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> | |
20 | #include <linux/vmalloc.h> | |
21 | #include "ctree.h" | |
22 | #include "disk-io.h" | |
23 | #include "locking.h" | |
24 | #include "free-space-tree.h" | |
25 | #include "transaction.h" | |
14e46e04 | 26 | #include "sysfs.h" |
a5ed9182 OS |
27 | |
28 | static int __add_block_group_free_space(struct btrfs_trans_handle *trans, | |
29 | struct btrfs_fs_info *fs_info, | |
30 | struct btrfs_block_group_cache *block_group, | |
31 | struct btrfs_path *path); | |
32 | ||
33 | void set_free_space_tree_thresholds(struct btrfs_block_group_cache *cache) | |
34 | { | |
35 | u32 bitmap_range; | |
36 | size_t bitmap_size; | |
37 | u64 num_bitmaps, total_bitmap_size; | |
38 | ||
39 | /* | |
40 | * We convert to bitmaps when the disk space required for using extents | |
41 | * exceeds that required for using bitmaps. | |
42 | */ | |
43 | bitmap_range = cache->sectorsize * BTRFS_FREE_SPACE_BITMAP_BITS; | |
44 | num_bitmaps = div_u64(cache->key.offset + bitmap_range - 1, | |
45 | bitmap_range); | |
46 | bitmap_size = sizeof(struct btrfs_item) + BTRFS_FREE_SPACE_BITMAP_SIZE; | |
47 | total_bitmap_size = num_bitmaps * bitmap_size; | |
48 | cache->bitmap_high_thresh = div_u64(total_bitmap_size, | |
49 | sizeof(struct btrfs_item)); | |
50 | ||
51 | /* | |
52 | * We allow for a small buffer between the high threshold and low | |
53 | * threshold to avoid thrashing back and forth between the two formats. | |
54 | */ | |
55 | if (cache->bitmap_high_thresh > 100) | |
56 | cache->bitmap_low_thresh = cache->bitmap_high_thresh - 100; | |
57 | else | |
58 | cache->bitmap_low_thresh = 0; | |
59 | } | |
60 | ||
61 | static int add_new_free_space_info(struct btrfs_trans_handle *trans, | |
62 | struct btrfs_fs_info *fs_info, | |
63 | struct btrfs_block_group_cache *block_group, | |
64 | struct btrfs_path *path) | |
65 | { | |
66 | struct btrfs_root *root = fs_info->free_space_root; | |
67 | struct btrfs_free_space_info *info; | |
68 | struct btrfs_key key; | |
69 | struct extent_buffer *leaf; | |
70 | int ret; | |
71 | ||
72 | key.objectid = block_group->key.objectid; | |
73 | key.type = BTRFS_FREE_SPACE_INFO_KEY; | |
74 | key.offset = block_group->key.offset; | |
75 | ||
76 | ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*info)); | |
77 | if (ret) | |
78 | goto out; | |
79 | ||
80 | leaf = path->nodes[0]; | |
81 | info = btrfs_item_ptr(leaf, path->slots[0], | |
82 | struct btrfs_free_space_info); | |
83 | btrfs_set_free_space_extent_count(leaf, info, 0); | |
84 | btrfs_set_free_space_flags(leaf, info, 0); | |
85 | btrfs_mark_buffer_dirty(leaf); | |
86 | ||
87 | ret = 0; | |
88 | out: | |
89 | btrfs_release_path(path); | |
90 | return ret; | |
91 | } | |
92 | ||
93 | struct btrfs_free_space_info * | |
94 | search_free_space_info(struct btrfs_trans_handle *trans, | |
95 | struct btrfs_fs_info *fs_info, | |
96 | struct btrfs_block_group_cache *block_group, | |
97 | struct btrfs_path *path, int cow) | |
98 | { | |
99 | struct btrfs_root *root = fs_info->free_space_root; | |
100 | struct btrfs_key key; | |
101 | int ret; | |
102 | ||
103 | key.objectid = block_group->key.objectid; | |
104 | key.type = BTRFS_FREE_SPACE_INFO_KEY; | |
105 | key.offset = block_group->key.offset; | |
106 | ||
107 | ret = btrfs_search_slot(trans, root, &key, path, 0, cow); | |
108 | if (ret < 0) | |
109 | return ERR_PTR(ret); | |
110 | if (ret != 0) { | |
111 | btrfs_warn(fs_info, "missing free space info for %llu\n", | |
112 | block_group->key.objectid); | |
113 | ASSERT(0); | |
114 | return ERR_PTR(-ENOENT); | |
115 | } | |
116 | ||
117 | return btrfs_item_ptr(path->nodes[0], path->slots[0], | |
118 | struct btrfs_free_space_info); | |
119 | } | |
120 | ||
121 | /* | |
122 | * btrfs_search_slot() but we're looking for the greatest key less than the | |
123 | * passed key. | |
124 | */ | |
125 | static int btrfs_search_prev_slot(struct btrfs_trans_handle *trans, | |
126 | struct btrfs_root *root, | |
127 | struct btrfs_key *key, struct btrfs_path *p, | |
128 | int ins_len, int cow) | |
129 | { | |
130 | int ret; | |
131 | ||
132 | ret = btrfs_search_slot(trans, root, key, p, ins_len, cow); | |
133 | if (ret < 0) | |
134 | return ret; | |
135 | ||
136 | if (ret == 0) { | |
137 | ASSERT(0); | |
138 | return -EIO; | |
139 | } | |
140 | ||
141 | if (p->slots[0] == 0) { | |
142 | ASSERT(0); | |
143 | return -EIO; | |
144 | } | |
145 | p->slots[0]--; | |
146 | ||
147 | return 0; | |
148 | } | |
149 | ||
150 | static inline u32 free_space_bitmap_size(u64 size, u32 sectorsize) | |
151 | { | |
152 | return DIV_ROUND_UP((u32)div_u64(size, sectorsize), BITS_PER_BYTE); | |
153 | } | |
154 | ||
155 | static unsigned long *alloc_bitmap(u32 bitmap_size) | |
156 | { | |
79b134a2 DS |
157 | void *mem; |
158 | ||
159 | /* | |
160 | * The allocation size varies, observed numbers were < 4K up to 16K. | |
161 | * Using vmalloc unconditionally would be too heavy, we'll try | |
162 | * contiguous allocations first. | |
163 | */ | |
164 | if (bitmap_size <= PAGE_SIZE) | |
165 | return kzalloc(bitmap_size, GFP_NOFS); | |
166 | ||
e1c0ebad | 167 | mem = kzalloc(bitmap_size, GFP_NOFS | __GFP_NOWARN); |
79b134a2 DS |
168 | if (mem) |
169 | return mem; | |
170 | ||
a5ed9182 OS |
171 | return __vmalloc(bitmap_size, GFP_NOFS | __GFP_HIGHMEM | __GFP_ZERO, |
172 | PAGE_KERNEL); | |
173 | } | |
174 | ||
175 | int convert_free_space_to_bitmaps(struct btrfs_trans_handle *trans, | |
176 | struct btrfs_fs_info *fs_info, | |
177 | struct btrfs_block_group_cache *block_group, | |
178 | struct btrfs_path *path) | |
179 | { | |
180 | struct btrfs_root *root = fs_info->free_space_root; | |
181 | struct btrfs_free_space_info *info; | |
182 | struct btrfs_key key, found_key; | |
183 | struct extent_buffer *leaf; | |
184 | unsigned long *bitmap; | |
185 | char *bitmap_cursor; | |
186 | u64 start, end; | |
187 | u64 bitmap_range, i; | |
188 | u32 bitmap_size, flags, expected_extent_count; | |
189 | u32 extent_count = 0; | |
190 | int done = 0, nr; | |
191 | int ret; | |
192 | ||
193 | bitmap_size = free_space_bitmap_size(block_group->key.offset, | |
194 | block_group->sectorsize); | |
195 | bitmap = alloc_bitmap(bitmap_size); | |
196 | if (!bitmap) { | |
197 | ret = -ENOMEM; | |
198 | goto out; | |
199 | } | |
200 | ||
201 | start = block_group->key.objectid; | |
202 | end = block_group->key.objectid + block_group->key.offset; | |
203 | ||
204 | key.objectid = end - 1; | |
205 | key.type = (u8)-1; | |
206 | key.offset = (u64)-1; | |
207 | ||
208 | while (!done) { | |
209 | ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1); | |
210 | if (ret) | |
211 | goto out; | |
212 | ||
213 | leaf = path->nodes[0]; | |
214 | nr = 0; | |
215 | path->slots[0]++; | |
216 | while (path->slots[0] > 0) { | |
217 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1); | |
218 | ||
219 | if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) { | |
220 | ASSERT(found_key.objectid == block_group->key.objectid); | |
221 | ASSERT(found_key.offset == block_group->key.offset); | |
222 | done = 1; | |
223 | break; | |
224 | } else if (found_key.type == BTRFS_FREE_SPACE_EXTENT_KEY) { | |
225 | u64 first, last; | |
226 | ||
227 | ASSERT(found_key.objectid >= start); | |
228 | ASSERT(found_key.objectid < end); | |
229 | ASSERT(found_key.objectid + found_key.offset <= end); | |
230 | ||
231 | first = div_u64(found_key.objectid - start, | |
232 | block_group->sectorsize); | |
233 | last = div_u64(found_key.objectid + found_key.offset - start, | |
234 | block_group->sectorsize); | |
235 | bitmap_set(bitmap, first, last - first); | |
236 | ||
237 | extent_count++; | |
238 | nr++; | |
239 | path->slots[0]--; | |
240 | } else { | |
241 | ASSERT(0); | |
242 | } | |
243 | } | |
244 | ||
245 | ret = btrfs_del_items(trans, root, path, path->slots[0], nr); | |
246 | if (ret) | |
247 | goto out; | |
248 | btrfs_release_path(path); | |
249 | } | |
250 | ||
251 | info = search_free_space_info(trans, fs_info, block_group, path, 1); | |
252 | if (IS_ERR(info)) { | |
253 | ret = PTR_ERR(info); | |
254 | goto out; | |
255 | } | |
256 | leaf = path->nodes[0]; | |
257 | flags = btrfs_free_space_flags(leaf, info); | |
258 | flags |= BTRFS_FREE_SPACE_USING_BITMAPS; | |
259 | btrfs_set_free_space_flags(leaf, info, flags); | |
260 | expected_extent_count = btrfs_free_space_extent_count(leaf, info); | |
261 | btrfs_mark_buffer_dirty(leaf); | |
262 | btrfs_release_path(path); | |
263 | ||
264 | if (extent_count != expected_extent_count) { | |
265 | btrfs_err(fs_info, "incorrect extent count for %llu; counted %u, expected %u", | |
266 | block_group->key.objectid, extent_count, | |
267 | expected_extent_count); | |
268 | ASSERT(0); | |
269 | ret = -EIO; | |
270 | goto out; | |
271 | } | |
272 | ||
273 | bitmap_cursor = (char *)bitmap; | |
274 | bitmap_range = block_group->sectorsize * BTRFS_FREE_SPACE_BITMAP_BITS; | |
275 | i = start; | |
276 | while (i < end) { | |
277 | unsigned long ptr; | |
278 | u64 extent_size; | |
279 | u32 data_size; | |
280 | ||
281 | extent_size = min(end - i, bitmap_range); | |
282 | data_size = free_space_bitmap_size(extent_size, | |
283 | block_group->sectorsize); | |
284 | ||
285 | key.objectid = i; | |
286 | key.type = BTRFS_FREE_SPACE_BITMAP_KEY; | |
287 | key.offset = extent_size; | |
288 | ||
289 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
290 | data_size); | |
291 | if (ret) | |
292 | goto out; | |
293 | ||
294 | leaf = path->nodes[0]; | |
295 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
296 | write_extent_buffer(leaf, bitmap_cursor, ptr, | |
297 | data_size); | |
298 | btrfs_mark_buffer_dirty(leaf); | |
299 | btrfs_release_path(path); | |
300 | ||
301 | i += extent_size; | |
302 | bitmap_cursor += data_size; | |
303 | } | |
304 | ||
305 | ret = 0; | |
306 | out: | |
79b134a2 | 307 | kvfree(bitmap); |
a5ed9182 OS |
308 | if (ret) |
309 | btrfs_abort_transaction(trans, root, ret); | |
310 | return ret; | |
311 | } | |
312 | ||
313 | int convert_free_space_to_extents(struct btrfs_trans_handle *trans, | |
314 | struct btrfs_fs_info *fs_info, | |
315 | struct btrfs_block_group_cache *block_group, | |
316 | struct btrfs_path *path) | |
317 | { | |
318 | struct btrfs_root *root = fs_info->free_space_root; | |
319 | struct btrfs_free_space_info *info; | |
320 | struct btrfs_key key, found_key; | |
321 | struct extent_buffer *leaf; | |
322 | unsigned long *bitmap; | |
323 | u64 start, end; | |
324 | /* Initialize to silence GCC. */ | |
325 | u64 extent_start = 0; | |
326 | u64 offset; | |
327 | u32 bitmap_size, flags, expected_extent_count; | |
328 | int prev_bit = 0, bit, bitnr; | |
329 | u32 extent_count = 0; | |
330 | int done = 0, nr; | |
331 | int ret; | |
332 | ||
333 | bitmap_size = free_space_bitmap_size(block_group->key.offset, | |
334 | block_group->sectorsize); | |
335 | bitmap = alloc_bitmap(bitmap_size); | |
336 | if (!bitmap) { | |
337 | ret = -ENOMEM; | |
338 | goto out; | |
339 | } | |
340 | ||
341 | start = block_group->key.objectid; | |
342 | end = block_group->key.objectid + block_group->key.offset; | |
343 | ||
344 | key.objectid = end - 1; | |
345 | key.type = (u8)-1; | |
346 | key.offset = (u64)-1; | |
347 | ||
348 | while (!done) { | |
349 | ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1); | |
350 | if (ret) | |
351 | goto out; | |
352 | ||
353 | leaf = path->nodes[0]; | |
354 | nr = 0; | |
355 | path->slots[0]++; | |
356 | while (path->slots[0] > 0) { | |
357 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1); | |
358 | ||
359 | if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) { | |
360 | ASSERT(found_key.objectid == block_group->key.objectid); | |
361 | ASSERT(found_key.offset == block_group->key.offset); | |
362 | done = 1; | |
363 | break; | |
364 | } else if (found_key.type == BTRFS_FREE_SPACE_BITMAP_KEY) { | |
365 | unsigned long ptr; | |
366 | char *bitmap_cursor; | |
367 | u32 bitmap_pos, data_size; | |
368 | ||
369 | ASSERT(found_key.objectid >= start); | |
370 | ASSERT(found_key.objectid < end); | |
371 | ASSERT(found_key.objectid + found_key.offset <= end); | |
372 | ||
373 | bitmap_pos = div_u64(found_key.objectid - start, | |
374 | block_group->sectorsize * | |
375 | BITS_PER_BYTE); | |
376 | bitmap_cursor = ((char *)bitmap) + bitmap_pos; | |
377 | data_size = free_space_bitmap_size(found_key.offset, | |
378 | block_group->sectorsize); | |
379 | ||
380 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0] - 1); | |
381 | read_extent_buffer(leaf, bitmap_cursor, ptr, | |
382 | data_size); | |
383 | ||
384 | nr++; | |
385 | path->slots[0]--; | |
386 | } else { | |
387 | ASSERT(0); | |
388 | } | |
389 | } | |
390 | ||
391 | ret = btrfs_del_items(trans, root, path, path->slots[0], nr); | |
392 | if (ret) | |
393 | goto out; | |
394 | btrfs_release_path(path); | |
395 | } | |
396 | ||
397 | info = search_free_space_info(trans, fs_info, block_group, path, 1); | |
398 | if (IS_ERR(info)) { | |
399 | ret = PTR_ERR(info); | |
400 | goto out; | |
401 | } | |
402 | leaf = path->nodes[0]; | |
403 | flags = btrfs_free_space_flags(leaf, info); | |
404 | flags &= ~BTRFS_FREE_SPACE_USING_BITMAPS; | |
405 | btrfs_set_free_space_flags(leaf, info, flags); | |
406 | expected_extent_count = btrfs_free_space_extent_count(leaf, info); | |
407 | btrfs_mark_buffer_dirty(leaf); | |
408 | btrfs_release_path(path); | |
409 | ||
410 | offset = start; | |
411 | bitnr = 0; | |
412 | while (offset < end) { | |
413 | bit = !!test_bit(bitnr, bitmap); | |
414 | if (prev_bit == 0 && bit == 1) { | |
415 | extent_start = offset; | |
416 | } else if (prev_bit == 1 && bit == 0) { | |
417 | key.objectid = extent_start; | |
418 | key.type = BTRFS_FREE_SPACE_EXTENT_KEY; | |
419 | key.offset = offset - extent_start; | |
420 | ||
421 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
422 | if (ret) | |
423 | goto out; | |
424 | btrfs_release_path(path); | |
425 | ||
426 | extent_count++; | |
427 | } | |
428 | prev_bit = bit; | |
429 | offset += block_group->sectorsize; | |
430 | bitnr++; | |
431 | } | |
432 | if (prev_bit == 1) { | |
433 | key.objectid = extent_start; | |
434 | key.type = BTRFS_FREE_SPACE_EXTENT_KEY; | |
435 | key.offset = end - extent_start; | |
436 | ||
437 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
438 | if (ret) | |
439 | goto out; | |
440 | btrfs_release_path(path); | |
441 | ||
442 | extent_count++; | |
443 | } | |
444 | ||
445 | if (extent_count != expected_extent_count) { | |
446 | btrfs_err(fs_info, "incorrect extent count for %llu; counted %u, expected %u", | |
447 | block_group->key.objectid, extent_count, | |
448 | expected_extent_count); | |
449 | ASSERT(0); | |
450 | ret = -EIO; | |
451 | goto out; | |
452 | } | |
453 | ||
454 | ret = 0; | |
455 | out: | |
79b134a2 | 456 | kvfree(bitmap); |
a5ed9182 OS |
457 | if (ret) |
458 | btrfs_abort_transaction(trans, root, ret); | |
459 | return ret; | |
460 | } | |
461 | ||
462 | static int update_free_space_extent_count(struct btrfs_trans_handle *trans, | |
463 | struct btrfs_fs_info *fs_info, | |
464 | struct btrfs_block_group_cache *block_group, | |
465 | struct btrfs_path *path, | |
466 | int new_extents) | |
467 | { | |
468 | struct btrfs_free_space_info *info; | |
469 | u32 flags; | |
470 | u32 extent_count; | |
471 | int ret = 0; | |
472 | ||
473 | if (new_extents == 0) | |
474 | return 0; | |
475 | ||
476 | info = search_free_space_info(trans, fs_info, block_group, path, 1); | |
477 | if (IS_ERR(info)) { | |
478 | ret = PTR_ERR(info); | |
479 | goto out; | |
480 | } | |
481 | flags = btrfs_free_space_flags(path->nodes[0], info); | |
482 | extent_count = btrfs_free_space_extent_count(path->nodes[0], info); | |
483 | ||
484 | extent_count += new_extents; | |
485 | btrfs_set_free_space_extent_count(path->nodes[0], info, extent_count); | |
486 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
487 | btrfs_release_path(path); | |
488 | ||
489 | if (!(flags & BTRFS_FREE_SPACE_USING_BITMAPS) && | |
490 | extent_count > block_group->bitmap_high_thresh) { | |
491 | ret = convert_free_space_to_bitmaps(trans, fs_info, block_group, | |
492 | path); | |
493 | } else if ((flags & BTRFS_FREE_SPACE_USING_BITMAPS) && | |
494 | extent_count < block_group->bitmap_low_thresh) { | |
495 | ret = convert_free_space_to_extents(trans, fs_info, block_group, | |
496 | path); | |
497 | } | |
498 | ||
499 | out: | |
500 | return ret; | |
501 | } | |
502 | ||
503 | int free_space_test_bit(struct btrfs_block_group_cache *block_group, | |
504 | struct btrfs_path *path, u64 offset) | |
505 | { | |
506 | struct extent_buffer *leaf; | |
507 | struct btrfs_key key; | |
508 | u64 found_start, found_end; | |
509 | unsigned long ptr, i; | |
510 | ||
511 | leaf = path->nodes[0]; | |
512 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
513 | ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY); | |
514 | ||
515 | found_start = key.objectid; | |
516 | found_end = key.objectid + key.offset; | |
517 | ASSERT(offset >= found_start && offset < found_end); | |
518 | ||
519 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
520 | i = div_u64(offset - found_start, block_group->sectorsize); | |
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 | { | |
528 | struct extent_buffer *leaf; | |
529 | struct btrfs_key key; | |
530 | u64 end = *start + *size; | |
531 | u64 found_start, found_end; | |
532 | unsigned long ptr, first, last; | |
533 | ||
534 | leaf = path->nodes[0]; | |
535 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
536 | ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY); | |
537 | ||
538 | found_start = key.objectid; | |
539 | found_end = key.objectid + key.offset; | |
540 | ASSERT(*start >= found_start && *start < found_end); | |
541 | ASSERT(end > found_start); | |
542 | ||
543 | if (end > found_end) | |
544 | end = found_end; | |
545 | ||
546 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
547 | first = div_u64(*start - found_start, block_group->sectorsize); | |
548 | last = div_u64(end - found_start, block_group->sectorsize); | |
549 | if (bit) | |
550 | extent_buffer_bitmap_set(leaf, ptr, first, last - first); | |
551 | else | |
552 | extent_buffer_bitmap_clear(leaf, ptr, first, last - first); | |
553 | btrfs_mark_buffer_dirty(leaf); | |
554 | ||
555 | *size -= end - *start; | |
556 | *start = end; | |
557 | } | |
558 | ||
559 | /* | |
560 | * We can't use btrfs_next_item() in modify_free_space_bitmap() because | |
561 | * btrfs_next_leaf() doesn't get the path for writing. We can forgo the fancy | |
562 | * tree walking in btrfs_next_leaf() anyways because we know exactly what we're | |
563 | * looking for. | |
564 | */ | |
565 | static int free_space_next_bitmap(struct btrfs_trans_handle *trans, | |
566 | struct btrfs_root *root, struct btrfs_path *p) | |
567 | { | |
568 | struct btrfs_key key; | |
569 | ||
570 | if (p->slots[0] + 1 < btrfs_header_nritems(p->nodes[0])) { | |
571 | p->slots[0]++; | |
572 | return 0; | |
573 | } | |
574 | ||
575 | btrfs_item_key_to_cpu(p->nodes[0], &key, p->slots[0]); | |
576 | btrfs_release_path(p); | |
577 | ||
578 | key.objectid += key.offset; | |
579 | key.type = (u8)-1; | |
580 | key.offset = (u64)-1; | |
581 | ||
582 | return btrfs_search_prev_slot(trans, root, &key, p, 0, 1); | |
583 | } | |
584 | ||
585 | /* | |
586 | * If remove is 1, then we are removing free space, thus clearing bits in the | |
587 | * bitmap. If remove is 0, then we are adding free space, thus setting bits in | |
588 | * the bitmap. | |
589 | */ | |
590 | static int modify_free_space_bitmap(struct btrfs_trans_handle *trans, | |
591 | struct btrfs_fs_info *fs_info, | |
592 | struct btrfs_block_group_cache *block_group, | |
593 | struct btrfs_path *path, | |
594 | u64 start, u64 size, int remove) | |
595 | { | |
596 | struct btrfs_root *root = fs_info->free_space_root; | |
597 | struct btrfs_key key; | |
598 | u64 end = start + size; | |
599 | u64 cur_start, cur_size; | |
600 | int prev_bit, next_bit; | |
601 | int new_extents; | |
602 | int ret; | |
603 | ||
604 | /* | |
605 | * Read the bit for the block immediately before the extent of space if | |
606 | * that block is within the block group. | |
607 | */ | |
608 | if (start > block_group->key.objectid) { | |
609 | u64 prev_block = start - block_group->sectorsize; | |
610 | ||
611 | key.objectid = prev_block; | |
612 | key.type = (u8)-1; | |
613 | key.offset = (u64)-1; | |
614 | ||
615 | ret = btrfs_search_prev_slot(trans, root, &key, path, 0, 1); | |
616 | if (ret) | |
617 | goto out; | |
618 | ||
619 | prev_bit = free_space_test_bit(block_group, path, prev_block); | |
620 | ||
621 | /* The previous block may have been in the previous bitmap. */ | |
622 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
623 | if (start >= key.objectid + key.offset) { | |
624 | ret = free_space_next_bitmap(trans, root, path); | |
625 | if (ret) | |
626 | goto out; | |
627 | } | |
628 | } else { | |
629 | key.objectid = start; | |
630 | key.type = (u8)-1; | |
631 | key.offset = (u64)-1; | |
632 | ||
633 | ret = btrfs_search_prev_slot(trans, root, &key, path, 0, 1); | |
634 | if (ret) | |
635 | goto out; | |
636 | ||
637 | prev_bit = -1; | |
638 | } | |
639 | ||
640 | /* | |
641 | * Iterate over all of the bitmaps overlapped by the extent of space, | |
642 | * clearing/setting bits as required. | |
643 | */ | |
644 | cur_start = start; | |
645 | cur_size = size; | |
646 | while (1) { | |
647 | free_space_set_bits(block_group, path, &cur_start, &cur_size, | |
648 | !remove); | |
649 | if (cur_size == 0) | |
650 | break; | |
651 | ret = free_space_next_bitmap(trans, root, path); | |
652 | if (ret) | |
653 | goto out; | |
654 | } | |
655 | ||
656 | /* | |
657 | * Read the bit for the block immediately after the extent of space if | |
658 | * that block is within the block group. | |
659 | */ | |
660 | if (end < block_group->key.objectid + block_group->key.offset) { | |
661 | /* The next block may be in the next bitmap. */ | |
662 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
663 | if (end >= key.objectid + key.offset) { | |
664 | ret = free_space_next_bitmap(trans, root, path); | |
665 | if (ret) | |
666 | goto out; | |
667 | } | |
668 | ||
669 | next_bit = free_space_test_bit(block_group, path, end); | |
670 | } else { | |
671 | next_bit = -1; | |
672 | } | |
673 | ||
674 | if (remove) { | |
675 | new_extents = -1; | |
676 | if (prev_bit == 1) { | |
677 | /* Leftover on the left. */ | |
678 | new_extents++; | |
679 | } | |
680 | if (next_bit == 1) { | |
681 | /* Leftover on the right. */ | |
682 | new_extents++; | |
683 | } | |
684 | } else { | |
685 | new_extents = 1; | |
686 | if (prev_bit == 1) { | |
687 | /* Merging with neighbor on the left. */ | |
688 | new_extents--; | |
689 | } | |
690 | if (next_bit == 1) { | |
691 | /* Merging with neighbor on the right. */ | |
692 | new_extents--; | |
693 | } | |
694 | } | |
695 | ||
696 | btrfs_release_path(path); | |
697 | ret = update_free_space_extent_count(trans, fs_info, block_group, path, | |
698 | new_extents); | |
699 | ||
700 | out: | |
701 | return ret; | |
702 | } | |
703 | ||
704 | static int remove_free_space_extent(struct btrfs_trans_handle *trans, | |
705 | struct btrfs_fs_info *fs_info, | |
706 | struct btrfs_block_group_cache *block_group, | |
707 | struct btrfs_path *path, | |
708 | u64 start, u64 size) | |
709 | { | |
710 | struct btrfs_root *root = fs_info->free_space_root; | |
711 | struct btrfs_key key; | |
712 | u64 found_start, found_end; | |
713 | u64 end = start + size; | |
714 | int new_extents = -1; | |
715 | int ret; | |
716 | ||
717 | key.objectid = start; | |
718 | key.type = (u8)-1; | |
719 | key.offset = (u64)-1; | |
720 | ||
721 | ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1); | |
722 | if (ret) | |
723 | goto out; | |
724 | ||
725 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
726 | ||
727 | ASSERT(key.type == BTRFS_FREE_SPACE_EXTENT_KEY); | |
728 | ||
729 | found_start = key.objectid; | |
730 | found_end = key.objectid + key.offset; | |
731 | ASSERT(start >= found_start && end <= found_end); | |
732 | ||
733 | /* | |
734 | * Okay, now that we've found the free space extent which contains the | |
735 | * free space that we are removing, there are four cases: | |
736 | * | |
737 | * 1. We're using the whole extent: delete the key we found and | |
738 | * decrement the free space extent count. | |
739 | * 2. We are using part of the extent starting at the beginning: delete | |
740 | * the key we found and insert a new key representing the leftover at | |
741 | * the end. There is no net change in the number of extents. | |
742 | * 3. We are using part of the extent ending at the end: delete the key | |
743 | * we found and insert a new key representing the leftover at the | |
744 | * beginning. There is no net change in the number of extents. | |
745 | * 4. We are using part of the extent in the middle: delete the key we | |
746 | * found and insert two new keys representing the leftovers on each | |
747 | * side. Where we used to have one extent, we now have two, so increment | |
748 | * the extent count. We may need to convert the block group to bitmaps | |
749 | * as a result. | |
750 | */ | |
751 | ||
752 | /* Delete the existing key (cases 1-4). */ | |
753 | ret = btrfs_del_item(trans, root, path); | |
754 | if (ret) | |
755 | goto out; | |
756 | ||
757 | /* Add a key for leftovers at the beginning (cases 3 and 4). */ | |
758 | if (start > found_start) { | |
759 | key.objectid = found_start; | |
760 | key.type = BTRFS_FREE_SPACE_EXTENT_KEY; | |
761 | key.offset = start - found_start; | |
762 | ||
763 | btrfs_release_path(path); | |
764 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
765 | if (ret) | |
766 | goto out; | |
767 | new_extents++; | |
768 | } | |
769 | ||
770 | /* Add a key for leftovers at the end (cases 2 and 4). */ | |
771 | if (end < found_end) { | |
772 | key.objectid = end; | |
773 | key.type = BTRFS_FREE_SPACE_EXTENT_KEY; | |
774 | key.offset = found_end - end; | |
775 | ||
776 | btrfs_release_path(path); | |
777 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
778 | if (ret) | |
779 | goto out; | |
780 | new_extents++; | |
781 | } | |
782 | ||
783 | btrfs_release_path(path); | |
784 | ret = update_free_space_extent_count(trans, fs_info, block_group, path, | |
785 | new_extents); | |
786 | ||
787 | out: | |
788 | return ret; | |
789 | } | |
790 | ||
791 | int __remove_from_free_space_tree(struct btrfs_trans_handle *trans, | |
792 | struct btrfs_fs_info *fs_info, | |
793 | struct btrfs_block_group_cache *block_group, | |
794 | struct btrfs_path *path, u64 start, u64 size) | |
795 | { | |
796 | struct btrfs_free_space_info *info; | |
797 | u32 flags; | |
798 | int ret; | |
799 | ||
800 | if (block_group->needs_free_space) { | |
801 | ret = __add_block_group_free_space(trans, fs_info, block_group, | |
802 | path); | |
803 | if (ret) | |
804 | return ret; | |
805 | } | |
806 | ||
807 | info = search_free_space_info(NULL, fs_info, block_group, path, 0); | |
808 | if (IS_ERR(info)) | |
809 | return PTR_ERR(info); | |
810 | flags = btrfs_free_space_flags(path->nodes[0], info); | |
811 | btrfs_release_path(path); | |
812 | ||
813 | if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) { | |
814 | return modify_free_space_bitmap(trans, fs_info, block_group, | |
815 | path, start, size, 1); | |
816 | } else { | |
817 | return remove_free_space_extent(trans, fs_info, block_group, | |
818 | path, start, size); | |
819 | } | |
820 | } | |
821 | ||
822 | int remove_from_free_space_tree(struct btrfs_trans_handle *trans, | |
823 | struct btrfs_fs_info *fs_info, | |
824 | u64 start, u64 size) | |
825 | { | |
826 | struct btrfs_block_group_cache *block_group; | |
827 | struct btrfs_path *path; | |
828 | int ret; | |
829 | ||
830 | if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) | |
831 | return 0; | |
832 | ||
833 | path = btrfs_alloc_path(); | |
834 | if (!path) { | |
835 | ret = -ENOMEM; | |
836 | goto out; | |
837 | } | |
838 | ||
839 | block_group = btrfs_lookup_block_group(fs_info, start); | |
840 | if (!block_group) { | |
841 | ASSERT(0); | |
842 | ret = -ENOENT; | |
843 | goto out; | |
844 | } | |
845 | ||
846 | mutex_lock(&block_group->free_space_lock); | |
847 | ret = __remove_from_free_space_tree(trans, fs_info, block_group, path, | |
848 | start, size); | |
849 | mutex_unlock(&block_group->free_space_lock); | |
850 | ||
851 | btrfs_put_block_group(block_group); | |
852 | out: | |
853 | btrfs_free_path(path); | |
854 | if (ret) | |
855 | btrfs_abort_transaction(trans, fs_info->free_space_root, ret); | |
856 | return ret; | |
857 | } | |
858 | ||
859 | static int add_free_space_extent(struct btrfs_trans_handle *trans, | |
860 | struct btrfs_fs_info *fs_info, | |
861 | struct btrfs_block_group_cache *block_group, | |
862 | struct btrfs_path *path, | |
863 | u64 start, u64 size) | |
864 | { | |
865 | struct btrfs_root *root = fs_info->free_space_root; | |
866 | struct btrfs_key key, new_key; | |
867 | u64 found_start, found_end; | |
868 | u64 end = start + size; | |
869 | int new_extents = 1; | |
870 | int ret; | |
871 | ||
872 | /* | |
873 | * We are adding a new extent of free space, but we need to merge | |
874 | * extents. There are four cases here: | |
875 | * | |
876 | * 1. The new extent does not have any immediate neighbors to merge | |
877 | * with: add the new key and increment the free space extent count. We | |
878 | * may need to convert the block group to bitmaps as a result. | |
879 | * 2. The new extent has an immediate neighbor before it: remove the | |
880 | * previous key and insert a new key combining both of them. There is no | |
881 | * net change in the number of extents. | |
882 | * 3. The new extent has an immediate neighbor after it: remove the next | |
883 | * key and insert a new key combining both of them. There is no net | |
884 | * change in the number of extents. | |
885 | * 4. The new extent has immediate neighbors on both sides: remove both | |
886 | * of the keys and insert a new key combining all of them. Where we used | |
887 | * to have two extents, we now have one, so decrement the extent count. | |
888 | */ | |
889 | ||
890 | new_key.objectid = start; | |
891 | new_key.type = BTRFS_FREE_SPACE_EXTENT_KEY; | |
892 | new_key.offset = size; | |
893 | ||
894 | /* Search for a neighbor on the left. */ | |
895 | if (start == block_group->key.objectid) | |
896 | goto right; | |
897 | key.objectid = start - 1; | |
898 | key.type = (u8)-1; | |
899 | key.offset = (u64)-1; | |
900 | ||
901 | ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1); | |
902 | if (ret) | |
903 | goto out; | |
904 | ||
905 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
906 | ||
907 | if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) { | |
908 | ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY); | |
909 | btrfs_release_path(path); | |
910 | goto right; | |
911 | } | |
912 | ||
913 | found_start = key.objectid; | |
914 | found_end = key.objectid + key.offset; | |
915 | ASSERT(found_start >= block_group->key.objectid && | |
916 | found_end > block_group->key.objectid); | |
917 | ASSERT(found_start < start && found_end <= start); | |
918 | ||
919 | /* | |
920 | * Delete the neighbor on the left and absorb it into the new key (cases | |
921 | * 2 and 4). | |
922 | */ | |
923 | if (found_end == start) { | |
924 | ret = btrfs_del_item(trans, root, path); | |
925 | if (ret) | |
926 | goto out; | |
927 | new_key.objectid = found_start; | |
928 | new_key.offset += key.offset; | |
929 | new_extents--; | |
930 | } | |
931 | btrfs_release_path(path); | |
932 | ||
933 | right: | |
934 | /* Search for a neighbor on the right. */ | |
935 | if (end == block_group->key.objectid + block_group->key.offset) | |
936 | goto insert; | |
937 | key.objectid = end; | |
938 | key.type = (u8)-1; | |
939 | key.offset = (u64)-1; | |
940 | ||
941 | ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1); | |
942 | if (ret) | |
943 | goto out; | |
944 | ||
945 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
946 | ||
947 | if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) { | |
948 | ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY); | |
949 | btrfs_release_path(path); | |
950 | goto insert; | |
951 | } | |
952 | ||
953 | found_start = key.objectid; | |
954 | found_end = key.objectid + key.offset; | |
955 | ASSERT(found_start >= block_group->key.objectid && | |
956 | found_end > block_group->key.objectid); | |
957 | ASSERT((found_start < start && found_end <= start) || | |
958 | (found_start >= end && found_end > end)); | |
959 | ||
960 | /* | |
961 | * Delete the neighbor on the right and absorb it into the new key | |
962 | * (cases 3 and 4). | |
963 | */ | |
964 | if (found_start == end) { | |
965 | ret = btrfs_del_item(trans, root, path); | |
966 | if (ret) | |
967 | goto out; | |
968 | new_key.offset += key.offset; | |
969 | new_extents--; | |
970 | } | |
971 | btrfs_release_path(path); | |
972 | ||
973 | insert: | |
974 | /* Insert the new key (cases 1-4). */ | |
975 | ret = btrfs_insert_empty_item(trans, root, path, &new_key, 0); | |
976 | if (ret) | |
977 | goto out; | |
978 | ||
979 | btrfs_release_path(path); | |
980 | ret = update_free_space_extent_count(trans, fs_info, block_group, path, | |
981 | new_extents); | |
982 | ||
983 | out: | |
984 | return ret; | |
985 | } | |
986 | ||
987 | int __add_to_free_space_tree(struct btrfs_trans_handle *trans, | |
988 | struct btrfs_fs_info *fs_info, | |
989 | struct btrfs_block_group_cache *block_group, | |
990 | struct btrfs_path *path, u64 start, u64 size) | |
991 | { | |
992 | struct btrfs_free_space_info *info; | |
993 | u32 flags; | |
994 | int ret; | |
995 | ||
996 | if (block_group->needs_free_space) { | |
997 | ret = __add_block_group_free_space(trans, fs_info, block_group, | |
998 | path); | |
999 | if (ret) | |
1000 | return ret; | |
1001 | } | |
1002 | ||
1003 | info = search_free_space_info(NULL, fs_info, block_group, path, 0); | |
1004 | if (IS_ERR(info)) | |
1005 | return PTR_ERR(info); | |
1006 | flags = btrfs_free_space_flags(path->nodes[0], info); | |
1007 | btrfs_release_path(path); | |
1008 | ||
1009 | if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) { | |
1010 | return modify_free_space_bitmap(trans, fs_info, block_group, | |
1011 | path, start, size, 0); | |
1012 | } else { | |
1013 | return add_free_space_extent(trans, fs_info, block_group, path, | |
1014 | start, size); | |
1015 | } | |
1016 | } | |
1017 | ||
1018 | int add_to_free_space_tree(struct btrfs_trans_handle *trans, | |
1019 | struct btrfs_fs_info *fs_info, | |
1020 | u64 start, u64 size) | |
1021 | { | |
1022 | struct btrfs_block_group_cache *block_group; | |
1023 | struct btrfs_path *path; | |
1024 | int ret; | |
1025 | ||
1026 | if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) | |
1027 | return 0; | |
1028 | ||
1029 | path = btrfs_alloc_path(); | |
1030 | if (!path) { | |
1031 | ret = -ENOMEM; | |
1032 | goto out; | |
1033 | } | |
1034 | ||
1035 | block_group = btrfs_lookup_block_group(fs_info, start); | |
1036 | if (!block_group) { | |
1037 | ASSERT(0); | |
1038 | ret = -ENOENT; | |
1039 | goto out; | |
1040 | } | |
1041 | ||
1042 | mutex_lock(&block_group->free_space_lock); | |
1043 | ret = __add_to_free_space_tree(trans, fs_info, block_group, path, start, | |
1044 | size); | |
1045 | mutex_unlock(&block_group->free_space_lock); | |
1046 | ||
1047 | btrfs_put_block_group(block_group); | |
1048 | out: | |
1049 | btrfs_free_path(path); | |
1050 | if (ret) | |
1051 | btrfs_abort_transaction(trans, fs_info->free_space_root, ret); | |
1052 | return ret; | |
1053 | } | |
1054 | ||
1055 | /* | |
1056 | * Populate the free space tree by walking the extent tree. Operations on the | |
1057 | * extent tree that happen as a result of writes to the free space tree will go | |
1058 | * through the normal add/remove hooks. | |
1059 | */ | |
1060 | static int populate_free_space_tree(struct btrfs_trans_handle *trans, | |
1061 | struct btrfs_fs_info *fs_info, | |
1062 | struct btrfs_block_group_cache *block_group) | |
1063 | { | |
1064 | struct btrfs_root *extent_root = fs_info->extent_root; | |
1065 | struct btrfs_path *path, *path2; | |
1066 | struct btrfs_key key; | |
1067 | u64 start, end; | |
1068 | int ret; | |
1069 | ||
1070 | path = btrfs_alloc_path(); | |
1071 | if (!path) | |
1072 | return -ENOMEM; | |
1073 | path->reada = 1; | |
1074 | ||
1075 | path2 = btrfs_alloc_path(); | |
1076 | if (!path2) { | |
1077 | btrfs_free_path(path); | |
1078 | return -ENOMEM; | |
1079 | } | |
1080 | ||
1081 | ret = add_new_free_space_info(trans, fs_info, block_group, path2); | |
1082 | if (ret) | |
1083 | goto out; | |
1084 | ||
511711af CM |
1085 | mutex_lock(&block_group->free_space_lock); |
1086 | ||
a5ed9182 OS |
1087 | /* |
1088 | * Iterate through all of the extent and metadata items in this block | |
1089 | * group, adding the free space between them and the free space at the | |
1090 | * end. Note that EXTENT_ITEM and METADATA_ITEM are less than | |
1091 | * BLOCK_GROUP_ITEM, so an extent may precede the block group that it's | |
1092 | * contained in. | |
1093 | */ | |
1094 | key.objectid = block_group->key.objectid; | |
1095 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
1096 | key.offset = 0; | |
1097 | ||
1098 | ret = btrfs_search_slot_for_read(extent_root, &key, path, 1, 0); | |
1099 | if (ret < 0) | |
511711af | 1100 | goto out_locked; |
a5ed9182 OS |
1101 | ASSERT(ret == 0); |
1102 | ||
1103 | start = block_group->key.objectid; | |
1104 | end = block_group->key.objectid + block_group->key.offset; | |
1105 | while (1) { | |
1106 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
1107 | ||
1108 | if (key.type == BTRFS_EXTENT_ITEM_KEY || | |
1109 | key.type == BTRFS_METADATA_ITEM_KEY) { | |
1110 | if (key.objectid >= end) | |
1111 | break; | |
1112 | ||
1113 | if (start < key.objectid) { | |
1114 | ret = __add_to_free_space_tree(trans, fs_info, | |
1115 | block_group, | |
1116 | path2, start, | |
1117 | key.objectid - | |
1118 | start); | |
1119 | if (ret) | |
511711af | 1120 | goto out_locked; |
a5ed9182 OS |
1121 | } |
1122 | start = key.objectid; | |
1123 | if (key.type == BTRFS_METADATA_ITEM_KEY) | |
1124 | start += fs_info->tree_root->nodesize; | |
1125 | else | |
1126 | start += key.offset; | |
1127 | } else if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { | |
1128 | if (key.objectid != block_group->key.objectid) | |
1129 | break; | |
1130 | } | |
1131 | ||
1132 | ret = btrfs_next_item(extent_root, path); | |
1133 | if (ret < 0) | |
511711af | 1134 | goto out_locked; |
a5ed9182 OS |
1135 | if (ret) |
1136 | break; | |
1137 | } | |
1138 | if (start < end) { | |
1139 | ret = __add_to_free_space_tree(trans, fs_info, block_group, | |
1140 | path2, start, end - start); | |
1141 | if (ret) | |
511711af | 1142 | goto out_locked; |
a5ed9182 OS |
1143 | } |
1144 | ||
1145 | ret = 0; | |
511711af CM |
1146 | out_locked: |
1147 | mutex_unlock(&block_group->free_space_lock); | |
a5ed9182 OS |
1148 | out: |
1149 | btrfs_free_path(path2); | |
1150 | btrfs_free_path(path); | |
1151 | return ret; | |
1152 | } | |
1153 | ||
1154 | int btrfs_create_free_space_tree(struct btrfs_fs_info *fs_info) | |
1155 | { | |
1156 | struct btrfs_trans_handle *trans; | |
1157 | struct btrfs_root *tree_root = fs_info->tree_root; | |
1158 | struct btrfs_root *free_space_root; | |
1159 | struct btrfs_block_group_cache *block_group; | |
1160 | struct rb_node *node; | |
1161 | int ret; | |
1162 | ||
1163 | trans = btrfs_start_transaction(tree_root, 0); | |
1164 | if (IS_ERR(trans)) | |
1165 | return PTR_ERR(trans); | |
1166 | ||
511711af | 1167 | fs_info->creating_free_space_tree = 1; |
a5ed9182 OS |
1168 | free_space_root = btrfs_create_tree(trans, fs_info, |
1169 | BTRFS_FREE_SPACE_TREE_OBJECTID); | |
1170 | if (IS_ERR(free_space_root)) { | |
1171 | ret = PTR_ERR(free_space_root); | |
1172 | goto abort; | |
1173 | } | |
1174 | fs_info->free_space_root = free_space_root; | |
1175 | ||
1176 | node = rb_first(&fs_info->block_group_cache_tree); | |
1177 | while (node) { | |
1178 | block_group = rb_entry(node, struct btrfs_block_group_cache, | |
1179 | cache_node); | |
1180 | ret = populate_free_space_tree(trans, fs_info, block_group); | |
1181 | if (ret) | |
1182 | goto abort; | |
1183 | node = rb_next(node); | |
1184 | } | |
1185 | ||
1186 | btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE); | |
14e46e04 DS |
1187 | btrfs_sysfs_feature_update(fs_info, |
1188 | BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE, FEAT_COMPAT_RO); | |
1189 | ||
511711af | 1190 | fs_info->creating_free_space_tree = 0; |
a5ed9182 OS |
1191 | |
1192 | ret = btrfs_commit_transaction(trans, tree_root); | |
1193 | if (ret) | |
1194 | return ret; | |
1195 | ||
1196 | return 0; | |
1197 | ||
1198 | abort: | |
511711af | 1199 | fs_info->creating_free_space_tree = 0; |
a5ed9182 OS |
1200 | btrfs_abort_transaction(trans, tree_root, ret); |
1201 | btrfs_end_transaction(trans, tree_root); | |
1202 | return ret; | |
1203 | } | |
1204 | ||
1205 | static int clear_free_space_tree(struct btrfs_trans_handle *trans, | |
1206 | struct btrfs_root *root) | |
1207 | { | |
1208 | struct btrfs_path *path; | |
1209 | struct btrfs_key key; | |
1210 | int nr; | |
1211 | int ret; | |
1212 | ||
1213 | path = btrfs_alloc_path(); | |
1214 | if (!path) | |
1215 | return -ENOMEM; | |
1216 | ||
1217 | path->leave_spinning = 1; | |
1218 | ||
1219 | key.objectid = 0; | |
1220 | key.type = 0; | |
1221 | key.offset = 0; | |
1222 | ||
1223 | while (1) { | |
1224 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1225 | if (ret < 0) | |
1226 | goto out; | |
1227 | ||
1228 | nr = btrfs_header_nritems(path->nodes[0]); | |
1229 | if (!nr) | |
1230 | break; | |
1231 | ||
1232 | path->slots[0] = 0; | |
1233 | ret = btrfs_del_items(trans, root, path, 0, nr); | |
1234 | if (ret) | |
1235 | goto out; | |
1236 | ||
1237 | btrfs_release_path(path); | |
1238 | } | |
1239 | ||
1240 | ret = 0; | |
1241 | out: | |
1242 | btrfs_free_path(path); | |
1243 | return ret; | |
1244 | } | |
1245 | ||
1246 | int btrfs_clear_free_space_tree(struct btrfs_fs_info *fs_info) | |
1247 | { | |
1248 | struct btrfs_trans_handle *trans; | |
1249 | struct btrfs_root *tree_root = fs_info->tree_root; | |
1250 | struct btrfs_root *free_space_root = fs_info->free_space_root; | |
1251 | int ret; | |
1252 | ||
1253 | trans = btrfs_start_transaction(tree_root, 0); | |
1254 | if (IS_ERR(trans)) | |
1255 | return PTR_ERR(trans); | |
1256 | ||
1257 | btrfs_clear_fs_compat_ro(fs_info, FREE_SPACE_TREE); | |
14e46e04 DS |
1258 | btrfs_sysfs_feature_update(fs_info, |
1259 | BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE, FEAT_COMPAT_RO); | |
1260 | ||
a5ed9182 OS |
1261 | fs_info->free_space_root = NULL; |
1262 | ||
1263 | ret = clear_free_space_tree(trans, free_space_root); | |
1264 | if (ret) | |
1265 | goto abort; | |
1266 | ||
1267 | ret = btrfs_del_root(trans, tree_root, &free_space_root->root_key); | |
1268 | if (ret) | |
1269 | goto abort; | |
1270 | ||
1271 | list_del(&free_space_root->dirty_list); | |
1272 | ||
1273 | btrfs_tree_lock(free_space_root->node); | |
1274 | clean_tree_block(trans, tree_root->fs_info, free_space_root->node); | |
1275 | btrfs_tree_unlock(free_space_root->node); | |
1276 | btrfs_free_tree_block(trans, free_space_root, free_space_root->node, | |
1277 | 0, 1); | |
1278 | ||
1279 | free_extent_buffer(free_space_root->node); | |
1280 | free_extent_buffer(free_space_root->commit_root); | |
1281 | kfree(free_space_root); | |
1282 | ||
1283 | ret = btrfs_commit_transaction(trans, tree_root); | |
1284 | if (ret) | |
1285 | return ret; | |
1286 | ||
1287 | return 0; | |
1288 | ||
1289 | abort: | |
1290 | btrfs_abort_transaction(trans, tree_root, ret); | |
1291 | btrfs_end_transaction(trans, tree_root); | |
1292 | return ret; | |
1293 | } | |
1294 | ||
1295 | static int __add_block_group_free_space(struct btrfs_trans_handle *trans, | |
1296 | struct btrfs_fs_info *fs_info, | |
1297 | struct btrfs_block_group_cache *block_group, | |
1298 | struct btrfs_path *path) | |
1299 | { | |
1300 | u64 start, end; | |
1301 | int ret; | |
1302 | ||
1303 | start = block_group->key.objectid; | |
1304 | end = block_group->key.objectid + block_group->key.offset; | |
1305 | ||
1306 | block_group->needs_free_space = 0; | |
1307 | ||
1308 | ret = add_new_free_space_info(trans, fs_info, block_group, path); | |
1309 | if (ret) | |
1310 | return ret; | |
1311 | ||
1312 | return __add_to_free_space_tree(trans, fs_info, block_group, path, | |
1313 | block_group->key.objectid, | |
1314 | block_group->key.offset); | |
1315 | } | |
1316 | ||
1317 | int add_block_group_free_space(struct btrfs_trans_handle *trans, | |
1318 | struct btrfs_fs_info *fs_info, | |
1319 | struct btrfs_block_group_cache *block_group) | |
1320 | { | |
1321 | struct btrfs_path *path = NULL; | |
1322 | int ret = 0; | |
1323 | ||
1324 | if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) | |
1325 | return 0; | |
1326 | ||
1327 | mutex_lock(&block_group->free_space_lock); | |
1328 | if (!block_group->needs_free_space) | |
1329 | goto out; | |
1330 | ||
1331 | path = btrfs_alloc_path(); | |
1332 | if (!path) { | |
1333 | ret = -ENOMEM; | |
1334 | goto out; | |
1335 | } | |
1336 | ||
1337 | ret = __add_block_group_free_space(trans, fs_info, block_group, path); | |
1338 | ||
1339 | out: | |
1340 | btrfs_free_path(path); | |
1341 | mutex_unlock(&block_group->free_space_lock); | |
1342 | if (ret) | |
1343 | btrfs_abort_transaction(trans, fs_info->free_space_root, ret); | |
1344 | return ret; | |
1345 | } | |
1346 | ||
1347 | int remove_block_group_free_space(struct btrfs_trans_handle *trans, | |
1348 | struct btrfs_fs_info *fs_info, | |
1349 | struct btrfs_block_group_cache *block_group) | |
1350 | { | |
1351 | struct btrfs_root *root = fs_info->free_space_root; | |
1352 | struct btrfs_path *path; | |
1353 | struct btrfs_key key, found_key; | |
1354 | struct extent_buffer *leaf; | |
1355 | u64 start, end; | |
1356 | int done = 0, nr; | |
1357 | int ret; | |
1358 | ||
1359 | if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) | |
1360 | return 0; | |
1361 | ||
1362 | if (block_group->needs_free_space) { | |
1363 | /* We never added this block group to the free space tree. */ | |
1364 | return 0; | |
1365 | } | |
1366 | ||
1367 | path = btrfs_alloc_path(); | |
1368 | if (!path) { | |
1369 | ret = -ENOMEM; | |
1370 | goto out; | |
1371 | } | |
1372 | ||
1373 | start = block_group->key.objectid; | |
1374 | end = block_group->key.objectid + block_group->key.offset; | |
1375 | ||
1376 | key.objectid = end - 1; | |
1377 | key.type = (u8)-1; | |
1378 | key.offset = (u64)-1; | |
1379 | ||
1380 | while (!done) { | |
1381 | ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1); | |
1382 | if (ret) | |
1383 | goto out; | |
1384 | ||
1385 | leaf = path->nodes[0]; | |
1386 | nr = 0; | |
1387 | path->slots[0]++; | |
1388 | while (path->slots[0] > 0) { | |
1389 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1); | |
1390 | ||
1391 | if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) { | |
1392 | ASSERT(found_key.objectid == block_group->key.objectid); | |
1393 | ASSERT(found_key.offset == block_group->key.offset); | |
1394 | done = 1; | |
1395 | nr++; | |
1396 | path->slots[0]--; | |
1397 | break; | |
1398 | } else if (found_key.type == BTRFS_FREE_SPACE_EXTENT_KEY || | |
1399 | found_key.type == BTRFS_FREE_SPACE_BITMAP_KEY) { | |
1400 | ASSERT(found_key.objectid >= start); | |
1401 | ASSERT(found_key.objectid < end); | |
1402 | ASSERT(found_key.objectid + found_key.offset <= end); | |
1403 | nr++; | |
1404 | path->slots[0]--; | |
1405 | } else { | |
1406 | ASSERT(0); | |
1407 | } | |
1408 | } | |
1409 | ||
1410 | ret = btrfs_del_items(trans, root, path, path->slots[0], nr); | |
1411 | if (ret) | |
1412 | goto out; | |
1413 | btrfs_release_path(path); | |
1414 | } | |
1415 | ||
1416 | ret = 0; | |
1417 | out: | |
1418 | btrfs_free_path(path); | |
1419 | if (ret) | |
1420 | btrfs_abort_transaction(trans, root, ret); | |
1421 | return ret; | |
1422 | } | |
1423 | ||
1424 | static int load_free_space_bitmaps(struct btrfs_caching_control *caching_ctl, | |
1425 | struct btrfs_path *path, | |
1426 | u32 expected_extent_count) | |
1427 | { | |
1428 | struct btrfs_block_group_cache *block_group; | |
1429 | struct btrfs_fs_info *fs_info; | |
1430 | struct btrfs_root *root; | |
1431 | struct btrfs_key key; | |
1432 | int prev_bit = 0, bit; | |
1433 | /* Initialize to silence GCC. */ | |
1434 | u64 extent_start = 0; | |
1435 | u64 end, offset; | |
1436 | u64 total_found = 0; | |
1437 | u32 extent_count = 0; | |
1438 | int ret; | |
1439 | ||
1440 | block_group = caching_ctl->block_group; | |
1441 | fs_info = block_group->fs_info; | |
1442 | root = fs_info->free_space_root; | |
1443 | ||
1444 | end = block_group->key.objectid + block_group->key.offset; | |
1445 | ||
1446 | while (1) { | |
1447 | ret = btrfs_next_item(root, path); | |
1448 | if (ret < 0) | |
1449 | goto out; | |
1450 | if (ret) | |
1451 | break; | |
1452 | ||
1453 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
1454 | ||
1455 | if (key.type == BTRFS_FREE_SPACE_INFO_KEY) | |
1456 | break; | |
1457 | ||
1458 | ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY); | |
1459 | ASSERT(key.objectid < end && key.objectid + key.offset <= end); | |
1460 | ||
1461 | caching_ctl->progress = key.objectid; | |
1462 | ||
1463 | offset = key.objectid; | |
1464 | while (offset < key.objectid + key.offset) { | |
1465 | bit = free_space_test_bit(block_group, path, offset); | |
1466 | if (prev_bit == 0 && bit == 1) { | |
1467 | extent_start = offset; | |
1468 | } else if (prev_bit == 1 && bit == 0) { | |
1469 | total_found += add_new_free_space(block_group, | |
1470 | fs_info, | |
1471 | extent_start, | |
1472 | offset); | |
1473 | if (total_found > CACHING_CTL_WAKE_UP) { | |
1474 | total_found = 0; | |
1475 | wake_up(&caching_ctl->wait); | |
1476 | } | |
1477 | extent_count++; | |
1478 | } | |
1479 | prev_bit = bit; | |
1480 | offset += block_group->sectorsize; | |
1481 | } | |
1482 | } | |
1483 | if (prev_bit == 1) { | |
1484 | total_found += add_new_free_space(block_group, fs_info, | |
1485 | extent_start, end); | |
1486 | extent_count++; | |
1487 | } | |
1488 | ||
1489 | if (extent_count != expected_extent_count) { | |
1490 | btrfs_err(fs_info, "incorrect extent count for %llu; counted %u, expected %u", | |
1491 | block_group->key.objectid, extent_count, | |
1492 | expected_extent_count); | |
1493 | ASSERT(0); | |
1494 | ret = -EIO; | |
1495 | goto out; | |
1496 | } | |
1497 | ||
1498 | caching_ctl->progress = (u64)-1; | |
1499 | ||
1500 | ret = 0; | |
1501 | out: | |
1502 | return ret; | |
1503 | } | |
1504 | ||
1505 | static int load_free_space_extents(struct btrfs_caching_control *caching_ctl, | |
1506 | struct btrfs_path *path, | |
1507 | u32 expected_extent_count) | |
1508 | { | |
1509 | struct btrfs_block_group_cache *block_group; | |
1510 | struct btrfs_fs_info *fs_info; | |
1511 | struct btrfs_root *root; | |
1512 | struct btrfs_key key; | |
1513 | u64 end; | |
1514 | u64 total_found = 0; | |
1515 | u32 extent_count = 0; | |
1516 | int ret; | |
1517 | ||
1518 | block_group = caching_ctl->block_group; | |
1519 | fs_info = block_group->fs_info; | |
1520 | root = fs_info->free_space_root; | |
1521 | ||
1522 | end = block_group->key.objectid + block_group->key.offset; | |
1523 | ||
1524 | while (1) { | |
1525 | ret = btrfs_next_item(root, path); | |
1526 | if (ret < 0) | |
1527 | goto out; | |
1528 | if (ret) | |
1529 | break; | |
1530 | ||
1531 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
1532 | ||
1533 | if (key.type == BTRFS_FREE_SPACE_INFO_KEY) | |
1534 | break; | |
1535 | ||
1536 | ASSERT(key.type == BTRFS_FREE_SPACE_EXTENT_KEY); | |
1537 | ASSERT(key.objectid < end && key.objectid + key.offset <= end); | |
1538 | ||
1539 | caching_ctl->progress = key.objectid; | |
1540 | ||
1541 | total_found += add_new_free_space(block_group, fs_info, | |
1542 | key.objectid, | |
1543 | key.objectid + key.offset); | |
1544 | if (total_found > CACHING_CTL_WAKE_UP) { | |
1545 | total_found = 0; | |
1546 | wake_up(&caching_ctl->wait); | |
1547 | } | |
1548 | extent_count++; | |
1549 | } | |
1550 | ||
1551 | if (extent_count != expected_extent_count) { | |
1552 | btrfs_err(fs_info, "incorrect extent count for %llu; counted %u, expected %u", | |
1553 | block_group->key.objectid, extent_count, | |
1554 | expected_extent_count); | |
1555 | ASSERT(0); | |
1556 | ret = -EIO; | |
1557 | goto out; | |
1558 | } | |
1559 | ||
1560 | caching_ctl->progress = (u64)-1; | |
1561 | ||
1562 | ret = 0; | |
1563 | out: | |
1564 | return ret; | |
1565 | } | |
1566 | ||
1567 | int load_free_space_tree(struct btrfs_caching_control *caching_ctl) | |
1568 | { | |
1569 | struct btrfs_block_group_cache *block_group; | |
1570 | struct btrfs_fs_info *fs_info; | |
1571 | struct btrfs_free_space_info *info; | |
1572 | struct btrfs_path *path; | |
1573 | u32 extent_count, flags; | |
1574 | int ret; | |
1575 | ||
1576 | block_group = caching_ctl->block_group; | |
1577 | fs_info = block_group->fs_info; | |
1578 | ||
1579 | path = btrfs_alloc_path(); | |
1580 | if (!path) | |
1581 | return -ENOMEM; | |
1582 | ||
1583 | /* | |
1584 | * Just like caching_thread() doesn't want to deadlock on the extent | |
1585 | * tree, we don't want to deadlock on the free space tree. | |
1586 | */ | |
1587 | path->skip_locking = 1; | |
1588 | path->search_commit_root = 1; | |
1589 | path->reada = 1; | |
1590 | ||
1591 | info = search_free_space_info(NULL, fs_info, block_group, path, 0); | |
1592 | if (IS_ERR(info)) { | |
1593 | ret = PTR_ERR(info); | |
1594 | goto out; | |
1595 | } | |
1596 | extent_count = btrfs_free_space_extent_count(path->nodes[0], info); | |
1597 | flags = btrfs_free_space_flags(path->nodes[0], info); | |
1598 | ||
1599 | /* | |
1600 | * We left path pointing to the free space info item, so now | |
1601 | * load_free_space_foo can just iterate through the free space tree from | |
1602 | * there. | |
1603 | */ | |
1604 | if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) | |
1605 | ret = load_free_space_bitmaps(caching_ctl, path, extent_count); | |
1606 | else | |
1607 | ret = load_free_space_extents(caching_ctl, path, extent_count); | |
1608 | ||
1609 | out: | |
1610 | btrfs_free_path(path); | |
1611 | return ret; | |
1612 | } |