (!path) {
844 ret = -ENOMEM;
845 goto out;
846 }
847
25a356d3 848 block_group = btrfs_lookup_block_group(trans->fs_info, start);
a5ed9182
OS
849 if (!block_group) {
850 ASSERT(0);
851 ret = -ENOENT;
852 goto out;
853 }
854
855 mutex_lock(&block_group->free_space_lock);
c31683a6
NB
856 ret = __remove_from_free_space_tree(trans, block_group, path, start,
857 size);
a5ed9182
OS
858 mutex_unlock(&block_group->free_space_lock);
859
860 btrfs_put_block_group(block_group);
861out:
862 btrfs_free_path(path);
863 if (ret)
66642832 864 btrfs_abort_transaction(trans, ret);
a5ed9182
OS
865 return ret;
866}
867
868static int add_free_space_extent(struct btrfs_trans_handle *trans,
32da5386 869 struct btrfs_block_group *block_group,
a5ed9182
OS
870 struct btrfs_path *path,
871 u64 start, u64 size)
872{
7939dd9f 873 struct btrfs_root *root = btrfs_free_space_root(block_group);
a5ed9182
OS
874 struct btrfs_key key, new_key;
875 u64 found_start, found_end;
876 u64 end = start + size;
877 int new_extents = 1;
878 int ret;
879
880 /*
881 * We are adding a new extent of free space, but we need to merge
882 * extents. There are four cases here:
883 *
884 * 1. The new extent does not have any immediate neighbors to merge
885 * with: add the new key and increment the free space extent count. We
886 * may need to convert the block group to bitmaps as a result.
887 * 2. The new extent has an immediate neighbor before it: remove the
888 * previous key and insert a new key combining both of them. There is no
889 * net change in the number of extents.
890 * 3. The new extent has an immediate neighbor after it: remove the next
891 * key and insert a new key combining both of them. There is no net
892 * change in the number of extents.
893 * 4. The new extent has immediate neighbors on both sides: remove both
894 * of the keys and insert a new key combining all of them. Where we used
895 * to have two extents, we now have one, so decrement the extent count.
896 */
897
898 new_key.objectid = start;
899 new_key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
900 new_key.offset = size;
901
902 /* Search for a neighbor on the left. */
b3470b5d 903 if (start == block_group->start)
a5ed9182
OS
904 goto right;
905 key.objectid = start - 1;
906 key.type = (u8)-1;
907 key.offset = (u64)-1;
908
909 ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
910 if (ret)
911 goto out;
912
913 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
914
915 if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) {
916 ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY);
917 btrfs_release_path(path);
918 goto right;
919 }
920
921 found_start = key.objectid;
922 found_end = key.objectid + key.offset;
b3470b5d
DS
923 ASSERT(found_start >= block_group->start &&
924 found_end > block_group->start);
a5ed9182
OS
925 ASSERT(found_start < start && found_end <= start);
926
927 /*
928 * Delete the neighbor on the left and absorb it into the new key (cases
929 * 2 and 4).
930 */
931 if (found_end == start) {
932 ret = btrfs_del_item(trans, root, path);
933 if (ret)
934 goto out;
935 new_key.objectid = found_start;
936 new_key.offset += key.offset;
937 new_extents--;
938 }
939 btrfs_release_path(path);
940
941right:
942 /* Search for a neighbor on the right. */
b3470b5d 943 if (end == block_group->start + block_group->length)
a5ed9182
OS
944 goto insert;
945 key.objectid = end;
946 key.type = (u8)-1;
947 key.offset = (u64)-1;
948
949 ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
950 if (ret)
951 goto out;
952
953 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
954
955 if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) {
956 ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY);
957 btrfs_release_path(path);
958 goto insert;
959 }
960
961 found_start = key.objectid;
962 found_end = key.objectid + key.offset;
b3470b5d
DS
963 ASSERT(found_start >= block_group->start &&
964 found_end > block_group->start);
a5ed9182
OS
965 ASSERT((found_start < start && found_end <= start) ||
966 (found_start >= end && found_end > end));
967
968 /*
969 * Delete the neighbor on the right and absorb it into the new key
970 * (cases 3 and 4).
971 */
972 if (found_start == end) {
973 ret = btrfs_del_item(trans, root, path);
974 if (ret)
975 goto out;
976 new_key.offset += key.offset;
977 new_extents--;
978 }
979 btrfs_release_path(path);
980
981insert:
982 /* Insert the new key (cases 1-4). */
983 ret = btrfs_insert_empty_item(trans, root, path, &new_key, 0);
984 if (ret)
985 goto out;
986
987 btrfs_release_path(path);
690d7682 988 ret = update_free_space_extent_count(trans, block_group, path,
a5ed9182
OS
989 new_extents);
990
991out:
992 return ret;
993}
994
ce9f967f 995EXPORT_FOR_TESTS
a5ed9182 996int __add_to_free_space_tree(struct btrfs_trans_handle *trans,
32da5386 997 struct btrfs_block_group *block_group,
a5ed9182
OS
998 struct btrfs_path *path, u64 start, u64 size)
999{
1000 struct btrfs_free_space_info *info;
1001 u32 flags;
1002 int ret;
1003
0d7764ff 1004 if (test_bit(BLOCK_GROUP_FLAG_NEEDS_FREE_SPACE, &block_group->runtime_flags)) {
9a7e0f92 1005 ret = __add_block_group_free_space(trans, block_group, path);
a5ed9182
OS
1006 if (ret)
1007 return ret;
1008 }
1009
2ccf545e 1010 info = search_free_space_info(NULL, block_group, path, 0);
a5ed9182
OS
1011 if (IS_ERR(info))
1012 return PTR_ERR(info);
1013 flags = btrfs_free_space_flags(path->nodes[0], info);
1014 btrfs_release_path(path);
1015
1016 if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) {
85a7ef13
NB
1017 return modify_free_space_bitmap(trans, block_group, path,
1018 start, size, 0);
a5ed9182 1019 } else {
5cb17822
NB
1020 return add_free_space_extent(trans, block_group, path, start,
1021 size);
a5ed9182
OS
1022 }
1023}
1024
1025int add_to_free_space_tree(struct btrfs_trans_handle *trans,
a5ed9182
OS
1026 u64 start, u64 size)
1027{
32da5386 1028 struct btrfs_block_group *block_group;
a5ed9182
OS
1029 struct btrfs_path *path;
1030 int ret;
1031
e7355e50 1032 if (!btrfs_fs_compat_ro(trans->fs_info, FREE_SPACE_TREE))
a5ed9182
OS
1033 return 0;
1034
1035 path = btrfs_alloc_path();
1036 if (!path) {
1037 ret = -ENOMEM;
1038 goto out;
1039 }
1040
e7355e50 1041 block_group = btrfs_lookup_block_group(trans->fs_info, start);
a5ed9182
OS
1042 if (!block_group) {
1043 ASSERT(0);
1044 ret = -ENOENT;
1045 goto out;
1046 }
1047
1048 mutex_lock(&block_group->free_space_lock);
2d5cffa1 1049 ret = __add_to_free_space_tree(trans, block_group, path, start, size);
a5ed9182
OS
1050 mutex_unlock(&block_group->free_space_lock);
1051
1052 btrfs_put_block_group(block_group);
1053out:
1054 btrfs_free_path(path);
1055 if (ret)
66642832 1056 btrfs_abort_transaction(trans, ret);
a5ed9182
OS
1057 return ret;
1058}
1059
1060/*
1061 * Populate the free space tree by walking the extent tree. Operations on the
1062 * extent tree that happen as a result of writes to the free space tree will go
1063 * through the normal add/remove hooks.
1064 */
1065static int populate_free_space_tree(struct btrfs_trans_handle *trans,
32da5386 1066 struct btrfs_block_group *block_group)
a5ed9182 1067{
29cbcf40 1068 struct btrfs_root *extent_root;
a5ed9182
OS
1069 struct btrfs_path *path, *path2;
1070 struct btrfs_key key;
1071 u64 start, end;
1072 int ret;
1073
1074 path = btrfs_alloc_path();
1075 if (!path)
1076 return -ENOMEM;
019599ad 1077 path->reada = READA_FORWARD;
a5ed9182
OS
1078
1079 path2 = btrfs_alloc_path();
1080 if (!path2) {
1081 btrfs_free_path(path);
1082 return -ENOMEM;
1083 }
1084
66afee18 1085 ret = add_new_free_space_info(trans, block_group, path2);
a5ed9182
OS
1086 if (ret)
1087 goto out;
1088
511711af
CM
1089 mutex_lock(&block_group->free_space_lock);
1090
a5ed9182
OS
1091 /*
1092 * Iterate through all of the extent and metadata items in this block
1093 * group, adding the free space between them and the free space at the
1094 * end. Note that EXTENT_ITEM and METADATA_ITEM are less than
1095 * BLOCK_GROUP_ITEM, so an extent may precede the block group that it's
1096 * contained in.
1097 */
b3470b5d 1098 key.objectid = block_group->start;
a5ed9182
OS
1099 key.type = BTRFS_EXTENT_ITEM_KEY;
1100 key.offset = 0;
1101
29cbcf40 1102 extent_root = btrfs_extent_root(trans->fs_info, key.objectid);
a5ed9182
OS
1103 ret = btrfs_search_slot_for_read(extent_root, &key, path, 1, 0);
1104 if (ret < 0)
511711af 1105 goto out_locked;
a5ed9182
OS
1106 ASSERT(ret == 0);
1107
b3470b5d
DS
1108 start = block_group->start;
1109 end = block_group->start + block_group->length;
a5ed9182
OS
1110 while (1) {
1111 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1112
1113 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
1114 key.type == BTRFS_METADATA_ITEM_KEY) {
1115 if (key.objectid >= end)
1116 break;
1117
1118 if (start < key.objectid) {
2d5cffa1 1119 ret = __add_to_free_space_tree(trans,
a5ed9182
OS
1120 block_group,
1121 path2, start,
1122 key.objectid -
1123 start);
1124 if (ret)
511711af 1125 goto out_locked;
a5ed9182
OS
1126 }
1127 start = key.objectid;
1128 if (key.type == BTRFS_METADATA_ITEM_KEY)
ffa9a9ef 1129 start += trans->fs_info->nodesize;
a5ed9182
OS
1130 else
1131 start += key.offset;
1132 } else if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
b3470b5d 1133 if (key.objectid != block_group->start)
a5ed9182
OS
1134 break;
1135 }
1136
1137 ret = btrfs_next_item(extent_root, path);
1138 if (ret < 0)
511711af 1139 goto out_locked;
a5ed9182
OS
1140 if (ret)
1141 break;
1142 }
1143 if (start < end) {
2d5cffa1
NB
1144 ret = __add_to_free_space_tree(trans, block_group, path2,
1145 start, end - start);
a5ed9182 1146 if (ret)
511711af 1147 goto out_locked;
a5ed9182
OS
1148 }
1149
1150 ret = 0;
511711af
CM
1151out_locked:
1152 mutex_unlock(&block_group->free_space_lock);
a5ed9182
OS
1153out:
1154 btrfs_free_path(path2);
1155 btrfs_free_path(path);
1156 return ret;
1157}
1158
1159int btrfs_create_free_space_tree(struct btrfs_fs_info *fs_info)
1160{
1161 struct btrfs_trans_handle *trans;
1162 struct btrfs_root *tree_root = fs_info->tree_root;
1163 struct btrfs_root *free_space_root;
32da5386 1164 struct btrfs_block_group *block_group;
a5ed9182
OS
1165 struct rb_node *node;
1166 int ret;
1167
1168 trans = btrfs_start_transaction(tree_root, 0);
1169 if (IS_ERR(trans))
1170 return PTR_ERR(trans);
1171
afcdd129 1172 set_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
2f96e402 1173 set_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags);
9b7a2440 1174 free_space_root = btrfs_create_tree(trans,
a5ed9182
OS
1175 BTRFS_FREE_SPACE_TREE_OBJECTID);
1176 if (IS_ERR(free_space_root)) {
1177 ret = PTR_ERR(free_space_root);
1178 goto abort;
1179 }
abed4aaa
JB
1180 ret = btrfs_global_root_insert(free_space_root);
1181 if (ret) {
1182 btrfs_put_root(free_space_root);
1183 goto abort;
1184 }
a5ed9182 1185
08dddb29 1186 node = rb_first_cached(&fs_info->block_group_cache_tree);
a5ed9182 1187 while (node) {
32da5386 1188 block_group = rb_entry(node, struct btrfs_block_group,
a5ed9182 1189 cache_node);
ffa9a9ef 1190 ret = populate_free_space_tree(trans, block_group);
a5ed9182
OS
1191 if (ret)
1192 goto abort;
1193 node = rb_next(node);
1194 }
1195
1196 btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE);
6675df31 1197 btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID);
afcdd129 1198 clear_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
2f96e402 1199 ret = btrfs_commit_transaction(trans);
a5ed9182 1200
2f96e402
JB
1201 /*
1202 * Now that we've committed the transaction any reading of our commit
1203 * root will be safe, so we can cache from the free space tree now.
1204 */
1205 clear_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags);
1206 return ret;
a5ed9182
OS
1207
1208abort:
afcdd129 1209 clear_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
2f96e402 1210 clear_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags);
66642832 1211 btrfs_abort_transaction(trans, ret);
3a45bb20 1212 btrfs_end_transaction(trans);
a5ed9182
OS
1213 return ret;
1214}
1215
1216static int clear_free_space_tree(struct btrfs_trans_handle *trans,
1217 struct btrfs_root *root)
1218{
1219 struct btrfs_path *path;
1220 struct btrfs_key key;
1221 int nr;
1222 int ret;
1223
1224 path = btrfs_alloc_path();
1225 if (!path)
1226 return -ENOMEM;
1227
a5ed9182
OS
1228 key.objectid = 0;
1229 key.type = 0;
1230 key.offset = 0;
1231
1232 while (1) {
1233 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1234 if (ret < 0)
1235 goto out;
1236
1237 nr = btrfs_header_nritems(path->nodes[0]);
1238 if (!nr)
1239 break;
1240
1241 path->slots[0] = 0;
1242 ret = btrfs_del_items(trans, root, path, 0, nr);
1243 if (ret)
1244 goto out;
1245
1246 btrfs_release_path(path);
1247 }
1248
1249 ret = 0;
1250out:
1251 btrfs_free_path(path);
1252 return ret;
1253}
1254
1d6a4fc8 1255int btrfs_delete_free_space_tree(struct btrfs_fs_info *fs_info)
a5ed9182
OS
1256{
1257 struct btrfs_trans_handle *trans;
1258 struct btrfs_root *tree_root = fs_info->tree_root;
abed4aaa
JB
1259 struct btrfs_key key = {
1260 .objectid = BTRFS_FREE_SPACE_TREE_OBJECTID,
1261 .type = BTRFS_ROOT_ITEM_KEY,
1262 .offset = 0,
1263 };
1264 struct btrfs_root *free_space_root = btrfs_global_root(fs_info, &key);
a5ed9182
OS
1265 int ret;
1266
1267 trans = btrfs_start_transaction(tree_root, 0);
1268 if (IS_ERR(trans))
1269 return PTR_ERR(trans);
1270
1271 btrfs_clear_fs_compat_ro(fs_info, FREE_SPACE_TREE);
6675df31 1272 btrfs_clear_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID);
a5ed9182
OS
1273
1274 ret = clear_free_space_tree(trans, free_space_root);
1275 if (ret)
1276 goto abort;
1277
ab9ce7d4 1278 ret = btrfs_del_root(trans, &free_space_root->root_key);
a5ed9182
OS
1279 if (ret)
1280 goto abort;
1281
abed4aaa 1282 btrfs_global_root_delete(free_space_root);
babebf02
FM
1283
1284 spin_lock(&fs_info->trans_lock);
a5ed9182 1285 list_del(&free_space_root->dirty_list);
babebf02 1286 spin_unlock(&fs_info->trans_lock);
a5ed9182
OS
1287
1288 btrfs_tree_lock(free_space_root->node);
190a8339 1289 btrfs_clear_buffer_dirty(trans, free_space_root->node);
a5ed9182 1290 btrfs_tree_unlock(free_space_root->node);
7a163608
FM
1291 btrfs_free_tree_block(trans, btrfs_root_id(free_space_root),
1292 free_space_root->node, 0, 1);
a5ed9182 1293
00246528 1294 btrfs_put_root(free_space_root);
a5ed9182 1295
0bef7109 1296 return btrfs_commit_transaction(trans);
a5ed9182
OS
1297
1298abort:
66642832 1299 btrfs_abort_transaction(trans, ret);
3a45bb20 1300 btrfs_end_transaction(trans);
a5ed9182
OS
1301 return ret;
1302}
1303
1d6a4fc8
QW
1304int btrfs_rebuild_free_space_tree(struct btrfs_fs_info *fs_info)
1305{
1306 struct btrfs_trans_handle *trans;
1307 struct btrfs_key key = {
1308 .objectid = BTRFS_FREE_SPACE_TREE_OBJECTID,
1309 .type = BTRFS_ROOT_ITEM_KEY,
1310 .offset = 0,
1311 };
1312 struct btrfs_root *free_space_root = btrfs_global_root(fs_info, &key);
1313 struct rb_node *node;
1314 int ret;
1315
1316 trans = btrfs_start_transaction(free_space_root, 1);
1317 if (IS_ERR(trans))
1318 return PTR_ERR(trans);
1319
1320 set_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1321 set_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags);
1322
1323 ret = clear_free_space_tree(trans, free_space_root);
1324 if (ret)
1325 goto abort;
1326
1327 node = rb_first_cached(&fs_info->block_group_cache_tree);
1328 while (node) {
1329 struct btrfs_block_group *block_group;
1330
1331 block_group = rb_entry(node, struct btrfs_block_group,
1332 cache_node);
1333 ret = populate_free_space_tree(trans, block_group);
1334 if (ret)
1335 goto abort;
1336 node = rb_next(node);
1337 }
1338
1339 btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE);
1340 btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID);
1341 clear_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1342
1343 ret = btrfs_commit_transaction(trans);
1344 clear_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags);
1345 return ret;
1346abort:
1347 btrfs_abort_transaction(trans, ret);
1348 btrfs_end_transaction(trans);
1349 return ret;
1350}
1351
a5ed9182 1352static int __add_block_group_free_space(struct btrfs_trans_handle *trans,
32da5386 1353 struct btrfs_block_group *block_group,
a5ed9182
OS
1354 struct btrfs_path *path)
1355{
a5ed9182
OS
1356 int ret;
1357
0d7764ff 1358 clear_bit(BLOCK_GROUP_FLAG_NEEDS_FREE_SPACE, &block_group->runtime_flags);
a5ed9182 1359
66afee18 1360 ret = add_new_free_space_info(trans, block_group, path);
a5ed9182
OS
1361 if (ret)
1362 return ret;
1363
2d5cffa1 1364 return __add_to_free_space_tree(trans, block_group, path,
b3470b5d
DS
1365 block_group->start,
1366 block_group->length);
a5ed9182
OS
1367}
1368
1369int add_block_group_free_space(struct btrfs_trans_handle *trans,
32da5386 1370 struct btrfs_block_group *block_group)
a5ed9182 1371{
e4e0711c 1372 struct btrfs_fs_info *fs_info = trans->fs_info;
a5ed9182
OS
1373 struct btrfs_path *path = NULL;
1374 int ret = 0;
1375
1376 if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
1377 return 0;
1378
1379 mutex_lock(&block_group->free_space_lock);
0d7764ff 1380 if (!test_bit(BLOCK_GROUP_FLAG_NEEDS_FREE_SPACE, &block_group->runtime_flags))
a5ed9182
OS
1381 goto out;
1382
1383 path = btrfs_alloc_path();
1384 if (!path) {
1385 ret = -ENOMEM;
1386 goto out;
1387 }
1388
9a7e0f92 1389 ret = __add_block_group_free_space(trans, block_group, path);
a5ed9182
OS
1390
1391out:
1392 btrfs_free_path(path);
1393 mutex_unlock(&block_group->free_space_lock);
1394 if (ret)
66642832 1395 btrfs_abort_transaction(trans, ret);
a5ed9182
OS
1396 return ret;
1397}
1398
1399int remove_block_group_free_space(struct btrfs_trans_handle *trans,
32da5386 1400 struct btrfs_block_group *block_group)
a5ed9182 1401{
7939dd9f 1402 struct btrfs_root *root = btrfs_free_space_root(block_group);
a5ed9182
OS
1403 struct btrfs_path *path;
1404 struct btrfs_key key, found_key;
1405 struct extent_buffer *leaf;
1406 u64 start, end;
1407 int done = 0, nr;
1408 int ret;
1409
f3f72779 1410 if (!btrfs_fs_compat_ro(trans->fs_info, FREE_SPACE_TREE))
a5ed9182
OS
1411 return 0;
1412
0d7764ff 1413 if (test_bit(BLOCK_GROUP_FLAG_NEEDS_FREE_SPACE, &block_group->runtime_flags)) {
a5ed9182
OS
1414 /* We never added this block group to the free space tree. */
1415 return 0;
1416 }
1417
1418 path = btrfs_alloc_path();
1419 if (!path) {
1420 ret = -ENOMEM;
1421 goto out;
1422 }
1423
b3470b5d
DS
1424 start = block_group->start;
1425 end = block_group->start + block_group->length;
a5ed9182
OS
1426
1427 key.objectid = end - 1;
1428 key.type = (u8)-1;
1429 key.offset = (u64)-1;
1430
1431 while (!done) {
1432 ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
1433 if (ret)
1434 goto out;
1435
1436 leaf = path->nodes[0];
1437 nr = 0;
1438 path->slots[0]++;
1439 while (path->slots[0] > 0) {
1440 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
1441
1442 if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
b3470b5d
DS
1443 ASSERT(found_key.objectid == block_group->start);
1444 ASSERT(found_key.offset == block_group->length);
a5ed9182
OS
1445 done = 1;
1446 nr++;
1447 path->slots[0]--;
1448 break;
1449 } else if (found_key.type == BTRFS_FREE_SPACE_EXTENT_KEY ||
1450 found_key.type == BTRFS_FREE_SPACE_BITMAP_KEY) {
1451 ASSERT(found_key.objectid >= start);
1452 ASSERT(found_key.objectid < end);
1453 ASSERT(found_key.objectid + found_key.offset <= end);
1454 nr++;
1455 path->slots[0]--;
1456 } else {
1457 ASSERT(0);
1458 }
1459 }
1460
1461 ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
1462 if (ret)
1463 goto out;
1464 btrfs_release_path(path);
1465 }
1466
1467 ret = 0;
1468out:
1469 btrfs_free_path(path);
1470 if (ret)
66642832 1471 btrfs_abort_transaction(trans, ret);
a5ed9182
OS
1472 return ret;
1473}
1474
1475static int load_free_space_bitmaps(struct btrfs_caching_control *caching_ctl,
1476 struct btrfs_path *path,
1477 u32 expected_extent_count)
1478{
32da5386 1479 struct btrfs_block_group *block_group;
a5ed9182
OS
1480 struct btrfs_fs_info *fs_info;
1481 struct btrfs_root *root;
1482 struct btrfs_key key;
1483 int prev_bit = 0, bit;
1484 /* Initialize to silence GCC. */
1485 u64 extent_start = 0;
1486 u64 end, offset;
1487 u64 total_found = 0;
1488 u32 extent_count = 0;
1489 int ret;
1490
1491 block_group = caching_ctl->block_group;
1492 fs_info = block_group->fs_info;
7939dd9f 1493 root = btrfs_free_space_root(block_group);
a5ed9182 1494
b3470b5d 1495 end = block_group->start + block_group->length;
a5ed9182
OS
1496
1497 while (1) {
1498 ret = btrfs_next_item(root, path);
1499 if (ret < 0)
1500 goto out;
1501 if (ret)
1502 break;
1503
1504 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1505
1506 if (key.type == BTRFS_FREE_SPACE_INFO_KEY)
1507 break;
1508
1509 ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
1510 ASSERT(key.objectid < end && key.objectid + key.offset <= end);
1511
a5ed9182
OS
1512 offset = key.objectid;
1513 while (offset < key.objectid + key.offset) {
1514 bit = free_space_test_bit(block_group, path, offset);
1515 if (prev_bit == 0 && bit == 1) {
1516 extent_start = offset;
1517 } else if (prev_bit == 1 && bit == 0) {
d8ccbd21
FM
1518 u64 space_added;
1519
3b9f0995
FM
1520 ret = btrfs_add_new_free_space(block_group,
1521 extent_start,
1522 offset,
1523 &space_added);
d8ccbd21
FM
1524 if (ret)
1525 goto out;
1526 total_found += space_added;
a5ed9182
OS
1527 if (total_found > CACHING_CTL_WAKE_UP) {
1528 total_found = 0;
1529 wake_up(&caching_ctl->wait);
1530 }
1531 extent_count++;
1532 }
1533 prev_bit = bit;
0b246afa 1534 offset += fs_info->sectorsize;
a5ed9182
OS
1535 }
1536 }
1537 if (prev_bit == 1) {
3b9f0995 1538 ret = btrfs_add_new_free_space(block_group, extent_start, end, NULL);
d8ccbd21
FM
1539 if (ret)
1540 goto out;
a5ed9182
OS
1541 extent_count++;
1542 }
1543
1544 if (extent_count != expected_extent_count) {
5d163e0e
JM
1545 btrfs_err(fs_info,
1546 "incorrect extent count for %llu; counted %u, expected %u",
b3470b5d 1547 block_group->start, extent_count,
a5ed9182
OS
1548 expected_extent_count);
1549 ASSERT(0);
1550 ret = -EIO;
1551 goto out;
1552 }
1553
a5ed9182
OS
1554 ret = 0;
1555out:
1556 return ret;
1557}
1558
1559static int load_free_space_extents(struct btrfs_caching_control *caching_ctl,
1560 struct btrfs_path *path,
1561 u32 expected_extent_count)
1562{
32da5386 1563 struct btrfs_block_group *block_group;
a5ed9182
OS
1564 struct btrfs_fs_info *fs_info;
1565 struct btrfs_root *root;
1566 struct btrfs_key key;
1567 u64 end;
1568 u64 total_found = 0;
1569 u32 extent_count = 0;
1570 int ret;
1571
1572 block_group = caching_ctl->block_group;
1573 fs_info = block_group->fs_info;
7939dd9f 1574 root = btrfs_free_space_root(block_group);
a5ed9182 1575
b3470b5d 1576 end = block_group->start + block_group->length;
a5ed9182
OS
1577
1578 while (1) {
d8ccbd21
FM
1579 u64 space_added;
1580
a5ed9182
OS
1581 ret = btrfs_next_item(root, path);
1582 if (ret < 0)
1583 goto out;
1584 if (ret)
1585 break;
1586
1587 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1588
1589 if (key.type == BTRFS_FREE_SPACE_INFO_KEY)
1590 break;
1591
1592 ASSERT(key.type == BTRFS_FREE_SPACE_EXTENT_KEY);
1593 ASSERT(key.objectid < end && key.objectid + key.offset <= end);
1594
3b9f0995
FM
1595 ret = btrfs_add_new_free_space(block_group, key.objectid,
1596 key.objectid + key.offset,
1597 &space_added);
d8ccbd21
FM
1598 if (ret)
1599 goto out;
1600 total_found += space_added;
a5ed9182
OS
1601 if (total_found > CACHING_CTL_WAKE_UP) {
1602 total_found = 0;
1603 wake_up(&caching_ctl->wait);
1604 }
1605 extent_count++;
1606 }
1607
1608 if (extent_count != expected_extent_count) {
5d163e0e
JM
1609 btrfs_err(fs_info,
1610 "incorrect extent count for %llu; counted %u, expected %u",
b3470b5d 1611 block_group->start, extent_count,
a5ed9182
OS
1612 expected_extent_count);
1613 ASSERT(0);
1614 ret = -EIO;
1615 goto out;
1616 }
1617
a5ed9182
OS
1618 ret = 0;
1619out:
1620 return ret;
1621}
1622
1623int load_free_space_tree(struct btrfs_caching_control *caching_ctl)
1624{
32da5386 1625 struct btrfs_block_group *block_group;
a5ed9182
OS
1626 struct btrfs_free_space_info *info;
1627 struct btrfs_path *path;
1628 u32 extent_count, flags;
1629 int ret;
1630
1631 block_group = caching_ctl->block_group;
a5ed9182
OS
1632
1633 path = btrfs_alloc_path();
1634 if (!path)
1635 return -ENOMEM;
1636
1637 /*
1638 * Just like caching_thread() doesn't want to deadlock on the extent
1639 * tree, we don't want to deadlock on the free space tree.
1640 */
1641 path->skip_locking = 1;
1642 path->search_commit_root = 1;
7ce311d5 1643 path->reada = READA_FORWARD;
a5ed9182 1644
2ccf545e 1645 info = search_free_space_info(NULL, block_group, path, 0);
a5ed9182
OS
1646 if (IS_ERR(info)) {
1647 ret = PTR_ERR(info);
1648 goto out;
1649 }
1650 extent_count = btrfs_free_space_extent_count(path->nodes[0], info);
1651 flags = btrfs_free_space_flags(path->nodes[0], info);
1652
1653 /*
1654 * We left path pointing to the free space info item, so now
1655 * load_free_space_foo can just iterate through the free space tree from
1656 * there.
1657 */
1658 if (flags & BTRFS_FREE_SPACE_USING_BITMAPS)
1659 ret = load_free_space_bitmaps(caching_ctl, path, extent_count);
1660 else
1661 ret = load_free_space_extents(caching_ctl, path, extent_count);
1662
1663out:
1664 btrfs_free_path(path);
1665 return ret;
1666}