]>
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" |
04216820 | 30 | #include "volumes.h" |
fa9c0d79 | 31 | |
96303081 JB |
32 | #define BITS_PER_BITMAP (PAGE_CACHE_SIZE * 8) |
33 | #define MAX_CACHE_BYTES_PER_GIG (32 * 1024) | |
0f9dd46c | 34 | |
55507ce3 FM |
35 | struct btrfs_trim_range { |
36 | u64 start; | |
37 | u64 bytes; | |
38 | struct list_head list; | |
39 | }; | |
40 | ||
34d52cb6 | 41 | static int link_free_space(struct btrfs_free_space_ctl *ctl, |
0cb59c99 | 42 | struct btrfs_free_space *info); |
cd023e7b JB |
43 | static void unlink_free_space(struct btrfs_free_space_ctl *ctl, |
44 | struct btrfs_free_space *info); | |
0cb59c99 | 45 | |
0414efae LZ |
46 | static struct inode *__lookup_free_space_inode(struct btrfs_root *root, |
47 | struct btrfs_path *path, | |
48 | u64 offset) | |
0af3d00b JB |
49 | { |
50 | struct btrfs_key key; | |
51 | struct btrfs_key location; | |
52 | struct btrfs_disk_key disk_key; | |
53 | struct btrfs_free_space_header *header; | |
54 | struct extent_buffer *leaf; | |
55 | struct inode *inode = NULL; | |
56 | int ret; | |
57 | ||
0af3d00b | 58 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; |
0414efae | 59 | key.offset = offset; |
0af3d00b JB |
60 | key.type = 0; |
61 | ||
62 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
63 | if (ret < 0) | |
64 | return ERR_PTR(ret); | |
65 | if (ret > 0) { | |
b3b4aa74 | 66 | btrfs_release_path(path); |
0af3d00b JB |
67 | return ERR_PTR(-ENOENT); |
68 | } | |
69 | ||
70 | leaf = path->nodes[0]; | |
71 | header = btrfs_item_ptr(leaf, path->slots[0], | |
72 | struct btrfs_free_space_header); | |
73 | btrfs_free_space_key(leaf, header, &disk_key); | |
74 | btrfs_disk_key_to_cpu(&location, &disk_key); | |
b3b4aa74 | 75 | btrfs_release_path(path); |
0af3d00b JB |
76 | |
77 | inode = btrfs_iget(root->fs_info->sb, &location, root, NULL); | |
78 | if (!inode) | |
79 | return ERR_PTR(-ENOENT); | |
80 | if (IS_ERR(inode)) | |
81 | return inode; | |
82 | if (is_bad_inode(inode)) { | |
83 | iput(inode); | |
84 | return ERR_PTR(-ENOENT); | |
85 | } | |
86 | ||
528c0327 AV |
87 | mapping_set_gfp_mask(inode->i_mapping, |
88 | mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS); | |
adae52b9 | 89 | |
0414efae LZ |
90 | return inode; |
91 | } | |
92 | ||
93 | struct inode *lookup_free_space_inode(struct btrfs_root *root, | |
94 | struct btrfs_block_group_cache | |
95 | *block_group, struct btrfs_path *path) | |
96 | { | |
97 | struct inode *inode = NULL; | |
5b0e95bf | 98 | u32 flags = BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW; |
0414efae LZ |
99 | |
100 | spin_lock(&block_group->lock); | |
101 | if (block_group->inode) | |
102 | inode = igrab(block_group->inode); | |
103 | spin_unlock(&block_group->lock); | |
104 | if (inode) | |
105 | return inode; | |
106 | ||
107 | inode = __lookup_free_space_inode(root, path, | |
108 | block_group->key.objectid); | |
109 | if (IS_ERR(inode)) | |
110 | return inode; | |
111 | ||
0af3d00b | 112 | spin_lock(&block_group->lock); |
5b0e95bf | 113 | if (!((BTRFS_I(inode)->flags & flags) == flags)) { |
c2cf52eb SK |
114 | btrfs_info(root->fs_info, |
115 | "Old style space inode found, converting."); | |
5b0e95bf JB |
116 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM | |
117 | BTRFS_INODE_NODATACOW; | |
2f356126 JB |
118 | block_group->disk_cache_state = BTRFS_DC_CLEAR; |
119 | } | |
120 | ||
300e4f8a | 121 | if (!block_group->iref) { |
0af3d00b JB |
122 | block_group->inode = igrab(inode); |
123 | block_group->iref = 1; | |
124 | } | |
125 | spin_unlock(&block_group->lock); | |
126 | ||
127 | return inode; | |
128 | } | |
129 | ||
48a3b636 ES |
130 | static int __create_free_space_inode(struct btrfs_root *root, |
131 | struct btrfs_trans_handle *trans, | |
132 | struct btrfs_path *path, | |
133 | u64 ino, u64 offset) | |
0af3d00b JB |
134 | { |
135 | struct btrfs_key key; | |
136 | struct btrfs_disk_key disk_key; | |
137 | struct btrfs_free_space_header *header; | |
138 | struct btrfs_inode_item *inode_item; | |
139 | struct extent_buffer *leaf; | |
5b0e95bf | 140 | u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC; |
0af3d00b JB |
141 | int ret; |
142 | ||
0414efae | 143 | ret = btrfs_insert_empty_inode(trans, root, path, ino); |
0af3d00b JB |
144 | if (ret) |
145 | return ret; | |
146 | ||
5b0e95bf JB |
147 | /* We inline crc's for the free disk space cache */ |
148 | if (ino != BTRFS_FREE_INO_OBJECTID) | |
149 | flags |= BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW; | |
150 | ||
0af3d00b JB |
151 | leaf = path->nodes[0]; |
152 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
153 | struct btrfs_inode_item); | |
154 | btrfs_item_key(leaf, &disk_key, path->slots[0]); | |
155 | memset_extent_buffer(leaf, 0, (unsigned long)inode_item, | |
156 | sizeof(*inode_item)); | |
157 | btrfs_set_inode_generation(leaf, inode_item, trans->transid); | |
158 | btrfs_set_inode_size(leaf, inode_item, 0); | |
159 | btrfs_set_inode_nbytes(leaf, inode_item, 0); | |
160 | btrfs_set_inode_uid(leaf, inode_item, 0); | |
161 | btrfs_set_inode_gid(leaf, inode_item, 0); | |
162 | btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600); | |
5b0e95bf | 163 | btrfs_set_inode_flags(leaf, inode_item, flags); |
0af3d00b JB |
164 | btrfs_set_inode_nlink(leaf, inode_item, 1); |
165 | btrfs_set_inode_transid(leaf, inode_item, trans->transid); | |
0414efae | 166 | btrfs_set_inode_block_group(leaf, inode_item, offset); |
0af3d00b | 167 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 168 | btrfs_release_path(path); |
0af3d00b JB |
169 | |
170 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
0414efae | 171 | key.offset = offset; |
0af3d00b JB |
172 | key.type = 0; |
173 | ||
174 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
175 | sizeof(struct btrfs_free_space_header)); | |
176 | if (ret < 0) { | |
b3b4aa74 | 177 | btrfs_release_path(path); |
0af3d00b JB |
178 | return ret; |
179 | } | |
180 | leaf = path->nodes[0]; | |
181 | header = btrfs_item_ptr(leaf, path->slots[0], | |
182 | struct btrfs_free_space_header); | |
183 | memset_extent_buffer(leaf, 0, (unsigned long)header, sizeof(*header)); | |
184 | btrfs_set_free_space_key(leaf, header, &disk_key); | |
185 | btrfs_mark_buffer_dirty(leaf); | |
b3b4aa74 | 186 | btrfs_release_path(path); |
0af3d00b JB |
187 | |
188 | return 0; | |
189 | } | |
190 | ||
0414efae LZ |
191 | int create_free_space_inode(struct btrfs_root *root, |
192 | struct btrfs_trans_handle *trans, | |
193 | struct btrfs_block_group_cache *block_group, | |
194 | struct btrfs_path *path) | |
195 | { | |
196 | int ret; | |
197 | u64 ino; | |
198 | ||
199 | ret = btrfs_find_free_objectid(root, &ino); | |
200 | if (ret < 0) | |
201 | return ret; | |
202 | ||
203 | return __create_free_space_inode(root, trans, path, ino, | |
204 | block_group->key.objectid); | |
205 | } | |
206 | ||
7b61cd92 MX |
207 | int btrfs_check_trunc_cache_free_space(struct btrfs_root *root, |
208 | struct btrfs_block_rsv *rsv) | |
0af3d00b | 209 | { |
c8174313 | 210 | u64 needed_bytes; |
7b61cd92 | 211 | int ret; |
c8174313 JB |
212 | |
213 | /* 1 for slack space, 1 for updating the inode */ | |
214 | needed_bytes = btrfs_calc_trunc_metadata_size(root, 1) + | |
215 | btrfs_calc_trans_metadata_size(root, 1); | |
216 | ||
7b61cd92 MX |
217 | spin_lock(&rsv->lock); |
218 | if (rsv->reserved < needed_bytes) | |
219 | ret = -ENOSPC; | |
220 | else | |
221 | ret = 0; | |
222 | spin_unlock(&rsv->lock); | |
4b286cd1 | 223 | return ret; |
7b61cd92 MX |
224 | } |
225 | ||
226 | int btrfs_truncate_free_space_cache(struct btrfs_root *root, | |
227 | struct btrfs_trans_handle *trans, | |
7b61cd92 MX |
228 | struct inode *inode) |
229 | { | |
7b61cd92 | 230 | int ret = 0; |
0af3d00b | 231 | |
0af3d00b | 232 | btrfs_i_size_write(inode, 0); |
7caef267 | 233 | truncate_pagecache(inode, 0); |
0af3d00b JB |
234 | |
235 | /* | |
236 | * We don't need an orphan item because truncating the free space cache | |
237 | * will never be split across transactions. | |
28ed1345 CM |
238 | * We don't need to check for -EAGAIN because we're a free space |
239 | * cache inode | |
0af3d00b JB |
240 | */ |
241 | ret = btrfs_truncate_inode_items(trans, root, inode, | |
242 | 0, BTRFS_EXTENT_DATA_KEY); | |
243 | if (ret) { | |
79787eaa | 244 | btrfs_abort_transaction(trans, root, ret); |
0af3d00b JB |
245 | return ret; |
246 | } | |
247 | ||
82d5902d | 248 | ret = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
249 | if (ret) |
250 | btrfs_abort_transaction(trans, root, ret); | |
c8174313 | 251 | |
82d5902d | 252 | return ret; |
0af3d00b JB |
253 | } |
254 | ||
9d66e233 JB |
255 | static int readahead_cache(struct inode *inode) |
256 | { | |
257 | struct file_ra_state *ra; | |
258 | unsigned long last_index; | |
259 | ||
260 | ra = kzalloc(sizeof(*ra), GFP_NOFS); | |
261 | if (!ra) | |
262 | return -ENOMEM; | |
263 | ||
264 | file_ra_state_init(ra, inode->i_mapping); | |
265 | last_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT; | |
266 | ||
267 | page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index); | |
268 | ||
269 | kfree(ra); | |
270 | ||
271 | return 0; | |
272 | } | |
273 | ||
a67509c3 JB |
274 | struct io_ctl { |
275 | void *cur, *orig; | |
276 | struct page *page; | |
277 | struct page **pages; | |
278 | struct btrfs_root *root; | |
279 | unsigned long size; | |
280 | int index; | |
281 | int num_pages; | |
5b0e95bf | 282 | unsigned check_crcs:1; |
a67509c3 JB |
283 | }; |
284 | ||
285 | static int io_ctl_init(struct io_ctl *io_ctl, struct inode *inode, | |
5349d6c3 | 286 | struct btrfs_root *root, int write) |
a67509c3 | 287 | { |
5349d6c3 MX |
288 | int num_pages; |
289 | int check_crcs = 0; | |
290 | ||
ed6078f7 | 291 | num_pages = DIV_ROUND_UP(i_size_read(inode), PAGE_CACHE_SIZE); |
5349d6c3 MX |
292 | |
293 | if (btrfs_ino(inode) != BTRFS_FREE_INO_OBJECTID) | |
294 | check_crcs = 1; | |
295 | ||
296 | /* Make sure we can fit our crcs into the first page */ | |
297 | if (write && check_crcs && | |
298 | (num_pages * sizeof(u32)) >= PAGE_CACHE_SIZE) | |
299 | return -ENOSPC; | |
300 | ||
a67509c3 | 301 | memset(io_ctl, 0, sizeof(struct io_ctl)); |
5349d6c3 | 302 | |
31e818fe | 303 | io_ctl->pages = kcalloc(num_pages, sizeof(struct page *), GFP_NOFS); |
a67509c3 JB |
304 | if (!io_ctl->pages) |
305 | return -ENOMEM; | |
5349d6c3 MX |
306 | |
307 | io_ctl->num_pages = num_pages; | |
a67509c3 | 308 | io_ctl->root = root; |
5349d6c3 MX |
309 | io_ctl->check_crcs = check_crcs; |
310 | ||
a67509c3 JB |
311 | return 0; |
312 | } | |
313 | ||
314 | static void io_ctl_free(struct io_ctl *io_ctl) | |
315 | { | |
316 | kfree(io_ctl->pages); | |
317 | } | |
318 | ||
319 | static void io_ctl_unmap_page(struct io_ctl *io_ctl) | |
320 | { | |
321 | if (io_ctl->cur) { | |
322 | kunmap(io_ctl->page); | |
323 | io_ctl->cur = NULL; | |
324 | io_ctl->orig = NULL; | |
325 | } | |
326 | } | |
327 | ||
328 | static void io_ctl_map_page(struct io_ctl *io_ctl, int clear) | |
329 | { | |
b12d6869 | 330 | ASSERT(io_ctl->index < io_ctl->num_pages); |
a67509c3 JB |
331 | io_ctl->page = io_ctl->pages[io_ctl->index++]; |
332 | io_ctl->cur = kmap(io_ctl->page); | |
333 | io_ctl->orig = io_ctl->cur; | |
334 | io_ctl->size = PAGE_CACHE_SIZE; | |
335 | if (clear) | |
336 | memset(io_ctl->cur, 0, PAGE_CACHE_SIZE); | |
337 | } | |
338 | ||
339 | static void io_ctl_drop_pages(struct io_ctl *io_ctl) | |
340 | { | |
341 | int i; | |
342 | ||
343 | io_ctl_unmap_page(io_ctl); | |
344 | ||
345 | for (i = 0; i < io_ctl->num_pages; i++) { | |
a1ee5a45 LZ |
346 | if (io_ctl->pages[i]) { |
347 | ClearPageChecked(io_ctl->pages[i]); | |
348 | unlock_page(io_ctl->pages[i]); | |
349 | page_cache_release(io_ctl->pages[i]); | |
350 | } | |
a67509c3 JB |
351 | } |
352 | } | |
353 | ||
354 | static int io_ctl_prepare_pages(struct io_ctl *io_ctl, struct inode *inode, | |
355 | int uptodate) | |
356 | { | |
357 | struct page *page; | |
358 | gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); | |
359 | int i; | |
360 | ||
361 | for (i = 0; i < io_ctl->num_pages; i++) { | |
362 | page = find_or_create_page(inode->i_mapping, i, mask); | |
363 | if (!page) { | |
364 | io_ctl_drop_pages(io_ctl); | |
365 | return -ENOMEM; | |
366 | } | |
367 | io_ctl->pages[i] = page; | |
368 | if (uptodate && !PageUptodate(page)) { | |
369 | btrfs_readpage(NULL, page); | |
370 | lock_page(page); | |
371 | if (!PageUptodate(page)) { | |
efe120a0 FH |
372 | btrfs_err(BTRFS_I(inode)->root->fs_info, |
373 | "error reading free space cache"); | |
a67509c3 JB |
374 | io_ctl_drop_pages(io_ctl); |
375 | return -EIO; | |
376 | } | |
377 | } | |
378 | } | |
379 | ||
f7d61dcd JB |
380 | for (i = 0; i < io_ctl->num_pages; i++) { |
381 | clear_page_dirty_for_io(io_ctl->pages[i]); | |
382 | set_page_extent_mapped(io_ctl->pages[i]); | |
383 | } | |
384 | ||
a67509c3 JB |
385 | return 0; |
386 | } | |
387 | ||
388 | static void io_ctl_set_generation(struct io_ctl *io_ctl, u64 generation) | |
389 | { | |
528c0327 | 390 | __le64 *val; |
a67509c3 JB |
391 | |
392 | io_ctl_map_page(io_ctl, 1); | |
393 | ||
394 | /* | |
5b0e95bf JB |
395 | * Skip the csum areas. If we don't check crcs then we just have a |
396 | * 64bit chunk at the front of the first page. | |
a67509c3 | 397 | */ |
5b0e95bf JB |
398 | if (io_ctl->check_crcs) { |
399 | io_ctl->cur += (sizeof(u32) * io_ctl->num_pages); | |
400 | io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages); | |
401 | } else { | |
402 | io_ctl->cur += sizeof(u64); | |
403 | io_ctl->size -= sizeof(u64) * 2; | |
404 | } | |
a67509c3 JB |
405 | |
406 | val = io_ctl->cur; | |
407 | *val = cpu_to_le64(generation); | |
408 | io_ctl->cur += sizeof(u64); | |
a67509c3 JB |
409 | } |
410 | ||
411 | static int io_ctl_check_generation(struct io_ctl *io_ctl, u64 generation) | |
412 | { | |
528c0327 | 413 | __le64 *gen; |
a67509c3 | 414 | |
5b0e95bf JB |
415 | /* |
416 | * Skip the crc area. If we don't check crcs then we just have a 64bit | |
417 | * chunk at the front of the first page. | |
418 | */ | |
419 | if (io_ctl->check_crcs) { | |
420 | io_ctl->cur += sizeof(u32) * io_ctl->num_pages; | |
421 | io_ctl->size -= sizeof(u64) + | |
422 | (sizeof(u32) * io_ctl->num_pages); | |
423 | } else { | |
424 | io_ctl->cur += sizeof(u64); | |
425 | io_ctl->size -= sizeof(u64) * 2; | |
426 | } | |
a67509c3 | 427 | |
a67509c3 JB |
428 | gen = io_ctl->cur; |
429 | if (le64_to_cpu(*gen) != generation) { | |
efe120a0 | 430 | printk_ratelimited(KERN_ERR "BTRFS: space cache generation " |
a67509c3 JB |
431 | "(%Lu) does not match inode (%Lu)\n", *gen, |
432 | generation); | |
433 | io_ctl_unmap_page(io_ctl); | |
434 | return -EIO; | |
435 | } | |
436 | io_ctl->cur += sizeof(u64); | |
5b0e95bf JB |
437 | return 0; |
438 | } | |
439 | ||
440 | static void io_ctl_set_crc(struct io_ctl *io_ctl, int index) | |
441 | { | |
442 | u32 *tmp; | |
443 | u32 crc = ~(u32)0; | |
444 | unsigned offset = 0; | |
445 | ||
446 | if (!io_ctl->check_crcs) { | |
447 | io_ctl_unmap_page(io_ctl); | |
448 | return; | |
449 | } | |
450 | ||
451 | if (index == 0) | |
cb54f257 | 452 | offset = sizeof(u32) * io_ctl->num_pages; |
5b0e95bf | 453 | |
b0496686 | 454 | crc = btrfs_csum_data(io_ctl->orig + offset, crc, |
5b0e95bf JB |
455 | PAGE_CACHE_SIZE - offset); |
456 | btrfs_csum_final(crc, (char *)&crc); | |
457 | io_ctl_unmap_page(io_ctl); | |
458 | tmp = kmap(io_ctl->pages[0]); | |
459 | tmp += index; | |
460 | *tmp = crc; | |
461 | kunmap(io_ctl->pages[0]); | |
462 | } | |
463 | ||
464 | static int io_ctl_check_crc(struct io_ctl *io_ctl, int index) | |
465 | { | |
466 | u32 *tmp, val; | |
467 | u32 crc = ~(u32)0; | |
468 | unsigned offset = 0; | |
469 | ||
470 | if (!io_ctl->check_crcs) { | |
471 | io_ctl_map_page(io_ctl, 0); | |
472 | return 0; | |
473 | } | |
474 | ||
475 | if (index == 0) | |
476 | offset = sizeof(u32) * io_ctl->num_pages; | |
477 | ||
478 | tmp = kmap(io_ctl->pages[0]); | |
479 | tmp += index; | |
480 | val = *tmp; | |
481 | kunmap(io_ctl->pages[0]); | |
482 | ||
483 | io_ctl_map_page(io_ctl, 0); | |
b0496686 | 484 | crc = btrfs_csum_data(io_ctl->orig + offset, crc, |
5b0e95bf JB |
485 | PAGE_CACHE_SIZE - offset); |
486 | btrfs_csum_final(crc, (char *)&crc); | |
487 | if (val != crc) { | |
efe120a0 | 488 | printk_ratelimited(KERN_ERR "BTRFS: csum mismatch on free " |
5b0e95bf JB |
489 | "space cache\n"); |
490 | io_ctl_unmap_page(io_ctl); | |
491 | return -EIO; | |
492 | } | |
493 | ||
a67509c3 JB |
494 | return 0; |
495 | } | |
496 | ||
497 | static int io_ctl_add_entry(struct io_ctl *io_ctl, u64 offset, u64 bytes, | |
498 | void *bitmap) | |
499 | { | |
500 | struct btrfs_free_space_entry *entry; | |
501 | ||
502 | if (!io_ctl->cur) | |
503 | return -ENOSPC; | |
504 | ||
505 | entry = io_ctl->cur; | |
506 | entry->offset = cpu_to_le64(offset); | |
507 | entry->bytes = cpu_to_le64(bytes); | |
508 | entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP : | |
509 | BTRFS_FREE_SPACE_EXTENT; | |
510 | io_ctl->cur += sizeof(struct btrfs_free_space_entry); | |
511 | io_ctl->size -= sizeof(struct btrfs_free_space_entry); | |
512 | ||
513 | if (io_ctl->size >= sizeof(struct btrfs_free_space_entry)) | |
514 | return 0; | |
515 | ||
5b0e95bf | 516 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); |
a67509c3 JB |
517 | |
518 | /* No more pages to map */ | |
519 | if (io_ctl->index >= io_ctl->num_pages) | |
520 | return 0; | |
521 | ||
522 | /* map the next page */ | |
523 | io_ctl_map_page(io_ctl, 1); | |
524 | return 0; | |
525 | } | |
526 | ||
527 | static int io_ctl_add_bitmap(struct io_ctl *io_ctl, void *bitmap) | |
528 | { | |
529 | if (!io_ctl->cur) | |
530 | return -ENOSPC; | |
531 | ||
532 | /* | |
533 | * If we aren't at the start of the current page, unmap this one and | |
534 | * map the next one if there is any left. | |
535 | */ | |
536 | if (io_ctl->cur != io_ctl->orig) { | |
5b0e95bf | 537 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); |
a67509c3 JB |
538 | if (io_ctl->index >= io_ctl->num_pages) |
539 | return -ENOSPC; | |
540 | io_ctl_map_page(io_ctl, 0); | |
541 | } | |
542 | ||
543 | memcpy(io_ctl->cur, bitmap, PAGE_CACHE_SIZE); | |
5b0e95bf | 544 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); |
a67509c3 JB |
545 | if (io_ctl->index < io_ctl->num_pages) |
546 | io_ctl_map_page(io_ctl, 0); | |
547 | return 0; | |
548 | } | |
549 | ||
550 | static void io_ctl_zero_remaining_pages(struct io_ctl *io_ctl) | |
551 | { | |
5b0e95bf JB |
552 | /* |
553 | * If we're not on the boundary we know we've modified the page and we | |
554 | * need to crc the page. | |
555 | */ | |
556 | if (io_ctl->cur != io_ctl->orig) | |
557 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); | |
558 | else | |
559 | io_ctl_unmap_page(io_ctl); | |
a67509c3 JB |
560 | |
561 | while (io_ctl->index < io_ctl->num_pages) { | |
562 | io_ctl_map_page(io_ctl, 1); | |
5b0e95bf | 563 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); |
a67509c3 JB |
564 | } |
565 | } | |
566 | ||
5b0e95bf JB |
567 | static int io_ctl_read_entry(struct io_ctl *io_ctl, |
568 | struct btrfs_free_space *entry, u8 *type) | |
a67509c3 JB |
569 | { |
570 | struct btrfs_free_space_entry *e; | |
2f120c05 JB |
571 | int ret; |
572 | ||
573 | if (!io_ctl->cur) { | |
574 | ret = io_ctl_check_crc(io_ctl, io_ctl->index); | |
575 | if (ret) | |
576 | return ret; | |
577 | } | |
a67509c3 JB |
578 | |
579 | e = io_ctl->cur; | |
580 | entry->offset = le64_to_cpu(e->offset); | |
581 | entry->bytes = le64_to_cpu(e->bytes); | |
5b0e95bf | 582 | *type = e->type; |
a67509c3 JB |
583 | io_ctl->cur += sizeof(struct btrfs_free_space_entry); |
584 | io_ctl->size -= sizeof(struct btrfs_free_space_entry); | |
585 | ||
586 | if (io_ctl->size >= sizeof(struct btrfs_free_space_entry)) | |
5b0e95bf | 587 | return 0; |
a67509c3 JB |
588 | |
589 | io_ctl_unmap_page(io_ctl); | |
590 | ||
2f120c05 | 591 | return 0; |
a67509c3 JB |
592 | } |
593 | ||
5b0e95bf JB |
594 | static int io_ctl_read_bitmap(struct io_ctl *io_ctl, |
595 | struct btrfs_free_space *entry) | |
a67509c3 | 596 | { |
5b0e95bf JB |
597 | int ret; |
598 | ||
5b0e95bf JB |
599 | ret = io_ctl_check_crc(io_ctl, io_ctl->index); |
600 | if (ret) | |
601 | return ret; | |
602 | ||
a67509c3 JB |
603 | memcpy(entry->bitmap, io_ctl->cur, PAGE_CACHE_SIZE); |
604 | io_ctl_unmap_page(io_ctl); | |
5b0e95bf JB |
605 | |
606 | return 0; | |
a67509c3 JB |
607 | } |
608 | ||
cd023e7b JB |
609 | /* |
610 | * Since we attach pinned extents after the fact we can have contiguous sections | |
611 | * of free space that are split up in entries. This poses a problem with the | |
612 | * tree logging stuff since it could have allocated across what appears to be 2 | |
613 | * entries since we would have merged the entries when adding the pinned extents | |
614 | * back to the free space cache. So run through the space cache that we just | |
615 | * loaded and merge contiguous entries. This will make the log replay stuff not | |
616 | * blow up and it will make for nicer allocator behavior. | |
617 | */ | |
618 | static void merge_space_tree(struct btrfs_free_space_ctl *ctl) | |
619 | { | |
620 | struct btrfs_free_space *e, *prev = NULL; | |
621 | struct rb_node *n; | |
622 | ||
623 | again: | |
624 | spin_lock(&ctl->tree_lock); | |
625 | for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) { | |
626 | e = rb_entry(n, struct btrfs_free_space, offset_index); | |
627 | if (!prev) | |
628 | goto next; | |
629 | if (e->bitmap || prev->bitmap) | |
630 | goto next; | |
631 | if (prev->offset + prev->bytes == e->offset) { | |
632 | unlink_free_space(ctl, prev); | |
633 | unlink_free_space(ctl, e); | |
634 | prev->bytes += e->bytes; | |
635 | kmem_cache_free(btrfs_free_space_cachep, e); | |
636 | link_free_space(ctl, prev); | |
637 | prev = NULL; | |
638 | spin_unlock(&ctl->tree_lock); | |
639 | goto again; | |
640 | } | |
641 | next: | |
642 | prev = e; | |
643 | } | |
644 | spin_unlock(&ctl->tree_lock); | |
645 | } | |
646 | ||
48a3b636 ES |
647 | static int __load_free_space_cache(struct btrfs_root *root, struct inode *inode, |
648 | struct btrfs_free_space_ctl *ctl, | |
649 | struct btrfs_path *path, u64 offset) | |
9d66e233 | 650 | { |
9d66e233 JB |
651 | struct btrfs_free_space_header *header; |
652 | struct extent_buffer *leaf; | |
a67509c3 | 653 | struct io_ctl io_ctl; |
9d66e233 | 654 | struct btrfs_key key; |
a67509c3 | 655 | struct btrfs_free_space *e, *n; |
b76808fc | 656 | LIST_HEAD(bitmaps); |
9d66e233 JB |
657 | u64 num_entries; |
658 | u64 num_bitmaps; | |
659 | u64 generation; | |
a67509c3 | 660 | u8 type; |
f6a39829 | 661 | int ret = 0; |
9d66e233 | 662 | |
9d66e233 | 663 | /* Nothing in the space cache, goodbye */ |
0414efae | 664 | if (!i_size_read(inode)) |
a67509c3 | 665 | return 0; |
9d66e233 JB |
666 | |
667 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
0414efae | 668 | key.offset = offset; |
9d66e233 JB |
669 | key.type = 0; |
670 | ||
671 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
0414efae | 672 | if (ret < 0) |
a67509c3 | 673 | return 0; |
0414efae | 674 | else if (ret > 0) { |
945d8962 | 675 | btrfs_release_path(path); |
a67509c3 | 676 | return 0; |
9d66e233 JB |
677 | } |
678 | ||
0414efae LZ |
679 | ret = -1; |
680 | ||
9d66e233 JB |
681 | leaf = path->nodes[0]; |
682 | header = btrfs_item_ptr(leaf, path->slots[0], | |
683 | struct btrfs_free_space_header); | |
684 | num_entries = btrfs_free_space_entries(leaf, header); | |
685 | num_bitmaps = btrfs_free_space_bitmaps(leaf, header); | |
686 | generation = btrfs_free_space_generation(leaf, header); | |
945d8962 | 687 | btrfs_release_path(path); |
9d66e233 | 688 | |
e570fd27 MX |
689 | if (!BTRFS_I(inode)->generation) { |
690 | btrfs_info(root->fs_info, | |
691 | "The free space cache file (%llu) is invalid. skip it\n", | |
692 | offset); | |
693 | return 0; | |
694 | } | |
695 | ||
9d66e233 | 696 | if (BTRFS_I(inode)->generation != generation) { |
c2cf52eb SK |
697 | btrfs_err(root->fs_info, |
698 | "free space inode generation (%llu) " | |
699 | "did not match free space cache generation (%llu)", | |
c1c9ff7c | 700 | BTRFS_I(inode)->generation, generation); |
a67509c3 | 701 | return 0; |
9d66e233 JB |
702 | } |
703 | ||
704 | if (!num_entries) | |
a67509c3 | 705 | return 0; |
9d66e233 | 706 | |
5349d6c3 | 707 | ret = io_ctl_init(&io_ctl, inode, root, 0); |
706efc66 LZ |
708 | if (ret) |
709 | return ret; | |
710 | ||
9d66e233 | 711 | ret = readahead_cache(inode); |
0414efae | 712 | if (ret) |
9d66e233 | 713 | goto out; |
9d66e233 | 714 | |
a67509c3 JB |
715 | ret = io_ctl_prepare_pages(&io_ctl, inode, 1); |
716 | if (ret) | |
717 | goto out; | |
9d66e233 | 718 | |
5b0e95bf JB |
719 | ret = io_ctl_check_crc(&io_ctl, 0); |
720 | if (ret) | |
721 | goto free_cache; | |
722 | ||
a67509c3 JB |
723 | ret = io_ctl_check_generation(&io_ctl, generation); |
724 | if (ret) | |
725 | goto free_cache; | |
9d66e233 | 726 | |
a67509c3 JB |
727 | while (num_entries) { |
728 | e = kmem_cache_zalloc(btrfs_free_space_cachep, | |
729 | GFP_NOFS); | |
730 | if (!e) | |
9d66e233 | 731 | goto free_cache; |
9d66e233 | 732 | |
5b0e95bf JB |
733 | ret = io_ctl_read_entry(&io_ctl, e, &type); |
734 | if (ret) { | |
735 | kmem_cache_free(btrfs_free_space_cachep, e); | |
736 | goto free_cache; | |
737 | } | |
738 | ||
a67509c3 JB |
739 | if (!e->bytes) { |
740 | kmem_cache_free(btrfs_free_space_cachep, e); | |
741 | goto free_cache; | |
9d66e233 | 742 | } |
a67509c3 JB |
743 | |
744 | if (type == BTRFS_FREE_SPACE_EXTENT) { | |
745 | spin_lock(&ctl->tree_lock); | |
746 | ret = link_free_space(ctl, e); | |
747 | spin_unlock(&ctl->tree_lock); | |
748 | if (ret) { | |
c2cf52eb SK |
749 | btrfs_err(root->fs_info, |
750 | "Duplicate entries in free space cache, dumping"); | |
a67509c3 | 751 | kmem_cache_free(btrfs_free_space_cachep, e); |
9d66e233 JB |
752 | goto free_cache; |
753 | } | |
a67509c3 | 754 | } else { |
b12d6869 | 755 | ASSERT(num_bitmaps); |
a67509c3 JB |
756 | num_bitmaps--; |
757 | e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS); | |
758 | if (!e->bitmap) { | |
759 | kmem_cache_free( | |
760 | btrfs_free_space_cachep, e); | |
9d66e233 JB |
761 | goto free_cache; |
762 | } | |
a67509c3 JB |
763 | spin_lock(&ctl->tree_lock); |
764 | ret = link_free_space(ctl, e); | |
765 | ctl->total_bitmaps++; | |
766 | ctl->op->recalc_thresholds(ctl); | |
767 | spin_unlock(&ctl->tree_lock); | |
768 | if (ret) { | |
c2cf52eb SK |
769 | btrfs_err(root->fs_info, |
770 | "Duplicate entries in free space cache, dumping"); | |
dc89e982 | 771 | kmem_cache_free(btrfs_free_space_cachep, e); |
9d66e233 JB |
772 | goto free_cache; |
773 | } | |
a67509c3 | 774 | list_add_tail(&e->list, &bitmaps); |
9d66e233 JB |
775 | } |
776 | ||
a67509c3 JB |
777 | num_entries--; |
778 | } | |
9d66e233 | 779 | |
2f120c05 JB |
780 | io_ctl_unmap_page(&io_ctl); |
781 | ||
a67509c3 JB |
782 | /* |
783 | * We add the bitmaps at the end of the entries in order that | |
784 | * the bitmap entries are added to the cache. | |
785 | */ | |
786 | list_for_each_entry_safe(e, n, &bitmaps, list) { | |
9d66e233 | 787 | list_del_init(&e->list); |
5b0e95bf JB |
788 | ret = io_ctl_read_bitmap(&io_ctl, e); |
789 | if (ret) | |
790 | goto free_cache; | |
9d66e233 JB |
791 | } |
792 | ||
a67509c3 | 793 | io_ctl_drop_pages(&io_ctl); |
cd023e7b | 794 | merge_space_tree(ctl); |
9d66e233 JB |
795 | ret = 1; |
796 | out: | |
a67509c3 | 797 | io_ctl_free(&io_ctl); |
9d66e233 | 798 | return ret; |
9d66e233 | 799 | free_cache: |
a67509c3 | 800 | io_ctl_drop_pages(&io_ctl); |
0414efae | 801 | __btrfs_remove_free_space_cache(ctl); |
9d66e233 JB |
802 | goto out; |
803 | } | |
804 | ||
0414efae LZ |
805 | int load_free_space_cache(struct btrfs_fs_info *fs_info, |
806 | struct btrfs_block_group_cache *block_group) | |
0cb59c99 | 807 | { |
34d52cb6 | 808 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0414efae LZ |
809 | struct btrfs_root *root = fs_info->tree_root; |
810 | struct inode *inode; | |
811 | struct btrfs_path *path; | |
5b0e95bf | 812 | int ret = 0; |
0414efae LZ |
813 | bool matched; |
814 | u64 used = btrfs_block_group_used(&block_group->item); | |
815 | ||
0414efae LZ |
816 | /* |
817 | * If this block group has been marked to be cleared for one reason or | |
818 | * another then we can't trust the on disk cache, so just return. | |
819 | */ | |
9d66e233 | 820 | spin_lock(&block_group->lock); |
0414efae LZ |
821 | if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) { |
822 | spin_unlock(&block_group->lock); | |
823 | return 0; | |
824 | } | |
9d66e233 | 825 | spin_unlock(&block_group->lock); |
0414efae LZ |
826 | |
827 | path = btrfs_alloc_path(); | |
828 | if (!path) | |
829 | return 0; | |
d53ba474 JB |
830 | path->search_commit_root = 1; |
831 | path->skip_locking = 1; | |
0414efae LZ |
832 | |
833 | inode = lookup_free_space_inode(root, block_group, path); | |
834 | if (IS_ERR(inode)) { | |
835 | btrfs_free_path(path); | |
836 | return 0; | |
837 | } | |
838 | ||
5b0e95bf JB |
839 | /* We may have converted the inode and made the cache invalid. */ |
840 | spin_lock(&block_group->lock); | |
841 | if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) { | |
842 | spin_unlock(&block_group->lock); | |
a7e221e9 | 843 | btrfs_free_path(path); |
5b0e95bf JB |
844 | goto out; |
845 | } | |
846 | spin_unlock(&block_group->lock); | |
847 | ||
0414efae LZ |
848 | ret = __load_free_space_cache(fs_info->tree_root, inode, ctl, |
849 | path, block_group->key.objectid); | |
850 | btrfs_free_path(path); | |
851 | if (ret <= 0) | |
852 | goto out; | |
853 | ||
854 | spin_lock(&ctl->tree_lock); | |
855 | matched = (ctl->free_space == (block_group->key.offset - used - | |
856 | block_group->bytes_super)); | |
857 | spin_unlock(&ctl->tree_lock); | |
858 | ||
859 | if (!matched) { | |
860 | __btrfs_remove_free_space_cache(ctl); | |
32d6b47f | 861 | btrfs_warn(fs_info, "block group %llu has wrong amount of free space", |
c2cf52eb | 862 | block_group->key.objectid); |
0414efae LZ |
863 | ret = -1; |
864 | } | |
865 | out: | |
866 | if (ret < 0) { | |
867 | /* This cache is bogus, make sure it gets cleared */ | |
868 | spin_lock(&block_group->lock); | |
869 | block_group->disk_cache_state = BTRFS_DC_CLEAR; | |
870 | spin_unlock(&block_group->lock); | |
82d5902d | 871 | ret = 0; |
0414efae | 872 | |
32d6b47f | 873 | btrfs_warn(fs_info, "failed to load free space cache for block group %llu, rebuild it now", |
c2cf52eb | 874 | block_group->key.objectid); |
0414efae LZ |
875 | } |
876 | ||
877 | iput(inode); | |
878 | return ret; | |
9d66e233 JB |
879 | } |
880 | ||
d4452bc5 CM |
881 | static noinline_for_stack |
882 | int write_cache_extent_entries(struct io_ctl *io_ctl, | |
883 | struct btrfs_free_space_ctl *ctl, | |
884 | struct btrfs_block_group_cache *block_group, | |
885 | int *entries, int *bitmaps, | |
886 | struct list_head *bitmap_list) | |
0cb59c99 | 887 | { |
c09544e0 | 888 | int ret; |
d4452bc5 CM |
889 | struct btrfs_free_cluster *cluster = NULL; |
890 | struct rb_node *node = rb_first(&ctl->free_space_offset); | |
55507ce3 | 891 | struct btrfs_trim_range *trim_entry; |
be1a12a0 | 892 | |
43be2146 | 893 | /* Get the cluster for this block_group if it exists */ |
d4452bc5 | 894 | if (block_group && !list_empty(&block_group->cluster_list)) { |
43be2146 JB |
895 | cluster = list_entry(block_group->cluster_list.next, |
896 | struct btrfs_free_cluster, | |
897 | block_group_list); | |
d4452bc5 | 898 | } |
43be2146 | 899 | |
f75b130e JB |
900 | if (!node && cluster) { |
901 | node = rb_first(&cluster->root); | |
902 | cluster = NULL; | |
903 | } | |
904 | ||
a67509c3 JB |
905 | /* Write out the extent entries */ |
906 | while (node) { | |
907 | struct btrfs_free_space *e; | |
0cb59c99 | 908 | |
a67509c3 | 909 | e = rb_entry(node, struct btrfs_free_space, offset_index); |
d4452bc5 | 910 | *entries += 1; |
0cb59c99 | 911 | |
d4452bc5 | 912 | ret = io_ctl_add_entry(io_ctl, e->offset, e->bytes, |
a67509c3 JB |
913 | e->bitmap); |
914 | if (ret) | |
d4452bc5 | 915 | goto fail; |
2f356126 | 916 | |
a67509c3 | 917 | if (e->bitmap) { |
d4452bc5 CM |
918 | list_add_tail(&e->list, bitmap_list); |
919 | *bitmaps += 1; | |
2f356126 | 920 | } |
a67509c3 JB |
921 | node = rb_next(node); |
922 | if (!node && cluster) { | |
923 | node = rb_first(&cluster->root); | |
924 | cluster = NULL; | |
43be2146 | 925 | } |
a67509c3 | 926 | } |
55507ce3 FM |
927 | |
928 | /* | |
929 | * Make sure we don't miss any range that was removed from our rbtree | |
930 | * because trimming is running. Otherwise after a umount+mount (or crash | |
931 | * after committing the transaction) we would leak free space and get | |
932 | * an inconsistent free space cache report from fsck. | |
933 | */ | |
934 | list_for_each_entry(trim_entry, &ctl->trimming_ranges, list) { | |
935 | ret = io_ctl_add_entry(io_ctl, trim_entry->start, | |
936 | trim_entry->bytes, NULL); | |
937 | if (ret) | |
938 | goto fail; | |
939 | *entries += 1; | |
940 | } | |
941 | ||
d4452bc5 CM |
942 | return 0; |
943 | fail: | |
944 | return -ENOSPC; | |
945 | } | |
946 | ||
947 | static noinline_for_stack int | |
948 | update_cache_item(struct btrfs_trans_handle *trans, | |
949 | struct btrfs_root *root, | |
950 | struct inode *inode, | |
951 | struct btrfs_path *path, u64 offset, | |
952 | int entries, int bitmaps) | |
953 | { | |
954 | struct btrfs_key key; | |
955 | struct btrfs_free_space_header *header; | |
956 | struct extent_buffer *leaf; | |
957 | int ret; | |
958 | ||
959 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
960 | key.offset = offset; | |
961 | key.type = 0; | |
962 | ||
963 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
964 | if (ret < 0) { | |
965 | clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1, | |
966 | EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL, | |
967 | GFP_NOFS); | |
968 | goto fail; | |
969 | } | |
970 | leaf = path->nodes[0]; | |
971 | if (ret > 0) { | |
972 | struct btrfs_key found_key; | |
973 | ASSERT(path->slots[0]); | |
974 | path->slots[0]--; | |
975 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
976 | if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID || | |
977 | found_key.offset != offset) { | |
978 | clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, | |
979 | inode->i_size - 1, | |
980 | EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, | |
981 | NULL, GFP_NOFS); | |
982 | btrfs_release_path(path); | |
983 | goto fail; | |
984 | } | |
985 | } | |
986 | ||
987 | BTRFS_I(inode)->generation = trans->transid; | |
988 | header = btrfs_item_ptr(leaf, path->slots[0], | |
989 | struct btrfs_free_space_header); | |
990 | btrfs_set_free_space_entries(leaf, header, entries); | |
991 | btrfs_set_free_space_bitmaps(leaf, header, bitmaps); | |
992 | btrfs_set_free_space_generation(leaf, header, trans->transid); | |
993 | btrfs_mark_buffer_dirty(leaf); | |
994 | btrfs_release_path(path); | |
995 | ||
996 | return 0; | |
997 | ||
998 | fail: | |
999 | return -1; | |
1000 | } | |
1001 | ||
1002 | static noinline_for_stack int | |
5349d6c3 MX |
1003 | write_pinned_extent_entries(struct btrfs_root *root, |
1004 | struct btrfs_block_group_cache *block_group, | |
1005 | struct io_ctl *io_ctl, | |
1006 | int *entries) | |
d4452bc5 CM |
1007 | { |
1008 | u64 start, extent_start, extent_end, len; | |
d4452bc5 CM |
1009 | struct extent_io_tree *unpin = NULL; |
1010 | int ret; | |
43be2146 | 1011 | |
5349d6c3 MX |
1012 | if (!block_group) |
1013 | return 0; | |
1014 | ||
a67509c3 JB |
1015 | /* |
1016 | * We want to add any pinned extents to our free space cache | |
1017 | * so we don't leak the space | |
d4452bc5 | 1018 | * |
db804f23 LZ |
1019 | * We shouldn't have switched the pinned extents yet so this is the |
1020 | * right one | |
1021 | */ | |
1022 | unpin = root->fs_info->pinned_extents; | |
1023 | ||
5349d6c3 | 1024 | start = block_group->key.objectid; |
db804f23 | 1025 | |
5349d6c3 | 1026 | while (start < block_group->key.objectid + block_group->key.offset) { |
db804f23 LZ |
1027 | ret = find_first_extent_bit(unpin, start, |
1028 | &extent_start, &extent_end, | |
e6138876 | 1029 | EXTENT_DIRTY, NULL); |
5349d6c3 MX |
1030 | if (ret) |
1031 | return 0; | |
0cb59c99 | 1032 | |
a67509c3 | 1033 | /* This pinned extent is out of our range */ |
db804f23 | 1034 | if (extent_start >= block_group->key.objectid + |
a67509c3 | 1035 | block_group->key.offset) |
5349d6c3 | 1036 | return 0; |
2f356126 | 1037 | |
db804f23 LZ |
1038 | extent_start = max(extent_start, start); |
1039 | extent_end = min(block_group->key.objectid + | |
1040 | block_group->key.offset, extent_end + 1); | |
1041 | len = extent_end - extent_start; | |
0cb59c99 | 1042 | |
d4452bc5 CM |
1043 | *entries += 1; |
1044 | ret = io_ctl_add_entry(io_ctl, extent_start, len, NULL); | |
a67509c3 | 1045 | if (ret) |
5349d6c3 | 1046 | return -ENOSPC; |
0cb59c99 | 1047 | |
db804f23 | 1048 | start = extent_end; |
a67509c3 | 1049 | } |
0cb59c99 | 1050 | |
5349d6c3 MX |
1051 | return 0; |
1052 | } | |
1053 | ||
1054 | static noinline_for_stack int | |
1055 | write_bitmap_entries(struct io_ctl *io_ctl, struct list_head *bitmap_list) | |
1056 | { | |
1057 | struct list_head *pos, *n; | |
1058 | int ret; | |
1059 | ||
0cb59c99 | 1060 | /* Write out the bitmaps */ |
d4452bc5 | 1061 | list_for_each_safe(pos, n, bitmap_list) { |
0cb59c99 JB |
1062 | struct btrfs_free_space *entry = |
1063 | list_entry(pos, struct btrfs_free_space, list); | |
1064 | ||
d4452bc5 | 1065 | ret = io_ctl_add_bitmap(io_ctl, entry->bitmap); |
a67509c3 | 1066 | if (ret) |
5349d6c3 | 1067 | return -ENOSPC; |
0cb59c99 | 1068 | list_del_init(&entry->list); |
be1a12a0 JB |
1069 | } |
1070 | ||
5349d6c3 MX |
1071 | return 0; |
1072 | } | |
0cb59c99 | 1073 | |
5349d6c3 MX |
1074 | static int flush_dirty_cache(struct inode *inode) |
1075 | { | |
1076 | int ret; | |
be1a12a0 | 1077 | |
0ef8b726 | 1078 | ret = btrfs_wait_ordered_range(inode, 0, (u64)-1); |
5349d6c3 | 1079 | if (ret) |
0ef8b726 JB |
1080 | clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1, |
1081 | EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL, | |
1082 | GFP_NOFS); | |
0cb59c99 | 1083 | |
5349d6c3 | 1084 | return ret; |
d4452bc5 CM |
1085 | } |
1086 | ||
1087 | static void noinline_for_stack | |
1088 | cleanup_write_cache_enospc(struct inode *inode, | |
1089 | struct io_ctl *io_ctl, | |
1090 | struct extent_state **cached_state, | |
1091 | struct list_head *bitmap_list) | |
1092 | { | |
1093 | struct list_head *pos, *n; | |
5349d6c3 | 1094 | |
d4452bc5 CM |
1095 | list_for_each_safe(pos, n, bitmap_list) { |
1096 | struct btrfs_free_space *entry = | |
1097 | list_entry(pos, struct btrfs_free_space, list); | |
1098 | list_del_init(&entry->list); | |
0cb59c99 | 1099 | } |
d4452bc5 CM |
1100 | io_ctl_drop_pages(io_ctl); |
1101 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0, | |
1102 | i_size_read(inode) - 1, cached_state, | |
1103 | GFP_NOFS); | |
1104 | } | |
549b4fdb | 1105 | |
d4452bc5 CM |
1106 | /** |
1107 | * __btrfs_write_out_cache - write out cached info to an inode | |
1108 | * @root - the root the inode belongs to | |
1109 | * @ctl - the free space cache we are going to write out | |
1110 | * @block_group - the block_group for this cache if it belongs to a block_group | |
1111 | * @trans - the trans handle | |
1112 | * @path - the path to use | |
1113 | * @offset - the offset for the key we'll insert | |
1114 | * | |
1115 | * This function writes out a free space cache struct to disk for quick recovery | |
1116 | * on mount. This will return 0 if it was successfull in writing the cache out, | |
1117 | * and -1 if it was not. | |
1118 | */ | |
1119 | static int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode, | |
1120 | struct btrfs_free_space_ctl *ctl, | |
1121 | struct btrfs_block_group_cache *block_group, | |
1122 | struct btrfs_trans_handle *trans, | |
1123 | struct btrfs_path *path, u64 offset) | |
1124 | { | |
1125 | struct extent_state *cached_state = NULL; | |
1126 | struct io_ctl io_ctl; | |
5349d6c3 | 1127 | LIST_HEAD(bitmap_list); |
d4452bc5 CM |
1128 | int entries = 0; |
1129 | int bitmaps = 0; | |
1130 | int ret; | |
d4452bc5 CM |
1131 | |
1132 | if (!i_size_read(inode)) | |
1133 | return -1; | |
1134 | ||
5349d6c3 | 1135 | ret = io_ctl_init(&io_ctl, inode, root, 1); |
d4452bc5 CM |
1136 | if (ret) |
1137 | return -1; | |
1138 | ||
e570fd27 MX |
1139 | if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA)) { |
1140 | down_write(&block_group->data_rwsem); | |
1141 | spin_lock(&block_group->lock); | |
1142 | if (block_group->delalloc_bytes) { | |
1143 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
1144 | spin_unlock(&block_group->lock); | |
1145 | up_write(&block_group->data_rwsem); | |
1146 | BTRFS_I(inode)->generation = 0; | |
1147 | ret = 0; | |
1148 | goto out; | |
1149 | } | |
1150 | spin_unlock(&block_group->lock); | |
1151 | } | |
1152 | ||
d4452bc5 CM |
1153 | /* Lock all pages first so we can lock the extent safely. */ |
1154 | io_ctl_prepare_pages(&io_ctl, inode, 0); | |
1155 | ||
1156 | lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1, | |
1157 | 0, &cached_state); | |
1158 | ||
d4452bc5 CM |
1159 | io_ctl_set_generation(&io_ctl, trans->transid); |
1160 | ||
55507ce3 | 1161 | mutex_lock(&ctl->cache_writeout_mutex); |
5349d6c3 | 1162 | /* Write out the extent entries in the free space cache */ |
d4452bc5 CM |
1163 | ret = write_cache_extent_entries(&io_ctl, ctl, |
1164 | block_group, &entries, &bitmaps, | |
1165 | &bitmap_list); | |
55507ce3 FM |
1166 | if (ret) { |
1167 | mutex_unlock(&ctl->cache_writeout_mutex); | |
d4452bc5 | 1168 | goto out_nospc; |
55507ce3 | 1169 | } |
d4452bc5 | 1170 | |
5349d6c3 MX |
1171 | /* |
1172 | * Some spaces that are freed in the current transaction are pinned, | |
1173 | * they will be added into free space cache after the transaction is | |
1174 | * committed, we shouldn't lose them. | |
1175 | */ | |
1176 | ret = write_pinned_extent_entries(root, block_group, &io_ctl, &entries); | |
55507ce3 FM |
1177 | if (ret) { |
1178 | mutex_unlock(&ctl->cache_writeout_mutex); | |
5349d6c3 | 1179 | goto out_nospc; |
55507ce3 | 1180 | } |
5349d6c3 | 1181 | |
55507ce3 FM |
1182 | /* |
1183 | * At last, we write out all the bitmaps and keep cache_writeout_mutex | |
1184 | * locked while doing it because a concurrent trim can be manipulating | |
1185 | * or freeing the bitmap. | |
1186 | */ | |
5349d6c3 | 1187 | ret = write_bitmap_entries(&io_ctl, &bitmap_list); |
55507ce3 | 1188 | mutex_unlock(&ctl->cache_writeout_mutex); |
5349d6c3 MX |
1189 | if (ret) |
1190 | goto out_nospc; | |
1191 | ||
1192 | /* Zero out the rest of the pages just to make sure */ | |
1193 | io_ctl_zero_remaining_pages(&io_ctl); | |
d4452bc5 | 1194 | |
5349d6c3 MX |
1195 | /* Everything is written out, now we dirty the pages in the file. */ |
1196 | ret = btrfs_dirty_pages(root, inode, io_ctl.pages, io_ctl.num_pages, | |
1197 | 0, i_size_read(inode), &cached_state); | |
1198 | if (ret) | |
d4452bc5 | 1199 | goto out_nospc; |
5349d6c3 | 1200 | |
e570fd27 MX |
1201 | if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA)) |
1202 | up_write(&block_group->data_rwsem); | |
5349d6c3 MX |
1203 | /* |
1204 | * Release the pages and unlock the extent, we will flush | |
1205 | * them out later | |
1206 | */ | |
1207 | io_ctl_drop_pages(&io_ctl); | |
1208 | ||
1209 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0, | |
1210 | i_size_read(inode) - 1, &cached_state, GFP_NOFS); | |
1211 | ||
1212 | /* Flush the dirty pages in the cache file. */ | |
1213 | ret = flush_dirty_cache(inode); | |
1214 | if (ret) | |
d4452bc5 CM |
1215 | goto out; |
1216 | ||
5349d6c3 MX |
1217 | /* Update the cache item to tell everyone this cache file is valid. */ |
1218 | ret = update_cache_item(trans, root, inode, path, offset, | |
d4452bc5 | 1219 | entries, bitmaps); |
2f356126 | 1220 | out: |
a67509c3 | 1221 | io_ctl_free(&io_ctl); |
5349d6c3 | 1222 | if (ret) { |
a67509c3 | 1223 | invalidate_inode_pages2(inode->i_mapping); |
0cb59c99 JB |
1224 | BTRFS_I(inode)->generation = 0; |
1225 | } | |
0cb59c99 | 1226 | btrfs_update_inode(trans, root, inode); |
5349d6c3 | 1227 | return ret; |
a67509c3 JB |
1228 | |
1229 | out_nospc: | |
d4452bc5 | 1230 | cleanup_write_cache_enospc(inode, &io_ctl, &cached_state, &bitmap_list); |
e570fd27 MX |
1231 | |
1232 | if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA)) | |
1233 | up_write(&block_group->data_rwsem); | |
1234 | ||
a67509c3 | 1235 | goto out; |
0414efae LZ |
1236 | } |
1237 | ||
1238 | int btrfs_write_out_cache(struct btrfs_root *root, | |
1239 | struct btrfs_trans_handle *trans, | |
1240 | struct btrfs_block_group_cache *block_group, | |
1241 | struct btrfs_path *path) | |
1242 | { | |
1243 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
1244 | struct inode *inode; | |
1245 | int ret = 0; | |
ce93ec54 | 1246 | enum btrfs_disk_cache_state dcs = BTRFS_DC_WRITTEN; |
0414efae LZ |
1247 | |
1248 | root = root->fs_info->tree_root; | |
1249 | ||
1250 | spin_lock(&block_group->lock); | |
1251 | if (block_group->disk_cache_state < BTRFS_DC_SETUP) { | |
1252 | spin_unlock(&block_group->lock); | |
1253 | return 0; | |
1254 | } | |
e570fd27 MX |
1255 | |
1256 | if (block_group->delalloc_bytes) { | |
1257 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
1258 | spin_unlock(&block_group->lock); | |
1259 | return 0; | |
1260 | } | |
0414efae LZ |
1261 | spin_unlock(&block_group->lock); |
1262 | ||
1263 | inode = lookup_free_space_inode(root, block_group, path); | |
1264 | if (IS_ERR(inode)) | |
1265 | return 0; | |
1266 | ||
1267 | ret = __btrfs_write_out_cache(root, inode, ctl, block_group, trans, | |
1268 | path, block_group->key.objectid); | |
c09544e0 | 1269 | if (ret) { |
ce93ec54 | 1270 | dcs = BTRFS_DC_ERROR; |
82d5902d | 1271 | ret = 0; |
c09544e0 | 1272 | #ifdef DEBUG |
c2cf52eb SK |
1273 | btrfs_err(root->fs_info, |
1274 | "failed to write free space cache for block group %llu", | |
1275 | block_group->key.objectid); | |
c09544e0 | 1276 | #endif |
0414efae LZ |
1277 | } |
1278 | ||
ce93ec54 JB |
1279 | spin_lock(&block_group->lock); |
1280 | block_group->disk_cache_state = dcs; | |
1281 | spin_unlock(&block_group->lock); | |
0cb59c99 JB |
1282 | iput(inode); |
1283 | return ret; | |
1284 | } | |
1285 | ||
34d52cb6 | 1286 | static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit, |
96303081 | 1287 | u64 offset) |
0f9dd46c | 1288 | { |
b12d6869 | 1289 | ASSERT(offset >= bitmap_start); |
96303081 | 1290 | offset -= bitmap_start; |
34d52cb6 | 1291 | return (unsigned long)(div_u64(offset, unit)); |
96303081 | 1292 | } |
0f9dd46c | 1293 | |
34d52cb6 | 1294 | static inline unsigned long bytes_to_bits(u64 bytes, u32 unit) |
96303081 | 1295 | { |
34d52cb6 | 1296 | return (unsigned long)(div_u64(bytes, unit)); |
96303081 | 1297 | } |
0f9dd46c | 1298 | |
34d52cb6 | 1299 | static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1300 | u64 offset) |
1301 | { | |
1302 | u64 bitmap_start; | |
b8b93add | 1303 | u32 bytes_per_bitmap; |
0f9dd46c | 1304 | |
34d52cb6 LZ |
1305 | bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit; |
1306 | bitmap_start = offset - ctl->start; | |
b8b93add | 1307 | bitmap_start = div_u64(bitmap_start, bytes_per_bitmap); |
96303081 | 1308 | bitmap_start *= bytes_per_bitmap; |
34d52cb6 | 1309 | bitmap_start += ctl->start; |
0f9dd46c | 1310 | |
96303081 | 1311 | return bitmap_start; |
0f9dd46c JB |
1312 | } |
1313 | ||
96303081 JB |
1314 | static int tree_insert_offset(struct rb_root *root, u64 offset, |
1315 | struct rb_node *node, int bitmap) | |
0f9dd46c JB |
1316 | { |
1317 | struct rb_node **p = &root->rb_node; | |
1318 | struct rb_node *parent = NULL; | |
1319 | struct btrfs_free_space *info; | |
1320 | ||
1321 | while (*p) { | |
1322 | parent = *p; | |
96303081 | 1323 | info = rb_entry(parent, struct btrfs_free_space, offset_index); |
0f9dd46c | 1324 | |
96303081 | 1325 | if (offset < info->offset) { |
0f9dd46c | 1326 | p = &(*p)->rb_left; |
96303081 | 1327 | } else if (offset > info->offset) { |
0f9dd46c | 1328 | p = &(*p)->rb_right; |
96303081 JB |
1329 | } else { |
1330 | /* | |
1331 | * we could have a bitmap entry and an extent entry | |
1332 | * share the same offset. If this is the case, we want | |
1333 | * the extent entry to always be found first if we do a | |
1334 | * linear search through the tree, since we want to have | |
1335 | * the quickest allocation time, and allocating from an | |
1336 | * extent is faster than allocating from a bitmap. So | |
1337 | * if we're inserting a bitmap and we find an entry at | |
1338 | * this offset, we want to go right, or after this entry | |
1339 | * logically. If we are inserting an extent and we've | |
1340 | * found a bitmap, we want to go left, or before | |
1341 | * logically. | |
1342 | */ | |
1343 | if (bitmap) { | |
207dde82 JB |
1344 | if (info->bitmap) { |
1345 | WARN_ON_ONCE(1); | |
1346 | return -EEXIST; | |
1347 | } | |
96303081 JB |
1348 | p = &(*p)->rb_right; |
1349 | } else { | |
207dde82 JB |
1350 | if (!info->bitmap) { |
1351 | WARN_ON_ONCE(1); | |
1352 | return -EEXIST; | |
1353 | } | |
96303081 JB |
1354 | p = &(*p)->rb_left; |
1355 | } | |
1356 | } | |
0f9dd46c JB |
1357 | } |
1358 | ||
1359 | rb_link_node(node, parent, p); | |
1360 | rb_insert_color(node, root); | |
1361 | ||
1362 | return 0; | |
1363 | } | |
1364 | ||
1365 | /* | |
70cb0743 JB |
1366 | * searches the tree for the given offset. |
1367 | * | |
96303081 JB |
1368 | * fuzzy - If this is set, then we are trying to make an allocation, and we just |
1369 | * want a section that has at least bytes size and comes at or after the given | |
1370 | * offset. | |
0f9dd46c | 1371 | */ |
96303081 | 1372 | static struct btrfs_free_space * |
34d52cb6 | 1373 | tree_search_offset(struct btrfs_free_space_ctl *ctl, |
96303081 | 1374 | u64 offset, int bitmap_only, int fuzzy) |
0f9dd46c | 1375 | { |
34d52cb6 | 1376 | struct rb_node *n = ctl->free_space_offset.rb_node; |
96303081 JB |
1377 | struct btrfs_free_space *entry, *prev = NULL; |
1378 | ||
1379 | /* find entry that is closest to the 'offset' */ | |
1380 | while (1) { | |
1381 | if (!n) { | |
1382 | entry = NULL; | |
1383 | break; | |
1384 | } | |
0f9dd46c | 1385 | |
0f9dd46c | 1386 | entry = rb_entry(n, struct btrfs_free_space, offset_index); |
96303081 | 1387 | prev = entry; |
0f9dd46c | 1388 | |
96303081 | 1389 | if (offset < entry->offset) |
0f9dd46c | 1390 | n = n->rb_left; |
96303081 | 1391 | else if (offset > entry->offset) |
0f9dd46c | 1392 | n = n->rb_right; |
96303081 | 1393 | else |
0f9dd46c | 1394 | break; |
0f9dd46c JB |
1395 | } |
1396 | ||
96303081 JB |
1397 | if (bitmap_only) { |
1398 | if (!entry) | |
1399 | return NULL; | |
1400 | if (entry->bitmap) | |
1401 | return entry; | |
0f9dd46c | 1402 | |
96303081 JB |
1403 | /* |
1404 | * bitmap entry and extent entry may share same offset, | |
1405 | * in that case, bitmap entry comes after extent entry. | |
1406 | */ | |
1407 | n = rb_next(n); | |
1408 | if (!n) | |
1409 | return NULL; | |
1410 | entry = rb_entry(n, struct btrfs_free_space, offset_index); | |
1411 | if (entry->offset != offset) | |
1412 | return NULL; | |
0f9dd46c | 1413 | |
96303081 JB |
1414 | WARN_ON(!entry->bitmap); |
1415 | return entry; | |
1416 | } else if (entry) { | |
1417 | if (entry->bitmap) { | |
0f9dd46c | 1418 | /* |
96303081 JB |
1419 | * if previous extent entry covers the offset, |
1420 | * we should return it instead of the bitmap entry | |
0f9dd46c | 1421 | */ |
de6c4115 MX |
1422 | n = rb_prev(&entry->offset_index); |
1423 | if (n) { | |
96303081 JB |
1424 | prev = rb_entry(n, struct btrfs_free_space, |
1425 | offset_index); | |
de6c4115 MX |
1426 | if (!prev->bitmap && |
1427 | prev->offset + prev->bytes > offset) | |
1428 | entry = prev; | |
0f9dd46c | 1429 | } |
96303081 JB |
1430 | } |
1431 | return entry; | |
1432 | } | |
1433 | ||
1434 | if (!prev) | |
1435 | return NULL; | |
1436 | ||
1437 | /* find last entry before the 'offset' */ | |
1438 | entry = prev; | |
1439 | if (entry->offset > offset) { | |
1440 | n = rb_prev(&entry->offset_index); | |
1441 | if (n) { | |
1442 | entry = rb_entry(n, struct btrfs_free_space, | |
1443 | offset_index); | |
b12d6869 | 1444 | ASSERT(entry->offset <= offset); |
0f9dd46c | 1445 | } else { |
96303081 JB |
1446 | if (fuzzy) |
1447 | return entry; | |
1448 | else | |
1449 | return NULL; | |
0f9dd46c JB |
1450 | } |
1451 | } | |
1452 | ||
96303081 | 1453 | if (entry->bitmap) { |
de6c4115 MX |
1454 | n = rb_prev(&entry->offset_index); |
1455 | if (n) { | |
96303081 JB |
1456 | prev = rb_entry(n, struct btrfs_free_space, |
1457 | offset_index); | |
de6c4115 MX |
1458 | if (!prev->bitmap && |
1459 | prev->offset + prev->bytes > offset) | |
1460 | return prev; | |
96303081 | 1461 | } |
34d52cb6 | 1462 | if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset) |
96303081 JB |
1463 | return entry; |
1464 | } else if (entry->offset + entry->bytes > offset) | |
1465 | return entry; | |
1466 | ||
1467 | if (!fuzzy) | |
1468 | return NULL; | |
1469 | ||
1470 | while (1) { | |
1471 | if (entry->bitmap) { | |
1472 | if (entry->offset + BITS_PER_BITMAP * | |
34d52cb6 | 1473 | ctl->unit > offset) |
96303081 JB |
1474 | break; |
1475 | } else { | |
1476 | if (entry->offset + entry->bytes > offset) | |
1477 | break; | |
1478 | } | |
1479 | ||
1480 | n = rb_next(&entry->offset_index); | |
1481 | if (!n) | |
1482 | return NULL; | |
1483 | entry = rb_entry(n, struct btrfs_free_space, offset_index); | |
1484 | } | |
1485 | return entry; | |
0f9dd46c JB |
1486 | } |
1487 | ||
f333adb5 | 1488 | static inline void |
34d52cb6 | 1489 | __unlink_free_space(struct btrfs_free_space_ctl *ctl, |
f333adb5 | 1490 | struct btrfs_free_space *info) |
0f9dd46c | 1491 | { |
34d52cb6 LZ |
1492 | rb_erase(&info->offset_index, &ctl->free_space_offset); |
1493 | ctl->free_extents--; | |
f333adb5 LZ |
1494 | } |
1495 | ||
34d52cb6 | 1496 | static void unlink_free_space(struct btrfs_free_space_ctl *ctl, |
f333adb5 LZ |
1497 | struct btrfs_free_space *info) |
1498 | { | |
34d52cb6 LZ |
1499 | __unlink_free_space(ctl, info); |
1500 | ctl->free_space -= info->bytes; | |
0f9dd46c JB |
1501 | } |
1502 | ||
34d52cb6 | 1503 | static int link_free_space(struct btrfs_free_space_ctl *ctl, |
0f9dd46c JB |
1504 | struct btrfs_free_space *info) |
1505 | { | |
1506 | int ret = 0; | |
1507 | ||
b12d6869 | 1508 | ASSERT(info->bytes || info->bitmap); |
34d52cb6 | 1509 | ret = tree_insert_offset(&ctl->free_space_offset, info->offset, |
96303081 | 1510 | &info->offset_index, (info->bitmap != NULL)); |
0f9dd46c JB |
1511 | if (ret) |
1512 | return ret; | |
1513 | ||
34d52cb6 LZ |
1514 | ctl->free_space += info->bytes; |
1515 | ctl->free_extents++; | |
96303081 JB |
1516 | return ret; |
1517 | } | |
1518 | ||
34d52cb6 | 1519 | static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl) |
96303081 | 1520 | { |
34d52cb6 | 1521 | struct btrfs_block_group_cache *block_group = ctl->private; |
25891f79 JB |
1522 | u64 max_bytes; |
1523 | u64 bitmap_bytes; | |
1524 | u64 extent_bytes; | |
8eb2d829 | 1525 | u64 size = block_group->key.offset; |
b8b93add DS |
1526 | u32 bytes_per_bg = BITS_PER_BITMAP * ctl->unit; |
1527 | u32 max_bitmaps = div_u64(size + bytes_per_bg - 1, bytes_per_bg); | |
34d52cb6 | 1528 | |
b8b93add | 1529 | max_bitmaps = max_t(u32, max_bitmaps, 1); |
dde5740f | 1530 | |
b12d6869 | 1531 | ASSERT(ctl->total_bitmaps <= max_bitmaps); |
96303081 JB |
1532 | |
1533 | /* | |
1534 | * The goal is to keep the total amount of memory used per 1gb of space | |
1535 | * at or below 32k, so we need to adjust how much memory we allow to be | |
1536 | * used by extent based free space tracking | |
1537 | */ | |
8eb2d829 LZ |
1538 | if (size < 1024 * 1024 * 1024) |
1539 | max_bytes = MAX_CACHE_BYTES_PER_GIG; | |
1540 | else | |
1541 | max_bytes = MAX_CACHE_BYTES_PER_GIG * | |
f8c269d7 | 1542 | div_u64(size, 1024 * 1024 * 1024); |
96303081 | 1543 | |
25891f79 JB |
1544 | /* |
1545 | * we want to account for 1 more bitmap than what we have so we can make | |
1546 | * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as | |
1547 | * we add more bitmaps. | |
1548 | */ | |
34d52cb6 | 1549 | bitmap_bytes = (ctl->total_bitmaps + 1) * PAGE_CACHE_SIZE; |
96303081 | 1550 | |
25891f79 | 1551 | if (bitmap_bytes >= max_bytes) { |
34d52cb6 | 1552 | ctl->extents_thresh = 0; |
25891f79 JB |
1553 | return; |
1554 | } | |
96303081 | 1555 | |
25891f79 | 1556 | /* |
f8c269d7 | 1557 | * we want the extent entry threshold to always be at most 1/2 the max |
25891f79 JB |
1558 | * bytes we can have, or whatever is less than that. |
1559 | */ | |
1560 | extent_bytes = max_bytes - bitmap_bytes; | |
f8c269d7 | 1561 | extent_bytes = min_t(u64, extent_bytes, max_bytes >> 1); |
96303081 | 1562 | |
34d52cb6 | 1563 | ctl->extents_thresh = |
f8c269d7 | 1564 | div_u64(extent_bytes, sizeof(struct btrfs_free_space)); |
96303081 JB |
1565 | } |
1566 | ||
bb3ac5a4 MX |
1567 | static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl, |
1568 | struct btrfs_free_space *info, | |
1569 | u64 offset, u64 bytes) | |
96303081 | 1570 | { |
f38b6e75 | 1571 | unsigned long start, count; |
96303081 | 1572 | |
34d52cb6 LZ |
1573 | start = offset_to_bit(info->offset, ctl->unit, offset); |
1574 | count = bytes_to_bits(bytes, ctl->unit); | |
b12d6869 | 1575 | ASSERT(start + count <= BITS_PER_BITMAP); |
96303081 | 1576 | |
f38b6e75 | 1577 | bitmap_clear(info->bitmap, start, count); |
96303081 JB |
1578 | |
1579 | info->bytes -= bytes; | |
bb3ac5a4 MX |
1580 | } |
1581 | ||
1582 | static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl, | |
1583 | struct btrfs_free_space *info, u64 offset, | |
1584 | u64 bytes) | |
1585 | { | |
1586 | __bitmap_clear_bits(ctl, info, offset, bytes); | |
34d52cb6 | 1587 | ctl->free_space -= bytes; |
96303081 JB |
1588 | } |
1589 | ||
34d52cb6 | 1590 | static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl, |
817d52f8 JB |
1591 | struct btrfs_free_space *info, u64 offset, |
1592 | u64 bytes) | |
96303081 | 1593 | { |
f38b6e75 | 1594 | unsigned long start, count; |
96303081 | 1595 | |
34d52cb6 LZ |
1596 | start = offset_to_bit(info->offset, ctl->unit, offset); |
1597 | count = bytes_to_bits(bytes, ctl->unit); | |
b12d6869 | 1598 | ASSERT(start + count <= BITS_PER_BITMAP); |
96303081 | 1599 | |
f38b6e75 | 1600 | bitmap_set(info->bitmap, start, count); |
96303081 JB |
1601 | |
1602 | info->bytes += bytes; | |
34d52cb6 | 1603 | ctl->free_space += bytes; |
96303081 JB |
1604 | } |
1605 | ||
a4820398 MX |
1606 | /* |
1607 | * If we can not find suitable extent, we will use bytes to record | |
1608 | * the size of the max extent. | |
1609 | */ | |
34d52cb6 | 1610 | static int search_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1611 | struct btrfs_free_space *bitmap_info, u64 *offset, |
1612 | u64 *bytes) | |
1613 | { | |
1614 | unsigned long found_bits = 0; | |
a4820398 | 1615 | unsigned long max_bits = 0; |
96303081 JB |
1616 | unsigned long bits, i; |
1617 | unsigned long next_zero; | |
a4820398 | 1618 | unsigned long extent_bits; |
96303081 | 1619 | |
34d52cb6 | 1620 | i = offset_to_bit(bitmap_info->offset, ctl->unit, |
96303081 | 1621 | max_t(u64, *offset, bitmap_info->offset)); |
34d52cb6 | 1622 | bits = bytes_to_bits(*bytes, ctl->unit); |
96303081 | 1623 | |
ebb3dad4 | 1624 | for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) { |
96303081 JB |
1625 | next_zero = find_next_zero_bit(bitmap_info->bitmap, |
1626 | BITS_PER_BITMAP, i); | |
a4820398 MX |
1627 | extent_bits = next_zero - i; |
1628 | if (extent_bits >= bits) { | |
1629 | found_bits = extent_bits; | |
96303081 | 1630 | break; |
a4820398 MX |
1631 | } else if (extent_bits > max_bits) { |
1632 | max_bits = extent_bits; | |
96303081 JB |
1633 | } |
1634 | i = next_zero; | |
1635 | } | |
1636 | ||
1637 | if (found_bits) { | |
34d52cb6 LZ |
1638 | *offset = (u64)(i * ctl->unit) + bitmap_info->offset; |
1639 | *bytes = (u64)(found_bits) * ctl->unit; | |
96303081 JB |
1640 | return 0; |
1641 | } | |
1642 | ||
a4820398 | 1643 | *bytes = (u64)(max_bits) * ctl->unit; |
96303081 JB |
1644 | return -1; |
1645 | } | |
1646 | ||
a4820398 | 1647 | /* Cache the size of the max extent in bytes */ |
34d52cb6 | 1648 | static struct btrfs_free_space * |
53b381b3 | 1649 | find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes, |
a4820398 | 1650 | unsigned long align, u64 *max_extent_size) |
96303081 JB |
1651 | { |
1652 | struct btrfs_free_space *entry; | |
1653 | struct rb_node *node; | |
53b381b3 DW |
1654 | u64 tmp; |
1655 | u64 align_off; | |
96303081 JB |
1656 | int ret; |
1657 | ||
34d52cb6 | 1658 | if (!ctl->free_space_offset.rb_node) |
a4820398 | 1659 | goto out; |
96303081 | 1660 | |
34d52cb6 | 1661 | entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1); |
96303081 | 1662 | if (!entry) |
a4820398 | 1663 | goto out; |
96303081 JB |
1664 | |
1665 | for (node = &entry->offset_index; node; node = rb_next(node)) { | |
1666 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
a4820398 MX |
1667 | if (entry->bytes < *bytes) { |
1668 | if (entry->bytes > *max_extent_size) | |
1669 | *max_extent_size = entry->bytes; | |
96303081 | 1670 | continue; |
a4820398 | 1671 | } |
96303081 | 1672 | |
53b381b3 DW |
1673 | /* make sure the space returned is big enough |
1674 | * to match our requested alignment | |
1675 | */ | |
1676 | if (*bytes >= align) { | |
a4820398 | 1677 | tmp = entry->offset - ctl->start + align - 1; |
47c5713f | 1678 | tmp = div64_u64(tmp, align); |
53b381b3 DW |
1679 | tmp = tmp * align + ctl->start; |
1680 | align_off = tmp - entry->offset; | |
1681 | } else { | |
1682 | align_off = 0; | |
1683 | tmp = entry->offset; | |
1684 | } | |
1685 | ||
a4820398 MX |
1686 | if (entry->bytes < *bytes + align_off) { |
1687 | if (entry->bytes > *max_extent_size) | |
1688 | *max_extent_size = entry->bytes; | |
53b381b3 | 1689 | continue; |
a4820398 | 1690 | } |
53b381b3 | 1691 | |
96303081 | 1692 | if (entry->bitmap) { |
a4820398 MX |
1693 | u64 size = *bytes; |
1694 | ||
1695 | ret = search_bitmap(ctl, entry, &tmp, &size); | |
53b381b3 DW |
1696 | if (!ret) { |
1697 | *offset = tmp; | |
a4820398 | 1698 | *bytes = size; |
96303081 | 1699 | return entry; |
a4820398 MX |
1700 | } else if (size > *max_extent_size) { |
1701 | *max_extent_size = size; | |
53b381b3 | 1702 | } |
96303081 JB |
1703 | continue; |
1704 | } | |
1705 | ||
53b381b3 DW |
1706 | *offset = tmp; |
1707 | *bytes = entry->bytes - align_off; | |
96303081 JB |
1708 | return entry; |
1709 | } | |
a4820398 | 1710 | out: |
96303081 JB |
1711 | return NULL; |
1712 | } | |
1713 | ||
34d52cb6 | 1714 | static void add_new_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1715 | struct btrfs_free_space *info, u64 offset) |
1716 | { | |
34d52cb6 | 1717 | info->offset = offset_to_bitmap(ctl, offset); |
f019f426 | 1718 | info->bytes = 0; |
f2d0f676 | 1719 | INIT_LIST_HEAD(&info->list); |
34d52cb6 LZ |
1720 | link_free_space(ctl, info); |
1721 | ctl->total_bitmaps++; | |
96303081 | 1722 | |
34d52cb6 | 1723 | ctl->op->recalc_thresholds(ctl); |
96303081 JB |
1724 | } |
1725 | ||
34d52cb6 | 1726 | static void free_bitmap(struct btrfs_free_space_ctl *ctl, |
edf6e2d1 LZ |
1727 | struct btrfs_free_space *bitmap_info) |
1728 | { | |
34d52cb6 | 1729 | unlink_free_space(ctl, bitmap_info); |
edf6e2d1 | 1730 | kfree(bitmap_info->bitmap); |
dc89e982 | 1731 | kmem_cache_free(btrfs_free_space_cachep, bitmap_info); |
34d52cb6 LZ |
1732 | ctl->total_bitmaps--; |
1733 | ctl->op->recalc_thresholds(ctl); | |
edf6e2d1 LZ |
1734 | } |
1735 | ||
34d52cb6 | 1736 | static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1737 | struct btrfs_free_space *bitmap_info, |
1738 | u64 *offset, u64 *bytes) | |
1739 | { | |
1740 | u64 end; | |
6606bb97 JB |
1741 | u64 search_start, search_bytes; |
1742 | int ret; | |
96303081 JB |
1743 | |
1744 | again: | |
34d52cb6 | 1745 | end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1; |
96303081 | 1746 | |
6606bb97 | 1747 | /* |
bdb7d303 JB |
1748 | * We need to search for bits in this bitmap. We could only cover some |
1749 | * of the extent in this bitmap thanks to how we add space, so we need | |
1750 | * to search for as much as it as we can and clear that amount, and then | |
1751 | * go searching for the next bit. | |
6606bb97 JB |
1752 | */ |
1753 | search_start = *offset; | |
bdb7d303 | 1754 | search_bytes = ctl->unit; |
13dbc089 | 1755 | search_bytes = min(search_bytes, end - search_start + 1); |
34d52cb6 | 1756 | ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes); |
b50c6e25 JB |
1757 | if (ret < 0 || search_start != *offset) |
1758 | return -EINVAL; | |
6606bb97 | 1759 | |
bdb7d303 JB |
1760 | /* We may have found more bits than what we need */ |
1761 | search_bytes = min(search_bytes, *bytes); | |
1762 | ||
1763 | /* Cannot clear past the end of the bitmap */ | |
1764 | search_bytes = min(search_bytes, end - search_start + 1); | |
1765 | ||
1766 | bitmap_clear_bits(ctl, bitmap_info, search_start, search_bytes); | |
1767 | *offset += search_bytes; | |
1768 | *bytes -= search_bytes; | |
96303081 JB |
1769 | |
1770 | if (*bytes) { | |
6606bb97 | 1771 | struct rb_node *next = rb_next(&bitmap_info->offset_index); |
edf6e2d1 | 1772 | if (!bitmap_info->bytes) |
34d52cb6 | 1773 | free_bitmap(ctl, bitmap_info); |
96303081 | 1774 | |
6606bb97 JB |
1775 | /* |
1776 | * no entry after this bitmap, but we still have bytes to | |
1777 | * remove, so something has gone wrong. | |
1778 | */ | |
1779 | if (!next) | |
96303081 JB |
1780 | return -EINVAL; |
1781 | ||
6606bb97 JB |
1782 | bitmap_info = rb_entry(next, struct btrfs_free_space, |
1783 | offset_index); | |
1784 | ||
1785 | /* | |
1786 | * if the next entry isn't a bitmap we need to return to let the | |
1787 | * extent stuff do its work. | |
1788 | */ | |
96303081 JB |
1789 | if (!bitmap_info->bitmap) |
1790 | return -EAGAIN; | |
1791 | ||
6606bb97 JB |
1792 | /* |
1793 | * Ok the next item is a bitmap, but it may not actually hold | |
1794 | * the information for the rest of this free space stuff, so | |
1795 | * look for it, and if we don't find it return so we can try | |
1796 | * everything over again. | |
1797 | */ | |
1798 | search_start = *offset; | |
bdb7d303 | 1799 | search_bytes = ctl->unit; |
34d52cb6 | 1800 | ret = search_bitmap(ctl, bitmap_info, &search_start, |
6606bb97 JB |
1801 | &search_bytes); |
1802 | if (ret < 0 || search_start != *offset) | |
1803 | return -EAGAIN; | |
1804 | ||
96303081 | 1805 | goto again; |
edf6e2d1 | 1806 | } else if (!bitmap_info->bytes) |
34d52cb6 | 1807 | free_bitmap(ctl, bitmap_info); |
96303081 JB |
1808 | |
1809 | return 0; | |
1810 | } | |
1811 | ||
2cdc342c JB |
1812 | static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl, |
1813 | struct btrfs_free_space *info, u64 offset, | |
1814 | u64 bytes) | |
1815 | { | |
1816 | u64 bytes_to_set = 0; | |
1817 | u64 end; | |
1818 | ||
1819 | end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit); | |
1820 | ||
1821 | bytes_to_set = min(end - offset, bytes); | |
1822 | ||
1823 | bitmap_set_bits(ctl, info, offset, bytes_to_set); | |
1824 | ||
1825 | return bytes_to_set; | |
1826 | ||
1827 | } | |
1828 | ||
34d52cb6 LZ |
1829 | static bool use_bitmap(struct btrfs_free_space_ctl *ctl, |
1830 | struct btrfs_free_space *info) | |
96303081 | 1831 | { |
34d52cb6 | 1832 | struct btrfs_block_group_cache *block_group = ctl->private; |
96303081 JB |
1833 | |
1834 | /* | |
1835 | * If we are below the extents threshold then we can add this as an | |
1836 | * extent, and don't have to deal with the bitmap | |
1837 | */ | |
34d52cb6 | 1838 | if (ctl->free_extents < ctl->extents_thresh) { |
32cb0840 JB |
1839 | /* |
1840 | * If this block group has some small extents we don't want to | |
1841 | * use up all of our free slots in the cache with them, we want | |
1842 | * to reserve them to larger extents, however if we have plent | |
1843 | * of cache left then go ahead an dadd them, no sense in adding | |
1844 | * the overhead of a bitmap if we don't have to. | |
1845 | */ | |
1846 | if (info->bytes <= block_group->sectorsize * 4) { | |
34d52cb6 LZ |
1847 | if (ctl->free_extents * 2 <= ctl->extents_thresh) |
1848 | return false; | |
32cb0840 | 1849 | } else { |
34d52cb6 | 1850 | return false; |
32cb0840 JB |
1851 | } |
1852 | } | |
96303081 JB |
1853 | |
1854 | /* | |
dde5740f JB |
1855 | * The original block groups from mkfs can be really small, like 8 |
1856 | * megabytes, so don't bother with a bitmap for those entries. However | |
1857 | * some block groups can be smaller than what a bitmap would cover but | |
1858 | * are still large enough that they could overflow the 32k memory limit, | |
1859 | * so allow those block groups to still be allowed to have a bitmap | |
1860 | * entry. | |
96303081 | 1861 | */ |
dde5740f | 1862 | if (((BITS_PER_BITMAP * ctl->unit) >> 1) > block_group->key.offset) |
34d52cb6 LZ |
1863 | return false; |
1864 | ||
1865 | return true; | |
1866 | } | |
1867 | ||
2cdc342c JB |
1868 | static struct btrfs_free_space_op free_space_op = { |
1869 | .recalc_thresholds = recalculate_thresholds, | |
1870 | .use_bitmap = use_bitmap, | |
1871 | }; | |
1872 | ||
34d52cb6 LZ |
1873 | static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl, |
1874 | struct btrfs_free_space *info) | |
1875 | { | |
1876 | struct btrfs_free_space *bitmap_info; | |
2cdc342c | 1877 | struct btrfs_block_group_cache *block_group = NULL; |
34d52cb6 | 1878 | int added = 0; |
2cdc342c | 1879 | u64 bytes, offset, bytes_added; |
34d52cb6 | 1880 | int ret; |
96303081 JB |
1881 | |
1882 | bytes = info->bytes; | |
1883 | offset = info->offset; | |
1884 | ||
34d52cb6 LZ |
1885 | if (!ctl->op->use_bitmap(ctl, info)) |
1886 | return 0; | |
1887 | ||
2cdc342c JB |
1888 | if (ctl->op == &free_space_op) |
1889 | block_group = ctl->private; | |
38e87880 | 1890 | again: |
2cdc342c JB |
1891 | /* |
1892 | * Since we link bitmaps right into the cluster we need to see if we | |
1893 | * have a cluster here, and if so and it has our bitmap we need to add | |
1894 | * the free space to that bitmap. | |
1895 | */ | |
1896 | if (block_group && !list_empty(&block_group->cluster_list)) { | |
1897 | struct btrfs_free_cluster *cluster; | |
1898 | struct rb_node *node; | |
1899 | struct btrfs_free_space *entry; | |
1900 | ||
1901 | cluster = list_entry(block_group->cluster_list.next, | |
1902 | struct btrfs_free_cluster, | |
1903 | block_group_list); | |
1904 | spin_lock(&cluster->lock); | |
1905 | node = rb_first(&cluster->root); | |
1906 | if (!node) { | |
1907 | spin_unlock(&cluster->lock); | |
38e87880 | 1908 | goto no_cluster_bitmap; |
2cdc342c JB |
1909 | } |
1910 | ||
1911 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
1912 | if (!entry->bitmap) { | |
1913 | spin_unlock(&cluster->lock); | |
38e87880 | 1914 | goto no_cluster_bitmap; |
2cdc342c JB |
1915 | } |
1916 | ||
1917 | if (entry->offset == offset_to_bitmap(ctl, offset)) { | |
1918 | bytes_added = add_bytes_to_bitmap(ctl, entry, | |
1919 | offset, bytes); | |
1920 | bytes -= bytes_added; | |
1921 | offset += bytes_added; | |
1922 | } | |
1923 | spin_unlock(&cluster->lock); | |
1924 | if (!bytes) { | |
1925 | ret = 1; | |
1926 | goto out; | |
1927 | } | |
1928 | } | |
38e87880 CM |
1929 | |
1930 | no_cluster_bitmap: | |
34d52cb6 | 1931 | bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), |
96303081 JB |
1932 | 1, 0); |
1933 | if (!bitmap_info) { | |
b12d6869 | 1934 | ASSERT(added == 0); |
96303081 JB |
1935 | goto new_bitmap; |
1936 | } | |
1937 | ||
2cdc342c JB |
1938 | bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes); |
1939 | bytes -= bytes_added; | |
1940 | offset += bytes_added; | |
1941 | added = 0; | |
96303081 JB |
1942 | |
1943 | if (!bytes) { | |
1944 | ret = 1; | |
1945 | goto out; | |
1946 | } else | |
1947 | goto again; | |
1948 | ||
1949 | new_bitmap: | |
1950 | if (info && info->bitmap) { | |
34d52cb6 | 1951 | add_new_bitmap(ctl, info, offset); |
96303081 JB |
1952 | added = 1; |
1953 | info = NULL; | |
1954 | goto again; | |
1955 | } else { | |
34d52cb6 | 1956 | spin_unlock(&ctl->tree_lock); |
96303081 JB |
1957 | |
1958 | /* no pre-allocated info, allocate a new one */ | |
1959 | if (!info) { | |
dc89e982 JB |
1960 | info = kmem_cache_zalloc(btrfs_free_space_cachep, |
1961 | GFP_NOFS); | |
96303081 | 1962 | if (!info) { |
34d52cb6 | 1963 | spin_lock(&ctl->tree_lock); |
96303081 JB |
1964 | ret = -ENOMEM; |
1965 | goto out; | |
1966 | } | |
1967 | } | |
1968 | ||
1969 | /* allocate the bitmap */ | |
1970 | info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS); | |
34d52cb6 | 1971 | spin_lock(&ctl->tree_lock); |
96303081 JB |
1972 | if (!info->bitmap) { |
1973 | ret = -ENOMEM; | |
1974 | goto out; | |
1975 | } | |
1976 | goto again; | |
1977 | } | |
1978 | ||
1979 | out: | |
1980 | if (info) { | |
1981 | if (info->bitmap) | |
1982 | kfree(info->bitmap); | |
dc89e982 | 1983 | kmem_cache_free(btrfs_free_space_cachep, info); |
96303081 | 1984 | } |
0f9dd46c JB |
1985 | |
1986 | return ret; | |
1987 | } | |
1988 | ||
945d8962 | 1989 | static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl, |
f333adb5 | 1990 | struct btrfs_free_space *info, bool update_stat) |
0f9dd46c | 1991 | { |
120d66ee LZ |
1992 | struct btrfs_free_space *left_info; |
1993 | struct btrfs_free_space *right_info; | |
1994 | bool merged = false; | |
1995 | u64 offset = info->offset; | |
1996 | u64 bytes = info->bytes; | |
6226cb0a | 1997 | |
0f9dd46c JB |
1998 | /* |
1999 | * first we want to see if there is free space adjacent to the range we | |
2000 | * are adding, if there is remove that struct and add a new one to | |
2001 | * cover the entire range | |
2002 | */ | |
34d52cb6 | 2003 | right_info = tree_search_offset(ctl, offset + bytes, 0, 0); |
96303081 JB |
2004 | if (right_info && rb_prev(&right_info->offset_index)) |
2005 | left_info = rb_entry(rb_prev(&right_info->offset_index), | |
2006 | struct btrfs_free_space, offset_index); | |
2007 | else | |
34d52cb6 | 2008 | left_info = tree_search_offset(ctl, offset - 1, 0, 0); |
0f9dd46c | 2009 | |
96303081 | 2010 | if (right_info && !right_info->bitmap) { |
f333adb5 | 2011 | if (update_stat) |
34d52cb6 | 2012 | unlink_free_space(ctl, right_info); |
f333adb5 | 2013 | else |
34d52cb6 | 2014 | __unlink_free_space(ctl, right_info); |
6226cb0a | 2015 | info->bytes += right_info->bytes; |
dc89e982 | 2016 | kmem_cache_free(btrfs_free_space_cachep, right_info); |
120d66ee | 2017 | merged = true; |
0f9dd46c JB |
2018 | } |
2019 | ||
96303081 JB |
2020 | if (left_info && !left_info->bitmap && |
2021 | left_info->offset + left_info->bytes == offset) { | |
f333adb5 | 2022 | if (update_stat) |
34d52cb6 | 2023 | unlink_free_space(ctl, left_info); |
f333adb5 | 2024 | else |
34d52cb6 | 2025 | __unlink_free_space(ctl, left_info); |
6226cb0a JB |
2026 | info->offset = left_info->offset; |
2027 | info->bytes += left_info->bytes; | |
dc89e982 | 2028 | kmem_cache_free(btrfs_free_space_cachep, left_info); |
120d66ee | 2029 | merged = true; |
0f9dd46c JB |
2030 | } |
2031 | ||
120d66ee LZ |
2032 | return merged; |
2033 | } | |
2034 | ||
20005523 FM |
2035 | static bool steal_from_bitmap_to_end(struct btrfs_free_space_ctl *ctl, |
2036 | struct btrfs_free_space *info, | |
2037 | bool update_stat) | |
2038 | { | |
2039 | struct btrfs_free_space *bitmap; | |
2040 | unsigned long i; | |
2041 | unsigned long j; | |
2042 | const u64 end = info->offset + info->bytes; | |
2043 | const u64 bitmap_offset = offset_to_bitmap(ctl, end); | |
2044 | u64 bytes; | |
2045 | ||
2046 | bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0); | |
2047 | if (!bitmap) | |
2048 | return false; | |
2049 | ||
2050 | i = offset_to_bit(bitmap->offset, ctl->unit, end); | |
2051 | j = find_next_zero_bit(bitmap->bitmap, BITS_PER_BITMAP, i); | |
2052 | if (j == i) | |
2053 | return false; | |
2054 | bytes = (j - i) * ctl->unit; | |
2055 | info->bytes += bytes; | |
2056 | ||
2057 | if (update_stat) | |
2058 | bitmap_clear_bits(ctl, bitmap, end, bytes); | |
2059 | else | |
2060 | __bitmap_clear_bits(ctl, bitmap, end, bytes); | |
2061 | ||
2062 | if (!bitmap->bytes) | |
2063 | free_bitmap(ctl, bitmap); | |
2064 | ||
2065 | return true; | |
2066 | } | |
2067 | ||
2068 | static bool steal_from_bitmap_to_front(struct btrfs_free_space_ctl *ctl, | |
2069 | struct btrfs_free_space *info, | |
2070 | bool update_stat) | |
2071 | { | |
2072 | struct btrfs_free_space *bitmap; | |
2073 | u64 bitmap_offset; | |
2074 | unsigned long i; | |
2075 | unsigned long j; | |
2076 | unsigned long prev_j; | |
2077 | u64 bytes; | |
2078 | ||
2079 | bitmap_offset = offset_to_bitmap(ctl, info->offset); | |
2080 | /* If we're on a boundary, try the previous logical bitmap. */ | |
2081 | if (bitmap_offset == info->offset) { | |
2082 | if (info->offset == 0) | |
2083 | return false; | |
2084 | bitmap_offset = offset_to_bitmap(ctl, info->offset - 1); | |
2085 | } | |
2086 | ||
2087 | bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0); | |
2088 | if (!bitmap) | |
2089 | return false; | |
2090 | ||
2091 | i = offset_to_bit(bitmap->offset, ctl->unit, info->offset) - 1; | |
2092 | j = 0; | |
2093 | prev_j = (unsigned long)-1; | |
2094 | for_each_clear_bit_from(j, bitmap->bitmap, BITS_PER_BITMAP) { | |
2095 | if (j > i) | |
2096 | break; | |
2097 | prev_j = j; | |
2098 | } | |
2099 | if (prev_j == i) | |
2100 | return false; | |
2101 | ||
2102 | if (prev_j == (unsigned long)-1) | |
2103 | bytes = (i + 1) * ctl->unit; | |
2104 | else | |
2105 | bytes = (i - prev_j) * ctl->unit; | |
2106 | ||
2107 | info->offset -= bytes; | |
2108 | info->bytes += bytes; | |
2109 | ||
2110 | if (update_stat) | |
2111 | bitmap_clear_bits(ctl, bitmap, info->offset, bytes); | |
2112 | else | |
2113 | __bitmap_clear_bits(ctl, bitmap, info->offset, bytes); | |
2114 | ||
2115 | if (!bitmap->bytes) | |
2116 | free_bitmap(ctl, bitmap); | |
2117 | ||
2118 | return true; | |
2119 | } | |
2120 | ||
2121 | /* | |
2122 | * We prefer always to allocate from extent entries, both for clustered and | |
2123 | * non-clustered allocation requests. So when attempting to add a new extent | |
2124 | * entry, try to see if there's adjacent free space in bitmap entries, and if | |
2125 | * there is, migrate that space from the bitmaps to the extent. | |
2126 | * Like this we get better chances of satisfying space allocation requests | |
2127 | * because we attempt to satisfy them based on a single cache entry, and never | |
2128 | * on 2 or more entries - even if the entries represent a contiguous free space | |
2129 | * region (e.g. 1 extent entry + 1 bitmap entry starting where the extent entry | |
2130 | * ends). | |
2131 | */ | |
2132 | static void steal_from_bitmap(struct btrfs_free_space_ctl *ctl, | |
2133 | struct btrfs_free_space *info, | |
2134 | bool update_stat) | |
2135 | { | |
2136 | /* | |
2137 | * Only work with disconnected entries, as we can change their offset, | |
2138 | * and must be extent entries. | |
2139 | */ | |
2140 | ASSERT(!info->bitmap); | |
2141 | ASSERT(RB_EMPTY_NODE(&info->offset_index)); | |
2142 | ||
2143 | if (ctl->total_bitmaps > 0) { | |
2144 | bool stole_end; | |
2145 | bool stole_front = false; | |
2146 | ||
2147 | stole_end = steal_from_bitmap_to_end(ctl, info, update_stat); | |
2148 | if (ctl->total_bitmaps > 0) | |
2149 | stole_front = steal_from_bitmap_to_front(ctl, info, | |
2150 | update_stat); | |
2151 | ||
2152 | if (stole_end || stole_front) | |
2153 | try_merge_free_space(ctl, info, update_stat); | |
2154 | } | |
2155 | } | |
2156 | ||
581bb050 LZ |
2157 | int __btrfs_add_free_space(struct btrfs_free_space_ctl *ctl, |
2158 | u64 offset, u64 bytes) | |
120d66ee LZ |
2159 | { |
2160 | struct btrfs_free_space *info; | |
2161 | int ret = 0; | |
2162 | ||
dc89e982 | 2163 | info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS); |
120d66ee LZ |
2164 | if (!info) |
2165 | return -ENOMEM; | |
2166 | ||
2167 | info->offset = offset; | |
2168 | info->bytes = bytes; | |
20005523 | 2169 | RB_CLEAR_NODE(&info->offset_index); |
120d66ee | 2170 | |
34d52cb6 | 2171 | spin_lock(&ctl->tree_lock); |
120d66ee | 2172 | |
34d52cb6 | 2173 | if (try_merge_free_space(ctl, info, true)) |
120d66ee LZ |
2174 | goto link; |
2175 | ||
2176 | /* | |
2177 | * There was no extent directly to the left or right of this new | |
2178 | * extent then we know we're going to have to allocate a new extent, so | |
2179 | * before we do that see if we need to drop this into a bitmap | |
2180 | */ | |
34d52cb6 | 2181 | ret = insert_into_bitmap(ctl, info); |
120d66ee LZ |
2182 | if (ret < 0) { |
2183 | goto out; | |
2184 | } else if (ret) { | |
2185 | ret = 0; | |
2186 | goto out; | |
2187 | } | |
2188 | link: | |
20005523 FM |
2189 | /* |
2190 | * Only steal free space from adjacent bitmaps if we're sure we're not | |
2191 | * going to add the new free space to existing bitmap entries - because | |
2192 | * that would mean unnecessary work that would be reverted. Therefore | |
2193 | * attempt to steal space from bitmaps if we're adding an extent entry. | |
2194 | */ | |
2195 | steal_from_bitmap(ctl, info, true); | |
2196 | ||
34d52cb6 | 2197 | ret = link_free_space(ctl, info); |
0f9dd46c | 2198 | if (ret) |
dc89e982 | 2199 | kmem_cache_free(btrfs_free_space_cachep, info); |
96303081 | 2200 | out: |
34d52cb6 | 2201 | spin_unlock(&ctl->tree_lock); |
6226cb0a | 2202 | |
0f9dd46c | 2203 | if (ret) { |
efe120a0 | 2204 | printk(KERN_CRIT "BTRFS: unable to add free space :%d\n", ret); |
b12d6869 | 2205 | ASSERT(ret != -EEXIST); |
0f9dd46c JB |
2206 | } |
2207 | ||
0f9dd46c JB |
2208 | return ret; |
2209 | } | |
2210 | ||
6226cb0a JB |
2211 | int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group, |
2212 | u64 offset, u64 bytes) | |
0f9dd46c | 2213 | { |
34d52cb6 | 2214 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0f9dd46c | 2215 | struct btrfs_free_space *info; |
b0175117 JB |
2216 | int ret; |
2217 | bool re_search = false; | |
0f9dd46c | 2218 | |
34d52cb6 | 2219 | spin_lock(&ctl->tree_lock); |
6226cb0a | 2220 | |
96303081 | 2221 | again: |
b0175117 | 2222 | ret = 0; |
bdb7d303 JB |
2223 | if (!bytes) |
2224 | goto out_lock; | |
2225 | ||
34d52cb6 | 2226 | info = tree_search_offset(ctl, offset, 0, 0); |
96303081 | 2227 | if (!info) { |
6606bb97 JB |
2228 | /* |
2229 | * oops didn't find an extent that matched the space we wanted | |
2230 | * to remove, look for a bitmap instead | |
2231 | */ | |
34d52cb6 | 2232 | info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), |
6606bb97 JB |
2233 | 1, 0); |
2234 | if (!info) { | |
b0175117 JB |
2235 | /* |
2236 | * If we found a partial bit of our free space in a | |
2237 | * bitmap but then couldn't find the other part this may | |
2238 | * be a problem, so WARN about it. | |
24a70313 | 2239 | */ |
b0175117 | 2240 | WARN_ON(re_search); |
6606bb97 JB |
2241 | goto out_lock; |
2242 | } | |
96303081 JB |
2243 | } |
2244 | ||
b0175117 | 2245 | re_search = false; |
bdb7d303 | 2246 | if (!info->bitmap) { |
34d52cb6 | 2247 | unlink_free_space(ctl, info); |
bdb7d303 JB |
2248 | if (offset == info->offset) { |
2249 | u64 to_free = min(bytes, info->bytes); | |
2250 | ||
2251 | info->bytes -= to_free; | |
2252 | info->offset += to_free; | |
2253 | if (info->bytes) { | |
2254 | ret = link_free_space(ctl, info); | |
2255 | WARN_ON(ret); | |
2256 | } else { | |
2257 | kmem_cache_free(btrfs_free_space_cachep, info); | |
2258 | } | |
0f9dd46c | 2259 | |
bdb7d303 JB |
2260 | offset += to_free; |
2261 | bytes -= to_free; | |
2262 | goto again; | |
2263 | } else { | |
2264 | u64 old_end = info->bytes + info->offset; | |
9b49c9b9 | 2265 | |
bdb7d303 | 2266 | info->bytes = offset - info->offset; |
34d52cb6 | 2267 | ret = link_free_space(ctl, info); |
96303081 JB |
2268 | WARN_ON(ret); |
2269 | if (ret) | |
2270 | goto out_lock; | |
96303081 | 2271 | |
bdb7d303 JB |
2272 | /* Not enough bytes in this entry to satisfy us */ |
2273 | if (old_end < offset + bytes) { | |
2274 | bytes -= old_end - offset; | |
2275 | offset = old_end; | |
2276 | goto again; | |
2277 | } else if (old_end == offset + bytes) { | |
2278 | /* all done */ | |
2279 | goto out_lock; | |
2280 | } | |
2281 | spin_unlock(&ctl->tree_lock); | |
2282 | ||
2283 | ret = btrfs_add_free_space(block_group, offset + bytes, | |
2284 | old_end - (offset + bytes)); | |
2285 | WARN_ON(ret); | |
2286 | goto out; | |
2287 | } | |
0f9dd46c | 2288 | } |
96303081 | 2289 | |
34d52cb6 | 2290 | ret = remove_from_bitmap(ctl, info, &offset, &bytes); |
b0175117 JB |
2291 | if (ret == -EAGAIN) { |
2292 | re_search = true; | |
96303081 | 2293 | goto again; |
b0175117 | 2294 | } |
96303081 | 2295 | out_lock: |
34d52cb6 | 2296 | spin_unlock(&ctl->tree_lock); |
0f9dd46c | 2297 | out: |
25179201 JB |
2298 | return ret; |
2299 | } | |
2300 | ||
0f9dd46c JB |
2301 | void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group, |
2302 | u64 bytes) | |
2303 | { | |
34d52cb6 | 2304 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0f9dd46c JB |
2305 | struct btrfs_free_space *info; |
2306 | struct rb_node *n; | |
2307 | int count = 0; | |
2308 | ||
34d52cb6 | 2309 | for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) { |
0f9dd46c | 2310 | info = rb_entry(n, struct btrfs_free_space, offset_index); |
f6175efa | 2311 | if (info->bytes >= bytes && !block_group->ro) |
0f9dd46c | 2312 | count++; |
efe120a0 FH |
2313 | btrfs_crit(block_group->fs_info, |
2314 | "entry offset %llu, bytes %llu, bitmap %s", | |
2315 | info->offset, info->bytes, | |
96303081 | 2316 | (info->bitmap) ? "yes" : "no"); |
0f9dd46c | 2317 | } |
efe120a0 | 2318 | btrfs_info(block_group->fs_info, "block group has cluster?: %s", |
96303081 | 2319 | list_empty(&block_group->cluster_list) ? "no" : "yes"); |
efe120a0 FH |
2320 | btrfs_info(block_group->fs_info, |
2321 | "%d blocks of free space at or bigger than bytes is", count); | |
0f9dd46c JB |
2322 | } |
2323 | ||
34d52cb6 | 2324 | void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group) |
0f9dd46c | 2325 | { |
34d52cb6 | 2326 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0f9dd46c | 2327 | |
34d52cb6 LZ |
2328 | spin_lock_init(&ctl->tree_lock); |
2329 | ctl->unit = block_group->sectorsize; | |
2330 | ctl->start = block_group->key.objectid; | |
2331 | ctl->private = block_group; | |
2332 | ctl->op = &free_space_op; | |
55507ce3 FM |
2333 | INIT_LIST_HEAD(&ctl->trimming_ranges); |
2334 | mutex_init(&ctl->cache_writeout_mutex); | |
0f9dd46c | 2335 | |
34d52cb6 LZ |
2336 | /* |
2337 | * we only want to have 32k of ram per block group for keeping | |
2338 | * track of free space, and if we pass 1/2 of that we want to | |
2339 | * start converting things over to using bitmaps | |
2340 | */ | |
2341 | ctl->extents_thresh = ((1024 * 32) / 2) / | |
2342 | sizeof(struct btrfs_free_space); | |
0f9dd46c JB |
2343 | } |
2344 | ||
fa9c0d79 CM |
2345 | /* |
2346 | * for a given cluster, put all of its extents back into the free | |
2347 | * space cache. If the block group passed doesn't match the block group | |
2348 | * pointed to by the cluster, someone else raced in and freed the | |
2349 | * cluster already. In that case, we just return without changing anything | |
2350 | */ | |
2351 | static int | |
2352 | __btrfs_return_cluster_to_free_space( | |
2353 | struct btrfs_block_group_cache *block_group, | |
2354 | struct btrfs_free_cluster *cluster) | |
2355 | { | |
34d52cb6 | 2356 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
fa9c0d79 CM |
2357 | struct btrfs_free_space *entry; |
2358 | struct rb_node *node; | |
2359 | ||
2360 | spin_lock(&cluster->lock); | |
2361 | if (cluster->block_group != block_group) | |
2362 | goto out; | |
2363 | ||
96303081 | 2364 | cluster->block_group = NULL; |
fa9c0d79 | 2365 | cluster->window_start = 0; |
96303081 | 2366 | list_del_init(&cluster->block_group_list); |
96303081 | 2367 | |
fa9c0d79 | 2368 | node = rb_first(&cluster->root); |
96303081 | 2369 | while (node) { |
4e69b598 JB |
2370 | bool bitmap; |
2371 | ||
fa9c0d79 CM |
2372 | entry = rb_entry(node, struct btrfs_free_space, offset_index); |
2373 | node = rb_next(&entry->offset_index); | |
2374 | rb_erase(&entry->offset_index, &cluster->root); | |
20005523 | 2375 | RB_CLEAR_NODE(&entry->offset_index); |
4e69b598 JB |
2376 | |
2377 | bitmap = (entry->bitmap != NULL); | |
20005523 | 2378 | if (!bitmap) { |
34d52cb6 | 2379 | try_merge_free_space(ctl, entry, false); |
20005523 FM |
2380 | steal_from_bitmap(ctl, entry, false); |
2381 | } | |
34d52cb6 | 2382 | tree_insert_offset(&ctl->free_space_offset, |
4e69b598 | 2383 | entry->offset, &entry->offset_index, bitmap); |
fa9c0d79 | 2384 | } |
6bef4d31 | 2385 | cluster->root = RB_ROOT; |
96303081 | 2386 | |
fa9c0d79 CM |
2387 | out: |
2388 | spin_unlock(&cluster->lock); | |
96303081 | 2389 | btrfs_put_block_group(block_group); |
fa9c0d79 CM |
2390 | return 0; |
2391 | } | |
2392 | ||
48a3b636 ES |
2393 | static void __btrfs_remove_free_space_cache_locked( |
2394 | struct btrfs_free_space_ctl *ctl) | |
0f9dd46c JB |
2395 | { |
2396 | struct btrfs_free_space *info; | |
2397 | struct rb_node *node; | |
581bb050 | 2398 | |
581bb050 LZ |
2399 | while ((node = rb_last(&ctl->free_space_offset)) != NULL) { |
2400 | info = rb_entry(node, struct btrfs_free_space, offset_index); | |
9b90f513 JB |
2401 | if (!info->bitmap) { |
2402 | unlink_free_space(ctl, info); | |
2403 | kmem_cache_free(btrfs_free_space_cachep, info); | |
2404 | } else { | |
2405 | free_bitmap(ctl, info); | |
2406 | } | |
351810c1 DS |
2407 | |
2408 | cond_resched_lock(&ctl->tree_lock); | |
581bb050 | 2409 | } |
09655373 CM |
2410 | } |
2411 | ||
2412 | void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl) | |
2413 | { | |
2414 | spin_lock(&ctl->tree_lock); | |
2415 | __btrfs_remove_free_space_cache_locked(ctl); | |
581bb050 LZ |
2416 | spin_unlock(&ctl->tree_lock); |
2417 | } | |
2418 | ||
2419 | void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group) | |
2420 | { | |
2421 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
fa9c0d79 | 2422 | struct btrfs_free_cluster *cluster; |
96303081 | 2423 | struct list_head *head; |
0f9dd46c | 2424 | |
34d52cb6 | 2425 | spin_lock(&ctl->tree_lock); |
96303081 JB |
2426 | while ((head = block_group->cluster_list.next) != |
2427 | &block_group->cluster_list) { | |
2428 | cluster = list_entry(head, struct btrfs_free_cluster, | |
2429 | block_group_list); | |
fa9c0d79 CM |
2430 | |
2431 | WARN_ON(cluster->block_group != block_group); | |
2432 | __btrfs_return_cluster_to_free_space(block_group, cluster); | |
351810c1 DS |
2433 | |
2434 | cond_resched_lock(&ctl->tree_lock); | |
fa9c0d79 | 2435 | } |
09655373 | 2436 | __btrfs_remove_free_space_cache_locked(ctl); |
34d52cb6 | 2437 | spin_unlock(&ctl->tree_lock); |
fa9c0d79 | 2438 | |
0f9dd46c JB |
2439 | } |
2440 | ||
6226cb0a | 2441 | u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group, |
a4820398 MX |
2442 | u64 offset, u64 bytes, u64 empty_size, |
2443 | u64 *max_extent_size) | |
0f9dd46c | 2444 | { |
34d52cb6 | 2445 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
6226cb0a | 2446 | struct btrfs_free_space *entry = NULL; |
96303081 | 2447 | u64 bytes_search = bytes + empty_size; |
6226cb0a | 2448 | u64 ret = 0; |
53b381b3 DW |
2449 | u64 align_gap = 0; |
2450 | u64 align_gap_len = 0; | |
0f9dd46c | 2451 | |
34d52cb6 | 2452 | spin_lock(&ctl->tree_lock); |
53b381b3 | 2453 | entry = find_free_space(ctl, &offset, &bytes_search, |
a4820398 | 2454 | block_group->full_stripe_len, max_extent_size); |
6226cb0a | 2455 | if (!entry) |
96303081 JB |
2456 | goto out; |
2457 | ||
2458 | ret = offset; | |
2459 | if (entry->bitmap) { | |
34d52cb6 | 2460 | bitmap_clear_bits(ctl, entry, offset, bytes); |
edf6e2d1 | 2461 | if (!entry->bytes) |
34d52cb6 | 2462 | free_bitmap(ctl, entry); |
96303081 | 2463 | } else { |
34d52cb6 | 2464 | unlink_free_space(ctl, entry); |
53b381b3 DW |
2465 | align_gap_len = offset - entry->offset; |
2466 | align_gap = entry->offset; | |
2467 | ||
2468 | entry->offset = offset + bytes; | |
2469 | WARN_ON(entry->bytes < bytes + align_gap_len); | |
2470 | ||
2471 | entry->bytes -= bytes + align_gap_len; | |
6226cb0a | 2472 | if (!entry->bytes) |
dc89e982 | 2473 | kmem_cache_free(btrfs_free_space_cachep, entry); |
6226cb0a | 2474 | else |
34d52cb6 | 2475 | link_free_space(ctl, entry); |
6226cb0a | 2476 | } |
96303081 | 2477 | out: |
34d52cb6 | 2478 | spin_unlock(&ctl->tree_lock); |
817d52f8 | 2479 | |
53b381b3 DW |
2480 | if (align_gap_len) |
2481 | __btrfs_add_free_space(ctl, align_gap, align_gap_len); | |
0f9dd46c JB |
2482 | return ret; |
2483 | } | |
fa9c0d79 CM |
2484 | |
2485 | /* | |
2486 | * given a cluster, put all of its extents back into the free space | |
2487 | * cache. If a block group is passed, this function will only free | |
2488 | * a cluster that belongs to the passed block group. | |
2489 | * | |
2490 | * Otherwise, it'll get a reference on the block group pointed to by the | |
2491 | * cluster and remove the cluster from it. | |
2492 | */ | |
2493 | int btrfs_return_cluster_to_free_space( | |
2494 | struct btrfs_block_group_cache *block_group, | |
2495 | struct btrfs_free_cluster *cluster) | |
2496 | { | |
34d52cb6 | 2497 | struct btrfs_free_space_ctl *ctl; |
fa9c0d79 CM |
2498 | int ret; |
2499 | ||
2500 | /* first, get a safe pointer to the block group */ | |
2501 | spin_lock(&cluster->lock); | |
2502 | if (!block_group) { | |
2503 | block_group = cluster->block_group; | |
2504 | if (!block_group) { | |
2505 | spin_unlock(&cluster->lock); | |
2506 | return 0; | |
2507 | } | |
2508 | } else if (cluster->block_group != block_group) { | |
2509 | /* someone else has already freed it don't redo their work */ | |
2510 | spin_unlock(&cluster->lock); | |
2511 | return 0; | |
2512 | } | |
2513 | atomic_inc(&block_group->count); | |
2514 | spin_unlock(&cluster->lock); | |
2515 | ||
34d52cb6 LZ |
2516 | ctl = block_group->free_space_ctl; |
2517 | ||
fa9c0d79 | 2518 | /* now return any extents the cluster had on it */ |
34d52cb6 | 2519 | spin_lock(&ctl->tree_lock); |
fa9c0d79 | 2520 | ret = __btrfs_return_cluster_to_free_space(block_group, cluster); |
34d52cb6 | 2521 | spin_unlock(&ctl->tree_lock); |
fa9c0d79 CM |
2522 | |
2523 | /* finally drop our ref */ | |
2524 | btrfs_put_block_group(block_group); | |
2525 | return ret; | |
2526 | } | |
2527 | ||
96303081 JB |
2528 | static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group, |
2529 | struct btrfs_free_cluster *cluster, | |
4e69b598 | 2530 | struct btrfs_free_space *entry, |
a4820398 MX |
2531 | u64 bytes, u64 min_start, |
2532 | u64 *max_extent_size) | |
96303081 | 2533 | { |
34d52cb6 | 2534 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
96303081 JB |
2535 | int err; |
2536 | u64 search_start = cluster->window_start; | |
2537 | u64 search_bytes = bytes; | |
2538 | u64 ret = 0; | |
2539 | ||
96303081 JB |
2540 | search_start = min_start; |
2541 | search_bytes = bytes; | |
2542 | ||
34d52cb6 | 2543 | err = search_bitmap(ctl, entry, &search_start, &search_bytes); |
a4820398 MX |
2544 | if (err) { |
2545 | if (search_bytes > *max_extent_size) | |
2546 | *max_extent_size = search_bytes; | |
4e69b598 | 2547 | return 0; |
a4820398 | 2548 | } |
96303081 JB |
2549 | |
2550 | ret = search_start; | |
bb3ac5a4 | 2551 | __bitmap_clear_bits(ctl, entry, ret, bytes); |
96303081 JB |
2552 | |
2553 | return ret; | |
2554 | } | |
2555 | ||
fa9c0d79 CM |
2556 | /* |
2557 | * given a cluster, try to allocate 'bytes' from it, returns 0 | |
2558 | * if it couldn't find anything suitably large, or a logical disk offset | |
2559 | * if things worked out | |
2560 | */ | |
2561 | u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group, | |
2562 | struct btrfs_free_cluster *cluster, u64 bytes, | |
a4820398 | 2563 | u64 min_start, u64 *max_extent_size) |
fa9c0d79 | 2564 | { |
34d52cb6 | 2565 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
fa9c0d79 CM |
2566 | struct btrfs_free_space *entry = NULL; |
2567 | struct rb_node *node; | |
2568 | u64 ret = 0; | |
2569 | ||
2570 | spin_lock(&cluster->lock); | |
2571 | if (bytes > cluster->max_size) | |
2572 | goto out; | |
2573 | ||
2574 | if (cluster->block_group != block_group) | |
2575 | goto out; | |
2576 | ||
2577 | node = rb_first(&cluster->root); | |
2578 | if (!node) | |
2579 | goto out; | |
2580 | ||
2581 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
67871254 | 2582 | while (1) { |
a4820398 MX |
2583 | if (entry->bytes < bytes && entry->bytes > *max_extent_size) |
2584 | *max_extent_size = entry->bytes; | |
2585 | ||
4e69b598 JB |
2586 | if (entry->bytes < bytes || |
2587 | (!entry->bitmap && entry->offset < min_start)) { | |
fa9c0d79 CM |
2588 | node = rb_next(&entry->offset_index); |
2589 | if (!node) | |
2590 | break; | |
2591 | entry = rb_entry(node, struct btrfs_free_space, | |
2592 | offset_index); | |
2593 | continue; | |
2594 | } | |
fa9c0d79 | 2595 | |
4e69b598 JB |
2596 | if (entry->bitmap) { |
2597 | ret = btrfs_alloc_from_bitmap(block_group, | |
2598 | cluster, entry, bytes, | |
a4820398 MX |
2599 | cluster->window_start, |
2600 | max_extent_size); | |
4e69b598 | 2601 | if (ret == 0) { |
4e69b598 JB |
2602 | node = rb_next(&entry->offset_index); |
2603 | if (!node) | |
2604 | break; | |
2605 | entry = rb_entry(node, struct btrfs_free_space, | |
2606 | offset_index); | |
2607 | continue; | |
2608 | } | |
9b230628 | 2609 | cluster->window_start += bytes; |
4e69b598 | 2610 | } else { |
4e69b598 JB |
2611 | ret = entry->offset; |
2612 | ||
2613 | entry->offset += bytes; | |
2614 | entry->bytes -= bytes; | |
2615 | } | |
fa9c0d79 | 2616 | |
5e71b5d5 | 2617 | if (entry->bytes == 0) |
fa9c0d79 | 2618 | rb_erase(&entry->offset_index, &cluster->root); |
fa9c0d79 CM |
2619 | break; |
2620 | } | |
2621 | out: | |
2622 | spin_unlock(&cluster->lock); | |
96303081 | 2623 | |
5e71b5d5 LZ |
2624 | if (!ret) |
2625 | return 0; | |
2626 | ||
34d52cb6 | 2627 | spin_lock(&ctl->tree_lock); |
5e71b5d5 | 2628 | |
34d52cb6 | 2629 | ctl->free_space -= bytes; |
5e71b5d5 | 2630 | if (entry->bytes == 0) { |
34d52cb6 | 2631 | ctl->free_extents--; |
4e69b598 JB |
2632 | if (entry->bitmap) { |
2633 | kfree(entry->bitmap); | |
34d52cb6 LZ |
2634 | ctl->total_bitmaps--; |
2635 | ctl->op->recalc_thresholds(ctl); | |
4e69b598 | 2636 | } |
dc89e982 | 2637 | kmem_cache_free(btrfs_free_space_cachep, entry); |
5e71b5d5 LZ |
2638 | } |
2639 | ||
34d52cb6 | 2640 | spin_unlock(&ctl->tree_lock); |
5e71b5d5 | 2641 | |
fa9c0d79 CM |
2642 | return ret; |
2643 | } | |
2644 | ||
96303081 JB |
2645 | static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group, |
2646 | struct btrfs_free_space *entry, | |
2647 | struct btrfs_free_cluster *cluster, | |
1bb91902 AO |
2648 | u64 offset, u64 bytes, |
2649 | u64 cont1_bytes, u64 min_bytes) | |
96303081 | 2650 | { |
34d52cb6 | 2651 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
96303081 JB |
2652 | unsigned long next_zero; |
2653 | unsigned long i; | |
1bb91902 AO |
2654 | unsigned long want_bits; |
2655 | unsigned long min_bits; | |
96303081 JB |
2656 | unsigned long found_bits; |
2657 | unsigned long start = 0; | |
2658 | unsigned long total_found = 0; | |
4e69b598 | 2659 | int ret; |
96303081 | 2660 | |
96009762 | 2661 | i = offset_to_bit(entry->offset, ctl->unit, |
96303081 | 2662 | max_t(u64, offset, entry->offset)); |
96009762 WSH |
2663 | want_bits = bytes_to_bits(bytes, ctl->unit); |
2664 | min_bits = bytes_to_bits(min_bytes, ctl->unit); | |
96303081 JB |
2665 | |
2666 | again: | |
2667 | found_bits = 0; | |
ebb3dad4 | 2668 | for_each_set_bit_from(i, entry->bitmap, BITS_PER_BITMAP) { |
96303081 JB |
2669 | next_zero = find_next_zero_bit(entry->bitmap, |
2670 | BITS_PER_BITMAP, i); | |
1bb91902 | 2671 | if (next_zero - i >= min_bits) { |
96303081 JB |
2672 | found_bits = next_zero - i; |
2673 | break; | |
2674 | } | |
2675 | i = next_zero; | |
2676 | } | |
2677 | ||
2678 | if (!found_bits) | |
4e69b598 | 2679 | return -ENOSPC; |
96303081 | 2680 | |
1bb91902 | 2681 | if (!total_found) { |
96303081 | 2682 | start = i; |
b78d09bc | 2683 | cluster->max_size = 0; |
96303081 JB |
2684 | } |
2685 | ||
2686 | total_found += found_bits; | |
2687 | ||
96009762 WSH |
2688 | if (cluster->max_size < found_bits * ctl->unit) |
2689 | cluster->max_size = found_bits * ctl->unit; | |
96303081 | 2690 | |
1bb91902 AO |
2691 | if (total_found < want_bits || cluster->max_size < cont1_bytes) { |
2692 | i = next_zero + 1; | |
96303081 JB |
2693 | goto again; |
2694 | } | |
2695 | ||
96009762 | 2696 | cluster->window_start = start * ctl->unit + entry->offset; |
34d52cb6 | 2697 | rb_erase(&entry->offset_index, &ctl->free_space_offset); |
4e69b598 JB |
2698 | ret = tree_insert_offset(&cluster->root, entry->offset, |
2699 | &entry->offset_index, 1); | |
b12d6869 | 2700 | ASSERT(!ret); /* -EEXIST; Logic error */ |
96303081 | 2701 | |
3f7de037 | 2702 | trace_btrfs_setup_cluster(block_group, cluster, |
96009762 | 2703 | total_found * ctl->unit, 1); |
96303081 JB |
2704 | return 0; |
2705 | } | |
2706 | ||
4e69b598 JB |
2707 | /* |
2708 | * This searches the block group for just extents to fill the cluster with. | |
1bb91902 AO |
2709 | * Try to find a cluster with at least bytes total bytes, at least one |
2710 | * extent of cont1_bytes, and other clusters of at least min_bytes. | |
4e69b598 | 2711 | */ |
3de85bb9 JB |
2712 | static noinline int |
2713 | setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group, | |
2714 | struct btrfs_free_cluster *cluster, | |
2715 | struct list_head *bitmaps, u64 offset, u64 bytes, | |
1bb91902 | 2716 | u64 cont1_bytes, u64 min_bytes) |
4e69b598 | 2717 | { |
34d52cb6 | 2718 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
4e69b598 JB |
2719 | struct btrfs_free_space *first = NULL; |
2720 | struct btrfs_free_space *entry = NULL; | |
4e69b598 JB |
2721 | struct btrfs_free_space *last; |
2722 | struct rb_node *node; | |
4e69b598 JB |
2723 | u64 window_free; |
2724 | u64 max_extent; | |
3f7de037 | 2725 | u64 total_size = 0; |
4e69b598 | 2726 | |
34d52cb6 | 2727 | entry = tree_search_offset(ctl, offset, 0, 1); |
4e69b598 JB |
2728 | if (!entry) |
2729 | return -ENOSPC; | |
2730 | ||
2731 | /* | |
2732 | * We don't want bitmaps, so just move along until we find a normal | |
2733 | * extent entry. | |
2734 | */ | |
1bb91902 AO |
2735 | while (entry->bitmap || entry->bytes < min_bytes) { |
2736 | if (entry->bitmap && list_empty(&entry->list)) | |
86d4a77b | 2737 | list_add_tail(&entry->list, bitmaps); |
4e69b598 JB |
2738 | node = rb_next(&entry->offset_index); |
2739 | if (!node) | |
2740 | return -ENOSPC; | |
2741 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
2742 | } | |
2743 | ||
4e69b598 JB |
2744 | window_free = entry->bytes; |
2745 | max_extent = entry->bytes; | |
2746 | first = entry; | |
2747 | last = entry; | |
4e69b598 | 2748 | |
1bb91902 AO |
2749 | for (node = rb_next(&entry->offset_index); node; |
2750 | node = rb_next(&entry->offset_index)) { | |
4e69b598 JB |
2751 | entry = rb_entry(node, struct btrfs_free_space, offset_index); |
2752 | ||
86d4a77b JB |
2753 | if (entry->bitmap) { |
2754 | if (list_empty(&entry->list)) | |
2755 | list_add_tail(&entry->list, bitmaps); | |
4e69b598 | 2756 | continue; |
86d4a77b JB |
2757 | } |
2758 | ||
1bb91902 AO |
2759 | if (entry->bytes < min_bytes) |
2760 | continue; | |
2761 | ||
2762 | last = entry; | |
2763 | window_free += entry->bytes; | |
2764 | if (entry->bytes > max_extent) | |
4e69b598 | 2765 | max_extent = entry->bytes; |
4e69b598 JB |
2766 | } |
2767 | ||
1bb91902 AO |
2768 | if (window_free < bytes || max_extent < cont1_bytes) |
2769 | return -ENOSPC; | |
2770 | ||
4e69b598 JB |
2771 | cluster->window_start = first->offset; |
2772 | ||
2773 | node = &first->offset_index; | |
2774 | ||
2775 | /* | |
2776 | * now we've found our entries, pull them out of the free space | |
2777 | * cache and put them into the cluster rbtree | |
2778 | */ | |
2779 | do { | |
2780 | int ret; | |
2781 | ||
2782 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
2783 | node = rb_next(&entry->offset_index); | |
1bb91902 | 2784 | if (entry->bitmap || entry->bytes < min_bytes) |
4e69b598 JB |
2785 | continue; |
2786 | ||
34d52cb6 | 2787 | rb_erase(&entry->offset_index, &ctl->free_space_offset); |
4e69b598 JB |
2788 | ret = tree_insert_offset(&cluster->root, entry->offset, |
2789 | &entry->offset_index, 0); | |
3f7de037 | 2790 | total_size += entry->bytes; |
b12d6869 | 2791 | ASSERT(!ret); /* -EEXIST; Logic error */ |
4e69b598 JB |
2792 | } while (node && entry != last); |
2793 | ||
2794 | cluster->max_size = max_extent; | |
3f7de037 | 2795 | trace_btrfs_setup_cluster(block_group, cluster, total_size, 0); |
4e69b598 JB |
2796 | return 0; |
2797 | } | |
2798 | ||
2799 | /* | |
2800 | * This specifically looks for bitmaps that may work in the cluster, we assume | |
2801 | * that we have already failed to find extents that will work. | |
2802 | */ | |
3de85bb9 JB |
2803 | static noinline int |
2804 | setup_cluster_bitmap(struct btrfs_block_group_cache *block_group, | |
2805 | struct btrfs_free_cluster *cluster, | |
2806 | struct list_head *bitmaps, u64 offset, u64 bytes, | |
1bb91902 | 2807 | u64 cont1_bytes, u64 min_bytes) |
4e69b598 | 2808 | { |
34d52cb6 | 2809 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
4e69b598 | 2810 | struct btrfs_free_space *entry; |
4e69b598 | 2811 | int ret = -ENOSPC; |
0f0fbf1d | 2812 | u64 bitmap_offset = offset_to_bitmap(ctl, offset); |
4e69b598 | 2813 | |
34d52cb6 | 2814 | if (ctl->total_bitmaps == 0) |
4e69b598 JB |
2815 | return -ENOSPC; |
2816 | ||
0f0fbf1d LZ |
2817 | /* |
2818 | * The bitmap that covers offset won't be in the list unless offset | |
2819 | * is just its start offset. | |
2820 | */ | |
2821 | entry = list_first_entry(bitmaps, struct btrfs_free_space, list); | |
2822 | if (entry->offset != bitmap_offset) { | |
2823 | entry = tree_search_offset(ctl, bitmap_offset, 1, 0); | |
2824 | if (entry && list_empty(&entry->list)) | |
2825 | list_add(&entry->list, bitmaps); | |
2826 | } | |
2827 | ||
86d4a77b | 2828 | list_for_each_entry(entry, bitmaps, list) { |
357b9784 | 2829 | if (entry->bytes < bytes) |
86d4a77b JB |
2830 | continue; |
2831 | ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset, | |
1bb91902 | 2832 | bytes, cont1_bytes, min_bytes); |
86d4a77b JB |
2833 | if (!ret) |
2834 | return 0; | |
2835 | } | |
2836 | ||
2837 | /* | |
52621cb6 LZ |
2838 | * The bitmaps list has all the bitmaps that record free space |
2839 | * starting after offset, so no more search is required. | |
86d4a77b | 2840 | */ |
52621cb6 | 2841 | return -ENOSPC; |
4e69b598 JB |
2842 | } |
2843 | ||
fa9c0d79 CM |
2844 | /* |
2845 | * here we try to find a cluster of blocks in a block group. The goal | |
1bb91902 | 2846 | * is to find at least bytes+empty_size. |
fa9c0d79 CM |
2847 | * We might not find them all in one contiguous area. |
2848 | * | |
2849 | * returns zero and sets up cluster if things worked out, otherwise | |
2850 | * it returns -enospc | |
2851 | */ | |
00361589 | 2852 | int btrfs_find_space_cluster(struct btrfs_root *root, |
fa9c0d79 CM |
2853 | struct btrfs_block_group_cache *block_group, |
2854 | struct btrfs_free_cluster *cluster, | |
2855 | u64 offset, u64 bytes, u64 empty_size) | |
2856 | { | |
34d52cb6 | 2857 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
86d4a77b | 2858 | struct btrfs_free_space *entry, *tmp; |
52621cb6 | 2859 | LIST_HEAD(bitmaps); |
fa9c0d79 | 2860 | u64 min_bytes; |
1bb91902 | 2861 | u64 cont1_bytes; |
fa9c0d79 CM |
2862 | int ret; |
2863 | ||
1bb91902 AO |
2864 | /* |
2865 | * Choose the minimum extent size we'll require for this | |
2866 | * cluster. For SSD_SPREAD, don't allow any fragmentation. | |
2867 | * For metadata, allow allocates with smaller extents. For | |
2868 | * data, keep it dense. | |
2869 | */ | |
451d7585 | 2870 | if (btrfs_test_opt(root, SSD_SPREAD)) { |
1bb91902 | 2871 | cont1_bytes = min_bytes = bytes + empty_size; |
451d7585 | 2872 | } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) { |
1bb91902 AO |
2873 | cont1_bytes = bytes; |
2874 | min_bytes = block_group->sectorsize; | |
2875 | } else { | |
2876 | cont1_bytes = max(bytes, (bytes + empty_size) >> 2); | |
2877 | min_bytes = block_group->sectorsize; | |
2878 | } | |
fa9c0d79 | 2879 | |
34d52cb6 | 2880 | spin_lock(&ctl->tree_lock); |
7d0d2e8e JB |
2881 | |
2882 | /* | |
2883 | * If we know we don't have enough space to make a cluster don't even | |
2884 | * bother doing all the work to try and find one. | |
2885 | */ | |
1bb91902 | 2886 | if (ctl->free_space < bytes) { |
34d52cb6 | 2887 | spin_unlock(&ctl->tree_lock); |
7d0d2e8e JB |
2888 | return -ENOSPC; |
2889 | } | |
2890 | ||
fa9c0d79 CM |
2891 | spin_lock(&cluster->lock); |
2892 | ||
2893 | /* someone already found a cluster, hooray */ | |
2894 | if (cluster->block_group) { | |
2895 | ret = 0; | |
2896 | goto out; | |
2897 | } | |
fa9c0d79 | 2898 | |
3f7de037 JB |
2899 | trace_btrfs_find_cluster(block_group, offset, bytes, empty_size, |
2900 | min_bytes); | |
2901 | ||
86d4a77b | 2902 | ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset, |
1bb91902 AO |
2903 | bytes + empty_size, |
2904 | cont1_bytes, min_bytes); | |
4e69b598 | 2905 | if (ret) |
86d4a77b | 2906 | ret = setup_cluster_bitmap(block_group, cluster, &bitmaps, |
1bb91902 AO |
2907 | offset, bytes + empty_size, |
2908 | cont1_bytes, min_bytes); | |
86d4a77b JB |
2909 | |
2910 | /* Clear our temporary list */ | |
2911 | list_for_each_entry_safe(entry, tmp, &bitmaps, list) | |
2912 | list_del_init(&entry->list); | |
fa9c0d79 | 2913 | |
4e69b598 JB |
2914 | if (!ret) { |
2915 | atomic_inc(&block_group->count); | |
2916 | list_add_tail(&cluster->block_group_list, | |
2917 | &block_group->cluster_list); | |
2918 | cluster->block_group = block_group; | |
3f7de037 JB |
2919 | } else { |
2920 | trace_btrfs_failed_cluster_setup(block_group); | |
fa9c0d79 | 2921 | } |
fa9c0d79 CM |
2922 | out: |
2923 | spin_unlock(&cluster->lock); | |
34d52cb6 | 2924 | spin_unlock(&ctl->tree_lock); |
fa9c0d79 CM |
2925 | |
2926 | return ret; | |
2927 | } | |
2928 | ||
2929 | /* | |
2930 | * simple code to zero out a cluster | |
2931 | */ | |
2932 | void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster) | |
2933 | { | |
2934 | spin_lock_init(&cluster->lock); | |
2935 | spin_lock_init(&cluster->refill_lock); | |
6bef4d31 | 2936 | cluster->root = RB_ROOT; |
fa9c0d79 CM |
2937 | cluster->max_size = 0; |
2938 | INIT_LIST_HEAD(&cluster->block_group_list); | |
2939 | cluster->block_group = NULL; | |
2940 | } | |
2941 | ||
7fe1e641 LZ |
2942 | static int do_trimming(struct btrfs_block_group_cache *block_group, |
2943 | u64 *total_trimmed, u64 start, u64 bytes, | |
55507ce3 FM |
2944 | u64 reserved_start, u64 reserved_bytes, |
2945 | struct btrfs_trim_range *trim_entry) | |
f7039b1d | 2946 | { |
7fe1e641 | 2947 | struct btrfs_space_info *space_info = block_group->space_info; |
f7039b1d | 2948 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
55507ce3 | 2949 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
7fe1e641 LZ |
2950 | int ret; |
2951 | int update = 0; | |
2952 | u64 trimmed = 0; | |
f7039b1d | 2953 | |
7fe1e641 LZ |
2954 | spin_lock(&space_info->lock); |
2955 | spin_lock(&block_group->lock); | |
2956 | if (!block_group->ro) { | |
2957 | block_group->reserved += reserved_bytes; | |
2958 | space_info->bytes_reserved += reserved_bytes; | |
2959 | update = 1; | |
2960 | } | |
2961 | spin_unlock(&block_group->lock); | |
2962 | spin_unlock(&space_info->lock); | |
2963 | ||
1edb647b FM |
2964 | ret = btrfs_discard_extent(fs_info->extent_root, |
2965 | start, bytes, &trimmed); | |
7fe1e641 LZ |
2966 | if (!ret) |
2967 | *total_trimmed += trimmed; | |
2968 | ||
55507ce3 | 2969 | mutex_lock(&ctl->cache_writeout_mutex); |
7fe1e641 | 2970 | btrfs_add_free_space(block_group, reserved_start, reserved_bytes); |
55507ce3 FM |
2971 | list_del(&trim_entry->list); |
2972 | mutex_unlock(&ctl->cache_writeout_mutex); | |
7fe1e641 LZ |
2973 | |
2974 | if (update) { | |
2975 | spin_lock(&space_info->lock); | |
2976 | spin_lock(&block_group->lock); | |
2977 | if (block_group->ro) | |
2978 | space_info->bytes_readonly += reserved_bytes; | |
2979 | block_group->reserved -= reserved_bytes; | |
2980 | space_info->bytes_reserved -= reserved_bytes; | |
2981 | spin_unlock(&space_info->lock); | |
2982 | spin_unlock(&block_group->lock); | |
2983 | } | |
2984 | ||
2985 | return ret; | |
2986 | } | |
2987 | ||
2988 | static int trim_no_bitmap(struct btrfs_block_group_cache *block_group, | |
2989 | u64 *total_trimmed, u64 start, u64 end, u64 minlen) | |
2990 | { | |
2991 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
2992 | struct btrfs_free_space *entry; | |
2993 | struct rb_node *node; | |
2994 | int ret = 0; | |
2995 | u64 extent_start; | |
2996 | u64 extent_bytes; | |
2997 | u64 bytes; | |
f7039b1d LD |
2998 | |
2999 | while (start < end) { | |
55507ce3 FM |
3000 | struct btrfs_trim_range trim_entry; |
3001 | ||
3002 | mutex_lock(&ctl->cache_writeout_mutex); | |
34d52cb6 | 3003 | spin_lock(&ctl->tree_lock); |
f7039b1d | 3004 | |
34d52cb6 LZ |
3005 | if (ctl->free_space < minlen) { |
3006 | spin_unlock(&ctl->tree_lock); | |
55507ce3 | 3007 | mutex_unlock(&ctl->cache_writeout_mutex); |
f7039b1d LD |
3008 | break; |
3009 | } | |
3010 | ||
34d52cb6 | 3011 | entry = tree_search_offset(ctl, start, 0, 1); |
7fe1e641 | 3012 | if (!entry) { |
34d52cb6 | 3013 | spin_unlock(&ctl->tree_lock); |
55507ce3 | 3014 | mutex_unlock(&ctl->cache_writeout_mutex); |
f7039b1d LD |
3015 | break; |
3016 | } | |
3017 | ||
7fe1e641 LZ |
3018 | /* skip bitmaps */ |
3019 | while (entry->bitmap) { | |
3020 | node = rb_next(&entry->offset_index); | |
3021 | if (!node) { | |
34d52cb6 | 3022 | spin_unlock(&ctl->tree_lock); |
55507ce3 | 3023 | mutex_unlock(&ctl->cache_writeout_mutex); |
7fe1e641 | 3024 | goto out; |
f7039b1d | 3025 | } |
7fe1e641 LZ |
3026 | entry = rb_entry(node, struct btrfs_free_space, |
3027 | offset_index); | |
f7039b1d LD |
3028 | } |
3029 | ||
7fe1e641 LZ |
3030 | if (entry->offset >= end) { |
3031 | spin_unlock(&ctl->tree_lock); | |
55507ce3 | 3032 | mutex_unlock(&ctl->cache_writeout_mutex); |
7fe1e641 | 3033 | break; |
f7039b1d LD |
3034 | } |
3035 | ||
7fe1e641 LZ |
3036 | extent_start = entry->offset; |
3037 | extent_bytes = entry->bytes; | |
3038 | start = max(start, extent_start); | |
3039 | bytes = min(extent_start + extent_bytes, end) - start; | |
3040 | if (bytes < minlen) { | |
3041 | spin_unlock(&ctl->tree_lock); | |
55507ce3 | 3042 | mutex_unlock(&ctl->cache_writeout_mutex); |
7fe1e641 | 3043 | goto next; |
f7039b1d LD |
3044 | } |
3045 | ||
7fe1e641 LZ |
3046 | unlink_free_space(ctl, entry); |
3047 | kmem_cache_free(btrfs_free_space_cachep, entry); | |
3048 | ||
34d52cb6 | 3049 | spin_unlock(&ctl->tree_lock); |
55507ce3 FM |
3050 | trim_entry.start = extent_start; |
3051 | trim_entry.bytes = extent_bytes; | |
3052 | list_add_tail(&trim_entry.list, &ctl->trimming_ranges); | |
3053 | mutex_unlock(&ctl->cache_writeout_mutex); | |
f7039b1d | 3054 | |
7fe1e641 | 3055 | ret = do_trimming(block_group, total_trimmed, start, bytes, |
55507ce3 | 3056 | extent_start, extent_bytes, &trim_entry); |
7fe1e641 LZ |
3057 | if (ret) |
3058 | break; | |
3059 | next: | |
3060 | start += bytes; | |
f7039b1d | 3061 | |
7fe1e641 LZ |
3062 | if (fatal_signal_pending(current)) { |
3063 | ret = -ERESTARTSYS; | |
3064 | break; | |
3065 | } | |
3066 | ||
3067 | cond_resched(); | |
3068 | } | |
3069 | out: | |
3070 | return ret; | |
3071 | } | |
3072 | ||
3073 | static int trim_bitmaps(struct btrfs_block_group_cache *block_group, | |
3074 | u64 *total_trimmed, u64 start, u64 end, u64 minlen) | |
3075 | { | |
3076 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
3077 | struct btrfs_free_space *entry; | |
3078 | int ret = 0; | |
3079 | int ret2; | |
3080 | u64 bytes; | |
3081 | u64 offset = offset_to_bitmap(ctl, start); | |
3082 | ||
3083 | while (offset < end) { | |
3084 | bool next_bitmap = false; | |
55507ce3 | 3085 | struct btrfs_trim_range trim_entry; |
7fe1e641 | 3086 | |
55507ce3 | 3087 | mutex_lock(&ctl->cache_writeout_mutex); |
7fe1e641 LZ |
3088 | spin_lock(&ctl->tree_lock); |
3089 | ||
3090 | if (ctl->free_space < minlen) { | |
3091 | spin_unlock(&ctl->tree_lock); | |
55507ce3 | 3092 | mutex_unlock(&ctl->cache_writeout_mutex); |
7fe1e641 LZ |
3093 | break; |
3094 | } | |
3095 | ||
3096 | entry = tree_search_offset(ctl, offset, 1, 0); | |
3097 | if (!entry) { | |
3098 | spin_unlock(&ctl->tree_lock); | |
55507ce3 | 3099 | mutex_unlock(&ctl->cache_writeout_mutex); |
7fe1e641 LZ |
3100 | next_bitmap = true; |
3101 | goto next; | |
3102 | } | |
3103 | ||
3104 | bytes = minlen; | |
3105 | ret2 = search_bitmap(ctl, entry, &start, &bytes); | |
3106 | if (ret2 || start >= end) { | |
3107 | spin_unlock(&ctl->tree_lock); | |
55507ce3 | 3108 | mutex_unlock(&ctl->cache_writeout_mutex); |
7fe1e641 LZ |
3109 | next_bitmap = true; |
3110 | goto next; | |
3111 | } | |
3112 | ||
3113 | bytes = min(bytes, end - start); | |
3114 | if (bytes < minlen) { | |
3115 | spin_unlock(&ctl->tree_lock); | |
55507ce3 | 3116 | mutex_unlock(&ctl->cache_writeout_mutex); |
7fe1e641 LZ |
3117 | goto next; |
3118 | } | |
3119 | ||
3120 | bitmap_clear_bits(ctl, entry, start, bytes); | |
3121 | if (entry->bytes == 0) | |
3122 | free_bitmap(ctl, entry); | |
3123 | ||
3124 | spin_unlock(&ctl->tree_lock); | |
55507ce3 FM |
3125 | trim_entry.start = start; |
3126 | trim_entry.bytes = bytes; | |
3127 | list_add_tail(&trim_entry.list, &ctl->trimming_ranges); | |
3128 | mutex_unlock(&ctl->cache_writeout_mutex); | |
7fe1e641 LZ |
3129 | |
3130 | ret = do_trimming(block_group, total_trimmed, start, bytes, | |
55507ce3 | 3131 | start, bytes, &trim_entry); |
7fe1e641 LZ |
3132 | if (ret) |
3133 | break; | |
3134 | next: | |
3135 | if (next_bitmap) { | |
3136 | offset += BITS_PER_BITMAP * ctl->unit; | |
3137 | } else { | |
3138 | start += bytes; | |
3139 | if (start >= offset + BITS_PER_BITMAP * ctl->unit) | |
3140 | offset += BITS_PER_BITMAP * ctl->unit; | |
f7039b1d | 3141 | } |
f7039b1d LD |
3142 | |
3143 | if (fatal_signal_pending(current)) { | |
3144 | ret = -ERESTARTSYS; | |
3145 | break; | |
3146 | } | |
3147 | ||
3148 | cond_resched(); | |
3149 | } | |
3150 | ||
3151 | return ret; | |
3152 | } | |
581bb050 | 3153 | |
7fe1e641 LZ |
3154 | int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group, |
3155 | u64 *trimmed, u64 start, u64 end, u64 minlen) | |
3156 | { | |
3157 | int ret; | |
3158 | ||
3159 | *trimmed = 0; | |
3160 | ||
04216820 FM |
3161 | spin_lock(&block_group->lock); |
3162 | if (block_group->removed) { | |
3163 | spin_unlock(&block_group->lock); | |
3164 | return 0; | |
3165 | } | |
3166 | atomic_inc(&block_group->trimming); | |
3167 | spin_unlock(&block_group->lock); | |
3168 | ||
7fe1e641 LZ |
3169 | ret = trim_no_bitmap(block_group, trimmed, start, end, minlen); |
3170 | if (ret) | |
04216820 | 3171 | goto out; |
7fe1e641 LZ |
3172 | |
3173 | ret = trim_bitmaps(block_group, trimmed, start, end, minlen); | |
04216820 FM |
3174 | out: |
3175 | spin_lock(&block_group->lock); | |
3176 | if (atomic_dec_and_test(&block_group->trimming) && | |
3177 | block_group->removed) { | |
3178 | struct extent_map_tree *em_tree; | |
3179 | struct extent_map *em; | |
3180 | ||
3181 | spin_unlock(&block_group->lock); | |
3182 | ||
a1e7e16e | 3183 | lock_chunks(block_group->fs_info->chunk_root); |
04216820 FM |
3184 | em_tree = &block_group->fs_info->mapping_tree.map_tree; |
3185 | write_lock(&em_tree->lock); | |
3186 | em = lookup_extent_mapping(em_tree, block_group->key.objectid, | |
3187 | 1); | |
3188 | BUG_ON(!em); /* logic error, can't happen */ | |
a1e7e16e FM |
3189 | /* |
3190 | * remove_extent_mapping() will delete us from the pinned_chunks | |
3191 | * list, which is protected by the chunk mutex. | |
3192 | */ | |
04216820 FM |
3193 | remove_extent_mapping(em_tree, em); |
3194 | write_unlock(&em_tree->lock); | |
04216820 FM |
3195 | unlock_chunks(block_group->fs_info->chunk_root); |
3196 | ||
3197 | /* once for us and once for the tree */ | |
3198 | free_extent_map(em); | |
3199 | free_extent_map(em); | |
946ddbe8 FM |
3200 | |
3201 | /* | |
3202 | * We've left one free space entry and other tasks trimming | |
3203 | * this block group have left 1 entry each one. Free them. | |
3204 | */ | |
3205 | __btrfs_remove_free_space_cache(block_group->free_space_ctl); | |
04216820 FM |
3206 | } else { |
3207 | spin_unlock(&block_group->lock); | |
3208 | } | |
7fe1e641 LZ |
3209 | |
3210 | return ret; | |
3211 | } | |
3212 | ||
581bb050 LZ |
3213 | /* |
3214 | * Find the left-most item in the cache tree, and then return the | |
3215 | * smallest inode number in the item. | |
3216 | * | |
3217 | * Note: the returned inode number may not be the smallest one in | |
3218 | * the tree, if the left-most item is a bitmap. | |
3219 | */ | |
3220 | u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root) | |
3221 | { | |
3222 | struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl; | |
3223 | struct btrfs_free_space *entry = NULL; | |
3224 | u64 ino = 0; | |
3225 | ||
3226 | spin_lock(&ctl->tree_lock); | |
3227 | ||
3228 | if (RB_EMPTY_ROOT(&ctl->free_space_offset)) | |
3229 | goto out; | |
3230 | ||
3231 | entry = rb_entry(rb_first(&ctl->free_space_offset), | |
3232 | struct btrfs_free_space, offset_index); | |
3233 | ||
3234 | if (!entry->bitmap) { | |
3235 | ino = entry->offset; | |
3236 | ||
3237 | unlink_free_space(ctl, entry); | |
3238 | entry->offset++; | |
3239 | entry->bytes--; | |
3240 | if (!entry->bytes) | |
3241 | kmem_cache_free(btrfs_free_space_cachep, entry); | |
3242 | else | |
3243 | link_free_space(ctl, entry); | |
3244 | } else { | |
3245 | u64 offset = 0; | |
3246 | u64 count = 1; | |
3247 | int ret; | |
3248 | ||
3249 | ret = search_bitmap(ctl, entry, &offset, &count); | |
79787eaa | 3250 | /* Logic error; Should be empty if it can't find anything */ |
b12d6869 | 3251 | ASSERT(!ret); |
581bb050 LZ |
3252 | |
3253 | ino = offset; | |
3254 | bitmap_clear_bits(ctl, entry, offset, 1); | |
3255 | if (entry->bytes == 0) | |
3256 | free_bitmap(ctl, entry); | |
3257 | } | |
3258 | out: | |
3259 | spin_unlock(&ctl->tree_lock); | |
3260 | ||
3261 | return ino; | |
3262 | } | |
82d5902d LZ |
3263 | |
3264 | struct inode *lookup_free_ino_inode(struct btrfs_root *root, | |
3265 | struct btrfs_path *path) | |
3266 | { | |
3267 | struct inode *inode = NULL; | |
3268 | ||
57cdc8db DS |
3269 | spin_lock(&root->ino_cache_lock); |
3270 | if (root->ino_cache_inode) | |
3271 | inode = igrab(root->ino_cache_inode); | |
3272 | spin_unlock(&root->ino_cache_lock); | |
82d5902d LZ |
3273 | if (inode) |
3274 | return inode; | |
3275 | ||
3276 | inode = __lookup_free_space_inode(root, path, 0); | |
3277 | if (IS_ERR(inode)) | |
3278 | return inode; | |
3279 | ||
57cdc8db | 3280 | spin_lock(&root->ino_cache_lock); |
7841cb28 | 3281 | if (!btrfs_fs_closing(root->fs_info)) |
57cdc8db DS |
3282 | root->ino_cache_inode = igrab(inode); |
3283 | spin_unlock(&root->ino_cache_lock); | |
82d5902d LZ |
3284 | |
3285 | return inode; | |
3286 | } | |
3287 | ||
3288 | int create_free_ino_inode(struct btrfs_root *root, | |
3289 | struct btrfs_trans_handle *trans, | |
3290 | struct btrfs_path *path) | |
3291 | { | |
3292 | return __create_free_space_inode(root, trans, path, | |
3293 | BTRFS_FREE_INO_OBJECTID, 0); | |
3294 | } | |
3295 | ||
3296 | int load_free_ino_cache(struct btrfs_fs_info *fs_info, struct btrfs_root *root) | |
3297 | { | |
3298 | struct btrfs_free_space_ctl *ctl = root->free_ino_ctl; | |
3299 | struct btrfs_path *path; | |
3300 | struct inode *inode; | |
3301 | int ret = 0; | |
3302 | u64 root_gen = btrfs_root_generation(&root->root_item); | |
3303 | ||
4b9465cb CM |
3304 | if (!btrfs_test_opt(root, INODE_MAP_CACHE)) |
3305 | return 0; | |
3306 | ||
82d5902d LZ |
3307 | /* |
3308 | * If we're unmounting then just return, since this does a search on the | |
3309 | * normal root and not the commit root and we could deadlock. | |
3310 | */ | |
7841cb28 | 3311 | if (btrfs_fs_closing(fs_info)) |
82d5902d LZ |
3312 | return 0; |
3313 | ||
3314 | path = btrfs_alloc_path(); | |
3315 | if (!path) | |
3316 | return 0; | |
3317 | ||
3318 | inode = lookup_free_ino_inode(root, path); | |
3319 | if (IS_ERR(inode)) | |
3320 | goto out; | |
3321 | ||
3322 | if (root_gen != BTRFS_I(inode)->generation) | |
3323 | goto out_put; | |
3324 | ||
3325 | ret = __load_free_space_cache(root, inode, ctl, path, 0); | |
3326 | ||
3327 | if (ret < 0) | |
c2cf52eb SK |
3328 | btrfs_err(fs_info, |
3329 | "failed to load free ino cache for root %llu", | |
3330 | root->root_key.objectid); | |
82d5902d LZ |
3331 | out_put: |
3332 | iput(inode); | |
3333 | out: | |
3334 | btrfs_free_path(path); | |
3335 | return ret; | |
3336 | } | |
3337 | ||
3338 | int btrfs_write_out_ino_cache(struct btrfs_root *root, | |
3339 | struct btrfs_trans_handle *trans, | |
53645a91 FDBM |
3340 | struct btrfs_path *path, |
3341 | struct inode *inode) | |
82d5902d LZ |
3342 | { |
3343 | struct btrfs_free_space_ctl *ctl = root->free_ino_ctl; | |
82d5902d LZ |
3344 | int ret; |
3345 | ||
4b9465cb CM |
3346 | if (!btrfs_test_opt(root, INODE_MAP_CACHE)) |
3347 | return 0; | |
3348 | ||
82d5902d | 3349 | ret = __btrfs_write_out_cache(root, inode, ctl, NULL, trans, path, 0); |
c09544e0 JB |
3350 | if (ret) { |
3351 | btrfs_delalloc_release_metadata(inode, inode->i_size); | |
3352 | #ifdef DEBUG | |
c2cf52eb SK |
3353 | btrfs_err(root->fs_info, |
3354 | "failed to write free ino cache for root %llu", | |
3355 | root->root_key.objectid); | |
c09544e0 JB |
3356 | #endif |
3357 | } | |
82d5902d | 3358 | |
82d5902d LZ |
3359 | return ret; |
3360 | } | |
74255aa0 JB |
3361 | |
3362 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS | |
dc11dd5d JB |
3363 | /* |
3364 | * Use this if you need to make a bitmap or extent entry specifically, it | |
3365 | * doesn't do any of the merging that add_free_space does, this acts a lot like | |
3366 | * how the free space cache loading stuff works, so you can get really weird | |
3367 | * configurations. | |
3368 | */ | |
3369 | int test_add_free_space_entry(struct btrfs_block_group_cache *cache, | |
3370 | u64 offset, u64 bytes, bool bitmap) | |
74255aa0 | 3371 | { |
dc11dd5d JB |
3372 | struct btrfs_free_space_ctl *ctl = cache->free_space_ctl; |
3373 | struct btrfs_free_space *info = NULL, *bitmap_info; | |
3374 | void *map = NULL; | |
3375 | u64 bytes_added; | |
3376 | int ret; | |
74255aa0 | 3377 | |
dc11dd5d JB |
3378 | again: |
3379 | if (!info) { | |
3380 | info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS); | |
3381 | if (!info) | |
3382 | return -ENOMEM; | |
74255aa0 JB |
3383 | } |
3384 | ||
dc11dd5d JB |
3385 | if (!bitmap) { |
3386 | spin_lock(&ctl->tree_lock); | |
3387 | info->offset = offset; | |
3388 | info->bytes = bytes; | |
3389 | ret = link_free_space(ctl, info); | |
3390 | spin_unlock(&ctl->tree_lock); | |
3391 | if (ret) | |
3392 | kmem_cache_free(btrfs_free_space_cachep, info); | |
3393 | return ret; | |
3394 | } | |
3395 | ||
3396 | if (!map) { | |
3397 | map = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS); | |
3398 | if (!map) { | |
3399 | kmem_cache_free(btrfs_free_space_cachep, info); | |
3400 | return -ENOMEM; | |
3401 | } | |
3402 | } | |
3403 | ||
3404 | spin_lock(&ctl->tree_lock); | |
3405 | bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), | |
3406 | 1, 0); | |
3407 | if (!bitmap_info) { | |
3408 | info->bitmap = map; | |
3409 | map = NULL; | |
3410 | add_new_bitmap(ctl, info, offset); | |
3411 | bitmap_info = info; | |
20005523 | 3412 | info = NULL; |
dc11dd5d | 3413 | } |
74255aa0 | 3414 | |
dc11dd5d JB |
3415 | bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes); |
3416 | bytes -= bytes_added; | |
3417 | offset += bytes_added; | |
3418 | spin_unlock(&ctl->tree_lock); | |
74255aa0 | 3419 | |
dc11dd5d JB |
3420 | if (bytes) |
3421 | goto again; | |
74255aa0 | 3422 | |
20005523 FM |
3423 | if (info) |
3424 | kmem_cache_free(btrfs_free_space_cachep, info); | |
dc11dd5d JB |
3425 | if (map) |
3426 | kfree(map); | |
3427 | return 0; | |
74255aa0 JB |
3428 | } |
3429 | ||
3430 | /* | |
3431 | * Checks to see if the given range is in the free space cache. This is really | |
3432 | * just used to check the absence of space, so if there is free space in the | |
3433 | * range at all we will return 1. | |
3434 | */ | |
dc11dd5d JB |
3435 | int test_check_exists(struct btrfs_block_group_cache *cache, |
3436 | u64 offset, u64 bytes) | |
74255aa0 JB |
3437 | { |
3438 | struct btrfs_free_space_ctl *ctl = cache->free_space_ctl; | |
3439 | struct btrfs_free_space *info; | |
3440 | int ret = 0; | |
3441 | ||
3442 | spin_lock(&ctl->tree_lock); | |
3443 | info = tree_search_offset(ctl, offset, 0, 0); | |
3444 | if (!info) { | |
3445 | info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), | |
3446 | 1, 0); | |
3447 | if (!info) | |
3448 | goto out; | |
3449 | } | |
3450 | ||
3451 | have_info: | |
3452 | if (info->bitmap) { | |
3453 | u64 bit_off, bit_bytes; | |
3454 | struct rb_node *n; | |
3455 | struct btrfs_free_space *tmp; | |
3456 | ||
3457 | bit_off = offset; | |
3458 | bit_bytes = ctl->unit; | |
3459 | ret = search_bitmap(ctl, info, &bit_off, &bit_bytes); | |
3460 | if (!ret) { | |
3461 | if (bit_off == offset) { | |
3462 | ret = 1; | |
3463 | goto out; | |
3464 | } else if (bit_off > offset && | |
3465 | offset + bytes > bit_off) { | |
3466 | ret = 1; | |
3467 | goto out; | |
3468 | } | |
3469 | } | |
3470 | ||
3471 | n = rb_prev(&info->offset_index); | |
3472 | while (n) { | |
3473 | tmp = rb_entry(n, struct btrfs_free_space, | |
3474 | offset_index); | |
3475 | if (tmp->offset + tmp->bytes < offset) | |
3476 | break; | |
3477 | if (offset + bytes < tmp->offset) { | |
3478 | n = rb_prev(&info->offset_index); | |
3479 | continue; | |
3480 | } | |
3481 | info = tmp; | |
3482 | goto have_info; | |
3483 | } | |
3484 | ||
3485 | n = rb_next(&info->offset_index); | |
3486 | while (n) { | |
3487 | tmp = rb_entry(n, struct btrfs_free_space, | |
3488 | offset_index); | |
3489 | if (offset + bytes < tmp->offset) | |
3490 | break; | |
3491 | if (tmp->offset + tmp->bytes < offset) { | |
3492 | n = rb_next(&info->offset_index); | |
3493 | continue; | |
3494 | } | |
3495 | info = tmp; | |
3496 | goto have_info; | |
3497 | } | |
3498 | ||
20005523 | 3499 | ret = 0; |
74255aa0 JB |
3500 | goto out; |
3501 | } | |
3502 | ||
3503 | if (info->offset == offset) { | |
3504 | ret = 1; | |
3505 | goto out; | |
3506 | } | |
3507 | ||
3508 | if (offset > info->offset && offset < info->offset + info->bytes) | |
3509 | ret = 1; | |
3510 | out: | |
3511 | spin_unlock(&ctl->tree_lock); | |
3512 | return ret; | |
3513 | } | |
dc11dd5d | 3514 | #endif /* CONFIG_BTRFS_FS_RUN_SANITY_TESTS */ |