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