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, | |
221 | struct btrfs_path *path, | |
222 | struct inode *inode) | |
223 | { | |
224 | loff_t oldsize; | |
225 | int ret = 0; | |
0af3d00b JB |
226 | |
227 | oldsize = i_size_read(inode); | |
228 | btrfs_i_size_write(inode, 0); | |
229 | truncate_pagecache(inode, oldsize, 0); | |
230 | ||
231 | /* | |
232 | * We don't need an orphan item because truncating the free space cache | |
233 | * will never be split across transactions. | |
234 | */ | |
235 | ret = btrfs_truncate_inode_items(trans, root, inode, | |
236 | 0, BTRFS_EXTENT_DATA_KEY); | |
237 | if (ret) { | |
79787eaa | 238 | btrfs_abort_transaction(trans, root, ret); |
0af3d00b JB |
239 | return ret; |
240 | } | |
241 | ||
82d5902d | 242 | ret = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
243 | if (ret) |
244 | btrfs_abort_transaction(trans, root, ret); | |
c8174313 | 245 | |
82d5902d | 246 | return ret; |
0af3d00b JB |
247 | } |
248 | ||
9d66e233 JB |
249 | static int readahead_cache(struct inode *inode) |
250 | { | |
251 | struct file_ra_state *ra; | |
252 | unsigned long last_index; | |
253 | ||
254 | ra = kzalloc(sizeof(*ra), GFP_NOFS); | |
255 | if (!ra) | |
256 | return -ENOMEM; | |
257 | ||
258 | file_ra_state_init(ra, inode->i_mapping); | |
259 | last_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT; | |
260 | ||
261 | page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index); | |
262 | ||
263 | kfree(ra); | |
264 | ||
265 | return 0; | |
266 | } | |
267 | ||
a67509c3 JB |
268 | struct io_ctl { |
269 | void *cur, *orig; | |
270 | struct page *page; | |
271 | struct page **pages; | |
272 | struct btrfs_root *root; | |
273 | unsigned long size; | |
274 | int index; | |
275 | int num_pages; | |
5b0e95bf | 276 | unsigned check_crcs:1; |
a67509c3 JB |
277 | }; |
278 | ||
279 | static int io_ctl_init(struct io_ctl *io_ctl, struct inode *inode, | |
280 | struct btrfs_root *root) | |
281 | { | |
282 | memset(io_ctl, 0, sizeof(struct io_ctl)); | |
283 | io_ctl->num_pages = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> | |
284 | PAGE_CACHE_SHIFT; | |
285 | io_ctl->pages = kzalloc(sizeof(struct page *) * io_ctl->num_pages, | |
286 | GFP_NOFS); | |
287 | if (!io_ctl->pages) | |
288 | return -ENOMEM; | |
289 | io_ctl->root = root; | |
5b0e95bf JB |
290 | if (btrfs_ino(inode) != BTRFS_FREE_INO_OBJECTID) |
291 | io_ctl->check_crcs = 1; | |
a67509c3 JB |
292 | return 0; |
293 | } | |
294 | ||
295 | static void io_ctl_free(struct io_ctl *io_ctl) | |
296 | { | |
297 | kfree(io_ctl->pages); | |
298 | } | |
299 | ||
300 | static void io_ctl_unmap_page(struct io_ctl *io_ctl) | |
301 | { | |
302 | if (io_ctl->cur) { | |
303 | kunmap(io_ctl->page); | |
304 | io_ctl->cur = NULL; | |
305 | io_ctl->orig = NULL; | |
306 | } | |
307 | } | |
308 | ||
309 | static void io_ctl_map_page(struct io_ctl *io_ctl, int clear) | |
310 | { | |
a67509c3 JB |
311 | BUG_ON(io_ctl->index >= io_ctl->num_pages); |
312 | io_ctl->page = io_ctl->pages[io_ctl->index++]; | |
313 | io_ctl->cur = kmap(io_ctl->page); | |
314 | io_ctl->orig = io_ctl->cur; | |
315 | io_ctl->size = PAGE_CACHE_SIZE; | |
316 | if (clear) | |
317 | memset(io_ctl->cur, 0, PAGE_CACHE_SIZE); | |
318 | } | |
319 | ||
320 | static void io_ctl_drop_pages(struct io_ctl *io_ctl) | |
321 | { | |
322 | int i; | |
323 | ||
324 | io_ctl_unmap_page(io_ctl); | |
325 | ||
326 | for (i = 0; i < io_ctl->num_pages; i++) { | |
a1ee5a45 LZ |
327 | if (io_ctl->pages[i]) { |
328 | ClearPageChecked(io_ctl->pages[i]); | |
329 | unlock_page(io_ctl->pages[i]); | |
330 | page_cache_release(io_ctl->pages[i]); | |
331 | } | |
a67509c3 JB |
332 | } |
333 | } | |
334 | ||
335 | static int io_ctl_prepare_pages(struct io_ctl *io_ctl, struct inode *inode, | |
336 | int uptodate) | |
337 | { | |
338 | struct page *page; | |
339 | gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); | |
340 | int i; | |
341 | ||
342 | for (i = 0; i < io_ctl->num_pages; i++) { | |
343 | page = find_or_create_page(inode->i_mapping, i, mask); | |
344 | if (!page) { | |
345 | io_ctl_drop_pages(io_ctl); | |
346 | return -ENOMEM; | |
347 | } | |
348 | io_ctl->pages[i] = page; | |
349 | if (uptodate && !PageUptodate(page)) { | |
350 | btrfs_readpage(NULL, page); | |
351 | lock_page(page); | |
352 | if (!PageUptodate(page)) { | |
353 | printk(KERN_ERR "btrfs: error reading free " | |
354 | "space cache\n"); | |
355 | io_ctl_drop_pages(io_ctl); | |
356 | return -EIO; | |
357 | } | |
358 | } | |
359 | } | |
360 | ||
f7d61dcd JB |
361 | for (i = 0; i < io_ctl->num_pages; i++) { |
362 | clear_page_dirty_for_io(io_ctl->pages[i]); | |
363 | set_page_extent_mapped(io_ctl->pages[i]); | |
364 | } | |
365 | ||
a67509c3 JB |
366 | return 0; |
367 | } | |
368 | ||
369 | static void io_ctl_set_generation(struct io_ctl *io_ctl, u64 generation) | |
370 | { | |
528c0327 | 371 | __le64 *val; |
a67509c3 JB |
372 | |
373 | io_ctl_map_page(io_ctl, 1); | |
374 | ||
375 | /* | |
5b0e95bf JB |
376 | * Skip the csum areas. If we don't check crcs then we just have a |
377 | * 64bit chunk at the front of the first page. | |
a67509c3 | 378 | */ |
5b0e95bf JB |
379 | if (io_ctl->check_crcs) { |
380 | io_ctl->cur += (sizeof(u32) * io_ctl->num_pages); | |
381 | io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages); | |
382 | } else { | |
383 | io_ctl->cur += sizeof(u64); | |
384 | io_ctl->size -= sizeof(u64) * 2; | |
385 | } | |
a67509c3 JB |
386 | |
387 | val = io_ctl->cur; | |
388 | *val = cpu_to_le64(generation); | |
389 | io_ctl->cur += sizeof(u64); | |
a67509c3 JB |
390 | } |
391 | ||
392 | static int io_ctl_check_generation(struct io_ctl *io_ctl, u64 generation) | |
393 | { | |
528c0327 | 394 | __le64 *gen; |
a67509c3 | 395 | |
5b0e95bf JB |
396 | /* |
397 | * Skip the crc area. If we don't check crcs then we just have a 64bit | |
398 | * chunk at the front of the first page. | |
399 | */ | |
400 | if (io_ctl->check_crcs) { | |
401 | io_ctl->cur += sizeof(u32) * io_ctl->num_pages; | |
402 | io_ctl->size -= sizeof(u64) + | |
403 | (sizeof(u32) * io_ctl->num_pages); | |
404 | } else { | |
405 | io_ctl->cur += sizeof(u64); | |
406 | io_ctl->size -= sizeof(u64) * 2; | |
407 | } | |
a67509c3 | 408 | |
a67509c3 JB |
409 | gen = io_ctl->cur; |
410 | if (le64_to_cpu(*gen) != generation) { | |
411 | printk_ratelimited(KERN_ERR "btrfs: space cache generation " | |
412 | "(%Lu) does not match inode (%Lu)\n", *gen, | |
413 | generation); | |
414 | io_ctl_unmap_page(io_ctl); | |
415 | return -EIO; | |
416 | } | |
417 | io_ctl->cur += sizeof(u64); | |
5b0e95bf JB |
418 | return 0; |
419 | } | |
420 | ||
421 | static void io_ctl_set_crc(struct io_ctl *io_ctl, int index) | |
422 | { | |
423 | u32 *tmp; | |
424 | u32 crc = ~(u32)0; | |
425 | unsigned offset = 0; | |
426 | ||
427 | if (!io_ctl->check_crcs) { | |
428 | io_ctl_unmap_page(io_ctl); | |
429 | return; | |
430 | } | |
431 | ||
432 | if (index == 0) | |
cb54f257 | 433 | offset = sizeof(u32) * io_ctl->num_pages; |
5b0e95bf | 434 | |
b0496686 | 435 | crc = btrfs_csum_data(io_ctl->orig + offset, crc, |
5b0e95bf JB |
436 | PAGE_CACHE_SIZE - offset); |
437 | btrfs_csum_final(crc, (char *)&crc); | |
438 | io_ctl_unmap_page(io_ctl); | |
439 | tmp = kmap(io_ctl->pages[0]); | |
440 | tmp += index; | |
441 | *tmp = crc; | |
442 | kunmap(io_ctl->pages[0]); | |
443 | } | |
444 | ||
445 | static int io_ctl_check_crc(struct io_ctl *io_ctl, int index) | |
446 | { | |
447 | u32 *tmp, val; | |
448 | u32 crc = ~(u32)0; | |
449 | unsigned offset = 0; | |
450 | ||
451 | if (!io_ctl->check_crcs) { | |
452 | io_ctl_map_page(io_ctl, 0); | |
453 | return 0; | |
454 | } | |
455 | ||
456 | if (index == 0) | |
457 | offset = sizeof(u32) * io_ctl->num_pages; | |
458 | ||
459 | tmp = kmap(io_ctl->pages[0]); | |
460 | tmp += index; | |
461 | val = *tmp; | |
462 | kunmap(io_ctl->pages[0]); | |
463 | ||
464 | io_ctl_map_page(io_ctl, 0); | |
b0496686 | 465 | crc = btrfs_csum_data(io_ctl->orig + offset, crc, |
5b0e95bf JB |
466 | PAGE_CACHE_SIZE - offset); |
467 | btrfs_csum_final(crc, (char *)&crc); | |
468 | if (val != crc) { | |
469 | printk_ratelimited(KERN_ERR "btrfs: csum mismatch on free " | |
470 | "space cache\n"); | |
471 | io_ctl_unmap_page(io_ctl); | |
472 | return -EIO; | |
473 | } | |
474 | ||
a67509c3 JB |
475 | return 0; |
476 | } | |
477 | ||
478 | static int io_ctl_add_entry(struct io_ctl *io_ctl, u64 offset, u64 bytes, | |
479 | void *bitmap) | |
480 | { | |
481 | struct btrfs_free_space_entry *entry; | |
482 | ||
483 | if (!io_ctl->cur) | |
484 | return -ENOSPC; | |
485 | ||
486 | entry = io_ctl->cur; | |
487 | entry->offset = cpu_to_le64(offset); | |
488 | entry->bytes = cpu_to_le64(bytes); | |
489 | entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP : | |
490 | BTRFS_FREE_SPACE_EXTENT; | |
491 | io_ctl->cur += sizeof(struct btrfs_free_space_entry); | |
492 | io_ctl->size -= sizeof(struct btrfs_free_space_entry); | |
493 | ||
494 | if (io_ctl->size >= sizeof(struct btrfs_free_space_entry)) | |
495 | return 0; | |
496 | ||
5b0e95bf | 497 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); |
a67509c3 JB |
498 | |
499 | /* No more pages to map */ | |
500 | if (io_ctl->index >= io_ctl->num_pages) | |
501 | return 0; | |
502 | ||
503 | /* map the next page */ | |
504 | io_ctl_map_page(io_ctl, 1); | |
505 | return 0; | |
506 | } | |
507 | ||
508 | static int io_ctl_add_bitmap(struct io_ctl *io_ctl, void *bitmap) | |
509 | { | |
510 | if (!io_ctl->cur) | |
511 | return -ENOSPC; | |
512 | ||
513 | /* | |
514 | * If we aren't at the start of the current page, unmap this one and | |
515 | * map the next one if there is any left. | |
516 | */ | |
517 | if (io_ctl->cur != io_ctl->orig) { | |
5b0e95bf | 518 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); |
a67509c3 JB |
519 | if (io_ctl->index >= io_ctl->num_pages) |
520 | return -ENOSPC; | |
521 | io_ctl_map_page(io_ctl, 0); | |
522 | } | |
523 | ||
524 | memcpy(io_ctl->cur, bitmap, PAGE_CACHE_SIZE); | |
5b0e95bf | 525 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); |
a67509c3 JB |
526 | if (io_ctl->index < io_ctl->num_pages) |
527 | io_ctl_map_page(io_ctl, 0); | |
528 | return 0; | |
529 | } | |
530 | ||
531 | static void io_ctl_zero_remaining_pages(struct io_ctl *io_ctl) | |
532 | { | |
5b0e95bf JB |
533 | /* |
534 | * If we're not on the boundary we know we've modified the page and we | |
535 | * need to crc the page. | |
536 | */ | |
537 | if (io_ctl->cur != io_ctl->orig) | |
538 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); | |
539 | else | |
540 | io_ctl_unmap_page(io_ctl); | |
a67509c3 JB |
541 | |
542 | while (io_ctl->index < io_ctl->num_pages) { | |
543 | io_ctl_map_page(io_ctl, 1); | |
5b0e95bf | 544 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); |
a67509c3 JB |
545 | } |
546 | } | |
547 | ||
5b0e95bf JB |
548 | static int io_ctl_read_entry(struct io_ctl *io_ctl, |
549 | struct btrfs_free_space *entry, u8 *type) | |
a67509c3 JB |
550 | { |
551 | struct btrfs_free_space_entry *e; | |
2f120c05 JB |
552 | int ret; |
553 | ||
554 | if (!io_ctl->cur) { | |
555 | ret = io_ctl_check_crc(io_ctl, io_ctl->index); | |
556 | if (ret) | |
557 | return ret; | |
558 | } | |
a67509c3 JB |
559 | |
560 | e = io_ctl->cur; | |
561 | entry->offset = le64_to_cpu(e->offset); | |
562 | entry->bytes = le64_to_cpu(e->bytes); | |
5b0e95bf | 563 | *type = e->type; |
a67509c3 JB |
564 | io_ctl->cur += sizeof(struct btrfs_free_space_entry); |
565 | io_ctl->size -= sizeof(struct btrfs_free_space_entry); | |
566 | ||
567 | if (io_ctl->size >= sizeof(struct btrfs_free_space_entry)) | |
5b0e95bf | 568 | return 0; |
a67509c3 JB |
569 | |
570 | io_ctl_unmap_page(io_ctl); | |
571 | ||
2f120c05 | 572 | return 0; |
a67509c3 JB |
573 | } |
574 | ||
5b0e95bf JB |
575 | static int io_ctl_read_bitmap(struct io_ctl *io_ctl, |
576 | struct btrfs_free_space *entry) | |
a67509c3 | 577 | { |
5b0e95bf JB |
578 | int ret; |
579 | ||
5b0e95bf JB |
580 | ret = io_ctl_check_crc(io_ctl, io_ctl->index); |
581 | if (ret) | |
582 | return ret; | |
583 | ||
a67509c3 JB |
584 | memcpy(entry->bitmap, io_ctl->cur, PAGE_CACHE_SIZE); |
585 | io_ctl_unmap_page(io_ctl); | |
5b0e95bf JB |
586 | |
587 | return 0; | |
a67509c3 JB |
588 | } |
589 | ||
cd023e7b JB |
590 | /* |
591 | * Since we attach pinned extents after the fact we can have contiguous sections | |
592 | * of free space that are split up in entries. This poses a problem with the | |
593 | * tree logging stuff since it could have allocated across what appears to be 2 | |
594 | * entries since we would have merged the entries when adding the pinned extents | |
595 | * back to the free space cache. So run through the space cache that we just | |
596 | * loaded and merge contiguous entries. This will make the log replay stuff not | |
597 | * blow up and it will make for nicer allocator behavior. | |
598 | */ | |
599 | static void merge_space_tree(struct btrfs_free_space_ctl *ctl) | |
600 | { | |
601 | struct btrfs_free_space *e, *prev = NULL; | |
602 | struct rb_node *n; | |
603 | ||
604 | again: | |
605 | spin_lock(&ctl->tree_lock); | |
606 | for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) { | |
607 | e = rb_entry(n, struct btrfs_free_space, offset_index); | |
608 | if (!prev) | |
609 | goto next; | |
610 | if (e->bitmap || prev->bitmap) | |
611 | goto next; | |
612 | if (prev->offset + prev->bytes == e->offset) { | |
613 | unlink_free_space(ctl, prev); | |
614 | unlink_free_space(ctl, e); | |
615 | prev->bytes += e->bytes; | |
616 | kmem_cache_free(btrfs_free_space_cachep, e); | |
617 | link_free_space(ctl, prev); | |
618 | prev = NULL; | |
619 | spin_unlock(&ctl->tree_lock); | |
620 | goto again; | |
621 | } | |
622 | next: | |
623 | prev = e; | |
624 | } | |
625 | spin_unlock(&ctl->tree_lock); | |
626 | } | |
627 | ||
48a3b636 ES |
628 | static int __load_free_space_cache(struct btrfs_root *root, struct inode *inode, |
629 | struct btrfs_free_space_ctl *ctl, | |
630 | struct btrfs_path *path, u64 offset) | |
9d66e233 | 631 | { |
9d66e233 JB |
632 | struct btrfs_free_space_header *header; |
633 | struct extent_buffer *leaf; | |
a67509c3 | 634 | struct io_ctl io_ctl; |
9d66e233 | 635 | struct btrfs_key key; |
a67509c3 | 636 | struct btrfs_free_space *e, *n; |
9d66e233 JB |
637 | struct list_head bitmaps; |
638 | u64 num_entries; | |
639 | u64 num_bitmaps; | |
640 | u64 generation; | |
a67509c3 | 641 | u8 type; |
f6a39829 | 642 | int ret = 0; |
9d66e233 JB |
643 | |
644 | INIT_LIST_HEAD(&bitmaps); | |
645 | ||
9d66e233 | 646 | /* Nothing in the space cache, goodbye */ |
0414efae | 647 | if (!i_size_read(inode)) |
a67509c3 | 648 | return 0; |
9d66e233 JB |
649 | |
650 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
0414efae | 651 | key.offset = offset; |
9d66e233 JB |
652 | key.type = 0; |
653 | ||
654 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
0414efae | 655 | if (ret < 0) |
a67509c3 | 656 | return 0; |
0414efae | 657 | else if (ret > 0) { |
945d8962 | 658 | btrfs_release_path(path); |
a67509c3 | 659 | return 0; |
9d66e233 JB |
660 | } |
661 | ||
0414efae LZ |
662 | ret = -1; |
663 | ||
9d66e233 JB |
664 | leaf = path->nodes[0]; |
665 | header = btrfs_item_ptr(leaf, path->slots[0], | |
666 | struct btrfs_free_space_header); | |
667 | num_entries = btrfs_free_space_entries(leaf, header); | |
668 | num_bitmaps = btrfs_free_space_bitmaps(leaf, header); | |
669 | generation = btrfs_free_space_generation(leaf, header); | |
945d8962 | 670 | btrfs_release_path(path); |
9d66e233 JB |
671 | |
672 | if (BTRFS_I(inode)->generation != generation) { | |
c2cf52eb SK |
673 | btrfs_err(root->fs_info, |
674 | "free space inode generation (%llu) " | |
675 | "did not match free space cache generation (%llu)", | |
676 | (unsigned long long)BTRFS_I(inode)->generation, | |
677 | (unsigned long long)generation); | |
a67509c3 | 678 | return 0; |
9d66e233 JB |
679 | } |
680 | ||
681 | if (!num_entries) | |
a67509c3 | 682 | return 0; |
9d66e233 | 683 | |
706efc66 LZ |
684 | ret = io_ctl_init(&io_ctl, inode, root); |
685 | if (ret) | |
686 | return ret; | |
687 | ||
9d66e233 | 688 | ret = readahead_cache(inode); |
0414efae | 689 | if (ret) |
9d66e233 | 690 | goto out; |
9d66e233 | 691 | |
a67509c3 JB |
692 | ret = io_ctl_prepare_pages(&io_ctl, inode, 1); |
693 | if (ret) | |
694 | goto out; | |
9d66e233 | 695 | |
5b0e95bf JB |
696 | ret = io_ctl_check_crc(&io_ctl, 0); |
697 | if (ret) | |
698 | goto free_cache; | |
699 | ||
a67509c3 JB |
700 | ret = io_ctl_check_generation(&io_ctl, generation); |
701 | if (ret) | |
702 | goto free_cache; | |
9d66e233 | 703 | |
a67509c3 JB |
704 | while (num_entries) { |
705 | e = kmem_cache_zalloc(btrfs_free_space_cachep, | |
706 | GFP_NOFS); | |
707 | if (!e) | |
9d66e233 | 708 | goto free_cache; |
9d66e233 | 709 | |
5b0e95bf JB |
710 | ret = io_ctl_read_entry(&io_ctl, e, &type); |
711 | if (ret) { | |
712 | kmem_cache_free(btrfs_free_space_cachep, e); | |
713 | goto free_cache; | |
714 | } | |
715 | ||
a67509c3 JB |
716 | if (!e->bytes) { |
717 | kmem_cache_free(btrfs_free_space_cachep, e); | |
718 | goto free_cache; | |
9d66e233 | 719 | } |
a67509c3 JB |
720 | |
721 | if (type == BTRFS_FREE_SPACE_EXTENT) { | |
722 | spin_lock(&ctl->tree_lock); | |
723 | ret = link_free_space(ctl, e); | |
724 | spin_unlock(&ctl->tree_lock); | |
725 | if (ret) { | |
c2cf52eb SK |
726 | btrfs_err(root->fs_info, |
727 | "Duplicate entries in free space cache, dumping"); | |
a67509c3 | 728 | kmem_cache_free(btrfs_free_space_cachep, e); |
9d66e233 JB |
729 | goto free_cache; |
730 | } | |
a67509c3 JB |
731 | } else { |
732 | BUG_ON(!num_bitmaps); | |
733 | num_bitmaps--; | |
734 | e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS); | |
735 | if (!e->bitmap) { | |
736 | kmem_cache_free( | |
737 | btrfs_free_space_cachep, e); | |
9d66e233 JB |
738 | goto free_cache; |
739 | } | |
a67509c3 JB |
740 | spin_lock(&ctl->tree_lock); |
741 | ret = link_free_space(ctl, e); | |
742 | ctl->total_bitmaps++; | |
743 | ctl->op->recalc_thresholds(ctl); | |
744 | spin_unlock(&ctl->tree_lock); | |
745 | if (ret) { | |
c2cf52eb SK |
746 | btrfs_err(root->fs_info, |
747 | "Duplicate entries in free space cache, dumping"); | |
dc89e982 | 748 | kmem_cache_free(btrfs_free_space_cachep, e); |
9d66e233 JB |
749 | goto free_cache; |
750 | } | |
a67509c3 | 751 | list_add_tail(&e->list, &bitmaps); |
9d66e233 JB |
752 | } |
753 | ||
a67509c3 JB |
754 | num_entries--; |
755 | } | |
9d66e233 | 756 | |
2f120c05 JB |
757 | io_ctl_unmap_page(&io_ctl); |
758 | ||
a67509c3 JB |
759 | /* |
760 | * We add the bitmaps at the end of the entries in order that | |
761 | * the bitmap entries are added to the cache. | |
762 | */ | |
763 | list_for_each_entry_safe(e, n, &bitmaps, list) { | |
9d66e233 | 764 | list_del_init(&e->list); |
5b0e95bf JB |
765 | ret = io_ctl_read_bitmap(&io_ctl, e); |
766 | if (ret) | |
767 | goto free_cache; | |
9d66e233 JB |
768 | } |
769 | ||
a67509c3 | 770 | io_ctl_drop_pages(&io_ctl); |
cd023e7b | 771 | merge_space_tree(ctl); |
9d66e233 JB |
772 | ret = 1; |
773 | out: | |
a67509c3 | 774 | io_ctl_free(&io_ctl); |
9d66e233 | 775 | return ret; |
9d66e233 | 776 | free_cache: |
a67509c3 | 777 | io_ctl_drop_pages(&io_ctl); |
0414efae | 778 | __btrfs_remove_free_space_cache(ctl); |
9d66e233 JB |
779 | goto out; |
780 | } | |
781 | ||
0414efae LZ |
782 | int load_free_space_cache(struct btrfs_fs_info *fs_info, |
783 | struct btrfs_block_group_cache *block_group) | |
0cb59c99 | 784 | { |
34d52cb6 | 785 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0414efae LZ |
786 | struct btrfs_root *root = fs_info->tree_root; |
787 | struct inode *inode; | |
788 | struct btrfs_path *path; | |
5b0e95bf | 789 | int ret = 0; |
0414efae LZ |
790 | bool matched; |
791 | u64 used = btrfs_block_group_used(&block_group->item); | |
792 | ||
0414efae LZ |
793 | /* |
794 | * If this block group has been marked to be cleared for one reason or | |
795 | * another then we can't trust the on disk cache, so just return. | |
796 | */ | |
9d66e233 | 797 | spin_lock(&block_group->lock); |
0414efae LZ |
798 | if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) { |
799 | spin_unlock(&block_group->lock); | |
800 | return 0; | |
801 | } | |
9d66e233 | 802 | spin_unlock(&block_group->lock); |
0414efae LZ |
803 | |
804 | path = btrfs_alloc_path(); | |
805 | if (!path) | |
806 | return 0; | |
d53ba474 JB |
807 | path->search_commit_root = 1; |
808 | path->skip_locking = 1; | |
0414efae LZ |
809 | |
810 | inode = lookup_free_space_inode(root, block_group, path); | |
811 | if (IS_ERR(inode)) { | |
812 | btrfs_free_path(path); | |
813 | return 0; | |
814 | } | |
815 | ||
5b0e95bf JB |
816 | /* We may have converted the inode and made the cache invalid. */ |
817 | spin_lock(&block_group->lock); | |
818 | if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) { | |
819 | spin_unlock(&block_group->lock); | |
a7e221e9 | 820 | btrfs_free_path(path); |
5b0e95bf JB |
821 | goto out; |
822 | } | |
823 | spin_unlock(&block_group->lock); | |
824 | ||
0414efae LZ |
825 | ret = __load_free_space_cache(fs_info->tree_root, inode, ctl, |
826 | path, block_group->key.objectid); | |
827 | btrfs_free_path(path); | |
828 | if (ret <= 0) | |
829 | goto out; | |
830 | ||
831 | spin_lock(&ctl->tree_lock); | |
832 | matched = (ctl->free_space == (block_group->key.offset - used - | |
833 | block_group->bytes_super)); | |
834 | spin_unlock(&ctl->tree_lock); | |
835 | ||
836 | if (!matched) { | |
837 | __btrfs_remove_free_space_cache(ctl); | |
c2cf52eb SK |
838 | btrfs_err(fs_info, "block group %llu has wrong amount of free space", |
839 | block_group->key.objectid); | |
0414efae LZ |
840 | ret = -1; |
841 | } | |
842 | out: | |
843 | if (ret < 0) { | |
844 | /* This cache is bogus, make sure it gets cleared */ | |
845 | spin_lock(&block_group->lock); | |
846 | block_group->disk_cache_state = BTRFS_DC_CLEAR; | |
847 | spin_unlock(&block_group->lock); | |
82d5902d | 848 | ret = 0; |
0414efae | 849 | |
c2cf52eb SK |
850 | btrfs_err(fs_info, "failed to load free space cache for block group %llu", |
851 | block_group->key.objectid); | |
0414efae LZ |
852 | } |
853 | ||
854 | iput(inode); | |
855 | return ret; | |
9d66e233 JB |
856 | } |
857 | ||
c09544e0 JB |
858 | /** |
859 | * __btrfs_write_out_cache - write out cached info to an inode | |
860 | * @root - the root the inode belongs to | |
861 | * @ctl - the free space cache we are going to write out | |
862 | * @block_group - the block_group for this cache if it belongs to a block_group | |
863 | * @trans - the trans handle | |
864 | * @path - the path to use | |
865 | * @offset - the offset for the key we'll insert | |
866 | * | |
867 | * This function writes out a free space cache struct to disk for quick recovery | |
868 | * on mount. This will return 0 if it was successfull in writing the cache out, | |
869 | * and -1 if it was not. | |
870 | */ | |
48a3b636 ES |
871 | static int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode, |
872 | struct btrfs_free_space_ctl *ctl, | |
873 | struct btrfs_block_group_cache *block_group, | |
874 | struct btrfs_trans_handle *trans, | |
875 | struct btrfs_path *path, u64 offset) | |
0cb59c99 JB |
876 | { |
877 | struct btrfs_free_space_header *header; | |
878 | struct extent_buffer *leaf; | |
0cb59c99 JB |
879 | struct rb_node *node; |
880 | struct list_head *pos, *n; | |
0cb59c99 | 881 | struct extent_state *cached_state = NULL; |
43be2146 JB |
882 | struct btrfs_free_cluster *cluster = NULL; |
883 | struct extent_io_tree *unpin = NULL; | |
a67509c3 | 884 | struct io_ctl io_ctl; |
0cb59c99 JB |
885 | struct list_head bitmap_list; |
886 | struct btrfs_key key; | |
db804f23 | 887 | u64 start, extent_start, extent_end, len; |
0cb59c99 JB |
888 | int entries = 0; |
889 | int bitmaps = 0; | |
c09544e0 JB |
890 | int ret; |
891 | int err = -1; | |
0cb59c99 | 892 | |
0cb59c99 JB |
893 | INIT_LIST_HEAD(&bitmap_list); |
894 | ||
0414efae LZ |
895 | if (!i_size_read(inode)) |
896 | return -1; | |
2b20982e | 897 | |
706efc66 LZ |
898 | ret = io_ctl_init(&io_ctl, inode, root); |
899 | if (ret) | |
900 | return -1; | |
be1a12a0 | 901 | |
43be2146 | 902 | /* Get the cluster for this block_group if it exists */ |
0414efae | 903 | if (block_group && !list_empty(&block_group->cluster_list)) |
43be2146 JB |
904 | cluster = list_entry(block_group->cluster_list.next, |
905 | struct btrfs_free_cluster, | |
906 | block_group_list); | |
907 | ||
a67509c3 JB |
908 | /* Lock all pages first so we can lock the extent safely. */ |
909 | io_ctl_prepare_pages(&io_ctl, inode, 0); | |
0cb59c99 | 910 | |
0cb59c99 | 911 | lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1, |
d0082371 | 912 | 0, &cached_state); |
0cb59c99 | 913 | |
f75b130e JB |
914 | node = rb_first(&ctl->free_space_offset); |
915 | if (!node && cluster) { | |
916 | node = rb_first(&cluster->root); | |
917 | cluster = NULL; | |
918 | } | |
919 | ||
5b0e95bf JB |
920 | /* Make sure we can fit our crcs into the first page */ |
921 | if (io_ctl.check_crcs && | |
73e1e61f | 922 | (io_ctl.num_pages * sizeof(u32)) >= PAGE_CACHE_SIZE) |
5b0e95bf | 923 | goto out_nospc; |
5b0e95bf | 924 | |
a67509c3 | 925 | io_ctl_set_generation(&io_ctl, trans->transid); |
43be2146 | 926 | |
a67509c3 JB |
927 | /* Write out the extent entries */ |
928 | while (node) { | |
929 | struct btrfs_free_space *e; | |
0cb59c99 | 930 | |
a67509c3 JB |
931 | e = rb_entry(node, struct btrfs_free_space, offset_index); |
932 | entries++; | |
0cb59c99 | 933 | |
a67509c3 JB |
934 | ret = io_ctl_add_entry(&io_ctl, e->offset, e->bytes, |
935 | e->bitmap); | |
936 | if (ret) | |
937 | goto out_nospc; | |
2f356126 | 938 | |
a67509c3 JB |
939 | if (e->bitmap) { |
940 | list_add_tail(&e->list, &bitmap_list); | |
941 | bitmaps++; | |
2f356126 | 942 | } |
a67509c3 JB |
943 | node = rb_next(node); |
944 | if (!node && cluster) { | |
945 | node = rb_first(&cluster->root); | |
946 | cluster = NULL; | |
43be2146 | 947 | } |
a67509c3 | 948 | } |
43be2146 | 949 | |
a67509c3 JB |
950 | /* |
951 | * We want to add any pinned extents to our free space cache | |
952 | * so we don't leak the space | |
953 | */ | |
db804f23 LZ |
954 | |
955 | /* | |
956 | * We shouldn't have switched the pinned extents yet so this is the | |
957 | * right one | |
958 | */ | |
959 | unpin = root->fs_info->pinned_extents; | |
960 | ||
961 | if (block_group) | |
962 | start = block_group->key.objectid; | |
963 | ||
a67509c3 JB |
964 | while (block_group && (start < block_group->key.objectid + |
965 | block_group->key.offset)) { | |
db804f23 LZ |
966 | ret = find_first_extent_bit(unpin, start, |
967 | &extent_start, &extent_end, | |
e6138876 | 968 | EXTENT_DIRTY, NULL); |
a67509c3 JB |
969 | if (ret) { |
970 | ret = 0; | |
971 | break; | |
0cb59c99 | 972 | } |
0cb59c99 | 973 | |
a67509c3 | 974 | /* This pinned extent is out of our range */ |
db804f23 | 975 | if (extent_start >= block_group->key.objectid + |
a67509c3 JB |
976 | block_group->key.offset) |
977 | break; | |
2f356126 | 978 | |
db804f23 LZ |
979 | extent_start = max(extent_start, start); |
980 | extent_end = min(block_group->key.objectid + | |
981 | block_group->key.offset, extent_end + 1); | |
982 | len = extent_end - extent_start; | |
0cb59c99 | 983 | |
a67509c3 | 984 | entries++; |
db804f23 | 985 | ret = io_ctl_add_entry(&io_ctl, extent_start, len, NULL); |
a67509c3 JB |
986 | if (ret) |
987 | goto out_nospc; | |
0cb59c99 | 988 | |
db804f23 | 989 | start = extent_end; |
a67509c3 | 990 | } |
0cb59c99 JB |
991 | |
992 | /* Write out the bitmaps */ | |
993 | list_for_each_safe(pos, n, &bitmap_list) { | |
0cb59c99 JB |
994 | struct btrfs_free_space *entry = |
995 | list_entry(pos, struct btrfs_free_space, list); | |
996 | ||
a67509c3 JB |
997 | ret = io_ctl_add_bitmap(&io_ctl, entry->bitmap); |
998 | if (ret) | |
999 | goto out_nospc; | |
0cb59c99 | 1000 | list_del_init(&entry->list); |
be1a12a0 JB |
1001 | } |
1002 | ||
0cb59c99 | 1003 | /* Zero out the rest of the pages just to make sure */ |
a67509c3 | 1004 | io_ctl_zero_remaining_pages(&io_ctl); |
0cb59c99 | 1005 | |
a67509c3 JB |
1006 | ret = btrfs_dirty_pages(root, inode, io_ctl.pages, io_ctl.num_pages, |
1007 | 0, i_size_read(inode), &cached_state); | |
1008 | io_ctl_drop_pages(&io_ctl); | |
0cb59c99 JB |
1009 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0, |
1010 | i_size_read(inode) - 1, &cached_state, GFP_NOFS); | |
1011 | ||
c09544e0 | 1012 | if (ret) |
2f356126 | 1013 | goto out; |
be1a12a0 | 1014 | |
be1a12a0 | 1015 | |
5fd02043 | 1016 | btrfs_wait_ordered_range(inode, 0, (u64)-1); |
0cb59c99 JB |
1017 | |
1018 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
0414efae | 1019 | key.offset = offset; |
0cb59c99 JB |
1020 | key.type = 0; |
1021 | ||
a9b5fcdd | 1022 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
0cb59c99 | 1023 | if (ret < 0) { |
a67509c3 | 1024 | clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1, |
5b0e95bf JB |
1025 | EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL, |
1026 | GFP_NOFS); | |
2f356126 | 1027 | goto out; |
0cb59c99 JB |
1028 | } |
1029 | leaf = path->nodes[0]; | |
1030 | if (ret > 0) { | |
1031 | struct btrfs_key found_key; | |
1032 | BUG_ON(!path->slots[0]); | |
1033 | path->slots[0]--; | |
1034 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1035 | if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID || | |
0414efae | 1036 | found_key.offset != offset) { |
a67509c3 JB |
1037 | clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, |
1038 | inode->i_size - 1, | |
5b0e95bf JB |
1039 | EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, |
1040 | NULL, GFP_NOFS); | |
b3b4aa74 | 1041 | btrfs_release_path(path); |
2f356126 | 1042 | goto out; |
0cb59c99 JB |
1043 | } |
1044 | } | |
549b4fdb JB |
1045 | |
1046 | BTRFS_I(inode)->generation = trans->transid; | |
0cb59c99 JB |
1047 | header = btrfs_item_ptr(leaf, path->slots[0], |
1048 | struct btrfs_free_space_header); | |
1049 | btrfs_set_free_space_entries(leaf, header, entries); | |
1050 | btrfs_set_free_space_bitmaps(leaf, header, bitmaps); | |
1051 | btrfs_set_free_space_generation(leaf, header, trans->transid); | |
1052 | btrfs_mark_buffer_dirty(leaf); | |
b3b4aa74 | 1053 | btrfs_release_path(path); |
0cb59c99 | 1054 | |
c09544e0 | 1055 | err = 0; |
2f356126 | 1056 | out: |
a67509c3 | 1057 | io_ctl_free(&io_ctl); |
c09544e0 | 1058 | if (err) { |
a67509c3 | 1059 | invalidate_inode_pages2(inode->i_mapping); |
0cb59c99 JB |
1060 | BTRFS_I(inode)->generation = 0; |
1061 | } | |
0cb59c99 | 1062 | btrfs_update_inode(trans, root, inode); |
c09544e0 | 1063 | return err; |
a67509c3 JB |
1064 | |
1065 | out_nospc: | |
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); | |
1070 | } | |
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, GFP_NOFS); | |
1074 | goto out; | |
0414efae LZ |
1075 | } |
1076 | ||
1077 | int btrfs_write_out_cache(struct btrfs_root *root, | |
1078 | struct btrfs_trans_handle *trans, | |
1079 | struct btrfs_block_group_cache *block_group, | |
1080 | struct btrfs_path *path) | |
1081 | { | |
1082 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
1083 | struct inode *inode; | |
1084 | int ret = 0; | |
1085 | ||
1086 | root = root->fs_info->tree_root; | |
1087 | ||
1088 | spin_lock(&block_group->lock); | |
1089 | if (block_group->disk_cache_state < BTRFS_DC_SETUP) { | |
1090 | spin_unlock(&block_group->lock); | |
1091 | return 0; | |
1092 | } | |
1093 | spin_unlock(&block_group->lock); | |
1094 | ||
1095 | inode = lookup_free_space_inode(root, block_group, path); | |
1096 | if (IS_ERR(inode)) | |
1097 | return 0; | |
1098 | ||
1099 | ret = __btrfs_write_out_cache(root, inode, ctl, block_group, trans, | |
1100 | path, block_group->key.objectid); | |
c09544e0 | 1101 | if (ret) { |
0414efae LZ |
1102 | spin_lock(&block_group->lock); |
1103 | block_group->disk_cache_state = BTRFS_DC_ERROR; | |
1104 | spin_unlock(&block_group->lock); | |
82d5902d | 1105 | ret = 0; |
c09544e0 | 1106 | #ifdef DEBUG |
c2cf52eb SK |
1107 | btrfs_err(root->fs_info, |
1108 | "failed to write free space cache for block group %llu", | |
1109 | block_group->key.objectid); | |
c09544e0 | 1110 | #endif |
0414efae LZ |
1111 | } |
1112 | ||
0cb59c99 JB |
1113 | iput(inode); |
1114 | return ret; | |
1115 | } | |
1116 | ||
34d52cb6 | 1117 | static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit, |
96303081 | 1118 | u64 offset) |
0f9dd46c | 1119 | { |
96303081 JB |
1120 | BUG_ON(offset < bitmap_start); |
1121 | offset -= bitmap_start; | |
34d52cb6 | 1122 | return (unsigned long)(div_u64(offset, unit)); |
96303081 | 1123 | } |
0f9dd46c | 1124 | |
34d52cb6 | 1125 | static inline unsigned long bytes_to_bits(u64 bytes, u32 unit) |
96303081 | 1126 | { |
34d52cb6 | 1127 | return (unsigned long)(div_u64(bytes, unit)); |
96303081 | 1128 | } |
0f9dd46c | 1129 | |
34d52cb6 | 1130 | static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1131 | u64 offset) |
1132 | { | |
1133 | u64 bitmap_start; | |
1134 | u64 bytes_per_bitmap; | |
0f9dd46c | 1135 | |
34d52cb6 LZ |
1136 | bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit; |
1137 | bitmap_start = offset - ctl->start; | |
96303081 JB |
1138 | bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap); |
1139 | bitmap_start *= bytes_per_bitmap; | |
34d52cb6 | 1140 | bitmap_start += ctl->start; |
0f9dd46c | 1141 | |
96303081 | 1142 | return bitmap_start; |
0f9dd46c JB |
1143 | } |
1144 | ||
96303081 JB |
1145 | static int tree_insert_offset(struct rb_root *root, u64 offset, |
1146 | struct rb_node *node, int bitmap) | |
0f9dd46c JB |
1147 | { |
1148 | struct rb_node **p = &root->rb_node; | |
1149 | struct rb_node *parent = NULL; | |
1150 | struct btrfs_free_space *info; | |
1151 | ||
1152 | while (*p) { | |
1153 | parent = *p; | |
96303081 | 1154 | info = rb_entry(parent, struct btrfs_free_space, offset_index); |
0f9dd46c | 1155 | |
96303081 | 1156 | if (offset < info->offset) { |
0f9dd46c | 1157 | p = &(*p)->rb_left; |
96303081 | 1158 | } else if (offset > info->offset) { |
0f9dd46c | 1159 | p = &(*p)->rb_right; |
96303081 JB |
1160 | } else { |
1161 | /* | |
1162 | * we could have a bitmap entry and an extent entry | |
1163 | * share the same offset. If this is the case, we want | |
1164 | * the extent entry to always be found first if we do a | |
1165 | * linear search through the tree, since we want to have | |
1166 | * the quickest allocation time, and allocating from an | |
1167 | * extent is faster than allocating from a bitmap. So | |
1168 | * if we're inserting a bitmap and we find an entry at | |
1169 | * this offset, we want to go right, or after this entry | |
1170 | * logically. If we are inserting an extent and we've | |
1171 | * found a bitmap, we want to go left, or before | |
1172 | * logically. | |
1173 | */ | |
1174 | if (bitmap) { | |
207dde82 JB |
1175 | if (info->bitmap) { |
1176 | WARN_ON_ONCE(1); | |
1177 | return -EEXIST; | |
1178 | } | |
96303081 JB |
1179 | p = &(*p)->rb_right; |
1180 | } else { | |
207dde82 JB |
1181 | if (!info->bitmap) { |
1182 | WARN_ON_ONCE(1); | |
1183 | return -EEXIST; | |
1184 | } | |
96303081 JB |
1185 | p = &(*p)->rb_left; |
1186 | } | |
1187 | } | |
0f9dd46c JB |
1188 | } |
1189 | ||
1190 | rb_link_node(node, parent, p); | |
1191 | rb_insert_color(node, root); | |
1192 | ||
1193 | return 0; | |
1194 | } | |
1195 | ||
1196 | /* | |
70cb0743 JB |
1197 | * searches the tree for the given offset. |
1198 | * | |
96303081 JB |
1199 | * fuzzy - If this is set, then we are trying to make an allocation, and we just |
1200 | * want a section that has at least bytes size and comes at or after the given | |
1201 | * offset. | |
0f9dd46c | 1202 | */ |
96303081 | 1203 | static struct btrfs_free_space * |
34d52cb6 | 1204 | tree_search_offset(struct btrfs_free_space_ctl *ctl, |
96303081 | 1205 | u64 offset, int bitmap_only, int fuzzy) |
0f9dd46c | 1206 | { |
34d52cb6 | 1207 | struct rb_node *n = ctl->free_space_offset.rb_node; |
96303081 JB |
1208 | struct btrfs_free_space *entry, *prev = NULL; |
1209 | ||
1210 | /* find entry that is closest to the 'offset' */ | |
1211 | while (1) { | |
1212 | if (!n) { | |
1213 | entry = NULL; | |
1214 | break; | |
1215 | } | |
0f9dd46c | 1216 | |
0f9dd46c | 1217 | entry = rb_entry(n, struct btrfs_free_space, offset_index); |
96303081 | 1218 | prev = entry; |
0f9dd46c | 1219 | |
96303081 | 1220 | if (offset < entry->offset) |
0f9dd46c | 1221 | n = n->rb_left; |
96303081 | 1222 | else if (offset > entry->offset) |
0f9dd46c | 1223 | n = n->rb_right; |
96303081 | 1224 | else |
0f9dd46c | 1225 | break; |
0f9dd46c JB |
1226 | } |
1227 | ||
96303081 JB |
1228 | if (bitmap_only) { |
1229 | if (!entry) | |
1230 | return NULL; | |
1231 | if (entry->bitmap) | |
1232 | return entry; | |
0f9dd46c | 1233 | |
96303081 JB |
1234 | /* |
1235 | * bitmap entry and extent entry may share same offset, | |
1236 | * in that case, bitmap entry comes after extent entry. | |
1237 | */ | |
1238 | n = rb_next(n); | |
1239 | if (!n) | |
1240 | return NULL; | |
1241 | entry = rb_entry(n, struct btrfs_free_space, offset_index); | |
1242 | if (entry->offset != offset) | |
1243 | return NULL; | |
0f9dd46c | 1244 | |
96303081 JB |
1245 | WARN_ON(!entry->bitmap); |
1246 | return entry; | |
1247 | } else if (entry) { | |
1248 | if (entry->bitmap) { | |
0f9dd46c | 1249 | /* |
96303081 JB |
1250 | * if previous extent entry covers the offset, |
1251 | * we should return it instead of the bitmap entry | |
0f9dd46c | 1252 | */ |
de6c4115 MX |
1253 | n = rb_prev(&entry->offset_index); |
1254 | if (n) { | |
96303081 JB |
1255 | prev = rb_entry(n, struct btrfs_free_space, |
1256 | offset_index); | |
de6c4115 MX |
1257 | if (!prev->bitmap && |
1258 | prev->offset + prev->bytes > offset) | |
1259 | entry = prev; | |
0f9dd46c | 1260 | } |
96303081 JB |
1261 | } |
1262 | return entry; | |
1263 | } | |
1264 | ||
1265 | if (!prev) | |
1266 | return NULL; | |
1267 | ||
1268 | /* find last entry before the 'offset' */ | |
1269 | entry = prev; | |
1270 | if (entry->offset > offset) { | |
1271 | n = rb_prev(&entry->offset_index); | |
1272 | if (n) { | |
1273 | entry = rb_entry(n, struct btrfs_free_space, | |
1274 | offset_index); | |
1275 | BUG_ON(entry->offset > offset); | |
0f9dd46c | 1276 | } else { |
96303081 JB |
1277 | if (fuzzy) |
1278 | return entry; | |
1279 | else | |
1280 | return NULL; | |
0f9dd46c JB |
1281 | } |
1282 | } | |
1283 | ||
96303081 | 1284 | if (entry->bitmap) { |
de6c4115 MX |
1285 | n = rb_prev(&entry->offset_index); |
1286 | if (n) { | |
96303081 JB |
1287 | prev = rb_entry(n, struct btrfs_free_space, |
1288 | offset_index); | |
de6c4115 MX |
1289 | if (!prev->bitmap && |
1290 | prev->offset + prev->bytes > offset) | |
1291 | return prev; | |
96303081 | 1292 | } |
34d52cb6 | 1293 | if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset) |
96303081 JB |
1294 | return entry; |
1295 | } else if (entry->offset + entry->bytes > offset) | |
1296 | return entry; | |
1297 | ||
1298 | if (!fuzzy) | |
1299 | return NULL; | |
1300 | ||
1301 | while (1) { | |
1302 | if (entry->bitmap) { | |
1303 | if (entry->offset + BITS_PER_BITMAP * | |
34d52cb6 | 1304 | ctl->unit > offset) |
96303081 JB |
1305 | break; |
1306 | } else { | |
1307 | if (entry->offset + entry->bytes > offset) | |
1308 | break; | |
1309 | } | |
1310 | ||
1311 | n = rb_next(&entry->offset_index); | |
1312 | if (!n) | |
1313 | return NULL; | |
1314 | entry = rb_entry(n, struct btrfs_free_space, offset_index); | |
1315 | } | |
1316 | return entry; | |
0f9dd46c JB |
1317 | } |
1318 | ||
f333adb5 | 1319 | static inline void |
34d52cb6 | 1320 | __unlink_free_space(struct btrfs_free_space_ctl *ctl, |
f333adb5 | 1321 | struct btrfs_free_space *info) |
0f9dd46c | 1322 | { |
34d52cb6 LZ |
1323 | rb_erase(&info->offset_index, &ctl->free_space_offset); |
1324 | ctl->free_extents--; | |
f333adb5 LZ |
1325 | } |
1326 | ||
34d52cb6 | 1327 | static void unlink_free_space(struct btrfs_free_space_ctl *ctl, |
f333adb5 LZ |
1328 | struct btrfs_free_space *info) |
1329 | { | |
34d52cb6 LZ |
1330 | __unlink_free_space(ctl, info); |
1331 | ctl->free_space -= info->bytes; | |
0f9dd46c JB |
1332 | } |
1333 | ||
34d52cb6 | 1334 | static int link_free_space(struct btrfs_free_space_ctl *ctl, |
0f9dd46c JB |
1335 | struct btrfs_free_space *info) |
1336 | { | |
1337 | int ret = 0; | |
1338 | ||
96303081 | 1339 | BUG_ON(!info->bitmap && !info->bytes); |
34d52cb6 | 1340 | ret = tree_insert_offset(&ctl->free_space_offset, info->offset, |
96303081 | 1341 | &info->offset_index, (info->bitmap != NULL)); |
0f9dd46c JB |
1342 | if (ret) |
1343 | return ret; | |
1344 | ||
34d52cb6 LZ |
1345 | ctl->free_space += info->bytes; |
1346 | ctl->free_extents++; | |
96303081 JB |
1347 | return ret; |
1348 | } | |
1349 | ||
34d52cb6 | 1350 | static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl) |
96303081 | 1351 | { |
34d52cb6 | 1352 | struct btrfs_block_group_cache *block_group = ctl->private; |
25891f79 JB |
1353 | u64 max_bytes; |
1354 | u64 bitmap_bytes; | |
1355 | u64 extent_bytes; | |
8eb2d829 | 1356 | u64 size = block_group->key.offset; |
96009762 | 1357 | u64 bytes_per_bg = BITS_PER_BITMAP * ctl->unit; |
34d52cb6 LZ |
1358 | int max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg); |
1359 | ||
dde5740f JB |
1360 | max_bitmaps = max(max_bitmaps, 1); |
1361 | ||
34d52cb6 | 1362 | BUG_ON(ctl->total_bitmaps > max_bitmaps); |
96303081 JB |
1363 | |
1364 | /* | |
1365 | * The goal is to keep the total amount of memory used per 1gb of space | |
1366 | * at or below 32k, so we need to adjust how much memory we allow to be | |
1367 | * used by extent based free space tracking | |
1368 | */ | |
8eb2d829 LZ |
1369 | if (size < 1024 * 1024 * 1024) |
1370 | max_bytes = MAX_CACHE_BYTES_PER_GIG; | |
1371 | else | |
1372 | max_bytes = MAX_CACHE_BYTES_PER_GIG * | |
1373 | div64_u64(size, 1024 * 1024 * 1024); | |
96303081 | 1374 | |
25891f79 JB |
1375 | /* |
1376 | * we want to account for 1 more bitmap than what we have so we can make | |
1377 | * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as | |
1378 | * we add more bitmaps. | |
1379 | */ | |
34d52cb6 | 1380 | bitmap_bytes = (ctl->total_bitmaps + 1) * PAGE_CACHE_SIZE; |
96303081 | 1381 | |
25891f79 | 1382 | if (bitmap_bytes >= max_bytes) { |
34d52cb6 | 1383 | ctl->extents_thresh = 0; |
25891f79 JB |
1384 | return; |
1385 | } | |
96303081 | 1386 | |
25891f79 JB |
1387 | /* |
1388 | * we want the extent entry threshold to always be at most 1/2 the maxw | |
1389 | * bytes we can have, or whatever is less than that. | |
1390 | */ | |
1391 | extent_bytes = max_bytes - bitmap_bytes; | |
1392 | extent_bytes = min_t(u64, extent_bytes, div64_u64(max_bytes, 2)); | |
96303081 | 1393 | |
34d52cb6 | 1394 | ctl->extents_thresh = |
25891f79 | 1395 | div64_u64(extent_bytes, (sizeof(struct btrfs_free_space))); |
96303081 JB |
1396 | } |
1397 | ||
bb3ac5a4 MX |
1398 | static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl, |
1399 | struct btrfs_free_space *info, | |
1400 | u64 offset, u64 bytes) | |
96303081 | 1401 | { |
f38b6e75 | 1402 | unsigned long start, count; |
96303081 | 1403 | |
34d52cb6 LZ |
1404 | start = offset_to_bit(info->offset, ctl->unit, offset); |
1405 | count = bytes_to_bits(bytes, ctl->unit); | |
f38b6e75 | 1406 | BUG_ON(start + count > BITS_PER_BITMAP); |
96303081 | 1407 | |
f38b6e75 | 1408 | bitmap_clear(info->bitmap, start, count); |
96303081 JB |
1409 | |
1410 | info->bytes -= bytes; | |
bb3ac5a4 MX |
1411 | } |
1412 | ||
1413 | static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl, | |
1414 | struct btrfs_free_space *info, u64 offset, | |
1415 | u64 bytes) | |
1416 | { | |
1417 | __bitmap_clear_bits(ctl, info, offset, bytes); | |
34d52cb6 | 1418 | ctl->free_space -= bytes; |
96303081 JB |
1419 | } |
1420 | ||
34d52cb6 | 1421 | static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl, |
817d52f8 JB |
1422 | struct btrfs_free_space *info, u64 offset, |
1423 | u64 bytes) | |
96303081 | 1424 | { |
f38b6e75 | 1425 | unsigned long start, count; |
96303081 | 1426 | |
34d52cb6 LZ |
1427 | start = offset_to_bit(info->offset, ctl->unit, offset); |
1428 | count = bytes_to_bits(bytes, ctl->unit); | |
f38b6e75 | 1429 | BUG_ON(start + count > BITS_PER_BITMAP); |
96303081 | 1430 | |
f38b6e75 | 1431 | bitmap_set(info->bitmap, start, count); |
96303081 JB |
1432 | |
1433 | info->bytes += bytes; | |
34d52cb6 | 1434 | ctl->free_space += bytes; |
96303081 JB |
1435 | } |
1436 | ||
34d52cb6 | 1437 | static int search_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1438 | struct btrfs_free_space *bitmap_info, u64 *offset, |
1439 | u64 *bytes) | |
1440 | { | |
1441 | unsigned long found_bits = 0; | |
1442 | unsigned long bits, i; | |
1443 | unsigned long next_zero; | |
1444 | ||
34d52cb6 | 1445 | i = offset_to_bit(bitmap_info->offset, ctl->unit, |
96303081 | 1446 | max_t(u64, *offset, bitmap_info->offset)); |
34d52cb6 | 1447 | bits = bytes_to_bits(*bytes, ctl->unit); |
96303081 | 1448 | |
ebb3dad4 | 1449 | for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) { |
96303081 JB |
1450 | next_zero = find_next_zero_bit(bitmap_info->bitmap, |
1451 | BITS_PER_BITMAP, i); | |
1452 | if ((next_zero - i) >= bits) { | |
1453 | found_bits = next_zero - i; | |
1454 | break; | |
1455 | } | |
1456 | i = next_zero; | |
1457 | } | |
1458 | ||
1459 | if (found_bits) { | |
34d52cb6 LZ |
1460 | *offset = (u64)(i * ctl->unit) + bitmap_info->offset; |
1461 | *bytes = (u64)(found_bits) * ctl->unit; | |
96303081 JB |
1462 | return 0; |
1463 | } | |
1464 | ||
1465 | return -1; | |
1466 | } | |
1467 | ||
34d52cb6 | 1468 | static struct btrfs_free_space * |
53b381b3 DW |
1469 | find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes, |
1470 | unsigned long align) | |
96303081 JB |
1471 | { |
1472 | struct btrfs_free_space *entry; | |
1473 | struct rb_node *node; | |
53b381b3 DW |
1474 | u64 ctl_off; |
1475 | u64 tmp; | |
1476 | u64 align_off; | |
96303081 JB |
1477 | int ret; |
1478 | ||
34d52cb6 | 1479 | if (!ctl->free_space_offset.rb_node) |
96303081 JB |
1480 | return NULL; |
1481 | ||
34d52cb6 | 1482 | entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1); |
96303081 JB |
1483 | if (!entry) |
1484 | return NULL; | |
1485 | ||
1486 | for (node = &entry->offset_index; node; node = rb_next(node)) { | |
1487 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
1488 | if (entry->bytes < *bytes) | |
1489 | continue; | |
1490 | ||
53b381b3 DW |
1491 | /* make sure the space returned is big enough |
1492 | * to match our requested alignment | |
1493 | */ | |
1494 | if (*bytes >= align) { | |
1495 | ctl_off = entry->offset - ctl->start; | |
1496 | tmp = ctl_off + align - 1;; | |
1497 | do_div(tmp, align); | |
1498 | tmp = tmp * align + ctl->start; | |
1499 | align_off = tmp - entry->offset; | |
1500 | } else { | |
1501 | align_off = 0; | |
1502 | tmp = entry->offset; | |
1503 | } | |
1504 | ||
1505 | if (entry->bytes < *bytes + align_off) | |
1506 | continue; | |
1507 | ||
96303081 | 1508 | if (entry->bitmap) { |
53b381b3 DW |
1509 | ret = search_bitmap(ctl, entry, &tmp, bytes); |
1510 | if (!ret) { | |
1511 | *offset = tmp; | |
96303081 | 1512 | return entry; |
53b381b3 | 1513 | } |
96303081 JB |
1514 | continue; |
1515 | } | |
1516 | ||
53b381b3 DW |
1517 | *offset = tmp; |
1518 | *bytes = entry->bytes - align_off; | |
96303081 JB |
1519 | return entry; |
1520 | } | |
1521 | ||
1522 | return NULL; | |
1523 | } | |
1524 | ||
34d52cb6 | 1525 | static void add_new_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1526 | struct btrfs_free_space *info, u64 offset) |
1527 | { | |
34d52cb6 | 1528 | info->offset = offset_to_bitmap(ctl, offset); |
f019f426 | 1529 | info->bytes = 0; |
f2d0f676 | 1530 | INIT_LIST_HEAD(&info->list); |
34d52cb6 LZ |
1531 | link_free_space(ctl, info); |
1532 | ctl->total_bitmaps++; | |
96303081 | 1533 | |
34d52cb6 | 1534 | ctl->op->recalc_thresholds(ctl); |
96303081 JB |
1535 | } |
1536 | ||
34d52cb6 | 1537 | static void free_bitmap(struct btrfs_free_space_ctl *ctl, |
edf6e2d1 LZ |
1538 | struct btrfs_free_space *bitmap_info) |
1539 | { | |
34d52cb6 | 1540 | unlink_free_space(ctl, bitmap_info); |
edf6e2d1 | 1541 | kfree(bitmap_info->bitmap); |
dc89e982 | 1542 | kmem_cache_free(btrfs_free_space_cachep, bitmap_info); |
34d52cb6 LZ |
1543 | ctl->total_bitmaps--; |
1544 | ctl->op->recalc_thresholds(ctl); | |
edf6e2d1 LZ |
1545 | } |
1546 | ||
34d52cb6 | 1547 | static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1548 | struct btrfs_free_space *bitmap_info, |
1549 | u64 *offset, u64 *bytes) | |
1550 | { | |
1551 | u64 end; | |
6606bb97 JB |
1552 | u64 search_start, search_bytes; |
1553 | int ret; | |
96303081 JB |
1554 | |
1555 | again: | |
34d52cb6 | 1556 | end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1; |
96303081 | 1557 | |
6606bb97 | 1558 | /* |
bdb7d303 JB |
1559 | * We need to search for bits in this bitmap. We could only cover some |
1560 | * of the extent in this bitmap thanks to how we add space, so we need | |
1561 | * to search for as much as it as we can and clear that amount, and then | |
1562 | * go searching for the next bit. | |
6606bb97 JB |
1563 | */ |
1564 | search_start = *offset; | |
bdb7d303 | 1565 | search_bytes = ctl->unit; |
13dbc089 | 1566 | search_bytes = min(search_bytes, end - search_start + 1); |
34d52cb6 | 1567 | ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes); |
b50c6e25 JB |
1568 | if (ret < 0 || search_start != *offset) |
1569 | return -EINVAL; | |
6606bb97 | 1570 | |
bdb7d303 JB |
1571 | /* We may have found more bits than what we need */ |
1572 | search_bytes = min(search_bytes, *bytes); | |
1573 | ||
1574 | /* Cannot clear past the end of the bitmap */ | |
1575 | search_bytes = min(search_bytes, end - search_start + 1); | |
1576 | ||
1577 | bitmap_clear_bits(ctl, bitmap_info, search_start, search_bytes); | |
1578 | *offset += search_bytes; | |
1579 | *bytes -= search_bytes; | |
96303081 JB |
1580 | |
1581 | if (*bytes) { | |
6606bb97 | 1582 | struct rb_node *next = rb_next(&bitmap_info->offset_index); |
edf6e2d1 | 1583 | if (!bitmap_info->bytes) |
34d52cb6 | 1584 | free_bitmap(ctl, bitmap_info); |
96303081 | 1585 | |
6606bb97 JB |
1586 | /* |
1587 | * no entry after this bitmap, but we still have bytes to | |
1588 | * remove, so something has gone wrong. | |
1589 | */ | |
1590 | if (!next) | |
96303081 JB |
1591 | return -EINVAL; |
1592 | ||
6606bb97 JB |
1593 | bitmap_info = rb_entry(next, struct btrfs_free_space, |
1594 | offset_index); | |
1595 | ||
1596 | /* | |
1597 | * if the next entry isn't a bitmap we need to return to let the | |
1598 | * extent stuff do its work. | |
1599 | */ | |
96303081 JB |
1600 | if (!bitmap_info->bitmap) |
1601 | return -EAGAIN; | |
1602 | ||
6606bb97 JB |
1603 | /* |
1604 | * Ok the next item is a bitmap, but it may not actually hold | |
1605 | * the information for the rest of this free space stuff, so | |
1606 | * look for it, and if we don't find it return so we can try | |
1607 | * everything over again. | |
1608 | */ | |
1609 | search_start = *offset; | |
bdb7d303 | 1610 | search_bytes = ctl->unit; |
34d52cb6 | 1611 | ret = search_bitmap(ctl, bitmap_info, &search_start, |
6606bb97 JB |
1612 | &search_bytes); |
1613 | if (ret < 0 || search_start != *offset) | |
1614 | return -EAGAIN; | |
1615 | ||
96303081 | 1616 | goto again; |
edf6e2d1 | 1617 | } else if (!bitmap_info->bytes) |
34d52cb6 | 1618 | free_bitmap(ctl, bitmap_info); |
96303081 JB |
1619 | |
1620 | return 0; | |
1621 | } | |
1622 | ||
2cdc342c JB |
1623 | static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl, |
1624 | struct btrfs_free_space *info, u64 offset, | |
1625 | u64 bytes) | |
1626 | { | |
1627 | u64 bytes_to_set = 0; | |
1628 | u64 end; | |
1629 | ||
1630 | end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit); | |
1631 | ||
1632 | bytes_to_set = min(end - offset, bytes); | |
1633 | ||
1634 | bitmap_set_bits(ctl, info, offset, bytes_to_set); | |
1635 | ||
1636 | return bytes_to_set; | |
1637 | ||
1638 | } | |
1639 | ||
34d52cb6 LZ |
1640 | static bool use_bitmap(struct btrfs_free_space_ctl *ctl, |
1641 | struct btrfs_free_space *info) | |
96303081 | 1642 | { |
34d52cb6 | 1643 | struct btrfs_block_group_cache *block_group = ctl->private; |
96303081 JB |
1644 | |
1645 | /* | |
1646 | * If we are below the extents threshold then we can add this as an | |
1647 | * extent, and don't have to deal with the bitmap | |
1648 | */ | |
34d52cb6 | 1649 | if (ctl->free_extents < ctl->extents_thresh) { |
32cb0840 JB |
1650 | /* |
1651 | * If this block group has some small extents we don't want to | |
1652 | * use up all of our free slots in the cache with them, we want | |
1653 | * to reserve them to larger extents, however if we have plent | |
1654 | * of cache left then go ahead an dadd them, no sense in adding | |
1655 | * the overhead of a bitmap if we don't have to. | |
1656 | */ | |
1657 | if (info->bytes <= block_group->sectorsize * 4) { | |
34d52cb6 LZ |
1658 | if (ctl->free_extents * 2 <= ctl->extents_thresh) |
1659 | return false; | |
32cb0840 | 1660 | } else { |
34d52cb6 | 1661 | return false; |
32cb0840 JB |
1662 | } |
1663 | } | |
96303081 JB |
1664 | |
1665 | /* | |
dde5740f JB |
1666 | * The original block groups from mkfs can be really small, like 8 |
1667 | * megabytes, so don't bother with a bitmap for those entries. However | |
1668 | * some block groups can be smaller than what a bitmap would cover but | |
1669 | * are still large enough that they could overflow the 32k memory limit, | |
1670 | * so allow those block groups to still be allowed to have a bitmap | |
1671 | * entry. | |
96303081 | 1672 | */ |
dde5740f | 1673 | if (((BITS_PER_BITMAP * ctl->unit) >> 1) > block_group->key.offset) |
34d52cb6 LZ |
1674 | return false; |
1675 | ||
1676 | return true; | |
1677 | } | |
1678 | ||
2cdc342c JB |
1679 | static struct btrfs_free_space_op free_space_op = { |
1680 | .recalc_thresholds = recalculate_thresholds, | |
1681 | .use_bitmap = use_bitmap, | |
1682 | }; | |
1683 | ||
34d52cb6 LZ |
1684 | static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl, |
1685 | struct btrfs_free_space *info) | |
1686 | { | |
1687 | struct btrfs_free_space *bitmap_info; | |
2cdc342c | 1688 | struct btrfs_block_group_cache *block_group = NULL; |
34d52cb6 | 1689 | int added = 0; |
2cdc342c | 1690 | u64 bytes, offset, bytes_added; |
34d52cb6 | 1691 | int ret; |
96303081 JB |
1692 | |
1693 | bytes = info->bytes; | |
1694 | offset = info->offset; | |
1695 | ||
34d52cb6 LZ |
1696 | if (!ctl->op->use_bitmap(ctl, info)) |
1697 | return 0; | |
1698 | ||
2cdc342c JB |
1699 | if (ctl->op == &free_space_op) |
1700 | block_group = ctl->private; | |
38e87880 | 1701 | again: |
2cdc342c JB |
1702 | /* |
1703 | * Since we link bitmaps right into the cluster we need to see if we | |
1704 | * have a cluster here, and if so and it has our bitmap we need to add | |
1705 | * the free space to that bitmap. | |
1706 | */ | |
1707 | if (block_group && !list_empty(&block_group->cluster_list)) { | |
1708 | struct btrfs_free_cluster *cluster; | |
1709 | struct rb_node *node; | |
1710 | struct btrfs_free_space *entry; | |
1711 | ||
1712 | cluster = list_entry(block_group->cluster_list.next, | |
1713 | struct btrfs_free_cluster, | |
1714 | block_group_list); | |
1715 | spin_lock(&cluster->lock); | |
1716 | node = rb_first(&cluster->root); | |
1717 | if (!node) { | |
1718 | spin_unlock(&cluster->lock); | |
38e87880 | 1719 | goto no_cluster_bitmap; |
2cdc342c JB |
1720 | } |
1721 | ||
1722 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
1723 | if (!entry->bitmap) { | |
1724 | spin_unlock(&cluster->lock); | |
38e87880 | 1725 | goto no_cluster_bitmap; |
2cdc342c JB |
1726 | } |
1727 | ||
1728 | if (entry->offset == offset_to_bitmap(ctl, offset)) { | |
1729 | bytes_added = add_bytes_to_bitmap(ctl, entry, | |
1730 | offset, bytes); | |
1731 | bytes -= bytes_added; | |
1732 | offset += bytes_added; | |
1733 | } | |
1734 | spin_unlock(&cluster->lock); | |
1735 | if (!bytes) { | |
1736 | ret = 1; | |
1737 | goto out; | |
1738 | } | |
1739 | } | |
38e87880 CM |
1740 | |
1741 | no_cluster_bitmap: | |
34d52cb6 | 1742 | bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), |
96303081 JB |
1743 | 1, 0); |
1744 | if (!bitmap_info) { | |
1745 | BUG_ON(added); | |
1746 | goto new_bitmap; | |
1747 | } | |
1748 | ||
2cdc342c JB |
1749 | bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes); |
1750 | bytes -= bytes_added; | |
1751 | offset += bytes_added; | |
1752 | added = 0; | |
96303081 JB |
1753 | |
1754 | if (!bytes) { | |
1755 | ret = 1; | |
1756 | goto out; | |
1757 | } else | |
1758 | goto again; | |
1759 | ||
1760 | new_bitmap: | |
1761 | if (info && info->bitmap) { | |
34d52cb6 | 1762 | add_new_bitmap(ctl, info, offset); |
96303081 JB |
1763 | added = 1; |
1764 | info = NULL; | |
1765 | goto again; | |
1766 | } else { | |
34d52cb6 | 1767 | spin_unlock(&ctl->tree_lock); |
96303081 JB |
1768 | |
1769 | /* no pre-allocated info, allocate a new one */ | |
1770 | if (!info) { | |
dc89e982 JB |
1771 | info = kmem_cache_zalloc(btrfs_free_space_cachep, |
1772 | GFP_NOFS); | |
96303081 | 1773 | if (!info) { |
34d52cb6 | 1774 | spin_lock(&ctl->tree_lock); |
96303081 JB |
1775 | ret = -ENOMEM; |
1776 | goto out; | |
1777 | } | |
1778 | } | |
1779 | ||
1780 | /* allocate the bitmap */ | |
1781 | info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS); | |
34d52cb6 | 1782 | spin_lock(&ctl->tree_lock); |
96303081 JB |
1783 | if (!info->bitmap) { |
1784 | ret = -ENOMEM; | |
1785 | goto out; | |
1786 | } | |
1787 | goto again; | |
1788 | } | |
1789 | ||
1790 | out: | |
1791 | if (info) { | |
1792 | if (info->bitmap) | |
1793 | kfree(info->bitmap); | |
dc89e982 | 1794 | kmem_cache_free(btrfs_free_space_cachep, info); |
96303081 | 1795 | } |
0f9dd46c JB |
1796 | |
1797 | return ret; | |
1798 | } | |
1799 | ||
945d8962 | 1800 | static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl, |
f333adb5 | 1801 | struct btrfs_free_space *info, bool update_stat) |
0f9dd46c | 1802 | { |
120d66ee LZ |
1803 | struct btrfs_free_space *left_info; |
1804 | struct btrfs_free_space *right_info; | |
1805 | bool merged = false; | |
1806 | u64 offset = info->offset; | |
1807 | u64 bytes = info->bytes; | |
6226cb0a | 1808 | |
0f9dd46c JB |
1809 | /* |
1810 | * first we want to see if there is free space adjacent to the range we | |
1811 | * are adding, if there is remove that struct and add a new one to | |
1812 | * cover the entire range | |
1813 | */ | |
34d52cb6 | 1814 | right_info = tree_search_offset(ctl, offset + bytes, 0, 0); |
96303081 JB |
1815 | if (right_info && rb_prev(&right_info->offset_index)) |
1816 | left_info = rb_entry(rb_prev(&right_info->offset_index), | |
1817 | struct btrfs_free_space, offset_index); | |
1818 | else | |
34d52cb6 | 1819 | left_info = tree_search_offset(ctl, offset - 1, 0, 0); |
0f9dd46c | 1820 | |
96303081 | 1821 | if (right_info && !right_info->bitmap) { |
f333adb5 | 1822 | if (update_stat) |
34d52cb6 | 1823 | unlink_free_space(ctl, right_info); |
f333adb5 | 1824 | else |
34d52cb6 | 1825 | __unlink_free_space(ctl, right_info); |
6226cb0a | 1826 | info->bytes += right_info->bytes; |
dc89e982 | 1827 | kmem_cache_free(btrfs_free_space_cachep, right_info); |
120d66ee | 1828 | merged = true; |
0f9dd46c JB |
1829 | } |
1830 | ||
96303081 JB |
1831 | if (left_info && !left_info->bitmap && |
1832 | left_info->offset + left_info->bytes == offset) { | |
f333adb5 | 1833 | if (update_stat) |
34d52cb6 | 1834 | unlink_free_space(ctl, left_info); |
f333adb5 | 1835 | else |
34d52cb6 | 1836 | __unlink_free_space(ctl, left_info); |
6226cb0a JB |
1837 | info->offset = left_info->offset; |
1838 | info->bytes += left_info->bytes; | |
dc89e982 | 1839 | kmem_cache_free(btrfs_free_space_cachep, left_info); |
120d66ee | 1840 | merged = true; |
0f9dd46c JB |
1841 | } |
1842 | ||
120d66ee LZ |
1843 | return merged; |
1844 | } | |
1845 | ||
581bb050 LZ |
1846 | int __btrfs_add_free_space(struct btrfs_free_space_ctl *ctl, |
1847 | u64 offset, u64 bytes) | |
120d66ee LZ |
1848 | { |
1849 | struct btrfs_free_space *info; | |
1850 | int ret = 0; | |
1851 | ||
dc89e982 | 1852 | info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS); |
120d66ee LZ |
1853 | if (!info) |
1854 | return -ENOMEM; | |
1855 | ||
1856 | info->offset = offset; | |
1857 | info->bytes = bytes; | |
1858 | ||
34d52cb6 | 1859 | spin_lock(&ctl->tree_lock); |
120d66ee | 1860 | |
34d52cb6 | 1861 | if (try_merge_free_space(ctl, info, true)) |
120d66ee LZ |
1862 | goto link; |
1863 | ||
1864 | /* | |
1865 | * There was no extent directly to the left or right of this new | |
1866 | * extent then we know we're going to have to allocate a new extent, so | |
1867 | * before we do that see if we need to drop this into a bitmap | |
1868 | */ | |
34d52cb6 | 1869 | ret = insert_into_bitmap(ctl, info); |
120d66ee LZ |
1870 | if (ret < 0) { |
1871 | goto out; | |
1872 | } else if (ret) { | |
1873 | ret = 0; | |
1874 | goto out; | |
1875 | } | |
1876 | link: | |
34d52cb6 | 1877 | ret = link_free_space(ctl, info); |
0f9dd46c | 1878 | if (ret) |
dc89e982 | 1879 | kmem_cache_free(btrfs_free_space_cachep, info); |
96303081 | 1880 | out: |
34d52cb6 | 1881 | spin_unlock(&ctl->tree_lock); |
6226cb0a | 1882 | |
0f9dd46c | 1883 | if (ret) { |
96303081 | 1884 | printk(KERN_CRIT "btrfs: unable to add free space :%d\n", ret); |
c293498b | 1885 | BUG_ON(ret == -EEXIST); |
0f9dd46c JB |
1886 | } |
1887 | ||
0f9dd46c JB |
1888 | return ret; |
1889 | } | |
1890 | ||
6226cb0a JB |
1891 | int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group, |
1892 | u64 offset, u64 bytes) | |
0f9dd46c | 1893 | { |
34d52cb6 | 1894 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0f9dd46c | 1895 | struct btrfs_free_space *info; |
b0175117 JB |
1896 | int ret; |
1897 | bool re_search = false; | |
0f9dd46c | 1898 | |
34d52cb6 | 1899 | spin_lock(&ctl->tree_lock); |
6226cb0a | 1900 | |
96303081 | 1901 | again: |
b0175117 | 1902 | ret = 0; |
bdb7d303 JB |
1903 | if (!bytes) |
1904 | goto out_lock; | |
1905 | ||
34d52cb6 | 1906 | info = tree_search_offset(ctl, offset, 0, 0); |
96303081 | 1907 | if (!info) { |
6606bb97 JB |
1908 | /* |
1909 | * oops didn't find an extent that matched the space we wanted | |
1910 | * to remove, look for a bitmap instead | |
1911 | */ | |
34d52cb6 | 1912 | info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), |
6606bb97 JB |
1913 | 1, 0); |
1914 | if (!info) { | |
b0175117 JB |
1915 | /* |
1916 | * If we found a partial bit of our free space in a | |
1917 | * bitmap but then couldn't find the other part this may | |
1918 | * be a problem, so WARN about it. | |
24a70313 | 1919 | */ |
b0175117 | 1920 | WARN_ON(re_search); |
6606bb97 JB |
1921 | goto out_lock; |
1922 | } | |
96303081 JB |
1923 | } |
1924 | ||
b0175117 | 1925 | re_search = false; |
bdb7d303 | 1926 | if (!info->bitmap) { |
34d52cb6 | 1927 | unlink_free_space(ctl, info); |
bdb7d303 JB |
1928 | if (offset == info->offset) { |
1929 | u64 to_free = min(bytes, info->bytes); | |
1930 | ||
1931 | info->bytes -= to_free; | |
1932 | info->offset += to_free; | |
1933 | if (info->bytes) { | |
1934 | ret = link_free_space(ctl, info); | |
1935 | WARN_ON(ret); | |
1936 | } else { | |
1937 | kmem_cache_free(btrfs_free_space_cachep, info); | |
1938 | } | |
0f9dd46c | 1939 | |
bdb7d303 JB |
1940 | offset += to_free; |
1941 | bytes -= to_free; | |
1942 | goto again; | |
1943 | } else { | |
1944 | u64 old_end = info->bytes + info->offset; | |
9b49c9b9 | 1945 | |
bdb7d303 | 1946 | info->bytes = offset - info->offset; |
34d52cb6 | 1947 | ret = link_free_space(ctl, info); |
96303081 JB |
1948 | WARN_ON(ret); |
1949 | if (ret) | |
1950 | goto out_lock; | |
96303081 | 1951 | |
bdb7d303 JB |
1952 | /* Not enough bytes in this entry to satisfy us */ |
1953 | if (old_end < offset + bytes) { | |
1954 | bytes -= old_end - offset; | |
1955 | offset = old_end; | |
1956 | goto again; | |
1957 | } else if (old_end == offset + bytes) { | |
1958 | /* all done */ | |
1959 | goto out_lock; | |
1960 | } | |
1961 | spin_unlock(&ctl->tree_lock); | |
1962 | ||
1963 | ret = btrfs_add_free_space(block_group, offset + bytes, | |
1964 | old_end - (offset + bytes)); | |
1965 | WARN_ON(ret); | |
1966 | goto out; | |
1967 | } | |
0f9dd46c | 1968 | } |
96303081 | 1969 | |
34d52cb6 | 1970 | ret = remove_from_bitmap(ctl, info, &offset, &bytes); |
b0175117 JB |
1971 | if (ret == -EAGAIN) { |
1972 | re_search = true; | |
96303081 | 1973 | goto again; |
b0175117 | 1974 | } |
96303081 | 1975 | out_lock: |
34d52cb6 | 1976 | spin_unlock(&ctl->tree_lock); |
0f9dd46c | 1977 | out: |
25179201 JB |
1978 | return ret; |
1979 | } | |
1980 | ||
0f9dd46c JB |
1981 | void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group, |
1982 | u64 bytes) | |
1983 | { | |
34d52cb6 | 1984 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0f9dd46c JB |
1985 | struct btrfs_free_space *info; |
1986 | struct rb_node *n; | |
1987 | int count = 0; | |
1988 | ||
34d52cb6 | 1989 | for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) { |
0f9dd46c | 1990 | info = rb_entry(n, struct btrfs_free_space, offset_index); |
f6175efa | 1991 | if (info->bytes >= bytes && !block_group->ro) |
0f9dd46c | 1992 | count++; |
96303081 | 1993 | printk(KERN_CRIT "entry offset %llu, bytes %llu, bitmap %s\n", |
21380931 | 1994 | (unsigned long long)info->offset, |
96303081 JB |
1995 | (unsigned long long)info->bytes, |
1996 | (info->bitmap) ? "yes" : "no"); | |
0f9dd46c | 1997 | } |
96303081 JB |
1998 | printk(KERN_INFO "block group has cluster?: %s\n", |
1999 | list_empty(&block_group->cluster_list) ? "no" : "yes"); | |
0f9dd46c JB |
2000 | printk(KERN_INFO "%d blocks of free space at or bigger than bytes is" |
2001 | "\n", count); | |
2002 | } | |
2003 | ||
34d52cb6 | 2004 | void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group) |
0f9dd46c | 2005 | { |
34d52cb6 | 2006 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0f9dd46c | 2007 | |
34d52cb6 LZ |
2008 | spin_lock_init(&ctl->tree_lock); |
2009 | ctl->unit = block_group->sectorsize; | |
2010 | ctl->start = block_group->key.objectid; | |
2011 | ctl->private = block_group; | |
2012 | ctl->op = &free_space_op; | |
0f9dd46c | 2013 | |
34d52cb6 LZ |
2014 | /* |
2015 | * we only want to have 32k of ram per block group for keeping | |
2016 | * track of free space, and if we pass 1/2 of that we want to | |
2017 | * start converting things over to using bitmaps | |
2018 | */ | |
2019 | ctl->extents_thresh = ((1024 * 32) / 2) / | |
2020 | sizeof(struct btrfs_free_space); | |
0f9dd46c JB |
2021 | } |
2022 | ||
fa9c0d79 CM |
2023 | /* |
2024 | * for a given cluster, put all of its extents back into the free | |
2025 | * space cache. If the block group passed doesn't match the block group | |
2026 | * pointed to by the cluster, someone else raced in and freed the | |
2027 | * cluster already. In that case, we just return without changing anything | |
2028 | */ | |
2029 | static int | |
2030 | __btrfs_return_cluster_to_free_space( | |
2031 | struct btrfs_block_group_cache *block_group, | |
2032 | struct btrfs_free_cluster *cluster) | |
2033 | { | |
34d52cb6 | 2034 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
fa9c0d79 CM |
2035 | struct btrfs_free_space *entry; |
2036 | struct rb_node *node; | |
2037 | ||
2038 | spin_lock(&cluster->lock); | |
2039 | if (cluster->block_group != block_group) | |
2040 | goto out; | |
2041 | ||
96303081 | 2042 | cluster->block_group = NULL; |
fa9c0d79 | 2043 | cluster->window_start = 0; |
96303081 | 2044 | list_del_init(&cluster->block_group_list); |
96303081 | 2045 | |
fa9c0d79 | 2046 | node = rb_first(&cluster->root); |
96303081 | 2047 | while (node) { |
4e69b598 JB |
2048 | bool bitmap; |
2049 | ||
fa9c0d79 CM |
2050 | entry = rb_entry(node, struct btrfs_free_space, offset_index); |
2051 | node = rb_next(&entry->offset_index); | |
2052 | rb_erase(&entry->offset_index, &cluster->root); | |
4e69b598 JB |
2053 | |
2054 | bitmap = (entry->bitmap != NULL); | |
2055 | if (!bitmap) | |
34d52cb6 LZ |
2056 | try_merge_free_space(ctl, entry, false); |
2057 | tree_insert_offset(&ctl->free_space_offset, | |
4e69b598 | 2058 | entry->offset, &entry->offset_index, bitmap); |
fa9c0d79 | 2059 | } |
6bef4d31 | 2060 | cluster->root = RB_ROOT; |
96303081 | 2061 | |
fa9c0d79 CM |
2062 | out: |
2063 | spin_unlock(&cluster->lock); | |
96303081 | 2064 | btrfs_put_block_group(block_group); |
fa9c0d79 CM |
2065 | return 0; |
2066 | } | |
2067 | ||
48a3b636 ES |
2068 | static void __btrfs_remove_free_space_cache_locked( |
2069 | struct btrfs_free_space_ctl *ctl) | |
0f9dd46c JB |
2070 | { |
2071 | struct btrfs_free_space *info; | |
2072 | struct rb_node *node; | |
581bb050 | 2073 | |
581bb050 LZ |
2074 | while ((node = rb_last(&ctl->free_space_offset)) != NULL) { |
2075 | info = rb_entry(node, struct btrfs_free_space, offset_index); | |
9b90f513 JB |
2076 | if (!info->bitmap) { |
2077 | unlink_free_space(ctl, info); | |
2078 | kmem_cache_free(btrfs_free_space_cachep, info); | |
2079 | } else { | |
2080 | free_bitmap(ctl, info); | |
2081 | } | |
581bb050 LZ |
2082 | if (need_resched()) { |
2083 | spin_unlock(&ctl->tree_lock); | |
2084 | cond_resched(); | |
2085 | spin_lock(&ctl->tree_lock); | |
2086 | } | |
2087 | } | |
09655373 CM |
2088 | } |
2089 | ||
2090 | void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl) | |
2091 | { | |
2092 | spin_lock(&ctl->tree_lock); | |
2093 | __btrfs_remove_free_space_cache_locked(ctl); | |
581bb050 LZ |
2094 | spin_unlock(&ctl->tree_lock); |
2095 | } | |
2096 | ||
2097 | void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group) | |
2098 | { | |
2099 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
fa9c0d79 | 2100 | struct btrfs_free_cluster *cluster; |
96303081 | 2101 | struct list_head *head; |
0f9dd46c | 2102 | |
34d52cb6 | 2103 | spin_lock(&ctl->tree_lock); |
96303081 JB |
2104 | while ((head = block_group->cluster_list.next) != |
2105 | &block_group->cluster_list) { | |
2106 | cluster = list_entry(head, struct btrfs_free_cluster, | |
2107 | block_group_list); | |
fa9c0d79 CM |
2108 | |
2109 | WARN_ON(cluster->block_group != block_group); | |
2110 | __btrfs_return_cluster_to_free_space(block_group, cluster); | |
96303081 | 2111 | if (need_resched()) { |
34d52cb6 | 2112 | spin_unlock(&ctl->tree_lock); |
96303081 | 2113 | cond_resched(); |
34d52cb6 | 2114 | spin_lock(&ctl->tree_lock); |
96303081 | 2115 | } |
fa9c0d79 | 2116 | } |
09655373 | 2117 | __btrfs_remove_free_space_cache_locked(ctl); |
34d52cb6 | 2118 | spin_unlock(&ctl->tree_lock); |
fa9c0d79 | 2119 | |
0f9dd46c JB |
2120 | } |
2121 | ||
6226cb0a JB |
2122 | u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group, |
2123 | u64 offset, u64 bytes, u64 empty_size) | |
0f9dd46c | 2124 | { |
34d52cb6 | 2125 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
6226cb0a | 2126 | struct btrfs_free_space *entry = NULL; |
96303081 | 2127 | u64 bytes_search = bytes + empty_size; |
6226cb0a | 2128 | u64 ret = 0; |
53b381b3 DW |
2129 | u64 align_gap = 0; |
2130 | u64 align_gap_len = 0; | |
0f9dd46c | 2131 | |
34d52cb6 | 2132 | spin_lock(&ctl->tree_lock); |
53b381b3 DW |
2133 | entry = find_free_space(ctl, &offset, &bytes_search, |
2134 | block_group->full_stripe_len); | |
6226cb0a | 2135 | if (!entry) |
96303081 JB |
2136 | goto out; |
2137 | ||
2138 | ret = offset; | |
2139 | if (entry->bitmap) { | |
34d52cb6 | 2140 | bitmap_clear_bits(ctl, entry, offset, bytes); |
edf6e2d1 | 2141 | if (!entry->bytes) |
34d52cb6 | 2142 | free_bitmap(ctl, entry); |
96303081 | 2143 | } else { |
53b381b3 | 2144 | |
34d52cb6 | 2145 | unlink_free_space(ctl, entry); |
53b381b3 DW |
2146 | align_gap_len = offset - entry->offset; |
2147 | align_gap = entry->offset; | |
2148 | ||
2149 | entry->offset = offset + bytes; | |
2150 | WARN_ON(entry->bytes < bytes + align_gap_len); | |
2151 | ||
2152 | entry->bytes -= bytes + align_gap_len; | |
6226cb0a | 2153 | if (!entry->bytes) |
dc89e982 | 2154 | kmem_cache_free(btrfs_free_space_cachep, entry); |
6226cb0a | 2155 | else |
34d52cb6 | 2156 | link_free_space(ctl, entry); |
6226cb0a | 2157 | } |
0f9dd46c | 2158 | |
96303081 | 2159 | out: |
34d52cb6 | 2160 | spin_unlock(&ctl->tree_lock); |
817d52f8 | 2161 | |
53b381b3 DW |
2162 | if (align_gap_len) |
2163 | __btrfs_add_free_space(ctl, align_gap, align_gap_len); | |
0f9dd46c JB |
2164 | return ret; |
2165 | } | |
fa9c0d79 CM |
2166 | |
2167 | /* | |
2168 | * given a cluster, put all of its extents back into the free space | |
2169 | * cache. If a block group is passed, this function will only free | |
2170 | * a cluster that belongs to the passed block group. | |
2171 | * | |
2172 | * Otherwise, it'll get a reference on the block group pointed to by the | |
2173 | * cluster and remove the cluster from it. | |
2174 | */ | |
2175 | int btrfs_return_cluster_to_free_space( | |
2176 | struct btrfs_block_group_cache *block_group, | |
2177 | struct btrfs_free_cluster *cluster) | |
2178 | { | |
34d52cb6 | 2179 | struct btrfs_free_space_ctl *ctl; |
fa9c0d79 CM |
2180 | int ret; |
2181 | ||
2182 | /* first, get a safe pointer to the block group */ | |
2183 | spin_lock(&cluster->lock); | |
2184 | if (!block_group) { | |
2185 | block_group = cluster->block_group; | |
2186 | if (!block_group) { | |
2187 | spin_unlock(&cluster->lock); | |
2188 | return 0; | |
2189 | } | |
2190 | } else if (cluster->block_group != block_group) { | |
2191 | /* someone else has already freed it don't redo their work */ | |
2192 | spin_unlock(&cluster->lock); | |
2193 | return 0; | |
2194 | } | |
2195 | atomic_inc(&block_group->count); | |
2196 | spin_unlock(&cluster->lock); | |
2197 | ||
34d52cb6 LZ |
2198 | ctl = block_group->free_space_ctl; |
2199 | ||
fa9c0d79 | 2200 | /* now return any extents the cluster had on it */ |
34d52cb6 | 2201 | spin_lock(&ctl->tree_lock); |
fa9c0d79 | 2202 | ret = __btrfs_return_cluster_to_free_space(block_group, cluster); |
34d52cb6 | 2203 | spin_unlock(&ctl->tree_lock); |
fa9c0d79 CM |
2204 | |
2205 | /* finally drop our ref */ | |
2206 | btrfs_put_block_group(block_group); | |
2207 | return ret; | |
2208 | } | |
2209 | ||
96303081 JB |
2210 | static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group, |
2211 | struct btrfs_free_cluster *cluster, | |
4e69b598 | 2212 | struct btrfs_free_space *entry, |
96303081 JB |
2213 | u64 bytes, u64 min_start) |
2214 | { | |
34d52cb6 | 2215 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
96303081 JB |
2216 | int err; |
2217 | u64 search_start = cluster->window_start; | |
2218 | u64 search_bytes = bytes; | |
2219 | u64 ret = 0; | |
2220 | ||
96303081 JB |
2221 | search_start = min_start; |
2222 | search_bytes = bytes; | |
2223 | ||
34d52cb6 | 2224 | err = search_bitmap(ctl, entry, &search_start, &search_bytes); |
96303081 | 2225 | if (err) |
4e69b598 | 2226 | return 0; |
96303081 JB |
2227 | |
2228 | ret = search_start; | |
bb3ac5a4 | 2229 | __bitmap_clear_bits(ctl, entry, ret, bytes); |
96303081 JB |
2230 | |
2231 | return ret; | |
2232 | } | |
2233 | ||
fa9c0d79 CM |
2234 | /* |
2235 | * given a cluster, try to allocate 'bytes' from it, returns 0 | |
2236 | * if it couldn't find anything suitably large, or a logical disk offset | |
2237 | * if things worked out | |
2238 | */ | |
2239 | u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group, | |
2240 | struct btrfs_free_cluster *cluster, u64 bytes, | |
2241 | u64 min_start) | |
2242 | { | |
34d52cb6 | 2243 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
fa9c0d79 CM |
2244 | struct btrfs_free_space *entry = NULL; |
2245 | struct rb_node *node; | |
2246 | u64 ret = 0; | |
2247 | ||
2248 | spin_lock(&cluster->lock); | |
2249 | if (bytes > cluster->max_size) | |
2250 | goto out; | |
2251 | ||
2252 | if (cluster->block_group != block_group) | |
2253 | goto out; | |
2254 | ||
2255 | node = rb_first(&cluster->root); | |
2256 | if (!node) | |
2257 | goto out; | |
2258 | ||
2259 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
fa9c0d79 | 2260 | while(1) { |
4e69b598 JB |
2261 | if (entry->bytes < bytes || |
2262 | (!entry->bitmap && entry->offset < min_start)) { | |
fa9c0d79 CM |
2263 | node = rb_next(&entry->offset_index); |
2264 | if (!node) | |
2265 | break; | |
2266 | entry = rb_entry(node, struct btrfs_free_space, | |
2267 | offset_index); | |
2268 | continue; | |
2269 | } | |
fa9c0d79 | 2270 | |
4e69b598 JB |
2271 | if (entry->bitmap) { |
2272 | ret = btrfs_alloc_from_bitmap(block_group, | |
2273 | cluster, entry, bytes, | |
0b4a9d24 | 2274 | cluster->window_start); |
4e69b598 | 2275 | if (ret == 0) { |
4e69b598 JB |
2276 | node = rb_next(&entry->offset_index); |
2277 | if (!node) | |
2278 | break; | |
2279 | entry = rb_entry(node, struct btrfs_free_space, | |
2280 | offset_index); | |
2281 | continue; | |
2282 | } | |
9b230628 | 2283 | cluster->window_start += bytes; |
4e69b598 | 2284 | } else { |
4e69b598 JB |
2285 | ret = entry->offset; |
2286 | ||
2287 | entry->offset += bytes; | |
2288 | entry->bytes -= bytes; | |
2289 | } | |
fa9c0d79 | 2290 | |
5e71b5d5 | 2291 | if (entry->bytes == 0) |
fa9c0d79 | 2292 | rb_erase(&entry->offset_index, &cluster->root); |
fa9c0d79 CM |
2293 | break; |
2294 | } | |
2295 | out: | |
2296 | spin_unlock(&cluster->lock); | |
96303081 | 2297 | |
5e71b5d5 LZ |
2298 | if (!ret) |
2299 | return 0; | |
2300 | ||
34d52cb6 | 2301 | spin_lock(&ctl->tree_lock); |
5e71b5d5 | 2302 | |
34d52cb6 | 2303 | ctl->free_space -= bytes; |
5e71b5d5 | 2304 | if (entry->bytes == 0) { |
34d52cb6 | 2305 | ctl->free_extents--; |
4e69b598 JB |
2306 | if (entry->bitmap) { |
2307 | kfree(entry->bitmap); | |
34d52cb6 LZ |
2308 | ctl->total_bitmaps--; |
2309 | ctl->op->recalc_thresholds(ctl); | |
4e69b598 | 2310 | } |
dc89e982 | 2311 | kmem_cache_free(btrfs_free_space_cachep, entry); |
5e71b5d5 LZ |
2312 | } |
2313 | ||
34d52cb6 | 2314 | spin_unlock(&ctl->tree_lock); |
5e71b5d5 | 2315 | |
fa9c0d79 CM |
2316 | return ret; |
2317 | } | |
2318 | ||
96303081 JB |
2319 | static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group, |
2320 | struct btrfs_free_space *entry, | |
2321 | struct btrfs_free_cluster *cluster, | |
1bb91902 AO |
2322 | u64 offset, u64 bytes, |
2323 | u64 cont1_bytes, u64 min_bytes) | |
96303081 | 2324 | { |
34d52cb6 | 2325 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
96303081 JB |
2326 | unsigned long next_zero; |
2327 | unsigned long i; | |
1bb91902 AO |
2328 | unsigned long want_bits; |
2329 | unsigned long min_bits; | |
96303081 JB |
2330 | unsigned long found_bits; |
2331 | unsigned long start = 0; | |
2332 | unsigned long total_found = 0; | |
4e69b598 | 2333 | int ret; |
96303081 | 2334 | |
96009762 | 2335 | i = offset_to_bit(entry->offset, ctl->unit, |
96303081 | 2336 | max_t(u64, offset, entry->offset)); |
96009762 WSH |
2337 | want_bits = bytes_to_bits(bytes, ctl->unit); |
2338 | min_bits = bytes_to_bits(min_bytes, ctl->unit); | |
96303081 JB |
2339 | |
2340 | again: | |
2341 | found_bits = 0; | |
ebb3dad4 | 2342 | for_each_set_bit_from(i, entry->bitmap, BITS_PER_BITMAP) { |
96303081 JB |
2343 | next_zero = find_next_zero_bit(entry->bitmap, |
2344 | BITS_PER_BITMAP, i); | |
1bb91902 | 2345 | if (next_zero - i >= min_bits) { |
96303081 JB |
2346 | found_bits = next_zero - i; |
2347 | break; | |
2348 | } | |
2349 | i = next_zero; | |
2350 | } | |
2351 | ||
2352 | if (!found_bits) | |
4e69b598 | 2353 | return -ENOSPC; |
96303081 | 2354 | |
1bb91902 | 2355 | if (!total_found) { |
96303081 | 2356 | start = i; |
b78d09bc | 2357 | cluster->max_size = 0; |
96303081 JB |
2358 | } |
2359 | ||
2360 | total_found += found_bits; | |
2361 | ||
96009762 WSH |
2362 | if (cluster->max_size < found_bits * ctl->unit) |
2363 | cluster->max_size = found_bits * ctl->unit; | |
96303081 | 2364 | |
1bb91902 AO |
2365 | if (total_found < want_bits || cluster->max_size < cont1_bytes) { |
2366 | i = next_zero + 1; | |
96303081 JB |
2367 | goto again; |
2368 | } | |
2369 | ||
96009762 | 2370 | cluster->window_start = start * ctl->unit + entry->offset; |
34d52cb6 | 2371 | rb_erase(&entry->offset_index, &ctl->free_space_offset); |
4e69b598 JB |
2372 | ret = tree_insert_offset(&cluster->root, entry->offset, |
2373 | &entry->offset_index, 1); | |
79787eaa | 2374 | BUG_ON(ret); /* -EEXIST; Logic error */ |
96303081 | 2375 | |
3f7de037 | 2376 | trace_btrfs_setup_cluster(block_group, cluster, |
96009762 | 2377 | total_found * ctl->unit, 1); |
96303081 JB |
2378 | return 0; |
2379 | } | |
2380 | ||
4e69b598 JB |
2381 | /* |
2382 | * This searches the block group for just extents to fill the cluster with. | |
1bb91902 AO |
2383 | * Try to find a cluster with at least bytes total bytes, at least one |
2384 | * extent of cont1_bytes, and other clusters of at least min_bytes. | |
4e69b598 | 2385 | */ |
3de85bb9 JB |
2386 | static noinline int |
2387 | setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group, | |
2388 | struct btrfs_free_cluster *cluster, | |
2389 | struct list_head *bitmaps, u64 offset, u64 bytes, | |
1bb91902 | 2390 | u64 cont1_bytes, u64 min_bytes) |
4e69b598 | 2391 | { |
34d52cb6 | 2392 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
4e69b598 JB |
2393 | struct btrfs_free_space *first = NULL; |
2394 | struct btrfs_free_space *entry = NULL; | |
4e69b598 JB |
2395 | struct btrfs_free_space *last; |
2396 | struct rb_node *node; | |
2397 | u64 window_start; | |
2398 | u64 window_free; | |
2399 | u64 max_extent; | |
3f7de037 | 2400 | u64 total_size = 0; |
4e69b598 | 2401 | |
34d52cb6 | 2402 | entry = tree_search_offset(ctl, offset, 0, 1); |
4e69b598 JB |
2403 | if (!entry) |
2404 | return -ENOSPC; | |
2405 | ||
2406 | /* | |
2407 | * We don't want bitmaps, so just move along until we find a normal | |
2408 | * extent entry. | |
2409 | */ | |
1bb91902 AO |
2410 | while (entry->bitmap || entry->bytes < min_bytes) { |
2411 | if (entry->bitmap && list_empty(&entry->list)) | |
86d4a77b | 2412 | list_add_tail(&entry->list, bitmaps); |
4e69b598 JB |
2413 | node = rb_next(&entry->offset_index); |
2414 | if (!node) | |
2415 | return -ENOSPC; | |
2416 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
2417 | } | |
2418 | ||
2419 | window_start = entry->offset; | |
2420 | window_free = entry->bytes; | |
2421 | max_extent = entry->bytes; | |
2422 | first = entry; | |
2423 | last = entry; | |
4e69b598 | 2424 | |
1bb91902 AO |
2425 | for (node = rb_next(&entry->offset_index); node; |
2426 | node = rb_next(&entry->offset_index)) { | |
4e69b598 JB |
2427 | entry = rb_entry(node, struct btrfs_free_space, offset_index); |
2428 | ||
86d4a77b JB |
2429 | if (entry->bitmap) { |
2430 | if (list_empty(&entry->list)) | |
2431 | list_add_tail(&entry->list, bitmaps); | |
4e69b598 | 2432 | continue; |
86d4a77b JB |
2433 | } |
2434 | ||
1bb91902 AO |
2435 | if (entry->bytes < min_bytes) |
2436 | continue; | |
2437 | ||
2438 | last = entry; | |
2439 | window_free += entry->bytes; | |
2440 | if (entry->bytes > max_extent) | |
4e69b598 | 2441 | max_extent = entry->bytes; |
4e69b598 JB |
2442 | } |
2443 | ||
1bb91902 AO |
2444 | if (window_free < bytes || max_extent < cont1_bytes) |
2445 | return -ENOSPC; | |
2446 | ||
4e69b598 JB |
2447 | cluster->window_start = first->offset; |
2448 | ||
2449 | node = &first->offset_index; | |
2450 | ||
2451 | /* | |
2452 | * now we've found our entries, pull them out of the free space | |
2453 | * cache and put them into the cluster rbtree | |
2454 | */ | |
2455 | do { | |
2456 | int ret; | |
2457 | ||
2458 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
2459 | node = rb_next(&entry->offset_index); | |
1bb91902 | 2460 | if (entry->bitmap || entry->bytes < min_bytes) |
4e69b598 JB |
2461 | continue; |
2462 | ||
34d52cb6 | 2463 | rb_erase(&entry->offset_index, &ctl->free_space_offset); |
4e69b598 JB |
2464 | ret = tree_insert_offset(&cluster->root, entry->offset, |
2465 | &entry->offset_index, 0); | |
3f7de037 | 2466 | total_size += entry->bytes; |
79787eaa | 2467 | BUG_ON(ret); /* -EEXIST; Logic error */ |
4e69b598 JB |
2468 | } while (node && entry != last); |
2469 | ||
2470 | cluster->max_size = max_extent; | |
3f7de037 | 2471 | trace_btrfs_setup_cluster(block_group, cluster, total_size, 0); |
4e69b598 JB |
2472 | return 0; |
2473 | } | |
2474 | ||
2475 | /* | |
2476 | * This specifically looks for bitmaps that may work in the cluster, we assume | |
2477 | * that we have already failed to find extents that will work. | |
2478 | */ | |
3de85bb9 JB |
2479 | static noinline int |
2480 | setup_cluster_bitmap(struct btrfs_block_group_cache *block_group, | |
2481 | struct btrfs_free_cluster *cluster, | |
2482 | struct list_head *bitmaps, u64 offset, u64 bytes, | |
1bb91902 | 2483 | u64 cont1_bytes, u64 min_bytes) |
4e69b598 | 2484 | { |
34d52cb6 | 2485 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
4e69b598 | 2486 | struct btrfs_free_space *entry; |
4e69b598 | 2487 | int ret = -ENOSPC; |
0f0fbf1d | 2488 | u64 bitmap_offset = offset_to_bitmap(ctl, offset); |
4e69b598 | 2489 | |
34d52cb6 | 2490 | if (ctl->total_bitmaps == 0) |
4e69b598 JB |
2491 | return -ENOSPC; |
2492 | ||
0f0fbf1d LZ |
2493 | /* |
2494 | * The bitmap that covers offset won't be in the list unless offset | |
2495 | * is just its start offset. | |
2496 | */ | |
2497 | entry = list_first_entry(bitmaps, struct btrfs_free_space, list); | |
2498 | if (entry->offset != bitmap_offset) { | |
2499 | entry = tree_search_offset(ctl, bitmap_offset, 1, 0); | |
2500 | if (entry && list_empty(&entry->list)) | |
2501 | list_add(&entry->list, bitmaps); | |
2502 | } | |
2503 | ||
86d4a77b | 2504 | list_for_each_entry(entry, bitmaps, list) { |
357b9784 | 2505 | if (entry->bytes < bytes) |
86d4a77b JB |
2506 | continue; |
2507 | ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset, | |
1bb91902 | 2508 | bytes, cont1_bytes, min_bytes); |
86d4a77b JB |
2509 | if (!ret) |
2510 | return 0; | |
2511 | } | |
2512 | ||
2513 | /* | |
52621cb6 LZ |
2514 | * The bitmaps list has all the bitmaps that record free space |
2515 | * starting after offset, so no more search is required. | |
86d4a77b | 2516 | */ |
52621cb6 | 2517 | return -ENOSPC; |
4e69b598 JB |
2518 | } |
2519 | ||
fa9c0d79 CM |
2520 | /* |
2521 | * here we try to find a cluster of blocks in a block group. The goal | |
1bb91902 | 2522 | * is to find at least bytes+empty_size. |
fa9c0d79 CM |
2523 | * We might not find them all in one contiguous area. |
2524 | * | |
2525 | * returns zero and sets up cluster if things worked out, otherwise | |
2526 | * it returns -enospc | |
2527 | */ | |
2528 | int btrfs_find_space_cluster(struct btrfs_trans_handle *trans, | |
451d7585 | 2529 | struct btrfs_root *root, |
fa9c0d79 CM |
2530 | struct btrfs_block_group_cache *block_group, |
2531 | struct btrfs_free_cluster *cluster, | |
2532 | u64 offset, u64 bytes, u64 empty_size) | |
2533 | { | |
34d52cb6 | 2534 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
86d4a77b | 2535 | struct btrfs_free_space *entry, *tmp; |
52621cb6 | 2536 | LIST_HEAD(bitmaps); |
fa9c0d79 | 2537 | u64 min_bytes; |
1bb91902 | 2538 | u64 cont1_bytes; |
fa9c0d79 CM |
2539 | int ret; |
2540 | ||
1bb91902 AO |
2541 | /* |
2542 | * Choose the minimum extent size we'll require for this | |
2543 | * cluster. For SSD_SPREAD, don't allow any fragmentation. | |
2544 | * For metadata, allow allocates with smaller extents. For | |
2545 | * data, keep it dense. | |
2546 | */ | |
451d7585 | 2547 | if (btrfs_test_opt(root, SSD_SPREAD)) { |
1bb91902 | 2548 | cont1_bytes = min_bytes = bytes + empty_size; |
451d7585 | 2549 | } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) { |
1bb91902 AO |
2550 | cont1_bytes = bytes; |
2551 | min_bytes = block_group->sectorsize; | |
2552 | } else { | |
2553 | cont1_bytes = max(bytes, (bytes + empty_size) >> 2); | |
2554 | min_bytes = block_group->sectorsize; | |
2555 | } | |
fa9c0d79 | 2556 | |
34d52cb6 | 2557 | spin_lock(&ctl->tree_lock); |
7d0d2e8e JB |
2558 | |
2559 | /* | |
2560 | * If we know we don't have enough space to make a cluster don't even | |
2561 | * bother doing all the work to try and find one. | |
2562 | */ | |
1bb91902 | 2563 | if (ctl->free_space < bytes) { |
34d52cb6 | 2564 | spin_unlock(&ctl->tree_lock); |
7d0d2e8e JB |
2565 | return -ENOSPC; |
2566 | } | |
2567 | ||
fa9c0d79 CM |
2568 | spin_lock(&cluster->lock); |
2569 | ||
2570 | /* someone already found a cluster, hooray */ | |
2571 | if (cluster->block_group) { | |
2572 | ret = 0; | |
2573 | goto out; | |
2574 | } | |
fa9c0d79 | 2575 | |
3f7de037 JB |
2576 | trace_btrfs_find_cluster(block_group, offset, bytes, empty_size, |
2577 | min_bytes); | |
2578 | ||
2579 | INIT_LIST_HEAD(&bitmaps); | |
86d4a77b | 2580 | ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset, |
1bb91902 AO |
2581 | bytes + empty_size, |
2582 | cont1_bytes, min_bytes); | |
4e69b598 | 2583 | if (ret) |
86d4a77b | 2584 | ret = setup_cluster_bitmap(block_group, cluster, &bitmaps, |
1bb91902 AO |
2585 | offset, bytes + empty_size, |
2586 | cont1_bytes, min_bytes); | |
86d4a77b JB |
2587 | |
2588 | /* Clear our temporary list */ | |
2589 | list_for_each_entry_safe(entry, tmp, &bitmaps, list) | |
2590 | list_del_init(&entry->list); | |
fa9c0d79 | 2591 | |
4e69b598 JB |
2592 | if (!ret) { |
2593 | atomic_inc(&block_group->count); | |
2594 | list_add_tail(&cluster->block_group_list, | |
2595 | &block_group->cluster_list); | |
2596 | cluster->block_group = block_group; | |
3f7de037 JB |
2597 | } else { |
2598 | trace_btrfs_failed_cluster_setup(block_group); | |
fa9c0d79 | 2599 | } |
fa9c0d79 CM |
2600 | out: |
2601 | spin_unlock(&cluster->lock); | |
34d52cb6 | 2602 | spin_unlock(&ctl->tree_lock); |
fa9c0d79 CM |
2603 | |
2604 | return ret; | |
2605 | } | |
2606 | ||
2607 | /* | |
2608 | * simple code to zero out a cluster | |
2609 | */ | |
2610 | void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster) | |
2611 | { | |
2612 | spin_lock_init(&cluster->lock); | |
2613 | spin_lock_init(&cluster->refill_lock); | |
6bef4d31 | 2614 | cluster->root = RB_ROOT; |
fa9c0d79 CM |
2615 | cluster->max_size = 0; |
2616 | INIT_LIST_HEAD(&cluster->block_group_list); | |
2617 | cluster->block_group = NULL; | |
2618 | } | |
2619 | ||
7fe1e641 LZ |
2620 | static int do_trimming(struct btrfs_block_group_cache *block_group, |
2621 | u64 *total_trimmed, u64 start, u64 bytes, | |
2622 | u64 reserved_start, u64 reserved_bytes) | |
f7039b1d | 2623 | { |
7fe1e641 | 2624 | struct btrfs_space_info *space_info = block_group->space_info; |
f7039b1d | 2625 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
7fe1e641 LZ |
2626 | int ret; |
2627 | int update = 0; | |
2628 | u64 trimmed = 0; | |
f7039b1d | 2629 | |
7fe1e641 LZ |
2630 | spin_lock(&space_info->lock); |
2631 | spin_lock(&block_group->lock); | |
2632 | if (!block_group->ro) { | |
2633 | block_group->reserved += reserved_bytes; | |
2634 | space_info->bytes_reserved += reserved_bytes; | |
2635 | update = 1; | |
2636 | } | |
2637 | spin_unlock(&block_group->lock); | |
2638 | spin_unlock(&space_info->lock); | |
2639 | ||
2640 | ret = btrfs_error_discard_extent(fs_info->extent_root, | |
2641 | start, bytes, &trimmed); | |
2642 | if (!ret) | |
2643 | *total_trimmed += trimmed; | |
2644 | ||
2645 | btrfs_add_free_space(block_group, reserved_start, reserved_bytes); | |
2646 | ||
2647 | if (update) { | |
2648 | spin_lock(&space_info->lock); | |
2649 | spin_lock(&block_group->lock); | |
2650 | if (block_group->ro) | |
2651 | space_info->bytes_readonly += reserved_bytes; | |
2652 | block_group->reserved -= reserved_bytes; | |
2653 | space_info->bytes_reserved -= reserved_bytes; | |
2654 | spin_unlock(&space_info->lock); | |
2655 | spin_unlock(&block_group->lock); | |
2656 | } | |
2657 | ||
2658 | return ret; | |
2659 | } | |
2660 | ||
2661 | static int trim_no_bitmap(struct btrfs_block_group_cache *block_group, | |
2662 | u64 *total_trimmed, u64 start, u64 end, u64 minlen) | |
2663 | { | |
2664 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
2665 | struct btrfs_free_space *entry; | |
2666 | struct rb_node *node; | |
2667 | int ret = 0; | |
2668 | u64 extent_start; | |
2669 | u64 extent_bytes; | |
2670 | u64 bytes; | |
f7039b1d LD |
2671 | |
2672 | while (start < end) { | |
34d52cb6 | 2673 | spin_lock(&ctl->tree_lock); |
f7039b1d | 2674 | |
34d52cb6 LZ |
2675 | if (ctl->free_space < minlen) { |
2676 | spin_unlock(&ctl->tree_lock); | |
f7039b1d LD |
2677 | break; |
2678 | } | |
2679 | ||
34d52cb6 | 2680 | entry = tree_search_offset(ctl, start, 0, 1); |
7fe1e641 | 2681 | if (!entry) { |
34d52cb6 | 2682 | spin_unlock(&ctl->tree_lock); |
f7039b1d LD |
2683 | break; |
2684 | } | |
2685 | ||
7fe1e641 LZ |
2686 | /* skip bitmaps */ |
2687 | while (entry->bitmap) { | |
2688 | node = rb_next(&entry->offset_index); | |
2689 | if (!node) { | |
34d52cb6 | 2690 | spin_unlock(&ctl->tree_lock); |
7fe1e641 | 2691 | goto out; |
f7039b1d | 2692 | } |
7fe1e641 LZ |
2693 | entry = rb_entry(node, struct btrfs_free_space, |
2694 | offset_index); | |
f7039b1d LD |
2695 | } |
2696 | ||
7fe1e641 LZ |
2697 | if (entry->offset >= end) { |
2698 | spin_unlock(&ctl->tree_lock); | |
2699 | break; | |
f7039b1d LD |
2700 | } |
2701 | ||
7fe1e641 LZ |
2702 | extent_start = entry->offset; |
2703 | extent_bytes = entry->bytes; | |
2704 | start = max(start, extent_start); | |
2705 | bytes = min(extent_start + extent_bytes, end) - start; | |
2706 | if (bytes < minlen) { | |
2707 | spin_unlock(&ctl->tree_lock); | |
2708 | goto next; | |
f7039b1d LD |
2709 | } |
2710 | ||
7fe1e641 LZ |
2711 | unlink_free_space(ctl, entry); |
2712 | kmem_cache_free(btrfs_free_space_cachep, entry); | |
2713 | ||
34d52cb6 | 2714 | spin_unlock(&ctl->tree_lock); |
f7039b1d | 2715 | |
7fe1e641 LZ |
2716 | ret = do_trimming(block_group, total_trimmed, start, bytes, |
2717 | extent_start, extent_bytes); | |
2718 | if (ret) | |
2719 | break; | |
2720 | next: | |
2721 | start += bytes; | |
f7039b1d | 2722 | |
7fe1e641 LZ |
2723 | if (fatal_signal_pending(current)) { |
2724 | ret = -ERESTARTSYS; | |
2725 | break; | |
2726 | } | |
2727 | ||
2728 | cond_resched(); | |
2729 | } | |
2730 | out: | |
2731 | return ret; | |
2732 | } | |
2733 | ||
2734 | static int trim_bitmaps(struct btrfs_block_group_cache *block_group, | |
2735 | u64 *total_trimmed, u64 start, u64 end, u64 minlen) | |
2736 | { | |
2737 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
2738 | struct btrfs_free_space *entry; | |
2739 | int ret = 0; | |
2740 | int ret2; | |
2741 | u64 bytes; | |
2742 | u64 offset = offset_to_bitmap(ctl, start); | |
2743 | ||
2744 | while (offset < end) { | |
2745 | bool next_bitmap = false; | |
2746 | ||
2747 | spin_lock(&ctl->tree_lock); | |
2748 | ||
2749 | if (ctl->free_space < minlen) { | |
2750 | spin_unlock(&ctl->tree_lock); | |
2751 | break; | |
2752 | } | |
2753 | ||
2754 | entry = tree_search_offset(ctl, offset, 1, 0); | |
2755 | if (!entry) { | |
2756 | spin_unlock(&ctl->tree_lock); | |
2757 | next_bitmap = true; | |
2758 | goto next; | |
2759 | } | |
2760 | ||
2761 | bytes = minlen; | |
2762 | ret2 = search_bitmap(ctl, entry, &start, &bytes); | |
2763 | if (ret2 || start >= end) { | |
2764 | spin_unlock(&ctl->tree_lock); | |
2765 | next_bitmap = true; | |
2766 | goto next; | |
2767 | } | |
2768 | ||
2769 | bytes = min(bytes, end - start); | |
2770 | if (bytes < minlen) { | |
2771 | spin_unlock(&ctl->tree_lock); | |
2772 | goto next; | |
2773 | } | |
2774 | ||
2775 | bitmap_clear_bits(ctl, entry, start, bytes); | |
2776 | if (entry->bytes == 0) | |
2777 | free_bitmap(ctl, entry); | |
2778 | ||
2779 | spin_unlock(&ctl->tree_lock); | |
2780 | ||
2781 | ret = do_trimming(block_group, total_trimmed, start, bytes, | |
2782 | start, bytes); | |
2783 | if (ret) | |
2784 | break; | |
2785 | next: | |
2786 | if (next_bitmap) { | |
2787 | offset += BITS_PER_BITMAP * ctl->unit; | |
2788 | } else { | |
2789 | start += bytes; | |
2790 | if (start >= offset + BITS_PER_BITMAP * ctl->unit) | |
2791 | offset += BITS_PER_BITMAP * ctl->unit; | |
f7039b1d | 2792 | } |
f7039b1d LD |
2793 | |
2794 | if (fatal_signal_pending(current)) { | |
2795 | ret = -ERESTARTSYS; | |
2796 | break; | |
2797 | } | |
2798 | ||
2799 | cond_resched(); | |
2800 | } | |
2801 | ||
2802 | return ret; | |
2803 | } | |
581bb050 | 2804 | |
7fe1e641 LZ |
2805 | int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group, |
2806 | u64 *trimmed, u64 start, u64 end, u64 minlen) | |
2807 | { | |
2808 | int ret; | |
2809 | ||
2810 | *trimmed = 0; | |
2811 | ||
2812 | ret = trim_no_bitmap(block_group, trimmed, start, end, minlen); | |
2813 | if (ret) | |
2814 | return ret; | |
2815 | ||
2816 | ret = trim_bitmaps(block_group, trimmed, start, end, minlen); | |
2817 | ||
2818 | return ret; | |
2819 | } | |
2820 | ||
581bb050 LZ |
2821 | /* |
2822 | * Find the left-most item in the cache tree, and then return the | |
2823 | * smallest inode number in the item. | |
2824 | * | |
2825 | * Note: the returned inode number may not be the smallest one in | |
2826 | * the tree, if the left-most item is a bitmap. | |
2827 | */ | |
2828 | u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root) | |
2829 | { | |
2830 | struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl; | |
2831 | struct btrfs_free_space *entry = NULL; | |
2832 | u64 ino = 0; | |
2833 | ||
2834 | spin_lock(&ctl->tree_lock); | |
2835 | ||
2836 | if (RB_EMPTY_ROOT(&ctl->free_space_offset)) | |
2837 | goto out; | |
2838 | ||
2839 | entry = rb_entry(rb_first(&ctl->free_space_offset), | |
2840 | struct btrfs_free_space, offset_index); | |
2841 | ||
2842 | if (!entry->bitmap) { | |
2843 | ino = entry->offset; | |
2844 | ||
2845 | unlink_free_space(ctl, entry); | |
2846 | entry->offset++; | |
2847 | entry->bytes--; | |
2848 | if (!entry->bytes) | |
2849 | kmem_cache_free(btrfs_free_space_cachep, entry); | |
2850 | else | |
2851 | link_free_space(ctl, entry); | |
2852 | } else { | |
2853 | u64 offset = 0; | |
2854 | u64 count = 1; | |
2855 | int ret; | |
2856 | ||
2857 | ret = search_bitmap(ctl, entry, &offset, &count); | |
79787eaa | 2858 | /* Logic error; Should be empty if it can't find anything */ |
581bb050 LZ |
2859 | BUG_ON(ret); |
2860 | ||
2861 | ino = offset; | |
2862 | bitmap_clear_bits(ctl, entry, offset, 1); | |
2863 | if (entry->bytes == 0) | |
2864 | free_bitmap(ctl, entry); | |
2865 | } | |
2866 | out: | |
2867 | spin_unlock(&ctl->tree_lock); | |
2868 | ||
2869 | return ino; | |
2870 | } | |
82d5902d LZ |
2871 | |
2872 | struct inode *lookup_free_ino_inode(struct btrfs_root *root, | |
2873 | struct btrfs_path *path) | |
2874 | { | |
2875 | struct inode *inode = NULL; | |
2876 | ||
2877 | spin_lock(&root->cache_lock); | |
2878 | if (root->cache_inode) | |
2879 | inode = igrab(root->cache_inode); | |
2880 | spin_unlock(&root->cache_lock); | |
2881 | if (inode) | |
2882 | return inode; | |
2883 | ||
2884 | inode = __lookup_free_space_inode(root, path, 0); | |
2885 | if (IS_ERR(inode)) | |
2886 | return inode; | |
2887 | ||
2888 | spin_lock(&root->cache_lock); | |
7841cb28 | 2889 | if (!btrfs_fs_closing(root->fs_info)) |
82d5902d LZ |
2890 | root->cache_inode = igrab(inode); |
2891 | spin_unlock(&root->cache_lock); | |
2892 | ||
2893 | return inode; | |
2894 | } | |
2895 | ||
2896 | int create_free_ino_inode(struct btrfs_root *root, | |
2897 | struct btrfs_trans_handle *trans, | |
2898 | struct btrfs_path *path) | |
2899 | { | |
2900 | return __create_free_space_inode(root, trans, path, | |
2901 | BTRFS_FREE_INO_OBJECTID, 0); | |
2902 | } | |
2903 | ||
2904 | int load_free_ino_cache(struct btrfs_fs_info *fs_info, struct btrfs_root *root) | |
2905 | { | |
2906 | struct btrfs_free_space_ctl *ctl = root->free_ino_ctl; | |
2907 | struct btrfs_path *path; | |
2908 | struct inode *inode; | |
2909 | int ret = 0; | |
2910 | u64 root_gen = btrfs_root_generation(&root->root_item); | |
2911 | ||
4b9465cb CM |
2912 | if (!btrfs_test_opt(root, INODE_MAP_CACHE)) |
2913 | return 0; | |
2914 | ||
82d5902d LZ |
2915 | /* |
2916 | * If we're unmounting then just return, since this does a search on the | |
2917 | * normal root and not the commit root and we could deadlock. | |
2918 | */ | |
7841cb28 | 2919 | if (btrfs_fs_closing(fs_info)) |
82d5902d LZ |
2920 | return 0; |
2921 | ||
2922 | path = btrfs_alloc_path(); | |
2923 | if (!path) | |
2924 | return 0; | |
2925 | ||
2926 | inode = lookup_free_ino_inode(root, path); | |
2927 | if (IS_ERR(inode)) | |
2928 | goto out; | |
2929 | ||
2930 | if (root_gen != BTRFS_I(inode)->generation) | |
2931 | goto out_put; | |
2932 | ||
2933 | ret = __load_free_space_cache(root, inode, ctl, path, 0); | |
2934 | ||
2935 | if (ret < 0) | |
c2cf52eb SK |
2936 | btrfs_err(fs_info, |
2937 | "failed to load free ino cache for root %llu", | |
2938 | root->root_key.objectid); | |
82d5902d LZ |
2939 | out_put: |
2940 | iput(inode); | |
2941 | out: | |
2942 | btrfs_free_path(path); | |
2943 | return ret; | |
2944 | } | |
2945 | ||
2946 | int btrfs_write_out_ino_cache(struct btrfs_root *root, | |
2947 | struct btrfs_trans_handle *trans, | |
2948 | struct btrfs_path *path) | |
2949 | { | |
2950 | struct btrfs_free_space_ctl *ctl = root->free_ino_ctl; | |
2951 | struct inode *inode; | |
2952 | int ret; | |
2953 | ||
4b9465cb CM |
2954 | if (!btrfs_test_opt(root, INODE_MAP_CACHE)) |
2955 | return 0; | |
2956 | ||
82d5902d LZ |
2957 | inode = lookup_free_ino_inode(root, path); |
2958 | if (IS_ERR(inode)) | |
2959 | return 0; | |
2960 | ||
2961 | ret = __btrfs_write_out_cache(root, inode, ctl, NULL, trans, path, 0); | |
c09544e0 JB |
2962 | if (ret) { |
2963 | btrfs_delalloc_release_metadata(inode, inode->i_size); | |
2964 | #ifdef DEBUG | |
c2cf52eb SK |
2965 | btrfs_err(root->fs_info, |
2966 | "failed to write free ino cache for root %llu", | |
2967 | root->root_key.objectid); | |
c09544e0 JB |
2968 | #endif |
2969 | } | |
82d5902d LZ |
2970 | |
2971 | iput(inode); | |
2972 | return ret; | |
2973 | } | |
74255aa0 JB |
2974 | |
2975 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS | |
2976 | static struct btrfs_block_group_cache *init_test_block_group(void) | |
2977 | { | |
2978 | struct btrfs_block_group_cache *cache; | |
2979 | ||
2980 | cache = kzalloc(sizeof(*cache), GFP_NOFS); | |
2981 | if (!cache) | |
2982 | return NULL; | |
2983 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), | |
2984 | GFP_NOFS); | |
2985 | if (!cache->free_space_ctl) { | |
2986 | kfree(cache); | |
2987 | return NULL; | |
2988 | } | |
2989 | ||
2990 | cache->key.objectid = 0; | |
2991 | cache->key.offset = 1024 * 1024 * 1024; | |
2992 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
2993 | cache->sectorsize = 4096; | |
2994 | ||
2995 | spin_lock_init(&cache->lock); | |
2996 | INIT_LIST_HEAD(&cache->list); | |
2997 | INIT_LIST_HEAD(&cache->cluster_list); | |
2998 | INIT_LIST_HEAD(&cache->new_bg_list); | |
2999 | ||
3000 | btrfs_init_free_space_ctl(cache); | |
3001 | ||
3002 | return cache; | |
3003 | } | |
3004 | ||
3005 | /* | |
3006 | * Checks to see if the given range is in the free space cache. This is really | |
3007 | * just used to check the absence of space, so if there is free space in the | |
3008 | * range at all we will return 1. | |
3009 | */ | |
3010 | static int check_exists(struct btrfs_block_group_cache *cache, u64 offset, | |
3011 | u64 bytes) | |
3012 | { | |
3013 | struct btrfs_free_space_ctl *ctl = cache->free_space_ctl; | |
3014 | struct btrfs_free_space *info; | |
3015 | int ret = 0; | |
3016 | ||
3017 | spin_lock(&ctl->tree_lock); | |
3018 | info = tree_search_offset(ctl, offset, 0, 0); | |
3019 | if (!info) { | |
3020 | info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), | |
3021 | 1, 0); | |
3022 | if (!info) | |
3023 | goto out; | |
3024 | } | |
3025 | ||
3026 | have_info: | |
3027 | if (info->bitmap) { | |
3028 | u64 bit_off, bit_bytes; | |
3029 | struct rb_node *n; | |
3030 | struct btrfs_free_space *tmp; | |
3031 | ||
3032 | bit_off = offset; | |
3033 | bit_bytes = ctl->unit; | |
3034 | ret = search_bitmap(ctl, info, &bit_off, &bit_bytes); | |
3035 | if (!ret) { | |
3036 | if (bit_off == offset) { | |
3037 | ret = 1; | |
3038 | goto out; | |
3039 | } else if (bit_off > offset && | |
3040 | offset + bytes > bit_off) { | |
3041 | ret = 1; | |
3042 | goto out; | |
3043 | } | |
3044 | } | |
3045 | ||
3046 | n = rb_prev(&info->offset_index); | |
3047 | while (n) { | |
3048 | tmp = rb_entry(n, struct btrfs_free_space, | |
3049 | offset_index); | |
3050 | if (tmp->offset + tmp->bytes < offset) | |
3051 | break; | |
3052 | if (offset + bytes < tmp->offset) { | |
3053 | n = rb_prev(&info->offset_index); | |
3054 | continue; | |
3055 | } | |
3056 | info = tmp; | |
3057 | goto have_info; | |
3058 | } | |
3059 | ||
3060 | n = rb_next(&info->offset_index); | |
3061 | while (n) { | |
3062 | tmp = rb_entry(n, struct btrfs_free_space, | |
3063 | offset_index); | |
3064 | if (offset + bytes < tmp->offset) | |
3065 | break; | |
3066 | if (tmp->offset + tmp->bytes < offset) { | |
3067 | n = rb_next(&info->offset_index); | |
3068 | continue; | |
3069 | } | |
3070 | info = tmp; | |
3071 | goto have_info; | |
3072 | } | |
3073 | ||
3074 | goto out; | |
3075 | } | |
3076 | ||
3077 | if (info->offset == offset) { | |
3078 | ret = 1; | |
3079 | goto out; | |
3080 | } | |
3081 | ||
3082 | if (offset > info->offset && offset < info->offset + info->bytes) | |
3083 | ret = 1; | |
3084 | out: | |
3085 | spin_unlock(&ctl->tree_lock); | |
3086 | return ret; | |
3087 | } | |
3088 | ||
3089 | /* | |
3090 | * Use this if you need to make a bitmap or extent entry specifically, it | |
3091 | * doesn't do any of the merging that add_free_space does, this acts a lot like | |
3092 | * how the free space cache loading stuff works, so you can get really weird | |
3093 | * configurations. | |
3094 | */ | |
3095 | static int add_free_space_entry(struct btrfs_block_group_cache *cache, | |
3096 | u64 offset, u64 bytes, bool bitmap) | |
3097 | { | |
3098 | struct btrfs_free_space_ctl *ctl = cache->free_space_ctl; | |
3099 | struct btrfs_free_space *info = NULL, *bitmap_info; | |
3100 | void *map = NULL; | |
3101 | u64 bytes_added; | |
3102 | int ret; | |
3103 | ||
3104 | again: | |
3105 | if (!info) { | |
3106 | info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS); | |
3107 | if (!info) | |
3108 | return -ENOMEM; | |
3109 | } | |
3110 | ||
3111 | if (!bitmap) { | |
3112 | spin_lock(&ctl->tree_lock); | |
3113 | info->offset = offset; | |
3114 | info->bytes = bytes; | |
3115 | ret = link_free_space(ctl, info); | |
3116 | spin_unlock(&ctl->tree_lock); | |
3117 | if (ret) | |
3118 | kmem_cache_free(btrfs_free_space_cachep, info); | |
3119 | return ret; | |
3120 | } | |
3121 | ||
3122 | if (!map) { | |
3123 | map = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS); | |
3124 | if (!map) { | |
3125 | kmem_cache_free(btrfs_free_space_cachep, info); | |
3126 | return -ENOMEM; | |
3127 | } | |
3128 | } | |
3129 | ||
3130 | spin_lock(&ctl->tree_lock); | |
3131 | bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), | |
3132 | 1, 0); | |
3133 | if (!bitmap_info) { | |
3134 | info->bitmap = map; | |
3135 | map = NULL; | |
3136 | add_new_bitmap(ctl, info, offset); | |
3137 | bitmap_info = info; | |
3138 | } | |
3139 | ||
3140 | bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes); | |
3141 | bytes -= bytes_added; | |
3142 | offset += bytes_added; | |
3143 | spin_unlock(&ctl->tree_lock); | |
3144 | ||
3145 | if (bytes) | |
3146 | goto again; | |
3147 | ||
3148 | if (map) | |
3149 | kfree(map); | |
3150 | return 0; | |
3151 | } | |
3152 | ||
905d0f56 DS |
3153 | #define test_msg(fmt, ...) printk(KERN_INFO "btrfs: selftest: " fmt, ##__VA_ARGS__) |
3154 | ||
74255aa0 JB |
3155 | /* |
3156 | * This test just does basic sanity checking, making sure we can add an exten | |
3157 | * entry and remove space from either end and the middle, and make sure we can | |
3158 | * remove space that covers adjacent extent entries. | |
3159 | */ | |
3160 | static int test_extents(struct btrfs_block_group_cache *cache) | |
3161 | { | |
3162 | int ret = 0; | |
3163 | ||
905d0f56 | 3164 | test_msg("Running extent only tests\n"); |
74255aa0 JB |
3165 | |
3166 | /* First just make sure we can remove an entire entry */ | |
3167 | ret = btrfs_add_free_space(cache, 0, 4 * 1024 * 1024); | |
3168 | if (ret) { | |
905d0f56 | 3169 | test_msg("Error adding initial extents %d\n", ret); |
74255aa0 JB |
3170 | return ret; |
3171 | } | |
3172 | ||
3173 | ret = btrfs_remove_free_space(cache, 0, 4 * 1024 * 1024); | |
3174 | if (ret) { | |
905d0f56 | 3175 | test_msg("Error removing extent %d\n", ret); |
74255aa0 JB |
3176 | return ret; |
3177 | } | |
3178 | ||
3179 | if (check_exists(cache, 0, 4 * 1024 * 1024)) { | |
905d0f56 | 3180 | test_msg("Full remove left some lingering space\n"); |
74255aa0 JB |
3181 | return -1; |
3182 | } | |
3183 | ||
3184 | /* Ok edge and middle cases now */ | |
3185 | ret = btrfs_add_free_space(cache, 0, 4 * 1024 * 1024); | |
3186 | if (ret) { | |
905d0f56 | 3187 | test_msg("Error adding half extent %d\n", ret); |
74255aa0 JB |
3188 | return ret; |
3189 | } | |
3190 | ||
3191 | ret = btrfs_remove_free_space(cache, 3 * 1024 * 1024, 1 * 1024 * 1024); | |
3192 | if (ret) { | |
905d0f56 | 3193 | test_msg("Error removing tail end %d\n", ret); |
74255aa0 JB |
3194 | return ret; |
3195 | } | |
3196 | ||
3197 | ret = btrfs_remove_free_space(cache, 0, 1 * 1024 * 1024); | |
3198 | if (ret) { | |
905d0f56 | 3199 | test_msg("Error removing front end %d\n", ret); |
74255aa0 JB |
3200 | return ret; |
3201 | } | |
3202 | ||
3203 | ret = btrfs_remove_free_space(cache, 2 * 1024 * 1024, 4096); | |
3204 | if (ret) { | |
e3a0dd98 | 3205 | test_msg("Error removing middle piece %d\n", ret); |
74255aa0 JB |
3206 | return ret; |
3207 | } | |
3208 | ||
3209 | if (check_exists(cache, 0, 1 * 1024 * 1024)) { | |
905d0f56 | 3210 | test_msg("Still have space at the front\n"); |
74255aa0 JB |
3211 | return -1; |
3212 | } | |
3213 | ||
3214 | if (check_exists(cache, 2 * 1024 * 1024, 4096)) { | |
905d0f56 | 3215 | test_msg("Still have space in the middle\n"); |
74255aa0 JB |
3216 | return -1; |
3217 | } | |
3218 | ||
3219 | if (check_exists(cache, 3 * 1024 * 1024, 1 * 1024 * 1024)) { | |
905d0f56 | 3220 | test_msg("Still have space at the end\n"); |
74255aa0 JB |
3221 | return -1; |
3222 | } | |
3223 | ||
3224 | /* Cleanup */ | |
3225 | __btrfs_remove_free_space_cache(cache->free_space_ctl); | |
3226 | ||
3227 | return 0; | |
3228 | } | |
3229 | ||
3230 | static int test_bitmaps(struct btrfs_block_group_cache *cache) | |
3231 | { | |
3232 | u64 next_bitmap_offset; | |
3233 | int ret; | |
3234 | ||
905d0f56 | 3235 | test_msg("Running bitmap only tests\n"); |
74255aa0 JB |
3236 | |
3237 | ret = add_free_space_entry(cache, 0, 4 * 1024 * 1024, 1); | |
3238 | if (ret) { | |
905d0f56 | 3239 | test_msg("Couldn't create a bitmap entry %d\n", ret); |
74255aa0 JB |
3240 | return ret; |
3241 | } | |
3242 | ||
3243 | ret = btrfs_remove_free_space(cache, 0, 4 * 1024 * 1024); | |
3244 | if (ret) { | |
905d0f56 | 3245 | test_msg("Error removing bitmap full range %d\n", ret); |
74255aa0 JB |
3246 | return ret; |
3247 | } | |
3248 | ||
3249 | if (check_exists(cache, 0, 4 * 1024 * 1024)) { | |
905d0f56 | 3250 | test_msg("Left some space in bitmap\n"); |
74255aa0 JB |
3251 | return -1; |
3252 | } | |
3253 | ||
3254 | ret = add_free_space_entry(cache, 0, 4 * 1024 * 1024, 1); | |
3255 | if (ret) { | |
905d0f56 | 3256 | test_msg("Couldn't add to our bitmap entry %d\n", ret); |
74255aa0 JB |
3257 | return ret; |
3258 | } | |
3259 | ||
3260 | ret = btrfs_remove_free_space(cache, 1 * 1024 * 1024, 2 * 1024 * 1024); | |
3261 | if (ret) { | |
905d0f56 | 3262 | test_msg("Couldn't remove middle chunk %d\n", ret); |
74255aa0 JB |
3263 | return ret; |
3264 | } | |
3265 | ||
3266 | /* | |
3267 | * The first bitmap we have starts at offset 0 so the next one is just | |
3268 | * at the end of the first bitmap. | |
3269 | */ | |
3270 | next_bitmap_offset = (u64)(BITS_PER_BITMAP * 4096); | |
3271 | ||
3272 | /* Test a bit straddling two bitmaps */ | |
3273 | ret = add_free_space_entry(cache, next_bitmap_offset - | |
3274 | (2 * 1024 * 1024), 4 * 1024 * 1024, 1); | |
3275 | if (ret) { | |
905d0f56 DS |
3276 | test_msg("Couldn't add space that straddles two bitmaps %d\n", |
3277 | ret); | |
74255aa0 JB |
3278 | return ret; |
3279 | } | |
3280 | ||
3281 | ret = btrfs_remove_free_space(cache, next_bitmap_offset - | |
3282 | (1 * 1024 * 1024), 2 * 1024 * 1024); | |
3283 | if (ret) { | |
905d0f56 | 3284 | test_msg("Couldn't remove overlapping space %d\n", ret); |
74255aa0 JB |
3285 | return ret; |
3286 | } | |
3287 | ||
3288 | if (check_exists(cache, next_bitmap_offset - (1 * 1024 * 1024), | |
3289 | 2 * 1024 * 1024)) { | |
905d0f56 | 3290 | test_msg("Left some space when removing overlapping\n"); |
74255aa0 JB |
3291 | return -1; |
3292 | } | |
3293 | ||
3294 | __btrfs_remove_free_space_cache(cache->free_space_ctl); | |
3295 | ||
3296 | return 0; | |
3297 | } | |
3298 | ||
3299 | /* This is the high grade jackassery */ | |
3300 | static int test_bitmaps_and_extents(struct btrfs_block_group_cache *cache) | |
3301 | { | |
3302 | u64 bitmap_offset = (u64)(BITS_PER_BITMAP * 4096); | |
3303 | int ret; | |
3304 | ||
905d0f56 | 3305 | test_msg("Running bitmap and extent tests\n"); |
74255aa0 JB |
3306 | |
3307 | /* | |
3308 | * First let's do something simple, an extent at the same offset as the | |
3309 | * bitmap, but the free space completely in the extent and then | |
3310 | * completely in the bitmap. | |
3311 | */ | |
3312 | ret = add_free_space_entry(cache, 4 * 1024 * 1024, 1 * 1024 * 1024, 1); | |
3313 | if (ret) { | |
905d0f56 | 3314 | test_msg("Couldn't create bitmap entry %d\n", ret); |
74255aa0 JB |
3315 | return ret; |
3316 | } | |
3317 | ||
3318 | ret = add_free_space_entry(cache, 0, 1 * 1024 * 1024, 0); | |
3319 | if (ret) { | |
905d0f56 | 3320 | test_msg("Couldn't add extent entry %d\n", ret); |
74255aa0 JB |
3321 | return ret; |
3322 | } | |
3323 | ||
3324 | ret = btrfs_remove_free_space(cache, 0, 1 * 1024 * 1024); | |
3325 | if (ret) { | |
905d0f56 | 3326 | test_msg("Couldn't remove extent entry %d\n", ret); |
74255aa0 JB |
3327 | return ret; |
3328 | } | |
3329 | ||
3330 | if (check_exists(cache, 0, 1 * 1024 * 1024)) { | |
905d0f56 | 3331 | test_msg("Left remnants after our remove\n"); |
74255aa0 JB |
3332 | return -1; |
3333 | } | |
3334 | ||
3335 | /* Now to add back the extent entry and remove from the bitmap */ | |
3336 | ret = add_free_space_entry(cache, 0, 1 * 1024 * 1024, 0); | |
3337 | if (ret) { | |
905d0f56 | 3338 | test_msg("Couldn't re-add extent entry %d\n", ret); |
74255aa0 JB |
3339 | return ret; |
3340 | } | |
3341 | ||
3342 | ret = btrfs_remove_free_space(cache, 4 * 1024 * 1024, 1 * 1024 * 1024); | |
3343 | if (ret) { | |
905d0f56 | 3344 | test_msg("Couldn't remove from bitmap %d\n", ret); |
74255aa0 JB |
3345 | return ret; |
3346 | } | |
3347 | ||
3348 | if (check_exists(cache, 4 * 1024 * 1024, 1 * 1024 * 1024)) { | |
905d0f56 | 3349 | test_msg("Left remnants in the bitmap\n"); |
74255aa0 JB |
3350 | return -1; |
3351 | } | |
3352 | ||
3353 | /* | |
3354 | * Ok so a little more evil, extent entry and bitmap at the same offset, | |
3355 | * removing an overlapping chunk. | |
3356 | */ | |
3357 | ret = add_free_space_entry(cache, 1 * 1024 * 1024, 4 * 1024 * 1024, 1); | |
3358 | if (ret) { | |
905d0f56 | 3359 | test_msg("Couldn't add to a bitmap %d\n", ret); |
74255aa0 JB |
3360 | return ret; |
3361 | } | |
3362 | ||
3363 | ret = btrfs_remove_free_space(cache, 512 * 1024, 3 * 1024 * 1024); | |
3364 | if (ret) { | |
905d0f56 | 3365 | test_msg("Couldn't remove overlapping space %d\n", ret); |
74255aa0 JB |
3366 | return ret; |
3367 | } | |
3368 | ||
3369 | if (check_exists(cache, 512 * 1024, 3 * 1024 * 1024)) { | |
905d0f56 | 3370 | test_msg("Left over peices after removing overlapping\n"); |
74255aa0 JB |
3371 | return -1; |
3372 | } | |
3373 | ||
3374 | __btrfs_remove_free_space_cache(cache->free_space_ctl); | |
3375 | ||
3376 | /* Now with the extent entry offset into the bitmap */ | |
3377 | ret = add_free_space_entry(cache, 4 * 1024 * 1024, 4 * 1024 * 1024, 1); | |
3378 | if (ret) { | |
905d0f56 | 3379 | test_msg("Couldn't add space to the bitmap %d\n", ret); |
74255aa0 JB |
3380 | return ret; |
3381 | } | |
3382 | ||
3383 | ret = add_free_space_entry(cache, 2 * 1024 * 1024, 2 * 1024 * 1024, 0); | |
3384 | if (ret) { | |
905d0f56 | 3385 | test_msg("Couldn't add extent to the cache %d\n", ret); |
74255aa0 JB |
3386 | return ret; |
3387 | } | |
3388 | ||
3389 | ret = btrfs_remove_free_space(cache, 3 * 1024 * 1024, 4 * 1024 * 1024); | |
3390 | if (ret) { | |
905d0f56 | 3391 | test_msg("Problem removing overlapping space %d\n", ret); |
74255aa0 JB |
3392 | return ret; |
3393 | } | |
3394 | ||
3395 | if (check_exists(cache, 3 * 1024 * 1024, 4 * 1024 * 1024)) { | |
905d0f56 | 3396 | test_msg("Left something behind when removing space"); |
74255aa0 JB |
3397 | return -1; |
3398 | } | |
3399 | ||
3400 | /* | |
3401 | * This has blown up in the past, the extent entry starts before the | |
3402 | * bitmap entry, but we're trying to remove an offset that falls | |
3403 | * completely within the bitmap range and is in both the extent entry | |
3404 | * and the bitmap entry, looks like this | |
3405 | * | |
3406 | * [ extent ] | |
3407 | * [ bitmap ] | |
3408 | * [ del ] | |
3409 | */ | |
3410 | __btrfs_remove_free_space_cache(cache->free_space_ctl); | |
3411 | ret = add_free_space_entry(cache, bitmap_offset + 4 * 1024 * 1024, | |
3412 | 4 * 1024 * 1024, 1); | |
3413 | if (ret) { | |
905d0f56 | 3414 | test_msg("Couldn't add bitmap %d\n", ret); |
74255aa0 JB |
3415 | return ret; |
3416 | } | |
3417 | ||
3418 | ret = add_free_space_entry(cache, bitmap_offset - 1 * 1024 * 1024, | |
3419 | 5 * 1024 * 1024, 0); | |
3420 | if (ret) { | |
905d0f56 | 3421 | test_msg("Couldn't add extent entry %d\n", ret); |
74255aa0 JB |
3422 | return ret; |
3423 | } | |
3424 | ||
3425 | ret = btrfs_remove_free_space(cache, bitmap_offset + 1 * 1024 * 1024, | |
3426 | 5 * 1024 * 1024); | |
3427 | if (ret) { | |
905d0f56 | 3428 | test_msg("Failed to free our space %d\n", ret); |
74255aa0 JB |
3429 | return ret; |
3430 | } | |
3431 | ||
3432 | if (check_exists(cache, bitmap_offset + 1 * 1024 * 1024, | |
3433 | 5 * 1024 * 1024)) { | |
905d0f56 | 3434 | test_msg("Left stuff over\n"); |
74255aa0 JB |
3435 | return -1; |
3436 | } | |
3437 | ||
3438 | __btrfs_remove_free_space_cache(cache->free_space_ctl); | |
3439 | ||
3440 | /* | |
3441 | * This blew up before, we have part of the free space in a bitmap and | |
3442 | * then the entirety of the rest of the space in an extent. This used | |
3443 | * to return -EAGAIN back from btrfs_remove_extent, make sure this | |
3444 | * doesn't happen. | |
3445 | */ | |
3446 | ret = add_free_space_entry(cache, 1 * 1024 * 1024, 2 * 1024 * 1024, 1); | |
3447 | if (ret) { | |
905d0f56 | 3448 | test_msg("Couldn't add bitmap entry %d\n", ret); |
74255aa0 JB |
3449 | return ret; |
3450 | } | |
3451 | ||
3452 | ret = add_free_space_entry(cache, 3 * 1024 * 1024, 1 * 1024 * 1024, 0); | |
3453 | if (ret) { | |
905d0f56 | 3454 | test_msg("Couldn't add extent entry %d\n", ret); |
74255aa0 JB |
3455 | return ret; |
3456 | } | |
3457 | ||
3458 | ret = btrfs_remove_free_space(cache, 1 * 1024 * 1024, 3 * 1024 * 1024); | |
3459 | if (ret) { | |
905d0f56 | 3460 | test_msg("Error removing bitmap and extent overlapping %d\n", ret); |
74255aa0 JB |
3461 | return ret; |
3462 | } | |
3463 | ||
3464 | __btrfs_remove_free_space_cache(cache->free_space_ctl); | |
3465 | return 0; | |
3466 | } | |
3467 | ||
3468 | void btrfs_test_free_space_cache(void) | |
3469 | { | |
3470 | struct btrfs_block_group_cache *cache; | |
3471 | ||
905d0f56 | 3472 | test_msg("Running btrfs free space cache tests\n"); |
74255aa0 JB |
3473 | |
3474 | cache = init_test_block_group(); | |
3475 | if (!cache) { | |
905d0f56 | 3476 | test_msg("Couldn't run the tests\n"); |
74255aa0 JB |
3477 | return; |
3478 | } | |
3479 | ||
3480 | if (test_extents(cache)) | |
3481 | goto out; | |
3482 | if (test_bitmaps(cache)) | |
3483 | goto out; | |
3484 | if (test_bitmaps_and_extents(cache)) | |
3485 | goto out; | |
3486 | out: | |
3487 | __btrfs_remove_free_space_cache(cache->free_space_ctl); | |
3488 | kfree(cache->free_space_ctl); | |
3489 | kfree(cache); | |
905d0f56 | 3490 | test_msg("Free space cache tests finished\n"); |
74255aa0 | 3491 | } |
905d0f56 | 3492 | #undef test_msg |
e6d29605 DS |
3493 | #else /* !CONFIG_BTRFS_FS_RUN_SANITY_TESTS */ |
3494 | void btrfs_test_free_space_cache(void) {} | |
3495 | #endif /* !CONFIG_BTRFS_FS_RUN_SANITY_TESTS */ |