Commit | Line | Data |
---|---|---|
a5ed9182 OS |
1 | /* |
2 | * Copyright (C) 2015 Facebook. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
19 | #include <linux/kernel.h> | |
25ff17e8 | 20 | #include <linux/sched/mm.h> |
a5ed9182 OS |
21 | #include "ctree.h" |
22 | #include "disk-io.h" | |
23 | #include "locking.h" | |
24 | #include "free-space-tree.h" | |
25 | #include "transaction.h" | |
26 | ||
27 | static int __add_block_group_free_space(struct btrfs_trans_handle *trans, | |
28 | struct btrfs_fs_info *fs_info, | |
29 | struct btrfs_block_group_cache *block_group, | |
30 | struct btrfs_path *path); | |
31 | ||
32 | void set_free_space_tree_thresholds(struct btrfs_block_group_cache *cache) | |
33 | { | |
34 | u32 bitmap_range; | |
35 | size_t bitmap_size; | |
36 | u64 num_bitmaps, total_bitmap_size; | |
37 | ||
38 | /* | |
39 | * We convert to bitmaps when the disk space required for using extents | |
40 | * exceeds that required for using bitmaps. | |
41 | */ | |
da17066c | 42 | bitmap_range = cache->fs_info->sectorsize * BTRFS_FREE_SPACE_BITMAP_BITS; |
a5ed9182 OS |
43 | num_bitmaps = div_u64(cache->key.offset + bitmap_range - 1, |
44 | bitmap_range); | |
45 | bitmap_size = sizeof(struct btrfs_item) + BTRFS_FREE_SPACE_BITMAP_SIZE; | |
46 | total_bitmap_size = num_bitmaps * bitmap_size; | |
47 | cache->bitmap_high_thresh = div_u64(total_bitmap_size, | |
48 | sizeof(struct btrfs_item)); | |
49 | ||
50 | /* | |
51 | * We allow for a small buffer between the high threshold and low | |
52 | * threshold to avoid thrashing back and forth between the two formats. | |
53 | */ | |
54 | if (cache->bitmap_high_thresh > 100) | |
55 | cache->bitmap_low_thresh = cache->bitmap_high_thresh - 100; | |
56 | else | |
57 | cache->bitmap_low_thresh = 0; | |
58 | } | |
59 | ||
60 | static int add_new_free_space_info(struct btrfs_trans_handle *trans, | |
61 | struct btrfs_fs_info *fs_info, | |
62 | struct btrfs_block_group_cache *block_group, | |
63 | struct btrfs_path *path) | |
64 | { | |
65 | struct btrfs_root *root = fs_info->free_space_root; | |
66 | struct btrfs_free_space_info *info; | |
67 | struct btrfs_key key; | |
68 | struct extent_buffer *leaf; | |
69 | int ret; | |
70 | ||
71 | key.objectid = block_group->key.objectid; | |
72 | key.type = BTRFS_FREE_SPACE_INFO_KEY; | |
73 | key.offset = block_group->key.offset; | |
74 | ||
75 | ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*info)); | |
76 | if (ret) | |
77 | goto out; | |
78 | ||
79 | leaf = path->nodes[0]; | |
80 | info = btrfs_item_ptr(leaf, path->slots[0], | |
81 | struct btrfs_free_space_info); | |
82 | btrfs_set_free_space_extent_count(leaf, info, 0); | |
83 | btrfs_set_free_space_flags(leaf, info, 0); | |
84 | btrfs_mark_buffer_dirty(leaf); | |
85 | ||
86 | ret = 0; | |
87 | out: | |
88 | btrfs_release_path(path); | |
89 | return ret; | |
90 | } | |
91 | ||
92 | struct btrfs_free_space_info * | |
93 | search_free_space_info(struct btrfs_trans_handle *trans, | |
94 | struct btrfs_fs_info *fs_info, | |
95 | struct btrfs_block_group_cache *block_group, | |
96 | struct btrfs_path *path, int cow) | |
97 | { | |
98 | struct btrfs_root *root = fs_info->free_space_root; | |
99 | struct btrfs_key key; | |
100 | int ret; | |
101 | ||
102 | key.objectid = block_group->key.objectid; | |
103 | key.type = BTRFS_FREE_SPACE_INFO_KEY; | |
104 | key.offset = block_group->key.offset; | |
105 | ||
106 | ret = btrfs_search_slot(trans, root, &key, path, 0, cow); | |
107 | if (ret < 0) | |
108 | return ERR_PTR(ret); | |
109 | if (ret != 0) { | |
ab8d0fc4 | 110 | btrfs_warn(fs_info, "missing free space info for %llu", |
a5ed9182 OS |
111 | block_group->key.objectid); |
112 | ASSERT(0); | |
113 | return ERR_PTR(-ENOENT); | |
114 | } | |
115 | ||
116 | return btrfs_item_ptr(path->nodes[0], path->slots[0], | |
117 | struct btrfs_free_space_info); | |
118 | } | |
119 | ||
120 | /* | |
121 | * btrfs_search_slot() but we're looking for the greatest key less than the | |
122 | * passed key. | |
123 | */ | |
124 | static int btrfs_search_prev_slot(struct btrfs_trans_handle *trans, | |
125 | struct btrfs_root *root, | |
126 | struct btrfs_key *key, struct btrfs_path *p, | |
127 | int ins_len, int cow) | |
128 | { | |
129 | int ret; | |
130 | ||
131 | ret = btrfs_search_slot(trans, root, key, p, ins_len, cow); | |
132 | if (ret < 0) | |
133 | return ret; | |
134 | ||
135 | if (ret == 0) { | |
136 | ASSERT(0); | |
137 | return -EIO; | |
138 | } | |
139 | ||
140 | if (p->slots[0] == 0) { | |
141 | ASSERT(0); | |
142 | return -EIO; | |
143 | } | |
144 | p->slots[0]--; | |
145 | ||
146 | return 0; | |
147 | } | |
148 | ||
149 | static inline u32 free_space_bitmap_size(u64 size, u32 sectorsize) | |
150 | { | |
151 | return DIV_ROUND_UP((u32)div_u64(size, sectorsize), BITS_PER_BYTE); | |
152 | } | |
153 | ||
2fe1d551 | 154 | static u8 *alloc_bitmap(u32 bitmap_size) |
a5ed9182 | 155 | { |
25ff17e8 OS |
156 | u8 *ret; |
157 | unsigned int nofs_flag; | |
79b134a2 DS |
158 | |
159 | /* | |
25ff17e8 OS |
160 | * GFP_NOFS doesn't work with kvmalloc(), but we really can't recurse |
161 | * into the filesystem as the free space bitmap can be modified in the | |
162 | * critical section of a transaction commit. | |
163 | * | |
164 | * TODO: push the memalloc_nofs_{save,restore}() to the caller where we | |
165 | * know that recursion is unsafe. | |
79b134a2 | 166 | */ |
25ff17e8 OS |
167 | nofs_flag = memalloc_nofs_save(); |
168 | ret = kvzalloc(bitmap_size, GFP_KERNEL); | |
169 | memalloc_nofs_restore(nofs_flag); | |
170 | return ret; | |
a5ed9182 OS |
171 | } |
172 | ||
173 | int convert_free_space_to_bitmaps(struct btrfs_trans_handle *trans, | |
174 | struct btrfs_fs_info *fs_info, | |
175 | struct btrfs_block_group_cache *block_group, | |
176 | struct btrfs_path *path) | |
177 | { | |
178 | struct btrfs_root *root = fs_info->free_space_root; | |
179 | struct btrfs_free_space_info *info; | |
180 | struct btrfs_key key, found_key; | |
181 | struct extent_buffer *leaf; | |
2fe1d551 | 182 | u8 *bitmap, *bitmap_cursor; |
a5ed9182 OS |
183 | u64 start, end; |
184 | u64 bitmap_range, i; | |
185 | u32 bitmap_size, flags, expected_extent_count; | |
186 | u32 extent_count = 0; | |
187 | int done = 0, nr; | |
188 | int ret; | |
189 | ||
190 | bitmap_size = free_space_bitmap_size(block_group->key.offset, | |
0b246afa | 191 | fs_info->sectorsize); |
a5ed9182 OS |
192 | bitmap = alloc_bitmap(bitmap_size); |
193 | if (!bitmap) { | |
194 | ret = -ENOMEM; | |
195 | goto out; | |
196 | } | |
197 | ||
198 | start = block_group->key.objectid; | |
199 | end = block_group->key.objectid + block_group->key.offset; | |
200 | ||
201 | key.objectid = end - 1; | |
202 | key.type = (u8)-1; | |
203 | key.offset = (u64)-1; | |
204 | ||
205 | while (!done) { | |
206 | ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1); | |
207 | if (ret) | |
208 | goto out; | |
209 | ||
210 | leaf = path->nodes[0]; | |
211 | nr = 0; | |
212 | path->slots[0]++; | |
213 | while (path->slots[0] > 0) { | |
214 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1); | |
215 | ||
216 | if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) { | |
217 | ASSERT(found_key.objectid == block_group->key.objectid); | |
218 | ASSERT(found_key.offset == block_group->key.offset); | |
219 | done = 1; | |
220 | break; | |
221 | } else if (found_key.type == BTRFS_FREE_SPACE_EXTENT_KEY) { | |
222 | u64 first, last; | |
223 | ||
224 | ASSERT(found_key.objectid >= start); | |
225 | ASSERT(found_key.objectid < end); | |
226 | ASSERT(found_key.objectid + found_key.offset <= end); | |
227 | ||
228 | first = div_u64(found_key.objectid - start, | |
0b246afa | 229 | fs_info->sectorsize); |
a5ed9182 | 230 | last = div_u64(found_key.objectid + found_key.offset - start, |
0b246afa | 231 | fs_info->sectorsize); |
2fe1d551 | 232 | le_bitmap_set(bitmap, first, last - first); |
a5ed9182 OS |
233 | |
234 | extent_count++; | |
235 | nr++; | |
236 | path->slots[0]--; | |
237 | } else { | |
238 | ASSERT(0); | |
239 | } | |
240 | } | |
241 | ||
242 | ret = btrfs_del_items(trans, root, path, path->slots[0], nr); | |
243 | if (ret) | |
244 | goto out; | |
245 | btrfs_release_path(path); | |
246 | } | |
247 | ||
248 | info = search_free_space_info(trans, fs_info, block_group, path, 1); | |
249 | if (IS_ERR(info)) { | |
250 | ret = PTR_ERR(info); | |
251 | goto out; | |
252 | } | |
253 | leaf = path->nodes[0]; | |
254 | flags = btrfs_free_space_flags(leaf, info); | |
255 | flags |= BTRFS_FREE_SPACE_USING_BITMAPS; | |
256 | btrfs_set_free_space_flags(leaf, info, flags); | |
257 | expected_extent_count = btrfs_free_space_extent_count(leaf, info); | |
258 | btrfs_mark_buffer_dirty(leaf); | |
259 | btrfs_release_path(path); | |
260 | ||
261 | if (extent_count != expected_extent_count) { | |
5d163e0e JM |
262 | btrfs_err(fs_info, |
263 | "incorrect extent count for %llu; counted %u, expected %u", | |
a5ed9182 OS |
264 | block_group->key.objectid, extent_count, |
265 | expected_extent_count); | |
266 | ASSERT(0); | |
267 | ret = -EIO; | |
268 | goto out; | |
269 | } | |
270 | ||
2fe1d551 | 271 | bitmap_cursor = bitmap; |
0b246afa | 272 | bitmap_range = fs_info->sectorsize * BTRFS_FREE_SPACE_BITMAP_BITS; |
a5ed9182 OS |
273 | i = start; |
274 | while (i < end) { | |
275 | unsigned long ptr; | |
276 | u64 extent_size; | |
277 | u32 data_size; | |
278 | ||
279 | extent_size = min(end - i, bitmap_range); | |
280 | data_size = free_space_bitmap_size(extent_size, | |
0b246afa | 281 | fs_info->sectorsize); |
a5ed9182 OS |
282 | |
283 | key.objectid = i; | |
284 | key.type = BTRFS_FREE_SPACE_BITMAP_KEY; | |
285 | key.offset = extent_size; | |
286 | ||
287 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
288 | data_size); | |
289 | if (ret) | |
290 | goto out; | |
291 | ||
292 | leaf = path->nodes[0]; | |
293 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
294 | write_extent_buffer(leaf, bitmap_cursor, ptr, | |
295 | data_size); | |
296 | btrfs_mark_buffer_dirty(leaf); | |
297 | btrfs_release_path(path); | |
298 | ||
299 | i += extent_size; | |
300 | bitmap_cursor += data_size; | |
301 | } | |
302 | ||
303 | ret = 0; | |
304 | out: | |
79b134a2 | 305 | kvfree(bitmap); |
a5ed9182 | 306 | if (ret) |
66642832 | 307 | btrfs_abort_transaction(trans, ret); |
a5ed9182 OS |
308 | return ret; |
309 | } | |
310 | ||
311 | int convert_free_space_to_extents(struct btrfs_trans_handle *trans, | |
312 | struct btrfs_fs_info *fs_info, | |
313 | struct btrfs_block_group_cache *block_group, | |
314 | struct btrfs_path *path) | |
315 | { | |
316 | struct btrfs_root *root = fs_info->free_space_root; | |
317 | struct btrfs_free_space_info *info; | |
318 | struct btrfs_key key, found_key; | |
319 | struct extent_buffer *leaf; | |
2fe1d551 | 320 | u8 *bitmap; |
a5ed9182 OS |
321 | u64 start, end; |
322 | /* Initialize to silence GCC. */ | |
323 | u64 extent_start = 0; | |
324 | u64 offset; | |
325 | u32 bitmap_size, flags, expected_extent_count; | |
326 | int prev_bit = 0, bit, bitnr; | |
327 | u32 extent_count = 0; | |
328 | int done = 0, nr; | |
329 | int ret; | |
330 | ||
331 | bitmap_size = free_space_bitmap_size(block_group->key.offset, | |
0b246afa | 332 | fs_info->sectorsize); |
a5ed9182 OS |
333 | bitmap = alloc_bitmap(bitmap_size); |
334 | if (!bitmap) { | |
335 | ret = -ENOMEM; | |
336 | goto out; | |
337 | } | |
338 | ||
339 | start = block_group->key.objectid; | |
340 | end = block_group->key.objectid + block_group->key.offset; | |
341 | ||
342 | key.objectid = end - 1; | |
343 | key.type = (u8)-1; | |
344 | key.offset = (u64)-1; | |
345 | ||
346 | while (!done) { | |
347 | ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1); | |
348 | if (ret) | |
349 | goto out; | |
350 | ||
351 | leaf = path->nodes[0]; | |
352 | nr = 0; | |
353 | path->slots[0]++; | |
354 | while (path->slots[0] > 0) { | |
355 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1); | |
356 | ||
357 | if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) { | |
358 | ASSERT(found_key.objectid == block_group->key.objectid); | |
359 | ASSERT(found_key.offset == block_group->key.offset); | |
360 | done = 1; | |
361 | break; | |
362 | } else if (found_key.type == BTRFS_FREE_SPACE_BITMAP_KEY) { | |
363 | unsigned long ptr; | |
2fe1d551 | 364 | u8 *bitmap_cursor; |
a5ed9182 OS |
365 | u32 bitmap_pos, data_size; |
366 | ||
367 | ASSERT(found_key.objectid >= start); | |
368 | ASSERT(found_key.objectid < end); | |
369 | ASSERT(found_key.objectid + found_key.offset <= end); | |
370 | ||
371 | bitmap_pos = div_u64(found_key.objectid - start, | |
0b246afa | 372 | fs_info->sectorsize * |
a5ed9182 | 373 | BITS_PER_BYTE); |
2fe1d551 | 374 | bitmap_cursor = bitmap + bitmap_pos; |
a5ed9182 | 375 | data_size = free_space_bitmap_size(found_key.offset, |
0b246afa | 376 | fs_info->sectorsize); |
a5ed9182 OS |
377 | |
378 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0] - 1); | |
379 | read_extent_buffer(leaf, bitmap_cursor, ptr, | |
380 | data_size); | |
381 | ||
382 | nr++; | |
383 | path->slots[0]--; | |
384 | } else { | |
385 | ASSERT(0); | |
386 | } | |
387 | } | |
388 | ||
389 | ret = btrfs_del_items(trans, root, path, path->slots[0], nr); | |
390 | if (ret) | |
391 | goto out; | |
392 | btrfs_release_path(path); | |
393 | } | |
394 | ||
395 | info = search_free_space_info(trans, fs_info, block_group, path, 1); | |
396 | if (IS_ERR(info)) { | |
397 | ret = PTR_ERR(info); | |
398 | goto out; | |
399 | } | |
400 | leaf = path->nodes[0]; | |
401 | flags = btrfs_free_space_flags(leaf, info); | |
402 | flags &= ~BTRFS_FREE_SPACE_USING_BITMAPS; | |
403 | btrfs_set_free_space_flags(leaf, info, flags); | |
404 | expected_extent_count = btrfs_free_space_extent_count(leaf, info); | |
405 | btrfs_mark_buffer_dirty(leaf); | |
406 | btrfs_release_path(path); | |
407 | ||
408 | offset = start; | |
409 | bitnr = 0; | |
410 | while (offset < end) { | |
2fe1d551 | 411 | bit = !!le_test_bit(bitnr, bitmap); |
a5ed9182 OS |
412 | if (prev_bit == 0 && bit == 1) { |
413 | extent_start = offset; | |
414 | } else if (prev_bit == 1 && bit == 0) { | |
415 | key.objectid = extent_start; | |
416 | key.type = BTRFS_FREE_SPACE_EXTENT_KEY; | |
417 | key.offset = offset - extent_start; | |
418 | ||
419 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
420 | if (ret) | |
421 | goto out; | |
422 | btrfs_release_path(path); | |
423 | ||
424 | extent_count++; | |
425 | } | |
426 | prev_bit = bit; | |
0b246afa | 427 | offset += fs_info->sectorsize; |
a5ed9182 OS |
428 | bitnr++; |
429 | } | |
430 | if (prev_bit == 1) { | |
431 | key.objectid = extent_start; | |
432 | key.type = BTRFS_FREE_SPACE_EXTENT_KEY; | |
433 | key.offset = end - extent_start; | |
434 | ||
435 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
436 | if (ret) | |
437 | goto out; | |
438 | btrfs_release_path(path); | |
439 | ||
440 | extent_count++; | |
441 | } | |
442 | ||
443 | if (extent_count != expected_extent_count) { | |
5d163e0e JM |
444 | btrfs_err(fs_info, |
445 | "incorrect extent count for %llu; counted %u, expected %u", | |
a5ed9182 OS |
446 | block_group->key.objectid, extent_count, |
447 | expected_extent_count); | |
448 | ASSERT(0); | |
449 | ret = -EIO; | |
450 | goto out; | |
451 | } | |
452 | ||
453 | ret = 0; | |
454 | out: | |
79b134a2 | 455 | kvfree(bitmap); |
a5ed9182 | 456 | if (ret) |
66642832 | 457 | btrfs_abort_transaction(trans, ret); |
a5ed9182 OS |
458 | return ret; |
459 | } | |
460 | ||
461 | static int update_free_space_extent_count(struct btrfs_trans_handle *trans, | |
462 | struct btrfs_fs_info *fs_info, | |
463 | struct btrfs_block_group_cache *block_group, | |
464 | struct btrfs_path *path, | |
465 | int new_extents) | |
466 | { | |
467 | struct btrfs_free_space_info *info; | |
468 | u32 flags; | |
469 | u32 extent_count; | |
470 | int ret = 0; | |
471 | ||
472 | if (new_extents == 0) | |
473 | return 0; | |
474 | ||
475 | info = search_free_space_info(trans, fs_info, block_group, path, 1); | |
476 | if (IS_ERR(info)) { | |
477 | ret = PTR_ERR(info); | |
478 | goto out; | |
479 | } | |
480 | flags = btrfs_free_space_flags(path->nodes[0], info); | |
481 | extent_count = btrfs_free_space_extent_count(path->nodes[0], info); | |
482 | ||
483 | extent_count += new_extents; | |
484 | btrfs_set_free_space_extent_count(path->nodes[0], info, extent_count); | |
485 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
486 | btrfs_release_path(path); | |
487 | ||
488 | if (!(flags & BTRFS_FREE_SPACE_USING_BITMAPS) && | |
489 | extent_count > block_group->bitmap_high_thresh) { | |
490 | ret = convert_free_space_to_bitmaps(trans, fs_info, block_group, | |
491 | path); | |
492 | } else if ((flags & BTRFS_FREE_SPACE_USING_BITMAPS) && | |
493 | extent_count < block_group->bitmap_low_thresh) { | |
494 | ret = convert_free_space_to_extents(trans, fs_info, block_group, | |
495 | path); | |
496 | } | |
497 | ||
498 | out: | |
499 | return ret; | |
500 | } | |
501 | ||
502 | int free_space_test_bit(struct btrfs_block_group_cache *block_group, | |
503 | struct btrfs_path *path, u64 offset) | |
504 | { | |
505 | struct extent_buffer *leaf; | |
506 | struct btrfs_key key; | |
507 | u64 found_start, found_end; | |
508 | unsigned long ptr, i; | |
509 | ||
510 | leaf = path->nodes[0]; | |
511 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
512 | ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY); | |
513 | ||
514 | found_start = key.objectid; | |
515 | found_end = key.objectid + key.offset; | |
516 | ASSERT(offset >= found_start && offset < found_end); | |
517 | ||
518 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
da17066c JM |
519 | i = div_u64(offset - found_start, |
520 | block_group->fs_info->sectorsize); | |
a5ed9182 OS |
521 | return !!extent_buffer_test_bit(leaf, ptr, i); |
522 | } | |
523 | ||
524 | static void free_space_set_bits(struct btrfs_block_group_cache *block_group, | |
525 | struct btrfs_path *path, u64 *start, u64 *size, | |
526 | int bit) | |
527 | { | |
0b246afa | 528 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
a5ed9182 OS |
529 | struct extent_buffer *leaf; |
530 | struct btrfs_key key; | |
531 | u64 end = *start + *size; | |
532 | u64 found_start, found_end; | |
533 | unsigned long ptr, first, last; | |
534 | ||
535 | leaf = path->nodes[0]; | |
536 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
537 | ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY); | |
538 | ||
539 | found_start = key.objectid; | |
540 | found_end = key.objectid + key.offset; | |
541 | ASSERT(*start >= found_start && *start < found_end); | |
542 | ASSERT(end > found_start); | |
543 | ||
544 | if (end > found_end) | |
545 | end = found_end; | |
546 | ||
547 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
0b246afa JM |
548 | first = div_u64(*start - found_start, fs_info->sectorsize); |
549 | last = div_u64(end - found_start, fs_info->sectorsize); | |
a5ed9182 OS |
550 | if (bit) |
551 | extent_buffer_bitmap_set(leaf, ptr, first, last - first); | |
552 | else | |
553 | extent_buffer_bitmap_clear(leaf, ptr, first, last - first); | |
554 | btrfs_mark_buffer_dirty(leaf); | |
555 | ||
556 | *size -= end - *start; | |
557 | *start = end; | |
558 | } | |
559 | ||
560 | /* | |
561 | * We can't use btrfs_next_item() in modify_free_space_bitmap() because | |
562 | * btrfs_next_leaf() doesn't get the path for writing. We can forgo the fancy | |
563 | * tree walking in btrfs_next_leaf() anyways because we know exactly what we're | |
564 | * looking for. | |
565 | */ | |
566 | static int free_space_next_bitmap(struct btrfs_trans_handle *trans, | |
567 | struct btrfs_root *root, struct btrfs_path *p) | |
568 | { | |
569 | struct btrfs_key key; | |
570 | ||
571 | if (p->slots[0] + 1 < btrfs_header_nritems(p->nodes[0])) { | |
572 | p->slots[0]++; | |
573 | return 0; | |
574 | } | |
575 | ||
576 | btrfs_item_key_to_cpu(p->nodes[0], &key, p->slots[0]); | |
577 | btrfs_release_path(p); | |
578 | ||
579 | key.objectid += key.offset; | |
580 | key.type = (u8)-1; | |
581 | key.offset = (u64)-1; | |
582 | ||
583 | return btrfs_search_prev_slot(trans, root, &key, p, 0, 1); | |
584 | } | |
585 | ||
586 | /* | |
587 | * If remove is 1, then we are removing free space, thus clearing bits in the | |
588 | * bitmap. If remove is 0, then we are adding free space, thus setting bits in | |
589 | * the bitmap. | |
590 | */ | |
591 | static int modify_free_space_bitmap(struct btrfs_trans_handle *trans, | |
592 | struct btrfs_fs_info *fs_info, | |
593 | struct btrfs_block_group_cache *block_group, | |
594 | struct btrfs_path *path, | |
595 | u64 start, u64 size, int remove) | |
596 | { | |
597 | struct btrfs_root *root = fs_info->free_space_root; | |
598 | struct btrfs_key key; | |
599 | u64 end = start + size; | |
600 | u64 cur_start, cur_size; | |
601 | int prev_bit, next_bit; | |
602 | int new_extents; | |
603 | int ret; | |
604 | ||
605 | /* | |
606 | * Read the bit for the block immediately before the extent of space if | |
607 | * that block is within the block group. | |
608 | */ | |
609 | if (start > block_group->key.objectid) { | |
da17066c | 610 | u64 prev_block = start - block_group->fs_info->sectorsize; |
a5ed9182 OS |
611 | |
612 | key.objectid = prev_block; | |
613 | key.type = (u8)-1; | |
614 | key.offset = (u64)-1; | |
615 | ||
616 | ret = btrfs_search_prev_slot(trans, root, &key, path, 0, 1); | |
617 | if (ret) | |
618 | goto out; | |
619 | ||
620 | prev_bit = free_space_test_bit(block_group, path, prev_block); | |
621 | ||
622 | /* The previous block may have been in the previous bitmap. */ | |
623 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
624 | if (start >= key.objectid + key.offset) { | |
625 | ret = free_space_next_bitmap(trans, root, path); | |
626 | if (ret) | |
627 | goto out; | |
628 | } | |
629 | } else { | |
630 | key.objectid = start; | |
631 | key.type = (u8)-1; | |
632 | key.offset = (u64)-1; | |
633 | ||
634 | ret = btrfs_search_prev_slot(trans, root, &key, path, 0, 1); | |
635 | if (ret) | |
636 | goto out; | |
637 | ||
638 | prev_bit = -1; | |
639 | } | |
640 | ||
641 | /* | |
642 | * Iterate over all of the bitmaps overlapped by the extent of space, | |
643 | * clearing/setting bits as required. | |
644 | */ | |
645 | cur_start = start; | |
646 | cur_size = size; | |
647 | while (1) { | |
648 | free_space_set_bits(block_group, path, &cur_start, &cur_size, | |
649 | !remove); | |
650 | if (cur_size == 0) | |
651 | break; | |
652 | ret = free_space_next_bitmap(trans, root, path); | |
653 | if (ret) | |
654 | goto out; | |
655 | } | |
656 | ||
657 | /* | |
658 | * Read the bit for the block immediately after the extent of space if | |
659 | * that block is within the block group. | |
660 | */ | |
661 | if (end < block_group->key.objectid + block_group->key.offset) { | |
662 | /* The next block may be in the next bitmap. */ | |
663 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
664 | if (end >= key.objectid + key.offset) { | |
665 | ret = free_space_next_bitmap(trans, root, path); | |
666 | if (ret) | |
667 | goto out; | |
668 | } | |
669 | ||
670 | next_bit = free_space_test_bit(block_group, path, end); | |
671 | } else { | |
672 | next_bit = -1; | |
673 | } | |
674 | ||
675 | if (remove) { | |
676 | new_extents = -1; | |
677 | if (prev_bit == 1) { | |
678 | /* Leftover on the left. */ | |
679 | new_extents++; | |
680 | } | |
681 | if (next_bit == 1) { | |
682 | /* Leftover on the right. */ | |
683 | new_extents++; | |
684 | } | |
685 | } else { | |
686 | new_extents = 1; | |
687 | if (prev_bit == 1) { | |
688 | /* Merging with neighbor on the left. */ | |
689 | new_extents--; | |
690 | } | |
691 | if (next_bit == 1) { | |
692 | /* Merging with neighbor on the right. */ | |
693 | new_extents--; | |
694 | } | |
695 | } | |
696 | ||
697 | btrfs_release_path(path); | |
698 | ret = update_free_space_extent_count(trans, fs_info, block_group, path, | |
699 | new_extents); | |
700 | ||
701 | out: | |
702 | return ret; | |
703 | } | |
704 | ||
705 | static int remove_free_space_extent(struct btrfs_trans_handle *trans, | |
706 | struct btrfs_fs_info *fs_info, | |
707 | struct btrfs_block_group_cache *block_group, | |
708 | struct btrfs_path *path, | |
709 | u64 start, u64 size) | |
710 | { | |
711 | struct btrfs_root *root = fs_info->free_space_root; | |
712 | struct btrfs_key key; | |
713 | u64 found_start, found_end; | |
714 | u64 end = start + size; | |
715 | int new_extents = -1; | |
716 | int ret; | |
717 | ||
718 | key.objectid = start; | |
719 | key.type = (u8)-1; | |
720 | key.offset = (u64)-1; | |
721 | ||
722 | ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1); | |
723 | if (ret) | |
724 | goto out; | |
725 | ||
726 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
727 | ||
728 | ASSERT(key.type == BTRFS_FREE_SPACE_EXTENT_KEY); | |
729 | ||
730 | found_start = key.objectid; | |
731 | found_end = key.objectid + key.offset; | |
732 | ASSERT(start >= found_start && end <= found_end); | |
733 | ||
734 | /* | |
735 | * Okay, now that we've found the free space extent which contains the | |
736 | * free space that we are removing, there are four cases: | |
737 | * | |
738 | * 1. We're using the whole extent: delete the key we found and | |
739 | * decrement the free space extent count. | |
740 | * 2. We are using part of the extent starting at the beginning: delete | |
741 | * the key we found and insert a new key representing the leftover at | |
742 | * the end. There is no net change in the number of extents. | |
743 | * 3. We are using part of the extent ending at the end: delete the key | |
744 | * we found and insert a new key representing the leftover at the | |
745 | * beginning. There is no net change in the number of extents. | |
746 | * 4. We are using part of the extent in the middle: delete the key we | |
747 | * found and insert two new keys representing the leftovers on each | |
748 | * side. Where we used to have one extent, we now have two, so increment | |
749 | * the extent count. We may need to convert the block group to bitmaps | |
750 | * as a result. | |
751 | */ | |
752 | ||
753 | /* Delete the existing key (cases 1-4). */ | |
754 | ret = btrfs_del_item(trans, root, path); | |
755 | if (ret) | |
756 | goto out; | |
757 | ||
758 | /* Add a key for leftovers at the beginning (cases 3 and 4). */ | |
759 | if (start > found_start) { | |
760 | key.objectid = found_start; | |
761 | key.type = BTRFS_FREE_SPACE_EXTENT_KEY; | |
762 | key.offset = start - found_start; | |
763 | ||
764 | btrfs_release_path(path); | |
765 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
766 | if (ret) | |
767 | goto out; | |
768 | new_extents++; | |
769 | } | |
770 | ||
771 | /* Add a key for leftovers at the end (cases 2 and 4). */ | |
772 | if (end < found_end) { | |
773 | key.objectid = end; | |
774 | key.type = BTRFS_FREE_SPACE_EXTENT_KEY; | |
775 | key.offset = found_end - end; | |
776 | ||
777 | btrfs_release_path(path); | |
778 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
779 | if (ret) | |
780 | goto out; | |
781 | new_extents++; | |
782 | } | |
783 | ||
784 | btrfs_release_path(path); | |
785 | ret = update_free_space_extent_count(trans, fs_info, block_group, path, | |
786 | new_extents); | |
787 | ||
788 | out: | |
789 | return ret; | |
790 | } | |
791 | ||
792 | int __remove_from_free_space_tree(struct btrfs_trans_handle *trans, | |
793 | struct btrfs_fs_info *fs_info, | |
794 | struct btrfs_block_group_cache *block_group, | |
795 | struct btrfs_path *path, u64 start, u64 size) | |
796 | { | |
797 | struct btrfs_free_space_info *info; | |
798 | u32 flags; | |
799 | int ret; | |
800 | ||
801 | if (block_group->needs_free_space) { | |
802 | ret = __add_block_group_free_space(trans, fs_info, block_group, | |
803 | path); | |
804 | if (ret) | |
805 | return ret; | |
806 | } | |
807 | ||
808 | info = search_free_space_info(NULL, fs_info, block_group, path, 0); | |
809 | if (IS_ERR(info)) | |
810 | return PTR_ERR(info); | |
811 | flags = btrfs_free_space_flags(path->nodes[0], info); | |
812 | btrfs_release_path(path); | |
813 | ||
814 | if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) { | |
815 | return modify_free_space_bitmap(trans, fs_info, block_group, | |
816 | path, start, size, 1); | |
817 | } else { | |
818 | return remove_free_space_extent(trans, fs_info, block_group, | |
819 | path, start, size); | |
820 | } | |
821 | } | |
822 | ||
823 | int remove_from_free_space_tree(struct btrfs_trans_handle *trans, | |
824 | struct btrfs_fs_info *fs_info, | |
825 | u64 start, u64 size) | |
826 | { | |
827 | struct btrfs_block_group_cache *block_group; | |
828 | struct btrfs_path *path; | |
829 | int ret; | |
830 | ||
831 | if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) | |
832 | return 0; | |
833 | ||
834 | path = btrfs_alloc_path(); | |
835 | if (!path) { | |
836 | ret = -ENOMEM; | |
837 | goto out; | |
838 | } | |
839 | ||
840 | block_group = btrfs_lookup_block_group(fs_info, start); | |
841 | if (!block_group) { | |
842 | ASSERT(0); | |
843 | ret = -ENOENT; | |
844 | goto out; | |
845 | } | |
846 | ||
847 | mutex_lock(&block_group->free_space_lock); | |
848 | ret = __remove_from_free_space_tree(trans, fs_info, block_group, path, | |
849 | start, size); | |
850 | mutex_unlock(&block_group->free_space_lock); | |
851 | ||
852 | btrfs_put_block_group(block_group); | |
853 | out: | |
854 | btrfs_free_path(path); | |
855 | if (ret) | |
66642832 | 856 | btrfs_abort_transaction(trans, ret); |
a5ed9182 OS |
857 | return ret; |
858 | } | |
859 | ||
860 | static int add_free_space_extent(struct btrfs_trans_handle *trans, | |
861 | struct btrfs_fs_info *fs_info, | |
862 | struct btrfs_block_group_cache *block_group, | |
863 | struct btrfs_path *path, | |
864 | u64 start, u64 size) | |
865 | { | |
866 | struct btrfs_root *root = fs_info->free_space_root; | |
867 | struct btrfs_key key, new_key; | |
868 | u64 found_start, found_end; | |
869 | u64 end = start + size; | |
870 | int new_extents = 1; | |
871 | int ret; | |
872 | ||
873 | /* | |
874 | * We are adding a new extent of free space, but we need to merge | |
875 | * extents. There are four cases here: | |
876 | * | |
877 | * 1. The new extent does not have any immediate neighbors to merge | |
878 | * with: add the new key and increment the free space extent count. We | |
879 | * may need to convert the block group to bitmaps as a result. | |
880 | * 2. The new extent has an immediate neighbor before it: remove the | |
881 | * previous key and insert a new key combining both of them. There is no | |
882 | * net change in the number of extents. | |
883 | * 3. The new extent has an immediate neighbor after it: remove the next | |
884 | * key and insert a new key combining both of them. There is no net | |
885 | * change in the number of extents. | |
886 | * 4. The new extent has immediate neighbors on both sides: remove both | |
887 | * of the keys and insert a new key combining all of them. Where we used | |
888 | * to have two extents, we now have one, so decrement the extent count. | |
889 | */ | |
890 | ||
891 | new_key.objectid = start; | |
892 | new_key.type = BTRFS_FREE_SPACE_EXTENT_KEY; | |
893 | new_key.offset = size; | |
894 | ||
895 | /* Search for a neighbor on the left. */ | |
896 | if (start == block_group->key.objectid) | |
897 | goto right; | |
898 | key.objectid = start - 1; | |
899 | key.type = (u8)-1; | |
900 | key.offset = (u64)-1; | |
901 | ||
902 | ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1); | |
903 | if (ret) | |
904 | goto out; | |
905 | ||
906 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
907 | ||
908 | if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) { | |
909 | ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY); | |
910 | btrfs_release_path(path); | |
911 | goto right; | |
912 | } | |
913 | ||
914 | found_start = key.objectid; | |
915 | found_end = key.objectid + key.offset; | |
916 | ASSERT(found_start >= block_group->key.objectid && | |
917 | found_end > block_group->key.objectid); | |
918 | ASSERT(found_start < start && found_end <= start); | |
919 | ||
920 | /* | |
921 | * Delete the neighbor on the left and absorb it into the new key (cases | |
922 | * 2 and 4). | |
923 | */ | |
924 | if (found_end == start) { | |
925 | ret = btrfs_del_item(trans, root, path); | |
926 | if (ret) | |
927 | goto out; | |
928 | new_key.objectid = found_start; | |
929 | new_key.offset += key.offset; | |
930 | new_extents--; | |
931 | } | |
932 | btrfs_release_path(path); | |
933 | ||
934 | right: | |
935 | /* Search for a neighbor on the right. */ | |
936 | if (end == block_group->key.objectid + block_group->key.offset) | |
937 | goto insert; | |
938 | key.objectid = end; | |
939 | key.type = (u8)-1; | |
940 | key.offset = (u64)-1; | |
941 | ||
942 | ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1); | |
943 | if (ret) | |
944 | goto out; | |
945 | ||
946 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
947 | ||
948 | if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) { | |
949 | ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY); | |
950 | btrfs_release_path(path); | |
951 | goto insert; | |
952 | } | |
953 | ||
954 | found_start = key.objectid; | |
955 | found_end = key.objectid + key.offset; | |
956 | ASSERT(found_start >= block_group->key.objectid && | |
957 | found_end > block_group->key.objectid); | |
958 | ASSERT((found_start < start && found_end <= start) || | |
959 | (found_start >= end && found_end > end)); | |
960 | ||
961 | /* | |
962 | * Delete the neighbor on the right and absorb it into the new key | |
963 | * (cases 3 and 4). | |
964 | */ | |
965 | if (found_start == end) { | |
966 | ret = btrfs_del_item(trans, root, path); | |
967 | if (ret) | |
968 | goto out; | |
969 | new_key.offset += key.offset; | |
970 | new_extents--; | |
971 | } | |
972 | btrfs_release_path(path); | |
973 | ||
974 | insert: | |
975 | /* Insert the new key (cases 1-4). */ | |
976 | ret = btrfs_insert_empty_item(trans, root, path, &new_key, 0); | |
977 | if (ret) | |
978 | goto out; | |
979 | ||
980 | btrfs_release_path(path); | |
981 | ret = update_free_space_extent_count(trans, fs_info, block_group, path, | |
982 | new_extents); | |
983 | ||
984 | out: | |
985 | return ret; | |
986 | } | |
987 | ||
988 | int __add_to_free_space_tree(struct btrfs_trans_handle *trans, | |
989 | struct btrfs_fs_info *fs_info, | |
990 | struct btrfs_block_group_cache *block_group, | |
991 | struct btrfs_path *path, u64 start, u64 size) | |
992 | { | |
993 | struct btrfs_free_space_info *info; | |
994 | u32 flags; | |
995 | int ret; | |
996 | ||
997 | if (block_group->needs_free_space) { | |
998 | ret = __add_block_group_free_space(trans, fs_info, block_group, | |
999 | path); | |
1000 | if (ret) | |
1001 | return ret; | |
1002 | } | |
1003 | ||
1004 | info = search_free_space_info(NULL, fs_info, block_group, path, 0); | |
1005 | if (IS_ERR(info)) | |
1006 | return PTR_ERR(info); | |
1007 | flags = btrfs_free_space_flags(path->nodes[0], info); | |
1008 | btrfs_release_path(path); | |
1009 | ||
1010 | if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) { | |
1011 | return modify_free_space_bitmap(trans, fs_info, block_group, | |
1012 | path, start, size, 0); | |
1013 | } else { | |
1014 | return add_free_space_extent(trans, fs_info, block_group, path, | |
1015 | start, size); | |
1016 | } | |
1017 | } | |
1018 | ||
1019 | int add_to_free_space_tree(struct btrfs_trans_handle *trans, | |
1020 | struct btrfs_fs_info *fs_info, | |
1021 | u64 start, u64 size) | |
1022 | { | |
1023 | struct btrfs_block_group_cache *block_group; | |
1024 | struct btrfs_path *path; | |
1025 | int ret; | |
1026 | ||
1027 | if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) | |
1028 | return 0; | |
1029 | ||
1030 | path = btrfs_alloc_path(); | |
1031 | if (!path) { | |
1032 | ret = -ENOMEM; | |
1033 | goto out; | |
1034 | } | |
1035 | ||
1036 | block_group = btrfs_lookup_block_group(fs_info, start); | |
1037 | if (!block_group) { | |
1038 | ASSERT(0); | |
1039 | ret = -ENOENT; | |
1040 | goto out; | |
1041 | } | |
1042 | ||
1043 | mutex_lock(&block_group->free_space_lock); | |
1044 | ret = __add_to_free_space_tree(trans, fs_info, block_group, path, start, | |
1045 | size); | |
1046 | mutex_unlock(&block_group->free_space_lock); | |
1047 | ||
1048 | btrfs_put_block_group(block_group); | |
1049 | out: | |
1050 | btrfs_free_path(path); | |
1051 | if (ret) | |
66642832 | 1052 | btrfs_abort_transaction(trans, ret); |
a5ed9182 OS |
1053 | return ret; |
1054 | } | |
1055 | ||
1056 | /* | |
1057 | * Populate the free space tree by walking the extent tree. Operations on the | |
1058 | * extent tree that happen as a result of writes to the free space tree will go | |
1059 | * through the normal add/remove hooks. | |
1060 | */ | |
1061 | static int populate_free_space_tree(struct btrfs_trans_handle *trans, | |
1062 | struct btrfs_fs_info *fs_info, | |
1063 | struct btrfs_block_group_cache *block_group) | |
1064 | { | |
1065 | struct btrfs_root *extent_root = fs_info->extent_root; | |
1066 | struct btrfs_path *path, *path2; | |
1067 | struct btrfs_key key; | |
1068 | u64 start, end; | |
1069 | int ret; | |
1070 | ||
1071 | path = btrfs_alloc_path(); | |
1072 | if (!path) | |
1073 | return -ENOMEM; | |
1074 | path->reada = 1; | |
1075 | ||
1076 | path2 = btrfs_alloc_path(); | |
1077 | if (!path2) { | |
1078 | btrfs_free_path(path); | |
1079 | return -ENOMEM; | |
1080 | } | |
1081 | ||
1082 | ret = add_new_free_space_info(trans, fs_info, block_group, path2); | |
1083 | if (ret) | |
1084 | goto out; | |
1085 | ||
511711af CM |
1086 | mutex_lock(&block_group->free_space_lock); |
1087 | ||
a5ed9182 OS |
1088 | /* |
1089 | * Iterate through all of the extent and metadata items in this block | |
1090 | * group, adding the free space between them and the free space at the | |
1091 | * end. Note that EXTENT_ITEM and METADATA_ITEM are less than | |
1092 | * BLOCK_GROUP_ITEM, so an extent may precede the block group that it's | |
1093 | * contained in. | |
1094 | */ | |
1095 | key.objectid = block_group->key.objectid; | |
1096 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
1097 | key.offset = 0; | |
1098 | ||
1099 | ret = btrfs_search_slot_for_read(extent_root, &key, path, 1, 0); | |
1100 | if (ret < 0) | |
511711af | 1101 | goto out_locked; |
a5ed9182 OS |
1102 | ASSERT(ret == 0); |
1103 | ||
1104 | start = block_group->key.objectid; | |
1105 | end = block_group->key.objectid + block_group->key.offset; | |
1106 | while (1) { | |
1107 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
1108 | ||
1109 | if (key.type == BTRFS_EXTENT_ITEM_KEY || | |
1110 | key.type == BTRFS_METADATA_ITEM_KEY) { | |
1111 | if (key.objectid >= end) | |
1112 | break; | |
1113 | ||
1114 | if (start < key.objectid) { | |
1115 | ret = __add_to_free_space_tree(trans, fs_info, | |
1116 | block_group, | |
1117 | path2, start, | |
1118 | key.objectid - | |
1119 | start); | |
1120 | if (ret) | |
511711af | 1121 | goto out_locked; |
a5ed9182 OS |
1122 | } |
1123 | start = key.objectid; | |
1124 | if (key.type == BTRFS_METADATA_ITEM_KEY) | |
da17066c | 1125 | start += fs_info->nodesize; |
a5ed9182 OS |
1126 | else |
1127 | start += key.offset; | |
1128 | } else if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { | |
1129 | if (key.objectid != block_group->key.objectid) | |
1130 | break; | |
1131 | } | |
1132 | ||
1133 | ret = btrfs_next_item(extent_root, path); | |
1134 | if (ret < 0) | |
511711af | 1135 | goto out_locked; |
a5ed9182 OS |
1136 | if (ret) |
1137 | break; | |
1138 | } | |
1139 | if (start < end) { | |
1140 | ret = __add_to_free_space_tree(trans, fs_info, block_group, | |
1141 | path2, start, end - start); | |
1142 | if (ret) | |
511711af | 1143 | goto out_locked; |
a5ed9182 OS |
1144 | } |
1145 | ||
1146 | ret = 0; | |
511711af CM |
1147 | out_locked: |
1148 | mutex_unlock(&block_group->free_space_lock); | |
a5ed9182 OS |
1149 | out: |
1150 | btrfs_free_path(path2); | |
1151 | btrfs_free_path(path); | |
1152 | return ret; | |
1153 | } | |
1154 | ||
1155 | int btrfs_create_free_space_tree(struct btrfs_fs_info *fs_info) | |
1156 | { | |
1157 | struct btrfs_trans_handle *trans; | |
1158 | struct btrfs_root *tree_root = fs_info->tree_root; | |
1159 | struct btrfs_root *free_space_root; | |
1160 | struct btrfs_block_group_cache *block_group; | |
1161 | struct rb_node *node; | |
1162 | int ret; | |
1163 | ||
1164 | trans = btrfs_start_transaction(tree_root, 0); | |
1165 | if (IS_ERR(trans)) | |
1166 | return PTR_ERR(trans); | |
1167 | ||
afcdd129 | 1168 | set_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags); |
a5ed9182 OS |
1169 | free_space_root = btrfs_create_tree(trans, fs_info, |
1170 | BTRFS_FREE_SPACE_TREE_OBJECTID); | |
1171 | if (IS_ERR(free_space_root)) { | |
1172 | ret = PTR_ERR(free_space_root); | |
1173 | goto abort; | |
1174 | } | |
1175 | fs_info->free_space_root = free_space_root; | |
1176 | ||
1177 | node = rb_first(&fs_info->block_group_cache_tree); | |
1178 | while (node) { | |
1179 | block_group = rb_entry(node, struct btrfs_block_group_cache, | |
1180 | cache_node); | |
1181 | ret = populate_free_space_tree(trans, fs_info, block_group); | |
1182 | if (ret) | |
1183 | goto abort; | |
1184 | node = rb_next(node); | |
1185 | } | |
1186 | ||
1187 | btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE); | |
6675df31 | 1188 | btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID); |
afcdd129 | 1189 | clear_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags); |
a5ed9182 | 1190 | |
0bef7109 | 1191 | return btrfs_commit_transaction(trans); |
a5ed9182 OS |
1192 | |
1193 | abort: | |
afcdd129 | 1194 | clear_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags); |
66642832 | 1195 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 1196 | btrfs_end_transaction(trans); |
a5ed9182 OS |
1197 | return ret; |
1198 | } | |
1199 | ||
1200 | static int clear_free_space_tree(struct btrfs_trans_handle *trans, | |
1201 | struct btrfs_root *root) | |
1202 | { | |
1203 | struct btrfs_path *path; | |
1204 | struct btrfs_key key; | |
1205 | int nr; | |
1206 | int ret; | |
1207 | ||
1208 | path = btrfs_alloc_path(); | |
1209 | if (!path) | |
1210 | return -ENOMEM; | |
1211 | ||
1212 | path->leave_spinning = 1; | |
1213 | ||
1214 | key.objectid = 0; | |
1215 | key.type = 0; | |
1216 | key.offset = 0; | |
1217 | ||
1218 | while (1) { | |
1219 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1220 | if (ret < 0) | |
1221 | goto out; | |
1222 | ||
1223 | nr = btrfs_header_nritems(path->nodes[0]); | |
1224 | if (!nr) | |
1225 | break; | |
1226 | ||
1227 | path->slots[0] = 0; | |
1228 | ret = btrfs_del_items(trans, root, path, 0, nr); | |
1229 | if (ret) | |
1230 | goto out; | |
1231 | ||
1232 | btrfs_release_path(path); | |
1233 | } | |
1234 | ||
1235 | ret = 0; | |
1236 | out: | |
1237 | btrfs_free_path(path); | |
1238 | return ret; | |
1239 | } | |
1240 | ||
1241 | int btrfs_clear_free_space_tree(struct btrfs_fs_info *fs_info) | |
1242 | { | |
1243 | struct btrfs_trans_handle *trans; | |
1244 | struct btrfs_root *tree_root = fs_info->tree_root; | |
1245 | struct btrfs_root *free_space_root = fs_info->free_space_root; | |
1246 | int ret; | |
1247 | ||
1248 | trans = btrfs_start_transaction(tree_root, 0); | |
1249 | if (IS_ERR(trans)) | |
1250 | return PTR_ERR(trans); | |
1251 | ||
1252 | btrfs_clear_fs_compat_ro(fs_info, FREE_SPACE_TREE); | |
6675df31 | 1253 | btrfs_clear_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID); |
a5ed9182 OS |
1254 | fs_info->free_space_root = NULL; |
1255 | ||
1256 | ret = clear_free_space_tree(trans, free_space_root); | |
1257 | if (ret) | |
1258 | goto abort; | |
1259 | ||
1cd5447e | 1260 | ret = btrfs_del_root(trans, fs_info, &free_space_root->root_key); |
a5ed9182 OS |
1261 | if (ret) |
1262 | goto abort; | |
1263 | ||
1264 | list_del(&free_space_root->dirty_list); | |
1265 | ||
1266 | btrfs_tree_lock(free_space_root->node); | |
7c302b49 | 1267 | clean_tree_block(fs_info, free_space_root->node); |
a5ed9182 OS |
1268 | btrfs_tree_unlock(free_space_root->node); |
1269 | btrfs_free_tree_block(trans, free_space_root, free_space_root->node, | |
1270 | 0, 1); | |
1271 | ||
1272 | free_extent_buffer(free_space_root->node); | |
1273 | free_extent_buffer(free_space_root->commit_root); | |
1274 | kfree(free_space_root); | |
1275 | ||
0bef7109 | 1276 | return btrfs_commit_transaction(trans); |
a5ed9182 OS |
1277 | |
1278 | abort: | |
66642832 | 1279 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 1280 | btrfs_end_transaction(trans); |
a5ed9182 OS |
1281 | return ret; |
1282 | } | |
1283 | ||
1284 | static int __add_block_group_free_space(struct btrfs_trans_handle *trans, | |
1285 | struct btrfs_fs_info *fs_info, | |
1286 | struct btrfs_block_group_cache *block_group, | |
1287 | struct btrfs_path *path) | |
1288 | { | |
a5ed9182 OS |
1289 | int ret; |
1290 | ||
a5ed9182 OS |
1291 | block_group->needs_free_space = 0; |
1292 | ||
1293 | ret = add_new_free_space_info(trans, fs_info, block_group, path); | |
1294 | if (ret) | |
1295 | return ret; | |
1296 | ||
1297 | return __add_to_free_space_tree(trans, fs_info, block_group, path, | |
1298 | block_group->key.objectid, | |
1299 | block_group->key.offset); | |
1300 | } | |
1301 | ||
1302 | int add_block_group_free_space(struct btrfs_trans_handle *trans, | |
1303 | struct btrfs_fs_info *fs_info, | |
1304 | struct btrfs_block_group_cache *block_group) | |
1305 | { | |
1306 | struct btrfs_path *path = NULL; | |
1307 | int ret = 0; | |
1308 | ||
1309 | if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) | |
1310 | return 0; | |
1311 | ||
1312 | mutex_lock(&block_group->free_space_lock); | |
1313 | if (!block_group->needs_free_space) | |
1314 | goto out; | |
1315 | ||
1316 | path = btrfs_alloc_path(); | |
1317 | if (!path) { | |
1318 | ret = -ENOMEM; | |
1319 | goto out; | |
1320 | } | |
1321 | ||
1322 | ret = __add_block_group_free_space(trans, fs_info, block_group, path); | |
1323 | ||
1324 | out: | |
1325 | btrfs_free_path(path); | |
1326 | mutex_unlock(&block_group->free_space_lock); | |
1327 | if (ret) | |
66642832 | 1328 | btrfs_abort_transaction(trans, ret); |
a5ed9182 OS |
1329 | return ret; |
1330 | } | |
1331 | ||
1332 | int remove_block_group_free_space(struct btrfs_trans_handle *trans, | |
1333 | struct btrfs_fs_info *fs_info, | |
1334 | struct btrfs_block_group_cache *block_group) | |
1335 | { | |
1336 | struct btrfs_root *root = fs_info->free_space_root; | |
1337 | struct btrfs_path *path; | |
1338 | struct btrfs_key key, found_key; | |
1339 | struct extent_buffer *leaf; | |
1340 | u64 start, end; | |
1341 | int done = 0, nr; | |
1342 | int ret; | |
1343 | ||
1344 | if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) | |
1345 | return 0; | |
1346 | ||
1347 | if (block_group->needs_free_space) { | |
1348 | /* We never added this block group to the free space tree. */ | |
1349 | return 0; | |
1350 | } | |
1351 | ||
1352 | path = btrfs_alloc_path(); | |
1353 | if (!path) { | |
1354 | ret = -ENOMEM; | |
1355 | goto out; | |
1356 | } | |
1357 | ||
1358 | start = block_group->key.objectid; | |
1359 | end = block_group->key.objectid + block_group->key.offset; | |
1360 | ||
1361 | key.objectid = end - 1; | |
1362 | key.type = (u8)-1; | |
1363 | key.offset = (u64)-1; | |
1364 | ||
1365 | while (!done) { | |
1366 | ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1); | |
1367 | if (ret) | |
1368 | goto out; | |
1369 | ||
1370 | leaf = path->nodes[0]; | |
1371 | nr = 0; | |
1372 | path->slots[0]++; | |
1373 | while (path->slots[0] > 0) { | |
1374 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1); | |
1375 | ||
1376 | if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) { | |
1377 | ASSERT(found_key.objectid == block_group->key.objectid); | |
1378 | ASSERT(found_key.offset == block_group->key.offset); | |
1379 | done = 1; | |
1380 | nr++; | |
1381 | path->slots[0]--; | |
1382 | break; | |
1383 | } else if (found_key.type == BTRFS_FREE_SPACE_EXTENT_KEY || | |
1384 | found_key.type == BTRFS_FREE_SPACE_BITMAP_KEY) { | |
1385 | ASSERT(found_key.objectid >= start); | |
1386 | ASSERT(found_key.objectid < end); | |
1387 | ASSERT(found_key.objectid + found_key.offset <= end); | |
1388 | nr++; | |
1389 | path->slots[0]--; | |
1390 | } else { | |
1391 | ASSERT(0); | |
1392 | } | |
1393 | } | |
1394 | ||
1395 | ret = btrfs_del_items(trans, root, path, path->slots[0], nr); | |
1396 | if (ret) | |
1397 | goto out; | |
1398 | btrfs_release_path(path); | |
1399 | } | |
1400 | ||
1401 | ret = 0; | |
1402 | out: | |
1403 | btrfs_free_path(path); | |
1404 | if (ret) | |
66642832 | 1405 | btrfs_abort_transaction(trans, ret); |
a5ed9182 OS |
1406 | return ret; |
1407 | } | |
1408 | ||
1409 | static int load_free_space_bitmaps(struct btrfs_caching_control *caching_ctl, | |
1410 | struct btrfs_path *path, | |
1411 | u32 expected_extent_count) | |
1412 | { | |
1413 | struct btrfs_block_group_cache *block_group; | |
1414 | struct btrfs_fs_info *fs_info; | |
1415 | struct btrfs_root *root; | |
1416 | struct btrfs_key key; | |
1417 | int prev_bit = 0, bit; | |
1418 | /* Initialize to silence GCC. */ | |
1419 | u64 extent_start = 0; | |
1420 | u64 end, offset; | |
1421 | u64 total_found = 0; | |
1422 | u32 extent_count = 0; | |
1423 | int ret; | |
1424 | ||
1425 | block_group = caching_ctl->block_group; | |
1426 | fs_info = block_group->fs_info; | |
1427 | root = fs_info->free_space_root; | |
1428 | ||
1429 | end = block_group->key.objectid + block_group->key.offset; | |
1430 | ||
1431 | while (1) { | |
1432 | ret = btrfs_next_item(root, path); | |
1433 | if (ret < 0) | |
1434 | goto out; | |
1435 | if (ret) | |
1436 | break; | |
1437 | ||
1438 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
1439 | ||
1440 | if (key.type == BTRFS_FREE_SPACE_INFO_KEY) | |
1441 | break; | |
1442 | ||
1443 | ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY); | |
1444 | ASSERT(key.objectid < end && key.objectid + key.offset <= end); | |
1445 | ||
1446 | caching_ctl->progress = key.objectid; | |
1447 | ||
1448 | offset = key.objectid; | |
1449 | while (offset < key.objectid + key.offset) { | |
1450 | bit = free_space_test_bit(block_group, path, offset); | |
1451 | if (prev_bit == 0 && bit == 1) { | |
1452 | extent_start = offset; | |
1453 | } else if (prev_bit == 1 && bit == 0) { | |
1454 | total_found += add_new_free_space(block_group, | |
1455 | fs_info, | |
1456 | extent_start, | |
1457 | offset); | |
1458 | if (total_found > CACHING_CTL_WAKE_UP) { | |
1459 | total_found = 0; | |
1460 | wake_up(&caching_ctl->wait); | |
1461 | } | |
1462 | extent_count++; | |
1463 | } | |
1464 | prev_bit = bit; | |
0b246afa | 1465 | offset += fs_info->sectorsize; |
a5ed9182 OS |
1466 | } |
1467 | } | |
1468 | if (prev_bit == 1) { | |
1469 | total_found += add_new_free_space(block_group, fs_info, | |
1470 | extent_start, end); | |
1471 | extent_count++; | |
1472 | } | |
1473 | ||
1474 | if (extent_count != expected_extent_count) { | |
5d163e0e JM |
1475 | btrfs_err(fs_info, |
1476 | "incorrect extent count for %llu; counted %u, expected %u", | |
a5ed9182 OS |
1477 | block_group->key.objectid, extent_count, |
1478 | expected_extent_count); | |
1479 | ASSERT(0); | |
1480 | ret = -EIO; | |
1481 | goto out; | |
1482 | } | |
1483 | ||
1484 | caching_ctl->progress = (u64)-1; | |
1485 | ||
1486 | ret = 0; | |
1487 | out: | |
1488 | return ret; | |
1489 | } | |
1490 | ||
1491 | static int load_free_space_extents(struct btrfs_caching_control *caching_ctl, | |
1492 | struct btrfs_path *path, | |
1493 | u32 expected_extent_count) | |
1494 | { | |
1495 | struct btrfs_block_group_cache *block_group; | |
1496 | struct btrfs_fs_info *fs_info; | |
1497 | struct btrfs_root *root; | |
1498 | struct btrfs_key key; | |
1499 | u64 end; | |
1500 | u64 total_found = 0; | |
1501 | u32 extent_count = 0; | |
1502 | int ret; | |
1503 | ||
1504 | block_group = caching_ctl->block_group; | |
1505 | fs_info = block_group->fs_info; | |
1506 | root = fs_info->free_space_root; | |
1507 | ||
1508 | end = block_group->key.objectid + block_group->key.offset; | |
1509 | ||
1510 | while (1) { | |
1511 | ret = btrfs_next_item(root, path); | |
1512 | if (ret < 0) | |
1513 | goto out; | |
1514 | if (ret) | |
1515 | break; | |
1516 | ||
1517 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
1518 | ||
1519 | if (key.type == BTRFS_FREE_SPACE_INFO_KEY) | |
1520 | break; | |
1521 | ||
1522 | ASSERT(key.type == BTRFS_FREE_SPACE_EXTENT_KEY); | |
1523 | ASSERT(key.objectid < end && key.objectid + key.offset <= end); | |
1524 | ||
1525 | caching_ctl->progress = key.objectid; | |
1526 | ||
1527 | total_found += add_new_free_space(block_group, fs_info, | |
1528 | key.objectid, | |
1529 | key.objectid + key.offset); | |
1530 | if (total_found > CACHING_CTL_WAKE_UP) { | |
1531 | total_found = 0; | |
1532 | wake_up(&caching_ctl->wait); | |
1533 | } | |
1534 | extent_count++; | |
1535 | } | |
1536 | ||
1537 | if (extent_count != expected_extent_count) { | |
5d163e0e JM |
1538 | btrfs_err(fs_info, |
1539 | "incorrect extent count for %llu; counted %u, expected %u", | |
a5ed9182 OS |
1540 | block_group->key.objectid, extent_count, |
1541 | expected_extent_count); | |
1542 | ASSERT(0); | |
1543 | ret = -EIO; | |
1544 | goto out; | |
1545 | } | |
1546 | ||
1547 | caching_ctl->progress = (u64)-1; | |
1548 | ||
1549 | ret = 0; | |
1550 | out: | |
1551 | return ret; | |
1552 | } | |
1553 | ||
1554 | int load_free_space_tree(struct btrfs_caching_control *caching_ctl) | |
1555 | { | |
1556 | struct btrfs_block_group_cache *block_group; | |
1557 | struct btrfs_fs_info *fs_info; | |
1558 | struct btrfs_free_space_info *info; | |
1559 | struct btrfs_path *path; | |
1560 | u32 extent_count, flags; | |
1561 | int ret; | |
1562 | ||
1563 | block_group = caching_ctl->block_group; | |
1564 | fs_info = block_group->fs_info; | |
1565 | ||
1566 | path = btrfs_alloc_path(); | |
1567 | if (!path) | |
1568 | return -ENOMEM; | |
1569 | ||
1570 | /* | |
1571 | * Just like caching_thread() doesn't want to deadlock on the extent | |
1572 | * tree, we don't want to deadlock on the free space tree. | |
1573 | */ | |
1574 | path->skip_locking = 1; | |
1575 | path->search_commit_root = 1; | |
1576 | path->reada = 1; | |
1577 | ||
1578 | info = search_free_space_info(NULL, fs_info, block_group, path, 0); | |
1579 | if (IS_ERR(info)) { | |
1580 | ret = PTR_ERR(info); | |
1581 | goto out; | |
1582 | } | |
1583 | extent_count = btrfs_free_space_extent_count(path->nodes[0], info); | |
1584 | flags = btrfs_free_space_flags(path->nodes[0], info); | |
1585 | ||
1586 | /* | |
1587 | * We left path pointing to the free space info item, so now | |
1588 | * load_free_space_foo can just iterate through the free space tree from | |
1589 | * there. | |
1590 | */ | |
1591 | if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) | |
1592 | ret = load_free_space_bitmaps(caching_ctl, path, extent_count); | |
1593 | else | |
1594 | ret = load_free_space_extents(caching_ctl, path, extent_count); | |
1595 | ||
1596 | out: | |
1597 | btrfs_free_path(path); | |
1598 | return ret; | |
1599 | } |