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