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); | |
afdb5718 JM |
45 | static int btrfs_wait_cache_io_root(struct btrfs_root *root, |
46 | struct btrfs_trans_handle *trans, | |
47 | struct btrfs_io_ctl *io_ctl, | |
48 | struct btrfs_path *path); | |
0cb59c99 | 49 | |
0414efae LZ |
50 | static struct inode *__lookup_free_space_inode(struct btrfs_root *root, |
51 | struct btrfs_path *path, | |
52 | u64 offset) | |
0af3d00b | 53 | { |
0b246afa | 54 | struct btrfs_fs_info *fs_info = root->fs_info; |
0af3d00b JB |
55 | struct btrfs_key key; |
56 | struct btrfs_key location; | |
57 | struct btrfs_disk_key disk_key; | |
58 | struct btrfs_free_space_header *header; | |
59 | struct extent_buffer *leaf; | |
60 | struct inode *inode = NULL; | |
61 | int ret; | |
62 | ||
0af3d00b | 63 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; |
0414efae | 64 | key.offset = offset; |
0af3d00b JB |
65 | key.type = 0; |
66 | ||
67 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
68 | if (ret < 0) | |
69 | return ERR_PTR(ret); | |
70 | if (ret > 0) { | |
b3b4aa74 | 71 | btrfs_release_path(path); |
0af3d00b JB |
72 | return ERR_PTR(-ENOENT); |
73 | } | |
74 | ||
75 | leaf = path->nodes[0]; | |
76 | header = btrfs_item_ptr(leaf, path->slots[0], | |
77 | struct btrfs_free_space_header); | |
78 | btrfs_free_space_key(leaf, header, &disk_key); | |
79 | btrfs_disk_key_to_cpu(&location, &disk_key); | |
b3b4aa74 | 80 | btrfs_release_path(path); |
0af3d00b | 81 | |
0b246afa | 82 | inode = btrfs_iget(fs_info->sb, &location, root, NULL); |
0af3d00b JB |
83 | if (IS_ERR(inode)) |
84 | return inode; | |
85 | if (is_bad_inode(inode)) { | |
86 | iput(inode); | |
87 | return ERR_PTR(-ENOENT); | |
88 | } | |
89 | ||
528c0327 | 90 | mapping_set_gfp_mask(inode->i_mapping, |
c62d2555 MH |
91 | mapping_gfp_constraint(inode->i_mapping, |
92 | ~(__GFP_FS | __GFP_HIGHMEM))); | |
adae52b9 | 93 | |
0414efae LZ |
94 | return inode; |
95 | } | |
96 | ||
97 | struct inode *lookup_free_space_inode(struct btrfs_root *root, | |
98 | struct btrfs_block_group_cache | |
99 | *block_group, struct btrfs_path *path) | |
100 | { | |
101 | struct inode *inode = NULL; | |
0b246afa | 102 | struct btrfs_fs_info *fs_info = root->fs_info; |
5b0e95bf | 103 | u32 flags = BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW; |
0414efae LZ |
104 | |
105 | spin_lock(&block_group->lock); | |
106 | if (block_group->inode) | |
107 | inode = igrab(block_group->inode); | |
108 | spin_unlock(&block_group->lock); | |
109 | if (inode) | |
110 | return inode; | |
111 | ||
112 | inode = __lookup_free_space_inode(root, path, | |
113 | block_group->key.objectid); | |
114 | if (IS_ERR(inode)) | |
115 | return inode; | |
116 | ||
0af3d00b | 117 | spin_lock(&block_group->lock); |
5b0e95bf | 118 | if (!((BTRFS_I(inode)->flags & flags) == flags)) { |
0b246afa | 119 | btrfs_info(fs_info, "Old style space inode found, converting."); |
5b0e95bf JB |
120 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM | |
121 | BTRFS_INODE_NODATACOW; | |
2f356126 JB |
122 | block_group->disk_cache_state = BTRFS_DC_CLEAR; |
123 | } | |
124 | ||
300e4f8a | 125 | if (!block_group->iref) { |
0af3d00b JB |
126 | block_group->inode = igrab(inode); |
127 | block_group->iref = 1; | |
128 | } | |
129 | spin_unlock(&block_group->lock); | |
130 | ||
131 | return inode; | |
132 | } | |
133 | ||
48a3b636 ES |
134 | static int __create_free_space_inode(struct btrfs_root *root, |
135 | struct btrfs_trans_handle *trans, | |
136 | struct btrfs_path *path, | |
137 | u64 ino, u64 offset) | |
0af3d00b JB |
138 | { |
139 | struct btrfs_key key; | |
140 | struct btrfs_disk_key disk_key; | |
141 | struct btrfs_free_space_header *header; | |
142 | struct btrfs_inode_item *inode_item; | |
143 | struct extent_buffer *leaf; | |
5b0e95bf | 144 | u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC; |
0af3d00b JB |
145 | int ret; |
146 | ||
0414efae | 147 | ret = btrfs_insert_empty_inode(trans, root, path, ino); |
0af3d00b JB |
148 | if (ret) |
149 | return ret; | |
150 | ||
5b0e95bf JB |
151 | /* We inline crc's for the free disk space cache */ |
152 | if (ino != BTRFS_FREE_INO_OBJECTID) | |
153 | flags |= BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW; | |
154 | ||
0af3d00b JB |
155 | leaf = path->nodes[0]; |
156 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
157 | struct btrfs_inode_item); | |
158 | btrfs_item_key(leaf, &disk_key, path->slots[0]); | |
b159fa28 | 159 | memzero_extent_buffer(leaf, (unsigned long)inode_item, |
0af3d00b JB |
160 | sizeof(*inode_item)); |
161 | btrfs_set_inode_generation(leaf, inode_item, trans->transid); | |
162 | btrfs_set_inode_size(leaf, inode_item, 0); | |
163 | btrfs_set_inode_nbytes(leaf, inode_item, 0); | |
164 | btrfs_set_inode_uid(leaf, inode_item, 0); | |
165 | btrfs_set_inode_gid(leaf, inode_item, 0); | |
166 | btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600); | |
5b0e95bf | 167 | btrfs_set_inode_flags(leaf, inode_item, flags); |
0af3d00b JB |
168 | btrfs_set_inode_nlink(leaf, inode_item, 1); |
169 | btrfs_set_inode_transid(leaf, inode_item, trans->transid); | |
0414efae | 170 | btrfs_set_inode_block_group(leaf, inode_item, offset); |
0af3d00b | 171 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 172 | btrfs_release_path(path); |
0af3d00b JB |
173 | |
174 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
0414efae | 175 | key.offset = offset; |
0af3d00b | 176 | key.type = 0; |
0af3d00b JB |
177 | ret = btrfs_insert_empty_item(trans, root, path, &key, |
178 | sizeof(struct btrfs_free_space_header)); | |
179 | if (ret < 0) { | |
b3b4aa74 | 180 | btrfs_release_path(path); |
0af3d00b JB |
181 | return ret; |
182 | } | |
c9dc4c65 | 183 | |
0af3d00b JB |
184 | leaf = path->nodes[0]; |
185 | header = btrfs_item_ptr(leaf, path->slots[0], | |
186 | struct btrfs_free_space_header); | |
b159fa28 | 187 | memzero_extent_buffer(leaf, (unsigned long)header, sizeof(*header)); |
0af3d00b JB |
188 | btrfs_set_free_space_key(leaf, header, &disk_key); |
189 | btrfs_mark_buffer_dirty(leaf); | |
b3b4aa74 | 190 | btrfs_release_path(path); |
0af3d00b JB |
191 | |
192 | return 0; | |
193 | } | |
194 | ||
0414efae LZ |
195 | int create_free_space_inode(struct btrfs_root *root, |
196 | struct btrfs_trans_handle *trans, | |
197 | struct btrfs_block_group_cache *block_group, | |
198 | struct btrfs_path *path) | |
199 | { | |
200 | int ret; | |
201 | u64 ino; | |
202 | ||
203 | ret = btrfs_find_free_objectid(root, &ino); | |
204 | if (ret < 0) | |
205 | return ret; | |
206 | ||
207 | return __create_free_space_inode(root, trans, path, ino, | |
208 | block_group->key.objectid); | |
209 | } | |
210 | ||
2ff7e61e | 211 | int btrfs_check_trunc_cache_free_space(struct btrfs_fs_info *fs_info, |
7b61cd92 | 212 | struct btrfs_block_rsv *rsv) |
0af3d00b | 213 | { |
c8174313 | 214 | u64 needed_bytes; |
7b61cd92 | 215 | int ret; |
c8174313 JB |
216 | |
217 | /* 1 for slack space, 1 for updating the inode */ | |
0b246afa JM |
218 | needed_bytes = btrfs_calc_trunc_metadata_size(fs_info, 1) + |
219 | btrfs_calc_trans_metadata_size(fs_info, 1); | |
c8174313 | 220 | |
7b61cd92 MX |
221 | spin_lock(&rsv->lock); |
222 | if (rsv->reserved < needed_bytes) | |
223 | ret = -ENOSPC; | |
224 | else | |
225 | ret = 0; | |
226 | spin_unlock(&rsv->lock); | |
4b286cd1 | 227 | return ret; |
7b61cd92 MX |
228 | } |
229 | ||
230 | int btrfs_truncate_free_space_cache(struct btrfs_root *root, | |
231 | struct btrfs_trans_handle *trans, | |
1bbc621e | 232 | struct btrfs_block_group_cache *block_group, |
7b61cd92 MX |
233 | struct inode *inode) |
234 | { | |
7b61cd92 | 235 | int ret = 0; |
1bbc621e | 236 | struct btrfs_path *path = btrfs_alloc_path(); |
35c76642 | 237 | bool locked = false; |
1bbc621e CM |
238 | |
239 | if (!path) { | |
240 | ret = -ENOMEM; | |
241 | goto fail; | |
242 | } | |
243 | ||
244 | if (block_group) { | |
35c76642 | 245 | locked = true; |
1bbc621e CM |
246 | mutex_lock(&trans->transaction->cache_write_mutex); |
247 | if (!list_empty(&block_group->io_list)) { | |
248 | list_del_init(&block_group->io_list); | |
249 | ||
afdb5718 | 250 | btrfs_wait_cache_io(trans, block_group, path); |
1bbc621e CM |
251 | btrfs_put_block_group(block_group); |
252 | } | |
253 | ||
254 | /* | |
255 | * now that we've truncated the cache away, its no longer | |
256 | * setup or written | |
257 | */ | |
258 | spin_lock(&block_group->lock); | |
259 | block_group->disk_cache_state = BTRFS_DC_CLEAR; | |
260 | spin_unlock(&block_group->lock); | |
261 | } | |
262 | btrfs_free_path(path); | |
0af3d00b | 263 | |
0af3d00b | 264 | btrfs_i_size_write(inode, 0); |
7caef267 | 265 | truncate_pagecache(inode, 0); |
0af3d00b JB |
266 | |
267 | /* | |
268 | * We don't need an orphan item because truncating the free space cache | |
269 | * will never be split across transactions. | |
28ed1345 CM |
270 | * We don't need to check for -EAGAIN because we're a free space |
271 | * cache inode | |
0af3d00b JB |
272 | */ |
273 | ret = btrfs_truncate_inode_items(trans, root, inode, | |
274 | 0, BTRFS_EXTENT_DATA_KEY); | |
35c76642 FM |
275 | if (ret) |
276 | goto fail; | |
0af3d00b | 277 | |
82d5902d | 278 | ret = btrfs_update_inode(trans, root, inode); |
1bbc621e | 279 | |
1bbc621e | 280 | fail: |
35c76642 FM |
281 | if (locked) |
282 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
79787eaa | 283 | if (ret) |
66642832 | 284 | btrfs_abort_transaction(trans, ret); |
c8174313 | 285 | |
82d5902d | 286 | return ret; |
0af3d00b JB |
287 | } |
288 | ||
1d480538 | 289 | static void readahead_cache(struct inode *inode) |
9d66e233 JB |
290 | { |
291 | struct file_ra_state *ra; | |
292 | unsigned long last_index; | |
293 | ||
294 | ra = kzalloc(sizeof(*ra), GFP_NOFS); | |
295 | if (!ra) | |
1d480538 | 296 | return; |
9d66e233 JB |
297 | |
298 | file_ra_state_init(ra, inode->i_mapping); | |
09cbfeaf | 299 | last_index = (i_size_read(inode) - 1) >> PAGE_SHIFT; |
9d66e233 JB |
300 | |
301 | page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index); | |
302 | ||
303 | kfree(ra); | |
9d66e233 JB |
304 | } |
305 | ||
4c6d1d85 | 306 | static int io_ctl_init(struct btrfs_io_ctl *io_ctl, struct inode *inode, |
f15376df | 307 | int write) |
a67509c3 | 308 | { |
5349d6c3 MX |
309 | int num_pages; |
310 | int check_crcs = 0; | |
311 | ||
09cbfeaf | 312 | num_pages = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
5349d6c3 | 313 | |
4a0cc7ca | 314 | if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FREE_INO_OBJECTID) |
5349d6c3 MX |
315 | check_crcs = 1; |
316 | ||
317 | /* Make sure we can fit our crcs into the first page */ | |
318 | if (write && check_crcs && | |
09cbfeaf | 319 | (num_pages * sizeof(u32)) >= PAGE_SIZE) |
5349d6c3 MX |
320 | return -ENOSPC; |
321 | ||
4c6d1d85 | 322 | memset(io_ctl, 0, sizeof(struct btrfs_io_ctl)); |
5349d6c3 | 323 | |
31e818fe | 324 | io_ctl->pages = kcalloc(num_pages, sizeof(struct page *), GFP_NOFS); |
a67509c3 JB |
325 | if (!io_ctl->pages) |
326 | return -ENOMEM; | |
5349d6c3 MX |
327 | |
328 | io_ctl->num_pages = num_pages; | |
f15376df | 329 | io_ctl->fs_info = btrfs_sb(inode->i_sb); |
5349d6c3 | 330 | io_ctl->check_crcs = check_crcs; |
c9dc4c65 | 331 | io_ctl->inode = inode; |
5349d6c3 | 332 | |
a67509c3 JB |
333 | return 0; |
334 | } | |
335 | ||
4c6d1d85 | 336 | static void io_ctl_free(struct btrfs_io_ctl *io_ctl) |
a67509c3 JB |
337 | { |
338 | kfree(io_ctl->pages); | |
c9dc4c65 | 339 | io_ctl->pages = NULL; |
a67509c3 JB |
340 | } |
341 | ||
4c6d1d85 | 342 | static void io_ctl_unmap_page(struct btrfs_io_ctl *io_ctl) |
a67509c3 JB |
343 | { |
344 | if (io_ctl->cur) { | |
a67509c3 JB |
345 | io_ctl->cur = NULL; |
346 | io_ctl->orig = NULL; | |
347 | } | |
348 | } | |
349 | ||
4c6d1d85 | 350 | static void io_ctl_map_page(struct btrfs_io_ctl *io_ctl, int clear) |
a67509c3 | 351 | { |
b12d6869 | 352 | ASSERT(io_ctl->index < io_ctl->num_pages); |
a67509c3 | 353 | io_ctl->page = io_ctl->pages[io_ctl->index++]; |
2b108268 | 354 | io_ctl->cur = page_address(io_ctl->page); |
a67509c3 | 355 | io_ctl->orig = io_ctl->cur; |
09cbfeaf | 356 | io_ctl->size = PAGE_SIZE; |
a67509c3 | 357 | if (clear) |
09cbfeaf | 358 | memset(io_ctl->cur, 0, PAGE_SIZE); |
a67509c3 JB |
359 | } |
360 | ||
4c6d1d85 | 361 | static void io_ctl_drop_pages(struct btrfs_io_ctl *io_ctl) |
a67509c3 JB |
362 | { |
363 | int i; | |
364 | ||
365 | io_ctl_unmap_page(io_ctl); | |
366 | ||
367 | for (i = 0; i < io_ctl->num_pages; i++) { | |
a1ee5a45 LZ |
368 | if (io_ctl->pages[i]) { |
369 | ClearPageChecked(io_ctl->pages[i]); | |
370 | unlock_page(io_ctl->pages[i]); | |
09cbfeaf | 371 | put_page(io_ctl->pages[i]); |
a1ee5a45 | 372 | } |
a67509c3 JB |
373 | } |
374 | } | |
375 | ||
4c6d1d85 | 376 | static int io_ctl_prepare_pages(struct btrfs_io_ctl *io_ctl, struct inode *inode, |
a67509c3 JB |
377 | int uptodate) |
378 | { | |
379 | struct page *page; | |
380 | gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); | |
381 | int i; | |
382 | ||
383 | for (i = 0; i < io_ctl->num_pages; i++) { | |
384 | page = find_or_create_page(inode->i_mapping, i, mask); | |
385 | if (!page) { | |
386 | io_ctl_drop_pages(io_ctl); | |
387 | return -ENOMEM; | |
388 | } | |
389 | io_ctl->pages[i] = page; | |
390 | if (uptodate && !PageUptodate(page)) { | |
391 | btrfs_readpage(NULL, page); | |
392 | lock_page(page); | |
393 | if (!PageUptodate(page)) { | |
efe120a0 FH |
394 | btrfs_err(BTRFS_I(inode)->root->fs_info, |
395 | "error reading free space cache"); | |
a67509c3 JB |
396 | io_ctl_drop_pages(io_ctl); |
397 | return -EIO; | |
398 | } | |
399 | } | |
400 | } | |
401 | ||
f7d61dcd JB |
402 | for (i = 0; i < io_ctl->num_pages; i++) { |
403 | clear_page_dirty_for_io(io_ctl->pages[i]); | |
404 | set_page_extent_mapped(io_ctl->pages[i]); | |
405 | } | |
406 | ||
a67509c3 JB |
407 | return 0; |
408 | } | |
409 | ||
4c6d1d85 | 410 | static void io_ctl_set_generation(struct btrfs_io_ctl *io_ctl, u64 generation) |
a67509c3 | 411 | { |
528c0327 | 412 | __le64 *val; |
a67509c3 JB |
413 | |
414 | io_ctl_map_page(io_ctl, 1); | |
415 | ||
416 | /* | |
5b0e95bf JB |
417 | * Skip the csum areas. If we don't check crcs then we just have a |
418 | * 64bit chunk at the front of the first page. | |
a67509c3 | 419 | */ |
5b0e95bf JB |
420 | if (io_ctl->check_crcs) { |
421 | io_ctl->cur += (sizeof(u32) * io_ctl->num_pages); | |
422 | io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages); | |
423 | } else { | |
424 | io_ctl->cur += sizeof(u64); | |
425 | io_ctl->size -= sizeof(u64) * 2; | |
426 | } | |
a67509c3 JB |
427 | |
428 | val = io_ctl->cur; | |
429 | *val = cpu_to_le64(generation); | |
430 | io_ctl->cur += sizeof(u64); | |
a67509c3 JB |
431 | } |
432 | ||
4c6d1d85 | 433 | static int io_ctl_check_generation(struct btrfs_io_ctl *io_ctl, u64 generation) |
a67509c3 | 434 | { |
528c0327 | 435 | __le64 *gen; |
a67509c3 | 436 | |
5b0e95bf JB |
437 | /* |
438 | * Skip the crc area. If we don't check crcs then we just have a 64bit | |
439 | * chunk at the front of the first page. | |
440 | */ | |
441 | if (io_ctl->check_crcs) { | |
442 | io_ctl->cur += sizeof(u32) * io_ctl->num_pages; | |
443 | io_ctl->size -= sizeof(u64) + | |
444 | (sizeof(u32) * io_ctl->num_pages); | |
445 | } else { | |
446 | io_ctl->cur += sizeof(u64); | |
447 | io_ctl->size -= sizeof(u64) * 2; | |
448 | } | |
a67509c3 | 449 | |
a67509c3 JB |
450 | gen = io_ctl->cur; |
451 | if (le64_to_cpu(*gen) != generation) { | |
f15376df | 452 | btrfs_err_rl(io_ctl->fs_info, |
94647322 DS |
453 | "space cache generation (%llu) does not match inode (%llu)", |
454 | *gen, generation); | |
a67509c3 JB |
455 | io_ctl_unmap_page(io_ctl); |
456 | return -EIO; | |
457 | } | |
458 | io_ctl->cur += sizeof(u64); | |
5b0e95bf JB |
459 | return 0; |
460 | } | |
461 | ||
4c6d1d85 | 462 | static void io_ctl_set_crc(struct btrfs_io_ctl *io_ctl, int index) |
5b0e95bf JB |
463 | { |
464 | u32 *tmp; | |
465 | u32 crc = ~(u32)0; | |
466 | unsigned offset = 0; | |
467 | ||
468 | if (!io_ctl->check_crcs) { | |
469 | io_ctl_unmap_page(io_ctl); | |
470 | return; | |
471 | } | |
472 | ||
473 | if (index == 0) | |
cb54f257 | 474 | offset = sizeof(u32) * io_ctl->num_pages; |
5b0e95bf | 475 | |
b0496686 | 476 | crc = btrfs_csum_data(io_ctl->orig + offset, crc, |
09cbfeaf | 477 | PAGE_SIZE - offset); |
0b5e3daf | 478 | btrfs_csum_final(crc, (u8 *)&crc); |
5b0e95bf | 479 | io_ctl_unmap_page(io_ctl); |
2b108268 | 480 | tmp = page_address(io_ctl->pages[0]); |
5b0e95bf JB |
481 | tmp += index; |
482 | *tmp = crc; | |
5b0e95bf JB |
483 | } |
484 | ||
4c6d1d85 | 485 | static int io_ctl_check_crc(struct btrfs_io_ctl *io_ctl, int index) |
5b0e95bf JB |
486 | { |
487 | u32 *tmp, val; | |
488 | u32 crc = ~(u32)0; | |
489 | unsigned offset = 0; | |
490 | ||
491 | if (!io_ctl->check_crcs) { | |
492 | io_ctl_map_page(io_ctl, 0); | |
493 | return 0; | |
494 | } | |
495 | ||
496 | if (index == 0) | |
497 | offset = sizeof(u32) * io_ctl->num_pages; | |
498 | ||
2b108268 | 499 | tmp = page_address(io_ctl->pages[0]); |
5b0e95bf JB |
500 | tmp += index; |
501 | val = *tmp; | |
5b0e95bf JB |
502 | |
503 | io_ctl_map_page(io_ctl, 0); | |
b0496686 | 504 | crc = btrfs_csum_data(io_ctl->orig + offset, crc, |
09cbfeaf | 505 | PAGE_SIZE - offset); |
0b5e3daf | 506 | btrfs_csum_final(crc, (u8 *)&crc); |
5b0e95bf | 507 | if (val != crc) { |
f15376df | 508 | btrfs_err_rl(io_ctl->fs_info, |
94647322 | 509 | "csum mismatch on free space cache"); |
5b0e95bf JB |
510 | io_ctl_unmap_page(io_ctl); |
511 | return -EIO; | |
512 | } | |
513 | ||
a67509c3 JB |
514 | return 0; |
515 | } | |
516 | ||
4c6d1d85 | 517 | static int io_ctl_add_entry(struct btrfs_io_ctl *io_ctl, u64 offset, u64 bytes, |
a67509c3 JB |
518 | void *bitmap) |
519 | { | |
520 | struct btrfs_free_space_entry *entry; | |
521 | ||
522 | if (!io_ctl->cur) | |
523 | return -ENOSPC; | |
524 | ||
525 | entry = io_ctl->cur; | |
526 | entry->offset = cpu_to_le64(offset); | |
527 | entry->bytes = cpu_to_le64(bytes); | |
528 | entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP : | |
529 | BTRFS_FREE_SPACE_EXTENT; | |
530 | io_ctl->cur += sizeof(struct btrfs_free_space_entry); | |
531 | io_ctl->size -= sizeof(struct btrfs_free_space_entry); | |
532 | ||
533 | if (io_ctl->size >= sizeof(struct btrfs_free_space_entry)) | |
534 | return 0; | |
535 | ||
5b0e95bf | 536 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); |
a67509c3 JB |
537 | |
538 | /* No more pages to map */ | |
539 | if (io_ctl->index >= io_ctl->num_pages) | |
540 | return 0; | |
541 | ||
542 | /* map the next page */ | |
543 | io_ctl_map_page(io_ctl, 1); | |
544 | return 0; | |
545 | } | |
546 | ||
4c6d1d85 | 547 | static int io_ctl_add_bitmap(struct btrfs_io_ctl *io_ctl, void *bitmap) |
a67509c3 JB |
548 | { |
549 | if (!io_ctl->cur) | |
550 | return -ENOSPC; | |
551 | ||
552 | /* | |
553 | * If we aren't at the start of the current page, unmap this one and | |
554 | * map the next one if there is any left. | |
555 | */ | |
556 | if (io_ctl->cur != io_ctl->orig) { | |
5b0e95bf | 557 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); |
a67509c3 JB |
558 | if (io_ctl->index >= io_ctl->num_pages) |
559 | return -ENOSPC; | |
560 | io_ctl_map_page(io_ctl, 0); | |
561 | } | |
562 | ||
09cbfeaf | 563 | memcpy(io_ctl->cur, bitmap, PAGE_SIZE); |
5b0e95bf | 564 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); |
a67509c3 JB |
565 | if (io_ctl->index < io_ctl->num_pages) |
566 | io_ctl_map_page(io_ctl, 0); | |
567 | return 0; | |
568 | } | |
569 | ||
4c6d1d85 | 570 | static void io_ctl_zero_remaining_pages(struct btrfs_io_ctl *io_ctl) |
a67509c3 | 571 | { |
5b0e95bf JB |
572 | /* |
573 | * If we're not on the boundary we know we've modified the page and we | |
574 | * need to crc the page. | |
575 | */ | |
576 | if (io_ctl->cur != io_ctl->orig) | |
577 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); | |
578 | else | |
579 | io_ctl_unmap_page(io_ctl); | |
a67509c3 JB |
580 | |
581 | while (io_ctl->index < io_ctl->num_pages) { | |
582 | io_ctl_map_page(io_ctl, 1); | |
5b0e95bf | 583 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); |
a67509c3 JB |
584 | } |
585 | } | |
586 | ||
4c6d1d85 | 587 | static int io_ctl_read_entry(struct btrfs_io_ctl *io_ctl, |
5b0e95bf | 588 | struct btrfs_free_space *entry, u8 *type) |
a67509c3 JB |
589 | { |
590 | struct btrfs_free_space_entry *e; | |
2f120c05 JB |
591 | int ret; |
592 | ||
593 | if (!io_ctl->cur) { | |
594 | ret = io_ctl_check_crc(io_ctl, io_ctl->index); | |
595 | if (ret) | |
596 | return ret; | |
597 | } | |
a67509c3 JB |
598 | |
599 | e = io_ctl->cur; | |
600 | entry->offset = le64_to_cpu(e->offset); | |
601 | entry->bytes = le64_to_cpu(e->bytes); | |
5b0e95bf | 602 | *type = e->type; |
a67509c3 JB |
603 | io_ctl->cur += sizeof(struct btrfs_free_space_entry); |
604 | io_ctl->size -= sizeof(struct btrfs_free_space_entry); | |
605 | ||
606 | if (io_ctl->size >= sizeof(struct btrfs_free_space_entry)) | |
5b0e95bf | 607 | return 0; |
a67509c3 JB |
608 | |
609 | io_ctl_unmap_page(io_ctl); | |
610 | ||
2f120c05 | 611 | return 0; |
a67509c3 JB |
612 | } |
613 | ||
4c6d1d85 | 614 | static int io_ctl_read_bitmap(struct btrfs_io_ctl *io_ctl, |
5b0e95bf | 615 | struct btrfs_free_space *entry) |
a67509c3 | 616 | { |
5b0e95bf JB |
617 | int ret; |
618 | ||
5b0e95bf JB |
619 | ret = io_ctl_check_crc(io_ctl, io_ctl->index); |
620 | if (ret) | |
621 | return ret; | |
622 | ||
09cbfeaf | 623 | memcpy(entry->bitmap, io_ctl->cur, PAGE_SIZE); |
a67509c3 | 624 | io_ctl_unmap_page(io_ctl); |
5b0e95bf JB |
625 | |
626 | return 0; | |
a67509c3 JB |
627 | } |
628 | ||
cd023e7b JB |
629 | /* |
630 | * Since we attach pinned extents after the fact we can have contiguous sections | |
631 | * of free space that are split up in entries. This poses a problem with the | |
632 | * tree logging stuff since it could have allocated across what appears to be 2 | |
633 | * entries since we would have merged the entries when adding the pinned extents | |
634 | * back to the free space cache. So run through the space cache that we just | |
635 | * loaded and merge contiguous entries. This will make the log replay stuff not | |
636 | * blow up and it will make for nicer allocator behavior. | |
637 | */ | |
638 | static void merge_space_tree(struct btrfs_free_space_ctl *ctl) | |
639 | { | |
640 | struct btrfs_free_space *e, *prev = NULL; | |
641 | struct rb_node *n; | |
642 | ||
643 | again: | |
644 | spin_lock(&ctl->tree_lock); | |
645 | for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) { | |
646 | e = rb_entry(n, struct btrfs_free_space, offset_index); | |
647 | if (!prev) | |
648 | goto next; | |
649 | if (e->bitmap || prev->bitmap) | |
650 | goto next; | |
651 | if (prev->offset + prev->bytes == e->offset) { | |
652 | unlink_free_space(ctl, prev); | |
653 | unlink_free_space(ctl, e); | |
654 | prev->bytes += e->bytes; | |
655 | kmem_cache_free(btrfs_free_space_cachep, e); | |
656 | link_free_space(ctl, prev); | |
657 | prev = NULL; | |
658 | spin_unlock(&ctl->tree_lock); | |
659 | goto again; | |
660 | } | |
661 | next: | |
662 | prev = e; | |
663 | } | |
664 | spin_unlock(&ctl->tree_lock); | |
665 | } | |
666 | ||
48a3b636 ES |
667 | static int __load_free_space_cache(struct btrfs_root *root, struct inode *inode, |
668 | struct btrfs_free_space_ctl *ctl, | |
669 | struct btrfs_path *path, u64 offset) | |
9d66e233 | 670 | { |
0b246afa | 671 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
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 | 710 | if (!BTRFS_I(inode)->generation) { |
0b246afa | 711 | btrfs_info(fs_info, |
e570fd27 MX |
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) { |
0b246afa JM |
718 | btrfs_err(fs_info, |
719 | "free space inode generation (%llu) did not match free space cache generation (%llu)", | |
720 | BTRFS_I(inode)->generation, generation); | |
a67509c3 | 721 | return 0; |
9d66e233 JB |
722 | } |
723 | ||
724 | if (!num_entries) | |
a67509c3 | 725 | return 0; |
9d66e233 | 726 | |
f15376df | 727 | ret = io_ctl_init(&io_ctl, inode, 0); |
706efc66 LZ |
728 | if (ret) |
729 | return ret; | |
730 | ||
1d480538 | 731 | readahead_cache(inode); |
9d66e233 | 732 | |
a67509c3 JB |
733 | ret = io_ctl_prepare_pages(&io_ctl, inode, 1); |
734 | if (ret) | |
735 | goto out; | |
9d66e233 | 736 | |
5b0e95bf JB |
737 | ret = io_ctl_check_crc(&io_ctl, 0); |
738 | if (ret) | |
739 | goto free_cache; | |
740 | ||
a67509c3 JB |
741 | ret = io_ctl_check_generation(&io_ctl, generation); |
742 | if (ret) | |
743 | goto free_cache; | |
9d66e233 | 744 | |
a67509c3 JB |
745 | while (num_entries) { |
746 | e = kmem_cache_zalloc(btrfs_free_space_cachep, | |
747 | GFP_NOFS); | |
748 | if (!e) | |
9d66e233 | 749 | goto free_cache; |
9d66e233 | 750 | |
5b0e95bf JB |
751 | ret = io_ctl_read_entry(&io_ctl, e, &type); |
752 | if (ret) { | |
753 | kmem_cache_free(btrfs_free_space_cachep, e); | |
754 | goto free_cache; | |
755 | } | |
756 | ||
a67509c3 JB |
757 | if (!e->bytes) { |
758 | kmem_cache_free(btrfs_free_space_cachep, e); | |
759 | goto free_cache; | |
9d66e233 | 760 | } |
a67509c3 JB |
761 | |
762 | if (type == BTRFS_FREE_SPACE_EXTENT) { | |
763 | spin_lock(&ctl->tree_lock); | |
764 | ret = link_free_space(ctl, e); | |
765 | spin_unlock(&ctl->tree_lock); | |
766 | if (ret) { | |
0b246afa | 767 | btrfs_err(fs_info, |
c2cf52eb | 768 | "Duplicate entries in free space cache, dumping"); |
a67509c3 | 769 | kmem_cache_free(btrfs_free_space_cachep, e); |
9d66e233 JB |
770 | goto free_cache; |
771 | } | |
a67509c3 | 772 | } else { |
b12d6869 | 773 | ASSERT(num_bitmaps); |
a67509c3 | 774 | num_bitmaps--; |
09cbfeaf | 775 | e->bitmap = kzalloc(PAGE_SIZE, GFP_NOFS); |
a67509c3 JB |
776 | if (!e->bitmap) { |
777 | kmem_cache_free( | |
778 | btrfs_free_space_cachep, e); | |
9d66e233 JB |
779 | goto free_cache; |
780 | } | |
a67509c3 JB |
781 | spin_lock(&ctl->tree_lock); |
782 | ret = link_free_space(ctl, e); | |
783 | ctl->total_bitmaps++; | |
784 | ctl->op->recalc_thresholds(ctl); | |
785 | spin_unlock(&ctl->tree_lock); | |
786 | if (ret) { | |
0b246afa | 787 | btrfs_err(fs_info, |
c2cf52eb | 788 | "Duplicate entries in free space cache, dumping"); |
dc89e982 | 789 | kmem_cache_free(btrfs_free_space_cachep, e); |
9d66e233 JB |
790 | goto free_cache; |
791 | } | |
a67509c3 | 792 | list_add_tail(&e->list, &bitmaps); |
9d66e233 JB |
793 | } |
794 | ||
a67509c3 JB |
795 | num_entries--; |
796 | } | |
9d66e233 | 797 | |
2f120c05 JB |
798 | io_ctl_unmap_page(&io_ctl); |
799 | ||
a67509c3 JB |
800 | /* |
801 | * We add the bitmaps at the end of the entries in order that | |
802 | * the bitmap entries are added to the cache. | |
803 | */ | |
804 | list_for_each_entry_safe(e, n, &bitmaps, list) { | |
9d66e233 | 805 | list_del_init(&e->list); |
5b0e95bf JB |
806 | ret = io_ctl_read_bitmap(&io_ctl, e); |
807 | if (ret) | |
808 | goto free_cache; | |
9d66e233 JB |
809 | } |
810 | ||
a67509c3 | 811 | io_ctl_drop_pages(&io_ctl); |
cd023e7b | 812 | merge_space_tree(ctl); |
9d66e233 JB |
813 | ret = 1; |
814 | out: | |
a67509c3 | 815 | io_ctl_free(&io_ctl); |
9d66e233 | 816 | return ret; |
9d66e233 | 817 | free_cache: |
a67509c3 | 818 | io_ctl_drop_pages(&io_ctl); |
0414efae | 819 | __btrfs_remove_free_space_cache(ctl); |
9d66e233 JB |
820 | goto out; |
821 | } | |
822 | ||
0414efae LZ |
823 | int load_free_space_cache(struct btrfs_fs_info *fs_info, |
824 | struct btrfs_block_group_cache *block_group) | |
0cb59c99 | 825 | { |
34d52cb6 | 826 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0414efae LZ |
827 | struct btrfs_root *root = fs_info->tree_root; |
828 | struct inode *inode; | |
829 | struct btrfs_path *path; | |
5b0e95bf | 830 | int ret = 0; |
0414efae LZ |
831 | bool matched; |
832 | u64 used = btrfs_block_group_used(&block_group->item); | |
833 | ||
0414efae LZ |
834 | /* |
835 | * If this block group has been marked to be cleared for one reason or | |
836 | * another then we can't trust the on disk cache, so just return. | |
837 | */ | |
9d66e233 | 838 | spin_lock(&block_group->lock); |
0414efae LZ |
839 | if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) { |
840 | spin_unlock(&block_group->lock); | |
841 | return 0; | |
842 | } | |
9d66e233 | 843 | spin_unlock(&block_group->lock); |
0414efae LZ |
844 | |
845 | path = btrfs_alloc_path(); | |
846 | if (!path) | |
847 | return 0; | |
d53ba474 JB |
848 | path->search_commit_root = 1; |
849 | path->skip_locking = 1; | |
0414efae LZ |
850 | |
851 | inode = lookup_free_space_inode(root, block_group, path); | |
852 | if (IS_ERR(inode)) { | |
853 | btrfs_free_path(path); | |
854 | return 0; | |
855 | } | |
856 | ||
5b0e95bf JB |
857 | /* We may have converted the inode and made the cache invalid. */ |
858 | spin_lock(&block_group->lock); | |
859 | if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) { | |
860 | spin_unlock(&block_group->lock); | |
a7e221e9 | 861 | btrfs_free_path(path); |
5b0e95bf JB |
862 | goto out; |
863 | } | |
864 | spin_unlock(&block_group->lock); | |
865 | ||
0414efae LZ |
866 | ret = __load_free_space_cache(fs_info->tree_root, inode, ctl, |
867 | path, block_group->key.objectid); | |
868 | btrfs_free_path(path); | |
869 | if (ret <= 0) | |
870 | goto out; | |
871 | ||
872 | spin_lock(&ctl->tree_lock); | |
873 | matched = (ctl->free_space == (block_group->key.offset - used - | |
874 | block_group->bytes_super)); | |
875 | spin_unlock(&ctl->tree_lock); | |
876 | ||
877 | if (!matched) { | |
878 | __btrfs_remove_free_space_cache(ctl); | |
5d163e0e JM |
879 | btrfs_warn(fs_info, |
880 | "block group %llu has wrong amount of free space", | |
881 | block_group->key.objectid); | |
0414efae LZ |
882 | ret = -1; |
883 | } | |
884 | out: | |
885 | if (ret < 0) { | |
886 | /* This cache is bogus, make sure it gets cleared */ | |
887 | spin_lock(&block_group->lock); | |
888 | block_group->disk_cache_state = BTRFS_DC_CLEAR; | |
889 | spin_unlock(&block_group->lock); | |
82d5902d | 890 | ret = 0; |
0414efae | 891 | |
5d163e0e JM |
892 | btrfs_warn(fs_info, |
893 | "failed to load free space cache for block group %llu, rebuilding it now", | |
894 | block_group->key.objectid); | |
0414efae LZ |
895 | } |
896 | ||
897 | iput(inode); | |
898 | return ret; | |
9d66e233 JB |
899 | } |
900 | ||
d4452bc5 | 901 | static noinline_for_stack |
4c6d1d85 | 902 | int write_cache_extent_entries(struct btrfs_io_ctl *io_ctl, |
d4452bc5 CM |
903 | struct btrfs_free_space_ctl *ctl, |
904 | struct btrfs_block_group_cache *block_group, | |
905 | int *entries, int *bitmaps, | |
906 | struct list_head *bitmap_list) | |
0cb59c99 | 907 | { |
c09544e0 | 908 | int ret; |
d4452bc5 | 909 | struct btrfs_free_cluster *cluster = NULL; |
1bbc621e | 910 | struct btrfs_free_cluster *cluster_locked = NULL; |
d4452bc5 | 911 | struct rb_node *node = rb_first(&ctl->free_space_offset); |
55507ce3 | 912 | struct btrfs_trim_range *trim_entry; |
be1a12a0 | 913 | |
43be2146 | 914 | /* Get the cluster for this block_group if it exists */ |
d4452bc5 | 915 | if (block_group && !list_empty(&block_group->cluster_list)) { |
43be2146 JB |
916 | cluster = list_entry(block_group->cluster_list.next, |
917 | struct btrfs_free_cluster, | |
918 | block_group_list); | |
d4452bc5 | 919 | } |
43be2146 | 920 | |
f75b130e | 921 | if (!node && cluster) { |
1bbc621e CM |
922 | cluster_locked = cluster; |
923 | spin_lock(&cluster_locked->lock); | |
f75b130e JB |
924 | node = rb_first(&cluster->root); |
925 | cluster = NULL; | |
926 | } | |
927 | ||
a67509c3 JB |
928 | /* Write out the extent entries */ |
929 | while (node) { | |
930 | struct btrfs_free_space *e; | |
0cb59c99 | 931 | |
a67509c3 | 932 | e = rb_entry(node, struct btrfs_free_space, offset_index); |
d4452bc5 | 933 | *entries += 1; |
0cb59c99 | 934 | |
d4452bc5 | 935 | ret = io_ctl_add_entry(io_ctl, e->offset, e->bytes, |
a67509c3 JB |
936 | e->bitmap); |
937 | if (ret) | |
d4452bc5 | 938 | goto fail; |
2f356126 | 939 | |
a67509c3 | 940 | if (e->bitmap) { |
d4452bc5 CM |
941 | list_add_tail(&e->list, bitmap_list); |
942 | *bitmaps += 1; | |
2f356126 | 943 | } |
a67509c3 JB |
944 | node = rb_next(node); |
945 | if (!node && cluster) { | |
946 | node = rb_first(&cluster->root); | |
1bbc621e CM |
947 | cluster_locked = cluster; |
948 | spin_lock(&cluster_locked->lock); | |
a67509c3 | 949 | cluster = NULL; |
43be2146 | 950 | } |
a67509c3 | 951 | } |
1bbc621e CM |
952 | if (cluster_locked) { |
953 | spin_unlock(&cluster_locked->lock); | |
954 | cluster_locked = NULL; | |
955 | } | |
55507ce3 FM |
956 | |
957 | /* | |
958 | * Make sure we don't miss any range that was removed from our rbtree | |
959 | * because trimming is running. Otherwise after a umount+mount (or crash | |
960 | * after committing the transaction) we would leak free space and get | |
961 | * an inconsistent free space cache report from fsck. | |
962 | */ | |
963 | list_for_each_entry(trim_entry, &ctl->trimming_ranges, list) { | |
964 | ret = io_ctl_add_entry(io_ctl, trim_entry->start, | |
965 | trim_entry->bytes, NULL); | |
966 | if (ret) | |
967 | goto fail; | |
968 | *entries += 1; | |
969 | } | |
970 | ||
d4452bc5 CM |
971 | return 0; |
972 | fail: | |
1bbc621e CM |
973 | if (cluster_locked) |
974 | spin_unlock(&cluster_locked->lock); | |
d4452bc5 CM |
975 | return -ENOSPC; |
976 | } | |
977 | ||
978 | static noinline_for_stack int | |
979 | update_cache_item(struct btrfs_trans_handle *trans, | |
980 | struct btrfs_root *root, | |
981 | struct inode *inode, | |
982 | struct btrfs_path *path, u64 offset, | |
983 | int entries, int bitmaps) | |
984 | { | |
985 | struct btrfs_key key; | |
986 | struct btrfs_free_space_header *header; | |
987 | struct extent_buffer *leaf; | |
988 | int ret; | |
989 | ||
990 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
991 | key.offset = offset; | |
992 | key.type = 0; | |
993 | ||
994 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
995 | if (ret < 0) { | |
996 | clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1, | |
997 | EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL, | |
998 | GFP_NOFS); | |
999 | goto fail; | |
1000 | } | |
1001 | leaf = path->nodes[0]; | |
1002 | if (ret > 0) { | |
1003 | struct btrfs_key found_key; | |
1004 | ASSERT(path->slots[0]); | |
1005 | path->slots[0]--; | |
1006 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1007 | if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID || | |
1008 | found_key.offset != offset) { | |
1009 | clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, | |
1010 | inode->i_size - 1, | |
1011 | EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, | |
1012 | NULL, GFP_NOFS); | |
1013 | btrfs_release_path(path); | |
1014 | goto fail; | |
1015 | } | |
1016 | } | |
1017 | ||
1018 | BTRFS_I(inode)->generation = trans->transid; | |
1019 | header = btrfs_item_ptr(leaf, path->slots[0], | |
1020 | struct btrfs_free_space_header); | |
1021 | btrfs_set_free_space_entries(leaf, header, entries); | |
1022 | btrfs_set_free_space_bitmaps(leaf, header, bitmaps); | |
1023 | btrfs_set_free_space_generation(leaf, header, trans->transid); | |
1024 | btrfs_mark_buffer_dirty(leaf); | |
1025 | btrfs_release_path(path); | |
1026 | ||
1027 | return 0; | |
1028 | ||
1029 | fail: | |
1030 | return -1; | |
1031 | } | |
1032 | ||
1033 | static noinline_for_stack int | |
2ff7e61e | 1034 | write_pinned_extent_entries(struct btrfs_fs_info *fs_info, |
5349d6c3 | 1035 | struct btrfs_block_group_cache *block_group, |
4c6d1d85 | 1036 | struct btrfs_io_ctl *io_ctl, |
5349d6c3 | 1037 | int *entries) |
d4452bc5 CM |
1038 | { |
1039 | u64 start, extent_start, extent_end, len; | |
d4452bc5 CM |
1040 | struct extent_io_tree *unpin = NULL; |
1041 | int ret; | |
43be2146 | 1042 | |
5349d6c3 MX |
1043 | if (!block_group) |
1044 | return 0; | |
1045 | ||
a67509c3 JB |
1046 | /* |
1047 | * We want to add any pinned extents to our free space cache | |
1048 | * so we don't leak the space | |
d4452bc5 | 1049 | * |
db804f23 LZ |
1050 | * We shouldn't have switched the pinned extents yet so this is the |
1051 | * right one | |
1052 | */ | |
0b246afa | 1053 | unpin = fs_info->pinned_extents; |
db804f23 | 1054 | |
5349d6c3 | 1055 | start = block_group->key.objectid; |
db804f23 | 1056 | |
5349d6c3 | 1057 | while (start < block_group->key.objectid + block_group->key.offset) { |
db804f23 LZ |
1058 | ret = find_first_extent_bit(unpin, start, |
1059 | &extent_start, &extent_end, | |
e6138876 | 1060 | EXTENT_DIRTY, NULL); |
5349d6c3 MX |
1061 | if (ret) |
1062 | return 0; | |
0cb59c99 | 1063 | |
a67509c3 | 1064 | /* This pinned extent is out of our range */ |
db804f23 | 1065 | if (extent_start >= block_group->key.objectid + |
a67509c3 | 1066 | block_group->key.offset) |
5349d6c3 | 1067 | return 0; |
2f356126 | 1068 | |
db804f23 LZ |
1069 | extent_start = max(extent_start, start); |
1070 | extent_end = min(block_group->key.objectid + | |
1071 | block_group->key.offset, extent_end + 1); | |
1072 | len = extent_end - extent_start; | |
0cb59c99 | 1073 | |
d4452bc5 CM |
1074 | *entries += 1; |
1075 | ret = io_ctl_add_entry(io_ctl, extent_start, len, NULL); | |
a67509c3 | 1076 | if (ret) |
5349d6c3 | 1077 | return -ENOSPC; |
0cb59c99 | 1078 | |
db804f23 | 1079 | start = extent_end; |
a67509c3 | 1080 | } |
0cb59c99 | 1081 | |
5349d6c3 MX |
1082 | return 0; |
1083 | } | |
1084 | ||
1085 | static noinline_for_stack int | |
4c6d1d85 | 1086 | write_bitmap_entries(struct btrfs_io_ctl *io_ctl, struct list_head *bitmap_list) |
5349d6c3 | 1087 | { |
7ae1681e | 1088 | struct btrfs_free_space *entry, *next; |
5349d6c3 MX |
1089 | int ret; |
1090 | ||
0cb59c99 | 1091 | /* Write out the bitmaps */ |
7ae1681e | 1092 | list_for_each_entry_safe(entry, next, bitmap_list, list) { |
d4452bc5 | 1093 | ret = io_ctl_add_bitmap(io_ctl, entry->bitmap); |
a67509c3 | 1094 | if (ret) |
5349d6c3 | 1095 | return -ENOSPC; |
0cb59c99 | 1096 | list_del_init(&entry->list); |
be1a12a0 JB |
1097 | } |
1098 | ||
5349d6c3 MX |
1099 | return 0; |
1100 | } | |
0cb59c99 | 1101 | |
5349d6c3 MX |
1102 | static int flush_dirty_cache(struct inode *inode) |
1103 | { | |
1104 | int ret; | |
be1a12a0 | 1105 | |
0ef8b726 | 1106 | ret = btrfs_wait_ordered_range(inode, 0, (u64)-1); |
5349d6c3 | 1107 | if (ret) |
0ef8b726 JB |
1108 | clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1, |
1109 | EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL, | |
1110 | GFP_NOFS); | |
0cb59c99 | 1111 | |
5349d6c3 | 1112 | return ret; |
d4452bc5 CM |
1113 | } |
1114 | ||
1115 | static void noinline_for_stack | |
a3bdccc4 | 1116 | cleanup_bitmap_list(struct list_head *bitmap_list) |
d4452bc5 | 1117 | { |
7ae1681e | 1118 | struct btrfs_free_space *entry, *next; |
5349d6c3 | 1119 | |
7ae1681e | 1120 | list_for_each_entry_safe(entry, next, bitmap_list, list) |
d4452bc5 | 1121 | list_del_init(&entry->list); |
a3bdccc4 CM |
1122 | } |
1123 | ||
1124 | static void noinline_for_stack | |
1125 | cleanup_write_cache_enospc(struct inode *inode, | |
1126 | struct btrfs_io_ctl *io_ctl, | |
7bf1a159 | 1127 | struct extent_state **cached_state) |
a3bdccc4 | 1128 | { |
d4452bc5 CM |
1129 | io_ctl_drop_pages(io_ctl); |
1130 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0, | |
1131 | i_size_read(inode) - 1, cached_state, | |
1132 | GFP_NOFS); | |
1133 | } | |
549b4fdb | 1134 | |
afdb5718 JM |
1135 | static int __btrfs_wait_cache_io(struct btrfs_root *root, |
1136 | struct btrfs_trans_handle *trans, | |
1137 | struct btrfs_block_group_cache *block_group, | |
1138 | struct btrfs_io_ctl *io_ctl, | |
1139 | struct btrfs_path *path, u64 offset) | |
c9dc4c65 CM |
1140 | { |
1141 | int ret; | |
1142 | struct inode *inode = io_ctl->inode; | |
0b246afa | 1143 | struct btrfs_fs_info *fs_info; |
c9dc4c65 | 1144 | |
1bbc621e CM |
1145 | if (!inode) |
1146 | return 0; | |
1147 | ||
0b246afa JM |
1148 | fs_info = btrfs_sb(inode->i_sb); |
1149 | ||
c9dc4c65 CM |
1150 | /* Flush the dirty pages in the cache file. */ |
1151 | ret = flush_dirty_cache(inode); | |
1152 | if (ret) | |
1153 | goto out; | |
1154 | ||
1155 | /* Update the cache item to tell everyone this cache file is valid. */ | |
1156 | ret = update_cache_item(trans, root, inode, path, offset, | |
1157 | io_ctl->entries, io_ctl->bitmaps); | |
1158 | out: | |
1159 | io_ctl_free(io_ctl); | |
1160 | if (ret) { | |
1161 | invalidate_inode_pages2(inode->i_mapping); | |
1162 | BTRFS_I(inode)->generation = 0; | |
1163 | if (block_group) { | |
1164 | #ifdef DEBUG | |
0b246afa JM |
1165 | btrfs_err(fs_info, |
1166 | "failed to write free space cache for block group %llu", | |
1167 | block_group->key.objectid); | |
c9dc4c65 CM |
1168 | #endif |
1169 | } | |
1170 | } | |
1171 | btrfs_update_inode(trans, root, inode); | |
1172 | ||
1173 | if (block_group) { | |
1bbc621e CM |
1174 | /* the dirty list is protected by the dirty_bgs_lock */ |
1175 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
1176 | ||
1177 | /* the disk_cache_state is protected by the block group lock */ | |
c9dc4c65 CM |
1178 | spin_lock(&block_group->lock); |
1179 | ||
1180 | /* | |
1181 | * only mark this as written if we didn't get put back on | |
1bbc621e CM |
1182 | * the dirty list while waiting for IO. Otherwise our |
1183 | * cache state won't be right, and we won't get written again | |
c9dc4c65 CM |
1184 | */ |
1185 | if (!ret && list_empty(&block_group->dirty_list)) | |
1186 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
1187 | else if (ret) | |
1188 | block_group->disk_cache_state = BTRFS_DC_ERROR; | |
1189 | ||
1190 | spin_unlock(&block_group->lock); | |
1bbc621e | 1191 | spin_unlock(&trans->transaction->dirty_bgs_lock); |
c9dc4c65 CM |
1192 | io_ctl->inode = NULL; |
1193 | iput(inode); | |
1194 | } | |
1195 | ||
1196 | return ret; | |
1197 | ||
1198 | } | |
1199 | ||
afdb5718 JM |
1200 | static int btrfs_wait_cache_io_root(struct btrfs_root *root, |
1201 | struct btrfs_trans_handle *trans, | |
1202 | struct btrfs_io_ctl *io_ctl, | |
1203 | struct btrfs_path *path) | |
1204 | { | |
1205 | return __btrfs_wait_cache_io(root, trans, NULL, io_ctl, path, 0); | |
1206 | } | |
1207 | ||
1208 | int btrfs_wait_cache_io(struct btrfs_trans_handle *trans, | |
1209 | struct btrfs_block_group_cache *block_group, | |
1210 | struct btrfs_path *path) | |
1211 | { | |
1212 | return __btrfs_wait_cache_io(block_group->fs_info->tree_root, trans, | |
1213 | block_group, &block_group->io_ctl, | |
1214 | path, block_group->key.objectid); | |
1215 | } | |
1216 | ||
d4452bc5 CM |
1217 | /** |
1218 | * __btrfs_write_out_cache - write out cached info to an inode | |
1219 | * @root - the root the inode belongs to | |
1220 | * @ctl - the free space cache we are going to write out | |
1221 | * @block_group - the block_group for this cache if it belongs to a block_group | |
1222 | * @trans - the trans handle | |
d4452bc5 CM |
1223 | * |
1224 | * This function writes out a free space cache struct to disk for quick recovery | |
8cd1e731 | 1225 | * on mount. This will return 0 if it was successful in writing the cache out, |
b8605454 | 1226 | * or an errno if it was not. |
d4452bc5 CM |
1227 | */ |
1228 | static int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode, | |
1229 | struct btrfs_free_space_ctl *ctl, | |
1230 | struct btrfs_block_group_cache *block_group, | |
c9dc4c65 | 1231 | struct btrfs_io_ctl *io_ctl, |
0e8d931a | 1232 | struct btrfs_trans_handle *trans) |
d4452bc5 | 1233 | { |
2ff7e61e | 1234 | struct btrfs_fs_info *fs_info = root->fs_info; |
d4452bc5 | 1235 | struct extent_state *cached_state = NULL; |
5349d6c3 | 1236 | LIST_HEAD(bitmap_list); |
d4452bc5 CM |
1237 | int entries = 0; |
1238 | int bitmaps = 0; | |
1239 | int ret; | |
c9dc4c65 | 1240 | int must_iput = 0; |
d4452bc5 CM |
1241 | |
1242 | if (!i_size_read(inode)) | |
b8605454 | 1243 | return -EIO; |
d4452bc5 | 1244 | |
c9dc4c65 | 1245 | WARN_ON(io_ctl->pages); |
f15376df | 1246 | ret = io_ctl_init(io_ctl, inode, 1); |
d4452bc5 | 1247 | if (ret) |
b8605454 | 1248 | return ret; |
d4452bc5 | 1249 | |
e570fd27 MX |
1250 | if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA)) { |
1251 | down_write(&block_group->data_rwsem); | |
1252 | spin_lock(&block_group->lock); | |
1253 | if (block_group->delalloc_bytes) { | |
1254 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
1255 | spin_unlock(&block_group->lock); | |
1256 | up_write(&block_group->data_rwsem); | |
1257 | BTRFS_I(inode)->generation = 0; | |
1258 | ret = 0; | |
c9dc4c65 | 1259 | must_iput = 1; |
e570fd27 MX |
1260 | goto out; |
1261 | } | |
1262 | spin_unlock(&block_group->lock); | |
1263 | } | |
1264 | ||
d4452bc5 | 1265 | /* Lock all pages first so we can lock the extent safely. */ |
b8605454 OS |
1266 | ret = io_ctl_prepare_pages(io_ctl, inode, 0); |
1267 | if (ret) | |
1268 | goto out; | |
d4452bc5 CM |
1269 | |
1270 | lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1, | |
ff13db41 | 1271 | &cached_state); |
d4452bc5 | 1272 | |
c9dc4c65 | 1273 | io_ctl_set_generation(io_ctl, trans->transid); |
d4452bc5 | 1274 | |
55507ce3 | 1275 | mutex_lock(&ctl->cache_writeout_mutex); |
5349d6c3 | 1276 | /* Write out the extent entries in the free space cache */ |
1bbc621e | 1277 | spin_lock(&ctl->tree_lock); |
c9dc4c65 | 1278 | ret = write_cache_extent_entries(io_ctl, ctl, |
d4452bc5 CM |
1279 | block_group, &entries, &bitmaps, |
1280 | &bitmap_list); | |
a3bdccc4 CM |
1281 | if (ret) |
1282 | goto out_nospc_locked; | |
d4452bc5 | 1283 | |
5349d6c3 MX |
1284 | /* |
1285 | * Some spaces that are freed in the current transaction are pinned, | |
1286 | * they will be added into free space cache after the transaction is | |
1287 | * committed, we shouldn't lose them. | |
1bbc621e CM |
1288 | * |
1289 | * If this changes while we are working we'll get added back to | |
1290 | * the dirty list and redo it. No locking needed | |
5349d6c3 | 1291 | */ |
2ff7e61e JM |
1292 | ret = write_pinned_extent_entries(fs_info, block_group, |
1293 | io_ctl, &entries); | |
a3bdccc4 CM |
1294 | if (ret) |
1295 | goto out_nospc_locked; | |
5349d6c3 | 1296 | |
55507ce3 FM |
1297 | /* |
1298 | * At last, we write out all the bitmaps and keep cache_writeout_mutex | |
1299 | * locked while doing it because a concurrent trim can be manipulating | |
1300 | * or freeing the bitmap. | |
1301 | */ | |
c9dc4c65 | 1302 | ret = write_bitmap_entries(io_ctl, &bitmap_list); |
1bbc621e | 1303 | spin_unlock(&ctl->tree_lock); |
55507ce3 | 1304 | mutex_unlock(&ctl->cache_writeout_mutex); |
5349d6c3 MX |
1305 | if (ret) |
1306 | goto out_nospc; | |
1307 | ||
1308 | /* Zero out the rest of the pages just to make sure */ | |
c9dc4c65 | 1309 | io_ctl_zero_remaining_pages(io_ctl); |
d4452bc5 | 1310 | |
5349d6c3 | 1311 | /* Everything is written out, now we dirty the pages in the file. */ |
2ff7e61e JM |
1312 | ret = btrfs_dirty_pages(inode, io_ctl->pages, io_ctl->num_pages, 0, |
1313 | i_size_read(inode), &cached_state); | |
5349d6c3 | 1314 | if (ret) |
d4452bc5 | 1315 | goto out_nospc; |
5349d6c3 | 1316 | |
e570fd27 MX |
1317 | if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA)) |
1318 | up_write(&block_group->data_rwsem); | |
5349d6c3 MX |
1319 | /* |
1320 | * Release the pages and unlock the extent, we will flush | |
1321 | * them out later | |
1322 | */ | |
c9dc4c65 | 1323 | io_ctl_drop_pages(io_ctl); |
5349d6c3 MX |
1324 | |
1325 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0, | |
1326 | i_size_read(inode) - 1, &cached_state, GFP_NOFS); | |
1327 | ||
c9dc4c65 CM |
1328 | /* |
1329 | * at this point the pages are under IO and we're happy, | |
1330 | * The caller is responsible for waiting on them and updating the | |
1331 | * the cache and the inode | |
1332 | */ | |
1333 | io_ctl->entries = entries; | |
1334 | io_ctl->bitmaps = bitmaps; | |
1335 | ||
1336 | ret = btrfs_fdatawrite_range(inode, 0, (u64)-1); | |
5349d6c3 | 1337 | if (ret) |
d4452bc5 CM |
1338 | goto out; |
1339 | ||
c9dc4c65 CM |
1340 | return 0; |
1341 | ||
2f356126 | 1342 | out: |
c9dc4c65 CM |
1343 | io_ctl->inode = NULL; |
1344 | io_ctl_free(io_ctl); | |
5349d6c3 | 1345 | if (ret) { |
a67509c3 | 1346 | invalidate_inode_pages2(inode->i_mapping); |
0cb59c99 JB |
1347 | BTRFS_I(inode)->generation = 0; |
1348 | } | |
0cb59c99 | 1349 | btrfs_update_inode(trans, root, inode); |
c9dc4c65 CM |
1350 | if (must_iput) |
1351 | iput(inode); | |
5349d6c3 | 1352 | return ret; |
a67509c3 | 1353 | |
a3bdccc4 CM |
1354 | out_nospc_locked: |
1355 | cleanup_bitmap_list(&bitmap_list); | |
1356 | spin_unlock(&ctl->tree_lock); | |
1357 | mutex_unlock(&ctl->cache_writeout_mutex); | |
1358 | ||
a67509c3 | 1359 | out_nospc: |
7bf1a159 | 1360 | cleanup_write_cache_enospc(inode, io_ctl, &cached_state); |
e570fd27 MX |
1361 | |
1362 | if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA)) | |
1363 | up_write(&block_group->data_rwsem); | |
1364 | ||
a67509c3 | 1365 | goto out; |
0414efae LZ |
1366 | } |
1367 | ||
5b4aacef | 1368 | int btrfs_write_out_cache(struct btrfs_fs_info *fs_info, |
0414efae LZ |
1369 | struct btrfs_trans_handle *trans, |
1370 | struct btrfs_block_group_cache *block_group, | |
1371 | struct btrfs_path *path) | |
1372 | { | |
5b4aacef | 1373 | struct btrfs_root *root = fs_info->tree_root; |
0414efae LZ |
1374 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
1375 | struct inode *inode; | |
1376 | int ret = 0; | |
1377 | ||
0414efae LZ |
1378 | spin_lock(&block_group->lock); |
1379 | if (block_group->disk_cache_state < BTRFS_DC_SETUP) { | |
1380 | spin_unlock(&block_group->lock); | |
e570fd27 MX |
1381 | return 0; |
1382 | } | |
0414efae LZ |
1383 | spin_unlock(&block_group->lock); |
1384 | ||
1385 | inode = lookup_free_space_inode(root, block_group, path); | |
1386 | if (IS_ERR(inode)) | |
1387 | return 0; | |
1388 | ||
c9dc4c65 | 1389 | ret = __btrfs_write_out_cache(root, inode, ctl, block_group, |
0e8d931a | 1390 | &block_group->io_ctl, trans); |
c09544e0 | 1391 | if (ret) { |
c09544e0 | 1392 | #ifdef DEBUG |
0b246afa JM |
1393 | btrfs_err(fs_info, |
1394 | "failed to write free space cache for block group %llu", | |
1395 | block_group->key.objectid); | |
c09544e0 | 1396 | #endif |
c9dc4c65 CM |
1397 | spin_lock(&block_group->lock); |
1398 | block_group->disk_cache_state = BTRFS_DC_ERROR; | |
1399 | spin_unlock(&block_group->lock); | |
1400 | ||
1401 | block_group->io_ctl.inode = NULL; | |
1402 | iput(inode); | |
0414efae LZ |
1403 | } |
1404 | ||
c9dc4c65 CM |
1405 | /* |
1406 | * if ret == 0 the caller is expected to call btrfs_wait_cache_io | |
1407 | * to wait for IO and put the inode | |
1408 | */ | |
1409 | ||
0cb59c99 JB |
1410 | return ret; |
1411 | } | |
1412 | ||
34d52cb6 | 1413 | static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit, |
96303081 | 1414 | u64 offset) |
0f9dd46c | 1415 | { |
b12d6869 | 1416 | ASSERT(offset >= bitmap_start); |
96303081 | 1417 | offset -= bitmap_start; |
34d52cb6 | 1418 | return (unsigned long)(div_u64(offset, unit)); |
96303081 | 1419 | } |
0f9dd46c | 1420 | |
34d52cb6 | 1421 | static inline unsigned long bytes_to_bits(u64 bytes, u32 unit) |
96303081 | 1422 | { |
34d52cb6 | 1423 | return (unsigned long)(div_u64(bytes, unit)); |
96303081 | 1424 | } |
0f9dd46c | 1425 | |
34d52cb6 | 1426 | static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1427 | u64 offset) |
1428 | { | |
1429 | u64 bitmap_start; | |
0ef6447a | 1430 | u64 bytes_per_bitmap; |
0f9dd46c | 1431 | |
34d52cb6 LZ |
1432 | bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit; |
1433 | bitmap_start = offset - ctl->start; | |
0ef6447a | 1434 | bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap); |
96303081 | 1435 | bitmap_start *= bytes_per_bitmap; |
34d52cb6 | 1436 | bitmap_start += ctl->start; |
0f9dd46c | 1437 | |
96303081 | 1438 | return bitmap_start; |
0f9dd46c JB |
1439 | } |
1440 | ||
96303081 JB |
1441 | static int tree_insert_offset(struct rb_root *root, u64 offset, |
1442 | struct rb_node *node, int bitmap) | |
0f9dd46c JB |
1443 | { |
1444 | struct rb_node **p = &root->rb_node; | |
1445 | struct rb_node *parent = NULL; | |
1446 | struct btrfs_free_space *info; | |
1447 | ||
1448 | while (*p) { | |
1449 | parent = *p; | |
96303081 | 1450 | info = rb_entry(parent, struct btrfs_free_space, offset_index); |
0f9dd46c | 1451 | |
96303081 | 1452 | if (offset < info->offset) { |
0f9dd46c | 1453 | p = &(*p)->rb_left; |
96303081 | 1454 | } else if (offset > info->offset) { |
0f9dd46c | 1455 | p = &(*p)->rb_right; |
96303081 JB |
1456 | } else { |
1457 | /* | |
1458 | * we could have a bitmap entry and an extent entry | |
1459 | * share the same offset. If this is the case, we want | |
1460 | * the extent entry to always be found first if we do a | |
1461 | * linear search through the tree, since we want to have | |
1462 | * the quickest allocation time, and allocating from an | |
1463 | * extent is faster than allocating from a bitmap. So | |
1464 | * if we're inserting a bitmap and we find an entry at | |
1465 | * this offset, we want to go right, or after this entry | |
1466 | * logically. If we are inserting an extent and we've | |
1467 | * found a bitmap, we want to go left, or before | |
1468 | * logically. | |
1469 | */ | |
1470 | if (bitmap) { | |
207dde82 JB |
1471 | if (info->bitmap) { |
1472 | WARN_ON_ONCE(1); | |
1473 | return -EEXIST; | |
1474 | } | |
96303081 JB |
1475 | p = &(*p)->rb_right; |
1476 | } else { | |
207dde82 JB |
1477 | if (!info->bitmap) { |
1478 | WARN_ON_ONCE(1); | |
1479 | return -EEXIST; | |
1480 | } | |
96303081 JB |
1481 | p = &(*p)->rb_left; |
1482 | } | |
1483 | } | |
0f9dd46c JB |
1484 | } |
1485 | ||
1486 | rb_link_node(node, parent, p); | |
1487 | rb_insert_color(node, root); | |
1488 | ||
1489 | return 0; | |
1490 | } | |
1491 | ||
1492 | /* | |
70cb0743 JB |
1493 | * searches the tree for the given offset. |
1494 | * | |
96303081 JB |
1495 | * fuzzy - If this is set, then we are trying to make an allocation, and we just |
1496 | * want a section that has at least bytes size and comes at or after the given | |
1497 | * offset. | |
0f9dd46c | 1498 | */ |
96303081 | 1499 | static struct btrfs_free_space * |
34d52cb6 | 1500 | tree_search_offset(struct btrfs_free_space_ctl *ctl, |
96303081 | 1501 | u64 offset, int bitmap_only, int fuzzy) |
0f9dd46c | 1502 | { |
34d52cb6 | 1503 | struct rb_node *n = ctl->free_space_offset.rb_node; |
96303081 JB |
1504 | struct btrfs_free_space *entry, *prev = NULL; |
1505 | ||
1506 | /* find entry that is closest to the 'offset' */ | |
1507 | while (1) { | |
1508 | if (!n) { | |
1509 | entry = NULL; | |
1510 | break; | |
1511 | } | |
0f9dd46c | 1512 | |
0f9dd46c | 1513 | entry = rb_entry(n, struct btrfs_free_space, offset_index); |
96303081 | 1514 | prev = entry; |
0f9dd46c | 1515 | |
96303081 | 1516 | if (offset < entry->offset) |
0f9dd46c | 1517 | n = n->rb_left; |
96303081 | 1518 | else if (offset > entry->offset) |
0f9dd46c | 1519 | n = n->rb_right; |
96303081 | 1520 | else |
0f9dd46c | 1521 | break; |
0f9dd46c JB |
1522 | } |
1523 | ||
96303081 JB |
1524 | if (bitmap_only) { |
1525 | if (!entry) | |
1526 | return NULL; | |
1527 | if (entry->bitmap) | |
1528 | return entry; | |
0f9dd46c | 1529 | |
96303081 JB |
1530 | /* |
1531 | * bitmap entry and extent entry may share same offset, | |
1532 | * in that case, bitmap entry comes after extent entry. | |
1533 | */ | |
1534 | n = rb_next(n); | |
1535 | if (!n) | |
1536 | return NULL; | |
1537 | entry = rb_entry(n, struct btrfs_free_space, offset_index); | |
1538 | if (entry->offset != offset) | |
1539 | return NULL; | |
0f9dd46c | 1540 | |
96303081 JB |
1541 | WARN_ON(!entry->bitmap); |
1542 | return entry; | |
1543 | } else if (entry) { | |
1544 | if (entry->bitmap) { | |
0f9dd46c | 1545 | /* |
96303081 JB |
1546 | * if previous extent entry covers the offset, |
1547 | * we should return it instead of the bitmap entry | |
0f9dd46c | 1548 | */ |
de6c4115 MX |
1549 | n = rb_prev(&entry->offset_index); |
1550 | if (n) { | |
96303081 JB |
1551 | prev = rb_entry(n, struct btrfs_free_space, |
1552 | offset_index); | |
de6c4115 MX |
1553 | if (!prev->bitmap && |
1554 | prev->offset + prev->bytes > offset) | |
1555 | entry = prev; | |
0f9dd46c | 1556 | } |
96303081 JB |
1557 | } |
1558 | return entry; | |
1559 | } | |
1560 | ||
1561 | if (!prev) | |
1562 | return NULL; | |
1563 | ||
1564 | /* find last entry before the 'offset' */ | |
1565 | entry = prev; | |
1566 | if (entry->offset > offset) { | |
1567 | n = rb_prev(&entry->offset_index); | |
1568 | if (n) { | |
1569 | entry = rb_entry(n, struct btrfs_free_space, | |
1570 | offset_index); | |
b12d6869 | 1571 | ASSERT(entry->offset <= offset); |
0f9dd46c | 1572 | } else { |
96303081 JB |
1573 | if (fuzzy) |
1574 | return entry; | |
1575 | else | |
1576 | return NULL; | |
0f9dd46c JB |
1577 | } |
1578 | } | |
1579 | ||
96303081 | 1580 | if (entry->bitmap) { |
de6c4115 MX |
1581 | n = rb_prev(&entry->offset_index); |
1582 | if (n) { | |
96303081 JB |
1583 | prev = rb_entry(n, struct btrfs_free_space, |
1584 | offset_index); | |
de6c4115 MX |
1585 | if (!prev->bitmap && |
1586 | prev->offset + prev->bytes > offset) | |
1587 | return prev; | |
96303081 | 1588 | } |
34d52cb6 | 1589 | if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset) |
96303081 JB |
1590 | return entry; |
1591 | } else if (entry->offset + entry->bytes > offset) | |
1592 | return entry; | |
1593 | ||
1594 | if (!fuzzy) | |
1595 | return NULL; | |
1596 | ||
1597 | while (1) { | |
1598 | if (entry->bitmap) { | |
1599 | if (entry->offset + BITS_PER_BITMAP * | |
34d52cb6 | 1600 | ctl->unit > offset) |
96303081 JB |
1601 | break; |
1602 | } else { | |
1603 | if (entry->offset + entry->bytes > offset) | |
1604 | break; | |
1605 | } | |
1606 | ||
1607 | n = rb_next(&entry->offset_index); | |
1608 | if (!n) | |
1609 | return NULL; | |
1610 | entry = rb_entry(n, struct btrfs_free_space, offset_index); | |
1611 | } | |
1612 | return entry; | |
0f9dd46c JB |
1613 | } |
1614 | ||
f333adb5 | 1615 | static inline void |
34d52cb6 | 1616 | __unlink_free_space(struct btrfs_free_space_ctl *ctl, |
f333adb5 | 1617 | struct btrfs_free_space *info) |
0f9dd46c | 1618 | { |
34d52cb6 LZ |
1619 | rb_erase(&info->offset_index, &ctl->free_space_offset); |
1620 | ctl->free_extents--; | |
f333adb5 LZ |
1621 | } |
1622 | ||
34d52cb6 | 1623 | static void unlink_free_space(struct btrfs_free_space_ctl *ctl, |
f333adb5 LZ |
1624 | struct btrfs_free_space *info) |
1625 | { | |
34d52cb6 LZ |
1626 | __unlink_free_space(ctl, info); |
1627 | ctl->free_space -= info->bytes; | |
0f9dd46c JB |
1628 | } |
1629 | ||
34d52cb6 | 1630 | static int link_free_space(struct btrfs_free_space_ctl *ctl, |
0f9dd46c JB |
1631 | struct btrfs_free_space *info) |
1632 | { | |
1633 | int ret = 0; | |
1634 | ||
b12d6869 | 1635 | ASSERT(info->bytes || info->bitmap); |
34d52cb6 | 1636 | ret = tree_insert_offset(&ctl->free_space_offset, info->offset, |
96303081 | 1637 | &info->offset_index, (info->bitmap != NULL)); |
0f9dd46c JB |
1638 | if (ret) |
1639 | return ret; | |
1640 | ||
34d52cb6 LZ |
1641 | ctl->free_space += info->bytes; |
1642 | ctl->free_extents++; | |
96303081 JB |
1643 | return ret; |
1644 | } | |
1645 | ||
34d52cb6 | 1646 | static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl) |
96303081 | 1647 | { |
34d52cb6 | 1648 | struct btrfs_block_group_cache *block_group = ctl->private; |
25891f79 JB |
1649 | u64 max_bytes; |
1650 | u64 bitmap_bytes; | |
1651 | u64 extent_bytes; | |
8eb2d829 | 1652 | u64 size = block_group->key.offset; |
0ef6447a FX |
1653 | u64 bytes_per_bg = BITS_PER_BITMAP * ctl->unit; |
1654 | u64 max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg); | |
34d52cb6 | 1655 | |
0ef6447a | 1656 | max_bitmaps = max_t(u64, max_bitmaps, 1); |
dde5740f | 1657 | |
b12d6869 | 1658 | ASSERT(ctl->total_bitmaps <= max_bitmaps); |
96303081 JB |
1659 | |
1660 | /* | |
1661 | * The goal is to keep the total amount of memory used per 1gb of space | |
1662 | * at or below 32k, so we need to adjust how much memory we allow to be | |
1663 | * used by extent based free space tracking | |
1664 | */ | |
ee22184b | 1665 | if (size < SZ_1G) |
8eb2d829 LZ |
1666 | max_bytes = MAX_CACHE_BYTES_PER_GIG; |
1667 | else | |
ee22184b | 1668 | max_bytes = MAX_CACHE_BYTES_PER_GIG * div_u64(size, SZ_1G); |
96303081 | 1669 | |
25891f79 JB |
1670 | /* |
1671 | * we want to account for 1 more bitmap than what we have so we can make | |
1672 | * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as | |
1673 | * we add more bitmaps. | |
1674 | */ | |
b9ef22de | 1675 | bitmap_bytes = (ctl->total_bitmaps + 1) * ctl->unit; |
96303081 | 1676 | |
25891f79 | 1677 | if (bitmap_bytes >= max_bytes) { |
34d52cb6 | 1678 | ctl->extents_thresh = 0; |
25891f79 JB |
1679 | return; |
1680 | } | |
96303081 | 1681 | |
25891f79 | 1682 | /* |
f8c269d7 | 1683 | * we want the extent entry threshold to always be at most 1/2 the max |
25891f79 JB |
1684 | * bytes we can have, or whatever is less than that. |
1685 | */ | |
1686 | extent_bytes = max_bytes - bitmap_bytes; | |
f8c269d7 | 1687 | extent_bytes = min_t(u64, extent_bytes, max_bytes >> 1); |
96303081 | 1688 | |
34d52cb6 | 1689 | ctl->extents_thresh = |
f8c269d7 | 1690 | div_u64(extent_bytes, sizeof(struct btrfs_free_space)); |
96303081 JB |
1691 | } |
1692 | ||
bb3ac5a4 MX |
1693 | static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl, |
1694 | struct btrfs_free_space *info, | |
1695 | u64 offset, u64 bytes) | |
96303081 | 1696 | { |
f38b6e75 | 1697 | unsigned long start, count; |
96303081 | 1698 | |
34d52cb6 LZ |
1699 | start = offset_to_bit(info->offset, ctl->unit, offset); |
1700 | count = bytes_to_bits(bytes, ctl->unit); | |
b12d6869 | 1701 | ASSERT(start + count <= BITS_PER_BITMAP); |
96303081 | 1702 | |
f38b6e75 | 1703 | bitmap_clear(info->bitmap, start, count); |
96303081 JB |
1704 | |
1705 | info->bytes -= bytes; | |
bb3ac5a4 MX |
1706 | } |
1707 | ||
1708 | static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl, | |
1709 | struct btrfs_free_space *info, u64 offset, | |
1710 | u64 bytes) | |
1711 | { | |
1712 | __bitmap_clear_bits(ctl, info, offset, bytes); | |
34d52cb6 | 1713 | ctl->free_space -= bytes; |
96303081 JB |
1714 | } |
1715 | ||
34d52cb6 | 1716 | static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl, |
817d52f8 JB |
1717 | struct btrfs_free_space *info, u64 offset, |
1718 | u64 bytes) | |
96303081 | 1719 | { |
f38b6e75 | 1720 | unsigned long start, count; |
96303081 | 1721 | |
34d52cb6 LZ |
1722 | start = offset_to_bit(info->offset, ctl->unit, offset); |
1723 | count = bytes_to_bits(bytes, ctl->unit); | |
b12d6869 | 1724 | ASSERT(start + count <= BITS_PER_BITMAP); |
96303081 | 1725 | |
f38b6e75 | 1726 | bitmap_set(info->bitmap, start, count); |
96303081 JB |
1727 | |
1728 | info->bytes += bytes; | |
34d52cb6 | 1729 | ctl->free_space += bytes; |
96303081 JB |
1730 | } |
1731 | ||
a4820398 MX |
1732 | /* |
1733 | * If we can not find suitable extent, we will use bytes to record | |
1734 | * the size of the max extent. | |
1735 | */ | |
34d52cb6 | 1736 | static int search_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 | 1737 | struct btrfs_free_space *bitmap_info, u64 *offset, |
0584f718 | 1738 | u64 *bytes, bool for_alloc) |
96303081 JB |
1739 | { |
1740 | unsigned long found_bits = 0; | |
a4820398 | 1741 | unsigned long max_bits = 0; |
96303081 JB |
1742 | unsigned long bits, i; |
1743 | unsigned long next_zero; | |
a4820398 | 1744 | unsigned long extent_bits; |
96303081 | 1745 | |
cef40483 JB |
1746 | /* |
1747 | * Skip searching the bitmap if we don't have a contiguous section that | |
1748 | * is large enough for this allocation. | |
1749 | */ | |
0584f718 JB |
1750 | if (for_alloc && |
1751 | bitmap_info->max_extent_size && | |
cef40483 JB |
1752 | bitmap_info->max_extent_size < *bytes) { |
1753 | *bytes = bitmap_info->max_extent_size; | |
1754 | return -1; | |
1755 | } | |
1756 | ||
34d52cb6 | 1757 | i = offset_to_bit(bitmap_info->offset, ctl->unit, |
96303081 | 1758 | max_t(u64, *offset, bitmap_info->offset)); |
34d52cb6 | 1759 | bits = bytes_to_bits(*bytes, ctl->unit); |
96303081 | 1760 | |
ebb3dad4 | 1761 | for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) { |
0584f718 JB |
1762 | if (for_alloc && bits == 1) { |
1763 | found_bits = 1; | |
1764 | break; | |
1765 | } | |
96303081 JB |
1766 | next_zero = find_next_zero_bit(bitmap_info->bitmap, |
1767 | BITS_PER_BITMAP, i); | |
a4820398 MX |
1768 | extent_bits = next_zero - i; |
1769 | if (extent_bits >= bits) { | |
1770 | found_bits = extent_bits; | |
96303081 | 1771 | break; |
a4820398 MX |
1772 | } else if (extent_bits > max_bits) { |
1773 | max_bits = extent_bits; | |
96303081 JB |
1774 | } |
1775 | i = next_zero; | |
1776 | } | |
1777 | ||
1778 | if (found_bits) { | |
34d52cb6 LZ |
1779 | *offset = (u64)(i * ctl->unit) + bitmap_info->offset; |
1780 | *bytes = (u64)(found_bits) * ctl->unit; | |
96303081 JB |
1781 | return 0; |
1782 | } | |
1783 | ||
a4820398 | 1784 | *bytes = (u64)(max_bits) * ctl->unit; |
cef40483 | 1785 | bitmap_info->max_extent_size = *bytes; |
96303081 JB |
1786 | return -1; |
1787 | } | |
1788 | ||
a4820398 | 1789 | /* Cache the size of the max extent in bytes */ |
34d52cb6 | 1790 | static struct btrfs_free_space * |
53b381b3 | 1791 | find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes, |
a4820398 | 1792 | unsigned long align, u64 *max_extent_size) |
96303081 JB |
1793 | { |
1794 | struct btrfs_free_space *entry; | |
1795 | struct rb_node *node; | |
53b381b3 DW |
1796 | u64 tmp; |
1797 | u64 align_off; | |
96303081 JB |
1798 | int ret; |
1799 | ||
34d52cb6 | 1800 | if (!ctl->free_space_offset.rb_node) |
a4820398 | 1801 | goto out; |
96303081 | 1802 | |
34d52cb6 | 1803 | entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1); |
96303081 | 1804 | if (!entry) |
a4820398 | 1805 | goto out; |
96303081 JB |
1806 | |
1807 | for (node = &entry->offset_index; node; node = rb_next(node)) { | |
1808 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
a4820398 MX |
1809 | if (entry->bytes < *bytes) { |
1810 | if (entry->bytes > *max_extent_size) | |
1811 | *max_extent_size = entry->bytes; | |
96303081 | 1812 | continue; |
a4820398 | 1813 | } |
96303081 | 1814 | |
53b381b3 DW |
1815 | /* make sure the space returned is big enough |
1816 | * to match our requested alignment | |
1817 | */ | |
1818 | if (*bytes >= align) { | |
a4820398 | 1819 | tmp = entry->offset - ctl->start + align - 1; |
47c5713f | 1820 | tmp = div64_u64(tmp, align); |
53b381b3 DW |
1821 | tmp = tmp * align + ctl->start; |
1822 | align_off = tmp - entry->offset; | |
1823 | } else { | |
1824 | align_off = 0; | |
1825 | tmp = entry->offset; | |
1826 | } | |
1827 | ||
a4820398 MX |
1828 | if (entry->bytes < *bytes + align_off) { |
1829 | if (entry->bytes > *max_extent_size) | |
1830 | *max_extent_size = entry->bytes; | |
53b381b3 | 1831 | continue; |
a4820398 | 1832 | } |
53b381b3 | 1833 | |
96303081 | 1834 | if (entry->bitmap) { |
a4820398 MX |
1835 | u64 size = *bytes; |
1836 | ||
0584f718 | 1837 | ret = search_bitmap(ctl, entry, &tmp, &size, true); |
53b381b3 DW |
1838 | if (!ret) { |
1839 | *offset = tmp; | |
a4820398 | 1840 | *bytes = size; |
96303081 | 1841 | return entry; |
a4820398 MX |
1842 | } else if (size > *max_extent_size) { |
1843 | *max_extent_size = size; | |
53b381b3 | 1844 | } |
96303081 JB |
1845 | continue; |
1846 | } | |
1847 | ||
53b381b3 DW |
1848 | *offset = tmp; |
1849 | *bytes = entry->bytes - align_off; | |
96303081 JB |
1850 | return entry; |
1851 | } | |
a4820398 | 1852 | out: |
96303081 JB |
1853 | return NULL; |
1854 | } | |
1855 | ||
34d52cb6 | 1856 | static void add_new_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1857 | struct btrfs_free_space *info, u64 offset) |
1858 | { | |
34d52cb6 | 1859 | info->offset = offset_to_bitmap(ctl, offset); |
f019f426 | 1860 | info->bytes = 0; |
f2d0f676 | 1861 | INIT_LIST_HEAD(&info->list); |
34d52cb6 LZ |
1862 | link_free_space(ctl, info); |
1863 | ctl->total_bitmaps++; | |
96303081 | 1864 | |
34d52cb6 | 1865 | ctl->op->recalc_thresholds(ctl); |
96303081 JB |
1866 | } |
1867 | ||
34d52cb6 | 1868 | static void free_bitmap(struct btrfs_free_space_ctl *ctl, |
edf6e2d1 LZ |
1869 | struct btrfs_free_space *bitmap_info) |
1870 | { | |
34d52cb6 | 1871 | unlink_free_space(ctl, bitmap_info); |
edf6e2d1 | 1872 | kfree(bitmap_info->bitmap); |
dc89e982 | 1873 | kmem_cache_free(btrfs_free_space_cachep, bitmap_info); |
34d52cb6 LZ |
1874 | ctl->total_bitmaps--; |
1875 | ctl->op->recalc_thresholds(ctl); | |
edf6e2d1 LZ |
1876 | } |
1877 | ||
34d52cb6 | 1878 | static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1879 | struct btrfs_free_space *bitmap_info, |
1880 | u64 *offset, u64 *bytes) | |
1881 | { | |
1882 | u64 end; | |
6606bb97 JB |
1883 | u64 search_start, search_bytes; |
1884 | int ret; | |
96303081 JB |
1885 | |
1886 | again: | |
34d52cb6 | 1887 | end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1; |
96303081 | 1888 | |
6606bb97 | 1889 | /* |
bdb7d303 JB |
1890 | * We need to search for bits in this bitmap. We could only cover some |
1891 | * of the extent in this bitmap thanks to how we add space, so we need | |
1892 | * to search for as much as it as we can and clear that amount, and then | |
1893 | * go searching for the next bit. | |
6606bb97 JB |
1894 | */ |
1895 | search_start = *offset; | |
bdb7d303 | 1896 | search_bytes = ctl->unit; |
13dbc089 | 1897 | search_bytes = min(search_bytes, end - search_start + 1); |
0584f718 JB |
1898 | ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes, |
1899 | false); | |
b50c6e25 JB |
1900 | if (ret < 0 || search_start != *offset) |
1901 | return -EINVAL; | |
6606bb97 | 1902 | |
bdb7d303 JB |
1903 | /* We may have found more bits than what we need */ |
1904 | search_bytes = min(search_bytes, *bytes); | |
1905 | ||
1906 | /* Cannot clear past the end of the bitmap */ | |
1907 | search_bytes = min(search_bytes, end - search_start + 1); | |
1908 | ||
1909 | bitmap_clear_bits(ctl, bitmap_info, search_start, search_bytes); | |
1910 | *offset += search_bytes; | |
1911 | *bytes -= search_bytes; | |
96303081 JB |
1912 | |
1913 | if (*bytes) { | |
6606bb97 | 1914 | struct rb_node *next = rb_next(&bitmap_info->offset_index); |
edf6e2d1 | 1915 | if (!bitmap_info->bytes) |
34d52cb6 | 1916 | free_bitmap(ctl, bitmap_info); |
96303081 | 1917 | |
6606bb97 JB |
1918 | /* |
1919 | * no entry after this bitmap, but we still have bytes to | |
1920 | * remove, so something has gone wrong. | |
1921 | */ | |
1922 | if (!next) | |
96303081 JB |
1923 | return -EINVAL; |
1924 | ||
6606bb97 JB |
1925 | bitmap_info = rb_entry(next, struct btrfs_free_space, |
1926 | offset_index); | |
1927 | ||
1928 | /* | |
1929 | * if the next entry isn't a bitmap we need to return to let the | |
1930 | * extent stuff do its work. | |
1931 | */ | |
96303081 JB |
1932 | if (!bitmap_info->bitmap) |
1933 | return -EAGAIN; | |
1934 | ||
6606bb97 JB |
1935 | /* |
1936 | * Ok the next item is a bitmap, but it may not actually hold | |
1937 | * the information for the rest of this free space stuff, so | |
1938 | * look for it, and if we don't find it return so we can try | |
1939 | * everything over again. | |
1940 | */ | |
1941 | search_start = *offset; | |
bdb7d303 | 1942 | search_bytes = ctl->unit; |
34d52cb6 | 1943 | ret = search_bitmap(ctl, bitmap_info, &search_start, |
0584f718 | 1944 | &search_bytes, false); |
6606bb97 JB |
1945 | if (ret < 0 || search_start != *offset) |
1946 | return -EAGAIN; | |
1947 | ||
96303081 | 1948 | goto again; |
edf6e2d1 | 1949 | } else if (!bitmap_info->bytes) |
34d52cb6 | 1950 | free_bitmap(ctl, bitmap_info); |
96303081 JB |
1951 | |
1952 | return 0; | |
1953 | } | |
1954 | ||
2cdc342c JB |
1955 | static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl, |
1956 | struct btrfs_free_space *info, u64 offset, | |
1957 | u64 bytes) | |
1958 | { | |
1959 | u64 bytes_to_set = 0; | |
1960 | u64 end; | |
1961 | ||
1962 | end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit); | |
1963 | ||
1964 | bytes_to_set = min(end - offset, bytes); | |
1965 | ||
1966 | bitmap_set_bits(ctl, info, offset, bytes_to_set); | |
1967 | ||
cef40483 JB |
1968 | /* |
1969 | * We set some bytes, we have no idea what the max extent size is | |
1970 | * anymore. | |
1971 | */ | |
1972 | info->max_extent_size = 0; | |
1973 | ||
2cdc342c JB |
1974 | return bytes_to_set; |
1975 | ||
1976 | } | |
1977 | ||
34d52cb6 LZ |
1978 | static bool use_bitmap(struct btrfs_free_space_ctl *ctl, |
1979 | struct btrfs_free_space *info) | |
96303081 | 1980 | { |
34d52cb6 | 1981 | struct btrfs_block_group_cache *block_group = ctl->private; |
0b246afa | 1982 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
d0bd4560 JB |
1983 | bool forced = false; |
1984 | ||
1985 | #ifdef CONFIG_BTRFS_DEBUG | |
2ff7e61e | 1986 | if (btrfs_should_fragment_free_space(block_group)) |
d0bd4560 JB |
1987 | forced = true; |
1988 | #endif | |
96303081 JB |
1989 | |
1990 | /* | |
1991 | * If we are below the extents threshold then we can add this as an | |
1992 | * extent, and don't have to deal with the bitmap | |
1993 | */ | |
d0bd4560 | 1994 | if (!forced && ctl->free_extents < ctl->extents_thresh) { |
32cb0840 JB |
1995 | /* |
1996 | * If this block group has some small extents we don't want to | |
1997 | * use up all of our free slots in the cache with them, we want | |
01327610 | 1998 | * to reserve them to larger extents, however if we have plenty |
32cb0840 JB |
1999 | * of cache left then go ahead an dadd them, no sense in adding |
2000 | * the overhead of a bitmap if we don't have to. | |
2001 | */ | |
0b246afa | 2002 | if (info->bytes <= fs_info->sectorsize * 4) { |
34d52cb6 LZ |
2003 | if (ctl->free_extents * 2 <= ctl->extents_thresh) |
2004 | return false; | |
32cb0840 | 2005 | } else { |
34d52cb6 | 2006 | return false; |
32cb0840 JB |
2007 | } |
2008 | } | |
96303081 JB |
2009 | |
2010 | /* | |
dde5740f JB |
2011 | * The original block groups from mkfs can be really small, like 8 |
2012 | * megabytes, so don't bother with a bitmap for those entries. However | |
2013 | * some block groups can be smaller than what a bitmap would cover but | |
2014 | * are still large enough that they could overflow the 32k memory limit, | |
2015 | * so allow those block groups to still be allowed to have a bitmap | |
2016 | * entry. | |
96303081 | 2017 | */ |
dde5740f | 2018 | if (((BITS_PER_BITMAP * ctl->unit) >> 1) > block_group->key.offset) |
34d52cb6 LZ |
2019 | return false; |
2020 | ||
2021 | return true; | |
2022 | } | |
2023 | ||
20e5506b | 2024 | static const struct btrfs_free_space_op free_space_op = { |
2cdc342c JB |
2025 | .recalc_thresholds = recalculate_thresholds, |
2026 | .use_bitmap = use_bitmap, | |
2027 | }; | |
2028 | ||
34d52cb6 LZ |
2029 | static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl, |
2030 | struct btrfs_free_space *info) | |
2031 | { | |
2032 | struct btrfs_free_space *bitmap_info; | |
2cdc342c | 2033 | struct btrfs_block_group_cache *block_group = NULL; |
34d52cb6 | 2034 | int added = 0; |
2cdc342c | 2035 | u64 bytes, offset, bytes_added; |
34d52cb6 | 2036 | int ret; |
96303081 JB |
2037 | |
2038 | bytes = info->bytes; | |
2039 | offset = info->offset; | |
2040 | ||
34d52cb6 LZ |
2041 | if (!ctl->op->use_bitmap(ctl, info)) |
2042 | return 0; | |
2043 | ||
2cdc342c JB |
2044 | if (ctl->op == &free_space_op) |
2045 | block_group = ctl->private; | |
38e87880 | 2046 | again: |
2cdc342c JB |
2047 | /* |
2048 | * Since we link bitmaps right into the cluster we need to see if we | |
2049 | * have a cluster here, and if so and it has our bitmap we need to add | |
2050 | * the free space to that bitmap. | |
2051 | */ | |
2052 | if (block_group && !list_empty(&block_group->cluster_list)) { | |
2053 | struct btrfs_free_cluster *cluster; | |
2054 | struct rb_node *node; | |
2055 | struct btrfs_free_space *entry; | |
2056 | ||
2057 | cluster = list_entry(block_group->cluster_list.next, | |
2058 | struct btrfs_free_cluster, | |
2059 | block_group_list); | |
2060 | spin_lock(&cluster->lock); | |
2061 | node = rb_first(&cluster->root); | |
2062 | if (!node) { | |
2063 | spin_unlock(&cluster->lock); | |
38e87880 | 2064 | goto no_cluster_bitmap; |
2cdc342c JB |
2065 | } |
2066 | ||
2067 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
2068 | if (!entry->bitmap) { | |
2069 | spin_unlock(&cluster->lock); | |
38e87880 | 2070 | goto no_cluster_bitmap; |
2cdc342c JB |
2071 | } |
2072 | ||
2073 | if (entry->offset == offset_to_bitmap(ctl, offset)) { | |
2074 | bytes_added = add_bytes_to_bitmap(ctl, entry, | |
2075 | offset, bytes); | |
2076 | bytes -= bytes_added; | |
2077 | offset += bytes_added; | |
2078 | } | |
2079 | spin_unlock(&cluster->lock); | |
2080 | if (!bytes) { | |
2081 | ret = 1; | |
2082 | goto out; | |
2083 | } | |
2084 | } | |
38e87880 CM |
2085 | |
2086 | no_cluster_bitmap: | |
34d52cb6 | 2087 | bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), |
96303081 JB |
2088 | 1, 0); |
2089 | if (!bitmap_info) { | |
b12d6869 | 2090 | ASSERT(added == 0); |
96303081 JB |
2091 | goto new_bitmap; |
2092 | } | |
2093 | ||
2cdc342c JB |
2094 | bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes); |
2095 | bytes -= bytes_added; | |
2096 | offset += bytes_added; | |
2097 | added = 0; | |
96303081 JB |
2098 | |
2099 | if (!bytes) { | |
2100 | ret = 1; | |
2101 | goto out; | |
2102 | } else | |
2103 | goto again; | |
2104 | ||
2105 | new_bitmap: | |
2106 | if (info && info->bitmap) { | |
34d52cb6 | 2107 | add_new_bitmap(ctl, info, offset); |
96303081 JB |
2108 | added = 1; |
2109 | info = NULL; | |
2110 | goto again; | |
2111 | } else { | |
34d52cb6 | 2112 | spin_unlock(&ctl->tree_lock); |
96303081 JB |
2113 | |
2114 | /* no pre-allocated info, allocate a new one */ | |
2115 | if (!info) { | |
dc89e982 JB |
2116 | info = kmem_cache_zalloc(btrfs_free_space_cachep, |
2117 | GFP_NOFS); | |
96303081 | 2118 | if (!info) { |
34d52cb6 | 2119 | spin_lock(&ctl->tree_lock); |
96303081 JB |
2120 | ret = -ENOMEM; |
2121 | goto out; | |
2122 | } | |
2123 | } | |
2124 | ||
2125 | /* allocate the bitmap */ | |
09cbfeaf | 2126 | info->bitmap = kzalloc(PAGE_SIZE, GFP_NOFS); |
34d52cb6 | 2127 | spin_lock(&ctl->tree_lock); |
96303081 JB |
2128 | if (!info->bitmap) { |
2129 | ret = -ENOMEM; | |
2130 | goto out; | |
2131 | } | |
2132 | goto again; | |
2133 | } | |
2134 | ||
2135 | out: | |
2136 | if (info) { | |
2137 | if (info->bitmap) | |
2138 | kfree(info->bitmap); | |
dc89e982 | 2139 | kmem_cache_free(btrfs_free_space_cachep, info); |
96303081 | 2140 | } |
0f9dd46c JB |
2141 | |
2142 | return ret; | |
2143 | } | |
2144 | ||
945d8962 | 2145 | static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl, |
f333adb5 | 2146 | struct btrfs_free_space *info, bool update_stat) |
0f9dd46c | 2147 | { |
120d66ee LZ |
2148 | struct btrfs_free_space *left_info; |
2149 | struct btrfs_free_space *right_info; | |
2150 | bool merged = false; | |
2151 | u64 offset = info->offset; | |
2152 | u64 bytes = info->bytes; | |
6226cb0a | 2153 | |
0f9dd46c JB |
2154 | /* |
2155 | * first we want to see if there is free space adjacent to the range we | |
2156 | * are adding, if there is remove that struct and add a new one to | |
2157 | * cover the entire range | |
2158 | */ | |
34d52cb6 | 2159 | right_info = tree_search_offset(ctl, offset + bytes, 0, 0); |
96303081 JB |
2160 | if (right_info && rb_prev(&right_info->offset_index)) |
2161 | left_info = rb_entry(rb_prev(&right_info->offset_index), | |
2162 | struct btrfs_free_space, offset_index); | |
2163 | else | |
34d52cb6 | 2164 | left_info = tree_search_offset(ctl, offset - 1, 0, 0); |
0f9dd46c | 2165 | |
96303081 | 2166 | if (right_info && !right_info->bitmap) { |
f333adb5 | 2167 | if (update_stat) |
34d52cb6 | 2168 | unlink_free_space(ctl, right_info); |
f333adb5 | 2169 | else |
34d52cb6 | 2170 | __unlink_free_space(ctl, right_info); |
6226cb0a | 2171 | info->bytes += right_info->bytes; |
dc89e982 | 2172 | kmem_cache_free(btrfs_free_space_cachep, right_info); |
120d66ee | 2173 | merged = true; |
0f9dd46c JB |
2174 | } |
2175 | ||
96303081 JB |
2176 | if (left_info && !left_info->bitmap && |
2177 | left_info->offset + left_info->bytes == offset) { | |
f333adb5 | 2178 | if (update_stat) |
34d52cb6 | 2179 | unlink_free_space(ctl, left_info); |
f333adb5 | 2180 | else |
34d52cb6 | 2181 | __unlink_free_space(ctl, left_info); |
6226cb0a JB |
2182 | info->offset = left_info->offset; |
2183 | info->bytes += left_info->bytes; | |
dc89e982 | 2184 | kmem_cache_free(btrfs_free_space_cachep, left_info); |
120d66ee | 2185 | merged = true; |
0f9dd46c JB |
2186 | } |
2187 | ||
120d66ee LZ |
2188 | return merged; |
2189 | } | |
2190 | ||
20005523 FM |
2191 | static bool steal_from_bitmap_to_end(struct btrfs_free_space_ctl *ctl, |
2192 | struct btrfs_free_space *info, | |
2193 | bool update_stat) | |
2194 | { | |
2195 | struct btrfs_free_space *bitmap; | |
2196 | unsigned long i; | |
2197 | unsigned long j; | |
2198 | const u64 end = info->offset + info->bytes; | |
2199 | const u64 bitmap_offset = offset_to_bitmap(ctl, end); | |
2200 | u64 bytes; | |
2201 | ||
2202 | bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0); | |
2203 | if (!bitmap) | |
2204 | return false; | |
2205 | ||
2206 | i = offset_to_bit(bitmap->offset, ctl->unit, end); | |
2207 | j = find_next_zero_bit(bitmap->bitmap, BITS_PER_BITMAP, i); | |
2208 | if (j == i) | |
2209 | return false; | |
2210 | bytes = (j - i) * ctl->unit; | |
2211 | info->bytes += bytes; | |
2212 | ||
2213 | if (update_stat) | |
2214 | bitmap_clear_bits(ctl, bitmap, end, bytes); | |
2215 | else | |
2216 | __bitmap_clear_bits(ctl, bitmap, end, bytes); | |
2217 | ||
2218 | if (!bitmap->bytes) | |
2219 | free_bitmap(ctl, bitmap); | |
2220 | ||
2221 | return true; | |
2222 | } | |
2223 | ||
2224 | static bool steal_from_bitmap_to_front(struct btrfs_free_space_ctl *ctl, | |
2225 | struct btrfs_free_space *info, | |
2226 | bool update_stat) | |
2227 | { | |
2228 | struct btrfs_free_space *bitmap; | |
2229 | u64 bitmap_offset; | |
2230 | unsigned long i; | |
2231 | unsigned long j; | |
2232 | unsigned long prev_j; | |
2233 | u64 bytes; | |
2234 | ||
2235 | bitmap_offset = offset_to_bitmap(ctl, info->offset); | |
2236 | /* If we're on a boundary, try the previous logical bitmap. */ | |
2237 | if (bitmap_offset == info->offset) { | |
2238 | if (info->offset == 0) | |
2239 | return false; | |
2240 | bitmap_offset = offset_to_bitmap(ctl, info->offset - 1); | |
2241 | } | |
2242 | ||
2243 | bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0); | |
2244 | if (!bitmap) | |
2245 | return false; | |
2246 | ||
2247 | i = offset_to_bit(bitmap->offset, ctl->unit, info->offset) - 1; | |
2248 | j = 0; | |
2249 | prev_j = (unsigned long)-1; | |
2250 | for_each_clear_bit_from(j, bitmap->bitmap, BITS_PER_BITMAP) { | |
2251 | if (j > i) | |
2252 | break; | |
2253 | prev_j = j; | |
2254 | } | |
2255 | if (prev_j == i) | |
2256 | return false; | |
2257 | ||
2258 | if (prev_j == (unsigned long)-1) | |
2259 | bytes = (i + 1) * ctl->unit; | |
2260 | else | |
2261 | bytes = (i - prev_j) * ctl->unit; | |
2262 | ||
2263 | info->offset -= bytes; | |
2264 | info->bytes += bytes; | |
2265 | ||
2266 | if (update_stat) | |
2267 | bitmap_clear_bits(ctl, bitmap, info->offset, bytes); | |
2268 | else | |
2269 | __bitmap_clear_bits(ctl, bitmap, info->offset, bytes); | |
2270 | ||
2271 | if (!bitmap->bytes) | |
2272 | free_bitmap(ctl, bitmap); | |
2273 | ||
2274 | return true; | |
2275 | } | |
2276 | ||
2277 | /* | |
2278 | * We prefer always to allocate from extent entries, both for clustered and | |
2279 | * non-clustered allocation requests. So when attempting to add a new extent | |
2280 | * entry, try to see if there's adjacent free space in bitmap entries, and if | |
2281 | * there is, migrate that space from the bitmaps to the extent. | |
2282 | * Like this we get better chances of satisfying space allocation requests | |
2283 | * because we attempt to satisfy them based on a single cache entry, and never | |
2284 | * on 2 or more entries - even if the entries represent a contiguous free space | |
2285 | * region (e.g. 1 extent entry + 1 bitmap entry starting where the extent entry | |
2286 | * ends). | |
2287 | */ | |
2288 | static void steal_from_bitmap(struct btrfs_free_space_ctl *ctl, | |
2289 | struct btrfs_free_space *info, | |
2290 | bool update_stat) | |
2291 | { | |
2292 | /* | |
2293 | * Only work with disconnected entries, as we can change their offset, | |
2294 | * and must be extent entries. | |
2295 | */ | |
2296 | ASSERT(!info->bitmap); | |
2297 | ASSERT(RB_EMPTY_NODE(&info->offset_index)); | |
2298 | ||
2299 | if (ctl->total_bitmaps > 0) { | |
2300 | bool stole_end; | |
2301 | bool stole_front = false; | |
2302 | ||
2303 | stole_end = steal_from_bitmap_to_end(ctl, info, update_stat); | |
2304 | if (ctl->total_bitmaps > 0) | |
2305 | stole_front = steal_from_bitmap_to_front(ctl, info, | |
2306 | update_stat); | |
2307 | ||
2308 | if (stole_end || stole_front) | |
2309 | try_merge_free_space(ctl, info, update_stat); | |
2310 | } | |
2311 | } | |
2312 | ||
ab8d0fc4 JM |
2313 | int __btrfs_add_free_space(struct btrfs_fs_info *fs_info, |
2314 | struct btrfs_free_space_ctl *ctl, | |
581bb050 | 2315 | u64 offset, u64 bytes) |
120d66ee LZ |
2316 | { |
2317 | struct btrfs_free_space *info; | |
2318 | int ret = 0; | |
2319 | ||
dc89e982 | 2320 | info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS); |
120d66ee LZ |
2321 | if (!info) |
2322 | return -ENOMEM; | |
2323 | ||
2324 | info->offset = offset; | |
2325 | info->bytes = bytes; | |
20005523 | 2326 | RB_CLEAR_NODE(&info->offset_index); |
120d66ee | 2327 | |
34d52cb6 | 2328 | spin_lock(&ctl->tree_lock); |
120d66ee | 2329 | |
34d52cb6 | 2330 | if (try_merge_free_space(ctl, info, true)) |
120d66ee LZ |
2331 | goto link; |
2332 | ||
2333 | /* | |
2334 | * There was no extent directly to the left or right of this new | |
2335 | * extent then we know we're going to have to allocate a new extent, so | |
2336 | * before we do that see if we need to drop this into a bitmap | |
2337 | */ | |
34d52cb6 | 2338 | ret = insert_into_bitmap(ctl, info); |
120d66ee LZ |
2339 | if (ret < 0) { |
2340 | goto out; | |
2341 | } else if (ret) { | |
2342 | ret = 0; | |
2343 | goto out; | |
2344 | } | |
2345 | link: | |
20005523 FM |
2346 | /* |
2347 | * Only steal free space from adjacent bitmaps if we're sure we're not | |
2348 | * going to add the new free space to existing bitmap entries - because | |
2349 | * that would mean unnecessary work that would be reverted. Therefore | |
2350 | * attempt to steal space from bitmaps if we're adding an extent entry. | |
2351 | */ | |
2352 | steal_from_bitmap(ctl, info, true); | |
2353 | ||
34d52cb6 | 2354 | ret = link_free_space(ctl, info); |
0f9dd46c | 2355 | if (ret) |
dc89e982 | 2356 | kmem_cache_free(btrfs_free_space_cachep, info); |
96303081 | 2357 | out: |
34d52cb6 | 2358 | spin_unlock(&ctl->tree_lock); |
6226cb0a | 2359 | |
0f9dd46c | 2360 | if (ret) { |
ab8d0fc4 | 2361 | btrfs_crit(fs_info, "unable to add free space :%d", ret); |
b12d6869 | 2362 | ASSERT(ret != -EEXIST); |
0f9dd46c JB |
2363 | } |
2364 | ||
0f9dd46c JB |
2365 | return ret; |
2366 | } | |
2367 | ||
6226cb0a JB |
2368 | int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group, |
2369 | u64 offset, u64 bytes) | |
0f9dd46c | 2370 | { |
34d52cb6 | 2371 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0f9dd46c | 2372 | struct btrfs_free_space *info; |
b0175117 JB |
2373 | int ret; |
2374 | bool re_search = false; | |
0f9dd46c | 2375 | |
34d52cb6 | 2376 | spin_lock(&ctl->tree_lock); |
6226cb0a | 2377 | |
96303081 | 2378 | again: |
b0175117 | 2379 | ret = 0; |
bdb7d303 JB |
2380 | if (!bytes) |
2381 | goto out_lock; | |
2382 | ||
34d52cb6 | 2383 | info = tree_search_offset(ctl, offset, 0, 0); |
96303081 | 2384 | if (!info) { |
6606bb97 JB |
2385 | /* |
2386 | * oops didn't find an extent that matched the space we wanted | |
2387 | * to remove, look for a bitmap instead | |
2388 | */ | |
34d52cb6 | 2389 | info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), |
6606bb97 JB |
2390 | 1, 0); |
2391 | if (!info) { | |
b0175117 JB |
2392 | /* |
2393 | * If we found a partial bit of our free space in a | |
2394 | * bitmap but then couldn't find the other part this may | |
2395 | * be a problem, so WARN about it. | |
24a70313 | 2396 | */ |
b0175117 | 2397 | WARN_ON(re_search); |
6606bb97 JB |
2398 | goto out_lock; |
2399 | } | |
96303081 JB |
2400 | } |
2401 | ||
b0175117 | 2402 | re_search = false; |
bdb7d303 | 2403 | if (!info->bitmap) { |
34d52cb6 | 2404 | unlink_free_space(ctl, info); |
bdb7d303 JB |
2405 | if (offset == info->offset) { |
2406 | u64 to_free = min(bytes, info->bytes); | |
2407 | ||
2408 | info->bytes -= to_free; | |
2409 | info->offset += to_free; | |
2410 | if (info->bytes) { | |
2411 | ret = link_free_space(ctl, info); | |
2412 | WARN_ON(ret); | |
2413 | } else { | |
2414 | kmem_cache_free(btrfs_free_space_cachep, info); | |
2415 | } | |
0f9dd46c | 2416 | |
bdb7d303 JB |
2417 | offset += to_free; |
2418 | bytes -= to_free; | |
2419 | goto again; | |
2420 | } else { | |
2421 | u64 old_end = info->bytes + info->offset; | |
9b49c9b9 | 2422 | |
bdb7d303 | 2423 | info->bytes = offset - info->offset; |
34d52cb6 | 2424 | ret = link_free_space(ctl, info); |
96303081 JB |
2425 | WARN_ON(ret); |
2426 | if (ret) | |
2427 | goto out_lock; | |
96303081 | 2428 | |
bdb7d303 JB |
2429 | /* Not enough bytes in this entry to satisfy us */ |
2430 | if (old_end < offset + bytes) { | |
2431 | bytes -= old_end - offset; | |
2432 | offset = old_end; | |
2433 | goto again; | |
2434 | } else if (old_end == offset + bytes) { | |
2435 | /* all done */ | |
2436 | goto out_lock; | |
2437 | } | |
2438 | spin_unlock(&ctl->tree_lock); | |
2439 | ||
2440 | ret = btrfs_add_free_space(block_group, offset + bytes, | |
2441 | old_end - (offset + bytes)); | |
2442 | WARN_ON(ret); | |
2443 | goto out; | |
2444 | } | |
0f9dd46c | 2445 | } |
96303081 | 2446 | |
34d52cb6 | 2447 | ret = remove_from_bitmap(ctl, info, &offset, &bytes); |
b0175117 JB |
2448 | if (ret == -EAGAIN) { |
2449 | re_search = true; | |
96303081 | 2450 | goto again; |
b0175117 | 2451 | } |
96303081 | 2452 | out_lock: |
34d52cb6 | 2453 | spin_unlock(&ctl->tree_lock); |
0f9dd46c | 2454 | out: |
25179201 JB |
2455 | return ret; |
2456 | } | |
2457 | ||
0f9dd46c JB |
2458 | void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group, |
2459 | u64 bytes) | |
2460 | { | |
0b246afa | 2461 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
34d52cb6 | 2462 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0f9dd46c JB |
2463 | struct btrfs_free_space *info; |
2464 | struct rb_node *n; | |
2465 | int count = 0; | |
2466 | ||
34d52cb6 | 2467 | for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) { |
0f9dd46c | 2468 | info = rb_entry(n, struct btrfs_free_space, offset_index); |
f6175efa | 2469 | if (info->bytes >= bytes && !block_group->ro) |
0f9dd46c | 2470 | count++; |
0b246afa | 2471 | btrfs_crit(fs_info, "entry offset %llu, bytes %llu, bitmap %s", |
efe120a0 | 2472 | info->offset, info->bytes, |
96303081 | 2473 | (info->bitmap) ? "yes" : "no"); |
0f9dd46c | 2474 | } |
0b246afa | 2475 | btrfs_info(fs_info, "block group has cluster?: %s", |
96303081 | 2476 | list_empty(&block_group->cluster_list) ? "no" : "yes"); |
0b246afa | 2477 | btrfs_info(fs_info, |
efe120a0 | 2478 | "%d blocks of free space at or bigger than bytes is", count); |
0f9dd46c JB |
2479 | } |
2480 | ||
34d52cb6 | 2481 | void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group) |
0f9dd46c | 2482 | { |
0b246afa | 2483 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
34d52cb6 | 2484 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0f9dd46c | 2485 | |
34d52cb6 | 2486 | spin_lock_init(&ctl->tree_lock); |
0b246afa | 2487 | ctl->unit = fs_info->sectorsize; |
34d52cb6 LZ |
2488 | ctl->start = block_group->key.objectid; |
2489 | ctl->private = block_group; | |
2490 | ctl->op = &free_space_op; | |
55507ce3 FM |
2491 | INIT_LIST_HEAD(&ctl->trimming_ranges); |
2492 | mutex_init(&ctl->cache_writeout_mutex); | |
0f9dd46c | 2493 | |
34d52cb6 LZ |
2494 | /* |
2495 | * we only want to have 32k of ram per block group for keeping | |
2496 | * track of free space, and if we pass 1/2 of that we want to | |
2497 | * start converting things over to using bitmaps | |
2498 | */ | |
ee22184b | 2499 | ctl->extents_thresh = (SZ_32K / 2) / sizeof(struct btrfs_free_space); |
0f9dd46c JB |
2500 | } |
2501 | ||
fa9c0d79 CM |
2502 | /* |
2503 | * for a given cluster, put all of its extents back into the free | |
2504 | * space cache. If the block group passed doesn't match the block group | |
2505 | * pointed to by the cluster, someone else raced in and freed the | |
2506 | * cluster already. In that case, we just return without changing anything | |
2507 | */ | |
2508 | static int | |
2509 | __btrfs_return_cluster_to_free_space( | |
2510 | struct btrfs_block_group_cache *block_group, | |
2511 | struct btrfs_free_cluster *cluster) | |
2512 | { | |
34d52cb6 | 2513 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
fa9c0d79 CM |
2514 | struct btrfs_free_space *entry; |
2515 | struct rb_node *node; | |
2516 | ||
2517 | spin_lock(&cluster->lock); | |
2518 | if (cluster->block_group != block_group) | |
2519 | goto out; | |
2520 | ||
96303081 | 2521 | cluster->block_group = NULL; |
fa9c0d79 | 2522 | cluster->window_start = 0; |
96303081 | 2523 | list_del_init(&cluster->block_group_list); |
96303081 | 2524 | |
fa9c0d79 | 2525 | node = rb_first(&cluster->root); |
96303081 | 2526 | while (node) { |
4e69b598 JB |
2527 | bool bitmap; |
2528 | ||
fa9c0d79 CM |
2529 | entry = rb_entry(node, struct btrfs_free_space, offset_index); |
2530 | node = rb_next(&entry->offset_index); | |
2531 | rb_erase(&entry->offset_index, &cluster->root); | |
20005523 | 2532 | RB_CLEAR_NODE(&entry->offset_index); |
4e69b598 JB |
2533 | |
2534 | bitmap = (entry->bitmap != NULL); | |
20005523 | 2535 | if (!bitmap) { |
34d52cb6 | 2536 | try_merge_free_space(ctl, entry, false); |
20005523 FM |
2537 | steal_from_bitmap(ctl, entry, false); |
2538 | } | |
34d52cb6 | 2539 | tree_insert_offset(&ctl->free_space_offset, |
4e69b598 | 2540 | entry->offset, &entry->offset_index, bitmap); |
fa9c0d79 | 2541 | } |
6bef4d31 | 2542 | cluster->root = RB_ROOT; |
96303081 | 2543 | |
fa9c0d79 CM |
2544 | out: |
2545 | spin_unlock(&cluster->lock); | |
96303081 | 2546 | btrfs_put_block_group(block_group); |
fa9c0d79 CM |
2547 | return 0; |
2548 | } | |
2549 | ||
48a3b636 ES |
2550 | static void __btrfs_remove_free_space_cache_locked( |
2551 | struct btrfs_free_space_ctl *ctl) | |
0f9dd46c JB |
2552 | { |
2553 | struct btrfs_free_space *info; | |
2554 | struct rb_node *node; | |
581bb050 | 2555 | |
581bb050 LZ |
2556 | while ((node = rb_last(&ctl->free_space_offset)) != NULL) { |
2557 | info = rb_entry(node, struct btrfs_free_space, offset_index); | |
9b90f513 JB |
2558 | if (!info->bitmap) { |
2559 | unlink_free_space(ctl, info); | |
2560 | kmem_cache_free(btrfs_free_space_cachep, info); | |
2561 | } else { | |
2562 | free_bitmap(ctl, info); | |
2563 | } | |
351810c1 DS |
2564 | |
2565 | cond_resched_lock(&ctl->tree_lock); | |
581bb050 | 2566 | } |
09655373 CM |
2567 | } |
2568 | ||
2569 | void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl) | |
2570 | { | |
2571 | spin_lock(&ctl->tree_lock); | |
2572 | __btrfs_remove_free_space_cache_locked(ctl); | |
581bb050 LZ |
2573 | spin_unlock(&ctl->tree_lock); |
2574 | } | |
2575 | ||
2576 | void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group) | |
2577 | { | |
2578 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
fa9c0d79 | 2579 | struct btrfs_free_cluster *cluster; |
96303081 | 2580 | struct list_head *head; |
0f9dd46c | 2581 | |
34d52cb6 | 2582 | spin_lock(&ctl->tree_lock); |
96303081 JB |
2583 | while ((head = block_group->cluster_list.next) != |
2584 | &block_group->cluster_list) { | |
2585 | cluster = list_entry(head, struct btrfs_free_cluster, | |
2586 | block_group_list); | |
fa9c0d79 CM |
2587 | |
2588 | WARN_ON(cluster->block_group != block_group); | |
2589 | __btrfs_return_cluster_to_free_space(block_group, cluster); | |
351810c1 DS |
2590 | |
2591 | cond_resched_lock(&ctl->tree_lock); | |
fa9c0d79 | 2592 | } |
09655373 | 2593 | __btrfs_remove_free_space_cache_locked(ctl); |
34d52cb6 | 2594 | spin_unlock(&ctl->tree_lock); |
fa9c0d79 | 2595 | |
0f9dd46c JB |
2596 | } |
2597 | ||
6226cb0a | 2598 | u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group, |
a4820398 MX |
2599 | u64 offset, u64 bytes, u64 empty_size, |
2600 | u64 *max_extent_size) | |
0f9dd46c | 2601 | { |
34d52cb6 | 2602 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
6226cb0a | 2603 | struct btrfs_free_space *entry = NULL; |
96303081 | 2604 | u64 bytes_search = bytes + empty_size; |
6226cb0a | 2605 | u64 ret = 0; |
53b381b3 DW |
2606 | u64 align_gap = 0; |
2607 | u64 align_gap_len = 0; | |
0f9dd46c | 2608 | |
34d52cb6 | 2609 | spin_lock(&ctl->tree_lock); |
53b381b3 | 2610 | entry = find_free_space(ctl, &offset, &bytes_search, |
a4820398 | 2611 | block_group->full_stripe_len, max_extent_size); |
6226cb0a | 2612 | if (!entry) |
96303081 JB |
2613 | goto out; |
2614 | ||
2615 | ret = offset; | |
2616 | if (entry->bitmap) { | |
34d52cb6 | 2617 | bitmap_clear_bits(ctl, entry, offset, bytes); |
edf6e2d1 | 2618 | if (!entry->bytes) |
34d52cb6 | 2619 | free_bitmap(ctl, entry); |
96303081 | 2620 | } else { |
34d52cb6 | 2621 | unlink_free_space(ctl, entry); |
53b381b3 DW |
2622 | align_gap_len = offset - entry->offset; |
2623 | align_gap = entry->offset; | |
2624 | ||
2625 | entry->offset = offset + bytes; | |
2626 | WARN_ON(entry->bytes < bytes + align_gap_len); | |
2627 | ||
2628 | entry->bytes -= bytes + align_gap_len; | |
6226cb0a | 2629 | if (!entry->bytes) |
dc89e982 | 2630 | kmem_cache_free(btrfs_free_space_cachep, entry); |
6226cb0a | 2631 | else |
34d52cb6 | 2632 | link_free_space(ctl, entry); |
6226cb0a | 2633 | } |
96303081 | 2634 | out: |
34d52cb6 | 2635 | spin_unlock(&ctl->tree_lock); |
817d52f8 | 2636 | |
53b381b3 | 2637 | if (align_gap_len) |
ab8d0fc4 JM |
2638 | __btrfs_add_free_space(block_group->fs_info, ctl, |
2639 | align_gap, align_gap_len); | |
0f9dd46c JB |
2640 | return ret; |
2641 | } | |
fa9c0d79 CM |
2642 | |
2643 | /* | |
2644 | * given a cluster, put all of its extents back into the free space | |
2645 | * cache. If a block group is passed, this function will only free | |
2646 | * a cluster that belongs to the passed block group. | |
2647 | * | |
2648 | * Otherwise, it'll get a reference on the block group pointed to by the | |
2649 | * cluster and remove the cluster from it. | |
2650 | */ | |
2651 | int btrfs_return_cluster_to_free_space( | |
2652 | struct btrfs_block_group_cache *block_group, | |
2653 | struct btrfs_free_cluster *cluster) | |
2654 | { | |
34d52cb6 | 2655 | struct btrfs_free_space_ctl *ctl; |
fa9c0d79 CM |
2656 | int ret; |
2657 | ||
2658 | /* first, get a safe pointer to the block group */ | |
2659 | spin_lock(&cluster->lock); | |
2660 | if (!block_group) { | |
2661 | block_group = cluster->block_group; | |
2662 | if (!block_group) { | |
2663 | spin_unlock(&cluster->lock); | |
2664 | return 0; | |
2665 | } | |
2666 | } else if (cluster->block_group != block_group) { | |
2667 | /* someone else has already freed it don't redo their work */ | |
2668 | spin_unlock(&cluster->lock); | |
2669 | return 0; | |
2670 | } | |
2671 | atomic_inc(&block_group->count); | |
2672 | spin_unlock(&cluster->lock); | |
2673 | ||
34d52cb6 LZ |
2674 | ctl = block_group->free_space_ctl; |
2675 | ||
fa9c0d79 | 2676 | /* now return any extents the cluster had on it */ |
34d52cb6 | 2677 | spin_lock(&ctl->tree_lock); |
fa9c0d79 | 2678 | ret = __btrfs_return_cluster_to_free_space(block_group, cluster); |
34d52cb6 | 2679 | spin_unlock(&ctl->tree_lock); |
fa9c0d79 CM |
2680 | |
2681 | /* finally drop our ref */ | |
2682 | btrfs_put_block_group(block_group); | |
2683 | return ret; | |
2684 | } | |
2685 | ||
96303081 JB |
2686 | static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group, |
2687 | struct btrfs_free_cluster *cluster, | |
4e69b598 | 2688 | struct btrfs_free_space *entry, |
a4820398 MX |
2689 | u64 bytes, u64 min_start, |
2690 | u64 *max_extent_size) | |
96303081 | 2691 | { |
34d52cb6 | 2692 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
96303081 JB |
2693 | int err; |
2694 | u64 search_start = cluster->window_start; | |
2695 | u64 search_bytes = bytes; | |
2696 | u64 ret = 0; | |
2697 | ||
96303081 JB |
2698 | search_start = min_start; |
2699 | search_bytes = bytes; | |
2700 | ||
0584f718 | 2701 | err = search_bitmap(ctl, entry, &search_start, &search_bytes, true); |
a4820398 MX |
2702 | if (err) { |
2703 | if (search_bytes > *max_extent_size) | |
2704 | *max_extent_size = search_bytes; | |
4e69b598 | 2705 | return 0; |
a4820398 | 2706 | } |
96303081 JB |
2707 | |
2708 | ret = search_start; | |
bb3ac5a4 | 2709 | __bitmap_clear_bits(ctl, entry, ret, bytes); |
96303081 JB |
2710 | |
2711 | return ret; | |
2712 | } | |
2713 | ||
fa9c0d79 CM |
2714 | /* |
2715 | * given a cluster, try to allocate 'bytes' from it, returns 0 | |
2716 | * if it couldn't find anything suitably large, or a logical disk offset | |
2717 | * if things worked out | |
2718 | */ | |
2719 | u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group, | |
2720 | struct btrfs_free_cluster *cluster, u64 bytes, | |
a4820398 | 2721 | u64 min_start, u64 *max_extent_size) |
fa9c0d79 | 2722 | { |
34d52cb6 | 2723 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
fa9c0d79 CM |
2724 | struct btrfs_free_space *entry = NULL; |
2725 | struct rb_node *node; | |
2726 | u64 ret = 0; | |
2727 | ||
2728 | spin_lock(&cluster->lock); | |
2729 | if (bytes > cluster->max_size) | |
2730 | goto out; | |
2731 | ||
2732 | if (cluster->block_group != block_group) | |
2733 | goto out; | |
2734 | ||
2735 | node = rb_first(&cluster->root); | |
2736 | if (!node) | |
2737 | goto out; | |
2738 | ||
2739 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
67871254 | 2740 | while (1) { |
a4820398 MX |
2741 | if (entry->bytes < bytes && entry->bytes > *max_extent_size) |
2742 | *max_extent_size = entry->bytes; | |
2743 | ||
4e69b598 JB |
2744 | if (entry->bytes < bytes || |
2745 | (!entry->bitmap && entry->offset < min_start)) { | |
fa9c0d79 CM |
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 | } | |
fa9c0d79 | 2753 | |
4e69b598 JB |
2754 | if (entry->bitmap) { |
2755 | ret = btrfs_alloc_from_bitmap(block_group, | |
2756 | cluster, entry, bytes, | |
a4820398 MX |
2757 | cluster->window_start, |
2758 | max_extent_size); | |
4e69b598 | 2759 | if (ret == 0) { |
4e69b598 JB |
2760 | node = rb_next(&entry->offset_index); |
2761 | if (!node) | |
2762 | break; | |
2763 | entry = rb_entry(node, struct btrfs_free_space, | |
2764 | offset_index); | |
2765 | continue; | |
2766 | } | |
9b230628 | 2767 | cluster->window_start += bytes; |
4e69b598 | 2768 | } else { |
4e69b598 JB |
2769 | ret = entry->offset; |
2770 | ||
2771 | entry->offset += bytes; | |
2772 | entry->bytes -= bytes; | |
2773 | } | |
fa9c0d79 | 2774 | |
5e71b5d5 | 2775 | if (entry->bytes == 0) |
fa9c0d79 | 2776 | rb_erase(&entry->offset_index, &cluster->root); |
fa9c0d79 CM |
2777 | break; |
2778 | } | |
2779 | out: | |
2780 | spin_unlock(&cluster->lock); | |
96303081 | 2781 | |
5e71b5d5 LZ |
2782 | if (!ret) |
2783 | return 0; | |
2784 | ||
34d52cb6 | 2785 | spin_lock(&ctl->tree_lock); |
5e71b5d5 | 2786 | |
34d52cb6 | 2787 | ctl->free_space -= bytes; |
5e71b5d5 | 2788 | if (entry->bytes == 0) { |
34d52cb6 | 2789 | ctl->free_extents--; |
4e69b598 JB |
2790 | if (entry->bitmap) { |
2791 | kfree(entry->bitmap); | |
34d52cb6 LZ |
2792 | ctl->total_bitmaps--; |
2793 | ctl->op->recalc_thresholds(ctl); | |
4e69b598 | 2794 | } |
dc89e982 | 2795 | kmem_cache_free(btrfs_free_space_cachep, entry); |
5e71b5d5 LZ |
2796 | } |
2797 | ||
34d52cb6 | 2798 | spin_unlock(&ctl->tree_lock); |
5e71b5d5 | 2799 | |
fa9c0d79 CM |
2800 | return ret; |
2801 | } | |
2802 | ||
96303081 JB |
2803 | static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group, |
2804 | struct btrfs_free_space *entry, | |
2805 | struct btrfs_free_cluster *cluster, | |
1bb91902 AO |
2806 | u64 offset, u64 bytes, |
2807 | u64 cont1_bytes, u64 min_bytes) | |
96303081 | 2808 | { |
34d52cb6 | 2809 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
96303081 JB |
2810 | unsigned long next_zero; |
2811 | unsigned long i; | |
1bb91902 AO |
2812 | unsigned long want_bits; |
2813 | unsigned long min_bits; | |
96303081 | 2814 | unsigned long found_bits; |
cef40483 | 2815 | unsigned long max_bits = 0; |
96303081 JB |
2816 | unsigned long start = 0; |
2817 | unsigned long total_found = 0; | |
4e69b598 | 2818 | int ret; |
96303081 | 2819 | |
96009762 | 2820 | i = offset_to_bit(entry->offset, ctl->unit, |
96303081 | 2821 | max_t(u64, offset, entry->offset)); |
96009762 WSH |
2822 | want_bits = bytes_to_bits(bytes, ctl->unit); |
2823 | min_bits = bytes_to_bits(min_bytes, ctl->unit); | |
96303081 | 2824 | |
cef40483 JB |
2825 | /* |
2826 | * Don't bother looking for a cluster in this bitmap if it's heavily | |
2827 | * fragmented. | |
2828 | */ | |
2829 | if (entry->max_extent_size && | |
2830 | entry->max_extent_size < cont1_bytes) | |
2831 | return -ENOSPC; | |
96303081 JB |
2832 | again: |
2833 | found_bits = 0; | |
ebb3dad4 | 2834 | for_each_set_bit_from(i, entry->bitmap, BITS_PER_BITMAP) { |
96303081 JB |
2835 | next_zero = find_next_zero_bit(entry->bitmap, |
2836 | BITS_PER_BITMAP, i); | |
1bb91902 | 2837 | if (next_zero - i >= min_bits) { |
96303081 | 2838 | found_bits = next_zero - i; |
cef40483 JB |
2839 | if (found_bits > max_bits) |
2840 | max_bits = found_bits; | |
96303081 JB |
2841 | break; |
2842 | } | |
cef40483 JB |
2843 | if (next_zero - i > max_bits) |
2844 | max_bits = next_zero - i; | |
96303081 JB |
2845 | i = next_zero; |
2846 | } | |
2847 | ||
cef40483 JB |
2848 | if (!found_bits) { |
2849 | entry->max_extent_size = (u64)max_bits * ctl->unit; | |
4e69b598 | 2850 | return -ENOSPC; |
cef40483 | 2851 | } |
96303081 | 2852 | |
1bb91902 | 2853 | if (!total_found) { |
96303081 | 2854 | start = i; |
b78d09bc | 2855 | cluster->max_size = 0; |
96303081 JB |
2856 | } |
2857 | ||
2858 | total_found += found_bits; | |
2859 | ||
96009762 WSH |
2860 | if (cluster->max_size < found_bits * ctl->unit) |
2861 | cluster->max_size = found_bits * ctl->unit; | |
96303081 | 2862 | |
1bb91902 AO |
2863 | if (total_found < want_bits || cluster->max_size < cont1_bytes) { |
2864 | i = next_zero + 1; | |
96303081 JB |
2865 | goto again; |
2866 | } | |
2867 | ||
96009762 | 2868 | cluster->window_start = start * ctl->unit + entry->offset; |
34d52cb6 | 2869 | rb_erase(&entry->offset_index, &ctl->free_space_offset); |
4e69b598 JB |
2870 | ret = tree_insert_offset(&cluster->root, entry->offset, |
2871 | &entry->offset_index, 1); | |
b12d6869 | 2872 | ASSERT(!ret); /* -EEXIST; Logic error */ |
96303081 | 2873 | |
3f7de037 | 2874 | trace_btrfs_setup_cluster(block_group, cluster, |
96009762 | 2875 | total_found * ctl->unit, 1); |
96303081 JB |
2876 | return 0; |
2877 | } | |
2878 | ||
4e69b598 JB |
2879 | /* |
2880 | * This searches the block group for just extents to fill the cluster with. | |
1bb91902 AO |
2881 | * Try to find a cluster with at least bytes total bytes, at least one |
2882 | * extent of cont1_bytes, and other clusters of at least min_bytes. | |
4e69b598 | 2883 | */ |
3de85bb9 JB |
2884 | static noinline int |
2885 | setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group, | |
2886 | struct btrfs_free_cluster *cluster, | |
2887 | struct list_head *bitmaps, u64 offset, u64 bytes, | |
1bb91902 | 2888 | u64 cont1_bytes, u64 min_bytes) |
4e69b598 | 2889 | { |
34d52cb6 | 2890 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
4e69b598 JB |
2891 | struct btrfs_free_space *first = NULL; |
2892 | struct btrfs_free_space *entry = NULL; | |
4e69b598 JB |
2893 | struct btrfs_free_space *last; |
2894 | struct rb_node *node; | |
4e69b598 JB |
2895 | u64 window_free; |
2896 | u64 max_extent; | |
3f7de037 | 2897 | u64 total_size = 0; |
4e69b598 | 2898 | |
34d52cb6 | 2899 | entry = tree_search_offset(ctl, offset, 0, 1); |
4e69b598 JB |
2900 | if (!entry) |
2901 | return -ENOSPC; | |
2902 | ||
2903 | /* | |
2904 | * We don't want bitmaps, so just move along until we find a normal | |
2905 | * extent entry. | |
2906 | */ | |
1bb91902 AO |
2907 | while (entry->bitmap || entry->bytes < min_bytes) { |
2908 | if (entry->bitmap && list_empty(&entry->list)) | |
86d4a77b | 2909 | list_add_tail(&entry->list, bitmaps); |
4e69b598 JB |
2910 | node = rb_next(&entry->offset_index); |
2911 | if (!node) | |
2912 | return -ENOSPC; | |
2913 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
2914 | } | |
2915 | ||
4e69b598 JB |
2916 | window_free = entry->bytes; |
2917 | max_extent = entry->bytes; | |
2918 | first = entry; | |
2919 | last = entry; | |
4e69b598 | 2920 | |
1bb91902 AO |
2921 | for (node = rb_next(&entry->offset_index); node; |
2922 | node = rb_next(&entry->offset_index)) { | |
4e69b598 JB |
2923 | entry = rb_entry(node, struct btrfs_free_space, offset_index); |
2924 | ||
86d4a77b JB |
2925 | if (entry->bitmap) { |
2926 | if (list_empty(&entry->list)) | |
2927 | list_add_tail(&entry->list, bitmaps); | |
4e69b598 | 2928 | continue; |
86d4a77b JB |
2929 | } |
2930 | ||
1bb91902 AO |
2931 | if (entry->bytes < min_bytes) |
2932 | continue; | |
2933 | ||
2934 | last = entry; | |
2935 | window_free += entry->bytes; | |
2936 | if (entry->bytes > max_extent) | |
4e69b598 | 2937 | max_extent = entry->bytes; |
4e69b598 JB |
2938 | } |
2939 | ||
1bb91902 AO |
2940 | if (window_free < bytes || max_extent < cont1_bytes) |
2941 | return -ENOSPC; | |
2942 | ||
4e69b598 JB |
2943 | cluster->window_start = first->offset; |
2944 | ||
2945 | node = &first->offset_index; | |
2946 | ||
2947 | /* | |
2948 | * now we've found our entries, pull them out of the free space | |
2949 | * cache and put them into the cluster rbtree | |
2950 | */ | |
2951 | do { | |
2952 | int ret; | |
2953 | ||
2954 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
2955 | node = rb_next(&entry->offset_index); | |
1bb91902 | 2956 | if (entry->bitmap || entry->bytes < min_bytes) |
4e69b598 JB |
2957 | continue; |
2958 | ||
34d52cb6 | 2959 | rb_erase(&entry->offset_index, &ctl->free_space_offset); |
4e69b598 JB |
2960 | ret = tree_insert_offset(&cluster->root, entry->offset, |
2961 | &entry->offset_index, 0); | |
3f7de037 | 2962 | total_size += entry->bytes; |
b12d6869 | 2963 | ASSERT(!ret); /* -EEXIST; Logic error */ |
4e69b598 JB |
2964 | } while (node && entry != last); |
2965 | ||
2966 | cluster->max_size = max_extent; | |
3f7de037 | 2967 | trace_btrfs_setup_cluster(block_group, cluster, total_size, 0); |
4e69b598 JB |
2968 | return 0; |
2969 | } | |
2970 | ||
2971 | /* | |
2972 | * This specifically looks for bitmaps that may work in the cluster, we assume | |
2973 | * that we have already failed to find extents that will work. | |
2974 | */ | |
3de85bb9 JB |
2975 | static noinline int |
2976 | setup_cluster_bitmap(struct btrfs_block_group_cache *block_group, | |
2977 | struct btrfs_free_cluster *cluster, | |
2978 | struct list_head *bitmaps, u64 offset, u64 bytes, | |
1bb91902 | 2979 | u64 cont1_bytes, u64 min_bytes) |
4e69b598 | 2980 | { |
34d52cb6 | 2981 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
1b9b922a | 2982 | struct btrfs_free_space *entry = NULL; |
4e69b598 | 2983 | int ret = -ENOSPC; |
0f0fbf1d | 2984 | u64 bitmap_offset = offset_to_bitmap(ctl, offset); |
4e69b598 | 2985 | |
34d52cb6 | 2986 | if (ctl->total_bitmaps == 0) |
4e69b598 JB |
2987 | return -ENOSPC; |
2988 | ||
0f0fbf1d LZ |
2989 | /* |
2990 | * The bitmap that covers offset won't be in the list unless offset | |
2991 | * is just its start offset. | |
2992 | */ | |
1b9b922a CM |
2993 | if (!list_empty(bitmaps)) |
2994 | entry = list_first_entry(bitmaps, struct btrfs_free_space, list); | |
2995 | ||
2996 | if (!entry || entry->offset != bitmap_offset) { | |
0f0fbf1d LZ |
2997 | entry = tree_search_offset(ctl, bitmap_offset, 1, 0); |
2998 | if (entry && list_empty(&entry->list)) | |
2999 | list_add(&entry->list, bitmaps); | |
3000 | } | |
3001 | ||
86d4a77b | 3002 | list_for_each_entry(entry, bitmaps, list) { |
357b9784 | 3003 | if (entry->bytes < bytes) |
86d4a77b JB |
3004 | continue; |
3005 | ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset, | |
1bb91902 | 3006 | bytes, cont1_bytes, min_bytes); |
86d4a77b JB |
3007 | if (!ret) |
3008 | return 0; | |
3009 | } | |
3010 | ||
3011 | /* | |
52621cb6 LZ |
3012 | * The bitmaps list has all the bitmaps that record free space |
3013 | * starting after offset, so no more search is required. | |
86d4a77b | 3014 | */ |
52621cb6 | 3015 | return -ENOSPC; |
4e69b598 JB |
3016 | } |
3017 | ||
fa9c0d79 CM |
3018 | /* |
3019 | * here we try to find a cluster of blocks in a block group. The goal | |
1bb91902 | 3020 | * is to find at least bytes+empty_size. |
fa9c0d79 CM |
3021 | * We might not find them all in one contiguous area. |
3022 | * | |
3023 | * returns zero and sets up cluster if things worked out, otherwise | |
3024 | * it returns -enospc | |
3025 | */ | |
2ff7e61e | 3026 | int btrfs_find_space_cluster(struct btrfs_fs_info *fs_info, |
fa9c0d79 CM |
3027 | struct btrfs_block_group_cache *block_group, |
3028 | struct btrfs_free_cluster *cluster, | |
3029 | u64 offset, u64 bytes, u64 empty_size) | |
3030 | { | |
34d52cb6 | 3031 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
86d4a77b | 3032 | struct btrfs_free_space *entry, *tmp; |
52621cb6 | 3033 | LIST_HEAD(bitmaps); |
fa9c0d79 | 3034 | u64 min_bytes; |
1bb91902 | 3035 | u64 cont1_bytes; |
fa9c0d79 CM |
3036 | int ret; |
3037 | ||
1bb91902 AO |
3038 | /* |
3039 | * Choose the minimum extent size we'll require for this | |
3040 | * cluster. For SSD_SPREAD, don't allow any fragmentation. | |
3041 | * For metadata, allow allocates with smaller extents. For | |
3042 | * data, keep it dense. | |
3043 | */ | |
0b246afa | 3044 | if (btrfs_test_opt(fs_info, SSD_SPREAD)) { |
1bb91902 | 3045 | cont1_bytes = min_bytes = bytes + empty_size; |
451d7585 | 3046 | } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) { |
1bb91902 | 3047 | cont1_bytes = bytes; |
0b246afa | 3048 | min_bytes = fs_info->sectorsize; |
1bb91902 AO |
3049 | } else { |
3050 | cont1_bytes = max(bytes, (bytes + empty_size) >> 2); | |
0b246afa | 3051 | min_bytes = fs_info->sectorsize; |
1bb91902 | 3052 | } |
fa9c0d79 | 3053 | |
34d52cb6 | 3054 | spin_lock(&ctl->tree_lock); |
7d0d2e8e JB |
3055 | |
3056 | /* | |
3057 | * If we know we don't have enough space to make a cluster don't even | |
3058 | * bother doing all the work to try and find one. | |
3059 | */ | |
1bb91902 | 3060 | if (ctl->free_space < bytes) { |
34d52cb6 | 3061 | spin_unlock(&ctl->tree_lock); |
7d0d2e8e JB |
3062 | return -ENOSPC; |
3063 | } | |
3064 | ||
fa9c0d79 CM |
3065 | spin_lock(&cluster->lock); |
3066 | ||
3067 | /* someone already found a cluster, hooray */ | |
3068 | if (cluster->block_group) { | |
3069 | ret = 0; | |
3070 | goto out; | |
3071 | } | |
fa9c0d79 | 3072 | |
3f7de037 JB |
3073 | trace_btrfs_find_cluster(block_group, offset, bytes, empty_size, |
3074 | min_bytes); | |
3075 | ||
86d4a77b | 3076 | ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset, |
1bb91902 AO |
3077 | bytes + empty_size, |
3078 | cont1_bytes, min_bytes); | |
4e69b598 | 3079 | if (ret) |
86d4a77b | 3080 | ret = setup_cluster_bitmap(block_group, cluster, &bitmaps, |
1bb91902 AO |
3081 | offset, bytes + empty_size, |
3082 | cont1_bytes, min_bytes); | |
86d4a77b JB |
3083 | |
3084 | /* Clear our temporary list */ | |
3085 | list_for_each_entry_safe(entry, tmp, &bitmaps, list) | |
3086 | list_del_init(&entry->list); | |
fa9c0d79 | 3087 | |
4e69b598 JB |
3088 | if (!ret) { |
3089 | atomic_inc(&block_group->count); | |
3090 | list_add_tail(&cluster->block_group_list, | |
3091 | &block_group->cluster_list); | |
3092 | cluster->block_group = block_group; | |
3f7de037 JB |
3093 | } else { |
3094 | trace_btrfs_failed_cluster_setup(block_group); | |
fa9c0d79 | 3095 | } |
fa9c0d79 CM |
3096 | out: |
3097 | spin_unlock(&cluster->lock); | |
34d52cb6 | 3098 | spin_unlock(&ctl->tree_lock); |
fa9c0d79 CM |
3099 | |
3100 | return ret; | |
3101 | } | |
3102 | ||
3103 | /* | |
3104 | * simple code to zero out a cluster | |
3105 | */ | |
3106 | void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster) | |
3107 | { | |
3108 | spin_lock_init(&cluster->lock); | |
3109 | spin_lock_init(&cluster->refill_lock); | |
6bef4d31 | 3110 | cluster->root = RB_ROOT; |
fa9c0d79 | 3111 | cluster->max_size = 0; |
c759c4e1 | 3112 | cluster->fragmented = false; |
fa9c0d79 CM |
3113 | INIT_LIST_HEAD(&cluster->block_group_list); |
3114 | cluster->block_group = NULL; | |
3115 | } | |
3116 | ||
7fe1e641 LZ |
3117 | static int do_trimming(struct btrfs_block_group_cache *block_group, |
3118 | u64 *total_trimmed, u64 start, u64 bytes, | |
55507ce3 FM |
3119 | u64 reserved_start, u64 reserved_bytes, |
3120 | struct btrfs_trim_range *trim_entry) | |
f7039b1d | 3121 | { |
7fe1e641 | 3122 | struct btrfs_space_info *space_info = block_group->space_info; |
f7039b1d | 3123 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
55507ce3 | 3124 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
7fe1e641 LZ |
3125 | int ret; |
3126 | int update = 0; | |
3127 | u64 trimmed = 0; | |
f7039b1d | 3128 | |
7fe1e641 LZ |
3129 | spin_lock(&space_info->lock); |
3130 | spin_lock(&block_group->lock); | |
3131 | if (!block_group->ro) { | |
3132 | block_group->reserved += reserved_bytes; | |
3133 | space_info->bytes_reserved += reserved_bytes; | |
3134 | update = 1; | |
3135 | } | |
3136 | spin_unlock(&block_group->lock); | |
3137 | spin_unlock(&space_info->lock); | |
3138 | ||
2ff7e61e | 3139 | ret = btrfs_discard_extent(fs_info, start, bytes, &trimmed); |
7fe1e641 LZ |
3140 | if (!ret) |
3141 | *total_trimmed += trimmed; | |
3142 | ||
55507ce3 | 3143 | mutex_lock(&ctl->cache_writeout_mutex); |
7fe1e641 | 3144 | btrfs_add_free_space(block_group, reserved_start, reserved_bytes); |
55507ce3 FM |
3145 | list_del(&trim_entry->list); |
3146 | mutex_unlock(&ctl->cache_writeout_mutex); | |
7fe1e641 LZ |
3147 | |
3148 | if (update) { | |
3149 | spin_lock(&space_info->lock); | |
3150 | spin_lock(&block_group->lock); | |
3151 | if (block_group->ro) | |
3152 | space_info->bytes_readonly += reserved_bytes; | |
3153 | block_group->reserved -= reserved_bytes; | |
3154 | space_info->bytes_reserved -= reserved_bytes; | |
3155 | spin_unlock(&space_info->lock); | |
3156 | spin_unlock(&block_group->lock); | |
3157 | } | |
3158 | ||
3159 | return ret; | |
3160 | } | |
3161 | ||
3162 | static int trim_no_bitmap(struct btrfs_block_group_cache *block_group, | |
3163 | u64 *total_trimmed, u64 start, u64 end, u64 minlen) | |
3164 | { | |
3165 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
3166 | struct btrfs_free_space *entry; | |
3167 | struct rb_node *node; | |
3168 | int ret = 0; | |
3169 | u64 extent_start; | |
3170 | u64 extent_bytes; | |
3171 | u64 bytes; | |
f7039b1d LD |
3172 | |
3173 | while (start < end) { | |
55507ce3 FM |
3174 | struct btrfs_trim_range trim_entry; |
3175 | ||
3176 | mutex_lock(&ctl->cache_writeout_mutex); | |
34d52cb6 | 3177 | spin_lock(&ctl->tree_lock); |
f7039b1d | 3178 | |
34d52cb6 LZ |
3179 | if (ctl->free_space < minlen) { |
3180 | spin_unlock(&ctl->tree_lock); | |
55507ce3 | 3181 | mutex_unlock(&ctl->cache_writeout_mutex); |
f7039b1d LD |
3182 | break; |
3183 | } | |
3184 | ||
34d52cb6 | 3185 | entry = tree_search_offset(ctl, start, 0, 1); |
7fe1e641 | 3186 | if (!entry) { |
34d52cb6 | 3187 | spin_unlock(&ctl->tree_lock); |
55507ce3 | 3188 | mutex_unlock(&ctl->cache_writeout_mutex); |
f7039b1d LD |
3189 | break; |
3190 | } | |
3191 | ||
7fe1e641 LZ |
3192 | /* skip bitmaps */ |
3193 | while (entry->bitmap) { | |
3194 | node = rb_next(&entry->offset_index); | |
3195 | if (!node) { | |
34d52cb6 | 3196 | spin_unlock(&ctl->tree_lock); |
55507ce3 | 3197 | mutex_unlock(&ctl->cache_writeout_mutex); |
7fe1e641 | 3198 | goto out; |
f7039b1d | 3199 | } |
7fe1e641 LZ |
3200 | entry = rb_entry(node, struct btrfs_free_space, |
3201 | offset_index); | |
f7039b1d LD |
3202 | } |
3203 | ||
7fe1e641 LZ |
3204 | if (entry->offset >= end) { |
3205 | spin_unlock(&ctl->tree_lock); | |
55507ce3 | 3206 | mutex_unlock(&ctl->cache_writeout_mutex); |
7fe1e641 | 3207 | break; |
f7039b1d LD |
3208 | } |
3209 | ||
7fe1e641 LZ |
3210 | extent_start = entry->offset; |
3211 | extent_bytes = entry->bytes; | |
3212 | start = max(start, extent_start); | |
3213 | bytes = min(extent_start + extent_bytes, end) - start; | |
3214 | if (bytes < minlen) { | |
3215 | spin_unlock(&ctl->tree_lock); | |
55507ce3 | 3216 | mutex_unlock(&ctl->cache_writeout_mutex); |
7fe1e641 | 3217 | goto next; |
f7039b1d LD |
3218 | } |
3219 | ||
7fe1e641 LZ |
3220 | unlink_free_space(ctl, entry); |
3221 | kmem_cache_free(btrfs_free_space_cachep, entry); | |
3222 | ||
34d52cb6 | 3223 | spin_unlock(&ctl->tree_lock); |
55507ce3 FM |
3224 | trim_entry.start = extent_start; |
3225 | trim_entry.bytes = extent_bytes; | |
3226 | list_add_tail(&trim_entry.list, &ctl->trimming_ranges); | |
3227 | mutex_unlock(&ctl->cache_writeout_mutex); | |
f7039b1d | 3228 | |
7fe1e641 | 3229 | ret = do_trimming(block_group, total_trimmed, start, bytes, |
55507ce3 | 3230 | extent_start, extent_bytes, &trim_entry); |
7fe1e641 LZ |
3231 | if (ret) |
3232 | break; | |
3233 | next: | |
3234 | start += bytes; | |
f7039b1d | 3235 | |
7fe1e641 LZ |
3236 | if (fatal_signal_pending(current)) { |
3237 | ret = -ERESTARTSYS; | |
3238 | break; | |
3239 | } | |
3240 | ||
3241 | cond_resched(); | |
3242 | } | |
3243 | out: | |
3244 | return ret; | |
3245 | } | |
3246 | ||
3247 | static int trim_bitmaps(struct btrfs_block_group_cache *block_group, | |
3248 | u64 *total_trimmed, u64 start, u64 end, u64 minlen) | |
3249 | { | |
3250 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
3251 | struct btrfs_free_space *entry; | |
3252 | int ret = 0; | |
3253 | int ret2; | |
3254 | u64 bytes; | |
3255 | u64 offset = offset_to_bitmap(ctl, start); | |
3256 | ||
3257 | while (offset < end) { | |
3258 | bool next_bitmap = false; | |
55507ce3 | 3259 | struct btrfs_trim_range trim_entry; |
7fe1e641 | 3260 | |
55507ce3 | 3261 | mutex_lock(&ctl->cache_writeout_mutex); |
7fe1e641 LZ |
3262 | spin_lock(&ctl->tree_lock); |
3263 | ||
3264 | if (ctl->free_space < minlen) { | |
3265 | spin_unlock(&ctl->tree_lock); | |
55507ce3 | 3266 | mutex_unlock(&ctl->cache_writeout_mutex); |
7fe1e641 LZ |
3267 | break; |
3268 | } | |
3269 | ||
3270 | entry = tree_search_offset(ctl, offset, 1, 0); | |
3271 | if (!entry) { | |
3272 | spin_unlock(&ctl->tree_lock); | |
55507ce3 | 3273 | mutex_unlock(&ctl->cache_writeout_mutex); |
7fe1e641 LZ |
3274 | next_bitmap = true; |
3275 | goto next; | |
3276 | } | |
3277 | ||
3278 | bytes = minlen; | |
0584f718 | 3279 | ret2 = search_bitmap(ctl, entry, &start, &bytes, false); |
7fe1e641 LZ |
3280 | if (ret2 || start >= end) { |
3281 | spin_unlock(&ctl->tree_lock); | |
55507ce3 | 3282 | mutex_unlock(&ctl->cache_writeout_mutex); |
7fe1e641 LZ |
3283 | next_bitmap = true; |
3284 | goto next; | |
3285 | } | |
3286 | ||
3287 | bytes = min(bytes, end - start); | |
3288 | if (bytes < minlen) { | |
3289 | spin_unlock(&ctl->tree_lock); | |
55507ce3 | 3290 | mutex_unlock(&ctl->cache_writeout_mutex); |
7fe1e641 LZ |
3291 | goto next; |
3292 | } | |
3293 | ||
3294 | bitmap_clear_bits(ctl, entry, start, bytes); | |
3295 | if (entry->bytes == 0) | |
3296 | free_bitmap(ctl, entry); | |
3297 | ||
3298 | spin_unlock(&ctl->tree_lock); | |
55507ce3 FM |
3299 | trim_entry.start = start; |
3300 | trim_entry.bytes = bytes; | |
3301 | list_add_tail(&trim_entry.list, &ctl->trimming_ranges); | |
3302 | mutex_unlock(&ctl->cache_writeout_mutex); | |
7fe1e641 LZ |
3303 | |
3304 | ret = do_trimming(block_group, total_trimmed, start, bytes, | |
55507ce3 | 3305 | start, bytes, &trim_entry); |
7fe1e641 LZ |
3306 | if (ret) |
3307 | break; | |
3308 | next: | |
3309 | if (next_bitmap) { | |
3310 | offset += BITS_PER_BITMAP * ctl->unit; | |
3311 | } else { | |
3312 | start += bytes; | |
3313 | if (start >= offset + BITS_PER_BITMAP * ctl->unit) | |
3314 | offset += BITS_PER_BITMAP * ctl->unit; | |
f7039b1d | 3315 | } |
f7039b1d LD |
3316 | |
3317 | if (fatal_signal_pending(current)) { | |
3318 | ret = -ERESTARTSYS; | |
3319 | break; | |
3320 | } | |
3321 | ||
3322 | cond_resched(); | |
3323 | } | |
3324 | ||
3325 | return ret; | |
3326 | } | |
581bb050 | 3327 | |
e33e17ee | 3328 | void btrfs_get_block_group_trimming(struct btrfs_block_group_cache *cache) |
7fe1e641 | 3329 | { |
e33e17ee JM |
3330 | atomic_inc(&cache->trimming); |
3331 | } | |
7fe1e641 | 3332 | |
e33e17ee JM |
3333 | void btrfs_put_block_group_trimming(struct btrfs_block_group_cache *block_group) |
3334 | { | |
0b246afa | 3335 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
e33e17ee JM |
3336 | struct extent_map_tree *em_tree; |
3337 | struct extent_map *em; | |
3338 | bool cleanup; | |
7fe1e641 | 3339 | |
04216820 | 3340 | spin_lock(&block_group->lock); |
e33e17ee JM |
3341 | cleanup = (atomic_dec_and_test(&block_group->trimming) && |
3342 | block_group->removed); | |
04216820 FM |
3343 | spin_unlock(&block_group->lock); |
3344 | ||
e33e17ee | 3345 | if (cleanup) { |
34441361 | 3346 | mutex_lock(&fs_info->chunk_mutex); |
0b246afa | 3347 | em_tree = &fs_info->mapping_tree.map_tree; |
04216820 FM |
3348 | write_lock(&em_tree->lock); |
3349 | em = lookup_extent_mapping(em_tree, block_group->key.objectid, | |
3350 | 1); | |
3351 | BUG_ON(!em); /* logic error, can't happen */ | |
a1e7e16e FM |
3352 | /* |
3353 | * remove_extent_mapping() will delete us from the pinned_chunks | |
3354 | * list, which is protected by the chunk mutex. | |
3355 | */ | |
04216820 FM |
3356 | remove_extent_mapping(em_tree, em); |
3357 | write_unlock(&em_tree->lock); | |
34441361 | 3358 | mutex_unlock(&fs_info->chunk_mutex); |
04216820 FM |
3359 | |
3360 | /* once for us and once for the tree */ | |
3361 | free_extent_map(em); | |
3362 | free_extent_map(em); | |
946ddbe8 FM |
3363 | |
3364 | /* | |
3365 | * We've left one free space entry and other tasks trimming | |
3366 | * this block group have left 1 entry each one. Free them. | |
3367 | */ | |
3368 | __btrfs_remove_free_space_cache(block_group->free_space_ctl); | |
e33e17ee JM |
3369 | } |
3370 | } | |
3371 | ||
3372 | int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group, | |
3373 | u64 *trimmed, u64 start, u64 end, u64 minlen) | |
3374 | { | |
3375 | int ret; | |
3376 | ||
3377 | *trimmed = 0; | |
3378 | ||
3379 | spin_lock(&block_group->lock); | |
3380 | if (block_group->removed) { | |
04216820 | 3381 | spin_unlock(&block_group->lock); |
e33e17ee | 3382 | return 0; |
04216820 | 3383 | } |
e33e17ee JM |
3384 | btrfs_get_block_group_trimming(block_group); |
3385 | spin_unlock(&block_group->lock); | |
3386 | ||
3387 | ret = trim_no_bitmap(block_group, trimmed, start, end, minlen); | |
3388 | if (ret) | |
3389 | goto out; | |
7fe1e641 | 3390 | |
e33e17ee JM |
3391 | ret = trim_bitmaps(block_group, trimmed, start, end, minlen); |
3392 | out: | |
3393 | btrfs_put_block_group_trimming(block_group); | |
7fe1e641 LZ |
3394 | return ret; |
3395 | } | |
3396 | ||
581bb050 LZ |
3397 | /* |
3398 | * Find the left-most item in the cache tree, and then return the | |
3399 | * smallest inode number in the item. | |
3400 | * | |
3401 | * Note: the returned inode number may not be the smallest one in | |
3402 | * the tree, if the left-most item is a bitmap. | |
3403 | */ | |
3404 | u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root) | |
3405 | { | |
3406 | struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl; | |
3407 | struct btrfs_free_space *entry = NULL; | |
3408 | u64 ino = 0; | |
3409 | ||
3410 | spin_lock(&ctl->tree_lock); | |
3411 | ||
3412 | if (RB_EMPTY_ROOT(&ctl->free_space_offset)) | |
3413 | goto out; | |
3414 | ||
3415 | entry = rb_entry(rb_first(&ctl->free_space_offset), | |
3416 | struct btrfs_free_space, offset_index); | |
3417 | ||
3418 | if (!entry->bitmap) { | |
3419 | ino = entry->offset; | |
3420 | ||
3421 | unlink_free_space(ctl, entry); | |
3422 | entry->offset++; | |
3423 | entry->bytes--; | |
3424 | if (!entry->bytes) | |
3425 | kmem_cache_free(btrfs_free_space_cachep, entry); | |
3426 | else | |
3427 | link_free_space(ctl, entry); | |
3428 | } else { | |
3429 | u64 offset = 0; | |
3430 | u64 count = 1; | |
3431 | int ret; | |
3432 | ||
0584f718 | 3433 | ret = search_bitmap(ctl, entry, &offset, &count, true); |
79787eaa | 3434 | /* Logic error; Should be empty if it can't find anything */ |
b12d6869 | 3435 | ASSERT(!ret); |
581bb050 LZ |
3436 | |
3437 | ino = offset; | |
3438 | bitmap_clear_bits(ctl, entry, offset, 1); | |
3439 | if (entry->bytes == 0) | |
3440 | free_bitmap(ctl, entry); | |
3441 | } | |
3442 | out: | |
3443 | spin_unlock(&ctl->tree_lock); | |
3444 | ||
3445 | return ino; | |
3446 | } | |
82d5902d LZ |
3447 | |
3448 | struct inode *lookup_free_ino_inode(struct btrfs_root *root, | |
3449 | struct btrfs_path *path) | |
3450 | { | |
3451 | struct inode *inode = NULL; | |
3452 | ||
57cdc8db DS |
3453 | spin_lock(&root->ino_cache_lock); |
3454 | if (root->ino_cache_inode) | |
3455 | inode = igrab(root->ino_cache_inode); | |
3456 | spin_unlock(&root->ino_cache_lock); | |
82d5902d LZ |
3457 | if (inode) |
3458 | return inode; | |
3459 | ||
3460 | inode = __lookup_free_space_inode(root, path, 0); | |
3461 | if (IS_ERR(inode)) | |
3462 | return inode; | |
3463 | ||
57cdc8db | 3464 | spin_lock(&root->ino_cache_lock); |
7841cb28 | 3465 | if (!btrfs_fs_closing(root->fs_info)) |
57cdc8db DS |
3466 | root->ino_cache_inode = igrab(inode); |
3467 | spin_unlock(&root->ino_cache_lock); | |
82d5902d LZ |
3468 | |
3469 | return inode; | |
3470 | } | |
3471 | ||
3472 | int create_free_ino_inode(struct btrfs_root *root, | |
3473 | struct btrfs_trans_handle *trans, | |
3474 | struct btrfs_path *path) | |
3475 | { | |
3476 | return __create_free_space_inode(root, trans, path, | |
3477 | BTRFS_FREE_INO_OBJECTID, 0); | |
3478 | } | |
3479 | ||
3480 | int load_free_ino_cache(struct btrfs_fs_info *fs_info, struct btrfs_root *root) | |
3481 | { | |
3482 | struct btrfs_free_space_ctl *ctl = root->free_ino_ctl; | |
3483 | struct btrfs_path *path; | |
3484 | struct inode *inode; | |
3485 | int ret = 0; | |
3486 | u64 root_gen = btrfs_root_generation(&root->root_item); | |
3487 | ||
0b246afa | 3488 | if (!btrfs_test_opt(fs_info, INODE_MAP_CACHE)) |
4b9465cb CM |
3489 | return 0; |
3490 | ||
82d5902d LZ |
3491 | /* |
3492 | * If we're unmounting then just return, since this does a search on the | |
3493 | * normal root and not the commit root and we could deadlock. | |
3494 | */ | |
7841cb28 | 3495 | if (btrfs_fs_closing(fs_info)) |
82d5902d LZ |
3496 | return 0; |
3497 | ||
3498 | path = btrfs_alloc_path(); | |
3499 | if (!path) | |
3500 | return 0; | |
3501 | ||
3502 | inode = lookup_free_ino_inode(root, path); | |
3503 | if (IS_ERR(inode)) | |
3504 | goto out; | |
3505 | ||
3506 | if (root_gen != BTRFS_I(inode)->generation) | |
3507 | goto out_put; | |
3508 | ||
3509 | ret = __load_free_space_cache(root, inode, ctl, path, 0); | |
3510 | ||
3511 | if (ret < 0) | |
c2cf52eb SK |
3512 | btrfs_err(fs_info, |
3513 | "failed to load free ino cache for root %llu", | |
3514 | root->root_key.objectid); | |
82d5902d LZ |
3515 | out_put: |
3516 | iput(inode); | |
3517 | out: | |
3518 | btrfs_free_path(path); | |
3519 | return ret; | |
3520 | } | |
3521 | ||
3522 | int btrfs_write_out_ino_cache(struct btrfs_root *root, | |
3523 | struct btrfs_trans_handle *trans, | |
53645a91 FDBM |
3524 | struct btrfs_path *path, |
3525 | struct inode *inode) | |
82d5902d | 3526 | { |
0b246afa | 3527 | struct btrfs_fs_info *fs_info = root->fs_info; |
82d5902d | 3528 | struct btrfs_free_space_ctl *ctl = root->free_ino_ctl; |
82d5902d | 3529 | int ret; |
c9dc4c65 | 3530 | struct btrfs_io_ctl io_ctl; |
e43699d4 | 3531 | bool release_metadata = true; |
82d5902d | 3532 | |
0b246afa | 3533 | if (!btrfs_test_opt(fs_info, INODE_MAP_CACHE)) |
4b9465cb CM |
3534 | return 0; |
3535 | ||
85db36cf | 3536 | memset(&io_ctl, 0, sizeof(io_ctl)); |
0e8d931a | 3537 | ret = __btrfs_write_out_cache(root, inode, ctl, NULL, &io_ctl, trans); |
e43699d4 FM |
3538 | if (!ret) { |
3539 | /* | |
3540 | * At this point writepages() didn't error out, so our metadata | |
3541 | * reservation is released when the writeback finishes, at | |
3542 | * inode.c:btrfs_finish_ordered_io(), regardless of it finishing | |
3543 | * with or without an error. | |
3544 | */ | |
3545 | release_metadata = false; | |
afdb5718 | 3546 | ret = btrfs_wait_cache_io_root(root, trans, &io_ctl, path); |
e43699d4 | 3547 | } |
85db36cf | 3548 | |
c09544e0 | 3549 | if (ret) { |
e43699d4 FM |
3550 | if (release_metadata) |
3551 | btrfs_delalloc_release_metadata(inode, inode->i_size); | |
c09544e0 | 3552 | #ifdef DEBUG |
0b246afa JM |
3553 | btrfs_err(fs_info, |
3554 | "failed to write free ino cache for root %llu", | |
3555 | root->root_key.objectid); | |
c09544e0 JB |
3556 | #endif |
3557 | } | |
82d5902d | 3558 | |
82d5902d LZ |
3559 | return ret; |
3560 | } | |
74255aa0 JB |
3561 | |
3562 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS | |
dc11dd5d JB |
3563 | /* |
3564 | * Use this if you need to make a bitmap or extent entry specifically, it | |
3565 | * doesn't do any of the merging that add_free_space does, this acts a lot like | |
3566 | * how the free space cache loading stuff works, so you can get really weird | |
3567 | * configurations. | |
3568 | */ | |
3569 | int test_add_free_space_entry(struct btrfs_block_group_cache *cache, | |
3570 | u64 offset, u64 bytes, bool bitmap) | |
74255aa0 | 3571 | { |
dc11dd5d JB |
3572 | struct btrfs_free_space_ctl *ctl = cache->free_space_ctl; |
3573 | struct btrfs_free_space *info = NULL, *bitmap_info; | |
3574 | void *map = NULL; | |
3575 | u64 bytes_added; | |
3576 | int ret; | |
74255aa0 | 3577 | |
dc11dd5d JB |
3578 | again: |
3579 | if (!info) { | |
3580 | info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS); | |
3581 | if (!info) | |
3582 | return -ENOMEM; | |
74255aa0 JB |
3583 | } |
3584 | ||
dc11dd5d JB |
3585 | if (!bitmap) { |
3586 | spin_lock(&ctl->tree_lock); | |
3587 | info->offset = offset; | |
3588 | info->bytes = bytes; | |
cef40483 | 3589 | info->max_extent_size = 0; |
dc11dd5d JB |
3590 | ret = link_free_space(ctl, info); |
3591 | spin_unlock(&ctl->tree_lock); | |
3592 | if (ret) | |
3593 | kmem_cache_free(btrfs_free_space_cachep, info); | |
3594 | return ret; | |
3595 | } | |
3596 | ||
3597 | if (!map) { | |
09cbfeaf | 3598 | map = kzalloc(PAGE_SIZE, GFP_NOFS); |
dc11dd5d JB |
3599 | if (!map) { |
3600 | kmem_cache_free(btrfs_free_space_cachep, info); | |
3601 | return -ENOMEM; | |
3602 | } | |
3603 | } | |
3604 | ||
3605 | spin_lock(&ctl->tree_lock); | |
3606 | bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), | |
3607 | 1, 0); | |
3608 | if (!bitmap_info) { | |
3609 | info->bitmap = map; | |
3610 | map = NULL; | |
3611 | add_new_bitmap(ctl, info, offset); | |
3612 | bitmap_info = info; | |
20005523 | 3613 | info = NULL; |
dc11dd5d | 3614 | } |
74255aa0 | 3615 | |
dc11dd5d | 3616 | bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes); |
cef40483 | 3617 | |
dc11dd5d JB |
3618 | bytes -= bytes_added; |
3619 | offset += bytes_added; | |
3620 | spin_unlock(&ctl->tree_lock); | |
74255aa0 | 3621 | |
dc11dd5d JB |
3622 | if (bytes) |
3623 | goto again; | |
74255aa0 | 3624 | |
20005523 FM |
3625 | if (info) |
3626 | kmem_cache_free(btrfs_free_space_cachep, info); | |
dc11dd5d JB |
3627 | if (map) |
3628 | kfree(map); | |
3629 | return 0; | |
74255aa0 JB |
3630 | } |
3631 | ||
3632 | /* | |
3633 | * Checks to see if the given range is in the free space cache. This is really | |
3634 | * just used to check the absence of space, so if there is free space in the | |
3635 | * range at all we will return 1. | |
3636 | */ | |
dc11dd5d JB |
3637 | int test_check_exists(struct btrfs_block_group_cache *cache, |
3638 | u64 offset, u64 bytes) | |
74255aa0 JB |
3639 | { |
3640 | struct btrfs_free_space_ctl *ctl = cache->free_space_ctl; | |
3641 | struct btrfs_free_space *info; | |
3642 | int ret = 0; | |
3643 | ||
3644 | spin_lock(&ctl->tree_lock); | |
3645 | info = tree_search_offset(ctl, offset, 0, 0); | |
3646 | if (!info) { | |
3647 | info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), | |
3648 | 1, 0); | |
3649 | if (!info) | |
3650 | goto out; | |
3651 | } | |
3652 | ||
3653 | have_info: | |
3654 | if (info->bitmap) { | |
3655 | u64 bit_off, bit_bytes; | |
3656 | struct rb_node *n; | |
3657 | struct btrfs_free_space *tmp; | |
3658 | ||
3659 | bit_off = offset; | |
3660 | bit_bytes = ctl->unit; | |
0584f718 | 3661 | ret = search_bitmap(ctl, info, &bit_off, &bit_bytes, false); |
74255aa0 JB |
3662 | if (!ret) { |
3663 | if (bit_off == offset) { | |
3664 | ret = 1; | |
3665 | goto out; | |
3666 | } else if (bit_off > offset && | |
3667 | offset + bytes > bit_off) { | |
3668 | ret = 1; | |
3669 | goto out; | |
3670 | } | |
3671 | } | |
3672 | ||
3673 | n = rb_prev(&info->offset_index); | |
3674 | while (n) { | |
3675 | tmp = rb_entry(n, struct btrfs_free_space, | |
3676 | offset_index); | |
3677 | if (tmp->offset + tmp->bytes < offset) | |
3678 | break; | |
3679 | if (offset + bytes < tmp->offset) { | |
5473e0c4 | 3680 | n = rb_prev(&tmp->offset_index); |
74255aa0 JB |
3681 | continue; |
3682 | } | |
3683 | info = tmp; | |
3684 | goto have_info; | |
3685 | } | |
3686 | ||
3687 | n = rb_next(&info->offset_index); | |
3688 | while (n) { | |
3689 | tmp = rb_entry(n, struct btrfs_free_space, | |
3690 | offset_index); | |
3691 | if (offset + bytes < tmp->offset) | |
3692 | break; | |
3693 | if (tmp->offset + tmp->bytes < offset) { | |
5473e0c4 | 3694 | n = rb_next(&tmp->offset_index); |
74255aa0 JB |
3695 | continue; |
3696 | } | |
3697 | info = tmp; | |
3698 | goto have_info; | |
3699 | } | |
3700 | ||
20005523 | 3701 | ret = 0; |
74255aa0 JB |
3702 | goto out; |
3703 | } | |
3704 | ||
3705 | if (info->offset == offset) { | |
3706 | ret = 1; | |
3707 | goto out; | |
3708 | } | |
3709 | ||
3710 | if (offset > info->offset && offset < info->offset + info->bytes) | |
3711 | ret = 1; | |
3712 | out: | |
3713 | spin_unlock(&ctl->tree_lock); | |
3714 | return ret; | |
3715 | } | |
dc11dd5d | 3716 | #endif /* CONFIG_BTRFS_FS_RUN_SANITY_TESTS */ |