]>
Commit | Line | Data |
---|---|---|
0f9dd46c JB |
1 | /* |
2 | * Copyright (C) 2008 Red Hat. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
96303081 | 19 | #include <linux/pagemap.h> |
0f9dd46c | 20 | #include <linux/sched.h> |
5a0e3ad6 | 21 | #include <linux/slab.h> |
96303081 | 22 | #include <linux/math64.h> |
6ab60601 | 23 | #include <linux/ratelimit.h> |
0f9dd46c | 24 | #include "ctree.h" |
fa9c0d79 CM |
25 | #include "free-space-cache.h" |
26 | #include "transaction.h" | |
0af3d00b | 27 | #include "disk-io.h" |
43be2146 | 28 | #include "extent_io.h" |
581bb050 | 29 | #include "inode-map.h" |
04216820 | 30 | #include "volumes.h" |
fa9c0d79 | 31 | |
96303081 JB |
32 | #define BITS_PER_BITMAP (PAGE_CACHE_SIZE * 8) |
33 | #define MAX_CACHE_BYTES_PER_GIG (32 * 1024) | |
0f9dd46c | 34 | |
55507ce3 FM |
35 | struct btrfs_trim_range { |
36 | u64 start; | |
37 | u64 bytes; | |
38 | struct list_head list; | |
39 | }; | |
40 | ||
34d52cb6 | 41 | static int link_free_space(struct btrfs_free_space_ctl *ctl, |
0cb59c99 | 42 | struct btrfs_free_space *info); |
cd023e7b JB |
43 | static void unlink_free_space(struct btrfs_free_space_ctl *ctl, |
44 | struct btrfs_free_space *info); | |
0cb59c99 | 45 | |
0414efae LZ |
46 | static struct inode *__lookup_free_space_inode(struct btrfs_root *root, |
47 | struct btrfs_path *path, | |
48 | u64 offset) | |
0af3d00b JB |
49 | { |
50 | struct btrfs_key key; | |
51 | struct btrfs_key location; | |
52 | struct btrfs_disk_key disk_key; | |
53 | struct btrfs_free_space_header *header; | |
54 | struct extent_buffer *leaf; | |
55 | struct inode *inode = NULL; | |
56 | int ret; | |
57 | ||
0af3d00b | 58 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; |
0414efae | 59 | key.offset = offset; |
0af3d00b JB |
60 | key.type = 0; |
61 | ||
62 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
63 | if (ret < 0) | |
64 | return ERR_PTR(ret); | |
65 | if (ret > 0) { | |
b3b4aa74 | 66 | btrfs_release_path(path); |
0af3d00b JB |
67 | return ERR_PTR(-ENOENT); |
68 | } | |
69 | ||
70 | leaf = path->nodes[0]; | |
71 | header = btrfs_item_ptr(leaf, path->slots[0], | |
72 | struct btrfs_free_space_header); | |
73 | btrfs_free_space_key(leaf, header, &disk_key); | |
74 | btrfs_disk_key_to_cpu(&location, &disk_key); | |
b3b4aa74 | 75 | btrfs_release_path(path); |
0af3d00b JB |
76 | |
77 | inode = btrfs_iget(root->fs_info->sb, &location, root, NULL); | |
78 | if (!inode) | |
79 | return ERR_PTR(-ENOENT); | |
80 | if (IS_ERR(inode)) | |
81 | return inode; | |
82 | if (is_bad_inode(inode)) { | |
83 | iput(inode); | |
84 | return ERR_PTR(-ENOENT); | |
85 | } | |
86 | ||
528c0327 AV |
87 | mapping_set_gfp_mask(inode->i_mapping, |
88 | mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS); | |
adae52b9 | 89 | |
0414efae LZ |
90 | return inode; |
91 | } | |
92 | ||
93 | struct inode *lookup_free_space_inode(struct btrfs_root *root, | |
94 | struct btrfs_block_group_cache | |
95 | *block_group, struct btrfs_path *path) | |
96 | { | |
97 | struct inode *inode = NULL; | |
5b0e95bf | 98 | u32 flags = BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW; |
0414efae LZ |
99 | |
100 | spin_lock(&block_group->lock); | |
101 | if (block_group->inode) | |
102 | inode = igrab(block_group->inode); | |
103 | spin_unlock(&block_group->lock); | |
104 | if (inode) | |
105 | return inode; | |
106 | ||
107 | inode = __lookup_free_space_inode(root, path, | |
108 | block_group->key.objectid); | |
109 | if (IS_ERR(inode)) | |
110 | return inode; | |
111 | ||
0af3d00b | 112 | spin_lock(&block_group->lock); |
5b0e95bf | 113 | if (!((BTRFS_I(inode)->flags & flags) == flags)) { |
c2cf52eb SK |
114 | btrfs_info(root->fs_info, |
115 | "Old style space inode found, converting."); | |
5b0e95bf JB |
116 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM | |
117 | BTRFS_INODE_NODATACOW; | |
2f356126 JB |
118 | block_group->disk_cache_state = BTRFS_DC_CLEAR; |
119 | } | |
120 | ||
300e4f8a | 121 | if (!block_group->iref) { |
0af3d00b JB |
122 | block_group->inode = igrab(inode); |
123 | block_group->iref = 1; | |
124 | } | |
125 | spin_unlock(&block_group->lock); | |
126 | ||
127 | return inode; | |
128 | } | |
129 | ||
48a3b636 ES |
130 | static int __create_free_space_inode(struct btrfs_root *root, |
131 | struct btrfs_trans_handle *trans, | |
132 | struct btrfs_path *path, | |
133 | u64 ino, u64 offset) | |
0af3d00b JB |
134 | { |
135 | struct btrfs_key key; | |
136 | struct btrfs_disk_key disk_key; | |
137 | struct btrfs_free_space_header *header; | |
138 | struct btrfs_inode_item *inode_item; | |
139 | struct extent_buffer *leaf; | |
5b0e95bf | 140 | u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC; |
0af3d00b JB |
141 | int ret; |
142 | ||
0414efae | 143 | ret = btrfs_insert_empty_inode(trans, root, path, ino); |
0af3d00b JB |
144 | if (ret) |
145 | return ret; | |
146 | ||
5b0e95bf JB |
147 | /* We inline crc's for the free disk space cache */ |
148 | if (ino != BTRFS_FREE_INO_OBJECTID) | |
149 | flags |= BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW; | |
150 | ||
0af3d00b JB |
151 | leaf = path->nodes[0]; |
152 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
153 | struct btrfs_inode_item); | |
154 | btrfs_item_key(leaf, &disk_key, path->slots[0]); | |
155 | memset_extent_buffer(leaf, 0, (unsigned long)inode_item, | |
156 | sizeof(*inode_item)); | |
157 | btrfs_set_inode_generation(leaf, inode_item, trans->transid); | |
158 | btrfs_set_inode_size(leaf, inode_item, 0); | |
159 | btrfs_set_inode_nbytes(leaf, inode_item, 0); | |
160 | btrfs_set_inode_uid(leaf, inode_item, 0); | |
161 | btrfs_set_inode_gid(leaf, inode_item, 0); | |
162 | btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600); | |
5b0e95bf | 163 | btrfs_set_inode_flags(leaf, inode_item, flags); |
0af3d00b JB |
164 | btrfs_set_inode_nlink(leaf, inode_item, 1); |
165 | btrfs_set_inode_transid(leaf, inode_item, trans->transid); | |
0414efae | 166 | btrfs_set_inode_block_group(leaf, inode_item, offset); |
0af3d00b | 167 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 168 | btrfs_release_path(path); |
0af3d00b JB |
169 | |
170 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
0414efae | 171 | key.offset = offset; |
0af3d00b JB |
172 | key.type = 0; |
173 | ||
174 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
175 | sizeof(struct btrfs_free_space_header)); | |
176 | if (ret < 0) { | |
b3b4aa74 | 177 | btrfs_release_path(path); |
0af3d00b JB |
178 | return ret; |
179 | } | |
180 | leaf = path->nodes[0]; | |
181 | header = btrfs_item_ptr(leaf, path->slots[0], | |
182 | struct btrfs_free_space_header); | |
183 | memset_extent_buffer(leaf, 0, (unsigned long)header, sizeof(*header)); | |
184 | btrfs_set_free_space_key(leaf, header, &disk_key); | |
185 | btrfs_mark_buffer_dirty(leaf); | |
b3b4aa74 | 186 | btrfs_release_path(path); |
0af3d00b JB |
187 | |
188 | return 0; | |
189 | } | |
190 | ||
0414efae LZ |
191 | int create_free_space_inode(struct btrfs_root *root, |
192 | struct btrfs_trans_handle *trans, | |
193 | struct btrfs_block_group_cache *block_group, | |
194 | struct btrfs_path *path) | |
195 | { | |
196 | int ret; | |
197 | u64 ino; | |
198 | ||
199 | ret = btrfs_find_free_objectid(root, &ino); | |
200 | if (ret < 0) | |
201 | return ret; | |
202 | ||
203 | return __create_free_space_inode(root, trans, path, ino, | |
204 | block_group->key.objectid); | |
205 | } | |
206 | ||
7b61cd92 MX |
207 | int btrfs_check_trunc_cache_free_space(struct btrfs_root *root, |
208 | struct btrfs_block_rsv *rsv) | |
0af3d00b | 209 | { |
c8174313 | 210 | u64 needed_bytes; |
7b61cd92 | 211 | int ret; |
c8174313 JB |
212 | |
213 | /* 1 for slack space, 1 for updating the inode */ | |
214 | needed_bytes = btrfs_calc_trunc_metadata_size(root, 1) + | |
215 | btrfs_calc_trans_metadata_size(root, 1); | |
216 | ||
7b61cd92 MX |
217 | spin_lock(&rsv->lock); |
218 | if (rsv->reserved < needed_bytes) | |
219 | ret = -ENOSPC; | |
220 | else | |
221 | ret = 0; | |
222 | spin_unlock(&rsv->lock); | |
4b286cd1 | 223 | return ret; |
7b61cd92 MX |
224 | } |
225 | ||
226 | int btrfs_truncate_free_space_cache(struct btrfs_root *root, | |
227 | struct btrfs_trans_handle *trans, | |
7b61cd92 MX |
228 | struct inode *inode) |
229 | { | |
7b61cd92 | 230 | int ret = 0; |
0af3d00b | 231 | |
0af3d00b | 232 | btrfs_i_size_write(inode, 0); |
7caef267 | 233 | truncate_pagecache(inode, 0); |
0af3d00b JB |
234 | |
235 | /* | |
236 | * We don't need an orphan item because truncating the free space cache | |
237 | * will never be split across transactions. | |
28ed1345 CM |
238 | * We don't need to check for -EAGAIN because we're a free space |
239 | * cache inode | |
0af3d00b JB |
240 | */ |
241 | ret = btrfs_truncate_inode_items(trans, root, inode, | |
242 | 0, BTRFS_EXTENT_DATA_KEY); | |
243 | if (ret) { | |
79787eaa | 244 | btrfs_abort_transaction(trans, root, ret); |
0af3d00b JB |
245 | return ret; |
246 | } | |
247 | ||
82d5902d | 248 | ret = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
249 | if (ret) |
250 | btrfs_abort_transaction(trans, root, ret); | |
c8174313 | 251 | |
82d5902d | 252 | return ret; |
0af3d00b JB |
253 | } |
254 | ||
9d66e233 JB |
255 | static int readahead_cache(struct inode *inode) |
256 | { | |
257 | struct file_ra_state *ra; | |
258 | unsigned long last_index; | |
259 | ||
260 | ra = kzalloc(sizeof(*ra), GFP_NOFS); | |
261 | if (!ra) | |
262 | return -ENOMEM; | |
263 | ||
264 | file_ra_state_init(ra, inode->i_mapping); | |
265 | last_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT; | |
266 | ||
267 | page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index); | |
268 | ||
269 | kfree(ra); | |
270 | ||
271 | return 0; | |
272 | } | |
273 | ||
4c6d1d85 | 274 | static int io_ctl_init(struct btrfs_io_ctl *io_ctl, struct inode *inode, |
5349d6c3 | 275 | struct btrfs_root *root, int write) |
a67509c3 | 276 | { |
5349d6c3 MX |
277 | int num_pages; |
278 | int check_crcs = 0; | |
279 | ||
ed6078f7 | 280 | num_pages = DIV_ROUND_UP(i_size_read(inode), PAGE_CACHE_SIZE); |
5349d6c3 MX |
281 | |
282 | if (btrfs_ino(inode) != BTRFS_FREE_INO_OBJECTID) | |
283 | check_crcs = 1; | |
284 | ||
285 | /* Make sure we can fit our crcs into the first page */ | |
286 | if (write && check_crcs && | |
287 | (num_pages * sizeof(u32)) >= PAGE_CACHE_SIZE) | |
288 | return -ENOSPC; | |
289 | ||
4c6d1d85 | 290 | memset(io_ctl, 0, sizeof(struct btrfs_io_ctl)); |
5349d6c3 | 291 | |
31e818fe | 292 | io_ctl->pages = kcalloc(num_pages, sizeof(struct page *), GFP_NOFS); |
a67509c3 JB |
293 | if (!io_ctl->pages) |
294 | return -ENOMEM; | |
5349d6c3 MX |
295 | |
296 | io_ctl->num_pages = num_pages; | |
a67509c3 | 297 | io_ctl->root = root; |
5349d6c3 MX |
298 | io_ctl->check_crcs = check_crcs; |
299 | ||
a67509c3 JB |
300 | return 0; |
301 | } | |
302 | ||
4c6d1d85 | 303 | static void io_ctl_free(struct btrfs_io_ctl *io_ctl) |
a67509c3 JB |
304 | { |
305 | kfree(io_ctl->pages); | |
306 | } | |
307 | ||
4c6d1d85 | 308 | static void io_ctl_unmap_page(struct btrfs_io_ctl *io_ctl) |
a67509c3 JB |
309 | { |
310 | if (io_ctl->cur) { | |
311 | kunmap(io_ctl->page); | |
312 | io_ctl->cur = NULL; | |
313 | io_ctl->orig = NULL; | |
314 | } | |
315 | } | |
316 | ||
4c6d1d85 | 317 | static void io_ctl_map_page(struct btrfs_io_ctl *io_ctl, int clear) |
a67509c3 | 318 | { |
b12d6869 | 319 | ASSERT(io_ctl->index < io_ctl->num_pages); |
a67509c3 JB |
320 | io_ctl->page = io_ctl->pages[io_ctl->index++]; |
321 | io_ctl->cur = kmap(io_ctl->page); | |
322 | io_ctl->orig = io_ctl->cur; | |
323 | io_ctl->size = PAGE_CACHE_SIZE; | |
324 | if (clear) | |
325 | memset(io_ctl->cur, 0, PAGE_CACHE_SIZE); | |
326 | } | |
327 | ||
4c6d1d85 | 328 | static void io_ctl_drop_pages(struct btrfs_io_ctl *io_ctl) |
a67509c3 JB |
329 | { |
330 | int i; | |
331 | ||
332 | io_ctl_unmap_page(io_ctl); | |
333 | ||
334 | for (i = 0; i < io_ctl->num_pages; i++) { | |
a1ee5a45 LZ |
335 | if (io_ctl->pages[i]) { |
336 | ClearPageChecked(io_ctl->pages[i]); | |
337 | unlock_page(io_ctl->pages[i]); | |
338 | page_cache_release(io_ctl->pages[i]); | |
339 | } | |
a67509c3 JB |
340 | } |
341 | } | |
342 | ||
4c6d1d85 | 343 | static int io_ctl_prepare_pages(struct btrfs_io_ctl *io_ctl, struct inode *inode, |
a67509c3 JB |
344 | int uptodate) |
345 | { | |
346 | struct page *page; | |
347 | gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); | |
348 | int i; | |
349 | ||
350 | for (i = 0; i < io_ctl->num_pages; i++) { | |
351 | page = find_or_create_page(inode->i_mapping, i, mask); | |
352 | if (!page) { | |
353 | io_ctl_drop_pages(io_ctl); | |
354 | return -ENOMEM; | |
355 | } | |
356 | io_ctl->pages[i] = page; | |
357 | if (uptodate && !PageUptodate(page)) { | |
358 | btrfs_readpage(NULL, page); | |
359 | lock_page(page); | |
360 | if (!PageUptodate(page)) { | |
efe120a0 FH |
361 | btrfs_err(BTRFS_I(inode)->root->fs_info, |
362 | "error reading free space cache"); | |
a67509c3 JB |
363 | io_ctl_drop_pages(io_ctl); |
364 | return -EIO; | |
365 | } | |
366 | } | |
367 | } | |
368 | ||
f7d61dcd JB |
369 | for (i = 0; i < io_ctl->num_pages; i++) { |
370 | clear_page_dirty_for_io(io_ctl->pages[i]); | |
371 | set_page_extent_mapped(io_ctl->pages[i]); | |
372 | } | |
373 | ||
a67509c3 JB |
374 | return 0; |
375 | } | |
376 | ||
4c6d1d85 | 377 | static void io_ctl_set_generation(struct btrfs_io_ctl *io_ctl, u64 generation) |
a67509c3 | 378 | { |
528c0327 | 379 | __le64 *val; |
a67509c3 JB |
380 | |
381 | io_ctl_map_page(io_ctl, 1); | |
382 | ||
383 | /* | |
5b0e95bf JB |
384 | * Skip the csum areas. If we don't check crcs then we just have a |
385 | * 64bit chunk at the front of the first page. | |
a67509c3 | 386 | */ |
5b0e95bf JB |
387 | if (io_ctl->check_crcs) { |
388 | io_ctl->cur += (sizeof(u32) * io_ctl->num_pages); | |
389 | io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages); | |
390 | } else { | |
391 | io_ctl->cur += sizeof(u64); | |
392 | io_ctl->size -= sizeof(u64) * 2; | |
393 | } | |
a67509c3 JB |
394 | |
395 | val = io_ctl->cur; | |
396 | *val = cpu_to_le64(generation); | |
397 | io_ctl->cur += sizeof(u64); | |
a67509c3 JB |
398 | } |
399 | ||
4c6d1d85 | 400 | static int io_ctl_check_generation(struct btrfs_io_ctl *io_ctl, u64 generation) |
a67509c3 | 401 | { |
528c0327 | 402 | __le64 *gen; |
a67509c3 | 403 | |
5b0e95bf JB |
404 | /* |
405 | * Skip the crc area. If we don't check crcs then we just have a 64bit | |
406 | * chunk at the front of the first page. | |
407 | */ | |
408 | if (io_ctl->check_crcs) { | |
409 | io_ctl->cur += sizeof(u32) * io_ctl->num_pages; | |
410 | io_ctl->size -= sizeof(u64) + | |
411 | (sizeof(u32) * io_ctl->num_pages); | |
412 | } else { | |
413 | io_ctl->cur += sizeof(u64); | |
414 | io_ctl->size -= sizeof(u64) * 2; | |
415 | } | |
a67509c3 | 416 | |
a67509c3 JB |
417 | gen = io_ctl->cur; |
418 | if (le64_to_cpu(*gen) != generation) { | |
efe120a0 | 419 | printk_ratelimited(KERN_ERR "BTRFS: space cache generation " |
a67509c3 JB |
420 | "(%Lu) does not match inode (%Lu)\n", *gen, |
421 | generation); | |
422 | io_ctl_unmap_page(io_ctl); | |
423 | return -EIO; | |
424 | } | |
425 | io_ctl->cur += sizeof(u64); | |
5b0e95bf JB |
426 | return 0; |
427 | } | |
428 | ||
4c6d1d85 | 429 | static void io_ctl_set_crc(struct btrfs_io_ctl *io_ctl, int index) |
5b0e95bf JB |
430 | { |
431 | u32 *tmp; | |
432 | u32 crc = ~(u32)0; | |
433 | unsigned offset = 0; | |
434 | ||
435 | if (!io_ctl->check_crcs) { | |
436 | io_ctl_unmap_page(io_ctl); | |
437 | return; | |
438 | } | |
439 | ||
440 | if (index == 0) | |
cb54f257 | 441 | offset = sizeof(u32) * io_ctl->num_pages; |
5b0e95bf | 442 | |
b0496686 | 443 | crc = btrfs_csum_data(io_ctl->orig + offset, crc, |
5b0e95bf JB |
444 | PAGE_CACHE_SIZE - offset); |
445 | btrfs_csum_final(crc, (char *)&crc); | |
446 | io_ctl_unmap_page(io_ctl); | |
447 | tmp = kmap(io_ctl->pages[0]); | |
448 | tmp += index; | |
449 | *tmp = crc; | |
450 | kunmap(io_ctl->pages[0]); | |
451 | } | |
452 | ||
4c6d1d85 | 453 | static int io_ctl_check_crc(struct btrfs_io_ctl *io_ctl, int index) |
5b0e95bf JB |
454 | { |
455 | u32 *tmp, val; | |
456 | u32 crc = ~(u32)0; | |
457 | unsigned offset = 0; | |
458 | ||
459 | if (!io_ctl->check_crcs) { | |
460 | io_ctl_map_page(io_ctl, 0); | |
461 | return 0; | |
462 | } | |
463 | ||
464 | if (index == 0) | |
465 | offset = sizeof(u32) * io_ctl->num_pages; | |
466 | ||
467 | tmp = kmap(io_ctl->pages[0]); | |
468 | tmp += index; | |
469 | val = *tmp; | |
470 | kunmap(io_ctl->pages[0]); | |
471 | ||
472 | io_ctl_map_page(io_ctl, 0); | |
b0496686 | 473 | crc = btrfs_csum_data(io_ctl->orig + offset, crc, |
5b0e95bf JB |
474 | PAGE_CACHE_SIZE - offset); |
475 | btrfs_csum_final(crc, (char *)&crc); | |
476 | if (val != crc) { | |
efe120a0 | 477 | printk_ratelimited(KERN_ERR "BTRFS: csum mismatch on free " |
5b0e95bf JB |
478 | "space cache\n"); |
479 | io_ctl_unmap_page(io_ctl); | |
480 | return -EIO; | |
481 | } | |
482 | ||
a67509c3 JB |
483 | return 0; |
484 | } | |
485 | ||
4c6d1d85 | 486 | static int io_ctl_add_entry(struct btrfs_io_ctl *io_ctl, u64 offset, u64 bytes, |
a67509c3 JB |
487 | void *bitmap) |
488 | { | |
489 | struct btrfs_free_space_entry *entry; | |
490 | ||
491 | if (!io_ctl->cur) | |
492 | return -ENOSPC; | |
493 | ||
494 | entry = io_ctl->cur; | |
495 | entry->offset = cpu_to_le64(offset); | |
496 | entry->bytes = cpu_to_le64(bytes); | |
497 | entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP : | |
498 | BTRFS_FREE_SPACE_EXTENT; | |
499 | io_ctl->cur += sizeof(struct btrfs_free_space_entry); | |
500 | io_ctl->size -= sizeof(struct btrfs_free_space_entry); | |
501 | ||
502 | if (io_ctl->size >= sizeof(struct btrfs_free_space_entry)) | |
503 | return 0; | |
504 | ||
5b0e95bf | 505 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); |
a67509c3 JB |
506 | |
507 | /* No more pages to map */ | |
508 | if (io_ctl->index >= io_ctl->num_pages) | |
509 | return 0; | |
510 | ||
511 | /* map the next page */ | |
512 | io_ctl_map_page(io_ctl, 1); | |
513 | return 0; | |
514 | } | |
515 | ||
4c6d1d85 | 516 | static int io_ctl_add_bitmap(struct btrfs_io_ctl *io_ctl, void *bitmap) |
a67509c3 JB |
517 | { |
518 | if (!io_ctl->cur) | |
519 | return -ENOSPC; | |
520 | ||
521 | /* | |
522 | * If we aren't at the start of the current page, unmap this one and | |
523 | * map the next one if there is any left. | |
524 | */ | |
525 | if (io_ctl->cur != io_ctl->orig) { | |
5b0e95bf | 526 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); |
a67509c3 JB |
527 | if (io_ctl->index >= io_ctl->num_pages) |
528 | return -ENOSPC; | |
529 | io_ctl_map_page(io_ctl, 0); | |
530 | } | |
531 | ||
532 | memcpy(io_ctl->cur, bitmap, PAGE_CACHE_SIZE); | |
5b0e95bf | 533 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); |
a67509c3 JB |
534 | if (io_ctl->index < io_ctl->num_pages) |
535 | io_ctl_map_page(io_ctl, 0); | |
536 | return 0; | |
537 | } | |
538 | ||
4c6d1d85 | 539 | static void io_ctl_zero_remaining_pages(struct btrfs_io_ctl *io_ctl) |
a67509c3 | 540 | { |
5b0e95bf JB |
541 | /* |
542 | * If we're not on the boundary we know we've modified the page and we | |
543 | * need to crc the page. | |
544 | */ | |
545 | if (io_ctl->cur != io_ctl->orig) | |
546 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); | |
547 | else | |
548 | io_ctl_unmap_page(io_ctl); | |
a67509c3 JB |
549 | |
550 | while (io_ctl->index < io_ctl->num_pages) { | |
551 | io_ctl_map_page(io_ctl, 1); | |
5b0e95bf | 552 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); |
a67509c3 JB |
553 | } |
554 | } | |
555 | ||
4c6d1d85 | 556 | static int io_ctl_read_entry(struct btrfs_io_ctl *io_ctl, |
5b0e95bf | 557 | struct btrfs_free_space *entry, u8 *type) |
a67509c3 JB |
558 | { |
559 | struct btrfs_free_space_entry *e; | |
2f120c05 JB |
560 | int ret; |
561 | ||
562 | if (!io_ctl->cur) { | |
563 | ret = io_ctl_check_crc(io_ctl, io_ctl->index); | |
564 | if (ret) | |
565 | return ret; | |
566 | } | |
a67509c3 JB |
567 | |
568 | e = io_ctl->cur; | |
569 | entry->offset = le64_to_cpu(e->offset); | |
570 | entry->bytes = le64_to_cpu(e->bytes); | |
5b0e95bf | 571 | *type = e->type; |
a67509c3 JB |
572 | io_ctl->cur += sizeof(struct btrfs_free_space_entry); |
573 | io_ctl->size -= sizeof(struct btrfs_free_space_entry); | |
574 | ||
575 | if (io_ctl->size >= sizeof(struct btrfs_free_space_entry)) | |
5b0e95bf | 576 | return 0; |
a67509c3 JB |
577 | |
578 | io_ctl_unmap_page(io_ctl); | |
579 | ||
2f120c05 | 580 | return 0; |
a67509c3 JB |
581 | } |
582 | ||
4c6d1d85 | 583 | static int io_ctl_read_bitmap(struct btrfs_io_ctl *io_ctl, |
5b0e95bf | 584 | struct btrfs_free_space *entry) |
a67509c3 | 585 | { |
5b0e95bf JB |
586 | int ret; |
587 | ||
5b0e95bf JB |
588 | ret = io_ctl_check_crc(io_ctl, io_ctl->index); |
589 | if (ret) | |
590 | return ret; | |
591 | ||
a67509c3 JB |
592 | memcpy(entry->bitmap, io_ctl->cur, PAGE_CACHE_SIZE); |
593 | io_ctl_unmap_page(io_ctl); | |
5b0e95bf JB |
594 | |
595 | return 0; | |
a67509c3 JB |
596 | } |
597 | ||
cd023e7b JB |
598 | /* |
599 | * Since we attach pinned extents after the fact we can have contiguous sections | |
600 | * of free space that are split up in entries. This poses a problem with the | |
601 | * tree logging stuff since it could have allocated across what appears to be 2 | |
602 | * entries since we would have merged the entries when adding the pinned extents | |
603 | * back to the free space cache. So run through the space cache that we just | |
604 | * loaded and merge contiguous entries. This will make the log replay stuff not | |
605 | * blow up and it will make for nicer allocator behavior. | |
606 | */ | |
607 | static void merge_space_tree(struct btrfs_free_space_ctl *ctl) | |
608 | { | |
609 | struct btrfs_free_space *e, *prev = NULL; | |
610 | struct rb_node *n; | |
611 | ||
612 | again: | |
613 | spin_lock(&ctl->tree_lock); | |
614 | for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) { | |
615 | e = rb_entry(n, struct btrfs_free_space, offset_index); | |
616 | if (!prev) | |
617 | goto next; | |
618 | if (e->bitmap || prev->bitmap) | |
619 | goto next; | |
620 | if (prev->offset + prev->bytes == e->offset) { | |
621 | unlink_free_space(ctl, prev); | |
622 | unlink_free_space(ctl, e); | |
623 | prev->bytes += e->bytes; | |
624 | kmem_cache_free(btrfs_free_space_cachep, e); | |
625 | link_free_space(ctl, prev); | |
626 | prev = NULL; | |
627 | spin_unlock(&ctl->tree_lock); | |
628 | goto again; | |
629 | } | |
630 | next: | |
631 | prev = e; | |
632 | } | |
633 | spin_unlock(&ctl->tree_lock); | |
634 | } | |
635 | ||
48a3b636 ES |
636 | static int __load_free_space_cache(struct btrfs_root *root, struct inode *inode, |
637 | struct btrfs_free_space_ctl *ctl, | |
638 | struct btrfs_path *path, u64 offset) | |
9d66e233 | 639 | { |
9d66e233 JB |
640 | struct btrfs_free_space_header *header; |
641 | struct extent_buffer *leaf; | |
4c6d1d85 | 642 | struct btrfs_io_ctl io_ctl; |
9d66e233 | 643 | struct btrfs_key key; |
a67509c3 | 644 | struct btrfs_free_space *e, *n; |
b76808fc | 645 | LIST_HEAD(bitmaps); |
9d66e233 JB |
646 | u64 num_entries; |
647 | u64 num_bitmaps; | |
648 | u64 generation; | |
a67509c3 | 649 | u8 type; |
f6a39829 | 650 | int ret = 0; |
9d66e233 | 651 | |
9d66e233 | 652 | /* Nothing in the space cache, goodbye */ |
0414efae | 653 | if (!i_size_read(inode)) |
a67509c3 | 654 | return 0; |
9d66e233 JB |
655 | |
656 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
0414efae | 657 | key.offset = offset; |
9d66e233 JB |
658 | key.type = 0; |
659 | ||
660 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
0414efae | 661 | if (ret < 0) |
a67509c3 | 662 | return 0; |
0414efae | 663 | else if (ret > 0) { |
945d8962 | 664 | btrfs_release_path(path); |
a67509c3 | 665 | return 0; |
9d66e233 JB |
666 | } |
667 | ||
0414efae LZ |
668 | ret = -1; |
669 | ||
9d66e233 JB |
670 | leaf = path->nodes[0]; |
671 | header = btrfs_item_ptr(leaf, path->slots[0], | |
672 | struct btrfs_free_space_header); | |
673 | num_entries = btrfs_free_space_entries(leaf, header); | |
674 | num_bitmaps = btrfs_free_space_bitmaps(leaf, header); | |
675 | generation = btrfs_free_space_generation(leaf, header); | |
945d8962 | 676 | btrfs_release_path(path); |
9d66e233 | 677 | |
e570fd27 MX |
678 | if (!BTRFS_I(inode)->generation) { |
679 | btrfs_info(root->fs_info, | |
680 | "The free space cache file (%llu) is invalid. skip it\n", | |
681 | offset); | |
682 | return 0; | |
683 | } | |
684 | ||
9d66e233 | 685 | if (BTRFS_I(inode)->generation != generation) { |
c2cf52eb SK |
686 | btrfs_err(root->fs_info, |
687 | "free space inode generation (%llu) " | |
688 | "did not match free space cache generation (%llu)", | |
c1c9ff7c | 689 | BTRFS_I(inode)->generation, generation); |
a67509c3 | 690 | return 0; |
9d66e233 JB |
691 | } |
692 | ||
693 | if (!num_entries) | |
a67509c3 | 694 | return 0; |
9d66e233 | 695 | |
5349d6c3 | 696 | ret = io_ctl_init(&io_ctl, inode, root, 0); |
706efc66 LZ |
697 | if (ret) |
698 | return ret; | |
699 | ||
9d66e233 | 700 | ret = readahead_cache(inode); |
0414efae | 701 | if (ret) |
9d66e233 | 702 | goto out; |
9d66e233 | 703 | |
a67509c3 JB |
704 | ret = io_ctl_prepare_pages(&io_ctl, inode, 1); |
705 | if (ret) | |
706 | goto out; | |
9d66e233 | 707 | |
5b0e95bf JB |
708 | ret = io_ctl_check_crc(&io_ctl, 0); |
709 | if (ret) | |
710 | goto free_cache; | |
711 | ||
a67509c3 JB |
712 | ret = io_ctl_check_generation(&io_ctl, generation); |
713 | if (ret) | |
714 | goto free_cache; | |
9d66e233 | 715 | |
a67509c3 JB |
716 | while (num_entries) { |
717 | e = kmem_cache_zalloc(btrfs_free_space_cachep, | |
718 | GFP_NOFS); | |
719 | if (!e) | |
9d66e233 | 720 | goto free_cache; |
9d66e233 | 721 | |
5b0e95bf JB |
722 | ret = io_ctl_read_entry(&io_ctl, e, &type); |
723 | if (ret) { | |
724 | kmem_cache_free(btrfs_free_space_cachep, e); | |
725 | goto free_cache; | |
726 | } | |
727 | ||
a67509c3 JB |
728 | if (!e->bytes) { |
729 | kmem_cache_free(btrfs_free_space_cachep, e); | |
730 | goto free_cache; | |
9d66e233 | 731 | } |
a67509c3 JB |
732 | |
733 | if (type == BTRFS_FREE_SPACE_EXTENT) { | |
734 | spin_lock(&ctl->tree_lock); | |
735 | ret = link_free_space(ctl, e); | |
736 | spin_unlock(&ctl->tree_lock); | |
737 | if (ret) { | |
c2cf52eb SK |
738 | btrfs_err(root->fs_info, |
739 | "Duplicate entries in free space cache, dumping"); | |
a67509c3 | 740 | kmem_cache_free(btrfs_free_space_cachep, e); |
9d66e233 JB |
741 | goto free_cache; |
742 | } | |
a67509c3 | 743 | } else { |
b12d6869 | 744 | ASSERT(num_bitmaps); |
a67509c3 JB |
745 | num_bitmaps--; |
746 | e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS); | |
747 | if (!e->bitmap) { | |
748 | kmem_cache_free( | |
749 | btrfs_free_space_cachep, e); | |
9d66e233 JB |
750 | goto free_cache; |
751 | } | |
a67509c3 JB |
752 | spin_lock(&ctl->tree_lock); |
753 | ret = link_free_space(ctl, e); | |
754 | ctl->total_bitmaps++; | |
755 | ctl->op->recalc_thresholds(ctl); | |
756 | spin_unlock(&ctl->tree_lock); | |
757 | if (ret) { | |
c2cf52eb SK |
758 | btrfs_err(root->fs_info, |
759 | "Duplicate entries in free space cache, dumping"); | |
dc89e982 | 760 | kmem_cache_free(btrfs_free_space_cachep, e); |
9d66e233 JB |
761 | goto free_cache; |
762 | } | |
a67509c3 | 763 | list_add_tail(&e->list, &bitmaps); |
9d66e233 JB |
764 | } |
765 | ||
a67509c3 JB |
766 | num_entries--; |
767 | } | |
9d66e233 | 768 | |
2f120c05 JB |
769 | io_ctl_unmap_page(&io_ctl); |
770 | ||
a67509c3 JB |
771 | /* |
772 | * We add the bitmaps at the end of the entries in order that | |
773 | * the bitmap entries are added to the cache. | |
774 | */ | |
775 | list_for_each_entry_safe(e, n, &bitmaps, list) { | |
9d66e233 | 776 | list_del_init(&e->list); |
5b0e95bf JB |
777 | ret = io_ctl_read_bitmap(&io_ctl, e); |
778 | if (ret) | |
779 | goto free_cache; | |
9d66e233 JB |
780 | } |
781 | ||
a67509c3 | 782 | io_ctl_drop_pages(&io_ctl); |
cd023e7b | 783 | merge_space_tree(ctl); |
9d66e233 JB |
784 | ret = 1; |
785 | out: | |
a67509c3 | 786 | io_ctl_free(&io_ctl); |
9d66e233 | 787 | return ret; |
9d66e233 | 788 | free_cache: |
a67509c3 | 789 | io_ctl_drop_pages(&io_ctl); |
0414efae | 790 | __btrfs_remove_free_space_cache(ctl); |
9d66e233 JB |
791 | goto out; |
792 | } | |
793 | ||
0414efae LZ |
794 | int load_free_space_cache(struct btrfs_fs_info *fs_info, |
795 | struct btrfs_block_group_cache *block_group) | |
0cb59c99 | 796 | { |
34d52cb6 | 797 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0414efae LZ |
798 | struct btrfs_root *root = fs_info->tree_root; |
799 | struct inode *inode; | |
800 | struct btrfs_path *path; | |
5b0e95bf | 801 | int ret = 0; |
0414efae LZ |
802 | bool matched; |
803 | u64 used = btrfs_block_group_used(&block_group->item); | |
804 | ||
0414efae LZ |
805 | /* |
806 | * If this block group has been marked to be cleared for one reason or | |
807 | * another then we can't trust the on disk cache, so just return. | |
808 | */ | |
9d66e233 | 809 | spin_lock(&block_group->lock); |
0414efae LZ |
810 | if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) { |
811 | spin_unlock(&block_group->lock); | |
812 | return 0; | |
813 | } | |
9d66e233 | 814 | spin_unlock(&block_group->lock); |
0414efae LZ |
815 | |
816 | path = btrfs_alloc_path(); | |
817 | if (!path) | |
818 | return 0; | |
d53ba474 JB |
819 | path->search_commit_root = 1; |
820 | path->skip_locking = 1; | |
0414efae LZ |
821 | |
822 | inode = lookup_free_space_inode(root, block_group, path); | |
823 | if (IS_ERR(inode)) { | |
824 | btrfs_free_path(path); | |
825 | return 0; | |
826 | } | |
827 | ||
5b0e95bf JB |
828 | /* We may have converted the inode and made the cache invalid. */ |
829 | spin_lock(&block_group->lock); | |
830 | if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) { | |
831 | spin_unlock(&block_group->lock); | |
a7e221e9 | 832 | btrfs_free_path(path); |
5b0e95bf JB |
833 | goto out; |
834 | } | |
835 | spin_unlock(&block_group->lock); | |
836 | ||
0414efae LZ |
837 | ret = __load_free_space_cache(fs_info->tree_root, inode, ctl, |
838 | path, block_group->key.objectid); | |
839 | btrfs_free_path(path); | |
840 | if (ret <= 0) | |
841 | goto out; | |
842 | ||
843 | spin_lock(&ctl->tree_lock); | |
844 | matched = (ctl->free_space == (block_group->key.offset - used - | |
845 | block_group->bytes_super)); | |
846 | spin_unlock(&ctl->tree_lock); | |
847 | ||
848 | if (!matched) { | |
849 | __btrfs_remove_free_space_cache(ctl); | |
32d6b47f | 850 | btrfs_warn(fs_info, "block group %llu has wrong amount of free space", |
c2cf52eb | 851 | block_group->key.objectid); |
0414efae LZ |
852 | ret = -1; |
853 | } | |
854 | out: | |
855 | if (ret < 0) { | |
856 | /* This cache is bogus, make sure it gets cleared */ | |
857 | spin_lock(&block_group->lock); | |
858 | block_group->disk_cache_state = BTRFS_DC_CLEAR; | |
859 | spin_unlock(&block_group->lock); | |
82d5902d | 860 | ret = 0; |
0414efae | 861 | |
32d6b47f | 862 | btrfs_warn(fs_info, "failed to load free space cache for block group %llu, rebuild it now", |
c2cf52eb | 863 | block_group->key.objectid); |
0414efae LZ |
864 | } |
865 | ||
866 | iput(inode); | |
867 | return ret; | |
9d66e233 JB |
868 | } |
869 | ||
d4452bc5 | 870 | static noinline_for_stack |
4c6d1d85 | 871 | int write_cache_extent_entries(struct btrfs_io_ctl *io_ctl, |
d4452bc5 CM |
872 | struct btrfs_free_space_ctl *ctl, |
873 | struct btrfs_block_group_cache *block_group, | |
874 | int *entries, int *bitmaps, | |
875 | struct list_head *bitmap_list) | |
0cb59c99 | 876 | { |
c09544e0 | 877 | int ret; |
d4452bc5 CM |
878 | struct btrfs_free_cluster *cluster = NULL; |
879 | struct rb_node *node = rb_first(&ctl->free_space_offset); | |
55507ce3 | 880 | struct btrfs_trim_range *trim_entry; |
be1a12a0 | 881 | |
43be2146 | 882 | /* Get the cluster for this block_group if it exists */ |
d4452bc5 | 883 | if (block_group && !list_empty(&block_group->cluster_list)) { |
43be2146 JB |
884 | cluster = list_entry(block_group->cluster_list.next, |
885 | struct btrfs_free_cluster, | |
886 | block_group_list); | |
d4452bc5 | 887 | } |
43be2146 | 888 | |
f75b130e JB |
889 | if (!node && cluster) { |
890 | node = rb_first(&cluster->root); | |
891 | cluster = NULL; | |
892 | } | |
893 | ||
a67509c3 JB |
894 | /* Write out the extent entries */ |
895 | while (node) { | |
896 | struct btrfs_free_space *e; | |
0cb59c99 | 897 | |
a67509c3 | 898 | e = rb_entry(node, struct btrfs_free_space, offset_index); |
d4452bc5 | 899 | *entries += 1; |
0cb59c99 | 900 | |
d4452bc5 | 901 | ret = io_ctl_add_entry(io_ctl, e->offset, e->bytes, |
a67509c3 JB |
902 | e->bitmap); |
903 | if (ret) | |
d4452bc5 | 904 | goto fail; |
2f356126 | 905 | |
a67509c3 | 906 | if (e->bitmap) { |
d4452bc5 CM |
907 | list_add_tail(&e->list, bitmap_list); |
908 | *bitmaps += 1; | |
2f356126 | 909 | } |
a67509c3 JB |
910 | node = rb_next(node); |
911 | if (!node && cluster) { | |
912 | node = rb_first(&cluster->root); | |
913 | cluster = NULL; | |
43be2146 | 914 | } |
a67509c3 | 915 | } |
55507ce3 FM |
916 | |
917 | /* | |
918 | * Make sure we don't miss any range that was removed from our rbtree | |
919 | * because trimming is running. Otherwise after a umount+mount (or crash | |
920 | * after committing the transaction) we would leak free space and get | |
921 | * an inconsistent free space cache report from fsck. | |
922 | */ | |
923 | list_for_each_entry(trim_entry, &ctl->trimming_ranges, list) { | |
924 | ret = io_ctl_add_entry(io_ctl, trim_entry->start, | |
925 | trim_entry->bytes, NULL); | |
926 | if (ret) | |
927 | goto fail; | |
928 | *entries += 1; | |
929 | } | |
930 | ||
d4452bc5 CM |
931 | return 0; |
932 | fail: | |
933 | return -ENOSPC; | |
934 | } | |
935 | ||
936 | static noinline_for_stack int | |
937 | update_cache_item(struct btrfs_trans_handle *trans, | |
938 | struct btrfs_root *root, | |
939 | struct inode *inode, | |
940 | struct btrfs_path *path, u64 offset, | |
941 | int entries, int bitmaps) | |
942 | { | |
943 | struct btrfs_key key; | |
944 | struct btrfs_free_space_header *header; | |
945 | struct extent_buffer *leaf; | |
946 | int ret; | |
947 | ||
948 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
949 | key.offset = offset; | |
950 | key.type = 0; | |
951 | ||
952 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
953 | if (ret < 0) { | |
954 | clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1, | |
955 | EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL, | |
956 | GFP_NOFS); | |
957 | goto fail; | |
958 | } | |
959 | leaf = path->nodes[0]; | |
960 | if (ret > 0) { | |
961 | struct btrfs_key found_key; | |
962 | ASSERT(path->slots[0]); | |
963 | path->slots[0]--; | |
964 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
965 | if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID || | |
966 | found_key.offset != offset) { | |
967 | clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, | |
968 | inode->i_size - 1, | |
969 | EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, | |
970 | NULL, GFP_NOFS); | |
971 | btrfs_release_path(path); | |
972 | goto fail; | |
973 | } | |
974 | } | |
975 | ||
976 | BTRFS_I(inode)->generation = trans->transid; | |
977 | header = btrfs_item_ptr(leaf, path->slots[0], | |
978 | struct btrfs_free_space_header); | |
979 | btrfs_set_free_space_entries(leaf, header, entries); | |
980 | btrfs_set_free_space_bitmaps(leaf, header, bitmaps); | |
981 | btrfs_set_free_space_generation(leaf, header, trans->transid); | |
982 | btrfs_mark_buffer_dirty(leaf); | |
983 | btrfs_release_path(path); | |
984 | ||
985 | return 0; | |
986 | ||
987 | fail: | |
988 | return -1; | |
989 | } | |
990 | ||
991 | static noinline_for_stack int | |
5349d6c3 MX |
992 | write_pinned_extent_entries(struct btrfs_root *root, |
993 | struct btrfs_block_group_cache *block_group, | |
4c6d1d85 | 994 | struct btrfs_io_ctl *io_ctl, |
5349d6c3 | 995 | int *entries) |
d4452bc5 CM |
996 | { |
997 | u64 start, extent_start, extent_end, len; | |
d4452bc5 CM |
998 | struct extent_io_tree *unpin = NULL; |
999 | int ret; | |
43be2146 | 1000 | |
5349d6c3 MX |
1001 | if (!block_group) |
1002 | return 0; | |
1003 | ||
a67509c3 JB |
1004 | /* |
1005 | * We want to add any pinned extents to our free space cache | |
1006 | * so we don't leak the space | |
d4452bc5 | 1007 | * |
db804f23 LZ |
1008 | * We shouldn't have switched the pinned extents yet so this is the |
1009 | * right one | |
1010 | */ | |
1011 | unpin = root->fs_info->pinned_extents; | |
1012 | ||
5349d6c3 | 1013 | start = block_group->key.objectid; |
db804f23 | 1014 | |
5349d6c3 | 1015 | while (start < block_group->key.objectid + block_group->key.offset) { |
db804f23 LZ |
1016 | ret = find_first_extent_bit(unpin, start, |
1017 | &extent_start, &extent_end, | |
e6138876 | 1018 | EXTENT_DIRTY, NULL); |
5349d6c3 MX |
1019 | if (ret) |
1020 | return 0; | |
0cb59c99 | 1021 | |
a67509c3 | 1022 | /* This pinned extent is out of our range */ |
db804f23 | 1023 | if (extent_start >= block_group->key.objectid + |
a67509c3 | 1024 | block_group->key.offset) |
5349d6c3 | 1025 | return 0; |
2f356126 | 1026 | |
db804f23 LZ |
1027 | extent_start = max(extent_start, start); |
1028 | extent_end = min(block_group->key.objectid + | |
1029 | block_group->key.offset, extent_end + 1); | |
1030 | len = extent_end - extent_start; | |
0cb59c99 | 1031 | |
d4452bc5 CM |
1032 | *entries += 1; |
1033 | ret = io_ctl_add_entry(io_ctl, extent_start, len, NULL); | |
a67509c3 | 1034 | if (ret) |
5349d6c3 | 1035 | return -ENOSPC; |
0cb59c99 | 1036 | |
db804f23 | 1037 | start = extent_end; |
a67509c3 | 1038 | } |
0cb59c99 | 1039 | |
5349d6c3 MX |
1040 | return 0; |
1041 | } | |
1042 | ||
1043 | static noinline_for_stack int | |
4c6d1d85 | 1044 | write_bitmap_entries(struct btrfs_io_ctl *io_ctl, struct list_head *bitmap_list) |
5349d6c3 MX |
1045 | { |
1046 | struct list_head *pos, *n; | |
1047 | int ret; | |
1048 | ||
0cb59c99 | 1049 | /* Write out the bitmaps */ |
d4452bc5 | 1050 | list_for_each_safe(pos, n, bitmap_list) { |
0cb59c99 JB |
1051 | struct btrfs_free_space *entry = |
1052 | list_entry(pos, struct btrfs_free_space, list); | |
1053 | ||
d4452bc5 | 1054 | ret = io_ctl_add_bitmap(io_ctl, entry->bitmap); |
a67509c3 | 1055 | if (ret) |
5349d6c3 | 1056 | return -ENOSPC; |
0cb59c99 | 1057 | list_del_init(&entry->list); |
be1a12a0 JB |
1058 | } |
1059 | ||
5349d6c3 MX |
1060 | return 0; |
1061 | } | |
0cb59c99 | 1062 | |
5349d6c3 MX |
1063 | static int flush_dirty_cache(struct inode *inode) |
1064 | { | |
1065 | int ret; | |
be1a12a0 | 1066 | |
0ef8b726 | 1067 | ret = btrfs_wait_ordered_range(inode, 0, (u64)-1); |
5349d6c3 | 1068 | if (ret) |
0ef8b726 JB |
1069 | clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1, |
1070 | EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL, | |
1071 | GFP_NOFS); | |
0cb59c99 | 1072 | |
5349d6c3 | 1073 | return ret; |
d4452bc5 CM |
1074 | } |
1075 | ||
1076 | static void noinline_for_stack | |
1077 | cleanup_write_cache_enospc(struct inode *inode, | |
4c6d1d85 | 1078 | struct btrfs_io_ctl *io_ctl, |
d4452bc5 CM |
1079 | struct extent_state **cached_state, |
1080 | struct list_head *bitmap_list) | |
1081 | { | |
1082 | struct list_head *pos, *n; | |
5349d6c3 | 1083 | |
d4452bc5 CM |
1084 | list_for_each_safe(pos, n, bitmap_list) { |
1085 | struct btrfs_free_space *entry = | |
1086 | list_entry(pos, struct btrfs_free_space, list); | |
1087 | list_del_init(&entry->list); | |
0cb59c99 | 1088 | } |
d4452bc5 CM |
1089 | io_ctl_drop_pages(io_ctl); |
1090 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0, | |
1091 | i_size_read(inode) - 1, cached_state, | |
1092 | GFP_NOFS); | |
1093 | } | |
549b4fdb | 1094 | |
d4452bc5 CM |
1095 | /** |
1096 | * __btrfs_write_out_cache - write out cached info to an inode | |
1097 | * @root - the root the inode belongs to | |
1098 | * @ctl - the free space cache we are going to write out | |
1099 | * @block_group - the block_group for this cache if it belongs to a block_group | |
1100 | * @trans - the trans handle | |
1101 | * @path - the path to use | |
1102 | * @offset - the offset for the key we'll insert | |
1103 | * | |
1104 | * This function writes out a free space cache struct to disk for quick recovery | |
1105 | * on mount. This will return 0 if it was successfull in writing the cache out, | |
1106 | * and -1 if it was not. | |
1107 | */ | |
1108 | static int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode, | |
1109 | struct btrfs_free_space_ctl *ctl, | |
1110 | struct btrfs_block_group_cache *block_group, | |
1111 | struct btrfs_trans_handle *trans, | |
1112 | struct btrfs_path *path, u64 offset) | |
1113 | { | |
1114 | struct extent_state *cached_state = NULL; | |
4c6d1d85 | 1115 | struct btrfs_io_ctl io_ctl; |
5349d6c3 | 1116 | LIST_HEAD(bitmap_list); |
d4452bc5 CM |
1117 | int entries = 0; |
1118 | int bitmaps = 0; | |
1119 | int ret; | |
d4452bc5 CM |
1120 | |
1121 | if (!i_size_read(inode)) | |
1122 | return -1; | |
1123 | ||
5349d6c3 | 1124 | ret = io_ctl_init(&io_ctl, inode, root, 1); |
d4452bc5 CM |
1125 | if (ret) |
1126 | return -1; | |
1127 | ||
e570fd27 MX |
1128 | if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA)) { |
1129 | down_write(&block_group->data_rwsem); | |
1130 | spin_lock(&block_group->lock); | |
1131 | if (block_group->delalloc_bytes) { | |
1132 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
1133 | spin_unlock(&block_group->lock); | |
1134 | up_write(&block_group->data_rwsem); | |
1135 | BTRFS_I(inode)->generation = 0; | |
1136 | ret = 0; | |
1137 | goto out; | |
1138 | } | |
1139 | spin_unlock(&block_group->lock); | |
1140 | } | |
1141 | ||
d4452bc5 CM |
1142 | /* Lock all pages first so we can lock the extent safely. */ |
1143 | io_ctl_prepare_pages(&io_ctl, inode, 0); | |
1144 | ||
1145 | lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1, | |
1146 | 0, &cached_state); | |
1147 | ||
d4452bc5 CM |
1148 | io_ctl_set_generation(&io_ctl, trans->transid); |
1149 | ||
55507ce3 | 1150 | mutex_lock(&ctl->cache_writeout_mutex); |
5349d6c3 | 1151 | /* Write out the extent entries in the free space cache */ |
d4452bc5 CM |
1152 | ret = write_cache_extent_entries(&io_ctl, ctl, |
1153 | block_group, &entries, &bitmaps, | |
1154 | &bitmap_list); | |
55507ce3 FM |
1155 | if (ret) { |
1156 | mutex_unlock(&ctl->cache_writeout_mutex); | |
d4452bc5 | 1157 | goto out_nospc; |
55507ce3 | 1158 | } |
d4452bc5 | 1159 | |
5349d6c3 MX |
1160 | /* |
1161 | * Some spaces that are freed in the current transaction are pinned, | |
1162 | * they will be added into free space cache after the transaction is | |
1163 | * committed, we shouldn't lose them. | |
1164 | */ | |
1165 | ret = write_pinned_extent_entries(root, block_group, &io_ctl, &entries); | |
55507ce3 FM |
1166 | if (ret) { |
1167 | mutex_unlock(&ctl->cache_writeout_mutex); | |
5349d6c3 | 1168 | goto out_nospc; |
55507ce3 | 1169 | } |
5349d6c3 | 1170 | |
55507ce3 FM |
1171 | /* |
1172 | * At last, we write out all the bitmaps and keep cache_writeout_mutex | |
1173 | * locked while doing it because a concurrent trim can be manipulating | |
1174 | * or freeing the bitmap. | |
1175 | */ | |
5349d6c3 | 1176 | ret = write_bitmap_entries(&io_ctl, &bitmap_list); |
55507ce3 | 1177 | mutex_unlock(&ctl->cache_writeout_mutex); |
5349d6c3 MX |
1178 | if (ret) |
1179 | goto out_nospc; | |
1180 | ||
1181 | /* Zero out the rest of the pages just to make sure */ | |
1182 | io_ctl_zero_remaining_pages(&io_ctl); | |
d4452bc5 | 1183 | |
5349d6c3 MX |
1184 | /* Everything is written out, now we dirty the pages in the file. */ |
1185 | ret = btrfs_dirty_pages(root, inode, io_ctl.pages, io_ctl.num_pages, | |
1186 | 0, i_size_read(inode), &cached_state); | |
1187 | if (ret) | |
d4452bc5 | 1188 | goto out_nospc; |
5349d6c3 | 1189 | |
e570fd27 MX |
1190 | if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA)) |
1191 | up_write(&block_group->data_rwsem); | |
5349d6c3 MX |
1192 | /* |
1193 | * Release the pages and unlock the extent, we will flush | |
1194 | * them out later | |
1195 | */ | |
1196 | io_ctl_drop_pages(&io_ctl); | |
1197 | ||
1198 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0, | |
1199 | i_size_read(inode) - 1, &cached_state, GFP_NOFS); | |
1200 | ||
1201 | /* Flush the dirty pages in the cache file. */ | |
1202 | ret = flush_dirty_cache(inode); | |
1203 | if (ret) | |
d4452bc5 CM |
1204 | goto out; |
1205 | ||
5349d6c3 MX |
1206 | /* Update the cache item to tell everyone this cache file is valid. */ |
1207 | ret = update_cache_item(trans, root, inode, path, offset, | |
d4452bc5 | 1208 | entries, bitmaps); |
2f356126 | 1209 | out: |
a67509c3 | 1210 | io_ctl_free(&io_ctl); |
5349d6c3 | 1211 | if (ret) { |
a67509c3 | 1212 | invalidate_inode_pages2(inode->i_mapping); |
0cb59c99 JB |
1213 | BTRFS_I(inode)->generation = 0; |
1214 | } | |
0cb59c99 | 1215 | btrfs_update_inode(trans, root, inode); |
5349d6c3 | 1216 | return ret; |
a67509c3 JB |
1217 | |
1218 | out_nospc: | |
d4452bc5 | 1219 | cleanup_write_cache_enospc(inode, &io_ctl, &cached_state, &bitmap_list); |
e570fd27 MX |
1220 | |
1221 | if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA)) | |
1222 | up_write(&block_group->data_rwsem); | |
1223 | ||
a67509c3 | 1224 | goto out; |
0414efae LZ |
1225 | } |
1226 | ||
1227 | int btrfs_write_out_cache(struct btrfs_root *root, | |
1228 | struct btrfs_trans_handle *trans, | |
1229 | struct btrfs_block_group_cache *block_group, | |
1230 | struct btrfs_path *path) | |
1231 | { | |
1232 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
1233 | struct inode *inode; | |
1234 | int ret = 0; | |
ce93ec54 | 1235 | enum btrfs_disk_cache_state dcs = BTRFS_DC_WRITTEN; |
0414efae LZ |
1236 | |
1237 | root = root->fs_info->tree_root; | |
1238 | ||
1239 | spin_lock(&block_group->lock); | |
1240 | if (block_group->disk_cache_state < BTRFS_DC_SETUP) { | |
1241 | spin_unlock(&block_group->lock); | |
1242 | return 0; | |
1243 | } | |
e570fd27 MX |
1244 | |
1245 | if (block_group->delalloc_bytes) { | |
1246 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
1247 | spin_unlock(&block_group->lock); | |
1248 | return 0; | |
1249 | } | |
0414efae LZ |
1250 | spin_unlock(&block_group->lock); |
1251 | ||
1252 | inode = lookup_free_space_inode(root, block_group, path); | |
1253 | if (IS_ERR(inode)) | |
1254 | return 0; | |
1255 | ||
1256 | ret = __btrfs_write_out_cache(root, inode, ctl, block_group, trans, | |
1257 | path, block_group->key.objectid); | |
c09544e0 | 1258 | if (ret) { |
ce93ec54 | 1259 | dcs = BTRFS_DC_ERROR; |
82d5902d | 1260 | ret = 0; |
c09544e0 | 1261 | #ifdef DEBUG |
c2cf52eb SK |
1262 | btrfs_err(root->fs_info, |
1263 | "failed to write free space cache for block group %llu", | |
1264 | block_group->key.objectid); | |
c09544e0 | 1265 | #endif |
0414efae LZ |
1266 | } |
1267 | ||
ce93ec54 JB |
1268 | spin_lock(&block_group->lock); |
1269 | block_group->disk_cache_state = dcs; | |
1270 | spin_unlock(&block_group->lock); | |
0cb59c99 JB |
1271 | iput(inode); |
1272 | return ret; | |
1273 | } | |
1274 | ||
34d52cb6 | 1275 | static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit, |
96303081 | 1276 | u64 offset) |
0f9dd46c | 1277 | { |
b12d6869 | 1278 | ASSERT(offset >= bitmap_start); |
96303081 | 1279 | offset -= bitmap_start; |
34d52cb6 | 1280 | return (unsigned long)(div_u64(offset, unit)); |
96303081 | 1281 | } |
0f9dd46c | 1282 | |
34d52cb6 | 1283 | static inline unsigned long bytes_to_bits(u64 bytes, u32 unit) |
96303081 | 1284 | { |
34d52cb6 | 1285 | return (unsigned long)(div_u64(bytes, unit)); |
96303081 | 1286 | } |
0f9dd46c | 1287 | |
34d52cb6 | 1288 | static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1289 | u64 offset) |
1290 | { | |
1291 | u64 bitmap_start; | |
b8b93add | 1292 | u32 bytes_per_bitmap; |
0f9dd46c | 1293 | |
34d52cb6 LZ |
1294 | bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit; |
1295 | bitmap_start = offset - ctl->start; | |
b8b93add | 1296 | bitmap_start = div_u64(bitmap_start, bytes_per_bitmap); |
96303081 | 1297 | bitmap_start *= bytes_per_bitmap; |
34d52cb6 | 1298 | bitmap_start += ctl->start; |
0f9dd46c | 1299 | |
96303081 | 1300 | return bitmap_start; |
0f9dd46c JB |
1301 | } |
1302 | ||
96303081 JB |
1303 | static int tree_insert_offset(struct rb_root *root, u64 offset, |
1304 | struct rb_node *node, int bitmap) | |
0f9dd46c JB |
1305 | { |
1306 | struct rb_node **p = &root->rb_node; | |
1307 | struct rb_node *parent = NULL; | |
1308 | struct btrfs_free_space *info; | |
1309 | ||
1310 | while (*p) { | |
1311 | parent = *p; | |
96303081 | 1312 | info = rb_entry(parent, struct btrfs_free_space, offset_index); |
0f9dd46c | 1313 | |
96303081 | 1314 | if (offset < info->offset) { |
0f9dd46c | 1315 | p = &(*p)->rb_left; |
96303081 | 1316 | } else if (offset > info->offset) { |
0f9dd46c | 1317 | p = &(*p)->rb_right; |
96303081 JB |
1318 | } else { |
1319 | /* | |
1320 | * we could have a bitmap entry and an extent entry | |
1321 | * share the same offset. If this is the case, we want | |
1322 | * the extent entry to always be found first if we do a | |
1323 | * linear search through the tree, since we want to have | |
1324 | * the quickest allocation time, and allocating from an | |
1325 | * extent is faster than allocating from a bitmap. So | |
1326 | * if we're inserting a bitmap and we find an entry at | |
1327 | * this offset, we want to go right, or after this entry | |
1328 | * logically. If we are inserting an extent and we've | |
1329 | * found a bitmap, we want to go left, or before | |
1330 | * logically. | |
1331 | */ | |
1332 | if (bitmap) { | |
207dde82 JB |
1333 | if (info->bitmap) { |
1334 | WARN_ON_ONCE(1); | |
1335 | return -EEXIST; | |
1336 | } | |
96303081 JB |
1337 | p = &(*p)->rb_right; |
1338 | } else { | |
207dde82 JB |
1339 | if (!info->bitmap) { |
1340 | WARN_ON_ONCE(1); | |
1341 | return -EEXIST; | |
1342 | } | |
96303081 JB |
1343 | p = &(*p)->rb_left; |
1344 | } | |
1345 | } | |
0f9dd46c JB |
1346 | } |
1347 | ||
1348 | rb_link_node(node, parent, p); | |
1349 | rb_insert_color(node, root); | |
1350 | ||
1351 | return 0; | |
1352 | } | |
1353 | ||
1354 | /* | |
70cb0743 JB |
1355 | * searches the tree for the given offset. |
1356 | * | |
96303081 JB |
1357 | * fuzzy - If this is set, then we are trying to make an allocation, and we just |
1358 | * want a section that has at least bytes size and comes at or after the given | |
1359 | * offset. | |
0f9dd46c | 1360 | */ |
96303081 | 1361 | static struct btrfs_free_space * |
34d52cb6 | 1362 | tree_search_offset(struct btrfs_free_space_ctl *ctl, |
96303081 | 1363 | u64 offset, int bitmap_only, int fuzzy) |
0f9dd46c | 1364 | { |
34d52cb6 | 1365 | struct rb_node *n = ctl->free_space_offset.rb_node; |
96303081 JB |
1366 | struct btrfs_free_space *entry, *prev = NULL; |
1367 | ||
1368 | /* find entry that is closest to the 'offset' */ | |
1369 | while (1) { | |
1370 | if (!n) { | |
1371 | entry = NULL; | |
1372 | break; | |
1373 | } | |
0f9dd46c | 1374 | |
0f9dd46c | 1375 | entry = rb_entry(n, struct btrfs_free_space, offset_index); |
96303081 | 1376 | prev = entry; |
0f9dd46c | 1377 | |
96303081 | 1378 | if (offset < entry->offset) |
0f9dd46c | 1379 | n = n->rb_left; |
96303081 | 1380 | else if (offset > entry->offset) |
0f9dd46c | 1381 | n = n->rb_right; |
96303081 | 1382 | else |
0f9dd46c | 1383 | break; |
0f9dd46c JB |
1384 | } |
1385 | ||
96303081 JB |
1386 | if (bitmap_only) { |
1387 | if (!entry) | |
1388 | return NULL; | |
1389 | if (entry->bitmap) | |
1390 | return entry; | |
0f9dd46c | 1391 | |
96303081 JB |
1392 | /* |
1393 | * bitmap entry and extent entry may share same offset, | |
1394 | * in that case, bitmap entry comes after extent entry. | |
1395 | */ | |
1396 | n = rb_next(n); | |
1397 | if (!n) | |
1398 | return NULL; | |
1399 | entry = rb_entry(n, struct btrfs_free_space, offset_index); | |
1400 | if (entry->offset != offset) | |
1401 | return NULL; | |
0f9dd46c | 1402 | |
96303081 JB |
1403 | WARN_ON(!entry->bitmap); |
1404 | return entry; | |
1405 | } else if (entry) { | |
1406 | if (entry->bitmap) { | |
0f9dd46c | 1407 | /* |
96303081 JB |
1408 | * if previous extent entry covers the offset, |
1409 | * we should return it instead of the bitmap entry | |
0f9dd46c | 1410 | */ |
de6c4115 MX |
1411 | n = rb_prev(&entry->offset_index); |
1412 | if (n) { | |
96303081 JB |
1413 | prev = rb_entry(n, struct btrfs_free_space, |
1414 | offset_index); | |
de6c4115 MX |
1415 | if (!prev->bitmap && |
1416 | prev->offset + prev->bytes > offset) | |
1417 | entry = prev; | |
0f9dd46c | 1418 | } |
96303081 JB |
1419 | } |
1420 | return entry; | |
1421 | } | |
1422 | ||
1423 | if (!prev) | |
1424 | return NULL; | |
1425 | ||
1426 | /* find last entry before the 'offset' */ | |
1427 | entry = prev; | |
1428 | if (entry->offset > offset) { | |
1429 | n = rb_prev(&entry->offset_index); | |
1430 | if (n) { | |
1431 | entry = rb_entry(n, struct btrfs_free_space, | |
1432 | offset_index); | |
b12d6869 | 1433 | ASSERT(entry->offset <= offset); |
0f9dd46c | 1434 | } else { |
96303081 JB |
1435 | if (fuzzy) |
1436 | return entry; | |
1437 | else | |
1438 | return NULL; | |
0f9dd46c JB |
1439 | } |
1440 | } | |
1441 | ||
96303081 | 1442 | if (entry->bitmap) { |
de6c4115 MX |
1443 | n = rb_prev(&entry->offset_index); |
1444 | if (n) { | |
96303081 JB |
1445 | prev = rb_entry(n, struct btrfs_free_space, |
1446 | offset_index); | |
de6c4115 MX |
1447 | if (!prev->bitmap && |
1448 | prev->offset + prev->bytes > offset) | |
1449 | return prev; | |
96303081 | 1450 | } |
34d52cb6 | 1451 | if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset) |
96303081 JB |
1452 | return entry; |
1453 | } else if (entry->offset + entry->bytes > offset) | |
1454 | return entry; | |
1455 | ||
1456 | if (!fuzzy) | |
1457 | return NULL; | |
1458 | ||
1459 | while (1) { | |
1460 | if (entry->bitmap) { | |
1461 | if (entry->offset + BITS_PER_BITMAP * | |
34d52cb6 | 1462 | ctl->unit > offset) |
96303081 JB |
1463 | break; |
1464 | } else { | |
1465 | if (entry->offset + entry->bytes > offset) | |
1466 | break; | |
1467 | } | |
1468 | ||
1469 | n = rb_next(&entry->offset_index); | |
1470 | if (!n) | |
1471 | return NULL; | |
1472 | entry = rb_entry(n, struct btrfs_free_space, offset_index); | |
1473 | } | |
1474 | return entry; | |
0f9dd46c JB |
1475 | } |
1476 | ||
f333adb5 | 1477 | static inline void |
34d52cb6 | 1478 | __unlink_free_space(struct btrfs_free_space_ctl *ctl, |
f333adb5 | 1479 | struct btrfs_free_space *info) |
0f9dd46c | 1480 | { |
34d52cb6 LZ |
1481 | rb_erase(&info->offset_index, &ctl->free_space_offset); |
1482 | ctl->free_extents--; | |
f333adb5 LZ |
1483 | } |
1484 | ||
34d52cb6 | 1485 | static void unlink_free_space(struct btrfs_free_space_ctl *ctl, |
f333adb5 LZ |
1486 | struct btrfs_free_space *info) |
1487 | { | |
34d52cb6 LZ |
1488 | __unlink_free_space(ctl, info); |
1489 | ctl->free_space -= info->bytes; | |
0f9dd46c JB |
1490 | } |
1491 | ||
34d52cb6 | 1492 | static int link_free_space(struct btrfs_free_space_ctl *ctl, |
0f9dd46c JB |
1493 | struct btrfs_free_space *info) |
1494 | { | |
1495 | int ret = 0; | |
1496 | ||
b12d6869 | 1497 | ASSERT(info->bytes || info->bitmap); |
34d52cb6 | 1498 | ret = tree_insert_offset(&ctl->free_space_offset, info->offset, |
96303081 | 1499 | &info->offset_index, (info->bitmap != NULL)); |
0f9dd46c JB |
1500 | if (ret) |
1501 | return ret; | |
1502 | ||
34d52cb6 LZ |
1503 | ctl->free_space += info->bytes; |
1504 | ctl->free_extents++; | |
96303081 JB |
1505 | return ret; |
1506 | } | |
1507 | ||
34d52cb6 | 1508 | static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl) |
96303081 | 1509 | { |
34d52cb6 | 1510 | struct btrfs_block_group_cache *block_group = ctl->private; |
25891f79 JB |
1511 | u64 max_bytes; |
1512 | u64 bitmap_bytes; | |
1513 | u64 extent_bytes; | |
8eb2d829 | 1514 | u64 size = block_group->key.offset; |
b8b93add DS |
1515 | u32 bytes_per_bg = BITS_PER_BITMAP * ctl->unit; |
1516 | u32 max_bitmaps = div_u64(size + bytes_per_bg - 1, bytes_per_bg); | |
34d52cb6 | 1517 | |
b8b93add | 1518 | max_bitmaps = max_t(u32, max_bitmaps, 1); |
dde5740f | 1519 | |
b12d6869 | 1520 | ASSERT(ctl->total_bitmaps <= max_bitmaps); |
96303081 JB |
1521 | |
1522 | /* | |
1523 | * The goal is to keep the total amount of memory used per 1gb of space | |
1524 | * at or below 32k, so we need to adjust how much memory we allow to be | |
1525 | * used by extent based free space tracking | |
1526 | */ | |
8eb2d829 LZ |
1527 | if (size < 1024 * 1024 * 1024) |
1528 | max_bytes = MAX_CACHE_BYTES_PER_GIG; | |
1529 | else | |
1530 | max_bytes = MAX_CACHE_BYTES_PER_GIG * | |
f8c269d7 | 1531 | div_u64(size, 1024 * 1024 * 1024); |
96303081 | 1532 | |
25891f79 JB |
1533 | /* |
1534 | * we want to account for 1 more bitmap than what we have so we can make | |
1535 | * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as | |
1536 | * we add more bitmaps. | |
1537 | */ | |
34d52cb6 | 1538 | bitmap_bytes = (ctl->total_bitmaps + 1) * PAGE_CACHE_SIZE; |
96303081 | 1539 | |
25891f79 | 1540 | if (bitmap_bytes >= max_bytes) { |
34d52cb6 | 1541 | ctl->extents_thresh = 0; |
25891f79 JB |
1542 | return; |
1543 | } | |
96303081 | 1544 | |
25891f79 | 1545 | /* |
f8c269d7 | 1546 | * we want the extent entry threshold to always be at most 1/2 the max |
25891f79 JB |
1547 | * bytes we can have, or whatever is less than that. |
1548 | */ | |
1549 | extent_bytes = max_bytes - bitmap_bytes; | |
f8c269d7 | 1550 | extent_bytes = min_t(u64, extent_bytes, max_bytes >> 1); |
96303081 | 1551 | |
34d52cb6 | 1552 | ctl->extents_thresh = |
f8c269d7 | 1553 | div_u64(extent_bytes, sizeof(struct btrfs_free_space)); |
96303081 JB |
1554 | } |
1555 | ||
bb3ac5a4 MX |
1556 | static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl, |
1557 | struct btrfs_free_space *info, | |
1558 | u64 offset, u64 bytes) | |
96303081 | 1559 | { |
f38b6e75 | 1560 | unsigned long start, count; |
96303081 | 1561 | |
34d52cb6 LZ |
1562 | start = offset_to_bit(info->offset, ctl->unit, offset); |
1563 | count = bytes_to_bits(bytes, ctl->unit); | |
b12d6869 | 1564 | ASSERT(start + count <= BITS_PER_BITMAP); |
96303081 | 1565 | |
f38b6e75 | 1566 | bitmap_clear(info->bitmap, start, count); |
96303081 JB |
1567 | |
1568 | info->bytes -= bytes; | |
bb3ac5a4 MX |
1569 | } |
1570 | ||
1571 | static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl, | |
1572 | struct btrfs_free_space *info, u64 offset, | |
1573 | u64 bytes) | |
1574 | { | |
1575 | __bitmap_clear_bits(ctl, info, offset, bytes); | |
34d52cb6 | 1576 | ctl->free_space -= bytes; |
96303081 JB |
1577 | } |
1578 | ||
34d52cb6 | 1579 | static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl, |
817d52f8 JB |
1580 | struct btrfs_free_space *info, u64 offset, |
1581 | u64 bytes) | |
96303081 | 1582 | { |
f38b6e75 | 1583 | unsigned long start, count; |
96303081 | 1584 | |
34d52cb6 LZ |
1585 | start = offset_to_bit(info->offset, ctl->unit, offset); |
1586 | count = bytes_to_bits(bytes, ctl->unit); | |
b12d6869 | 1587 | ASSERT(start + count <= BITS_PER_BITMAP); |
96303081 | 1588 | |
f38b6e75 | 1589 | bitmap_set(info->bitmap, start, count); |
96303081 JB |
1590 | |
1591 | info->bytes += bytes; | |
34d52cb6 | 1592 | ctl->free_space += bytes; |
96303081 JB |
1593 | } |
1594 | ||
a4820398 MX |
1595 | /* |
1596 | * If we can not find suitable extent, we will use bytes to record | |
1597 | * the size of the max extent. | |
1598 | */ | |
34d52cb6 | 1599 | static int search_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1600 | struct btrfs_free_space *bitmap_info, u64 *offset, |
1601 | u64 *bytes) | |
1602 | { | |
1603 | unsigned long found_bits = 0; | |
a4820398 | 1604 | unsigned long max_bits = 0; |
96303081 JB |
1605 | unsigned long bits, i; |
1606 | unsigned long next_zero; | |
a4820398 | 1607 | unsigned long extent_bits; |
96303081 | 1608 | |
34d52cb6 | 1609 | i = offset_to_bit(bitmap_info->offset, ctl->unit, |
96303081 | 1610 | max_t(u64, *offset, bitmap_info->offset)); |
34d52cb6 | 1611 | bits = bytes_to_bits(*bytes, ctl->unit); |
96303081 | 1612 | |
ebb3dad4 | 1613 | for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) { |
96303081 JB |
1614 | next_zero = find_next_zero_bit(bitmap_info->bitmap, |
1615 | BITS_PER_BITMAP, i); | |
a4820398 MX |
1616 | extent_bits = next_zero - i; |
1617 | if (extent_bits >= bits) { | |
1618 | found_bits = extent_bits; | |
96303081 | 1619 | break; |
a4820398 MX |
1620 | } else if (extent_bits > max_bits) { |
1621 | max_bits = extent_bits; | |
96303081 JB |
1622 | } |
1623 | i = next_zero; | |
1624 | } | |
1625 | ||
1626 | if (found_bits) { | |
34d52cb6 LZ |
1627 | *offset = (u64)(i * ctl->unit) + bitmap_info->offset; |
1628 | *bytes = (u64)(found_bits) * ctl->unit; | |
96303081 JB |
1629 | return 0; |
1630 | } | |
1631 | ||
a4820398 | 1632 | *bytes = (u64)(max_bits) * ctl->unit; |
96303081 JB |
1633 | return -1; |
1634 | } | |
1635 | ||
a4820398 | 1636 | /* Cache the size of the max extent in bytes */ |
34d52cb6 | 1637 | static struct btrfs_free_space * |
53b381b3 | 1638 | find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes, |
a4820398 | 1639 | unsigned long align, u64 *max_extent_size) |
96303081 JB |
1640 | { |
1641 | struct btrfs_free_space *entry; | |
1642 | struct rb_node *node; | |
53b381b3 DW |
1643 | u64 tmp; |
1644 | u64 align_off; | |
96303081 JB |
1645 | int ret; |
1646 | ||
34d52cb6 | 1647 | if (!ctl->free_space_offset.rb_node) |
a4820398 | 1648 | goto out; |
96303081 | 1649 | |
34d52cb6 | 1650 | entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1); |
96303081 | 1651 | if (!entry) |
a4820398 | 1652 | goto out; |
96303081 JB |
1653 | |
1654 | for (node = &entry->offset_index; node; node = rb_next(node)) { | |
1655 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
a4820398 MX |
1656 | if (entry->bytes < *bytes) { |
1657 | if (entry->bytes > *max_extent_size) | |
1658 | *max_extent_size = entry->bytes; | |
96303081 | 1659 | continue; |
a4820398 | 1660 | } |
96303081 | 1661 | |
53b381b3 DW |
1662 | /* make sure the space returned is big enough |
1663 | * to match our requested alignment | |
1664 | */ | |
1665 | if (*bytes >= align) { | |
a4820398 | 1666 | tmp = entry->offset - ctl->start + align - 1; |
47c5713f | 1667 | tmp = div64_u64(tmp, align); |
53b381b3 DW |
1668 | tmp = tmp * align + ctl->start; |
1669 | align_off = tmp - entry->offset; | |
1670 | } else { | |
1671 | align_off = 0; | |
1672 | tmp = entry->offset; | |
1673 | } | |
1674 | ||
a4820398 MX |
1675 | if (entry->bytes < *bytes + align_off) { |
1676 | if (entry->bytes > *max_extent_size) | |
1677 | *max_extent_size = entry->bytes; | |
53b381b3 | 1678 | continue; |
a4820398 | 1679 | } |
53b381b3 | 1680 | |
96303081 | 1681 | if (entry->bitmap) { |
a4820398 MX |
1682 | u64 size = *bytes; |
1683 | ||
1684 | ret = search_bitmap(ctl, entry, &tmp, &size); | |
53b381b3 DW |
1685 | if (!ret) { |
1686 | *offset = tmp; | |
a4820398 | 1687 | *bytes = size; |
96303081 | 1688 | return entry; |
a4820398 MX |
1689 | } else if (size > *max_extent_size) { |
1690 | *max_extent_size = size; | |
53b381b3 | 1691 | } |
96303081 JB |
1692 | continue; |
1693 | } | |
1694 | ||
53b381b3 DW |
1695 | *offset = tmp; |
1696 | *bytes = entry->bytes - align_off; | |
96303081 JB |
1697 | return entry; |
1698 | } | |
a4820398 | 1699 | out: |
96303081 JB |
1700 | return NULL; |
1701 | } | |
1702 | ||
34d52cb6 | 1703 | static void add_new_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1704 | struct btrfs_free_space *info, u64 offset) |
1705 | { | |
34d52cb6 | 1706 | info->offset = offset_to_bitmap(ctl, offset); |
f019f426 | 1707 | info->bytes = 0; |
f2d0f676 | 1708 | INIT_LIST_HEAD(&info->list); |
34d52cb6 LZ |
1709 | link_free_space(ctl, info); |
1710 | ctl->total_bitmaps++; | |
96303081 | 1711 | |
34d52cb6 | 1712 | ctl->op->recalc_thresholds(ctl); |
96303081 JB |
1713 | } |
1714 | ||
34d52cb6 | 1715 | static void free_bitmap(struct btrfs_free_space_ctl *ctl, |
edf6e2d1 LZ |
1716 | struct btrfs_free_space *bitmap_info) |
1717 | { | |
34d52cb6 | 1718 | unlink_free_space(ctl, bitmap_info); |
edf6e2d1 | 1719 | kfree(bitmap_info->bitmap); |
dc89e982 | 1720 | kmem_cache_free(btrfs_free_space_cachep, bitmap_info); |
34d52cb6 LZ |
1721 | ctl->total_bitmaps--; |
1722 | ctl->op->recalc_thresholds(ctl); | |
edf6e2d1 LZ |
1723 | } |
1724 | ||
34d52cb6 | 1725 | static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1726 | struct btrfs_free_space *bitmap_info, |
1727 | u64 *offset, u64 *bytes) | |
1728 | { | |
1729 | u64 end; | |
6606bb97 JB |
1730 | u64 search_start, search_bytes; |
1731 | int ret; | |
96303081 JB |
1732 | |
1733 | again: | |
34d52cb6 | 1734 | end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1; |
96303081 | 1735 | |
6606bb97 | 1736 | /* |
bdb7d303 JB |
1737 | * We need to search for bits in this bitmap. We could only cover some |
1738 | * of the extent in this bitmap thanks to how we add space, so we need | |
1739 | * to search for as much as it as we can and clear that amount, and then | |
1740 | * go searching for the next bit. | |
6606bb97 JB |
1741 | */ |
1742 | search_start = *offset; | |
bdb7d303 | 1743 | search_bytes = ctl->unit; |
13dbc089 | 1744 | search_bytes = min(search_bytes, end - search_start + 1); |
34d52cb6 | 1745 | ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes); |
b50c6e25 JB |
1746 | if (ret < 0 || search_start != *offset) |
1747 | return -EINVAL; | |
6606bb97 | 1748 | |
bdb7d303 JB |
1749 | /* We may have found more bits than what we need */ |
1750 | search_bytes = min(search_bytes, *bytes); | |
1751 | ||
1752 | /* Cannot clear past the end of the bitmap */ | |
1753 | search_bytes = min(search_bytes, end - search_start + 1); | |
1754 | ||
1755 | bitmap_clear_bits(ctl, bitmap_info, search_start, search_bytes); | |
1756 | *offset += search_bytes; | |
1757 | *bytes -= search_bytes; | |
96303081 JB |
1758 | |
1759 | if (*bytes) { | |
6606bb97 | 1760 | struct rb_node *next = rb_next(&bitmap_info->offset_index); |
edf6e2d1 | 1761 | if (!bitmap_info->bytes) |
34d52cb6 | 1762 | free_bitmap(ctl, bitmap_info); |
96303081 | 1763 | |
6606bb97 JB |
1764 | /* |
1765 | * no entry after this bitmap, but we still have bytes to | |
1766 | * remove, so something has gone wrong. | |
1767 | */ | |
1768 | if (!next) | |
96303081 JB |
1769 | return -EINVAL; |
1770 | ||
6606bb97 JB |
1771 | bitmap_info = rb_entry(next, struct btrfs_free_space, |
1772 | offset_index); | |
1773 | ||
1774 | /* | |
1775 | * if the next entry isn't a bitmap we need to return to let the | |
1776 | * extent stuff do its work. | |
1777 | */ | |
96303081 JB |
1778 | if (!bitmap_info->bitmap) |
1779 | return -EAGAIN; | |
1780 | ||
6606bb97 JB |
1781 | /* |
1782 | * Ok the next item is a bitmap, but it may not actually hold | |
1783 | * the information for the rest of this free space stuff, so | |
1784 | * look for it, and if we don't find it return so we can try | |
1785 | * everything over again. | |
1786 | */ | |
1787 | search_start = *offset; | |
bdb7d303 | 1788 | search_bytes = ctl->unit; |
34d52cb6 | 1789 | ret = search_bitmap(ctl, bitmap_info, &search_start, |
6606bb97 JB |
1790 | &search_bytes); |
1791 | if (ret < 0 || search_start != *offset) | |
1792 | return -EAGAIN; | |
1793 | ||
96303081 | 1794 | goto again; |
edf6e2d1 | 1795 | } else if (!bitmap_info->bytes) |
34d52cb6 | 1796 | free_bitmap(ctl, bitmap_info); |
96303081 JB |
1797 | |
1798 | return 0; | |
1799 | } | |
1800 | ||
2cdc342c JB |
1801 | static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl, |
1802 | struct btrfs_free_space *info, u64 offset, | |
1803 | u64 bytes) | |
1804 | { | |
1805 | u64 bytes_to_set = 0; | |
1806 | u64 end; | |
1807 | ||
1808 | end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit); | |
1809 | ||
1810 | bytes_to_set = min(end - offset, bytes); | |
1811 | ||
1812 | bitmap_set_bits(ctl, info, offset, bytes_to_set); | |
1813 | ||
1814 | return bytes_to_set; | |
1815 | ||
1816 | } | |
1817 | ||
34d52cb6 LZ |
1818 | static bool use_bitmap(struct btrfs_free_space_ctl *ctl, |
1819 | struct btrfs_free_space *info) | |
96303081 | 1820 | { |
34d52cb6 | 1821 | struct btrfs_block_group_cache *block_group = ctl->private; |
96303081 JB |
1822 | |
1823 | /* | |
1824 | * If we are below the extents threshold then we can add this as an | |
1825 | * extent, and don't have to deal with the bitmap | |
1826 | */ | |
34d52cb6 | 1827 | if (ctl->free_extents < ctl->extents_thresh) { |
32cb0840 JB |
1828 | /* |
1829 | * If this block group has some small extents we don't want to | |
1830 | * use up all of our free slots in the cache with them, we want | |
1831 | * to reserve them to larger extents, however if we have plent | |
1832 | * of cache left then go ahead an dadd them, no sense in adding | |
1833 | * the overhead of a bitmap if we don't have to. | |
1834 | */ | |
1835 | if (info->bytes <= block_group->sectorsize * 4) { | |
34d52cb6 LZ |
1836 | if (ctl->free_extents * 2 <= ctl->extents_thresh) |
1837 | return false; | |
32cb0840 | 1838 | } else { |
34d52cb6 | 1839 | return false; |
32cb0840 JB |
1840 | } |
1841 | } | |
96303081 JB |
1842 | |
1843 | /* | |
dde5740f JB |
1844 | * The original block groups from mkfs can be really small, like 8 |
1845 | * megabytes, so don't bother with a bitmap for those entries. However | |
1846 | * some block groups can be smaller than what a bitmap would cover but | |
1847 | * are still large enough that they could overflow the 32k memory limit, | |
1848 | * so allow those block groups to still be allowed to have a bitmap | |
1849 | * entry. | |
96303081 | 1850 | */ |
dde5740f | 1851 | if (((BITS_PER_BITMAP * ctl->unit) >> 1) > block_group->key.offset) |
34d52cb6 LZ |
1852 | return false; |
1853 | ||
1854 | return true; | |
1855 | } | |
1856 | ||
2cdc342c JB |
1857 | static struct btrfs_free_space_op free_space_op = { |
1858 | .recalc_thresholds = recalculate_thresholds, | |
1859 | .use_bitmap = use_bitmap, | |
1860 | }; | |
1861 | ||
34d52cb6 LZ |
1862 | static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl, |
1863 | struct btrfs_free_space *info) | |
1864 | { | |
1865 | struct btrfs_free_space *bitmap_info; | |
2cdc342c | 1866 | struct btrfs_block_group_cache *block_group = NULL; |
34d52cb6 | 1867 | int added = 0; |
2cdc342c | 1868 | u64 bytes, offset, bytes_added; |
34d52cb6 | 1869 | int ret; |
96303081 JB |
1870 | |
1871 | bytes = info->bytes; | |
1872 | offset = info->offset; | |
1873 | ||
34d52cb6 LZ |
1874 | if (!ctl->op->use_bitmap(ctl, info)) |
1875 | return 0; | |
1876 | ||
2cdc342c JB |
1877 | if (ctl->op == &free_space_op) |
1878 | block_group = ctl->private; | |
38e87880 | 1879 | again: |
2cdc342c JB |
1880 | /* |
1881 | * Since we link bitmaps right into the cluster we need to see if we | |
1882 | * have a cluster here, and if so and it has our bitmap we need to add | |
1883 | * the free space to that bitmap. | |
1884 | */ | |
1885 | if (block_group && !list_empty(&block_group->cluster_list)) { | |
1886 | struct btrfs_free_cluster *cluster; | |
1887 | struct rb_node *node; | |
1888 | struct btrfs_free_space *entry; | |
1889 | ||
1890 | cluster = list_entry(block_group->cluster_list.next, | |
1891 | struct btrfs_free_cluster, | |
1892 | block_group_list); | |
1893 | spin_lock(&cluster->lock); | |
1894 | node = rb_first(&cluster->root); | |
1895 | if (!node) { | |
1896 | spin_unlock(&cluster->lock); | |
38e87880 | 1897 | goto no_cluster_bitmap; |
2cdc342c JB |
1898 | } |
1899 | ||
1900 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
1901 | if (!entry->bitmap) { | |
1902 | spin_unlock(&cluster->lock); | |
38e87880 | 1903 | goto no_cluster_bitmap; |
2cdc342c JB |
1904 | } |
1905 | ||
1906 | if (entry->offset == offset_to_bitmap(ctl, offset)) { | |
1907 | bytes_added = add_bytes_to_bitmap(ctl, entry, | |
1908 | offset, bytes); | |
1909 | bytes -= bytes_added; | |
1910 | offset += bytes_added; | |
1911 | } | |
1912 | spin_unlock(&cluster->lock); | |
1913 | if (!bytes) { | |
1914 | ret = 1; | |
1915 | goto out; | |
1916 | } | |
1917 | } | |
38e87880 CM |
1918 | |
1919 | no_cluster_bitmap: | |
34d52cb6 | 1920 | bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), |
96303081 JB |
1921 | 1, 0); |
1922 | if (!bitmap_info) { | |
b12d6869 | 1923 | ASSERT(added == 0); |
96303081 JB |
1924 | goto new_bitmap; |
1925 | } | |
1926 | ||
2cdc342c JB |
1927 | bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes); |
1928 | bytes -= bytes_added; | |
1929 | offset += bytes_added; | |
1930 | added = 0; | |
96303081 JB |
1931 | |
1932 | if (!bytes) { | |
1933 | ret = 1; | |
1934 | goto out; | |
1935 | } else | |
1936 | goto again; | |
1937 | ||
1938 | new_bitmap: | |
1939 | if (info && info->bitmap) { | |
34d52cb6 | 1940 | add_new_bitmap(ctl, info, offset); |
96303081 JB |
1941 | added = 1; |
1942 | info = NULL; | |
1943 | goto again; | |
1944 | } else { | |
34d52cb6 | 1945 | spin_unlock(&ctl->tree_lock); |
96303081 JB |
1946 | |
1947 | /* no pre-allocated info, allocate a new one */ | |
1948 | if (!info) { | |
dc89e982 JB |
1949 | info = kmem_cache_zalloc(btrfs_free_space_cachep, |
1950 | GFP_NOFS); | |
96303081 | 1951 | if (!info) { |
34d52cb6 | 1952 | spin_lock(&ctl->tree_lock); |
96303081 JB |
1953 | ret = -ENOMEM; |
1954 | goto out; | |
1955 | } | |
1956 | } | |
1957 | ||
1958 | /* allocate the bitmap */ | |
1959 | info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS); | |
34d52cb6 | 1960 | spin_lock(&ctl->tree_lock); |
96303081 JB |
1961 | if (!info->bitmap) { |
1962 | ret = -ENOMEM; | |
1963 | goto out; | |
1964 | } | |
1965 | goto again; | |
1966 | } | |
1967 | ||
1968 | out: | |
1969 | if (info) { | |
1970 | if (info->bitmap) | |
1971 | kfree(info->bitmap); | |
dc89e982 | 1972 | kmem_cache_free(btrfs_free_space_cachep, info); |
96303081 | 1973 | } |
0f9dd46c JB |
1974 | |
1975 | return ret; | |
1976 | } | |
1977 | ||
945d8962 | 1978 | static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl, |
f333adb5 | 1979 | struct btrfs_free_space *info, bool update_stat) |
0f9dd46c | 1980 | { |
120d66ee LZ |
1981 | struct btrfs_free_space *left_info; |
1982 | struct btrfs_free_space *right_info; | |
1983 | bool merged = false; | |
1984 | u64 offset = info->offset; | |
1985 | u64 bytes = info->bytes; | |
6226cb0a | 1986 | |
0f9dd46c JB |
1987 | /* |
1988 | * first we want to see if there is free space adjacent to the range we | |
1989 | * are adding, if there is remove that struct and add a new one to | |
1990 | * cover the entire range | |
1991 | */ | |
34d52cb6 | 1992 | right_info = tree_search_offset(ctl, offset + bytes, 0, 0); |
96303081 JB |
1993 | if (right_info && rb_prev(&right_info->offset_index)) |
1994 | left_info = rb_entry(rb_prev(&right_info->offset_index), | |
1995 | struct btrfs_free_space, offset_index); | |
1996 | else | |
34d52cb6 | 1997 | left_info = tree_search_offset(ctl, offset - 1, 0, 0); |
0f9dd46c | 1998 | |
96303081 | 1999 | if (right_info && !right_info->bitmap) { |
f333adb5 | 2000 | if (update_stat) |
34d52cb6 | 2001 | unlink_free_space(ctl, right_info); |
f333adb5 | 2002 | else |
34d52cb6 | 2003 | __unlink_free_space(ctl, right_info); |
6226cb0a | 2004 | info->bytes += right_info->bytes; |
dc89e982 | 2005 | kmem_cache_free(btrfs_free_space_cachep, right_info); |
120d66ee | 2006 | merged = true; |
0f9dd46c JB |
2007 | } |
2008 | ||
96303081 JB |
2009 | if (left_info && !left_info->bitmap && |
2010 | left_info->offset + left_info->bytes == offset) { | |
f333adb5 | 2011 | if (update_stat) |
34d52cb6 | 2012 | unlink_free_space(ctl, left_info); |
f333adb5 | 2013 | else |
34d52cb6 | 2014 | __unlink_free_space(ctl, left_info); |
6226cb0a JB |
2015 | info->offset = left_info->offset; |
2016 | info->bytes += left_info->bytes; | |
dc89e982 | 2017 | kmem_cache_free(btrfs_free_space_cachep, left_info); |
120d66ee | 2018 | merged = true; |
0f9dd46c JB |
2019 | } |
2020 | ||
120d66ee LZ |
2021 | return merged; |
2022 | } | |
2023 | ||
20005523 FM |
2024 | static bool steal_from_bitmap_to_end(struct btrfs_free_space_ctl *ctl, |
2025 | struct btrfs_free_space *info, | |
2026 | bool update_stat) | |
2027 | { | |
2028 | struct btrfs_free_space *bitmap; | |
2029 | unsigned long i; | |
2030 | unsigned long j; | |
2031 | const u64 end = info->offset + info->bytes; | |
2032 | const u64 bitmap_offset = offset_to_bitmap(ctl, end); | |
2033 | u64 bytes; | |
2034 | ||
2035 | bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0); | |
2036 | if (!bitmap) | |
2037 | return false; | |
2038 | ||
2039 | i = offset_to_bit(bitmap->offset, ctl->unit, end); | |
2040 | j = find_next_zero_bit(bitmap->bitmap, BITS_PER_BITMAP, i); | |
2041 | if (j == i) | |
2042 | return false; | |
2043 | bytes = (j - i) * ctl->unit; | |
2044 | info->bytes += bytes; | |
2045 | ||
2046 | if (update_stat) | |
2047 | bitmap_clear_bits(ctl, bitmap, end, bytes); | |
2048 | else | |
2049 | __bitmap_clear_bits(ctl, bitmap, end, bytes); | |
2050 | ||
2051 | if (!bitmap->bytes) | |
2052 | free_bitmap(ctl, bitmap); | |
2053 | ||
2054 | return true; | |
2055 | } | |
2056 | ||
2057 | static bool steal_from_bitmap_to_front(struct btrfs_free_space_ctl *ctl, | |
2058 | struct btrfs_free_space *info, | |
2059 | bool update_stat) | |
2060 | { | |
2061 | struct btrfs_free_space *bitmap; | |
2062 | u64 bitmap_offset; | |
2063 | unsigned long i; | |
2064 | unsigned long j; | |
2065 | unsigned long prev_j; | |
2066 | u64 bytes; | |
2067 | ||
2068 | bitmap_offset = offset_to_bitmap(ctl, info->offset); | |
2069 | /* If we're on a boundary, try the previous logical bitmap. */ | |
2070 | if (bitmap_offset == info->offset) { | |
2071 | if (info->offset == 0) | |
2072 | return false; | |
2073 | bitmap_offset = offset_to_bitmap(ctl, info->offset - 1); | |
2074 | } | |
2075 | ||
2076 | bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0); | |
2077 | if (!bitmap) | |
2078 | return false; | |
2079 | ||
2080 | i = offset_to_bit(bitmap->offset, ctl->unit, info->offset) - 1; | |
2081 | j = 0; | |
2082 | prev_j = (unsigned long)-1; | |
2083 | for_each_clear_bit_from(j, bitmap->bitmap, BITS_PER_BITMAP) { | |
2084 | if (j > i) | |
2085 | break; | |
2086 | prev_j = j; | |
2087 | } | |
2088 | if (prev_j == i) | |
2089 | return false; | |
2090 | ||
2091 | if (prev_j == (unsigned long)-1) | |
2092 | bytes = (i + 1) * ctl->unit; | |
2093 | else | |
2094 | bytes = (i - prev_j) * ctl->unit; | |
2095 | ||
2096 | info->offset -= bytes; | |
2097 | info->bytes += bytes; | |
2098 | ||
2099 | if (update_stat) | |
2100 | bitmap_clear_bits(ctl, bitmap, info->offset, bytes); | |
2101 | else | |
2102 | __bitmap_clear_bits(ctl, bitmap, info->offset, bytes); | |
2103 | ||
2104 | if (!bitmap->bytes) | |
2105 | free_bitmap(ctl, bitmap); | |
2106 | ||
2107 | return true; | |
2108 | } | |
2109 | ||
2110 | /* | |
2111 | * We prefer always to allocate from extent entries, both for clustered and | |
2112 | * non-clustered allocation requests. So when attempting to add a new extent | |
2113 | * entry, try to see if there's adjacent free space in bitmap entries, and if | |
2114 | * there is, migrate that space from the bitmaps to the extent. | |
2115 | * Like this we get better chances of satisfying space allocation requests | |
2116 | * because we attempt to satisfy them based on a single cache entry, and never | |
2117 | * on 2 or more entries - even if the entries represent a contiguous free space | |
2118 | * region (e.g. 1 extent entry + 1 bitmap entry starting where the extent entry | |
2119 | * ends). | |
2120 | */ | |
2121 | static void steal_from_bitmap(struct btrfs_free_space_ctl *ctl, | |
2122 | struct btrfs_free_space *info, | |
2123 | bool update_stat) | |
2124 | { | |
2125 | /* | |
2126 | * Only work with disconnected entries, as we can change their offset, | |
2127 | * and must be extent entries. | |
2128 | */ | |
2129 | ASSERT(!info->bitmap); | |
2130 | ASSERT(RB_EMPTY_NODE(&info->offset_index)); | |
2131 | ||
2132 | if (ctl->total_bitmaps > 0) { | |
2133 | bool stole_end; | |
2134 | bool stole_front = false; | |
2135 | ||
2136 | stole_end = steal_from_bitmap_to_end(ctl, info, update_stat); | |
2137 | if (ctl->total_bitmaps > 0) | |
2138 | stole_front = steal_from_bitmap_to_front(ctl, info, | |
2139 | update_stat); | |
2140 | ||
2141 | if (stole_end || stole_front) | |
2142 | try_merge_free_space(ctl, info, update_stat); | |
2143 | } | |
2144 | } | |
2145 | ||
581bb050 LZ |
2146 | int __btrfs_add_free_space(struct btrfs_free_space_ctl *ctl, |
2147 | u64 offset, u64 bytes) | |
120d66ee LZ |
2148 | { |
2149 | struct btrfs_free_space *info; | |
2150 | int ret = 0; | |
2151 | ||
dc89e982 | 2152 | info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS); |
120d66ee LZ |
2153 | if (!info) |
2154 | return -ENOMEM; | |
2155 | ||
2156 | info->offset = offset; | |
2157 | info->bytes = bytes; | |
20005523 | 2158 | RB_CLEAR_NODE(&info->offset_index); |
120d66ee | 2159 | |
34d52cb6 | 2160 | spin_lock(&ctl->tree_lock); |
120d66ee | 2161 | |
34d52cb6 | 2162 | if (try_merge_free_space(ctl, info, true)) |
120d66ee LZ |
2163 | goto link; |
2164 | ||
2165 | /* | |
2166 | * There was no extent directly to the left or right of this new | |
2167 | * extent then we know we're going to have to allocate a new extent, so | |
2168 | * before we do that see if we need to drop this into a bitmap | |
2169 | */ | |
34d52cb6 | 2170 | ret = insert_into_bitmap(ctl, info); |
120d66ee LZ |
2171 | if (ret < 0) { |
2172 | goto out; | |
2173 | } else if (ret) { | |
2174 | ret = 0; | |
2175 | goto out; | |
2176 | } | |
2177 | link: | |
20005523 FM |
2178 | /* |
2179 | * Only steal free space from adjacent bitmaps if we're sure we're not | |
2180 | * going to add the new free space to existing bitmap entries - because | |
2181 | * that would mean unnecessary work that would be reverted. Therefore | |
2182 | * attempt to steal space from bitmaps if we're adding an extent entry. | |
2183 | */ | |
2184 | steal_from_bitmap(ctl, info, true); | |
2185 | ||
34d52cb6 | 2186 | ret = link_free_space(ctl, info); |
0f9dd46c | 2187 | if (ret) |
dc89e982 | 2188 | kmem_cache_free(btrfs_free_space_cachep, info); |
96303081 | 2189 | out: |
34d52cb6 | 2190 | spin_unlock(&ctl->tree_lock); |
6226cb0a | 2191 | |
0f9dd46c | 2192 | if (ret) { |
efe120a0 | 2193 | printk(KERN_CRIT "BTRFS: unable to add free space :%d\n", ret); |
b12d6869 | 2194 | ASSERT(ret != -EEXIST); |
0f9dd46c JB |
2195 | } |
2196 | ||
0f9dd46c JB |
2197 | return ret; |
2198 | } | |
2199 | ||
6226cb0a JB |
2200 | int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group, |
2201 | u64 offset, u64 bytes) | |
0f9dd46c | 2202 | { |
34d52cb6 | 2203 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0f9dd46c | 2204 | struct btrfs_free_space *info; |
b0175117 JB |
2205 | int ret; |
2206 | bool re_search = false; | |
0f9dd46c | 2207 | |
34d52cb6 | 2208 | spin_lock(&ctl->tree_lock); |
6226cb0a | 2209 | |
96303081 | 2210 | again: |
b0175117 | 2211 | ret = 0; |
bdb7d303 JB |
2212 | if (!bytes) |
2213 | goto out_lock; | |
2214 | ||
34d52cb6 | 2215 | info = tree_search_offset(ctl, offset, 0, 0); |
96303081 | 2216 | if (!info) { |
6606bb97 JB |
2217 | /* |
2218 | * oops didn't find an extent that matched the space we wanted | |
2219 | * to remove, look for a bitmap instead | |
2220 | */ | |
34d52cb6 | 2221 | info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), |
6606bb97 JB |
2222 | 1, 0); |
2223 | if (!info) { | |
b0175117 JB |
2224 | /* |
2225 | * If we found a partial bit of our free space in a | |
2226 | * bitmap but then couldn't find the other part this may | |
2227 | * be a problem, so WARN about it. | |
24a70313 | 2228 | */ |
b0175117 | 2229 | WARN_ON(re_search); |
6606bb97 JB |
2230 | goto out_lock; |
2231 | } | |
96303081 JB |
2232 | } |
2233 | ||
b0175117 | 2234 | re_search = false; |
bdb7d303 | 2235 | if (!info->bitmap) { |
34d52cb6 | 2236 | unlink_free_space(ctl, info); |
bdb7d303 JB |
2237 | if (offset == info->offset) { |
2238 | u64 to_free = min(bytes, info->bytes); | |
2239 | ||
2240 | info->bytes -= to_free; | |
2241 | info->offset += to_free; | |
2242 | if (info->bytes) { | |
2243 | ret = link_free_space(ctl, info); | |
2244 | WARN_ON(ret); | |
2245 | } else { | |
2246 | kmem_cache_free(btrfs_free_space_cachep, info); | |
2247 | } | |
0f9dd46c | 2248 | |
bdb7d303 JB |
2249 | offset += to_free; |
2250 | bytes -= to_free; | |
2251 | goto again; | |
2252 | } else { | |
2253 | u64 old_end = info->bytes + info->offset; | |
9b49c9b9 | 2254 | |
bdb7d303 | 2255 | info->bytes = offset - info->offset; |
34d52cb6 | 2256 | ret = link_free_space(ctl, info); |
96303081 JB |
2257 | WARN_ON(ret); |
2258 | if (ret) | |
2259 | goto out_lock; | |
96303081 | 2260 | |
bdb7d303 JB |
2261 | /* Not enough bytes in this entry to satisfy us */ |
2262 | if (old_end < offset + bytes) { | |
2263 | bytes -= old_end - offset; | |
2264 | offset = old_end; | |
2265 | goto again; | |
2266 | } else if (old_end == offset + bytes) { | |
2267 | /* all done */ | |
2268 | goto out_lock; | |
2269 | } | |
2270 | spin_unlock(&ctl->tree_lock); | |
2271 | ||
2272 | ret = btrfs_add_free_space(block_group, offset + bytes, | |
2273 | old_end - (offset + bytes)); | |
2274 | WARN_ON(ret); | |
2275 | goto out; | |
2276 | } | |
0f9dd46c | 2277 | } |
96303081 | 2278 | |
34d52cb6 | 2279 | ret = remove_from_bitmap(ctl, info, &offset, &bytes); |
b0175117 JB |
2280 | if (ret == -EAGAIN) { |
2281 | re_search = true; | |
96303081 | 2282 | goto again; |
b0175117 | 2283 | } |
96303081 | 2284 | out_lock: |
34d52cb6 | 2285 | spin_unlock(&ctl->tree_lock); |
0f9dd46c | 2286 | out: |
25179201 JB |
2287 | return ret; |
2288 | } | |
2289 | ||
0f9dd46c JB |
2290 | void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group, |
2291 | u64 bytes) | |
2292 | { | |
34d52cb6 | 2293 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0f9dd46c JB |
2294 | struct btrfs_free_space *info; |
2295 | struct rb_node *n; | |
2296 | int count = 0; | |
2297 | ||
34d52cb6 | 2298 | for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) { |
0f9dd46c | 2299 | info = rb_entry(n, struct btrfs_free_space, offset_index); |
f6175efa | 2300 | if (info->bytes >= bytes && !block_group->ro) |
0f9dd46c | 2301 | count++; |
efe120a0 FH |
2302 | btrfs_crit(block_group->fs_info, |
2303 | "entry offset %llu, bytes %llu, bitmap %s", | |
2304 | info->offset, info->bytes, | |
96303081 | 2305 | (info->bitmap) ? "yes" : "no"); |
0f9dd46c | 2306 | } |
efe120a0 | 2307 | btrfs_info(block_group->fs_info, "block group has cluster?: %s", |
96303081 | 2308 | list_empty(&block_group->cluster_list) ? "no" : "yes"); |
efe120a0 FH |
2309 | btrfs_info(block_group->fs_info, |
2310 | "%d blocks of free space at or bigger than bytes is", count); | |
0f9dd46c JB |
2311 | } |
2312 | ||
34d52cb6 | 2313 | void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group) |
0f9dd46c | 2314 | { |
34d52cb6 | 2315 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0f9dd46c | 2316 | |
34d52cb6 LZ |
2317 | spin_lock_init(&ctl->tree_lock); |
2318 | ctl->unit = block_group->sectorsize; | |
2319 | ctl->start = block_group->key.objectid; | |
2320 | ctl->private = block_group; | |
2321 | ctl->op = &free_space_op; | |
55507ce3 FM |
2322 | INIT_LIST_HEAD(&ctl->trimming_ranges); |
2323 | mutex_init(&ctl->cache_writeout_mutex); | |
0f9dd46c | 2324 | |
34d52cb6 LZ |
2325 | /* |
2326 | * we only want to have 32k of ram per block group for keeping | |
2327 | * track of free space, and if we pass 1/2 of that we want to | |
2328 | * start converting things over to using bitmaps | |
2329 | */ | |
2330 | ctl->extents_thresh = ((1024 * 32) / 2) / | |
2331 | sizeof(struct btrfs_free_space); | |
0f9dd46c JB |
2332 | } |
2333 | ||
fa9c0d79 CM |
2334 | /* |
2335 | * for a given cluster, put all of its extents back into the free | |
2336 | * space cache. If the block group passed doesn't match the block group | |
2337 | * pointed to by the cluster, someone else raced in and freed the | |
2338 | * cluster already. In that case, we just return without changing anything | |
2339 | */ | |
2340 | static int | |
2341 | __btrfs_return_cluster_to_free_space( | |
2342 | struct btrfs_block_group_cache *block_group, | |
2343 | struct btrfs_free_cluster *cluster) | |
2344 | { | |
34d52cb6 | 2345 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
fa9c0d79 CM |
2346 | struct btrfs_free_space *entry; |
2347 | struct rb_node *node; | |
2348 | ||
2349 | spin_lock(&cluster->lock); | |
2350 | if (cluster->block_group != block_group) | |
2351 | goto out; | |
2352 | ||
96303081 | 2353 | cluster->block_group = NULL; |
fa9c0d79 | 2354 | cluster->window_start = 0; |
96303081 | 2355 | list_del_init(&cluster->block_group_list); |
96303081 | 2356 | |
fa9c0d79 | 2357 | node = rb_first(&cluster->root); |
96303081 | 2358 | while (node) { |
4e69b598 JB |
2359 | bool bitmap; |
2360 | ||
fa9c0d79 CM |
2361 | entry = rb_entry(node, struct btrfs_free_space, offset_index); |
2362 | node = rb_next(&entry->offset_index); | |
2363 | rb_erase(&entry->offset_index, &cluster->root); | |
20005523 | 2364 | RB_CLEAR_NODE(&entry->offset_index); |
4e69b598 JB |
2365 | |
2366 | bitmap = (entry->bitmap != NULL); | |
20005523 | 2367 | if (!bitmap) { |
34d52cb6 | 2368 | try_merge_free_space(ctl, entry, false); |
20005523 FM |
2369 | steal_from_bitmap(ctl, entry, false); |
2370 | } | |
34d52cb6 | 2371 | tree_insert_offset(&ctl->free_space_offset, |
4e69b598 | 2372 | entry->offset, &entry->offset_index, bitmap); |
fa9c0d79 | 2373 | } |
6bef4d31 | 2374 | cluster->root = RB_ROOT; |
96303081 | 2375 | |
fa9c0d79 CM |
2376 | out: |
2377 | spin_unlock(&cluster->lock); | |
96303081 | 2378 | btrfs_put_block_group(block_group); |
fa9c0d79 CM |
2379 | return 0; |
2380 | } | |
2381 | ||
48a3b636 ES |
2382 | static void __btrfs_remove_free_space_cache_locked( |
2383 | struct btrfs_free_space_ctl *ctl) | |
0f9dd46c JB |
2384 | { |
2385 | struct btrfs_free_space *info; | |
2386 | struct rb_node *node; | |
581bb050 | 2387 | |
581bb050 LZ |
2388 | while ((node = rb_last(&ctl->free_space_offset)) != NULL) { |
2389 | info = rb_entry(node, struct btrfs_free_space, offset_index); | |
9b90f513 JB |
2390 | if (!info->bitmap) { |
2391 | unlink_free_space(ctl, info); | |
2392 | kmem_cache_free(btrfs_free_space_cachep, info); | |
2393 | } else { | |
2394 | free_bitmap(ctl, info); | |
2395 | } | |
351810c1 DS |
2396 | |
2397 | cond_resched_lock(&ctl->tree_lock); | |
581bb050 | 2398 | } |
09655373 CM |
2399 | } |
2400 | ||
2401 | void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl) | |
2402 | { | |
2403 | spin_lock(&ctl->tree_lock); | |
2404 | __btrfs_remove_free_space_cache_locked(ctl); | |
581bb050 LZ |
2405 | spin_unlock(&ctl->tree_lock); |
2406 | } | |
2407 | ||
2408 | void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group) | |
2409 | { | |
2410 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
fa9c0d79 | 2411 | struct btrfs_free_cluster *cluster; |
96303081 | 2412 | struct list_head *head; |
0f9dd46c | 2413 | |
34d52cb6 | 2414 | spin_lock(&ctl->tree_lock); |
96303081 JB |
2415 | while ((head = block_group->cluster_list.next) != |
2416 | &block_group->cluster_list) { | |
2417 | cluster = list_entry(head, struct btrfs_free_cluster, | |
2418 | block_group_list); | |
fa9c0d79 CM |
2419 | |
2420 | WARN_ON(cluster->block_group != block_group); | |
2421 | __btrfs_return_cluster_to_free_space(block_group, cluster); | |
351810c1 DS |
2422 | |
2423 | cond_resched_lock(&ctl->tree_lock); | |
fa9c0d79 | 2424 | } |
09655373 | 2425 | __btrfs_remove_free_space_cache_locked(ctl); |
34d52cb6 | 2426 | spin_unlock(&ctl->tree_lock); |
fa9c0d79 | 2427 | |
0f9dd46c JB |
2428 | } |
2429 | ||
6226cb0a | 2430 | u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group, |
a4820398 MX |
2431 | u64 offset, u64 bytes, u64 empty_size, |
2432 | u64 *max_extent_size) | |
0f9dd46c | 2433 | { |
34d52cb6 | 2434 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
6226cb0a | 2435 | struct btrfs_free_space *entry = NULL; |
96303081 | 2436 | u64 bytes_search = bytes + empty_size; |
6226cb0a | 2437 | u64 ret = 0; |
53b381b3 DW |
2438 | u64 align_gap = 0; |
2439 | u64 align_gap_len = 0; | |
0f9dd46c | 2440 | |
34d52cb6 | 2441 | spin_lock(&ctl->tree_lock); |
53b381b3 | 2442 | entry = find_free_space(ctl, &offset, &bytes_search, |
a4820398 | 2443 | block_group->full_stripe_len, max_extent_size); |
6226cb0a | 2444 | if (!entry) |
96303081 JB |
2445 | goto out; |
2446 | ||
2447 | ret = offset; | |
2448 | if (entry->bitmap) { | |
34d52cb6 | 2449 | bitmap_clear_bits(ctl, entry, offset, bytes); |
edf6e2d1 | 2450 | if (!entry->bytes) |
34d52cb6 | 2451 | free_bitmap(ctl, entry); |
96303081 | 2452 | } else { |
34d52cb6 | 2453 | unlink_free_space(ctl, entry); |
53b381b3 DW |
2454 | align_gap_len = offset - entry->offset; |
2455 | align_gap = entry->offset; | |
2456 | ||
2457 | entry->offset = offset + bytes; | |
2458 | WARN_ON(entry->bytes < bytes + align_gap_len); | |
2459 | ||
2460 | entry->bytes -= bytes + align_gap_len; | |
6226cb0a | 2461 | if (!entry->bytes) |
dc89e982 | 2462 | kmem_cache_free(btrfs_free_space_cachep, entry); |
6226cb0a | 2463 | else |
34d52cb6 | 2464 | link_free_space(ctl, entry); |
6226cb0a | 2465 | } |
96303081 | 2466 | out: |
34d52cb6 | 2467 | spin_unlock(&ctl->tree_lock); |
817d52f8 | 2468 | |
53b381b3 DW |
2469 | if (align_gap_len) |
2470 | __btrfs_add_free_space(ctl, align_gap, align_gap_len); | |
0f9dd46c JB |
2471 | return ret; |
2472 | } | |
fa9c0d79 CM |
2473 | |
2474 | /* | |
2475 | * given a cluster, put all of its extents back into the free space | |
2476 | * cache. If a block group is passed, this function will only free | |
2477 | * a cluster that belongs to the passed block group. | |
2478 | * | |
2479 | * Otherwise, it'll get a reference on the block group pointed to by the | |
2480 | * cluster and remove the cluster from it. | |
2481 | */ | |
2482 | int btrfs_return_cluster_to_free_space( | |
2483 | struct btrfs_block_group_cache *block_group, | |
2484 | struct btrfs_free_cluster *cluster) | |
2485 | { | |
34d52cb6 | 2486 | struct btrfs_free_space_ctl *ctl; |
fa9c0d79 CM |
2487 | int ret; |
2488 | ||
2489 | /* first, get a safe pointer to the block group */ | |
2490 | spin_lock(&cluster->lock); | |
2491 | if (!block_group) { | |
2492 | block_group = cluster->block_group; | |
2493 | if (!block_group) { | |
2494 | spin_unlock(&cluster->lock); | |
2495 | return 0; | |
2496 | } | |
2497 | } else if (cluster->block_group != block_group) { | |
2498 | /* someone else has already freed it don't redo their work */ | |
2499 | spin_unlock(&cluster->lock); | |
2500 | return 0; | |
2501 | } | |
2502 | atomic_inc(&block_group->count); | |
2503 | spin_unlock(&cluster->lock); | |
2504 | ||
34d52cb6 LZ |
2505 | ctl = block_group->free_space_ctl; |
2506 | ||
fa9c0d79 | 2507 | /* now return any extents the cluster had on it */ |
34d52cb6 | 2508 | spin_lock(&ctl->tree_lock); |
fa9c0d79 | 2509 | ret = __btrfs_return_cluster_to_free_space(block_group, cluster); |
34d52cb6 | 2510 | spin_unlock(&ctl->tree_lock); |
fa9c0d79 CM |
2511 | |
2512 | /* finally drop our ref */ | |
2513 | btrfs_put_block_group(block_group); | |
2514 | return ret; | |
2515 | } | |
2516 | ||
96303081 JB |
2517 | static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group, |
2518 | struct btrfs_free_cluster *cluster, | |
4e69b598 | 2519 | struct btrfs_free_space *entry, |
a4820398 MX |
2520 | u64 bytes, u64 min_start, |
2521 | u64 *max_extent_size) | |
96303081 | 2522 | { |
34d52cb6 | 2523 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
96303081 JB |
2524 | int err; |
2525 | u64 search_start = cluster->window_start; | |
2526 | u64 search_bytes = bytes; | |
2527 | u64 ret = 0; | |
2528 | ||
96303081 JB |
2529 | search_start = min_start; |
2530 | search_bytes = bytes; | |
2531 | ||
34d52cb6 | 2532 | err = search_bitmap(ctl, entry, &search_start, &search_bytes); |
a4820398 MX |
2533 | if (err) { |
2534 | if (search_bytes > *max_extent_size) | |
2535 | *max_extent_size = search_bytes; | |
4e69b598 | 2536 | return 0; |
a4820398 | 2537 | } |
96303081 JB |
2538 | |
2539 | ret = search_start; | |
bb3ac5a4 | 2540 | __bitmap_clear_bits(ctl, entry, ret, bytes); |
96303081 JB |
2541 | |
2542 | return ret; | |
2543 | } | |
2544 | ||
fa9c0d79 CM |
2545 | /* |
2546 | * given a cluster, try to allocate 'bytes' from it, returns 0 | |
2547 | * if it couldn't find anything suitably large, or a logical disk offset | |
2548 | * if things worked out | |
2549 | */ | |
2550 | u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group, | |
2551 | struct btrfs_free_cluster *cluster, u64 bytes, | |
a4820398 | 2552 | u64 min_start, u64 *max_extent_size) |
fa9c0d79 | 2553 | { |
34d52cb6 | 2554 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
fa9c0d79 CM |
2555 | struct btrfs_free_space *entry = NULL; |
2556 | struct rb_node *node; | |
2557 | u64 ret = 0; | |
2558 | ||
2559 | spin_lock(&cluster->lock); | |
2560 | if (bytes > cluster->max_size) | |
2561 | goto out; | |
2562 | ||
2563 | if (cluster->block_group != block_group) | |
2564 | goto out; | |
2565 | ||
2566 | node = rb_first(&cluster->root); | |
2567 | if (!node) | |
2568 | goto out; | |
2569 | ||
2570 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
67871254 | 2571 | while (1) { |
a4820398 MX |
2572 | if (entry->bytes < bytes && entry->bytes > *max_extent_size) |
2573 | *max_extent_size = entry->bytes; | |
2574 | ||
4e69b598 JB |
2575 | if (entry->bytes < bytes || |
2576 | (!entry->bitmap && entry->offset < min_start)) { | |
fa9c0d79 CM |
2577 | node = rb_next(&entry->offset_index); |
2578 | if (!node) | |
2579 | break; | |
2580 | entry = rb_entry(node, struct btrfs_free_space, | |
2581 | offset_index); | |
2582 | continue; | |
2583 | } | |
fa9c0d79 | 2584 | |
4e69b598 JB |
2585 | if (entry->bitmap) { |
2586 | ret = btrfs_alloc_from_bitmap(block_group, | |
2587 | cluster, entry, bytes, | |
a4820398 MX |
2588 | cluster->window_start, |
2589 | max_extent_size); | |
4e69b598 | 2590 | if (ret == 0) { |
4e69b598 JB |
2591 | node = rb_next(&entry->offset_index); |
2592 | if (!node) | |
2593 | break; | |
2594 | entry = rb_entry(node, struct btrfs_free_space, | |
2595 | offset_index); | |
2596 | continue; | |
2597 | } | |
9b230628 | 2598 | cluster->window_start += bytes; |
4e69b598 | 2599 | } else { |
4e69b598 JB |
2600 | ret = entry->offset; |
2601 | ||
2602 | entry->offset += bytes; | |
2603 | entry->bytes -= bytes; | |
2604 | } | |
fa9c0d79 | 2605 | |
5e71b5d5 | 2606 | if (entry->bytes == 0) |
fa9c0d79 | 2607 | rb_erase(&entry->offset_index, &cluster->root); |
fa9c0d79 CM |
2608 | break; |
2609 | } | |
2610 | out: | |
2611 | spin_unlock(&cluster->lock); | |
96303081 | 2612 | |
5e71b5d5 LZ |
2613 | if (!ret) |
2614 | return 0; | |
2615 | ||
34d52cb6 | 2616 | spin_lock(&ctl->tree_lock); |
5e71b5d5 | 2617 | |
34d52cb6 | 2618 | ctl->free_space -= bytes; |
5e71b5d5 | 2619 | if (entry->bytes == 0) { |
34d52cb6 | 2620 | ctl->free_extents--; |
4e69b598 JB |
2621 | if (entry->bitmap) { |
2622 | kfree(entry->bitmap); | |
34d52cb6 LZ |
2623 | ctl->total_bitmaps--; |
2624 | ctl->op->recalc_thresholds(ctl); | |
4e69b598 | 2625 | } |
dc89e982 | 2626 | kmem_cache_free(btrfs_free_space_cachep, entry); |
5e71b5d5 LZ |
2627 | } |
2628 | ||
34d52cb6 | 2629 | spin_unlock(&ctl->tree_lock); |
5e71b5d5 | 2630 | |
fa9c0d79 CM |
2631 | return ret; |
2632 | } | |
2633 | ||
96303081 JB |
2634 | static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group, |
2635 | struct btrfs_free_space *entry, | |
2636 | struct btrfs_free_cluster *cluster, | |
1bb91902 AO |
2637 | u64 offset, u64 bytes, |
2638 | u64 cont1_bytes, u64 min_bytes) | |
96303081 | 2639 | { |
34d52cb6 | 2640 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
96303081 JB |
2641 | unsigned long next_zero; |
2642 | unsigned long i; | |
1bb91902 AO |
2643 | unsigned long want_bits; |
2644 | unsigned long min_bits; | |
96303081 JB |
2645 | unsigned long found_bits; |
2646 | unsigned long start = 0; | |
2647 | unsigned long total_found = 0; | |
4e69b598 | 2648 | int ret; |
96303081 | 2649 | |
96009762 | 2650 | i = offset_to_bit(entry->offset, ctl->unit, |
96303081 | 2651 | max_t(u64, offset, entry->offset)); |
96009762 WSH |
2652 | want_bits = bytes_to_bits(bytes, ctl->unit); |
2653 | min_bits = bytes_to_bits(min_bytes, ctl->unit); | |
96303081 JB |
2654 | |
2655 | again: | |
2656 | found_bits = 0; | |
ebb3dad4 | 2657 | for_each_set_bit_from(i, entry->bitmap, BITS_PER_BITMAP) { |
96303081 JB |
2658 | next_zero = find_next_zero_bit(entry->bitmap, |
2659 | BITS_PER_BITMAP, i); | |
1bb91902 | 2660 | if (next_zero - i >= min_bits) { |
96303081 JB |
2661 | found_bits = next_zero - i; |
2662 | break; | |
2663 | } | |
2664 | i = next_zero; | |
2665 | } | |
2666 | ||
2667 | if (!found_bits) | |
4e69b598 | 2668 | return -ENOSPC; |
96303081 | 2669 | |
1bb91902 | 2670 | if (!total_found) { |
96303081 | 2671 | start = i; |
b78d09bc | 2672 | cluster->max_size = 0; |
96303081 JB |
2673 | } |
2674 | ||
2675 | total_found += found_bits; | |
2676 | ||
96009762 WSH |
2677 | if (cluster->max_size < found_bits * ctl->unit) |
2678 | cluster->max_size = found_bits * ctl->unit; | |
96303081 | 2679 | |
1bb91902 AO |
2680 | if (total_found < want_bits || cluster->max_size < cont1_bytes) { |
2681 | i = next_zero + 1; | |
96303081 JB |
2682 | goto again; |
2683 | } | |
2684 | ||
96009762 | 2685 | cluster->window_start = start * ctl->unit + entry->offset; |
34d52cb6 | 2686 | rb_erase(&entry->offset_index, &ctl->free_space_offset); |
4e69b598 JB |
2687 | ret = tree_insert_offset(&cluster->root, entry->offset, |
2688 | &entry->offset_index, 1); | |
b12d6869 | 2689 | ASSERT(!ret); /* -EEXIST; Logic error */ |
96303081 | 2690 | |
3f7de037 | 2691 | trace_btrfs_setup_cluster(block_group, cluster, |
96009762 | 2692 | total_found * ctl->unit, 1); |
96303081 JB |
2693 | return 0; |
2694 | } | |
2695 | ||
4e69b598 JB |
2696 | /* |
2697 | * This searches the block group for just extents to fill the cluster with. | |
1bb91902 AO |
2698 | * Try to find a cluster with at least bytes total bytes, at least one |
2699 | * extent of cont1_bytes, and other clusters of at least min_bytes. | |
4e69b598 | 2700 | */ |
3de85bb9 JB |
2701 | static noinline int |
2702 | setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group, | |
2703 | struct btrfs_free_cluster *cluster, | |
2704 | struct list_head *bitmaps, u64 offset, u64 bytes, | |
1bb91902 | 2705 | u64 cont1_bytes, u64 min_bytes) |
4e69b598 | 2706 | { |
34d52cb6 | 2707 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
4e69b598 JB |
2708 | struct btrfs_free_space *first = NULL; |
2709 | struct btrfs_free_space *entry = NULL; | |
4e69b598 JB |
2710 | struct btrfs_free_space *last; |
2711 | struct rb_node *node; | |
4e69b598 JB |
2712 | u64 window_free; |
2713 | u64 max_extent; | |
3f7de037 | 2714 | u64 total_size = 0; |
4e69b598 | 2715 | |
34d52cb6 | 2716 | entry = tree_search_offset(ctl, offset, 0, 1); |
4e69b598 JB |
2717 | if (!entry) |
2718 | return -ENOSPC; | |
2719 | ||
2720 | /* | |
2721 | * We don't want bitmaps, so just move along until we find a normal | |
2722 | * extent entry. | |
2723 | */ | |
1bb91902 AO |
2724 | while (entry->bitmap || entry->bytes < min_bytes) { |
2725 | if (entry->bitmap && list_empty(&entry->list)) | |
86d4a77b | 2726 | list_add_tail(&entry->list, bitmaps); |
4e69b598 JB |
2727 | node = rb_next(&entry->offset_index); |
2728 | if (!node) | |
2729 | return -ENOSPC; | |
2730 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
2731 | } | |
2732 | ||
4e69b598 JB |
2733 | window_free = entry->bytes; |
2734 | max_extent = entry->bytes; | |
2735 | first = entry; | |
2736 | last = entry; | |
4e69b598 | 2737 | |
1bb91902 AO |
2738 | for (node = rb_next(&entry->offset_index); node; |
2739 | node = rb_next(&entry->offset_index)) { | |
4e69b598 JB |
2740 | entry = rb_entry(node, struct btrfs_free_space, offset_index); |
2741 | ||
86d4a77b JB |
2742 | if (entry->bitmap) { |
2743 | if (list_empty(&entry->list)) | |
2744 | list_add_tail(&entry->list, bitmaps); | |
4e69b598 | 2745 | continue; |
86d4a77b JB |
2746 | } |
2747 | ||
1bb91902 AO |
2748 | if (entry->bytes < min_bytes) |
2749 | continue; | |
2750 | ||
2751 | last = entry; | |
2752 | window_free += entry->bytes; | |
2753 | if (entry->bytes > max_extent) | |
4e69b598 | 2754 | max_extent = entry->bytes; |
4e69b598 JB |
2755 | } |
2756 | ||
1bb91902 AO |
2757 | if (window_free < bytes || max_extent < cont1_bytes) |
2758 | return -ENOSPC; | |
2759 | ||
4e69b598 JB |
2760 | cluster->window_start = first->offset; |
2761 | ||
2762 | node = &first->offset_index; | |
2763 | ||
2764 | /* | |
2765 | * now we've found our entries, pull them out of the free space | |
2766 | * cache and put them into the cluster rbtree | |
2767 | */ | |
2768 | do { | |
2769 | int ret; | |
2770 | ||
2771 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
2772 | node = rb_next(&entry->offset_index); | |
1bb91902 | 2773 | if (entry->bitmap || entry->bytes < min_bytes) |
4e69b598 JB |
2774 | continue; |
2775 | ||
34d52cb6 | 2776 | rb_erase(&entry->offset_index, &ctl->free_space_offset); |
4e69b598 JB |
2777 | ret = tree_insert_offset(&cluster->root, entry->offset, |
2778 | &entry->offset_index, 0); | |
3f7de037 | 2779 | total_size += entry->bytes; |
b12d6869 | 2780 | ASSERT(!ret); /* -EEXIST; Logic error */ |
4e69b598 JB |
2781 | } while (node && entry != last); |
2782 | ||
2783 | cluster->max_size = max_extent; | |
3f7de037 | 2784 | trace_btrfs_setup_cluster(block_group, cluster, total_size, 0); |
4e69b598 JB |
2785 | return 0; |
2786 | } | |
2787 | ||
2788 | /* | |
2789 | * This specifically looks for bitmaps that may work in the cluster, we assume | |
2790 | * that we have already failed to find extents that will work. | |
2791 | */ | |
3de85bb9 JB |
2792 | static noinline int |
2793 | setup_cluster_bitmap(struct btrfs_block_group_cache *block_group, | |
2794 | struct btrfs_free_cluster *cluster, | |
2795 | struct list_head *bitmaps, u64 offset, u64 bytes, | |
1bb91902 | 2796 | u64 cont1_bytes, u64 min_bytes) |
4e69b598 | 2797 | { |
34d52cb6 | 2798 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
4e69b598 | 2799 | struct btrfs_free_space *entry; |
4e69b598 | 2800 | int ret = -ENOSPC; |
0f0fbf1d | 2801 | u64 bitmap_offset = offset_to_bitmap(ctl, offset); |
4e69b598 | 2802 | |
34d52cb6 | 2803 | if (ctl->total_bitmaps == 0) |
4e69b598 JB |
2804 | return -ENOSPC; |
2805 | ||
0f0fbf1d LZ |
2806 | /* |
2807 | * The bitmap that covers offset won't be in the list unless offset | |
2808 | * is just its start offset. | |
2809 | */ | |
2810 | entry = list_first_entry(bitmaps, struct btrfs_free_space, list); | |
2811 | if (entry->offset != bitmap_offset) { | |
2812 | entry = tree_search_offset(ctl, bitmap_offset, 1, 0); | |
2813 | if (entry && list_empty(&entry->list)) | |
2814 | list_add(&entry->list, bitmaps); | |
2815 | } | |
2816 | ||
86d4a77b | 2817 | list_for_each_entry(entry, bitmaps, list) { |
357b9784 | 2818 | if (entry->bytes < bytes) |
86d4a77b JB |
2819 | continue; |
2820 | ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset, | |
1bb91902 | 2821 | bytes, cont1_bytes, min_bytes); |
86d4a77b JB |
2822 | if (!ret) |
2823 | return 0; | |
2824 | } | |
2825 | ||
2826 | /* | |
52621cb6 LZ |
2827 | * The bitmaps list has all the bitmaps that record free space |
2828 | * starting after offset, so no more search is required. | |
86d4a77b | 2829 | */ |
52621cb6 | 2830 | return -ENOSPC; |
4e69b598 JB |
2831 | } |
2832 | ||
fa9c0d79 CM |
2833 | /* |
2834 | * here we try to find a cluster of blocks in a block group. The goal | |
1bb91902 | 2835 | * is to find at least bytes+empty_size. |
fa9c0d79 CM |
2836 | * We might not find them all in one contiguous area. |
2837 | * | |
2838 | * returns zero and sets up cluster if things worked out, otherwise | |
2839 | * it returns -enospc | |
2840 | */ | |
00361589 | 2841 | int btrfs_find_space_cluster(struct btrfs_root *root, |
fa9c0d79 CM |
2842 | struct btrfs_block_group_cache *block_group, |
2843 | struct btrfs_free_cluster *cluster, | |
2844 | u64 offset, u64 bytes, u64 empty_size) | |
2845 | { | |
34d52cb6 | 2846 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
86d4a77b | 2847 | struct btrfs_free_space *entry, *tmp; |
52621cb6 | 2848 | LIST_HEAD(bitmaps); |
fa9c0d79 | 2849 | u64 min_bytes; |
1bb91902 | 2850 | u64 cont1_bytes; |
fa9c0d79 CM |
2851 | int ret; |
2852 | ||
1bb91902 AO |
2853 | /* |
2854 | * Choose the minimum extent size we'll require for this | |
2855 | * cluster. For SSD_SPREAD, don't allow any fragmentation. | |
2856 | * For metadata, allow allocates with smaller extents. For | |
2857 | * data, keep it dense. | |
2858 | */ | |
451d7585 | 2859 | if (btrfs_test_opt(root, SSD_SPREAD)) { |
1bb91902 | 2860 | cont1_bytes = min_bytes = bytes + empty_size; |
451d7585 | 2861 | } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) { |
1bb91902 AO |
2862 | cont1_bytes = bytes; |
2863 | min_bytes = block_group->sectorsize; | |
2864 | } else { | |
2865 | cont1_bytes = max(bytes, (bytes + empty_size) >> 2); | |
2866 | min_bytes = block_group->sectorsize; | |
2867 | } | |
fa9c0d79 | 2868 | |
34d52cb6 | 2869 | spin_lock(&ctl->tree_lock); |
7d0d2e8e JB |
2870 | |
2871 | /* | |
2872 | * If we know we don't have enough space to make a cluster don't even | |
2873 | * bother doing all the work to try and find one. | |
2874 | */ | |
1bb91902 | 2875 | if (ctl->free_space < bytes) { |
34d52cb6 | 2876 | spin_unlock(&ctl->tree_lock); |
7d0d2e8e JB |
2877 | return -ENOSPC; |
2878 | } | |
2879 | ||
fa9c0d79 CM |
2880 | spin_lock(&cluster->lock); |
2881 | ||
2882 | /* someone already found a cluster, hooray */ | |
2883 | if (cluster->block_group) { | |
2884 | ret = 0; | |
2885 | goto out; | |
2886 | } | |
fa9c0d79 | 2887 | |
3f7de037 JB |
2888 | trace_btrfs_find_cluster(block_group, offset, bytes, empty_size, |
2889 | min_bytes); | |
2890 | ||
86d4a77b | 2891 | ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset, |
1bb91902 AO |
2892 | bytes + empty_size, |
2893 | cont1_bytes, min_bytes); | |
4e69b598 | 2894 | if (ret) |
86d4a77b | 2895 | ret = setup_cluster_bitmap(block_group, cluster, &bitmaps, |
1bb91902 AO |
2896 | offset, bytes + empty_size, |
2897 | cont1_bytes, min_bytes); | |
86d4a77b JB |
2898 | |
2899 | /* Clear our temporary list */ | |
2900 | list_for_each_entry_safe(entry, tmp, &bitmaps, list) | |
2901 | list_del_init(&entry->list); | |
fa9c0d79 | 2902 | |
4e69b598 JB |
2903 | if (!ret) { |
2904 | atomic_inc(&block_group->count); | |
2905 | list_add_tail(&cluster->block_group_list, | |
2906 | &block_group->cluster_list); | |
2907 | cluster->block_group = block_group; | |
3f7de037 JB |
2908 | } else { |
2909 | trace_btrfs_failed_cluster_setup(block_group); | |
fa9c0d79 | 2910 | } |
fa9c0d79 CM |
2911 | out: |
2912 | spin_unlock(&cluster->lock); | |
34d52cb6 | 2913 | spin_unlock(&ctl->tree_lock); |
fa9c0d79 CM |
2914 | |
2915 | return ret; | |
2916 | } | |
2917 | ||
2918 | /* | |
2919 | * simple code to zero out a cluster | |
2920 | */ | |
2921 | void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster) | |
2922 | { | |
2923 | spin_lock_init(&cluster->lock); | |
2924 | spin_lock_init(&cluster->refill_lock); | |
6bef4d31 | 2925 | cluster->root = RB_ROOT; |
fa9c0d79 CM |
2926 | cluster->max_size = 0; |
2927 | INIT_LIST_HEAD(&cluster->block_group_list); | |
2928 | cluster->block_group = NULL; | |
2929 | } | |
2930 | ||
7fe1e641 LZ |
2931 | static int do_trimming(struct btrfs_block_group_cache *block_group, |
2932 | u64 *total_trimmed, u64 start, u64 bytes, | |
55507ce3 FM |
2933 | u64 reserved_start, u64 reserved_bytes, |
2934 | struct btrfs_trim_range *trim_entry) | |
f7039b1d | 2935 | { |
7fe1e641 | 2936 | struct btrfs_space_info *space_info = block_group->space_info; |
f7039b1d | 2937 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
55507ce3 | 2938 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
7fe1e641 LZ |
2939 | int ret; |
2940 | int update = 0; | |
2941 | u64 trimmed = 0; | |
f7039b1d | 2942 | |
7fe1e641 LZ |
2943 | spin_lock(&space_info->lock); |
2944 | spin_lock(&block_group->lock); | |
2945 | if (!block_group->ro) { | |
2946 | block_group->reserved += reserved_bytes; | |
2947 | space_info->bytes_reserved += reserved_bytes; | |
2948 | update = 1; | |
2949 | } | |
2950 | spin_unlock(&block_group->lock); | |
2951 | spin_unlock(&space_info->lock); | |
2952 | ||
1edb647b FM |
2953 | ret = btrfs_discard_extent(fs_info->extent_root, |
2954 | start, bytes, &trimmed); | |
7fe1e641 LZ |
2955 | if (!ret) |
2956 | *total_trimmed += trimmed; | |
2957 | ||
55507ce3 | 2958 | mutex_lock(&ctl->cache_writeout_mutex); |
7fe1e641 | 2959 | btrfs_add_free_space(block_group, reserved_start, reserved_bytes); |
55507ce3 FM |
2960 | list_del(&trim_entry->list); |
2961 | mutex_unlock(&ctl->cache_writeout_mutex); | |
7fe1e641 LZ |
2962 | |
2963 | if (update) { | |
2964 | spin_lock(&space_info->lock); | |
2965 | spin_lock(&block_group->lock); | |
2966 | if (block_group->ro) | |
2967 | space_info->bytes_readonly += reserved_bytes; | |
2968 | block_group->reserved -= reserved_bytes; | |
2969 | space_info->bytes_reserved -= reserved_bytes; | |
2970 | spin_unlock(&space_info->lock); | |
2971 | spin_unlock(&block_group->lock); | |
2972 | } | |
2973 | ||
2974 | return ret; | |
2975 | } | |
2976 | ||
2977 | static int trim_no_bitmap(struct btrfs_block_group_cache *block_group, | |
2978 | u64 *total_trimmed, u64 start, u64 end, u64 minlen) | |
2979 | { | |
2980 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
2981 | struct btrfs_free_space *entry; | |
2982 | struct rb_node *node; | |
2983 | int ret = 0; | |
2984 | u64 extent_start; | |
2985 | u64 extent_bytes; | |
2986 | u64 bytes; | |
f7039b1d LD |
2987 | |
2988 | while (start < end) { | |
55507ce3 FM |
2989 | struct btrfs_trim_range trim_entry; |
2990 | ||
2991 | mutex_lock(&ctl->cache_writeout_mutex); | |
34d52cb6 | 2992 | spin_lock(&ctl->tree_lock); |
f7039b1d | 2993 | |
34d52cb6 LZ |
2994 | if (ctl->free_space < minlen) { |
2995 | spin_unlock(&ctl->tree_lock); | |
55507ce3 | 2996 | mutex_unlock(&ctl->cache_writeout_mutex); |
f7039b1d LD |
2997 | break; |
2998 | } | |
2999 | ||
34d52cb6 | 3000 | entry = tree_search_offset(ctl, start, 0, 1); |
7fe1e641 | 3001 | if (!entry) { |
34d52cb6 | 3002 | spin_unlock(&ctl->tree_lock); |
55507ce3 | 3003 | mutex_unlock(&ctl->cache_writeout_mutex); |
f7039b1d LD |
3004 | break; |
3005 | } | |
3006 | ||
7fe1e641 LZ |
3007 | /* skip bitmaps */ |
3008 | while (entry->bitmap) { | |
3009 | node = rb_next(&entry->offset_index); | |
3010 | if (!node) { | |
34d52cb6 | 3011 | spin_unlock(&ctl->tree_lock); |
55507ce3 | 3012 | mutex_unlock(&ctl->cache_writeout_mutex); |
7fe1e641 | 3013 | goto out; |
f7039b1d | 3014 | } |
7fe1e641 LZ |
3015 | entry = rb_entry(node, struct btrfs_free_space, |
3016 | offset_index); | |
f7039b1d LD |
3017 | } |
3018 | ||
7fe1e641 LZ |
3019 | if (entry->offset >= end) { |
3020 | spin_unlock(&ctl->tree_lock); | |
55507ce3 | 3021 | mutex_unlock(&ctl->cache_writeout_mutex); |
7fe1e641 | 3022 | break; |
f7039b1d LD |
3023 | } |
3024 | ||
7fe1e641 LZ |
3025 | extent_start = entry->offset; |
3026 | extent_bytes = entry->bytes; | |
3027 | start = max(start, extent_start); | |
3028 | bytes = min(extent_start + extent_bytes, end) - start; | |
3029 | if (bytes < minlen) { | |
3030 | spin_unlock(&ctl->tree_lock); | |
55507ce3 | 3031 | mutex_unlock(&ctl->cache_writeout_mutex); |
7fe1e641 | 3032 | goto next; |
f7039b1d LD |
3033 | } |
3034 | ||
7fe1e641 LZ |
3035 | unlink_free_space(ctl, entry); |
3036 | kmem_cache_free(btrfs_free_space_cachep, entry); | |
3037 | ||
34d52cb6 | 3038 | spin_unlock(&ctl->tree_lock); |
55507ce3 FM |
3039 | trim_entry.start = extent_start; |
3040 | trim_entry.bytes = extent_bytes; | |
3041 | list_add_tail(&trim_entry.list, &ctl->trimming_ranges); | |
3042 | mutex_unlock(&ctl->cache_writeout_mutex); | |
f7039b1d | 3043 | |
7fe1e641 | 3044 | ret = do_trimming(block_group, total_trimmed, start, bytes, |
55507ce3 | 3045 | extent_start, extent_bytes, &trim_entry); |
7fe1e641 LZ |
3046 | if (ret) |
3047 | break; | |
3048 | next: | |
3049 | start += bytes; | |
f7039b1d | 3050 | |
7fe1e641 LZ |
3051 | if (fatal_signal_pending(current)) { |
3052 | ret = -ERESTARTSYS; | |
3053 | break; | |
3054 | } | |
3055 | ||
3056 | cond_resched(); | |
3057 | } | |
3058 | out: | |
3059 | return ret; | |
3060 | } | |
3061 | ||
3062 | static int trim_bitmaps(struct btrfs_block_group_cache *block_group, | |
3063 | u64 *total_trimmed, u64 start, u64 end, u64 minlen) | |
3064 | { | |
3065 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
3066 | struct btrfs_free_space *entry; | |
3067 | int ret = 0; | |
3068 | int ret2; | |
3069 | u64 bytes; | |
3070 | u64 offset = offset_to_bitmap(ctl, start); | |
3071 | ||
3072 | while (offset < end) { | |
3073 | bool next_bitmap = false; | |
55507ce3 | 3074 | struct btrfs_trim_range trim_entry; |
7fe1e641 | 3075 | |
55507ce3 | 3076 | mutex_lock(&ctl->cache_writeout_mutex); |
7fe1e641 LZ |
3077 | spin_lock(&ctl->tree_lock); |
3078 | ||
3079 | if (ctl->free_space < minlen) { | |
3080 | spin_unlock(&ctl->tree_lock); | |
55507ce3 | 3081 | mutex_unlock(&ctl->cache_writeout_mutex); |
7fe1e641 LZ |
3082 | break; |
3083 | } | |
3084 | ||
3085 | entry = tree_search_offset(ctl, offset, 1, 0); | |
3086 | if (!entry) { | |
3087 | spin_unlock(&ctl->tree_lock); | |
55507ce3 | 3088 | mutex_unlock(&ctl->cache_writeout_mutex); |
7fe1e641 LZ |
3089 | next_bitmap = true; |
3090 | goto next; | |
3091 | } | |
3092 | ||
3093 | bytes = minlen; | |
3094 | ret2 = search_bitmap(ctl, entry, &start, &bytes); | |
3095 | if (ret2 || start >= end) { | |
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 = min(bytes, end - start); | |
3103 | if (bytes < minlen) { | |
3104 | spin_unlock(&ctl->tree_lock); | |
55507ce3 | 3105 | mutex_unlock(&ctl->cache_writeout_mutex); |
7fe1e641 LZ |
3106 | goto next; |
3107 | } | |
3108 | ||
3109 | bitmap_clear_bits(ctl, entry, start, bytes); | |
3110 | if (entry->bytes == 0) | |
3111 | free_bitmap(ctl, entry); | |
3112 | ||
3113 | spin_unlock(&ctl->tree_lock); | |
55507ce3 FM |
3114 | trim_entry.start = start; |
3115 | trim_entry.bytes = bytes; | |
3116 | list_add_tail(&trim_entry.list, &ctl->trimming_ranges); | |
3117 | mutex_unlock(&ctl->cache_writeout_mutex); | |
7fe1e641 LZ |
3118 | |
3119 | ret = do_trimming(block_group, total_trimmed, start, bytes, | |
55507ce3 | 3120 | start, bytes, &trim_entry); |
7fe1e641 LZ |
3121 | if (ret) |
3122 | break; | |
3123 | next: | |
3124 | if (next_bitmap) { | |
3125 | offset += BITS_PER_BITMAP * ctl->unit; | |
3126 | } else { | |
3127 | start += bytes; | |
3128 | if (start >= offset + BITS_PER_BITMAP * ctl->unit) | |
3129 | offset += BITS_PER_BITMAP * ctl->unit; | |
f7039b1d | 3130 | } |
f7039b1d LD |
3131 | |
3132 | if (fatal_signal_pending(current)) { | |
3133 | ret = -ERESTARTSYS; | |
3134 | break; | |
3135 | } | |
3136 | ||
3137 | cond_resched(); | |
3138 | } | |
3139 | ||
3140 | return ret; | |
3141 | } | |
581bb050 | 3142 | |
7fe1e641 LZ |
3143 | int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group, |
3144 | u64 *trimmed, u64 start, u64 end, u64 minlen) | |
3145 | { | |
3146 | int ret; | |
3147 | ||
3148 | *trimmed = 0; | |
3149 | ||
04216820 FM |
3150 | spin_lock(&block_group->lock); |
3151 | if (block_group->removed) { | |
3152 | spin_unlock(&block_group->lock); | |
3153 | return 0; | |
3154 | } | |
3155 | atomic_inc(&block_group->trimming); | |
3156 | spin_unlock(&block_group->lock); | |
3157 | ||
7fe1e641 LZ |
3158 | ret = trim_no_bitmap(block_group, trimmed, start, end, minlen); |
3159 | if (ret) | |
04216820 | 3160 | goto out; |
7fe1e641 LZ |
3161 | |
3162 | ret = trim_bitmaps(block_group, trimmed, start, end, minlen); | |
04216820 FM |
3163 | out: |
3164 | spin_lock(&block_group->lock); | |
3165 | if (atomic_dec_and_test(&block_group->trimming) && | |
3166 | block_group->removed) { | |
3167 | struct extent_map_tree *em_tree; | |
3168 | struct extent_map *em; | |
3169 | ||
3170 | spin_unlock(&block_group->lock); | |
3171 | ||
a1e7e16e | 3172 | lock_chunks(block_group->fs_info->chunk_root); |
04216820 FM |
3173 | em_tree = &block_group->fs_info->mapping_tree.map_tree; |
3174 | write_lock(&em_tree->lock); | |
3175 | em = lookup_extent_mapping(em_tree, block_group->key.objectid, | |
3176 | 1); | |
3177 | BUG_ON(!em); /* logic error, can't happen */ | |
a1e7e16e FM |
3178 | /* |
3179 | * remove_extent_mapping() will delete us from the pinned_chunks | |
3180 | * list, which is protected by the chunk mutex. | |
3181 | */ | |
04216820 FM |
3182 | remove_extent_mapping(em_tree, em); |
3183 | write_unlock(&em_tree->lock); | |
04216820 FM |
3184 | unlock_chunks(block_group->fs_info->chunk_root); |
3185 | ||
3186 | /* once for us and once for the tree */ | |
3187 | free_extent_map(em); | |
3188 | free_extent_map(em); | |
946ddbe8 FM |
3189 | |
3190 | /* | |
3191 | * We've left one free space entry and other tasks trimming | |
3192 | * this block group have left 1 entry each one. Free them. | |
3193 | */ | |
3194 | __btrfs_remove_free_space_cache(block_group->free_space_ctl); | |
04216820 FM |
3195 | } else { |
3196 | spin_unlock(&block_group->lock); | |
3197 | } | |
7fe1e641 LZ |
3198 | |
3199 | return ret; | |
3200 | } | |
3201 | ||
581bb050 LZ |
3202 | /* |
3203 | * Find the left-most item in the cache tree, and then return the | |
3204 | * smallest inode number in the item. | |
3205 | * | |
3206 | * Note: the returned inode number may not be the smallest one in | |
3207 | * the tree, if the left-most item is a bitmap. | |
3208 | */ | |
3209 | u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root) | |
3210 | { | |
3211 | struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl; | |
3212 | struct btrfs_free_space *entry = NULL; | |
3213 | u64 ino = 0; | |
3214 | ||
3215 | spin_lock(&ctl->tree_lock); | |
3216 | ||
3217 | if (RB_EMPTY_ROOT(&ctl->free_space_offset)) | |
3218 | goto out; | |
3219 | ||
3220 | entry = rb_entry(rb_first(&ctl->free_space_offset), | |
3221 | struct btrfs_free_space, offset_index); | |
3222 | ||
3223 | if (!entry->bitmap) { | |
3224 | ino = entry->offset; | |
3225 | ||
3226 | unlink_free_space(ctl, entry); | |
3227 | entry->offset++; | |
3228 | entry->bytes--; | |
3229 | if (!entry->bytes) | |
3230 | kmem_cache_free(btrfs_free_space_cachep, entry); | |
3231 | else | |
3232 | link_free_space(ctl, entry); | |
3233 | } else { | |
3234 | u64 offset = 0; | |
3235 | u64 count = 1; | |
3236 | int ret; | |
3237 | ||
3238 | ret = search_bitmap(ctl, entry, &offset, &count); | |
79787eaa | 3239 | /* Logic error; Should be empty if it can't find anything */ |
b12d6869 | 3240 | ASSERT(!ret); |
581bb050 LZ |
3241 | |
3242 | ino = offset; | |
3243 | bitmap_clear_bits(ctl, entry, offset, 1); | |
3244 | if (entry->bytes == 0) | |
3245 | free_bitmap(ctl, entry); | |
3246 | } | |
3247 | out: | |
3248 | spin_unlock(&ctl->tree_lock); | |
3249 | ||
3250 | return ino; | |
3251 | } | |
82d5902d LZ |
3252 | |
3253 | struct inode *lookup_free_ino_inode(struct btrfs_root *root, | |
3254 | struct btrfs_path *path) | |
3255 | { | |
3256 | struct inode *inode = NULL; | |
3257 | ||
57cdc8db DS |
3258 | spin_lock(&root->ino_cache_lock); |
3259 | if (root->ino_cache_inode) | |
3260 | inode = igrab(root->ino_cache_inode); | |
3261 | spin_unlock(&root->ino_cache_lock); | |
82d5902d LZ |
3262 | if (inode) |
3263 | return inode; | |
3264 | ||
3265 | inode = __lookup_free_space_inode(root, path, 0); | |
3266 | if (IS_ERR(inode)) | |
3267 | return inode; | |
3268 | ||
57cdc8db | 3269 | spin_lock(&root->ino_cache_lock); |
7841cb28 | 3270 | if (!btrfs_fs_closing(root->fs_info)) |
57cdc8db DS |
3271 | root->ino_cache_inode = igrab(inode); |
3272 | spin_unlock(&root->ino_cache_lock); | |
82d5902d LZ |
3273 | |
3274 | return inode; | |
3275 | } | |
3276 | ||
3277 | int create_free_ino_inode(struct btrfs_root *root, | |
3278 | struct btrfs_trans_handle *trans, | |
3279 | struct btrfs_path *path) | |
3280 | { | |
3281 | return __create_free_space_inode(root, trans, path, | |
3282 | BTRFS_FREE_INO_OBJECTID, 0); | |
3283 | } | |
3284 | ||
3285 | int load_free_ino_cache(struct btrfs_fs_info *fs_info, struct btrfs_root *root) | |
3286 | { | |
3287 | struct btrfs_free_space_ctl *ctl = root->free_ino_ctl; | |
3288 | struct btrfs_path *path; | |
3289 | struct inode *inode; | |
3290 | int ret = 0; | |
3291 | u64 root_gen = btrfs_root_generation(&root->root_item); | |
3292 | ||
4b9465cb CM |
3293 | if (!btrfs_test_opt(root, INODE_MAP_CACHE)) |
3294 | return 0; | |
3295 | ||
82d5902d LZ |
3296 | /* |
3297 | * If we're unmounting then just return, since this does a search on the | |
3298 | * normal root and not the commit root and we could deadlock. | |
3299 | */ | |
7841cb28 | 3300 | if (btrfs_fs_closing(fs_info)) |
82d5902d LZ |
3301 | return 0; |
3302 | ||
3303 | path = btrfs_alloc_path(); | |
3304 | if (!path) | |
3305 | return 0; | |
3306 | ||
3307 | inode = lookup_free_ino_inode(root, path); | |
3308 | if (IS_ERR(inode)) | |
3309 | goto out; | |
3310 | ||
3311 | if (root_gen != BTRFS_I(inode)->generation) | |
3312 | goto out_put; | |
3313 | ||
3314 | ret = __load_free_space_cache(root, inode, ctl, path, 0); | |
3315 | ||
3316 | if (ret < 0) | |
c2cf52eb SK |
3317 | btrfs_err(fs_info, |
3318 | "failed to load free ino cache for root %llu", | |
3319 | root->root_key.objectid); | |
82d5902d LZ |
3320 | out_put: |
3321 | iput(inode); | |
3322 | out: | |
3323 | btrfs_free_path(path); | |
3324 | return ret; | |
3325 | } | |
3326 | ||
3327 | int btrfs_write_out_ino_cache(struct btrfs_root *root, | |
3328 | struct btrfs_trans_handle *trans, | |
53645a91 FDBM |
3329 | struct btrfs_path *path, |
3330 | struct inode *inode) | |
82d5902d LZ |
3331 | { |
3332 | struct btrfs_free_space_ctl *ctl = root->free_ino_ctl; | |
82d5902d LZ |
3333 | int ret; |
3334 | ||
4b9465cb CM |
3335 | if (!btrfs_test_opt(root, INODE_MAP_CACHE)) |
3336 | return 0; | |
3337 | ||
82d5902d | 3338 | ret = __btrfs_write_out_cache(root, inode, ctl, NULL, trans, path, 0); |
c09544e0 JB |
3339 | if (ret) { |
3340 | btrfs_delalloc_release_metadata(inode, inode->i_size); | |
3341 | #ifdef DEBUG | |
c2cf52eb SK |
3342 | btrfs_err(root->fs_info, |
3343 | "failed to write free ino cache for root %llu", | |
3344 | root->root_key.objectid); | |
c09544e0 JB |
3345 | #endif |
3346 | } | |
82d5902d | 3347 | |
82d5902d LZ |
3348 | return ret; |
3349 | } | |
74255aa0 JB |
3350 | |
3351 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS | |
dc11dd5d JB |
3352 | /* |
3353 | * Use this if you need to make a bitmap or extent entry specifically, it | |
3354 | * doesn't do any of the merging that add_free_space does, this acts a lot like | |
3355 | * how the free space cache loading stuff works, so you can get really weird | |
3356 | * configurations. | |
3357 | */ | |
3358 | int test_add_free_space_entry(struct btrfs_block_group_cache *cache, | |
3359 | u64 offset, u64 bytes, bool bitmap) | |
74255aa0 | 3360 | { |
dc11dd5d JB |
3361 | struct btrfs_free_space_ctl *ctl = cache->free_space_ctl; |
3362 | struct btrfs_free_space *info = NULL, *bitmap_info; | |
3363 | void *map = NULL; | |
3364 | u64 bytes_added; | |
3365 | int ret; | |
74255aa0 | 3366 | |
dc11dd5d JB |
3367 | again: |
3368 | if (!info) { | |
3369 | info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS); | |
3370 | if (!info) | |
3371 | return -ENOMEM; | |
74255aa0 JB |
3372 | } |
3373 | ||
dc11dd5d JB |
3374 | if (!bitmap) { |
3375 | spin_lock(&ctl->tree_lock); | |
3376 | info->offset = offset; | |
3377 | info->bytes = bytes; | |
3378 | ret = link_free_space(ctl, info); | |
3379 | spin_unlock(&ctl->tree_lock); | |
3380 | if (ret) | |
3381 | kmem_cache_free(btrfs_free_space_cachep, info); | |
3382 | return ret; | |
3383 | } | |
3384 | ||
3385 | if (!map) { | |
3386 | map = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS); | |
3387 | if (!map) { | |
3388 | kmem_cache_free(btrfs_free_space_cachep, info); | |
3389 | return -ENOMEM; | |
3390 | } | |
3391 | } | |
3392 | ||
3393 | spin_lock(&ctl->tree_lock); | |
3394 | bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), | |
3395 | 1, 0); | |
3396 | if (!bitmap_info) { | |
3397 | info->bitmap = map; | |
3398 | map = NULL; | |
3399 | add_new_bitmap(ctl, info, offset); | |
3400 | bitmap_info = info; | |
20005523 | 3401 | info = NULL; |
dc11dd5d | 3402 | } |
74255aa0 | 3403 | |
dc11dd5d JB |
3404 | bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes); |
3405 | bytes -= bytes_added; | |
3406 | offset += bytes_added; | |
3407 | spin_unlock(&ctl->tree_lock); | |
74255aa0 | 3408 | |
dc11dd5d JB |
3409 | if (bytes) |
3410 | goto again; | |
74255aa0 | 3411 | |
20005523 FM |
3412 | if (info) |
3413 | kmem_cache_free(btrfs_free_space_cachep, info); | |
dc11dd5d JB |
3414 | if (map) |
3415 | kfree(map); | |
3416 | return 0; | |
74255aa0 JB |
3417 | } |
3418 | ||
3419 | /* | |
3420 | * Checks to see if the given range is in the free space cache. This is really | |
3421 | * just used to check the absence of space, so if there is free space in the | |
3422 | * range at all we will return 1. | |
3423 | */ | |
dc11dd5d JB |
3424 | int test_check_exists(struct btrfs_block_group_cache *cache, |
3425 | u64 offset, u64 bytes) | |
74255aa0 JB |
3426 | { |
3427 | struct btrfs_free_space_ctl *ctl = cache->free_space_ctl; | |
3428 | struct btrfs_free_space *info; | |
3429 | int ret = 0; | |
3430 | ||
3431 | spin_lock(&ctl->tree_lock); | |
3432 | info = tree_search_offset(ctl, offset, 0, 0); | |
3433 | if (!info) { | |
3434 | info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), | |
3435 | 1, 0); | |
3436 | if (!info) | |
3437 | goto out; | |
3438 | } | |
3439 | ||
3440 | have_info: | |
3441 | if (info->bitmap) { | |
3442 | u64 bit_off, bit_bytes; | |
3443 | struct rb_node *n; | |
3444 | struct btrfs_free_space *tmp; | |
3445 | ||
3446 | bit_off = offset; | |
3447 | bit_bytes = ctl->unit; | |
3448 | ret = search_bitmap(ctl, info, &bit_off, &bit_bytes); | |
3449 | if (!ret) { | |
3450 | if (bit_off == offset) { | |
3451 | ret = 1; | |
3452 | goto out; | |
3453 | } else if (bit_off > offset && | |
3454 | offset + bytes > bit_off) { | |
3455 | ret = 1; | |
3456 | goto out; | |
3457 | } | |
3458 | } | |
3459 | ||
3460 | n = rb_prev(&info->offset_index); | |
3461 | while (n) { | |
3462 | tmp = rb_entry(n, struct btrfs_free_space, | |
3463 | offset_index); | |
3464 | if (tmp->offset + tmp->bytes < offset) | |
3465 | break; | |
3466 | if (offset + bytes < tmp->offset) { | |
3467 | n = rb_prev(&info->offset_index); | |
3468 | continue; | |
3469 | } | |
3470 | info = tmp; | |
3471 | goto have_info; | |
3472 | } | |
3473 | ||
3474 | n = rb_next(&info->offset_index); | |
3475 | while (n) { | |
3476 | tmp = rb_entry(n, struct btrfs_free_space, | |
3477 | offset_index); | |
3478 | if (offset + bytes < tmp->offset) | |
3479 | break; | |
3480 | if (tmp->offset + tmp->bytes < offset) { | |
3481 | n = rb_next(&info->offset_index); | |
3482 | continue; | |
3483 | } | |
3484 | info = tmp; | |
3485 | goto have_info; | |
3486 | } | |
3487 | ||
20005523 | 3488 | ret = 0; |
74255aa0 JB |
3489 | goto out; |
3490 | } | |
3491 | ||
3492 | if (info->offset == offset) { | |
3493 | ret = 1; | |
3494 | goto out; | |
3495 | } | |
3496 | ||
3497 | if (offset > info->offset && offset < info->offset + info->bytes) | |
3498 | ret = 1; | |
3499 | out: | |
3500 | spin_unlock(&ctl->tree_lock); | |
3501 | return ret; | |
3502 | } | |
dc11dd5d | 3503 | #endif /* CONFIG_BTRFS_FS_RUN_SANITY_TESTS */ |