]>
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> |
5a0e3ad6 | 21 | #include <linux/slab.h> |
96303081 | 22 | #include <linux/math64.h> |
0f9dd46c | 23 | #include "ctree.h" |
fa9c0d79 CM |
24 | #include "free-space-cache.h" |
25 | #include "transaction.h" | |
0af3d00b | 26 | #include "disk-io.h" |
fa9c0d79 | 27 | |
96303081 JB |
28 | #define BITS_PER_BITMAP (PAGE_CACHE_SIZE * 8) |
29 | #define MAX_CACHE_BYTES_PER_GIG (32 * 1024) | |
0f9dd46c | 30 | |
0cb59c99 JB |
31 | static void recalculate_thresholds(struct btrfs_block_group_cache |
32 | *block_group); | |
33 | static int link_free_space(struct btrfs_block_group_cache *block_group, | |
34 | struct btrfs_free_space *info); | |
35 | ||
0af3d00b JB |
36 | struct inode *lookup_free_space_inode(struct btrfs_root *root, |
37 | struct btrfs_block_group_cache | |
38 | *block_group, struct btrfs_path *path) | |
39 | { | |
40 | struct btrfs_key key; | |
41 | struct btrfs_key location; | |
42 | struct btrfs_disk_key disk_key; | |
43 | struct btrfs_free_space_header *header; | |
44 | struct extent_buffer *leaf; | |
45 | struct inode *inode = NULL; | |
46 | int ret; | |
47 | ||
48 | spin_lock(&block_group->lock); | |
49 | if (block_group->inode) | |
50 | inode = igrab(block_group->inode); | |
51 | spin_unlock(&block_group->lock); | |
52 | if (inode) | |
53 | return inode; | |
54 | ||
55 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
56 | key.offset = block_group->key.objectid; | |
57 | key.type = 0; | |
58 | ||
59 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
60 | if (ret < 0) | |
61 | return ERR_PTR(ret); | |
62 | if (ret > 0) { | |
63 | btrfs_release_path(root, path); | |
64 | return ERR_PTR(-ENOENT); | |
65 | } | |
66 | ||
67 | leaf = path->nodes[0]; | |
68 | header = btrfs_item_ptr(leaf, path->slots[0], | |
69 | struct btrfs_free_space_header); | |
70 | btrfs_free_space_key(leaf, header, &disk_key); | |
71 | btrfs_disk_key_to_cpu(&location, &disk_key); | |
72 | btrfs_release_path(root, path); | |
73 | ||
74 | inode = btrfs_iget(root->fs_info->sb, &location, root, NULL); | |
75 | if (!inode) | |
76 | return ERR_PTR(-ENOENT); | |
77 | if (IS_ERR(inode)) | |
78 | return inode; | |
79 | if (is_bad_inode(inode)) { | |
80 | iput(inode); | |
81 | return ERR_PTR(-ENOENT); | |
82 | } | |
83 | ||
84 | spin_lock(&block_group->lock); | |
85 | if (!root->fs_info->closing) { | |
86 | block_group->inode = igrab(inode); | |
87 | block_group->iref = 1; | |
88 | } | |
89 | spin_unlock(&block_group->lock); | |
90 | ||
91 | return inode; | |
92 | } | |
93 | ||
94 | int create_free_space_inode(struct btrfs_root *root, | |
95 | struct btrfs_trans_handle *trans, | |
96 | struct btrfs_block_group_cache *block_group, | |
97 | struct btrfs_path *path) | |
98 | { | |
99 | struct btrfs_key key; | |
100 | struct btrfs_disk_key disk_key; | |
101 | struct btrfs_free_space_header *header; | |
102 | struct btrfs_inode_item *inode_item; | |
103 | struct extent_buffer *leaf; | |
104 | u64 objectid; | |
105 | int ret; | |
106 | ||
107 | ret = btrfs_find_free_objectid(trans, root, 0, &objectid); | |
108 | if (ret < 0) | |
109 | return ret; | |
110 | ||
111 | ret = btrfs_insert_empty_inode(trans, root, path, objectid); | |
112 | if (ret) | |
113 | return ret; | |
114 | ||
115 | leaf = path->nodes[0]; | |
116 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
117 | struct btrfs_inode_item); | |
118 | btrfs_item_key(leaf, &disk_key, path->slots[0]); | |
119 | memset_extent_buffer(leaf, 0, (unsigned long)inode_item, | |
120 | sizeof(*inode_item)); | |
121 | btrfs_set_inode_generation(leaf, inode_item, trans->transid); | |
122 | btrfs_set_inode_size(leaf, inode_item, 0); | |
123 | btrfs_set_inode_nbytes(leaf, inode_item, 0); | |
124 | btrfs_set_inode_uid(leaf, inode_item, 0); | |
125 | btrfs_set_inode_gid(leaf, inode_item, 0); | |
126 | btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600); | |
127 | btrfs_set_inode_flags(leaf, inode_item, BTRFS_INODE_NOCOMPRESS | | |
128 | BTRFS_INODE_PREALLOC | BTRFS_INODE_NODATASUM); | |
129 | btrfs_set_inode_nlink(leaf, inode_item, 1); | |
130 | btrfs_set_inode_transid(leaf, inode_item, trans->transid); | |
131 | btrfs_set_inode_block_group(leaf, inode_item, | |
132 | block_group->key.objectid); | |
133 | btrfs_mark_buffer_dirty(leaf); | |
134 | btrfs_release_path(root, path); | |
135 | ||
136 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
137 | key.offset = block_group->key.objectid; | |
138 | key.type = 0; | |
139 | ||
140 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
141 | sizeof(struct btrfs_free_space_header)); | |
142 | if (ret < 0) { | |
143 | btrfs_release_path(root, path); | |
144 | return ret; | |
145 | } | |
146 | leaf = path->nodes[0]; | |
147 | header = btrfs_item_ptr(leaf, path->slots[0], | |
148 | struct btrfs_free_space_header); | |
149 | memset_extent_buffer(leaf, 0, (unsigned long)header, sizeof(*header)); | |
150 | btrfs_set_free_space_key(leaf, header, &disk_key); | |
151 | btrfs_mark_buffer_dirty(leaf); | |
152 | btrfs_release_path(root, path); | |
153 | ||
154 | return 0; | |
155 | } | |
156 | ||
157 | int btrfs_truncate_free_space_cache(struct btrfs_root *root, | |
158 | struct btrfs_trans_handle *trans, | |
159 | struct btrfs_path *path, | |
160 | struct inode *inode) | |
161 | { | |
162 | loff_t oldsize; | |
163 | int ret = 0; | |
164 | ||
165 | trans->block_rsv = root->orphan_block_rsv; | |
166 | ret = btrfs_block_rsv_check(trans, root, | |
167 | root->orphan_block_rsv, | |
168 | 0, 5); | |
169 | if (ret) | |
170 | return ret; | |
171 | ||
172 | oldsize = i_size_read(inode); | |
173 | btrfs_i_size_write(inode, 0); | |
174 | truncate_pagecache(inode, oldsize, 0); | |
175 | ||
176 | /* | |
177 | * We don't need an orphan item because truncating the free space cache | |
178 | * will never be split across transactions. | |
179 | */ | |
180 | ret = btrfs_truncate_inode_items(trans, root, inode, | |
181 | 0, BTRFS_EXTENT_DATA_KEY); | |
182 | if (ret) { | |
183 | WARN_ON(1); | |
184 | return ret; | |
185 | } | |
186 | ||
187 | return btrfs_update_inode(trans, root, inode); | |
188 | } | |
189 | ||
9d66e233 JB |
190 | static int readahead_cache(struct inode *inode) |
191 | { | |
192 | struct file_ra_state *ra; | |
193 | unsigned long last_index; | |
194 | ||
195 | ra = kzalloc(sizeof(*ra), GFP_NOFS); | |
196 | if (!ra) | |
197 | return -ENOMEM; | |
198 | ||
199 | file_ra_state_init(ra, inode->i_mapping); | |
200 | last_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT; | |
201 | ||
202 | page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index); | |
203 | ||
204 | kfree(ra); | |
205 | ||
206 | return 0; | |
207 | } | |
208 | ||
209 | int load_free_space_cache(struct btrfs_fs_info *fs_info, | |
210 | struct btrfs_block_group_cache *block_group) | |
211 | { | |
212 | struct btrfs_root *root = fs_info->tree_root; | |
213 | struct inode *inode; | |
214 | struct btrfs_free_space_header *header; | |
215 | struct extent_buffer *leaf; | |
216 | struct page *page; | |
217 | struct btrfs_path *path; | |
218 | u32 *checksums = NULL, *crc; | |
219 | char *disk_crcs = NULL; | |
220 | struct btrfs_key key; | |
221 | struct list_head bitmaps; | |
222 | u64 num_entries; | |
223 | u64 num_bitmaps; | |
224 | u64 generation; | |
225 | u32 cur_crc = ~(u32)0; | |
226 | pgoff_t index = 0; | |
227 | unsigned long first_page_offset; | |
228 | int num_checksums; | |
229 | int ret = 0; | |
230 | ||
231 | /* | |
232 | * If we're unmounting then just return, since this does a search on the | |
233 | * normal root and not the commit root and we could deadlock. | |
234 | */ | |
235 | smp_mb(); | |
236 | if (fs_info->closing) | |
237 | return 0; | |
238 | ||
239 | /* | |
240 | * If this block group has been marked to be cleared for one reason or | |
241 | * another then we can't trust the on disk cache, so just return. | |
242 | */ | |
243 | spin_lock(&block_group->lock); | |
244 | if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) { | |
9d66e233 JB |
245 | spin_unlock(&block_group->lock); |
246 | return 0; | |
247 | } | |
248 | spin_unlock(&block_group->lock); | |
249 | ||
250 | INIT_LIST_HEAD(&bitmaps); | |
251 | ||
252 | path = btrfs_alloc_path(); | |
253 | if (!path) | |
254 | return 0; | |
255 | ||
256 | inode = lookup_free_space_inode(root, block_group, path); | |
257 | if (IS_ERR(inode)) { | |
258 | btrfs_free_path(path); | |
259 | return 0; | |
260 | } | |
261 | ||
262 | /* Nothing in the space cache, goodbye */ | |
263 | if (!i_size_read(inode)) { | |
264 | btrfs_free_path(path); | |
265 | goto out; | |
266 | } | |
267 | ||
268 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
269 | key.offset = block_group->key.objectid; | |
270 | key.type = 0; | |
271 | ||
272 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
273 | if (ret) { | |
274 | btrfs_free_path(path); | |
275 | goto out; | |
276 | } | |
277 | ||
278 | leaf = path->nodes[0]; | |
279 | header = btrfs_item_ptr(leaf, path->slots[0], | |
280 | struct btrfs_free_space_header); | |
281 | num_entries = btrfs_free_space_entries(leaf, header); | |
282 | num_bitmaps = btrfs_free_space_bitmaps(leaf, header); | |
283 | generation = btrfs_free_space_generation(leaf, header); | |
284 | btrfs_free_path(path); | |
285 | ||
286 | if (BTRFS_I(inode)->generation != generation) { | |
287 | printk(KERN_ERR "btrfs: free space inode generation (%llu) did" | |
288 | " not match free space cache generation (%llu) for " | |
289 | "block group %llu\n", | |
290 | (unsigned long long)BTRFS_I(inode)->generation, | |
291 | (unsigned long long)generation, | |
292 | (unsigned long long)block_group->key.objectid); | |
2b20982e | 293 | goto free_cache; |
9d66e233 JB |
294 | } |
295 | ||
296 | if (!num_entries) | |
297 | goto out; | |
298 | ||
299 | /* Setup everything for doing checksumming */ | |
300 | num_checksums = i_size_read(inode) / PAGE_CACHE_SIZE; | |
301 | checksums = crc = kzalloc(sizeof(u32) * num_checksums, GFP_NOFS); | |
302 | if (!checksums) | |
303 | goto out; | |
304 | first_page_offset = (sizeof(u32) * num_checksums) + sizeof(u64); | |
305 | disk_crcs = kzalloc(first_page_offset, GFP_NOFS); | |
306 | if (!disk_crcs) | |
307 | goto out; | |
308 | ||
309 | ret = readahead_cache(inode); | |
310 | if (ret) { | |
311 | ret = 0; | |
312 | goto out; | |
313 | } | |
314 | ||
315 | while (1) { | |
316 | struct btrfs_free_space_entry *entry; | |
317 | struct btrfs_free_space *e; | |
318 | void *addr; | |
319 | unsigned long offset = 0; | |
320 | unsigned long start_offset = 0; | |
321 | int need_loop = 0; | |
322 | ||
323 | if (!num_entries && !num_bitmaps) | |
324 | break; | |
325 | ||
326 | if (index == 0) { | |
327 | start_offset = first_page_offset; | |
328 | offset = start_offset; | |
329 | } | |
330 | ||
331 | page = grab_cache_page(inode->i_mapping, index); | |
332 | if (!page) { | |
333 | ret = 0; | |
334 | goto free_cache; | |
335 | } | |
336 | ||
337 | if (!PageUptodate(page)) { | |
338 | btrfs_readpage(NULL, page); | |
339 | lock_page(page); | |
340 | if (!PageUptodate(page)) { | |
341 | unlock_page(page); | |
342 | page_cache_release(page); | |
343 | printk(KERN_ERR "btrfs: error reading free " | |
344 | "space cache: %llu\n", | |
345 | (unsigned long long) | |
346 | block_group->key.objectid); | |
347 | goto free_cache; | |
348 | } | |
349 | } | |
350 | addr = kmap(page); | |
351 | ||
352 | if (index == 0) { | |
353 | u64 *gen; | |
354 | ||
355 | memcpy(disk_crcs, addr, first_page_offset); | |
356 | gen = addr + (sizeof(u32) * num_checksums); | |
357 | if (*gen != BTRFS_I(inode)->generation) { | |
358 | printk(KERN_ERR "btrfs: space cache generation" | |
359 | " (%llu) does not match inode (%llu) " | |
360 | "for block group %llu\n", | |
361 | (unsigned long long)*gen, | |
362 | (unsigned long long) | |
363 | BTRFS_I(inode)->generation, | |
364 | (unsigned long long) | |
365 | block_group->key.objectid); | |
366 | kunmap(page); | |
367 | unlock_page(page); | |
368 | page_cache_release(page); | |
369 | goto free_cache; | |
370 | } | |
371 | crc = (u32 *)disk_crcs; | |
372 | } | |
373 | entry = addr + start_offset; | |
374 | ||
375 | /* First lets check our crc before we do anything fun */ | |
376 | cur_crc = ~(u32)0; | |
377 | cur_crc = btrfs_csum_data(root, addr + start_offset, cur_crc, | |
378 | PAGE_CACHE_SIZE - start_offset); | |
379 | btrfs_csum_final(cur_crc, (char *)&cur_crc); | |
380 | if (cur_crc != *crc) { | |
381 | printk(KERN_ERR "btrfs: crc mismatch for page %lu in " | |
382 | "block group %llu\n", index, | |
383 | (unsigned long long)block_group->key.objectid); | |
384 | kunmap(page); | |
385 | unlock_page(page); | |
386 | page_cache_release(page); | |
387 | goto free_cache; | |
388 | } | |
389 | crc++; | |
390 | ||
391 | while (1) { | |
392 | if (!num_entries) | |
393 | break; | |
394 | ||
395 | need_loop = 1; | |
dc89e982 JB |
396 | e = kmem_cache_zalloc(btrfs_free_space_cachep, |
397 | GFP_NOFS); | |
9d66e233 JB |
398 | if (!e) { |
399 | kunmap(page); | |
400 | unlock_page(page); | |
401 | page_cache_release(page); | |
402 | goto free_cache; | |
403 | } | |
404 | ||
405 | e->offset = le64_to_cpu(entry->offset); | |
406 | e->bytes = le64_to_cpu(entry->bytes); | |
407 | if (!e->bytes) { | |
408 | kunmap(page); | |
dc89e982 | 409 | kmem_cache_free(btrfs_free_space_cachep, e); |
9d66e233 JB |
410 | unlock_page(page); |
411 | page_cache_release(page); | |
412 | goto free_cache; | |
413 | } | |
414 | ||
415 | if (entry->type == BTRFS_FREE_SPACE_EXTENT) { | |
416 | spin_lock(&block_group->tree_lock); | |
417 | ret = link_free_space(block_group, e); | |
418 | spin_unlock(&block_group->tree_lock); | |
419 | BUG_ON(ret); | |
420 | } else { | |
421 | e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS); | |
422 | if (!e->bitmap) { | |
423 | kunmap(page); | |
dc89e982 JB |
424 | kmem_cache_free( |
425 | btrfs_free_space_cachep, e); | |
9d66e233 JB |
426 | unlock_page(page); |
427 | page_cache_release(page); | |
428 | goto free_cache; | |
429 | } | |
430 | spin_lock(&block_group->tree_lock); | |
431 | ret = link_free_space(block_group, e); | |
432 | block_group->total_bitmaps++; | |
433 | recalculate_thresholds(block_group); | |
434 | spin_unlock(&block_group->tree_lock); | |
435 | list_add_tail(&e->list, &bitmaps); | |
436 | } | |
437 | ||
438 | num_entries--; | |
439 | offset += sizeof(struct btrfs_free_space_entry); | |
440 | if (offset + sizeof(struct btrfs_free_space_entry) >= | |
441 | PAGE_CACHE_SIZE) | |
442 | break; | |
443 | entry++; | |
444 | } | |
445 | ||
446 | /* | |
447 | * We read an entry out of this page, we need to move on to the | |
448 | * next page. | |
449 | */ | |
450 | if (need_loop) { | |
451 | kunmap(page); | |
452 | goto next; | |
453 | } | |
454 | ||
455 | /* | |
456 | * We add the bitmaps at the end of the entries in order that | |
457 | * the bitmap entries are added to the cache. | |
458 | */ | |
459 | e = list_entry(bitmaps.next, struct btrfs_free_space, list); | |
460 | list_del_init(&e->list); | |
461 | memcpy(e->bitmap, addr, PAGE_CACHE_SIZE); | |
462 | kunmap(page); | |
463 | num_bitmaps--; | |
464 | next: | |
465 | unlock_page(page); | |
466 | page_cache_release(page); | |
467 | index++; | |
468 | } | |
469 | ||
470 | ret = 1; | |
471 | out: | |
472 | kfree(checksums); | |
473 | kfree(disk_crcs); | |
474 | iput(inode); | |
475 | return ret; | |
476 | ||
477 | free_cache: | |
478 | /* This cache is bogus, make sure it gets cleared */ | |
479 | spin_lock(&block_group->lock); | |
480 | block_group->disk_cache_state = BTRFS_DC_CLEAR; | |
481 | spin_unlock(&block_group->lock); | |
482 | btrfs_remove_free_space_cache(block_group); | |
483 | goto out; | |
484 | } | |
485 | ||
0cb59c99 JB |
486 | int btrfs_write_out_cache(struct btrfs_root *root, |
487 | struct btrfs_trans_handle *trans, | |
488 | struct btrfs_block_group_cache *block_group, | |
489 | struct btrfs_path *path) | |
490 | { | |
491 | struct btrfs_free_space_header *header; | |
492 | struct extent_buffer *leaf; | |
493 | struct inode *inode; | |
494 | struct rb_node *node; | |
495 | struct list_head *pos, *n; | |
496 | struct page *page; | |
497 | struct extent_state *cached_state = NULL; | |
498 | struct list_head bitmap_list; | |
499 | struct btrfs_key key; | |
500 | u64 bytes = 0; | |
501 | u32 *crc, *checksums; | |
502 | pgoff_t index = 0, last_index = 0; | |
503 | unsigned long first_page_offset; | |
504 | int num_checksums; | |
505 | int entries = 0; | |
506 | int bitmaps = 0; | |
507 | int ret = 0; | |
508 | ||
509 | root = root->fs_info->tree_root; | |
510 | ||
511 | INIT_LIST_HEAD(&bitmap_list); | |
512 | ||
513 | spin_lock(&block_group->lock); | |
514 | if (block_group->disk_cache_state < BTRFS_DC_SETUP) { | |
515 | spin_unlock(&block_group->lock); | |
516 | return 0; | |
517 | } | |
518 | spin_unlock(&block_group->lock); | |
519 | ||
520 | inode = lookup_free_space_inode(root, block_group, path); | |
521 | if (IS_ERR(inode)) | |
522 | return 0; | |
523 | ||
524 | if (!i_size_read(inode)) { | |
525 | iput(inode); | |
526 | return 0; | |
527 | } | |
528 | ||
2b20982e JB |
529 | node = rb_first(&block_group->free_space_offset); |
530 | if (!node) { | |
531 | iput(inode); | |
532 | return 0; | |
533 | } | |
534 | ||
0cb59c99 JB |
535 | last_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT; |
536 | filemap_write_and_wait(inode->i_mapping); | |
537 | btrfs_wait_ordered_range(inode, inode->i_size & | |
538 | ~(root->sectorsize - 1), (u64)-1); | |
539 | ||
540 | /* We need a checksum per page. */ | |
541 | num_checksums = i_size_read(inode) / PAGE_CACHE_SIZE; | |
542 | crc = checksums = kzalloc(sizeof(u32) * num_checksums, GFP_NOFS); | |
543 | if (!crc) { | |
544 | iput(inode); | |
545 | return 0; | |
546 | } | |
547 | ||
548 | /* Since the first page has all of our checksums and our generation we | |
549 | * need to calculate the offset into the page that we can start writing | |
550 | * our entries. | |
551 | */ | |
552 | first_page_offset = (sizeof(u32) * num_checksums) + sizeof(u64); | |
553 | ||
0cb59c99 JB |
554 | /* |
555 | * Lock all pages first so we can lock the extent safely. | |
556 | * | |
557 | * NOTE: Because we hold the ref the entire time we're going to write to | |
558 | * the page find_get_page should never fail, so we don't do a check | |
559 | * after find_get_page at this point. Just putting this here so people | |
560 | * know and don't freak out. | |
561 | */ | |
562 | while (index <= last_index) { | |
563 | page = grab_cache_page(inode->i_mapping, index); | |
564 | if (!page) { | |
565 | pgoff_t i = 0; | |
566 | ||
567 | while (i < index) { | |
568 | page = find_get_page(inode->i_mapping, i); | |
569 | unlock_page(page); | |
570 | page_cache_release(page); | |
571 | page_cache_release(page); | |
572 | i++; | |
573 | } | |
574 | goto out_free; | |
575 | } | |
576 | index++; | |
577 | } | |
578 | ||
579 | index = 0; | |
580 | lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1, | |
581 | 0, &cached_state, GFP_NOFS); | |
582 | ||
583 | /* Write out the extent entries */ | |
584 | do { | |
585 | struct btrfs_free_space_entry *entry; | |
586 | void *addr; | |
587 | unsigned long offset = 0; | |
588 | unsigned long start_offset = 0; | |
589 | ||
590 | if (index == 0) { | |
591 | start_offset = first_page_offset; | |
592 | offset = start_offset; | |
593 | } | |
594 | ||
595 | page = find_get_page(inode->i_mapping, index); | |
596 | ||
597 | addr = kmap(page); | |
598 | entry = addr + start_offset; | |
599 | ||
600 | memset(addr, 0, PAGE_CACHE_SIZE); | |
601 | while (1) { | |
602 | struct btrfs_free_space *e; | |
603 | ||
604 | e = rb_entry(node, struct btrfs_free_space, offset_index); | |
605 | entries++; | |
606 | ||
607 | entry->offset = cpu_to_le64(e->offset); | |
608 | entry->bytes = cpu_to_le64(e->bytes); | |
609 | if (e->bitmap) { | |
610 | entry->type = BTRFS_FREE_SPACE_BITMAP; | |
611 | list_add_tail(&e->list, &bitmap_list); | |
612 | bitmaps++; | |
613 | } else { | |
614 | entry->type = BTRFS_FREE_SPACE_EXTENT; | |
615 | } | |
616 | node = rb_next(node); | |
617 | if (!node) | |
618 | break; | |
619 | offset += sizeof(struct btrfs_free_space_entry); | |
620 | if (offset + sizeof(struct btrfs_free_space_entry) >= | |
621 | PAGE_CACHE_SIZE) | |
622 | break; | |
623 | entry++; | |
624 | } | |
625 | *crc = ~(u32)0; | |
626 | *crc = btrfs_csum_data(root, addr + start_offset, *crc, | |
627 | PAGE_CACHE_SIZE - start_offset); | |
628 | kunmap(page); | |
629 | ||
630 | btrfs_csum_final(*crc, (char *)crc); | |
631 | crc++; | |
632 | ||
633 | bytes += PAGE_CACHE_SIZE; | |
634 | ||
635 | ClearPageChecked(page); | |
636 | set_page_extent_mapped(page); | |
637 | SetPageUptodate(page); | |
638 | set_page_dirty(page); | |
639 | ||
640 | /* | |
641 | * We need to release our reference we got for grab_cache_page, | |
642 | * except for the first page which will hold our checksums, we | |
643 | * do that below. | |
644 | */ | |
645 | if (index != 0) { | |
646 | unlock_page(page); | |
647 | page_cache_release(page); | |
648 | } | |
649 | ||
650 | page_cache_release(page); | |
651 | ||
652 | index++; | |
653 | } while (node); | |
654 | ||
655 | /* Write out the bitmaps */ | |
656 | list_for_each_safe(pos, n, &bitmap_list) { | |
657 | void *addr; | |
658 | struct btrfs_free_space *entry = | |
659 | list_entry(pos, struct btrfs_free_space, list); | |
660 | ||
661 | page = find_get_page(inode->i_mapping, index); | |
662 | ||
663 | addr = kmap(page); | |
664 | memcpy(addr, entry->bitmap, PAGE_CACHE_SIZE); | |
665 | *crc = ~(u32)0; | |
666 | *crc = btrfs_csum_data(root, addr, *crc, PAGE_CACHE_SIZE); | |
667 | kunmap(page); | |
668 | btrfs_csum_final(*crc, (char *)crc); | |
669 | crc++; | |
670 | bytes += PAGE_CACHE_SIZE; | |
671 | ||
672 | ClearPageChecked(page); | |
673 | set_page_extent_mapped(page); | |
674 | SetPageUptodate(page); | |
675 | set_page_dirty(page); | |
676 | unlock_page(page); | |
677 | page_cache_release(page); | |
678 | page_cache_release(page); | |
679 | list_del_init(&entry->list); | |
680 | index++; | |
681 | } | |
682 | ||
683 | /* Zero out the rest of the pages just to make sure */ | |
684 | while (index <= last_index) { | |
685 | void *addr; | |
686 | ||
687 | page = find_get_page(inode->i_mapping, index); | |
688 | ||
689 | addr = kmap(page); | |
690 | memset(addr, 0, PAGE_CACHE_SIZE); | |
691 | kunmap(page); | |
692 | ClearPageChecked(page); | |
693 | set_page_extent_mapped(page); | |
694 | SetPageUptodate(page); | |
695 | set_page_dirty(page); | |
696 | unlock_page(page); | |
697 | page_cache_release(page); | |
698 | page_cache_release(page); | |
699 | bytes += PAGE_CACHE_SIZE; | |
700 | index++; | |
701 | } | |
702 | ||
703 | btrfs_set_extent_delalloc(inode, 0, bytes - 1, &cached_state); | |
704 | ||
705 | /* Write the checksums and trans id to the first page */ | |
706 | { | |
707 | void *addr; | |
708 | u64 *gen; | |
709 | ||
710 | page = find_get_page(inode->i_mapping, 0); | |
711 | ||
712 | addr = kmap(page); | |
713 | memcpy(addr, checksums, sizeof(u32) * num_checksums); | |
714 | gen = addr + (sizeof(u32) * num_checksums); | |
715 | *gen = trans->transid; | |
716 | kunmap(page); | |
717 | ClearPageChecked(page); | |
718 | set_page_extent_mapped(page); | |
719 | SetPageUptodate(page); | |
720 | set_page_dirty(page); | |
721 | unlock_page(page); | |
722 | page_cache_release(page); | |
723 | page_cache_release(page); | |
724 | } | |
725 | BTRFS_I(inode)->generation = trans->transid; | |
726 | ||
727 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0, | |
728 | i_size_read(inode) - 1, &cached_state, GFP_NOFS); | |
729 | ||
730 | filemap_write_and_wait(inode->i_mapping); | |
731 | ||
732 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
733 | key.offset = block_group->key.objectid; | |
734 | key.type = 0; | |
735 | ||
736 | ret = btrfs_search_slot(trans, root, &key, path, 1, 1); | |
737 | if (ret < 0) { | |
738 | ret = 0; | |
739 | clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, bytes - 1, | |
740 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
741 | EXTENT_DO_ACCOUNTING, 0, 0, NULL, GFP_NOFS); | |
742 | goto out_free; | |
743 | } | |
744 | leaf = path->nodes[0]; | |
745 | if (ret > 0) { | |
746 | struct btrfs_key found_key; | |
747 | BUG_ON(!path->slots[0]); | |
748 | path->slots[0]--; | |
749 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
750 | if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID || | |
751 | found_key.offset != block_group->key.objectid) { | |
752 | ret = 0; | |
753 | clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, bytes - 1, | |
754 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
755 | EXTENT_DO_ACCOUNTING, 0, 0, NULL, | |
756 | GFP_NOFS); | |
757 | btrfs_release_path(root, path); | |
758 | goto out_free; | |
759 | } | |
760 | } | |
761 | header = btrfs_item_ptr(leaf, path->slots[0], | |
762 | struct btrfs_free_space_header); | |
763 | btrfs_set_free_space_entries(leaf, header, entries); | |
764 | btrfs_set_free_space_bitmaps(leaf, header, bitmaps); | |
765 | btrfs_set_free_space_generation(leaf, header, trans->transid); | |
766 | btrfs_mark_buffer_dirty(leaf); | |
767 | btrfs_release_path(root, path); | |
768 | ||
769 | ret = 1; | |
770 | ||
771 | out_free: | |
772 | if (ret == 0) { | |
773 | invalidate_inode_pages2_range(inode->i_mapping, 0, index); | |
774 | spin_lock(&block_group->lock); | |
775 | block_group->disk_cache_state = BTRFS_DC_ERROR; | |
776 | spin_unlock(&block_group->lock); | |
777 | BTRFS_I(inode)->generation = 0; | |
778 | } | |
779 | kfree(checksums); | |
780 | btrfs_update_inode(trans, root, inode); | |
781 | iput(inode); | |
782 | return ret; | |
783 | } | |
784 | ||
96303081 JB |
785 | static inline unsigned long offset_to_bit(u64 bitmap_start, u64 sectorsize, |
786 | u64 offset) | |
0f9dd46c | 787 | { |
96303081 JB |
788 | BUG_ON(offset < bitmap_start); |
789 | offset -= bitmap_start; | |
790 | return (unsigned long)(div64_u64(offset, sectorsize)); | |
791 | } | |
0f9dd46c | 792 | |
96303081 JB |
793 | static inline unsigned long bytes_to_bits(u64 bytes, u64 sectorsize) |
794 | { | |
795 | return (unsigned long)(div64_u64(bytes, sectorsize)); | |
796 | } | |
0f9dd46c | 797 | |
96303081 JB |
798 | static inline u64 offset_to_bitmap(struct btrfs_block_group_cache *block_group, |
799 | u64 offset) | |
800 | { | |
801 | u64 bitmap_start; | |
802 | u64 bytes_per_bitmap; | |
0f9dd46c | 803 | |
96303081 JB |
804 | bytes_per_bitmap = BITS_PER_BITMAP * block_group->sectorsize; |
805 | bitmap_start = offset - block_group->key.objectid; | |
806 | bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap); | |
807 | bitmap_start *= bytes_per_bitmap; | |
808 | bitmap_start += block_group->key.objectid; | |
0f9dd46c | 809 | |
96303081 | 810 | return bitmap_start; |
0f9dd46c JB |
811 | } |
812 | ||
96303081 JB |
813 | static int tree_insert_offset(struct rb_root *root, u64 offset, |
814 | struct rb_node *node, int bitmap) | |
0f9dd46c JB |
815 | { |
816 | struct rb_node **p = &root->rb_node; | |
817 | struct rb_node *parent = NULL; | |
818 | struct btrfs_free_space *info; | |
819 | ||
820 | while (*p) { | |
821 | parent = *p; | |
96303081 | 822 | info = rb_entry(parent, struct btrfs_free_space, offset_index); |
0f9dd46c | 823 | |
96303081 | 824 | if (offset < info->offset) { |
0f9dd46c | 825 | p = &(*p)->rb_left; |
96303081 | 826 | } else if (offset > info->offset) { |
0f9dd46c | 827 | p = &(*p)->rb_right; |
96303081 JB |
828 | } else { |
829 | /* | |
830 | * we could have a bitmap entry and an extent entry | |
831 | * share the same offset. If this is the case, we want | |
832 | * the extent entry to always be found first if we do a | |
833 | * linear search through the tree, since we want to have | |
834 | * the quickest allocation time, and allocating from an | |
835 | * extent is faster than allocating from a bitmap. So | |
836 | * if we're inserting a bitmap and we find an entry at | |
837 | * this offset, we want to go right, or after this entry | |
838 | * logically. If we are inserting an extent and we've | |
839 | * found a bitmap, we want to go left, or before | |
840 | * logically. | |
841 | */ | |
842 | if (bitmap) { | |
843 | WARN_ON(info->bitmap); | |
844 | p = &(*p)->rb_right; | |
845 | } else { | |
846 | WARN_ON(!info->bitmap); | |
847 | p = &(*p)->rb_left; | |
848 | } | |
849 | } | |
0f9dd46c JB |
850 | } |
851 | ||
852 | rb_link_node(node, parent, p); | |
853 | rb_insert_color(node, root); | |
854 | ||
855 | return 0; | |
856 | } | |
857 | ||
858 | /* | |
70cb0743 JB |
859 | * searches the tree for the given offset. |
860 | * | |
96303081 JB |
861 | * fuzzy - If this is set, then we are trying to make an allocation, and we just |
862 | * want a section that has at least bytes size and comes at or after the given | |
863 | * offset. | |
0f9dd46c | 864 | */ |
96303081 JB |
865 | static struct btrfs_free_space * |
866 | tree_search_offset(struct btrfs_block_group_cache *block_group, | |
867 | u64 offset, int bitmap_only, int fuzzy) | |
0f9dd46c | 868 | { |
96303081 JB |
869 | struct rb_node *n = block_group->free_space_offset.rb_node; |
870 | struct btrfs_free_space *entry, *prev = NULL; | |
871 | ||
872 | /* find entry that is closest to the 'offset' */ | |
873 | while (1) { | |
874 | if (!n) { | |
875 | entry = NULL; | |
876 | break; | |
877 | } | |
0f9dd46c | 878 | |
0f9dd46c | 879 | entry = rb_entry(n, struct btrfs_free_space, offset_index); |
96303081 | 880 | prev = entry; |
0f9dd46c | 881 | |
96303081 | 882 | if (offset < entry->offset) |
0f9dd46c | 883 | n = n->rb_left; |
96303081 | 884 | else if (offset > entry->offset) |
0f9dd46c | 885 | n = n->rb_right; |
96303081 | 886 | else |
0f9dd46c | 887 | break; |
0f9dd46c JB |
888 | } |
889 | ||
96303081 JB |
890 | if (bitmap_only) { |
891 | if (!entry) | |
892 | return NULL; | |
893 | if (entry->bitmap) | |
894 | return entry; | |
0f9dd46c | 895 | |
96303081 JB |
896 | /* |
897 | * bitmap entry and extent entry may share same offset, | |
898 | * in that case, bitmap entry comes after extent entry. | |
899 | */ | |
900 | n = rb_next(n); | |
901 | if (!n) | |
902 | return NULL; | |
903 | entry = rb_entry(n, struct btrfs_free_space, offset_index); | |
904 | if (entry->offset != offset) | |
905 | return NULL; | |
0f9dd46c | 906 | |
96303081 JB |
907 | WARN_ON(!entry->bitmap); |
908 | return entry; | |
909 | } else if (entry) { | |
910 | if (entry->bitmap) { | |
0f9dd46c | 911 | /* |
96303081 JB |
912 | * if previous extent entry covers the offset, |
913 | * we should return it instead of the bitmap entry | |
0f9dd46c | 914 | */ |
96303081 JB |
915 | n = &entry->offset_index; |
916 | while (1) { | |
917 | n = rb_prev(n); | |
918 | if (!n) | |
919 | break; | |
920 | prev = rb_entry(n, struct btrfs_free_space, | |
921 | offset_index); | |
922 | if (!prev->bitmap) { | |
923 | if (prev->offset + prev->bytes > offset) | |
924 | entry = prev; | |
925 | break; | |
926 | } | |
0f9dd46c | 927 | } |
96303081 JB |
928 | } |
929 | return entry; | |
930 | } | |
931 | ||
932 | if (!prev) | |
933 | return NULL; | |
934 | ||
935 | /* find last entry before the 'offset' */ | |
936 | entry = prev; | |
937 | if (entry->offset > offset) { | |
938 | n = rb_prev(&entry->offset_index); | |
939 | if (n) { | |
940 | entry = rb_entry(n, struct btrfs_free_space, | |
941 | offset_index); | |
942 | BUG_ON(entry->offset > offset); | |
0f9dd46c | 943 | } else { |
96303081 JB |
944 | if (fuzzy) |
945 | return entry; | |
946 | else | |
947 | return NULL; | |
0f9dd46c JB |
948 | } |
949 | } | |
950 | ||
96303081 JB |
951 | if (entry->bitmap) { |
952 | n = &entry->offset_index; | |
953 | while (1) { | |
954 | n = rb_prev(n); | |
955 | if (!n) | |
956 | break; | |
957 | prev = rb_entry(n, struct btrfs_free_space, | |
958 | offset_index); | |
959 | if (!prev->bitmap) { | |
960 | if (prev->offset + prev->bytes > offset) | |
961 | return prev; | |
962 | break; | |
963 | } | |
964 | } | |
965 | if (entry->offset + BITS_PER_BITMAP * | |
966 | block_group->sectorsize > offset) | |
967 | return entry; | |
968 | } else if (entry->offset + entry->bytes > offset) | |
969 | return entry; | |
970 | ||
971 | if (!fuzzy) | |
972 | return NULL; | |
973 | ||
974 | while (1) { | |
975 | if (entry->bitmap) { | |
976 | if (entry->offset + BITS_PER_BITMAP * | |
977 | block_group->sectorsize > offset) | |
978 | break; | |
979 | } else { | |
980 | if (entry->offset + entry->bytes > offset) | |
981 | break; | |
982 | } | |
983 | ||
984 | n = rb_next(&entry->offset_index); | |
985 | if (!n) | |
986 | return NULL; | |
987 | entry = rb_entry(n, struct btrfs_free_space, offset_index); | |
988 | } | |
989 | return entry; | |
0f9dd46c JB |
990 | } |
991 | ||
f333adb5 LZ |
992 | static inline void |
993 | __unlink_free_space(struct btrfs_block_group_cache *block_group, | |
994 | struct btrfs_free_space *info) | |
0f9dd46c JB |
995 | { |
996 | rb_erase(&info->offset_index, &block_group->free_space_offset); | |
96303081 | 997 | block_group->free_extents--; |
f333adb5 LZ |
998 | } |
999 | ||
1000 | static void unlink_free_space(struct btrfs_block_group_cache *block_group, | |
1001 | struct btrfs_free_space *info) | |
1002 | { | |
1003 | __unlink_free_space(block_group, info); | |
817d52f8 | 1004 | block_group->free_space -= info->bytes; |
0f9dd46c JB |
1005 | } |
1006 | ||
1007 | static int link_free_space(struct btrfs_block_group_cache *block_group, | |
1008 | struct btrfs_free_space *info) | |
1009 | { | |
1010 | int ret = 0; | |
1011 | ||
96303081 | 1012 | BUG_ON(!info->bitmap && !info->bytes); |
0f9dd46c | 1013 | ret = tree_insert_offset(&block_group->free_space_offset, info->offset, |
96303081 | 1014 | &info->offset_index, (info->bitmap != NULL)); |
0f9dd46c JB |
1015 | if (ret) |
1016 | return ret; | |
1017 | ||
817d52f8 | 1018 | block_group->free_space += info->bytes; |
96303081 JB |
1019 | block_group->free_extents++; |
1020 | return ret; | |
1021 | } | |
1022 | ||
1023 | static void recalculate_thresholds(struct btrfs_block_group_cache *block_group) | |
1024 | { | |
25891f79 JB |
1025 | u64 max_bytes; |
1026 | u64 bitmap_bytes; | |
1027 | u64 extent_bytes; | |
8eb2d829 | 1028 | u64 size = block_group->key.offset; |
96303081 JB |
1029 | |
1030 | /* | |
1031 | * The goal is to keep the total amount of memory used per 1gb of space | |
1032 | * at or below 32k, so we need to adjust how much memory we allow to be | |
1033 | * used by extent based free space tracking | |
1034 | */ | |
8eb2d829 LZ |
1035 | if (size < 1024 * 1024 * 1024) |
1036 | max_bytes = MAX_CACHE_BYTES_PER_GIG; | |
1037 | else | |
1038 | max_bytes = MAX_CACHE_BYTES_PER_GIG * | |
1039 | div64_u64(size, 1024 * 1024 * 1024); | |
96303081 | 1040 | |
25891f79 JB |
1041 | /* |
1042 | * we want to account for 1 more bitmap than what we have so we can make | |
1043 | * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as | |
1044 | * we add more bitmaps. | |
1045 | */ | |
1046 | bitmap_bytes = (block_group->total_bitmaps + 1) * PAGE_CACHE_SIZE; | |
96303081 | 1047 | |
25891f79 JB |
1048 | if (bitmap_bytes >= max_bytes) { |
1049 | block_group->extents_thresh = 0; | |
1050 | return; | |
1051 | } | |
96303081 | 1052 | |
25891f79 JB |
1053 | /* |
1054 | * we want the extent entry threshold to always be at most 1/2 the maxw | |
1055 | * bytes we can have, or whatever is less than that. | |
1056 | */ | |
1057 | extent_bytes = max_bytes - bitmap_bytes; | |
1058 | extent_bytes = min_t(u64, extent_bytes, div64_u64(max_bytes, 2)); | |
96303081 | 1059 | |
25891f79 JB |
1060 | block_group->extents_thresh = |
1061 | div64_u64(extent_bytes, (sizeof(struct btrfs_free_space))); | |
96303081 JB |
1062 | } |
1063 | ||
817d52f8 JB |
1064 | static void bitmap_clear_bits(struct btrfs_block_group_cache *block_group, |
1065 | struct btrfs_free_space *info, u64 offset, | |
1066 | u64 bytes) | |
96303081 JB |
1067 | { |
1068 | unsigned long start, end; | |
1069 | unsigned long i; | |
1070 | ||
817d52f8 JB |
1071 | start = offset_to_bit(info->offset, block_group->sectorsize, offset); |
1072 | end = start + bytes_to_bits(bytes, block_group->sectorsize); | |
96303081 JB |
1073 | BUG_ON(end > BITS_PER_BITMAP); |
1074 | ||
1075 | for (i = start; i < end; i++) | |
1076 | clear_bit(i, info->bitmap); | |
1077 | ||
1078 | info->bytes -= bytes; | |
817d52f8 | 1079 | block_group->free_space -= bytes; |
96303081 JB |
1080 | } |
1081 | ||
817d52f8 JB |
1082 | static void bitmap_set_bits(struct btrfs_block_group_cache *block_group, |
1083 | struct btrfs_free_space *info, u64 offset, | |
1084 | u64 bytes) | |
96303081 JB |
1085 | { |
1086 | unsigned long start, end; | |
1087 | unsigned long i; | |
1088 | ||
817d52f8 JB |
1089 | start = offset_to_bit(info->offset, block_group->sectorsize, offset); |
1090 | end = start + bytes_to_bits(bytes, block_group->sectorsize); | |
96303081 JB |
1091 | BUG_ON(end > BITS_PER_BITMAP); |
1092 | ||
1093 | for (i = start; i < end; i++) | |
1094 | set_bit(i, info->bitmap); | |
1095 | ||
1096 | info->bytes += bytes; | |
817d52f8 | 1097 | block_group->free_space += bytes; |
96303081 JB |
1098 | } |
1099 | ||
1100 | static int search_bitmap(struct btrfs_block_group_cache *block_group, | |
1101 | struct btrfs_free_space *bitmap_info, u64 *offset, | |
1102 | u64 *bytes) | |
1103 | { | |
1104 | unsigned long found_bits = 0; | |
1105 | unsigned long bits, i; | |
1106 | unsigned long next_zero; | |
1107 | ||
1108 | i = offset_to_bit(bitmap_info->offset, block_group->sectorsize, | |
1109 | max_t(u64, *offset, bitmap_info->offset)); | |
1110 | bits = bytes_to_bits(*bytes, block_group->sectorsize); | |
1111 | ||
1112 | for (i = find_next_bit(bitmap_info->bitmap, BITS_PER_BITMAP, i); | |
1113 | i < BITS_PER_BITMAP; | |
1114 | i = find_next_bit(bitmap_info->bitmap, BITS_PER_BITMAP, i + 1)) { | |
1115 | next_zero = find_next_zero_bit(bitmap_info->bitmap, | |
1116 | BITS_PER_BITMAP, i); | |
1117 | if ((next_zero - i) >= bits) { | |
1118 | found_bits = next_zero - i; | |
1119 | break; | |
1120 | } | |
1121 | i = next_zero; | |
1122 | } | |
1123 | ||
1124 | if (found_bits) { | |
1125 | *offset = (u64)(i * block_group->sectorsize) + | |
1126 | bitmap_info->offset; | |
1127 | *bytes = (u64)(found_bits) * block_group->sectorsize; | |
1128 | return 0; | |
1129 | } | |
1130 | ||
1131 | return -1; | |
1132 | } | |
1133 | ||
1134 | static struct btrfs_free_space *find_free_space(struct btrfs_block_group_cache | |
1135 | *block_group, u64 *offset, | |
1136 | u64 *bytes, int debug) | |
1137 | { | |
1138 | struct btrfs_free_space *entry; | |
1139 | struct rb_node *node; | |
1140 | int ret; | |
1141 | ||
1142 | if (!block_group->free_space_offset.rb_node) | |
1143 | return NULL; | |
1144 | ||
1145 | entry = tree_search_offset(block_group, | |
1146 | offset_to_bitmap(block_group, *offset), | |
1147 | 0, 1); | |
1148 | if (!entry) | |
1149 | return NULL; | |
1150 | ||
1151 | for (node = &entry->offset_index; node; node = rb_next(node)) { | |
1152 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
1153 | if (entry->bytes < *bytes) | |
1154 | continue; | |
1155 | ||
1156 | if (entry->bitmap) { | |
1157 | ret = search_bitmap(block_group, entry, offset, bytes); | |
1158 | if (!ret) | |
1159 | return entry; | |
1160 | continue; | |
1161 | } | |
1162 | ||
1163 | *offset = entry->offset; | |
1164 | *bytes = entry->bytes; | |
1165 | return entry; | |
1166 | } | |
1167 | ||
1168 | return NULL; | |
1169 | } | |
1170 | ||
1171 | static void add_new_bitmap(struct btrfs_block_group_cache *block_group, | |
1172 | struct btrfs_free_space *info, u64 offset) | |
1173 | { | |
1174 | u64 bytes_per_bg = BITS_PER_BITMAP * block_group->sectorsize; | |
1175 | int max_bitmaps = (int)div64_u64(block_group->key.offset + | |
1176 | bytes_per_bg - 1, bytes_per_bg); | |
1177 | BUG_ON(block_group->total_bitmaps >= max_bitmaps); | |
1178 | ||
1179 | info->offset = offset_to_bitmap(block_group, offset); | |
f019f426 | 1180 | info->bytes = 0; |
96303081 JB |
1181 | link_free_space(block_group, info); |
1182 | block_group->total_bitmaps++; | |
1183 | ||
1184 | recalculate_thresholds(block_group); | |
1185 | } | |
1186 | ||
edf6e2d1 LZ |
1187 | static void free_bitmap(struct btrfs_block_group_cache *block_group, |
1188 | struct btrfs_free_space *bitmap_info) | |
1189 | { | |
1190 | unlink_free_space(block_group, bitmap_info); | |
1191 | kfree(bitmap_info->bitmap); | |
dc89e982 | 1192 | kmem_cache_free(btrfs_free_space_cachep, bitmap_info); |
edf6e2d1 LZ |
1193 | block_group->total_bitmaps--; |
1194 | recalculate_thresholds(block_group); | |
1195 | } | |
1196 | ||
96303081 JB |
1197 | static noinline int remove_from_bitmap(struct btrfs_block_group_cache *block_group, |
1198 | struct btrfs_free_space *bitmap_info, | |
1199 | u64 *offset, u64 *bytes) | |
1200 | { | |
1201 | u64 end; | |
6606bb97 JB |
1202 | u64 search_start, search_bytes; |
1203 | int ret; | |
96303081 JB |
1204 | |
1205 | again: | |
1206 | end = bitmap_info->offset + | |
1207 | (u64)(BITS_PER_BITMAP * block_group->sectorsize) - 1; | |
1208 | ||
6606bb97 JB |
1209 | /* |
1210 | * XXX - this can go away after a few releases. | |
1211 | * | |
1212 | * since the only user of btrfs_remove_free_space is the tree logging | |
1213 | * stuff, and the only way to test that is under crash conditions, we | |
1214 | * want to have this debug stuff here just in case somethings not | |
1215 | * working. Search the bitmap for the space we are trying to use to | |
1216 | * make sure its actually there. If its not there then we need to stop | |
1217 | * because something has gone wrong. | |
1218 | */ | |
1219 | search_start = *offset; | |
1220 | search_bytes = *bytes; | |
13dbc089 | 1221 | search_bytes = min(search_bytes, end - search_start + 1); |
6606bb97 JB |
1222 | ret = search_bitmap(block_group, bitmap_info, &search_start, |
1223 | &search_bytes); | |
1224 | BUG_ON(ret < 0 || search_start != *offset); | |
1225 | ||
96303081 | 1226 | if (*offset > bitmap_info->offset && *offset + *bytes > end) { |
817d52f8 JB |
1227 | bitmap_clear_bits(block_group, bitmap_info, *offset, |
1228 | end - *offset + 1); | |
96303081 JB |
1229 | *bytes -= end - *offset + 1; |
1230 | *offset = end + 1; | |
1231 | } else if (*offset >= bitmap_info->offset && *offset + *bytes <= end) { | |
817d52f8 | 1232 | bitmap_clear_bits(block_group, bitmap_info, *offset, *bytes); |
96303081 JB |
1233 | *bytes = 0; |
1234 | } | |
1235 | ||
1236 | if (*bytes) { | |
6606bb97 | 1237 | struct rb_node *next = rb_next(&bitmap_info->offset_index); |
edf6e2d1 LZ |
1238 | if (!bitmap_info->bytes) |
1239 | free_bitmap(block_group, bitmap_info); | |
96303081 | 1240 | |
6606bb97 JB |
1241 | /* |
1242 | * no entry after this bitmap, but we still have bytes to | |
1243 | * remove, so something has gone wrong. | |
1244 | */ | |
1245 | if (!next) | |
96303081 JB |
1246 | return -EINVAL; |
1247 | ||
6606bb97 JB |
1248 | bitmap_info = rb_entry(next, struct btrfs_free_space, |
1249 | offset_index); | |
1250 | ||
1251 | /* | |
1252 | * if the next entry isn't a bitmap we need to return to let the | |
1253 | * extent stuff do its work. | |
1254 | */ | |
96303081 JB |
1255 | if (!bitmap_info->bitmap) |
1256 | return -EAGAIN; | |
1257 | ||
6606bb97 JB |
1258 | /* |
1259 | * Ok the next item is a bitmap, but it may not actually hold | |
1260 | * the information for the rest of this free space stuff, so | |
1261 | * look for it, and if we don't find it return so we can try | |
1262 | * everything over again. | |
1263 | */ | |
1264 | search_start = *offset; | |
1265 | search_bytes = *bytes; | |
1266 | ret = search_bitmap(block_group, bitmap_info, &search_start, | |
1267 | &search_bytes); | |
1268 | if (ret < 0 || search_start != *offset) | |
1269 | return -EAGAIN; | |
1270 | ||
96303081 | 1271 | goto again; |
edf6e2d1 LZ |
1272 | } else if (!bitmap_info->bytes) |
1273 | free_bitmap(block_group, bitmap_info); | |
96303081 JB |
1274 | |
1275 | return 0; | |
1276 | } | |
1277 | ||
1278 | static int insert_into_bitmap(struct btrfs_block_group_cache *block_group, | |
1279 | struct btrfs_free_space *info) | |
1280 | { | |
1281 | struct btrfs_free_space *bitmap_info; | |
1282 | int added = 0; | |
1283 | u64 bytes, offset, end; | |
1284 | int ret; | |
1285 | ||
1286 | /* | |
1287 | * If we are below the extents threshold then we can add this as an | |
1288 | * extent, and don't have to deal with the bitmap | |
1289 | */ | |
1290 | if (block_group->free_extents < block_group->extents_thresh && | |
1291 | info->bytes > block_group->sectorsize * 4) | |
1292 | return 0; | |
1293 | ||
1294 | /* | |
1295 | * some block groups are so tiny they can't be enveloped by a bitmap, so | |
1296 | * don't even bother to create a bitmap for this | |
1297 | */ | |
1298 | if (BITS_PER_BITMAP * block_group->sectorsize > | |
1299 | block_group->key.offset) | |
1300 | return 0; | |
1301 | ||
1302 | bytes = info->bytes; | |
1303 | offset = info->offset; | |
1304 | ||
1305 | again: | |
1306 | bitmap_info = tree_search_offset(block_group, | |
1307 | offset_to_bitmap(block_group, offset), | |
1308 | 1, 0); | |
1309 | if (!bitmap_info) { | |
1310 | BUG_ON(added); | |
1311 | goto new_bitmap; | |
1312 | } | |
1313 | ||
1314 | end = bitmap_info->offset + | |
1315 | (u64)(BITS_PER_BITMAP * block_group->sectorsize); | |
1316 | ||
1317 | if (offset >= bitmap_info->offset && offset + bytes > end) { | |
817d52f8 JB |
1318 | bitmap_set_bits(block_group, bitmap_info, offset, |
1319 | end - offset); | |
96303081 JB |
1320 | bytes -= end - offset; |
1321 | offset = end; | |
1322 | added = 0; | |
1323 | } else if (offset >= bitmap_info->offset && offset + bytes <= end) { | |
817d52f8 | 1324 | bitmap_set_bits(block_group, bitmap_info, offset, bytes); |
96303081 JB |
1325 | bytes = 0; |
1326 | } else { | |
1327 | BUG(); | |
1328 | } | |
1329 | ||
1330 | if (!bytes) { | |
1331 | ret = 1; | |
1332 | goto out; | |
1333 | } else | |
1334 | goto again; | |
1335 | ||
1336 | new_bitmap: | |
1337 | if (info && info->bitmap) { | |
1338 | add_new_bitmap(block_group, info, offset); | |
1339 | added = 1; | |
1340 | info = NULL; | |
1341 | goto again; | |
1342 | } else { | |
1343 | spin_unlock(&block_group->tree_lock); | |
1344 | ||
1345 | /* no pre-allocated info, allocate a new one */ | |
1346 | if (!info) { | |
dc89e982 JB |
1347 | info = kmem_cache_zalloc(btrfs_free_space_cachep, |
1348 | GFP_NOFS); | |
96303081 JB |
1349 | if (!info) { |
1350 | spin_lock(&block_group->tree_lock); | |
1351 | ret = -ENOMEM; | |
1352 | goto out; | |
1353 | } | |
1354 | } | |
1355 | ||
1356 | /* allocate the bitmap */ | |
1357 | info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS); | |
1358 | spin_lock(&block_group->tree_lock); | |
1359 | if (!info->bitmap) { | |
1360 | ret = -ENOMEM; | |
1361 | goto out; | |
1362 | } | |
1363 | goto again; | |
1364 | } | |
1365 | ||
1366 | out: | |
1367 | if (info) { | |
1368 | if (info->bitmap) | |
1369 | kfree(info->bitmap); | |
dc89e982 | 1370 | kmem_cache_free(btrfs_free_space_cachep, info); |
96303081 | 1371 | } |
0f9dd46c JB |
1372 | |
1373 | return ret; | |
1374 | } | |
1375 | ||
120d66ee | 1376 | bool try_merge_free_space(struct btrfs_block_group_cache *block_group, |
f333adb5 | 1377 | struct btrfs_free_space *info, bool update_stat) |
0f9dd46c | 1378 | { |
120d66ee LZ |
1379 | struct btrfs_free_space *left_info; |
1380 | struct btrfs_free_space *right_info; | |
1381 | bool merged = false; | |
1382 | u64 offset = info->offset; | |
1383 | u64 bytes = info->bytes; | |
6226cb0a | 1384 | |
0f9dd46c JB |
1385 | /* |
1386 | * first we want to see if there is free space adjacent to the range we | |
1387 | * are adding, if there is remove that struct and add a new one to | |
1388 | * cover the entire range | |
1389 | */ | |
96303081 JB |
1390 | right_info = tree_search_offset(block_group, offset + bytes, 0, 0); |
1391 | if (right_info && rb_prev(&right_info->offset_index)) | |
1392 | left_info = rb_entry(rb_prev(&right_info->offset_index), | |
1393 | struct btrfs_free_space, offset_index); | |
1394 | else | |
1395 | left_info = tree_search_offset(block_group, offset - 1, 0, 0); | |
0f9dd46c | 1396 | |
96303081 | 1397 | if (right_info && !right_info->bitmap) { |
f333adb5 LZ |
1398 | if (update_stat) |
1399 | unlink_free_space(block_group, right_info); | |
1400 | else | |
1401 | __unlink_free_space(block_group, right_info); | |
6226cb0a | 1402 | info->bytes += right_info->bytes; |
dc89e982 | 1403 | kmem_cache_free(btrfs_free_space_cachep, right_info); |
120d66ee | 1404 | merged = true; |
0f9dd46c JB |
1405 | } |
1406 | ||
96303081 JB |
1407 | if (left_info && !left_info->bitmap && |
1408 | left_info->offset + left_info->bytes == offset) { | |
f333adb5 LZ |
1409 | if (update_stat) |
1410 | unlink_free_space(block_group, left_info); | |
1411 | else | |
1412 | __unlink_free_space(block_group, left_info); | |
6226cb0a JB |
1413 | info->offset = left_info->offset; |
1414 | info->bytes += left_info->bytes; | |
dc89e982 | 1415 | kmem_cache_free(btrfs_free_space_cachep, left_info); |
120d66ee | 1416 | merged = true; |
0f9dd46c JB |
1417 | } |
1418 | ||
120d66ee LZ |
1419 | return merged; |
1420 | } | |
1421 | ||
1422 | int btrfs_add_free_space(struct btrfs_block_group_cache *block_group, | |
1423 | u64 offset, u64 bytes) | |
1424 | { | |
1425 | struct btrfs_free_space *info; | |
1426 | int ret = 0; | |
1427 | ||
dc89e982 | 1428 | info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS); |
120d66ee LZ |
1429 | if (!info) |
1430 | return -ENOMEM; | |
1431 | ||
1432 | info->offset = offset; | |
1433 | info->bytes = bytes; | |
1434 | ||
1435 | spin_lock(&block_group->tree_lock); | |
1436 | ||
f333adb5 | 1437 | if (try_merge_free_space(block_group, info, true)) |
120d66ee LZ |
1438 | goto link; |
1439 | ||
1440 | /* | |
1441 | * There was no extent directly to the left or right of this new | |
1442 | * extent then we know we're going to have to allocate a new extent, so | |
1443 | * before we do that see if we need to drop this into a bitmap | |
1444 | */ | |
1445 | ret = insert_into_bitmap(block_group, info); | |
1446 | if (ret < 0) { | |
1447 | goto out; | |
1448 | } else if (ret) { | |
1449 | ret = 0; | |
1450 | goto out; | |
1451 | } | |
1452 | link: | |
0f9dd46c JB |
1453 | ret = link_free_space(block_group, info); |
1454 | if (ret) | |
dc89e982 | 1455 | kmem_cache_free(btrfs_free_space_cachep, info); |
96303081 | 1456 | out: |
6226cb0a JB |
1457 | spin_unlock(&block_group->tree_lock); |
1458 | ||
0f9dd46c | 1459 | if (ret) { |
96303081 | 1460 | printk(KERN_CRIT "btrfs: unable to add free space :%d\n", ret); |
c293498b | 1461 | BUG_ON(ret == -EEXIST); |
0f9dd46c JB |
1462 | } |
1463 | ||
0f9dd46c JB |
1464 | return ret; |
1465 | } | |
1466 | ||
6226cb0a JB |
1467 | int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group, |
1468 | u64 offset, u64 bytes) | |
0f9dd46c JB |
1469 | { |
1470 | struct btrfs_free_space *info; | |
96303081 | 1471 | struct btrfs_free_space *next_info = NULL; |
0f9dd46c JB |
1472 | int ret = 0; |
1473 | ||
6226cb0a JB |
1474 | spin_lock(&block_group->tree_lock); |
1475 | ||
96303081 JB |
1476 | again: |
1477 | info = tree_search_offset(block_group, offset, 0, 0); | |
1478 | if (!info) { | |
6606bb97 JB |
1479 | /* |
1480 | * oops didn't find an extent that matched the space we wanted | |
1481 | * to remove, look for a bitmap instead | |
1482 | */ | |
1483 | info = tree_search_offset(block_group, | |
1484 | offset_to_bitmap(block_group, offset), | |
1485 | 1, 0); | |
1486 | if (!info) { | |
1487 | WARN_ON(1); | |
1488 | goto out_lock; | |
1489 | } | |
96303081 JB |
1490 | } |
1491 | ||
1492 | if (info->bytes < bytes && rb_next(&info->offset_index)) { | |
1493 | u64 end; | |
1494 | next_info = rb_entry(rb_next(&info->offset_index), | |
1495 | struct btrfs_free_space, | |
1496 | offset_index); | |
1497 | ||
1498 | if (next_info->bitmap) | |
1499 | end = next_info->offset + BITS_PER_BITMAP * | |
1500 | block_group->sectorsize - 1; | |
1501 | else | |
1502 | end = next_info->offset + next_info->bytes; | |
1503 | ||
1504 | if (next_info->bytes < bytes || | |
1505 | next_info->offset > offset || offset > end) { | |
1506 | printk(KERN_CRIT "Found free space at %llu, size %llu," | |
1507 | " trying to use %llu\n", | |
1508 | (unsigned long long)info->offset, | |
1509 | (unsigned long long)info->bytes, | |
1510 | (unsigned long long)bytes); | |
0f9dd46c JB |
1511 | WARN_ON(1); |
1512 | ret = -EINVAL; | |
96303081 | 1513 | goto out_lock; |
0f9dd46c | 1514 | } |
0f9dd46c | 1515 | |
96303081 JB |
1516 | info = next_info; |
1517 | } | |
1518 | ||
1519 | if (info->bytes == bytes) { | |
1520 | unlink_free_space(block_group, info); | |
1521 | if (info->bitmap) { | |
1522 | kfree(info->bitmap); | |
1523 | block_group->total_bitmaps--; | |
0f9dd46c | 1524 | } |
dc89e982 | 1525 | kmem_cache_free(btrfs_free_space_cachep, info); |
96303081 JB |
1526 | goto out_lock; |
1527 | } | |
0f9dd46c | 1528 | |
96303081 JB |
1529 | if (!info->bitmap && info->offset == offset) { |
1530 | unlink_free_space(block_group, info); | |
0f9dd46c JB |
1531 | info->offset += bytes; |
1532 | info->bytes -= bytes; | |
96303081 JB |
1533 | link_free_space(block_group, info); |
1534 | goto out_lock; | |
1535 | } | |
0f9dd46c | 1536 | |
96303081 JB |
1537 | if (!info->bitmap && info->offset <= offset && |
1538 | info->offset + info->bytes >= offset + bytes) { | |
9b49c9b9 CM |
1539 | u64 old_start = info->offset; |
1540 | /* | |
1541 | * we're freeing space in the middle of the info, | |
1542 | * this can happen during tree log replay | |
1543 | * | |
1544 | * first unlink the old info and then | |
1545 | * insert it again after the hole we're creating | |
1546 | */ | |
1547 | unlink_free_space(block_group, info); | |
1548 | if (offset + bytes < info->offset + info->bytes) { | |
1549 | u64 old_end = info->offset + info->bytes; | |
1550 | ||
1551 | info->offset = offset + bytes; | |
1552 | info->bytes = old_end - info->offset; | |
1553 | ret = link_free_space(block_group, info); | |
96303081 JB |
1554 | WARN_ON(ret); |
1555 | if (ret) | |
1556 | goto out_lock; | |
9b49c9b9 CM |
1557 | } else { |
1558 | /* the hole we're creating ends at the end | |
1559 | * of the info struct, just free the info | |
1560 | */ | |
dc89e982 | 1561 | kmem_cache_free(btrfs_free_space_cachep, info); |
9b49c9b9 | 1562 | } |
6226cb0a | 1563 | spin_unlock(&block_group->tree_lock); |
96303081 JB |
1564 | |
1565 | /* step two, insert a new info struct to cover | |
1566 | * anything before the hole | |
9b49c9b9 | 1567 | */ |
6226cb0a JB |
1568 | ret = btrfs_add_free_space(block_group, old_start, |
1569 | offset - old_start); | |
96303081 JB |
1570 | WARN_ON(ret); |
1571 | goto out; | |
0f9dd46c | 1572 | } |
96303081 JB |
1573 | |
1574 | ret = remove_from_bitmap(block_group, info, &offset, &bytes); | |
1575 | if (ret == -EAGAIN) | |
1576 | goto again; | |
1577 | BUG_ON(ret); | |
1578 | out_lock: | |
1579 | spin_unlock(&block_group->tree_lock); | |
0f9dd46c | 1580 | out: |
25179201 JB |
1581 | return ret; |
1582 | } | |
1583 | ||
0f9dd46c JB |
1584 | void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group, |
1585 | u64 bytes) | |
1586 | { | |
1587 | struct btrfs_free_space *info; | |
1588 | struct rb_node *n; | |
1589 | int count = 0; | |
1590 | ||
1591 | for (n = rb_first(&block_group->free_space_offset); n; n = rb_next(n)) { | |
1592 | info = rb_entry(n, struct btrfs_free_space, offset_index); | |
1593 | if (info->bytes >= bytes) | |
1594 | count++; | |
96303081 | 1595 | printk(KERN_CRIT "entry offset %llu, bytes %llu, bitmap %s\n", |
21380931 | 1596 | (unsigned long long)info->offset, |
96303081 JB |
1597 | (unsigned long long)info->bytes, |
1598 | (info->bitmap) ? "yes" : "no"); | |
0f9dd46c | 1599 | } |
96303081 JB |
1600 | printk(KERN_INFO "block group has cluster?: %s\n", |
1601 | list_empty(&block_group->cluster_list) ? "no" : "yes"); | |
0f9dd46c JB |
1602 | printk(KERN_INFO "%d blocks of free space at or bigger than bytes is" |
1603 | "\n", count); | |
1604 | } | |
1605 | ||
1606 | u64 btrfs_block_group_free_space(struct btrfs_block_group_cache *block_group) | |
1607 | { | |
1608 | struct btrfs_free_space *info; | |
1609 | struct rb_node *n; | |
1610 | u64 ret = 0; | |
1611 | ||
1612 | for (n = rb_first(&block_group->free_space_offset); n; | |
1613 | n = rb_next(n)) { | |
1614 | info = rb_entry(n, struct btrfs_free_space, offset_index); | |
1615 | ret += info->bytes; | |
1616 | } | |
1617 | ||
1618 | return ret; | |
1619 | } | |
1620 | ||
fa9c0d79 CM |
1621 | /* |
1622 | * for a given cluster, put all of its extents back into the free | |
1623 | * space cache. If the block group passed doesn't match the block group | |
1624 | * pointed to by the cluster, someone else raced in and freed the | |
1625 | * cluster already. In that case, we just return without changing anything | |
1626 | */ | |
1627 | static int | |
1628 | __btrfs_return_cluster_to_free_space( | |
1629 | struct btrfs_block_group_cache *block_group, | |
1630 | struct btrfs_free_cluster *cluster) | |
1631 | { | |
1632 | struct btrfs_free_space *entry; | |
1633 | struct rb_node *node; | |
96303081 | 1634 | bool bitmap; |
fa9c0d79 CM |
1635 | |
1636 | spin_lock(&cluster->lock); | |
1637 | if (cluster->block_group != block_group) | |
1638 | goto out; | |
1639 | ||
96303081 JB |
1640 | bitmap = cluster->points_to_bitmap; |
1641 | cluster->block_group = NULL; | |
fa9c0d79 | 1642 | cluster->window_start = 0; |
96303081 JB |
1643 | list_del_init(&cluster->block_group_list); |
1644 | cluster->points_to_bitmap = false; | |
1645 | ||
1646 | if (bitmap) | |
1647 | goto out; | |
1648 | ||
fa9c0d79 | 1649 | node = rb_first(&cluster->root); |
96303081 | 1650 | while (node) { |
fa9c0d79 CM |
1651 | entry = rb_entry(node, struct btrfs_free_space, offset_index); |
1652 | node = rb_next(&entry->offset_index); | |
1653 | rb_erase(&entry->offset_index, &cluster->root); | |
96303081 | 1654 | BUG_ON(entry->bitmap); |
f333adb5 | 1655 | try_merge_free_space(block_group, entry, false); |
96303081 JB |
1656 | tree_insert_offset(&block_group->free_space_offset, |
1657 | entry->offset, &entry->offset_index, 0); | |
fa9c0d79 | 1658 | } |
6bef4d31 | 1659 | cluster->root = RB_ROOT; |
96303081 | 1660 | |
fa9c0d79 CM |
1661 | out: |
1662 | spin_unlock(&cluster->lock); | |
96303081 | 1663 | btrfs_put_block_group(block_group); |
fa9c0d79 CM |
1664 | return 0; |
1665 | } | |
1666 | ||
0f9dd46c JB |
1667 | void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group) |
1668 | { | |
1669 | struct btrfs_free_space *info; | |
1670 | struct rb_node *node; | |
fa9c0d79 | 1671 | struct btrfs_free_cluster *cluster; |
96303081 | 1672 | struct list_head *head; |
0f9dd46c | 1673 | |
6226cb0a | 1674 | spin_lock(&block_group->tree_lock); |
96303081 JB |
1675 | while ((head = block_group->cluster_list.next) != |
1676 | &block_group->cluster_list) { | |
1677 | cluster = list_entry(head, struct btrfs_free_cluster, | |
1678 | block_group_list); | |
fa9c0d79 CM |
1679 | |
1680 | WARN_ON(cluster->block_group != block_group); | |
1681 | __btrfs_return_cluster_to_free_space(block_group, cluster); | |
96303081 JB |
1682 | if (need_resched()) { |
1683 | spin_unlock(&block_group->tree_lock); | |
1684 | cond_resched(); | |
1685 | spin_lock(&block_group->tree_lock); | |
1686 | } | |
fa9c0d79 CM |
1687 | } |
1688 | ||
96303081 JB |
1689 | while ((node = rb_last(&block_group->free_space_offset)) != NULL) { |
1690 | info = rb_entry(node, struct btrfs_free_space, offset_index); | |
0f9dd46c | 1691 | unlink_free_space(block_group, info); |
96303081 JB |
1692 | if (info->bitmap) |
1693 | kfree(info->bitmap); | |
dc89e982 | 1694 | kmem_cache_free(btrfs_free_space_cachep, info); |
0f9dd46c | 1695 | if (need_resched()) { |
6226cb0a | 1696 | spin_unlock(&block_group->tree_lock); |
0f9dd46c | 1697 | cond_resched(); |
6226cb0a | 1698 | spin_lock(&block_group->tree_lock); |
0f9dd46c JB |
1699 | } |
1700 | } | |
96303081 | 1701 | |
6226cb0a | 1702 | spin_unlock(&block_group->tree_lock); |
0f9dd46c JB |
1703 | } |
1704 | ||
6226cb0a JB |
1705 | u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group, |
1706 | u64 offset, u64 bytes, u64 empty_size) | |
0f9dd46c | 1707 | { |
6226cb0a | 1708 | struct btrfs_free_space *entry = NULL; |
96303081 | 1709 | u64 bytes_search = bytes + empty_size; |
6226cb0a | 1710 | u64 ret = 0; |
0f9dd46c | 1711 | |
6226cb0a | 1712 | spin_lock(&block_group->tree_lock); |
96303081 | 1713 | entry = find_free_space(block_group, &offset, &bytes_search, 0); |
6226cb0a | 1714 | if (!entry) |
96303081 JB |
1715 | goto out; |
1716 | ||
1717 | ret = offset; | |
1718 | if (entry->bitmap) { | |
817d52f8 | 1719 | bitmap_clear_bits(block_group, entry, offset, bytes); |
edf6e2d1 LZ |
1720 | if (!entry->bytes) |
1721 | free_bitmap(block_group, entry); | |
96303081 | 1722 | } else { |
6226cb0a | 1723 | unlink_free_space(block_group, entry); |
6226cb0a JB |
1724 | entry->offset += bytes; |
1725 | entry->bytes -= bytes; | |
6226cb0a | 1726 | if (!entry->bytes) |
dc89e982 | 1727 | kmem_cache_free(btrfs_free_space_cachep, entry); |
6226cb0a JB |
1728 | else |
1729 | link_free_space(block_group, entry); | |
1730 | } | |
0f9dd46c | 1731 | |
96303081 JB |
1732 | out: |
1733 | spin_unlock(&block_group->tree_lock); | |
817d52f8 | 1734 | |
0f9dd46c JB |
1735 | return ret; |
1736 | } | |
fa9c0d79 CM |
1737 | |
1738 | /* | |
1739 | * given a cluster, put all of its extents back into the free space | |
1740 | * cache. If a block group is passed, this function will only free | |
1741 | * a cluster that belongs to the passed block group. | |
1742 | * | |
1743 | * Otherwise, it'll get a reference on the block group pointed to by the | |
1744 | * cluster and remove the cluster from it. | |
1745 | */ | |
1746 | int btrfs_return_cluster_to_free_space( | |
1747 | struct btrfs_block_group_cache *block_group, | |
1748 | struct btrfs_free_cluster *cluster) | |
1749 | { | |
1750 | int ret; | |
1751 | ||
1752 | /* first, get a safe pointer to the block group */ | |
1753 | spin_lock(&cluster->lock); | |
1754 | if (!block_group) { | |
1755 | block_group = cluster->block_group; | |
1756 | if (!block_group) { | |
1757 | spin_unlock(&cluster->lock); | |
1758 | return 0; | |
1759 | } | |
1760 | } else if (cluster->block_group != block_group) { | |
1761 | /* someone else has already freed it don't redo their work */ | |
1762 | spin_unlock(&cluster->lock); | |
1763 | return 0; | |
1764 | } | |
1765 | atomic_inc(&block_group->count); | |
1766 | spin_unlock(&cluster->lock); | |
1767 | ||
1768 | /* now return any extents the cluster had on it */ | |
1769 | spin_lock(&block_group->tree_lock); | |
1770 | ret = __btrfs_return_cluster_to_free_space(block_group, cluster); | |
1771 | spin_unlock(&block_group->tree_lock); | |
1772 | ||
1773 | /* finally drop our ref */ | |
1774 | btrfs_put_block_group(block_group); | |
1775 | return ret; | |
1776 | } | |
1777 | ||
96303081 JB |
1778 | static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group, |
1779 | struct btrfs_free_cluster *cluster, | |
1780 | u64 bytes, u64 min_start) | |
1781 | { | |
1782 | struct btrfs_free_space *entry; | |
1783 | int err; | |
1784 | u64 search_start = cluster->window_start; | |
1785 | u64 search_bytes = bytes; | |
1786 | u64 ret = 0; | |
1787 | ||
1788 | spin_lock(&block_group->tree_lock); | |
1789 | spin_lock(&cluster->lock); | |
1790 | ||
1791 | if (!cluster->points_to_bitmap) | |
1792 | goto out; | |
1793 | ||
1794 | if (cluster->block_group != block_group) | |
1795 | goto out; | |
1796 | ||
6606bb97 JB |
1797 | /* |
1798 | * search_start is the beginning of the bitmap, but at some point it may | |
1799 | * be a good idea to point to the actual start of the free area in the | |
1800 | * bitmap, so do the offset_to_bitmap trick anyway, and set bitmap_only | |
1801 | * to 1 to make sure we get the bitmap entry | |
1802 | */ | |
1803 | entry = tree_search_offset(block_group, | |
1804 | offset_to_bitmap(block_group, search_start), | |
1805 | 1, 0); | |
96303081 JB |
1806 | if (!entry || !entry->bitmap) |
1807 | goto out; | |
1808 | ||
1809 | search_start = min_start; | |
1810 | search_bytes = bytes; | |
1811 | ||
1812 | err = search_bitmap(block_group, entry, &search_start, | |
1813 | &search_bytes); | |
1814 | if (err) | |
1815 | goto out; | |
1816 | ||
1817 | ret = search_start; | |
817d52f8 | 1818 | bitmap_clear_bits(block_group, entry, ret, bytes); |
70b7da30 LZ |
1819 | if (entry->bytes == 0) |
1820 | free_bitmap(block_group, entry); | |
96303081 JB |
1821 | out: |
1822 | spin_unlock(&cluster->lock); | |
1823 | spin_unlock(&block_group->tree_lock); | |
1824 | ||
1825 | return ret; | |
1826 | } | |
1827 | ||
fa9c0d79 CM |
1828 | /* |
1829 | * given a cluster, try to allocate 'bytes' from it, returns 0 | |
1830 | * if it couldn't find anything suitably large, or a logical disk offset | |
1831 | * if things worked out | |
1832 | */ | |
1833 | u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group, | |
1834 | struct btrfs_free_cluster *cluster, u64 bytes, | |
1835 | u64 min_start) | |
1836 | { | |
1837 | struct btrfs_free_space *entry = NULL; | |
1838 | struct rb_node *node; | |
1839 | u64 ret = 0; | |
1840 | ||
96303081 JB |
1841 | if (cluster->points_to_bitmap) |
1842 | return btrfs_alloc_from_bitmap(block_group, cluster, bytes, | |
1843 | min_start); | |
1844 | ||
fa9c0d79 CM |
1845 | spin_lock(&cluster->lock); |
1846 | if (bytes > cluster->max_size) | |
1847 | goto out; | |
1848 | ||
1849 | if (cluster->block_group != block_group) | |
1850 | goto out; | |
1851 | ||
1852 | node = rb_first(&cluster->root); | |
1853 | if (!node) | |
1854 | goto out; | |
1855 | ||
1856 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
1857 | ||
1858 | while(1) { | |
1859 | if (entry->bytes < bytes || entry->offset < min_start) { | |
1860 | struct rb_node *node; | |
1861 | ||
1862 | node = rb_next(&entry->offset_index); | |
1863 | if (!node) | |
1864 | break; | |
1865 | entry = rb_entry(node, struct btrfs_free_space, | |
1866 | offset_index); | |
1867 | continue; | |
1868 | } | |
1869 | ret = entry->offset; | |
1870 | ||
1871 | entry->offset += bytes; | |
1872 | entry->bytes -= bytes; | |
1873 | ||
5e71b5d5 | 1874 | if (entry->bytes == 0) |
fa9c0d79 | 1875 | rb_erase(&entry->offset_index, &cluster->root); |
fa9c0d79 CM |
1876 | break; |
1877 | } | |
1878 | out: | |
1879 | spin_unlock(&cluster->lock); | |
96303081 | 1880 | |
5e71b5d5 LZ |
1881 | if (!ret) |
1882 | return 0; | |
1883 | ||
1884 | spin_lock(&block_group->tree_lock); | |
1885 | ||
1886 | block_group->free_space -= bytes; | |
1887 | if (entry->bytes == 0) { | |
1888 | block_group->free_extents--; | |
dc89e982 | 1889 | kmem_cache_free(btrfs_free_space_cachep, entry); |
5e71b5d5 LZ |
1890 | } |
1891 | ||
1892 | spin_unlock(&block_group->tree_lock); | |
1893 | ||
fa9c0d79 CM |
1894 | return ret; |
1895 | } | |
1896 | ||
96303081 JB |
1897 | static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group, |
1898 | struct btrfs_free_space *entry, | |
1899 | struct btrfs_free_cluster *cluster, | |
1900 | u64 offset, u64 bytes, u64 min_bytes) | |
1901 | { | |
1902 | unsigned long next_zero; | |
1903 | unsigned long i; | |
1904 | unsigned long search_bits; | |
1905 | unsigned long total_bits; | |
1906 | unsigned long found_bits; | |
1907 | unsigned long start = 0; | |
1908 | unsigned long total_found = 0; | |
1909 | bool found = false; | |
1910 | ||
1911 | i = offset_to_bit(entry->offset, block_group->sectorsize, | |
1912 | max_t(u64, offset, entry->offset)); | |
1913 | search_bits = bytes_to_bits(min_bytes, block_group->sectorsize); | |
1914 | total_bits = bytes_to_bits(bytes, block_group->sectorsize); | |
1915 | ||
1916 | again: | |
1917 | found_bits = 0; | |
1918 | for (i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, i); | |
1919 | i < BITS_PER_BITMAP; | |
1920 | i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, i + 1)) { | |
1921 | next_zero = find_next_zero_bit(entry->bitmap, | |
1922 | BITS_PER_BITMAP, i); | |
1923 | if (next_zero - i >= search_bits) { | |
1924 | found_bits = next_zero - i; | |
1925 | break; | |
1926 | } | |
1927 | i = next_zero; | |
1928 | } | |
1929 | ||
1930 | if (!found_bits) | |
1931 | return -1; | |
1932 | ||
1933 | if (!found) { | |
1934 | start = i; | |
1935 | found = true; | |
1936 | } | |
1937 | ||
1938 | total_found += found_bits; | |
1939 | ||
1940 | if (cluster->max_size < found_bits * block_group->sectorsize) | |
1941 | cluster->max_size = found_bits * block_group->sectorsize; | |
1942 | ||
1943 | if (total_found < total_bits) { | |
1944 | i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, next_zero); | |
1945 | if (i - start > total_bits * 2) { | |
1946 | total_found = 0; | |
1947 | cluster->max_size = 0; | |
1948 | found = false; | |
1949 | } | |
1950 | goto again; | |
1951 | } | |
1952 | ||
1953 | cluster->window_start = start * block_group->sectorsize + | |
1954 | entry->offset; | |
1955 | cluster->points_to_bitmap = true; | |
1956 | ||
1957 | return 0; | |
1958 | } | |
1959 | ||
fa9c0d79 CM |
1960 | /* |
1961 | * here we try to find a cluster of blocks in a block group. The goal | |
1962 | * is to find at least bytes free and up to empty_size + bytes free. | |
1963 | * We might not find them all in one contiguous area. | |
1964 | * | |
1965 | * returns zero and sets up cluster if things worked out, otherwise | |
1966 | * it returns -enospc | |
1967 | */ | |
1968 | int btrfs_find_space_cluster(struct btrfs_trans_handle *trans, | |
451d7585 | 1969 | struct btrfs_root *root, |
fa9c0d79 CM |
1970 | struct btrfs_block_group_cache *block_group, |
1971 | struct btrfs_free_cluster *cluster, | |
1972 | u64 offset, u64 bytes, u64 empty_size) | |
1973 | { | |
1974 | struct btrfs_free_space *entry = NULL; | |
1975 | struct rb_node *node; | |
1976 | struct btrfs_free_space *next; | |
96303081 | 1977 | struct btrfs_free_space *last = NULL; |
fa9c0d79 CM |
1978 | u64 min_bytes; |
1979 | u64 window_start; | |
1980 | u64 window_free; | |
1981 | u64 max_extent = 0; | |
96303081 | 1982 | bool found_bitmap = false; |
fa9c0d79 CM |
1983 | int ret; |
1984 | ||
1985 | /* for metadata, allow allocates with more holes */ | |
451d7585 CM |
1986 | if (btrfs_test_opt(root, SSD_SPREAD)) { |
1987 | min_bytes = bytes + empty_size; | |
1988 | } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) { | |
fa9c0d79 CM |
1989 | /* |
1990 | * we want to do larger allocations when we are | |
1991 | * flushing out the delayed refs, it helps prevent | |
1992 | * making more work as we go along. | |
1993 | */ | |
1994 | if (trans->transaction->delayed_refs.flushing) | |
1995 | min_bytes = max(bytes, (bytes + empty_size) >> 1); | |
1996 | else | |
1997 | min_bytes = max(bytes, (bytes + empty_size) >> 4); | |
1998 | } else | |
1999 | min_bytes = max(bytes, (bytes + empty_size) >> 2); | |
2000 | ||
2001 | spin_lock(&block_group->tree_lock); | |
2002 | spin_lock(&cluster->lock); | |
2003 | ||
2004 | /* someone already found a cluster, hooray */ | |
2005 | if (cluster->block_group) { | |
2006 | ret = 0; | |
2007 | goto out; | |
2008 | } | |
2009 | again: | |
96303081 | 2010 | entry = tree_search_offset(block_group, offset, found_bitmap, 1); |
fa9c0d79 CM |
2011 | if (!entry) { |
2012 | ret = -ENOSPC; | |
2013 | goto out; | |
2014 | } | |
96303081 JB |
2015 | |
2016 | /* | |
2017 | * If found_bitmap is true, we exhausted our search for extent entries, | |
2018 | * and we just want to search all of the bitmaps that we can find, and | |
2019 | * ignore any extent entries we find. | |
2020 | */ | |
2021 | while (entry->bitmap || found_bitmap || | |
2022 | (!entry->bitmap && entry->bytes < min_bytes)) { | |
2023 | struct rb_node *node = rb_next(&entry->offset_index); | |
2024 | ||
2025 | if (entry->bitmap && entry->bytes > bytes + empty_size) { | |
2026 | ret = btrfs_bitmap_cluster(block_group, entry, cluster, | |
2027 | offset, bytes + empty_size, | |
2028 | min_bytes); | |
2029 | if (!ret) | |
2030 | goto got_it; | |
2031 | } | |
2032 | ||
2033 | if (!node) { | |
2034 | ret = -ENOSPC; | |
2035 | goto out; | |
2036 | } | |
2037 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
2038 | } | |
2039 | ||
2040 | /* | |
2041 | * We already searched all the extent entries from the passed in offset | |
2042 | * to the end and didn't find enough space for the cluster, and we also | |
2043 | * didn't find any bitmaps that met our criteria, just go ahead and exit | |
2044 | */ | |
2045 | if (found_bitmap) { | |
2046 | ret = -ENOSPC; | |
2047 | goto out; | |
2048 | } | |
2049 | ||
2050 | cluster->points_to_bitmap = false; | |
fa9c0d79 CM |
2051 | window_start = entry->offset; |
2052 | window_free = entry->bytes; | |
2053 | last = entry; | |
2054 | max_extent = entry->bytes; | |
2055 | ||
96303081 | 2056 | while (1) { |
fa9c0d79 CM |
2057 | /* out window is just right, lets fill it */ |
2058 | if (window_free >= bytes + empty_size) | |
2059 | break; | |
2060 | ||
2061 | node = rb_next(&last->offset_index); | |
2062 | if (!node) { | |
96303081 JB |
2063 | if (found_bitmap) |
2064 | goto again; | |
fa9c0d79 CM |
2065 | ret = -ENOSPC; |
2066 | goto out; | |
2067 | } | |
2068 | next = rb_entry(node, struct btrfs_free_space, offset_index); | |
2069 | ||
96303081 JB |
2070 | /* |
2071 | * we found a bitmap, so if this search doesn't result in a | |
2072 | * cluster, we know to go and search again for the bitmaps and | |
2073 | * start looking for space there | |
2074 | */ | |
2075 | if (next->bitmap) { | |
2076 | if (!found_bitmap) | |
2077 | offset = next->offset; | |
2078 | found_bitmap = true; | |
2079 | last = next; | |
2080 | continue; | |
2081 | } | |
2082 | ||
fa9c0d79 CM |
2083 | /* |
2084 | * we haven't filled the empty size and the window is | |
2085 | * very large. reset and try again | |
2086 | */ | |
c6044801 CM |
2087 | if (next->offset - (last->offset + last->bytes) > 128 * 1024 || |
2088 | next->offset - window_start > (bytes + empty_size) * 2) { | |
fa9c0d79 CM |
2089 | entry = next; |
2090 | window_start = entry->offset; | |
2091 | window_free = entry->bytes; | |
2092 | last = entry; | |
01dea1ef | 2093 | max_extent = entry->bytes; |
fa9c0d79 CM |
2094 | } else { |
2095 | last = next; | |
2096 | window_free += next->bytes; | |
2097 | if (entry->bytes > max_extent) | |
2098 | max_extent = entry->bytes; | |
2099 | } | |
2100 | } | |
2101 | ||
2102 | cluster->window_start = entry->offset; | |
2103 | ||
2104 | /* | |
2105 | * now we've found our entries, pull them out of the free space | |
2106 | * cache and put them into the cluster rbtree | |
2107 | * | |
2108 | * The cluster includes an rbtree, but only uses the offset index | |
2109 | * of each free space cache entry. | |
2110 | */ | |
96303081 | 2111 | while (1) { |
fa9c0d79 | 2112 | node = rb_next(&entry->offset_index); |
96303081 JB |
2113 | if (entry->bitmap && node) { |
2114 | entry = rb_entry(node, struct btrfs_free_space, | |
2115 | offset_index); | |
2116 | continue; | |
2117 | } else if (entry->bitmap && !node) { | |
2118 | break; | |
2119 | } | |
2120 | ||
2121 | rb_erase(&entry->offset_index, &block_group->free_space_offset); | |
fa9c0d79 | 2122 | ret = tree_insert_offset(&cluster->root, entry->offset, |
96303081 | 2123 | &entry->offset_index, 0); |
fa9c0d79 CM |
2124 | BUG_ON(ret); |
2125 | ||
2126 | if (!node || entry == last) | |
2127 | break; | |
2128 | ||
2129 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
2130 | } | |
96303081 | 2131 | |
fa9c0d79 | 2132 | cluster->max_size = max_extent; |
96303081 JB |
2133 | got_it: |
2134 | ret = 0; | |
fa9c0d79 CM |
2135 | atomic_inc(&block_group->count); |
2136 | list_add_tail(&cluster->block_group_list, &block_group->cluster_list); | |
2137 | cluster->block_group = block_group; | |
2138 | out: | |
2139 | spin_unlock(&cluster->lock); | |
2140 | spin_unlock(&block_group->tree_lock); | |
2141 | ||
2142 | return ret; | |
2143 | } | |
2144 | ||
2145 | /* | |
2146 | * simple code to zero out a cluster | |
2147 | */ | |
2148 | void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster) | |
2149 | { | |
2150 | spin_lock_init(&cluster->lock); | |
2151 | spin_lock_init(&cluster->refill_lock); | |
6bef4d31 | 2152 | cluster->root = RB_ROOT; |
fa9c0d79 | 2153 | cluster->max_size = 0; |
96303081 | 2154 | cluster->points_to_bitmap = false; |
fa9c0d79 CM |
2155 | INIT_LIST_HEAD(&cluster->block_group_list); |
2156 | cluster->block_group = NULL; | |
2157 | } | |
2158 |