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