btrfs: move transaction abort to the error site in btrfs_delete_free_space_tree()
[linux-2.6-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);
50564b65 92 btrfs_mark_buffer_dirty(trans, leaf);
a5ed9182
OS
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);
50564b65 290 btrfs_mark_buffer_dirty(trans, leaf);
a5ed9182
OS
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);
50564b65 327 btrfs_mark_buffer_dirty(trans, leaf);
a5ed9182
OS
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);
50564b65 433 btrfs_mark_buffer_dirty(trans, leaf);
a5ed9182
OS
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);
50564b65 498 btrfs_mark_buffer_dirty(trans, path->nodes[0]);
a5ed9182
OS
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
50564b65
FM
536static void free_space_set_bits(struct btrfs_trans_handle *trans,
537 struct btrfs_block_group *block_group,
a5ed9182
OS
538 struct btrfs_path *path, u64 *start, u64 *size,
539 int bit)
540{
0b246afa 541 struct btrfs_fs_info *fs_info = block_group->fs_info;
a5ed9182
OS
542 struct extent_buffer *leaf;
543 struct btrfs_key key;
544 u64 end = *start + *size;
545 u64 found_start, found_end;
546 unsigned long ptr, first, last;
547
548 leaf = path->nodes[0];
549 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
550 ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
551
552 found_start = key.objectid;
553 found_end = key.objectid + key.offset;
554 ASSERT(*start >= found_start && *start < found_end);
555 ASSERT(end > found_start);
556
557 if (end > found_end)
558 end = found_end;
559
560 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
ab108d99
DS
561 first = (*start - found_start) >> fs_info->sectorsize_bits;
562 last = (end - found_start) >> fs_info->sectorsize_bits;
a5ed9182
OS
563 if (bit)
564 extent_buffer_bitmap_set(leaf, ptr, first, last - first);
565 else
566 extent_buffer_bitmap_clear(leaf, ptr, first, last - first);
50564b65 567 btrfs_mark_buffer_dirty(trans, leaf);
a5ed9182
OS
568
569 *size -= end - *start;
570 *start = end;
571}
572
573/*
574 * We can't use btrfs_next_item() in modify_free_space_bitmap() because
575 * btrfs_next_leaf() doesn't get the path for writing. We can forgo the fancy
576 * tree walking in btrfs_next_leaf() anyways because we know exactly what we're
577 * looking for.
578 */
579static int free_space_next_bitmap(struct btrfs_trans_handle *trans,
580 struct btrfs_root *root, struct btrfs_path *p)
581{
582 struct btrfs_key key;
583
584 if (p->slots[0] + 1 < btrfs_header_nritems(p->nodes[0])) {
585 p->slots[0]++;
586 return 0;
587 }
588
589 btrfs_item_key_to_cpu(p->nodes[0], &key, p->slots[0]);
590 btrfs_release_path(p);
591
592 key.objectid += key.offset;
593 key.type = (u8)-1;
594 key.offset = (u64)-1;
595
596 return btrfs_search_prev_slot(trans, root, &key, p, 0, 1);
597}
598
599/*
600 * If remove is 1, then we are removing free space, thus clearing bits in the
601 * bitmap. If remove is 0, then we are adding free space, thus setting bits in
602 * the bitmap.
603 */
604static int modify_free_space_bitmap(struct btrfs_trans_handle *trans,
32da5386 605 struct btrfs_block_group *block_group,
a5ed9182
OS
606 struct btrfs_path *path,
607 u64 start, u64 size, int remove)
608{
7939dd9f 609 struct btrfs_root *root = btrfs_free_space_root(block_group);
a5ed9182
OS
610 struct btrfs_key key;
611 u64 end = start + size;
612 u64 cur_start, cur_size;
613 int prev_bit, next_bit;
614 int new_extents;
615 int ret;
616
617 /*
618 * Read the bit for the block immediately before the extent of space if
619 * that block is within the block group.
620 */
b3470b5d 621 if (start > block_group->start) {
da17066c 622 u64 prev_block = start - block_group->fs_info->sectorsize;
a5ed9182
OS
623
624 key.objectid = prev_block;
625 key.type = (u8)-1;
626 key.offset = (u64)-1;
627
628 ret = btrfs_search_prev_slot(trans, root, &key, path, 0, 1);
629 if (ret)
630 goto out;
631
632 prev_bit = free_space_test_bit(block_group, path, prev_block);
633
634 /* The previous block may have been in the previous bitmap. */
635 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
636 if (start >= key.objectid + key.offset) {
637 ret = free_space_next_bitmap(trans, root, path);
638 if (ret)
639 goto out;
640 }
641 } else {
642 key.objectid = start;
643 key.type = (u8)-1;
644 key.offset = (u64)-1;
645
646 ret = btrfs_search_prev_slot(trans, root, &key, path, 0, 1);
647 if (ret)
648 goto out;
649
650 prev_bit = -1;
651 }
652
653 /*
654 * Iterate over all of the bitmaps overlapped by the extent of space,
655 * clearing/setting bits as required.
656 */
657 cur_start = start;
658 cur_size = size;
659 while (1) {
50564b65 660 free_space_set_bits(trans, block_group, path, &cur_start, &cur_size,
a5ed9182
OS
661 !remove);
662 if (cur_size == 0)
663 break;
664 ret = free_space_next_bitmap(trans, root, path);
665 if (ret)
666 goto out;
667 }
668
669 /*
670 * Read the bit for the block immediately after the extent of space if
671 * that block is within the block group.
672 */
b3470b5d 673 if (end < block_group->start + block_group->length) {
a5ed9182
OS
674 /* The next block may be in the next bitmap. */
675 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
676 if (end >= key.objectid + key.offset) {
677 ret = free_space_next_bitmap(trans, root, path);
678 if (ret)
679 goto out;
680 }
681
682 next_bit = free_space_test_bit(block_group, path, end);
683 } else {
684 next_bit = -1;
685 }
686
687 if (remove) {
688 new_extents = -1;
689 if (prev_bit == 1) {
690 /* Leftover on the left. */
691 new_extents++;
692 }
693 if (next_bit == 1) {
694 /* Leftover on the right. */
695 new_extents++;
696 }
697 } else {
698 new_extents = 1;
699 if (prev_bit == 1) {
700 /* Merging with neighbor on the left. */
701 new_extents--;
702 }
703 if (next_bit == 1) {
704 /* Merging with neighbor on the right. */
705 new_extents--;
706 }
707 }
708
709 btrfs_release_path(path);
690d7682 710 ret = update_free_space_extent_count(trans, block_group, path,
a5ed9182
OS
711 new_extents);
712
713out:
714 return ret;
715}
716
717static int remove_free_space_extent(struct btrfs_trans_handle *trans,
32da5386 718 struct btrfs_block_group *block_group,
a5ed9182
OS
719 struct btrfs_path *path,
720 u64 start, u64 size)
721{
7939dd9f 722 struct btrfs_root *root = btrfs_free_space_root(block_group);
a5ed9182
OS
723 struct btrfs_key key;
724 u64 found_start, found_end;
725 u64 end = start + size;
726 int new_extents = -1;
727 int ret;
728
729 key.objectid = start;
730 key.type = (u8)-1;
731 key.offset = (u64)-1;
732
733 ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
734 if (ret)
735 goto out;
736
737 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
738
739 ASSERT(key.type == BTRFS_FREE_SPACE_EXTENT_KEY);
740
741 found_start = key.objectid;
742 found_end = key.objectid + key.offset;
743 ASSERT(start >= found_start && end <= found_end);
744
745 /*
746 * Okay, now that we've found the free space extent which contains the
747 * free space that we are removing, there are four cases:
748 *
749 * 1. We're using the whole extent: delete the key we found and
750 * decrement the free space extent count.
751 * 2. We are using part of the extent starting at the beginning: delete
752 * the key we found and insert a new key representing the leftover at
753 * the end. There is no net change in the number of extents.
754 * 3. We are using part of the extent ending at the end: delete the key
755 * we found and insert a new key representing the leftover at the
756 * beginning. There is no net change in the number of extents.
757 * 4. We are using part of the extent in the middle: delete the key we
758 * found and insert two new keys representing the leftovers on each
759 * side. Where we used to have one extent, we now have two, so increment
760 * the extent count. We may need to convert the block group to bitmaps
761 * as a result.
762 */
763
764 /* Delete the existing key (cases 1-4). */
765 ret = btrfs_del_item(trans, root, path);
766 if (ret)
767 goto out;
768
769 /* Add a key for leftovers at the beginning (cases 3 and 4). */
770 if (start > found_start) {
771 key.objectid = found_start;
772 key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
773 key.offset = start - found_start;
774
775 btrfs_release_path(path);
776 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
777 if (ret)
778 goto out;
779 new_extents++;
780 }
781
782 /* Add a key for leftovers at the end (cases 2 and 4). */
783 if (end < found_end) {
784 key.objectid = end;
785 key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
786 key.offset = found_end - end;
787
788 btrfs_release_path(path);
789 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
790 if (ret)
791 goto out;
792 new_extents++;
793 }
794
795 btrfs_release_path(path);
690d7682 796 ret = update_free_space_extent_count(trans, block_group, path,
a5ed9182
OS
797 new_extents);
798
799out:
800 return ret;
801}
802
ce9f967f 803EXPORT_FOR_TESTS
a5ed9182 804int __remove_from_free_space_tree(struct btrfs_trans_handle *trans,
32da5386 805 struct btrfs_block_group *block_group,
a5ed9182
OS
806 struct btrfs_path *path, u64 start, u64 size)
807{
808 struct btrfs_free_space_info *info;
809 u32 flags;
810 int ret;
811
0d7764ff 812 if (test_bit(BLOCK_GROUP_FLAG_NEEDS_FREE_SPACE, &block_group->runtime_flags)) {
9a7e0f92 813 ret = __add_block_group_free_space(trans, block_group, path);
a5ed9182
OS
814 if (ret)
815 return ret;
816 }
817
2ccf545e 818 info = search_free_space_info(NULL, block_group, path, 0);
a5ed9182
OS
819 if (IS_ERR(info))
820 return PTR_ERR(info);
821 flags = btrfs_free_space_flags(path->nodes[0], info);
822 btrfs_release_path(path);
823
824 if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) {
85a7ef13
NB
825 return modify_free_space_bitmap(trans, block_group, path,
826 start, size, 1);
a5ed9182 827 } else {
e581168d
NB
828 return remove_free_space_extent(trans, block_group, path,
829 start, size);
a5ed9182
OS
830 }
831}
832
833int remove_from_free_space_tree(struct btrfs_trans_handle *trans,
a5ed9182
OS
834 u64 start, u64 size)
835{
32da5386 836 struct btrfs_block_group *block_group;
a5ed9182
OS
837 struct btrfs_path *path;
838 int ret;
839
25a356d3 840 if (!btrfs_fs_compat_ro(trans->fs_info, FREE_SPACE_TREE))
a5ed9182
OS
841 return 0;
842
843 path = btrfs_alloc_path();
844 if (!path) {
845 ret = -ENOMEM;
846 goto out;
847 }
848
25a356d3 849 block_group = btrfs_lookup_block_group(trans->fs_info, start);
a5ed9182
OS
850 if (!block_group) {
851 ASSERT(0);
852 ret = -ENOENT;
853 goto out;
854 }
855
856 mutex_lock(&block_group->free_space_lock);
c31683a6
NB
857 ret = __remove_from_free_space_tree(trans, block_group, path, start,
858 size);
a5ed9182
OS
859 mutex_unlock(&block_group->free_space_lock);
860
861 btrfs_put_block_group(block_group);
862out:
863 btrfs_free_path(path);
864 if (ret)
66642832 865 btrfs_abort_transaction(trans, ret);
a5ed9182
OS
866 return ret;
867}
868
869static int add_free_space_extent(struct btrfs_trans_handle *trans,
32da5386 870 struct btrfs_block_group *block_group,
a5ed9182
OS
871 struct btrfs_path *path,
872 u64 start, u64 size)
873{
7939dd9f 874 struct btrfs_root *root = btrfs_free_space_root(block_group);
a5ed9182
OS
875 struct btrfs_key key, new_key;
876 u64 found_start, found_end;
877 u64 end = start + size;
878 int new_extents = 1;
879 int ret;
880
881 /*
882 * We are adding a new extent of free space, but we need to merge
883 * extents. There are four cases here:
884 *
885 * 1. The new extent does not have any immediate neighbors to merge
886 * with: add the new key and increment the free space extent count. We
887 * may need to convert the block group to bitmaps as a result.
888 * 2. The new extent has an immediate neighbor before it: remove the
889 * previous key and insert a new key combining both of them. There is no
890 * net change in the number of extents.
891 * 3. The new extent has an immediate neighbor after it: remove the next
892 * key and insert a new key combining both of them. There is no net
893 * change in the number of extents.
894 * 4. The new extent has immediate neighbors on both sides: remove both
895 * of the keys and insert a new key combining all of them. Where we used
896 * to have two extents, we now have one, so decrement the extent count.
897 */
898
899 new_key.objectid = start;
900 new_key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
901 new_key.offset = size;
902
903 /* Search for a neighbor on the left. */
b3470b5d 904 if (start == block_group->start)
a5ed9182
OS
905 goto right;
906 key.objectid = start - 1;
907 key.type = (u8)-1;
908 key.offset = (u64)-1;
909
910 ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
911 if (ret)
912 goto out;
913
914 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
915
916 if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) {
917 ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY);
918 btrfs_release_path(path);
919 goto right;
920 }
921
922 found_start = key.objectid;
923 found_end = key.objectid + key.offset;
b3470b5d
DS
924 ASSERT(found_start >= block_group->start &&
925 found_end > block_group->start);
a5ed9182
OS
926 ASSERT(found_start < start && found_end <= start);
927
928 /*
929 * Delete the neighbor on the left and absorb it into the new key (cases
930 * 2 and 4).
931 */
932 if (found_end == start) {
933 ret = btrfs_del_item(trans, root, path);
934 if (ret)
935 goto out;
936 new_key.objectid = found_start;
937 new_key.offset += key.offset;
938 new_extents--;
939 }
940 btrfs_release_path(path);
941
942right:
943 /* Search for a neighbor on the right. */
b3470b5d 944 if (end == block_group->start + block_group->length)
a5ed9182
OS
945 goto insert;
946 key.objectid = end;
947 key.type = (u8)-1;
948 key.offset = (u64)-1;
949
950 ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
951 if (ret)
952 goto out;
953
954 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
955
956 if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) {
957 ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY);
958 btrfs_release_path(path);
959 goto insert;
960 }
961
962 found_start = key.objectid;
963 found_end = key.objectid + key.offset;
b3470b5d
DS
964 ASSERT(found_start >= block_group->start &&
965 found_end > block_group->start);
a5ed9182
OS
966 ASSERT((found_start < start && found_end <= start) ||
967 (found_start >= end && found_end > end));
968
969 /*
970 * Delete the neighbor on the right and absorb it into the new key
971 * (cases 3 and 4).
972 */
973 if (found_start == end) {
974 ret = btrfs_del_item(trans, root, path);
975 if (ret)
976 goto out;
977 new_key.offset += key.offset;
978 new_extents--;
979 }
980 btrfs_release_path(path);
981
982insert:
983 /* Insert the new key (cases 1-4). */
984 ret = btrfs_insert_empty_item(trans, root, path, &new_key, 0);
985 if (ret)
986 goto out;
987
988 btrfs_release_path(path);
690d7682 989 ret = update_free_space_extent_count(trans, block_group, path,
a5ed9182
OS
990 new_extents);
991
992out:
993 return ret;
994}
995
ce9f967f 996EXPORT_FOR_TESTS
a5ed9182 997int __add_to_free_space_tree(struct btrfs_trans_handle *trans,
32da5386 998 struct btrfs_block_group *block_group,
a5ed9182
OS
999 struct btrfs_path *path, u64 start, u64 size)
1000{
1001 struct btrfs_free_space_info *info;
1002 u32 flags;
1003 int ret;
1004
0d7764ff 1005 if (test_bit(BLOCK_GROUP_FLAG_NEEDS_FREE_SPACE, &block_group->runtime_flags)) {
9a7e0f92 1006 ret = __add_block_group_free_space(trans, block_group, path);
a5ed9182
OS
1007 if (ret)
1008 return ret;
1009 }
1010
2ccf545e 1011 info = search_free_space_info(NULL, block_group, path, 0);
a5ed9182
OS
1012 if (IS_ERR(info))
1013 return PTR_ERR(info);
1014 flags = btrfs_free_space_flags(path->nodes[0], info);
1015 btrfs_release_path(path);
1016
1017 if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) {
85a7ef13
NB
1018 return modify_free_space_bitmap(trans, block_group, path,
1019 start, size, 0);
a5ed9182 1020 } else {
5cb17822
NB
1021 return add_free_space_extent(trans, block_group, path, start,
1022 size);
a5ed9182
OS
1023 }
1024}
1025
1026int add_to_free_space_tree(struct btrfs_trans_handle *trans,
a5ed9182
OS
1027 u64 start, u64 size)
1028{
32da5386 1029 struct btrfs_block_group *block_group;
a5ed9182
OS
1030 struct btrfs_path *path;
1031 int ret;
1032
e7355e50 1033 if (!btrfs_fs_compat_ro(trans->fs_info, FREE_SPACE_TREE))
a5ed9182
OS
1034 return 0;
1035
1036 path = btrfs_alloc_path();
1037 if (!path) {
1038 ret = -ENOMEM;
1039 goto out;
1040 }
1041
e7355e50 1042 block_group = btrfs_lookup_block_group(trans->fs_info, start);
a5ed9182
OS
1043 if (!block_group) {
1044 ASSERT(0);
1045 ret = -ENOENT;
1046 goto out;
1047 }
1048
1049 mutex_lock(&block_group->free_space_lock);
2d5cffa1 1050 ret = __add_to_free_space_tree(trans, block_group, path, start, size);
a5ed9182
OS
1051 mutex_unlock(&block_group->free_space_lock);
1052
1053 btrfs_put_block_group(block_group);
1054out:
1055 btrfs_free_path(path);
1056 if (ret)
66642832 1057 btrfs_abort_transaction(trans, ret);
a5ed9182
OS
1058 return ret;
1059}
1060
1061/*
1062 * Populate the free space tree by walking the extent tree. Operations on the
1063 * extent tree that happen as a result of writes to the free space tree will go
1064 * through the normal add/remove hooks.
1065 */
1066static int populate_free_space_tree(struct btrfs_trans_handle *trans,
32da5386 1067 struct btrfs_block_group *block_group)
a5ed9182 1068{
29cbcf40 1069 struct btrfs_root *extent_root;
a5ed9182
OS
1070 struct btrfs_path *path, *path2;
1071 struct btrfs_key key;
1072 u64 start, end;
1073 int ret;
1074
1075 path = btrfs_alloc_path();
1076 if (!path)
1077 return -ENOMEM;
019599ad 1078 path->reada = READA_FORWARD;
a5ed9182
OS
1079
1080 path2 = btrfs_alloc_path();
1081 if (!path2) {
1082 btrfs_free_path(path);
1083 return -ENOMEM;
1084 }
1085
66afee18 1086 ret = add_new_free_space_info(trans, block_group, path2);
a5ed9182
OS
1087 if (ret)
1088 goto out;
1089
511711af
CM
1090 mutex_lock(&block_group->free_space_lock);
1091
a5ed9182
OS
1092 /*
1093 * Iterate through all of the extent and metadata items in this block
1094 * group, adding the free space between them and the free space at the
1095 * end. Note that EXTENT_ITEM and METADATA_ITEM are less than
1096 * BLOCK_GROUP_ITEM, so an extent may precede the block group that it's
1097 * contained in.
1098 */
b3470b5d 1099 key.objectid = block_group->start;
a5ed9182
OS
1100 key.type = BTRFS_EXTENT_ITEM_KEY;
1101 key.offset = 0;
1102
29cbcf40 1103 extent_root = btrfs_extent_root(trans->fs_info, key.objectid);
a5ed9182
OS
1104 ret = btrfs_search_slot_for_read(extent_root, &key, path, 1, 0);
1105 if (ret < 0)
511711af 1106 goto out_locked;
a5ed9182
OS
1107 ASSERT(ret == 0);
1108
b3470b5d
DS
1109 start = block_group->start;
1110 end = block_group->start + block_group->length;
a5ed9182
OS
1111 while (1) {
1112 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1113
1114 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
1115 key.type == BTRFS_METADATA_ITEM_KEY) {
1116 if (key.objectid >= end)
1117 break;
1118
1119 if (start < key.objectid) {
2d5cffa1 1120 ret = __add_to_free_space_tree(trans,
a5ed9182
OS
1121 block_group,
1122 path2, start,
1123 key.objectid -
1124 start);
1125 if (ret)
511711af 1126 goto out_locked;
a5ed9182
OS
1127 }
1128 start = key.objectid;
1129 if (key.type == BTRFS_METADATA_ITEM_KEY)
ffa9a9ef 1130 start += trans->fs_info->nodesize;
a5ed9182
OS
1131 else
1132 start += key.offset;
1133 } else if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
b3470b5d 1134 if (key.objectid != block_group->start)
a5ed9182
OS
1135 break;
1136 }
1137
1138 ret = btrfs_next_item(extent_root, path);
1139 if (ret < 0)
511711af 1140 goto out_locked;
a5ed9182
OS
1141 if (ret)
1142 break;
1143 }
1144 if (start < end) {
2d5cffa1
NB
1145 ret = __add_to_free_space_tree(trans, block_group, path2,
1146 start, end - start);
a5ed9182 1147 if (ret)
511711af 1148 goto out_locked;
a5ed9182
OS
1149 }
1150
1151 ret = 0;
511711af
CM
1152out_locked:
1153 mutex_unlock(&block_group->free_space_lock);
a5ed9182
OS
1154out:
1155 btrfs_free_path(path2);
1156 btrfs_free_path(path);
1157 return ret;
1158}
1159
1160int btrfs_create_free_space_tree(struct btrfs_fs_info *fs_info)
1161{
1162 struct btrfs_trans_handle *trans;
1163 struct btrfs_root *tree_root = fs_info->tree_root;
1164 struct btrfs_root *free_space_root;
32da5386 1165 struct btrfs_block_group *block_group;
a5ed9182
OS
1166 struct rb_node *node;
1167 int ret;
1168
1169 trans = btrfs_start_transaction(tree_root, 0);
1170 if (IS_ERR(trans))
1171 return PTR_ERR(trans);
1172
afcdd129 1173 set_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
2f96e402 1174 set_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags);
9b7a2440 1175 free_space_root = btrfs_create_tree(trans,
a5ed9182
OS
1176 BTRFS_FREE_SPACE_TREE_OBJECTID);
1177 if (IS_ERR(free_space_root)) {
1178 ret = PTR_ERR(free_space_root);
1179 goto abort;
1180 }
abed4aaa
JB
1181 ret = btrfs_global_root_insert(free_space_root);
1182 if (ret) {
1183 btrfs_put_root(free_space_root);
1184 goto abort;
1185 }
a5ed9182 1186
08dddb29 1187 node = rb_first_cached(&fs_info->block_group_cache_tree);
a5ed9182 1188 while (node) {
32da5386 1189 block_group = rb_entry(node, struct btrfs_block_group,
a5ed9182 1190 cache_node);
ffa9a9ef 1191 ret = populate_free_space_tree(trans, block_group);
a5ed9182
OS
1192 if (ret)
1193 goto abort;
1194 node = rb_next(node);
1195 }
1196
1197 btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE);
6675df31 1198 btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID);
afcdd129 1199 clear_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
2f96e402 1200 ret = btrfs_commit_transaction(trans);
a5ed9182 1201
2f96e402
JB
1202 /*
1203 * Now that we've committed the transaction any reading of our commit
1204 * root will be safe, so we can cache from the free space tree now.
1205 */
1206 clear_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags);
1207 return ret;
a5ed9182
OS
1208
1209abort:
afcdd129 1210 clear_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
2f96e402 1211 clear_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags);
66642832 1212 btrfs_abort_transaction(trans, ret);
3a45bb20 1213 btrfs_end_transaction(trans);
a5ed9182
OS
1214 return ret;
1215}
1216
1217static int clear_free_space_tree(struct btrfs_trans_handle *trans,
1218 struct btrfs_root *root)
1219{
1220 struct btrfs_path *path;
1221 struct btrfs_key key;
1222 int nr;
1223 int ret;
1224
1225 path = btrfs_alloc_path();
1226 if (!path)
1227 return -ENOMEM;
1228
a5ed9182
OS
1229 key.objectid = 0;
1230 key.type = 0;
1231 key.offset = 0;
1232
1233 while (1) {
1234 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1235 if (ret < 0)
1236 goto out;
1237
1238 nr = btrfs_header_nritems(path->nodes[0]);
1239 if (!nr)
1240 break;
1241
1242 path->slots[0] = 0;
1243 ret = btrfs_del_items(trans, root, path, 0, nr);
1244 if (ret)
1245 goto out;
1246
1247 btrfs_release_path(path);
1248 }
1249
1250 ret = 0;
1251out:
1252 btrfs_free_path(path);
1253 return ret;
1254}
1255
1d6a4fc8 1256int btrfs_delete_free_space_tree(struct btrfs_fs_info *fs_info)
a5ed9182
OS
1257{
1258 struct btrfs_trans_handle *trans;
1259 struct btrfs_root *tree_root = fs_info->tree_root;
abed4aaa
JB
1260 struct btrfs_key key = {
1261 .objectid = BTRFS_FREE_SPACE_TREE_OBJECTID,
1262 .type = BTRFS_ROOT_ITEM_KEY,
1263 .offset = 0,
1264 };
1265 struct btrfs_root *free_space_root = btrfs_global_root(fs_info, &key);
a5ed9182
OS
1266 int ret;
1267
1268 trans = btrfs_start_transaction(tree_root, 0);
1269 if (IS_ERR(trans))
1270 return PTR_ERR(trans);
1271
1272 btrfs_clear_fs_compat_ro(fs_info, FREE_SPACE_TREE);
6675df31 1273 btrfs_clear_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID);
a5ed9182
OS
1274
1275 ret = clear_free_space_tree(trans, free_space_root);
0896ce75
DS
1276 if (ret) {
1277 btrfs_abort_transaction(trans, ret);
1278 btrfs_end_transaction(trans);
1279 return ret;
1280 }
a5ed9182 1281
ab9ce7d4 1282 ret = btrfs_del_root(trans, &free_space_root->root_key);
0896ce75
DS
1283 if (ret) {
1284 btrfs_abort_transaction(trans, ret);
1285 btrfs_end_transaction(trans);
1286 return ret;
1287 }
a5ed9182 1288
abed4aaa 1289 btrfs_global_root_delete(free_space_root);
babebf02
FM
1290
1291 spin_lock(&fs_info->trans_lock);
a5ed9182 1292 list_del(&free_space_root->dirty_list);
babebf02 1293 spin_unlock(&fs_info->trans_lock);
a5ed9182
OS
1294
1295 btrfs_tree_lock(free_space_root->node);
190a8339 1296 btrfs_clear_buffer_dirty(trans, free_space_root->node);
a5ed9182 1297 btrfs_tree_unlock(free_space_root->node);
7a163608
FM
1298 btrfs_free_tree_block(trans, btrfs_root_id(free_space_root),
1299 free_space_root->node, 0, 1);
a5ed9182 1300
00246528 1301 btrfs_put_root(free_space_root);
a5ed9182 1302
0bef7109 1303 return btrfs_commit_transaction(trans);
a5ed9182
OS
1304}
1305
1d6a4fc8
QW
1306int btrfs_rebuild_free_space_tree(struct btrfs_fs_info *fs_info)
1307{
1308 struct btrfs_trans_handle *trans;
1309 struct btrfs_key key = {
1310 .objectid = BTRFS_FREE_SPACE_TREE_OBJECTID,
1311 .type = BTRFS_ROOT_ITEM_KEY,
1312 .offset = 0,
1313 };
1314 struct btrfs_root *free_space_root = btrfs_global_root(fs_info, &key);
1315 struct rb_node *node;
1316 int ret;
1317
1318 trans = btrfs_start_transaction(free_space_root, 1);
1319 if (IS_ERR(trans))
1320 return PTR_ERR(trans);
1321
1322 set_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1323 set_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags);
1324
1325 ret = clear_free_space_tree(trans, free_space_root);
1326 if (ret)
1327 goto abort;
1328
1329 node = rb_first_cached(&fs_info->block_group_cache_tree);
1330 while (node) {
1331 struct btrfs_block_group *block_group;
1332
1333 block_group = rb_entry(node, struct btrfs_block_group,
1334 cache_node);
1335 ret = populate_free_space_tree(trans, block_group);
1336 if (ret)
1337 goto abort;
1338 node = rb_next(node);
1339 }
1340
1341 btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE);
1342 btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID);
1343 clear_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1344
1345 ret = btrfs_commit_transaction(trans);
1346 clear_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags);
1347 return ret;
1348abort:
1349 btrfs_abort_transaction(trans, ret);
1350 btrfs_end_transaction(trans);
1351 return ret;
1352}
1353
a5ed9182 1354static int __add_block_group_free_space(struct btrfs_trans_handle *trans,
32da5386 1355 struct btrfs_block_group *block_group,
a5ed9182
OS
1356 struct btrfs_path *path)
1357{
a5ed9182
OS
1358 int ret;
1359
0d7764ff 1360 clear_bit(BLOCK_GROUP_FLAG_NEEDS_FREE_SPACE, &block_group->runtime_flags);
a5ed9182 1361
66afee18 1362 ret = add_new_free_space_info(trans, block_group, path);
a5ed9182
OS
1363 if (ret)
1364 return ret;
1365
2d5cffa1 1366 return __add_to_free_space_tree(trans, block_group, path,
b3470b5d
DS
1367 block_group->start,
1368 block_group->length);
a5ed9182
OS
1369}
1370
1371int add_block_group_free_space(struct btrfs_trans_handle *trans,
32da5386 1372 struct btrfs_block_group *block_group)
a5ed9182 1373{
e4e0711c 1374 struct btrfs_fs_info *fs_info = trans->fs_info;
a5ed9182
OS
1375 struct btrfs_path *path = NULL;
1376 int ret = 0;
1377
1378 if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
1379 return 0;
1380
1381 mutex_lock(&block_group->free_space_lock);
0d7764ff 1382 if (!test_bit(BLOCK_GROUP_FLAG_NEEDS_FREE_SPACE, &block_group->runtime_flags))
a5ed9182
OS
1383 goto out;
1384
1385 path = btrfs_alloc_path();
1386 if (!path) {
1387 ret = -ENOMEM;
1388 goto out;
1389 }
1390
9a7e0f92 1391 ret = __add_block_group_free_space(trans, block_group, path);
a5ed9182
OS
1392
1393out:
1394 btrfs_free_path(path);
1395 mutex_unlock(&block_group->free_space_lock);
1396 if (ret)
66642832 1397 btrfs_abort_transaction(trans, ret);
a5ed9182
OS
1398 return ret;
1399}
1400
1401int remove_block_group_free_space(struct btrfs_trans_handle *trans,
32da5386 1402 struct btrfs_block_group *block_group)
a5ed9182 1403{
7939dd9f 1404 struct btrfs_root *root = btrfs_free_space_root(block_group);
a5ed9182
OS
1405 struct btrfs_path *path;
1406 struct btrfs_key key, found_key;
1407 struct extent_buffer *leaf;
1408 u64 start, end;
1409 int done = 0, nr;
1410 int ret;
1411
f3f72779 1412 if (!btrfs_fs_compat_ro(trans->fs_info, FREE_SPACE_TREE))
a5ed9182
OS
1413 return 0;
1414
0d7764ff 1415 if (test_bit(BLOCK_GROUP_FLAG_NEEDS_FREE_SPACE, &block_group->runtime_flags)) {
a5ed9182
OS
1416 /* We never added this block group to the free space tree. */
1417 return 0;
1418 }
1419
1420 path = btrfs_alloc_path();
1421 if (!path) {
1422 ret = -ENOMEM;
1423 goto out;
1424 }
1425
b3470b5d
DS
1426 start = block_group->start;
1427 end = block_group->start + block_group->length;
a5ed9182
OS
1428
1429 key.objectid = end - 1;
1430 key.type = (u8)-1;
1431 key.offset = (u64)-1;
1432
1433 while (!done) {
1434 ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
1435 if (ret)
1436 goto out;
1437
1438 leaf = path->nodes[0];
1439 nr = 0;
1440 path->slots[0]++;
1441 while (path->slots[0] > 0) {
1442 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
1443
1444 if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
b3470b5d
DS
1445 ASSERT(found_key.objectid == block_group->start);
1446 ASSERT(found_key.offset == block_group->length);
a5ed9182
OS
1447 done = 1;
1448 nr++;
1449 path->slots[0]--;
1450 break;
1451 } else if (found_key.type == BTRFS_FREE_SPACE_EXTENT_KEY ||
1452 found_key.type == BTRFS_FREE_SPACE_BITMAP_KEY) {
1453 ASSERT(found_key.objectid >= start);
1454 ASSERT(found_key.objectid < end);
1455 ASSERT(found_key.objectid + found_key.offset <= end);
1456 nr++;
1457 path->slots[0]--;
1458 } else {
1459 ASSERT(0);
1460 }
1461 }
1462
1463 ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
1464 if (ret)
1465 goto out;
1466 btrfs_release_path(path);
1467 }
1468
1469 ret = 0;
1470out:
1471 btrfs_free_path(path);
1472 if (ret)
66642832 1473 btrfs_abort_transaction(trans, ret);
a5ed9182
OS
1474 return ret;
1475}
1476
1477static int load_free_space_bitmaps(struct btrfs_caching_control *caching_ctl,
1478 struct btrfs_path *path,
1479 u32 expected_extent_count)
1480{
32da5386 1481 struct btrfs_block_group *block_group;
a5ed9182
OS
1482 struct btrfs_fs_info *fs_info;
1483 struct btrfs_root *root;
1484 struct btrfs_key key;
1485 int prev_bit = 0, bit;
1486 /* Initialize to silence GCC. */
1487 u64 extent_start = 0;
1488 u64 end, offset;
1489 u64 total_found = 0;
1490 u32 extent_count = 0;
1491 int ret;
1492
1493 block_group = caching_ctl->block_group;
1494 fs_info = block_group->fs_info;
7939dd9f 1495 root = btrfs_free_space_root(block_group);
a5ed9182 1496
b3470b5d 1497 end = block_group->start + block_group->length;
a5ed9182
OS
1498
1499 while (1) {
1500 ret = btrfs_next_item(root, path);
1501 if (ret < 0)
1502 goto out;
1503 if (ret)
1504 break;
1505
1506 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1507
1508 if (key.type == BTRFS_FREE_SPACE_INFO_KEY)
1509 break;
1510
1511 ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
1512 ASSERT(key.objectid < end && key.objectid + key.offset <= end);
1513
a5ed9182
OS
1514 offset = key.objectid;
1515 while (offset < key.objectid + key.offset) {
1516 bit = free_space_test_bit(block_group, path, offset);
1517 if (prev_bit == 0 && bit == 1) {
1518 extent_start = offset;
1519 } else if (prev_bit == 1 && bit == 0) {
d8ccbd21
FM
1520 u64 space_added;
1521
3b9f0995
FM
1522 ret = btrfs_add_new_free_space(block_group,
1523 extent_start,
1524 offset,
1525 &space_added);
d8ccbd21
FM
1526 if (ret)
1527 goto out;
1528 total_found += space_added;
a5ed9182
OS
1529 if (total_found > CACHING_CTL_WAKE_UP) {
1530 total_found = 0;
1531 wake_up(&caching_ctl->wait);
1532 }
1533 extent_count++;
1534 }
1535 prev_bit = bit;
0b246afa 1536 offset += fs_info->sectorsize;
a5ed9182
OS
1537 }
1538 }
1539 if (prev_bit == 1) {
3b9f0995 1540 ret = btrfs_add_new_free_space(block_group, extent_start, end, NULL);
d8ccbd21
FM
1541 if (ret)
1542 goto out;
a5ed9182
OS
1543 extent_count++;
1544 }
1545
1546 if (extent_count != expected_extent_count) {
5d163e0e
JM
1547 btrfs_err(fs_info,
1548 "incorrect extent count for %llu; counted %u, expected %u",
b3470b5d 1549 block_group->start, extent_count,
a5ed9182
OS
1550 expected_extent_count);
1551 ASSERT(0);
1552 ret = -EIO;
1553 goto out;
1554 }
1555
a5ed9182
OS
1556 ret = 0;
1557out:
1558 return ret;
1559}
1560
1561static int load_free_space_extents(struct btrfs_caching_control *caching_ctl,
1562 struct btrfs_path *path,
1563 u32 expected_extent_count)
1564{
32da5386 1565 struct btrfs_block_group *block_group;
a5ed9182
OS
1566 struct btrfs_fs_info *fs_info;
1567 struct btrfs_root *root;
1568 struct btrfs_key key;
1569 u64 end;
1570 u64 total_found = 0;
1571 u32 extent_count = 0;
1572 int ret;
1573
1574 block_group = caching_ctl->block_group;
1575 fs_info = block_group->fs_info;
7939dd9f 1576 root = btrfs_free_space_root(block_group);
a5ed9182 1577
b3470b5d 1578 end = block_group->start + block_group->length;
a5ed9182
OS
1579
1580 while (1) {
d8ccbd21
FM
1581 u64 space_added;
1582
a5ed9182
OS
1583 ret = btrfs_next_item(root, path);
1584 if (ret < 0)
1585 goto out;
1586 if (ret)
1587 break;
1588
1589 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1590
1591 if (key.type == BTRFS_FREE_SPACE_INFO_KEY)
1592 break;
1593
1594 ASSERT(key.type == BTRFS_FREE_SPACE_EXTENT_KEY);
1595 ASSERT(key.objectid < end && key.objectid + key.offset <= end);
1596
3b9f0995
FM
1597 ret = btrfs_add_new_free_space(block_group, key.objectid,
1598 key.objectid + key.offset,
1599 &space_added);
d8ccbd21
FM
1600 if (ret)
1601 goto out;
1602 total_found += space_added;
a5ed9182
OS
1603 if (total_found > CACHING_CTL_WAKE_UP) {
1604 total_found = 0;
1605 wake_up(&caching_ctl->wait);
1606 }
1607 extent_count++;
1608 }
1609
1610 if (extent_count != expected_extent_count) {
5d163e0e
JM
1611 btrfs_err(fs_info,
1612 "incorrect extent count for %llu; counted %u, expected %u",
b3470b5d 1613 block_group->start, extent_count,
a5ed9182
OS
1614 expected_extent_count);
1615 ASSERT(0);
1616 ret = -EIO;
1617 goto out;
1618 }
1619
a5ed9182
OS
1620 ret = 0;
1621out:
1622 return ret;
1623}
1624
1625int load_free_space_tree(struct btrfs_caching_control *caching_ctl)
1626{
32da5386 1627 struct btrfs_block_group *block_group;
a5ed9182
OS
1628 struct btrfs_free_space_info *info;
1629 struct btrfs_path *path;
1630 u32 extent_count, flags;
1631 int ret;
1632
1633 block_group = caching_ctl->block_group;
a5ed9182
OS
1634
1635 path = btrfs_alloc_path();
1636 if (!path)
1637 return -ENOMEM;
1638
1639 /*
1640 * Just like caching_thread() doesn't want to deadlock on the extent
1641 * tree, we don't want to deadlock on the free space tree.
1642 */
1643 path->skip_locking = 1;
1644 path->search_commit_root = 1;
7ce311d5 1645 path->reada = READA_FORWARD;
a5ed9182 1646
2ccf545e 1647 info = search_free_space_info(NULL, block_group, path, 0);
a5ed9182
OS
1648 if (IS_ERR(info)) {
1649 ret = PTR_ERR(info);
1650 goto out;
1651 }
1652 extent_count = btrfs_free_space_extent_count(path->nodes[0], info);
1653 flags = btrfs_free_space_flags(path->nodes[0], info);
1654
1655 /*
1656 * We left path pointing to the free space info item, so now
1657 * load_free_space_foo can just iterate through the free space tree from
1658 * there.
1659 */
1660 if (flags & BTRFS_FREE_SPACE_USING_BITMAPS)
1661 ret = load_free_space_bitmaps(caching_ctl, path, extent_count);
1662 else
1663 ret = load_free_space_extents(caching_ctl, path, extent_count);
1664
1665out:
1666 btrfs_free_path(path);
1667 return ret;
1668}