Commit | Line | Data |
---|---|---|
0f9dd46c JB |
1 | /* |
2 | * Copyright (C) 2008 Red Hat. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
96303081 | 19 | #include <linux/pagemap.h> |
0f9dd46c | 20 | #include <linux/sched.h> |
5a0e3ad6 | 21 | #include <linux/slab.h> |
96303081 | 22 | #include <linux/math64.h> |
0f9dd46c | 23 | #include "ctree.h" |
fa9c0d79 CM |
24 | #include "free-space-cache.h" |
25 | #include "transaction.h" | |
0af3d00b | 26 | #include "disk-io.h" |
43be2146 | 27 | #include "extent_io.h" |
581bb050 | 28 | #include "inode-map.h" |
fa9c0d79 | 29 | |
96303081 JB |
30 | #define BITS_PER_BITMAP (PAGE_CACHE_SIZE * 8) |
31 | #define MAX_CACHE_BYTES_PER_GIG (32 * 1024) | |
0f9dd46c | 32 | |
34d52cb6 | 33 | static int link_free_space(struct btrfs_free_space_ctl *ctl, |
0cb59c99 JB |
34 | struct btrfs_free_space *info); |
35 | ||
0414efae LZ |
36 | static struct inode *__lookup_free_space_inode(struct btrfs_root *root, |
37 | struct btrfs_path *path, | |
38 | u64 offset) | |
0af3d00b JB |
39 | { |
40 | struct btrfs_key key; | |
41 | struct btrfs_key location; | |
42 | struct btrfs_disk_key disk_key; | |
43 | struct btrfs_free_space_header *header; | |
44 | struct extent_buffer *leaf; | |
45 | struct inode *inode = NULL; | |
46 | int ret; | |
47 | ||
0af3d00b | 48 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; |
0414efae | 49 | key.offset = offset; |
0af3d00b JB |
50 | key.type = 0; |
51 | ||
52 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
53 | if (ret < 0) | |
54 | return ERR_PTR(ret); | |
55 | if (ret > 0) { | |
56 | btrfs_release_path(root, path); | |
57 | return ERR_PTR(-ENOENT); | |
58 | } | |
59 | ||
60 | leaf = path->nodes[0]; | |
61 | header = btrfs_item_ptr(leaf, path->slots[0], | |
62 | struct btrfs_free_space_header); | |
63 | btrfs_free_space_key(leaf, header, &disk_key); | |
64 | btrfs_disk_key_to_cpu(&location, &disk_key); | |
65 | btrfs_release_path(root, path); | |
66 | ||
67 | inode = btrfs_iget(root->fs_info->sb, &location, root, NULL); | |
68 | if (!inode) | |
69 | return ERR_PTR(-ENOENT); | |
70 | if (IS_ERR(inode)) | |
71 | return inode; | |
72 | if (is_bad_inode(inode)) { | |
73 | iput(inode); | |
74 | return ERR_PTR(-ENOENT); | |
75 | } | |
76 | ||
adae52b9 MX |
77 | inode->i_mapping->flags &= ~__GFP_FS; |
78 | ||
0414efae LZ |
79 | return inode; |
80 | } | |
81 | ||
82 | struct inode *lookup_free_space_inode(struct btrfs_root *root, | |
83 | struct btrfs_block_group_cache | |
84 | *block_group, struct btrfs_path *path) | |
85 | { | |
86 | struct inode *inode = NULL; | |
87 | ||
88 | spin_lock(&block_group->lock); | |
89 | if (block_group->inode) | |
90 | inode = igrab(block_group->inode); | |
91 | spin_unlock(&block_group->lock); | |
92 | if (inode) | |
93 | return inode; | |
94 | ||
95 | inode = __lookup_free_space_inode(root, path, | |
96 | block_group->key.objectid); | |
97 | if (IS_ERR(inode)) | |
98 | return inode; | |
99 | ||
0af3d00b JB |
100 | spin_lock(&block_group->lock); |
101 | if (!root->fs_info->closing) { | |
102 | block_group->inode = igrab(inode); | |
103 | block_group->iref = 1; | |
104 | } | |
105 | spin_unlock(&block_group->lock); | |
106 | ||
107 | return inode; | |
108 | } | |
109 | ||
0414efae LZ |
110 | int __create_free_space_inode(struct btrfs_root *root, |
111 | struct btrfs_trans_handle *trans, | |
112 | struct btrfs_path *path, u64 ino, u64 offset) | |
0af3d00b JB |
113 | { |
114 | struct btrfs_key key; | |
115 | struct btrfs_disk_key disk_key; | |
116 | struct btrfs_free_space_header *header; | |
117 | struct btrfs_inode_item *inode_item; | |
118 | struct extent_buffer *leaf; | |
0af3d00b JB |
119 | int ret; |
120 | ||
0414efae | 121 | ret = btrfs_insert_empty_inode(trans, root, path, ino); |
0af3d00b JB |
122 | if (ret) |
123 | return ret; | |
124 | ||
125 | leaf = path->nodes[0]; | |
126 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
127 | struct btrfs_inode_item); | |
128 | btrfs_item_key(leaf, &disk_key, path->slots[0]); | |
129 | memset_extent_buffer(leaf, 0, (unsigned long)inode_item, | |
130 | sizeof(*inode_item)); | |
131 | btrfs_set_inode_generation(leaf, inode_item, trans->transid); | |
132 | btrfs_set_inode_size(leaf, inode_item, 0); | |
133 | btrfs_set_inode_nbytes(leaf, inode_item, 0); | |
134 | btrfs_set_inode_uid(leaf, inode_item, 0); | |
135 | btrfs_set_inode_gid(leaf, inode_item, 0); | |
136 | btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600); | |
137 | btrfs_set_inode_flags(leaf, inode_item, BTRFS_INODE_NOCOMPRESS | | |
138 | BTRFS_INODE_PREALLOC | BTRFS_INODE_NODATASUM); | |
139 | btrfs_set_inode_nlink(leaf, inode_item, 1); | |
140 | btrfs_set_inode_transid(leaf, inode_item, trans->transid); | |
0414efae | 141 | btrfs_set_inode_block_group(leaf, inode_item, offset); |
0af3d00b JB |
142 | btrfs_mark_buffer_dirty(leaf); |
143 | btrfs_release_path(root, path); | |
144 | ||
145 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
0414efae | 146 | key.offset = offset; |
0af3d00b JB |
147 | key.type = 0; |
148 | ||
149 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
150 | sizeof(struct btrfs_free_space_header)); | |
151 | if (ret < 0) { | |
152 | btrfs_release_path(root, path); | |
153 | return ret; | |
154 | } | |
155 | leaf = path->nodes[0]; | |
156 | header = btrfs_item_ptr(leaf, path->slots[0], | |
157 | struct btrfs_free_space_header); | |
158 | memset_extent_buffer(leaf, 0, (unsigned long)header, sizeof(*header)); | |
159 | btrfs_set_free_space_key(leaf, header, &disk_key); | |
160 | btrfs_mark_buffer_dirty(leaf); | |
161 | btrfs_release_path(root, path); | |
162 | ||
163 | return 0; | |
164 | } | |
165 | ||
0414efae LZ |
166 | int create_free_space_inode(struct btrfs_root *root, |
167 | struct btrfs_trans_handle *trans, | |
168 | struct btrfs_block_group_cache *block_group, | |
169 | struct btrfs_path *path) | |
170 | { | |
171 | int ret; | |
172 | u64 ino; | |
173 | ||
174 | ret = btrfs_find_free_objectid(root, &ino); | |
175 | if (ret < 0) | |
176 | return ret; | |
177 | ||
178 | return __create_free_space_inode(root, trans, path, ino, | |
179 | block_group->key.objectid); | |
180 | } | |
181 | ||
0af3d00b JB |
182 | int btrfs_truncate_free_space_cache(struct btrfs_root *root, |
183 | struct btrfs_trans_handle *trans, | |
184 | struct btrfs_path *path, | |
185 | struct inode *inode) | |
186 | { | |
187 | loff_t oldsize; | |
188 | int ret = 0; | |
189 | ||
190 | trans->block_rsv = root->orphan_block_rsv; | |
191 | ret = btrfs_block_rsv_check(trans, root, | |
192 | root->orphan_block_rsv, | |
193 | 0, 5); | |
194 | if (ret) | |
195 | return ret; | |
196 | ||
197 | oldsize = i_size_read(inode); | |
198 | btrfs_i_size_write(inode, 0); | |
199 | truncate_pagecache(inode, oldsize, 0); | |
200 | ||
201 | /* | |
202 | * We don't need an orphan item because truncating the free space cache | |
203 | * will never be split across transactions. | |
204 | */ | |
205 | ret = btrfs_truncate_inode_items(trans, root, inode, | |
206 | 0, BTRFS_EXTENT_DATA_KEY); | |
207 | if (ret) { | |
208 | WARN_ON(1); | |
209 | return ret; | |
210 | } | |
211 | ||
212 | return btrfs_update_inode(trans, root, inode); | |
213 | } | |
214 | ||
9d66e233 JB |
215 | static int readahead_cache(struct inode *inode) |
216 | { | |
217 | struct file_ra_state *ra; | |
218 | unsigned long last_index; | |
219 | ||
220 | ra = kzalloc(sizeof(*ra), GFP_NOFS); | |
221 | if (!ra) | |
222 | return -ENOMEM; | |
223 | ||
224 | file_ra_state_init(ra, inode->i_mapping); | |
225 | last_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT; | |
226 | ||
227 | page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index); | |
228 | ||
229 | kfree(ra); | |
230 | ||
231 | return 0; | |
232 | } | |
233 | ||
0414efae LZ |
234 | int __load_free_space_cache(struct btrfs_root *root, struct inode *inode, |
235 | struct btrfs_free_space_ctl *ctl, | |
236 | struct btrfs_path *path, u64 offset) | |
9d66e233 | 237 | { |
9d66e233 JB |
238 | struct btrfs_free_space_header *header; |
239 | struct extent_buffer *leaf; | |
240 | struct page *page; | |
9d66e233 JB |
241 | u32 *checksums = NULL, *crc; |
242 | char *disk_crcs = NULL; | |
243 | struct btrfs_key key; | |
244 | struct list_head bitmaps; | |
245 | u64 num_entries; | |
246 | u64 num_bitmaps; | |
247 | u64 generation; | |
248 | u32 cur_crc = ~(u32)0; | |
249 | pgoff_t index = 0; | |
250 | unsigned long first_page_offset; | |
251 | int num_checksums; | |
0414efae | 252 | int ret = 0, ret2; |
9d66e233 JB |
253 | |
254 | INIT_LIST_HEAD(&bitmaps); | |
255 | ||
9d66e233 | 256 | /* Nothing in the space cache, goodbye */ |
0414efae | 257 | if (!i_size_read(inode)) |
9d66e233 | 258 | goto out; |
9d66e233 JB |
259 | |
260 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
0414efae | 261 | key.offset = offset; |
9d66e233 JB |
262 | key.type = 0; |
263 | ||
264 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
0414efae LZ |
265 | if (ret < 0) |
266 | goto out; | |
267 | else if (ret > 0) { | |
268 | btrfs_release_path(root, path); | |
269 | ret = 0; | |
9d66e233 JB |
270 | goto out; |
271 | } | |
272 | ||
0414efae LZ |
273 | ret = -1; |
274 | ||
9d66e233 JB |
275 | leaf = path->nodes[0]; |
276 | header = btrfs_item_ptr(leaf, path->slots[0], | |
277 | struct btrfs_free_space_header); | |
278 | num_entries = btrfs_free_space_entries(leaf, header); | |
279 | num_bitmaps = btrfs_free_space_bitmaps(leaf, header); | |
280 | generation = btrfs_free_space_generation(leaf, header); | |
0414efae | 281 | btrfs_release_path(root, path); |
9d66e233 JB |
282 | |
283 | if (BTRFS_I(inode)->generation != generation) { | |
284 | printk(KERN_ERR "btrfs: free space inode generation (%llu) did" | |
0414efae | 285 | " not match free space cache generation (%llu)\n", |
9d66e233 | 286 | (unsigned long long)BTRFS_I(inode)->generation, |
0414efae LZ |
287 | (unsigned long long)generation); |
288 | goto out; | |
9d66e233 JB |
289 | } |
290 | ||
291 | if (!num_entries) | |
292 | goto out; | |
293 | ||
294 | /* Setup everything for doing checksumming */ | |
295 | num_checksums = i_size_read(inode) / PAGE_CACHE_SIZE; | |
296 | checksums = crc = kzalloc(sizeof(u32) * num_checksums, GFP_NOFS); | |
297 | if (!checksums) | |
298 | goto out; | |
299 | first_page_offset = (sizeof(u32) * num_checksums) + sizeof(u64); | |
300 | disk_crcs = kzalloc(first_page_offset, GFP_NOFS); | |
301 | if (!disk_crcs) | |
302 | goto out; | |
303 | ||
304 | ret = readahead_cache(inode); | |
0414efae | 305 | if (ret) |
9d66e233 | 306 | goto out; |
9d66e233 JB |
307 | |
308 | while (1) { | |
309 | struct btrfs_free_space_entry *entry; | |
310 | struct btrfs_free_space *e; | |
311 | void *addr; | |
312 | unsigned long offset = 0; | |
313 | unsigned long start_offset = 0; | |
314 | int need_loop = 0; | |
315 | ||
316 | if (!num_entries && !num_bitmaps) | |
317 | break; | |
318 | ||
319 | if (index == 0) { | |
320 | start_offset = first_page_offset; | |
321 | offset = start_offset; | |
322 | } | |
323 | ||
324 | page = grab_cache_page(inode->i_mapping, index); | |
0414efae | 325 | if (!page) |
9d66e233 | 326 | goto free_cache; |
9d66e233 JB |
327 | |
328 | if (!PageUptodate(page)) { | |
329 | btrfs_readpage(NULL, page); | |
330 | lock_page(page); | |
331 | if (!PageUptodate(page)) { | |
332 | unlock_page(page); | |
333 | page_cache_release(page); | |
334 | printk(KERN_ERR "btrfs: error reading free " | |
0414efae | 335 | "space cache\n"); |
9d66e233 JB |
336 | goto free_cache; |
337 | } | |
338 | } | |
339 | addr = kmap(page); | |
340 | ||
341 | if (index == 0) { | |
342 | u64 *gen; | |
343 | ||
344 | memcpy(disk_crcs, addr, first_page_offset); | |
345 | gen = addr + (sizeof(u32) * num_checksums); | |
346 | if (*gen != BTRFS_I(inode)->generation) { | |
347 | printk(KERN_ERR "btrfs: space cache generation" | |
0414efae | 348 | " (%llu) does not match inode (%llu)\n", |
9d66e233 JB |
349 | (unsigned long long)*gen, |
350 | (unsigned long long) | |
0414efae | 351 | BTRFS_I(inode)->generation); |
9d66e233 JB |
352 | kunmap(page); |
353 | unlock_page(page); | |
354 | page_cache_release(page); | |
355 | goto free_cache; | |
356 | } | |
357 | crc = (u32 *)disk_crcs; | |
358 | } | |
359 | entry = addr + start_offset; | |
360 | ||
361 | /* First lets check our crc before we do anything fun */ | |
362 | cur_crc = ~(u32)0; | |
363 | cur_crc = btrfs_csum_data(root, addr + start_offset, cur_crc, | |
364 | PAGE_CACHE_SIZE - start_offset); | |
365 | btrfs_csum_final(cur_crc, (char *)&cur_crc); | |
366 | if (cur_crc != *crc) { | |
0414efae LZ |
367 | printk(KERN_ERR "btrfs: crc mismatch for page %lu\n", |
368 | index); | |
9d66e233 JB |
369 | kunmap(page); |
370 | unlock_page(page); | |
371 | page_cache_release(page); | |
372 | goto free_cache; | |
373 | } | |
374 | crc++; | |
375 | ||
376 | while (1) { | |
377 | if (!num_entries) | |
378 | break; | |
379 | ||
380 | need_loop = 1; | |
dc89e982 JB |
381 | e = kmem_cache_zalloc(btrfs_free_space_cachep, |
382 | GFP_NOFS); | |
9d66e233 JB |
383 | if (!e) { |
384 | kunmap(page); | |
385 | unlock_page(page); | |
386 | page_cache_release(page); | |
387 | goto free_cache; | |
388 | } | |
389 | ||
390 | e->offset = le64_to_cpu(entry->offset); | |
391 | e->bytes = le64_to_cpu(entry->bytes); | |
392 | if (!e->bytes) { | |
393 | kunmap(page); | |
dc89e982 | 394 | kmem_cache_free(btrfs_free_space_cachep, e); |
9d66e233 JB |
395 | unlock_page(page); |
396 | page_cache_release(page); | |
397 | goto free_cache; | |
398 | } | |
399 | ||
400 | if (entry->type == BTRFS_FREE_SPACE_EXTENT) { | |
34d52cb6 LZ |
401 | spin_lock(&ctl->tree_lock); |
402 | ret = link_free_space(ctl, e); | |
403 | spin_unlock(&ctl->tree_lock); | |
9d66e233 JB |
404 | BUG_ON(ret); |
405 | } else { | |
406 | e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS); | |
407 | if (!e->bitmap) { | |
408 | kunmap(page); | |
dc89e982 JB |
409 | kmem_cache_free( |
410 | btrfs_free_space_cachep, e); | |
9d66e233 JB |
411 | unlock_page(page); |
412 | page_cache_release(page); | |
413 | goto free_cache; | |
414 | } | |
34d52cb6 | 415 | spin_lock(&ctl->tree_lock); |
0414efae | 416 | ret2 = link_free_space(ctl, e); |
34d52cb6 LZ |
417 | ctl->total_bitmaps++; |
418 | ctl->op->recalc_thresholds(ctl); | |
419 | spin_unlock(&ctl->tree_lock); | |
9d66e233 JB |
420 | list_add_tail(&e->list, &bitmaps); |
421 | } | |
422 | ||
423 | num_entries--; | |
424 | offset += sizeof(struct btrfs_free_space_entry); | |
425 | if (offset + sizeof(struct btrfs_free_space_entry) >= | |
426 | PAGE_CACHE_SIZE) | |
427 | break; | |
428 | entry++; | |
429 | } | |
430 | ||
431 | /* | |
432 | * We read an entry out of this page, we need to move on to the | |
433 | * next page. | |
434 | */ | |
435 | if (need_loop) { | |
436 | kunmap(page); | |
437 | goto next; | |
438 | } | |
439 | ||
440 | /* | |
441 | * We add the bitmaps at the end of the entries in order that | |
442 | * the bitmap entries are added to the cache. | |
443 | */ | |
444 | e = list_entry(bitmaps.next, struct btrfs_free_space, list); | |
445 | list_del_init(&e->list); | |
446 | memcpy(e->bitmap, addr, PAGE_CACHE_SIZE); | |
447 | kunmap(page); | |
448 | num_bitmaps--; | |
449 | next: | |
450 | unlock_page(page); | |
451 | page_cache_release(page); | |
452 | index++; | |
453 | } | |
454 | ||
455 | ret = 1; | |
456 | out: | |
457 | kfree(checksums); | |
458 | kfree(disk_crcs); | |
9d66e233 | 459 | return ret; |
9d66e233 | 460 | free_cache: |
0414efae | 461 | __btrfs_remove_free_space_cache(ctl); |
9d66e233 JB |
462 | goto out; |
463 | } | |
464 | ||
0414efae LZ |
465 | int load_free_space_cache(struct btrfs_fs_info *fs_info, |
466 | struct btrfs_block_group_cache *block_group) | |
0cb59c99 | 467 | { |
34d52cb6 | 468 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0414efae LZ |
469 | struct btrfs_root *root = fs_info->tree_root; |
470 | struct inode *inode; | |
471 | struct btrfs_path *path; | |
472 | int ret; | |
473 | bool matched; | |
474 | u64 used = btrfs_block_group_used(&block_group->item); | |
475 | ||
476 | /* | |
477 | * If we're unmounting then just return, since this does a search on the | |
478 | * normal root and not the commit root and we could deadlock. | |
479 | */ | |
480 | smp_mb(); | |
481 | if (fs_info->closing) | |
482 | return 0; | |
483 | ||
484 | /* | |
485 | * If this block group has been marked to be cleared for one reason or | |
486 | * another then we can't trust the on disk cache, so just return. | |
487 | */ | |
488 | spin_lock(&block_group->lock); | |
489 | if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) { | |
490 | spin_unlock(&block_group->lock); | |
491 | return 0; | |
492 | } | |
493 | spin_unlock(&block_group->lock); | |
494 | ||
495 | path = btrfs_alloc_path(); | |
496 | if (!path) | |
497 | return 0; | |
498 | ||
499 | inode = lookup_free_space_inode(root, block_group, path); | |
500 | if (IS_ERR(inode)) { | |
501 | btrfs_free_path(path); | |
502 | return 0; | |
503 | } | |
504 | ||
505 | ret = __load_free_space_cache(fs_info->tree_root, inode, ctl, | |
506 | path, block_group->key.objectid); | |
507 | btrfs_free_path(path); | |
508 | if (ret <= 0) | |
509 | goto out; | |
510 | ||
511 | spin_lock(&ctl->tree_lock); | |
512 | matched = (ctl->free_space == (block_group->key.offset - used - | |
513 | block_group->bytes_super)); | |
514 | spin_unlock(&ctl->tree_lock); | |
515 | ||
516 | if (!matched) { | |
517 | __btrfs_remove_free_space_cache(ctl); | |
518 | printk(KERN_ERR "block group %llu has an wrong amount of free " | |
519 | "space\n", block_group->key.objectid); | |
520 | ret = -1; | |
521 | } | |
522 | out: | |
523 | if (ret < 0) { | |
524 | /* This cache is bogus, make sure it gets cleared */ | |
525 | spin_lock(&block_group->lock); | |
526 | block_group->disk_cache_state = BTRFS_DC_CLEAR; | |
527 | spin_unlock(&block_group->lock); | |
528 | ||
529 | printk(KERN_ERR "btrfs: failed to load free space cache " | |
530 | "for block group %llu\n", block_group->key.objectid); | |
531 | } | |
532 | ||
533 | iput(inode); | |
534 | return ret; | |
535 | } | |
536 | ||
537 | int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode, | |
538 | struct btrfs_free_space_ctl *ctl, | |
539 | struct btrfs_block_group_cache *block_group, | |
540 | struct btrfs_trans_handle *trans, | |
541 | struct btrfs_path *path, u64 offset) | |
542 | { | |
0cb59c99 JB |
543 | struct btrfs_free_space_header *header; |
544 | struct extent_buffer *leaf; | |
0cb59c99 JB |
545 | struct rb_node *node; |
546 | struct list_head *pos, *n; | |
be1a12a0 | 547 | struct page **pages; |
0cb59c99 JB |
548 | struct page *page; |
549 | struct extent_state *cached_state = NULL; | |
43be2146 JB |
550 | struct btrfs_free_cluster *cluster = NULL; |
551 | struct extent_io_tree *unpin = NULL; | |
0cb59c99 JB |
552 | struct list_head bitmap_list; |
553 | struct btrfs_key key; | |
43be2146 | 554 | u64 start, end, len; |
0cb59c99 JB |
555 | u64 bytes = 0; |
556 | u32 *crc, *checksums; | |
0cb59c99 | 557 | unsigned long first_page_offset; |
be1a12a0 | 558 | int index = 0, num_pages = 0; |
0cb59c99 JB |
559 | int entries = 0; |
560 | int bitmaps = 0; | |
0414efae | 561 | int ret = -1; |
43be2146 | 562 | bool next_page = false; |
be1a12a0 | 563 | bool out_of_space = false; |
0cb59c99 | 564 | |
0cb59c99 JB |
565 | INIT_LIST_HEAD(&bitmap_list); |
566 | ||
34d52cb6 | 567 | node = rb_first(&ctl->free_space_offset); |
0414efae | 568 | if (!node) |
2b20982e | 569 | return 0; |
0414efae LZ |
570 | |
571 | if (!i_size_read(inode)) | |
572 | return -1; | |
2b20982e | 573 | |
be1a12a0 JB |
574 | num_pages = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> |
575 | PAGE_CACHE_SHIFT; | |
0cb59c99 JB |
576 | filemap_write_and_wait(inode->i_mapping); |
577 | btrfs_wait_ordered_range(inode, inode->i_size & | |
578 | ~(root->sectorsize - 1), (u64)-1); | |
579 | ||
580 | /* We need a checksum per page. */ | |
be1a12a0 | 581 | crc = checksums = kzalloc(sizeof(u32) * num_pages, GFP_NOFS); |
0414efae LZ |
582 | if (!crc) |
583 | return -1; | |
0cb59c99 | 584 | |
be1a12a0 JB |
585 | pages = kzalloc(sizeof(struct page *) * num_pages, GFP_NOFS); |
586 | if (!pages) { | |
587 | kfree(crc); | |
0414efae | 588 | return -1; |
be1a12a0 JB |
589 | } |
590 | ||
0cb59c99 JB |
591 | /* Since the first page has all of our checksums and our generation we |
592 | * need to calculate the offset into the page that we can start writing | |
593 | * our entries. | |
594 | */ | |
be1a12a0 | 595 | first_page_offset = (sizeof(u32) * num_pages) + sizeof(u64); |
0cb59c99 | 596 | |
43be2146 | 597 | /* Get the cluster for this block_group if it exists */ |
0414efae | 598 | if (block_group && !list_empty(&block_group->cluster_list)) |
43be2146 JB |
599 | cluster = list_entry(block_group->cluster_list.next, |
600 | struct btrfs_free_cluster, | |
601 | block_group_list); | |
602 | ||
603 | /* | |
604 | * We shouldn't have switched the pinned extents yet so this is the | |
605 | * right one | |
606 | */ | |
607 | unpin = root->fs_info->pinned_extents; | |
608 | ||
0cb59c99 JB |
609 | /* |
610 | * Lock all pages first so we can lock the extent safely. | |
611 | * | |
612 | * NOTE: Because we hold the ref the entire time we're going to write to | |
613 | * the page find_get_page should never fail, so we don't do a check | |
614 | * after find_get_page at this point. Just putting this here so people | |
615 | * know and don't freak out. | |
616 | */ | |
be1a12a0 | 617 | while (index < num_pages) { |
0cb59c99 JB |
618 | page = grab_cache_page(inode->i_mapping, index); |
619 | if (!page) { | |
be1a12a0 | 620 | int i; |
0cb59c99 | 621 | |
be1a12a0 JB |
622 | for (i = 0; i < num_pages; i++) { |
623 | unlock_page(pages[i]); | |
624 | page_cache_release(pages[i]); | |
0cb59c99 JB |
625 | } |
626 | goto out_free; | |
627 | } | |
be1a12a0 | 628 | pages[index] = page; |
0cb59c99 JB |
629 | index++; |
630 | } | |
631 | ||
632 | index = 0; | |
633 | lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1, | |
634 | 0, &cached_state, GFP_NOFS); | |
635 | ||
43be2146 JB |
636 | /* |
637 | * When searching for pinned extents, we need to start at our start | |
638 | * offset. | |
639 | */ | |
0414efae LZ |
640 | if (block_group) |
641 | start = block_group->key.objectid; | |
43be2146 | 642 | |
0cb59c99 JB |
643 | /* Write out the extent entries */ |
644 | do { | |
645 | struct btrfs_free_space_entry *entry; | |
646 | void *addr; | |
647 | unsigned long offset = 0; | |
648 | unsigned long start_offset = 0; | |
649 | ||
43be2146 JB |
650 | next_page = false; |
651 | ||
0cb59c99 JB |
652 | if (index == 0) { |
653 | start_offset = first_page_offset; | |
654 | offset = start_offset; | |
655 | } | |
656 | ||
be1a12a0 JB |
657 | if (index >= num_pages) { |
658 | out_of_space = true; | |
659 | break; | |
660 | } | |
661 | ||
662 | page = pages[index]; | |
0cb59c99 JB |
663 | |
664 | addr = kmap(page); | |
665 | entry = addr + start_offset; | |
666 | ||
667 | memset(addr, 0, PAGE_CACHE_SIZE); | |
43be2146 | 668 | while (node && !next_page) { |
0cb59c99 JB |
669 | struct btrfs_free_space *e; |
670 | ||
671 | e = rb_entry(node, struct btrfs_free_space, offset_index); | |
672 | entries++; | |
673 | ||
674 | entry->offset = cpu_to_le64(e->offset); | |
675 | entry->bytes = cpu_to_le64(e->bytes); | |
676 | if (e->bitmap) { | |
677 | entry->type = BTRFS_FREE_SPACE_BITMAP; | |
678 | list_add_tail(&e->list, &bitmap_list); | |
679 | bitmaps++; | |
680 | } else { | |
681 | entry->type = BTRFS_FREE_SPACE_EXTENT; | |
682 | } | |
683 | node = rb_next(node); | |
43be2146 JB |
684 | if (!node && cluster) { |
685 | node = rb_first(&cluster->root); | |
686 | cluster = NULL; | |
687 | } | |
0cb59c99 JB |
688 | offset += sizeof(struct btrfs_free_space_entry); |
689 | if (offset + sizeof(struct btrfs_free_space_entry) >= | |
690 | PAGE_CACHE_SIZE) | |
43be2146 JB |
691 | next_page = true; |
692 | entry++; | |
693 | } | |
694 | ||
695 | /* | |
696 | * We want to add any pinned extents to our free space cache | |
697 | * so we don't leak the space | |
698 | */ | |
0414efae LZ |
699 | while (block_group && !next_page && |
700 | (start < block_group->key.objectid + | |
701 | block_group->key.offset)) { | |
43be2146 JB |
702 | ret = find_first_extent_bit(unpin, start, &start, &end, |
703 | EXTENT_DIRTY); | |
704 | if (ret) { | |
705 | ret = 0; | |
706 | break; | |
707 | } | |
708 | ||
709 | /* This pinned extent is out of our range */ | |
710 | if (start >= block_group->key.objectid + | |
711 | block_group->key.offset) | |
0cb59c99 | 712 | break; |
43be2146 JB |
713 | |
714 | len = block_group->key.objectid + | |
715 | block_group->key.offset - start; | |
716 | len = min(len, end + 1 - start); | |
717 | ||
718 | entries++; | |
719 | entry->offset = cpu_to_le64(start); | |
720 | entry->bytes = cpu_to_le64(len); | |
721 | entry->type = BTRFS_FREE_SPACE_EXTENT; | |
722 | ||
723 | start = end + 1; | |
724 | offset += sizeof(struct btrfs_free_space_entry); | |
725 | if (offset + sizeof(struct btrfs_free_space_entry) >= | |
726 | PAGE_CACHE_SIZE) | |
727 | next_page = true; | |
0cb59c99 JB |
728 | entry++; |
729 | } | |
730 | *crc = ~(u32)0; | |
731 | *crc = btrfs_csum_data(root, addr + start_offset, *crc, | |
732 | PAGE_CACHE_SIZE - start_offset); | |
733 | kunmap(page); | |
734 | ||
735 | btrfs_csum_final(*crc, (char *)crc); | |
736 | crc++; | |
737 | ||
738 | bytes += PAGE_CACHE_SIZE; | |
739 | ||
0cb59c99 | 740 | index++; |
43be2146 | 741 | } while (node || next_page); |
0cb59c99 JB |
742 | |
743 | /* Write out the bitmaps */ | |
744 | list_for_each_safe(pos, n, &bitmap_list) { | |
745 | void *addr; | |
746 | struct btrfs_free_space *entry = | |
747 | list_entry(pos, struct btrfs_free_space, list); | |
748 | ||
be1a12a0 JB |
749 | if (index >= num_pages) { |
750 | out_of_space = true; | |
751 | break; | |
752 | } | |
f65647c2 | 753 | page = pages[index]; |
0cb59c99 JB |
754 | |
755 | addr = kmap(page); | |
756 | memcpy(addr, entry->bitmap, PAGE_CACHE_SIZE); | |
757 | *crc = ~(u32)0; | |
758 | *crc = btrfs_csum_data(root, addr, *crc, PAGE_CACHE_SIZE); | |
759 | kunmap(page); | |
760 | btrfs_csum_final(*crc, (char *)crc); | |
761 | crc++; | |
762 | bytes += PAGE_CACHE_SIZE; | |
763 | ||
0cb59c99 JB |
764 | list_del_init(&entry->list); |
765 | index++; | |
766 | } | |
767 | ||
be1a12a0 JB |
768 | if (out_of_space) { |
769 | btrfs_drop_pages(pages, num_pages); | |
770 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0, | |
771 | i_size_read(inode) - 1, &cached_state, | |
772 | GFP_NOFS); | |
773 | ret = 0; | |
774 | goto out_free; | |
775 | } | |
776 | ||
0cb59c99 | 777 | /* Zero out the rest of the pages just to make sure */ |
be1a12a0 | 778 | while (index < num_pages) { |
0cb59c99 JB |
779 | void *addr; |
780 | ||
be1a12a0 | 781 | page = pages[index]; |
0cb59c99 JB |
782 | addr = kmap(page); |
783 | memset(addr, 0, PAGE_CACHE_SIZE); | |
784 | kunmap(page); | |
0cb59c99 JB |
785 | bytes += PAGE_CACHE_SIZE; |
786 | index++; | |
787 | } | |
788 | ||
0cb59c99 JB |
789 | /* Write the checksums and trans id to the first page */ |
790 | { | |
791 | void *addr; | |
792 | u64 *gen; | |
793 | ||
be1a12a0 | 794 | page = pages[0]; |
0cb59c99 JB |
795 | |
796 | addr = kmap(page); | |
be1a12a0 JB |
797 | memcpy(addr, checksums, sizeof(u32) * num_pages); |
798 | gen = addr + (sizeof(u32) * num_pages); | |
0cb59c99 JB |
799 | *gen = trans->transid; |
800 | kunmap(page); | |
0cb59c99 | 801 | } |
0cb59c99 | 802 | |
be1a12a0 JB |
803 | ret = btrfs_dirty_pages(root, inode, pages, num_pages, 0, |
804 | bytes, &cached_state); | |
805 | btrfs_drop_pages(pages, num_pages); | |
0cb59c99 JB |
806 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0, |
807 | i_size_read(inode) - 1, &cached_state, GFP_NOFS); | |
808 | ||
be1a12a0 JB |
809 | if (ret) { |
810 | ret = 0; | |
811 | goto out_free; | |
812 | } | |
813 | ||
814 | BTRFS_I(inode)->generation = trans->transid; | |
815 | ||
0cb59c99 JB |
816 | filemap_write_and_wait(inode->i_mapping); |
817 | ||
818 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
0414efae | 819 | key.offset = offset; |
0cb59c99 JB |
820 | key.type = 0; |
821 | ||
822 | ret = btrfs_search_slot(trans, root, &key, path, 1, 1); | |
823 | if (ret < 0) { | |
0414efae | 824 | ret = -1; |
0cb59c99 JB |
825 | clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, bytes - 1, |
826 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
827 | EXTENT_DO_ACCOUNTING, 0, 0, NULL, GFP_NOFS); | |
828 | goto out_free; | |
829 | } | |
830 | leaf = path->nodes[0]; | |
831 | if (ret > 0) { | |
832 | struct btrfs_key found_key; | |
833 | BUG_ON(!path->slots[0]); | |
834 | path->slots[0]--; | |
835 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
836 | if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID || | |
0414efae LZ |
837 | found_key.offset != offset) { |
838 | ret = -1; | |
0cb59c99 JB |
839 | clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, bytes - 1, |
840 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
841 | EXTENT_DO_ACCOUNTING, 0, 0, NULL, | |
842 | GFP_NOFS); | |
843 | btrfs_release_path(root, path); | |
844 | goto out_free; | |
845 | } | |
846 | } | |
847 | header = btrfs_item_ptr(leaf, path->slots[0], | |
848 | struct btrfs_free_space_header); | |
849 | btrfs_set_free_space_entries(leaf, header, entries); | |
850 | btrfs_set_free_space_bitmaps(leaf, header, bitmaps); | |
851 | btrfs_set_free_space_generation(leaf, header, trans->transid); | |
852 | btrfs_mark_buffer_dirty(leaf); | |
853 | btrfs_release_path(root, path); | |
854 | ||
855 | ret = 1; | |
856 | ||
857 | out_free: | |
0414efae | 858 | if (ret != 1) { |
0cb59c99 | 859 | invalidate_inode_pages2_range(inode->i_mapping, 0, index); |
0cb59c99 JB |
860 | BTRFS_I(inode)->generation = 0; |
861 | } | |
862 | kfree(checksums); | |
be1a12a0 | 863 | kfree(pages); |
0cb59c99 | 864 | btrfs_update_inode(trans, root, inode); |
0414efae LZ |
865 | return ret; |
866 | } | |
867 | ||
868 | int btrfs_write_out_cache(struct btrfs_root *root, | |
869 | struct btrfs_trans_handle *trans, | |
870 | struct btrfs_block_group_cache *block_group, | |
871 | struct btrfs_path *path) | |
872 | { | |
873 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
874 | struct inode *inode; | |
875 | int ret = 0; | |
876 | ||
877 | root = root->fs_info->tree_root; | |
878 | ||
879 | spin_lock(&block_group->lock); | |
880 | if (block_group->disk_cache_state < BTRFS_DC_SETUP) { | |
881 | spin_unlock(&block_group->lock); | |
882 | return 0; | |
883 | } | |
884 | spin_unlock(&block_group->lock); | |
885 | ||
886 | inode = lookup_free_space_inode(root, block_group, path); | |
887 | if (IS_ERR(inode)) | |
888 | return 0; | |
889 | ||
890 | ret = __btrfs_write_out_cache(root, inode, ctl, block_group, trans, | |
891 | path, block_group->key.objectid); | |
892 | if (ret < 0) { | |
893 | spin_lock(&block_group->lock); | |
894 | block_group->disk_cache_state = BTRFS_DC_ERROR; | |
895 | spin_unlock(&block_group->lock); | |
896 | ||
897 | printk(KERN_ERR "btrfs: failed to write free space cace " | |
898 | "for block group %llu\n", block_group->key.objectid); | |
899 | } | |
900 | ||
0cb59c99 JB |
901 | iput(inode); |
902 | return ret; | |
903 | } | |
904 | ||
34d52cb6 | 905 | static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit, |
96303081 | 906 | u64 offset) |
0f9dd46c | 907 | { |
96303081 JB |
908 | BUG_ON(offset < bitmap_start); |
909 | offset -= bitmap_start; | |
34d52cb6 | 910 | return (unsigned long)(div_u64(offset, unit)); |
96303081 | 911 | } |
0f9dd46c | 912 | |
34d52cb6 | 913 | static inline unsigned long bytes_to_bits(u64 bytes, u32 unit) |
96303081 | 914 | { |
34d52cb6 | 915 | return (unsigned long)(div_u64(bytes, unit)); |
96303081 | 916 | } |
0f9dd46c | 917 | |
34d52cb6 | 918 | static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
919 | u64 offset) |
920 | { | |
921 | u64 bitmap_start; | |
922 | u64 bytes_per_bitmap; | |
0f9dd46c | 923 | |
34d52cb6 LZ |
924 | bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit; |
925 | bitmap_start = offset - ctl->start; | |
96303081 JB |
926 | bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap); |
927 | bitmap_start *= bytes_per_bitmap; | |
34d52cb6 | 928 | bitmap_start += ctl->start; |
0f9dd46c | 929 | |
96303081 | 930 | return bitmap_start; |
0f9dd46c JB |
931 | } |
932 | ||
96303081 JB |
933 | static int tree_insert_offset(struct rb_root *root, u64 offset, |
934 | struct rb_node *node, int bitmap) | |
0f9dd46c JB |
935 | { |
936 | struct rb_node **p = &root->rb_node; | |
937 | struct rb_node *parent = NULL; | |
938 | struct btrfs_free_space *info; | |
939 | ||
940 | while (*p) { | |
941 | parent = *p; | |
96303081 | 942 | info = rb_entry(parent, struct btrfs_free_space, offset_index); |
0f9dd46c | 943 | |
96303081 | 944 | if (offset < info->offset) { |
0f9dd46c | 945 | p = &(*p)->rb_left; |
96303081 | 946 | } else if (offset > info->offset) { |
0f9dd46c | 947 | p = &(*p)->rb_right; |
96303081 JB |
948 | } else { |
949 | /* | |
950 | * we could have a bitmap entry and an extent entry | |
951 | * share the same offset. If this is the case, we want | |
952 | * the extent entry to always be found first if we do a | |
953 | * linear search through the tree, since we want to have | |
954 | * the quickest allocation time, and allocating from an | |
955 | * extent is faster than allocating from a bitmap. So | |
956 | * if we're inserting a bitmap and we find an entry at | |
957 | * this offset, we want to go right, or after this entry | |
958 | * logically. If we are inserting an extent and we've | |
959 | * found a bitmap, we want to go left, or before | |
960 | * logically. | |
961 | */ | |
962 | if (bitmap) { | |
963 | WARN_ON(info->bitmap); | |
964 | p = &(*p)->rb_right; | |
965 | } else { | |
966 | WARN_ON(!info->bitmap); | |
967 | p = &(*p)->rb_left; | |
968 | } | |
969 | } | |
0f9dd46c JB |
970 | } |
971 | ||
972 | rb_link_node(node, parent, p); | |
973 | rb_insert_color(node, root); | |
974 | ||
975 | return 0; | |
976 | } | |
977 | ||
978 | /* | |
70cb0743 JB |
979 | * searches the tree for the given offset. |
980 | * | |
96303081 JB |
981 | * fuzzy - If this is set, then we are trying to make an allocation, and we just |
982 | * want a section that has at least bytes size and comes at or after the given | |
983 | * offset. | |
0f9dd46c | 984 | */ |
96303081 | 985 | static struct btrfs_free_space * |
34d52cb6 | 986 | tree_search_offset(struct btrfs_free_space_ctl *ctl, |
96303081 | 987 | u64 offset, int bitmap_only, int fuzzy) |
0f9dd46c | 988 | { |
34d52cb6 | 989 | struct rb_node *n = ctl->free_space_offset.rb_node; |
96303081 JB |
990 | struct btrfs_free_space *entry, *prev = NULL; |
991 | ||
992 | /* find entry that is closest to the 'offset' */ | |
993 | while (1) { | |
994 | if (!n) { | |
995 | entry = NULL; | |
996 | break; | |
997 | } | |
0f9dd46c | 998 | |
0f9dd46c | 999 | entry = rb_entry(n, struct btrfs_free_space, offset_index); |
96303081 | 1000 | prev = entry; |
0f9dd46c | 1001 | |
96303081 | 1002 | if (offset < entry->offset) |
0f9dd46c | 1003 | n = n->rb_left; |
96303081 | 1004 | else if (offset > entry->offset) |
0f9dd46c | 1005 | n = n->rb_right; |
96303081 | 1006 | else |
0f9dd46c | 1007 | break; |
0f9dd46c JB |
1008 | } |
1009 | ||
96303081 JB |
1010 | if (bitmap_only) { |
1011 | if (!entry) | |
1012 | return NULL; | |
1013 | if (entry->bitmap) | |
1014 | return entry; | |
0f9dd46c | 1015 | |
96303081 JB |
1016 | /* |
1017 | * bitmap entry and extent entry may share same offset, | |
1018 | * in that case, bitmap entry comes after extent entry. | |
1019 | */ | |
1020 | n = rb_next(n); | |
1021 | if (!n) | |
1022 | return NULL; | |
1023 | entry = rb_entry(n, struct btrfs_free_space, offset_index); | |
1024 | if (entry->offset != offset) | |
1025 | return NULL; | |
0f9dd46c | 1026 | |
96303081 JB |
1027 | WARN_ON(!entry->bitmap); |
1028 | return entry; | |
1029 | } else if (entry) { | |
1030 | if (entry->bitmap) { | |
0f9dd46c | 1031 | /* |
96303081 JB |
1032 | * if previous extent entry covers the offset, |
1033 | * we should return it instead of the bitmap entry | |
0f9dd46c | 1034 | */ |
96303081 JB |
1035 | n = &entry->offset_index; |
1036 | while (1) { | |
1037 | n = rb_prev(n); | |
1038 | if (!n) | |
1039 | break; | |
1040 | prev = rb_entry(n, struct btrfs_free_space, | |
1041 | offset_index); | |
1042 | if (!prev->bitmap) { | |
1043 | if (prev->offset + prev->bytes > offset) | |
1044 | entry = prev; | |
1045 | break; | |
1046 | } | |
0f9dd46c | 1047 | } |
96303081 JB |
1048 | } |
1049 | return entry; | |
1050 | } | |
1051 | ||
1052 | if (!prev) | |
1053 | return NULL; | |
1054 | ||
1055 | /* find last entry before the 'offset' */ | |
1056 | entry = prev; | |
1057 | if (entry->offset > offset) { | |
1058 | n = rb_prev(&entry->offset_index); | |
1059 | if (n) { | |
1060 | entry = rb_entry(n, struct btrfs_free_space, | |
1061 | offset_index); | |
1062 | BUG_ON(entry->offset > offset); | |
0f9dd46c | 1063 | } else { |
96303081 JB |
1064 | if (fuzzy) |
1065 | return entry; | |
1066 | else | |
1067 | return NULL; | |
0f9dd46c JB |
1068 | } |
1069 | } | |
1070 | ||
96303081 JB |
1071 | if (entry->bitmap) { |
1072 | n = &entry->offset_index; | |
1073 | while (1) { | |
1074 | n = rb_prev(n); | |
1075 | if (!n) | |
1076 | break; | |
1077 | prev = rb_entry(n, struct btrfs_free_space, | |
1078 | offset_index); | |
1079 | if (!prev->bitmap) { | |
1080 | if (prev->offset + prev->bytes > offset) | |
1081 | return prev; | |
1082 | break; | |
1083 | } | |
1084 | } | |
34d52cb6 | 1085 | if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset) |
96303081 JB |
1086 | return entry; |
1087 | } else if (entry->offset + entry->bytes > offset) | |
1088 | return entry; | |
1089 | ||
1090 | if (!fuzzy) | |
1091 | return NULL; | |
1092 | ||
1093 | while (1) { | |
1094 | if (entry->bitmap) { | |
1095 | if (entry->offset + BITS_PER_BITMAP * | |
34d52cb6 | 1096 | ctl->unit > offset) |
96303081 JB |
1097 | break; |
1098 | } else { | |
1099 | if (entry->offset + entry->bytes > offset) | |
1100 | break; | |
1101 | } | |
1102 | ||
1103 | n = rb_next(&entry->offset_index); | |
1104 | if (!n) | |
1105 | return NULL; | |
1106 | entry = rb_entry(n, struct btrfs_free_space, offset_index); | |
1107 | } | |
1108 | return entry; | |
0f9dd46c JB |
1109 | } |
1110 | ||
f333adb5 | 1111 | static inline void |
34d52cb6 | 1112 | __unlink_free_space(struct btrfs_free_space_ctl *ctl, |
f333adb5 | 1113 | struct btrfs_free_space *info) |
0f9dd46c | 1114 | { |
34d52cb6 LZ |
1115 | rb_erase(&info->offset_index, &ctl->free_space_offset); |
1116 | ctl->free_extents--; | |
f333adb5 LZ |
1117 | } |
1118 | ||
34d52cb6 | 1119 | static void unlink_free_space(struct btrfs_free_space_ctl *ctl, |
f333adb5 LZ |
1120 | struct btrfs_free_space *info) |
1121 | { | |
34d52cb6 LZ |
1122 | __unlink_free_space(ctl, info); |
1123 | ctl->free_space -= info->bytes; | |
0f9dd46c JB |
1124 | } |
1125 | ||
34d52cb6 | 1126 | static int link_free_space(struct btrfs_free_space_ctl *ctl, |
0f9dd46c JB |
1127 | struct btrfs_free_space *info) |
1128 | { | |
1129 | int ret = 0; | |
1130 | ||
96303081 | 1131 | BUG_ON(!info->bitmap && !info->bytes); |
34d52cb6 | 1132 | ret = tree_insert_offset(&ctl->free_space_offset, info->offset, |
96303081 | 1133 | &info->offset_index, (info->bitmap != NULL)); |
0f9dd46c JB |
1134 | if (ret) |
1135 | return ret; | |
1136 | ||
34d52cb6 LZ |
1137 | ctl->free_space += info->bytes; |
1138 | ctl->free_extents++; | |
96303081 JB |
1139 | return ret; |
1140 | } | |
1141 | ||
34d52cb6 | 1142 | static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl) |
96303081 | 1143 | { |
34d52cb6 | 1144 | struct btrfs_block_group_cache *block_group = ctl->private; |
25891f79 JB |
1145 | u64 max_bytes; |
1146 | u64 bitmap_bytes; | |
1147 | u64 extent_bytes; | |
8eb2d829 | 1148 | u64 size = block_group->key.offset; |
34d52cb6 LZ |
1149 | u64 bytes_per_bg = BITS_PER_BITMAP * block_group->sectorsize; |
1150 | int max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg); | |
1151 | ||
1152 | BUG_ON(ctl->total_bitmaps > max_bitmaps); | |
96303081 JB |
1153 | |
1154 | /* | |
1155 | * The goal is to keep the total amount of memory used per 1gb of space | |
1156 | * at or below 32k, so we need to adjust how much memory we allow to be | |
1157 | * used by extent based free space tracking | |
1158 | */ | |
8eb2d829 LZ |
1159 | if (size < 1024 * 1024 * 1024) |
1160 | max_bytes = MAX_CACHE_BYTES_PER_GIG; | |
1161 | else | |
1162 | max_bytes = MAX_CACHE_BYTES_PER_GIG * | |
1163 | div64_u64(size, 1024 * 1024 * 1024); | |
96303081 | 1164 | |
25891f79 JB |
1165 | /* |
1166 | * we want to account for 1 more bitmap than what we have so we can make | |
1167 | * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as | |
1168 | * we add more bitmaps. | |
1169 | */ | |
34d52cb6 | 1170 | bitmap_bytes = (ctl->total_bitmaps + 1) * PAGE_CACHE_SIZE; |
96303081 | 1171 | |
25891f79 | 1172 | if (bitmap_bytes >= max_bytes) { |
34d52cb6 | 1173 | ctl->extents_thresh = 0; |
25891f79 JB |
1174 | return; |
1175 | } | |
96303081 | 1176 | |
25891f79 JB |
1177 | /* |
1178 | * we want the extent entry threshold to always be at most 1/2 the maxw | |
1179 | * bytes we can have, or whatever is less than that. | |
1180 | */ | |
1181 | extent_bytes = max_bytes - bitmap_bytes; | |
1182 | extent_bytes = min_t(u64, extent_bytes, div64_u64(max_bytes, 2)); | |
96303081 | 1183 | |
34d52cb6 | 1184 | ctl->extents_thresh = |
25891f79 | 1185 | div64_u64(extent_bytes, (sizeof(struct btrfs_free_space))); |
96303081 JB |
1186 | } |
1187 | ||
34d52cb6 | 1188 | static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl, |
817d52f8 JB |
1189 | struct btrfs_free_space *info, u64 offset, |
1190 | u64 bytes) | |
96303081 | 1191 | { |
f38b6e75 | 1192 | unsigned long start, count; |
96303081 | 1193 | |
34d52cb6 LZ |
1194 | start = offset_to_bit(info->offset, ctl->unit, offset); |
1195 | count = bytes_to_bits(bytes, ctl->unit); | |
f38b6e75 | 1196 | BUG_ON(start + count > BITS_PER_BITMAP); |
96303081 | 1197 | |
f38b6e75 | 1198 | bitmap_clear(info->bitmap, start, count); |
96303081 JB |
1199 | |
1200 | info->bytes -= bytes; | |
34d52cb6 | 1201 | ctl->free_space -= bytes; |
96303081 JB |
1202 | } |
1203 | ||
34d52cb6 | 1204 | static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl, |
817d52f8 JB |
1205 | struct btrfs_free_space *info, u64 offset, |
1206 | u64 bytes) | |
96303081 | 1207 | { |
f38b6e75 | 1208 | unsigned long start, count; |
96303081 | 1209 | |
34d52cb6 LZ |
1210 | start = offset_to_bit(info->offset, ctl->unit, offset); |
1211 | count = bytes_to_bits(bytes, ctl->unit); | |
f38b6e75 | 1212 | BUG_ON(start + count > BITS_PER_BITMAP); |
96303081 | 1213 | |
f38b6e75 | 1214 | bitmap_set(info->bitmap, start, count); |
96303081 JB |
1215 | |
1216 | info->bytes += bytes; | |
34d52cb6 | 1217 | ctl->free_space += bytes; |
96303081 JB |
1218 | } |
1219 | ||
34d52cb6 | 1220 | static int search_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1221 | struct btrfs_free_space *bitmap_info, u64 *offset, |
1222 | u64 *bytes) | |
1223 | { | |
1224 | unsigned long found_bits = 0; | |
1225 | unsigned long bits, i; | |
1226 | unsigned long next_zero; | |
1227 | ||
34d52cb6 | 1228 | i = offset_to_bit(bitmap_info->offset, ctl->unit, |
96303081 | 1229 | max_t(u64, *offset, bitmap_info->offset)); |
34d52cb6 | 1230 | bits = bytes_to_bits(*bytes, ctl->unit); |
96303081 JB |
1231 | |
1232 | for (i = find_next_bit(bitmap_info->bitmap, BITS_PER_BITMAP, i); | |
1233 | i < BITS_PER_BITMAP; | |
1234 | i = find_next_bit(bitmap_info->bitmap, BITS_PER_BITMAP, i + 1)) { | |
1235 | next_zero = find_next_zero_bit(bitmap_info->bitmap, | |
1236 | BITS_PER_BITMAP, i); | |
1237 | if ((next_zero - i) >= bits) { | |
1238 | found_bits = next_zero - i; | |
1239 | break; | |
1240 | } | |
1241 | i = next_zero; | |
1242 | } | |
1243 | ||
1244 | if (found_bits) { | |
34d52cb6 LZ |
1245 | *offset = (u64)(i * ctl->unit) + bitmap_info->offset; |
1246 | *bytes = (u64)(found_bits) * ctl->unit; | |
96303081 JB |
1247 | return 0; |
1248 | } | |
1249 | ||
1250 | return -1; | |
1251 | } | |
1252 | ||
34d52cb6 LZ |
1253 | static struct btrfs_free_space * |
1254 | find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes) | |
96303081 JB |
1255 | { |
1256 | struct btrfs_free_space *entry; | |
1257 | struct rb_node *node; | |
1258 | int ret; | |
1259 | ||
34d52cb6 | 1260 | if (!ctl->free_space_offset.rb_node) |
96303081 JB |
1261 | return NULL; |
1262 | ||
34d52cb6 | 1263 | entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1); |
96303081 JB |
1264 | if (!entry) |
1265 | return NULL; | |
1266 | ||
1267 | for (node = &entry->offset_index; node; node = rb_next(node)) { | |
1268 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
1269 | if (entry->bytes < *bytes) | |
1270 | continue; | |
1271 | ||
1272 | if (entry->bitmap) { | |
34d52cb6 | 1273 | ret = search_bitmap(ctl, entry, offset, bytes); |
96303081 JB |
1274 | if (!ret) |
1275 | return entry; | |
1276 | continue; | |
1277 | } | |
1278 | ||
1279 | *offset = entry->offset; | |
1280 | *bytes = entry->bytes; | |
1281 | return entry; | |
1282 | } | |
1283 | ||
1284 | return NULL; | |
1285 | } | |
1286 | ||
34d52cb6 | 1287 | static void add_new_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1288 | struct btrfs_free_space *info, u64 offset) |
1289 | { | |
34d52cb6 | 1290 | info->offset = offset_to_bitmap(ctl, offset); |
f019f426 | 1291 | info->bytes = 0; |
34d52cb6 LZ |
1292 | link_free_space(ctl, info); |
1293 | ctl->total_bitmaps++; | |
96303081 | 1294 | |
34d52cb6 | 1295 | ctl->op->recalc_thresholds(ctl); |
96303081 JB |
1296 | } |
1297 | ||
34d52cb6 | 1298 | static void free_bitmap(struct btrfs_free_space_ctl *ctl, |
edf6e2d1 LZ |
1299 | struct btrfs_free_space *bitmap_info) |
1300 | { | |
34d52cb6 | 1301 | unlink_free_space(ctl, bitmap_info); |
edf6e2d1 | 1302 | kfree(bitmap_info->bitmap); |
dc89e982 | 1303 | kmem_cache_free(btrfs_free_space_cachep, bitmap_info); |
34d52cb6 LZ |
1304 | ctl->total_bitmaps--; |
1305 | ctl->op->recalc_thresholds(ctl); | |
edf6e2d1 LZ |
1306 | } |
1307 | ||
34d52cb6 | 1308 | static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1309 | struct btrfs_free_space *bitmap_info, |
1310 | u64 *offset, u64 *bytes) | |
1311 | { | |
1312 | u64 end; | |
6606bb97 JB |
1313 | u64 search_start, search_bytes; |
1314 | int ret; | |
96303081 JB |
1315 | |
1316 | again: | |
34d52cb6 | 1317 | end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1; |
96303081 | 1318 | |
6606bb97 JB |
1319 | /* |
1320 | * XXX - this can go away after a few releases. | |
1321 | * | |
1322 | * since the only user of btrfs_remove_free_space is the tree logging | |
1323 | * stuff, and the only way to test that is under crash conditions, we | |
1324 | * want to have this debug stuff here just in case somethings not | |
1325 | * working. Search the bitmap for the space we are trying to use to | |
1326 | * make sure its actually there. If its not there then we need to stop | |
1327 | * because something has gone wrong. | |
1328 | */ | |
1329 | search_start = *offset; | |
1330 | search_bytes = *bytes; | |
13dbc089 | 1331 | search_bytes = min(search_bytes, end - search_start + 1); |
34d52cb6 | 1332 | ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes); |
6606bb97 JB |
1333 | BUG_ON(ret < 0 || search_start != *offset); |
1334 | ||
96303081 | 1335 | if (*offset > bitmap_info->offset && *offset + *bytes > end) { |
34d52cb6 | 1336 | bitmap_clear_bits(ctl, bitmap_info, *offset, end - *offset + 1); |
96303081 JB |
1337 | *bytes -= end - *offset + 1; |
1338 | *offset = end + 1; | |
1339 | } else if (*offset >= bitmap_info->offset && *offset + *bytes <= end) { | |
34d52cb6 | 1340 | bitmap_clear_bits(ctl, bitmap_info, *offset, *bytes); |
96303081 JB |
1341 | *bytes = 0; |
1342 | } | |
1343 | ||
1344 | if (*bytes) { | |
6606bb97 | 1345 | struct rb_node *next = rb_next(&bitmap_info->offset_index); |
edf6e2d1 | 1346 | if (!bitmap_info->bytes) |
34d52cb6 | 1347 | free_bitmap(ctl, bitmap_info); |
96303081 | 1348 | |
6606bb97 JB |
1349 | /* |
1350 | * no entry after this bitmap, but we still have bytes to | |
1351 | * remove, so something has gone wrong. | |
1352 | */ | |
1353 | if (!next) | |
96303081 JB |
1354 | return -EINVAL; |
1355 | ||
6606bb97 JB |
1356 | bitmap_info = rb_entry(next, struct btrfs_free_space, |
1357 | offset_index); | |
1358 | ||
1359 | /* | |
1360 | * if the next entry isn't a bitmap we need to return to let the | |
1361 | * extent stuff do its work. | |
1362 | */ | |
96303081 JB |
1363 | if (!bitmap_info->bitmap) |
1364 | return -EAGAIN; | |
1365 | ||
6606bb97 JB |
1366 | /* |
1367 | * Ok the next item is a bitmap, but it may not actually hold | |
1368 | * the information for the rest of this free space stuff, so | |
1369 | * look for it, and if we don't find it return so we can try | |
1370 | * everything over again. | |
1371 | */ | |
1372 | search_start = *offset; | |
1373 | search_bytes = *bytes; | |
34d52cb6 | 1374 | ret = search_bitmap(ctl, bitmap_info, &search_start, |
6606bb97 JB |
1375 | &search_bytes); |
1376 | if (ret < 0 || search_start != *offset) | |
1377 | return -EAGAIN; | |
1378 | ||
96303081 | 1379 | goto again; |
edf6e2d1 | 1380 | } else if (!bitmap_info->bytes) |
34d52cb6 | 1381 | free_bitmap(ctl, bitmap_info); |
96303081 JB |
1382 | |
1383 | return 0; | |
1384 | } | |
1385 | ||
34d52cb6 LZ |
1386 | static bool use_bitmap(struct btrfs_free_space_ctl *ctl, |
1387 | struct btrfs_free_space *info) | |
96303081 | 1388 | { |
34d52cb6 | 1389 | struct btrfs_block_group_cache *block_group = ctl->private; |
96303081 JB |
1390 | |
1391 | /* | |
1392 | * If we are below the extents threshold then we can add this as an | |
1393 | * extent, and don't have to deal with the bitmap | |
1394 | */ | |
34d52cb6 | 1395 | if (ctl->free_extents < ctl->extents_thresh) { |
32cb0840 JB |
1396 | /* |
1397 | * If this block group has some small extents we don't want to | |
1398 | * use up all of our free slots in the cache with them, we want | |
1399 | * to reserve them to larger extents, however if we have plent | |
1400 | * of cache left then go ahead an dadd them, no sense in adding | |
1401 | * the overhead of a bitmap if we don't have to. | |
1402 | */ | |
1403 | if (info->bytes <= block_group->sectorsize * 4) { | |
34d52cb6 LZ |
1404 | if (ctl->free_extents * 2 <= ctl->extents_thresh) |
1405 | return false; | |
32cb0840 | 1406 | } else { |
34d52cb6 | 1407 | return false; |
32cb0840 JB |
1408 | } |
1409 | } | |
96303081 JB |
1410 | |
1411 | /* | |
1412 | * some block groups are so tiny they can't be enveloped by a bitmap, so | |
1413 | * don't even bother to create a bitmap for this | |
1414 | */ | |
1415 | if (BITS_PER_BITMAP * block_group->sectorsize > | |
1416 | block_group->key.offset) | |
34d52cb6 LZ |
1417 | return false; |
1418 | ||
1419 | return true; | |
1420 | } | |
1421 | ||
1422 | static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl, | |
1423 | struct btrfs_free_space *info) | |
1424 | { | |
1425 | struct btrfs_free_space *bitmap_info; | |
1426 | int added = 0; | |
1427 | u64 bytes, offset, end; | |
1428 | int ret; | |
96303081 JB |
1429 | |
1430 | bytes = info->bytes; | |
1431 | offset = info->offset; | |
1432 | ||
34d52cb6 LZ |
1433 | if (!ctl->op->use_bitmap(ctl, info)) |
1434 | return 0; | |
1435 | ||
96303081 | 1436 | again: |
34d52cb6 | 1437 | bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), |
96303081 JB |
1438 | 1, 0); |
1439 | if (!bitmap_info) { | |
1440 | BUG_ON(added); | |
1441 | goto new_bitmap; | |
1442 | } | |
1443 | ||
34d52cb6 | 1444 | end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit); |
96303081 JB |
1445 | |
1446 | if (offset >= bitmap_info->offset && offset + bytes > end) { | |
34d52cb6 | 1447 | bitmap_set_bits(ctl, bitmap_info, offset, end - offset); |
96303081 JB |
1448 | bytes -= end - offset; |
1449 | offset = end; | |
1450 | added = 0; | |
1451 | } else if (offset >= bitmap_info->offset && offset + bytes <= end) { | |
34d52cb6 | 1452 | bitmap_set_bits(ctl, bitmap_info, offset, bytes); |
96303081 JB |
1453 | bytes = 0; |
1454 | } else { | |
1455 | BUG(); | |
1456 | } | |
1457 | ||
1458 | if (!bytes) { | |
1459 | ret = 1; | |
1460 | goto out; | |
1461 | } else | |
1462 | goto again; | |
1463 | ||
1464 | new_bitmap: | |
1465 | if (info && info->bitmap) { | |
34d52cb6 | 1466 | add_new_bitmap(ctl, info, offset); |
96303081 JB |
1467 | added = 1; |
1468 | info = NULL; | |
1469 | goto again; | |
1470 | } else { | |
34d52cb6 | 1471 | spin_unlock(&ctl->tree_lock); |
96303081 JB |
1472 | |
1473 | /* no pre-allocated info, allocate a new one */ | |
1474 | if (!info) { | |
dc89e982 JB |
1475 | info = kmem_cache_zalloc(btrfs_free_space_cachep, |
1476 | GFP_NOFS); | |
96303081 | 1477 | if (!info) { |
34d52cb6 | 1478 | spin_lock(&ctl->tree_lock); |
96303081 JB |
1479 | ret = -ENOMEM; |
1480 | goto out; | |
1481 | } | |
1482 | } | |
1483 | ||
1484 | /* allocate the bitmap */ | |
1485 | info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS); | |
34d52cb6 | 1486 | spin_lock(&ctl->tree_lock); |
96303081 JB |
1487 | if (!info->bitmap) { |
1488 | ret = -ENOMEM; | |
1489 | goto out; | |
1490 | } | |
1491 | goto again; | |
1492 | } | |
1493 | ||
1494 | out: | |
1495 | if (info) { | |
1496 | if (info->bitmap) | |
1497 | kfree(info->bitmap); | |
dc89e982 | 1498 | kmem_cache_free(btrfs_free_space_cachep, info); |
96303081 | 1499 | } |
0f9dd46c JB |
1500 | |
1501 | return ret; | |
1502 | } | |
1503 | ||
34d52cb6 | 1504 | bool try_merge_free_space(struct btrfs_free_space_ctl *ctl, |
f333adb5 | 1505 | struct btrfs_free_space *info, bool update_stat) |
0f9dd46c | 1506 | { |
120d66ee LZ |
1507 | struct btrfs_free_space *left_info; |
1508 | struct btrfs_free_space *right_info; | |
1509 | bool merged = false; | |
1510 | u64 offset = info->offset; | |
1511 | u64 bytes = info->bytes; | |
6226cb0a | 1512 | |
0f9dd46c JB |
1513 | /* |
1514 | * first we want to see if there is free space adjacent to the range we | |
1515 | * are adding, if there is remove that struct and add a new one to | |
1516 | * cover the entire range | |
1517 | */ | |
34d52cb6 | 1518 | right_info = tree_search_offset(ctl, offset + bytes, 0, 0); |
96303081 JB |
1519 | if (right_info && rb_prev(&right_info->offset_index)) |
1520 | left_info = rb_entry(rb_prev(&right_info->offset_index), | |
1521 | struct btrfs_free_space, offset_index); | |
1522 | else | |
34d52cb6 | 1523 | left_info = tree_search_offset(ctl, offset - 1, 0, 0); |
0f9dd46c | 1524 | |
96303081 | 1525 | if (right_info && !right_info->bitmap) { |
f333adb5 | 1526 | if (update_stat) |
34d52cb6 | 1527 | unlink_free_space(ctl, right_info); |
f333adb5 | 1528 | else |
34d52cb6 | 1529 | __unlink_free_space(ctl, right_info); |
6226cb0a | 1530 | info->bytes += right_info->bytes; |
dc89e982 | 1531 | kmem_cache_free(btrfs_free_space_cachep, right_info); |
120d66ee | 1532 | merged = true; |
0f9dd46c JB |
1533 | } |
1534 | ||
96303081 JB |
1535 | if (left_info && !left_info->bitmap && |
1536 | left_info->offset + left_info->bytes == offset) { | |
f333adb5 | 1537 | if (update_stat) |
34d52cb6 | 1538 | unlink_free_space(ctl, left_info); |
f333adb5 | 1539 | else |
34d52cb6 | 1540 | __unlink_free_space(ctl, left_info); |
6226cb0a JB |
1541 | info->offset = left_info->offset; |
1542 | info->bytes += left_info->bytes; | |
dc89e982 | 1543 | kmem_cache_free(btrfs_free_space_cachep, left_info); |
120d66ee | 1544 | merged = true; |
0f9dd46c JB |
1545 | } |
1546 | ||
120d66ee LZ |
1547 | return merged; |
1548 | } | |
1549 | ||
581bb050 LZ |
1550 | int __btrfs_add_free_space(struct btrfs_free_space_ctl *ctl, |
1551 | u64 offset, u64 bytes) | |
120d66ee LZ |
1552 | { |
1553 | struct btrfs_free_space *info; | |
1554 | int ret = 0; | |
1555 | ||
dc89e982 | 1556 | info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS); |
120d66ee LZ |
1557 | if (!info) |
1558 | return -ENOMEM; | |
1559 | ||
1560 | info->offset = offset; | |
1561 | info->bytes = bytes; | |
1562 | ||
34d52cb6 | 1563 | spin_lock(&ctl->tree_lock); |
120d66ee | 1564 | |
34d52cb6 | 1565 | if (try_merge_free_space(ctl, info, true)) |
120d66ee LZ |
1566 | goto link; |
1567 | ||
1568 | /* | |
1569 | * There was no extent directly to the left or right of this new | |
1570 | * extent then we know we're going to have to allocate a new extent, so | |
1571 | * before we do that see if we need to drop this into a bitmap | |
1572 | */ | |
34d52cb6 | 1573 | ret = insert_into_bitmap(ctl, info); |
120d66ee LZ |
1574 | if (ret < 0) { |
1575 | goto out; | |
1576 | } else if (ret) { | |
1577 | ret = 0; | |
1578 | goto out; | |
1579 | } | |
1580 | link: | |
34d52cb6 | 1581 | ret = link_free_space(ctl, info); |
0f9dd46c | 1582 | if (ret) |
dc89e982 | 1583 | kmem_cache_free(btrfs_free_space_cachep, info); |
96303081 | 1584 | out: |
34d52cb6 | 1585 | spin_unlock(&ctl->tree_lock); |
6226cb0a | 1586 | |
0f9dd46c | 1587 | if (ret) { |
96303081 | 1588 | printk(KERN_CRIT "btrfs: unable to add free space :%d\n", ret); |
c293498b | 1589 | BUG_ON(ret == -EEXIST); |
0f9dd46c JB |
1590 | } |
1591 | ||
0f9dd46c JB |
1592 | return ret; |
1593 | } | |
1594 | ||
6226cb0a JB |
1595 | int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group, |
1596 | u64 offset, u64 bytes) | |
0f9dd46c | 1597 | { |
34d52cb6 | 1598 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0f9dd46c | 1599 | struct btrfs_free_space *info; |
96303081 | 1600 | struct btrfs_free_space *next_info = NULL; |
0f9dd46c JB |
1601 | int ret = 0; |
1602 | ||
34d52cb6 | 1603 | spin_lock(&ctl->tree_lock); |
6226cb0a | 1604 | |
96303081 | 1605 | again: |
34d52cb6 | 1606 | info = tree_search_offset(ctl, offset, 0, 0); |
96303081 | 1607 | if (!info) { |
6606bb97 JB |
1608 | /* |
1609 | * oops didn't find an extent that matched the space we wanted | |
1610 | * to remove, look for a bitmap instead | |
1611 | */ | |
34d52cb6 | 1612 | info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), |
6606bb97 JB |
1613 | 1, 0); |
1614 | if (!info) { | |
1615 | WARN_ON(1); | |
1616 | goto out_lock; | |
1617 | } | |
96303081 JB |
1618 | } |
1619 | ||
1620 | if (info->bytes < bytes && rb_next(&info->offset_index)) { | |
1621 | u64 end; | |
1622 | next_info = rb_entry(rb_next(&info->offset_index), | |
1623 | struct btrfs_free_space, | |
1624 | offset_index); | |
1625 | ||
1626 | if (next_info->bitmap) | |
34d52cb6 LZ |
1627 | end = next_info->offset + |
1628 | BITS_PER_BITMAP * ctl->unit - 1; | |
96303081 JB |
1629 | else |
1630 | end = next_info->offset + next_info->bytes; | |
1631 | ||
1632 | if (next_info->bytes < bytes || | |
1633 | next_info->offset > offset || offset > end) { | |
1634 | printk(KERN_CRIT "Found free space at %llu, size %llu," | |
1635 | " trying to use %llu\n", | |
1636 | (unsigned long long)info->offset, | |
1637 | (unsigned long long)info->bytes, | |
1638 | (unsigned long long)bytes); | |
0f9dd46c JB |
1639 | WARN_ON(1); |
1640 | ret = -EINVAL; | |
96303081 | 1641 | goto out_lock; |
0f9dd46c | 1642 | } |
0f9dd46c | 1643 | |
96303081 JB |
1644 | info = next_info; |
1645 | } | |
1646 | ||
1647 | if (info->bytes == bytes) { | |
34d52cb6 | 1648 | unlink_free_space(ctl, info); |
96303081 JB |
1649 | if (info->bitmap) { |
1650 | kfree(info->bitmap); | |
34d52cb6 | 1651 | ctl->total_bitmaps--; |
0f9dd46c | 1652 | } |
dc89e982 | 1653 | kmem_cache_free(btrfs_free_space_cachep, info); |
96303081 JB |
1654 | goto out_lock; |
1655 | } | |
0f9dd46c | 1656 | |
96303081 | 1657 | if (!info->bitmap && info->offset == offset) { |
34d52cb6 | 1658 | unlink_free_space(ctl, info); |
0f9dd46c JB |
1659 | info->offset += bytes; |
1660 | info->bytes -= bytes; | |
34d52cb6 | 1661 | link_free_space(ctl, info); |
96303081 JB |
1662 | goto out_lock; |
1663 | } | |
0f9dd46c | 1664 | |
96303081 JB |
1665 | if (!info->bitmap && info->offset <= offset && |
1666 | info->offset + info->bytes >= offset + bytes) { | |
9b49c9b9 CM |
1667 | u64 old_start = info->offset; |
1668 | /* | |
1669 | * we're freeing space in the middle of the info, | |
1670 | * this can happen during tree log replay | |
1671 | * | |
1672 | * first unlink the old info and then | |
1673 | * insert it again after the hole we're creating | |
1674 | */ | |
34d52cb6 | 1675 | unlink_free_space(ctl, info); |
9b49c9b9 CM |
1676 | if (offset + bytes < info->offset + info->bytes) { |
1677 | u64 old_end = info->offset + info->bytes; | |
1678 | ||
1679 | info->offset = offset + bytes; | |
1680 | info->bytes = old_end - info->offset; | |
34d52cb6 | 1681 | ret = link_free_space(ctl, info); |
96303081 JB |
1682 | WARN_ON(ret); |
1683 | if (ret) | |
1684 | goto out_lock; | |
9b49c9b9 CM |
1685 | } else { |
1686 | /* the hole we're creating ends at the end | |
1687 | * of the info struct, just free the info | |
1688 | */ | |
dc89e982 | 1689 | kmem_cache_free(btrfs_free_space_cachep, info); |
9b49c9b9 | 1690 | } |
34d52cb6 | 1691 | spin_unlock(&ctl->tree_lock); |
96303081 JB |
1692 | |
1693 | /* step two, insert a new info struct to cover | |
1694 | * anything before the hole | |
9b49c9b9 | 1695 | */ |
6226cb0a JB |
1696 | ret = btrfs_add_free_space(block_group, old_start, |
1697 | offset - old_start); | |
96303081 JB |
1698 | WARN_ON(ret); |
1699 | goto out; | |
0f9dd46c | 1700 | } |
96303081 | 1701 | |
34d52cb6 | 1702 | ret = remove_from_bitmap(ctl, info, &offset, &bytes); |
96303081 JB |
1703 | if (ret == -EAGAIN) |
1704 | goto again; | |
1705 | BUG_ON(ret); | |
1706 | out_lock: | |
34d52cb6 | 1707 | spin_unlock(&ctl->tree_lock); |
0f9dd46c | 1708 | out: |
25179201 JB |
1709 | return ret; |
1710 | } | |
1711 | ||
0f9dd46c JB |
1712 | void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group, |
1713 | u64 bytes) | |
1714 | { | |
34d52cb6 | 1715 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0f9dd46c JB |
1716 | struct btrfs_free_space *info; |
1717 | struct rb_node *n; | |
1718 | int count = 0; | |
1719 | ||
34d52cb6 | 1720 | for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) { |
0f9dd46c JB |
1721 | info = rb_entry(n, struct btrfs_free_space, offset_index); |
1722 | if (info->bytes >= bytes) | |
1723 | count++; | |
96303081 | 1724 | printk(KERN_CRIT "entry offset %llu, bytes %llu, bitmap %s\n", |
21380931 | 1725 | (unsigned long long)info->offset, |
96303081 JB |
1726 | (unsigned long long)info->bytes, |
1727 | (info->bitmap) ? "yes" : "no"); | |
0f9dd46c | 1728 | } |
96303081 JB |
1729 | printk(KERN_INFO "block group has cluster?: %s\n", |
1730 | list_empty(&block_group->cluster_list) ? "no" : "yes"); | |
0f9dd46c JB |
1731 | printk(KERN_INFO "%d blocks of free space at or bigger than bytes is" |
1732 | "\n", count); | |
1733 | } | |
1734 | ||
34d52cb6 LZ |
1735 | static struct btrfs_free_space_op free_space_op = { |
1736 | .recalc_thresholds = recalculate_thresholds, | |
1737 | .use_bitmap = use_bitmap, | |
1738 | }; | |
1739 | ||
1740 | void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group) | |
1741 | { | |
1742 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
1743 | ||
1744 | spin_lock_init(&ctl->tree_lock); | |
1745 | ctl->unit = block_group->sectorsize; | |
1746 | ctl->start = block_group->key.objectid; | |
1747 | ctl->private = block_group; | |
1748 | ctl->op = &free_space_op; | |
1749 | ||
1750 | /* | |
1751 | * we only want to have 32k of ram per block group for keeping | |
1752 | * track of free space, and if we pass 1/2 of that we want to | |
1753 | * start converting things over to using bitmaps | |
1754 | */ | |
1755 | ctl->extents_thresh = ((1024 * 32) / 2) / | |
1756 | sizeof(struct btrfs_free_space); | |
1757 | } | |
1758 | ||
fa9c0d79 CM |
1759 | /* |
1760 | * for a given cluster, put all of its extents back into the free | |
1761 | * space cache. If the block group passed doesn't match the block group | |
1762 | * pointed to by the cluster, someone else raced in and freed the | |
1763 | * cluster already. In that case, we just return without changing anything | |
1764 | */ | |
1765 | static int | |
1766 | __btrfs_return_cluster_to_free_space( | |
1767 | struct btrfs_block_group_cache *block_group, | |
1768 | struct btrfs_free_cluster *cluster) | |
1769 | { | |
34d52cb6 | 1770 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
fa9c0d79 CM |
1771 | struct btrfs_free_space *entry; |
1772 | struct rb_node *node; | |
1773 | ||
1774 | spin_lock(&cluster->lock); | |
1775 | if (cluster->block_group != block_group) | |
1776 | goto out; | |
1777 | ||
96303081 | 1778 | cluster->block_group = NULL; |
fa9c0d79 | 1779 | cluster->window_start = 0; |
96303081 | 1780 | list_del_init(&cluster->block_group_list); |
96303081 | 1781 | |
fa9c0d79 | 1782 | node = rb_first(&cluster->root); |
96303081 | 1783 | while (node) { |
4e69b598 JB |
1784 | bool bitmap; |
1785 | ||
fa9c0d79 CM |
1786 | entry = rb_entry(node, struct btrfs_free_space, offset_index); |
1787 | node = rb_next(&entry->offset_index); | |
1788 | rb_erase(&entry->offset_index, &cluster->root); | |
4e69b598 JB |
1789 | |
1790 | bitmap = (entry->bitmap != NULL); | |
1791 | if (!bitmap) | |
34d52cb6 LZ |
1792 | try_merge_free_space(ctl, entry, false); |
1793 | tree_insert_offset(&ctl->free_space_offset, | |
4e69b598 | 1794 | entry->offset, &entry->offset_index, bitmap); |
fa9c0d79 | 1795 | } |
6bef4d31 | 1796 | cluster->root = RB_ROOT; |
96303081 | 1797 | |
fa9c0d79 CM |
1798 | out: |
1799 | spin_unlock(&cluster->lock); | |
96303081 | 1800 | btrfs_put_block_group(block_group); |
fa9c0d79 CM |
1801 | return 0; |
1802 | } | |
1803 | ||
581bb050 | 1804 | void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl) |
0f9dd46c JB |
1805 | { |
1806 | struct btrfs_free_space *info; | |
1807 | struct rb_node *node; | |
581bb050 LZ |
1808 | |
1809 | spin_lock(&ctl->tree_lock); | |
1810 | while ((node = rb_last(&ctl->free_space_offset)) != NULL) { | |
1811 | info = rb_entry(node, struct btrfs_free_space, offset_index); | |
1812 | unlink_free_space(ctl, info); | |
1813 | kfree(info->bitmap); | |
1814 | kmem_cache_free(btrfs_free_space_cachep, info); | |
1815 | if (need_resched()) { | |
1816 | spin_unlock(&ctl->tree_lock); | |
1817 | cond_resched(); | |
1818 | spin_lock(&ctl->tree_lock); | |
1819 | } | |
1820 | } | |
1821 | spin_unlock(&ctl->tree_lock); | |
1822 | } | |
1823 | ||
1824 | void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group) | |
1825 | { | |
1826 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
fa9c0d79 | 1827 | struct btrfs_free_cluster *cluster; |
96303081 | 1828 | struct list_head *head; |
0f9dd46c | 1829 | |
34d52cb6 | 1830 | spin_lock(&ctl->tree_lock); |
96303081 JB |
1831 | while ((head = block_group->cluster_list.next) != |
1832 | &block_group->cluster_list) { | |
1833 | cluster = list_entry(head, struct btrfs_free_cluster, | |
1834 | block_group_list); | |
fa9c0d79 CM |
1835 | |
1836 | WARN_ON(cluster->block_group != block_group); | |
1837 | __btrfs_return_cluster_to_free_space(block_group, cluster); | |
96303081 | 1838 | if (need_resched()) { |
34d52cb6 | 1839 | spin_unlock(&ctl->tree_lock); |
96303081 | 1840 | cond_resched(); |
34d52cb6 | 1841 | spin_lock(&ctl->tree_lock); |
96303081 | 1842 | } |
fa9c0d79 | 1843 | } |
34d52cb6 | 1844 | spin_unlock(&ctl->tree_lock); |
581bb050 LZ |
1845 | |
1846 | __btrfs_remove_free_space_cache(ctl); | |
0f9dd46c JB |
1847 | } |
1848 | ||
6226cb0a JB |
1849 | u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group, |
1850 | u64 offset, u64 bytes, u64 empty_size) | |
0f9dd46c | 1851 | { |
34d52cb6 | 1852 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
6226cb0a | 1853 | struct btrfs_free_space *entry = NULL; |
96303081 | 1854 | u64 bytes_search = bytes + empty_size; |
6226cb0a | 1855 | u64 ret = 0; |
0f9dd46c | 1856 | |
34d52cb6 LZ |
1857 | spin_lock(&ctl->tree_lock); |
1858 | entry = find_free_space(ctl, &offset, &bytes_search); | |
6226cb0a | 1859 | if (!entry) |
96303081 JB |
1860 | goto out; |
1861 | ||
1862 | ret = offset; | |
1863 | if (entry->bitmap) { | |
34d52cb6 | 1864 | bitmap_clear_bits(ctl, entry, offset, bytes); |
edf6e2d1 | 1865 | if (!entry->bytes) |
34d52cb6 | 1866 | free_bitmap(ctl, entry); |
96303081 | 1867 | } else { |
34d52cb6 | 1868 | unlink_free_space(ctl, entry); |
6226cb0a JB |
1869 | entry->offset += bytes; |
1870 | entry->bytes -= bytes; | |
6226cb0a | 1871 | if (!entry->bytes) |
dc89e982 | 1872 | kmem_cache_free(btrfs_free_space_cachep, entry); |
6226cb0a | 1873 | else |
34d52cb6 | 1874 | link_free_space(ctl, entry); |
6226cb0a | 1875 | } |
0f9dd46c | 1876 | |
96303081 | 1877 | out: |
34d52cb6 | 1878 | spin_unlock(&ctl->tree_lock); |
817d52f8 | 1879 | |
0f9dd46c JB |
1880 | return ret; |
1881 | } | |
fa9c0d79 CM |
1882 | |
1883 | /* | |
1884 | * given a cluster, put all of its extents back into the free space | |
1885 | * cache. If a block group is passed, this function will only free | |
1886 | * a cluster that belongs to the passed block group. | |
1887 | * | |
1888 | * Otherwise, it'll get a reference on the block group pointed to by the | |
1889 | * cluster and remove the cluster from it. | |
1890 | */ | |
1891 | int btrfs_return_cluster_to_free_space( | |
1892 | struct btrfs_block_group_cache *block_group, | |
1893 | struct btrfs_free_cluster *cluster) | |
1894 | { | |
34d52cb6 | 1895 | struct btrfs_free_space_ctl *ctl; |
fa9c0d79 CM |
1896 | int ret; |
1897 | ||
1898 | /* first, get a safe pointer to the block group */ | |
1899 | spin_lock(&cluster->lock); | |
1900 | if (!block_group) { | |
1901 | block_group = cluster->block_group; | |
1902 | if (!block_group) { | |
1903 | spin_unlock(&cluster->lock); | |
1904 | return 0; | |
1905 | } | |
1906 | } else if (cluster->block_group != block_group) { | |
1907 | /* someone else has already freed it don't redo their work */ | |
1908 | spin_unlock(&cluster->lock); | |
1909 | return 0; | |
1910 | } | |
1911 | atomic_inc(&block_group->count); | |
1912 | spin_unlock(&cluster->lock); | |
1913 | ||
34d52cb6 LZ |
1914 | ctl = block_group->free_space_ctl; |
1915 | ||
fa9c0d79 | 1916 | /* now return any extents the cluster had on it */ |
34d52cb6 | 1917 | spin_lock(&ctl->tree_lock); |
fa9c0d79 | 1918 | ret = __btrfs_return_cluster_to_free_space(block_group, cluster); |
34d52cb6 | 1919 | spin_unlock(&ctl->tree_lock); |
fa9c0d79 CM |
1920 | |
1921 | /* finally drop our ref */ | |
1922 | btrfs_put_block_group(block_group); | |
1923 | return ret; | |
1924 | } | |
1925 | ||
96303081 JB |
1926 | static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group, |
1927 | struct btrfs_free_cluster *cluster, | |
4e69b598 | 1928 | struct btrfs_free_space *entry, |
96303081 JB |
1929 | u64 bytes, u64 min_start) |
1930 | { | |
34d52cb6 | 1931 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
96303081 JB |
1932 | int err; |
1933 | u64 search_start = cluster->window_start; | |
1934 | u64 search_bytes = bytes; | |
1935 | u64 ret = 0; | |
1936 | ||
96303081 JB |
1937 | search_start = min_start; |
1938 | search_bytes = bytes; | |
1939 | ||
34d52cb6 | 1940 | err = search_bitmap(ctl, entry, &search_start, &search_bytes); |
96303081 | 1941 | if (err) |
4e69b598 | 1942 | return 0; |
96303081 JB |
1943 | |
1944 | ret = search_start; | |
34d52cb6 | 1945 | bitmap_clear_bits(ctl, entry, ret, bytes); |
96303081 JB |
1946 | |
1947 | return ret; | |
1948 | } | |
1949 | ||
fa9c0d79 CM |
1950 | /* |
1951 | * given a cluster, try to allocate 'bytes' from it, returns 0 | |
1952 | * if it couldn't find anything suitably large, or a logical disk offset | |
1953 | * if things worked out | |
1954 | */ | |
1955 | u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group, | |
1956 | struct btrfs_free_cluster *cluster, u64 bytes, | |
1957 | u64 min_start) | |
1958 | { | |
34d52cb6 | 1959 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
fa9c0d79 CM |
1960 | struct btrfs_free_space *entry = NULL; |
1961 | struct rb_node *node; | |
1962 | u64 ret = 0; | |
1963 | ||
1964 | spin_lock(&cluster->lock); | |
1965 | if (bytes > cluster->max_size) | |
1966 | goto out; | |
1967 | ||
1968 | if (cluster->block_group != block_group) | |
1969 | goto out; | |
1970 | ||
1971 | node = rb_first(&cluster->root); | |
1972 | if (!node) | |
1973 | goto out; | |
1974 | ||
1975 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
fa9c0d79 | 1976 | while(1) { |
4e69b598 JB |
1977 | if (entry->bytes < bytes || |
1978 | (!entry->bitmap && entry->offset < min_start)) { | |
fa9c0d79 CM |
1979 | struct rb_node *node; |
1980 | ||
1981 | node = rb_next(&entry->offset_index); | |
1982 | if (!node) | |
1983 | break; | |
1984 | entry = rb_entry(node, struct btrfs_free_space, | |
1985 | offset_index); | |
1986 | continue; | |
1987 | } | |
fa9c0d79 | 1988 | |
4e69b598 JB |
1989 | if (entry->bitmap) { |
1990 | ret = btrfs_alloc_from_bitmap(block_group, | |
1991 | cluster, entry, bytes, | |
1992 | min_start); | |
1993 | if (ret == 0) { | |
1994 | struct rb_node *node; | |
1995 | node = rb_next(&entry->offset_index); | |
1996 | if (!node) | |
1997 | break; | |
1998 | entry = rb_entry(node, struct btrfs_free_space, | |
1999 | offset_index); | |
2000 | continue; | |
2001 | } | |
2002 | } else { | |
2003 | ||
2004 | ret = entry->offset; | |
2005 | ||
2006 | entry->offset += bytes; | |
2007 | entry->bytes -= bytes; | |
2008 | } | |
fa9c0d79 | 2009 | |
5e71b5d5 | 2010 | if (entry->bytes == 0) |
fa9c0d79 | 2011 | rb_erase(&entry->offset_index, &cluster->root); |
fa9c0d79 CM |
2012 | break; |
2013 | } | |
2014 | out: | |
2015 | spin_unlock(&cluster->lock); | |
96303081 | 2016 | |
5e71b5d5 LZ |
2017 | if (!ret) |
2018 | return 0; | |
2019 | ||
34d52cb6 | 2020 | spin_lock(&ctl->tree_lock); |
5e71b5d5 | 2021 | |
34d52cb6 | 2022 | ctl->free_space -= bytes; |
5e71b5d5 | 2023 | if (entry->bytes == 0) { |
34d52cb6 | 2024 | ctl->free_extents--; |
4e69b598 JB |
2025 | if (entry->bitmap) { |
2026 | kfree(entry->bitmap); | |
34d52cb6 LZ |
2027 | ctl->total_bitmaps--; |
2028 | ctl->op->recalc_thresholds(ctl); | |
4e69b598 | 2029 | } |
dc89e982 | 2030 | kmem_cache_free(btrfs_free_space_cachep, entry); |
5e71b5d5 LZ |
2031 | } |
2032 | ||
34d52cb6 | 2033 | spin_unlock(&ctl->tree_lock); |
5e71b5d5 | 2034 | |
fa9c0d79 CM |
2035 | return ret; |
2036 | } | |
2037 | ||
96303081 JB |
2038 | static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group, |
2039 | struct btrfs_free_space *entry, | |
2040 | struct btrfs_free_cluster *cluster, | |
2041 | u64 offset, u64 bytes, u64 min_bytes) | |
2042 | { | |
34d52cb6 | 2043 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
96303081 JB |
2044 | unsigned long next_zero; |
2045 | unsigned long i; | |
2046 | unsigned long search_bits; | |
2047 | unsigned long total_bits; | |
2048 | unsigned long found_bits; | |
2049 | unsigned long start = 0; | |
2050 | unsigned long total_found = 0; | |
4e69b598 | 2051 | int ret; |
96303081 JB |
2052 | bool found = false; |
2053 | ||
2054 | i = offset_to_bit(entry->offset, block_group->sectorsize, | |
2055 | max_t(u64, offset, entry->offset)); | |
d0a365e8 JB |
2056 | search_bits = bytes_to_bits(bytes, block_group->sectorsize); |
2057 | total_bits = bytes_to_bits(min_bytes, block_group->sectorsize); | |
96303081 JB |
2058 | |
2059 | again: | |
2060 | found_bits = 0; | |
2061 | for (i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, i); | |
2062 | i < BITS_PER_BITMAP; | |
2063 | i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, i + 1)) { | |
2064 | next_zero = find_next_zero_bit(entry->bitmap, | |
2065 | BITS_PER_BITMAP, i); | |
2066 | if (next_zero - i >= search_bits) { | |
2067 | found_bits = next_zero - i; | |
2068 | break; | |
2069 | } | |
2070 | i = next_zero; | |
2071 | } | |
2072 | ||
2073 | if (!found_bits) | |
4e69b598 | 2074 | return -ENOSPC; |
96303081 JB |
2075 | |
2076 | if (!found) { | |
2077 | start = i; | |
2078 | found = true; | |
2079 | } | |
2080 | ||
2081 | total_found += found_bits; | |
2082 | ||
2083 | if (cluster->max_size < found_bits * block_group->sectorsize) | |
2084 | cluster->max_size = found_bits * block_group->sectorsize; | |
2085 | ||
2086 | if (total_found < total_bits) { | |
2087 | i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, next_zero); | |
2088 | if (i - start > total_bits * 2) { | |
2089 | total_found = 0; | |
2090 | cluster->max_size = 0; | |
2091 | found = false; | |
2092 | } | |
2093 | goto again; | |
2094 | } | |
2095 | ||
2096 | cluster->window_start = start * block_group->sectorsize + | |
2097 | entry->offset; | |
34d52cb6 | 2098 | rb_erase(&entry->offset_index, &ctl->free_space_offset); |
4e69b598 JB |
2099 | ret = tree_insert_offset(&cluster->root, entry->offset, |
2100 | &entry->offset_index, 1); | |
2101 | BUG_ON(ret); | |
96303081 JB |
2102 | |
2103 | return 0; | |
2104 | } | |
2105 | ||
4e69b598 JB |
2106 | /* |
2107 | * This searches the block group for just extents to fill the cluster with. | |
2108 | */ | |
2109 | static int setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group, | |
2110 | struct btrfs_free_cluster *cluster, | |
2111 | u64 offset, u64 bytes, u64 min_bytes) | |
2112 | { | |
34d52cb6 | 2113 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
4e69b598 JB |
2114 | struct btrfs_free_space *first = NULL; |
2115 | struct btrfs_free_space *entry = NULL; | |
2116 | struct btrfs_free_space *prev = NULL; | |
2117 | struct btrfs_free_space *last; | |
2118 | struct rb_node *node; | |
2119 | u64 window_start; | |
2120 | u64 window_free; | |
2121 | u64 max_extent; | |
2122 | u64 max_gap = 128 * 1024; | |
2123 | ||
34d52cb6 | 2124 | entry = tree_search_offset(ctl, offset, 0, 1); |
4e69b598 JB |
2125 | if (!entry) |
2126 | return -ENOSPC; | |
2127 | ||
2128 | /* | |
2129 | * We don't want bitmaps, so just move along until we find a normal | |
2130 | * extent entry. | |
2131 | */ | |
2132 | while (entry->bitmap) { | |
2133 | node = rb_next(&entry->offset_index); | |
2134 | if (!node) | |
2135 | return -ENOSPC; | |
2136 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
2137 | } | |
2138 | ||
2139 | window_start = entry->offset; | |
2140 | window_free = entry->bytes; | |
2141 | max_extent = entry->bytes; | |
2142 | first = entry; | |
2143 | last = entry; | |
2144 | prev = entry; | |
2145 | ||
2146 | while (window_free <= min_bytes) { | |
2147 | node = rb_next(&entry->offset_index); | |
2148 | if (!node) | |
2149 | return -ENOSPC; | |
2150 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
2151 | ||
2152 | if (entry->bitmap) | |
2153 | continue; | |
2154 | /* | |
2155 | * we haven't filled the empty size and the window is | |
2156 | * very large. reset and try again | |
2157 | */ | |
2158 | if (entry->offset - (prev->offset + prev->bytes) > max_gap || | |
2159 | entry->offset - window_start > (min_bytes * 2)) { | |
2160 | first = entry; | |
2161 | window_start = entry->offset; | |
2162 | window_free = entry->bytes; | |
2163 | last = entry; | |
2164 | max_extent = entry->bytes; | |
2165 | } else { | |
2166 | last = entry; | |
2167 | window_free += entry->bytes; | |
2168 | if (entry->bytes > max_extent) | |
2169 | max_extent = entry->bytes; | |
2170 | } | |
2171 | prev = entry; | |
2172 | } | |
2173 | ||
2174 | cluster->window_start = first->offset; | |
2175 | ||
2176 | node = &first->offset_index; | |
2177 | ||
2178 | /* | |
2179 | * now we've found our entries, pull them out of the free space | |
2180 | * cache and put them into the cluster rbtree | |
2181 | */ | |
2182 | do { | |
2183 | int ret; | |
2184 | ||
2185 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
2186 | node = rb_next(&entry->offset_index); | |
2187 | if (entry->bitmap) | |
2188 | continue; | |
2189 | ||
34d52cb6 | 2190 | rb_erase(&entry->offset_index, &ctl->free_space_offset); |
4e69b598 JB |
2191 | ret = tree_insert_offset(&cluster->root, entry->offset, |
2192 | &entry->offset_index, 0); | |
2193 | BUG_ON(ret); | |
2194 | } while (node && entry != last); | |
2195 | ||
2196 | cluster->max_size = max_extent; | |
2197 | ||
2198 | return 0; | |
2199 | } | |
2200 | ||
2201 | /* | |
2202 | * This specifically looks for bitmaps that may work in the cluster, we assume | |
2203 | * that we have already failed to find extents that will work. | |
2204 | */ | |
2205 | static int setup_cluster_bitmap(struct btrfs_block_group_cache *block_group, | |
2206 | struct btrfs_free_cluster *cluster, | |
2207 | u64 offset, u64 bytes, u64 min_bytes) | |
2208 | { | |
34d52cb6 | 2209 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
4e69b598 JB |
2210 | struct btrfs_free_space *entry; |
2211 | struct rb_node *node; | |
2212 | int ret = -ENOSPC; | |
2213 | ||
34d52cb6 | 2214 | if (ctl->total_bitmaps == 0) |
4e69b598 JB |
2215 | return -ENOSPC; |
2216 | ||
34d52cb6 | 2217 | entry = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), 0, 1); |
4e69b598 JB |
2218 | if (!entry) |
2219 | return -ENOSPC; | |
2220 | ||
2221 | node = &entry->offset_index; | |
2222 | do { | |
2223 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
2224 | node = rb_next(&entry->offset_index); | |
2225 | if (!entry->bitmap) | |
2226 | continue; | |
2227 | if (entry->bytes < min_bytes) | |
2228 | continue; | |
2229 | ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset, | |
2230 | bytes, min_bytes); | |
2231 | } while (ret && node); | |
2232 | ||
2233 | return ret; | |
2234 | } | |
2235 | ||
fa9c0d79 CM |
2236 | /* |
2237 | * here we try to find a cluster of blocks in a block group. The goal | |
2238 | * is to find at least bytes free and up to empty_size + bytes free. | |
2239 | * We might not find them all in one contiguous area. | |
2240 | * | |
2241 | * returns zero and sets up cluster if things worked out, otherwise | |
2242 | * it returns -enospc | |
2243 | */ | |
2244 | int btrfs_find_space_cluster(struct btrfs_trans_handle *trans, | |
451d7585 | 2245 | struct btrfs_root *root, |
fa9c0d79 CM |
2246 | struct btrfs_block_group_cache *block_group, |
2247 | struct btrfs_free_cluster *cluster, | |
2248 | u64 offset, u64 bytes, u64 empty_size) | |
2249 | { | |
34d52cb6 | 2250 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
fa9c0d79 | 2251 | u64 min_bytes; |
fa9c0d79 CM |
2252 | int ret; |
2253 | ||
2254 | /* for metadata, allow allocates with more holes */ | |
451d7585 CM |
2255 | if (btrfs_test_opt(root, SSD_SPREAD)) { |
2256 | min_bytes = bytes + empty_size; | |
2257 | } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) { | |
fa9c0d79 CM |
2258 | /* |
2259 | * we want to do larger allocations when we are | |
2260 | * flushing out the delayed refs, it helps prevent | |
2261 | * making more work as we go along. | |
2262 | */ | |
2263 | if (trans->transaction->delayed_refs.flushing) | |
2264 | min_bytes = max(bytes, (bytes + empty_size) >> 1); | |
2265 | else | |
2266 | min_bytes = max(bytes, (bytes + empty_size) >> 4); | |
2267 | } else | |
2268 | min_bytes = max(bytes, (bytes + empty_size) >> 2); | |
2269 | ||
34d52cb6 | 2270 | spin_lock(&ctl->tree_lock); |
7d0d2e8e JB |
2271 | |
2272 | /* | |
2273 | * If we know we don't have enough space to make a cluster don't even | |
2274 | * bother doing all the work to try and find one. | |
2275 | */ | |
34d52cb6 LZ |
2276 | if (ctl->free_space < min_bytes) { |
2277 | spin_unlock(&ctl->tree_lock); | |
7d0d2e8e JB |
2278 | return -ENOSPC; |
2279 | } | |
2280 | ||
fa9c0d79 CM |
2281 | spin_lock(&cluster->lock); |
2282 | ||
2283 | /* someone already found a cluster, hooray */ | |
2284 | if (cluster->block_group) { | |
2285 | ret = 0; | |
2286 | goto out; | |
2287 | } | |
fa9c0d79 | 2288 | |
4e69b598 JB |
2289 | ret = setup_cluster_no_bitmap(block_group, cluster, offset, bytes, |
2290 | min_bytes); | |
2291 | if (ret) | |
2292 | ret = setup_cluster_bitmap(block_group, cluster, offset, | |
2293 | bytes, min_bytes); | |
fa9c0d79 | 2294 | |
4e69b598 JB |
2295 | if (!ret) { |
2296 | atomic_inc(&block_group->count); | |
2297 | list_add_tail(&cluster->block_group_list, | |
2298 | &block_group->cluster_list); | |
2299 | cluster->block_group = block_group; | |
fa9c0d79 | 2300 | } |
fa9c0d79 CM |
2301 | out: |
2302 | spin_unlock(&cluster->lock); | |
34d52cb6 | 2303 | spin_unlock(&ctl->tree_lock); |
fa9c0d79 CM |
2304 | |
2305 | return ret; | |
2306 | } | |
2307 | ||
2308 | /* | |
2309 | * simple code to zero out a cluster | |
2310 | */ | |
2311 | void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster) | |
2312 | { | |
2313 | spin_lock_init(&cluster->lock); | |
2314 | spin_lock_init(&cluster->refill_lock); | |
6bef4d31 | 2315 | cluster->root = RB_ROOT; |
fa9c0d79 CM |
2316 | cluster->max_size = 0; |
2317 | INIT_LIST_HEAD(&cluster->block_group_list); | |
2318 | cluster->block_group = NULL; | |
2319 | } | |
2320 | ||
f7039b1d LD |
2321 | int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group, |
2322 | u64 *trimmed, u64 start, u64 end, u64 minlen) | |
2323 | { | |
34d52cb6 | 2324 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
f7039b1d LD |
2325 | struct btrfs_free_space *entry = NULL; |
2326 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
2327 | u64 bytes = 0; | |
2328 | u64 actually_trimmed; | |
2329 | int ret = 0; | |
2330 | ||
2331 | *trimmed = 0; | |
2332 | ||
2333 | while (start < end) { | |
34d52cb6 | 2334 | spin_lock(&ctl->tree_lock); |
f7039b1d | 2335 | |
34d52cb6 LZ |
2336 | if (ctl->free_space < minlen) { |
2337 | spin_unlock(&ctl->tree_lock); | |
f7039b1d LD |
2338 | break; |
2339 | } | |
2340 | ||
34d52cb6 | 2341 | entry = tree_search_offset(ctl, start, 0, 1); |
f7039b1d | 2342 | if (!entry) |
34d52cb6 LZ |
2343 | entry = tree_search_offset(ctl, |
2344 | offset_to_bitmap(ctl, start), | |
f7039b1d LD |
2345 | 1, 1); |
2346 | ||
2347 | if (!entry || entry->offset >= end) { | |
34d52cb6 | 2348 | spin_unlock(&ctl->tree_lock); |
f7039b1d LD |
2349 | break; |
2350 | } | |
2351 | ||
2352 | if (entry->bitmap) { | |
34d52cb6 | 2353 | ret = search_bitmap(ctl, entry, &start, &bytes); |
f7039b1d LD |
2354 | if (!ret) { |
2355 | if (start >= end) { | |
34d52cb6 | 2356 | spin_unlock(&ctl->tree_lock); |
f7039b1d LD |
2357 | break; |
2358 | } | |
2359 | bytes = min(bytes, end - start); | |
34d52cb6 | 2360 | bitmap_clear_bits(ctl, entry, start, bytes); |
f7039b1d | 2361 | if (entry->bytes == 0) |
34d52cb6 | 2362 | free_bitmap(ctl, entry); |
f7039b1d LD |
2363 | } else { |
2364 | start = entry->offset + BITS_PER_BITMAP * | |
2365 | block_group->sectorsize; | |
34d52cb6 | 2366 | spin_unlock(&ctl->tree_lock); |
f7039b1d LD |
2367 | ret = 0; |
2368 | continue; | |
2369 | } | |
2370 | } else { | |
2371 | start = entry->offset; | |
2372 | bytes = min(entry->bytes, end - start); | |
34d52cb6 | 2373 | unlink_free_space(ctl, entry); |
f7039b1d LD |
2374 | kfree(entry); |
2375 | } | |
2376 | ||
34d52cb6 | 2377 | spin_unlock(&ctl->tree_lock); |
f7039b1d LD |
2378 | |
2379 | if (bytes >= minlen) { | |
2380 | int update_ret; | |
2381 | update_ret = btrfs_update_reserved_bytes(block_group, | |
2382 | bytes, 1, 1); | |
2383 | ||
2384 | ret = btrfs_error_discard_extent(fs_info->extent_root, | |
2385 | start, | |
2386 | bytes, | |
2387 | &actually_trimmed); | |
2388 | ||
34d52cb6 | 2389 | btrfs_add_free_space(block_group, start, bytes); |
f7039b1d LD |
2390 | if (!update_ret) |
2391 | btrfs_update_reserved_bytes(block_group, | |
2392 | bytes, 0, 1); | |
2393 | ||
2394 | if (ret) | |
2395 | break; | |
2396 | *trimmed += actually_trimmed; | |
2397 | } | |
2398 | start += bytes; | |
2399 | bytes = 0; | |
2400 | ||
2401 | if (fatal_signal_pending(current)) { | |
2402 | ret = -ERESTARTSYS; | |
2403 | break; | |
2404 | } | |
2405 | ||
2406 | cond_resched(); | |
2407 | } | |
2408 | ||
2409 | return ret; | |
2410 | } | |
581bb050 LZ |
2411 | |
2412 | /* | |
2413 | * Find the left-most item in the cache tree, and then return the | |
2414 | * smallest inode number in the item. | |
2415 | * | |
2416 | * Note: the returned inode number may not be the smallest one in | |
2417 | * the tree, if the left-most item is a bitmap. | |
2418 | */ | |
2419 | u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root) | |
2420 | { | |
2421 | struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl; | |
2422 | struct btrfs_free_space *entry = NULL; | |
2423 | u64 ino = 0; | |
2424 | ||
2425 | spin_lock(&ctl->tree_lock); | |
2426 | ||
2427 | if (RB_EMPTY_ROOT(&ctl->free_space_offset)) | |
2428 | goto out; | |
2429 | ||
2430 | entry = rb_entry(rb_first(&ctl->free_space_offset), | |
2431 | struct btrfs_free_space, offset_index); | |
2432 | ||
2433 | if (!entry->bitmap) { | |
2434 | ino = entry->offset; | |
2435 | ||
2436 | unlink_free_space(ctl, entry); | |
2437 | entry->offset++; | |
2438 | entry->bytes--; | |
2439 | if (!entry->bytes) | |
2440 | kmem_cache_free(btrfs_free_space_cachep, entry); | |
2441 | else | |
2442 | link_free_space(ctl, entry); | |
2443 | } else { | |
2444 | u64 offset = 0; | |
2445 | u64 count = 1; | |
2446 | int ret; | |
2447 | ||
2448 | ret = search_bitmap(ctl, entry, &offset, &count); | |
2449 | BUG_ON(ret); | |
2450 | ||
2451 | ino = offset; | |
2452 | bitmap_clear_bits(ctl, entry, offset, 1); | |
2453 | if (entry->bytes == 0) | |
2454 | free_bitmap(ctl, entry); | |
2455 | } | |
2456 | out: | |
2457 | spin_unlock(&ctl->tree_lock); | |
2458 | ||
2459 | return ino; | |
2460 | } |