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> |
6ab60601 | 23 | #include <linux/ratelimit.h> |
0f9dd46c | 24 | #include "ctree.h" |
fa9c0d79 CM |
25 | #include "free-space-cache.h" |
26 | #include "transaction.h" | |
0af3d00b | 27 | #include "disk-io.h" |
43be2146 | 28 | #include "extent_io.h" |
581bb050 | 29 | #include "inode-map.h" |
fa9c0d79 | 30 | |
96303081 JB |
31 | #define BITS_PER_BITMAP (PAGE_CACHE_SIZE * 8) |
32 | #define MAX_CACHE_BYTES_PER_GIG (32 * 1024) | |
0f9dd46c | 33 | |
34d52cb6 | 34 | static int link_free_space(struct btrfs_free_space_ctl *ctl, |
0cb59c99 JB |
35 | struct btrfs_free_space *info); |
36 | ||
0414efae LZ |
37 | static struct inode *__lookup_free_space_inode(struct btrfs_root *root, |
38 | struct btrfs_path *path, | |
39 | u64 offset) | |
0af3d00b JB |
40 | { |
41 | struct btrfs_key key; | |
42 | struct btrfs_key location; | |
43 | struct btrfs_disk_key disk_key; | |
44 | struct btrfs_free_space_header *header; | |
45 | struct extent_buffer *leaf; | |
46 | struct inode *inode = NULL; | |
47 | int ret; | |
48 | ||
0af3d00b | 49 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; |
0414efae | 50 | key.offset = offset; |
0af3d00b JB |
51 | key.type = 0; |
52 | ||
53 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
54 | if (ret < 0) | |
55 | return ERR_PTR(ret); | |
56 | if (ret > 0) { | |
b3b4aa74 | 57 | btrfs_release_path(path); |
0af3d00b JB |
58 | return ERR_PTR(-ENOENT); |
59 | } | |
60 | ||
61 | leaf = path->nodes[0]; | |
62 | header = btrfs_item_ptr(leaf, path->slots[0], | |
63 | struct btrfs_free_space_header); | |
64 | btrfs_free_space_key(leaf, header, &disk_key); | |
65 | btrfs_disk_key_to_cpu(&location, &disk_key); | |
b3b4aa74 | 66 | btrfs_release_path(path); |
0af3d00b JB |
67 | |
68 | inode = btrfs_iget(root->fs_info->sb, &location, root, NULL); | |
69 | if (!inode) | |
70 | return ERR_PTR(-ENOENT); | |
71 | if (IS_ERR(inode)) | |
72 | return inode; | |
73 | if (is_bad_inode(inode)) { | |
74 | iput(inode); | |
75 | return ERR_PTR(-ENOENT); | |
76 | } | |
77 | ||
adae52b9 MX |
78 | inode->i_mapping->flags &= ~__GFP_FS; |
79 | ||
0414efae LZ |
80 | return inode; |
81 | } | |
82 | ||
83 | struct inode *lookup_free_space_inode(struct btrfs_root *root, | |
84 | struct btrfs_block_group_cache | |
85 | *block_group, struct btrfs_path *path) | |
86 | { | |
87 | struct inode *inode = NULL; | |
88 | ||
89 | spin_lock(&block_group->lock); | |
90 | if (block_group->inode) | |
91 | inode = igrab(block_group->inode); | |
92 | spin_unlock(&block_group->lock); | |
93 | if (inode) | |
94 | return inode; | |
95 | ||
96 | inode = __lookup_free_space_inode(root, path, | |
97 | block_group->key.objectid); | |
98 | if (IS_ERR(inode)) | |
99 | return inode; | |
100 | ||
0af3d00b | 101 | spin_lock(&block_group->lock); |
2f356126 JB |
102 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) { |
103 | printk(KERN_INFO "Old style space inode found, converting.\n"); | |
104 | BTRFS_I(inode)->flags &= ~BTRFS_INODE_NODATASUM; | |
105 | block_group->disk_cache_state = BTRFS_DC_CLEAR; | |
106 | } | |
107 | ||
300e4f8a | 108 | if (!block_group->iref) { |
0af3d00b JB |
109 | block_group->inode = igrab(inode); |
110 | block_group->iref = 1; | |
111 | } | |
112 | spin_unlock(&block_group->lock); | |
113 | ||
114 | return inode; | |
115 | } | |
116 | ||
0414efae LZ |
117 | int __create_free_space_inode(struct btrfs_root *root, |
118 | struct btrfs_trans_handle *trans, | |
119 | struct btrfs_path *path, u64 ino, u64 offset) | |
0af3d00b JB |
120 | { |
121 | struct btrfs_key key; | |
122 | struct btrfs_disk_key disk_key; | |
123 | struct btrfs_free_space_header *header; | |
124 | struct btrfs_inode_item *inode_item; | |
125 | struct extent_buffer *leaf; | |
0af3d00b JB |
126 | int ret; |
127 | ||
0414efae | 128 | ret = btrfs_insert_empty_inode(trans, root, path, ino); |
0af3d00b JB |
129 | if (ret) |
130 | return ret; | |
131 | ||
132 | leaf = path->nodes[0]; | |
133 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
134 | struct btrfs_inode_item); | |
135 | btrfs_item_key(leaf, &disk_key, path->slots[0]); | |
136 | memset_extent_buffer(leaf, 0, (unsigned long)inode_item, | |
137 | sizeof(*inode_item)); | |
138 | btrfs_set_inode_generation(leaf, inode_item, trans->transid); | |
139 | btrfs_set_inode_size(leaf, inode_item, 0); | |
140 | btrfs_set_inode_nbytes(leaf, inode_item, 0); | |
141 | btrfs_set_inode_uid(leaf, inode_item, 0); | |
142 | btrfs_set_inode_gid(leaf, inode_item, 0); | |
143 | btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600); | |
144 | btrfs_set_inode_flags(leaf, inode_item, BTRFS_INODE_NOCOMPRESS | | |
2f356126 | 145 | BTRFS_INODE_PREALLOC); |
0af3d00b JB |
146 | btrfs_set_inode_nlink(leaf, inode_item, 1); |
147 | btrfs_set_inode_transid(leaf, inode_item, trans->transid); | |
0414efae | 148 | btrfs_set_inode_block_group(leaf, inode_item, offset); |
0af3d00b | 149 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 150 | btrfs_release_path(path); |
0af3d00b JB |
151 | |
152 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
0414efae | 153 | key.offset = offset; |
0af3d00b JB |
154 | key.type = 0; |
155 | ||
156 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
157 | sizeof(struct btrfs_free_space_header)); | |
158 | if (ret < 0) { | |
b3b4aa74 | 159 | btrfs_release_path(path); |
0af3d00b JB |
160 | return ret; |
161 | } | |
162 | leaf = path->nodes[0]; | |
163 | header = btrfs_item_ptr(leaf, path->slots[0], | |
164 | struct btrfs_free_space_header); | |
165 | memset_extent_buffer(leaf, 0, (unsigned long)header, sizeof(*header)); | |
166 | btrfs_set_free_space_key(leaf, header, &disk_key); | |
167 | btrfs_mark_buffer_dirty(leaf); | |
b3b4aa74 | 168 | btrfs_release_path(path); |
0af3d00b JB |
169 | |
170 | return 0; | |
171 | } | |
172 | ||
0414efae LZ |
173 | int create_free_space_inode(struct btrfs_root *root, |
174 | struct btrfs_trans_handle *trans, | |
175 | struct btrfs_block_group_cache *block_group, | |
176 | struct btrfs_path *path) | |
177 | { | |
178 | int ret; | |
179 | u64 ino; | |
180 | ||
181 | ret = btrfs_find_free_objectid(root, &ino); | |
182 | if (ret < 0) | |
183 | return ret; | |
184 | ||
185 | return __create_free_space_inode(root, trans, path, ino, | |
186 | block_group->key.objectid); | |
187 | } | |
188 | ||
0af3d00b JB |
189 | int btrfs_truncate_free_space_cache(struct btrfs_root *root, |
190 | struct btrfs_trans_handle *trans, | |
191 | struct btrfs_path *path, | |
192 | struct inode *inode) | |
193 | { | |
65450aa6 | 194 | struct btrfs_block_rsv *rsv; |
0af3d00b JB |
195 | loff_t oldsize; |
196 | int ret = 0; | |
197 | ||
65450aa6 | 198 | rsv = trans->block_rsv; |
0af3d00b | 199 | trans->block_rsv = root->orphan_block_rsv; |
4a92b1b8 | 200 | ret = btrfs_block_rsv_check(root, root->orphan_block_rsv, 0, 5, 0); |
0af3d00b JB |
201 | if (ret) |
202 | return ret; | |
203 | ||
204 | oldsize = i_size_read(inode); | |
205 | btrfs_i_size_write(inode, 0); | |
206 | truncate_pagecache(inode, oldsize, 0); | |
207 | ||
208 | /* | |
209 | * We don't need an orphan item because truncating the free space cache | |
210 | * will never be split across transactions. | |
211 | */ | |
212 | ret = btrfs_truncate_inode_items(trans, root, inode, | |
213 | 0, BTRFS_EXTENT_DATA_KEY); | |
65450aa6 LB |
214 | |
215 | trans->block_rsv = rsv; | |
0af3d00b JB |
216 | if (ret) { |
217 | WARN_ON(1); | |
218 | return ret; | |
219 | } | |
220 | ||
82d5902d LZ |
221 | ret = btrfs_update_inode(trans, root, inode); |
222 | return ret; | |
0af3d00b JB |
223 | } |
224 | ||
9d66e233 JB |
225 | static int readahead_cache(struct inode *inode) |
226 | { | |
227 | struct file_ra_state *ra; | |
228 | unsigned long last_index; | |
229 | ||
230 | ra = kzalloc(sizeof(*ra), GFP_NOFS); | |
231 | if (!ra) | |
232 | return -ENOMEM; | |
233 | ||
234 | file_ra_state_init(ra, inode->i_mapping); | |
235 | last_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT; | |
236 | ||
237 | page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index); | |
238 | ||
239 | kfree(ra); | |
240 | ||
241 | return 0; | |
242 | } | |
243 | ||
0414efae LZ |
244 | int __load_free_space_cache(struct btrfs_root *root, struct inode *inode, |
245 | struct btrfs_free_space_ctl *ctl, | |
246 | struct btrfs_path *path, u64 offset) | |
9d66e233 | 247 | { |
9d66e233 JB |
248 | struct btrfs_free_space_header *header; |
249 | struct extent_buffer *leaf; | |
250 | struct page *page; | |
9d66e233 JB |
251 | struct btrfs_key key; |
252 | struct list_head bitmaps; | |
253 | u64 num_entries; | |
254 | u64 num_bitmaps; | |
255 | u64 generation; | |
9d66e233 | 256 | pgoff_t index = 0; |
3b16a4e3 | 257 | gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); |
f6a39829 | 258 | int ret = 0; |
9d66e233 JB |
259 | |
260 | INIT_LIST_HEAD(&bitmaps); | |
261 | ||
9d66e233 | 262 | /* Nothing in the space cache, goodbye */ |
0414efae | 263 | if (!i_size_read(inode)) |
9d66e233 | 264 | goto out; |
9d66e233 JB |
265 | |
266 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
0414efae | 267 | key.offset = offset; |
9d66e233 JB |
268 | key.type = 0; |
269 | ||
270 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
0414efae LZ |
271 | if (ret < 0) |
272 | goto out; | |
273 | else if (ret > 0) { | |
945d8962 | 274 | btrfs_release_path(path); |
0414efae | 275 | ret = 0; |
9d66e233 JB |
276 | goto out; |
277 | } | |
278 | ||
0414efae LZ |
279 | ret = -1; |
280 | ||
9d66e233 JB |
281 | leaf = path->nodes[0]; |
282 | header = btrfs_item_ptr(leaf, path->slots[0], | |
283 | struct btrfs_free_space_header); | |
284 | num_entries = btrfs_free_space_entries(leaf, header); | |
285 | num_bitmaps = btrfs_free_space_bitmaps(leaf, header); | |
286 | generation = btrfs_free_space_generation(leaf, header); | |
945d8962 | 287 | btrfs_release_path(path); |
9d66e233 JB |
288 | |
289 | if (BTRFS_I(inode)->generation != generation) { | |
290 | printk(KERN_ERR "btrfs: free space inode generation (%llu) did" | |
0414efae | 291 | " not match free space cache generation (%llu)\n", |
9d66e233 | 292 | (unsigned long long)BTRFS_I(inode)->generation, |
0414efae LZ |
293 | (unsigned long long)generation); |
294 | goto out; | |
9d66e233 JB |
295 | } |
296 | ||
297 | if (!num_entries) | |
298 | goto out; | |
299 | ||
9d66e233 | 300 | ret = readahead_cache(inode); |
0414efae | 301 | if (ret) |
9d66e233 | 302 | goto out; |
9d66e233 JB |
303 | |
304 | while (1) { | |
305 | struct btrfs_free_space_entry *entry; | |
306 | struct btrfs_free_space *e; | |
307 | void *addr; | |
308 | unsigned long offset = 0; | |
9d66e233 JB |
309 | int need_loop = 0; |
310 | ||
311 | if (!num_entries && !num_bitmaps) | |
312 | break; | |
313 | ||
3b16a4e3 | 314 | page = find_or_create_page(inode->i_mapping, index, mask); |
0414efae | 315 | if (!page) |
9d66e233 | 316 | goto free_cache; |
9d66e233 JB |
317 | |
318 | if (!PageUptodate(page)) { | |
319 | btrfs_readpage(NULL, page); | |
320 | lock_page(page); | |
321 | if (!PageUptodate(page)) { | |
322 | unlock_page(page); | |
323 | page_cache_release(page); | |
324 | printk(KERN_ERR "btrfs: error reading free " | |
0414efae | 325 | "space cache\n"); |
9d66e233 JB |
326 | goto free_cache; |
327 | } | |
328 | } | |
329 | addr = kmap(page); | |
330 | ||
331 | if (index == 0) { | |
332 | u64 *gen; | |
333 | ||
2f356126 JB |
334 | /* |
335 | * We put a bogus crc in the front of the first page in | |
336 | * case old kernels try to mount a fs with the new | |
337 | * format to make sure they discard the cache. | |
338 | */ | |
339 | addr += sizeof(u64); | |
340 | offset += sizeof(u64); | |
341 | ||
342 | gen = addr; | |
9d66e233 | 343 | if (*gen != BTRFS_I(inode)->generation) { |
6ab60601 JB |
344 | printk_ratelimited(KERN_ERR "btrfs: space cache" |
345 | " generation (%llu) does not match " | |
346 | "inode (%llu)\n", | |
347 | (unsigned long long)*gen, | |
348 | (unsigned long long) | |
349 | BTRFS_I(inode)->generation); | |
9d66e233 JB |
350 | kunmap(page); |
351 | unlock_page(page); | |
352 | page_cache_release(page); | |
353 | goto free_cache; | |
354 | } | |
2f356126 JB |
355 | addr += sizeof(u64); |
356 | offset += sizeof(u64); | |
9d66e233 | 357 | } |
2f356126 | 358 | entry = addr; |
9d66e233 JB |
359 | |
360 | while (1) { | |
361 | if (!num_entries) | |
362 | break; | |
363 | ||
364 | need_loop = 1; | |
dc89e982 JB |
365 | e = kmem_cache_zalloc(btrfs_free_space_cachep, |
366 | GFP_NOFS); | |
9d66e233 JB |
367 | if (!e) { |
368 | kunmap(page); | |
369 | unlock_page(page); | |
370 | page_cache_release(page); | |
371 | goto free_cache; | |
372 | } | |
373 | ||
374 | e->offset = le64_to_cpu(entry->offset); | |
375 | e->bytes = le64_to_cpu(entry->bytes); | |
376 | if (!e->bytes) { | |
377 | kunmap(page); | |
dc89e982 | 378 | kmem_cache_free(btrfs_free_space_cachep, e); |
9d66e233 JB |
379 | unlock_page(page); |
380 | page_cache_release(page); | |
381 | goto free_cache; | |
382 | } | |
383 | ||
384 | if (entry->type == BTRFS_FREE_SPACE_EXTENT) { | |
34d52cb6 LZ |
385 | spin_lock(&ctl->tree_lock); |
386 | ret = link_free_space(ctl, e); | |
387 | spin_unlock(&ctl->tree_lock); | |
207dde82 JB |
388 | if (ret) { |
389 | printk(KERN_ERR "Duplicate entries in " | |
390 | "free space cache, dumping\n"); | |
391 | kunmap(page); | |
392 | unlock_page(page); | |
393 | page_cache_release(page); | |
394 | goto free_cache; | |
395 | } | |
9d66e233 JB |
396 | } else { |
397 | e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS); | |
398 | if (!e->bitmap) { | |
399 | kunmap(page); | |
dc89e982 JB |
400 | kmem_cache_free( |
401 | btrfs_free_space_cachep, e); | |
9d66e233 JB |
402 | unlock_page(page); |
403 | page_cache_release(page); | |
404 | goto free_cache; | |
405 | } | |
34d52cb6 | 406 | spin_lock(&ctl->tree_lock); |
f6a39829 | 407 | ret = link_free_space(ctl, e); |
34d52cb6 LZ |
408 | ctl->total_bitmaps++; |
409 | ctl->op->recalc_thresholds(ctl); | |
410 | spin_unlock(&ctl->tree_lock); | |
207dde82 JB |
411 | if (ret) { |
412 | printk(KERN_ERR "Duplicate entries in " | |
413 | "free space cache, dumping\n"); | |
414 | kunmap(page); | |
415 | unlock_page(page); | |
416 | page_cache_release(page); | |
417 | goto free_cache; | |
418 | } | |
f6a39829 | 419 | list_add_tail(&e->list, &bitmaps); |
9d66e233 JB |
420 | } |
421 | ||
422 | num_entries--; | |
423 | offset += sizeof(struct btrfs_free_space_entry); | |
424 | if (offset + sizeof(struct btrfs_free_space_entry) >= | |
425 | PAGE_CACHE_SIZE) | |
426 | break; | |
427 | entry++; | |
428 | } | |
429 | ||
430 | /* | |
431 | * We read an entry out of this page, we need to move on to the | |
432 | * next page. | |
433 | */ | |
434 | if (need_loop) { | |
435 | kunmap(page); | |
436 | goto next; | |
437 | } | |
438 | ||
439 | /* | |
440 | * We add the bitmaps at the end of the entries in order that | |
441 | * the bitmap entries are added to the cache. | |
442 | */ | |
443 | e = list_entry(bitmaps.next, struct btrfs_free_space, list); | |
444 | list_del_init(&e->list); | |
445 | memcpy(e->bitmap, addr, PAGE_CACHE_SIZE); | |
446 | kunmap(page); | |
447 | num_bitmaps--; | |
448 | next: | |
449 | unlock_page(page); | |
450 | page_cache_release(page); | |
451 | index++; | |
452 | } | |
453 | ||
454 | ret = 1; | |
455 | out: | |
9d66e233 | 456 | return ret; |
9d66e233 | 457 | free_cache: |
0414efae | 458 | __btrfs_remove_free_space_cache(ctl); |
9d66e233 JB |
459 | goto out; |
460 | } | |
461 | ||
0414efae LZ |
462 | int load_free_space_cache(struct btrfs_fs_info *fs_info, |
463 | struct btrfs_block_group_cache *block_group) | |
0cb59c99 | 464 | { |
34d52cb6 | 465 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0414efae LZ |
466 | struct btrfs_root *root = fs_info->tree_root; |
467 | struct inode *inode; | |
468 | struct btrfs_path *path; | |
469 | int ret; | |
470 | bool matched; | |
471 | u64 used = btrfs_block_group_used(&block_group->item); | |
472 | ||
473 | /* | |
474 | * If we're unmounting then just return, since this does a search on the | |
475 | * normal root and not the commit root and we could deadlock. | |
476 | */ | |
7841cb28 | 477 | if (btrfs_fs_closing(fs_info)) |
0414efae LZ |
478 | return 0; |
479 | ||
480 | /* | |
481 | * If this block group has been marked to be cleared for one reason or | |
482 | * another then we can't trust the on disk cache, so just return. | |
483 | */ | |
9d66e233 | 484 | spin_lock(&block_group->lock); |
0414efae LZ |
485 | if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) { |
486 | spin_unlock(&block_group->lock); | |
487 | return 0; | |
488 | } | |
9d66e233 | 489 | spin_unlock(&block_group->lock); |
0414efae LZ |
490 | |
491 | path = btrfs_alloc_path(); | |
492 | if (!path) | |
493 | return 0; | |
494 | ||
495 | inode = lookup_free_space_inode(root, block_group, path); | |
496 | if (IS_ERR(inode)) { | |
497 | btrfs_free_path(path); | |
498 | return 0; | |
499 | } | |
500 | ||
501 | ret = __load_free_space_cache(fs_info->tree_root, inode, ctl, | |
502 | path, block_group->key.objectid); | |
503 | btrfs_free_path(path); | |
504 | if (ret <= 0) | |
505 | goto out; | |
506 | ||
507 | spin_lock(&ctl->tree_lock); | |
508 | matched = (ctl->free_space == (block_group->key.offset - used - | |
509 | block_group->bytes_super)); | |
510 | spin_unlock(&ctl->tree_lock); | |
511 | ||
512 | if (!matched) { | |
513 | __btrfs_remove_free_space_cache(ctl); | |
514 | printk(KERN_ERR "block group %llu has an wrong amount of free " | |
515 | "space\n", block_group->key.objectid); | |
516 | ret = -1; | |
517 | } | |
518 | out: | |
519 | if (ret < 0) { | |
520 | /* This cache is bogus, make sure it gets cleared */ | |
521 | spin_lock(&block_group->lock); | |
522 | block_group->disk_cache_state = BTRFS_DC_CLEAR; | |
523 | spin_unlock(&block_group->lock); | |
82d5902d | 524 | ret = 0; |
0414efae LZ |
525 | |
526 | printk(KERN_ERR "btrfs: failed to load free space cache " | |
527 | "for block group %llu\n", block_group->key.objectid); | |
528 | } | |
529 | ||
530 | iput(inode); | |
531 | return ret; | |
9d66e233 JB |
532 | } |
533 | ||
c09544e0 JB |
534 | /** |
535 | * __btrfs_write_out_cache - write out cached info to an inode | |
536 | * @root - the root the inode belongs to | |
537 | * @ctl - the free space cache we are going to write out | |
538 | * @block_group - the block_group for this cache if it belongs to a block_group | |
539 | * @trans - the trans handle | |
540 | * @path - the path to use | |
541 | * @offset - the offset for the key we'll insert | |
542 | * | |
543 | * This function writes out a free space cache struct to disk for quick recovery | |
544 | * on mount. This will return 0 if it was successfull in writing the cache out, | |
545 | * and -1 if it was not. | |
546 | */ | |
0414efae LZ |
547 | int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode, |
548 | struct btrfs_free_space_ctl *ctl, | |
549 | struct btrfs_block_group_cache *block_group, | |
550 | struct btrfs_trans_handle *trans, | |
551 | struct btrfs_path *path, u64 offset) | |
0cb59c99 JB |
552 | { |
553 | struct btrfs_free_space_header *header; | |
554 | struct extent_buffer *leaf; | |
0cb59c99 JB |
555 | struct rb_node *node; |
556 | struct list_head *pos, *n; | |
be1a12a0 | 557 | struct page **pages; |
0cb59c99 JB |
558 | struct page *page; |
559 | struct extent_state *cached_state = NULL; | |
43be2146 JB |
560 | struct btrfs_free_cluster *cluster = NULL; |
561 | struct extent_io_tree *unpin = NULL; | |
0cb59c99 JB |
562 | struct list_head bitmap_list; |
563 | struct btrfs_key key; | |
43be2146 | 564 | u64 start, end, len; |
0cb59c99 | 565 | u64 bytes = 0; |
2f356126 | 566 | u32 crc = ~(u32)0; |
3b16a4e3 | 567 | gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); |
be1a12a0 | 568 | int index = 0, num_pages = 0; |
0cb59c99 JB |
569 | int entries = 0; |
570 | int bitmaps = 0; | |
c09544e0 JB |
571 | int ret; |
572 | int err = -1; | |
43be2146 | 573 | bool next_page = false; |
be1a12a0 | 574 | bool out_of_space = false; |
0cb59c99 | 575 | |
0cb59c99 JB |
576 | INIT_LIST_HEAD(&bitmap_list); |
577 | ||
34d52cb6 | 578 | node = rb_first(&ctl->free_space_offset); |
0414efae | 579 | if (!node) |
c09544e0 | 580 | return -1; |
0cb59c99 | 581 | |
0414efae LZ |
582 | if (!i_size_read(inode)) |
583 | return -1; | |
2b20982e | 584 | |
be1a12a0 JB |
585 | num_pages = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> |
586 | PAGE_CACHE_SHIFT; | |
211f96c2 | 587 | |
0cb59c99 JB |
588 | filemap_write_and_wait(inode->i_mapping); |
589 | btrfs_wait_ordered_range(inode, inode->i_size & | |
590 | ~(root->sectorsize - 1), (u64)-1); | |
591 | ||
be1a12a0 | 592 | pages = kzalloc(sizeof(struct page *) * num_pages, GFP_NOFS); |
2f356126 | 593 | if (!pages) |
0414efae | 594 | return -1; |
be1a12a0 | 595 | |
43be2146 | 596 | /* Get the cluster for this block_group if it exists */ |
0414efae | 597 | if (block_group && !list_empty(&block_group->cluster_list)) |
43be2146 JB |
598 | cluster = list_entry(block_group->cluster_list.next, |
599 | struct btrfs_free_cluster, | |
600 | block_group_list); | |
601 | ||
602 | /* | |
603 | * We shouldn't have switched the pinned extents yet so this is the | |
604 | * right one | |
605 | */ | |
606 | unpin = root->fs_info->pinned_extents; | |
607 | ||
0cb59c99 JB |
608 | /* |
609 | * Lock all pages first so we can lock the extent safely. | |
610 | * | |
611 | * NOTE: Because we hold the ref the entire time we're going to write to | |
612 | * the page find_get_page should never fail, so we don't do a check | |
613 | * after find_get_page at this point. Just putting this here so people | |
614 | * know and don't freak out. | |
615 | */ | |
be1a12a0 | 616 | while (index < num_pages) { |
3b16a4e3 | 617 | page = find_or_create_page(inode->i_mapping, index, mask); |
0cb59c99 | 618 | if (!page) { |
be1a12a0 | 619 | int i; |
0cb59c99 | 620 | |
be1a12a0 JB |
621 | for (i = 0; i < num_pages; i++) { |
622 | unlock_page(pages[i]); | |
623 | page_cache_release(pages[i]); | |
0cb59c99 | 624 | } |
2f356126 | 625 | goto out; |
0cb59c99 | 626 | } |
be1a12a0 | 627 | pages[index] = page; |
0cb59c99 JB |
628 | index++; |
629 | } | |
630 | ||
631 | index = 0; | |
632 | lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1, | |
633 | 0, &cached_state, GFP_NOFS); | |
634 | ||
43be2146 JB |
635 | /* |
636 | * When searching for pinned extents, we need to start at our start | |
637 | * offset. | |
638 | */ | |
0414efae LZ |
639 | if (block_group) |
640 | start = block_group->key.objectid; | |
43be2146 | 641 | |
0cb59c99 JB |
642 | /* Write out the extent entries */ |
643 | do { | |
644 | struct btrfs_free_space_entry *entry; | |
2f356126 | 645 | void *addr, *orig; |
0cb59c99 | 646 | unsigned long offset = 0; |
0cb59c99 | 647 | |
43be2146 JB |
648 | next_page = false; |
649 | ||
be1a12a0 JB |
650 | if (index >= num_pages) { |
651 | out_of_space = true; | |
652 | break; | |
653 | } | |
654 | ||
655 | page = pages[index]; | |
0cb59c99 | 656 | |
2f356126 JB |
657 | orig = addr = kmap(page); |
658 | if (index == 0) { | |
659 | u64 *gen; | |
0cb59c99 | 660 | |
2f356126 JB |
661 | /* |
662 | * We're going to put in a bogus crc for this page to | |
663 | * make sure that old kernels who aren't aware of this | |
664 | * format will be sure to discard the cache. | |
665 | */ | |
666 | addr += sizeof(u64); | |
667 | offset += sizeof(u64); | |
668 | ||
669 | gen = addr; | |
670 | *gen = trans->transid; | |
671 | addr += sizeof(u64); | |
672 | offset += sizeof(u64); | |
673 | } | |
674 | entry = addr; | |
675 | ||
676 | memset(addr, 0, PAGE_CACHE_SIZE - offset); | |
43be2146 | 677 | while (node && !next_page) { |
0cb59c99 JB |
678 | struct btrfs_free_space *e; |
679 | ||
680 | e = rb_entry(node, struct btrfs_free_space, offset_index); | |
681 | entries++; | |
682 | ||
683 | entry->offset = cpu_to_le64(e->offset); | |
684 | entry->bytes = cpu_to_le64(e->bytes); | |
685 | if (e->bitmap) { | |
686 | entry->type = BTRFS_FREE_SPACE_BITMAP; | |
687 | list_add_tail(&e->list, &bitmap_list); | |
688 | bitmaps++; | |
689 | } else { | |
690 | entry->type = BTRFS_FREE_SPACE_EXTENT; | |
691 | } | |
692 | node = rb_next(node); | |
43be2146 JB |
693 | if (!node && cluster) { |
694 | node = rb_first(&cluster->root); | |
695 | cluster = NULL; | |
696 | } | |
0cb59c99 JB |
697 | offset += sizeof(struct btrfs_free_space_entry); |
698 | if (offset + sizeof(struct btrfs_free_space_entry) >= | |
699 | PAGE_CACHE_SIZE) | |
43be2146 JB |
700 | next_page = true; |
701 | entry++; | |
702 | } | |
703 | ||
704 | /* | |
705 | * We want to add any pinned extents to our free space cache | |
706 | * so we don't leak the space | |
707 | */ | |
0414efae LZ |
708 | while (block_group && !next_page && |
709 | (start < block_group->key.objectid + | |
710 | block_group->key.offset)) { | |
43be2146 JB |
711 | ret = find_first_extent_bit(unpin, start, &start, &end, |
712 | EXTENT_DIRTY); | |
713 | if (ret) { | |
714 | ret = 0; | |
715 | break; | |
716 | } | |
717 | ||
718 | /* This pinned extent is out of our range */ | |
719 | if (start >= block_group->key.objectid + | |
720 | block_group->key.offset) | |
0cb59c99 | 721 | break; |
43be2146 JB |
722 | |
723 | len = block_group->key.objectid + | |
724 | block_group->key.offset - start; | |
725 | len = min(len, end + 1 - start); | |
726 | ||
727 | entries++; | |
728 | entry->offset = cpu_to_le64(start); | |
729 | entry->bytes = cpu_to_le64(len); | |
730 | entry->type = BTRFS_FREE_SPACE_EXTENT; | |
731 | ||
732 | start = end + 1; | |
733 | offset += sizeof(struct btrfs_free_space_entry); | |
734 | if (offset + sizeof(struct btrfs_free_space_entry) >= | |
735 | PAGE_CACHE_SIZE) | |
736 | next_page = true; | |
0cb59c99 JB |
737 | entry++; |
738 | } | |
0cb59c99 | 739 | |
2f356126 JB |
740 | /* Generate bogus crc value */ |
741 | if (index == 0) { | |
742 | u32 *tmp; | |
743 | crc = btrfs_csum_data(root, orig + sizeof(u64), crc, | |
744 | PAGE_CACHE_SIZE - sizeof(u64)); | |
745 | btrfs_csum_final(crc, (char *)&crc); | |
746 | crc++; | |
747 | tmp = orig; | |
748 | *tmp = crc; | |
749 | } | |
750 | ||
751 | kunmap(page); | |
0cb59c99 JB |
752 | |
753 | bytes += PAGE_CACHE_SIZE; | |
754 | ||
0cb59c99 | 755 | index++; |
43be2146 | 756 | } while (node || next_page); |
0cb59c99 JB |
757 | |
758 | /* Write out the bitmaps */ | |
759 | list_for_each_safe(pos, n, &bitmap_list) { | |
760 | void *addr; | |
761 | struct btrfs_free_space *entry = | |
762 | list_entry(pos, struct btrfs_free_space, list); | |
763 | ||
be1a12a0 JB |
764 | if (index >= num_pages) { |
765 | out_of_space = true; | |
766 | break; | |
767 | } | |
f65647c2 | 768 | page = pages[index]; |
0cb59c99 JB |
769 | |
770 | addr = kmap(page); | |
771 | memcpy(addr, entry->bitmap, PAGE_CACHE_SIZE); | |
0cb59c99 | 772 | kunmap(page); |
0cb59c99 JB |
773 | bytes += PAGE_CACHE_SIZE; |
774 | ||
0cb59c99 JB |
775 | list_del_init(&entry->list); |
776 | index++; | |
777 | } | |
778 | ||
be1a12a0 JB |
779 | if (out_of_space) { |
780 | btrfs_drop_pages(pages, num_pages); | |
781 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0, | |
782 | i_size_read(inode) - 1, &cached_state, | |
783 | GFP_NOFS); | |
2f356126 | 784 | goto out; |
be1a12a0 JB |
785 | } |
786 | ||
0cb59c99 | 787 | /* Zero out the rest of the pages just to make sure */ |
be1a12a0 | 788 | while (index < num_pages) { |
0cb59c99 JB |
789 | void *addr; |
790 | ||
be1a12a0 | 791 | page = pages[index]; |
0cb59c99 JB |
792 | addr = kmap(page); |
793 | memset(addr, 0, PAGE_CACHE_SIZE); | |
794 | kunmap(page); | |
0cb59c99 JB |
795 | bytes += PAGE_CACHE_SIZE; |
796 | index++; | |
797 | } | |
798 | ||
be1a12a0 JB |
799 | ret = btrfs_dirty_pages(root, inode, pages, num_pages, 0, |
800 | bytes, &cached_state); | |
801 | btrfs_drop_pages(pages, num_pages); | |
0cb59c99 JB |
802 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0, |
803 | i_size_read(inode) - 1, &cached_state, GFP_NOFS); | |
804 | ||
c09544e0 | 805 | if (ret) |
2f356126 | 806 | goto out; |
be1a12a0 JB |
807 | |
808 | BTRFS_I(inode)->generation = trans->transid; | |
809 | ||
0cb59c99 JB |
810 | filemap_write_and_wait(inode->i_mapping); |
811 | ||
812 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
0414efae | 813 | key.offset = offset; |
0cb59c99 JB |
814 | key.type = 0; |
815 | ||
a9b5fcdd | 816 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
0cb59c99 | 817 | if (ret < 0) { |
0cb59c99 JB |
818 | clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, bytes - 1, |
819 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
820 | EXTENT_DO_ACCOUNTING, 0, 0, NULL, GFP_NOFS); | |
2f356126 | 821 | goto out; |
0cb59c99 JB |
822 | } |
823 | leaf = path->nodes[0]; | |
824 | if (ret > 0) { | |
825 | struct btrfs_key found_key; | |
826 | BUG_ON(!path->slots[0]); | |
827 | path->slots[0]--; | |
828 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
829 | if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID || | |
0414efae | 830 | found_key.offset != offset) { |
0cb59c99 JB |
831 | clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, bytes - 1, |
832 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
833 | EXTENT_DO_ACCOUNTING, 0, 0, NULL, | |
834 | GFP_NOFS); | |
b3b4aa74 | 835 | btrfs_release_path(path); |
2f356126 | 836 | goto out; |
0cb59c99 JB |
837 | } |
838 | } | |
839 | header = btrfs_item_ptr(leaf, path->slots[0], | |
840 | struct btrfs_free_space_header); | |
841 | btrfs_set_free_space_entries(leaf, header, entries); | |
842 | btrfs_set_free_space_bitmaps(leaf, header, bitmaps); | |
843 | btrfs_set_free_space_generation(leaf, header, trans->transid); | |
844 | btrfs_mark_buffer_dirty(leaf); | |
b3b4aa74 | 845 | btrfs_release_path(path); |
0cb59c99 | 846 | |
c09544e0 | 847 | err = 0; |
2f356126 | 848 | out: |
211f96c2 | 849 | kfree(pages); |
c09544e0 | 850 | if (err) { |
0cb59c99 | 851 | invalidate_inode_pages2_range(inode->i_mapping, 0, index); |
0cb59c99 JB |
852 | BTRFS_I(inode)->generation = 0; |
853 | } | |
0cb59c99 | 854 | btrfs_update_inode(trans, root, inode); |
c09544e0 | 855 | return err; |
0414efae LZ |
856 | } |
857 | ||
858 | int btrfs_write_out_cache(struct btrfs_root *root, | |
859 | struct btrfs_trans_handle *trans, | |
860 | struct btrfs_block_group_cache *block_group, | |
861 | struct btrfs_path *path) | |
862 | { | |
863 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
864 | struct inode *inode; | |
865 | int ret = 0; | |
866 | ||
867 | root = root->fs_info->tree_root; | |
868 | ||
869 | spin_lock(&block_group->lock); | |
870 | if (block_group->disk_cache_state < BTRFS_DC_SETUP) { | |
871 | spin_unlock(&block_group->lock); | |
872 | return 0; | |
873 | } | |
874 | spin_unlock(&block_group->lock); | |
875 | ||
876 | inode = lookup_free_space_inode(root, block_group, path); | |
877 | if (IS_ERR(inode)) | |
878 | return 0; | |
879 | ||
880 | ret = __btrfs_write_out_cache(root, inode, ctl, block_group, trans, | |
881 | path, block_group->key.objectid); | |
c09544e0 JB |
882 | if (ret) { |
883 | btrfs_delalloc_release_metadata(inode, inode->i_size); | |
0414efae LZ |
884 | spin_lock(&block_group->lock); |
885 | block_group->disk_cache_state = BTRFS_DC_ERROR; | |
886 | spin_unlock(&block_group->lock); | |
82d5902d | 887 | ret = 0; |
c09544e0 | 888 | #ifdef DEBUG |
0414efae LZ |
889 | printk(KERN_ERR "btrfs: failed to write free space cace " |
890 | "for block group %llu\n", block_group->key.objectid); | |
c09544e0 | 891 | #endif |
0414efae LZ |
892 | } |
893 | ||
0cb59c99 JB |
894 | iput(inode); |
895 | return ret; | |
896 | } | |
897 | ||
34d52cb6 | 898 | static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit, |
96303081 | 899 | u64 offset) |
0f9dd46c | 900 | { |
96303081 JB |
901 | BUG_ON(offset < bitmap_start); |
902 | offset -= bitmap_start; | |
34d52cb6 | 903 | return (unsigned long)(div_u64(offset, unit)); |
96303081 | 904 | } |
0f9dd46c | 905 | |
34d52cb6 | 906 | static inline unsigned long bytes_to_bits(u64 bytes, u32 unit) |
96303081 | 907 | { |
34d52cb6 | 908 | return (unsigned long)(div_u64(bytes, unit)); |
96303081 | 909 | } |
0f9dd46c | 910 | |
34d52cb6 | 911 | static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
912 | u64 offset) |
913 | { | |
914 | u64 bitmap_start; | |
915 | u64 bytes_per_bitmap; | |
0f9dd46c | 916 | |
34d52cb6 LZ |
917 | bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit; |
918 | bitmap_start = offset - ctl->start; | |
96303081 JB |
919 | bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap); |
920 | bitmap_start *= bytes_per_bitmap; | |
34d52cb6 | 921 | bitmap_start += ctl->start; |
0f9dd46c | 922 | |
96303081 | 923 | return bitmap_start; |
0f9dd46c JB |
924 | } |
925 | ||
96303081 JB |
926 | static int tree_insert_offset(struct rb_root *root, u64 offset, |
927 | struct rb_node *node, int bitmap) | |
0f9dd46c JB |
928 | { |
929 | struct rb_node **p = &root->rb_node; | |
930 | struct rb_node *parent = NULL; | |
931 | struct btrfs_free_space *info; | |
932 | ||
933 | while (*p) { | |
934 | parent = *p; | |
96303081 | 935 | info = rb_entry(parent, struct btrfs_free_space, offset_index); |
0f9dd46c | 936 | |
96303081 | 937 | if (offset < info->offset) { |
0f9dd46c | 938 | p = &(*p)->rb_left; |
96303081 | 939 | } else if (offset > info->offset) { |
0f9dd46c | 940 | p = &(*p)->rb_right; |
96303081 JB |
941 | } else { |
942 | /* | |
943 | * we could have a bitmap entry and an extent entry | |
944 | * share the same offset. If this is the case, we want | |
945 | * the extent entry to always be found first if we do a | |
946 | * linear search through the tree, since we want to have | |
947 | * the quickest allocation time, and allocating from an | |
948 | * extent is faster than allocating from a bitmap. So | |
949 | * if we're inserting a bitmap and we find an entry at | |
950 | * this offset, we want to go right, or after this entry | |
951 | * logically. If we are inserting an extent and we've | |
952 | * found a bitmap, we want to go left, or before | |
953 | * logically. | |
954 | */ | |
955 | if (bitmap) { | |
207dde82 JB |
956 | if (info->bitmap) { |
957 | WARN_ON_ONCE(1); | |
958 | return -EEXIST; | |
959 | } | |
96303081 JB |
960 | p = &(*p)->rb_right; |
961 | } else { | |
207dde82 JB |
962 | if (!info->bitmap) { |
963 | WARN_ON_ONCE(1); | |
964 | return -EEXIST; | |
965 | } | |
96303081 JB |
966 | p = &(*p)->rb_left; |
967 | } | |
968 | } | |
0f9dd46c JB |
969 | } |
970 | ||
971 | rb_link_node(node, parent, p); | |
972 | rb_insert_color(node, root); | |
973 | ||
974 | return 0; | |
975 | } | |
976 | ||
977 | /* | |
70cb0743 JB |
978 | * searches the tree for the given offset. |
979 | * | |
96303081 JB |
980 | * fuzzy - If this is set, then we are trying to make an allocation, and we just |
981 | * want a section that has at least bytes size and comes at or after the given | |
982 | * offset. | |
0f9dd46c | 983 | */ |
96303081 | 984 | static struct btrfs_free_space * |
34d52cb6 | 985 | tree_search_offset(struct btrfs_free_space_ctl *ctl, |
96303081 | 986 | u64 offset, int bitmap_only, int fuzzy) |
0f9dd46c | 987 | { |
34d52cb6 | 988 | struct rb_node *n = ctl->free_space_offset.rb_node; |
96303081 JB |
989 | struct btrfs_free_space *entry, *prev = NULL; |
990 | ||
991 | /* find entry that is closest to the 'offset' */ | |
992 | while (1) { | |
993 | if (!n) { | |
994 | entry = NULL; | |
995 | break; | |
996 | } | |
0f9dd46c | 997 | |
0f9dd46c | 998 | entry = rb_entry(n, struct btrfs_free_space, offset_index); |
96303081 | 999 | prev = entry; |
0f9dd46c | 1000 | |
96303081 | 1001 | if (offset < entry->offset) |
0f9dd46c | 1002 | n = n->rb_left; |
96303081 | 1003 | else if (offset > entry->offset) |
0f9dd46c | 1004 | n = n->rb_right; |
96303081 | 1005 | else |
0f9dd46c | 1006 | break; |
0f9dd46c JB |
1007 | } |
1008 | ||
96303081 JB |
1009 | if (bitmap_only) { |
1010 | if (!entry) | |
1011 | return NULL; | |
1012 | if (entry->bitmap) | |
1013 | return entry; | |
0f9dd46c | 1014 | |
96303081 JB |
1015 | /* |
1016 | * bitmap entry and extent entry may share same offset, | |
1017 | * in that case, bitmap entry comes after extent entry. | |
1018 | */ | |
1019 | n = rb_next(n); | |
1020 | if (!n) | |
1021 | return NULL; | |
1022 | entry = rb_entry(n, struct btrfs_free_space, offset_index); | |
1023 | if (entry->offset != offset) | |
1024 | return NULL; | |
0f9dd46c | 1025 | |
96303081 JB |
1026 | WARN_ON(!entry->bitmap); |
1027 | return entry; | |
1028 | } else if (entry) { | |
1029 | if (entry->bitmap) { | |
0f9dd46c | 1030 | /* |
96303081 JB |
1031 | * if previous extent entry covers the offset, |
1032 | * we should return it instead of the bitmap entry | |
0f9dd46c | 1033 | */ |
96303081 JB |
1034 | n = &entry->offset_index; |
1035 | while (1) { | |
1036 | n = rb_prev(n); | |
1037 | if (!n) | |
1038 | break; | |
1039 | prev = rb_entry(n, struct btrfs_free_space, | |
1040 | offset_index); | |
1041 | if (!prev->bitmap) { | |
1042 | if (prev->offset + prev->bytes > offset) | |
1043 | entry = prev; | |
1044 | break; | |
1045 | } | |
0f9dd46c | 1046 | } |
96303081 JB |
1047 | } |
1048 | return entry; | |
1049 | } | |
1050 | ||
1051 | if (!prev) | |
1052 | return NULL; | |
1053 | ||
1054 | /* find last entry before the 'offset' */ | |
1055 | entry = prev; | |
1056 | if (entry->offset > offset) { | |
1057 | n = rb_prev(&entry->offset_index); | |
1058 | if (n) { | |
1059 | entry = rb_entry(n, struct btrfs_free_space, | |
1060 | offset_index); | |
1061 | BUG_ON(entry->offset > offset); | |
0f9dd46c | 1062 | } else { |
96303081 JB |
1063 | if (fuzzy) |
1064 | return entry; | |
1065 | else | |
1066 | return NULL; | |
0f9dd46c JB |
1067 | } |
1068 | } | |
1069 | ||
96303081 JB |
1070 | if (entry->bitmap) { |
1071 | n = &entry->offset_index; | |
1072 | while (1) { | |
1073 | n = rb_prev(n); | |
1074 | if (!n) | |
1075 | break; | |
1076 | prev = rb_entry(n, struct btrfs_free_space, | |
1077 | offset_index); | |
1078 | if (!prev->bitmap) { | |
1079 | if (prev->offset + prev->bytes > offset) | |
1080 | return prev; | |
1081 | break; | |
1082 | } | |
1083 | } | |
34d52cb6 | 1084 | if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset) |
96303081 JB |
1085 | return entry; |
1086 | } else if (entry->offset + entry->bytes > offset) | |
1087 | return entry; | |
1088 | ||
1089 | if (!fuzzy) | |
1090 | return NULL; | |
1091 | ||
1092 | while (1) { | |
1093 | if (entry->bitmap) { | |
1094 | if (entry->offset + BITS_PER_BITMAP * | |
34d52cb6 | 1095 | ctl->unit > offset) |
96303081 JB |
1096 | break; |
1097 | } else { | |
1098 | if (entry->offset + entry->bytes > offset) | |
1099 | break; | |
1100 | } | |
1101 | ||
1102 | n = rb_next(&entry->offset_index); | |
1103 | if (!n) | |
1104 | return NULL; | |
1105 | entry = rb_entry(n, struct btrfs_free_space, offset_index); | |
1106 | } | |
1107 | return entry; | |
0f9dd46c JB |
1108 | } |
1109 | ||
f333adb5 | 1110 | static inline void |
34d52cb6 | 1111 | __unlink_free_space(struct btrfs_free_space_ctl *ctl, |
f333adb5 | 1112 | struct btrfs_free_space *info) |
0f9dd46c | 1113 | { |
34d52cb6 LZ |
1114 | rb_erase(&info->offset_index, &ctl->free_space_offset); |
1115 | ctl->free_extents--; | |
f333adb5 LZ |
1116 | } |
1117 | ||
34d52cb6 | 1118 | static void unlink_free_space(struct btrfs_free_space_ctl *ctl, |
f333adb5 LZ |
1119 | struct btrfs_free_space *info) |
1120 | { | |
34d52cb6 LZ |
1121 | __unlink_free_space(ctl, info); |
1122 | ctl->free_space -= info->bytes; | |
0f9dd46c JB |
1123 | } |
1124 | ||
34d52cb6 | 1125 | static int link_free_space(struct btrfs_free_space_ctl *ctl, |
0f9dd46c JB |
1126 | struct btrfs_free_space *info) |
1127 | { | |
1128 | int ret = 0; | |
1129 | ||
96303081 | 1130 | BUG_ON(!info->bitmap && !info->bytes); |
34d52cb6 | 1131 | ret = tree_insert_offset(&ctl->free_space_offset, info->offset, |
96303081 | 1132 | &info->offset_index, (info->bitmap != NULL)); |
0f9dd46c JB |
1133 | if (ret) |
1134 | return ret; | |
1135 | ||
34d52cb6 LZ |
1136 | ctl->free_space += info->bytes; |
1137 | ctl->free_extents++; | |
96303081 JB |
1138 | return ret; |
1139 | } | |
1140 | ||
34d52cb6 | 1141 | static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl) |
96303081 | 1142 | { |
34d52cb6 | 1143 | struct btrfs_block_group_cache *block_group = ctl->private; |
25891f79 JB |
1144 | u64 max_bytes; |
1145 | u64 bitmap_bytes; | |
1146 | u64 extent_bytes; | |
8eb2d829 | 1147 | u64 size = block_group->key.offset; |
34d52cb6 LZ |
1148 | u64 bytes_per_bg = BITS_PER_BITMAP * block_group->sectorsize; |
1149 | int max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg); | |
1150 | ||
1151 | BUG_ON(ctl->total_bitmaps > max_bitmaps); | |
96303081 JB |
1152 | |
1153 | /* | |
1154 | * The goal is to keep the total amount of memory used per 1gb of space | |
1155 | * at or below 32k, so we need to adjust how much memory we allow to be | |
1156 | * used by extent based free space tracking | |
1157 | */ | |
8eb2d829 LZ |
1158 | if (size < 1024 * 1024 * 1024) |
1159 | max_bytes = MAX_CACHE_BYTES_PER_GIG; | |
1160 | else | |
1161 | max_bytes = MAX_CACHE_BYTES_PER_GIG * | |
1162 | div64_u64(size, 1024 * 1024 * 1024); | |
96303081 | 1163 | |
25891f79 JB |
1164 | /* |
1165 | * we want to account for 1 more bitmap than what we have so we can make | |
1166 | * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as | |
1167 | * we add more bitmaps. | |
1168 | */ | |
34d52cb6 | 1169 | bitmap_bytes = (ctl->total_bitmaps + 1) * PAGE_CACHE_SIZE; |
96303081 | 1170 | |
25891f79 | 1171 | if (bitmap_bytes >= max_bytes) { |
34d52cb6 | 1172 | ctl->extents_thresh = 0; |
25891f79 JB |
1173 | return; |
1174 | } | |
96303081 | 1175 | |
25891f79 JB |
1176 | /* |
1177 | * we want the extent entry threshold to always be at most 1/2 the maxw | |
1178 | * bytes we can have, or whatever is less than that. | |
1179 | */ | |
1180 | extent_bytes = max_bytes - bitmap_bytes; | |
1181 | extent_bytes = min_t(u64, extent_bytes, div64_u64(max_bytes, 2)); | |
96303081 | 1182 | |
34d52cb6 | 1183 | ctl->extents_thresh = |
25891f79 | 1184 | div64_u64(extent_bytes, (sizeof(struct btrfs_free_space))); |
96303081 JB |
1185 | } |
1186 | ||
bb3ac5a4 MX |
1187 | static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl, |
1188 | struct btrfs_free_space *info, | |
1189 | u64 offset, u64 bytes) | |
96303081 | 1190 | { |
f38b6e75 | 1191 | unsigned long start, count; |
96303081 | 1192 | |
34d52cb6 LZ |
1193 | start = offset_to_bit(info->offset, ctl->unit, offset); |
1194 | count = bytes_to_bits(bytes, ctl->unit); | |
f38b6e75 | 1195 | BUG_ON(start + count > BITS_PER_BITMAP); |
96303081 | 1196 | |
f38b6e75 | 1197 | bitmap_clear(info->bitmap, start, count); |
96303081 JB |
1198 | |
1199 | info->bytes -= bytes; | |
bb3ac5a4 MX |
1200 | } |
1201 | ||
1202 | static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl, | |
1203 | struct btrfs_free_space *info, u64 offset, | |
1204 | u64 bytes) | |
1205 | { | |
1206 | __bitmap_clear_bits(ctl, info, offset, bytes); | |
34d52cb6 | 1207 | ctl->free_space -= bytes; |
96303081 JB |
1208 | } |
1209 | ||
34d52cb6 | 1210 | static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl, |
817d52f8 JB |
1211 | struct btrfs_free_space *info, u64 offset, |
1212 | u64 bytes) | |
96303081 | 1213 | { |
f38b6e75 | 1214 | unsigned long start, count; |
96303081 | 1215 | |
34d52cb6 LZ |
1216 | start = offset_to_bit(info->offset, ctl->unit, offset); |
1217 | count = bytes_to_bits(bytes, ctl->unit); | |
f38b6e75 | 1218 | BUG_ON(start + count > BITS_PER_BITMAP); |
96303081 | 1219 | |
f38b6e75 | 1220 | bitmap_set(info->bitmap, start, count); |
96303081 JB |
1221 | |
1222 | info->bytes += bytes; | |
34d52cb6 | 1223 | ctl->free_space += bytes; |
96303081 JB |
1224 | } |
1225 | ||
34d52cb6 | 1226 | static int search_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1227 | struct btrfs_free_space *bitmap_info, u64 *offset, |
1228 | u64 *bytes) | |
1229 | { | |
1230 | unsigned long found_bits = 0; | |
1231 | unsigned long bits, i; | |
1232 | unsigned long next_zero; | |
1233 | ||
34d52cb6 | 1234 | i = offset_to_bit(bitmap_info->offset, ctl->unit, |
96303081 | 1235 | max_t(u64, *offset, bitmap_info->offset)); |
34d52cb6 | 1236 | bits = bytes_to_bits(*bytes, ctl->unit); |
96303081 JB |
1237 | |
1238 | for (i = find_next_bit(bitmap_info->bitmap, BITS_PER_BITMAP, i); | |
1239 | i < BITS_PER_BITMAP; | |
1240 | i = find_next_bit(bitmap_info->bitmap, BITS_PER_BITMAP, i + 1)) { | |
1241 | next_zero = find_next_zero_bit(bitmap_info->bitmap, | |
1242 | BITS_PER_BITMAP, i); | |
1243 | if ((next_zero - i) >= bits) { | |
1244 | found_bits = next_zero - i; | |
1245 | break; | |
1246 | } | |
1247 | i = next_zero; | |
1248 | } | |
1249 | ||
1250 | if (found_bits) { | |
34d52cb6 LZ |
1251 | *offset = (u64)(i * ctl->unit) + bitmap_info->offset; |
1252 | *bytes = (u64)(found_bits) * ctl->unit; | |
96303081 JB |
1253 | return 0; |
1254 | } | |
1255 | ||
1256 | return -1; | |
1257 | } | |
1258 | ||
34d52cb6 LZ |
1259 | static struct btrfs_free_space * |
1260 | find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes) | |
96303081 JB |
1261 | { |
1262 | struct btrfs_free_space *entry; | |
1263 | struct rb_node *node; | |
1264 | int ret; | |
1265 | ||
34d52cb6 | 1266 | if (!ctl->free_space_offset.rb_node) |
96303081 JB |
1267 | return NULL; |
1268 | ||
34d52cb6 | 1269 | entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1); |
96303081 JB |
1270 | if (!entry) |
1271 | return NULL; | |
1272 | ||
1273 | for (node = &entry->offset_index; node; node = rb_next(node)) { | |
1274 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
1275 | if (entry->bytes < *bytes) | |
1276 | continue; | |
1277 | ||
1278 | if (entry->bitmap) { | |
34d52cb6 | 1279 | ret = search_bitmap(ctl, entry, offset, bytes); |
96303081 JB |
1280 | if (!ret) |
1281 | return entry; | |
1282 | continue; | |
1283 | } | |
1284 | ||
1285 | *offset = entry->offset; | |
1286 | *bytes = entry->bytes; | |
1287 | return entry; | |
1288 | } | |
1289 | ||
1290 | return NULL; | |
1291 | } | |
1292 | ||
34d52cb6 | 1293 | static void add_new_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1294 | struct btrfs_free_space *info, u64 offset) |
1295 | { | |
34d52cb6 | 1296 | info->offset = offset_to_bitmap(ctl, offset); |
f019f426 | 1297 | info->bytes = 0; |
34d52cb6 LZ |
1298 | link_free_space(ctl, info); |
1299 | ctl->total_bitmaps++; | |
96303081 | 1300 | |
34d52cb6 | 1301 | ctl->op->recalc_thresholds(ctl); |
96303081 JB |
1302 | } |
1303 | ||
34d52cb6 | 1304 | static void free_bitmap(struct btrfs_free_space_ctl *ctl, |
edf6e2d1 LZ |
1305 | struct btrfs_free_space *bitmap_info) |
1306 | { | |
34d52cb6 | 1307 | unlink_free_space(ctl, bitmap_info); |
edf6e2d1 | 1308 | kfree(bitmap_info->bitmap); |
dc89e982 | 1309 | kmem_cache_free(btrfs_free_space_cachep, bitmap_info); |
34d52cb6 LZ |
1310 | ctl->total_bitmaps--; |
1311 | ctl->op->recalc_thresholds(ctl); | |
edf6e2d1 LZ |
1312 | } |
1313 | ||
34d52cb6 | 1314 | static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1315 | struct btrfs_free_space *bitmap_info, |
1316 | u64 *offset, u64 *bytes) | |
1317 | { | |
1318 | u64 end; | |
6606bb97 JB |
1319 | u64 search_start, search_bytes; |
1320 | int ret; | |
96303081 JB |
1321 | |
1322 | again: | |
34d52cb6 | 1323 | end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1; |
96303081 | 1324 | |
6606bb97 JB |
1325 | /* |
1326 | * XXX - this can go away after a few releases. | |
1327 | * | |
1328 | * since the only user of btrfs_remove_free_space is the tree logging | |
1329 | * stuff, and the only way to test that is under crash conditions, we | |
1330 | * want to have this debug stuff here just in case somethings not | |
1331 | * working. Search the bitmap for the space we are trying to use to | |
1332 | * make sure its actually there. If its not there then we need to stop | |
1333 | * because something has gone wrong. | |
1334 | */ | |
1335 | search_start = *offset; | |
1336 | search_bytes = *bytes; | |
13dbc089 | 1337 | search_bytes = min(search_bytes, end - search_start + 1); |
34d52cb6 | 1338 | ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes); |
6606bb97 JB |
1339 | BUG_ON(ret < 0 || search_start != *offset); |
1340 | ||
96303081 | 1341 | if (*offset > bitmap_info->offset && *offset + *bytes > end) { |
34d52cb6 | 1342 | bitmap_clear_bits(ctl, bitmap_info, *offset, end - *offset + 1); |
96303081 JB |
1343 | *bytes -= end - *offset + 1; |
1344 | *offset = end + 1; | |
1345 | } else if (*offset >= bitmap_info->offset && *offset + *bytes <= end) { | |
34d52cb6 | 1346 | bitmap_clear_bits(ctl, bitmap_info, *offset, *bytes); |
96303081 JB |
1347 | *bytes = 0; |
1348 | } | |
1349 | ||
1350 | if (*bytes) { | |
6606bb97 | 1351 | struct rb_node *next = rb_next(&bitmap_info->offset_index); |
edf6e2d1 | 1352 | if (!bitmap_info->bytes) |
34d52cb6 | 1353 | free_bitmap(ctl, bitmap_info); |
96303081 | 1354 | |
6606bb97 JB |
1355 | /* |
1356 | * no entry after this bitmap, but we still have bytes to | |
1357 | * remove, so something has gone wrong. | |
1358 | */ | |
1359 | if (!next) | |
96303081 JB |
1360 | return -EINVAL; |
1361 | ||
6606bb97 JB |
1362 | bitmap_info = rb_entry(next, struct btrfs_free_space, |
1363 | offset_index); | |
1364 | ||
1365 | /* | |
1366 | * if the next entry isn't a bitmap we need to return to let the | |
1367 | * extent stuff do its work. | |
1368 | */ | |
96303081 JB |
1369 | if (!bitmap_info->bitmap) |
1370 | return -EAGAIN; | |
1371 | ||
6606bb97 JB |
1372 | /* |
1373 | * Ok the next item is a bitmap, but it may not actually hold | |
1374 | * the information for the rest of this free space stuff, so | |
1375 | * look for it, and if we don't find it return so we can try | |
1376 | * everything over again. | |
1377 | */ | |
1378 | search_start = *offset; | |
1379 | search_bytes = *bytes; | |
34d52cb6 | 1380 | ret = search_bitmap(ctl, bitmap_info, &search_start, |
6606bb97 JB |
1381 | &search_bytes); |
1382 | if (ret < 0 || search_start != *offset) | |
1383 | return -EAGAIN; | |
1384 | ||
96303081 | 1385 | goto again; |
edf6e2d1 | 1386 | } else if (!bitmap_info->bytes) |
34d52cb6 | 1387 | free_bitmap(ctl, bitmap_info); |
96303081 JB |
1388 | |
1389 | return 0; | |
1390 | } | |
1391 | ||
2cdc342c JB |
1392 | static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl, |
1393 | struct btrfs_free_space *info, u64 offset, | |
1394 | u64 bytes) | |
1395 | { | |
1396 | u64 bytes_to_set = 0; | |
1397 | u64 end; | |
1398 | ||
1399 | end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit); | |
1400 | ||
1401 | bytes_to_set = min(end - offset, bytes); | |
1402 | ||
1403 | bitmap_set_bits(ctl, info, offset, bytes_to_set); | |
1404 | ||
1405 | return bytes_to_set; | |
1406 | ||
1407 | } | |
1408 | ||
34d52cb6 LZ |
1409 | static bool use_bitmap(struct btrfs_free_space_ctl *ctl, |
1410 | struct btrfs_free_space *info) | |
96303081 | 1411 | { |
34d52cb6 | 1412 | struct btrfs_block_group_cache *block_group = ctl->private; |
96303081 JB |
1413 | |
1414 | /* | |
1415 | * If we are below the extents threshold then we can add this as an | |
1416 | * extent, and don't have to deal with the bitmap | |
1417 | */ | |
34d52cb6 | 1418 | if (ctl->free_extents < ctl->extents_thresh) { |
32cb0840 JB |
1419 | /* |
1420 | * If this block group has some small extents we don't want to | |
1421 | * use up all of our free slots in the cache with them, we want | |
1422 | * to reserve them to larger extents, however if we have plent | |
1423 | * of cache left then go ahead an dadd them, no sense in adding | |
1424 | * the overhead of a bitmap if we don't have to. | |
1425 | */ | |
1426 | if (info->bytes <= block_group->sectorsize * 4) { | |
34d52cb6 LZ |
1427 | if (ctl->free_extents * 2 <= ctl->extents_thresh) |
1428 | return false; | |
32cb0840 | 1429 | } else { |
34d52cb6 | 1430 | return false; |
32cb0840 JB |
1431 | } |
1432 | } | |
96303081 JB |
1433 | |
1434 | /* | |
1435 | * some block groups are so tiny they can't be enveloped by a bitmap, so | |
1436 | * don't even bother to create a bitmap for this | |
1437 | */ | |
1438 | if (BITS_PER_BITMAP * block_group->sectorsize > | |
1439 | block_group->key.offset) | |
34d52cb6 LZ |
1440 | return false; |
1441 | ||
1442 | return true; | |
1443 | } | |
1444 | ||
2cdc342c JB |
1445 | static struct btrfs_free_space_op free_space_op = { |
1446 | .recalc_thresholds = recalculate_thresholds, | |
1447 | .use_bitmap = use_bitmap, | |
1448 | }; | |
1449 | ||
34d52cb6 LZ |
1450 | static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl, |
1451 | struct btrfs_free_space *info) | |
1452 | { | |
1453 | struct btrfs_free_space *bitmap_info; | |
2cdc342c | 1454 | struct btrfs_block_group_cache *block_group = NULL; |
34d52cb6 | 1455 | int added = 0; |
2cdc342c | 1456 | u64 bytes, offset, bytes_added; |
34d52cb6 | 1457 | int ret; |
96303081 JB |
1458 | |
1459 | bytes = info->bytes; | |
1460 | offset = info->offset; | |
1461 | ||
34d52cb6 LZ |
1462 | if (!ctl->op->use_bitmap(ctl, info)) |
1463 | return 0; | |
1464 | ||
2cdc342c JB |
1465 | if (ctl->op == &free_space_op) |
1466 | block_group = ctl->private; | |
38e87880 | 1467 | again: |
2cdc342c JB |
1468 | /* |
1469 | * Since we link bitmaps right into the cluster we need to see if we | |
1470 | * have a cluster here, and if so and it has our bitmap we need to add | |
1471 | * the free space to that bitmap. | |
1472 | */ | |
1473 | if (block_group && !list_empty(&block_group->cluster_list)) { | |
1474 | struct btrfs_free_cluster *cluster; | |
1475 | struct rb_node *node; | |
1476 | struct btrfs_free_space *entry; | |
1477 | ||
1478 | cluster = list_entry(block_group->cluster_list.next, | |
1479 | struct btrfs_free_cluster, | |
1480 | block_group_list); | |
1481 | spin_lock(&cluster->lock); | |
1482 | node = rb_first(&cluster->root); | |
1483 | if (!node) { | |
1484 | spin_unlock(&cluster->lock); | |
38e87880 | 1485 | goto no_cluster_bitmap; |
2cdc342c JB |
1486 | } |
1487 | ||
1488 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
1489 | if (!entry->bitmap) { | |
1490 | spin_unlock(&cluster->lock); | |
38e87880 | 1491 | goto no_cluster_bitmap; |
2cdc342c JB |
1492 | } |
1493 | ||
1494 | if (entry->offset == offset_to_bitmap(ctl, offset)) { | |
1495 | bytes_added = add_bytes_to_bitmap(ctl, entry, | |
1496 | offset, bytes); | |
1497 | bytes -= bytes_added; | |
1498 | offset += bytes_added; | |
1499 | } | |
1500 | spin_unlock(&cluster->lock); | |
1501 | if (!bytes) { | |
1502 | ret = 1; | |
1503 | goto out; | |
1504 | } | |
1505 | } | |
38e87880 CM |
1506 | |
1507 | no_cluster_bitmap: | |
34d52cb6 | 1508 | bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), |
96303081 JB |
1509 | 1, 0); |
1510 | if (!bitmap_info) { | |
1511 | BUG_ON(added); | |
1512 | goto new_bitmap; | |
1513 | } | |
1514 | ||
2cdc342c JB |
1515 | bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes); |
1516 | bytes -= bytes_added; | |
1517 | offset += bytes_added; | |
1518 | added = 0; | |
96303081 JB |
1519 | |
1520 | if (!bytes) { | |
1521 | ret = 1; | |
1522 | goto out; | |
1523 | } else | |
1524 | goto again; | |
1525 | ||
1526 | new_bitmap: | |
1527 | if (info && info->bitmap) { | |
34d52cb6 | 1528 | add_new_bitmap(ctl, info, offset); |
96303081 JB |
1529 | added = 1; |
1530 | info = NULL; | |
1531 | goto again; | |
1532 | } else { | |
34d52cb6 | 1533 | spin_unlock(&ctl->tree_lock); |
96303081 JB |
1534 | |
1535 | /* no pre-allocated info, allocate a new one */ | |
1536 | if (!info) { | |
dc89e982 JB |
1537 | info = kmem_cache_zalloc(btrfs_free_space_cachep, |
1538 | GFP_NOFS); | |
96303081 | 1539 | if (!info) { |
34d52cb6 | 1540 | spin_lock(&ctl->tree_lock); |
96303081 JB |
1541 | ret = -ENOMEM; |
1542 | goto out; | |
1543 | } | |
1544 | } | |
1545 | ||
1546 | /* allocate the bitmap */ | |
1547 | info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS); | |
34d52cb6 | 1548 | spin_lock(&ctl->tree_lock); |
96303081 JB |
1549 | if (!info->bitmap) { |
1550 | ret = -ENOMEM; | |
1551 | goto out; | |
1552 | } | |
1553 | goto again; | |
1554 | } | |
1555 | ||
1556 | out: | |
1557 | if (info) { | |
1558 | if (info->bitmap) | |
1559 | kfree(info->bitmap); | |
dc89e982 | 1560 | kmem_cache_free(btrfs_free_space_cachep, info); |
96303081 | 1561 | } |
0f9dd46c JB |
1562 | |
1563 | return ret; | |
1564 | } | |
1565 | ||
945d8962 | 1566 | static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl, |
f333adb5 | 1567 | struct btrfs_free_space *info, bool update_stat) |
0f9dd46c | 1568 | { |
120d66ee LZ |
1569 | struct btrfs_free_space *left_info; |
1570 | struct btrfs_free_space *right_info; | |
1571 | bool merged = false; | |
1572 | u64 offset = info->offset; | |
1573 | u64 bytes = info->bytes; | |
6226cb0a | 1574 | |
0f9dd46c JB |
1575 | /* |
1576 | * first we want to see if there is free space adjacent to the range we | |
1577 | * are adding, if there is remove that struct and add a new one to | |
1578 | * cover the entire range | |
1579 | */ | |
34d52cb6 | 1580 | right_info = tree_search_offset(ctl, offset + bytes, 0, 0); |
96303081 JB |
1581 | if (right_info && rb_prev(&right_info->offset_index)) |
1582 | left_info = rb_entry(rb_prev(&right_info->offset_index), | |
1583 | struct btrfs_free_space, offset_index); | |
1584 | else | |
34d52cb6 | 1585 | left_info = tree_search_offset(ctl, offset - 1, 0, 0); |
0f9dd46c | 1586 | |
96303081 | 1587 | if (right_info && !right_info->bitmap) { |
f333adb5 | 1588 | if (update_stat) |
34d52cb6 | 1589 | unlink_free_space(ctl, right_info); |
f333adb5 | 1590 | else |
34d52cb6 | 1591 | __unlink_free_space(ctl, right_info); |
6226cb0a | 1592 | info->bytes += right_info->bytes; |
dc89e982 | 1593 | kmem_cache_free(btrfs_free_space_cachep, right_info); |
120d66ee | 1594 | merged = true; |
0f9dd46c JB |
1595 | } |
1596 | ||
96303081 JB |
1597 | if (left_info && !left_info->bitmap && |
1598 | left_info->offset + left_info->bytes == offset) { | |
f333adb5 | 1599 | if (update_stat) |
34d52cb6 | 1600 | unlink_free_space(ctl, left_info); |
f333adb5 | 1601 | else |
34d52cb6 | 1602 | __unlink_free_space(ctl, left_info); |
6226cb0a JB |
1603 | info->offset = left_info->offset; |
1604 | info->bytes += left_info->bytes; | |
dc89e982 | 1605 | kmem_cache_free(btrfs_free_space_cachep, left_info); |
120d66ee | 1606 | merged = true; |
0f9dd46c JB |
1607 | } |
1608 | ||
120d66ee LZ |
1609 | return merged; |
1610 | } | |
1611 | ||
581bb050 LZ |
1612 | int __btrfs_add_free_space(struct btrfs_free_space_ctl *ctl, |
1613 | u64 offset, u64 bytes) | |
120d66ee LZ |
1614 | { |
1615 | struct btrfs_free_space *info; | |
1616 | int ret = 0; | |
1617 | ||
dc89e982 | 1618 | info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS); |
120d66ee LZ |
1619 | if (!info) |
1620 | return -ENOMEM; | |
1621 | ||
1622 | info->offset = offset; | |
1623 | info->bytes = bytes; | |
1624 | ||
34d52cb6 | 1625 | spin_lock(&ctl->tree_lock); |
120d66ee | 1626 | |
34d52cb6 | 1627 | if (try_merge_free_space(ctl, info, true)) |
120d66ee LZ |
1628 | goto link; |
1629 | ||
1630 | /* | |
1631 | * There was no extent directly to the left or right of this new | |
1632 | * extent then we know we're going to have to allocate a new extent, so | |
1633 | * before we do that see if we need to drop this into a bitmap | |
1634 | */ | |
34d52cb6 | 1635 | ret = insert_into_bitmap(ctl, info); |
120d66ee LZ |
1636 | if (ret < 0) { |
1637 | goto out; | |
1638 | } else if (ret) { | |
1639 | ret = 0; | |
1640 | goto out; | |
1641 | } | |
1642 | link: | |
34d52cb6 | 1643 | ret = link_free_space(ctl, info); |
0f9dd46c | 1644 | if (ret) |
dc89e982 | 1645 | kmem_cache_free(btrfs_free_space_cachep, info); |
96303081 | 1646 | out: |
34d52cb6 | 1647 | spin_unlock(&ctl->tree_lock); |
6226cb0a | 1648 | |
0f9dd46c | 1649 | if (ret) { |
96303081 | 1650 | printk(KERN_CRIT "btrfs: unable to add free space :%d\n", ret); |
c293498b | 1651 | BUG_ON(ret == -EEXIST); |
0f9dd46c JB |
1652 | } |
1653 | ||
0f9dd46c JB |
1654 | return ret; |
1655 | } | |
1656 | ||
6226cb0a JB |
1657 | int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group, |
1658 | u64 offset, u64 bytes) | |
0f9dd46c | 1659 | { |
34d52cb6 | 1660 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0f9dd46c | 1661 | struct btrfs_free_space *info; |
96303081 | 1662 | struct btrfs_free_space *next_info = NULL; |
0f9dd46c JB |
1663 | int ret = 0; |
1664 | ||
34d52cb6 | 1665 | spin_lock(&ctl->tree_lock); |
6226cb0a | 1666 | |
96303081 | 1667 | again: |
34d52cb6 | 1668 | info = tree_search_offset(ctl, offset, 0, 0); |
96303081 | 1669 | if (!info) { |
6606bb97 JB |
1670 | /* |
1671 | * oops didn't find an extent that matched the space we wanted | |
1672 | * to remove, look for a bitmap instead | |
1673 | */ | |
34d52cb6 | 1674 | info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), |
6606bb97 JB |
1675 | 1, 0); |
1676 | if (!info) { | |
1677 | WARN_ON(1); | |
1678 | goto out_lock; | |
1679 | } | |
96303081 JB |
1680 | } |
1681 | ||
1682 | if (info->bytes < bytes && rb_next(&info->offset_index)) { | |
1683 | u64 end; | |
1684 | next_info = rb_entry(rb_next(&info->offset_index), | |
1685 | struct btrfs_free_space, | |
1686 | offset_index); | |
1687 | ||
1688 | if (next_info->bitmap) | |
34d52cb6 LZ |
1689 | end = next_info->offset + |
1690 | BITS_PER_BITMAP * ctl->unit - 1; | |
96303081 JB |
1691 | else |
1692 | end = next_info->offset + next_info->bytes; | |
1693 | ||
1694 | if (next_info->bytes < bytes || | |
1695 | next_info->offset > offset || offset > end) { | |
1696 | printk(KERN_CRIT "Found free space at %llu, size %llu," | |
1697 | " trying to use %llu\n", | |
1698 | (unsigned long long)info->offset, | |
1699 | (unsigned long long)info->bytes, | |
1700 | (unsigned long long)bytes); | |
0f9dd46c JB |
1701 | WARN_ON(1); |
1702 | ret = -EINVAL; | |
96303081 | 1703 | goto out_lock; |
0f9dd46c | 1704 | } |
0f9dd46c | 1705 | |
96303081 JB |
1706 | info = next_info; |
1707 | } | |
1708 | ||
1709 | if (info->bytes == bytes) { | |
34d52cb6 | 1710 | unlink_free_space(ctl, info); |
96303081 JB |
1711 | if (info->bitmap) { |
1712 | kfree(info->bitmap); | |
34d52cb6 | 1713 | ctl->total_bitmaps--; |
0f9dd46c | 1714 | } |
dc89e982 | 1715 | kmem_cache_free(btrfs_free_space_cachep, info); |
96303081 JB |
1716 | goto out_lock; |
1717 | } | |
0f9dd46c | 1718 | |
96303081 | 1719 | if (!info->bitmap && info->offset == offset) { |
34d52cb6 | 1720 | unlink_free_space(ctl, info); |
0f9dd46c JB |
1721 | info->offset += bytes; |
1722 | info->bytes -= bytes; | |
34d52cb6 | 1723 | link_free_space(ctl, info); |
96303081 JB |
1724 | goto out_lock; |
1725 | } | |
0f9dd46c | 1726 | |
96303081 JB |
1727 | if (!info->bitmap && info->offset <= offset && |
1728 | info->offset + info->bytes >= offset + bytes) { | |
9b49c9b9 CM |
1729 | u64 old_start = info->offset; |
1730 | /* | |
1731 | * we're freeing space in the middle of the info, | |
1732 | * this can happen during tree log replay | |
1733 | * | |
1734 | * first unlink the old info and then | |
1735 | * insert it again after the hole we're creating | |
1736 | */ | |
34d52cb6 | 1737 | unlink_free_space(ctl, info); |
9b49c9b9 CM |
1738 | if (offset + bytes < info->offset + info->bytes) { |
1739 | u64 old_end = info->offset + info->bytes; | |
1740 | ||
1741 | info->offset = offset + bytes; | |
1742 | info->bytes = old_end - info->offset; | |
34d52cb6 | 1743 | ret = link_free_space(ctl, info); |
96303081 JB |
1744 | WARN_ON(ret); |
1745 | if (ret) | |
1746 | goto out_lock; | |
9b49c9b9 CM |
1747 | } else { |
1748 | /* the hole we're creating ends at the end | |
1749 | * of the info struct, just free the info | |
1750 | */ | |
dc89e982 | 1751 | kmem_cache_free(btrfs_free_space_cachep, info); |
9b49c9b9 | 1752 | } |
34d52cb6 | 1753 | spin_unlock(&ctl->tree_lock); |
96303081 JB |
1754 | |
1755 | /* step two, insert a new info struct to cover | |
1756 | * anything before the hole | |
9b49c9b9 | 1757 | */ |
6226cb0a JB |
1758 | ret = btrfs_add_free_space(block_group, old_start, |
1759 | offset - old_start); | |
96303081 JB |
1760 | WARN_ON(ret); |
1761 | goto out; | |
0f9dd46c | 1762 | } |
96303081 | 1763 | |
34d52cb6 | 1764 | ret = remove_from_bitmap(ctl, info, &offset, &bytes); |
96303081 JB |
1765 | if (ret == -EAGAIN) |
1766 | goto again; | |
1767 | BUG_ON(ret); | |
1768 | out_lock: | |
34d52cb6 | 1769 | spin_unlock(&ctl->tree_lock); |
0f9dd46c | 1770 | out: |
25179201 JB |
1771 | return ret; |
1772 | } | |
1773 | ||
0f9dd46c JB |
1774 | void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group, |
1775 | u64 bytes) | |
1776 | { | |
34d52cb6 | 1777 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0f9dd46c JB |
1778 | struct btrfs_free_space *info; |
1779 | struct rb_node *n; | |
1780 | int count = 0; | |
1781 | ||
34d52cb6 | 1782 | for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) { |
0f9dd46c JB |
1783 | info = rb_entry(n, struct btrfs_free_space, offset_index); |
1784 | if (info->bytes >= bytes) | |
1785 | count++; | |
96303081 | 1786 | printk(KERN_CRIT "entry offset %llu, bytes %llu, bitmap %s\n", |
21380931 | 1787 | (unsigned long long)info->offset, |
96303081 JB |
1788 | (unsigned long long)info->bytes, |
1789 | (info->bitmap) ? "yes" : "no"); | |
0f9dd46c | 1790 | } |
96303081 JB |
1791 | printk(KERN_INFO "block group has cluster?: %s\n", |
1792 | list_empty(&block_group->cluster_list) ? "no" : "yes"); | |
0f9dd46c JB |
1793 | printk(KERN_INFO "%d blocks of free space at or bigger than bytes is" |
1794 | "\n", count); | |
1795 | } | |
1796 | ||
34d52cb6 | 1797 | void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group) |
0f9dd46c | 1798 | { |
34d52cb6 | 1799 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0f9dd46c | 1800 | |
34d52cb6 LZ |
1801 | spin_lock_init(&ctl->tree_lock); |
1802 | ctl->unit = block_group->sectorsize; | |
1803 | ctl->start = block_group->key.objectid; | |
1804 | ctl->private = block_group; | |
1805 | ctl->op = &free_space_op; | |
0f9dd46c | 1806 | |
34d52cb6 LZ |
1807 | /* |
1808 | * we only want to have 32k of ram per block group for keeping | |
1809 | * track of free space, and if we pass 1/2 of that we want to | |
1810 | * start converting things over to using bitmaps | |
1811 | */ | |
1812 | ctl->extents_thresh = ((1024 * 32) / 2) / | |
1813 | sizeof(struct btrfs_free_space); | |
0f9dd46c JB |
1814 | } |
1815 | ||
fa9c0d79 CM |
1816 | /* |
1817 | * for a given cluster, put all of its extents back into the free | |
1818 | * space cache. If the block group passed doesn't match the block group | |
1819 | * pointed to by the cluster, someone else raced in and freed the | |
1820 | * cluster already. In that case, we just return without changing anything | |
1821 | */ | |
1822 | static int | |
1823 | __btrfs_return_cluster_to_free_space( | |
1824 | struct btrfs_block_group_cache *block_group, | |
1825 | struct btrfs_free_cluster *cluster) | |
1826 | { | |
34d52cb6 | 1827 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
fa9c0d79 CM |
1828 | struct btrfs_free_space *entry; |
1829 | struct rb_node *node; | |
1830 | ||
1831 | spin_lock(&cluster->lock); | |
1832 | if (cluster->block_group != block_group) | |
1833 | goto out; | |
1834 | ||
96303081 | 1835 | cluster->block_group = NULL; |
fa9c0d79 | 1836 | cluster->window_start = 0; |
96303081 | 1837 | list_del_init(&cluster->block_group_list); |
96303081 | 1838 | |
fa9c0d79 | 1839 | node = rb_first(&cluster->root); |
96303081 | 1840 | while (node) { |
4e69b598 JB |
1841 | bool bitmap; |
1842 | ||
fa9c0d79 CM |
1843 | entry = rb_entry(node, struct btrfs_free_space, offset_index); |
1844 | node = rb_next(&entry->offset_index); | |
1845 | rb_erase(&entry->offset_index, &cluster->root); | |
4e69b598 JB |
1846 | |
1847 | bitmap = (entry->bitmap != NULL); | |
1848 | if (!bitmap) | |
34d52cb6 LZ |
1849 | try_merge_free_space(ctl, entry, false); |
1850 | tree_insert_offset(&ctl->free_space_offset, | |
4e69b598 | 1851 | entry->offset, &entry->offset_index, bitmap); |
fa9c0d79 | 1852 | } |
6bef4d31 | 1853 | cluster->root = RB_ROOT; |
96303081 | 1854 | |
fa9c0d79 CM |
1855 | out: |
1856 | spin_unlock(&cluster->lock); | |
96303081 | 1857 | btrfs_put_block_group(block_group); |
fa9c0d79 CM |
1858 | return 0; |
1859 | } | |
1860 | ||
09655373 | 1861 | void __btrfs_remove_free_space_cache_locked(struct btrfs_free_space_ctl *ctl) |
0f9dd46c JB |
1862 | { |
1863 | struct btrfs_free_space *info; | |
1864 | struct rb_node *node; | |
581bb050 | 1865 | |
581bb050 LZ |
1866 | while ((node = rb_last(&ctl->free_space_offset)) != NULL) { |
1867 | info = rb_entry(node, struct btrfs_free_space, offset_index); | |
9b90f513 JB |
1868 | if (!info->bitmap) { |
1869 | unlink_free_space(ctl, info); | |
1870 | kmem_cache_free(btrfs_free_space_cachep, info); | |
1871 | } else { | |
1872 | free_bitmap(ctl, info); | |
1873 | } | |
581bb050 LZ |
1874 | if (need_resched()) { |
1875 | spin_unlock(&ctl->tree_lock); | |
1876 | cond_resched(); | |
1877 | spin_lock(&ctl->tree_lock); | |
1878 | } | |
1879 | } | |
09655373 CM |
1880 | } |
1881 | ||
1882 | void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl) | |
1883 | { | |
1884 | spin_lock(&ctl->tree_lock); | |
1885 | __btrfs_remove_free_space_cache_locked(ctl); | |
581bb050 LZ |
1886 | spin_unlock(&ctl->tree_lock); |
1887 | } | |
1888 | ||
1889 | void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group) | |
1890 | { | |
1891 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
fa9c0d79 | 1892 | struct btrfs_free_cluster *cluster; |
96303081 | 1893 | struct list_head *head; |
0f9dd46c | 1894 | |
34d52cb6 | 1895 | spin_lock(&ctl->tree_lock); |
96303081 JB |
1896 | while ((head = block_group->cluster_list.next) != |
1897 | &block_group->cluster_list) { | |
1898 | cluster = list_entry(head, struct btrfs_free_cluster, | |
1899 | block_group_list); | |
fa9c0d79 CM |
1900 | |
1901 | WARN_ON(cluster->block_group != block_group); | |
1902 | __btrfs_return_cluster_to_free_space(block_group, cluster); | |
96303081 | 1903 | if (need_resched()) { |
34d52cb6 | 1904 | spin_unlock(&ctl->tree_lock); |
96303081 | 1905 | cond_resched(); |
34d52cb6 | 1906 | spin_lock(&ctl->tree_lock); |
96303081 | 1907 | } |
fa9c0d79 | 1908 | } |
09655373 | 1909 | __btrfs_remove_free_space_cache_locked(ctl); |
34d52cb6 | 1910 | spin_unlock(&ctl->tree_lock); |
fa9c0d79 | 1911 | |
0f9dd46c JB |
1912 | } |
1913 | ||
6226cb0a JB |
1914 | u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group, |
1915 | u64 offset, u64 bytes, u64 empty_size) | |
0f9dd46c | 1916 | { |
34d52cb6 | 1917 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
6226cb0a | 1918 | struct btrfs_free_space *entry = NULL; |
96303081 | 1919 | u64 bytes_search = bytes + empty_size; |
6226cb0a | 1920 | u64 ret = 0; |
0f9dd46c | 1921 | |
34d52cb6 LZ |
1922 | spin_lock(&ctl->tree_lock); |
1923 | entry = find_free_space(ctl, &offset, &bytes_search); | |
6226cb0a | 1924 | if (!entry) |
96303081 JB |
1925 | goto out; |
1926 | ||
1927 | ret = offset; | |
1928 | if (entry->bitmap) { | |
34d52cb6 | 1929 | bitmap_clear_bits(ctl, entry, offset, bytes); |
edf6e2d1 | 1930 | if (!entry->bytes) |
34d52cb6 | 1931 | free_bitmap(ctl, entry); |
96303081 | 1932 | } else { |
34d52cb6 | 1933 | unlink_free_space(ctl, entry); |
6226cb0a JB |
1934 | entry->offset += bytes; |
1935 | entry->bytes -= bytes; | |
6226cb0a | 1936 | if (!entry->bytes) |
dc89e982 | 1937 | kmem_cache_free(btrfs_free_space_cachep, entry); |
6226cb0a | 1938 | else |
34d52cb6 | 1939 | link_free_space(ctl, entry); |
6226cb0a | 1940 | } |
0f9dd46c | 1941 | |
96303081 | 1942 | out: |
34d52cb6 | 1943 | spin_unlock(&ctl->tree_lock); |
817d52f8 | 1944 | |
0f9dd46c JB |
1945 | return ret; |
1946 | } | |
fa9c0d79 CM |
1947 | |
1948 | /* | |
1949 | * given a cluster, put all of its extents back into the free space | |
1950 | * cache. If a block group is passed, this function will only free | |
1951 | * a cluster that belongs to the passed block group. | |
1952 | * | |
1953 | * Otherwise, it'll get a reference on the block group pointed to by the | |
1954 | * cluster and remove the cluster from it. | |
1955 | */ | |
1956 | int btrfs_return_cluster_to_free_space( | |
1957 | struct btrfs_block_group_cache *block_group, | |
1958 | struct btrfs_free_cluster *cluster) | |
1959 | { | |
34d52cb6 | 1960 | struct btrfs_free_space_ctl *ctl; |
fa9c0d79 CM |
1961 | int ret; |
1962 | ||
1963 | /* first, get a safe pointer to the block group */ | |
1964 | spin_lock(&cluster->lock); | |
1965 | if (!block_group) { | |
1966 | block_group = cluster->block_group; | |
1967 | if (!block_group) { | |
1968 | spin_unlock(&cluster->lock); | |
1969 | return 0; | |
1970 | } | |
1971 | } else if (cluster->block_group != block_group) { | |
1972 | /* someone else has already freed it don't redo their work */ | |
1973 | spin_unlock(&cluster->lock); | |
1974 | return 0; | |
1975 | } | |
1976 | atomic_inc(&block_group->count); | |
1977 | spin_unlock(&cluster->lock); | |
1978 | ||
34d52cb6 LZ |
1979 | ctl = block_group->free_space_ctl; |
1980 | ||
fa9c0d79 | 1981 | /* now return any extents the cluster had on it */ |
34d52cb6 | 1982 | spin_lock(&ctl->tree_lock); |
fa9c0d79 | 1983 | ret = __btrfs_return_cluster_to_free_space(block_group, cluster); |
34d52cb6 | 1984 | spin_unlock(&ctl->tree_lock); |
fa9c0d79 CM |
1985 | |
1986 | /* finally drop our ref */ | |
1987 | btrfs_put_block_group(block_group); | |
1988 | return ret; | |
1989 | } | |
1990 | ||
96303081 JB |
1991 | static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group, |
1992 | struct btrfs_free_cluster *cluster, | |
4e69b598 | 1993 | struct btrfs_free_space *entry, |
96303081 JB |
1994 | u64 bytes, u64 min_start) |
1995 | { | |
34d52cb6 | 1996 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
96303081 JB |
1997 | int err; |
1998 | u64 search_start = cluster->window_start; | |
1999 | u64 search_bytes = bytes; | |
2000 | u64 ret = 0; | |
2001 | ||
96303081 JB |
2002 | search_start = min_start; |
2003 | search_bytes = bytes; | |
2004 | ||
34d52cb6 | 2005 | err = search_bitmap(ctl, entry, &search_start, &search_bytes); |
96303081 | 2006 | if (err) |
4e69b598 | 2007 | return 0; |
96303081 JB |
2008 | |
2009 | ret = search_start; | |
bb3ac5a4 | 2010 | __bitmap_clear_bits(ctl, entry, ret, bytes); |
96303081 JB |
2011 | |
2012 | return ret; | |
2013 | } | |
2014 | ||
fa9c0d79 CM |
2015 | /* |
2016 | * given a cluster, try to allocate 'bytes' from it, returns 0 | |
2017 | * if it couldn't find anything suitably large, or a logical disk offset | |
2018 | * if things worked out | |
2019 | */ | |
2020 | u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group, | |
2021 | struct btrfs_free_cluster *cluster, u64 bytes, | |
2022 | u64 min_start) | |
2023 | { | |
34d52cb6 | 2024 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
fa9c0d79 CM |
2025 | struct btrfs_free_space *entry = NULL; |
2026 | struct rb_node *node; | |
2027 | u64 ret = 0; | |
2028 | ||
2029 | spin_lock(&cluster->lock); | |
2030 | if (bytes > cluster->max_size) | |
2031 | goto out; | |
2032 | ||
2033 | if (cluster->block_group != block_group) | |
2034 | goto out; | |
2035 | ||
2036 | node = rb_first(&cluster->root); | |
2037 | if (!node) | |
2038 | goto out; | |
2039 | ||
2040 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
fa9c0d79 | 2041 | while(1) { |
4e69b598 JB |
2042 | if (entry->bytes < bytes || |
2043 | (!entry->bitmap && entry->offset < min_start)) { | |
fa9c0d79 CM |
2044 | node = rb_next(&entry->offset_index); |
2045 | if (!node) | |
2046 | break; | |
2047 | entry = rb_entry(node, struct btrfs_free_space, | |
2048 | offset_index); | |
2049 | continue; | |
2050 | } | |
fa9c0d79 | 2051 | |
4e69b598 JB |
2052 | if (entry->bitmap) { |
2053 | ret = btrfs_alloc_from_bitmap(block_group, | |
2054 | cluster, entry, bytes, | |
2055 | min_start); | |
2056 | if (ret == 0) { | |
4e69b598 JB |
2057 | node = rb_next(&entry->offset_index); |
2058 | if (!node) | |
2059 | break; | |
2060 | entry = rb_entry(node, struct btrfs_free_space, | |
2061 | offset_index); | |
2062 | continue; | |
2063 | } | |
2064 | } else { | |
4e69b598 JB |
2065 | ret = entry->offset; |
2066 | ||
2067 | entry->offset += bytes; | |
2068 | entry->bytes -= bytes; | |
2069 | } | |
fa9c0d79 | 2070 | |
5e71b5d5 | 2071 | if (entry->bytes == 0) |
fa9c0d79 | 2072 | rb_erase(&entry->offset_index, &cluster->root); |
fa9c0d79 CM |
2073 | break; |
2074 | } | |
2075 | out: | |
2076 | spin_unlock(&cluster->lock); | |
96303081 | 2077 | |
5e71b5d5 LZ |
2078 | if (!ret) |
2079 | return 0; | |
2080 | ||
34d52cb6 | 2081 | spin_lock(&ctl->tree_lock); |
5e71b5d5 | 2082 | |
34d52cb6 | 2083 | ctl->free_space -= bytes; |
5e71b5d5 | 2084 | if (entry->bytes == 0) { |
34d52cb6 | 2085 | ctl->free_extents--; |
4e69b598 JB |
2086 | if (entry->bitmap) { |
2087 | kfree(entry->bitmap); | |
34d52cb6 LZ |
2088 | ctl->total_bitmaps--; |
2089 | ctl->op->recalc_thresholds(ctl); | |
4e69b598 | 2090 | } |
dc89e982 | 2091 | kmem_cache_free(btrfs_free_space_cachep, entry); |
5e71b5d5 LZ |
2092 | } |
2093 | ||
34d52cb6 | 2094 | spin_unlock(&ctl->tree_lock); |
5e71b5d5 | 2095 | |
fa9c0d79 CM |
2096 | return ret; |
2097 | } | |
2098 | ||
96303081 JB |
2099 | static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group, |
2100 | struct btrfs_free_space *entry, | |
2101 | struct btrfs_free_cluster *cluster, | |
2102 | u64 offset, u64 bytes, u64 min_bytes) | |
2103 | { | |
34d52cb6 | 2104 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
96303081 JB |
2105 | unsigned long next_zero; |
2106 | unsigned long i; | |
2107 | unsigned long search_bits; | |
2108 | unsigned long total_bits; | |
2109 | unsigned long found_bits; | |
2110 | unsigned long start = 0; | |
2111 | unsigned long total_found = 0; | |
4e69b598 | 2112 | int ret; |
96303081 JB |
2113 | bool found = false; |
2114 | ||
2115 | i = offset_to_bit(entry->offset, block_group->sectorsize, | |
2116 | max_t(u64, offset, entry->offset)); | |
d0a365e8 JB |
2117 | search_bits = bytes_to_bits(bytes, block_group->sectorsize); |
2118 | total_bits = bytes_to_bits(min_bytes, block_group->sectorsize); | |
96303081 JB |
2119 | |
2120 | again: | |
2121 | found_bits = 0; | |
2122 | for (i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, i); | |
2123 | i < BITS_PER_BITMAP; | |
2124 | i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, i + 1)) { | |
2125 | next_zero = find_next_zero_bit(entry->bitmap, | |
2126 | BITS_PER_BITMAP, i); | |
2127 | if (next_zero - i >= search_bits) { | |
2128 | found_bits = next_zero - i; | |
2129 | break; | |
2130 | } | |
2131 | i = next_zero; | |
2132 | } | |
2133 | ||
2134 | if (!found_bits) | |
4e69b598 | 2135 | return -ENOSPC; |
96303081 JB |
2136 | |
2137 | if (!found) { | |
2138 | start = i; | |
2139 | found = true; | |
2140 | } | |
2141 | ||
2142 | total_found += found_bits; | |
2143 | ||
2144 | if (cluster->max_size < found_bits * block_group->sectorsize) | |
2145 | cluster->max_size = found_bits * block_group->sectorsize; | |
2146 | ||
2147 | if (total_found < total_bits) { | |
2148 | i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, next_zero); | |
2149 | if (i - start > total_bits * 2) { | |
2150 | total_found = 0; | |
2151 | cluster->max_size = 0; | |
2152 | found = false; | |
2153 | } | |
2154 | goto again; | |
2155 | } | |
2156 | ||
2157 | cluster->window_start = start * block_group->sectorsize + | |
2158 | entry->offset; | |
34d52cb6 | 2159 | rb_erase(&entry->offset_index, &ctl->free_space_offset); |
4e69b598 JB |
2160 | ret = tree_insert_offset(&cluster->root, entry->offset, |
2161 | &entry->offset_index, 1); | |
2162 | BUG_ON(ret); | |
96303081 JB |
2163 | |
2164 | return 0; | |
2165 | } | |
2166 | ||
4e69b598 JB |
2167 | /* |
2168 | * This searches the block group for just extents to fill the cluster with. | |
2169 | */ | |
3de85bb9 JB |
2170 | static noinline int |
2171 | setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group, | |
2172 | struct btrfs_free_cluster *cluster, | |
2173 | struct list_head *bitmaps, u64 offset, u64 bytes, | |
2174 | u64 min_bytes) | |
4e69b598 | 2175 | { |
34d52cb6 | 2176 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
4e69b598 JB |
2177 | struct btrfs_free_space *first = NULL; |
2178 | struct btrfs_free_space *entry = NULL; | |
2179 | struct btrfs_free_space *prev = NULL; | |
2180 | struct btrfs_free_space *last; | |
2181 | struct rb_node *node; | |
2182 | u64 window_start; | |
2183 | u64 window_free; | |
2184 | u64 max_extent; | |
2185 | u64 max_gap = 128 * 1024; | |
2186 | ||
34d52cb6 | 2187 | entry = tree_search_offset(ctl, offset, 0, 1); |
4e69b598 JB |
2188 | if (!entry) |
2189 | return -ENOSPC; | |
2190 | ||
2191 | /* | |
2192 | * We don't want bitmaps, so just move along until we find a normal | |
2193 | * extent entry. | |
2194 | */ | |
2195 | while (entry->bitmap) { | |
86d4a77b JB |
2196 | if (list_empty(&entry->list)) |
2197 | list_add_tail(&entry->list, bitmaps); | |
4e69b598 JB |
2198 | node = rb_next(&entry->offset_index); |
2199 | if (!node) | |
2200 | return -ENOSPC; | |
2201 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
2202 | } | |
2203 | ||
2204 | window_start = entry->offset; | |
2205 | window_free = entry->bytes; | |
2206 | max_extent = entry->bytes; | |
2207 | first = entry; | |
2208 | last = entry; | |
2209 | prev = entry; | |
2210 | ||
2211 | while (window_free <= min_bytes) { | |
2212 | node = rb_next(&entry->offset_index); | |
2213 | if (!node) | |
2214 | return -ENOSPC; | |
2215 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
2216 | ||
86d4a77b JB |
2217 | if (entry->bitmap) { |
2218 | if (list_empty(&entry->list)) | |
2219 | list_add_tail(&entry->list, bitmaps); | |
4e69b598 | 2220 | continue; |
86d4a77b JB |
2221 | } |
2222 | ||
4e69b598 JB |
2223 | /* |
2224 | * we haven't filled the empty size and the window is | |
2225 | * very large. reset and try again | |
2226 | */ | |
2227 | if (entry->offset - (prev->offset + prev->bytes) > max_gap || | |
2228 | entry->offset - window_start > (min_bytes * 2)) { | |
2229 | first = entry; | |
2230 | window_start = entry->offset; | |
2231 | window_free = entry->bytes; | |
2232 | last = entry; | |
2233 | max_extent = entry->bytes; | |
2234 | } else { | |
2235 | last = entry; | |
2236 | window_free += entry->bytes; | |
2237 | if (entry->bytes > max_extent) | |
2238 | max_extent = entry->bytes; | |
2239 | } | |
2240 | prev = entry; | |
2241 | } | |
2242 | ||
2243 | cluster->window_start = first->offset; | |
2244 | ||
2245 | node = &first->offset_index; | |
2246 | ||
2247 | /* | |
2248 | * now we've found our entries, pull them out of the free space | |
2249 | * cache and put them into the cluster rbtree | |
2250 | */ | |
2251 | do { | |
2252 | int ret; | |
2253 | ||
2254 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
2255 | node = rb_next(&entry->offset_index); | |
2256 | if (entry->bitmap) | |
2257 | continue; | |
2258 | ||
34d52cb6 | 2259 | rb_erase(&entry->offset_index, &ctl->free_space_offset); |
4e69b598 JB |
2260 | ret = tree_insert_offset(&cluster->root, entry->offset, |
2261 | &entry->offset_index, 0); | |
2262 | BUG_ON(ret); | |
2263 | } while (node && entry != last); | |
2264 | ||
2265 | cluster->max_size = max_extent; | |
2266 | ||
2267 | return 0; | |
2268 | } | |
2269 | ||
2270 | /* | |
2271 | * This specifically looks for bitmaps that may work in the cluster, we assume | |
2272 | * that we have already failed to find extents that will work. | |
2273 | */ | |
3de85bb9 JB |
2274 | static noinline int |
2275 | setup_cluster_bitmap(struct btrfs_block_group_cache *block_group, | |
2276 | struct btrfs_free_cluster *cluster, | |
2277 | struct list_head *bitmaps, u64 offset, u64 bytes, | |
2278 | u64 min_bytes) | |
4e69b598 | 2279 | { |
34d52cb6 | 2280 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
4e69b598 JB |
2281 | struct btrfs_free_space *entry; |
2282 | struct rb_node *node; | |
2283 | int ret = -ENOSPC; | |
2284 | ||
34d52cb6 | 2285 | if (ctl->total_bitmaps == 0) |
4e69b598 JB |
2286 | return -ENOSPC; |
2287 | ||
86d4a77b JB |
2288 | /* |
2289 | * First check our cached list of bitmaps and see if there is an entry | |
2290 | * here that will work. | |
2291 | */ | |
2292 | list_for_each_entry(entry, bitmaps, list) { | |
2293 | if (entry->bytes < min_bytes) | |
2294 | continue; | |
2295 | ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset, | |
2296 | bytes, min_bytes); | |
2297 | if (!ret) | |
2298 | return 0; | |
2299 | } | |
2300 | ||
2301 | /* | |
2302 | * If we do have entries on our list and we are here then we didn't find | |
2303 | * anything, so go ahead and get the next entry after the last entry in | |
2304 | * this list and start the search from there. | |
2305 | */ | |
2306 | if (!list_empty(bitmaps)) { | |
2307 | entry = list_entry(bitmaps->prev, struct btrfs_free_space, | |
2308 | list); | |
2309 | node = rb_next(&entry->offset_index); | |
2310 | if (!node) | |
2311 | return -ENOSPC; | |
2312 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
2313 | goto search; | |
2314 | } | |
2315 | ||
34d52cb6 | 2316 | entry = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), 0, 1); |
4e69b598 JB |
2317 | if (!entry) |
2318 | return -ENOSPC; | |
2319 | ||
86d4a77b | 2320 | search: |
4e69b598 JB |
2321 | node = &entry->offset_index; |
2322 | do { | |
2323 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
2324 | node = rb_next(&entry->offset_index); | |
2325 | if (!entry->bitmap) | |
2326 | continue; | |
2327 | if (entry->bytes < min_bytes) | |
2328 | continue; | |
2329 | ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset, | |
2330 | bytes, min_bytes); | |
2331 | } while (ret && node); | |
2332 | ||
2333 | return ret; | |
2334 | } | |
2335 | ||
fa9c0d79 CM |
2336 | /* |
2337 | * here we try to find a cluster of blocks in a block group. The goal | |
2338 | * is to find at least bytes free and up to empty_size + bytes free. | |
2339 | * We might not find them all in one contiguous area. | |
2340 | * | |
2341 | * returns zero and sets up cluster if things worked out, otherwise | |
2342 | * it returns -enospc | |
2343 | */ | |
2344 | int btrfs_find_space_cluster(struct btrfs_trans_handle *trans, | |
451d7585 | 2345 | struct btrfs_root *root, |
fa9c0d79 CM |
2346 | struct btrfs_block_group_cache *block_group, |
2347 | struct btrfs_free_cluster *cluster, | |
2348 | u64 offset, u64 bytes, u64 empty_size) | |
2349 | { | |
34d52cb6 | 2350 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
86d4a77b JB |
2351 | struct list_head bitmaps; |
2352 | struct btrfs_free_space *entry, *tmp; | |
fa9c0d79 | 2353 | u64 min_bytes; |
fa9c0d79 CM |
2354 | int ret; |
2355 | ||
2356 | /* for metadata, allow allocates with more holes */ | |
451d7585 CM |
2357 | if (btrfs_test_opt(root, SSD_SPREAD)) { |
2358 | min_bytes = bytes + empty_size; | |
2359 | } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) { | |
fa9c0d79 CM |
2360 | /* |
2361 | * we want to do larger allocations when we are | |
2362 | * flushing out the delayed refs, it helps prevent | |
2363 | * making more work as we go along. | |
2364 | */ | |
2365 | if (trans->transaction->delayed_refs.flushing) | |
2366 | min_bytes = max(bytes, (bytes + empty_size) >> 1); | |
2367 | else | |
2368 | min_bytes = max(bytes, (bytes + empty_size) >> 4); | |
2369 | } else | |
2370 | min_bytes = max(bytes, (bytes + empty_size) >> 2); | |
2371 | ||
34d52cb6 | 2372 | spin_lock(&ctl->tree_lock); |
7d0d2e8e JB |
2373 | |
2374 | /* | |
2375 | * If we know we don't have enough space to make a cluster don't even | |
2376 | * bother doing all the work to try and find one. | |
2377 | */ | |
34d52cb6 LZ |
2378 | if (ctl->free_space < min_bytes) { |
2379 | spin_unlock(&ctl->tree_lock); | |
7d0d2e8e JB |
2380 | return -ENOSPC; |
2381 | } | |
2382 | ||
fa9c0d79 CM |
2383 | spin_lock(&cluster->lock); |
2384 | ||
2385 | /* someone already found a cluster, hooray */ | |
2386 | if (cluster->block_group) { | |
2387 | ret = 0; | |
2388 | goto out; | |
2389 | } | |
fa9c0d79 | 2390 | |
86d4a77b JB |
2391 | INIT_LIST_HEAD(&bitmaps); |
2392 | ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset, | |
2393 | bytes, min_bytes); | |
4e69b598 | 2394 | if (ret) |
86d4a77b JB |
2395 | ret = setup_cluster_bitmap(block_group, cluster, &bitmaps, |
2396 | offset, bytes, min_bytes); | |
2397 | ||
2398 | /* Clear our temporary list */ | |
2399 | list_for_each_entry_safe(entry, tmp, &bitmaps, list) | |
2400 | list_del_init(&entry->list); | |
fa9c0d79 | 2401 | |
4e69b598 JB |
2402 | if (!ret) { |
2403 | atomic_inc(&block_group->count); | |
2404 | list_add_tail(&cluster->block_group_list, | |
2405 | &block_group->cluster_list); | |
2406 | cluster->block_group = block_group; | |
fa9c0d79 | 2407 | } |
fa9c0d79 CM |
2408 | out: |
2409 | spin_unlock(&cluster->lock); | |
34d52cb6 | 2410 | spin_unlock(&ctl->tree_lock); |
fa9c0d79 CM |
2411 | |
2412 | return ret; | |
2413 | } | |
2414 | ||
2415 | /* | |
2416 | * simple code to zero out a cluster | |
2417 | */ | |
2418 | void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster) | |
2419 | { | |
2420 | spin_lock_init(&cluster->lock); | |
2421 | spin_lock_init(&cluster->refill_lock); | |
6bef4d31 | 2422 | cluster->root = RB_ROOT; |
fa9c0d79 CM |
2423 | cluster->max_size = 0; |
2424 | INIT_LIST_HEAD(&cluster->block_group_list); | |
2425 | cluster->block_group = NULL; | |
2426 | } | |
2427 | ||
f7039b1d LD |
2428 | int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group, |
2429 | u64 *trimmed, u64 start, u64 end, u64 minlen) | |
2430 | { | |
34d52cb6 | 2431 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
f7039b1d LD |
2432 | struct btrfs_free_space *entry = NULL; |
2433 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
2434 | u64 bytes = 0; | |
2435 | u64 actually_trimmed; | |
2436 | int ret = 0; | |
2437 | ||
2438 | *trimmed = 0; | |
2439 | ||
2440 | while (start < end) { | |
34d52cb6 | 2441 | spin_lock(&ctl->tree_lock); |
f7039b1d | 2442 | |
34d52cb6 LZ |
2443 | if (ctl->free_space < minlen) { |
2444 | spin_unlock(&ctl->tree_lock); | |
f7039b1d LD |
2445 | break; |
2446 | } | |
2447 | ||
34d52cb6 | 2448 | entry = tree_search_offset(ctl, start, 0, 1); |
f7039b1d | 2449 | if (!entry) |
34d52cb6 LZ |
2450 | entry = tree_search_offset(ctl, |
2451 | offset_to_bitmap(ctl, start), | |
f7039b1d LD |
2452 | 1, 1); |
2453 | ||
2454 | if (!entry || entry->offset >= end) { | |
34d52cb6 | 2455 | spin_unlock(&ctl->tree_lock); |
f7039b1d LD |
2456 | break; |
2457 | } | |
2458 | ||
2459 | if (entry->bitmap) { | |
34d52cb6 | 2460 | ret = search_bitmap(ctl, entry, &start, &bytes); |
f7039b1d LD |
2461 | if (!ret) { |
2462 | if (start >= end) { | |
34d52cb6 | 2463 | spin_unlock(&ctl->tree_lock); |
f7039b1d LD |
2464 | break; |
2465 | } | |
2466 | bytes = min(bytes, end - start); | |
34d52cb6 | 2467 | bitmap_clear_bits(ctl, entry, start, bytes); |
f7039b1d | 2468 | if (entry->bytes == 0) |
34d52cb6 | 2469 | free_bitmap(ctl, entry); |
f7039b1d LD |
2470 | } else { |
2471 | start = entry->offset + BITS_PER_BITMAP * | |
2472 | block_group->sectorsize; | |
34d52cb6 | 2473 | spin_unlock(&ctl->tree_lock); |
f7039b1d LD |
2474 | ret = 0; |
2475 | continue; | |
2476 | } | |
2477 | } else { | |
2478 | start = entry->offset; | |
2479 | bytes = min(entry->bytes, end - start); | |
34d52cb6 | 2480 | unlink_free_space(ctl, entry); |
f789b684 | 2481 | kmem_cache_free(btrfs_free_space_cachep, entry); |
f7039b1d LD |
2482 | } |
2483 | ||
34d52cb6 | 2484 | spin_unlock(&ctl->tree_lock); |
f7039b1d LD |
2485 | |
2486 | if (bytes >= minlen) { | |
fb25e914 JB |
2487 | struct btrfs_space_info *space_info; |
2488 | int update = 0; | |
2489 | ||
2490 | space_info = block_group->space_info; | |
2491 | spin_lock(&space_info->lock); | |
2492 | spin_lock(&block_group->lock); | |
2493 | if (!block_group->ro) { | |
2494 | block_group->reserved += bytes; | |
2495 | space_info->bytes_reserved += bytes; | |
2496 | update = 1; | |
2497 | } | |
2498 | spin_unlock(&block_group->lock); | |
2499 | spin_unlock(&space_info->lock); | |
f7039b1d LD |
2500 | |
2501 | ret = btrfs_error_discard_extent(fs_info->extent_root, | |
2502 | start, | |
2503 | bytes, | |
2504 | &actually_trimmed); | |
2505 | ||
34d52cb6 | 2506 | btrfs_add_free_space(block_group, start, bytes); |
fb25e914 JB |
2507 | if (update) { |
2508 | spin_lock(&space_info->lock); | |
2509 | spin_lock(&block_group->lock); | |
2510 | if (block_group->ro) | |
2511 | space_info->bytes_readonly += bytes; | |
2512 | block_group->reserved -= bytes; | |
2513 | space_info->bytes_reserved -= bytes; | |
2514 | spin_unlock(&space_info->lock); | |
2515 | spin_unlock(&block_group->lock); | |
2516 | } | |
f7039b1d LD |
2517 | |
2518 | if (ret) | |
2519 | break; | |
2520 | *trimmed += actually_trimmed; | |
2521 | } | |
2522 | start += bytes; | |
2523 | bytes = 0; | |
2524 | ||
2525 | if (fatal_signal_pending(current)) { | |
2526 | ret = -ERESTARTSYS; | |
2527 | break; | |
2528 | } | |
2529 | ||
2530 | cond_resched(); | |
2531 | } | |
2532 | ||
2533 | return ret; | |
2534 | } | |
581bb050 LZ |
2535 | |
2536 | /* | |
2537 | * Find the left-most item in the cache tree, and then return the | |
2538 | * smallest inode number in the item. | |
2539 | * | |
2540 | * Note: the returned inode number may not be the smallest one in | |
2541 | * the tree, if the left-most item is a bitmap. | |
2542 | */ | |
2543 | u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root) | |
2544 | { | |
2545 | struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl; | |
2546 | struct btrfs_free_space *entry = NULL; | |
2547 | u64 ino = 0; | |
2548 | ||
2549 | spin_lock(&ctl->tree_lock); | |
2550 | ||
2551 | if (RB_EMPTY_ROOT(&ctl->free_space_offset)) | |
2552 | goto out; | |
2553 | ||
2554 | entry = rb_entry(rb_first(&ctl->free_space_offset), | |
2555 | struct btrfs_free_space, offset_index); | |
2556 | ||
2557 | if (!entry->bitmap) { | |
2558 | ino = entry->offset; | |
2559 | ||
2560 | unlink_free_space(ctl, entry); | |
2561 | entry->offset++; | |
2562 | entry->bytes--; | |
2563 | if (!entry->bytes) | |
2564 | kmem_cache_free(btrfs_free_space_cachep, entry); | |
2565 | else | |
2566 | link_free_space(ctl, entry); | |
2567 | } else { | |
2568 | u64 offset = 0; | |
2569 | u64 count = 1; | |
2570 | int ret; | |
2571 | ||
2572 | ret = search_bitmap(ctl, entry, &offset, &count); | |
2573 | BUG_ON(ret); | |
2574 | ||
2575 | ino = offset; | |
2576 | bitmap_clear_bits(ctl, entry, offset, 1); | |
2577 | if (entry->bytes == 0) | |
2578 | free_bitmap(ctl, entry); | |
2579 | } | |
2580 | out: | |
2581 | spin_unlock(&ctl->tree_lock); | |
2582 | ||
2583 | return ino; | |
2584 | } | |
82d5902d LZ |
2585 | |
2586 | struct inode *lookup_free_ino_inode(struct btrfs_root *root, | |
2587 | struct btrfs_path *path) | |
2588 | { | |
2589 | struct inode *inode = NULL; | |
2590 | ||
2591 | spin_lock(&root->cache_lock); | |
2592 | if (root->cache_inode) | |
2593 | inode = igrab(root->cache_inode); | |
2594 | spin_unlock(&root->cache_lock); | |
2595 | if (inode) | |
2596 | return inode; | |
2597 | ||
2598 | inode = __lookup_free_space_inode(root, path, 0); | |
2599 | if (IS_ERR(inode)) | |
2600 | return inode; | |
2601 | ||
2602 | spin_lock(&root->cache_lock); | |
7841cb28 | 2603 | if (!btrfs_fs_closing(root->fs_info)) |
82d5902d LZ |
2604 | root->cache_inode = igrab(inode); |
2605 | spin_unlock(&root->cache_lock); | |
2606 | ||
2607 | return inode; | |
2608 | } | |
2609 | ||
2610 | int create_free_ino_inode(struct btrfs_root *root, | |
2611 | struct btrfs_trans_handle *trans, | |
2612 | struct btrfs_path *path) | |
2613 | { | |
2614 | return __create_free_space_inode(root, trans, path, | |
2615 | BTRFS_FREE_INO_OBJECTID, 0); | |
2616 | } | |
2617 | ||
2618 | int load_free_ino_cache(struct btrfs_fs_info *fs_info, struct btrfs_root *root) | |
2619 | { | |
2620 | struct btrfs_free_space_ctl *ctl = root->free_ino_ctl; | |
2621 | struct btrfs_path *path; | |
2622 | struct inode *inode; | |
2623 | int ret = 0; | |
2624 | u64 root_gen = btrfs_root_generation(&root->root_item); | |
2625 | ||
4b9465cb CM |
2626 | if (!btrfs_test_opt(root, INODE_MAP_CACHE)) |
2627 | return 0; | |
2628 | ||
82d5902d LZ |
2629 | /* |
2630 | * If we're unmounting then just return, since this does a search on the | |
2631 | * normal root and not the commit root and we could deadlock. | |
2632 | */ | |
7841cb28 | 2633 | if (btrfs_fs_closing(fs_info)) |
82d5902d LZ |
2634 | return 0; |
2635 | ||
2636 | path = btrfs_alloc_path(); | |
2637 | if (!path) | |
2638 | return 0; | |
2639 | ||
2640 | inode = lookup_free_ino_inode(root, path); | |
2641 | if (IS_ERR(inode)) | |
2642 | goto out; | |
2643 | ||
2644 | if (root_gen != BTRFS_I(inode)->generation) | |
2645 | goto out_put; | |
2646 | ||
2647 | ret = __load_free_space_cache(root, inode, ctl, path, 0); | |
2648 | ||
2649 | if (ret < 0) | |
2650 | printk(KERN_ERR "btrfs: failed to load free ino cache for " | |
2651 | "root %llu\n", root->root_key.objectid); | |
2652 | out_put: | |
2653 | iput(inode); | |
2654 | out: | |
2655 | btrfs_free_path(path); | |
2656 | return ret; | |
2657 | } | |
2658 | ||
2659 | int btrfs_write_out_ino_cache(struct btrfs_root *root, | |
2660 | struct btrfs_trans_handle *trans, | |
2661 | struct btrfs_path *path) | |
2662 | { | |
2663 | struct btrfs_free_space_ctl *ctl = root->free_ino_ctl; | |
2664 | struct inode *inode; | |
2665 | int ret; | |
2666 | ||
4b9465cb CM |
2667 | if (!btrfs_test_opt(root, INODE_MAP_CACHE)) |
2668 | return 0; | |
2669 | ||
82d5902d LZ |
2670 | inode = lookup_free_ino_inode(root, path); |
2671 | if (IS_ERR(inode)) | |
2672 | return 0; | |
2673 | ||
2674 | ret = __btrfs_write_out_cache(root, inode, ctl, NULL, trans, path, 0); | |
c09544e0 JB |
2675 | if (ret) { |
2676 | btrfs_delalloc_release_metadata(inode, inode->i_size); | |
2677 | #ifdef DEBUG | |
82d5902d LZ |
2678 | printk(KERN_ERR "btrfs: failed to write free ino cache " |
2679 | "for root %llu\n", root->root_key.objectid); | |
c09544e0 JB |
2680 | #endif |
2681 | } | |
82d5902d LZ |
2682 | |
2683 | iput(inode); | |
2684 | return ret; | |
2685 | } |