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