]>
Commit | Line | Data |
---|---|---|
0f9dd46c JB |
1 | /* |
2 | * Copyright (C) 2008 Red Hat. 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 | ||
96303081 | 19 | #include <linux/pagemap.h> |
0f9dd46c | 20 | #include <linux/sched.h> |
96303081 | 21 | #include <linux/math64.h> |
0f9dd46c | 22 | #include "ctree.h" |
fa9c0d79 CM |
23 | #include "free-space-cache.h" |
24 | #include "transaction.h" | |
25 | ||
96303081 JB |
26 | #define BITS_PER_BITMAP (PAGE_CACHE_SIZE * 8) |
27 | #define MAX_CACHE_BYTES_PER_GIG (32 * 1024) | |
0f9dd46c | 28 | |
96303081 JB |
29 | static inline unsigned long offset_to_bit(u64 bitmap_start, u64 sectorsize, |
30 | u64 offset) | |
0f9dd46c | 31 | { |
96303081 JB |
32 | BUG_ON(offset < bitmap_start); |
33 | offset -= bitmap_start; | |
34 | return (unsigned long)(div64_u64(offset, sectorsize)); | |
35 | } | |
0f9dd46c | 36 | |
96303081 JB |
37 | static inline unsigned long bytes_to_bits(u64 bytes, u64 sectorsize) |
38 | { | |
39 | return (unsigned long)(div64_u64(bytes, sectorsize)); | |
40 | } | |
0f9dd46c | 41 | |
96303081 JB |
42 | static inline u64 offset_to_bitmap(struct btrfs_block_group_cache *block_group, |
43 | u64 offset) | |
44 | { | |
45 | u64 bitmap_start; | |
46 | u64 bytes_per_bitmap; | |
0f9dd46c | 47 | |
96303081 JB |
48 | bytes_per_bitmap = BITS_PER_BITMAP * block_group->sectorsize; |
49 | bitmap_start = offset - block_group->key.objectid; | |
50 | bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap); | |
51 | bitmap_start *= bytes_per_bitmap; | |
52 | bitmap_start += block_group->key.objectid; | |
0f9dd46c | 53 | |
96303081 | 54 | return bitmap_start; |
0f9dd46c JB |
55 | } |
56 | ||
96303081 JB |
57 | static int tree_insert_offset(struct rb_root *root, u64 offset, |
58 | struct rb_node *node, int bitmap) | |
0f9dd46c JB |
59 | { |
60 | struct rb_node **p = &root->rb_node; | |
61 | struct rb_node *parent = NULL; | |
62 | struct btrfs_free_space *info; | |
63 | ||
64 | while (*p) { | |
65 | parent = *p; | |
96303081 | 66 | info = rb_entry(parent, struct btrfs_free_space, offset_index); |
0f9dd46c | 67 | |
96303081 | 68 | if (offset < info->offset) { |
0f9dd46c | 69 | p = &(*p)->rb_left; |
96303081 | 70 | } else if (offset > info->offset) { |
0f9dd46c | 71 | p = &(*p)->rb_right; |
96303081 JB |
72 | } else { |
73 | /* | |
74 | * we could have a bitmap entry and an extent entry | |
75 | * share the same offset. If this is the case, we want | |
76 | * the extent entry to always be found first if we do a | |
77 | * linear search through the tree, since we want to have | |
78 | * the quickest allocation time, and allocating from an | |
79 | * extent is faster than allocating from a bitmap. So | |
80 | * if we're inserting a bitmap and we find an entry at | |
81 | * this offset, we want to go right, or after this entry | |
82 | * logically. If we are inserting an extent and we've | |
83 | * found a bitmap, we want to go left, or before | |
84 | * logically. | |
85 | */ | |
86 | if (bitmap) { | |
87 | WARN_ON(info->bitmap); | |
88 | p = &(*p)->rb_right; | |
89 | } else { | |
90 | WARN_ON(!info->bitmap); | |
91 | p = &(*p)->rb_left; | |
92 | } | |
93 | } | |
0f9dd46c JB |
94 | } |
95 | ||
96 | rb_link_node(node, parent, p); | |
97 | rb_insert_color(node, root); | |
98 | ||
99 | return 0; | |
100 | } | |
101 | ||
102 | /* | |
70cb0743 JB |
103 | * searches the tree for the given offset. |
104 | * | |
96303081 JB |
105 | * fuzzy - If this is set, then we are trying to make an allocation, and we just |
106 | * want a section that has at least bytes size and comes at or after the given | |
107 | * offset. | |
0f9dd46c | 108 | */ |
96303081 JB |
109 | static struct btrfs_free_space * |
110 | tree_search_offset(struct btrfs_block_group_cache *block_group, | |
111 | u64 offset, int bitmap_only, int fuzzy) | |
0f9dd46c | 112 | { |
96303081 JB |
113 | struct rb_node *n = block_group->free_space_offset.rb_node; |
114 | struct btrfs_free_space *entry, *prev = NULL; | |
115 | ||
116 | /* find entry that is closest to the 'offset' */ | |
117 | while (1) { | |
118 | if (!n) { | |
119 | entry = NULL; | |
120 | break; | |
121 | } | |
0f9dd46c | 122 | |
0f9dd46c | 123 | entry = rb_entry(n, struct btrfs_free_space, offset_index); |
96303081 | 124 | prev = entry; |
0f9dd46c | 125 | |
96303081 | 126 | if (offset < entry->offset) |
0f9dd46c | 127 | n = n->rb_left; |
96303081 | 128 | else if (offset > entry->offset) |
0f9dd46c | 129 | n = n->rb_right; |
96303081 | 130 | else |
0f9dd46c | 131 | break; |
0f9dd46c JB |
132 | } |
133 | ||
96303081 JB |
134 | if (bitmap_only) { |
135 | if (!entry) | |
136 | return NULL; | |
137 | if (entry->bitmap) | |
138 | return entry; | |
0f9dd46c | 139 | |
96303081 JB |
140 | /* |
141 | * bitmap entry and extent entry may share same offset, | |
142 | * in that case, bitmap entry comes after extent entry. | |
143 | */ | |
144 | n = rb_next(n); | |
145 | if (!n) | |
146 | return NULL; | |
147 | entry = rb_entry(n, struct btrfs_free_space, offset_index); | |
148 | if (entry->offset != offset) | |
149 | return NULL; | |
0f9dd46c | 150 | |
96303081 JB |
151 | WARN_ON(!entry->bitmap); |
152 | return entry; | |
153 | } else if (entry) { | |
154 | if (entry->bitmap) { | |
0f9dd46c | 155 | /* |
96303081 JB |
156 | * if previous extent entry covers the offset, |
157 | * we should return it instead of the bitmap entry | |
0f9dd46c | 158 | */ |
96303081 JB |
159 | n = &entry->offset_index; |
160 | while (1) { | |
161 | n = rb_prev(n); | |
162 | if (!n) | |
163 | break; | |
164 | prev = rb_entry(n, struct btrfs_free_space, | |
165 | offset_index); | |
166 | if (!prev->bitmap) { | |
167 | if (prev->offset + prev->bytes > offset) | |
168 | entry = prev; | |
169 | break; | |
170 | } | |
0f9dd46c | 171 | } |
96303081 JB |
172 | } |
173 | return entry; | |
174 | } | |
175 | ||
176 | if (!prev) | |
177 | return NULL; | |
178 | ||
179 | /* find last entry before the 'offset' */ | |
180 | entry = prev; | |
181 | if (entry->offset > offset) { | |
182 | n = rb_prev(&entry->offset_index); | |
183 | if (n) { | |
184 | entry = rb_entry(n, struct btrfs_free_space, | |
185 | offset_index); | |
186 | BUG_ON(entry->offset > offset); | |
0f9dd46c | 187 | } else { |
96303081 JB |
188 | if (fuzzy) |
189 | return entry; | |
190 | else | |
191 | return NULL; | |
0f9dd46c JB |
192 | } |
193 | } | |
194 | ||
96303081 JB |
195 | if (entry->bitmap) { |
196 | n = &entry->offset_index; | |
197 | while (1) { | |
198 | n = rb_prev(n); | |
199 | if (!n) | |
200 | break; | |
201 | prev = rb_entry(n, struct btrfs_free_space, | |
202 | offset_index); | |
203 | if (!prev->bitmap) { | |
204 | if (prev->offset + prev->bytes > offset) | |
205 | return prev; | |
206 | break; | |
207 | } | |
208 | } | |
209 | if (entry->offset + BITS_PER_BITMAP * | |
210 | block_group->sectorsize > offset) | |
211 | return entry; | |
212 | } else if (entry->offset + entry->bytes > offset) | |
213 | return entry; | |
214 | ||
215 | if (!fuzzy) | |
216 | return NULL; | |
217 | ||
218 | while (1) { | |
219 | if (entry->bitmap) { | |
220 | if (entry->offset + BITS_PER_BITMAP * | |
221 | block_group->sectorsize > offset) | |
222 | break; | |
223 | } else { | |
224 | if (entry->offset + entry->bytes > offset) | |
225 | break; | |
226 | } | |
227 | ||
228 | n = rb_next(&entry->offset_index); | |
229 | if (!n) | |
230 | return NULL; | |
231 | entry = rb_entry(n, struct btrfs_free_space, offset_index); | |
232 | } | |
233 | return entry; | |
0f9dd46c JB |
234 | } |
235 | ||
236 | static void unlink_free_space(struct btrfs_block_group_cache *block_group, | |
237 | struct btrfs_free_space *info) | |
238 | { | |
239 | rb_erase(&info->offset_index, &block_group->free_space_offset); | |
96303081 | 240 | block_group->free_extents--; |
0f9dd46c JB |
241 | } |
242 | ||
243 | static int link_free_space(struct btrfs_block_group_cache *block_group, | |
244 | struct btrfs_free_space *info) | |
245 | { | |
246 | int ret = 0; | |
247 | ||
96303081 | 248 | BUG_ON(!info->bitmap && !info->bytes); |
0f9dd46c | 249 | ret = tree_insert_offset(&block_group->free_space_offset, info->offset, |
96303081 | 250 | &info->offset_index, (info->bitmap != NULL)); |
0f9dd46c JB |
251 | if (ret) |
252 | return ret; | |
253 | ||
96303081 JB |
254 | block_group->free_extents++; |
255 | return ret; | |
256 | } | |
257 | ||
258 | static void recalculate_thresholds(struct btrfs_block_group_cache *block_group) | |
259 | { | |
260 | u64 max_bytes, possible_bytes; | |
261 | ||
262 | /* | |
263 | * The goal is to keep the total amount of memory used per 1gb of space | |
264 | * at or below 32k, so we need to adjust how much memory we allow to be | |
265 | * used by extent based free space tracking | |
266 | */ | |
267 | max_bytes = MAX_CACHE_BYTES_PER_GIG * | |
268 | (div64_u64(block_group->key.offset, 1024 * 1024 * 1024)); | |
269 | ||
270 | possible_bytes = (block_group->total_bitmaps * PAGE_CACHE_SIZE) + | |
271 | (sizeof(struct btrfs_free_space) * | |
272 | block_group->extents_thresh); | |
273 | ||
274 | if (possible_bytes > max_bytes) { | |
275 | int extent_bytes = max_bytes - | |
276 | (block_group->total_bitmaps * PAGE_CACHE_SIZE); | |
277 | ||
278 | if (extent_bytes <= 0) { | |
279 | block_group->extents_thresh = 0; | |
280 | return; | |
281 | } | |
282 | ||
283 | block_group->extents_thresh = extent_bytes / | |
284 | (sizeof(struct btrfs_free_space)); | |
285 | } | |
286 | } | |
287 | ||
288 | static void bitmap_clear_bits(struct btrfs_free_space *info, u64 offset, u64 bytes, | |
289 | u64 sectorsize) | |
290 | { | |
291 | unsigned long start, end; | |
292 | unsigned long i; | |
293 | ||
294 | start = offset_to_bit(info->offset, sectorsize, offset); | |
295 | end = start + bytes_to_bits(bytes, sectorsize); | |
296 | BUG_ON(end > BITS_PER_BITMAP); | |
297 | ||
298 | for (i = start; i < end; i++) | |
299 | clear_bit(i, info->bitmap); | |
300 | ||
301 | info->bytes -= bytes; | |
302 | } | |
303 | ||
304 | static void bitmap_set_bits(struct btrfs_free_space *info, u64 offset, u64 bytes, | |
305 | u64 sectorsize) | |
306 | { | |
307 | unsigned long start, end; | |
308 | unsigned long i; | |
309 | ||
310 | start = offset_to_bit(info->offset, sectorsize, offset); | |
311 | end = start + bytes_to_bits(bytes, sectorsize); | |
312 | BUG_ON(end > BITS_PER_BITMAP); | |
313 | ||
314 | for (i = start; i < end; i++) | |
315 | set_bit(i, info->bitmap); | |
316 | ||
317 | info->bytes += bytes; | |
318 | } | |
319 | ||
320 | static int search_bitmap(struct btrfs_block_group_cache *block_group, | |
321 | struct btrfs_free_space *bitmap_info, u64 *offset, | |
322 | u64 *bytes) | |
323 | { | |
324 | unsigned long found_bits = 0; | |
325 | unsigned long bits, i; | |
326 | unsigned long next_zero; | |
327 | ||
328 | i = offset_to_bit(bitmap_info->offset, block_group->sectorsize, | |
329 | max_t(u64, *offset, bitmap_info->offset)); | |
330 | bits = bytes_to_bits(*bytes, block_group->sectorsize); | |
331 | ||
332 | for (i = find_next_bit(bitmap_info->bitmap, BITS_PER_BITMAP, i); | |
333 | i < BITS_PER_BITMAP; | |
334 | i = find_next_bit(bitmap_info->bitmap, BITS_PER_BITMAP, i + 1)) { | |
335 | next_zero = find_next_zero_bit(bitmap_info->bitmap, | |
336 | BITS_PER_BITMAP, i); | |
337 | if ((next_zero - i) >= bits) { | |
338 | found_bits = next_zero - i; | |
339 | break; | |
340 | } | |
341 | i = next_zero; | |
342 | } | |
343 | ||
344 | if (found_bits) { | |
345 | *offset = (u64)(i * block_group->sectorsize) + | |
346 | bitmap_info->offset; | |
347 | *bytes = (u64)(found_bits) * block_group->sectorsize; | |
348 | return 0; | |
349 | } | |
350 | ||
351 | return -1; | |
352 | } | |
353 | ||
354 | static struct btrfs_free_space *find_free_space(struct btrfs_block_group_cache | |
355 | *block_group, u64 *offset, | |
356 | u64 *bytes, int debug) | |
357 | { | |
358 | struct btrfs_free_space *entry; | |
359 | struct rb_node *node; | |
360 | int ret; | |
361 | ||
362 | if (!block_group->free_space_offset.rb_node) | |
363 | return NULL; | |
364 | ||
365 | entry = tree_search_offset(block_group, | |
366 | offset_to_bitmap(block_group, *offset), | |
367 | 0, 1); | |
368 | if (!entry) | |
369 | return NULL; | |
370 | ||
371 | for (node = &entry->offset_index; node; node = rb_next(node)) { | |
372 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
373 | if (entry->bytes < *bytes) | |
374 | continue; | |
375 | ||
376 | if (entry->bitmap) { | |
377 | ret = search_bitmap(block_group, entry, offset, bytes); | |
378 | if (!ret) | |
379 | return entry; | |
380 | continue; | |
381 | } | |
382 | ||
383 | *offset = entry->offset; | |
384 | *bytes = entry->bytes; | |
385 | return entry; | |
386 | } | |
387 | ||
388 | return NULL; | |
389 | } | |
390 | ||
391 | static void add_new_bitmap(struct btrfs_block_group_cache *block_group, | |
392 | struct btrfs_free_space *info, u64 offset) | |
393 | { | |
394 | u64 bytes_per_bg = BITS_PER_BITMAP * block_group->sectorsize; | |
395 | int max_bitmaps = (int)div64_u64(block_group->key.offset + | |
396 | bytes_per_bg - 1, bytes_per_bg); | |
397 | BUG_ON(block_group->total_bitmaps >= max_bitmaps); | |
398 | ||
399 | info->offset = offset_to_bitmap(block_group, offset); | |
400 | link_free_space(block_group, info); | |
401 | block_group->total_bitmaps++; | |
402 | ||
403 | recalculate_thresholds(block_group); | |
404 | } | |
405 | ||
406 | static noinline int remove_from_bitmap(struct btrfs_block_group_cache *block_group, | |
407 | struct btrfs_free_space *bitmap_info, | |
408 | u64 *offset, u64 *bytes) | |
409 | { | |
410 | u64 end; | |
411 | ||
412 | again: | |
413 | end = bitmap_info->offset + | |
414 | (u64)(BITS_PER_BITMAP * block_group->sectorsize) - 1; | |
415 | ||
416 | if (*offset > bitmap_info->offset && *offset + *bytes > end) { | |
417 | bitmap_clear_bits(bitmap_info, *offset, | |
418 | end - *offset + 1, block_group->sectorsize); | |
419 | *bytes -= end - *offset + 1; | |
420 | *offset = end + 1; | |
421 | } else if (*offset >= bitmap_info->offset && *offset + *bytes <= end) { | |
422 | bitmap_clear_bits(bitmap_info, *offset, | |
423 | *bytes, block_group->sectorsize); | |
424 | *bytes = 0; | |
425 | } | |
426 | ||
427 | if (*bytes) { | |
428 | if (!bitmap_info->bytes) { | |
429 | unlink_free_space(block_group, bitmap_info); | |
430 | kfree(bitmap_info->bitmap); | |
431 | kfree(bitmap_info); | |
432 | block_group->total_bitmaps--; | |
433 | recalculate_thresholds(block_group); | |
434 | } | |
435 | ||
436 | bitmap_info = tree_search_offset(block_group, | |
437 | offset_to_bitmap(block_group, | |
438 | *offset), | |
439 | 1, 0); | |
440 | if (!bitmap_info) | |
441 | return -EINVAL; | |
442 | ||
443 | if (!bitmap_info->bitmap) | |
444 | return -EAGAIN; | |
445 | ||
446 | goto again; | |
447 | } else if (!bitmap_info->bytes) { | |
448 | unlink_free_space(block_group, bitmap_info); | |
449 | kfree(bitmap_info->bitmap); | |
450 | kfree(bitmap_info); | |
451 | block_group->total_bitmaps--; | |
452 | recalculate_thresholds(block_group); | |
453 | } | |
454 | ||
455 | return 0; | |
456 | } | |
457 | ||
458 | static int insert_into_bitmap(struct btrfs_block_group_cache *block_group, | |
459 | struct btrfs_free_space *info) | |
460 | { | |
461 | struct btrfs_free_space *bitmap_info; | |
462 | int added = 0; | |
463 | u64 bytes, offset, end; | |
464 | int ret; | |
465 | ||
466 | /* | |
467 | * If we are below the extents threshold then we can add this as an | |
468 | * extent, and don't have to deal with the bitmap | |
469 | */ | |
470 | if (block_group->free_extents < block_group->extents_thresh && | |
471 | info->bytes > block_group->sectorsize * 4) | |
472 | return 0; | |
473 | ||
474 | /* | |
475 | * some block groups are so tiny they can't be enveloped by a bitmap, so | |
476 | * don't even bother to create a bitmap for this | |
477 | */ | |
478 | if (BITS_PER_BITMAP * block_group->sectorsize > | |
479 | block_group->key.offset) | |
480 | return 0; | |
481 | ||
482 | bytes = info->bytes; | |
483 | offset = info->offset; | |
484 | ||
485 | again: | |
486 | bitmap_info = tree_search_offset(block_group, | |
487 | offset_to_bitmap(block_group, offset), | |
488 | 1, 0); | |
489 | if (!bitmap_info) { | |
490 | BUG_ON(added); | |
491 | goto new_bitmap; | |
492 | } | |
493 | ||
494 | end = bitmap_info->offset + | |
495 | (u64)(BITS_PER_BITMAP * block_group->sectorsize); | |
496 | ||
497 | if (offset >= bitmap_info->offset && offset + bytes > end) { | |
498 | bitmap_set_bits(bitmap_info, offset, end - offset, | |
499 | block_group->sectorsize); | |
500 | bytes -= end - offset; | |
501 | offset = end; | |
502 | added = 0; | |
503 | } else if (offset >= bitmap_info->offset && offset + bytes <= end) { | |
504 | bitmap_set_bits(bitmap_info, offset, bytes, | |
505 | block_group->sectorsize); | |
506 | bytes = 0; | |
507 | } else { | |
508 | BUG(); | |
509 | } | |
510 | ||
511 | if (!bytes) { | |
512 | ret = 1; | |
513 | goto out; | |
514 | } else | |
515 | goto again; | |
516 | ||
517 | new_bitmap: | |
518 | if (info && info->bitmap) { | |
519 | add_new_bitmap(block_group, info, offset); | |
520 | added = 1; | |
521 | info = NULL; | |
522 | goto again; | |
523 | } else { | |
524 | spin_unlock(&block_group->tree_lock); | |
525 | ||
526 | /* no pre-allocated info, allocate a new one */ | |
527 | if (!info) { | |
528 | info = kzalloc(sizeof(struct btrfs_free_space), | |
529 | GFP_NOFS); | |
530 | if (!info) { | |
531 | spin_lock(&block_group->tree_lock); | |
532 | ret = -ENOMEM; | |
533 | goto out; | |
534 | } | |
535 | } | |
536 | ||
537 | /* allocate the bitmap */ | |
538 | info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS); | |
539 | spin_lock(&block_group->tree_lock); | |
540 | if (!info->bitmap) { | |
541 | ret = -ENOMEM; | |
542 | goto out; | |
543 | } | |
544 | goto again; | |
545 | } | |
546 | ||
547 | out: | |
548 | if (info) { | |
549 | if (info->bitmap) | |
550 | kfree(info->bitmap); | |
551 | kfree(info); | |
552 | } | |
0f9dd46c JB |
553 | |
554 | return ret; | |
555 | } | |
556 | ||
6226cb0a JB |
557 | int btrfs_add_free_space(struct btrfs_block_group_cache *block_group, |
558 | u64 offset, u64 bytes) | |
0f9dd46c | 559 | { |
96303081 JB |
560 | struct btrfs_free_space *right_info = NULL; |
561 | struct btrfs_free_space *left_info = NULL; | |
0f9dd46c | 562 | struct btrfs_free_space *info = NULL; |
0f9dd46c JB |
563 | int ret = 0; |
564 | ||
6226cb0a JB |
565 | info = kzalloc(sizeof(struct btrfs_free_space), GFP_NOFS); |
566 | if (!info) | |
567 | return -ENOMEM; | |
568 | ||
569 | info->offset = offset; | |
570 | info->bytes = bytes; | |
571 | ||
572 | spin_lock(&block_group->tree_lock); | |
573 | ||
0f9dd46c JB |
574 | /* |
575 | * first we want to see if there is free space adjacent to the range we | |
576 | * are adding, if there is remove that struct and add a new one to | |
577 | * cover the entire range | |
578 | */ | |
96303081 JB |
579 | right_info = tree_search_offset(block_group, offset + bytes, 0, 0); |
580 | if (right_info && rb_prev(&right_info->offset_index)) | |
581 | left_info = rb_entry(rb_prev(&right_info->offset_index), | |
582 | struct btrfs_free_space, offset_index); | |
583 | else | |
584 | left_info = tree_search_offset(block_group, offset - 1, 0, 0); | |
0f9dd46c | 585 | |
96303081 JB |
586 | /* |
587 | * If there was no extent directly to the left or right of this new | |
588 | * extent then we know we're going to have to allocate a new extent, so | |
589 | * before we do that see if we need to drop this into a bitmap | |
590 | */ | |
591 | if ((!left_info || left_info->bitmap) && | |
592 | (!right_info || right_info->bitmap)) { | |
593 | ret = insert_into_bitmap(block_group, info); | |
594 | ||
595 | if (ret < 0) { | |
596 | goto out; | |
597 | } else if (ret) { | |
598 | ret = 0; | |
599 | goto out; | |
600 | } | |
601 | } | |
602 | ||
603 | if (right_info && !right_info->bitmap) { | |
0f9dd46c | 604 | unlink_free_space(block_group, right_info); |
6226cb0a JB |
605 | info->bytes += right_info->bytes; |
606 | kfree(right_info); | |
0f9dd46c JB |
607 | } |
608 | ||
96303081 JB |
609 | if (left_info && !left_info->bitmap && |
610 | left_info->offset + left_info->bytes == offset) { | |
0f9dd46c | 611 | unlink_free_space(block_group, left_info); |
6226cb0a JB |
612 | info->offset = left_info->offset; |
613 | info->bytes += left_info->bytes; | |
614 | kfree(left_info); | |
0f9dd46c JB |
615 | } |
616 | ||
0f9dd46c JB |
617 | ret = link_free_space(block_group, info); |
618 | if (ret) | |
619 | kfree(info); | |
96303081 | 620 | out: |
6226cb0a JB |
621 | spin_unlock(&block_group->tree_lock); |
622 | ||
0f9dd46c | 623 | if (ret) { |
96303081 | 624 | printk(KERN_CRIT "btrfs: unable to add free space :%d\n", ret); |
c293498b | 625 | BUG_ON(ret == -EEXIST); |
0f9dd46c JB |
626 | } |
627 | ||
0f9dd46c JB |
628 | return ret; |
629 | } | |
630 | ||
6226cb0a JB |
631 | int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group, |
632 | u64 offset, u64 bytes) | |
0f9dd46c JB |
633 | { |
634 | struct btrfs_free_space *info; | |
96303081 | 635 | struct btrfs_free_space *next_info = NULL; |
0f9dd46c JB |
636 | int ret = 0; |
637 | ||
6226cb0a JB |
638 | spin_lock(&block_group->tree_lock); |
639 | ||
96303081 JB |
640 | again: |
641 | info = tree_search_offset(block_group, offset, 0, 0); | |
642 | if (!info) { | |
643 | WARN_ON(1); | |
644 | goto out_lock; | |
645 | } | |
646 | ||
647 | if (info->bytes < bytes && rb_next(&info->offset_index)) { | |
648 | u64 end; | |
649 | next_info = rb_entry(rb_next(&info->offset_index), | |
650 | struct btrfs_free_space, | |
651 | offset_index); | |
652 | ||
653 | if (next_info->bitmap) | |
654 | end = next_info->offset + BITS_PER_BITMAP * | |
655 | block_group->sectorsize - 1; | |
656 | else | |
657 | end = next_info->offset + next_info->bytes; | |
658 | ||
659 | if (next_info->bytes < bytes || | |
660 | next_info->offset > offset || offset > end) { | |
661 | printk(KERN_CRIT "Found free space at %llu, size %llu," | |
662 | " trying to use %llu\n", | |
663 | (unsigned long long)info->offset, | |
664 | (unsigned long long)info->bytes, | |
665 | (unsigned long long)bytes); | |
0f9dd46c JB |
666 | WARN_ON(1); |
667 | ret = -EINVAL; | |
96303081 | 668 | goto out_lock; |
0f9dd46c | 669 | } |
0f9dd46c | 670 | |
96303081 JB |
671 | info = next_info; |
672 | } | |
673 | ||
674 | if (info->bytes == bytes) { | |
675 | unlink_free_space(block_group, info); | |
676 | if (info->bitmap) { | |
677 | kfree(info->bitmap); | |
678 | block_group->total_bitmaps--; | |
0f9dd46c | 679 | } |
96303081 JB |
680 | kfree(info); |
681 | goto out_lock; | |
682 | } | |
0f9dd46c | 683 | |
96303081 JB |
684 | if (!info->bitmap && info->offset == offset) { |
685 | unlink_free_space(block_group, info); | |
0f9dd46c JB |
686 | info->offset += bytes; |
687 | info->bytes -= bytes; | |
96303081 JB |
688 | link_free_space(block_group, info); |
689 | goto out_lock; | |
690 | } | |
0f9dd46c | 691 | |
96303081 JB |
692 | if (!info->bitmap && info->offset <= offset && |
693 | info->offset + info->bytes >= offset + bytes) { | |
9b49c9b9 CM |
694 | u64 old_start = info->offset; |
695 | /* | |
696 | * we're freeing space in the middle of the info, | |
697 | * this can happen during tree log replay | |
698 | * | |
699 | * first unlink the old info and then | |
700 | * insert it again after the hole we're creating | |
701 | */ | |
702 | unlink_free_space(block_group, info); | |
703 | if (offset + bytes < info->offset + info->bytes) { | |
704 | u64 old_end = info->offset + info->bytes; | |
705 | ||
706 | info->offset = offset + bytes; | |
707 | info->bytes = old_end - info->offset; | |
708 | ret = link_free_space(block_group, info); | |
96303081 JB |
709 | WARN_ON(ret); |
710 | if (ret) | |
711 | goto out_lock; | |
9b49c9b9 CM |
712 | } else { |
713 | /* the hole we're creating ends at the end | |
714 | * of the info struct, just free the info | |
715 | */ | |
716 | kfree(info); | |
717 | } | |
6226cb0a | 718 | spin_unlock(&block_group->tree_lock); |
96303081 JB |
719 | |
720 | /* step two, insert a new info struct to cover | |
721 | * anything before the hole | |
9b49c9b9 | 722 | */ |
6226cb0a JB |
723 | ret = btrfs_add_free_space(block_group, old_start, |
724 | offset - old_start); | |
96303081 JB |
725 | WARN_ON(ret); |
726 | goto out; | |
0f9dd46c | 727 | } |
96303081 JB |
728 | |
729 | ret = remove_from_bitmap(block_group, info, &offset, &bytes); | |
730 | if (ret == -EAGAIN) | |
731 | goto again; | |
732 | BUG_ON(ret); | |
733 | out_lock: | |
734 | spin_unlock(&block_group->tree_lock); | |
0f9dd46c | 735 | out: |
25179201 JB |
736 | return ret; |
737 | } | |
738 | ||
0f9dd46c JB |
739 | void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group, |
740 | u64 bytes) | |
741 | { | |
742 | struct btrfs_free_space *info; | |
743 | struct rb_node *n; | |
744 | int count = 0; | |
745 | ||
746 | for (n = rb_first(&block_group->free_space_offset); n; n = rb_next(n)) { | |
747 | info = rb_entry(n, struct btrfs_free_space, offset_index); | |
748 | if (info->bytes >= bytes) | |
749 | count++; | |
96303081 | 750 | printk(KERN_CRIT "entry offset %llu, bytes %llu, bitmap %s\n", |
21380931 | 751 | (unsigned long long)info->offset, |
96303081 JB |
752 | (unsigned long long)info->bytes, |
753 | (info->bitmap) ? "yes" : "no"); | |
0f9dd46c | 754 | } |
96303081 JB |
755 | printk(KERN_INFO "block group has cluster?: %s\n", |
756 | list_empty(&block_group->cluster_list) ? "no" : "yes"); | |
0f9dd46c JB |
757 | printk(KERN_INFO "%d blocks of free space at or bigger than bytes is" |
758 | "\n", count); | |
759 | } | |
760 | ||
761 | u64 btrfs_block_group_free_space(struct btrfs_block_group_cache *block_group) | |
762 | { | |
763 | struct btrfs_free_space *info; | |
764 | struct rb_node *n; | |
765 | u64 ret = 0; | |
766 | ||
767 | for (n = rb_first(&block_group->free_space_offset); n; | |
768 | n = rb_next(n)) { | |
769 | info = rb_entry(n, struct btrfs_free_space, offset_index); | |
770 | ret += info->bytes; | |
771 | } | |
772 | ||
773 | return ret; | |
774 | } | |
775 | ||
fa9c0d79 CM |
776 | /* |
777 | * for a given cluster, put all of its extents back into the free | |
778 | * space cache. If the block group passed doesn't match the block group | |
779 | * pointed to by the cluster, someone else raced in and freed the | |
780 | * cluster already. In that case, we just return without changing anything | |
781 | */ | |
782 | static int | |
783 | __btrfs_return_cluster_to_free_space( | |
784 | struct btrfs_block_group_cache *block_group, | |
785 | struct btrfs_free_cluster *cluster) | |
786 | { | |
787 | struct btrfs_free_space *entry; | |
788 | struct rb_node *node; | |
96303081 | 789 | bool bitmap; |
fa9c0d79 CM |
790 | |
791 | spin_lock(&cluster->lock); | |
792 | if (cluster->block_group != block_group) | |
793 | goto out; | |
794 | ||
96303081 JB |
795 | bitmap = cluster->points_to_bitmap; |
796 | cluster->block_group = NULL; | |
fa9c0d79 | 797 | cluster->window_start = 0; |
96303081 JB |
798 | list_del_init(&cluster->block_group_list); |
799 | cluster->points_to_bitmap = false; | |
800 | ||
801 | if (bitmap) | |
802 | goto out; | |
803 | ||
fa9c0d79 | 804 | node = rb_first(&cluster->root); |
96303081 | 805 | while (node) { |
fa9c0d79 CM |
806 | entry = rb_entry(node, struct btrfs_free_space, offset_index); |
807 | node = rb_next(&entry->offset_index); | |
808 | rb_erase(&entry->offset_index, &cluster->root); | |
96303081 JB |
809 | BUG_ON(entry->bitmap); |
810 | tree_insert_offset(&block_group->free_space_offset, | |
811 | entry->offset, &entry->offset_index, 0); | |
fa9c0d79 | 812 | } |
fa9c0d79 | 813 | cluster->root.rb_node = NULL; |
96303081 | 814 | |
fa9c0d79 CM |
815 | out: |
816 | spin_unlock(&cluster->lock); | |
96303081 | 817 | btrfs_put_block_group(block_group); |
fa9c0d79 CM |
818 | return 0; |
819 | } | |
820 | ||
0f9dd46c JB |
821 | void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group) |
822 | { | |
823 | struct btrfs_free_space *info; | |
824 | struct rb_node *node; | |
fa9c0d79 | 825 | struct btrfs_free_cluster *cluster; |
96303081 | 826 | struct list_head *head; |
0f9dd46c | 827 | |
6226cb0a | 828 | spin_lock(&block_group->tree_lock); |
96303081 JB |
829 | while ((head = block_group->cluster_list.next) != |
830 | &block_group->cluster_list) { | |
831 | cluster = list_entry(head, struct btrfs_free_cluster, | |
832 | block_group_list); | |
fa9c0d79 CM |
833 | |
834 | WARN_ON(cluster->block_group != block_group); | |
835 | __btrfs_return_cluster_to_free_space(block_group, cluster); | |
96303081 JB |
836 | if (need_resched()) { |
837 | spin_unlock(&block_group->tree_lock); | |
838 | cond_resched(); | |
839 | spin_lock(&block_group->tree_lock); | |
840 | } | |
fa9c0d79 CM |
841 | } |
842 | ||
96303081 JB |
843 | while ((node = rb_last(&block_group->free_space_offset)) != NULL) { |
844 | info = rb_entry(node, struct btrfs_free_space, offset_index); | |
0f9dd46c | 845 | unlink_free_space(block_group, info); |
96303081 JB |
846 | if (info->bitmap) |
847 | kfree(info->bitmap); | |
0f9dd46c JB |
848 | kfree(info); |
849 | if (need_resched()) { | |
6226cb0a | 850 | spin_unlock(&block_group->tree_lock); |
0f9dd46c | 851 | cond_resched(); |
6226cb0a | 852 | spin_lock(&block_group->tree_lock); |
0f9dd46c JB |
853 | } |
854 | } | |
96303081 | 855 | |
6226cb0a | 856 | spin_unlock(&block_group->tree_lock); |
0f9dd46c JB |
857 | } |
858 | ||
6226cb0a JB |
859 | u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group, |
860 | u64 offset, u64 bytes, u64 empty_size) | |
0f9dd46c | 861 | { |
6226cb0a | 862 | struct btrfs_free_space *entry = NULL; |
96303081 | 863 | u64 bytes_search = bytes + empty_size; |
6226cb0a | 864 | u64 ret = 0; |
0f9dd46c | 865 | |
6226cb0a | 866 | spin_lock(&block_group->tree_lock); |
96303081 | 867 | entry = find_free_space(block_group, &offset, &bytes_search, 0); |
6226cb0a | 868 | if (!entry) |
96303081 JB |
869 | goto out; |
870 | ||
871 | ret = offset; | |
872 | if (entry->bitmap) { | |
873 | bitmap_clear_bits(entry, offset, bytes, | |
874 | block_group->sectorsize); | |
875 | if (!entry->bytes) { | |
876 | unlink_free_space(block_group, entry); | |
877 | kfree(entry->bitmap); | |
878 | kfree(entry); | |
879 | block_group->total_bitmaps--; | |
880 | recalculate_thresholds(block_group); | |
881 | } | |
882 | } else { | |
6226cb0a | 883 | unlink_free_space(block_group, entry); |
6226cb0a JB |
884 | entry->offset += bytes; |
885 | entry->bytes -= bytes; | |
6226cb0a JB |
886 | if (!entry->bytes) |
887 | kfree(entry); | |
888 | else | |
889 | link_free_space(block_group, entry); | |
890 | } | |
0f9dd46c | 891 | |
96303081 JB |
892 | out: |
893 | spin_unlock(&block_group->tree_lock); | |
0f9dd46c JB |
894 | return ret; |
895 | } | |
fa9c0d79 CM |
896 | |
897 | /* | |
898 | * given a cluster, put all of its extents back into the free space | |
899 | * cache. If a block group is passed, this function will only free | |
900 | * a cluster that belongs to the passed block group. | |
901 | * | |
902 | * Otherwise, it'll get a reference on the block group pointed to by the | |
903 | * cluster and remove the cluster from it. | |
904 | */ | |
905 | int btrfs_return_cluster_to_free_space( | |
906 | struct btrfs_block_group_cache *block_group, | |
907 | struct btrfs_free_cluster *cluster) | |
908 | { | |
909 | int ret; | |
910 | ||
911 | /* first, get a safe pointer to the block group */ | |
912 | spin_lock(&cluster->lock); | |
913 | if (!block_group) { | |
914 | block_group = cluster->block_group; | |
915 | if (!block_group) { | |
916 | spin_unlock(&cluster->lock); | |
917 | return 0; | |
918 | } | |
919 | } else if (cluster->block_group != block_group) { | |
920 | /* someone else has already freed it don't redo their work */ | |
921 | spin_unlock(&cluster->lock); | |
922 | return 0; | |
923 | } | |
924 | atomic_inc(&block_group->count); | |
925 | spin_unlock(&cluster->lock); | |
926 | ||
927 | /* now return any extents the cluster had on it */ | |
928 | spin_lock(&block_group->tree_lock); | |
929 | ret = __btrfs_return_cluster_to_free_space(block_group, cluster); | |
930 | spin_unlock(&block_group->tree_lock); | |
931 | ||
932 | /* finally drop our ref */ | |
933 | btrfs_put_block_group(block_group); | |
934 | return ret; | |
935 | } | |
936 | ||
96303081 JB |
937 | static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group, |
938 | struct btrfs_free_cluster *cluster, | |
939 | u64 bytes, u64 min_start) | |
940 | { | |
941 | struct btrfs_free_space *entry; | |
942 | int err; | |
943 | u64 search_start = cluster->window_start; | |
944 | u64 search_bytes = bytes; | |
945 | u64 ret = 0; | |
946 | ||
947 | spin_lock(&block_group->tree_lock); | |
948 | spin_lock(&cluster->lock); | |
949 | ||
950 | if (!cluster->points_to_bitmap) | |
951 | goto out; | |
952 | ||
953 | if (cluster->block_group != block_group) | |
954 | goto out; | |
955 | ||
956 | entry = tree_search_offset(block_group, search_start, 0, 0); | |
957 | ||
958 | if (!entry || !entry->bitmap) | |
959 | goto out; | |
960 | ||
961 | search_start = min_start; | |
962 | search_bytes = bytes; | |
963 | ||
964 | err = search_bitmap(block_group, entry, &search_start, | |
965 | &search_bytes); | |
966 | if (err) | |
967 | goto out; | |
968 | ||
969 | ret = search_start; | |
970 | bitmap_clear_bits(entry, ret, bytes, block_group->sectorsize); | |
971 | out: | |
972 | spin_unlock(&cluster->lock); | |
973 | spin_unlock(&block_group->tree_lock); | |
974 | ||
975 | return ret; | |
976 | } | |
977 | ||
fa9c0d79 CM |
978 | /* |
979 | * given a cluster, try to allocate 'bytes' from it, returns 0 | |
980 | * if it couldn't find anything suitably large, or a logical disk offset | |
981 | * if things worked out | |
982 | */ | |
983 | u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group, | |
984 | struct btrfs_free_cluster *cluster, u64 bytes, | |
985 | u64 min_start) | |
986 | { | |
987 | struct btrfs_free_space *entry = NULL; | |
988 | struct rb_node *node; | |
989 | u64 ret = 0; | |
990 | ||
96303081 JB |
991 | if (cluster->points_to_bitmap) |
992 | return btrfs_alloc_from_bitmap(block_group, cluster, bytes, | |
993 | min_start); | |
994 | ||
fa9c0d79 CM |
995 | spin_lock(&cluster->lock); |
996 | if (bytes > cluster->max_size) | |
997 | goto out; | |
998 | ||
999 | if (cluster->block_group != block_group) | |
1000 | goto out; | |
1001 | ||
1002 | node = rb_first(&cluster->root); | |
1003 | if (!node) | |
1004 | goto out; | |
1005 | ||
1006 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
1007 | ||
1008 | while(1) { | |
1009 | if (entry->bytes < bytes || entry->offset < min_start) { | |
1010 | struct rb_node *node; | |
1011 | ||
1012 | node = rb_next(&entry->offset_index); | |
1013 | if (!node) | |
1014 | break; | |
1015 | entry = rb_entry(node, struct btrfs_free_space, | |
1016 | offset_index); | |
1017 | continue; | |
1018 | } | |
1019 | ret = entry->offset; | |
1020 | ||
1021 | entry->offset += bytes; | |
1022 | entry->bytes -= bytes; | |
1023 | ||
1024 | if (entry->bytes == 0) { | |
1025 | rb_erase(&entry->offset_index, &cluster->root); | |
1026 | kfree(entry); | |
1027 | } | |
1028 | break; | |
1029 | } | |
1030 | out: | |
1031 | spin_unlock(&cluster->lock); | |
96303081 | 1032 | |
fa9c0d79 CM |
1033 | return ret; |
1034 | } | |
1035 | ||
96303081 JB |
1036 | static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group, |
1037 | struct btrfs_free_space *entry, | |
1038 | struct btrfs_free_cluster *cluster, | |
1039 | u64 offset, u64 bytes, u64 min_bytes) | |
1040 | { | |
1041 | unsigned long next_zero; | |
1042 | unsigned long i; | |
1043 | unsigned long search_bits; | |
1044 | unsigned long total_bits; | |
1045 | unsigned long found_bits; | |
1046 | unsigned long start = 0; | |
1047 | unsigned long total_found = 0; | |
1048 | bool found = false; | |
1049 | ||
1050 | i = offset_to_bit(entry->offset, block_group->sectorsize, | |
1051 | max_t(u64, offset, entry->offset)); | |
1052 | search_bits = bytes_to_bits(min_bytes, block_group->sectorsize); | |
1053 | total_bits = bytes_to_bits(bytes, block_group->sectorsize); | |
1054 | ||
1055 | again: | |
1056 | found_bits = 0; | |
1057 | for (i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, i); | |
1058 | i < BITS_PER_BITMAP; | |
1059 | i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, i + 1)) { | |
1060 | next_zero = find_next_zero_bit(entry->bitmap, | |
1061 | BITS_PER_BITMAP, i); | |
1062 | if (next_zero - i >= search_bits) { | |
1063 | found_bits = next_zero - i; | |
1064 | break; | |
1065 | } | |
1066 | i = next_zero; | |
1067 | } | |
1068 | ||
1069 | if (!found_bits) | |
1070 | return -1; | |
1071 | ||
1072 | if (!found) { | |
1073 | start = i; | |
1074 | found = true; | |
1075 | } | |
1076 | ||
1077 | total_found += found_bits; | |
1078 | ||
1079 | if (cluster->max_size < found_bits * block_group->sectorsize) | |
1080 | cluster->max_size = found_bits * block_group->sectorsize; | |
1081 | ||
1082 | if (total_found < total_bits) { | |
1083 | i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, next_zero); | |
1084 | if (i - start > total_bits * 2) { | |
1085 | total_found = 0; | |
1086 | cluster->max_size = 0; | |
1087 | found = false; | |
1088 | } | |
1089 | goto again; | |
1090 | } | |
1091 | ||
1092 | cluster->window_start = start * block_group->sectorsize + | |
1093 | entry->offset; | |
1094 | cluster->points_to_bitmap = true; | |
1095 | ||
1096 | return 0; | |
1097 | } | |
1098 | ||
fa9c0d79 CM |
1099 | /* |
1100 | * here we try to find a cluster of blocks in a block group. The goal | |
1101 | * is to find at least bytes free and up to empty_size + bytes free. | |
1102 | * We might not find them all in one contiguous area. | |
1103 | * | |
1104 | * returns zero and sets up cluster if things worked out, otherwise | |
1105 | * it returns -enospc | |
1106 | */ | |
1107 | int btrfs_find_space_cluster(struct btrfs_trans_handle *trans, | |
451d7585 | 1108 | struct btrfs_root *root, |
fa9c0d79 CM |
1109 | struct btrfs_block_group_cache *block_group, |
1110 | struct btrfs_free_cluster *cluster, | |
1111 | u64 offset, u64 bytes, u64 empty_size) | |
1112 | { | |
1113 | struct btrfs_free_space *entry = NULL; | |
1114 | struct rb_node *node; | |
1115 | struct btrfs_free_space *next; | |
96303081 | 1116 | struct btrfs_free_space *last = NULL; |
fa9c0d79 CM |
1117 | u64 min_bytes; |
1118 | u64 window_start; | |
1119 | u64 window_free; | |
1120 | u64 max_extent = 0; | |
96303081 | 1121 | bool found_bitmap = false; |
fa9c0d79 CM |
1122 | int ret; |
1123 | ||
1124 | /* for metadata, allow allocates with more holes */ | |
451d7585 CM |
1125 | if (btrfs_test_opt(root, SSD_SPREAD)) { |
1126 | min_bytes = bytes + empty_size; | |
1127 | } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) { | |
fa9c0d79 CM |
1128 | /* |
1129 | * we want to do larger allocations when we are | |
1130 | * flushing out the delayed refs, it helps prevent | |
1131 | * making more work as we go along. | |
1132 | */ | |
1133 | if (trans->transaction->delayed_refs.flushing) | |
1134 | min_bytes = max(bytes, (bytes + empty_size) >> 1); | |
1135 | else | |
1136 | min_bytes = max(bytes, (bytes + empty_size) >> 4); | |
1137 | } else | |
1138 | min_bytes = max(bytes, (bytes + empty_size) >> 2); | |
1139 | ||
1140 | spin_lock(&block_group->tree_lock); | |
1141 | spin_lock(&cluster->lock); | |
1142 | ||
1143 | /* someone already found a cluster, hooray */ | |
1144 | if (cluster->block_group) { | |
1145 | ret = 0; | |
1146 | goto out; | |
1147 | } | |
1148 | again: | |
96303081 | 1149 | entry = tree_search_offset(block_group, offset, found_bitmap, 1); |
fa9c0d79 CM |
1150 | if (!entry) { |
1151 | ret = -ENOSPC; | |
1152 | goto out; | |
1153 | } | |
96303081 JB |
1154 | |
1155 | /* | |
1156 | * If found_bitmap is true, we exhausted our search for extent entries, | |
1157 | * and we just want to search all of the bitmaps that we can find, and | |
1158 | * ignore any extent entries we find. | |
1159 | */ | |
1160 | while (entry->bitmap || found_bitmap || | |
1161 | (!entry->bitmap && entry->bytes < min_bytes)) { | |
1162 | struct rb_node *node = rb_next(&entry->offset_index); | |
1163 | ||
1164 | if (entry->bitmap && entry->bytes > bytes + empty_size) { | |
1165 | ret = btrfs_bitmap_cluster(block_group, entry, cluster, | |
1166 | offset, bytes + empty_size, | |
1167 | min_bytes); | |
1168 | if (!ret) | |
1169 | goto got_it; | |
1170 | } | |
1171 | ||
1172 | if (!node) { | |
1173 | ret = -ENOSPC; | |
1174 | goto out; | |
1175 | } | |
1176 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
1177 | } | |
1178 | ||
1179 | /* | |
1180 | * We already searched all the extent entries from the passed in offset | |
1181 | * to the end and didn't find enough space for the cluster, and we also | |
1182 | * didn't find any bitmaps that met our criteria, just go ahead and exit | |
1183 | */ | |
1184 | if (found_bitmap) { | |
1185 | ret = -ENOSPC; | |
1186 | goto out; | |
1187 | } | |
1188 | ||
1189 | cluster->points_to_bitmap = false; | |
fa9c0d79 CM |
1190 | window_start = entry->offset; |
1191 | window_free = entry->bytes; | |
1192 | last = entry; | |
1193 | max_extent = entry->bytes; | |
1194 | ||
96303081 | 1195 | while (1) { |
fa9c0d79 CM |
1196 | /* out window is just right, lets fill it */ |
1197 | if (window_free >= bytes + empty_size) | |
1198 | break; | |
1199 | ||
1200 | node = rb_next(&last->offset_index); | |
1201 | if (!node) { | |
96303081 JB |
1202 | if (found_bitmap) |
1203 | goto again; | |
fa9c0d79 CM |
1204 | ret = -ENOSPC; |
1205 | goto out; | |
1206 | } | |
1207 | next = rb_entry(node, struct btrfs_free_space, offset_index); | |
1208 | ||
96303081 JB |
1209 | /* |
1210 | * we found a bitmap, so if this search doesn't result in a | |
1211 | * cluster, we know to go and search again for the bitmaps and | |
1212 | * start looking for space there | |
1213 | */ | |
1214 | if (next->bitmap) { | |
1215 | if (!found_bitmap) | |
1216 | offset = next->offset; | |
1217 | found_bitmap = true; | |
1218 | last = next; | |
1219 | continue; | |
1220 | } | |
1221 | ||
fa9c0d79 CM |
1222 | /* |
1223 | * we haven't filled the empty size and the window is | |
1224 | * very large. reset and try again | |
1225 | */ | |
c6044801 CM |
1226 | if (next->offset - (last->offset + last->bytes) > 128 * 1024 || |
1227 | next->offset - window_start > (bytes + empty_size) * 2) { | |
fa9c0d79 CM |
1228 | entry = next; |
1229 | window_start = entry->offset; | |
1230 | window_free = entry->bytes; | |
1231 | last = entry; | |
1232 | max_extent = 0; | |
fa9c0d79 CM |
1233 | } else { |
1234 | last = next; | |
1235 | window_free += next->bytes; | |
1236 | if (entry->bytes > max_extent) | |
1237 | max_extent = entry->bytes; | |
1238 | } | |
1239 | } | |
1240 | ||
1241 | cluster->window_start = entry->offset; | |
1242 | ||
1243 | /* | |
1244 | * now we've found our entries, pull them out of the free space | |
1245 | * cache and put them into the cluster rbtree | |
1246 | * | |
1247 | * The cluster includes an rbtree, but only uses the offset index | |
1248 | * of each free space cache entry. | |
1249 | */ | |
96303081 | 1250 | while (1) { |
fa9c0d79 | 1251 | node = rb_next(&entry->offset_index); |
96303081 JB |
1252 | if (entry->bitmap && node) { |
1253 | entry = rb_entry(node, struct btrfs_free_space, | |
1254 | offset_index); | |
1255 | continue; | |
1256 | } else if (entry->bitmap && !node) { | |
1257 | break; | |
1258 | } | |
1259 | ||
1260 | rb_erase(&entry->offset_index, &block_group->free_space_offset); | |
fa9c0d79 | 1261 | ret = tree_insert_offset(&cluster->root, entry->offset, |
96303081 | 1262 | &entry->offset_index, 0); |
fa9c0d79 CM |
1263 | BUG_ON(ret); |
1264 | ||
1265 | if (!node || entry == last) | |
1266 | break; | |
1267 | ||
1268 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
1269 | } | |
96303081 | 1270 | |
fa9c0d79 | 1271 | cluster->max_size = max_extent; |
96303081 JB |
1272 | got_it: |
1273 | ret = 0; | |
fa9c0d79 CM |
1274 | atomic_inc(&block_group->count); |
1275 | list_add_tail(&cluster->block_group_list, &block_group->cluster_list); | |
1276 | cluster->block_group = block_group; | |
1277 | out: | |
1278 | spin_unlock(&cluster->lock); | |
1279 | spin_unlock(&block_group->tree_lock); | |
1280 | ||
1281 | return ret; | |
1282 | } | |
1283 | ||
1284 | /* | |
1285 | * simple code to zero out a cluster | |
1286 | */ | |
1287 | void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster) | |
1288 | { | |
1289 | spin_lock_init(&cluster->lock); | |
1290 | spin_lock_init(&cluster->refill_lock); | |
1291 | cluster->root.rb_node = NULL; | |
1292 | cluster->max_size = 0; | |
96303081 | 1293 | cluster->points_to_bitmap = false; |
fa9c0d79 CM |
1294 | INIT_LIST_HEAD(&cluster->block_group_list); |
1295 | cluster->block_group = NULL; | |
1296 | } | |
1297 |