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 | |
0ef6447a | 32 | #define BITS_PER_BITMAP (PAGE_SIZE * 8UL) |
ee22184b | 33 | #define MAX_CACHE_BYTES_PER_GIG SZ_32K |
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 | 87 | mapping_set_gfp_mask(inode->i_mapping, |
c62d2555 MH |
88 | mapping_gfp_constraint(inode->i_mapping, |
89 | ~(__GFP_FS | __GFP_HIGHMEM))); | |
adae52b9 | 90 | |
0414efae LZ |
91 | return inode; |
92 | } | |
93 | ||
94 | struct inode *lookup_free_space_inode(struct btrfs_root *root, | |
95 | struct btrfs_block_group_cache | |
96 | *block_group, struct btrfs_path *path) | |
97 | { | |
98 | struct inode *inode = NULL; | |
5b0e95bf | 99 | u32 flags = BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW; |
0414efae LZ |
100 | |
101 | spin_lock(&block_group->lock); | |
102 | if (block_group->inode) | |
103 | inode = igrab(block_group->inode); | |
104 | spin_unlock(&block_group->lock); | |
105 | if (inode) | |
106 | return inode; | |
107 | ||
108 | inode = __lookup_free_space_inode(root, path, | |
109 | block_group->key.objectid); | |
110 | if (IS_ERR(inode)) | |
111 | return inode; | |
112 | ||
0af3d00b | 113 | spin_lock(&block_group->lock); |
5b0e95bf | 114 | if (!((BTRFS_I(inode)->flags & flags) == flags)) { |
c2cf52eb SK |
115 | btrfs_info(root->fs_info, |
116 | "Old style space inode found, converting."); | |
5b0e95bf JB |
117 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM | |
118 | BTRFS_INODE_NODATACOW; | |
2f356126 JB |
119 | block_group->disk_cache_state = BTRFS_DC_CLEAR; |
120 | } | |
121 | ||
300e4f8a | 122 | if (!block_group->iref) { |
0af3d00b JB |
123 | block_group->inode = igrab(inode); |
124 | block_group->iref = 1; | |
125 | } | |
126 | spin_unlock(&block_group->lock); | |
127 | ||
128 | return inode; | |
129 | } | |
130 | ||
48a3b636 ES |
131 | static int __create_free_space_inode(struct btrfs_root *root, |
132 | struct btrfs_trans_handle *trans, | |
133 | struct btrfs_path *path, | |
134 | u64 ino, u64 offset) | |
0af3d00b JB |
135 | { |
136 | struct btrfs_key key; | |
137 | struct btrfs_disk_key disk_key; | |
138 | struct btrfs_free_space_header *header; | |
139 | struct btrfs_inode_item *inode_item; | |
140 | struct extent_buffer *leaf; | |
5b0e95bf | 141 | u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC; |
0af3d00b JB |
142 | int ret; |
143 | ||
0414efae | 144 | ret = btrfs_insert_empty_inode(trans, root, path, ino); |
0af3d00b JB |
145 | if (ret) |
146 | return ret; | |
147 | ||
5b0e95bf JB |
148 | /* We inline crc's for the free disk space cache */ |
149 | if (ino != BTRFS_FREE_INO_OBJECTID) | |
150 | flags |= BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW; | |
151 | ||
0af3d00b JB |
152 | leaf = path->nodes[0]; |
153 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
154 | struct btrfs_inode_item); | |
155 | btrfs_item_key(leaf, &disk_key, path->slots[0]); | |
156 | memset_extent_buffer(leaf, 0, (unsigned long)inode_item, | |
157 | sizeof(*inode_item)); | |
158 | btrfs_set_inode_generation(leaf, inode_item, trans->transid); | |
159 | btrfs_set_inode_size(leaf, inode_item, 0); | |
160 | btrfs_set_inode_nbytes(leaf, inode_item, 0); | |
161 | btrfs_set_inode_uid(leaf, inode_item, 0); | |
162 | btrfs_set_inode_gid(leaf, inode_item, 0); | |
163 | btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600); | |
5b0e95bf | 164 | btrfs_set_inode_flags(leaf, inode_item, flags); |
0af3d00b JB |
165 | btrfs_set_inode_nlink(leaf, inode_item, 1); |
166 | btrfs_set_inode_transid(leaf, inode_item, trans->transid); | |
0414efae | 167 | btrfs_set_inode_block_group(leaf, inode_item, offset); |
0af3d00b | 168 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 169 | btrfs_release_path(path); |
0af3d00b JB |
170 | |
171 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
0414efae | 172 | key.offset = offset; |
0af3d00b | 173 | key.type = 0; |
0af3d00b JB |
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 | } | |
c9dc4c65 | 180 | |
0af3d00b JB |
181 | leaf = path->nodes[0]; |
182 | header = btrfs_item_ptr(leaf, path->slots[0], | |
183 | struct btrfs_free_space_header); | |
184 | memset_extent_buffer(leaf, 0, (unsigned long)header, sizeof(*header)); | |
185 | btrfs_set_free_space_key(leaf, header, &disk_key); | |
186 | btrfs_mark_buffer_dirty(leaf); | |
b3b4aa74 | 187 | btrfs_release_path(path); |
0af3d00b JB |
188 | |
189 | return 0; | |
190 | } | |
191 | ||
0414efae LZ |
192 | int create_free_space_inode(struct btrfs_root *root, |
193 | struct btrfs_trans_handle *trans, | |
194 | struct btrfs_block_group_cache *block_group, | |
195 | struct btrfs_path *path) | |
196 | { | |
197 | int ret; | |
198 | u64 ino; | |
199 | ||
200 | ret = btrfs_find_free_objectid(root, &ino); | |
201 | if (ret < 0) | |
202 | return ret; | |
203 | ||
204 | return __create_free_space_inode(root, trans, path, ino, | |
205 | block_group->key.objectid); | |
206 | } | |
207 | ||
7b61cd92 MX |
208 | int btrfs_check_trunc_cache_free_space(struct btrfs_root *root, |
209 | struct btrfs_block_rsv *rsv) | |
0af3d00b | 210 | { |
c8174313 | 211 | u64 needed_bytes; |
7b61cd92 | 212 | int ret; |
c8174313 JB |
213 | |
214 | /* 1 for slack space, 1 for updating the inode */ | |
215 | needed_bytes = btrfs_calc_trunc_metadata_size(root, 1) + | |
216 | btrfs_calc_trans_metadata_size(root, 1); | |
217 | ||
7b61cd92 MX |
218 | spin_lock(&rsv->lock); |
219 | if (rsv->reserved < needed_bytes) | |
220 | ret = -ENOSPC; | |
221 | else | |
222 | ret = 0; | |
223 | spin_unlock(&rsv->lock); | |
4b286cd1 | 224 | return ret; |
7b61cd92 MX |
225 | } |
226 | ||
227 | int btrfs_truncate_free_space_cache(struct btrfs_root *root, | |
228 | struct btrfs_trans_handle *trans, | |
1bbc621e | 229 | struct btrfs_block_group_cache *block_group, |
7b61cd92 MX |
230 | struct inode *inode) |
231 | { | |
7b61cd92 | 232 | int ret = 0; |
1bbc621e | 233 | struct btrfs_path *path = btrfs_alloc_path(); |
35c76642 | 234 | bool locked = false; |
1bbc621e CM |
235 | |
236 | if (!path) { | |
237 | ret = -ENOMEM; | |
238 | goto fail; | |
239 | } | |
240 | ||
241 | if (block_group) { | |
35c76642 | 242 | locked = true; |
1bbc621e CM |
243 | mutex_lock(&trans->transaction->cache_write_mutex); |
244 | if (!list_empty(&block_group->io_list)) { | |
245 | list_del_init(&block_group->io_list); | |
246 | ||
247 | btrfs_wait_cache_io(root, trans, block_group, | |
248 | &block_group->io_ctl, path, | |
249 | block_group->key.objectid); | |
250 | btrfs_put_block_group(block_group); | |
251 | } | |
252 | ||
253 | /* | |
254 | * now that we've truncated the cache away, its no longer | |
255 | * setup or written | |
256 | */ | |
257 | spin_lock(&block_group->lock); | |
258 | block_group->disk_cache_state = BTRFS_DC_CLEAR; | |
259 | spin_unlock(&block_group->lock); | |
260 | } | |
261 | btrfs_free_path(path); | |
0af3d00b | 262 | |
0af3d00b | 263 | btrfs_i_size_write(inode, 0); |
7caef267 | 264 | truncate_pagecache(inode, 0); |
0af3d00b JB |
265 | |
266 | /* | |
267 | * We don't need an orphan item because truncating the free space cache | |
268 | * will never be split across transactions. | |
28ed1345 CM |
269 | * We don't need to check for -EAGAIN because we're a free space |
270 | * cache inode | |
0af3d00b JB |
271 | */ |
272 | ret = btrfs_truncate_inode_items(trans, root, inode, | |
273 | 0, BTRFS_EXTENT_DATA_KEY); | |
35c76642 FM |
274 | if (ret) |
275 | goto fail; | |
0af3d00b | 276 | |
82d5902d | 277 | ret = btrfs_update_inode(trans, root, inode); |
1bbc621e | 278 | |
1bbc621e | 279 | fail: |
35c76642 FM |
280 | if (locked) |
281 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
79787eaa | 282 | if (ret) |
66642832 | 283 | btrfs_abort_transaction(trans, ret); |
c8174313 | 284 | |
82d5902d | 285 | return ret; |
0af3d00b JB |
286 | } |
287 | ||
9d66e233 JB |
288 | static int readahead_cache(struct inode *inode) |
289 | { | |
290 | struct file_ra_state *ra; | |
291 | unsigned long last_index; | |
292 | ||
293 | ra = kzalloc(sizeof(*ra), GFP_NOFS); | |
294 | if (!ra) | |
295 | return -ENOMEM; | |
296 | ||
297 | file_ra_state_init(ra, inode->i_mapping); | |
09cbfeaf | 298 | last_index = (i_size_read(inode) - 1) >> PAGE_SHIFT; |
9d66e233 JB |
299 | |
300 | page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index); | |
301 | ||
302 | kfree(ra); | |
303 | ||
304 | return 0; | |
305 | } | |
306 | ||
4c6d1d85 | 307 | static int io_ctl_init(struct btrfs_io_ctl *io_ctl, struct inode *inode, |
5349d6c3 | 308 | struct btrfs_root *root, int write) |
a67509c3 | 309 | { |
5349d6c3 MX |
310 | int num_pages; |
311 | int check_crcs = 0; | |
312 | ||
09cbfeaf | 313 | num_pages = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
5349d6c3 MX |
314 | |
315 | if (btrfs_ino(inode) != BTRFS_FREE_INO_OBJECTID) | |
316 | check_crcs = 1; | |
317 | ||
318 | /* Make sure we can fit our crcs into the first page */ | |
319 | if (write && check_crcs && | |
09cbfeaf | 320 | (num_pages * sizeof(u32)) >= PAGE_SIZE) |
5349d6c3 MX |
321 | return -ENOSPC; |
322 | ||
4c6d1d85 | 323 | memset(io_ctl, 0, sizeof(struct btrfs_io_ctl)); |
5349d6c3 | 324 | |
31e818fe | 325 | io_ctl->pages = kcalloc(num_pages, sizeof(struct page *), GFP_NOFS); |
a67509c3 JB |
326 | if (!io_ctl->pages) |
327 | return -ENOMEM; | |
5349d6c3 MX |
328 | |
329 | io_ctl->num_pages = num_pages; | |
a67509c3 | 330 | io_ctl->root = root; |
5349d6c3 | 331 | io_ctl->check_crcs = check_crcs; |
c9dc4c65 | 332 | io_ctl->inode = inode; |
5349d6c3 | 333 | |
a67509c3 JB |
334 | return 0; |
335 | } | |
336 | ||
4c6d1d85 | 337 | static void io_ctl_free(struct btrfs_io_ctl *io_ctl) |
a67509c3 JB |
338 | { |
339 | kfree(io_ctl->pages); | |
c9dc4c65 | 340 | io_ctl->pages = NULL; |
a67509c3 JB |
341 | } |
342 | ||
4c6d1d85 | 343 | static void io_ctl_unmap_page(struct btrfs_io_ctl *io_ctl) |
a67509c3 JB |
344 | { |
345 | if (io_ctl->cur) { | |
a67509c3 JB |
346 | io_ctl->cur = NULL; |
347 | io_ctl->orig = NULL; | |
348 | } | |
349 | } | |
350 | ||
4c6d1d85 | 351 | static void io_ctl_map_page(struct btrfs_io_ctl *io_ctl, int clear) |
a67509c3 | 352 | { |
b12d6869 | 353 | ASSERT(io_ctl->index < io_ctl->num_pages); |
a67509c3 | 354 | io_ctl->page = io_ctl->pages[io_ctl->index++]; |
2b108268 | 355 | io_ctl->cur = page_address(io_ctl->page); |
a67509c3 | 356 | io_ctl->orig = io_ctl->cur; |
09cbfeaf | 357 | io_ctl->size = PAGE_SIZE; |
a67509c3 | 358 | if (clear) |
09cbfeaf | 359 | memset(io_ctl->cur, 0, PAGE_SIZE); |
a67509c3 JB |
360 | } |
361 | ||
4c6d1d85 | 362 | static void io_ctl_drop_pages(struct btrfs_io_ctl *io_ctl) |
a67509c3 JB |
363 | { |
364 | int i; | |
365 | ||
366 | io_ctl_unmap_page(io_ctl); | |
367 | ||
368 | for (i = 0; i < io_ctl->num_pages; i++) { | |
a1ee5a45 LZ |
369 | if (io_ctl->pages[i]) { |
370 | ClearPageChecked(io_ctl->pages[i]); | |
371 | unlock_page(io_ctl->pages[i]); | |
09cbfeaf | 372 | put_page(io_ctl->pages[i]); |
a1ee5a45 | 373 | } |
a67509c3 JB |
374 | } |
375 | } | |
376 | ||
4c6d1d85 | 377 | static int io_ctl_prepare_pages(struct btrfs_io_ctl *io_ctl, struct inode *inode, |
a67509c3 JB |
378 | int uptodate) |
379 | { | |
380 | struct page *page; | |
381 | gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); | |
382 | int i; | |
383 | ||
384 | for (i = 0; i < io_ctl->num_pages; i++) { | |
385 | page = find_or_create_page(inode->i_mapping, i, mask); | |
386 | if (!page) { | |
387 | io_ctl_drop_pages(io_ctl); | |
388 | return -ENOMEM; | |
389 | } | |
390 | io_ctl->pages[i] = page; | |
391 | if (uptodate && !PageUptodate(page)) { | |
392 | btrfs_readpage(NULL, page); | |
393 | lock_page(page); | |
394 | if (!PageUptodate(page)) { | |
efe120a0 FH |
395 | btrfs_err(BTRFS_I(inode)->root->fs_info, |
396 | "error reading free space cache"); | |
a67509c3 JB |
397 | io_ctl_drop_pages(io_ctl); |
398 | return -EIO; | |
399 | } | |
400 | } | |
401 | } | |
402 | ||
f7d61dcd JB |
403 | for (i = 0; i < io_ctl->num_pages; i++) { |
404 | clear_page_dirty_for_io(io_ctl->pages[i]); | |
405 | set_page_extent_mapped(io_ctl->pages[i]); | |
406 | } | |
407 | ||
a67509c3 JB |
408 | return 0; |
409 | } | |
410 | ||
4c6d1d85 | 411 | static void io_ctl_set_generation(struct btrfs_io_ctl *io_ctl, u64 generation) |
a67509c3 | 412 | { |
528c0327 | 413 | __le64 *val; |
a67509c3 JB |
414 | |
415 | io_ctl_map_page(io_ctl, 1); | |
416 | ||
417 | /* | |
5b0e95bf JB |
418 | * Skip the csum areas. If we don't check crcs then we just have a |
419 | * 64bit chunk at the front of the first page. | |
a67509c3 | 420 | */ |
5b0e95bf JB |
421 | if (io_ctl->check_crcs) { |
422 | io_ctl->cur += (sizeof(u32) * io_ctl->num_pages); | |
423 | io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages); | |
424 | } else { | |
425 | io_ctl->cur += sizeof(u64); | |
426 | io_ctl->size -= sizeof(u64) * 2; | |
427 | } | |
a67509c3 JB |
428 | |
429 | val = io_ctl->cur; | |
430 | *val = cpu_to_le64(generation); | |
431 | io_ctl->cur += sizeof(u64); | |
a67509c3 JB |
432 | } |
433 | ||
4c6d1d85 | 434 | static int io_ctl_check_generation(struct btrfs_io_ctl *io_ctl, u64 generation) |
a67509c3 | 435 | { |
528c0327 | 436 | __le64 *gen; |
a67509c3 | 437 | |
5b0e95bf JB |
438 | /* |
439 | * Skip the crc area. If we don't check crcs then we just have a 64bit | |
440 | * chunk at the front of the first page. | |
441 | */ | |
442 | if (io_ctl->check_crcs) { | |
443 | io_ctl->cur += sizeof(u32) * io_ctl->num_pages; | |
444 | io_ctl->size -= sizeof(u64) + | |
445 | (sizeof(u32) * io_ctl->num_pages); | |
446 | } else { | |
447 | io_ctl->cur += sizeof(u64); | |
448 | io_ctl->size -= sizeof(u64) * 2; | |
449 | } | |
a67509c3 | 450 | |
a67509c3 JB |
451 | gen = io_ctl->cur; |
452 | if (le64_to_cpu(*gen) != generation) { | |
94647322 DS |
453 | btrfs_err_rl(io_ctl->root->fs_info, |
454 | "space cache generation (%llu) does not match inode (%llu)", | |
455 | *gen, generation); | |
a67509c3 JB |
456 | io_ctl_unmap_page(io_ctl); |
457 | return -EIO; | |
458 | } | |
459 | io_ctl->cur += sizeof(u64); | |
5b0e95bf JB |
460 | return 0; |
461 | } | |
462 | ||
4c6d1d85 | 463 | static void io_ctl_set_crc(struct btrfs_io_ctl *io_ctl, int index) |
5b0e95bf JB |
464 | { |
465 | u32 *tmp; | |
466 | u32 crc = ~(u32)0; | |
467 | unsigned offset = 0; | |
468 | ||
469 | if (!io_ctl->check_crcs) { | |
470 | io_ctl_unmap_page(io_ctl); | |
471 | return; | |
472 | } | |
473 | ||
474 | if (index == 0) | |
cb54f257 | 475 | offset = sizeof(u32) * io_ctl->num_pages; |
5b0e95bf | 476 | |
b0496686 | 477 | crc = btrfs_csum_data(io_ctl->orig + offset, crc, |
09cbfeaf | 478 | PAGE_SIZE - offset); |
5b0e95bf JB |
479 | btrfs_csum_final(crc, (char *)&crc); |
480 | io_ctl_unmap_page(io_ctl); | |
2b108268 | 481 | tmp = page_address(io_ctl->pages[0]); |
5b0e95bf JB |
482 | tmp += index; |
483 | *tmp = crc; | |
5b0e95bf JB |
484 | } |
485 | ||
4c6d1d85 | 486 | static int io_ctl_check_crc(struct btrfs_io_ctl *io_ctl, int index) |
5b0e95bf JB |
487 | { |
488 | u32 *tmp, val; | |
489 | u32 crc = ~(u32)0; | |
490 | unsigned offset = 0; | |
491 | ||
492 | if (!io_ctl->check_crcs) { | |
493 | io_ctl_map_page(io_ctl, 0); | |
494 | return 0; | |
495 | } | |
496 | ||
497 | if (index == 0) | |
498 | offset = sizeof(u32) * io_ctl->num_pages; | |
499 | ||
2b108268 | 500 | tmp = page_address(io_ctl->pages[0]); |
5b0e95bf JB |
501 | tmp += index; |
502 | val = *tmp; | |
5b0e95bf JB |
503 | |
504 | io_ctl_map_page(io_ctl, 0); | |
b0496686 | 505 | crc = btrfs_csum_data(io_ctl->orig + offset, crc, |
09cbfeaf | 506 | PAGE_SIZE - offset); |
5b0e95bf JB |
507 | btrfs_csum_final(crc, (char *)&crc); |
508 | if (val != crc) { | |
94647322 DS |
509 | btrfs_err_rl(io_ctl->root->fs_info, |
510 | "csum mismatch on free space cache"); | |
5b0e95bf JB |
511 | io_ctl_unmap_page(io_ctl); |
512 | return -EIO; | |
513 | } | |
514 | ||
a67509c3 JB |
515 | return 0; |
516 | } | |
517 | ||
4c6d1d85 | 518 | static int io_ctl_add_entry(struct btrfs_io_ctl *io_ctl, u64 offset, u64 bytes, |
a67509c3 JB |
519 | void *bitmap) |
520 | { | |
521 | struct btrfs_free_space_entry *entry; | |
522 | ||
523 | if (!io_ctl->cur) | |
524 | return -ENOSPC; | |
525 | ||
526 | entry = io_ctl->cur; | |
527 | entry->offset = cpu_to_le64(offset); | |
528 | entry->bytes = cpu_to_le64(bytes); | |
529 | entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP : | |
530 | BTRFS_FREE_SPACE_EXTENT; | |
531 | io_ctl->cur += sizeof(struct btrfs_free_space_entry); | |
532 | io_ctl->size -= sizeof(struct btrfs_free_space_entry); | |
533 | ||
534 | if (io_ctl->size >= sizeof(struct btrfs_free_space_entry)) | |
535 | return 0; | |
536 | ||
5b0e95bf | 537 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); |
a67509c3 JB |
538 | |
539 | /* No more pages to map */ | |
540 | if (io_ctl->index >= io_ctl->num_pages) | |
541 | return 0; | |
542 | ||
543 | /* map the next page */ | |
544 | io_ctl_map_page(io_ctl, 1); | |
545 | return 0; | |
546 | } | |
547 | ||
4c6d1d85 | 548 | static int io_ctl_add_bitmap(struct btrfs_io_ctl *io_ctl, void *bitmap) |
a67509c3 JB |
549 | { |
550 | if (!io_ctl->cur) | |
551 | return -ENOSPC; | |
552 | ||
553 | /* | |
554 | * If we aren't at the start of the current page, unmap this one and | |
555 | * map the next one if there is any left. | |
556 | */ | |
557 | if (io_ctl->cur != io_ctl->orig) { | |
5b0e95bf | 558 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); |
a67509c3 JB |
559 | if (io_ctl->index >= io_ctl->num_pages) |
560 | return -ENOSPC; | |
561 | io_ctl_map_page(io_ctl, 0); | |
562 | } | |
563 | ||
09cbfeaf | 564 | memcpy(io_ctl->cur, bitmap, PAGE_SIZE); |
5b0e95bf | 565 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); |
a67509c3 JB |
566 | if (io_ctl->index < io_ctl->num_pages) |
567 | io_ctl_map_page(io_ctl, 0); | |
568 | return 0; | |
569 | } | |
570 | ||
4c6d1d85 | 571 | static void io_ctl_zero_remaining_pages(struct btrfs_io_ctl *io_ctl) |
a67509c3 | 572 | { |
5b0e95bf JB |
573 | /* |
574 | * If we're not on the boundary we know we've modified the page and we | |
575 | * need to crc the page. | |
576 | */ | |
577 | if (io_ctl->cur != io_ctl->orig) | |
578 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); | |
579 | else | |
580 | io_ctl_unmap_page(io_ctl); | |
a67509c3 JB |
581 | |
582 | while (io_ctl->index < io_ctl->num_pages) { | |
583 | io_ctl_map_page(io_ctl, 1); | |
5b0e95bf | 584 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); |
a67509c3 JB |
585 | } |
586 | } | |
587 | ||
4c6d1d85 | 588 | static int io_ctl_read_entry(struct btrfs_io_ctl *io_ctl, |
5b0e95bf | 589 | struct btrfs_free_space *entry, u8 *type) |
a67509c3 JB |
590 | { |
591 | struct btrfs_free_space_entry *e; | |
2f120c05 JB |
592 | int ret; |
593 | ||
594 | if (!io_ctl->cur) { | |
595 | ret = io_ctl_check_crc(io_ctl, io_ctl->index); | |
596 | if (ret) | |
597 | return ret; | |
598 | } | |
a67509c3 JB |
599 | |
600 | e = io_ctl->cur; | |
601 | entry->offset = le64_to_cpu(e->offset); | |
602 | entry->bytes = le64_to_cpu(e->bytes); | |
5b0e95bf | 603 | *type = e->type; |
a67509c3 JB |
604 | io_ctl->cur += sizeof(struct btrfs_free_space_entry); |
605 | io_ctl->size -= sizeof(struct btrfs_free_space_entry); | |
606 | ||
607 | if (io_ctl->size >= sizeof(struct btrfs_free_space_entry)) | |
5b0e95bf | 608 | return 0; |
a67509c3 JB |
609 | |
610 | io_ctl_unmap_page(io_ctl); | |
611 | ||
2f120c05 | 612 | return 0; |
a67509c3 JB |
613 | } |
614 | ||
4c6d1d85 | 615 | static int io_ctl_read_bitmap(struct btrfs_io_ctl *io_ctl, |
5b0e95bf | 616 | struct btrfs_free_space *entry) |
a67509c3 | 617 | { |
5b0e95bf JB |
618 | int ret; |
619 | ||
5b0e95bf JB |
620 | ret = io_ctl_check_crc(io_ctl, io_ctl->index); |
621 | if (ret) | |
622 | return ret; | |
623 | ||
09cbfeaf | 624 | memcpy(entry->bitmap, io_ctl->cur, PAGE_SIZE); |
a67509c3 | 625 | io_ctl_unmap_page(io_ctl); |
5b0e95bf JB |
626 | |
627 | return 0; | |
a67509c3 JB |
628 | } |
629 | ||
cd023e7b JB |
630 | /* |
631 | * Since we attach pinned extents after the fact we can have contiguous sections | |
632 | * of free space that are split up in entries. This poses a problem with the | |
633 | * tree logging stuff since it could have allocated across what appears to be 2 | |
634 | * entries since we would have merged the entries when adding the pinned extents | |
635 | * back to the free space cache. So run through the space cache that we just | |
636 | * loaded and merge contiguous entries. This will make the log replay stuff not | |
637 | * blow up and it will make for nicer allocator behavior. | |
638 | */ | |
639 | static void merge_space_tree(struct btrfs_free_space_ctl *ctl) | |
640 | { | |
641 | struct btrfs_free_space *e, *prev = NULL; | |
642 | struct rb_node *n; | |
643 | ||
644 | again: | |
645 | spin_lock(&ctl->tree_lock); | |
646 | for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) { | |
647 | e = rb_entry(n, struct btrfs_free_space, offset_index); | |
648 | if (!prev) | |
649 | goto next; | |
650 | if (e->bitmap || prev->bitmap) | |
651 | goto next; | |
652 | if (prev->offset + prev->bytes == e->offset) { | |
653 | unlink_free_space(ctl, prev); | |
654 | unlink_free_space(ctl, e); | |
655 | prev->bytes += e->bytes; | |
656 | kmem_cache_free(btrfs_free_space_cachep, e); | |
657 | link_free_space(ctl, prev); | |
658 | prev = NULL; | |
659 | spin_unlock(&ctl->tree_lock); | |
660 | goto again; | |
661 | } | |
662 | next: | |
663 | prev = e; | |
664 | } | |
665 | spin_unlock(&ctl->tree_lock); | |
666 | } | |
667 | ||
48a3b636 ES |
668 | static int __load_free_space_cache(struct btrfs_root *root, struct inode *inode, |
669 | struct btrfs_free_space_ctl *ctl, | |
670 | struct btrfs_path *path, u64 offset) | |
9d66e233 | 671 | { |
9d66e233 JB |
672 | struct btrfs_free_space_header *header; |
673 | struct extent_buffer *leaf; | |
4c6d1d85 | 674 | struct btrfs_io_ctl io_ctl; |
9d66e233 | 675 | struct btrfs_key key; |
a67509c3 | 676 | struct btrfs_free_space *e, *n; |
b76808fc | 677 | LIST_HEAD(bitmaps); |
9d66e233 JB |
678 | u64 num_entries; |
679 | u64 num_bitmaps; | |
680 | u64 generation; | |
a67509c3 | 681 | u8 type; |
f6a39829 | 682 | int ret = 0; |
9d66e233 | 683 | |
9d66e233 | 684 | /* Nothing in the space cache, goodbye */ |
0414efae | 685 | if (!i_size_read(inode)) |
a67509c3 | 686 | return 0; |
9d66e233 JB |
687 | |
688 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
0414efae | 689 | key.offset = offset; |
9d66e233 JB |
690 | key.type = 0; |
691 | ||
692 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
0414efae | 693 | if (ret < 0) |
a67509c3 | 694 | return 0; |
0414efae | 695 | else if (ret > 0) { |
945d8962 | 696 | btrfs_release_path(path); |
a67509c3 | 697 | return 0; |
9d66e233 JB |
698 | } |
699 | ||
0414efae LZ |
700 | ret = -1; |
701 | ||
9d66e233 JB |
702 | leaf = path->nodes[0]; |
703 | header = btrfs_item_ptr(leaf, path->slots[0], | |
704 | struct btrfs_free_space_header); | |
705 | num_entries = btrfs_free_space_entries(leaf, header); | |
706 | num_bitmaps = btrfs_free_space_bitmaps(leaf, header); | |
707 | generation = btrfs_free_space_generation(leaf, header); | |
945d8962 | 708 | btrfs_release_path(path); |
9d66e233 | 709 | |
e570fd27 MX |
710 | if (!BTRFS_I(inode)->generation) { |
711 | btrfs_info(root->fs_info, | |
712 | "The free space cache file (%llu) is invalid. skip it\n", | |
713 | offset); | |
714 | return 0; | |
715 | } | |
716 | ||
9d66e233 | 717 | if (BTRFS_I(inode)->generation != generation) { |
c2cf52eb | 718 | btrfs_err(root->fs_info, |
5d163e0e | 719 | "free space inode generation (%llu) did not match free space cache generation (%llu)", |
c1c9ff7c | 720 | BTRFS_I(inode)->generation, generation); |
a67509c3 | 721 | return 0; |
9d66e233 JB |
722 | } |
723 | ||
724 | if (!num_entries) | |
a67509c3 | 725 | return 0; |
9d66e233 | 726 | |
5349d6c3 | 727 | ret = io_ctl_init(&io_ctl, inode, root, 0); |
706efc66 LZ |
728 | if (ret) |
729 | return ret; | |
730 | ||
9d66e233 | 731 | ret = readahead_cache(inode); |
0414efae | 732 | if (ret) |
9d66e233 | 733 | goto out; |
9d66e233 | 734 | |
a67509c3 JB |
735 | ret = io_ctl_prepare_pages(&io_ctl, inode, 1); |
736 | if (ret) | |
737 | goto out; | |
9d66e233 | 738 | |
5b0e95bf JB |
739 | ret = io_ctl_check_crc(&io_ctl, 0); |
740 | if (ret) | |
741 | goto free_cache; | |
742 | ||
a67509c3 JB |
743 | ret = io_ctl_check_generation(&io_ctl, generation); |
744 | if (ret) | |
745 | goto free_cache; | |
9d66e233 | 746 | |
a67509c3 JB |
747 | while (num_entries) { |
748 | e = kmem_cache_zalloc(btrfs_free_space_cachep, | |
749 | GFP_NOFS); | |
750 | if (!e) | |
9d66e233 | 751 | goto free_cache; |
9d66e233 | 752 | |
5b0e95bf JB |
753 | ret = io_ctl_read_entry(&io_ctl, e, &type); |
754 | if (ret) { | |
755 | kmem_cache_free(btrfs_free_space_cachep, e); | |
756 | goto free_cache; | |
757 | } | |
758 | ||
a67509c3 JB |
759 | if (!e->bytes) { |
760 | kmem_cache_free(btrfs_free_space_cachep, e); | |
761 | goto free_cache; | |
9d66e233 | 762 | } |
a67509c3 JB |
763 | |
764 | if (type == BTRFS_FREE_SPACE_EXTENT) { | |
765 | spin_lock(&ctl->tree_lock); | |
766 | ret = link_free_space(ctl, e); | |
767 | spin_unlock(&ctl->tree_lock); | |
768 | if (ret) { | |
c2cf52eb SK |
769 | btrfs_err(root->fs_info, |
770 | "Duplicate entries in free space cache, dumping"); | |
a67509c3 | 771 | kmem_cache_free(btrfs_free_space_cachep, e); |
9d66e233 JB |
772 | goto free_cache; |
773 | } | |
a67509c3 | 774 | } else { |
b12d6869 | 775 | ASSERT(num_bitmaps); |
a67509c3 | 776 | num_bitmaps--; |
09cbfeaf | 777 | e->bitmap = kzalloc(PAGE_SIZE, GFP_NOFS); |
a67509c3 JB |
778 | if (!e->bitmap) { |
779 | kmem_cache_free( | |
780 | btrfs_free_space_cachep, e); | |
9d66e233 JB |
781 | goto free_cache; |
782 | } | |
a67509c3 JB |
783 | spin_lock(&ctl->tree_lock); |
784 | ret = link_free_space(ctl, e); | |
785 | ctl->total_bitmaps++; | |
786 | ctl->op->recalc_thresholds(ctl); | |
787 | spin_unlock(&ctl->tree_lock); | |
788 | if (ret) { | |
c2cf52eb SK |
789 | btrfs_err(root->fs_info, |
790 | "Duplicate entries in free space cache, dumping"); | |
dc89e982 | 791 | kmem_cache_free(btrfs_free_space_cachep, e); |
9d66e233 JB |
792 | goto free_cache; |
793 | } | |
a67509c3 | 794 | list_add_tail(&e->list, &bitmaps); |
9d66e233 JB |
795 | } |
796 | ||
a67509c3 JB |
797 | num_entries--; |
798 | } | |
9d66e233 | 799 | |
2f120c05 JB |
800 | io_ctl_unmap_page(&io_ctl); |
801 | ||
a67509c3 JB |
802 | /* |
803 | * We add the bitmaps at the end of the entries in order that | |
804 | * the bitmap entries are added to the cache. | |
805 | */ | |
806 | list_for_each_entry_safe(e, n, &bitmaps, list) { | |
9d66e233 | 807 | list_del_init(&e->list); |
5b0e95bf JB |
808 | ret = io_ctl_read_bitmap(&io_ctl, e); |
809 | if (ret) | |
810 | goto free_cache; | |
9d66e233 JB |
811 | } |
812 | ||
a67509c3 | 813 | io_ctl_drop_pages(&io_ctl); |
cd023e7b | 814 | merge_space_tree(ctl); |
9d66e233 JB |
815 | ret = 1; |
816 | out: | |
a67509c3 | 817 | io_ctl_free(&io_ctl); |
9d66e233 | 818 | return ret; |
9d66e233 | 819 | free_cache: |
a67509c3 | 820 | io_ctl_drop_pages(&io_ctl); |
0414efae | 821 | __btrfs_remove_free_space_cache(ctl); |
9d66e233 JB |
822 | goto out; |
823 | } | |
824 | ||
0414efae LZ |
825 | int load_free_space_cache(struct btrfs_fs_info *fs_info, |
826 | struct btrfs_block_group_cache *block_group) | |
0cb59c99 | 827 | { |
34d52cb6 | 828 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0414efae LZ |
829 | struct btrfs_root *root = fs_info->tree_root; |
830 | struct inode *inode; | |
831 | struct btrfs_path *path; | |
5b0e95bf | 832 | int ret = 0; |
0414efae LZ |
833 | bool matched; |
834 | u64 used = btrfs_block_group_used(&block_group->item); | |
835 | ||
0414efae LZ |
836 | /* |
837 | * If this block group has been marked to be cleared for one reason or | |
838 | * another then we can't trust the on disk cache, so just return. | |
839 | */ | |
9d66e233 | 840 | spin_lock(&block_group->lock); |
0414efae LZ |
841 | if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) { |
842 | spin_unlock(&block_group->lock); | |
843 | return 0; | |
844 | } | |
9d66e233 | 845 | spin_unlock(&block_group->lock); |
0414efae LZ |
846 | |
847 | path = btrfs_alloc_path(); | |
848 | if (!path) | |
849 | return 0; | |
d53ba474 JB |
850 | path->search_commit_root = 1; |
851 | path->skip_locking = 1; | |
0414efae LZ |
852 | |
853 | inode = lookup_free_space_inode(root, block_group, path); | |
854 | if (IS_ERR(inode)) { | |
855 | btrfs_free_path(path); | |
856 | return 0; | |
857 | } | |
858 | ||
5b0e95bf JB |
859 | /* We may have converted the inode and made the cache invalid. */ |
860 | spin_lock(&block_group->lock); | |
861 | if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) { | |
862 | spin_unlock(&block_group->lock); | |
a7e221e9 | 863 | btrfs_free_path(path); |
5b0e95bf JB |
864 | goto out; |
865 | } | |
866 | spin_unlock(&block_group->lock); | |
867 | ||
0414efae LZ |
868 | ret = __load_free_space_cache(fs_info->tree_root, inode, ctl, |
869 | path, block_group->key.objectid); | |
870 | btrfs_free_path(path); | |
871 | if (ret <= 0) | |
872 | goto out; | |
873 | ||
874 | spin_lock(&ctl->tree_lock); | |
875 | matched = (ctl->free_space == (block_group->key.offset - used - | |
876 | block_group->bytes_super)); | |
877 | spin_unlock(&ctl->tree_lock); | |
878 | ||
879 | if (!matched) { | |
880 | __btrfs_remove_free_space_cache(ctl); | |
5d163e0e JM |
881 | btrfs_warn(fs_info, |
882 | "block group %llu has wrong amount of free space", | |
883 | block_group->key.objectid); | |
0414efae LZ |
884 | ret = -1; |
885 | } | |
886 | out: | |
887 | if (ret < 0) { | |
888 | /* This cache is bogus, make sure it gets cleared */ | |
889 | spin_lock(&block_group->lock); | |
890 | block_group->disk_cache_state = BTRFS_DC_CLEAR; | |
891 | spin_unlock(&block_group->lock); | |
82d5902d | 892 | ret = 0; |
0414efae | 893 | |
5d163e0e JM |
894 | btrfs_warn(fs_info, |
895 | "failed to load free space cache for block group %llu, rebuilding it now", | |
896 | block_group->key.objectid); | |
0414efae LZ |
897 | } |
898 | ||
899 | iput(inode); | |
900 | return ret; | |
9d66e233 JB |
901 | } |
902 | ||
d4452bc5 | 903 | static noinline_for_stack |
4c6d1d85 | 904 | int write_cache_extent_entries(struct btrfs_io_ctl *io_ctl, |
d4452bc5 CM |
905 | struct btrfs_free_space_ctl *ctl, |
906 | struct btrfs_block_group_cache *block_group, | |
907 | int *entries, int *bitmaps, | |
908 | struct list_head *bitmap_list) | |
0cb59c99 | 909 | { |
c09544e0 | 910 | int ret; |
d4452bc5 | 911 | struct btrfs_free_cluster *cluster = NULL; |
1bbc621e | 912 | struct btrfs_free_cluster *cluster_locked = NULL; |
d4452bc5 | 913 | struct rb_node *node = rb_first(&ctl->free_space_offset); |
55507ce3 | 914 | struct btrfs_trim_range *trim_entry; |
be1a12a0 | 915 | |
43be2146 | 916 | /* Get the cluster for this block_group if it exists */ |
d4452bc5 | 917 | if (block_group && !list_empty(&block_group->cluster_list)) { |
43be2146 JB |
918 | cluster = list_entry(block_group->cluster_list.next, |
919 | struct btrfs_free_cluster, | |
920 | block_group_list); | |
d4452bc5 | 921 | } |
43be2146 | 922 | |
f75b130e | 923 | if (!node && cluster) { |
1bbc621e CM |
924 | cluster_locked = cluster; |
925 | spin_lock(&cluster_locked->lock); | |
f75b130e JB |
926 | node = rb_first(&cluster->root); |
927 | cluster = NULL; | |
928 | } | |
929 | ||
a67509c3 JB |
930 | /* Write out the extent entries */ |
931 | while (node) { | |
932 | struct btrfs_free_space *e; | |
0cb59c99 | 933 | |
a67509c3 | 934 | e = rb_entry(node, struct btrfs_free_space, offset_index); |
d4452bc5 | 935 | *entries += 1; |
0cb59c99 | 936 | |
d4452bc5 | 937 | ret = io_ctl_add_entry(io_ctl, e->offset, e->bytes, |
a67509c3 JB |
938 | e->bitmap); |
939 | if (ret) | |
d4452bc5 | 940 | goto fail; |
2f356126 | 941 | |
a67509c3 | 942 | if (e->bitmap) { |
d4452bc5 CM |
943 | list_add_tail(&e->list, bitmap_list); |
944 | *bitmaps += 1; | |
2f356126 | 945 | } |
a67509c3 JB |
946 | node = rb_next(node); |
947 | if (!node && cluster) { | |
948 | node = rb_first(&cluster->root); | |
1bbc621e CM |
949 | cluster_locked = cluster; |
950 | spin_lock(&cluster_locked->lock); | |
a67509c3 | 951 | cluster = NULL; |
43be2146 | 952 | } |
a67509c3 | 953 | } |
1bbc621e CM |
954 | if (cluster_locked) { |
955 | spin_unlock(&cluster_locked->lock); | |
956 | cluster_locked = NULL; | |
957 | } | |
55507ce3 FM |
958 | |
959 | /* | |
960 | * Make sure we don't miss any range that was removed from our rbtree | |
961 | * because trimming is running. Otherwise after a umount+mount (or crash | |
962 | * after committing the transaction) we would leak free space and get | |
963 | * an inconsistent free space cache report from fsck. | |
964 | */ | |
965 | list_for_each_entry(trim_entry, &ctl->trimming_ranges, list) { | |
966 | ret = io_ctl_add_entry(io_ctl, trim_entry->start, | |
967 | trim_entry->bytes, NULL); | |
968 | if (ret) | |
969 | goto fail; | |
970 | *entries += 1; | |
971 | } | |
972 | ||
d4452bc5 CM |
973 | return 0; |
974 | fail: | |
1bbc621e CM |
975 | if (cluster_locked) |
976 | spin_unlock(&cluster_locked->lock); | |
d4452bc5 CM |
977 | return -ENOSPC; |
978 | } | |
979 | ||
980 | static noinline_for_stack int | |
981 | update_cache_item(struct btrfs_trans_handle *trans, | |
982 | struct btrfs_root *root, | |
983 | struct inode *inode, | |
984 | struct btrfs_path *path, u64 offset, | |
985 | int entries, int bitmaps) | |
986 | { | |
987 | struct btrfs_key key; | |
988 | struct btrfs_free_space_header *header; | |
989 | struct extent_buffer *leaf; | |
990 | int ret; | |
991 | ||
992 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
993 | key.offset = offset; | |
994 | key.type = 0; | |
995 | ||
996 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
997 | if (ret < 0) { | |
998 | clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1, | |
999 | EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL, | |
1000 | GFP_NOFS); | |
1001 | goto fail; | |
1002 | } | |
1003 | leaf = path->nodes[0]; | |
1004 | if (ret > 0) { | |
1005 | struct btrfs_key found_key; | |
1006 | ASSERT(path->slots[0]); | |
1007 | path->slots[0]--; | |
1008 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1009 | if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID || | |
1010 | found_key.offset != offset) { | |
1011 | clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, | |
1012 | inode->i_size - 1, | |
1013 | EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, | |
1014 | NULL, GFP_NOFS); | |
1015 | btrfs_release_path(path); | |
1016 | goto fail; | |
1017 | } | |
1018 | } | |
1019 | ||
1020 | BTRFS_I(inode)->generation = trans->transid; | |
1021 | header = btrfs_item_ptr(leaf, path->slots[0], | |
1022 | struct btrfs_free_space_header); | |
1023 | btrfs_set_free_space_entries(leaf, header, entries); | |
1024 | btrfs_set_free_space_bitmaps(leaf, header, bitmaps); | |
1025 | btrfs_set_free_space_generation(leaf, header, trans->transid); | |
1026 | btrfs_mark_buffer_dirty(leaf); | |
1027 | btrfs_release_path(path); | |
1028 | ||
1029 | return 0; | |
1030 | ||
1031 | fail: | |
1032 | return -1; | |
1033 | } | |
1034 | ||
1035 | static noinline_for_stack int | |
5349d6c3 MX |
1036 | write_pinned_extent_entries(struct btrfs_root *root, |
1037 | struct btrfs_block_group_cache *block_group, | |
4c6d1d85 | 1038 | struct btrfs_io_ctl *io_ctl, |
5349d6c3 | 1039 | int *entries) |
d4452bc5 CM |
1040 | { |
1041 | u64 start, extent_start, extent_end, len; | |
d4452bc5 CM |
1042 | struct extent_io_tree *unpin = NULL; |
1043 | int ret; | |
43be2146 | 1044 | |
5349d6c3 MX |
1045 | if (!block_group) |
1046 | return 0; | |
1047 | ||
a67509c3 JB |
1048 | /* |
1049 | * We want to add any pinned extents to our free space cache | |
1050 | * so we don't leak the space | |
d4452bc5 | 1051 | * |
db804f23 LZ |
1052 | * We shouldn't have switched the pinned extents yet so this is the |
1053 | * right one | |
1054 | */ | |
1055 | unpin = root->fs_info->pinned_extents; | |
1056 | ||
5349d6c3 | 1057 | start = block_group->key.objectid; |
db804f23 | 1058 | |
5349d6c3 | 1059 | while (start < block_group->key.objectid + block_group->key.offset) { |
db804f23 LZ |
1060 | ret = find_first_extent_bit(unpin, start, |
1061 | &extent_start, &extent_end, | |
e6138876 | 1062 | EXTENT_DIRTY, NULL); |
5349d6c3 MX |
1063 | if (ret) |
1064 | return 0; | |
0cb59c99 | 1065 | |
a67509c3 | 1066 | /* This pinned extent is out of our range */ |
db804f23 | 1067 | if (extent_start >= block_group->key.objectid + |
a67509c3 | 1068 | block_group->key.offset) |
5349d6c3 | 1069 | return 0; |
2f356126 | 1070 | |
db804f23 LZ |
1071 | extent_start = max(extent_start, start); |
1072 | extent_end = min(block_group->key.objectid + | |
1073 | block_group->key.offset, extent_end + 1); | |
1074 | len = extent_end - extent_start; | |
0cb59c99 | 1075 | |
d4452bc5 CM |
1076 | *entries += 1; |
1077 | ret = io_ctl_add_entry(io_ctl, extent_start, len, NULL); | |
a67509c3 | 1078 | if (ret) |
5349d6c3 | 1079 | return -ENOSPC; |
0cb59c99 | 1080 | |
db804f23 | 1081 | start = extent_end; |
a67509c3 | 1082 | } |
0cb59c99 | 1083 | |
5349d6c3 MX |
1084 | return 0; |
1085 | } | |
1086 | ||
1087 | static noinline_for_stack int | |
4c6d1d85 | 1088 | write_bitmap_entries(struct btrfs_io_ctl *io_ctl, struct list_head *bitmap_list) |
5349d6c3 | 1089 | { |
7ae1681e | 1090 | struct btrfs_free_space *entry, *next; |
5349d6c3 MX |
1091 | int ret; |
1092 | ||
0cb59c99 | 1093 | /* Write out the bitmaps */ |
7ae1681e | 1094 | list_for_each_entry_safe(entry, next, bitmap_list, list) { |
d4452bc5 | 1095 | ret = io_ctl_add_bitmap(io_ctl, entry->bitmap); |
a67509c3 | 1096 | if (ret) |
5349d6c3 | 1097 | return -ENOSPC; |
0cb59c99 | 1098 | list_del_init(&entry->list); |
be1a12a0 JB |
1099 | } |
1100 | ||
5349d6c3 MX |
1101 | return 0; |
1102 | } | |
0cb59c99 | 1103 | |
5349d6c3 MX |
1104 | static int flush_dirty_cache(struct inode *inode) |
1105 | { | |
1106 | int ret; | |
be1a12a0 | 1107 | |
0ef8b726 | 1108 | ret = btrfs_wait_ordered_range(inode, 0, (u64)-1); |
5349d6c3 | 1109 | if (ret) |
0ef8b726 JB |
1110 | clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1, |
1111 | EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL, | |
1112 | GFP_NOFS); | |
0cb59c99 | 1113 | |
5349d6c3 | 1114 | return ret; |
d4452bc5 CM |
1115 | } |
1116 | ||
1117 | static void noinline_for_stack | |
a3bdccc4 | 1118 | cleanup_bitmap_list(struct list_head *bitmap_list) |
d4452bc5 | 1119 | { |
7ae1681e | 1120 | struct btrfs_free_space *entry, *next; |
5349d6c3 | 1121 | |
7ae1681e | 1122 | list_for_each_entry_safe(entry, next, bitmap_list, list) |
d4452bc5 | 1123 | list_del_init(&entry->list); |
a3bdccc4 CM |
1124 | } |
1125 | ||
1126 | static void noinline_for_stack | |
1127 | cleanup_write_cache_enospc(struct inode *inode, | |
1128 | struct btrfs_io_ctl *io_ctl, | |
1129 | struct extent_state **cached_state, | |
1130 | struct list_head *bitmap_list) | |
1131 | { | |
d4452bc5 CM |
1132 | io_ctl_drop_pages(io_ctl); |
1133 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0, | |
1134 | i_size_read(inode) - 1, cached_state, | |
1135 | GFP_NOFS); | |
1136 | } | |
549b4fdb | 1137 | |
c9dc4c65 CM |
1138 | int btrfs_wait_cache_io(struct btrfs_root *root, |
1139 | struct btrfs_trans_handle *trans, | |
1140 | struct btrfs_block_group_cache *block_group, | |
1141 | struct btrfs_io_ctl *io_ctl, | |
1142 | struct btrfs_path *path, u64 offset) | |
1143 | { | |
1144 | int ret; | |
1145 | struct inode *inode = io_ctl->inode; | |
1146 | ||
1bbc621e CM |
1147 | if (!inode) |
1148 | return 0; | |
1149 | ||
85db36cf CM |
1150 | if (block_group) |
1151 | root = root->fs_info->tree_root; | |
c9dc4c65 CM |
1152 | |
1153 | /* Flush the dirty pages in the cache file. */ | |
1154 | ret = flush_dirty_cache(inode); | |
1155 | if (ret) | |
1156 | goto out; | |
1157 | ||
1158 | /* Update the cache item to tell everyone this cache file is valid. */ | |
1159 | ret = update_cache_item(trans, root, inode, path, offset, | |
1160 | io_ctl->entries, io_ctl->bitmaps); | |
1161 | out: | |
1162 | io_ctl_free(io_ctl); | |
1163 | if (ret) { | |
1164 | invalidate_inode_pages2(inode->i_mapping); | |
1165 | BTRFS_I(inode)->generation = 0; | |
1166 | if (block_group) { | |
1167 | #ifdef DEBUG | |
1168 | btrfs_err(root->fs_info, | |
1169 | "failed to write free space cache for block group %llu", | |
1170 | block_group->key.objectid); | |
1171 | #endif | |
1172 | } | |
1173 | } | |
1174 | btrfs_update_inode(trans, root, inode); | |
1175 | ||
1176 | if (block_group) { | |
1bbc621e CM |
1177 | /* the dirty list is protected by the dirty_bgs_lock */ |
1178 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
1179 | ||
1180 | /* the disk_cache_state is protected by the block group lock */ | |
c9dc4c65 CM |
1181 | spin_lock(&block_group->lock); |
1182 | ||
1183 | /* | |
1184 | * only mark this as written if we didn't get put back on | |
1bbc621e CM |
1185 | * the dirty list while waiting for IO. Otherwise our |
1186 | * cache state won't be right, and we won't get written again | |
c9dc4c65 CM |
1187 | */ |
1188 | if (!ret && list_empty(&block_group->dirty_list)) | |
1189 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
1190 | else if (ret) | |
1191 | block_group->disk_cache_state = BTRFS_DC_ERROR; | |
1192 | ||
1193 | spin_unlock(&block_group->lock); | |
1bbc621e | 1194 | spin_unlock(&trans->transaction->dirty_bgs_lock); |
c9dc4c65 CM |
1195 | io_ctl->inode = NULL; |
1196 | iput(inode); | |
1197 | } | |
1198 | ||
1199 | return ret; | |
1200 | ||
1201 | } | |
1202 | ||
d4452bc5 CM |
1203 | /** |
1204 | * __btrfs_write_out_cache - write out cached info to an inode | |
1205 | * @root - the root the inode belongs to | |
1206 | * @ctl - the free space cache we are going to write out | |
1207 | * @block_group - the block_group for this cache if it belongs to a block_group | |
1208 | * @trans - the trans handle | |
1209 | * @path - the path to use | |
1210 | * @offset - the offset for the key we'll insert | |
1211 | * | |
1212 | * This function writes out a free space cache struct to disk for quick recovery | |
8cd1e731 | 1213 | * on mount. This will return 0 if it was successful in writing the cache out, |
b8605454 | 1214 | * or an errno if it was not. |
d4452bc5 CM |
1215 | */ |
1216 | static int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode, | |
1217 | struct btrfs_free_space_ctl *ctl, | |
1218 | struct btrfs_block_group_cache *block_group, | |
c9dc4c65 | 1219 | struct btrfs_io_ctl *io_ctl, |
d4452bc5 CM |
1220 | struct btrfs_trans_handle *trans, |
1221 | struct btrfs_path *path, u64 offset) | |
1222 | { | |
1223 | struct extent_state *cached_state = NULL; | |
5349d6c3 | 1224 | LIST_HEAD(bitmap_list); |
d4452bc5 CM |
1225 | int entries = 0; |
1226 | int bitmaps = 0; | |
1227 | int ret; | |
c9dc4c65 | 1228 | int must_iput = 0; |
d4452bc5 CM |
1229 | |
1230 | if (!i_size_read(inode)) | |
b8605454 | 1231 | return -EIO; |
d4452bc5 | 1232 | |
c9dc4c65 CM |
1233 | WARN_ON(io_ctl->pages); |
1234 | ret = io_ctl_init(io_ctl, inode, root, 1); | |
d4452bc5 | 1235 | if (ret) |
b8605454 | 1236 | return ret; |
d4452bc5 | 1237 | |
e570fd27 MX |
1238 | if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA)) { |
1239 | down_write(&block_group->data_rwsem); | |
1240 | spin_lock(&block_group->lock); | |
1241 | if (block_group->delalloc_bytes) { | |
1242 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
1243 | spin_unlock(&block_group->lock); | |
1244 | up_write(&block_group->data_rwsem); | |
1245 | BTRFS_I(inode)->generation = 0; | |
1246 | ret = 0; | |
c9dc4c65 | 1247 | must_iput = 1; |
e570fd27 MX |
1248 | goto out; |
1249 | } | |
1250 | spin_unlock(&block_group->lock); | |
1251 | } | |
1252 | ||
d4452bc5 | 1253 | /* Lock all pages first so we can lock the extent safely. */ |
b8605454 OS |
1254 | ret = io_ctl_prepare_pages(io_ctl, inode, 0); |
1255 | if (ret) | |
1256 | goto out; | |
d4452bc5 CM |
1257 | |
1258 | lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1, | |
ff13db41 | 1259 | &cached_state); |
d4452bc5 | 1260 | |
c9dc4c65 | 1261 | io_ctl_set_generation(io_ctl, trans->transid); |
d4452bc5 | 1262 | |
55507ce3 | 1263 | mutex_lock(&ctl->cache_writeout_mutex); |
5349d6c3 | 1264 | /* Write out the extent entries in the free space cache */ |
1bbc621e | 1265 | spin_lock(&ctl->tree_lock); |
c9dc4c65 | 1266 | ret = write_cache_extent_entries(io_ctl, ctl, |
d4452bc5 CM |
1267 | block_group, &entries, &bitmaps, |
1268 | &bitmap_list); | |
a3bdccc4 CM |
1269 | if (ret) |
1270 | goto out_nospc_locked; | |
d4452bc5 | 1271 | |
5349d6c3 MX |
1272 | /* |
1273 | * Some spaces that are freed in the current transaction are pinned, | |
1274 | * they will be added into free space cache after the transaction is | |
1275 | * committed, we shouldn't lose them. | |
1bbc621e CM |
1276 | * |
1277 | * If this changes while we are working we'll get added back to | |
1278 | * the dirty list and redo it. No locking needed | |
5349d6c3 | 1279 | */ |
c9dc4c65 | 1280 | ret = write_pinned_extent_entries(root, block_group, io_ctl, &entries); |
a3bdccc4 CM |
1281 | if (ret) |
1282 | goto out_nospc_locked; | |
5349d6c3 | 1283 | |
55507ce3 FM |
1284 | /* |
1285 | * At last, we write out all the bitmaps and keep cache_writeout_mutex | |
1286 | * locked while doing it because a concurrent trim can be manipulating | |
1287 | * or freeing the bitmap. | |
1288 | */ | |
c9dc4c65 | 1289 | ret = write_bitmap_entries(io_ctl, &bitmap_list); |
1bbc621e | 1290 | spin_unlock(&ctl->tree_lock); |
55507ce3 | 1291 | mutex_unlock(&ctl->cache_writeout_mutex); |
5349d6c3 MX |
1292 | if (ret) |
1293 | goto out_nospc; | |
1294 | ||
1295 | /* Zero out the rest of the pages just to make sure */ | |
c9dc4c65 | 1296 | io_ctl_zero_remaining_pages(io_ctl); |
d4452bc5 | 1297 | |
5349d6c3 | 1298 | /* Everything is written out, now we dirty the pages in the file. */ |
c9dc4c65 | 1299 | ret = btrfs_dirty_pages(root, inode, io_ctl->pages, io_ctl->num_pages, |
5349d6c3 MX |
1300 | 0, i_size_read(inode), &cached_state); |
1301 | if (ret) | |
d4452bc5 | 1302 | goto out_nospc; |
5349d6c3 | 1303 | |
e570fd27 MX |
1304 | if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA)) |
1305 | up_write(&block_group->data_rwsem); | |
5349d6c3 MX |
1306 | /* |
1307 | * Release the pages and unlock the extent, we will flush | |
1308 | * them out later | |
1309 | */ | |
c9dc4c65 | 1310 | io_ctl_drop_pages(io_ctl); |
5349d6c3 MX |
1311 | |
1312 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0, | |
1313 | i_size_read(inode) - 1, &cached_state, GFP_NOFS); | |
1314 | ||
c9dc4c65 CM |
1315 | /* |
1316 | * at this point the pages are under IO and we're happy, | |
1317 | * The caller is responsible for waiting on them and updating the | |
1318 | * the cache and the inode | |
1319 | */ | |
1320 | io_ctl->entries = entries; | |
1321 | io_ctl->bitmaps = bitmaps; | |
1322 | ||
1323 | ret = btrfs_fdatawrite_range(inode, 0, (u64)-1); | |
5349d6c3 | 1324 | if (ret) |
d4452bc5 CM |
1325 | goto out; |
1326 | ||
c9dc4c65 CM |
1327 | return 0; |
1328 | ||
2f356126 | 1329 | out: |
c9dc4c65 CM |
1330 | io_ctl->inode = NULL; |
1331 | io_ctl_free(io_ctl); | |
5349d6c3 | 1332 | if (ret) { |
a67509c3 | 1333 | invalidate_inode_pages2(inode->i_mapping); |
0cb59c99 JB |
1334 | BTRFS_I(inode)->generation = 0; |
1335 | } | |
0cb59c99 | 1336 | btrfs_update_inode(trans, root, inode); |
c9dc4c65 CM |
1337 | if (must_iput) |
1338 | iput(inode); | |
5349d6c3 | 1339 | return ret; |
a67509c3 | 1340 | |
a3bdccc4 CM |
1341 | out_nospc_locked: |
1342 | cleanup_bitmap_list(&bitmap_list); | |
1343 | spin_unlock(&ctl->tree_lock); | |
1344 | mutex_unlock(&ctl->cache_writeout_mutex); | |
1345 | ||
a67509c3 | 1346 | out_nospc: |
c9dc4c65 | 1347 | cleanup_write_cache_enospc(inode, io_ctl, &cached_state, &bitmap_list); |
e570fd27 MX |
1348 | |
1349 | if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA)) | |
1350 | up_write(&block_group->data_rwsem); | |
1351 | ||
a67509c3 | 1352 | goto out; |
0414efae LZ |
1353 | } |
1354 | ||
1355 | int btrfs_write_out_cache(struct btrfs_root *root, | |
1356 | struct btrfs_trans_handle *trans, | |
1357 | struct btrfs_block_group_cache *block_group, | |
1358 | struct btrfs_path *path) | |
1359 | { | |
1360 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
1361 | struct inode *inode; | |
1362 | int ret = 0; | |
1363 | ||
1364 | root = root->fs_info->tree_root; | |
1365 | ||
1366 | spin_lock(&block_group->lock); | |
1367 | if (block_group->disk_cache_state < BTRFS_DC_SETUP) { | |
1368 | spin_unlock(&block_group->lock); | |
e570fd27 MX |
1369 | return 0; |
1370 | } | |
0414efae LZ |
1371 | spin_unlock(&block_group->lock); |
1372 | ||
1373 | inode = lookup_free_space_inode(root, block_group, path); | |
1374 | if (IS_ERR(inode)) | |
1375 | return 0; | |
1376 | ||
c9dc4c65 CM |
1377 | ret = __btrfs_write_out_cache(root, inode, ctl, block_group, |
1378 | &block_group->io_ctl, trans, | |
0414efae | 1379 | path, block_group->key.objectid); |
c09544e0 | 1380 | if (ret) { |
c09544e0 | 1381 | #ifdef DEBUG |
c2cf52eb SK |
1382 | btrfs_err(root->fs_info, |
1383 | "failed to write free space cache for block group %llu", | |
1384 | block_group->key.objectid); | |
c09544e0 | 1385 | #endif |
c9dc4c65 CM |
1386 | spin_lock(&block_group->lock); |
1387 | block_group->disk_cache_state = BTRFS_DC_ERROR; | |
1388 | spin_unlock(&block_group->lock); | |
1389 | ||
1390 | block_group->io_ctl.inode = NULL; | |
1391 | iput(inode); | |
0414efae LZ |
1392 | } |
1393 | ||
c9dc4c65 CM |
1394 | /* |
1395 | * if ret == 0 the caller is expected to call btrfs_wait_cache_io | |
1396 | * to wait for IO and put the inode | |
1397 | */ | |
1398 | ||
0cb59c99 JB |
1399 | return ret; |
1400 | } | |
1401 | ||
34d52cb6 | 1402 | static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit, |
96303081 | 1403 | u64 offset) |
0f9dd46c | 1404 | { |
b12d6869 | 1405 | ASSERT(offset >= bitmap_start); |
96303081 | 1406 | offset -= bitmap_start; |
34d52cb6 | 1407 | return (unsigned long)(div_u64(offset, unit)); |
96303081 | 1408 | } |
0f9dd46c | 1409 | |
34d52cb6 | 1410 | static inline unsigned long bytes_to_bits(u64 bytes, u32 unit) |
96303081 | 1411 | { |
34d52cb6 | 1412 | return (unsigned long)(div_u64(bytes, unit)); |
96303081 | 1413 | } |
0f9dd46c | 1414 | |
34d52cb6 | 1415 | static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1416 | u64 offset) |
1417 | { | |
1418 | u64 bitmap_start; | |
0ef6447a | 1419 | u64 bytes_per_bitmap; |
0f9dd46c | 1420 | |
34d52cb6 LZ |
1421 | bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit; |
1422 | bitmap_start = offset - ctl->start; | |
0ef6447a | 1423 | bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap); |
96303081 | 1424 | bitmap_start *= bytes_per_bitmap; |
34d52cb6 | 1425 | bitmap_start += ctl->start; |
0f9dd46c | 1426 | |
96303081 | 1427 | return bitmap_start; |
0f9dd46c JB |
1428 | } |
1429 | ||
96303081 JB |
1430 | static int tree_insert_offset(struct rb_root *root, u64 offset, |
1431 | struct rb_node *node, int bitmap) | |
0f9dd46c JB |
1432 | { |
1433 | struct rb_node **p = &root->rb_node; | |
1434 | struct rb_node *parent = NULL; | |
1435 | struct btrfs_free_space *info; | |
1436 | ||
1437 | while (*p) { | |
1438 | parent = *p; | |
96303081 | 1439 | info = rb_entry(parent, struct btrfs_free_space, offset_index); |
0f9dd46c | 1440 | |
96303081 | 1441 | if (offset < info->offset) { |
0f9dd46c | 1442 | p = &(*p)->rb_left; |
96303081 | 1443 | } else if (offset > info->offset) { |
0f9dd46c | 1444 | p = &(*p)->rb_right; |
96303081 JB |
1445 | } else { |
1446 | /* | |
1447 | * we could have a bitmap entry and an extent entry | |
1448 | * share the same offset. If this is the case, we want | |
1449 | * the extent entry to always be found first if we do a | |
1450 | * linear search through the tree, since we want to have | |
1451 | * the quickest allocation time, and allocating from an | |
1452 | * extent is faster than allocating from a bitmap. So | |
1453 | * if we're inserting a bitmap and we find an entry at | |
1454 | * this offset, we want to go right, or after this entry | |
1455 | * logically. If we are inserting an extent and we've | |
1456 | * found a bitmap, we want to go left, or before | |
1457 | * logically. | |
1458 | */ | |
1459 | if (bitmap) { | |
207dde82 JB |
1460 | if (info->bitmap) { |
1461 | WARN_ON_ONCE(1); | |
1462 | return -EEXIST; | |
1463 | } | |
96303081 JB |
1464 | p = &(*p)->rb_right; |
1465 | } else { | |
207dde82 JB |
1466 | if (!info->bitmap) { |
1467 | WARN_ON_ONCE(1); | |
1468 | return -EEXIST; | |
1469 | } | |
96303081 JB |
1470 | p = &(*p)->rb_left; |
1471 | } | |
1472 | } | |
0f9dd46c JB |
1473 | } |
1474 | ||
1475 | rb_link_node(node, parent, p); | |
1476 | rb_insert_color(node, root); | |
1477 | ||
1478 | return 0; | |
1479 | } | |
1480 | ||
1481 | /* | |
70cb0743 JB |
1482 | * searches the tree for the given offset. |
1483 | * | |
96303081 JB |
1484 | * fuzzy - If this is set, then we are trying to make an allocation, and we just |
1485 | * want a section that has at least bytes size and comes at or after the given | |
1486 | * offset. | |
0f9dd46c | 1487 | */ |
96303081 | 1488 | static struct btrfs_free_space * |
34d52cb6 | 1489 | tree_search_offset(struct btrfs_free_space_ctl *ctl, |
96303081 | 1490 | u64 offset, int bitmap_only, int fuzzy) |
0f9dd46c | 1491 | { |
34d52cb6 | 1492 | struct rb_node *n = ctl->free_space_offset.rb_node; |
96303081 JB |
1493 | struct btrfs_free_space *entry, *prev = NULL; |
1494 | ||
1495 | /* find entry that is closest to the 'offset' */ | |
1496 | while (1) { | |
1497 | if (!n) { | |
1498 | entry = NULL; | |
1499 | break; | |
1500 | } | |
0f9dd46c | 1501 | |
0f9dd46c | 1502 | entry = rb_entry(n, struct btrfs_free_space, offset_index); |
96303081 | 1503 | prev = entry; |
0f9dd46c | 1504 | |
96303081 | 1505 | if (offset < entry->offset) |
0f9dd46c | 1506 | n = n->rb_left; |
96303081 | 1507 | else if (offset > entry->offset) |
0f9dd46c | 1508 | n = n->rb_right; |
96303081 | 1509 | else |
0f9dd46c | 1510 | break; |
0f9dd46c JB |
1511 | } |
1512 | ||
96303081 JB |
1513 | if (bitmap_only) { |
1514 | if (!entry) | |
1515 | return NULL; | |
1516 | if (entry->bitmap) | |
1517 | return entry; | |
0f9dd46c | 1518 | |
96303081 JB |
1519 | /* |
1520 | * bitmap entry and extent entry may share same offset, | |
1521 | * in that case, bitmap entry comes after extent entry. | |
1522 | */ | |
1523 | n = rb_next(n); | |
1524 | if (!n) | |
1525 | return NULL; | |
1526 | entry = rb_entry(n, struct btrfs_free_space, offset_index); | |
1527 | if (entry->offset != offset) | |
1528 | return NULL; | |
0f9dd46c | 1529 | |
96303081 JB |
1530 | WARN_ON(!entry->bitmap); |
1531 | return entry; | |
1532 | } else if (entry) { | |
1533 | if (entry->bitmap) { | |
0f9dd46c | 1534 | /* |
96303081 JB |
1535 | * if previous extent entry covers the offset, |
1536 | * we should return it instead of the bitmap entry | |
0f9dd46c | 1537 | */ |
de6c4115 MX |
1538 | n = rb_prev(&entry->offset_index); |
1539 | if (n) { | |
96303081 JB |
1540 | prev = rb_entry(n, struct btrfs_free_space, |
1541 | offset_index); | |
de6c4115 MX |
1542 | if (!prev->bitmap && |
1543 | prev->offset + prev->bytes > offset) | |
1544 | entry = prev; | |
0f9dd46c | 1545 | } |
96303081 JB |
1546 | } |
1547 | return entry; | |
1548 | } | |
1549 | ||
1550 | if (!prev) | |
1551 | return NULL; | |
1552 | ||
1553 | /* find last entry before the 'offset' */ | |
1554 | entry = prev; | |
1555 | if (entry->offset > offset) { | |
1556 | n = rb_prev(&entry->offset_index); | |
1557 | if (n) { | |
1558 | entry = rb_entry(n, struct btrfs_free_space, | |
1559 | offset_index); | |
b12d6869 | 1560 | ASSERT(entry->offset <= offset); |
0f9dd46c | 1561 | } else { |
96303081 JB |
1562 | if (fuzzy) |
1563 | return entry; | |
1564 | else | |
1565 | return NULL; | |
0f9dd46c JB |
1566 | } |
1567 | } | |
1568 | ||
96303081 | 1569 | if (entry->bitmap) { |
de6c4115 MX |
1570 | n = rb_prev(&entry->offset_index); |
1571 | if (n) { | |
96303081 JB |
1572 | prev = rb_entry(n, struct btrfs_free_space, |
1573 | offset_index); | |
de6c4115 MX |
1574 | if (!prev->bitmap && |
1575 | prev->offset + prev->bytes > offset) | |
1576 | return prev; | |
96303081 | 1577 | } |
34d52cb6 | 1578 | if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset) |
96303081 JB |
1579 | return entry; |
1580 | } else if (entry->offset + entry->bytes > offset) | |
1581 | return entry; | |
1582 | ||
1583 | if (!fuzzy) | |
1584 | return NULL; | |
1585 | ||
1586 | while (1) { | |
1587 | if (entry->bitmap) { | |
1588 | if (entry->offset + BITS_PER_BITMAP * | |
34d52cb6 | 1589 | ctl->unit > offset) |
96303081 JB |
1590 | break; |
1591 | } else { | |
1592 | if (entry->offset + entry->bytes > offset) | |
1593 | break; | |
1594 | } | |
1595 | ||
1596 | n = rb_next(&entry->offset_index); | |
1597 | if (!n) | |
1598 | return NULL; | |
1599 | entry = rb_entry(n, struct btrfs_free_space, offset_index); | |
1600 | } | |
1601 | return entry; | |
0f9dd46c JB |
1602 | } |
1603 | ||
f333adb5 | 1604 | static inline void |
34d52cb6 | 1605 | __unlink_free_space(struct btrfs_free_space_ctl *ctl, |
f333adb5 | 1606 | struct btrfs_free_space *info) |
0f9dd46c | 1607 | { |
34d52cb6 LZ |
1608 | rb_erase(&info->offset_index, &ctl->free_space_offset); |
1609 | ctl->free_extents--; | |
f333adb5 LZ |
1610 | } |
1611 | ||
34d52cb6 | 1612 | static void unlink_free_space(struct btrfs_free_space_ctl *ctl, |
f333adb5 LZ |
1613 | struct btrfs_free_space *info) |
1614 | { | |
34d52cb6 LZ |
1615 | __unlink_free_space(ctl, info); |
1616 | ctl->free_space -= info->bytes; | |
0f9dd46c JB |
1617 | } |
1618 | ||
34d52cb6 | 1619 | static int link_free_space(struct btrfs_free_space_ctl *ctl, |
0f9dd46c JB |
1620 | struct btrfs_free_space *info) |
1621 | { | |
1622 | int ret = 0; | |
1623 | ||
b12d6869 | 1624 | ASSERT(info->bytes || info->bitmap); |
34d52cb6 | 1625 | ret = tree_insert_offset(&ctl->free_space_offset, info->offset, |
96303081 | 1626 | &info->offset_index, (info->bitmap != NULL)); |
0f9dd46c JB |
1627 | if (ret) |
1628 | return ret; | |
1629 | ||
34d52cb6 LZ |
1630 | ctl->free_space += info->bytes; |
1631 | ctl->free_extents++; | |
96303081 JB |
1632 | return ret; |
1633 | } | |
1634 | ||
34d52cb6 | 1635 | static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl) |
96303081 | 1636 | { |
34d52cb6 | 1637 | struct btrfs_block_group_cache *block_group = ctl->private; |
25891f79 JB |
1638 | u64 max_bytes; |
1639 | u64 bitmap_bytes; | |
1640 | u64 extent_bytes; | |
8eb2d829 | 1641 | u64 size = block_group->key.offset; |
0ef6447a FX |
1642 | u64 bytes_per_bg = BITS_PER_BITMAP * ctl->unit; |
1643 | u64 max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg); | |
34d52cb6 | 1644 | |
0ef6447a | 1645 | max_bitmaps = max_t(u64, max_bitmaps, 1); |
dde5740f | 1646 | |
b12d6869 | 1647 | ASSERT(ctl->total_bitmaps <= max_bitmaps); |
96303081 JB |
1648 | |
1649 | /* | |
1650 | * The goal is to keep the total amount of memory used per 1gb of space | |
1651 | * at or below 32k, so we need to adjust how much memory we allow to be | |
1652 | * used by extent based free space tracking | |
1653 | */ | |
ee22184b | 1654 | if (size < SZ_1G) |
8eb2d829 LZ |
1655 | max_bytes = MAX_CACHE_BYTES_PER_GIG; |
1656 | else | |
ee22184b | 1657 | max_bytes = MAX_CACHE_BYTES_PER_GIG * div_u64(size, SZ_1G); |
96303081 | 1658 | |
25891f79 JB |
1659 | /* |
1660 | * we want to account for 1 more bitmap than what we have so we can make | |
1661 | * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as | |
1662 | * we add more bitmaps. | |
1663 | */ | |
b9ef22de | 1664 | bitmap_bytes = (ctl->total_bitmaps + 1) * ctl->unit; |
96303081 | 1665 | |
25891f79 | 1666 | if (bitmap_bytes >= max_bytes) { |
34d52cb6 | 1667 | ctl->extents_thresh = 0; |
25891f79 JB |
1668 | return; |
1669 | } | |
96303081 | 1670 | |
25891f79 | 1671 | /* |
f8c269d7 | 1672 | * we want the extent entry threshold to always be at most 1/2 the max |
25891f79 JB |
1673 | * bytes we can have, or whatever is less than that. |
1674 | */ | |
1675 | extent_bytes = max_bytes - bitmap_bytes; | |
f8c269d7 | 1676 | extent_bytes = min_t(u64, extent_bytes, max_bytes >> 1); |
96303081 | 1677 | |
34d52cb6 | 1678 | ctl->extents_thresh = |
f8c269d7 | 1679 | div_u64(extent_bytes, sizeof(struct btrfs_free_space)); |
96303081 JB |
1680 | } |
1681 | ||
bb3ac5a4 MX |
1682 | static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl, |
1683 | struct btrfs_free_space *info, | |
1684 | u64 offset, u64 bytes) | |
96303081 | 1685 | { |
f38b6e75 | 1686 | unsigned long start, count; |
96303081 | 1687 | |
34d52cb6 LZ |
1688 | start = offset_to_bit(info->offset, ctl->unit, offset); |
1689 | count = bytes_to_bits(bytes, ctl->unit); | |
b12d6869 | 1690 | ASSERT(start + count <= BITS_PER_BITMAP); |
96303081 | 1691 | |
f38b6e75 | 1692 | bitmap_clear(info->bitmap, start, count); |
96303081 JB |
1693 | |
1694 | info->bytes -= bytes; | |
bb3ac5a4 MX |
1695 | } |
1696 | ||
1697 | static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl, | |
1698 | struct btrfs_free_space *info, u64 offset, | |
1699 | u64 bytes) | |
1700 | { | |
1701 | __bitmap_clear_bits(ctl, info, offset, bytes); | |
34d52cb6 | 1702 | ctl->free_space -= bytes; |
96303081 JB |
1703 | } |
1704 | ||
34d52cb6 | 1705 | static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl, |
817d52f8 JB |
1706 | struct btrfs_free_space *info, u64 offset, |
1707 | u64 bytes) | |
96303081 | 1708 | { |
f38b6e75 | 1709 | unsigned long start, count; |
96303081 | 1710 | |
34d52cb6 LZ |
1711 | start = offset_to_bit(info->offset, ctl->unit, offset); |
1712 | count = bytes_to_bits(bytes, ctl->unit); | |
b12d6869 | 1713 | ASSERT(start + count <= BITS_PER_BITMAP); |
96303081 | 1714 | |
f38b6e75 | 1715 | bitmap_set(info->bitmap, start, count); |
96303081 JB |
1716 | |
1717 | info->bytes += bytes; | |
34d52cb6 | 1718 | ctl->free_space += bytes; |
96303081 JB |
1719 | } |
1720 | ||
a4820398 MX |
1721 | /* |
1722 | * If we can not find suitable extent, we will use bytes to record | |
1723 | * the size of the max extent. | |
1724 | */ | |
34d52cb6 | 1725 | static int search_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 | 1726 | struct btrfs_free_space *bitmap_info, u64 *offset, |
0584f718 | 1727 | u64 *bytes, bool for_alloc) |
96303081 JB |
1728 | { |
1729 | unsigned long found_bits = 0; | |
a4820398 | 1730 | unsigned long max_bits = 0; |
96303081 JB |
1731 | unsigned long bits, i; |
1732 | unsigned long next_zero; | |
a4820398 | 1733 | unsigned long extent_bits; |
96303081 | 1734 | |
cef40483 JB |
1735 | /* |
1736 | * Skip searching the bitmap if we don't have a contiguous section that | |
1737 | * is large enough for this allocation. | |
1738 | */ | |
0584f718 JB |
1739 | if (for_alloc && |
1740 | bitmap_info->max_extent_size && | |
cef40483 JB |
1741 | bitmap_info->max_extent_size < *bytes) { |
1742 | *bytes = bitmap_info->max_extent_size; | |
1743 | return -1; | |
1744 | } | |
1745 | ||
34d52cb6 | 1746 | i = offset_to_bit(bitmap_info->offset, ctl->unit, |
96303081 | 1747 | max_t(u64, *offset, bitmap_info->offset)); |
34d52cb6 | 1748 | bits = bytes_to_bits(*bytes, ctl->unit); |
96303081 | 1749 | |
ebb3dad4 | 1750 | for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) { |
0584f718 JB |
1751 | if (for_alloc && bits == 1) { |
1752 | found_bits = 1; | |
1753 | break; | |
1754 | } | |
96303081 JB |
1755 | next_zero = find_next_zero_bit(bitmap_info->bitmap, |
1756 | BITS_PER_BITMAP, i); | |
a4820398 MX |
1757 | extent_bits = next_zero - i; |
1758 | if (extent_bits >= bits) { | |
1759 | found_bits = extent_bits; | |
96303081 | 1760 | break; |
a4820398 MX |
1761 | } else if (extent_bits > max_bits) { |
1762 | max_bits = extent_bits; | |
96303081 JB |
1763 | } |
1764 | i = next_zero; | |
1765 | } | |
1766 | ||
1767 | if (found_bits) { | |
34d52cb6 LZ |
1768 | *offset = (u64)(i * ctl->unit) + bitmap_info->offset; |
1769 | *bytes = (u64)(found_bits) * ctl->unit; | |
96303081 JB |
1770 | return 0; |
1771 | } | |
1772 | ||
a4820398 | 1773 | *bytes = (u64)(max_bits) * ctl->unit; |
cef40483 | 1774 | bitmap_info->max_extent_size = *bytes; |
96303081 JB |
1775 | return -1; |
1776 | } | |
1777 | ||
a4820398 | 1778 | /* Cache the size of the max extent in bytes */ |
34d52cb6 | 1779 | static struct btrfs_free_space * |
53b381b3 | 1780 | find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes, |
a4820398 | 1781 | unsigned long align, u64 *max_extent_size) |
96303081 JB |
1782 | { |
1783 | struct btrfs_free_space *entry; | |
1784 | struct rb_node *node; | |
53b381b3 DW |
1785 | u64 tmp; |
1786 | u64 align_off; | |
96303081 JB |
1787 | int ret; |
1788 | ||
34d52cb6 | 1789 | if (!ctl->free_space_offset.rb_node) |
a4820398 | 1790 | goto out; |
96303081 | 1791 | |
34d52cb6 | 1792 | entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1); |
96303081 | 1793 | if (!entry) |
a4820398 | 1794 | goto out; |
96303081 JB |
1795 | |
1796 | for (node = &entry->offset_index; node; node = rb_next(node)) { | |
1797 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
a4820398 MX |
1798 | if (entry->bytes < *bytes) { |
1799 | if (entry->bytes > *max_extent_size) | |
1800 | *max_extent_size = entry->bytes; | |
96303081 | 1801 | continue; |
a4820398 | 1802 | } |
96303081 | 1803 | |
53b381b3 DW |
1804 | /* make sure the space returned is big enough |
1805 | * to match our requested alignment | |
1806 | */ | |
1807 | if (*bytes >= align) { | |
a4820398 | 1808 | tmp = entry->offset - ctl->start + align - 1; |
47c5713f | 1809 | tmp = div64_u64(tmp, align); |
53b381b3 DW |
1810 | tmp = tmp * align + ctl->start; |
1811 | align_off = tmp - entry->offset; | |
1812 | } else { | |
1813 | align_off = 0; | |
1814 | tmp = entry->offset; | |
1815 | } | |
1816 | ||
a4820398 MX |
1817 | if (entry->bytes < *bytes + align_off) { |
1818 | if (entry->bytes > *max_extent_size) | |
1819 | *max_extent_size = entry->bytes; | |
53b381b3 | 1820 | continue; |
a4820398 | 1821 | } |
53b381b3 | 1822 | |
96303081 | 1823 | if (entry->bitmap) { |
a4820398 MX |
1824 | u64 size = *bytes; |
1825 | ||
0584f718 | 1826 | ret = search_bitmap(ctl, entry, &tmp, &size, true); |
53b381b3 DW |
1827 | if (!ret) { |
1828 | *offset = tmp; | |
a4820398 | 1829 | *bytes = size; |
96303081 | 1830 | return entry; |
a4820398 MX |
1831 | } else if (size > *max_extent_size) { |
1832 | *max_extent_size = size; | |
53b381b3 | 1833 | } |
96303081 JB |
1834 | continue; |
1835 | } | |
1836 | ||
53b381b3 DW |
1837 | *offset = tmp; |
1838 | *bytes = entry->bytes - align_off; | |
96303081 JB |
1839 | return entry; |
1840 | } | |
a4820398 | 1841 | out: |
96303081 JB |
1842 | return NULL; |
1843 | } | |
1844 | ||
34d52cb6 | 1845 | static void add_new_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1846 | struct btrfs_free_space *info, u64 offset) |
1847 | { | |
34d52cb6 | 1848 | info->offset = offset_to_bitmap(ctl, offset); |
f019f426 | 1849 | info->bytes = 0; |
f2d0f676 | 1850 | INIT_LIST_HEAD(&info->list); |
34d52cb6 LZ |
1851 | link_free_space(ctl, info); |
1852 | ctl->total_bitmaps++; | |
96303081 | 1853 | |
34d52cb6 | 1854 | ctl->op->recalc_thresholds(ctl); |
96303081 JB |
1855 | } |
1856 | ||
34d52cb6 | 1857 | static void free_bitmap(struct btrfs_free_space_ctl *ctl, |
edf6e2d1 LZ |
1858 | struct btrfs_free_space *bitmap_info) |
1859 | { | |
34d52cb6 | 1860 | unlink_free_space(ctl, bitmap_info); |
edf6e2d1 | 1861 | kfree(bitmap_info->bitmap); |
dc89e982 | 1862 | kmem_cache_free(btrfs_free_space_cachep, bitmap_info); |
34d52cb6 LZ |
1863 | ctl->total_bitmaps--; |
1864 | ctl->op->recalc_thresholds(ctl); | |
edf6e2d1 LZ |
1865 | } |
1866 | ||
34d52cb6 | 1867 | static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1868 | struct btrfs_free_space *bitmap_info, |
1869 | u64 *offset, u64 *bytes) | |
1870 | { | |
1871 | u64 end; | |
6606bb97 JB |
1872 | u64 search_start, search_bytes; |
1873 | int ret; | |
96303081 JB |
1874 | |
1875 | again: | |
34d52cb6 | 1876 | end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1; |
96303081 | 1877 | |
6606bb97 | 1878 | /* |
bdb7d303 JB |
1879 | * We need to search for bits in this bitmap. We could only cover some |
1880 | * of the extent in this bitmap thanks to how we add space, so we need | |
1881 | * to search for as much as it as we can and clear that amount, and then | |
1882 | * go searching for the next bit. | |
6606bb97 JB |
1883 | */ |
1884 | search_start = *offset; | |
bdb7d303 | 1885 | search_bytes = ctl->unit; |
13dbc089 | 1886 | search_bytes = min(search_bytes, end - search_start + 1); |
0584f718 JB |
1887 | ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes, |
1888 | false); | |
b50c6e25 JB |
1889 | if (ret < 0 || search_start != *offset) |
1890 | return -EINVAL; | |
6606bb97 | 1891 | |
bdb7d303 JB |
1892 | /* We may have found more bits than what we need */ |
1893 | search_bytes = min(search_bytes, *bytes); | |
1894 | ||
1895 | /* Cannot clear past the end of the bitmap */ | |
1896 | search_bytes = min(search_bytes, end - search_start + 1); | |
1897 | ||
1898 | bitmap_clear_bits(ctl, bitmap_info, search_start, search_bytes); | |
1899 | *offset += search_bytes; | |
1900 | *bytes -= search_bytes; | |
96303081 JB |
1901 | |
1902 | if (*bytes) { | |
6606bb97 | 1903 | struct rb_node *next = rb_next(&bitmap_info->offset_index); |
edf6e2d1 | 1904 | if (!bitmap_info->bytes) |
34d52cb6 | 1905 | free_bitmap(ctl, bitmap_info); |
96303081 | 1906 | |
6606bb97 JB |
1907 | /* |
1908 | * no entry after this bitmap, but we still have bytes to | |
1909 | * remove, so something has gone wrong. | |
1910 | */ | |
1911 | if (!next) | |
96303081 JB |
1912 | return -EINVAL; |
1913 | ||
6606bb97 JB |
1914 | bitmap_info = rb_entry(next, struct btrfs_free_space, |
1915 | offset_index); | |
1916 | ||
1917 | /* | |
1918 | * if the next entry isn't a bitmap we need to return to let the | |
1919 | * extent stuff do its work. | |
1920 | */ | |
96303081 JB |
1921 | if (!bitmap_info->bitmap) |
1922 | return -EAGAIN; | |
1923 | ||
6606bb97 JB |
1924 | /* |
1925 | * Ok the next item is a bitmap, but it may not actually hold | |
1926 | * the information for the rest of this free space stuff, so | |
1927 | * look for it, and if we don't find it return so we can try | |
1928 | * everything over again. | |
1929 | */ | |
1930 | search_start = *offset; | |
bdb7d303 | 1931 | search_bytes = ctl->unit; |
34d52cb6 | 1932 | ret = search_bitmap(ctl, bitmap_info, &search_start, |
0584f718 | 1933 | &search_bytes, false); |
6606bb97 JB |
1934 | if (ret < 0 || search_start != *offset) |
1935 | return -EAGAIN; | |
1936 | ||
96303081 | 1937 | goto again; |
edf6e2d1 | 1938 | } else if (!bitmap_info->bytes) |
34d52cb6 | 1939 | free_bitmap(ctl, bitmap_info); |
96303081 JB |
1940 | |
1941 | return 0; | |
1942 | } | |
1943 | ||
2cdc342c JB |
1944 | static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl, |
1945 | struct btrfs_free_space *info, u64 offset, | |
1946 | u64 bytes) | |
1947 | { | |
1948 | u64 bytes_to_set = 0; | |
1949 | u64 end; | |
1950 | ||
1951 | end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit); | |
1952 | ||
1953 | bytes_to_set = min(end - offset, bytes); | |
1954 | ||
1955 | bitmap_set_bits(ctl, info, offset, bytes_to_set); | |
1956 | ||
cef40483 JB |
1957 | /* |
1958 | * We set some bytes, we have no idea what the max extent size is | |
1959 | * anymore. | |
1960 | */ | |
1961 | info->max_extent_size = 0; | |
1962 | ||
2cdc342c JB |
1963 | return bytes_to_set; |
1964 | ||
1965 | } | |
1966 | ||
34d52cb6 LZ |
1967 | static bool use_bitmap(struct btrfs_free_space_ctl *ctl, |
1968 | struct btrfs_free_space *info) | |
96303081 | 1969 | { |
34d52cb6 | 1970 | struct btrfs_block_group_cache *block_group = ctl->private; |
d0bd4560 JB |
1971 | bool forced = false; |
1972 | ||
1973 | #ifdef CONFIG_BTRFS_DEBUG | |
1974 | if (btrfs_should_fragment_free_space(block_group->fs_info->extent_root, | |
1975 | block_group)) | |
1976 | forced = true; | |
1977 | #endif | |
96303081 JB |
1978 | |
1979 | /* | |
1980 | * If we are below the extents threshold then we can add this as an | |
1981 | * extent, and don't have to deal with the bitmap | |
1982 | */ | |
d0bd4560 | 1983 | if (!forced && ctl->free_extents < ctl->extents_thresh) { |
32cb0840 JB |
1984 | /* |
1985 | * If this block group has some small extents we don't want to | |
1986 | * use up all of our free slots in the cache with them, we want | |
01327610 | 1987 | * to reserve them to larger extents, however if we have plenty |
32cb0840 JB |
1988 | * of cache left then go ahead an dadd them, no sense in adding |
1989 | * the overhead of a bitmap if we don't have to. | |
1990 | */ | |
1991 | if (info->bytes <= block_group->sectorsize * 4) { | |
34d52cb6 LZ |
1992 | if (ctl->free_extents * 2 <= ctl->extents_thresh) |
1993 | return false; | |
32cb0840 | 1994 | } else { |
34d52cb6 | 1995 | return false; |
32cb0840 JB |
1996 | } |
1997 | } | |
96303081 JB |
1998 | |
1999 | /* | |
dde5740f JB |
2000 | * The original block groups from mkfs can be really small, like 8 |
2001 | * megabytes, so don't bother with a bitmap for those entries. However | |
2002 | * some block groups can be smaller than what a bitmap would cover but | |
2003 | * are still large enough that they could overflow the 32k memory limit, | |
2004 | * so allow those block groups to still be allowed to have a bitmap | |
2005 | * entry. | |
96303081 | 2006 | */ |
dde5740f | 2007 | if (((BITS_PER_BITMAP * ctl->unit) >> 1) > block_group->key.offset) |
34d52cb6 LZ |
2008 | return false; |
2009 | ||
2010 | return true; | |
2011 | } | |
2012 | ||
20e5506b | 2013 | static const struct btrfs_free_space_op free_space_op = { |
2cdc342c JB |
2014 | .recalc_thresholds = recalculate_thresholds, |
2015 | .use_bitmap = use_bitmap, | |
2016 | }; | |
2017 | ||
34d52cb6 LZ |
2018 | static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl, |
2019 | struct btrfs_free_space *info) | |
2020 | { | |
2021 | struct btrfs_free_space *bitmap_info; | |
2cdc342c | 2022 | struct btrfs_block_group_cache *block_group = NULL; |
34d52cb6 | 2023 | int added = 0; |
2cdc342c | 2024 | u64 bytes, offset, bytes_added; |
34d52cb6 | 2025 | int ret; |
96303081 JB |
2026 | |
2027 | bytes = info->bytes; | |
2028 | offset = info->offset; | |
2029 | ||
34d52cb6 LZ |
2030 | if (!ctl->op->use_bitmap(ctl, info)) |
2031 | return 0; | |
2032 | ||
2cdc342c JB |
2033 | if (ctl->op == &free_space_op) |
2034 | block_group = ctl->private; | |
38e87880 | 2035 | again: |
2cdc342c JB |
2036 | /* |
2037 | * Since we link bitmaps right into the cluster we need to see if we | |
2038 | * have a cluster here, and if so and it has our bitmap we need to add | |
2039 | * the free space to that bitmap. | |
2040 | */ | |
2041 | if (block_group && !list_empty(&block_group->cluster_list)) { | |
2042 | struct btrfs_free_cluster *cluster; | |
2043 | struct rb_node *node; | |
2044 | struct btrfs_free_space *entry; | |
2045 | ||
2046 | cluster = list_entry(block_group->cluster_list.next, | |
2047 | struct btrfs_free_cluster, | |
2048 | block_group_list); | |
2049 | spin_lock(&cluster->lock); | |
2050 | node = rb_first(&cluster->root); | |
2051 | if (!node) { | |
2052 | spin_unlock(&cluster->lock); | |
38e87880 | 2053 | goto no_cluster_bitmap; |
2cdc342c JB |
2054 | } |
2055 | ||
2056 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
2057 | if (!entry->bitmap) { | |
2058 | spin_unlock(&cluster->lock); | |
38e87880 | 2059 | goto no_cluster_bitmap; |
2cdc342c JB |
2060 | } |
2061 | ||
2062 | if (entry->offset == offset_to_bitmap(ctl, offset)) { | |
2063 | bytes_added = add_bytes_to_bitmap(ctl, entry, | |
2064 | offset, bytes); | |
2065 | bytes -= bytes_added; | |
2066 | offset += bytes_added; | |
2067 | } | |
2068 | spin_unlock(&cluster->lock); | |
2069 | if (!bytes) { | |
2070 | ret = 1; | |
2071 | goto out; | |
2072 | } | |
2073 | } | |
38e87880 CM |
2074 | |
2075 | no_cluster_bitmap: | |
34d52cb6 | 2076 | bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), |
96303081 JB |
2077 | 1, 0); |
2078 | if (!bitmap_info) { | |
b12d6869 | 2079 | ASSERT(added == 0); |
96303081 JB |
2080 | goto new_bitmap; |
2081 | } | |
2082 | ||
2cdc342c JB |
2083 | bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes); |
2084 | bytes -= bytes_added; | |
2085 | offset += bytes_added; | |
2086 | added = 0; | |
96303081 JB |
2087 | |
2088 | if (!bytes) { | |
2089 | ret = 1; | |
2090 | goto out; | |
2091 | } else | |
2092 | goto again; | |
2093 | ||
2094 | new_bitmap: | |
2095 | if (info && info->bitmap) { | |
34d52cb6 | 2096 | add_new_bitmap(ctl, info, offset); |
96303081 JB |
2097 | added = 1; |
2098 | info = NULL; | |
2099 | goto again; | |
2100 | } else { | |
34d52cb6 | 2101 | spin_unlock(&ctl->tree_lock); |
96303081 JB |
2102 | |
2103 | /* no pre-allocated info, allocate a new one */ | |
2104 | if (!info) { | |
dc89e982 JB |
2105 | info = kmem_cache_zalloc(btrfs_free_space_cachep, |
2106 | GFP_NOFS); | |
96303081 | 2107 | if (!info) { |
34d52cb6 | 2108 | spin_lock(&ctl->tree_lock); |
96303081 JB |
2109 | ret = -ENOMEM; |
2110 | goto out; | |
2111 | } | |
2112 | } | |
2113 | ||
2114 | /* allocate the bitmap */ | |
09cbfeaf | 2115 | info->bitmap = kzalloc(PAGE_SIZE, GFP_NOFS); |
34d52cb6 | 2116 | spin_lock(&ctl->tree_lock); |
96303081 JB |
2117 | if (!info->bitmap) { |
2118 | ret = -ENOMEM; | |
2119 | goto out; | |
2120 | } | |
2121 | goto again; | |
2122 | } | |
2123 | ||
2124 | out: | |
2125 | if (info) { | |
2126 | if (info->bitmap) | |
2127 | kfree(info->bitmap); | |
dc89e982 | 2128 | kmem_cache_free(btrfs_free_space_cachep, info); |
96303081 | 2129 | } |
0f9dd46c JB |
2130 | |
2131 | return ret; | |
2132 | } | |
2133 | ||
945d8962 | 2134 | static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl, |
f333adb5 | 2135 | struct btrfs_free_space *info, bool update_stat) |
0f9dd46c | 2136 | { |
120d66ee LZ |
2137 | struct btrfs_free_space *left_info; |
2138 | struct btrfs_free_space *right_info; | |
2139 | bool merged = false; | |
2140 | u64 offset = info->offset; | |
2141 | u64 bytes = info->bytes; | |
6226cb0a | 2142 | |
0f9dd46c JB |
2143 | /* |
2144 | * first we want to see if there is free space adjacent to the range we | |
2145 | * are adding, if there is remove that struct and add a new one to | |
2146 | * cover the entire range | |
2147 | */ | |
34d52cb6 | 2148 | right_info = tree_search_offset(ctl, offset + bytes, 0, 0); |
96303081 JB |
2149 | if (right_info && rb_prev(&right_info->offset_index)) |
2150 | left_info = rb_entry(rb_prev(&right_info->offset_index), | |
2151 | struct btrfs_free_space, offset_index); | |
2152 | else | |
34d52cb6 | 2153 | left_info = tree_search_offset(ctl, offset - 1, 0, 0); |
0f9dd46c | 2154 | |
96303081 | 2155 | if (right_info && !right_info->bitmap) { |
f333adb5 | 2156 | if (update_stat) |
34d52cb6 | 2157 | unlink_free_space(ctl, right_info); |
f333adb5 | 2158 | else |
34d52cb6 | 2159 | __unlink_free_space(ctl, right_info); |
6226cb0a | 2160 | info->bytes += right_info->bytes; |
dc89e982 | 2161 | kmem_cache_free(btrfs_free_space_cachep, right_info); |
120d66ee | 2162 | merged = true; |
0f9dd46c JB |
2163 | } |
2164 | ||
96303081 JB |
2165 | if (left_info && !left_info->bitmap && |
2166 | left_info->offset + left_info->bytes == offset) { | |
f333adb5 | 2167 | if (update_stat) |
34d52cb6 | 2168 | unlink_free_space(ctl, left_info); |
f333adb5 | 2169 | else |
34d52cb6 | 2170 | __unlink_free_space(ctl, left_info); |
6226cb0a JB |
2171 | info->offset = left_info->offset; |
2172 | info->bytes += left_info->bytes; | |
dc89e982 | 2173 | kmem_cache_free(btrfs_free_space_cachep, left_info); |
120d66ee | 2174 | merged = true; |
0f9dd46c JB |
2175 | } |
2176 | ||
120d66ee LZ |
2177 | return merged; |
2178 | } | |
2179 | ||
20005523 FM |
2180 | static bool steal_from_bitmap_to_end(struct btrfs_free_space_ctl *ctl, |
2181 | struct btrfs_free_space *info, | |
2182 | bool update_stat) | |
2183 | { | |
2184 | struct btrfs_free_space *bitmap; | |
2185 | unsigned long i; | |
2186 | unsigned long j; | |
2187 | const u64 end = info->offset + info->bytes; | |
2188 | const u64 bitmap_offset = offset_to_bitmap(ctl, end); | |
2189 | u64 bytes; | |
2190 | ||
2191 | bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0); | |
2192 | if (!bitmap) | |
2193 | return false; | |
2194 | ||
2195 | i = offset_to_bit(bitmap->offset, ctl->unit, end); | |
2196 | j = find_next_zero_bit(bitmap->bitmap, BITS_PER_BITMAP, i); | |
2197 | if (j == i) | |
2198 | return false; | |
2199 | bytes = (j - i) * ctl->unit; | |
2200 | info->bytes += bytes; | |
2201 | ||
2202 | if (update_stat) | |
2203 | bitmap_clear_bits(ctl, bitmap, end, bytes); | |
2204 | else | |
2205 | __bitmap_clear_bits(ctl, bitmap, end, bytes); | |
2206 | ||
2207 | if (!bitmap->bytes) | |
2208 | free_bitmap(ctl, bitmap); | |
2209 | ||
2210 | return true; | |
2211 | } | |
2212 | ||
2213 | static bool steal_from_bitmap_to_front(struct btrfs_free_space_ctl *ctl, | |
2214 | struct btrfs_free_space *info, | |
2215 | bool update_stat) | |
2216 | { | |
2217 | struct btrfs_free_space *bitmap; | |
2218 | u64 bitmap_offset; | |
2219 | unsigned long i; | |
2220 | unsigned long j; | |
2221 | unsigned long prev_j; | |
2222 | u64 bytes; | |
2223 | ||
2224 | bitmap_offset = offset_to_bitmap(ctl, info->offset); | |
2225 | /* If we're on a boundary, try the previous logical bitmap. */ | |
2226 | if (bitmap_offset == info->offset) { | |
2227 | if (info->offset == 0) | |
2228 | return false; | |
2229 | bitmap_offset = offset_to_bitmap(ctl, info->offset - 1); | |
2230 | } | |
2231 | ||
2232 | bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0); | |
2233 | if (!bitmap) | |
2234 | return false; | |
2235 | ||
2236 | i = offset_to_bit(bitmap->offset, ctl->unit, info->offset) - 1; | |
2237 | j = 0; | |
2238 | prev_j = (unsigned long)-1; | |
2239 | for_each_clear_bit_from(j, bitmap->bitmap, BITS_PER_BITMAP) { | |
2240 | if (j > i) | |
2241 | break; | |
2242 | prev_j = j; | |
2243 | } | |
2244 | if (prev_j == i) | |
2245 | return false; | |
2246 | ||
2247 | if (prev_j == (unsigned long)-1) | |
2248 | bytes = (i + 1) * ctl->unit; | |
2249 | else | |
2250 | bytes = (i - prev_j) * ctl->unit; | |
2251 | ||
2252 | info->offset -= bytes; | |
2253 | info->bytes += bytes; | |
2254 | ||
2255 | if (update_stat) | |
2256 | bitmap_clear_bits(ctl, bitmap, info->offset, bytes); | |
2257 | else | |
2258 | __bitmap_clear_bits(ctl, bitmap, info->offset, bytes); | |
2259 | ||
2260 | if (!bitmap->bytes) | |
2261 | free_bitmap(ctl, bitmap); | |
2262 | ||
2263 | return true; | |
2264 | } | |
2265 | ||
2266 | /* | |
2267 | * We prefer always to allocate from extent entries, both for clustered and | |
2268 | * non-clustered allocation requests. So when attempting to add a new extent | |
2269 | * entry, try to see if there's adjacent free space in bitmap entries, and if | |
2270 | * there is, migrate that space from the bitmaps to the extent. | |
2271 | * Like this we get better chances of satisfying space allocation requests | |
2272 | * because we attempt to satisfy them based on a single cache entry, and never | |
2273 | * on 2 or more entries - even if the entries represent a contiguous free space | |
2274 | * region (e.g. 1 extent entry + 1 bitmap entry starting where the extent entry | |
2275 | * ends). | |
2276 | */ | |
2277 | static void steal_from_bitmap(struct btrfs_free_space_ctl *ctl, | |
2278 | struct btrfs_free_space *info, | |
2279 | bool update_stat) | |
2280 | { | |
2281 | /* | |
2282 | * Only work with disconnected entries, as we can change their offset, | |
2283 | * and must be extent entries. | |
2284 | */ | |
2285 | ASSERT(!info->bitmap); | |
2286 | ASSERT(RB_EMPTY_NODE(&info->offset_index)); | |
2287 | ||
2288 | if (ctl->total_bitmaps > 0) { | |
2289 | bool stole_end; | |
2290 | bool stole_front = false; | |
2291 | ||
2292 | stole_end = steal_from_bitmap_to_end(ctl, info, update_stat); | |
2293 | if (ctl->total_bitmaps > 0) | |
2294 | stole_front = steal_from_bitmap_to_front(ctl, info, | |
2295 | update_stat); | |
2296 | ||
2297 | if (stole_end || stole_front) | |
2298 | try_merge_free_space(ctl, info, update_stat); | |
2299 | } | |
2300 | } | |
2301 | ||
581bb050 LZ |
2302 | int __btrfs_add_free_space(struct btrfs_free_space_ctl *ctl, |
2303 | u64 offset, u64 bytes) | |
120d66ee LZ |
2304 | { |
2305 | struct btrfs_free_space *info; | |
2306 | int ret = 0; | |
2307 | ||
dc89e982 | 2308 | info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS); |
120d66ee LZ |
2309 | if (!info) |
2310 | return -ENOMEM; | |
2311 | ||
2312 | info->offset = offset; | |
2313 | info->bytes = bytes; | |
20005523 | 2314 | RB_CLEAR_NODE(&info->offset_index); |
120d66ee | 2315 | |
34d52cb6 | 2316 | spin_lock(&ctl->tree_lock); |
120d66ee | 2317 | |
34d52cb6 | 2318 | if (try_merge_free_space(ctl, info, true)) |
120d66ee LZ |
2319 | goto link; |
2320 | ||
2321 | /* | |
2322 | * There was no extent directly to the left or right of this new | |
2323 | * extent then we know we're going to have to allocate a new extent, so | |
2324 | * before we do that see if we need to drop this into a bitmap | |
2325 | */ | |
34d52cb6 | 2326 | ret = insert_into_bitmap(ctl, info); |
120d66ee LZ |
2327 | if (ret < 0) { |
2328 | goto out; | |
2329 | } else if (ret) { | |
2330 | ret = 0; | |
2331 | goto out; | |
2332 | } | |
2333 | link: | |
20005523 FM |
2334 | /* |
2335 | * Only steal free space from adjacent bitmaps if we're sure we're not | |
2336 | * going to add the new free space to existing bitmap entries - because | |
2337 | * that would mean unnecessary work that would be reverted. Therefore | |
2338 | * attempt to steal space from bitmaps if we're adding an extent entry. | |
2339 | */ | |
2340 | steal_from_bitmap(ctl, info, true); | |
2341 | ||
34d52cb6 | 2342 | ret = link_free_space(ctl, info); |
0f9dd46c | 2343 | if (ret) |
dc89e982 | 2344 | kmem_cache_free(btrfs_free_space_cachep, info); |
96303081 | 2345 | out: |
34d52cb6 | 2346 | spin_unlock(&ctl->tree_lock); |
6226cb0a | 2347 | |
0f9dd46c | 2348 | if (ret) { |
62e85577 | 2349 | pr_crit("BTRFS: unable to add free space :%d\n", ret); |
b12d6869 | 2350 | ASSERT(ret != -EEXIST); |
0f9dd46c JB |
2351 | } |
2352 | ||
0f9dd46c JB |
2353 | return ret; |
2354 | } | |
2355 | ||
6226cb0a JB |
2356 | int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group, |
2357 | u64 offset, u64 bytes) | |
0f9dd46c | 2358 | { |
34d52cb6 | 2359 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0f9dd46c | 2360 | struct btrfs_free_space *info; |
b0175117 JB |
2361 | int ret; |
2362 | bool re_search = false; | |
0f9dd46c | 2363 | |
34d52cb6 | 2364 | spin_lock(&ctl->tree_lock); |
6226cb0a | 2365 | |
96303081 | 2366 | again: |
b0175117 | 2367 | ret = 0; |
bdb7d303 JB |
2368 | if (!bytes) |
2369 | goto out_lock; | |
2370 | ||
34d52cb6 | 2371 | info = tree_search_offset(ctl, offset, 0, 0); |
96303081 | 2372 | if (!info) { |
6606bb97 JB |
2373 | /* |
2374 | * oops didn't find an extent that matched the space we wanted | |
2375 | * to remove, look for a bitmap instead | |
2376 | */ | |
34d52cb6 | 2377 | info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), |
6606bb97 JB |
2378 | 1, 0); |
2379 | if (!info) { | |
b0175117 JB |
2380 | /* |
2381 | * If we found a partial bit of our free space in a | |
2382 | * bitmap but then couldn't find the other part this may | |
2383 | * be a problem, so WARN about it. | |
24a70313 | 2384 | */ |
b0175117 | 2385 | WARN_ON(re_search); |
6606bb97 JB |
2386 | goto out_lock; |
2387 | } | |
96303081 JB |
2388 | } |
2389 | ||
b0175117 | 2390 | re_search = false; |
bdb7d303 | 2391 | if (!info->bitmap) { |
34d52cb6 | 2392 | unlink_free_space(ctl, info); |
bdb7d303 JB |
2393 | if (offset == info->offset) { |
2394 | u64 to_free = min(bytes, info->bytes); | |
2395 | ||
2396 | info->bytes -= to_free; | |
2397 | info->offset += to_free; | |
2398 | if (info->bytes) { | |
2399 | ret = link_free_space(ctl, info); | |
2400 | WARN_ON(ret); | |
2401 | } else { | |
2402 | kmem_cache_free(btrfs_free_space_cachep, info); | |
2403 | } | |
0f9dd46c | 2404 | |
bdb7d303 JB |
2405 | offset += to_free; |
2406 | bytes -= to_free; | |
2407 | goto again; | |
2408 | } else { | |
2409 | u64 old_end = info->bytes + info->offset; | |
9b49c9b9 | 2410 | |
bdb7d303 | 2411 | info->bytes = offset - info->offset; |
34d52cb6 | 2412 | ret = link_free_space(ctl, info); |
96303081 JB |
2413 | WARN_ON(ret); |
2414 | if (ret) | |
2415 | goto out_lock; | |
96303081 | 2416 | |
bdb7d303 JB |
2417 | /* Not enough bytes in this entry to satisfy us */ |
2418 | if (old_end < offset + bytes) { | |
2419 | bytes -= old_end - offset; | |
2420 | offset = old_end; | |
2421 | goto again; | |
2422 | } else if (old_end == offset + bytes) { | |
2423 | /* all done */ | |
2424 | goto out_lock; | |
2425 | } | |
2426 | spin_unlock(&ctl->tree_lock); | |
2427 | ||
2428 | ret = btrfs_add_free_space(block_group, offset + bytes, | |
2429 | old_end - (offset + bytes)); | |
2430 | WARN_ON(ret); | |
2431 | goto out; | |
2432 | } | |
0f9dd46c | 2433 | } |
96303081 | 2434 | |
34d52cb6 | 2435 | ret = remove_from_bitmap(ctl, info, &offset, &bytes); |
b0175117 JB |
2436 | if (ret == -EAGAIN) { |
2437 | re_search = true; | |
96303081 | 2438 | goto again; |
b0175117 | 2439 | } |
96303081 | 2440 | out_lock: |
34d52cb6 | 2441 | spin_unlock(&ctl->tree_lock); |
0f9dd46c | 2442 | out: |
25179201 JB |
2443 | return ret; |
2444 | } | |
2445 | ||
0f9dd46c JB |
2446 | void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group, |
2447 | u64 bytes) | |
2448 | { | |
34d52cb6 | 2449 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0f9dd46c JB |
2450 | struct btrfs_free_space *info; |
2451 | struct rb_node *n; | |
2452 | int count = 0; | |
2453 | ||
34d52cb6 | 2454 | for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) { |
0f9dd46c | 2455 | info = rb_entry(n, struct btrfs_free_space, offset_index); |
f6175efa | 2456 | if (info->bytes >= bytes && !block_group->ro) |
0f9dd46c | 2457 | count++; |
efe120a0 FH |
2458 | btrfs_crit(block_group->fs_info, |
2459 | "entry offset %llu, bytes %llu, bitmap %s", | |
2460 | info->offset, info->bytes, | |
96303081 | 2461 | (info->bitmap) ? "yes" : "no"); |
0f9dd46c | 2462 | } |
efe120a0 | 2463 | btrfs_info(block_group->fs_info, "block group has cluster?: %s", |
96303081 | 2464 | list_empty(&block_group->cluster_list) ? "no" : "yes"); |
efe120a0 FH |
2465 | btrfs_info(block_group->fs_info, |
2466 | "%d blocks of free space at or bigger than bytes is", count); | |
0f9dd46c JB |
2467 | } |
2468 | ||
34d52cb6 | 2469 | void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group) |
0f9dd46c | 2470 | { |
34d52cb6 | 2471 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0f9dd46c | 2472 | |
34d52cb6 LZ |
2473 | spin_lock_init(&ctl->tree_lock); |
2474 | ctl->unit = block_group->sectorsize; | |
2475 | ctl->start = block_group->key.objectid; | |
2476 | ctl->private = block_group; | |
2477 | ctl->op = &free_space_op; | |
55507ce3 FM |
2478 | INIT_LIST_HEAD(&ctl->trimming_ranges); |
2479 | mutex_init(&ctl->cache_writeout_mutex); | |
0f9dd46c | 2480 | |
34d52cb6 LZ |
2481 | /* |
2482 | * we only want to have 32k of ram per block group for keeping | |
2483 | * track of free space, and if we pass 1/2 of that we want to | |
2484 | * start converting things over to using bitmaps | |
2485 | */ | |
ee22184b | 2486 | ctl->extents_thresh = (SZ_32K / 2) / sizeof(struct btrfs_free_space); |
0f9dd46c JB |
2487 | } |
2488 | ||
fa9c0d79 CM |
2489 | /* |
2490 | * for a given cluster, put all of its extents back into the free | |
2491 | * space cache. If the block group passed doesn't match the block group | |
2492 | * pointed to by the cluster, someone else raced in and freed the | |
2493 | * cluster already. In that case, we just return without changing anything | |
2494 | */ | |
2495 | static int | |
2496 | __btrfs_return_cluster_to_free_space( | |
2497 | struct btrfs_block_group_cache *block_group, | |
2498 | struct btrfs_free_cluster *cluster) | |
2499 | { | |
34d52cb6 | 2500 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
fa9c0d79 CM |
2501 | struct btrfs_free_space *entry; |
2502 | struct rb_node *node; | |
2503 | ||
2504 | spin_lock(&cluster->lock); | |
2505 | if (cluster->block_group != block_group) | |
2506 | goto out; | |
2507 | ||
96303081 | 2508 | cluster->block_group = NULL; |
fa9c0d79 | 2509 | cluster->window_start = 0; |
96303081 | 2510 | list_del_init(&cluster->block_group_list); |
96303081 | 2511 | |
fa9c0d79 | 2512 | node = rb_first(&cluster->root); |
96303081 | 2513 | while (node) { |
4e69b598 JB |
2514 | bool bitmap; |
2515 | ||
fa9c0d79 CM |
2516 | entry = rb_entry(node, struct btrfs_free_space, offset_index); |
2517 | node = rb_next(&entry->offset_index); | |
2518 | rb_erase(&entry->offset_index, &cluster->root); | |
20005523 | 2519 | RB_CLEAR_NODE(&entry->offset_index); |
4e69b598 JB |
2520 | |
2521 | bitmap = (entry->bitmap != NULL); | |
20005523 | 2522 | if (!bitmap) { |
34d52cb6 | 2523 | try_merge_free_space(ctl, entry, false); |
20005523 FM |
2524 | steal_from_bitmap(ctl, entry, false); |
2525 | } | |
34d52cb6 | 2526 | tree_insert_offset(&ctl->free_space_offset, |
4e69b598 | 2527 | entry->offset, &entry->offset_index, bitmap); |
fa9c0d79 | 2528 | } |
6bef4d31 | 2529 | cluster->root = RB_ROOT; |
96303081 | 2530 | |
fa9c0d79 CM |
2531 | out: |
2532 | spin_unlock(&cluster->lock); | |
96303081 | 2533 | btrfs_put_block_group(block_group); |
fa9c0d79 CM |
2534 | return 0; |
2535 | } | |
2536 | ||
48a3b636 ES |
2537 | static void __btrfs_remove_free_space_cache_locked( |
2538 | struct btrfs_free_space_ctl *ctl) | |
0f9dd46c JB |
2539 | { |
2540 | struct btrfs_free_space *info; | |
2541 | struct rb_node *node; | |
581bb050 | 2542 | |
581bb050 LZ |
2543 | while ((node = rb_last(&ctl->free_space_offset)) != NULL) { |
2544 | info = rb_entry(node, struct btrfs_free_space, offset_index); | |
9b90f513 JB |
2545 | if (!info->bitmap) { |
2546 | unlink_free_space(ctl, info); | |
2547 | kmem_cache_free(btrfs_free_space_cachep, info); | |
2548 | } else { | |
2549 | free_bitmap(ctl, info); | |
2550 | } | |
351810c1 DS |
2551 | |
2552 | cond_resched_lock(&ctl->tree_lock); | |
581bb050 | 2553 | } |
09655373 CM |
2554 | } |
2555 | ||
2556 | void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl) | |
2557 | { | |
2558 | spin_lock(&ctl->tree_lock); | |
2559 | __btrfs_remove_free_space_cache_locked(ctl); | |
581bb050 LZ |
2560 | spin_unlock(&ctl->tree_lock); |
2561 | } | |
2562 | ||
2563 | void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group) | |
2564 | { | |
2565 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
fa9c0d79 | 2566 | struct btrfs_free_cluster *cluster; |
96303081 | 2567 | struct list_head *head; |
0f9dd46c | 2568 | |
34d52cb6 | 2569 | spin_lock(&ctl->tree_lock); |
96303081 JB |
2570 | while ((head = block_group->cluster_list.next) != |
2571 | &block_group->cluster_list) { | |
2572 | cluster = list_entry(head, struct btrfs_free_cluster, | |
2573 | block_group_list); | |
fa9c0d79 CM |
2574 | |
2575 | WARN_ON(cluster->block_group != block_group); | |
2576 | __btrfs_return_cluster_to_free_space(block_group, cluster); | |
351810c1 DS |
2577 | |
2578 | cond_resched_lock(&ctl->tree_lock); | |
fa9c0d79 | 2579 | } |
09655373 | 2580 | __btrfs_remove_free_space_cache_locked(ctl); |
34d52cb6 | 2581 | spin_unlock(&ctl->tree_lock); |
fa9c0d79 | 2582 | |
0f9dd46c JB |
2583 | } |
2584 | ||
6226cb0a | 2585 | u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group, |
a4820398 MX |
2586 | u64 offset, u64 bytes, u64 empty_size, |
2587 | u64 *max_extent_size) | |
0f9dd46c | 2588 | { |
34d52cb6 | 2589 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
6226cb0a | 2590 | struct btrfs_free_space *entry = NULL; |
96303081 | 2591 | u64 bytes_search = bytes + empty_size; |
6226cb0a | 2592 | u64 ret = 0; |
53b381b3 DW |
2593 | u64 align_gap = 0; |
2594 | u64 align_gap_len = 0; | |
0f9dd46c | 2595 | |
34d52cb6 | 2596 | spin_lock(&ctl->tree_lock); |
53b381b3 | 2597 | entry = find_free_space(ctl, &offset, &bytes_search, |
a4820398 | 2598 | block_group->full_stripe_len, max_extent_size); |
6226cb0a | 2599 | if (!entry) |
96303081 JB |
2600 | goto out; |
2601 | ||
2602 | ret = offset; | |
2603 | if (entry->bitmap) { | |
34d52cb6 | 2604 | bitmap_clear_bits(ctl, entry, offset, bytes); |
edf6e2d1 | 2605 | if (!entry->bytes) |
34d52cb6 | 2606 | free_bitmap(ctl, entry); |
96303081 | 2607 | } else { |
34d52cb6 | 2608 | unlink_free_space(ctl, entry); |
53b381b3 DW |
2609 | align_gap_len = offset - entry->offset; |
2610 | align_gap = entry->offset; | |
2611 | ||
2612 | entry->offset = offset + bytes; | |
2613 | WARN_ON(entry->bytes < bytes + align_gap_len); | |
2614 | ||
2615 | entry->bytes -= bytes + align_gap_len; | |
6226cb0a | 2616 | if (!entry->bytes) |
dc89e982 | 2617 | kmem_cache_free(btrfs_free_space_cachep, entry); |
6226cb0a | 2618 | else |
34d52cb6 | 2619 | link_free_space(ctl, entry); |
6226cb0a | 2620 | } |
96303081 | 2621 | out: |
34d52cb6 | 2622 | spin_unlock(&ctl->tree_lock); |
817d52f8 | 2623 | |
53b381b3 DW |
2624 | if (align_gap_len) |
2625 | __btrfs_add_free_space(ctl, align_gap, align_gap_len); | |
0f9dd46c JB |
2626 | return ret; |
2627 | } | |
fa9c0d79 CM |
2628 | |
2629 | /* | |
2630 | * given a cluster, put all of its extents back into the free space | |
2631 | * cache. If a block group is passed, this function will only free | |
2632 | * a cluster that belongs to the passed block group. | |
2633 | * | |
2634 | * Otherwise, it'll get a reference on the block group pointed to by the | |
2635 | * cluster and remove the cluster from it. | |
2636 | */ | |
2637 | int btrfs_return_cluster_to_free_space( | |
2638 | struct btrfs_block_group_cache *block_group, | |
2639 | struct btrfs_free_cluster *cluster) | |
2640 | { | |
34d52cb6 | 2641 | struct btrfs_free_space_ctl *ctl; |
fa9c0d79 CM |
2642 | int ret; |
2643 | ||
2644 | /* first, get a safe pointer to the block group */ | |
2645 | spin_lock(&cluster->lock); | |
2646 | if (!block_group) { | |
2647 | block_group = cluster->block_group; | |
2648 | if (!block_group) { | |
2649 | spin_unlock(&cluster->lock); | |
2650 | return 0; | |
2651 | } | |
2652 | } else if (cluster->block_group != block_group) { | |
2653 | /* someone else has already freed it don't redo their work */ | |
2654 | spin_unlock(&cluster->lock); | |
2655 | return 0; | |
2656 | } | |
2657 | atomic_inc(&block_group->count); | |
2658 | spin_unlock(&cluster->lock); | |
2659 | ||
34d52cb6 LZ |
2660 | ctl = block_group->free_space_ctl; |
2661 | ||
fa9c0d79 | 2662 | /* now return any extents the cluster had on it */ |
34d52cb6 | 2663 | spin_lock(&ctl->tree_lock); |
fa9c0d79 | 2664 | ret = __btrfs_return_cluster_to_free_space(block_group, cluster); |
34d52cb6 | 2665 | spin_unlock(&ctl->tree_lock); |
fa9c0d79 CM |
2666 | |
2667 | /* finally drop our ref */ | |
2668 | btrfs_put_block_group(block_group); | |
2669 | return ret; | |
2670 | } | |
2671 | ||
96303081 JB |
2672 | static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group, |
2673 | struct btrfs_free_cluster *cluster, | |
4e69b598 | 2674 | struct btrfs_free_space *entry, |
a4820398 MX |
2675 | u64 bytes, u64 min_start, |
2676 | u64 *max_extent_size) | |
96303081 | 2677 | { |
34d52cb6 | 2678 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
96303081 JB |
2679 | int err; |
2680 | u64 search_start = cluster->window_start; | |
2681 | u64 search_bytes = bytes; | |
2682 | u64 ret = 0; | |
2683 | ||
96303081 JB |
2684 | search_start = min_start; |
2685 | search_bytes = bytes; | |
2686 | ||
0584f718 | 2687 | err = search_bitmap(ctl, entry, &search_start, &search_bytes, true); |
a4820398 MX |
2688 | if (err) { |
2689 | if (search_bytes > *max_extent_size) | |
2690 | *max_extent_size = search_bytes; | |
4e69b598 | 2691 | return 0; |
a4820398 | 2692 | } |
96303081 JB |
2693 | |
2694 | ret = search_start; | |
bb3ac5a4 | 2695 | __bitmap_clear_bits(ctl, entry, ret, bytes); |
96303081 JB |
2696 | |
2697 | return ret; | |
2698 | } | |
2699 | ||
fa9c0d79 CM |
2700 | /* |
2701 | * given a cluster, try to allocate 'bytes' from it, returns 0 | |
2702 | * if it couldn't find anything suitably large, or a logical disk offset | |
2703 | * if things worked out | |
2704 | */ | |
2705 | u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group, | |
2706 | struct btrfs_free_cluster *cluster, u64 bytes, | |
a4820398 | 2707 | u64 min_start, u64 *max_extent_size) |
fa9c0d79 | 2708 | { |
34d52cb6 | 2709 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
fa9c0d79 CM |
2710 | struct btrfs_free_space *entry = NULL; |
2711 | struct rb_node *node; | |
2712 | u64 ret = 0; | |
2713 | ||
2714 | spin_lock(&cluster->lock); | |
2715 | if (bytes > cluster->max_size) | |
2716 | goto out; | |
2717 | ||
2718 | if (cluster->block_group != block_group) | |
2719 | goto out; | |
2720 | ||
2721 | node = rb_first(&cluster->root); | |
2722 | if (!node) | |
2723 | goto out; | |
2724 | ||
2725 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
67871254 | 2726 | while (1) { |
a4820398 MX |
2727 | if (entry->bytes < bytes && entry->bytes > *max_extent_size) |
2728 | *max_extent_size = entry->bytes; | |
2729 | ||
4e69b598 JB |
2730 | if (entry->bytes < bytes || |
2731 | (!entry->bitmap && entry->offset < min_start)) { | |
fa9c0d79 CM |
2732 | node = rb_next(&entry->offset_index); |
2733 | if (!node) | |
2734 | break; | |
2735 | entry = rb_entry(node, struct btrfs_free_space, | |
2736 | offset_index); | |
2737 | continue; | |
2738 | } | |
fa9c0d79 | 2739 | |
4e69b598 JB |
2740 | if (entry->bitmap) { |
2741 | ret = btrfs_alloc_from_bitmap(block_group, | |
2742 | cluster, entry, bytes, | |
a4820398 MX |
2743 | cluster->window_start, |
2744 | max_extent_size); | |
4e69b598 | 2745 | if (ret == 0) { |
4e69b598 JB |
2746 | node = rb_next(&entry->offset_index); |
2747 | if (!node) | |
2748 | break; | |
2749 | entry = rb_entry(node, struct btrfs_free_space, | |
2750 | offset_index); | |
2751 | continue; | |
2752 | } | |
9b230628 | 2753 | cluster->window_start += bytes; |
4e69b598 | 2754 | } else { |
4e69b598 JB |
2755 | ret = entry->offset; |
2756 | ||
2757 | entry->offset += bytes; | |
2758 | entry->bytes -= bytes; | |
2759 | } | |
fa9c0d79 | 2760 | |
5e71b5d5 | 2761 | if (entry->bytes == 0) |
fa9c0d79 | 2762 | rb_erase(&entry->offset_index, &cluster->root); |
fa9c0d79 CM |
2763 | break; |
2764 | } | |
2765 | out: | |
2766 | spin_unlock(&cluster->lock); | |
96303081 | 2767 | |
5e71b5d5 LZ |
2768 | if (!ret) |
2769 | return 0; | |
2770 | ||
34d52cb6 | 2771 | spin_lock(&ctl->tree_lock); |
5e71b5d5 | 2772 | |
34d52cb6 | 2773 | ctl->free_space -= bytes; |
5e71b5d5 | 2774 | if (entry->bytes == 0) { |
34d52cb6 | 2775 | ctl->free_extents--; |
4e69b598 JB |
2776 | if (entry->bitmap) { |
2777 | kfree(entry->bitmap); | |
34d52cb6 LZ |
2778 | ctl->total_bitmaps--; |
2779 | ctl->op->recalc_thresholds(ctl); | |
4e69b598 | 2780 | } |
dc89e982 | 2781 | kmem_cache_free(btrfs_free_space_cachep, entry); |
5e71b5d5 LZ |
2782 | } |
2783 | ||
34d52cb6 | 2784 | spin_unlock(&ctl->tree_lock); |
5e71b5d5 | 2785 | |
fa9c0d79 CM |
2786 | return ret; |
2787 | } | |
2788 | ||
96303081 JB |
2789 | static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group, |
2790 | struct btrfs_free_space *entry, | |
2791 | struct btrfs_free_cluster *cluster, | |
1bb91902 AO |
2792 | u64 offset, u64 bytes, |
2793 | u64 cont1_bytes, u64 min_bytes) | |
96303081 | 2794 | { |
34d52cb6 | 2795 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
96303081 JB |
2796 | unsigned long next_zero; |
2797 | unsigned long i; | |
1bb91902 AO |
2798 | unsigned long want_bits; |
2799 | unsigned long min_bits; | |
96303081 | 2800 | unsigned long found_bits; |
cef40483 | 2801 | unsigned long max_bits = 0; |
96303081 JB |
2802 | unsigned long start = 0; |
2803 | unsigned long total_found = 0; | |
4e69b598 | 2804 | int ret; |
96303081 | 2805 | |
96009762 | 2806 | i = offset_to_bit(entry->offset, ctl->unit, |
96303081 | 2807 | max_t(u64, offset, entry->offset)); |
96009762 WSH |
2808 | want_bits = bytes_to_bits(bytes, ctl->unit); |
2809 | min_bits = bytes_to_bits(min_bytes, ctl->unit); | |
96303081 | 2810 | |
cef40483 JB |
2811 | /* |
2812 | * Don't bother looking for a cluster in this bitmap if it's heavily | |
2813 | * fragmented. | |
2814 | */ | |
2815 | if (entry->max_extent_size && | |
2816 | entry->max_extent_size < cont1_bytes) | |
2817 | return -ENOSPC; | |
96303081 JB |
2818 | again: |
2819 | found_bits = 0; | |
ebb3dad4 | 2820 | for_each_set_bit_from(i, entry->bitmap, BITS_PER_BITMAP) { |
96303081 JB |
2821 | next_zero = find_next_zero_bit(entry->bitmap, |
2822 | BITS_PER_BITMAP, i); | |
1bb91902 | 2823 | if (next_zero - i >= min_bits) { |
96303081 | 2824 | found_bits = next_zero - i; |
cef40483 JB |
2825 | if (found_bits > max_bits) |
2826 | max_bits = found_bits; | |
96303081 JB |
2827 | break; |
2828 | } | |
cef40483 JB |
2829 | if (next_zero - i > max_bits) |
2830 | max_bits = next_zero - i; | |
96303081 JB |
2831 | i = next_zero; |
2832 | } | |
2833 | ||
cef40483 JB |
2834 | if (!found_bits) { |
2835 | entry->max_extent_size = (u64)max_bits * ctl->unit; | |
4e69b598 | 2836 | return -ENOSPC; |
cef40483 | 2837 | } |
96303081 | 2838 | |
1bb91902 | 2839 | if (!total_found) { |
96303081 | 2840 | start = i; |
b78d09bc | 2841 | cluster->max_size = 0; |
96303081 JB |
2842 | } |
2843 | ||
2844 | total_found += found_bits; | |
2845 | ||
96009762 WSH |
2846 | if (cluster->max_size < found_bits * ctl->unit) |
2847 | cluster->max_size = found_bits * ctl->unit; | |
96303081 | 2848 | |
1bb91902 AO |
2849 | if (total_found < want_bits || cluster->max_size < cont1_bytes) { |
2850 | i = next_zero + 1; | |
96303081 JB |
2851 | goto again; |
2852 | } | |
2853 | ||
96009762 | 2854 | cluster->window_start = start * ctl->unit + entry->offset; |
34d52cb6 | 2855 | rb_erase(&entry->offset_index, &ctl->free_space_offset); |
4e69b598 JB |
2856 | ret = tree_insert_offset(&cluster->root, entry->offset, |
2857 | &entry->offset_index, 1); | |
b12d6869 | 2858 | ASSERT(!ret); /* -EEXIST; Logic error */ |
96303081 | 2859 | |
3f7de037 | 2860 | trace_btrfs_setup_cluster(block_group, cluster, |
96009762 | 2861 | total_found * ctl->unit, 1); |
96303081 JB |
2862 | return 0; |
2863 | } | |
2864 | ||
4e69b598 JB |
2865 | /* |
2866 | * This searches the block group for just extents to fill the cluster with. | |
1bb91902 AO |
2867 | * Try to find a cluster with at least bytes total bytes, at least one |
2868 | * extent of cont1_bytes, and other clusters of at least min_bytes. | |
4e69b598 | 2869 | */ |
3de85bb9 JB |
2870 | static noinline int |
2871 | setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group, | |
2872 | struct btrfs_free_cluster *cluster, | |
2873 | struct list_head *bitmaps, u64 offset, u64 bytes, | |
1bb91902 | 2874 | u64 cont1_bytes, u64 min_bytes) |
4e69b598 | 2875 | { |
34d52cb6 | 2876 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
4e69b598 JB |
2877 | struct btrfs_free_space *first = NULL; |
2878 | struct btrfs_free_space *entry = NULL; | |
4e69b598 JB |
2879 | struct btrfs_free_space *last; |
2880 | struct rb_node *node; | |
4e69b598 JB |
2881 | u64 window_free; |
2882 | u64 max_extent; | |
3f7de037 | 2883 | u64 total_size = 0; |
4e69b598 | 2884 | |
34d52cb6 | 2885 | entry = tree_search_offset(ctl, offset, 0, 1); |
4e69b598 JB |
2886 | if (!entry) |
2887 | return -ENOSPC; | |
2888 | ||
2889 | /* | |
2890 | * We don't want bitmaps, so just move along until we find a normal | |
2891 | * extent entry. | |
2892 | */ | |
1bb91902 AO |
2893 | while (entry->bitmap || entry->bytes < min_bytes) { |
2894 | if (entry->bitmap && list_empty(&entry->list)) | |
86d4a77b | 2895 | list_add_tail(&entry->list, bitmaps); |
4e69b598 JB |
2896 | node = rb_next(&entry->offset_index); |
2897 | if (!node) | |
2898 | return -ENOSPC; | |
2899 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
2900 | } | |
2901 | ||
4e69b598 JB |
2902 | window_free = entry->bytes; |
2903 | max_extent = entry->bytes; | |
2904 | first = entry; | |
2905 | last = entry; | |
4e69b598 | 2906 | |
1bb91902 AO |
2907 | for (node = rb_next(&entry->offset_index); node; |
2908 | node = rb_next(&entry->offset_index)) { | |
4e69b598 JB |
2909 | entry = rb_entry(node, struct btrfs_free_space, offset_index); |
2910 | ||
86d4a77b JB |
2911 | if (entry->bitmap) { |
2912 | if (list_empty(&entry->list)) | |
2913 | list_add_tail(&entry->list, bitmaps); | |
4e69b598 | 2914 | continue; |
86d4a77b JB |
2915 | } |
2916 | ||
1bb91902 AO |
2917 | if (entry->bytes < min_bytes) |
2918 | continue; | |
2919 | ||
2920 | last = entry; | |
2921 | window_free += entry->bytes; | |
2922 | if (entry->bytes > max_extent) | |
4e69b598 | 2923 | max_extent = entry->bytes; |
4e69b598 JB |
2924 | } |
2925 | ||
1bb91902 AO |
2926 | if (window_free < bytes || max_extent < cont1_bytes) |
2927 | return -ENOSPC; | |
2928 | ||
4e69b598 JB |
2929 | cluster->window_start = first->offset; |
2930 | ||
2931 | node = &first->offset_index; | |
2932 | ||
2933 | /* | |
2934 | * now we've found our entries, pull them out of the free space | |
2935 | * cache and put them into the cluster rbtree | |
2936 | */ | |
2937 | do { | |
2938 | int ret; | |
2939 | ||
2940 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
2941 | node = rb_next(&entry->offset_index); | |
1bb91902 | 2942 | if (entry->bitmap || entry->bytes < min_bytes) |
4e69b598 JB |
2943 | continue; |
2944 | ||
34d52cb6 | 2945 | rb_erase(&entry->offset_index, &ctl->free_space_offset); |
4e69b598 JB |
2946 | ret = tree_insert_offset(&cluster->root, entry->offset, |
2947 | &entry->offset_index, 0); | |
3f7de037 | 2948 | total_size += entry->bytes; |
b12d6869 | 2949 | ASSERT(!ret); /* -EEXIST; Logic error */ |
4e69b598 JB |
2950 | } while (node && entry != last); |
2951 | ||
2952 | cluster->max_size = max_extent; | |
3f7de037 | 2953 | trace_btrfs_setup_cluster(block_group, cluster, total_size, 0); |
4e69b598 JB |
2954 | return 0; |
2955 | } | |
2956 | ||
2957 | /* | |
2958 | * This specifically looks for bitmaps that may work in the cluster, we assume | |
2959 | * that we have already failed to find extents that will work. | |
2960 | */ | |
3de85bb9 JB |
2961 | static noinline int |
2962 | setup_cluster_bitmap(struct btrfs_block_group_cache *block_group, | |
2963 | struct btrfs_free_cluster *cluster, | |
2964 | struct list_head *bitmaps, u64 offset, u64 bytes, | |
1bb91902 | 2965 | u64 cont1_bytes, u64 min_bytes) |
4e69b598 | 2966 | { |
34d52cb6 | 2967 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
1b9b922a | 2968 | struct btrfs_free_space *entry = NULL; |
4e69b598 | 2969 | int ret = -ENOSPC; |
0f0fbf1d | 2970 | u64 bitmap_offset = offset_to_bitmap(ctl, offset); |
4e69b598 | 2971 | |
34d52cb6 | 2972 | if (ctl->total_bitmaps == 0) |
4e69b598 JB |
2973 | return -ENOSPC; |
2974 | ||
0f0fbf1d LZ |
2975 | /* |
2976 | * The bitmap that covers offset won't be in the list unless offset | |
2977 | * is just its start offset. | |
2978 | */ | |
1b9b922a CM |
2979 | if (!list_empty(bitmaps)) |
2980 | entry = list_first_entry(bitmaps, struct btrfs_free_space, list); | |
2981 | ||
2982 | if (!entry || entry->offset != bitmap_offset) { | |
0f0fbf1d LZ |
2983 | entry = tree_search_offset(ctl, bitmap_offset, 1, 0); |
2984 | if (entry && list_empty(&entry->list)) | |
2985 | list_add(&entry->list, bitmaps); | |
2986 | } | |
2987 | ||
86d4a77b | 2988 | list_for_each_entry(entry, bitmaps, list) { |
357b9784 | 2989 | if (entry->bytes < bytes) |
86d4a77b JB |
2990 | continue; |
2991 | ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset, | |
1bb91902 | 2992 | bytes, cont1_bytes, min_bytes); |
86d4a77b JB |
2993 | if (!ret) |
2994 | return 0; | |
2995 | } | |
2996 | ||
2997 | /* | |
52621cb6 LZ |
2998 | * The bitmaps list has all the bitmaps that record free space |
2999 | * starting after offset, so no more search is required. | |
86d4a77b | 3000 | */ |
52621cb6 | 3001 | return -ENOSPC; |
4e69b598 JB |
3002 | } |
3003 | ||
fa9c0d79 CM |
3004 | /* |
3005 | * here we try to find a cluster of blocks in a block group. The goal | |
1bb91902 | 3006 | * is to find at least bytes+empty_size. |
fa9c0d79 CM |
3007 | * We might not find them all in one contiguous area. |
3008 | * | |
3009 | * returns zero and sets up cluster if things worked out, otherwise | |
3010 | * it returns -enospc | |
3011 | */ | |
00361589 | 3012 | int btrfs_find_space_cluster(struct btrfs_root *root, |
fa9c0d79 CM |
3013 | struct btrfs_block_group_cache *block_group, |
3014 | struct btrfs_free_cluster *cluster, | |
3015 | u64 offset, u64 bytes, u64 empty_size) | |
3016 | { | |
34d52cb6 | 3017 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
86d4a77b | 3018 | struct btrfs_free_space *entry, *tmp; |
52621cb6 | 3019 | LIST_HEAD(bitmaps); |
fa9c0d79 | 3020 | u64 min_bytes; |
1bb91902 | 3021 | u64 cont1_bytes; |
fa9c0d79 CM |
3022 | int ret; |
3023 | ||
1bb91902 AO |
3024 | /* |
3025 | * Choose the minimum extent size we'll require for this | |
3026 | * cluster. For SSD_SPREAD, don't allow any fragmentation. | |
3027 | * For metadata, allow allocates with smaller extents. For | |
3028 | * data, keep it dense. | |
3029 | */ | |
3cdde224 | 3030 | if (btrfs_test_opt(root->fs_info, SSD_SPREAD)) { |
1bb91902 | 3031 | cont1_bytes = min_bytes = bytes + empty_size; |
451d7585 | 3032 | } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) { |
1bb91902 AO |
3033 | cont1_bytes = bytes; |
3034 | min_bytes = block_group->sectorsize; | |
3035 | } else { | |
3036 | cont1_bytes = max(bytes, (bytes + empty_size) >> 2); | |
3037 | min_bytes = block_group->sectorsize; | |
3038 | } | |
fa9c0d79 | 3039 | |
34d52cb6 | 3040 | spin_lock(&ctl->tree_lock); |
7d0d2e8e JB |
3041 | |
3042 | /* | |
3043 | * If we know we don't have enough space to make a cluster don't even | |
3044 | * bother doing all the work to try and find one. | |
3045 | */ | |
1bb91902 | 3046 | if (ctl->free_space < bytes) { |
34d52cb6 | 3047 | spin_unlock(&ctl->tree_lock); |
7d0d2e8e JB |
3048 | return -ENOSPC; |
3049 | } | |
3050 | ||
fa9c0d79 CM |
3051 | spin_lock(&cluster->lock); |
3052 | ||
3053 | /* someone already found a cluster, hooray */ | |
3054 | if (cluster->block_group) { | |
3055 | ret = 0; | |
3056 | goto out; | |
3057 | } | |
fa9c0d79 | 3058 | |
3f7de037 JB |
3059 | trace_btrfs_find_cluster(block_group, offset, bytes, empty_size, |
3060 | min_bytes); | |
3061 | ||
86d4a77b | 3062 | ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset, |
1bb91902 AO |
3063 | bytes + empty_size, |
3064 | cont1_bytes, min_bytes); | |
4e69b598 | 3065 | if (ret) |
86d4a77b | 3066 | ret = setup_cluster_bitmap(block_group, cluster, &bitmaps, |
1bb91902 AO |
3067 | offset, bytes + empty_size, |
3068 | cont1_bytes, min_bytes); | |
86d4a77b JB |
3069 | |
3070 | /* Clear our temporary list */ | |
3071 | list_for_each_entry_safe(entry, tmp, &bitmaps, list) | |
3072 | list_del_init(&entry->list); | |
fa9c0d79 | 3073 | |
4e69b598 JB |
3074 | if (!ret) { |
3075 | atomic_inc(&block_group->count); | |
3076 | list_add_tail(&cluster->block_group_list, | |
3077 | &block_group->cluster_list); | |
3078 | cluster->block_group = block_group; | |
3f7de037 JB |
3079 | } else { |
3080 | trace_btrfs_failed_cluster_setup(block_group); | |
fa9c0d79 | 3081 | } |
fa9c0d79 CM |
3082 | out: |
3083 | spin_unlock(&cluster->lock); | |
34d52cb6 | 3084 | spin_unlock(&ctl->tree_lock); |
fa9c0d79 CM |
3085 | |
3086 | return ret; | |
3087 | } | |
3088 | ||
3089 | /* | |
3090 | * simple code to zero out a cluster | |
3091 | */ | |
3092 | void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster) | |
3093 | { | |
3094 | spin_lock_init(&cluster->lock); | |
3095 | spin_lock_init(&cluster->refill_lock); | |
6bef4d31 | 3096 | cluster->root = RB_ROOT; |
fa9c0d79 | 3097 | cluster->max_size = 0; |
c759c4e1 | 3098 | cluster->fragmented = false; |
fa9c0d79 CM |
3099 | INIT_LIST_HEAD(&cluster->block_group_list); |
3100 | cluster->block_group = NULL; | |
3101 | } | |
3102 | ||
7fe1e641 LZ |
3103 | static int do_trimming(struct btrfs_block_group_cache *block_group, |
3104 | u64 *total_trimmed, u64 start, u64 bytes, | |
55507ce3 FM |
3105 | u64 reserved_start, u64 reserved_bytes, |
3106 | struct btrfs_trim_range *trim_entry) | |
f7039b1d | 3107 | { |
7fe1e641 | 3108 | struct btrfs_space_info *space_info = block_group->space_info; |
f7039b1d | 3109 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
55507ce3 | 3110 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
7fe1e641 LZ |
3111 | int ret; |
3112 | int update = 0; | |
3113 | u64 trimmed = 0; | |
f7039b1d | 3114 | |
7fe1e641 LZ |
3115 | spin_lock(&space_info->lock); |
3116 | spin_lock(&block_group->lock); | |
3117 | if (!block_group->ro) { | |
3118 | block_group->reserved += reserved_bytes; | |
3119 | space_info->bytes_reserved += reserved_bytes; | |
3120 | update = 1; | |
3121 | } | |
3122 | spin_unlock(&block_group->lock); | |
3123 | spin_unlock(&space_info->lock); | |
3124 | ||
1edb647b FM |
3125 | ret = btrfs_discard_extent(fs_info->extent_root, |
3126 | start, bytes, &trimmed); | |
7fe1e641 LZ |
3127 | if (!ret) |
3128 | *total_trimmed += trimmed; | |
3129 | ||
55507ce3 | 3130 | mutex_lock(&ctl->cache_writeout_mutex); |
7fe1e641 | 3131 | btrfs_add_free_space(block_group, reserved_start, reserved_bytes); |
55507ce3 FM |
3132 | list_del(&trim_entry->list); |
3133 | mutex_unlock(&ctl->cache_writeout_mutex); | |
7fe1e641 LZ |
3134 | |
3135 | if (update) { | |
3136 | spin_lock(&space_info->lock); | |
3137 | spin_lock(&block_group->lock); | |
3138 | if (block_group->ro) | |
3139 | space_info->bytes_readonly += reserved_bytes; | |
3140 | block_group->reserved -= reserved_bytes; | |
3141 | space_info->bytes_reserved -= reserved_bytes; | |
3142 | spin_unlock(&space_info->lock); | |
3143 | spin_unlock(&block_group->lock); | |
3144 | } | |
3145 | ||
3146 | return ret; | |
3147 | } | |
3148 | ||
3149 | static int trim_no_bitmap(struct btrfs_block_group_cache *block_group, | |
3150 | u64 *total_trimmed, u64 start, u64 end, u64 minlen) | |
3151 | { | |
3152 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
3153 | struct btrfs_free_space *entry; | |
3154 | struct rb_node *node; | |
3155 | int ret = 0; | |
3156 | u64 extent_start; | |
3157 | u64 extent_bytes; | |
3158 | u64 bytes; | |
f7039b1d LD |
3159 | |
3160 | while (start < end) { | |
55507ce3 FM |
3161 | struct btrfs_trim_range trim_entry; |
3162 | ||
3163 | mutex_lock(&ctl->cache_writeout_mutex); | |
34d52cb6 | 3164 | spin_lock(&ctl->tree_lock); |
f7039b1d | 3165 | |
34d52cb6 LZ |
3166 | if (ctl->free_space < minlen) { |
3167 | spin_unlock(&ctl->tree_lock); | |
55507ce3 | 3168 | mutex_unlock(&ctl->cache_writeout_mutex); |
f7039b1d LD |
3169 | break; |
3170 | } | |
3171 | ||
34d52cb6 | 3172 | entry = tree_search_offset(ctl, start, 0, 1); |
7fe1e641 | 3173 | if (!entry) { |
34d52cb6 | 3174 | spin_unlock(&ctl->tree_lock); |
55507ce3 | 3175 | mutex_unlock(&ctl->cache_writeout_mutex); |
f7039b1d LD |
3176 | break; |
3177 | } | |
3178 | ||
7fe1e641 LZ |
3179 | /* skip bitmaps */ |
3180 | while (entry->bitmap) { | |
3181 | node = rb_next(&entry->offset_index); | |
3182 | if (!node) { | |
34d52cb6 | 3183 | spin_unlock(&ctl->tree_lock); |
55507ce3 | 3184 | mutex_unlock(&ctl->cache_writeout_mutex); |
7fe1e641 | 3185 | goto out; |
f7039b1d | 3186 | } |
7fe1e641 LZ |
3187 | entry = rb_entry(node, struct btrfs_free_space, |
3188 | offset_index); | |
f7039b1d LD |
3189 | } |
3190 | ||
7fe1e641 LZ |
3191 | if (entry->offset >= end) { |
3192 | spin_unlock(&ctl->tree_lock); | |
55507ce3 | 3193 | mutex_unlock(&ctl->cache_writeout_mutex); |
7fe1e641 | 3194 | break; |
f7039b1d LD |
3195 | } |
3196 | ||
7fe1e641 LZ |
3197 | extent_start = entry->offset; |
3198 | extent_bytes = entry->bytes; | |
3199 | start = max(start, extent_start); | |
3200 | bytes = min(extent_start + extent_bytes, end) - start; | |
3201 | if (bytes < minlen) { | |
3202 | spin_unlock(&ctl->tree_lock); | |
55507ce3 | 3203 | mutex_unlock(&ctl->cache_writeout_mutex); |
7fe1e641 | 3204 | goto next; |
f7039b1d LD |
3205 | } |
3206 | ||
7fe1e641 LZ |
3207 | unlink_free_space(ctl, entry); |
3208 | kmem_cache_free(btrfs_free_space_cachep, entry); | |
3209 | ||
34d52cb6 | 3210 | spin_unlock(&ctl->tree_lock); |
55507ce3 FM |
3211 | trim_entry.start = extent_start; |
3212 | trim_entry.bytes = extent_bytes; | |
3213 | list_add_tail(&trim_entry.list, &ctl->trimming_ranges); | |
3214 | mutex_unlock(&ctl->cache_writeout_mutex); | |
f7039b1d | 3215 | |
7fe1e641 | 3216 | ret = do_trimming(block_group, total_trimmed, start, bytes, |
55507ce3 | 3217 | extent_start, extent_bytes, &trim_entry); |
7fe1e641 LZ |
3218 | if (ret) |
3219 | break; | |
3220 | next: | |
3221 | start += bytes; | |
f7039b1d | 3222 | |
7fe1e641 LZ |
3223 | if (fatal_signal_pending(current)) { |
3224 | ret = -ERESTARTSYS; | |
3225 | break; | |
3226 | } | |
3227 | ||
3228 | cond_resched(); | |
3229 | } | |
3230 | out: | |
3231 | return ret; | |
3232 | } | |
3233 | ||
3234 | static int trim_bitmaps(struct btrfs_block_group_cache *block_group, | |
3235 | u64 *total_trimmed, u64 start, u64 end, u64 minlen) | |
3236 | { | |
3237 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
3238 | struct btrfs_free_space *entry; | |
3239 | int ret = 0; | |
3240 | int ret2; | |
3241 | u64 bytes; | |
3242 | u64 offset = offset_to_bitmap(ctl, start); | |
3243 | ||
3244 | while (offset < end) { | |
3245 | bool next_bitmap = false; | |
55507ce3 | 3246 | struct btrfs_trim_range trim_entry; |
7fe1e641 | 3247 | |
55507ce3 | 3248 | mutex_lock(&ctl->cache_writeout_mutex); |
7fe1e641 LZ |
3249 | spin_lock(&ctl->tree_lock); |
3250 | ||
3251 | if (ctl->free_space < minlen) { | |
3252 | spin_unlock(&ctl->tree_lock); | |
55507ce3 | 3253 | mutex_unlock(&ctl->cache_writeout_mutex); |
7fe1e641 LZ |
3254 | break; |
3255 | } | |
3256 | ||
3257 | entry = tree_search_offset(ctl, offset, 1, 0); | |
3258 | if (!entry) { | |
3259 | spin_unlock(&ctl->tree_lock); | |
55507ce3 | 3260 | mutex_unlock(&ctl->cache_writeout_mutex); |
7fe1e641 LZ |
3261 | next_bitmap = true; |
3262 | goto next; | |
3263 | } | |
3264 | ||
3265 | bytes = minlen; | |
0584f718 | 3266 | ret2 = search_bitmap(ctl, entry, &start, &bytes, false); |
7fe1e641 LZ |
3267 | if (ret2 || start >= end) { |
3268 | spin_unlock(&ctl->tree_lock); | |
55507ce3 | 3269 | mutex_unlock(&ctl->cache_writeout_mutex); |
7fe1e641 LZ |
3270 | next_bitmap = true; |
3271 | goto next; | |
3272 | } | |
3273 | ||
3274 | bytes = min(bytes, end - start); | |
3275 | if (bytes < minlen) { | |
3276 | spin_unlock(&ctl->tree_lock); | |
55507ce3 | 3277 | mutex_unlock(&ctl->cache_writeout_mutex); |
7fe1e641 LZ |
3278 | goto next; |
3279 | } | |
3280 | ||
3281 | bitmap_clear_bits(ctl, entry, start, bytes); | |
3282 | if (entry->bytes == 0) | |
3283 | free_bitmap(ctl, entry); | |
3284 | ||
3285 | spin_unlock(&ctl->tree_lock); | |
55507ce3 FM |
3286 | trim_entry.start = start; |
3287 | trim_entry.bytes = bytes; | |
3288 | list_add_tail(&trim_entry.list, &ctl->trimming_ranges); | |
3289 | mutex_unlock(&ctl->cache_writeout_mutex); | |
7fe1e641 LZ |
3290 | |
3291 | ret = do_trimming(block_group, total_trimmed, start, bytes, | |
55507ce3 | 3292 | start, bytes, &trim_entry); |
7fe1e641 LZ |
3293 | if (ret) |
3294 | break; | |
3295 | next: | |
3296 | if (next_bitmap) { | |
3297 | offset += BITS_PER_BITMAP * ctl->unit; | |
3298 | } else { | |
3299 | start += bytes; | |
3300 | if (start >= offset + BITS_PER_BITMAP * ctl->unit) | |
3301 | offset += BITS_PER_BITMAP * ctl->unit; | |
f7039b1d | 3302 | } |
f7039b1d LD |
3303 | |
3304 | if (fatal_signal_pending(current)) { | |
3305 | ret = -ERESTARTSYS; | |
3306 | break; | |
3307 | } | |
3308 | ||
3309 | cond_resched(); | |
3310 | } | |
3311 | ||
3312 | return ret; | |
3313 | } | |
581bb050 | 3314 | |
e33e17ee | 3315 | void btrfs_get_block_group_trimming(struct btrfs_block_group_cache *cache) |
7fe1e641 | 3316 | { |
e33e17ee JM |
3317 | atomic_inc(&cache->trimming); |
3318 | } | |
7fe1e641 | 3319 | |
e33e17ee JM |
3320 | void btrfs_put_block_group_trimming(struct btrfs_block_group_cache *block_group) |
3321 | { | |
3322 | struct extent_map_tree *em_tree; | |
3323 | struct extent_map *em; | |
3324 | bool cleanup; | |
7fe1e641 | 3325 | |
04216820 | 3326 | spin_lock(&block_group->lock); |
e33e17ee JM |
3327 | cleanup = (atomic_dec_and_test(&block_group->trimming) && |
3328 | block_group->removed); | |
04216820 FM |
3329 | spin_unlock(&block_group->lock); |
3330 | ||
e33e17ee | 3331 | if (cleanup) { |
a1e7e16e | 3332 | lock_chunks(block_group->fs_info->chunk_root); |
04216820 FM |
3333 | em_tree = &block_group->fs_info->mapping_tree.map_tree; |
3334 | write_lock(&em_tree->lock); | |
3335 | em = lookup_extent_mapping(em_tree, block_group->key.objectid, | |
3336 | 1); | |
3337 | BUG_ON(!em); /* logic error, can't happen */ | |
a1e7e16e FM |
3338 | /* |
3339 | * remove_extent_mapping() will delete us from the pinned_chunks | |
3340 | * list, which is protected by the chunk mutex. | |
3341 | */ | |
04216820 FM |
3342 | remove_extent_mapping(em_tree, em); |
3343 | write_unlock(&em_tree->lock); | |
04216820 FM |
3344 | unlock_chunks(block_group->fs_info->chunk_root); |
3345 | ||
3346 | /* once for us and once for the tree */ | |
3347 | free_extent_map(em); | |
3348 | free_extent_map(em); | |
946ddbe8 FM |
3349 | |
3350 | /* | |
3351 | * We've left one free space entry and other tasks trimming | |
3352 | * this block group have left 1 entry each one. Free them. | |
3353 | */ | |
3354 | __btrfs_remove_free_space_cache(block_group->free_space_ctl); | |
e33e17ee JM |
3355 | } |
3356 | } | |
3357 | ||
3358 | int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group, | |
3359 | u64 *trimmed, u64 start, u64 end, u64 minlen) | |
3360 | { | |
3361 | int ret; | |
3362 | ||
3363 | *trimmed = 0; | |
3364 | ||
3365 | spin_lock(&block_group->lock); | |
3366 | if (block_group->removed) { | |
04216820 | 3367 | spin_unlock(&block_group->lock); |
e33e17ee | 3368 | return 0; |
04216820 | 3369 | } |
e33e17ee JM |
3370 | btrfs_get_block_group_trimming(block_group); |
3371 | spin_unlock(&block_group->lock); | |
3372 | ||
3373 | ret = trim_no_bitmap(block_group, trimmed, start, end, minlen); | |
3374 | if (ret) | |
3375 | goto out; | |
7fe1e641 | 3376 | |
e33e17ee JM |
3377 | ret = trim_bitmaps(block_group, trimmed, start, end, minlen); |
3378 | out: | |
3379 | btrfs_put_block_group_trimming(block_group); | |
7fe1e641 LZ |
3380 | return ret; |
3381 | } | |
3382 | ||
581bb050 LZ |
3383 | /* |
3384 | * Find the left-most item in the cache tree, and then return the | |
3385 | * smallest inode number in the item. | |
3386 | * | |
3387 | * Note: the returned inode number may not be the smallest one in | |
3388 | * the tree, if the left-most item is a bitmap. | |
3389 | */ | |
3390 | u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root) | |
3391 | { | |
3392 | struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl; | |
3393 | struct btrfs_free_space *entry = NULL; | |
3394 | u64 ino = 0; | |
3395 | ||
3396 | spin_lock(&ctl->tree_lock); | |
3397 | ||
3398 | if (RB_EMPTY_ROOT(&ctl->free_space_offset)) | |
3399 | goto out; | |
3400 | ||
3401 | entry = rb_entry(rb_first(&ctl->free_space_offset), | |
3402 | struct btrfs_free_space, offset_index); | |
3403 | ||
3404 | if (!entry->bitmap) { | |
3405 | ino = entry->offset; | |
3406 | ||
3407 | unlink_free_space(ctl, entry); | |
3408 | entry->offset++; | |
3409 | entry->bytes--; | |
3410 | if (!entry->bytes) | |
3411 | kmem_cache_free(btrfs_free_space_cachep, entry); | |
3412 | else | |
3413 | link_free_space(ctl, entry); | |
3414 | } else { | |
3415 | u64 offset = 0; | |
3416 | u64 count = 1; | |
3417 | int ret; | |
3418 | ||
0584f718 | 3419 | ret = search_bitmap(ctl, entry, &offset, &count, true); |
79787eaa | 3420 | /* Logic error; Should be empty if it can't find anything */ |
b12d6869 | 3421 | ASSERT(!ret); |
581bb050 LZ |
3422 | |
3423 | ino = offset; | |
3424 | bitmap_clear_bits(ctl, entry, offset, 1); | |
3425 | if (entry->bytes == 0) | |
3426 | free_bitmap(ctl, entry); | |
3427 | } | |
3428 | out: | |
3429 | spin_unlock(&ctl->tree_lock); | |
3430 | ||
3431 | return ino; | |
3432 | } | |
82d5902d LZ |
3433 | |
3434 | struct inode *lookup_free_ino_inode(struct btrfs_root *root, | |
3435 | struct btrfs_path *path) | |
3436 | { | |
3437 | struct inode *inode = NULL; | |
3438 | ||
57cdc8db DS |
3439 | spin_lock(&root->ino_cache_lock); |
3440 | if (root->ino_cache_inode) | |
3441 | inode = igrab(root->ino_cache_inode); | |
3442 | spin_unlock(&root->ino_cache_lock); | |
82d5902d LZ |
3443 | if (inode) |
3444 | return inode; | |
3445 | ||
3446 | inode = __lookup_free_space_inode(root, path, 0); | |
3447 | if (IS_ERR(inode)) | |
3448 | return inode; | |
3449 | ||
57cdc8db | 3450 | spin_lock(&root->ino_cache_lock); |
7841cb28 | 3451 | if (!btrfs_fs_closing(root->fs_info)) |
57cdc8db DS |
3452 | root->ino_cache_inode = igrab(inode); |
3453 | spin_unlock(&root->ino_cache_lock); | |
82d5902d LZ |
3454 | |
3455 | return inode; | |
3456 | } | |
3457 | ||
3458 | int create_free_ino_inode(struct btrfs_root *root, | |
3459 | struct btrfs_trans_handle *trans, | |
3460 | struct btrfs_path *path) | |
3461 | { | |
3462 | return __create_free_space_inode(root, trans, path, | |
3463 | BTRFS_FREE_INO_OBJECTID, 0); | |
3464 | } | |
3465 | ||
3466 | int load_free_ino_cache(struct btrfs_fs_info *fs_info, struct btrfs_root *root) | |
3467 | { | |
3468 | struct btrfs_free_space_ctl *ctl = root->free_ino_ctl; | |
3469 | struct btrfs_path *path; | |
3470 | struct inode *inode; | |
3471 | int ret = 0; | |
3472 | u64 root_gen = btrfs_root_generation(&root->root_item); | |
3473 | ||
3cdde224 | 3474 | if (!btrfs_test_opt(root->fs_info, INODE_MAP_CACHE)) |
4b9465cb CM |
3475 | return 0; |
3476 | ||
82d5902d LZ |
3477 | /* |
3478 | * If we're unmounting then just return, since this does a search on the | |
3479 | * normal root and not the commit root and we could deadlock. | |
3480 | */ | |
7841cb28 | 3481 | if (btrfs_fs_closing(fs_info)) |
82d5902d LZ |
3482 | return 0; |
3483 | ||
3484 | path = btrfs_alloc_path(); | |
3485 | if (!path) | |
3486 | return 0; | |
3487 | ||
3488 | inode = lookup_free_ino_inode(root, path); | |
3489 | if (IS_ERR(inode)) | |
3490 | goto out; | |
3491 | ||
3492 | if (root_gen != BTRFS_I(inode)->generation) | |
3493 | goto out_put; | |
3494 | ||
3495 | ret = __load_free_space_cache(root, inode, ctl, path, 0); | |
3496 | ||
3497 | if (ret < 0) | |
c2cf52eb SK |
3498 | btrfs_err(fs_info, |
3499 | "failed to load free ino cache for root %llu", | |
3500 | root->root_key.objectid); | |
82d5902d LZ |
3501 | out_put: |
3502 | iput(inode); | |
3503 | out: | |
3504 | btrfs_free_path(path); | |
3505 | return ret; | |
3506 | } | |
3507 | ||
3508 | int btrfs_write_out_ino_cache(struct btrfs_root *root, | |
3509 | struct btrfs_trans_handle *trans, | |
53645a91 FDBM |
3510 | struct btrfs_path *path, |
3511 | struct inode *inode) | |
82d5902d LZ |
3512 | { |
3513 | struct btrfs_free_space_ctl *ctl = root->free_ino_ctl; | |
82d5902d | 3514 | int ret; |
c9dc4c65 | 3515 | struct btrfs_io_ctl io_ctl; |
e43699d4 | 3516 | bool release_metadata = true; |
82d5902d | 3517 | |
3cdde224 | 3518 | if (!btrfs_test_opt(root->fs_info, INODE_MAP_CACHE)) |
4b9465cb CM |
3519 | return 0; |
3520 | ||
85db36cf | 3521 | memset(&io_ctl, 0, sizeof(io_ctl)); |
c9dc4c65 | 3522 | ret = __btrfs_write_out_cache(root, inode, ctl, NULL, &io_ctl, |
85db36cf | 3523 | trans, path, 0); |
e43699d4 FM |
3524 | if (!ret) { |
3525 | /* | |
3526 | * At this point writepages() didn't error out, so our metadata | |
3527 | * reservation is released when the writeback finishes, at | |
3528 | * inode.c:btrfs_finish_ordered_io(), regardless of it finishing | |
3529 | * with or without an error. | |
3530 | */ | |
3531 | release_metadata = false; | |
85db36cf | 3532 | ret = btrfs_wait_cache_io(root, trans, NULL, &io_ctl, path, 0); |
e43699d4 | 3533 | } |
85db36cf | 3534 | |
c09544e0 | 3535 | if (ret) { |
e43699d4 FM |
3536 | if (release_metadata) |
3537 | btrfs_delalloc_release_metadata(inode, inode->i_size); | |
c09544e0 | 3538 | #ifdef DEBUG |
c2cf52eb SK |
3539 | btrfs_err(root->fs_info, |
3540 | "failed to write free ino cache for root %llu", | |
3541 | root->root_key.objectid); | |
c09544e0 JB |
3542 | #endif |
3543 | } | |
82d5902d | 3544 | |
82d5902d LZ |
3545 | return ret; |
3546 | } | |
74255aa0 JB |
3547 | |
3548 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS | |
dc11dd5d JB |
3549 | /* |
3550 | * Use this if you need to make a bitmap or extent entry specifically, it | |
3551 | * doesn't do any of the merging that add_free_space does, this acts a lot like | |
3552 | * how the free space cache loading stuff works, so you can get really weird | |
3553 | * configurations. | |
3554 | */ | |
3555 | int test_add_free_space_entry(struct btrfs_block_group_cache *cache, | |
3556 | u64 offset, u64 bytes, bool bitmap) | |
74255aa0 | 3557 | { |
dc11dd5d JB |
3558 | struct btrfs_free_space_ctl *ctl = cache->free_space_ctl; |
3559 | struct btrfs_free_space *info = NULL, *bitmap_info; | |
3560 | void *map = NULL; | |
3561 | u64 bytes_added; | |
3562 | int ret; | |
74255aa0 | 3563 | |
dc11dd5d JB |
3564 | again: |
3565 | if (!info) { | |
3566 | info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS); | |
3567 | if (!info) | |
3568 | return -ENOMEM; | |
74255aa0 JB |
3569 | } |
3570 | ||
dc11dd5d JB |
3571 | if (!bitmap) { |
3572 | spin_lock(&ctl->tree_lock); | |
3573 | info->offset = offset; | |
3574 | info->bytes = bytes; | |
cef40483 | 3575 | info->max_extent_size = 0; |
dc11dd5d JB |
3576 | ret = link_free_space(ctl, info); |
3577 | spin_unlock(&ctl->tree_lock); | |
3578 | if (ret) | |
3579 | kmem_cache_free(btrfs_free_space_cachep, info); | |
3580 | return ret; | |
3581 | } | |
3582 | ||
3583 | if (!map) { | |
09cbfeaf | 3584 | map = kzalloc(PAGE_SIZE, GFP_NOFS); |
dc11dd5d JB |
3585 | if (!map) { |
3586 | kmem_cache_free(btrfs_free_space_cachep, info); | |
3587 | return -ENOMEM; | |
3588 | } | |
3589 | } | |
3590 | ||
3591 | spin_lock(&ctl->tree_lock); | |
3592 | bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), | |
3593 | 1, 0); | |
3594 | if (!bitmap_info) { | |
3595 | info->bitmap = map; | |
3596 | map = NULL; | |
3597 | add_new_bitmap(ctl, info, offset); | |
3598 | bitmap_info = info; | |
20005523 | 3599 | info = NULL; |
dc11dd5d | 3600 | } |
74255aa0 | 3601 | |
dc11dd5d | 3602 | bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes); |
cef40483 | 3603 | |
dc11dd5d JB |
3604 | bytes -= bytes_added; |
3605 | offset += bytes_added; | |
3606 | spin_unlock(&ctl->tree_lock); | |
74255aa0 | 3607 | |
dc11dd5d JB |
3608 | if (bytes) |
3609 | goto again; | |
74255aa0 | 3610 | |
20005523 FM |
3611 | if (info) |
3612 | kmem_cache_free(btrfs_free_space_cachep, info); | |
dc11dd5d JB |
3613 | if (map) |
3614 | kfree(map); | |
3615 | return 0; | |
74255aa0 JB |
3616 | } |
3617 | ||
3618 | /* | |
3619 | * Checks to see if the given range is in the free space cache. This is really | |
3620 | * just used to check the absence of space, so if there is free space in the | |
3621 | * range at all we will return 1. | |
3622 | */ | |
dc11dd5d JB |
3623 | int test_check_exists(struct btrfs_block_group_cache *cache, |
3624 | u64 offset, u64 bytes) | |
74255aa0 JB |
3625 | { |
3626 | struct btrfs_free_space_ctl *ctl = cache->free_space_ctl; | |
3627 | struct btrfs_free_space *info; | |
3628 | int ret = 0; | |
3629 | ||
3630 | spin_lock(&ctl->tree_lock); | |
3631 | info = tree_search_offset(ctl, offset, 0, 0); | |
3632 | if (!info) { | |
3633 | info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), | |
3634 | 1, 0); | |
3635 | if (!info) | |
3636 | goto out; | |
3637 | } | |
3638 | ||
3639 | have_info: | |
3640 | if (info->bitmap) { | |
3641 | u64 bit_off, bit_bytes; | |
3642 | struct rb_node *n; | |
3643 | struct btrfs_free_space *tmp; | |
3644 | ||
3645 | bit_off = offset; | |
3646 | bit_bytes = ctl->unit; | |
0584f718 | 3647 | ret = search_bitmap(ctl, info, &bit_off, &bit_bytes, false); |
74255aa0 JB |
3648 | if (!ret) { |
3649 | if (bit_off == offset) { | |
3650 | ret = 1; | |
3651 | goto out; | |
3652 | } else if (bit_off > offset && | |
3653 | offset + bytes > bit_off) { | |
3654 | ret = 1; | |
3655 | goto out; | |
3656 | } | |
3657 | } | |
3658 | ||
3659 | n = rb_prev(&info->offset_index); | |
3660 | while (n) { | |
3661 | tmp = rb_entry(n, struct btrfs_free_space, | |
3662 | offset_index); | |
3663 | if (tmp->offset + tmp->bytes < offset) | |
3664 | break; | |
3665 | if (offset + bytes < tmp->offset) { | |
5473e0c4 | 3666 | n = rb_prev(&tmp->offset_index); |
74255aa0 JB |
3667 | continue; |
3668 | } | |
3669 | info = tmp; | |
3670 | goto have_info; | |
3671 | } | |
3672 | ||
3673 | n = rb_next(&info->offset_index); | |
3674 | while (n) { | |
3675 | tmp = rb_entry(n, struct btrfs_free_space, | |
3676 | offset_index); | |
3677 | if (offset + bytes < tmp->offset) | |
3678 | break; | |
3679 | if (tmp->offset + tmp->bytes < offset) { | |
5473e0c4 | 3680 | n = rb_next(&tmp->offset_index); |
74255aa0 JB |
3681 | continue; |
3682 | } | |
3683 | info = tmp; | |
3684 | goto have_info; | |
3685 | } | |
3686 | ||
20005523 | 3687 | ret = 0; |
74255aa0 JB |
3688 | goto out; |
3689 | } | |
3690 | ||
3691 | if (info->offset == offset) { | |
3692 | ret = 1; | |
3693 | goto out; | |
3694 | } | |
3695 | ||
3696 | if (offset > info->offset && offset < info->offset + info->bytes) | |
3697 | ret = 1; | |
3698 | out: | |
3699 | spin_unlock(&ctl->tree_lock); | |
3700 | return ret; | |
3701 | } | |
dc11dd5d | 3702 | #endif /* CONFIG_BTRFS_FS_RUN_SANITY_TESTS */ |