]>
Commit | Line | Data |
---|---|---|
0f9dd46c JB |
1 | /* |
2 | * Copyright (C) 2008 Red Hat. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
96303081 | 19 | #include <linux/pagemap.h> |
0f9dd46c | 20 | #include <linux/sched.h> |
5a0e3ad6 | 21 | #include <linux/slab.h> |
96303081 | 22 | #include <linux/math64.h> |
6ab60601 | 23 | #include <linux/ratelimit.h> |
0f9dd46c | 24 | #include "ctree.h" |
fa9c0d79 CM |
25 | #include "free-space-cache.h" |
26 | #include "transaction.h" | |
0af3d00b | 27 | #include "disk-io.h" |
43be2146 | 28 | #include "extent_io.h" |
581bb050 | 29 | #include "inode-map.h" |
fa9c0d79 | 30 | |
96303081 JB |
31 | #define BITS_PER_BITMAP (PAGE_CACHE_SIZE * 8) |
32 | #define MAX_CACHE_BYTES_PER_GIG (32 * 1024) | |
0f9dd46c | 33 | |
34d52cb6 | 34 | static int link_free_space(struct btrfs_free_space_ctl *ctl, |
0cb59c99 JB |
35 | struct btrfs_free_space *info); |
36 | ||
0414efae LZ |
37 | static struct inode *__lookup_free_space_inode(struct btrfs_root *root, |
38 | struct btrfs_path *path, | |
39 | u64 offset) | |
0af3d00b JB |
40 | { |
41 | struct btrfs_key key; | |
42 | struct btrfs_key location; | |
43 | struct btrfs_disk_key disk_key; | |
44 | struct btrfs_free_space_header *header; | |
45 | struct extent_buffer *leaf; | |
46 | struct inode *inode = NULL; | |
47 | int ret; | |
48 | ||
0af3d00b | 49 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; |
0414efae | 50 | key.offset = offset; |
0af3d00b JB |
51 | key.type = 0; |
52 | ||
53 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
54 | if (ret < 0) | |
55 | return ERR_PTR(ret); | |
56 | if (ret > 0) { | |
b3b4aa74 | 57 | btrfs_release_path(path); |
0af3d00b JB |
58 | return ERR_PTR(-ENOENT); |
59 | } | |
60 | ||
61 | leaf = path->nodes[0]; | |
62 | header = btrfs_item_ptr(leaf, path->slots[0], | |
63 | struct btrfs_free_space_header); | |
64 | btrfs_free_space_key(leaf, header, &disk_key); | |
65 | btrfs_disk_key_to_cpu(&location, &disk_key); | |
b3b4aa74 | 66 | btrfs_release_path(path); |
0af3d00b JB |
67 | |
68 | inode = btrfs_iget(root->fs_info->sb, &location, root, NULL); | |
69 | if (!inode) | |
70 | return ERR_PTR(-ENOENT); | |
71 | if (IS_ERR(inode)) | |
72 | return inode; | |
73 | if (is_bad_inode(inode)) { | |
74 | iput(inode); | |
75 | return ERR_PTR(-ENOENT); | |
76 | } | |
77 | ||
adae52b9 MX |
78 | inode->i_mapping->flags &= ~__GFP_FS; |
79 | ||
0414efae LZ |
80 | return inode; |
81 | } | |
82 | ||
83 | struct inode *lookup_free_space_inode(struct btrfs_root *root, | |
84 | struct btrfs_block_group_cache | |
85 | *block_group, struct btrfs_path *path) | |
86 | { | |
87 | struct inode *inode = NULL; | |
5b0e95bf | 88 | u32 flags = BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW; |
0414efae LZ |
89 | |
90 | spin_lock(&block_group->lock); | |
91 | if (block_group->inode) | |
92 | inode = igrab(block_group->inode); | |
93 | spin_unlock(&block_group->lock); | |
94 | if (inode) | |
95 | return inode; | |
96 | ||
97 | inode = __lookup_free_space_inode(root, path, | |
98 | block_group->key.objectid); | |
99 | if (IS_ERR(inode)) | |
100 | return inode; | |
101 | ||
0af3d00b | 102 | spin_lock(&block_group->lock); |
5b0e95bf | 103 | if (!((BTRFS_I(inode)->flags & flags) == flags)) { |
2f356126 | 104 | printk(KERN_INFO "Old style space inode found, converting.\n"); |
5b0e95bf JB |
105 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM | |
106 | BTRFS_INODE_NODATACOW; | |
2f356126 JB |
107 | block_group->disk_cache_state = BTRFS_DC_CLEAR; |
108 | } | |
109 | ||
300e4f8a | 110 | if (!block_group->iref) { |
0af3d00b JB |
111 | block_group->inode = igrab(inode); |
112 | block_group->iref = 1; | |
113 | } | |
114 | spin_unlock(&block_group->lock); | |
115 | ||
116 | return inode; | |
117 | } | |
118 | ||
0414efae LZ |
119 | int __create_free_space_inode(struct btrfs_root *root, |
120 | struct btrfs_trans_handle *trans, | |
121 | struct btrfs_path *path, u64 ino, u64 offset) | |
0af3d00b JB |
122 | { |
123 | struct btrfs_key key; | |
124 | struct btrfs_disk_key disk_key; | |
125 | struct btrfs_free_space_header *header; | |
126 | struct btrfs_inode_item *inode_item; | |
127 | struct extent_buffer *leaf; | |
5b0e95bf | 128 | u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC; |
0af3d00b JB |
129 | int ret; |
130 | ||
0414efae | 131 | ret = btrfs_insert_empty_inode(trans, root, path, ino); |
0af3d00b JB |
132 | if (ret) |
133 | return ret; | |
134 | ||
5b0e95bf JB |
135 | /* We inline crc's for the free disk space cache */ |
136 | if (ino != BTRFS_FREE_INO_OBJECTID) | |
137 | flags |= BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW; | |
138 | ||
0af3d00b JB |
139 | leaf = path->nodes[0]; |
140 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
141 | struct btrfs_inode_item); | |
142 | btrfs_item_key(leaf, &disk_key, path->slots[0]); | |
143 | memset_extent_buffer(leaf, 0, (unsigned long)inode_item, | |
144 | sizeof(*inode_item)); | |
145 | btrfs_set_inode_generation(leaf, inode_item, trans->transid); | |
146 | btrfs_set_inode_size(leaf, inode_item, 0); | |
147 | btrfs_set_inode_nbytes(leaf, inode_item, 0); | |
148 | btrfs_set_inode_uid(leaf, inode_item, 0); | |
149 | btrfs_set_inode_gid(leaf, inode_item, 0); | |
150 | btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600); | |
5b0e95bf | 151 | btrfs_set_inode_flags(leaf, inode_item, flags); |
0af3d00b JB |
152 | btrfs_set_inode_nlink(leaf, inode_item, 1); |
153 | btrfs_set_inode_transid(leaf, inode_item, trans->transid); | |
0414efae | 154 | btrfs_set_inode_block_group(leaf, inode_item, offset); |
0af3d00b | 155 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 156 | btrfs_release_path(path); |
0af3d00b JB |
157 | |
158 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
0414efae | 159 | key.offset = offset; |
0af3d00b JB |
160 | key.type = 0; |
161 | ||
162 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
163 | sizeof(struct btrfs_free_space_header)); | |
164 | if (ret < 0) { | |
b3b4aa74 | 165 | btrfs_release_path(path); |
0af3d00b JB |
166 | return ret; |
167 | } | |
168 | leaf = path->nodes[0]; | |
169 | header = btrfs_item_ptr(leaf, path->slots[0], | |
170 | struct btrfs_free_space_header); | |
171 | memset_extent_buffer(leaf, 0, (unsigned long)header, sizeof(*header)); | |
172 | btrfs_set_free_space_key(leaf, header, &disk_key); | |
173 | btrfs_mark_buffer_dirty(leaf); | |
b3b4aa74 | 174 | btrfs_release_path(path); |
0af3d00b JB |
175 | |
176 | return 0; | |
177 | } | |
178 | ||
0414efae LZ |
179 | int create_free_space_inode(struct btrfs_root *root, |
180 | struct btrfs_trans_handle *trans, | |
181 | struct btrfs_block_group_cache *block_group, | |
182 | struct btrfs_path *path) | |
183 | { | |
184 | int ret; | |
185 | u64 ino; | |
186 | ||
187 | ret = btrfs_find_free_objectid(root, &ino); | |
188 | if (ret < 0) | |
189 | return ret; | |
190 | ||
191 | return __create_free_space_inode(root, trans, path, ino, | |
192 | block_group->key.objectid); | |
193 | } | |
194 | ||
0af3d00b JB |
195 | int btrfs_truncate_free_space_cache(struct btrfs_root *root, |
196 | struct btrfs_trans_handle *trans, | |
197 | struct btrfs_path *path, | |
198 | struct inode *inode) | |
199 | { | |
65450aa6 | 200 | struct btrfs_block_rsv *rsv; |
0af3d00b JB |
201 | loff_t oldsize; |
202 | int ret = 0; | |
203 | ||
65450aa6 | 204 | rsv = trans->block_rsv; |
0af3d00b | 205 | trans->block_rsv = root->orphan_block_rsv; |
36ba022a | 206 | ret = btrfs_block_rsv_check(root, root->orphan_block_rsv, 5); |
0af3d00b JB |
207 | if (ret) |
208 | return ret; | |
209 | ||
210 | oldsize = i_size_read(inode); | |
211 | btrfs_i_size_write(inode, 0); | |
212 | truncate_pagecache(inode, oldsize, 0); | |
213 | ||
214 | /* | |
215 | * We don't need an orphan item because truncating the free space cache | |
216 | * will never be split across transactions. | |
217 | */ | |
218 | ret = btrfs_truncate_inode_items(trans, root, inode, | |
219 | 0, BTRFS_EXTENT_DATA_KEY); | |
65450aa6 LB |
220 | |
221 | trans->block_rsv = rsv; | |
0af3d00b JB |
222 | if (ret) { |
223 | WARN_ON(1); | |
224 | return ret; | |
225 | } | |
226 | ||
82d5902d LZ |
227 | ret = btrfs_update_inode(trans, root, inode); |
228 | return ret; | |
0af3d00b JB |
229 | } |
230 | ||
9d66e233 JB |
231 | static int readahead_cache(struct inode *inode) |
232 | { | |
233 | struct file_ra_state *ra; | |
234 | unsigned long last_index; | |
235 | ||
236 | ra = kzalloc(sizeof(*ra), GFP_NOFS); | |
237 | if (!ra) | |
238 | return -ENOMEM; | |
239 | ||
240 | file_ra_state_init(ra, inode->i_mapping); | |
241 | last_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT; | |
242 | ||
243 | page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index); | |
244 | ||
245 | kfree(ra); | |
246 | ||
247 | return 0; | |
248 | } | |
249 | ||
a67509c3 JB |
250 | struct io_ctl { |
251 | void *cur, *orig; | |
252 | struct page *page; | |
253 | struct page **pages; | |
254 | struct btrfs_root *root; | |
255 | unsigned long size; | |
256 | int index; | |
257 | int num_pages; | |
5b0e95bf | 258 | unsigned check_crcs:1; |
a67509c3 JB |
259 | }; |
260 | ||
261 | static int io_ctl_init(struct io_ctl *io_ctl, struct inode *inode, | |
262 | struct btrfs_root *root) | |
263 | { | |
264 | memset(io_ctl, 0, sizeof(struct io_ctl)); | |
265 | io_ctl->num_pages = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> | |
266 | PAGE_CACHE_SHIFT; | |
267 | io_ctl->pages = kzalloc(sizeof(struct page *) * io_ctl->num_pages, | |
268 | GFP_NOFS); | |
269 | if (!io_ctl->pages) | |
270 | return -ENOMEM; | |
271 | io_ctl->root = root; | |
5b0e95bf JB |
272 | if (btrfs_ino(inode) != BTRFS_FREE_INO_OBJECTID) |
273 | io_ctl->check_crcs = 1; | |
a67509c3 JB |
274 | return 0; |
275 | } | |
276 | ||
277 | static void io_ctl_free(struct io_ctl *io_ctl) | |
278 | { | |
279 | kfree(io_ctl->pages); | |
280 | } | |
281 | ||
282 | static void io_ctl_unmap_page(struct io_ctl *io_ctl) | |
283 | { | |
284 | if (io_ctl->cur) { | |
285 | kunmap(io_ctl->page); | |
286 | io_ctl->cur = NULL; | |
287 | io_ctl->orig = NULL; | |
288 | } | |
289 | } | |
290 | ||
291 | static void io_ctl_map_page(struct io_ctl *io_ctl, int clear) | |
292 | { | |
293 | WARN_ON(io_ctl->cur); | |
294 | BUG_ON(io_ctl->index >= io_ctl->num_pages); | |
295 | io_ctl->page = io_ctl->pages[io_ctl->index++]; | |
296 | io_ctl->cur = kmap(io_ctl->page); | |
297 | io_ctl->orig = io_ctl->cur; | |
298 | io_ctl->size = PAGE_CACHE_SIZE; | |
299 | if (clear) | |
300 | memset(io_ctl->cur, 0, PAGE_CACHE_SIZE); | |
301 | } | |
302 | ||
303 | static void io_ctl_drop_pages(struct io_ctl *io_ctl) | |
304 | { | |
305 | int i; | |
306 | ||
307 | io_ctl_unmap_page(io_ctl); | |
308 | ||
309 | for (i = 0; i < io_ctl->num_pages; i++) { | |
310 | ClearPageChecked(io_ctl->pages[i]); | |
311 | unlock_page(io_ctl->pages[i]); | |
312 | page_cache_release(io_ctl->pages[i]); | |
313 | } | |
314 | } | |
315 | ||
316 | static int io_ctl_prepare_pages(struct io_ctl *io_ctl, struct inode *inode, | |
317 | int uptodate) | |
318 | { | |
319 | struct page *page; | |
320 | gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); | |
321 | int i; | |
322 | ||
323 | for (i = 0; i < io_ctl->num_pages; i++) { | |
324 | page = find_or_create_page(inode->i_mapping, i, mask); | |
325 | if (!page) { | |
326 | io_ctl_drop_pages(io_ctl); | |
327 | return -ENOMEM; | |
328 | } | |
329 | io_ctl->pages[i] = page; | |
330 | if (uptodate && !PageUptodate(page)) { | |
331 | btrfs_readpage(NULL, page); | |
332 | lock_page(page); | |
333 | if (!PageUptodate(page)) { | |
334 | printk(KERN_ERR "btrfs: error reading free " | |
335 | "space cache\n"); | |
336 | io_ctl_drop_pages(io_ctl); | |
337 | return -EIO; | |
338 | } | |
339 | } | |
340 | } | |
341 | ||
342 | return 0; | |
343 | } | |
344 | ||
345 | static void io_ctl_set_generation(struct io_ctl *io_ctl, u64 generation) | |
346 | { | |
347 | u64 *val; | |
348 | ||
349 | io_ctl_map_page(io_ctl, 1); | |
350 | ||
351 | /* | |
5b0e95bf JB |
352 | * Skip the csum areas. If we don't check crcs then we just have a |
353 | * 64bit chunk at the front of the first page. | |
a67509c3 | 354 | */ |
5b0e95bf JB |
355 | if (io_ctl->check_crcs) { |
356 | io_ctl->cur += (sizeof(u32) * io_ctl->num_pages); | |
357 | io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages); | |
358 | } else { | |
359 | io_ctl->cur += sizeof(u64); | |
360 | io_ctl->size -= sizeof(u64) * 2; | |
361 | } | |
a67509c3 JB |
362 | |
363 | val = io_ctl->cur; | |
364 | *val = cpu_to_le64(generation); | |
365 | io_ctl->cur += sizeof(u64); | |
a67509c3 JB |
366 | } |
367 | ||
368 | static int io_ctl_check_generation(struct io_ctl *io_ctl, u64 generation) | |
369 | { | |
370 | u64 *gen; | |
371 | ||
5b0e95bf JB |
372 | /* |
373 | * Skip the crc area. If we don't check crcs then we just have a 64bit | |
374 | * chunk at the front of the first page. | |
375 | */ | |
376 | if (io_ctl->check_crcs) { | |
377 | io_ctl->cur += sizeof(u32) * io_ctl->num_pages; | |
378 | io_ctl->size -= sizeof(u64) + | |
379 | (sizeof(u32) * io_ctl->num_pages); | |
380 | } else { | |
381 | io_ctl->cur += sizeof(u64); | |
382 | io_ctl->size -= sizeof(u64) * 2; | |
383 | } | |
a67509c3 | 384 | |
a67509c3 JB |
385 | gen = io_ctl->cur; |
386 | if (le64_to_cpu(*gen) != generation) { | |
387 | printk_ratelimited(KERN_ERR "btrfs: space cache generation " | |
388 | "(%Lu) does not match inode (%Lu)\n", *gen, | |
389 | generation); | |
390 | io_ctl_unmap_page(io_ctl); | |
391 | return -EIO; | |
392 | } | |
393 | io_ctl->cur += sizeof(u64); | |
5b0e95bf JB |
394 | return 0; |
395 | } | |
396 | ||
397 | static void io_ctl_set_crc(struct io_ctl *io_ctl, int index) | |
398 | { | |
399 | u32 *tmp; | |
400 | u32 crc = ~(u32)0; | |
401 | unsigned offset = 0; | |
402 | ||
403 | if (!io_ctl->check_crcs) { | |
404 | io_ctl_unmap_page(io_ctl); | |
405 | return; | |
406 | } | |
407 | ||
408 | if (index == 0) | |
409 | offset = sizeof(u32) * io_ctl->num_pages;; | |
410 | ||
411 | crc = btrfs_csum_data(io_ctl->root, io_ctl->orig + offset, crc, | |
412 | PAGE_CACHE_SIZE - offset); | |
413 | btrfs_csum_final(crc, (char *)&crc); | |
414 | io_ctl_unmap_page(io_ctl); | |
415 | tmp = kmap(io_ctl->pages[0]); | |
416 | tmp += index; | |
417 | *tmp = crc; | |
418 | kunmap(io_ctl->pages[0]); | |
419 | } | |
420 | ||
421 | static int io_ctl_check_crc(struct io_ctl *io_ctl, int index) | |
422 | { | |
423 | u32 *tmp, val; | |
424 | u32 crc = ~(u32)0; | |
425 | unsigned offset = 0; | |
426 | ||
427 | if (!io_ctl->check_crcs) { | |
428 | io_ctl_map_page(io_ctl, 0); | |
429 | return 0; | |
430 | } | |
431 | ||
432 | if (index == 0) | |
433 | offset = sizeof(u32) * io_ctl->num_pages; | |
434 | ||
435 | tmp = kmap(io_ctl->pages[0]); | |
436 | tmp += index; | |
437 | val = *tmp; | |
438 | kunmap(io_ctl->pages[0]); | |
439 | ||
440 | io_ctl_map_page(io_ctl, 0); | |
441 | crc = btrfs_csum_data(io_ctl->root, io_ctl->orig + offset, crc, | |
442 | PAGE_CACHE_SIZE - offset); | |
443 | btrfs_csum_final(crc, (char *)&crc); | |
444 | if (val != crc) { | |
445 | printk_ratelimited(KERN_ERR "btrfs: csum mismatch on free " | |
446 | "space cache\n"); | |
447 | io_ctl_unmap_page(io_ctl); | |
448 | return -EIO; | |
449 | } | |
450 | ||
a67509c3 JB |
451 | return 0; |
452 | } | |
453 | ||
454 | static int io_ctl_add_entry(struct io_ctl *io_ctl, u64 offset, u64 bytes, | |
455 | void *bitmap) | |
456 | { | |
457 | struct btrfs_free_space_entry *entry; | |
458 | ||
459 | if (!io_ctl->cur) | |
460 | return -ENOSPC; | |
461 | ||
462 | entry = io_ctl->cur; | |
463 | entry->offset = cpu_to_le64(offset); | |
464 | entry->bytes = cpu_to_le64(bytes); | |
465 | entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP : | |
466 | BTRFS_FREE_SPACE_EXTENT; | |
467 | io_ctl->cur += sizeof(struct btrfs_free_space_entry); | |
468 | io_ctl->size -= sizeof(struct btrfs_free_space_entry); | |
469 | ||
470 | if (io_ctl->size >= sizeof(struct btrfs_free_space_entry)) | |
471 | return 0; | |
472 | ||
5b0e95bf | 473 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); |
a67509c3 JB |
474 | |
475 | /* No more pages to map */ | |
476 | if (io_ctl->index >= io_ctl->num_pages) | |
477 | return 0; | |
478 | ||
479 | /* map the next page */ | |
480 | io_ctl_map_page(io_ctl, 1); | |
481 | return 0; | |
482 | } | |
483 | ||
484 | static int io_ctl_add_bitmap(struct io_ctl *io_ctl, void *bitmap) | |
485 | { | |
486 | if (!io_ctl->cur) | |
487 | return -ENOSPC; | |
488 | ||
489 | /* | |
490 | * If we aren't at the start of the current page, unmap this one and | |
491 | * map the next one if there is any left. | |
492 | */ | |
493 | if (io_ctl->cur != io_ctl->orig) { | |
5b0e95bf | 494 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); |
a67509c3 JB |
495 | if (io_ctl->index >= io_ctl->num_pages) |
496 | return -ENOSPC; | |
497 | io_ctl_map_page(io_ctl, 0); | |
498 | } | |
499 | ||
500 | memcpy(io_ctl->cur, bitmap, PAGE_CACHE_SIZE); | |
5b0e95bf | 501 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); |
a67509c3 JB |
502 | if (io_ctl->index < io_ctl->num_pages) |
503 | io_ctl_map_page(io_ctl, 0); | |
504 | return 0; | |
505 | } | |
506 | ||
507 | static void io_ctl_zero_remaining_pages(struct io_ctl *io_ctl) | |
508 | { | |
5b0e95bf JB |
509 | /* |
510 | * If we're not on the boundary we know we've modified the page and we | |
511 | * need to crc the page. | |
512 | */ | |
513 | if (io_ctl->cur != io_ctl->orig) | |
514 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); | |
515 | else | |
516 | io_ctl_unmap_page(io_ctl); | |
a67509c3 JB |
517 | |
518 | while (io_ctl->index < io_ctl->num_pages) { | |
519 | io_ctl_map_page(io_ctl, 1); | |
5b0e95bf | 520 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); |
a67509c3 JB |
521 | } |
522 | } | |
523 | ||
5b0e95bf JB |
524 | static int io_ctl_read_entry(struct io_ctl *io_ctl, |
525 | struct btrfs_free_space *entry, u8 *type) | |
a67509c3 JB |
526 | { |
527 | struct btrfs_free_space_entry *e; | |
a67509c3 JB |
528 | |
529 | e = io_ctl->cur; | |
530 | entry->offset = le64_to_cpu(e->offset); | |
531 | entry->bytes = le64_to_cpu(e->bytes); | |
5b0e95bf | 532 | *type = e->type; |
a67509c3 JB |
533 | io_ctl->cur += sizeof(struct btrfs_free_space_entry); |
534 | io_ctl->size -= sizeof(struct btrfs_free_space_entry); | |
535 | ||
536 | if (io_ctl->size >= sizeof(struct btrfs_free_space_entry)) | |
5b0e95bf | 537 | return 0; |
a67509c3 JB |
538 | |
539 | io_ctl_unmap_page(io_ctl); | |
540 | ||
541 | if (io_ctl->index >= io_ctl->num_pages) | |
5b0e95bf | 542 | return 0; |
a67509c3 | 543 | |
5b0e95bf | 544 | return io_ctl_check_crc(io_ctl, io_ctl->index); |
a67509c3 JB |
545 | } |
546 | ||
5b0e95bf JB |
547 | static int io_ctl_read_bitmap(struct io_ctl *io_ctl, |
548 | struct btrfs_free_space *entry) | |
a67509c3 | 549 | { |
5b0e95bf JB |
550 | int ret; |
551 | ||
552 | if (io_ctl->cur && io_ctl->cur != io_ctl->orig) | |
a67509c3 | 553 | io_ctl_unmap_page(io_ctl); |
5b0e95bf JB |
554 | |
555 | ret = io_ctl_check_crc(io_ctl, io_ctl->index); | |
556 | if (ret) | |
557 | return ret; | |
558 | ||
a67509c3 JB |
559 | memcpy(entry->bitmap, io_ctl->cur, PAGE_CACHE_SIZE); |
560 | io_ctl_unmap_page(io_ctl); | |
5b0e95bf JB |
561 | |
562 | return 0; | |
a67509c3 JB |
563 | } |
564 | ||
0414efae LZ |
565 | int __load_free_space_cache(struct btrfs_root *root, struct inode *inode, |
566 | struct btrfs_free_space_ctl *ctl, | |
567 | struct btrfs_path *path, u64 offset) | |
9d66e233 | 568 | { |
9d66e233 JB |
569 | struct btrfs_free_space_header *header; |
570 | struct extent_buffer *leaf; | |
a67509c3 | 571 | struct io_ctl io_ctl; |
9d66e233 | 572 | struct btrfs_key key; |
a67509c3 | 573 | struct btrfs_free_space *e, *n; |
9d66e233 JB |
574 | struct list_head bitmaps; |
575 | u64 num_entries; | |
576 | u64 num_bitmaps; | |
577 | u64 generation; | |
a67509c3 | 578 | u8 type; |
f6a39829 | 579 | int ret = 0; |
9d66e233 JB |
580 | |
581 | INIT_LIST_HEAD(&bitmaps); | |
582 | ||
9d66e233 | 583 | /* Nothing in the space cache, goodbye */ |
0414efae | 584 | if (!i_size_read(inode)) |
a67509c3 | 585 | return 0; |
9d66e233 JB |
586 | |
587 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
0414efae | 588 | key.offset = offset; |
9d66e233 JB |
589 | key.type = 0; |
590 | ||
591 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
0414efae | 592 | if (ret < 0) |
a67509c3 | 593 | return 0; |
0414efae | 594 | else if (ret > 0) { |
945d8962 | 595 | btrfs_release_path(path); |
a67509c3 | 596 | return 0; |
9d66e233 JB |
597 | } |
598 | ||
0414efae LZ |
599 | ret = -1; |
600 | ||
9d66e233 JB |
601 | leaf = path->nodes[0]; |
602 | header = btrfs_item_ptr(leaf, path->slots[0], | |
603 | struct btrfs_free_space_header); | |
604 | num_entries = btrfs_free_space_entries(leaf, header); | |
605 | num_bitmaps = btrfs_free_space_bitmaps(leaf, header); | |
606 | generation = btrfs_free_space_generation(leaf, header); | |
945d8962 | 607 | btrfs_release_path(path); |
9d66e233 JB |
608 | |
609 | if (BTRFS_I(inode)->generation != generation) { | |
610 | printk(KERN_ERR "btrfs: free space inode generation (%llu) did" | |
0414efae | 611 | " not match free space cache generation (%llu)\n", |
9d66e233 | 612 | (unsigned long long)BTRFS_I(inode)->generation, |
0414efae | 613 | (unsigned long long)generation); |
a67509c3 | 614 | return 0; |
9d66e233 JB |
615 | } |
616 | ||
617 | if (!num_entries) | |
a67509c3 | 618 | return 0; |
9d66e233 | 619 | |
a67509c3 | 620 | io_ctl_init(&io_ctl, inode, root); |
9d66e233 | 621 | ret = readahead_cache(inode); |
0414efae | 622 | if (ret) |
9d66e233 | 623 | goto out; |
9d66e233 | 624 | |
a67509c3 JB |
625 | ret = io_ctl_prepare_pages(&io_ctl, inode, 1); |
626 | if (ret) | |
627 | goto out; | |
9d66e233 | 628 | |
5b0e95bf JB |
629 | ret = io_ctl_check_crc(&io_ctl, 0); |
630 | if (ret) | |
631 | goto free_cache; | |
632 | ||
a67509c3 JB |
633 | ret = io_ctl_check_generation(&io_ctl, generation); |
634 | if (ret) | |
635 | goto free_cache; | |
9d66e233 | 636 | |
a67509c3 JB |
637 | while (num_entries) { |
638 | e = kmem_cache_zalloc(btrfs_free_space_cachep, | |
639 | GFP_NOFS); | |
640 | if (!e) | |
9d66e233 | 641 | goto free_cache; |
9d66e233 | 642 | |
5b0e95bf JB |
643 | ret = io_ctl_read_entry(&io_ctl, e, &type); |
644 | if (ret) { | |
645 | kmem_cache_free(btrfs_free_space_cachep, e); | |
646 | goto free_cache; | |
647 | } | |
648 | ||
a67509c3 JB |
649 | if (!e->bytes) { |
650 | kmem_cache_free(btrfs_free_space_cachep, e); | |
651 | goto free_cache; | |
9d66e233 | 652 | } |
a67509c3 JB |
653 | |
654 | if (type == BTRFS_FREE_SPACE_EXTENT) { | |
655 | spin_lock(&ctl->tree_lock); | |
656 | ret = link_free_space(ctl, e); | |
657 | spin_unlock(&ctl->tree_lock); | |
658 | if (ret) { | |
659 | printk(KERN_ERR "Duplicate entries in " | |
660 | "free space cache, dumping\n"); | |
661 | kmem_cache_free(btrfs_free_space_cachep, e); | |
9d66e233 JB |
662 | goto free_cache; |
663 | } | |
a67509c3 JB |
664 | } else { |
665 | BUG_ON(!num_bitmaps); | |
666 | num_bitmaps--; | |
667 | e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS); | |
668 | if (!e->bitmap) { | |
669 | kmem_cache_free( | |
670 | btrfs_free_space_cachep, e); | |
9d66e233 JB |
671 | goto free_cache; |
672 | } | |
a67509c3 JB |
673 | spin_lock(&ctl->tree_lock); |
674 | ret = link_free_space(ctl, e); | |
675 | ctl->total_bitmaps++; | |
676 | ctl->op->recalc_thresholds(ctl); | |
677 | spin_unlock(&ctl->tree_lock); | |
678 | if (ret) { | |
679 | printk(KERN_ERR "Duplicate entries in " | |
680 | "free space cache, dumping\n"); | |
dc89e982 | 681 | kmem_cache_free(btrfs_free_space_cachep, e); |
9d66e233 JB |
682 | goto free_cache; |
683 | } | |
a67509c3 | 684 | list_add_tail(&e->list, &bitmaps); |
9d66e233 JB |
685 | } |
686 | ||
a67509c3 JB |
687 | num_entries--; |
688 | } | |
9d66e233 | 689 | |
a67509c3 JB |
690 | /* |
691 | * We add the bitmaps at the end of the entries in order that | |
692 | * the bitmap entries are added to the cache. | |
693 | */ | |
694 | list_for_each_entry_safe(e, n, &bitmaps, list) { | |
9d66e233 | 695 | list_del_init(&e->list); |
5b0e95bf JB |
696 | ret = io_ctl_read_bitmap(&io_ctl, e); |
697 | if (ret) | |
698 | goto free_cache; | |
9d66e233 JB |
699 | } |
700 | ||
a67509c3 | 701 | io_ctl_drop_pages(&io_ctl); |
9d66e233 JB |
702 | ret = 1; |
703 | out: | |
a67509c3 | 704 | io_ctl_free(&io_ctl); |
9d66e233 | 705 | return ret; |
9d66e233 | 706 | free_cache: |
a67509c3 | 707 | io_ctl_drop_pages(&io_ctl); |
0414efae | 708 | __btrfs_remove_free_space_cache(ctl); |
9d66e233 JB |
709 | goto out; |
710 | } | |
711 | ||
0414efae LZ |
712 | int load_free_space_cache(struct btrfs_fs_info *fs_info, |
713 | struct btrfs_block_group_cache *block_group) | |
0cb59c99 | 714 | { |
34d52cb6 | 715 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0414efae LZ |
716 | struct btrfs_root *root = fs_info->tree_root; |
717 | struct inode *inode; | |
718 | struct btrfs_path *path; | |
5b0e95bf | 719 | int ret = 0; |
0414efae LZ |
720 | bool matched; |
721 | u64 used = btrfs_block_group_used(&block_group->item); | |
722 | ||
723 | /* | |
724 | * If we're unmounting then just return, since this does a search on the | |
725 | * normal root and not the commit root and we could deadlock. | |
726 | */ | |
7841cb28 | 727 | if (btrfs_fs_closing(fs_info)) |
0414efae LZ |
728 | return 0; |
729 | ||
730 | /* | |
731 | * If this block group has been marked to be cleared for one reason or | |
732 | * another then we can't trust the on disk cache, so just return. | |
733 | */ | |
9d66e233 | 734 | spin_lock(&block_group->lock); |
0414efae LZ |
735 | if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) { |
736 | spin_unlock(&block_group->lock); | |
737 | return 0; | |
738 | } | |
9d66e233 | 739 | spin_unlock(&block_group->lock); |
0414efae LZ |
740 | |
741 | path = btrfs_alloc_path(); | |
742 | if (!path) | |
743 | return 0; | |
744 | ||
745 | inode = lookup_free_space_inode(root, block_group, path); | |
746 | if (IS_ERR(inode)) { | |
747 | btrfs_free_path(path); | |
748 | return 0; | |
749 | } | |
750 | ||
5b0e95bf JB |
751 | /* We may have converted the inode and made the cache invalid. */ |
752 | spin_lock(&block_group->lock); | |
753 | if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) { | |
754 | spin_unlock(&block_group->lock); | |
755 | goto out; | |
756 | } | |
757 | spin_unlock(&block_group->lock); | |
758 | ||
0414efae LZ |
759 | ret = __load_free_space_cache(fs_info->tree_root, inode, ctl, |
760 | path, block_group->key.objectid); | |
761 | btrfs_free_path(path); | |
762 | if (ret <= 0) | |
763 | goto out; | |
764 | ||
765 | spin_lock(&ctl->tree_lock); | |
766 | matched = (ctl->free_space == (block_group->key.offset - used - | |
767 | block_group->bytes_super)); | |
768 | spin_unlock(&ctl->tree_lock); | |
769 | ||
770 | if (!matched) { | |
771 | __btrfs_remove_free_space_cache(ctl); | |
772 | printk(KERN_ERR "block group %llu has an wrong amount of free " | |
773 | "space\n", block_group->key.objectid); | |
774 | ret = -1; | |
775 | } | |
776 | out: | |
777 | if (ret < 0) { | |
778 | /* This cache is bogus, make sure it gets cleared */ | |
779 | spin_lock(&block_group->lock); | |
780 | block_group->disk_cache_state = BTRFS_DC_CLEAR; | |
781 | spin_unlock(&block_group->lock); | |
82d5902d | 782 | ret = 0; |
0414efae LZ |
783 | |
784 | printk(KERN_ERR "btrfs: failed to load free space cache " | |
785 | "for block group %llu\n", block_group->key.objectid); | |
786 | } | |
787 | ||
788 | iput(inode); | |
789 | return ret; | |
9d66e233 JB |
790 | } |
791 | ||
c09544e0 JB |
792 | /** |
793 | * __btrfs_write_out_cache - write out cached info to an inode | |
794 | * @root - the root the inode belongs to | |
795 | * @ctl - the free space cache we are going to write out | |
796 | * @block_group - the block_group for this cache if it belongs to a block_group | |
797 | * @trans - the trans handle | |
798 | * @path - the path to use | |
799 | * @offset - the offset for the key we'll insert | |
800 | * | |
801 | * This function writes out a free space cache struct to disk for quick recovery | |
802 | * on mount. This will return 0 if it was successfull in writing the cache out, | |
803 | * and -1 if it was not. | |
804 | */ | |
0414efae LZ |
805 | int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode, |
806 | struct btrfs_free_space_ctl *ctl, | |
807 | struct btrfs_block_group_cache *block_group, | |
808 | struct btrfs_trans_handle *trans, | |
809 | struct btrfs_path *path, u64 offset) | |
0cb59c99 JB |
810 | { |
811 | struct btrfs_free_space_header *header; | |
812 | struct extent_buffer *leaf; | |
0cb59c99 JB |
813 | struct rb_node *node; |
814 | struct list_head *pos, *n; | |
0cb59c99 | 815 | struct extent_state *cached_state = NULL; |
43be2146 JB |
816 | struct btrfs_free_cluster *cluster = NULL; |
817 | struct extent_io_tree *unpin = NULL; | |
a67509c3 | 818 | struct io_ctl io_ctl; |
0cb59c99 JB |
819 | struct list_head bitmap_list; |
820 | struct btrfs_key key; | |
43be2146 | 821 | u64 start, end, len; |
0cb59c99 JB |
822 | int entries = 0; |
823 | int bitmaps = 0; | |
c09544e0 JB |
824 | int ret; |
825 | int err = -1; | |
0cb59c99 | 826 | |
0cb59c99 JB |
827 | INIT_LIST_HEAD(&bitmap_list); |
828 | ||
0414efae LZ |
829 | if (!i_size_read(inode)) |
830 | return -1; | |
2b20982e | 831 | |
a67509c3 | 832 | io_ctl_init(&io_ctl, inode, root); |
be1a12a0 | 833 | |
43be2146 | 834 | /* Get the cluster for this block_group if it exists */ |
0414efae | 835 | if (block_group && !list_empty(&block_group->cluster_list)) |
43be2146 JB |
836 | cluster = list_entry(block_group->cluster_list.next, |
837 | struct btrfs_free_cluster, | |
838 | block_group_list); | |
839 | ||
840 | /* | |
841 | * We shouldn't have switched the pinned extents yet so this is the | |
842 | * right one | |
843 | */ | |
844 | unpin = root->fs_info->pinned_extents; | |
845 | ||
a67509c3 JB |
846 | /* Lock all pages first so we can lock the extent safely. */ |
847 | io_ctl_prepare_pages(&io_ctl, inode, 0); | |
0cb59c99 | 848 | |
0cb59c99 JB |
849 | lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1, |
850 | 0, &cached_state, GFP_NOFS); | |
851 | ||
43be2146 JB |
852 | /* |
853 | * When searching for pinned extents, we need to start at our start | |
854 | * offset. | |
855 | */ | |
0414efae LZ |
856 | if (block_group) |
857 | start = block_group->key.objectid; | |
43be2146 | 858 | |
f75b130e JB |
859 | node = rb_first(&ctl->free_space_offset); |
860 | if (!node && cluster) { | |
861 | node = rb_first(&cluster->root); | |
862 | cluster = NULL; | |
863 | } | |
864 | ||
5b0e95bf JB |
865 | /* Make sure we can fit our crcs into the first page */ |
866 | if (io_ctl.check_crcs && | |
867 | (io_ctl.num_pages * sizeof(u32)) >= PAGE_CACHE_SIZE) { | |
868 | WARN_ON(1); | |
869 | goto out_nospc; | |
870 | } | |
871 | ||
a67509c3 | 872 | io_ctl_set_generation(&io_ctl, trans->transid); |
43be2146 | 873 | |
a67509c3 JB |
874 | /* Write out the extent entries */ |
875 | while (node) { | |
876 | struct btrfs_free_space *e; | |
0cb59c99 | 877 | |
a67509c3 JB |
878 | e = rb_entry(node, struct btrfs_free_space, offset_index); |
879 | entries++; | |
0cb59c99 | 880 | |
a67509c3 JB |
881 | ret = io_ctl_add_entry(&io_ctl, e->offset, e->bytes, |
882 | e->bitmap); | |
883 | if (ret) | |
884 | goto out_nospc; | |
2f356126 | 885 | |
a67509c3 JB |
886 | if (e->bitmap) { |
887 | list_add_tail(&e->list, &bitmap_list); | |
888 | bitmaps++; | |
2f356126 | 889 | } |
a67509c3 JB |
890 | node = rb_next(node); |
891 | if (!node && cluster) { | |
892 | node = rb_first(&cluster->root); | |
893 | cluster = NULL; | |
43be2146 | 894 | } |
a67509c3 | 895 | } |
43be2146 | 896 | |
a67509c3 JB |
897 | /* |
898 | * We want to add any pinned extents to our free space cache | |
899 | * so we don't leak the space | |
900 | */ | |
901 | while (block_group && (start < block_group->key.objectid + | |
902 | block_group->key.offset)) { | |
903 | ret = find_first_extent_bit(unpin, start, &start, &end, | |
904 | EXTENT_DIRTY); | |
905 | if (ret) { | |
906 | ret = 0; | |
907 | break; | |
0cb59c99 | 908 | } |
0cb59c99 | 909 | |
a67509c3 JB |
910 | /* This pinned extent is out of our range */ |
911 | if (start >= block_group->key.objectid + | |
912 | block_group->key.offset) | |
913 | break; | |
2f356126 | 914 | |
a67509c3 JB |
915 | len = block_group->key.objectid + |
916 | block_group->key.offset - start; | |
917 | len = min(len, end + 1 - start); | |
0cb59c99 | 918 | |
a67509c3 JB |
919 | entries++; |
920 | ret = io_ctl_add_entry(&io_ctl, start, len, NULL); | |
921 | if (ret) | |
922 | goto out_nospc; | |
0cb59c99 | 923 | |
a67509c3 JB |
924 | start = end + 1; |
925 | } | |
0cb59c99 JB |
926 | |
927 | /* Write out the bitmaps */ | |
928 | list_for_each_safe(pos, n, &bitmap_list) { | |
0cb59c99 JB |
929 | struct btrfs_free_space *entry = |
930 | list_entry(pos, struct btrfs_free_space, list); | |
931 | ||
a67509c3 JB |
932 | ret = io_ctl_add_bitmap(&io_ctl, entry->bitmap); |
933 | if (ret) | |
934 | goto out_nospc; | |
0cb59c99 | 935 | list_del_init(&entry->list); |
be1a12a0 JB |
936 | } |
937 | ||
0cb59c99 | 938 | /* Zero out the rest of the pages just to make sure */ |
a67509c3 | 939 | io_ctl_zero_remaining_pages(&io_ctl); |
0cb59c99 | 940 | |
a67509c3 JB |
941 | ret = btrfs_dirty_pages(root, inode, io_ctl.pages, io_ctl.num_pages, |
942 | 0, i_size_read(inode), &cached_state); | |
943 | io_ctl_drop_pages(&io_ctl); | |
0cb59c99 JB |
944 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0, |
945 | i_size_read(inode) - 1, &cached_state, GFP_NOFS); | |
946 | ||
c09544e0 | 947 | if (ret) |
2f356126 | 948 | goto out; |
be1a12a0 | 949 | |
be1a12a0 | 950 | |
549b4fdb JB |
951 | ret = filemap_write_and_wait(inode->i_mapping); |
952 | if (ret) | |
953 | goto out; | |
0cb59c99 JB |
954 | |
955 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
0414efae | 956 | key.offset = offset; |
0cb59c99 JB |
957 | key.type = 0; |
958 | ||
a9b5fcdd | 959 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
0cb59c99 | 960 | if (ret < 0) { |
a67509c3 | 961 | clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1, |
5b0e95bf JB |
962 | EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL, |
963 | GFP_NOFS); | |
2f356126 | 964 | goto out; |
0cb59c99 JB |
965 | } |
966 | leaf = path->nodes[0]; | |
967 | if (ret > 0) { | |
968 | struct btrfs_key found_key; | |
969 | BUG_ON(!path->slots[0]); | |
970 | path->slots[0]--; | |
971 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
972 | if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID || | |
0414efae | 973 | found_key.offset != offset) { |
a67509c3 JB |
974 | clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, |
975 | inode->i_size - 1, | |
5b0e95bf JB |
976 | EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, |
977 | NULL, GFP_NOFS); | |
b3b4aa74 | 978 | btrfs_release_path(path); |
2f356126 | 979 | goto out; |
0cb59c99 JB |
980 | } |
981 | } | |
549b4fdb JB |
982 | |
983 | BTRFS_I(inode)->generation = trans->transid; | |
0cb59c99 JB |
984 | header = btrfs_item_ptr(leaf, path->slots[0], |
985 | struct btrfs_free_space_header); | |
986 | btrfs_set_free_space_entries(leaf, header, entries); | |
987 | btrfs_set_free_space_bitmaps(leaf, header, bitmaps); | |
988 | btrfs_set_free_space_generation(leaf, header, trans->transid); | |
989 | btrfs_mark_buffer_dirty(leaf); | |
b3b4aa74 | 990 | btrfs_release_path(path); |
0cb59c99 | 991 | |
c09544e0 | 992 | err = 0; |
2f356126 | 993 | out: |
a67509c3 | 994 | io_ctl_free(&io_ctl); |
c09544e0 | 995 | if (err) { |
a67509c3 | 996 | invalidate_inode_pages2(inode->i_mapping); |
0cb59c99 JB |
997 | BTRFS_I(inode)->generation = 0; |
998 | } | |
0cb59c99 | 999 | btrfs_update_inode(trans, root, inode); |
c09544e0 | 1000 | return err; |
a67509c3 JB |
1001 | |
1002 | out_nospc: | |
1003 | list_for_each_safe(pos, n, &bitmap_list) { | |
1004 | struct btrfs_free_space *entry = | |
1005 | list_entry(pos, struct btrfs_free_space, list); | |
1006 | list_del_init(&entry->list); | |
1007 | } | |
1008 | io_ctl_drop_pages(&io_ctl); | |
1009 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0, | |
1010 | i_size_read(inode) - 1, &cached_state, GFP_NOFS); | |
1011 | goto out; | |
0414efae LZ |
1012 | } |
1013 | ||
1014 | int btrfs_write_out_cache(struct btrfs_root *root, | |
1015 | struct btrfs_trans_handle *trans, | |
1016 | struct btrfs_block_group_cache *block_group, | |
1017 | struct btrfs_path *path) | |
1018 | { | |
1019 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
1020 | struct inode *inode; | |
1021 | int ret = 0; | |
1022 | ||
1023 | root = root->fs_info->tree_root; | |
1024 | ||
1025 | spin_lock(&block_group->lock); | |
1026 | if (block_group->disk_cache_state < BTRFS_DC_SETUP) { | |
1027 | spin_unlock(&block_group->lock); | |
1028 | return 0; | |
1029 | } | |
1030 | spin_unlock(&block_group->lock); | |
1031 | ||
1032 | inode = lookup_free_space_inode(root, block_group, path); | |
1033 | if (IS_ERR(inode)) | |
1034 | return 0; | |
1035 | ||
1036 | ret = __btrfs_write_out_cache(root, inode, ctl, block_group, trans, | |
1037 | path, block_group->key.objectid); | |
c09544e0 | 1038 | if (ret) { |
0414efae LZ |
1039 | spin_lock(&block_group->lock); |
1040 | block_group->disk_cache_state = BTRFS_DC_ERROR; | |
1041 | spin_unlock(&block_group->lock); | |
82d5902d | 1042 | ret = 0; |
c09544e0 | 1043 | #ifdef DEBUG |
0414efae LZ |
1044 | printk(KERN_ERR "btrfs: failed to write free space cace " |
1045 | "for block group %llu\n", block_group->key.objectid); | |
c09544e0 | 1046 | #endif |
0414efae LZ |
1047 | } |
1048 | ||
0cb59c99 JB |
1049 | iput(inode); |
1050 | return ret; | |
1051 | } | |
1052 | ||
34d52cb6 | 1053 | static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit, |
96303081 | 1054 | u64 offset) |
0f9dd46c | 1055 | { |
96303081 JB |
1056 | BUG_ON(offset < bitmap_start); |
1057 | offset -= bitmap_start; | |
34d52cb6 | 1058 | return (unsigned long)(div_u64(offset, unit)); |
96303081 | 1059 | } |
0f9dd46c | 1060 | |
34d52cb6 | 1061 | static inline unsigned long bytes_to_bits(u64 bytes, u32 unit) |
96303081 | 1062 | { |
34d52cb6 | 1063 | return (unsigned long)(div_u64(bytes, unit)); |
96303081 | 1064 | } |
0f9dd46c | 1065 | |
34d52cb6 | 1066 | static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1067 | u64 offset) |
1068 | { | |
1069 | u64 bitmap_start; | |
1070 | u64 bytes_per_bitmap; | |
0f9dd46c | 1071 | |
34d52cb6 LZ |
1072 | bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit; |
1073 | bitmap_start = offset - ctl->start; | |
96303081 JB |
1074 | bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap); |
1075 | bitmap_start *= bytes_per_bitmap; | |
34d52cb6 | 1076 | bitmap_start += ctl->start; |
0f9dd46c | 1077 | |
96303081 | 1078 | return bitmap_start; |
0f9dd46c JB |
1079 | } |
1080 | ||
96303081 JB |
1081 | static int tree_insert_offset(struct rb_root *root, u64 offset, |
1082 | struct rb_node *node, int bitmap) | |
0f9dd46c JB |
1083 | { |
1084 | struct rb_node **p = &root->rb_node; | |
1085 | struct rb_node *parent = NULL; | |
1086 | struct btrfs_free_space *info; | |
1087 | ||
1088 | while (*p) { | |
1089 | parent = *p; | |
96303081 | 1090 | info = rb_entry(parent, struct btrfs_free_space, offset_index); |
0f9dd46c | 1091 | |
96303081 | 1092 | if (offset < info->offset) { |
0f9dd46c | 1093 | p = &(*p)->rb_left; |
96303081 | 1094 | } else if (offset > info->offset) { |
0f9dd46c | 1095 | p = &(*p)->rb_right; |
96303081 JB |
1096 | } else { |
1097 | /* | |
1098 | * we could have a bitmap entry and an extent entry | |
1099 | * share the same offset. If this is the case, we want | |
1100 | * the extent entry to always be found first if we do a | |
1101 | * linear search through the tree, since we want to have | |
1102 | * the quickest allocation time, and allocating from an | |
1103 | * extent is faster than allocating from a bitmap. So | |
1104 | * if we're inserting a bitmap and we find an entry at | |
1105 | * this offset, we want to go right, or after this entry | |
1106 | * logically. If we are inserting an extent and we've | |
1107 | * found a bitmap, we want to go left, or before | |
1108 | * logically. | |
1109 | */ | |
1110 | if (bitmap) { | |
207dde82 JB |
1111 | if (info->bitmap) { |
1112 | WARN_ON_ONCE(1); | |
1113 | return -EEXIST; | |
1114 | } | |
96303081 JB |
1115 | p = &(*p)->rb_right; |
1116 | } else { | |
207dde82 JB |
1117 | if (!info->bitmap) { |
1118 | WARN_ON_ONCE(1); | |
1119 | return -EEXIST; | |
1120 | } | |
96303081 JB |
1121 | p = &(*p)->rb_left; |
1122 | } | |
1123 | } | |
0f9dd46c JB |
1124 | } |
1125 | ||
1126 | rb_link_node(node, parent, p); | |
1127 | rb_insert_color(node, root); | |
1128 | ||
1129 | return 0; | |
1130 | } | |
1131 | ||
1132 | /* | |
70cb0743 JB |
1133 | * searches the tree for the given offset. |
1134 | * | |
96303081 JB |
1135 | * fuzzy - If this is set, then we are trying to make an allocation, and we just |
1136 | * want a section that has at least bytes size and comes at or after the given | |
1137 | * offset. | |
0f9dd46c | 1138 | */ |
96303081 | 1139 | static struct btrfs_free_space * |
34d52cb6 | 1140 | tree_search_offset(struct btrfs_free_space_ctl *ctl, |
96303081 | 1141 | u64 offset, int bitmap_only, int fuzzy) |
0f9dd46c | 1142 | { |
34d52cb6 | 1143 | struct rb_node *n = ctl->free_space_offset.rb_node; |
96303081 JB |
1144 | struct btrfs_free_space *entry, *prev = NULL; |
1145 | ||
1146 | /* find entry that is closest to the 'offset' */ | |
1147 | while (1) { | |
1148 | if (!n) { | |
1149 | entry = NULL; | |
1150 | break; | |
1151 | } | |
0f9dd46c | 1152 | |
0f9dd46c | 1153 | entry = rb_entry(n, struct btrfs_free_space, offset_index); |
96303081 | 1154 | prev = entry; |
0f9dd46c | 1155 | |
96303081 | 1156 | if (offset < entry->offset) |
0f9dd46c | 1157 | n = n->rb_left; |
96303081 | 1158 | else if (offset > entry->offset) |
0f9dd46c | 1159 | n = n->rb_right; |
96303081 | 1160 | else |
0f9dd46c | 1161 | break; |
0f9dd46c JB |
1162 | } |
1163 | ||
96303081 JB |
1164 | if (bitmap_only) { |
1165 | if (!entry) | |
1166 | return NULL; | |
1167 | if (entry->bitmap) | |
1168 | return entry; | |
0f9dd46c | 1169 | |
96303081 JB |
1170 | /* |
1171 | * bitmap entry and extent entry may share same offset, | |
1172 | * in that case, bitmap entry comes after extent entry. | |
1173 | */ | |
1174 | n = rb_next(n); | |
1175 | if (!n) | |
1176 | return NULL; | |
1177 | entry = rb_entry(n, struct btrfs_free_space, offset_index); | |
1178 | if (entry->offset != offset) | |
1179 | return NULL; | |
0f9dd46c | 1180 | |
96303081 JB |
1181 | WARN_ON(!entry->bitmap); |
1182 | return entry; | |
1183 | } else if (entry) { | |
1184 | if (entry->bitmap) { | |
0f9dd46c | 1185 | /* |
96303081 JB |
1186 | * if previous extent entry covers the offset, |
1187 | * we should return it instead of the bitmap entry | |
0f9dd46c | 1188 | */ |
96303081 JB |
1189 | n = &entry->offset_index; |
1190 | while (1) { | |
1191 | n = rb_prev(n); | |
1192 | if (!n) | |
1193 | break; | |
1194 | prev = rb_entry(n, struct btrfs_free_space, | |
1195 | offset_index); | |
1196 | if (!prev->bitmap) { | |
1197 | if (prev->offset + prev->bytes > offset) | |
1198 | entry = prev; | |
1199 | break; | |
1200 | } | |
0f9dd46c | 1201 | } |
96303081 JB |
1202 | } |
1203 | return entry; | |
1204 | } | |
1205 | ||
1206 | if (!prev) | |
1207 | return NULL; | |
1208 | ||
1209 | /* find last entry before the 'offset' */ | |
1210 | entry = prev; | |
1211 | if (entry->offset > offset) { | |
1212 | n = rb_prev(&entry->offset_index); | |
1213 | if (n) { | |
1214 | entry = rb_entry(n, struct btrfs_free_space, | |
1215 | offset_index); | |
1216 | BUG_ON(entry->offset > offset); | |
0f9dd46c | 1217 | } else { |
96303081 JB |
1218 | if (fuzzy) |
1219 | return entry; | |
1220 | else | |
1221 | return NULL; | |
0f9dd46c JB |
1222 | } |
1223 | } | |
1224 | ||
96303081 JB |
1225 | if (entry->bitmap) { |
1226 | n = &entry->offset_index; | |
1227 | while (1) { | |
1228 | n = rb_prev(n); | |
1229 | if (!n) | |
1230 | break; | |
1231 | prev = rb_entry(n, struct btrfs_free_space, | |
1232 | offset_index); | |
1233 | if (!prev->bitmap) { | |
1234 | if (prev->offset + prev->bytes > offset) | |
1235 | return prev; | |
1236 | break; | |
1237 | } | |
1238 | } | |
34d52cb6 | 1239 | if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset) |
96303081 JB |
1240 | return entry; |
1241 | } else if (entry->offset + entry->bytes > offset) | |
1242 | return entry; | |
1243 | ||
1244 | if (!fuzzy) | |
1245 | return NULL; | |
1246 | ||
1247 | while (1) { | |
1248 | if (entry->bitmap) { | |
1249 | if (entry->offset + BITS_PER_BITMAP * | |
34d52cb6 | 1250 | ctl->unit > offset) |
96303081 JB |
1251 | break; |
1252 | } else { | |
1253 | if (entry->offset + entry->bytes > offset) | |
1254 | break; | |
1255 | } | |
1256 | ||
1257 | n = rb_next(&entry->offset_index); | |
1258 | if (!n) | |
1259 | return NULL; | |
1260 | entry = rb_entry(n, struct btrfs_free_space, offset_index); | |
1261 | } | |
1262 | return entry; | |
0f9dd46c JB |
1263 | } |
1264 | ||
f333adb5 | 1265 | static inline void |
34d52cb6 | 1266 | __unlink_free_space(struct btrfs_free_space_ctl *ctl, |
f333adb5 | 1267 | struct btrfs_free_space *info) |
0f9dd46c | 1268 | { |
34d52cb6 LZ |
1269 | rb_erase(&info->offset_index, &ctl->free_space_offset); |
1270 | ctl->free_extents--; | |
f333adb5 LZ |
1271 | } |
1272 | ||
34d52cb6 | 1273 | static void unlink_free_space(struct btrfs_free_space_ctl *ctl, |
f333adb5 LZ |
1274 | struct btrfs_free_space *info) |
1275 | { | |
34d52cb6 LZ |
1276 | __unlink_free_space(ctl, info); |
1277 | ctl->free_space -= info->bytes; | |
0f9dd46c JB |
1278 | } |
1279 | ||
34d52cb6 | 1280 | static int link_free_space(struct btrfs_free_space_ctl *ctl, |
0f9dd46c JB |
1281 | struct btrfs_free_space *info) |
1282 | { | |
1283 | int ret = 0; | |
1284 | ||
96303081 | 1285 | BUG_ON(!info->bitmap && !info->bytes); |
34d52cb6 | 1286 | ret = tree_insert_offset(&ctl->free_space_offset, info->offset, |
96303081 | 1287 | &info->offset_index, (info->bitmap != NULL)); |
0f9dd46c JB |
1288 | if (ret) |
1289 | return ret; | |
1290 | ||
34d52cb6 LZ |
1291 | ctl->free_space += info->bytes; |
1292 | ctl->free_extents++; | |
96303081 JB |
1293 | return ret; |
1294 | } | |
1295 | ||
34d52cb6 | 1296 | static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl) |
96303081 | 1297 | { |
34d52cb6 | 1298 | struct btrfs_block_group_cache *block_group = ctl->private; |
25891f79 JB |
1299 | u64 max_bytes; |
1300 | u64 bitmap_bytes; | |
1301 | u64 extent_bytes; | |
8eb2d829 | 1302 | u64 size = block_group->key.offset; |
34d52cb6 LZ |
1303 | u64 bytes_per_bg = BITS_PER_BITMAP * block_group->sectorsize; |
1304 | int max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg); | |
1305 | ||
1306 | BUG_ON(ctl->total_bitmaps > max_bitmaps); | |
96303081 JB |
1307 | |
1308 | /* | |
1309 | * The goal is to keep the total amount of memory used per 1gb of space | |
1310 | * at or below 32k, so we need to adjust how much memory we allow to be | |
1311 | * used by extent based free space tracking | |
1312 | */ | |
8eb2d829 LZ |
1313 | if (size < 1024 * 1024 * 1024) |
1314 | max_bytes = MAX_CACHE_BYTES_PER_GIG; | |
1315 | else | |
1316 | max_bytes = MAX_CACHE_BYTES_PER_GIG * | |
1317 | div64_u64(size, 1024 * 1024 * 1024); | |
96303081 | 1318 | |
25891f79 JB |
1319 | /* |
1320 | * we want to account for 1 more bitmap than what we have so we can make | |
1321 | * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as | |
1322 | * we add more bitmaps. | |
1323 | */ | |
34d52cb6 | 1324 | bitmap_bytes = (ctl->total_bitmaps + 1) * PAGE_CACHE_SIZE; |
96303081 | 1325 | |
25891f79 | 1326 | if (bitmap_bytes >= max_bytes) { |
34d52cb6 | 1327 | ctl->extents_thresh = 0; |
25891f79 JB |
1328 | return; |
1329 | } | |
96303081 | 1330 | |
25891f79 JB |
1331 | /* |
1332 | * we want the extent entry threshold to always be at most 1/2 the maxw | |
1333 | * bytes we can have, or whatever is less than that. | |
1334 | */ | |
1335 | extent_bytes = max_bytes - bitmap_bytes; | |
1336 | extent_bytes = min_t(u64, extent_bytes, div64_u64(max_bytes, 2)); | |
96303081 | 1337 | |
34d52cb6 | 1338 | ctl->extents_thresh = |
25891f79 | 1339 | div64_u64(extent_bytes, (sizeof(struct btrfs_free_space))); |
96303081 JB |
1340 | } |
1341 | ||
bb3ac5a4 MX |
1342 | static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl, |
1343 | struct btrfs_free_space *info, | |
1344 | u64 offset, u64 bytes) | |
96303081 | 1345 | { |
f38b6e75 | 1346 | unsigned long start, count; |
96303081 | 1347 | |
34d52cb6 LZ |
1348 | start = offset_to_bit(info->offset, ctl->unit, offset); |
1349 | count = bytes_to_bits(bytes, ctl->unit); | |
f38b6e75 | 1350 | BUG_ON(start + count > BITS_PER_BITMAP); |
96303081 | 1351 | |
f38b6e75 | 1352 | bitmap_clear(info->bitmap, start, count); |
96303081 JB |
1353 | |
1354 | info->bytes -= bytes; | |
bb3ac5a4 MX |
1355 | } |
1356 | ||
1357 | static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl, | |
1358 | struct btrfs_free_space *info, u64 offset, | |
1359 | u64 bytes) | |
1360 | { | |
1361 | __bitmap_clear_bits(ctl, info, offset, bytes); | |
34d52cb6 | 1362 | ctl->free_space -= bytes; |
96303081 JB |
1363 | } |
1364 | ||
34d52cb6 | 1365 | static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl, |
817d52f8 JB |
1366 | struct btrfs_free_space *info, u64 offset, |
1367 | u64 bytes) | |
96303081 | 1368 | { |
f38b6e75 | 1369 | unsigned long start, count; |
96303081 | 1370 | |
34d52cb6 LZ |
1371 | start = offset_to_bit(info->offset, ctl->unit, offset); |
1372 | count = bytes_to_bits(bytes, ctl->unit); | |
f38b6e75 | 1373 | BUG_ON(start + count > BITS_PER_BITMAP); |
96303081 | 1374 | |
f38b6e75 | 1375 | bitmap_set(info->bitmap, start, count); |
96303081 JB |
1376 | |
1377 | info->bytes += bytes; | |
34d52cb6 | 1378 | ctl->free_space += bytes; |
96303081 JB |
1379 | } |
1380 | ||
34d52cb6 | 1381 | static int search_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1382 | struct btrfs_free_space *bitmap_info, u64 *offset, |
1383 | u64 *bytes) | |
1384 | { | |
1385 | unsigned long found_bits = 0; | |
1386 | unsigned long bits, i; | |
1387 | unsigned long next_zero; | |
1388 | ||
34d52cb6 | 1389 | i = offset_to_bit(bitmap_info->offset, ctl->unit, |
96303081 | 1390 | max_t(u64, *offset, bitmap_info->offset)); |
34d52cb6 | 1391 | bits = bytes_to_bits(*bytes, ctl->unit); |
96303081 JB |
1392 | |
1393 | for (i = find_next_bit(bitmap_info->bitmap, BITS_PER_BITMAP, i); | |
1394 | i < BITS_PER_BITMAP; | |
1395 | i = find_next_bit(bitmap_info->bitmap, BITS_PER_BITMAP, i + 1)) { | |
1396 | next_zero = find_next_zero_bit(bitmap_info->bitmap, | |
1397 | BITS_PER_BITMAP, i); | |
1398 | if ((next_zero - i) >= bits) { | |
1399 | found_bits = next_zero - i; | |
1400 | break; | |
1401 | } | |
1402 | i = next_zero; | |
1403 | } | |
1404 | ||
1405 | if (found_bits) { | |
34d52cb6 LZ |
1406 | *offset = (u64)(i * ctl->unit) + bitmap_info->offset; |
1407 | *bytes = (u64)(found_bits) * ctl->unit; | |
96303081 JB |
1408 | return 0; |
1409 | } | |
1410 | ||
1411 | return -1; | |
1412 | } | |
1413 | ||
34d52cb6 LZ |
1414 | static struct btrfs_free_space * |
1415 | find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes) | |
96303081 JB |
1416 | { |
1417 | struct btrfs_free_space *entry; | |
1418 | struct rb_node *node; | |
1419 | int ret; | |
1420 | ||
34d52cb6 | 1421 | if (!ctl->free_space_offset.rb_node) |
96303081 JB |
1422 | return NULL; |
1423 | ||
34d52cb6 | 1424 | entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1); |
96303081 JB |
1425 | if (!entry) |
1426 | return NULL; | |
1427 | ||
1428 | for (node = &entry->offset_index; node; node = rb_next(node)) { | |
1429 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
1430 | if (entry->bytes < *bytes) | |
1431 | continue; | |
1432 | ||
1433 | if (entry->bitmap) { | |
34d52cb6 | 1434 | ret = search_bitmap(ctl, entry, offset, bytes); |
96303081 JB |
1435 | if (!ret) |
1436 | return entry; | |
1437 | continue; | |
1438 | } | |
1439 | ||
1440 | *offset = entry->offset; | |
1441 | *bytes = entry->bytes; | |
1442 | return entry; | |
1443 | } | |
1444 | ||
1445 | return NULL; | |
1446 | } | |
1447 | ||
34d52cb6 | 1448 | static void add_new_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1449 | struct btrfs_free_space *info, u64 offset) |
1450 | { | |
34d52cb6 | 1451 | info->offset = offset_to_bitmap(ctl, offset); |
f019f426 | 1452 | info->bytes = 0; |
34d52cb6 LZ |
1453 | link_free_space(ctl, info); |
1454 | ctl->total_bitmaps++; | |
96303081 | 1455 | |
34d52cb6 | 1456 | ctl->op->recalc_thresholds(ctl); |
96303081 JB |
1457 | } |
1458 | ||
34d52cb6 | 1459 | static void free_bitmap(struct btrfs_free_space_ctl *ctl, |
edf6e2d1 LZ |
1460 | struct btrfs_free_space *bitmap_info) |
1461 | { | |
34d52cb6 | 1462 | unlink_free_space(ctl, bitmap_info); |
edf6e2d1 | 1463 | kfree(bitmap_info->bitmap); |
dc89e982 | 1464 | kmem_cache_free(btrfs_free_space_cachep, bitmap_info); |
34d52cb6 LZ |
1465 | ctl->total_bitmaps--; |
1466 | ctl->op->recalc_thresholds(ctl); | |
edf6e2d1 LZ |
1467 | } |
1468 | ||
34d52cb6 | 1469 | static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1470 | struct btrfs_free_space *bitmap_info, |
1471 | u64 *offset, u64 *bytes) | |
1472 | { | |
1473 | u64 end; | |
6606bb97 JB |
1474 | u64 search_start, search_bytes; |
1475 | int ret; | |
96303081 JB |
1476 | |
1477 | again: | |
34d52cb6 | 1478 | end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1; |
96303081 | 1479 | |
6606bb97 JB |
1480 | /* |
1481 | * XXX - this can go away after a few releases. | |
1482 | * | |
1483 | * since the only user of btrfs_remove_free_space is the tree logging | |
1484 | * stuff, and the only way to test that is under crash conditions, we | |
1485 | * want to have this debug stuff here just in case somethings not | |
1486 | * working. Search the bitmap for the space we are trying to use to | |
1487 | * make sure its actually there. If its not there then we need to stop | |
1488 | * because something has gone wrong. | |
1489 | */ | |
1490 | search_start = *offset; | |
1491 | search_bytes = *bytes; | |
13dbc089 | 1492 | search_bytes = min(search_bytes, end - search_start + 1); |
34d52cb6 | 1493 | ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes); |
6606bb97 JB |
1494 | BUG_ON(ret < 0 || search_start != *offset); |
1495 | ||
96303081 | 1496 | if (*offset > bitmap_info->offset && *offset + *bytes > end) { |
34d52cb6 | 1497 | bitmap_clear_bits(ctl, bitmap_info, *offset, end - *offset + 1); |
96303081 JB |
1498 | *bytes -= end - *offset + 1; |
1499 | *offset = end + 1; | |
1500 | } else if (*offset >= bitmap_info->offset && *offset + *bytes <= end) { | |
34d52cb6 | 1501 | bitmap_clear_bits(ctl, bitmap_info, *offset, *bytes); |
96303081 JB |
1502 | *bytes = 0; |
1503 | } | |
1504 | ||
1505 | if (*bytes) { | |
6606bb97 | 1506 | struct rb_node *next = rb_next(&bitmap_info->offset_index); |
edf6e2d1 | 1507 | if (!bitmap_info->bytes) |
34d52cb6 | 1508 | free_bitmap(ctl, bitmap_info); |
96303081 | 1509 | |
6606bb97 JB |
1510 | /* |
1511 | * no entry after this bitmap, but we still have bytes to | |
1512 | * remove, so something has gone wrong. | |
1513 | */ | |
1514 | if (!next) | |
96303081 JB |
1515 | return -EINVAL; |
1516 | ||
6606bb97 JB |
1517 | bitmap_info = rb_entry(next, struct btrfs_free_space, |
1518 | offset_index); | |
1519 | ||
1520 | /* | |
1521 | * if the next entry isn't a bitmap we need to return to let the | |
1522 | * extent stuff do its work. | |
1523 | */ | |
96303081 JB |
1524 | if (!bitmap_info->bitmap) |
1525 | return -EAGAIN; | |
1526 | ||
6606bb97 JB |
1527 | /* |
1528 | * Ok the next item is a bitmap, but it may not actually hold | |
1529 | * the information for the rest of this free space stuff, so | |
1530 | * look for it, and if we don't find it return so we can try | |
1531 | * everything over again. | |
1532 | */ | |
1533 | search_start = *offset; | |
1534 | search_bytes = *bytes; | |
34d52cb6 | 1535 | ret = search_bitmap(ctl, bitmap_info, &search_start, |
6606bb97 JB |
1536 | &search_bytes); |
1537 | if (ret < 0 || search_start != *offset) | |
1538 | return -EAGAIN; | |
1539 | ||
96303081 | 1540 | goto again; |
edf6e2d1 | 1541 | } else if (!bitmap_info->bytes) |
34d52cb6 | 1542 | free_bitmap(ctl, bitmap_info); |
96303081 JB |
1543 | |
1544 | return 0; | |
1545 | } | |
1546 | ||
2cdc342c JB |
1547 | static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl, |
1548 | struct btrfs_free_space *info, u64 offset, | |
1549 | u64 bytes) | |
1550 | { | |
1551 | u64 bytes_to_set = 0; | |
1552 | u64 end; | |
1553 | ||
1554 | end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit); | |
1555 | ||
1556 | bytes_to_set = min(end - offset, bytes); | |
1557 | ||
1558 | bitmap_set_bits(ctl, info, offset, bytes_to_set); | |
1559 | ||
1560 | return bytes_to_set; | |
1561 | ||
1562 | } | |
1563 | ||
34d52cb6 LZ |
1564 | static bool use_bitmap(struct btrfs_free_space_ctl *ctl, |
1565 | struct btrfs_free_space *info) | |
96303081 | 1566 | { |
34d52cb6 | 1567 | struct btrfs_block_group_cache *block_group = ctl->private; |
96303081 JB |
1568 | |
1569 | /* | |
1570 | * If we are below the extents threshold then we can add this as an | |
1571 | * extent, and don't have to deal with the bitmap | |
1572 | */ | |
34d52cb6 | 1573 | if (ctl->free_extents < ctl->extents_thresh) { |
32cb0840 JB |
1574 | /* |
1575 | * If this block group has some small extents we don't want to | |
1576 | * use up all of our free slots in the cache with them, we want | |
1577 | * to reserve them to larger extents, however if we have plent | |
1578 | * of cache left then go ahead an dadd them, no sense in adding | |
1579 | * the overhead of a bitmap if we don't have to. | |
1580 | */ | |
1581 | if (info->bytes <= block_group->sectorsize * 4) { | |
34d52cb6 LZ |
1582 | if (ctl->free_extents * 2 <= ctl->extents_thresh) |
1583 | return false; | |
32cb0840 | 1584 | } else { |
34d52cb6 | 1585 | return false; |
32cb0840 JB |
1586 | } |
1587 | } | |
96303081 JB |
1588 | |
1589 | /* | |
1590 | * some block groups are so tiny they can't be enveloped by a bitmap, so | |
1591 | * don't even bother to create a bitmap for this | |
1592 | */ | |
1593 | if (BITS_PER_BITMAP * block_group->sectorsize > | |
1594 | block_group->key.offset) | |
34d52cb6 LZ |
1595 | return false; |
1596 | ||
1597 | return true; | |
1598 | } | |
1599 | ||
2cdc342c JB |
1600 | static struct btrfs_free_space_op free_space_op = { |
1601 | .recalc_thresholds = recalculate_thresholds, | |
1602 | .use_bitmap = use_bitmap, | |
1603 | }; | |
1604 | ||
34d52cb6 LZ |
1605 | static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl, |
1606 | struct btrfs_free_space *info) | |
1607 | { | |
1608 | struct btrfs_free_space *bitmap_info; | |
2cdc342c | 1609 | struct btrfs_block_group_cache *block_group = NULL; |
34d52cb6 | 1610 | int added = 0; |
2cdc342c | 1611 | u64 bytes, offset, bytes_added; |
34d52cb6 | 1612 | int ret; |
96303081 JB |
1613 | |
1614 | bytes = info->bytes; | |
1615 | offset = info->offset; | |
1616 | ||
34d52cb6 LZ |
1617 | if (!ctl->op->use_bitmap(ctl, info)) |
1618 | return 0; | |
1619 | ||
2cdc342c JB |
1620 | if (ctl->op == &free_space_op) |
1621 | block_group = ctl->private; | |
38e87880 | 1622 | again: |
2cdc342c JB |
1623 | /* |
1624 | * Since we link bitmaps right into the cluster we need to see if we | |
1625 | * have a cluster here, and if so and it has our bitmap we need to add | |
1626 | * the free space to that bitmap. | |
1627 | */ | |
1628 | if (block_group && !list_empty(&block_group->cluster_list)) { | |
1629 | struct btrfs_free_cluster *cluster; | |
1630 | struct rb_node *node; | |
1631 | struct btrfs_free_space *entry; | |
1632 | ||
1633 | cluster = list_entry(block_group->cluster_list.next, | |
1634 | struct btrfs_free_cluster, | |
1635 | block_group_list); | |
1636 | spin_lock(&cluster->lock); | |
1637 | node = rb_first(&cluster->root); | |
1638 | if (!node) { | |
1639 | spin_unlock(&cluster->lock); | |
38e87880 | 1640 | goto no_cluster_bitmap; |
2cdc342c JB |
1641 | } |
1642 | ||
1643 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
1644 | if (!entry->bitmap) { | |
1645 | spin_unlock(&cluster->lock); | |
38e87880 | 1646 | goto no_cluster_bitmap; |
2cdc342c JB |
1647 | } |
1648 | ||
1649 | if (entry->offset == offset_to_bitmap(ctl, offset)) { | |
1650 | bytes_added = add_bytes_to_bitmap(ctl, entry, | |
1651 | offset, bytes); | |
1652 | bytes -= bytes_added; | |
1653 | offset += bytes_added; | |
1654 | } | |
1655 | spin_unlock(&cluster->lock); | |
1656 | if (!bytes) { | |
1657 | ret = 1; | |
1658 | goto out; | |
1659 | } | |
1660 | } | |
38e87880 CM |
1661 | |
1662 | no_cluster_bitmap: | |
34d52cb6 | 1663 | bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), |
96303081 JB |
1664 | 1, 0); |
1665 | if (!bitmap_info) { | |
1666 | BUG_ON(added); | |
1667 | goto new_bitmap; | |
1668 | } | |
1669 | ||
2cdc342c JB |
1670 | bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes); |
1671 | bytes -= bytes_added; | |
1672 | offset += bytes_added; | |
1673 | added = 0; | |
96303081 JB |
1674 | |
1675 | if (!bytes) { | |
1676 | ret = 1; | |
1677 | goto out; | |
1678 | } else | |
1679 | goto again; | |
1680 | ||
1681 | new_bitmap: | |
1682 | if (info && info->bitmap) { | |
34d52cb6 | 1683 | add_new_bitmap(ctl, info, offset); |
96303081 JB |
1684 | added = 1; |
1685 | info = NULL; | |
1686 | goto again; | |
1687 | } else { | |
34d52cb6 | 1688 | spin_unlock(&ctl->tree_lock); |
96303081 JB |
1689 | |
1690 | /* no pre-allocated info, allocate a new one */ | |
1691 | if (!info) { | |
dc89e982 JB |
1692 | info = kmem_cache_zalloc(btrfs_free_space_cachep, |
1693 | GFP_NOFS); | |
96303081 | 1694 | if (!info) { |
34d52cb6 | 1695 | spin_lock(&ctl->tree_lock); |
96303081 JB |
1696 | ret = -ENOMEM; |
1697 | goto out; | |
1698 | } | |
1699 | } | |
1700 | ||
1701 | /* allocate the bitmap */ | |
1702 | info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS); | |
34d52cb6 | 1703 | spin_lock(&ctl->tree_lock); |
96303081 JB |
1704 | if (!info->bitmap) { |
1705 | ret = -ENOMEM; | |
1706 | goto out; | |
1707 | } | |
1708 | goto again; | |
1709 | } | |
1710 | ||
1711 | out: | |
1712 | if (info) { | |
1713 | if (info->bitmap) | |
1714 | kfree(info->bitmap); | |
dc89e982 | 1715 | kmem_cache_free(btrfs_free_space_cachep, info); |
96303081 | 1716 | } |
0f9dd46c JB |
1717 | |
1718 | return ret; | |
1719 | } | |
1720 | ||
945d8962 | 1721 | static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl, |
f333adb5 | 1722 | struct btrfs_free_space *info, bool update_stat) |
0f9dd46c | 1723 | { |
120d66ee LZ |
1724 | struct btrfs_free_space *left_info; |
1725 | struct btrfs_free_space *right_info; | |
1726 | bool merged = false; | |
1727 | u64 offset = info->offset; | |
1728 | u64 bytes = info->bytes; | |
6226cb0a | 1729 | |
0f9dd46c JB |
1730 | /* |
1731 | * first we want to see if there is free space adjacent to the range we | |
1732 | * are adding, if there is remove that struct and add a new one to | |
1733 | * cover the entire range | |
1734 | */ | |
34d52cb6 | 1735 | right_info = tree_search_offset(ctl, offset + bytes, 0, 0); |
96303081 JB |
1736 | if (right_info && rb_prev(&right_info->offset_index)) |
1737 | left_info = rb_entry(rb_prev(&right_info->offset_index), | |
1738 | struct btrfs_free_space, offset_index); | |
1739 | else | |
34d52cb6 | 1740 | left_info = tree_search_offset(ctl, offset - 1, 0, 0); |
0f9dd46c | 1741 | |
96303081 | 1742 | if (right_info && !right_info->bitmap) { |
f333adb5 | 1743 | if (update_stat) |
34d52cb6 | 1744 | unlink_free_space(ctl, right_info); |
f333adb5 | 1745 | else |
34d52cb6 | 1746 | __unlink_free_space(ctl, right_info); |
6226cb0a | 1747 | info->bytes += right_info->bytes; |
dc89e982 | 1748 | kmem_cache_free(btrfs_free_space_cachep, right_info); |
120d66ee | 1749 | merged = true; |
0f9dd46c JB |
1750 | } |
1751 | ||
96303081 JB |
1752 | if (left_info && !left_info->bitmap && |
1753 | left_info->offset + left_info->bytes == offset) { | |
f333adb5 | 1754 | if (update_stat) |
34d52cb6 | 1755 | unlink_free_space(ctl, left_info); |
f333adb5 | 1756 | else |
34d52cb6 | 1757 | __unlink_free_space(ctl, left_info); |
6226cb0a JB |
1758 | info->offset = left_info->offset; |
1759 | info->bytes += left_info->bytes; | |
dc89e982 | 1760 | kmem_cache_free(btrfs_free_space_cachep, left_info); |
120d66ee | 1761 | merged = true; |
0f9dd46c JB |
1762 | } |
1763 | ||
120d66ee LZ |
1764 | return merged; |
1765 | } | |
1766 | ||
581bb050 LZ |
1767 | int __btrfs_add_free_space(struct btrfs_free_space_ctl *ctl, |
1768 | u64 offset, u64 bytes) | |
120d66ee LZ |
1769 | { |
1770 | struct btrfs_free_space *info; | |
1771 | int ret = 0; | |
1772 | ||
dc89e982 | 1773 | info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS); |
120d66ee LZ |
1774 | if (!info) |
1775 | return -ENOMEM; | |
1776 | ||
1777 | info->offset = offset; | |
1778 | info->bytes = bytes; | |
1779 | ||
34d52cb6 | 1780 | spin_lock(&ctl->tree_lock); |
120d66ee | 1781 | |
34d52cb6 | 1782 | if (try_merge_free_space(ctl, info, true)) |
120d66ee LZ |
1783 | goto link; |
1784 | ||
1785 | /* | |
1786 | * There was no extent directly to the left or right of this new | |
1787 | * extent then we know we're going to have to allocate a new extent, so | |
1788 | * before we do that see if we need to drop this into a bitmap | |
1789 | */ | |
34d52cb6 | 1790 | ret = insert_into_bitmap(ctl, info); |
120d66ee LZ |
1791 | if (ret < 0) { |
1792 | goto out; | |
1793 | } else if (ret) { | |
1794 | ret = 0; | |
1795 | goto out; | |
1796 | } | |
1797 | link: | |
34d52cb6 | 1798 | ret = link_free_space(ctl, info); |
0f9dd46c | 1799 | if (ret) |
dc89e982 | 1800 | kmem_cache_free(btrfs_free_space_cachep, info); |
96303081 | 1801 | out: |
34d52cb6 | 1802 | spin_unlock(&ctl->tree_lock); |
6226cb0a | 1803 | |
0f9dd46c | 1804 | if (ret) { |
96303081 | 1805 | printk(KERN_CRIT "btrfs: unable to add free space :%d\n", ret); |
c293498b | 1806 | BUG_ON(ret == -EEXIST); |
0f9dd46c JB |
1807 | } |
1808 | ||
0f9dd46c JB |
1809 | return ret; |
1810 | } | |
1811 | ||
6226cb0a JB |
1812 | int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group, |
1813 | u64 offset, u64 bytes) | |
0f9dd46c | 1814 | { |
34d52cb6 | 1815 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0f9dd46c | 1816 | struct btrfs_free_space *info; |
96303081 | 1817 | struct btrfs_free_space *next_info = NULL; |
0f9dd46c JB |
1818 | int ret = 0; |
1819 | ||
34d52cb6 | 1820 | spin_lock(&ctl->tree_lock); |
6226cb0a | 1821 | |
96303081 | 1822 | again: |
34d52cb6 | 1823 | info = tree_search_offset(ctl, offset, 0, 0); |
96303081 | 1824 | if (!info) { |
6606bb97 JB |
1825 | /* |
1826 | * oops didn't find an extent that matched the space we wanted | |
1827 | * to remove, look for a bitmap instead | |
1828 | */ | |
34d52cb6 | 1829 | info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), |
6606bb97 JB |
1830 | 1, 0); |
1831 | if (!info) { | |
1832 | WARN_ON(1); | |
1833 | goto out_lock; | |
1834 | } | |
96303081 JB |
1835 | } |
1836 | ||
1837 | if (info->bytes < bytes && rb_next(&info->offset_index)) { | |
1838 | u64 end; | |
1839 | next_info = rb_entry(rb_next(&info->offset_index), | |
1840 | struct btrfs_free_space, | |
1841 | offset_index); | |
1842 | ||
1843 | if (next_info->bitmap) | |
34d52cb6 LZ |
1844 | end = next_info->offset + |
1845 | BITS_PER_BITMAP * ctl->unit - 1; | |
96303081 JB |
1846 | else |
1847 | end = next_info->offset + next_info->bytes; | |
1848 | ||
1849 | if (next_info->bytes < bytes || | |
1850 | next_info->offset > offset || offset > end) { | |
1851 | printk(KERN_CRIT "Found free space at %llu, size %llu," | |
1852 | " trying to use %llu\n", | |
1853 | (unsigned long long)info->offset, | |
1854 | (unsigned long long)info->bytes, | |
1855 | (unsigned long long)bytes); | |
0f9dd46c JB |
1856 | WARN_ON(1); |
1857 | ret = -EINVAL; | |
96303081 | 1858 | goto out_lock; |
0f9dd46c | 1859 | } |
0f9dd46c | 1860 | |
96303081 JB |
1861 | info = next_info; |
1862 | } | |
1863 | ||
1864 | if (info->bytes == bytes) { | |
34d52cb6 | 1865 | unlink_free_space(ctl, info); |
96303081 JB |
1866 | if (info->bitmap) { |
1867 | kfree(info->bitmap); | |
34d52cb6 | 1868 | ctl->total_bitmaps--; |
0f9dd46c | 1869 | } |
dc89e982 | 1870 | kmem_cache_free(btrfs_free_space_cachep, info); |
96303081 JB |
1871 | goto out_lock; |
1872 | } | |
0f9dd46c | 1873 | |
96303081 | 1874 | if (!info->bitmap && info->offset == offset) { |
34d52cb6 | 1875 | unlink_free_space(ctl, info); |
0f9dd46c JB |
1876 | info->offset += bytes; |
1877 | info->bytes -= bytes; | |
34d52cb6 | 1878 | link_free_space(ctl, info); |
96303081 JB |
1879 | goto out_lock; |
1880 | } | |
0f9dd46c | 1881 | |
96303081 JB |
1882 | if (!info->bitmap && info->offset <= offset && |
1883 | info->offset + info->bytes >= offset + bytes) { | |
9b49c9b9 CM |
1884 | u64 old_start = info->offset; |
1885 | /* | |
1886 | * we're freeing space in the middle of the info, | |
1887 | * this can happen during tree log replay | |
1888 | * | |
1889 | * first unlink the old info and then | |
1890 | * insert it again after the hole we're creating | |
1891 | */ | |
34d52cb6 | 1892 | unlink_free_space(ctl, info); |
9b49c9b9 CM |
1893 | if (offset + bytes < info->offset + info->bytes) { |
1894 | u64 old_end = info->offset + info->bytes; | |
1895 | ||
1896 | info->offset = offset + bytes; | |
1897 | info->bytes = old_end - info->offset; | |
34d52cb6 | 1898 | ret = link_free_space(ctl, info); |
96303081 JB |
1899 | WARN_ON(ret); |
1900 | if (ret) | |
1901 | goto out_lock; | |
9b49c9b9 CM |
1902 | } else { |
1903 | /* the hole we're creating ends at the end | |
1904 | * of the info struct, just free the info | |
1905 | */ | |
dc89e982 | 1906 | kmem_cache_free(btrfs_free_space_cachep, info); |
9b49c9b9 | 1907 | } |
34d52cb6 | 1908 | spin_unlock(&ctl->tree_lock); |
96303081 JB |
1909 | |
1910 | /* step two, insert a new info struct to cover | |
1911 | * anything before the hole | |
9b49c9b9 | 1912 | */ |
6226cb0a JB |
1913 | ret = btrfs_add_free_space(block_group, old_start, |
1914 | offset - old_start); | |
96303081 JB |
1915 | WARN_ON(ret); |
1916 | goto out; | |
0f9dd46c | 1917 | } |
96303081 | 1918 | |
34d52cb6 | 1919 | ret = remove_from_bitmap(ctl, info, &offset, &bytes); |
96303081 JB |
1920 | if (ret == -EAGAIN) |
1921 | goto again; | |
1922 | BUG_ON(ret); | |
1923 | out_lock: | |
34d52cb6 | 1924 | spin_unlock(&ctl->tree_lock); |
0f9dd46c | 1925 | out: |
25179201 JB |
1926 | return ret; |
1927 | } | |
1928 | ||
0f9dd46c JB |
1929 | void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group, |
1930 | u64 bytes) | |
1931 | { | |
34d52cb6 | 1932 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0f9dd46c JB |
1933 | struct btrfs_free_space *info; |
1934 | struct rb_node *n; | |
1935 | int count = 0; | |
1936 | ||
34d52cb6 | 1937 | for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) { |
0f9dd46c JB |
1938 | info = rb_entry(n, struct btrfs_free_space, offset_index); |
1939 | if (info->bytes >= bytes) | |
1940 | count++; | |
96303081 | 1941 | printk(KERN_CRIT "entry offset %llu, bytes %llu, bitmap %s\n", |
21380931 | 1942 | (unsigned long long)info->offset, |
96303081 JB |
1943 | (unsigned long long)info->bytes, |
1944 | (info->bitmap) ? "yes" : "no"); | |
0f9dd46c | 1945 | } |
96303081 JB |
1946 | printk(KERN_INFO "block group has cluster?: %s\n", |
1947 | list_empty(&block_group->cluster_list) ? "no" : "yes"); | |
0f9dd46c JB |
1948 | printk(KERN_INFO "%d blocks of free space at or bigger than bytes is" |
1949 | "\n", count); | |
1950 | } | |
1951 | ||
34d52cb6 | 1952 | void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group) |
0f9dd46c | 1953 | { |
34d52cb6 | 1954 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0f9dd46c | 1955 | |
34d52cb6 LZ |
1956 | spin_lock_init(&ctl->tree_lock); |
1957 | ctl->unit = block_group->sectorsize; | |
1958 | ctl->start = block_group->key.objectid; | |
1959 | ctl->private = block_group; | |
1960 | ctl->op = &free_space_op; | |
0f9dd46c | 1961 | |
34d52cb6 LZ |
1962 | /* |
1963 | * we only want to have 32k of ram per block group for keeping | |
1964 | * track of free space, and if we pass 1/2 of that we want to | |
1965 | * start converting things over to using bitmaps | |
1966 | */ | |
1967 | ctl->extents_thresh = ((1024 * 32) / 2) / | |
1968 | sizeof(struct btrfs_free_space); | |
0f9dd46c JB |
1969 | } |
1970 | ||
fa9c0d79 CM |
1971 | /* |
1972 | * for a given cluster, put all of its extents back into the free | |
1973 | * space cache. If the block group passed doesn't match the block group | |
1974 | * pointed to by the cluster, someone else raced in and freed the | |
1975 | * cluster already. In that case, we just return without changing anything | |
1976 | */ | |
1977 | static int | |
1978 | __btrfs_return_cluster_to_free_space( | |
1979 | struct btrfs_block_group_cache *block_group, | |
1980 | struct btrfs_free_cluster *cluster) | |
1981 | { | |
34d52cb6 | 1982 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
fa9c0d79 CM |
1983 | struct btrfs_free_space *entry; |
1984 | struct rb_node *node; | |
1985 | ||
1986 | spin_lock(&cluster->lock); | |
1987 | if (cluster->block_group != block_group) | |
1988 | goto out; | |
1989 | ||
96303081 | 1990 | cluster->block_group = NULL; |
fa9c0d79 | 1991 | cluster->window_start = 0; |
96303081 | 1992 | list_del_init(&cluster->block_group_list); |
96303081 | 1993 | |
fa9c0d79 | 1994 | node = rb_first(&cluster->root); |
96303081 | 1995 | while (node) { |
4e69b598 JB |
1996 | bool bitmap; |
1997 | ||
fa9c0d79 CM |
1998 | entry = rb_entry(node, struct btrfs_free_space, offset_index); |
1999 | node = rb_next(&entry->offset_index); | |
2000 | rb_erase(&entry->offset_index, &cluster->root); | |
4e69b598 JB |
2001 | |
2002 | bitmap = (entry->bitmap != NULL); | |
2003 | if (!bitmap) | |
34d52cb6 LZ |
2004 | try_merge_free_space(ctl, entry, false); |
2005 | tree_insert_offset(&ctl->free_space_offset, | |
4e69b598 | 2006 | entry->offset, &entry->offset_index, bitmap); |
fa9c0d79 | 2007 | } |
6bef4d31 | 2008 | cluster->root = RB_ROOT; |
96303081 | 2009 | |
fa9c0d79 CM |
2010 | out: |
2011 | spin_unlock(&cluster->lock); | |
96303081 | 2012 | btrfs_put_block_group(block_group); |
fa9c0d79 CM |
2013 | return 0; |
2014 | } | |
2015 | ||
09655373 | 2016 | void __btrfs_remove_free_space_cache_locked(struct btrfs_free_space_ctl *ctl) |
0f9dd46c JB |
2017 | { |
2018 | struct btrfs_free_space *info; | |
2019 | struct rb_node *node; | |
581bb050 | 2020 | |
581bb050 LZ |
2021 | while ((node = rb_last(&ctl->free_space_offset)) != NULL) { |
2022 | info = rb_entry(node, struct btrfs_free_space, offset_index); | |
9b90f513 JB |
2023 | if (!info->bitmap) { |
2024 | unlink_free_space(ctl, info); | |
2025 | kmem_cache_free(btrfs_free_space_cachep, info); | |
2026 | } else { | |
2027 | free_bitmap(ctl, info); | |
2028 | } | |
581bb050 LZ |
2029 | if (need_resched()) { |
2030 | spin_unlock(&ctl->tree_lock); | |
2031 | cond_resched(); | |
2032 | spin_lock(&ctl->tree_lock); | |
2033 | } | |
2034 | } | |
09655373 CM |
2035 | } |
2036 | ||
2037 | void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl) | |
2038 | { | |
2039 | spin_lock(&ctl->tree_lock); | |
2040 | __btrfs_remove_free_space_cache_locked(ctl); | |
581bb050 LZ |
2041 | spin_unlock(&ctl->tree_lock); |
2042 | } | |
2043 | ||
2044 | void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group) | |
2045 | { | |
2046 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
fa9c0d79 | 2047 | struct btrfs_free_cluster *cluster; |
96303081 | 2048 | struct list_head *head; |
0f9dd46c | 2049 | |
34d52cb6 | 2050 | spin_lock(&ctl->tree_lock); |
96303081 JB |
2051 | while ((head = block_group->cluster_list.next) != |
2052 | &block_group->cluster_list) { | |
2053 | cluster = list_entry(head, struct btrfs_free_cluster, | |
2054 | block_group_list); | |
fa9c0d79 CM |
2055 | |
2056 | WARN_ON(cluster->block_group != block_group); | |
2057 | __btrfs_return_cluster_to_free_space(block_group, cluster); | |
96303081 | 2058 | if (need_resched()) { |
34d52cb6 | 2059 | spin_unlock(&ctl->tree_lock); |
96303081 | 2060 | cond_resched(); |
34d52cb6 | 2061 | spin_lock(&ctl->tree_lock); |
96303081 | 2062 | } |
fa9c0d79 | 2063 | } |
09655373 | 2064 | __btrfs_remove_free_space_cache_locked(ctl); |
34d52cb6 | 2065 | spin_unlock(&ctl->tree_lock); |
fa9c0d79 | 2066 | |
0f9dd46c JB |
2067 | } |
2068 | ||
6226cb0a JB |
2069 | u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group, |
2070 | u64 offset, u64 bytes, u64 empty_size) | |
0f9dd46c | 2071 | { |
34d52cb6 | 2072 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
6226cb0a | 2073 | struct btrfs_free_space *entry = NULL; |
96303081 | 2074 | u64 bytes_search = bytes + empty_size; |
6226cb0a | 2075 | u64 ret = 0; |
0f9dd46c | 2076 | |
34d52cb6 LZ |
2077 | spin_lock(&ctl->tree_lock); |
2078 | entry = find_free_space(ctl, &offset, &bytes_search); | |
6226cb0a | 2079 | if (!entry) |
96303081 JB |
2080 | goto out; |
2081 | ||
2082 | ret = offset; | |
2083 | if (entry->bitmap) { | |
34d52cb6 | 2084 | bitmap_clear_bits(ctl, entry, offset, bytes); |
edf6e2d1 | 2085 | if (!entry->bytes) |
34d52cb6 | 2086 | free_bitmap(ctl, entry); |
96303081 | 2087 | } else { |
34d52cb6 | 2088 | unlink_free_space(ctl, entry); |
6226cb0a JB |
2089 | entry->offset += bytes; |
2090 | entry->bytes -= bytes; | |
6226cb0a | 2091 | if (!entry->bytes) |
dc89e982 | 2092 | kmem_cache_free(btrfs_free_space_cachep, entry); |
6226cb0a | 2093 | else |
34d52cb6 | 2094 | link_free_space(ctl, entry); |
6226cb0a | 2095 | } |
0f9dd46c | 2096 | |
96303081 | 2097 | out: |
34d52cb6 | 2098 | spin_unlock(&ctl->tree_lock); |
817d52f8 | 2099 | |
0f9dd46c JB |
2100 | return ret; |
2101 | } | |
fa9c0d79 CM |
2102 | |
2103 | /* | |
2104 | * given a cluster, put all of its extents back into the free space | |
2105 | * cache. If a block group is passed, this function will only free | |
2106 | * a cluster that belongs to the passed block group. | |
2107 | * | |
2108 | * Otherwise, it'll get a reference on the block group pointed to by the | |
2109 | * cluster and remove the cluster from it. | |
2110 | */ | |
2111 | int btrfs_return_cluster_to_free_space( | |
2112 | struct btrfs_block_group_cache *block_group, | |
2113 | struct btrfs_free_cluster *cluster) | |
2114 | { | |
34d52cb6 | 2115 | struct btrfs_free_space_ctl *ctl; |
fa9c0d79 CM |
2116 | int ret; |
2117 | ||
2118 | /* first, get a safe pointer to the block group */ | |
2119 | spin_lock(&cluster->lock); | |
2120 | if (!block_group) { | |
2121 | block_group = cluster->block_group; | |
2122 | if (!block_group) { | |
2123 | spin_unlock(&cluster->lock); | |
2124 | return 0; | |
2125 | } | |
2126 | } else if (cluster->block_group != block_group) { | |
2127 | /* someone else has already freed it don't redo their work */ | |
2128 | spin_unlock(&cluster->lock); | |
2129 | return 0; | |
2130 | } | |
2131 | atomic_inc(&block_group->count); | |
2132 | spin_unlock(&cluster->lock); | |
2133 | ||
34d52cb6 LZ |
2134 | ctl = block_group->free_space_ctl; |
2135 | ||
fa9c0d79 | 2136 | /* now return any extents the cluster had on it */ |
34d52cb6 | 2137 | spin_lock(&ctl->tree_lock); |
fa9c0d79 | 2138 | ret = __btrfs_return_cluster_to_free_space(block_group, cluster); |
34d52cb6 | 2139 | spin_unlock(&ctl->tree_lock); |
fa9c0d79 CM |
2140 | |
2141 | /* finally drop our ref */ | |
2142 | btrfs_put_block_group(block_group); | |
2143 | return ret; | |
2144 | } | |
2145 | ||
96303081 JB |
2146 | static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group, |
2147 | struct btrfs_free_cluster *cluster, | |
4e69b598 | 2148 | struct btrfs_free_space *entry, |
96303081 JB |
2149 | u64 bytes, u64 min_start) |
2150 | { | |
34d52cb6 | 2151 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
96303081 JB |
2152 | int err; |
2153 | u64 search_start = cluster->window_start; | |
2154 | u64 search_bytes = bytes; | |
2155 | u64 ret = 0; | |
2156 | ||
96303081 JB |
2157 | search_start = min_start; |
2158 | search_bytes = bytes; | |
2159 | ||
34d52cb6 | 2160 | err = search_bitmap(ctl, entry, &search_start, &search_bytes); |
96303081 | 2161 | if (err) |
4e69b598 | 2162 | return 0; |
96303081 JB |
2163 | |
2164 | ret = search_start; | |
bb3ac5a4 | 2165 | __bitmap_clear_bits(ctl, entry, ret, bytes); |
96303081 JB |
2166 | |
2167 | return ret; | |
2168 | } | |
2169 | ||
fa9c0d79 CM |
2170 | /* |
2171 | * given a cluster, try to allocate 'bytes' from it, returns 0 | |
2172 | * if it couldn't find anything suitably large, or a logical disk offset | |
2173 | * if things worked out | |
2174 | */ | |
2175 | u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group, | |
2176 | struct btrfs_free_cluster *cluster, u64 bytes, | |
2177 | u64 min_start) | |
2178 | { | |
34d52cb6 | 2179 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
fa9c0d79 CM |
2180 | struct btrfs_free_space *entry = NULL; |
2181 | struct rb_node *node; | |
2182 | u64 ret = 0; | |
2183 | ||
2184 | spin_lock(&cluster->lock); | |
2185 | if (bytes > cluster->max_size) | |
2186 | goto out; | |
2187 | ||
2188 | if (cluster->block_group != block_group) | |
2189 | goto out; | |
2190 | ||
2191 | node = rb_first(&cluster->root); | |
2192 | if (!node) | |
2193 | goto out; | |
2194 | ||
2195 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
fa9c0d79 | 2196 | while(1) { |
4e69b598 JB |
2197 | if (entry->bytes < bytes || |
2198 | (!entry->bitmap && entry->offset < min_start)) { | |
fa9c0d79 CM |
2199 | node = rb_next(&entry->offset_index); |
2200 | if (!node) | |
2201 | break; | |
2202 | entry = rb_entry(node, struct btrfs_free_space, | |
2203 | offset_index); | |
2204 | continue; | |
2205 | } | |
fa9c0d79 | 2206 | |
4e69b598 JB |
2207 | if (entry->bitmap) { |
2208 | ret = btrfs_alloc_from_bitmap(block_group, | |
2209 | cluster, entry, bytes, | |
2210 | min_start); | |
2211 | if (ret == 0) { | |
4e69b598 JB |
2212 | node = rb_next(&entry->offset_index); |
2213 | if (!node) | |
2214 | break; | |
2215 | entry = rb_entry(node, struct btrfs_free_space, | |
2216 | offset_index); | |
2217 | continue; | |
2218 | } | |
2219 | } else { | |
4e69b598 JB |
2220 | ret = entry->offset; |
2221 | ||
2222 | entry->offset += bytes; | |
2223 | entry->bytes -= bytes; | |
2224 | } | |
fa9c0d79 | 2225 | |
5e71b5d5 | 2226 | if (entry->bytes == 0) |
fa9c0d79 | 2227 | rb_erase(&entry->offset_index, &cluster->root); |
fa9c0d79 CM |
2228 | break; |
2229 | } | |
2230 | out: | |
2231 | spin_unlock(&cluster->lock); | |
96303081 | 2232 | |
5e71b5d5 LZ |
2233 | if (!ret) |
2234 | return 0; | |
2235 | ||
34d52cb6 | 2236 | spin_lock(&ctl->tree_lock); |
5e71b5d5 | 2237 | |
34d52cb6 | 2238 | ctl->free_space -= bytes; |
5e71b5d5 | 2239 | if (entry->bytes == 0) { |
34d52cb6 | 2240 | ctl->free_extents--; |
4e69b598 JB |
2241 | if (entry->bitmap) { |
2242 | kfree(entry->bitmap); | |
34d52cb6 LZ |
2243 | ctl->total_bitmaps--; |
2244 | ctl->op->recalc_thresholds(ctl); | |
4e69b598 | 2245 | } |
dc89e982 | 2246 | kmem_cache_free(btrfs_free_space_cachep, entry); |
5e71b5d5 LZ |
2247 | } |
2248 | ||
34d52cb6 | 2249 | spin_unlock(&ctl->tree_lock); |
5e71b5d5 | 2250 | |
fa9c0d79 CM |
2251 | return ret; |
2252 | } | |
2253 | ||
96303081 JB |
2254 | static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group, |
2255 | struct btrfs_free_space *entry, | |
2256 | struct btrfs_free_cluster *cluster, | |
2257 | u64 offset, u64 bytes, u64 min_bytes) | |
2258 | { | |
34d52cb6 | 2259 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
96303081 JB |
2260 | unsigned long next_zero; |
2261 | unsigned long i; | |
2262 | unsigned long search_bits; | |
2263 | unsigned long total_bits; | |
2264 | unsigned long found_bits; | |
2265 | unsigned long start = 0; | |
2266 | unsigned long total_found = 0; | |
4e69b598 | 2267 | int ret; |
96303081 JB |
2268 | bool found = false; |
2269 | ||
2270 | i = offset_to_bit(entry->offset, block_group->sectorsize, | |
2271 | max_t(u64, offset, entry->offset)); | |
d0a365e8 JB |
2272 | search_bits = bytes_to_bits(bytes, block_group->sectorsize); |
2273 | total_bits = bytes_to_bits(min_bytes, block_group->sectorsize); | |
96303081 JB |
2274 | |
2275 | again: | |
2276 | found_bits = 0; | |
2277 | for (i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, i); | |
2278 | i < BITS_PER_BITMAP; | |
2279 | i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, i + 1)) { | |
2280 | next_zero = find_next_zero_bit(entry->bitmap, | |
2281 | BITS_PER_BITMAP, i); | |
2282 | if (next_zero - i >= search_bits) { | |
2283 | found_bits = next_zero - i; | |
2284 | break; | |
2285 | } | |
2286 | i = next_zero; | |
2287 | } | |
2288 | ||
2289 | if (!found_bits) | |
4e69b598 | 2290 | return -ENOSPC; |
96303081 JB |
2291 | |
2292 | if (!found) { | |
2293 | start = i; | |
2294 | found = true; | |
2295 | } | |
2296 | ||
2297 | total_found += found_bits; | |
2298 | ||
2299 | if (cluster->max_size < found_bits * block_group->sectorsize) | |
2300 | cluster->max_size = found_bits * block_group->sectorsize; | |
2301 | ||
2302 | if (total_found < total_bits) { | |
2303 | i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, next_zero); | |
2304 | if (i - start > total_bits * 2) { | |
2305 | total_found = 0; | |
2306 | cluster->max_size = 0; | |
2307 | found = false; | |
2308 | } | |
2309 | goto again; | |
2310 | } | |
2311 | ||
2312 | cluster->window_start = start * block_group->sectorsize + | |
2313 | entry->offset; | |
34d52cb6 | 2314 | rb_erase(&entry->offset_index, &ctl->free_space_offset); |
4e69b598 JB |
2315 | ret = tree_insert_offset(&cluster->root, entry->offset, |
2316 | &entry->offset_index, 1); | |
2317 | BUG_ON(ret); | |
96303081 JB |
2318 | |
2319 | return 0; | |
2320 | } | |
2321 | ||
4e69b598 JB |
2322 | /* |
2323 | * This searches the block group for just extents to fill the cluster with. | |
2324 | */ | |
3de85bb9 JB |
2325 | static noinline int |
2326 | setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group, | |
2327 | struct btrfs_free_cluster *cluster, | |
2328 | struct list_head *bitmaps, u64 offset, u64 bytes, | |
2329 | u64 min_bytes) | |
4e69b598 | 2330 | { |
34d52cb6 | 2331 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
4e69b598 JB |
2332 | struct btrfs_free_space *first = NULL; |
2333 | struct btrfs_free_space *entry = NULL; | |
2334 | struct btrfs_free_space *prev = NULL; | |
2335 | struct btrfs_free_space *last; | |
2336 | struct rb_node *node; | |
2337 | u64 window_start; | |
2338 | u64 window_free; | |
2339 | u64 max_extent; | |
2340 | u64 max_gap = 128 * 1024; | |
2341 | ||
34d52cb6 | 2342 | entry = tree_search_offset(ctl, offset, 0, 1); |
4e69b598 JB |
2343 | if (!entry) |
2344 | return -ENOSPC; | |
2345 | ||
2346 | /* | |
2347 | * We don't want bitmaps, so just move along until we find a normal | |
2348 | * extent entry. | |
2349 | */ | |
2350 | while (entry->bitmap) { | |
86d4a77b JB |
2351 | if (list_empty(&entry->list)) |
2352 | list_add_tail(&entry->list, bitmaps); | |
4e69b598 JB |
2353 | node = rb_next(&entry->offset_index); |
2354 | if (!node) | |
2355 | return -ENOSPC; | |
2356 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
2357 | } | |
2358 | ||
2359 | window_start = entry->offset; | |
2360 | window_free = entry->bytes; | |
2361 | max_extent = entry->bytes; | |
2362 | first = entry; | |
2363 | last = entry; | |
2364 | prev = entry; | |
2365 | ||
2366 | while (window_free <= min_bytes) { | |
2367 | node = rb_next(&entry->offset_index); | |
2368 | if (!node) | |
2369 | return -ENOSPC; | |
2370 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
2371 | ||
86d4a77b JB |
2372 | if (entry->bitmap) { |
2373 | if (list_empty(&entry->list)) | |
2374 | list_add_tail(&entry->list, bitmaps); | |
4e69b598 | 2375 | continue; |
86d4a77b JB |
2376 | } |
2377 | ||
4e69b598 JB |
2378 | /* |
2379 | * we haven't filled the empty size and the window is | |
2380 | * very large. reset and try again | |
2381 | */ | |
2382 | if (entry->offset - (prev->offset + prev->bytes) > max_gap || | |
2383 | entry->offset - window_start > (min_bytes * 2)) { | |
2384 | first = entry; | |
2385 | window_start = entry->offset; | |
2386 | window_free = entry->bytes; | |
2387 | last = entry; | |
2388 | max_extent = entry->bytes; | |
2389 | } else { | |
2390 | last = entry; | |
2391 | window_free += entry->bytes; | |
2392 | if (entry->bytes > max_extent) | |
2393 | max_extent = entry->bytes; | |
2394 | } | |
2395 | prev = entry; | |
2396 | } | |
2397 | ||
2398 | cluster->window_start = first->offset; | |
2399 | ||
2400 | node = &first->offset_index; | |
2401 | ||
2402 | /* | |
2403 | * now we've found our entries, pull them out of the free space | |
2404 | * cache and put them into the cluster rbtree | |
2405 | */ | |
2406 | do { | |
2407 | int ret; | |
2408 | ||
2409 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
2410 | node = rb_next(&entry->offset_index); | |
2411 | if (entry->bitmap) | |
2412 | continue; | |
2413 | ||
34d52cb6 | 2414 | rb_erase(&entry->offset_index, &ctl->free_space_offset); |
4e69b598 JB |
2415 | ret = tree_insert_offset(&cluster->root, entry->offset, |
2416 | &entry->offset_index, 0); | |
2417 | BUG_ON(ret); | |
2418 | } while (node && entry != last); | |
2419 | ||
2420 | cluster->max_size = max_extent; | |
2421 | ||
2422 | return 0; | |
2423 | } | |
2424 | ||
2425 | /* | |
2426 | * This specifically looks for bitmaps that may work in the cluster, we assume | |
2427 | * that we have already failed to find extents that will work. | |
2428 | */ | |
3de85bb9 JB |
2429 | static noinline int |
2430 | setup_cluster_bitmap(struct btrfs_block_group_cache *block_group, | |
2431 | struct btrfs_free_cluster *cluster, | |
2432 | struct list_head *bitmaps, u64 offset, u64 bytes, | |
2433 | u64 min_bytes) | |
4e69b598 | 2434 | { |
34d52cb6 | 2435 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
4e69b598 JB |
2436 | struct btrfs_free_space *entry; |
2437 | struct rb_node *node; | |
2438 | int ret = -ENOSPC; | |
2439 | ||
34d52cb6 | 2440 | if (ctl->total_bitmaps == 0) |
4e69b598 JB |
2441 | return -ENOSPC; |
2442 | ||
86d4a77b JB |
2443 | /* |
2444 | * First check our cached list of bitmaps and see if there is an entry | |
2445 | * here that will work. | |
2446 | */ | |
2447 | list_for_each_entry(entry, bitmaps, list) { | |
2448 | if (entry->bytes < min_bytes) | |
2449 | continue; | |
2450 | ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset, | |
2451 | bytes, min_bytes); | |
2452 | if (!ret) | |
2453 | return 0; | |
2454 | } | |
2455 | ||
2456 | /* | |
2457 | * If we do have entries on our list and we are here then we didn't find | |
2458 | * anything, so go ahead and get the next entry after the last entry in | |
2459 | * this list and start the search from there. | |
2460 | */ | |
2461 | if (!list_empty(bitmaps)) { | |
2462 | entry = list_entry(bitmaps->prev, struct btrfs_free_space, | |
2463 | list); | |
2464 | node = rb_next(&entry->offset_index); | |
2465 | if (!node) | |
2466 | return -ENOSPC; | |
2467 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
2468 | goto search; | |
2469 | } | |
2470 | ||
34d52cb6 | 2471 | entry = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), 0, 1); |
4e69b598 JB |
2472 | if (!entry) |
2473 | return -ENOSPC; | |
2474 | ||
86d4a77b | 2475 | search: |
4e69b598 JB |
2476 | node = &entry->offset_index; |
2477 | do { | |
2478 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
2479 | node = rb_next(&entry->offset_index); | |
2480 | if (!entry->bitmap) | |
2481 | continue; | |
2482 | if (entry->bytes < min_bytes) | |
2483 | continue; | |
2484 | ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset, | |
2485 | bytes, min_bytes); | |
2486 | } while (ret && node); | |
2487 | ||
2488 | return ret; | |
2489 | } | |
2490 | ||
fa9c0d79 CM |
2491 | /* |
2492 | * here we try to find a cluster of blocks in a block group. The goal | |
2493 | * is to find at least bytes free and up to empty_size + bytes free. | |
2494 | * We might not find them all in one contiguous area. | |
2495 | * | |
2496 | * returns zero and sets up cluster if things worked out, otherwise | |
2497 | * it returns -enospc | |
2498 | */ | |
2499 | int btrfs_find_space_cluster(struct btrfs_trans_handle *trans, | |
451d7585 | 2500 | struct btrfs_root *root, |
fa9c0d79 CM |
2501 | struct btrfs_block_group_cache *block_group, |
2502 | struct btrfs_free_cluster *cluster, | |
2503 | u64 offset, u64 bytes, u64 empty_size) | |
2504 | { | |
34d52cb6 | 2505 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
86d4a77b JB |
2506 | struct list_head bitmaps; |
2507 | struct btrfs_free_space *entry, *tmp; | |
fa9c0d79 | 2508 | u64 min_bytes; |
fa9c0d79 CM |
2509 | int ret; |
2510 | ||
2511 | /* for metadata, allow allocates with more holes */ | |
451d7585 CM |
2512 | if (btrfs_test_opt(root, SSD_SPREAD)) { |
2513 | min_bytes = bytes + empty_size; | |
2514 | } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) { | |
fa9c0d79 CM |
2515 | /* |
2516 | * we want to do larger allocations when we are | |
2517 | * flushing out the delayed refs, it helps prevent | |
2518 | * making more work as we go along. | |
2519 | */ | |
2520 | if (trans->transaction->delayed_refs.flushing) | |
2521 | min_bytes = max(bytes, (bytes + empty_size) >> 1); | |
2522 | else | |
2523 | min_bytes = max(bytes, (bytes + empty_size) >> 4); | |
2524 | } else | |
2525 | min_bytes = max(bytes, (bytes + empty_size) >> 2); | |
2526 | ||
34d52cb6 | 2527 | spin_lock(&ctl->tree_lock); |
7d0d2e8e JB |
2528 | |
2529 | /* | |
2530 | * If we know we don't have enough space to make a cluster don't even | |
2531 | * bother doing all the work to try and find one. | |
2532 | */ | |
34d52cb6 LZ |
2533 | if (ctl->free_space < min_bytes) { |
2534 | spin_unlock(&ctl->tree_lock); | |
7d0d2e8e JB |
2535 | return -ENOSPC; |
2536 | } | |
2537 | ||
fa9c0d79 CM |
2538 | spin_lock(&cluster->lock); |
2539 | ||
2540 | /* someone already found a cluster, hooray */ | |
2541 | if (cluster->block_group) { | |
2542 | ret = 0; | |
2543 | goto out; | |
2544 | } | |
fa9c0d79 | 2545 | |
86d4a77b JB |
2546 | INIT_LIST_HEAD(&bitmaps); |
2547 | ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset, | |
2548 | bytes, min_bytes); | |
4e69b598 | 2549 | if (ret) |
86d4a77b JB |
2550 | ret = setup_cluster_bitmap(block_group, cluster, &bitmaps, |
2551 | offset, bytes, min_bytes); | |
2552 | ||
2553 | /* Clear our temporary list */ | |
2554 | list_for_each_entry_safe(entry, tmp, &bitmaps, list) | |
2555 | list_del_init(&entry->list); | |
fa9c0d79 | 2556 | |
4e69b598 JB |
2557 | if (!ret) { |
2558 | atomic_inc(&block_group->count); | |
2559 | list_add_tail(&cluster->block_group_list, | |
2560 | &block_group->cluster_list); | |
2561 | cluster->block_group = block_group; | |
fa9c0d79 | 2562 | } |
fa9c0d79 CM |
2563 | out: |
2564 | spin_unlock(&cluster->lock); | |
34d52cb6 | 2565 | spin_unlock(&ctl->tree_lock); |
fa9c0d79 CM |
2566 | |
2567 | return ret; | |
2568 | } | |
2569 | ||
2570 | /* | |
2571 | * simple code to zero out a cluster | |
2572 | */ | |
2573 | void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster) | |
2574 | { | |
2575 | spin_lock_init(&cluster->lock); | |
2576 | spin_lock_init(&cluster->refill_lock); | |
6bef4d31 | 2577 | cluster->root = RB_ROOT; |
fa9c0d79 CM |
2578 | cluster->max_size = 0; |
2579 | INIT_LIST_HEAD(&cluster->block_group_list); | |
2580 | cluster->block_group = NULL; | |
2581 | } | |
2582 | ||
f7039b1d LD |
2583 | int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group, |
2584 | u64 *trimmed, u64 start, u64 end, u64 minlen) | |
2585 | { | |
34d52cb6 | 2586 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
f7039b1d LD |
2587 | struct btrfs_free_space *entry = NULL; |
2588 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
2589 | u64 bytes = 0; | |
2590 | u64 actually_trimmed; | |
2591 | int ret = 0; | |
2592 | ||
2593 | *trimmed = 0; | |
2594 | ||
2595 | while (start < end) { | |
34d52cb6 | 2596 | spin_lock(&ctl->tree_lock); |
f7039b1d | 2597 | |
34d52cb6 LZ |
2598 | if (ctl->free_space < minlen) { |
2599 | spin_unlock(&ctl->tree_lock); | |
f7039b1d LD |
2600 | break; |
2601 | } | |
2602 | ||
34d52cb6 | 2603 | entry = tree_search_offset(ctl, start, 0, 1); |
f7039b1d | 2604 | if (!entry) |
34d52cb6 LZ |
2605 | entry = tree_search_offset(ctl, |
2606 | offset_to_bitmap(ctl, start), | |
f7039b1d LD |
2607 | 1, 1); |
2608 | ||
2609 | if (!entry || entry->offset >= end) { | |
34d52cb6 | 2610 | spin_unlock(&ctl->tree_lock); |
f7039b1d LD |
2611 | break; |
2612 | } | |
2613 | ||
2614 | if (entry->bitmap) { | |
34d52cb6 | 2615 | ret = search_bitmap(ctl, entry, &start, &bytes); |
f7039b1d LD |
2616 | if (!ret) { |
2617 | if (start >= end) { | |
34d52cb6 | 2618 | spin_unlock(&ctl->tree_lock); |
f7039b1d LD |
2619 | break; |
2620 | } | |
2621 | bytes = min(bytes, end - start); | |
34d52cb6 | 2622 | bitmap_clear_bits(ctl, entry, start, bytes); |
f7039b1d | 2623 | if (entry->bytes == 0) |
34d52cb6 | 2624 | free_bitmap(ctl, entry); |
f7039b1d LD |
2625 | } else { |
2626 | start = entry->offset + BITS_PER_BITMAP * | |
2627 | block_group->sectorsize; | |
34d52cb6 | 2628 | spin_unlock(&ctl->tree_lock); |
f7039b1d LD |
2629 | ret = 0; |
2630 | continue; | |
2631 | } | |
2632 | } else { | |
2633 | start = entry->offset; | |
2634 | bytes = min(entry->bytes, end - start); | |
34d52cb6 | 2635 | unlink_free_space(ctl, entry); |
f789b684 | 2636 | kmem_cache_free(btrfs_free_space_cachep, entry); |
f7039b1d LD |
2637 | } |
2638 | ||
34d52cb6 | 2639 | spin_unlock(&ctl->tree_lock); |
f7039b1d LD |
2640 | |
2641 | if (bytes >= minlen) { | |
fb25e914 JB |
2642 | struct btrfs_space_info *space_info; |
2643 | int update = 0; | |
2644 | ||
2645 | space_info = block_group->space_info; | |
2646 | spin_lock(&space_info->lock); | |
2647 | spin_lock(&block_group->lock); | |
2648 | if (!block_group->ro) { | |
2649 | block_group->reserved += bytes; | |
2650 | space_info->bytes_reserved += bytes; | |
2651 | update = 1; | |
2652 | } | |
2653 | spin_unlock(&block_group->lock); | |
2654 | spin_unlock(&space_info->lock); | |
f7039b1d LD |
2655 | |
2656 | ret = btrfs_error_discard_extent(fs_info->extent_root, | |
2657 | start, | |
2658 | bytes, | |
2659 | &actually_trimmed); | |
2660 | ||
34d52cb6 | 2661 | btrfs_add_free_space(block_group, start, bytes); |
fb25e914 JB |
2662 | if (update) { |
2663 | spin_lock(&space_info->lock); | |
2664 | spin_lock(&block_group->lock); | |
2665 | if (block_group->ro) | |
2666 | space_info->bytes_readonly += bytes; | |
2667 | block_group->reserved -= bytes; | |
2668 | space_info->bytes_reserved -= bytes; | |
2669 | spin_unlock(&space_info->lock); | |
2670 | spin_unlock(&block_group->lock); | |
2671 | } | |
f7039b1d LD |
2672 | |
2673 | if (ret) | |
2674 | break; | |
2675 | *trimmed += actually_trimmed; | |
2676 | } | |
2677 | start += bytes; | |
2678 | bytes = 0; | |
2679 | ||
2680 | if (fatal_signal_pending(current)) { | |
2681 | ret = -ERESTARTSYS; | |
2682 | break; | |
2683 | } | |
2684 | ||
2685 | cond_resched(); | |
2686 | } | |
2687 | ||
2688 | return ret; | |
2689 | } | |
581bb050 LZ |
2690 | |
2691 | /* | |
2692 | * Find the left-most item in the cache tree, and then return the | |
2693 | * smallest inode number in the item. | |
2694 | * | |
2695 | * Note: the returned inode number may not be the smallest one in | |
2696 | * the tree, if the left-most item is a bitmap. | |
2697 | */ | |
2698 | u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root) | |
2699 | { | |
2700 | struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl; | |
2701 | struct btrfs_free_space *entry = NULL; | |
2702 | u64 ino = 0; | |
2703 | ||
2704 | spin_lock(&ctl->tree_lock); | |
2705 | ||
2706 | if (RB_EMPTY_ROOT(&ctl->free_space_offset)) | |
2707 | goto out; | |
2708 | ||
2709 | entry = rb_entry(rb_first(&ctl->free_space_offset), | |
2710 | struct btrfs_free_space, offset_index); | |
2711 | ||
2712 | if (!entry->bitmap) { | |
2713 | ino = entry->offset; | |
2714 | ||
2715 | unlink_free_space(ctl, entry); | |
2716 | entry->offset++; | |
2717 | entry->bytes--; | |
2718 | if (!entry->bytes) | |
2719 | kmem_cache_free(btrfs_free_space_cachep, entry); | |
2720 | else | |
2721 | link_free_space(ctl, entry); | |
2722 | } else { | |
2723 | u64 offset = 0; | |
2724 | u64 count = 1; | |
2725 | int ret; | |
2726 | ||
2727 | ret = search_bitmap(ctl, entry, &offset, &count); | |
2728 | BUG_ON(ret); | |
2729 | ||
2730 | ino = offset; | |
2731 | bitmap_clear_bits(ctl, entry, offset, 1); | |
2732 | if (entry->bytes == 0) | |
2733 | free_bitmap(ctl, entry); | |
2734 | } | |
2735 | out: | |
2736 | spin_unlock(&ctl->tree_lock); | |
2737 | ||
2738 | return ino; | |
2739 | } | |
82d5902d LZ |
2740 | |
2741 | struct inode *lookup_free_ino_inode(struct btrfs_root *root, | |
2742 | struct btrfs_path *path) | |
2743 | { | |
2744 | struct inode *inode = NULL; | |
2745 | ||
2746 | spin_lock(&root->cache_lock); | |
2747 | if (root->cache_inode) | |
2748 | inode = igrab(root->cache_inode); | |
2749 | spin_unlock(&root->cache_lock); | |
2750 | if (inode) | |
2751 | return inode; | |
2752 | ||
2753 | inode = __lookup_free_space_inode(root, path, 0); | |
2754 | if (IS_ERR(inode)) | |
2755 | return inode; | |
2756 | ||
2757 | spin_lock(&root->cache_lock); | |
7841cb28 | 2758 | if (!btrfs_fs_closing(root->fs_info)) |
82d5902d LZ |
2759 | root->cache_inode = igrab(inode); |
2760 | spin_unlock(&root->cache_lock); | |
2761 | ||
2762 | return inode; | |
2763 | } | |
2764 | ||
2765 | int create_free_ino_inode(struct btrfs_root *root, | |
2766 | struct btrfs_trans_handle *trans, | |
2767 | struct btrfs_path *path) | |
2768 | { | |
2769 | return __create_free_space_inode(root, trans, path, | |
2770 | BTRFS_FREE_INO_OBJECTID, 0); | |
2771 | } | |
2772 | ||
2773 | int load_free_ino_cache(struct btrfs_fs_info *fs_info, struct btrfs_root *root) | |
2774 | { | |
2775 | struct btrfs_free_space_ctl *ctl = root->free_ino_ctl; | |
2776 | struct btrfs_path *path; | |
2777 | struct inode *inode; | |
2778 | int ret = 0; | |
2779 | u64 root_gen = btrfs_root_generation(&root->root_item); | |
2780 | ||
4b9465cb CM |
2781 | if (!btrfs_test_opt(root, INODE_MAP_CACHE)) |
2782 | return 0; | |
2783 | ||
82d5902d LZ |
2784 | /* |
2785 | * If we're unmounting then just return, since this does a search on the | |
2786 | * normal root and not the commit root and we could deadlock. | |
2787 | */ | |
7841cb28 | 2788 | if (btrfs_fs_closing(fs_info)) |
82d5902d LZ |
2789 | return 0; |
2790 | ||
2791 | path = btrfs_alloc_path(); | |
2792 | if (!path) | |
2793 | return 0; | |
2794 | ||
2795 | inode = lookup_free_ino_inode(root, path); | |
2796 | if (IS_ERR(inode)) | |
2797 | goto out; | |
2798 | ||
2799 | if (root_gen != BTRFS_I(inode)->generation) | |
2800 | goto out_put; | |
2801 | ||
2802 | ret = __load_free_space_cache(root, inode, ctl, path, 0); | |
2803 | ||
2804 | if (ret < 0) | |
2805 | printk(KERN_ERR "btrfs: failed to load free ino cache for " | |
2806 | "root %llu\n", root->root_key.objectid); | |
2807 | out_put: | |
2808 | iput(inode); | |
2809 | out: | |
2810 | btrfs_free_path(path); | |
2811 | return ret; | |
2812 | } | |
2813 | ||
2814 | int btrfs_write_out_ino_cache(struct btrfs_root *root, | |
2815 | struct btrfs_trans_handle *trans, | |
2816 | struct btrfs_path *path) | |
2817 | { | |
2818 | struct btrfs_free_space_ctl *ctl = root->free_ino_ctl; | |
2819 | struct inode *inode; | |
2820 | int ret; | |
2821 | ||
4b9465cb CM |
2822 | if (!btrfs_test_opt(root, INODE_MAP_CACHE)) |
2823 | return 0; | |
2824 | ||
82d5902d LZ |
2825 | inode = lookup_free_ino_inode(root, path); |
2826 | if (IS_ERR(inode)) | |
2827 | return 0; | |
2828 | ||
2829 | ret = __btrfs_write_out_cache(root, inode, ctl, NULL, trans, path, 0); | |
c09544e0 JB |
2830 | if (ret) { |
2831 | btrfs_delalloc_release_metadata(inode, inode->i_size); | |
2832 | #ifdef DEBUG | |
82d5902d LZ |
2833 | printk(KERN_ERR "btrfs: failed to write free ino cache " |
2834 | "for root %llu\n", root->root_key.objectid); | |
c09544e0 JB |
2835 | #endif |
2836 | } | |
82d5902d LZ |
2837 | |
2838 | iput(inode); | |
2839 | return ret; | |
2840 | } |