]>
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 JB |
35 | struct btrfs_free_space *info); |
36 | ||
0414efae LZ |
37 | static struct inode *__lookup_free_space_inode(struct btrfs_root *root, |
38 | struct btrfs_path *path, | |
39 | u64 offset) | |
0af3d00b JB |
40 | { |
41 | struct btrfs_key key; | |
42 | struct btrfs_key location; | |
43 | struct btrfs_disk_key disk_key; | |
44 | struct btrfs_free_space_header *header; | |
45 | struct extent_buffer *leaf; | |
46 | struct inode *inode = NULL; | |
47 | int ret; | |
48 | ||
0af3d00b | 49 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; |
0414efae | 50 | key.offset = offset; |
0af3d00b JB |
51 | key.type = 0; |
52 | ||
53 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
54 | if (ret < 0) | |
55 | return ERR_PTR(ret); | |
56 | if (ret > 0) { | |
b3b4aa74 | 57 | btrfs_release_path(path); |
0af3d00b JB |
58 | return ERR_PTR(-ENOENT); |
59 | } | |
60 | ||
61 | leaf = path->nodes[0]; | |
62 | header = btrfs_item_ptr(leaf, path->slots[0], | |
63 | struct btrfs_free_space_header); | |
64 | btrfs_free_space_key(leaf, header, &disk_key); | |
65 | btrfs_disk_key_to_cpu(&location, &disk_key); | |
b3b4aa74 | 66 | btrfs_release_path(path); |
0af3d00b JB |
67 | |
68 | inode = btrfs_iget(root->fs_info->sb, &location, root, NULL); | |
69 | if (!inode) | |
70 | return ERR_PTR(-ENOENT); | |
71 | if (IS_ERR(inode)) | |
72 | return inode; | |
73 | if (is_bad_inode(inode)) { | |
74 | iput(inode); | |
75 | return ERR_PTR(-ENOENT); | |
76 | } | |
77 | ||
adae52b9 MX |
78 | inode->i_mapping->flags &= ~__GFP_FS; |
79 | ||
0414efae LZ |
80 | return inode; |
81 | } | |
82 | ||
83 | struct inode *lookup_free_space_inode(struct btrfs_root *root, | |
84 | struct btrfs_block_group_cache | |
85 | *block_group, struct btrfs_path *path) | |
86 | { | |
87 | struct inode *inode = NULL; | |
5b0e95bf | 88 | u32 flags = BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW; |
0414efae LZ |
89 | |
90 | spin_lock(&block_group->lock); | |
91 | if (block_group->inode) | |
92 | inode = igrab(block_group->inode); | |
93 | spin_unlock(&block_group->lock); | |
94 | if (inode) | |
95 | return inode; | |
96 | ||
97 | inode = __lookup_free_space_inode(root, path, | |
98 | block_group->key.objectid); | |
99 | if (IS_ERR(inode)) | |
100 | return inode; | |
101 | ||
0af3d00b | 102 | spin_lock(&block_group->lock); |
5b0e95bf | 103 | if (!((BTRFS_I(inode)->flags & flags) == flags)) { |
2f356126 | 104 | printk(KERN_INFO "Old style space inode found, converting.\n"); |
5b0e95bf JB |
105 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM | |
106 | BTRFS_INODE_NODATACOW; | |
2f356126 JB |
107 | block_group->disk_cache_state = BTRFS_DC_CLEAR; |
108 | } | |
109 | ||
300e4f8a | 110 | if (!block_group->iref) { |
0af3d00b JB |
111 | block_group->inode = igrab(inode); |
112 | block_group->iref = 1; | |
113 | } | |
114 | spin_unlock(&block_group->lock); | |
115 | ||
116 | return inode; | |
117 | } | |
118 | ||
0414efae LZ |
119 | int __create_free_space_inode(struct btrfs_root *root, |
120 | struct btrfs_trans_handle *trans, | |
121 | struct btrfs_path *path, u64 ino, u64 offset) | |
0af3d00b JB |
122 | { |
123 | struct btrfs_key key; | |
124 | struct btrfs_disk_key disk_key; | |
125 | struct btrfs_free_space_header *header; | |
126 | struct btrfs_inode_item *inode_item; | |
127 | struct extent_buffer *leaf; | |
5b0e95bf | 128 | u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC; |
0af3d00b JB |
129 | int ret; |
130 | ||
0414efae | 131 | ret = btrfs_insert_empty_inode(trans, root, path, ino); |
0af3d00b JB |
132 | if (ret) |
133 | return ret; | |
134 | ||
5b0e95bf JB |
135 | /* We inline crc's for the free disk space cache */ |
136 | if (ino != BTRFS_FREE_INO_OBJECTID) | |
137 | flags |= BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW; | |
138 | ||
0af3d00b JB |
139 | leaf = path->nodes[0]; |
140 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
141 | struct btrfs_inode_item); | |
142 | btrfs_item_key(leaf, &disk_key, path->slots[0]); | |
143 | memset_extent_buffer(leaf, 0, (unsigned long)inode_item, | |
144 | sizeof(*inode_item)); | |
145 | btrfs_set_inode_generation(leaf, inode_item, trans->transid); | |
146 | btrfs_set_inode_size(leaf, inode_item, 0); | |
147 | btrfs_set_inode_nbytes(leaf, inode_item, 0); | |
148 | btrfs_set_inode_uid(leaf, inode_item, 0); | |
149 | btrfs_set_inode_gid(leaf, inode_item, 0); | |
150 | btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600); | |
5b0e95bf | 151 | btrfs_set_inode_flags(leaf, inode_item, flags); |
0af3d00b JB |
152 | btrfs_set_inode_nlink(leaf, inode_item, 1); |
153 | btrfs_set_inode_transid(leaf, inode_item, trans->transid); | |
0414efae | 154 | btrfs_set_inode_block_group(leaf, inode_item, offset); |
0af3d00b | 155 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 156 | btrfs_release_path(path); |
0af3d00b JB |
157 | |
158 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
0414efae | 159 | key.offset = offset; |
0af3d00b JB |
160 | key.type = 0; |
161 | ||
162 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
163 | sizeof(struct btrfs_free_space_header)); | |
164 | if (ret < 0) { | |
b3b4aa74 | 165 | btrfs_release_path(path); |
0af3d00b JB |
166 | return ret; |
167 | } | |
168 | leaf = path->nodes[0]; | |
169 | header = btrfs_item_ptr(leaf, path->slots[0], | |
170 | struct btrfs_free_space_header); | |
171 | memset_extent_buffer(leaf, 0, (unsigned long)header, sizeof(*header)); | |
172 | btrfs_set_free_space_key(leaf, header, &disk_key); | |
173 | btrfs_mark_buffer_dirty(leaf); | |
b3b4aa74 | 174 | btrfs_release_path(path); |
0af3d00b JB |
175 | |
176 | return 0; | |
177 | } | |
178 | ||
0414efae LZ |
179 | int create_free_space_inode(struct btrfs_root *root, |
180 | struct btrfs_trans_handle *trans, | |
181 | struct btrfs_block_group_cache *block_group, | |
182 | struct btrfs_path *path) | |
183 | { | |
184 | int ret; | |
185 | u64 ino; | |
186 | ||
187 | ret = btrfs_find_free_objectid(root, &ino); | |
188 | if (ret < 0) | |
189 | return ret; | |
190 | ||
191 | return __create_free_space_inode(root, trans, path, ino, | |
192 | block_group->key.objectid); | |
193 | } | |
194 | ||
0af3d00b JB |
195 | int btrfs_truncate_free_space_cache(struct btrfs_root *root, |
196 | struct btrfs_trans_handle *trans, | |
197 | struct btrfs_path *path, | |
198 | struct inode *inode) | |
199 | { | |
65450aa6 | 200 | struct btrfs_block_rsv *rsv; |
c8174313 | 201 | u64 needed_bytes; |
0af3d00b JB |
202 | loff_t oldsize; |
203 | int ret = 0; | |
204 | ||
65450aa6 | 205 | rsv = trans->block_rsv; |
c8174313 JB |
206 | trans->block_rsv = &root->fs_info->global_block_rsv; |
207 | ||
208 | /* 1 for slack space, 1 for updating the inode */ | |
209 | needed_bytes = btrfs_calc_trunc_metadata_size(root, 1) + | |
210 | btrfs_calc_trans_metadata_size(root, 1); | |
211 | ||
212 | spin_lock(&trans->block_rsv->lock); | |
213 | if (trans->block_rsv->reserved < needed_bytes) { | |
214 | spin_unlock(&trans->block_rsv->lock); | |
215 | trans->block_rsv = rsv; | |
216 | return -ENOSPC; | |
217 | } | |
218 | spin_unlock(&trans->block_rsv->lock); | |
0af3d00b JB |
219 | |
220 | oldsize = i_size_read(inode); | |
221 | btrfs_i_size_write(inode, 0); | |
222 | truncate_pagecache(inode, oldsize, 0); | |
223 | ||
224 | /* | |
225 | * We don't need an orphan item because truncating the free space cache | |
226 | * will never be split across transactions. | |
227 | */ | |
228 | ret = btrfs_truncate_inode_items(trans, root, inode, | |
229 | 0, BTRFS_EXTENT_DATA_KEY); | |
65450aa6 | 230 | |
0af3d00b | 231 | if (ret) { |
c8174313 | 232 | trans->block_rsv = rsv; |
0af3d00b JB |
233 | WARN_ON(1); |
234 | return ret; | |
235 | } | |
236 | ||
82d5902d | 237 | ret = btrfs_update_inode(trans, root, inode); |
c8174313 JB |
238 | trans->block_rsv = rsv; |
239 | ||
82d5902d | 240 | return ret; |
0af3d00b JB |
241 | } |
242 | ||
9d66e233 JB |
243 | static int readahead_cache(struct inode *inode) |
244 | { | |
245 | struct file_ra_state *ra; | |
246 | unsigned long last_index; | |
247 | ||
248 | ra = kzalloc(sizeof(*ra), GFP_NOFS); | |
249 | if (!ra) | |
250 | return -ENOMEM; | |
251 | ||
252 | file_ra_state_init(ra, inode->i_mapping); | |
253 | last_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT; | |
254 | ||
255 | page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index); | |
256 | ||
257 | kfree(ra); | |
258 | ||
259 | return 0; | |
260 | } | |
261 | ||
a67509c3 JB |
262 | struct io_ctl { |
263 | void *cur, *orig; | |
264 | struct page *page; | |
265 | struct page **pages; | |
266 | struct btrfs_root *root; | |
267 | unsigned long size; | |
268 | int index; | |
269 | int num_pages; | |
5b0e95bf | 270 | unsigned check_crcs:1; |
a67509c3 JB |
271 | }; |
272 | ||
273 | static int io_ctl_init(struct io_ctl *io_ctl, struct inode *inode, | |
274 | struct btrfs_root *root) | |
275 | { | |
276 | memset(io_ctl, 0, sizeof(struct io_ctl)); | |
277 | io_ctl->num_pages = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> | |
278 | PAGE_CACHE_SHIFT; | |
279 | io_ctl->pages = kzalloc(sizeof(struct page *) * io_ctl->num_pages, | |
280 | GFP_NOFS); | |
281 | if (!io_ctl->pages) | |
282 | return -ENOMEM; | |
283 | io_ctl->root = root; | |
5b0e95bf JB |
284 | if (btrfs_ino(inode) != BTRFS_FREE_INO_OBJECTID) |
285 | io_ctl->check_crcs = 1; | |
a67509c3 JB |
286 | return 0; |
287 | } | |
288 | ||
289 | static void io_ctl_free(struct io_ctl *io_ctl) | |
290 | { | |
291 | kfree(io_ctl->pages); | |
292 | } | |
293 | ||
294 | static void io_ctl_unmap_page(struct io_ctl *io_ctl) | |
295 | { | |
296 | if (io_ctl->cur) { | |
297 | kunmap(io_ctl->page); | |
298 | io_ctl->cur = NULL; | |
299 | io_ctl->orig = NULL; | |
300 | } | |
301 | } | |
302 | ||
303 | static void io_ctl_map_page(struct io_ctl *io_ctl, int clear) | |
304 | { | |
305 | WARN_ON(io_ctl->cur); | |
306 | BUG_ON(io_ctl->index >= io_ctl->num_pages); | |
307 | io_ctl->page = io_ctl->pages[io_ctl->index++]; | |
308 | io_ctl->cur = kmap(io_ctl->page); | |
309 | io_ctl->orig = io_ctl->cur; | |
310 | io_ctl->size = PAGE_CACHE_SIZE; | |
311 | if (clear) | |
312 | memset(io_ctl->cur, 0, PAGE_CACHE_SIZE); | |
313 | } | |
314 | ||
315 | static void io_ctl_drop_pages(struct io_ctl *io_ctl) | |
316 | { | |
317 | int i; | |
318 | ||
319 | io_ctl_unmap_page(io_ctl); | |
320 | ||
321 | for (i = 0; i < io_ctl->num_pages; i++) { | |
322 | ClearPageChecked(io_ctl->pages[i]); | |
323 | unlock_page(io_ctl->pages[i]); | |
324 | page_cache_release(io_ctl->pages[i]); | |
325 | } | |
326 | } | |
327 | ||
328 | static int io_ctl_prepare_pages(struct io_ctl *io_ctl, struct inode *inode, | |
329 | int uptodate) | |
330 | { | |
331 | struct page *page; | |
332 | gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); | |
333 | int i; | |
334 | ||
335 | for (i = 0; i < io_ctl->num_pages; i++) { | |
336 | page = find_or_create_page(inode->i_mapping, i, mask); | |
337 | if (!page) { | |
338 | io_ctl_drop_pages(io_ctl); | |
339 | return -ENOMEM; | |
340 | } | |
341 | io_ctl->pages[i] = page; | |
342 | if (uptodate && !PageUptodate(page)) { | |
343 | btrfs_readpage(NULL, page); | |
344 | lock_page(page); | |
345 | if (!PageUptodate(page)) { | |
346 | printk(KERN_ERR "btrfs: error reading free " | |
347 | "space cache\n"); | |
348 | io_ctl_drop_pages(io_ctl); | |
349 | return -EIO; | |
350 | } | |
351 | } | |
352 | } | |
353 | ||
f7d61dcd JB |
354 | for (i = 0; i < io_ctl->num_pages; i++) { |
355 | clear_page_dirty_for_io(io_ctl->pages[i]); | |
356 | set_page_extent_mapped(io_ctl->pages[i]); | |
357 | } | |
358 | ||
a67509c3 JB |
359 | return 0; |
360 | } | |
361 | ||
362 | static void io_ctl_set_generation(struct io_ctl *io_ctl, u64 generation) | |
363 | { | |
364 | u64 *val; | |
365 | ||
366 | io_ctl_map_page(io_ctl, 1); | |
367 | ||
368 | /* | |
5b0e95bf JB |
369 | * Skip the csum areas. If we don't check crcs then we just have a |
370 | * 64bit chunk at the front of the first page. | |
a67509c3 | 371 | */ |
5b0e95bf JB |
372 | if (io_ctl->check_crcs) { |
373 | io_ctl->cur += (sizeof(u32) * io_ctl->num_pages); | |
374 | io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages); | |
375 | } else { | |
376 | io_ctl->cur += sizeof(u64); | |
377 | io_ctl->size -= sizeof(u64) * 2; | |
378 | } | |
a67509c3 JB |
379 | |
380 | val = io_ctl->cur; | |
381 | *val = cpu_to_le64(generation); | |
382 | io_ctl->cur += sizeof(u64); | |
a67509c3 JB |
383 | } |
384 | ||
385 | static int io_ctl_check_generation(struct io_ctl *io_ctl, u64 generation) | |
386 | { | |
387 | u64 *gen; | |
388 | ||
5b0e95bf JB |
389 | /* |
390 | * Skip the crc area. If we don't check crcs then we just have a 64bit | |
391 | * chunk at the front of the first page. | |
392 | */ | |
393 | if (io_ctl->check_crcs) { | |
394 | io_ctl->cur += sizeof(u32) * io_ctl->num_pages; | |
395 | io_ctl->size -= sizeof(u64) + | |
396 | (sizeof(u32) * io_ctl->num_pages); | |
397 | } else { | |
398 | io_ctl->cur += sizeof(u64); | |
399 | io_ctl->size -= sizeof(u64) * 2; | |
400 | } | |
a67509c3 | 401 | |
a67509c3 JB |
402 | gen = io_ctl->cur; |
403 | if (le64_to_cpu(*gen) != generation) { | |
404 | printk_ratelimited(KERN_ERR "btrfs: space cache generation " | |
405 | "(%Lu) does not match inode (%Lu)\n", *gen, | |
406 | generation); | |
407 | io_ctl_unmap_page(io_ctl); | |
408 | return -EIO; | |
409 | } | |
410 | io_ctl->cur += sizeof(u64); | |
5b0e95bf JB |
411 | return 0; |
412 | } | |
413 | ||
414 | static void io_ctl_set_crc(struct io_ctl *io_ctl, int index) | |
415 | { | |
416 | u32 *tmp; | |
417 | u32 crc = ~(u32)0; | |
418 | unsigned offset = 0; | |
419 | ||
420 | if (!io_ctl->check_crcs) { | |
421 | io_ctl_unmap_page(io_ctl); | |
422 | return; | |
423 | } | |
424 | ||
425 | if (index == 0) | |
426 | offset = sizeof(u32) * io_ctl->num_pages;; | |
427 | ||
428 | crc = btrfs_csum_data(io_ctl->root, io_ctl->orig + offset, crc, | |
429 | PAGE_CACHE_SIZE - offset); | |
430 | btrfs_csum_final(crc, (char *)&crc); | |
431 | io_ctl_unmap_page(io_ctl); | |
432 | tmp = kmap(io_ctl->pages[0]); | |
433 | tmp += index; | |
434 | *tmp = crc; | |
435 | kunmap(io_ctl->pages[0]); | |
436 | } | |
437 | ||
438 | static int io_ctl_check_crc(struct io_ctl *io_ctl, int index) | |
439 | { | |
440 | u32 *tmp, val; | |
441 | u32 crc = ~(u32)0; | |
442 | unsigned offset = 0; | |
443 | ||
444 | if (!io_ctl->check_crcs) { | |
445 | io_ctl_map_page(io_ctl, 0); | |
446 | return 0; | |
447 | } | |
448 | ||
449 | if (index == 0) | |
450 | offset = sizeof(u32) * io_ctl->num_pages; | |
451 | ||
452 | tmp = kmap(io_ctl->pages[0]); | |
453 | tmp += index; | |
454 | val = *tmp; | |
455 | kunmap(io_ctl->pages[0]); | |
456 | ||
457 | io_ctl_map_page(io_ctl, 0); | |
458 | crc = btrfs_csum_data(io_ctl->root, io_ctl->orig + offset, crc, | |
459 | PAGE_CACHE_SIZE - offset); | |
460 | btrfs_csum_final(crc, (char *)&crc); | |
461 | if (val != crc) { | |
462 | printk_ratelimited(KERN_ERR "btrfs: csum mismatch on free " | |
463 | "space cache\n"); | |
464 | io_ctl_unmap_page(io_ctl); | |
465 | return -EIO; | |
466 | } | |
467 | ||
a67509c3 JB |
468 | return 0; |
469 | } | |
470 | ||
471 | static int io_ctl_add_entry(struct io_ctl *io_ctl, u64 offset, u64 bytes, | |
472 | void *bitmap) | |
473 | { | |
474 | struct btrfs_free_space_entry *entry; | |
475 | ||
476 | if (!io_ctl->cur) | |
477 | return -ENOSPC; | |
478 | ||
479 | entry = io_ctl->cur; | |
480 | entry->offset = cpu_to_le64(offset); | |
481 | entry->bytes = cpu_to_le64(bytes); | |
482 | entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP : | |
483 | BTRFS_FREE_SPACE_EXTENT; | |
484 | io_ctl->cur += sizeof(struct btrfs_free_space_entry); | |
485 | io_ctl->size -= sizeof(struct btrfs_free_space_entry); | |
486 | ||
487 | if (io_ctl->size >= sizeof(struct btrfs_free_space_entry)) | |
488 | return 0; | |
489 | ||
5b0e95bf | 490 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); |
a67509c3 JB |
491 | |
492 | /* No more pages to map */ | |
493 | if (io_ctl->index >= io_ctl->num_pages) | |
494 | return 0; | |
495 | ||
496 | /* map the next page */ | |
497 | io_ctl_map_page(io_ctl, 1); | |
498 | return 0; | |
499 | } | |
500 | ||
501 | static int io_ctl_add_bitmap(struct io_ctl *io_ctl, void *bitmap) | |
502 | { | |
503 | if (!io_ctl->cur) | |
504 | return -ENOSPC; | |
505 | ||
506 | /* | |
507 | * If we aren't at the start of the current page, unmap this one and | |
508 | * map the next one if there is any left. | |
509 | */ | |
510 | if (io_ctl->cur != io_ctl->orig) { | |
5b0e95bf | 511 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); |
a67509c3 JB |
512 | if (io_ctl->index >= io_ctl->num_pages) |
513 | return -ENOSPC; | |
514 | io_ctl_map_page(io_ctl, 0); | |
515 | } | |
516 | ||
517 | memcpy(io_ctl->cur, bitmap, PAGE_CACHE_SIZE); | |
5b0e95bf | 518 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); |
a67509c3 JB |
519 | if (io_ctl->index < io_ctl->num_pages) |
520 | io_ctl_map_page(io_ctl, 0); | |
521 | return 0; | |
522 | } | |
523 | ||
524 | static void io_ctl_zero_remaining_pages(struct io_ctl *io_ctl) | |
525 | { | |
5b0e95bf JB |
526 | /* |
527 | * If we're not on the boundary we know we've modified the page and we | |
528 | * need to crc the page. | |
529 | */ | |
530 | if (io_ctl->cur != io_ctl->orig) | |
531 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); | |
532 | else | |
533 | io_ctl_unmap_page(io_ctl); | |
a67509c3 JB |
534 | |
535 | while (io_ctl->index < io_ctl->num_pages) { | |
536 | io_ctl_map_page(io_ctl, 1); | |
5b0e95bf | 537 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); |
a67509c3 JB |
538 | } |
539 | } | |
540 | ||
5b0e95bf JB |
541 | static int io_ctl_read_entry(struct io_ctl *io_ctl, |
542 | struct btrfs_free_space *entry, u8 *type) | |
a67509c3 JB |
543 | { |
544 | struct btrfs_free_space_entry *e; | |
2f120c05 JB |
545 | int ret; |
546 | ||
547 | if (!io_ctl->cur) { | |
548 | ret = io_ctl_check_crc(io_ctl, io_ctl->index); | |
549 | if (ret) | |
550 | return ret; | |
551 | } | |
a67509c3 JB |
552 | |
553 | e = io_ctl->cur; | |
554 | entry->offset = le64_to_cpu(e->offset); | |
555 | entry->bytes = le64_to_cpu(e->bytes); | |
5b0e95bf | 556 | *type = e->type; |
a67509c3 JB |
557 | io_ctl->cur += sizeof(struct btrfs_free_space_entry); |
558 | io_ctl->size -= sizeof(struct btrfs_free_space_entry); | |
559 | ||
560 | if (io_ctl->size >= sizeof(struct btrfs_free_space_entry)) | |
5b0e95bf | 561 | return 0; |
a67509c3 JB |
562 | |
563 | io_ctl_unmap_page(io_ctl); | |
564 | ||
2f120c05 | 565 | return 0; |
a67509c3 JB |
566 | } |
567 | ||
5b0e95bf JB |
568 | static int io_ctl_read_bitmap(struct io_ctl *io_ctl, |
569 | struct btrfs_free_space *entry) | |
a67509c3 | 570 | { |
5b0e95bf JB |
571 | int ret; |
572 | ||
5b0e95bf JB |
573 | ret = io_ctl_check_crc(io_ctl, io_ctl->index); |
574 | if (ret) | |
575 | return ret; | |
576 | ||
a67509c3 JB |
577 | memcpy(entry->bitmap, io_ctl->cur, PAGE_CACHE_SIZE); |
578 | io_ctl_unmap_page(io_ctl); | |
5b0e95bf JB |
579 | |
580 | return 0; | |
a67509c3 JB |
581 | } |
582 | ||
0414efae LZ |
583 | int __load_free_space_cache(struct btrfs_root *root, struct inode *inode, |
584 | struct btrfs_free_space_ctl *ctl, | |
585 | struct btrfs_path *path, u64 offset) | |
9d66e233 | 586 | { |
9d66e233 JB |
587 | struct btrfs_free_space_header *header; |
588 | struct extent_buffer *leaf; | |
a67509c3 | 589 | struct io_ctl io_ctl; |
9d66e233 | 590 | struct btrfs_key key; |
a67509c3 | 591 | struct btrfs_free_space *e, *n; |
9d66e233 JB |
592 | struct list_head bitmaps; |
593 | u64 num_entries; | |
594 | u64 num_bitmaps; | |
595 | u64 generation; | |
a67509c3 | 596 | u8 type; |
f6a39829 | 597 | int ret = 0; |
9d66e233 JB |
598 | |
599 | INIT_LIST_HEAD(&bitmaps); | |
600 | ||
9d66e233 | 601 | /* Nothing in the space cache, goodbye */ |
0414efae | 602 | if (!i_size_read(inode)) |
a67509c3 | 603 | return 0; |
9d66e233 JB |
604 | |
605 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
0414efae | 606 | key.offset = offset; |
9d66e233 JB |
607 | key.type = 0; |
608 | ||
609 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
0414efae | 610 | if (ret < 0) |
a67509c3 | 611 | return 0; |
0414efae | 612 | else if (ret > 0) { |
945d8962 | 613 | btrfs_release_path(path); |
a67509c3 | 614 | return 0; |
9d66e233 JB |
615 | } |
616 | ||
0414efae LZ |
617 | ret = -1; |
618 | ||
9d66e233 JB |
619 | leaf = path->nodes[0]; |
620 | header = btrfs_item_ptr(leaf, path->slots[0], | |
621 | struct btrfs_free_space_header); | |
622 | num_entries = btrfs_free_space_entries(leaf, header); | |
623 | num_bitmaps = btrfs_free_space_bitmaps(leaf, header); | |
624 | generation = btrfs_free_space_generation(leaf, header); | |
945d8962 | 625 | btrfs_release_path(path); |
9d66e233 JB |
626 | |
627 | if (BTRFS_I(inode)->generation != generation) { | |
628 | printk(KERN_ERR "btrfs: free space inode generation (%llu) did" | |
0414efae | 629 | " not match free space cache generation (%llu)\n", |
9d66e233 | 630 | (unsigned long long)BTRFS_I(inode)->generation, |
0414efae | 631 | (unsigned long long)generation); |
a67509c3 | 632 | return 0; |
9d66e233 JB |
633 | } |
634 | ||
635 | if (!num_entries) | |
a67509c3 | 636 | return 0; |
9d66e233 | 637 | |
a67509c3 | 638 | io_ctl_init(&io_ctl, inode, root); |
9d66e233 | 639 | ret = readahead_cache(inode); |
0414efae | 640 | if (ret) |
9d66e233 | 641 | goto out; |
9d66e233 | 642 | |
a67509c3 JB |
643 | ret = io_ctl_prepare_pages(&io_ctl, inode, 1); |
644 | if (ret) | |
645 | goto out; | |
9d66e233 | 646 | |
5b0e95bf JB |
647 | ret = io_ctl_check_crc(&io_ctl, 0); |
648 | if (ret) | |
649 | goto free_cache; | |
650 | ||
a67509c3 JB |
651 | ret = io_ctl_check_generation(&io_ctl, generation); |
652 | if (ret) | |
653 | goto free_cache; | |
9d66e233 | 654 | |
a67509c3 JB |
655 | while (num_entries) { |
656 | e = kmem_cache_zalloc(btrfs_free_space_cachep, | |
657 | GFP_NOFS); | |
658 | if (!e) | |
9d66e233 | 659 | goto free_cache; |
9d66e233 | 660 | |
5b0e95bf JB |
661 | ret = io_ctl_read_entry(&io_ctl, e, &type); |
662 | if (ret) { | |
663 | kmem_cache_free(btrfs_free_space_cachep, e); | |
664 | goto free_cache; | |
665 | } | |
666 | ||
a67509c3 JB |
667 | if (!e->bytes) { |
668 | kmem_cache_free(btrfs_free_space_cachep, e); | |
669 | goto free_cache; | |
9d66e233 | 670 | } |
a67509c3 JB |
671 | |
672 | if (type == BTRFS_FREE_SPACE_EXTENT) { | |
673 | spin_lock(&ctl->tree_lock); | |
674 | ret = link_free_space(ctl, e); | |
675 | spin_unlock(&ctl->tree_lock); | |
676 | if (ret) { | |
677 | printk(KERN_ERR "Duplicate entries in " | |
678 | "free space cache, dumping\n"); | |
679 | kmem_cache_free(btrfs_free_space_cachep, e); | |
9d66e233 JB |
680 | goto free_cache; |
681 | } | |
a67509c3 JB |
682 | } else { |
683 | BUG_ON(!num_bitmaps); | |
684 | num_bitmaps--; | |
685 | e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS); | |
686 | if (!e->bitmap) { | |
687 | kmem_cache_free( | |
688 | btrfs_free_space_cachep, e); | |
9d66e233 JB |
689 | goto free_cache; |
690 | } | |
a67509c3 JB |
691 | spin_lock(&ctl->tree_lock); |
692 | ret = link_free_space(ctl, e); | |
693 | ctl->total_bitmaps++; | |
694 | ctl->op->recalc_thresholds(ctl); | |
695 | spin_unlock(&ctl->tree_lock); | |
696 | if (ret) { | |
697 | printk(KERN_ERR "Duplicate entries in " | |
698 | "free space cache, dumping\n"); | |
dc89e982 | 699 | kmem_cache_free(btrfs_free_space_cachep, e); |
9d66e233 JB |
700 | goto free_cache; |
701 | } | |
a67509c3 | 702 | list_add_tail(&e->list, &bitmaps); |
9d66e233 JB |
703 | } |
704 | ||
a67509c3 JB |
705 | num_entries--; |
706 | } | |
9d66e233 | 707 | |
2f120c05 JB |
708 | io_ctl_unmap_page(&io_ctl); |
709 | ||
a67509c3 JB |
710 | /* |
711 | * We add the bitmaps at the end of the entries in order that | |
712 | * the bitmap entries are added to the cache. | |
713 | */ | |
714 | list_for_each_entry_safe(e, n, &bitmaps, list) { | |
9d66e233 | 715 | list_del_init(&e->list); |
5b0e95bf JB |
716 | ret = io_ctl_read_bitmap(&io_ctl, e); |
717 | if (ret) | |
718 | goto free_cache; | |
9d66e233 JB |
719 | } |
720 | ||
a67509c3 | 721 | io_ctl_drop_pages(&io_ctl); |
9d66e233 JB |
722 | ret = 1; |
723 | out: | |
a67509c3 | 724 | io_ctl_free(&io_ctl); |
9d66e233 | 725 | return ret; |
9d66e233 | 726 | free_cache: |
a67509c3 | 727 | io_ctl_drop_pages(&io_ctl); |
0414efae | 728 | __btrfs_remove_free_space_cache(ctl); |
9d66e233 JB |
729 | goto out; |
730 | } | |
731 | ||
0414efae LZ |
732 | int load_free_space_cache(struct btrfs_fs_info *fs_info, |
733 | struct btrfs_block_group_cache *block_group) | |
0cb59c99 | 734 | { |
34d52cb6 | 735 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0414efae LZ |
736 | struct btrfs_root *root = fs_info->tree_root; |
737 | struct inode *inode; | |
738 | struct btrfs_path *path; | |
5b0e95bf | 739 | int ret = 0; |
0414efae LZ |
740 | bool matched; |
741 | u64 used = btrfs_block_group_used(&block_group->item); | |
742 | ||
743 | /* | |
744 | * If we're unmounting then just return, since this does a search on the | |
745 | * normal root and not the commit root and we could deadlock. | |
746 | */ | |
7841cb28 | 747 | if (btrfs_fs_closing(fs_info)) |
0414efae LZ |
748 | return 0; |
749 | ||
750 | /* | |
751 | * If this block group has been marked to be cleared for one reason or | |
752 | * another then we can't trust the on disk cache, so just return. | |
753 | */ | |
9d66e233 | 754 | spin_lock(&block_group->lock); |
0414efae LZ |
755 | if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) { |
756 | spin_unlock(&block_group->lock); | |
757 | return 0; | |
758 | } | |
9d66e233 | 759 | spin_unlock(&block_group->lock); |
0414efae LZ |
760 | |
761 | path = btrfs_alloc_path(); | |
762 | if (!path) | |
763 | return 0; | |
764 | ||
765 | inode = lookup_free_space_inode(root, block_group, path); | |
766 | if (IS_ERR(inode)) { | |
767 | btrfs_free_path(path); | |
768 | return 0; | |
769 | } | |
770 | ||
5b0e95bf JB |
771 | /* We may have converted the inode and made the cache invalid. */ |
772 | spin_lock(&block_group->lock); | |
773 | if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) { | |
774 | spin_unlock(&block_group->lock); | |
775 | goto out; | |
776 | } | |
777 | spin_unlock(&block_group->lock); | |
778 | ||
0414efae LZ |
779 | ret = __load_free_space_cache(fs_info->tree_root, inode, ctl, |
780 | path, block_group->key.objectid); | |
781 | btrfs_free_path(path); | |
782 | if (ret <= 0) | |
783 | goto out; | |
784 | ||
785 | spin_lock(&ctl->tree_lock); | |
786 | matched = (ctl->free_space == (block_group->key.offset - used - | |
787 | block_group->bytes_super)); | |
788 | spin_unlock(&ctl->tree_lock); | |
789 | ||
790 | if (!matched) { | |
791 | __btrfs_remove_free_space_cache(ctl); | |
792 | printk(KERN_ERR "block group %llu has an wrong amount of free " | |
793 | "space\n", block_group->key.objectid); | |
794 | ret = -1; | |
795 | } | |
796 | out: | |
797 | if (ret < 0) { | |
798 | /* This cache is bogus, make sure it gets cleared */ | |
799 | spin_lock(&block_group->lock); | |
800 | block_group->disk_cache_state = BTRFS_DC_CLEAR; | |
801 | spin_unlock(&block_group->lock); | |
82d5902d | 802 | ret = 0; |
0414efae LZ |
803 | |
804 | printk(KERN_ERR "btrfs: failed to load free space cache " | |
805 | "for block group %llu\n", block_group->key.objectid); | |
806 | } | |
807 | ||
808 | iput(inode); | |
809 | return ret; | |
9d66e233 JB |
810 | } |
811 | ||
c09544e0 JB |
812 | /** |
813 | * __btrfs_write_out_cache - write out cached info to an inode | |
814 | * @root - the root the inode belongs to | |
815 | * @ctl - the free space cache we are going to write out | |
816 | * @block_group - the block_group for this cache if it belongs to a block_group | |
817 | * @trans - the trans handle | |
818 | * @path - the path to use | |
819 | * @offset - the offset for the key we'll insert | |
820 | * | |
821 | * This function writes out a free space cache struct to disk for quick recovery | |
822 | * on mount. This will return 0 if it was successfull in writing the cache out, | |
823 | * and -1 if it was not. | |
824 | */ | |
0414efae LZ |
825 | int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode, |
826 | struct btrfs_free_space_ctl *ctl, | |
827 | struct btrfs_block_group_cache *block_group, | |
828 | struct btrfs_trans_handle *trans, | |
829 | struct btrfs_path *path, u64 offset) | |
0cb59c99 JB |
830 | { |
831 | struct btrfs_free_space_header *header; | |
832 | struct extent_buffer *leaf; | |
0cb59c99 JB |
833 | struct rb_node *node; |
834 | struct list_head *pos, *n; | |
0cb59c99 | 835 | struct extent_state *cached_state = NULL; |
43be2146 JB |
836 | struct btrfs_free_cluster *cluster = NULL; |
837 | struct extent_io_tree *unpin = NULL; | |
a67509c3 | 838 | struct io_ctl io_ctl; |
0cb59c99 JB |
839 | struct list_head bitmap_list; |
840 | struct btrfs_key key; | |
43be2146 | 841 | u64 start, end, len; |
0cb59c99 JB |
842 | int entries = 0; |
843 | int bitmaps = 0; | |
c09544e0 JB |
844 | int ret; |
845 | int err = -1; | |
0cb59c99 | 846 | |
0cb59c99 JB |
847 | INIT_LIST_HEAD(&bitmap_list); |
848 | ||
0414efae LZ |
849 | if (!i_size_read(inode)) |
850 | return -1; | |
2b20982e | 851 | |
a67509c3 | 852 | io_ctl_init(&io_ctl, inode, root); |
be1a12a0 | 853 | |
43be2146 | 854 | /* Get the cluster for this block_group if it exists */ |
0414efae | 855 | if (block_group && !list_empty(&block_group->cluster_list)) |
43be2146 JB |
856 | cluster = list_entry(block_group->cluster_list.next, |
857 | struct btrfs_free_cluster, | |
858 | block_group_list); | |
859 | ||
860 | /* | |
861 | * We shouldn't have switched the pinned extents yet so this is the | |
862 | * right one | |
863 | */ | |
864 | unpin = root->fs_info->pinned_extents; | |
865 | ||
a67509c3 JB |
866 | /* Lock all pages first so we can lock the extent safely. */ |
867 | io_ctl_prepare_pages(&io_ctl, inode, 0); | |
0cb59c99 | 868 | |
0cb59c99 JB |
869 | lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1, |
870 | 0, &cached_state, GFP_NOFS); | |
871 | ||
43be2146 JB |
872 | /* |
873 | * When searching for pinned extents, we need to start at our start | |
874 | * offset. | |
875 | */ | |
0414efae LZ |
876 | if (block_group) |
877 | start = block_group->key.objectid; | |
43be2146 | 878 | |
f75b130e JB |
879 | node = rb_first(&ctl->free_space_offset); |
880 | if (!node && cluster) { | |
881 | node = rb_first(&cluster->root); | |
882 | cluster = NULL; | |
883 | } | |
884 | ||
5b0e95bf JB |
885 | /* Make sure we can fit our crcs into the first page */ |
886 | if (io_ctl.check_crcs && | |
887 | (io_ctl.num_pages * sizeof(u32)) >= PAGE_CACHE_SIZE) { | |
888 | WARN_ON(1); | |
889 | goto out_nospc; | |
890 | } | |
891 | ||
a67509c3 | 892 | io_ctl_set_generation(&io_ctl, trans->transid); |
43be2146 | 893 | |
a67509c3 JB |
894 | /* Write out the extent entries */ |
895 | while (node) { | |
896 | struct btrfs_free_space *e; | |
0cb59c99 | 897 | |
a67509c3 JB |
898 | e = rb_entry(node, struct btrfs_free_space, offset_index); |
899 | entries++; | |
0cb59c99 | 900 | |
a67509c3 JB |
901 | ret = io_ctl_add_entry(&io_ctl, e->offset, e->bytes, |
902 | e->bitmap); | |
903 | if (ret) | |
904 | goto out_nospc; | |
2f356126 | 905 | |
a67509c3 JB |
906 | if (e->bitmap) { |
907 | list_add_tail(&e->list, &bitmap_list); | |
908 | bitmaps++; | |
2f356126 | 909 | } |
a67509c3 JB |
910 | node = rb_next(node); |
911 | if (!node && cluster) { | |
912 | node = rb_first(&cluster->root); | |
913 | cluster = NULL; | |
43be2146 | 914 | } |
a67509c3 | 915 | } |
43be2146 | 916 | |
a67509c3 JB |
917 | /* |
918 | * We want to add any pinned extents to our free space cache | |
919 | * so we don't leak the space | |
920 | */ | |
921 | while (block_group && (start < block_group->key.objectid + | |
922 | block_group->key.offset)) { | |
923 | ret = find_first_extent_bit(unpin, start, &start, &end, | |
924 | EXTENT_DIRTY); | |
925 | if (ret) { | |
926 | ret = 0; | |
927 | break; | |
0cb59c99 | 928 | } |
0cb59c99 | 929 | |
a67509c3 JB |
930 | /* This pinned extent is out of our range */ |
931 | if (start >= block_group->key.objectid + | |
932 | block_group->key.offset) | |
933 | break; | |
2f356126 | 934 | |
a67509c3 JB |
935 | len = block_group->key.objectid + |
936 | block_group->key.offset - start; | |
937 | len = min(len, end + 1 - start); | |
0cb59c99 | 938 | |
a67509c3 JB |
939 | entries++; |
940 | ret = io_ctl_add_entry(&io_ctl, start, len, NULL); | |
941 | if (ret) | |
942 | goto out_nospc; | |
0cb59c99 | 943 | |
a67509c3 JB |
944 | start = end + 1; |
945 | } | |
0cb59c99 JB |
946 | |
947 | /* Write out the bitmaps */ | |
948 | list_for_each_safe(pos, n, &bitmap_list) { | |
0cb59c99 JB |
949 | struct btrfs_free_space *entry = |
950 | list_entry(pos, struct btrfs_free_space, list); | |
951 | ||
a67509c3 JB |
952 | ret = io_ctl_add_bitmap(&io_ctl, entry->bitmap); |
953 | if (ret) | |
954 | goto out_nospc; | |
0cb59c99 | 955 | list_del_init(&entry->list); |
be1a12a0 JB |
956 | } |
957 | ||
0cb59c99 | 958 | /* Zero out the rest of the pages just to make sure */ |
a67509c3 | 959 | io_ctl_zero_remaining_pages(&io_ctl); |
0cb59c99 | 960 | |
a67509c3 JB |
961 | ret = btrfs_dirty_pages(root, inode, io_ctl.pages, io_ctl.num_pages, |
962 | 0, i_size_read(inode), &cached_state); | |
963 | io_ctl_drop_pages(&io_ctl); | |
0cb59c99 JB |
964 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0, |
965 | i_size_read(inode) - 1, &cached_state, GFP_NOFS); | |
966 | ||
c09544e0 | 967 | if (ret) |
2f356126 | 968 | goto out; |
be1a12a0 | 969 | |
be1a12a0 | 970 | |
549b4fdb JB |
971 | ret = filemap_write_and_wait(inode->i_mapping); |
972 | if (ret) | |
973 | goto out; | |
0cb59c99 JB |
974 | |
975 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
0414efae | 976 | key.offset = offset; |
0cb59c99 JB |
977 | key.type = 0; |
978 | ||
a9b5fcdd | 979 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
0cb59c99 | 980 | if (ret < 0) { |
a67509c3 | 981 | clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1, |
5b0e95bf JB |
982 | EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL, |
983 | GFP_NOFS); | |
2f356126 | 984 | goto out; |
0cb59c99 JB |
985 | } |
986 | leaf = path->nodes[0]; | |
987 | if (ret > 0) { | |
988 | struct btrfs_key found_key; | |
989 | BUG_ON(!path->slots[0]); | |
990 | path->slots[0]--; | |
991 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
992 | if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID || | |
0414efae | 993 | found_key.offset != offset) { |
a67509c3 JB |
994 | clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, |
995 | inode->i_size - 1, | |
5b0e95bf JB |
996 | EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, |
997 | NULL, GFP_NOFS); | |
b3b4aa74 | 998 | btrfs_release_path(path); |
2f356126 | 999 | goto out; |
0cb59c99 JB |
1000 | } |
1001 | } | |
549b4fdb JB |
1002 | |
1003 | BTRFS_I(inode)->generation = trans->transid; | |
0cb59c99 JB |
1004 | header = btrfs_item_ptr(leaf, path->slots[0], |
1005 | struct btrfs_free_space_header); | |
1006 | btrfs_set_free_space_entries(leaf, header, entries); | |
1007 | btrfs_set_free_space_bitmaps(leaf, header, bitmaps); | |
1008 | btrfs_set_free_space_generation(leaf, header, trans->transid); | |
1009 | btrfs_mark_buffer_dirty(leaf); | |
b3b4aa74 | 1010 | btrfs_release_path(path); |
0cb59c99 | 1011 | |
c09544e0 | 1012 | err = 0; |
2f356126 | 1013 | out: |
a67509c3 | 1014 | io_ctl_free(&io_ctl); |
c09544e0 | 1015 | if (err) { |
a67509c3 | 1016 | invalidate_inode_pages2(inode->i_mapping); |
0cb59c99 JB |
1017 | BTRFS_I(inode)->generation = 0; |
1018 | } | |
0cb59c99 | 1019 | btrfs_update_inode(trans, root, inode); |
c09544e0 | 1020 | return err; |
a67509c3 JB |
1021 | |
1022 | out_nospc: | |
1023 | list_for_each_safe(pos, n, &bitmap_list) { | |
1024 | struct btrfs_free_space *entry = | |
1025 | list_entry(pos, struct btrfs_free_space, list); | |
1026 | list_del_init(&entry->list); | |
1027 | } | |
1028 | io_ctl_drop_pages(&io_ctl); | |
1029 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0, | |
1030 | i_size_read(inode) - 1, &cached_state, GFP_NOFS); | |
1031 | goto out; | |
0414efae LZ |
1032 | } |
1033 | ||
1034 | int btrfs_write_out_cache(struct btrfs_root *root, | |
1035 | struct btrfs_trans_handle *trans, | |
1036 | struct btrfs_block_group_cache *block_group, | |
1037 | struct btrfs_path *path) | |
1038 | { | |
1039 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
1040 | struct inode *inode; | |
1041 | int ret = 0; | |
1042 | ||
1043 | root = root->fs_info->tree_root; | |
1044 | ||
1045 | spin_lock(&block_group->lock); | |
1046 | if (block_group->disk_cache_state < BTRFS_DC_SETUP) { | |
1047 | spin_unlock(&block_group->lock); | |
1048 | return 0; | |
1049 | } | |
1050 | spin_unlock(&block_group->lock); | |
1051 | ||
1052 | inode = lookup_free_space_inode(root, block_group, path); | |
1053 | if (IS_ERR(inode)) | |
1054 | return 0; | |
1055 | ||
1056 | ret = __btrfs_write_out_cache(root, inode, ctl, block_group, trans, | |
1057 | path, block_group->key.objectid); | |
c09544e0 | 1058 | if (ret) { |
0414efae LZ |
1059 | spin_lock(&block_group->lock); |
1060 | block_group->disk_cache_state = BTRFS_DC_ERROR; | |
1061 | spin_unlock(&block_group->lock); | |
82d5902d | 1062 | ret = 0; |
c09544e0 | 1063 | #ifdef DEBUG |
0414efae LZ |
1064 | printk(KERN_ERR "btrfs: failed to write free space cace " |
1065 | "for block group %llu\n", block_group->key.objectid); | |
c09544e0 | 1066 | #endif |
0414efae LZ |
1067 | } |
1068 | ||
0cb59c99 JB |
1069 | iput(inode); |
1070 | return ret; | |
1071 | } | |
1072 | ||
34d52cb6 | 1073 | static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit, |
96303081 | 1074 | u64 offset) |
0f9dd46c | 1075 | { |
96303081 JB |
1076 | BUG_ON(offset < bitmap_start); |
1077 | offset -= bitmap_start; | |
34d52cb6 | 1078 | return (unsigned long)(div_u64(offset, unit)); |
96303081 | 1079 | } |
0f9dd46c | 1080 | |
34d52cb6 | 1081 | static inline unsigned long bytes_to_bits(u64 bytes, u32 unit) |
96303081 | 1082 | { |
34d52cb6 | 1083 | return (unsigned long)(div_u64(bytes, unit)); |
96303081 | 1084 | } |
0f9dd46c | 1085 | |
34d52cb6 | 1086 | static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1087 | u64 offset) |
1088 | { | |
1089 | u64 bitmap_start; | |
1090 | u64 bytes_per_bitmap; | |
0f9dd46c | 1091 | |
34d52cb6 LZ |
1092 | bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit; |
1093 | bitmap_start = offset - ctl->start; | |
96303081 JB |
1094 | bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap); |
1095 | bitmap_start *= bytes_per_bitmap; | |
34d52cb6 | 1096 | bitmap_start += ctl->start; |
0f9dd46c | 1097 | |
96303081 | 1098 | return bitmap_start; |
0f9dd46c JB |
1099 | } |
1100 | ||
96303081 JB |
1101 | static int tree_insert_offset(struct rb_root *root, u64 offset, |
1102 | struct rb_node *node, int bitmap) | |
0f9dd46c JB |
1103 | { |
1104 | struct rb_node **p = &root->rb_node; | |
1105 | struct rb_node *parent = NULL; | |
1106 | struct btrfs_free_space *info; | |
1107 | ||
1108 | while (*p) { | |
1109 | parent = *p; | |
96303081 | 1110 | info = rb_entry(parent, struct btrfs_free_space, offset_index); |
0f9dd46c | 1111 | |
96303081 | 1112 | if (offset < info->offset) { |
0f9dd46c | 1113 | p = &(*p)->rb_left; |
96303081 | 1114 | } else if (offset > info->offset) { |
0f9dd46c | 1115 | p = &(*p)->rb_right; |
96303081 JB |
1116 | } else { |
1117 | /* | |
1118 | * we could have a bitmap entry and an extent entry | |
1119 | * share the same offset. If this is the case, we want | |
1120 | * the extent entry to always be found first if we do a | |
1121 | * linear search through the tree, since we want to have | |
1122 | * the quickest allocation time, and allocating from an | |
1123 | * extent is faster than allocating from a bitmap. So | |
1124 | * if we're inserting a bitmap and we find an entry at | |
1125 | * this offset, we want to go right, or after this entry | |
1126 | * logically. If we are inserting an extent and we've | |
1127 | * found a bitmap, we want to go left, or before | |
1128 | * logically. | |
1129 | */ | |
1130 | if (bitmap) { | |
207dde82 JB |
1131 | if (info->bitmap) { |
1132 | WARN_ON_ONCE(1); | |
1133 | return -EEXIST; | |
1134 | } | |
96303081 JB |
1135 | p = &(*p)->rb_right; |
1136 | } else { | |
207dde82 JB |
1137 | if (!info->bitmap) { |
1138 | WARN_ON_ONCE(1); | |
1139 | return -EEXIST; | |
1140 | } | |
96303081 JB |
1141 | p = &(*p)->rb_left; |
1142 | } | |
1143 | } | |
0f9dd46c JB |
1144 | } |
1145 | ||
1146 | rb_link_node(node, parent, p); | |
1147 | rb_insert_color(node, root); | |
1148 | ||
1149 | return 0; | |
1150 | } | |
1151 | ||
1152 | /* | |
70cb0743 JB |
1153 | * searches the tree for the given offset. |
1154 | * | |
96303081 JB |
1155 | * fuzzy - If this is set, then we are trying to make an allocation, and we just |
1156 | * want a section that has at least bytes size and comes at or after the given | |
1157 | * offset. | |
0f9dd46c | 1158 | */ |
96303081 | 1159 | static struct btrfs_free_space * |
34d52cb6 | 1160 | tree_search_offset(struct btrfs_free_space_ctl *ctl, |
96303081 | 1161 | u64 offset, int bitmap_only, int fuzzy) |
0f9dd46c | 1162 | { |
34d52cb6 | 1163 | struct rb_node *n = ctl->free_space_offset.rb_node; |
96303081 JB |
1164 | struct btrfs_free_space *entry, *prev = NULL; |
1165 | ||
1166 | /* find entry that is closest to the 'offset' */ | |
1167 | while (1) { | |
1168 | if (!n) { | |
1169 | entry = NULL; | |
1170 | break; | |
1171 | } | |
0f9dd46c | 1172 | |
0f9dd46c | 1173 | entry = rb_entry(n, struct btrfs_free_space, offset_index); |
96303081 | 1174 | prev = entry; |
0f9dd46c | 1175 | |
96303081 | 1176 | if (offset < entry->offset) |
0f9dd46c | 1177 | n = n->rb_left; |
96303081 | 1178 | else if (offset > entry->offset) |
0f9dd46c | 1179 | n = n->rb_right; |
96303081 | 1180 | else |
0f9dd46c | 1181 | break; |
0f9dd46c JB |
1182 | } |
1183 | ||
96303081 JB |
1184 | if (bitmap_only) { |
1185 | if (!entry) | |
1186 | return NULL; | |
1187 | if (entry->bitmap) | |
1188 | return entry; | |
0f9dd46c | 1189 | |
96303081 JB |
1190 | /* |
1191 | * bitmap entry and extent entry may share same offset, | |
1192 | * in that case, bitmap entry comes after extent entry. | |
1193 | */ | |
1194 | n = rb_next(n); | |
1195 | if (!n) | |
1196 | return NULL; | |
1197 | entry = rb_entry(n, struct btrfs_free_space, offset_index); | |
1198 | if (entry->offset != offset) | |
1199 | return NULL; | |
0f9dd46c | 1200 | |
96303081 JB |
1201 | WARN_ON(!entry->bitmap); |
1202 | return entry; | |
1203 | } else if (entry) { | |
1204 | if (entry->bitmap) { | |
0f9dd46c | 1205 | /* |
96303081 JB |
1206 | * if previous extent entry covers the offset, |
1207 | * we should return it instead of the bitmap entry | |
0f9dd46c | 1208 | */ |
96303081 JB |
1209 | n = &entry->offset_index; |
1210 | while (1) { | |
1211 | n = rb_prev(n); | |
1212 | if (!n) | |
1213 | break; | |
1214 | prev = rb_entry(n, struct btrfs_free_space, | |
1215 | offset_index); | |
1216 | if (!prev->bitmap) { | |
1217 | if (prev->offset + prev->bytes > offset) | |
1218 | entry = prev; | |
1219 | break; | |
1220 | } | |
0f9dd46c | 1221 | } |
96303081 JB |
1222 | } |
1223 | return entry; | |
1224 | } | |
1225 | ||
1226 | if (!prev) | |
1227 | return NULL; | |
1228 | ||
1229 | /* find last entry before the 'offset' */ | |
1230 | entry = prev; | |
1231 | if (entry->offset > offset) { | |
1232 | n = rb_prev(&entry->offset_index); | |
1233 | if (n) { | |
1234 | entry = rb_entry(n, struct btrfs_free_space, | |
1235 | offset_index); | |
1236 | BUG_ON(entry->offset > offset); | |
0f9dd46c | 1237 | } else { |
96303081 JB |
1238 | if (fuzzy) |
1239 | return entry; | |
1240 | else | |
1241 | return NULL; | |
0f9dd46c JB |
1242 | } |
1243 | } | |
1244 | ||
96303081 JB |
1245 | if (entry->bitmap) { |
1246 | n = &entry->offset_index; | |
1247 | while (1) { | |
1248 | n = rb_prev(n); | |
1249 | if (!n) | |
1250 | break; | |
1251 | prev = rb_entry(n, struct btrfs_free_space, | |
1252 | offset_index); | |
1253 | if (!prev->bitmap) { | |
1254 | if (prev->offset + prev->bytes > offset) | |
1255 | return prev; | |
1256 | break; | |
1257 | } | |
1258 | } | |
34d52cb6 | 1259 | if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset) |
96303081 JB |
1260 | return entry; |
1261 | } else if (entry->offset + entry->bytes > offset) | |
1262 | return entry; | |
1263 | ||
1264 | if (!fuzzy) | |
1265 | return NULL; | |
1266 | ||
1267 | while (1) { | |
1268 | if (entry->bitmap) { | |
1269 | if (entry->offset + BITS_PER_BITMAP * | |
34d52cb6 | 1270 | ctl->unit > offset) |
96303081 JB |
1271 | break; |
1272 | } else { | |
1273 | if (entry->offset + entry->bytes > offset) | |
1274 | break; | |
1275 | } | |
1276 | ||
1277 | n = rb_next(&entry->offset_index); | |
1278 | if (!n) | |
1279 | return NULL; | |
1280 | entry = rb_entry(n, struct btrfs_free_space, offset_index); | |
1281 | } | |
1282 | return entry; | |
0f9dd46c JB |
1283 | } |
1284 | ||
f333adb5 | 1285 | static inline void |
34d52cb6 | 1286 | __unlink_free_space(struct btrfs_free_space_ctl *ctl, |
f333adb5 | 1287 | struct btrfs_free_space *info) |
0f9dd46c | 1288 | { |
34d52cb6 LZ |
1289 | rb_erase(&info->offset_index, &ctl->free_space_offset); |
1290 | ctl->free_extents--; | |
f333adb5 LZ |
1291 | } |
1292 | ||
34d52cb6 | 1293 | static void unlink_free_space(struct btrfs_free_space_ctl *ctl, |
f333adb5 LZ |
1294 | struct btrfs_free_space *info) |
1295 | { | |
34d52cb6 LZ |
1296 | __unlink_free_space(ctl, info); |
1297 | ctl->free_space -= info->bytes; | |
0f9dd46c JB |
1298 | } |
1299 | ||
34d52cb6 | 1300 | static int link_free_space(struct btrfs_free_space_ctl *ctl, |
0f9dd46c JB |
1301 | struct btrfs_free_space *info) |
1302 | { | |
1303 | int ret = 0; | |
1304 | ||
96303081 | 1305 | BUG_ON(!info->bitmap && !info->bytes); |
34d52cb6 | 1306 | ret = tree_insert_offset(&ctl->free_space_offset, info->offset, |
96303081 | 1307 | &info->offset_index, (info->bitmap != NULL)); |
0f9dd46c JB |
1308 | if (ret) |
1309 | return ret; | |
1310 | ||
34d52cb6 LZ |
1311 | ctl->free_space += info->bytes; |
1312 | ctl->free_extents++; | |
96303081 JB |
1313 | return ret; |
1314 | } | |
1315 | ||
34d52cb6 | 1316 | static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl) |
96303081 | 1317 | { |
34d52cb6 | 1318 | struct btrfs_block_group_cache *block_group = ctl->private; |
25891f79 JB |
1319 | u64 max_bytes; |
1320 | u64 bitmap_bytes; | |
1321 | u64 extent_bytes; | |
8eb2d829 | 1322 | u64 size = block_group->key.offset; |
34d52cb6 LZ |
1323 | u64 bytes_per_bg = BITS_PER_BITMAP * block_group->sectorsize; |
1324 | int max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg); | |
1325 | ||
1326 | BUG_ON(ctl->total_bitmaps > max_bitmaps); | |
96303081 JB |
1327 | |
1328 | /* | |
1329 | * The goal is to keep the total amount of memory used per 1gb of space | |
1330 | * at or below 32k, so we need to adjust how much memory we allow to be | |
1331 | * used by extent based free space tracking | |
1332 | */ | |
8eb2d829 LZ |
1333 | if (size < 1024 * 1024 * 1024) |
1334 | max_bytes = MAX_CACHE_BYTES_PER_GIG; | |
1335 | else | |
1336 | max_bytes = MAX_CACHE_BYTES_PER_GIG * | |
1337 | div64_u64(size, 1024 * 1024 * 1024); | |
96303081 | 1338 | |
25891f79 JB |
1339 | /* |
1340 | * we want to account for 1 more bitmap than what we have so we can make | |
1341 | * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as | |
1342 | * we add more bitmaps. | |
1343 | */ | |
34d52cb6 | 1344 | bitmap_bytes = (ctl->total_bitmaps + 1) * PAGE_CACHE_SIZE; |
96303081 | 1345 | |
25891f79 | 1346 | if (bitmap_bytes >= max_bytes) { |
34d52cb6 | 1347 | ctl->extents_thresh = 0; |
25891f79 JB |
1348 | return; |
1349 | } | |
96303081 | 1350 | |
25891f79 JB |
1351 | /* |
1352 | * we want the extent entry threshold to always be at most 1/2 the maxw | |
1353 | * bytes we can have, or whatever is less than that. | |
1354 | */ | |
1355 | extent_bytes = max_bytes - bitmap_bytes; | |
1356 | extent_bytes = min_t(u64, extent_bytes, div64_u64(max_bytes, 2)); | |
96303081 | 1357 | |
34d52cb6 | 1358 | ctl->extents_thresh = |
25891f79 | 1359 | div64_u64(extent_bytes, (sizeof(struct btrfs_free_space))); |
96303081 JB |
1360 | } |
1361 | ||
bb3ac5a4 MX |
1362 | static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl, |
1363 | struct btrfs_free_space *info, | |
1364 | u64 offset, u64 bytes) | |
96303081 | 1365 | { |
f38b6e75 | 1366 | unsigned long start, count; |
96303081 | 1367 | |
34d52cb6 LZ |
1368 | start = offset_to_bit(info->offset, ctl->unit, offset); |
1369 | count = bytes_to_bits(bytes, ctl->unit); | |
f38b6e75 | 1370 | BUG_ON(start + count > BITS_PER_BITMAP); |
96303081 | 1371 | |
f38b6e75 | 1372 | bitmap_clear(info->bitmap, start, count); |
96303081 JB |
1373 | |
1374 | info->bytes -= bytes; | |
bb3ac5a4 MX |
1375 | } |
1376 | ||
1377 | static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl, | |
1378 | struct btrfs_free_space *info, u64 offset, | |
1379 | u64 bytes) | |
1380 | { | |
1381 | __bitmap_clear_bits(ctl, info, offset, bytes); | |
34d52cb6 | 1382 | ctl->free_space -= bytes; |
96303081 JB |
1383 | } |
1384 | ||
34d52cb6 | 1385 | static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl, |
817d52f8 JB |
1386 | struct btrfs_free_space *info, u64 offset, |
1387 | u64 bytes) | |
96303081 | 1388 | { |
f38b6e75 | 1389 | unsigned long start, count; |
96303081 | 1390 | |
34d52cb6 LZ |
1391 | start = offset_to_bit(info->offset, ctl->unit, offset); |
1392 | count = bytes_to_bits(bytes, ctl->unit); | |
f38b6e75 | 1393 | BUG_ON(start + count > BITS_PER_BITMAP); |
96303081 | 1394 | |
f38b6e75 | 1395 | bitmap_set(info->bitmap, start, count); |
96303081 JB |
1396 | |
1397 | info->bytes += bytes; | |
34d52cb6 | 1398 | ctl->free_space += bytes; |
96303081 JB |
1399 | } |
1400 | ||
34d52cb6 | 1401 | static int search_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1402 | struct btrfs_free_space *bitmap_info, u64 *offset, |
1403 | u64 *bytes) | |
1404 | { | |
1405 | unsigned long found_bits = 0; | |
1406 | unsigned long bits, i; | |
1407 | unsigned long next_zero; | |
1408 | ||
34d52cb6 | 1409 | i = offset_to_bit(bitmap_info->offset, ctl->unit, |
96303081 | 1410 | max_t(u64, *offset, bitmap_info->offset)); |
34d52cb6 | 1411 | bits = bytes_to_bits(*bytes, ctl->unit); |
96303081 JB |
1412 | |
1413 | for (i = find_next_bit(bitmap_info->bitmap, BITS_PER_BITMAP, i); | |
1414 | i < BITS_PER_BITMAP; | |
1415 | i = find_next_bit(bitmap_info->bitmap, BITS_PER_BITMAP, i + 1)) { | |
1416 | next_zero = find_next_zero_bit(bitmap_info->bitmap, | |
1417 | BITS_PER_BITMAP, i); | |
1418 | if ((next_zero - i) >= bits) { | |
1419 | found_bits = next_zero - i; | |
1420 | break; | |
1421 | } | |
1422 | i = next_zero; | |
1423 | } | |
1424 | ||
1425 | if (found_bits) { | |
34d52cb6 LZ |
1426 | *offset = (u64)(i * ctl->unit) + bitmap_info->offset; |
1427 | *bytes = (u64)(found_bits) * ctl->unit; | |
96303081 JB |
1428 | return 0; |
1429 | } | |
1430 | ||
1431 | return -1; | |
1432 | } | |
1433 | ||
34d52cb6 LZ |
1434 | static struct btrfs_free_space * |
1435 | find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes) | |
96303081 JB |
1436 | { |
1437 | struct btrfs_free_space *entry; | |
1438 | struct rb_node *node; | |
1439 | int ret; | |
1440 | ||
34d52cb6 | 1441 | if (!ctl->free_space_offset.rb_node) |
96303081 JB |
1442 | return NULL; |
1443 | ||
34d52cb6 | 1444 | entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1); |
96303081 JB |
1445 | if (!entry) |
1446 | return NULL; | |
1447 | ||
1448 | for (node = &entry->offset_index; node; node = rb_next(node)) { | |
1449 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
1450 | if (entry->bytes < *bytes) | |
1451 | continue; | |
1452 | ||
1453 | if (entry->bitmap) { | |
34d52cb6 | 1454 | ret = search_bitmap(ctl, entry, offset, bytes); |
96303081 JB |
1455 | if (!ret) |
1456 | return entry; | |
1457 | continue; | |
1458 | } | |
1459 | ||
1460 | *offset = entry->offset; | |
1461 | *bytes = entry->bytes; | |
1462 | return entry; | |
1463 | } | |
1464 | ||
1465 | return NULL; | |
1466 | } | |
1467 | ||
34d52cb6 | 1468 | static void add_new_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1469 | struct btrfs_free_space *info, u64 offset) |
1470 | { | |
34d52cb6 | 1471 | info->offset = offset_to_bitmap(ctl, offset); |
f019f426 | 1472 | info->bytes = 0; |
34d52cb6 LZ |
1473 | link_free_space(ctl, info); |
1474 | ctl->total_bitmaps++; | |
96303081 | 1475 | |
34d52cb6 | 1476 | ctl->op->recalc_thresholds(ctl); |
96303081 JB |
1477 | } |
1478 | ||
34d52cb6 | 1479 | static void free_bitmap(struct btrfs_free_space_ctl *ctl, |
edf6e2d1 LZ |
1480 | struct btrfs_free_space *bitmap_info) |
1481 | { | |
34d52cb6 | 1482 | unlink_free_space(ctl, bitmap_info); |
edf6e2d1 | 1483 | kfree(bitmap_info->bitmap); |
dc89e982 | 1484 | kmem_cache_free(btrfs_free_space_cachep, bitmap_info); |
34d52cb6 LZ |
1485 | ctl->total_bitmaps--; |
1486 | ctl->op->recalc_thresholds(ctl); | |
edf6e2d1 LZ |
1487 | } |
1488 | ||
34d52cb6 | 1489 | static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1490 | struct btrfs_free_space *bitmap_info, |
1491 | u64 *offset, u64 *bytes) | |
1492 | { | |
1493 | u64 end; | |
6606bb97 JB |
1494 | u64 search_start, search_bytes; |
1495 | int ret; | |
96303081 JB |
1496 | |
1497 | again: | |
34d52cb6 | 1498 | end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1; |
96303081 | 1499 | |
6606bb97 JB |
1500 | /* |
1501 | * XXX - this can go away after a few releases. | |
1502 | * | |
1503 | * since the only user of btrfs_remove_free_space is the tree logging | |
1504 | * stuff, and the only way to test that is under crash conditions, we | |
1505 | * want to have this debug stuff here just in case somethings not | |
1506 | * working. Search the bitmap for the space we are trying to use to | |
1507 | * make sure its actually there. If its not there then we need to stop | |
1508 | * because something has gone wrong. | |
1509 | */ | |
1510 | search_start = *offset; | |
1511 | search_bytes = *bytes; | |
13dbc089 | 1512 | search_bytes = min(search_bytes, end - search_start + 1); |
34d52cb6 | 1513 | ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes); |
6606bb97 JB |
1514 | BUG_ON(ret < 0 || search_start != *offset); |
1515 | ||
96303081 | 1516 | if (*offset > bitmap_info->offset && *offset + *bytes > end) { |
34d52cb6 | 1517 | bitmap_clear_bits(ctl, bitmap_info, *offset, end - *offset + 1); |
96303081 JB |
1518 | *bytes -= end - *offset + 1; |
1519 | *offset = end + 1; | |
1520 | } else if (*offset >= bitmap_info->offset && *offset + *bytes <= end) { | |
34d52cb6 | 1521 | bitmap_clear_bits(ctl, bitmap_info, *offset, *bytes); |
96303081 JB |
1522 | *bytes = 0; |
1523 | } | |
1524 | ||
1525 | if (*bytes) { | |
6606bb97 | 1526 | struct rb_node *next = rb_next(&bitmap_info->offset_index); |
edf6e2d1 | 1527 | if (!bitmap_info->bytes) |
34d52cb6 | 1528 | free_bitmap(ctl, bitmap_info); |
96303081 | 1529 | |
6606bb97 JB |
1530 | /* |
1531 | * no entry after this bitmap, but we still have bytes to | |
1532 | * remove, so something has gone wrong. | |
1533 | */ | |
1534 | if (!next) | |
96303081 JB |
1535 | return -EINVAL; |
1536 | ||
6606bb97 JB |
1537 | bitmap_info = rb_entry(next, struct btrfs_free_space, |
1538 | offset_index); | |
1539 | ||
1540 | /* | |
1541 | * if the next entry isn't a bitmap we need to return to let the | |
1542 | * extent stuff do its work. | |
1543 | */ | |
96303081 JB |
1544 | if (!bitmap_info->bitmap) |
1545 | return -EAGAIN; | |
1546 | ||
6606bb97 JB |
1547 | /* |
1548 | * Ok the next item is a bitmap, but it may not actually hold | |
1549 | * the information for the rest of this free space stuff, so | |
1550 | * look for it, and if we don't find it return so we can try | |
1551 | * everything over again. | |
1552 | */ | |
1553 | search_start = *offset; | |
1554 | search_bytes = *bytes; | |
34d52cb6 | 1555 | ret = search_bitmap(ctl, bitmap_info, &search_start, |
6606bb97 JB |
1556 | &search_bytes); |
1557 | if (ret < 0 || search_start != *offset) | |
1558 | return -EAGAIN; | |
1559 | ||
96303081 | 1560 | goto again; |
edf6e2d1 | 1561 | } else if (!bitmap_info->bytes) |
34d52cb6 | 1562 | free_bitmap(ctl, bitmap_info); |
96303081 JB |
1563 | |
1564 | return 0; | |
1565 | } | |
1566 | ||
2cdc342c JB |
1567 | static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl, |
1568 | struct btrfs_free_space *info, u64 offset, | |
1569 | u64 bytes) | |
1570 | { | |
1571 | u64 bytes_to_set = 0; | |
1572 | u64 end; | |
1573 | ||
1574 | end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit); | |
1575 | ||
1576 | bytes_to_set = min(end - offset, bytes); | |
1577 | ||
1578 | bitmap_set_bits(ctl, info, offset, bytes_to_set); | |
1579 | ||
1580 | return bytes_to_set; | |
1581 | ||
1582 | } | |
1583 | ||
34d52cb6 LZ |
1584 | static bool use_bitmap(struct btrfs_free_space_ctl *ctl, |
1585 | struct btrfs_free_space *info) | |
96303081 | 1586 | { |
34d52cb6 | 1587 | struct btrfs_block_group_cache *block_group = ctl->private; |
96303081 JB |
1588 | |
1589 | /* | |
1590 | * If we are below the extents threshold then we can add this as an | |
1591 | * extent, and don't have to deal with the bitmap | |
1592 | */ | |
34d52cb6 | 1593 | if (ctl->free_extents < ctl->extents_thresh) { |
32cb0840 JB |
1594 | /* |
1595 | * If this block group has some small extents we don't want to | |
1596 | * use up all of our free slots in the cache with them, we want | |
1597 | * to reserve them to larger extents, however if we have plent | |
1598 | * of cache left then go ahead an dadd them, no sense in adding | |
1599 | * the overhead of a bitmap if we don't have to. | |
1600 | */ | |
1601 | if (info->bytes <= block_group->sectorsize * 4) { | |
34d52cb6 LZ |
1602 | if (ctl->free_extents * 2 <= ctl->extents_thresh) |
1603 | return false; | |
32cb0840 | 1604 | } else { |
34d52cb6 | 1605 | return false; |
32cb0840 JB |
1606 | } |
1607 | } | |
96303081 JB |
1608 | |
1609 | /* | |
1610 | * some block groups are so tiny they can't be enveloped by a bitmap, so | |
1611 | * don't even bother to create a bitmap for this | |
1612 | */ | |
1613 | if (BITS_PER_BITMAP * block_group->sectorsize > | |
1614 | block_group->key.offset) | |
34d52cb6 LZ |
1615 | return false; |
1616 | ||
1617 | return true; | |
1618 | } | |
1619 | ||
2cdc342c JB |
1620 | static struct btrfs_free_space_op free_space_op = { |
1621 | .recalc_thresholds = recalculate_thresholds, | |
1622 | .use_bitmap = use_bitmap, | |
1623 | }; | |
1624 | ||
34d52cb6 LZ |
1625 | static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl, |
1626 | struct btrfs_free_space *info) | |
1627 | { | |
1628 | struct btrfs_free_space *bitmap_info; | |
2cdc342c | 1629 | struct btrfs_block_group_cache *block_group = NULL; |
34d52cb6 | 1630 | int added = 0; |
2cdc342c | 1631 | u64 bytes, offset, bytes_added; |
34d52cb6 | 1632 | int ret; |
96303081 JB |
1633 | |
1634 | bytes = info->bytes; | |
1635 | offset = info->offset; | |
1636 | ||
34d52cb6 LZ |
1637 | if (!ctl->op->use_bitmap(ctl, info)) |
1638 | return 0; | |
1639 | ||
2cdc342c JB |
1640 | if (ctl->op == &free_space_op) |
1641 | block_group = ctl->private; | |
38e87880 | 1642 | again: |
2cdc342c JB |
1643 | /* |
1644 | * Since we link bitmaps right into the cluster we need to see if we | |
1645 | * have a cluster here, and if so and it has our bitmap we need to add | |
1646 | * the free space to that bitmap. | |
1647 | */ | |
1648 | if (block_group && !list_empty(&block_group->cluster_list)) { | |
1649 | struct btrfs_free_cluster *cluster; | |
1650 | struct rb_node *node; | |
1651 | struct btrfs_free_space *entry; | |
1652 | ||
1653 | cluster = list_entry(block_group->cluster_list.next, | |
1654 | struct btrfs_free_cluster, | |
1655 | block_group_list); | |
1656 | spin_lock(&cluster->lock); | |
1657 | node = rb_first(&cluster->root); | |
1658 | if (!node) { | |
1659 | spin_unlock(&cluster->lock); | |
38e87880 | 1660 | goto no_cluster_bitmap; |
2cdc342c JB |
1661 | } |
1662 | ||
1663 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
1664 | if (!entry->bitmap) { | |
1665 | spin_unlock(&cluster->lock); | |
38e87880 | 1666 | goto no_cluster_bitmap; |
2cdc342c JB |
1667 | } |
1668 | ||
1669 | if (entry->offset == offset_to_bitmap(ctl, offset)) { | |
1670 | bytes_added = add_bytes_to_bitmap(ctl, entry, | |
1671 | offset, bytes); | |
1672 | bytes -= bytes_added; | |
1673 | offset += bytes_added; | |
1674 | } | |
1675 | spin_unlock(&cluster->lock); | |
1676 | if (!bytes) { | |
1677 | ret = 1; | |
1678 | goto out; | |
1679 | } | |
1680 | } | |
38e87880 CM |
1681 | |
1682 | no_cluster_bitmap: | |
34d52cb6 | 1683 | bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), |
96303081 JB |
1684 | 1, 0); |
1685 | if (!bitmap_info) { | |
1686 | BUG_ON(added); | |
1687 | goto new_bitmap; | |
1688 | } | |
1689 | ||
2cdc342c JB |
1690 | bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes); |
1691 | bytes -= bytes_added; | |
1692 | offset += bytes_added; | |
1693 | added = 0; | |
96303081 JB |
1694 | |
1695 | if (!bytes) { | |
1696 | ret = 1; | |
1697 | goto out; | |
1698 | } else | |
1699 | goto again; | |
1700 | ||
1701 | new_bitmap: | |
1702 | if (info && info->bitmap) { | |
34d52cb6 | 1703 | add_new_bitmap(ctl, info, offset); |
96303081 JB |
1704 | added = 1; |
1705 | info = NULL; | |
1706 | goto again; | |
1707 | } else { | |
34d52cb6 | 1708 | spin_unlock(&ctl->tree_lock); |
96303081 JB |
1709 | |
1710 | /* no pre-allocated info, allocate a new one */ | |
1711 | if (!info) { | |
dc89e982 JB |
1712 | info = kmem_cache_zalloc(btrfs_free_space_cachep, |
1713 | GFP_NOFS); | |
96303081 | 1714 | if (!info) { |
34d52cb6 | 1715 | spin_lock(&ctl->tree_lock); |
96303081 JB |
1716 | ret = -ENOMEM; |
1717 | goto out; | |
1718 | } | |
1719 | } | |
1720 | ||
1721 | /* allocate the bitmap */ | |
1722 | info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS); | |
34d52cb6 | 1723 | spin_lock(&ctl->tree_lock); |
96303081 JB |
1724 | if (!info->bitmap) { |
1725 | ret = -ENOMEM; | |
1726 | goto out; | |
1727 | } | |
1728 | goto again; | |
1729 | } | |
1730 | ||
1731 | out: | |
1732 | if (info) { | |
1733 | if (info->bitmap) | |
1734 | kfree(info->bitmap); | |
dc89e982 | 1735 | kmem_cache_free(btrfs_free_space_cachep, info); |
96303081 | 1736 | } |
0f9dd46c JB |
1737 | |
1738 | return ret; | |
1739 | } | |
1740 | ||
945d8962 | 1741 | static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl, |
f333adb5 | 1742 | struct btrfs_free_space *info, bool update_stat) |
0f9dd46c | 1743 | { |
120d66ee LZ |
1744 | struct btrfs_free_space *left_info; |
1745 | struct btrfs_free_space *right_info; | |
1746 | bool merged = false; | |
1747 | u64 offset = info->offset; | |
1748 | u64 bytes = info->bytes; | |
6226cb0a | 1749 | |
0f9dd46c JB |
1750 | /* |
1751 | * first we want to see if there is free space adjacent to the range we | |
1752 | * are adding, if there is remove that struct and add a new one to | |
1753 | * cover the entire range | |
1754 | */ | |
34d52cb6 | 1755 | right_info = tree_search_offset(ctl, offset + bytes, 0, 0); |
96303081 JB |
1756 | if (right_info && rb_prev(&right_info->offset_index)) |
1757 | left_info = rb_entry(rb_prev(&right_info->offset_index), | |
1758 | struct btrfs_free_space, offset_index); | |
1759 | else | |
34d52cb6 | 1760 | left_info = tree_search_offset(ctl, offset - 1, 0, 0); |
0f9dd46c | 1761 | |
96303081 | 1762 | if (right_info && !right_info->bitmap) { |
f333adb5 | 1763 | if (update_stat) |
34d52cb6 | 1764 | unlink_free_space(ctl, right_info); |
f333adb5 | 1765 | else |
34d52cb6 | 1766 | __unlink_free_space(ctl, right_info); |
6226cb0a | 1767 | info->bytes += right_info->bytes; |
dc89e982 | 1768 | kmem_cache_free(btrfs_free_space_cachep, right_info); |
120d66ee | 1769 | merged = true; |
0f9dd46c JB |
1770 | } |
1771 | ||
96303081 JB |
1772 | if (left_info && !left_info->bitmap && |
1773 | left_info->offset + left_info->bytes == offset) { | |
f333adb5 | 1774 | if (update_stat) |
34d52cb6 | 1775 | unlink_free_space(ctl, left_info); |
f333adb5 | 1776 | else |
34d52cb6 | 1777 | __unlink_free_space(ctl, left_info); |
6226cb0a JB |
1778 | info->offset = left_info->offset; |
1779 | info->bytes += left_info->bytes; | |
dc89e982 | 1780 | kmem_cache_free(btrfs_free_space_cachep, left_info); |
120d66ee | 1781 | merged = true; |
0f9dd46c JB |
1782 | } |
1783 | ||
120d66ee LZ |
1784 | return merged; |
1785 | } | |
1786 | ||
581bb050 LZ |
1787 | int __btrfs_add_free_space(struct btrfs_free_space_ctl *ctl, |
1788 | u64 offset, u64 bytes) | |
120d66ee LZ |
1789 | { |
1790 | struct btrfs_free_space *info; | |
1791 | int ret = 0; | |
1792 | ||
dc89e982 | 1793 | info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS); |
120d66ee LZ |
1794 | if (!info) |
1795 | return -ENOMEM; | |
1796 | ||
1797 | info->offset = offset; | |
1798 | info->bytes = bytes; | |
1799 | ||
34d52cb6 | 1800 | spin_lock(&ctl->tree_lock); |
120d66ee | 1801 | |
34d52cb6 | 1802 | if (try_merge_free_space(ctl, info, true)) |
120d66ee LZ |
1803 | goto link; |
1804 | ||
1805 | /* | |
1806 | * There was no extent directly to the left or right of this new | |
1807 | * extent then we know we're going to have to allocate a new extent, so | |
1808 | * before we do that see if we need to drop this into a bitmap | |
1809 | */ | |
34d52cb6 | 1810 | ret = insert_into_bitmap(ctl, info); |
120d66ee LZ |
1811 | if (ret < 0) { |
1812 | goto out; | |
1813 | } else if (ret) { | |
1814 | ret = 0; | |
1815 | goto out; | |
1816 | } | |
1817 | link: | |
34d52cb6 | 1818 | ret = link_free_space(ctl, info); |
0f9dd46c | 1819 | if (ret) |
dc89e982 | 1820 | kmem_cache_free(btrfs_free_space_cachep, info); |
96303081 | 1821 | out: |
34d52cb6 | 1822 | spin_unlock(&ctl->tree_lock); |
6226cb0a | 1823 | |
0f9dd46c | 1824 | if (ret) { |
96303081 | 1825 | printk(KERN_CRIT "btrfs: unable to add free space :%d\n", ret); |
c293498b | 1826 | BUG_ON(ret == -EEXIST); |
0f9dd46c JB |
1827 | } |
1828 | ||
0f9dd46c JB |
1829 | return ret; |
1830 | } | |
1831 | ||
6226cb0a JB |
1832 | int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group, |
1833 | u64 offset, u64 bytes) | |
0f9dd46c | 1834 | { |
34d52cb6 | 1835 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0f9dd46c | 1836 | struct btrfs_free_space *info; |
96303081 | 1837 | struct btrfs_free_space *next_info = NULL; |
0f9dd46c JB |
1838 | int ret = 0; |
1839 | ||
34d52cb6 | 1840 | spin_lock(&ctl->tree_lock); |
6226cb0a | 1841 | |
96303081 | 1842 | again: |
34d52cb6 | 1843 | info = tree_search_offset(ctl, offset, 0, 0); |
96303081 | 1844 | if (!info) { |
6606bb97 JB |
1845 | /* |
1846 | * oops didn't find an extent that matched the space we wanted | |
1847 | * to remove, look for a bitmap instead | |
1848 | */ | |
34d52cb6 | 1849 | info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), |
6606bb97 JB |
1850 | 1, 0); |
1851 | if (!info) { | |
1852 | WARN_ON(1); | |
1853 | goto out_lock; | |
1854 | } | |
96303081 JB |
1855 | } |
1856 | ||
1857 | if (info->bytes < bytes && rb_next(&info->offset_index)) { | |
1858 | u64 end; | |
1859 | next_info = rb_entry(rb_next(&info->offset_index), | |
1860 | struct btrfs_free_space, | |
1861 | offset_index); | |
1862 | ||
1863 | if (next_info->bitmap) | |
34d52cb6 LZ |
1864 | end = next_info->offset + |
1865 | BITS_PER_BITMAP * ctl->unit - 1; | |
96303081 JB |
1866 | else |
1867 | end = next_info->offset + next_info->bytes; | |
1868 | ||
1869 | if (next_info->bytes < bytes || | |
1870 | next_info->offset > offset || offset > end) { | |
1871 | printk(KERN_CRIT "Found free space at %llu, size %llu," | |
1872 | " trying to use %llu\n", | |
1873 | (unsigned long long)info->offset, | |
1874 | (unsigned long long)info->bytes, | |
1875 | (unsigned long long)bytes); | |
0f9dd46c JB |
1876 | WARN_ON(1); |
1877 | ret = -EINVAL; | |
96303081 | 1878 | goto out_lock; |
0f9dd46c | 1879 | } |
0f9dd46c | 1880 | |
96303081 JB |
1881 | info = next_info; |
1882 | } | |
1883 | ||
1884 | if (info->bytes == bytes) { | |
34d52cb6 | 1885 | unlink_free_space(ctl, info); |
96303081 JB |
1886 | if (info->bitmap) { |
1887 | kfree(info->bitmap); | |
34d52cb6 | 1888 | ctl->total_bitmaps--; |
0f9dd46c | 1889 | } |
dc89e982 | 1890 | kmem_cache_free(btrfs_free_space_cachep, info); |
1eae31e9 | 1891 | ret = 0; |
96303081 JB |
1892 | goto out_lock; |
1893 | } | |
0f9dd46c | 1894 | |
96303081 | 1895 | if (!info->bitmap && info->offset == offset) { |
34d52cb6 | 1896 | unlink_free_space(ctl, info); |
0f9dd46c JB |
1897 | info->offset += bytes; |
1898 | info->bytes -= bytes; | |
1eae31e9 CM |
1899 | ret = link_free_space(ctl, info); |
1900 | WARN_ON(ret); | |
96303081 JB |
1901 | goto out_lock; |
1902 | } | |
0f9dd46c | 1903 | |
96303081 JB |
1904 | if (!info->bitmap && info->offset <= offset && |
1905 | info->offset + info->bytes >= offset + bytes) { | |
9b49c9b9 CM |
1906 | u64 old_start = info->offset; |
1907 | /* | |
1908 | * we're freeing space in the middle of the info, | |
1909 | * this can happen during tree log replay | |
1910 | * | |
1911 | * first unlink the old info and then | |
1912 | * insert it again after the hole we're creating | |
1913 | */ | |
34d52cb6 | 1914 | unlink_free_space(ctl, info); |
9b49c9b9 CM |
1915 | if (offset + bytes < info->offset + info->bytes) { |
1916 | u64 old_end = info->offset + info->bytes; | |
1917 | ||
1918 | info->offset = offset + bytes; | |
1919 | info->bytes = old_end - info->offset; | |
34d52cb6 | 1920 | ret = link_free_space(ctl, info); |
96303081 JB |
1921 | WARN_ON(ret); |
1922 | if (ret) | |
1923 | goto out_lock; | |
9b49c9b9 CM |
1924 | } else { |
1925 | /* the hole we're creating ends at the end | |
1926 | * of the info struct, just free the info | |
1927 | */ | |
dc89e982 | 1928 | kmem_cache_free(btrfs_free_space_cachep, info); |
9b49c9b9 | 1929 | } |
34d52cb6 | 1930 | spin_unlock(&ctl->tree_lock); |
96303081 JB |
1931 | |
1932 | /* step two, insert a new info struct to cover | |
1933 | * anything before the hole | |
9b49c9b9 | 1934 | */ |
6226cb0a JB |
1935 | ret = btrfs_add_free_space(block_group, old_start, |
1936 | offset - old_start); | |
96303081 JB |
1937 | WARN_ON(ret); |
1938 | goto out; | |
0f9dd46c | 1939 | } |
96303081 | 1940 | |
34d52cb6 | 1941 | ret = remove_from_bitmap(ctl, info, &offset, &bytes); |
96303081 JB |
1942 | if (ret == -EAGAIN) |
1943 | goto again; | |
1944 | BUG_ON(ret); | |
1945 | out_lock: | |
34d52cb6 | 1946 | spin_unlock(&ctl->tree_lock); |
0f9dd46c | 1947 | out: |
25179201 JB |
1948 | return ret; |
1949 | } | |
1950 | ||
0f9dd46c JB |
1951 | void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group, |
1952 | u64 bytes) | |
1953 | { | |
34d52cb6 | 1954 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0f9dd46c JB |
1955 | struct btrfs_free_space *info; |
1956 | struct rb_node *n; | |
1957 | int count = 0; | |
1958 | ||
34d52cb6 | 1959 | for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) { |
0f9dd46c JB |
1960 | info = rb_entry(n, struct btrfs_free_space, offset_index); |
1961 | if (info->bytes >= bytes) | |
1962 | count++; | |
96303081 | 1963 | printk(KERN_CRIT "entry offset %llu, bytes %llu, bitmap %s\n", |
21380931 | 1964 | (unsigned long long)info->offset, |
96303081 JB |
1965 | (unsigned long long)info->bytes, |
1966 | (info->bitmap) ? "yes" : "no"); | |
0f9dd46c | 1967 | } |
96303081 JB |
1968 | printk(KERN_INFO "block group has cluster?: %s\n", |
1969 | list_empty(&block_group->cluster_list) ? "no" : "yes"); | |
0f9dd46c JB |
1970 | printk(KERN_INFO "%d blocks of free space at or bigger than bytes is" |
1971 | "\n", count); | |
1972 | } | |
1973 | ||
34d52cb6 | 1974 | void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group) |
0f9dd46c | 1975 | { |
34d52cb6 | 1976 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0f9dd46c | 1977 | |
34d52cb6 LZ |
1978 | spin_lock_init(&ctl->tree_lock); |
1979 | ctl->unit = block_group->sectorsize; | |
1980 | ctl->start = block_group->key.objectid; | |
1981 | ctl->private = block_group; | |
1982 | ctl->op = &free_space_op; | |
0f9dd46c | 1983 | |
34d52cb6 LZ |
1984 | /* |
1985 | * we only want to have 32k of ram per block group for keeping | |
1986 | * track of free space, and if we pass 1/2 of that we want to | |
1987 | * start converting things over to using bitmaps | |
1988 | */ | |
1989 | ctl->extents_thresh = ((1024 * 32) / 2) / | |
1990 | sizeof(struct btrfs_free_space); | |
0f9dd46c JB |
1991 | } |
1992 | ||
fa9c0d79 CM |
1993 | /* |
1994 | * for a given cluster, put all of its extents back into the free | |
1995 | * space cache. If the block group passed doesn't match the block group | |
1996 | * pointed to by the cluster, someone else raced in and freed the | |
1997 | * cluster already. In that case, we just return without changing anything | |
1998 | */ | |
1999 | static int | |
2000 | __btrfs_return_cluster_to_free_space( | |
2001 | struct btrfs_block_group_cache *block_group, | |
2002 | struct btrfs_free_cluster *cluster) | |
2003 | { | |
34d52cb6 | 2004 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
fa9c0d79 CM |
2005 | struct btrfs_free_space *entry; |
2006 | struct rb_node *node; | |
2007 | ||
2008 | spin_lock(&cluster->lock); | |
2009 | if (cluster->block_group != block_group) | |
2010 | goto out; | |
2011 | ||
96303081 | 2012 | cluster->block_group = NULL; |
fa9c0d79 | 2013 | cluster->window_start = 0; |
96303081 | 2014 | list_del_init(&cluster->block_group_list); |
96303081 | 2015 | |
fa9c0d79 | 2016 | node = rb_first(&cluster->root); |
96303081 | 2017 | while (node) { |
4e69b598 JB |
2018 | bool bitmap; |
2019 | ||
fa9c0d79 CM |
2020 | entry = rb_entry(node, struct btrfs_free_space, offset_index); |
2021 | node = rb_next(&entry->offset_index); | |
2022 | rb_erase(&entry->offset_index, &cluster->root); | |
4e69b598 JB |
2023 | |
2024 | bitmap = (entry->bitmap != NULL); | |
2025 | if (!bitmap) | |
34d52cb6 LZ |
2026 | try_merge_free_space(ctl, entry, false); |
2027 | tree_insert_offset(&ctl->free_space_offset, | |
4e69b598 | 2028 | entry->offset, &entry->offset_index, bitmap); |
fa9c0d79 | 2029 | } |
6bef4d31 | 2030 | cluster->root = RB_ROOT; |
96303081 | 2031 | |
fa9c0d79 CM |
2032 | out: |
2033 | spin_unlock(&cluster->lock); | |
96303081 | 2034 | btrfs_put_block_group(block_group); |
fa9c0d79 CM |
2035 | return 0; |
2036 | } | |
2037 | ||
09655373 | 2038 | void __btrfs_remove_free_space_cache_locked(struct btrfs_free_space_ctl *ctl) |
0f9dd46c JB |
2039 | { |
2040 | struct btrfs_free_space *info; | |
2041 | struct rb_node *node; | |
581bb050 | 2042 | |
581bb050 LZ |
2043 | while ((node = rb_last(&ctl->free_space_offset)) != NULL) { |
2044 | info = rb_entry(node, struct btrfs_free_space, offset_index); | |
9b90f513 JB |
2045 | if (!info->bitmap) { |
2046 | unlink_free_space(ctl, info); | |
2047 | kmem_cache_free(btrfs_free_space_cachep, info); | |
2048 | } else { | |
2049 | free_bitmap(ctl, info); | |
2050 | } | |
581bb050 LZ |
2051 | if (need_resched()) { |
2052 | spin_unlock(&ctl->tree_lock); | |
2053 | cond_resched(); | |
2054 | spin_lock(&ctl->tree_lock); | |
2055 | } | |
2056 | } | |
09655373 CM |
2057 | } |
2058 | ||
2059 | void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl) | |
2060 | { | |
2061 | spin_lock(&ctl->tree_lock); | |
2062 | __btrfs_remove_free_space_cache_locked(ctl); | |
581bb050 LZ |
2063 | spin_unlock(&ctl->tree_lock); |
2064 | } | |
2065 | ||
2066 | void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group) | |
2067 | { | |
2068 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
fa9c0d79 | 2069 | struct btrfs_free_cluster *cluster; |
96303081 | 2070 | struct list_head *head; |
0f9dd46c | 2071 | |
34d52cb6 | 2072 | spin_lock(&ctl->tree_lock); |
96303081 JB |
2073 | while ((head = block_group->cluster_list.next) != |
2074 | &block_group->cluster_list) { | |
2075 | cluster = list_entry(head, struct btrfs_free_cluster, | |
2076 | block_group_list); | |
fa9c0d79 CM |
2077 | |
2078 | WARN_ON(cluster->block_group != block_group); | |
2079 | __btrfs_return_cluster_to_free_space(block_group, cluster); | |
96303081 | 2080 | if (need_resched()) { |
34d52cb6 | 2081 | spin_unlock(&ctl->tree_lock); |
96303081 | 2082 | cond_resched(); |
34d52cb6 | 2083 | spin_lock(&ctl->tree_lock); |
96303081 | 2084 | } |
fa9c0d79 | 2085 | } |
09655373 | 2086 | __btrfs_remove_free_space_cache_locked(ctl); |
34d52cb6 | 2087 | spin_unlock(&ctl->tree_lock); |
fa9c0d79 | 2088 | |
0f9dd46c JB |
2089 | } |
2090 | ||
6226cb0a JB |
2091 | u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group, |
2092 | u64 offset, u64 bytes, u64 empty_size) | |
0f9dd46c | 2093 | { |
34d52cb6 | 2094 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
6226cb0a | 2095 | struct btrfs_free_space *entry = NULL; |
96303081 | 2096 | u64 bytes_search = bytes + empty_size; |
6226cb0a | 2097 | u64 ret = 0; |
0f9dd46c | 2098 | |
34d52cb6 LZ |
2099 | spin_lock(&ctl->tree_lock); |
2100 | entry = find_free_space(ctl, &offset, &bytes_search); | |
6226cb0a | 2101 | if (!entry) |
96303081 JB |
2102 | goto out; |
2103 | ||
2104 | ret = offset; | |
2105 | if (entry->bitmap) { | |
34d52cb6 | 2106 | bitmap_clear_bits(ctl, entry, offset, bytes); |
edf6e2d1 | 2107 | if (!entry->bytes) |
34d52cb6 | 2108 | free_bitmap(ctl, entry); |
96303081 | 2109 | } else { |
34d52cb6 | 2110 | unlink_free_space(ctl, entry); |
6226cb0a JB |
2111 | entry->offset += bytes; |
2112 | entry->bytes -= bytes; | |
6226cb0a | 2113 | if (!entry->bytes) |
dc89e982 | 2114 | kmem_cache_free(btrfs_free_space_cachep, entry); |
6226cb0a | 2115 | else |
34d52cb6 | 2116 | link_free_space(ctl, entry); |
6226cb0a | 2117 | } |
0f9dd46c | 2118 | |
96303081 | 2119 | out: |
34d52cb6 | 2120 | spin_unlock(&ctl->tree_lock); |
817d52f8 | 2121 | |
0f9dd46c JB |
2122 | return ret; |
2123 | } | |
fa9c0d79 CM |
2124 | |
2125 | /* | |
2126 | * given a cluster, put all of its extents back into the free space | |
2127 | * cache. If a block group is passed, this function will only free | |
2128 | * a cluster that belongs to the passed block group. | |
2129 | * | |
2130 | * Otherwise, it'll get a reference on the block group pointed to by the | |
2131 | * cluster and remove the cluster from it. | |
2132 | */ | |
2133 | int btrfs_return_cluster_to_free_space( | |
2134 | struct btrfs_block_group_cache *block_group, | |
2135 | struct btrfs_free_cluster *cluster) | |
2136 | { | |
34d52cb6 | 2137 | struct btrfs_free_space_ctl *ctl; |
fa9c0d79 CM |
2138 | int ret; |
2139 | ||
2140 | /* first, get a safe pointer to the block group */ | |
2141 | spin_lock(&cluster->lock); | |
2142 | if (!block_group) { | |
2143 | block_group = cluster->block_group; | |
2144 | if (!block_group) { | |
2145 | spin_unlock(&cluster->lock); | |
2146 | return 0; | |
2147 | } | |
2148 | } else if (cluster->block_group != block_group) { | |
2149 | /* someone else has already freed it don't redo their work */ | |
2150 | spin_unlock(&cluster->lock); | |
2151 | return 0; | |
2152 | } | |
2153 | atomic_inc(&block_group->count); | |
2154 | spin_unlock(&cluster->lock); | |
2155 | ||
34d52cb6 LZ |
2156 | ctl = block_group->free_space_ctl; |
2157 | ||
fa9c0d79 | 2158 | /* now return any extents the cluster had on it */ |
34d52cb6 | 2159 | spin_lock(&ctl->tree_lock); |
fa9c0d79 | 2160 | ret = __btrfs_return_cluster_to_free_space(block_group, cluster); |
34d52cb6 | 2161 | spin_unlock(&ctl->tree_lock); |
fa9c0d79 CM |
2162 | |
2163 | /* finally drop our ref */ | |
2164 | btrfs_put_block_group(block_group); | |
2165 | return ret; | |
2166 | } | |
2167 | ||
96303081 JB |
2168 | static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group, |
2169 | struct btrfs_free_cluster *cluster, | |
4e69b598 | 2170 | struct btrfs_free_space *entry, |
96303081 JB |
2171 | u64 bytes, u64 min_start) |
2172 | { | |
34d52cb6 | 2173 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
96303081 JB |
2174 | int err; |
2175 | u64 search_start = cluster->window_start; | |
2176 | u64 search_bytes = bytes; | |
2177 | u64 ret = 0; | |
2178 | ||
96303081 JB |
2179 | search_start = min_start; |
2180 | search_bytes = bytes; | |
2181 | ||
34d52cb6 | 2182 | err = search_bitmap(ctl, entry, &search_start, &search_bytes); |
96303081 | 2183 | if (err) |
4e69b598 | 2184 | return 0; |
96303081 JB |
2185 | |
2186 | ret = search_start; | |
bb3ac5a4 | 2187 | __bitmap_clear_bits(ctl, entry, ret, bytes); |
96303081 JB |
2188 | |
2189 | return ret; | |
2190 | } | |
2191 | ||
fa9c0d79 CM |
2192 | /* |
2193 | * given a cluster, try to allocate 'bytes' from it, returns 0 | |
2194 | * if it couldn't find anything suitably large, or a logical disk offset | |
2195 | * if things worked out | |
2196 | */ | |
2197 | u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group, | |
2198 | struct btrfs_free_cluster *cluster, u64 bytes, | |
2199 | u64 min_start) | |
2200 | { | |
34d52cb6 | 2201 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
fa9c0d79 CM |
2202 | struct btrfs_free_space *entry = NULL; |
2203 | struct rb_node *node; | |
2204 | u64 ret = 0; | |
2205 | ||
2206 | spin_lock(&cluster->lock); | |
2207 | if (bytes > cluster->max_size) | |
2208 | goto out; | |
2209 | ||
2210 | if (cluster->block_group != block_group) | |
2211 | goto out; | |
2212 | ||
2213 | node = rb_first(&cluster->root); | |
2214 | if (!node) | |
2215 | goto out; | |
2216 | ||
2217 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
fa9c0d79 | 2218 | while(1) { |
4e69b598 JB |
2219 | if (entry->bytes < bytes || |
2220 | (!entry->bitmap && entry->offset < min_start)) { | |
fa9c0d79 CM |
2221 | node = rb_next(&entry->offset_index); |
2222 | if (!node) | |
2223 | break; | |
2224 | entry = rb_entry(node, struct btrfs_free_space, | |
2225 | offset_index); | |
2226 | continue; | |
2227 | } | |
fa9c0d79 | 2228 | |
4e69b598 JB |
2229 | if (entry->bitmap) { |
2230 | ret = btrfs_alloc_from_bitmap(block_group, | |
2231 | cluster, entry, bytes, | |
2232 | min_start); | |
2233 | if (ret == 0) { | |
4e69b598 JB |
2234 | node = rb_next(&entry->offset_index); |
2235 | if (!node) | |
2236 | break; | |
2237 | entry = rb_entry(node, struct btrfs_free_space, | |
2238 | offset_index); | |
2239 | continue; | |
2240 | } | |
2241 | } else { | |
4e69b598 JB |
2242 | ret = entry->offset; |
2243 | ||
2244 | entry->offset += bytes; | |
2245 | entry->bytes -= bytes; | |
2246 | } | |
fa9c0d79 | 2247 | |
5e71b5d5 | 2248 | if (entry->bytes == 0) |
fa9c0d79 | 2249 | rb_erase(&entry->offset_index, &cluster->root); |
fa9c0d79 CM |
2250 | break; |
2251 | } | |
2252 | out: | |
2253 | spin_unlock(&cluster->lock); | |
96303081 | 2254 | |
5e71b5d5 LZ |
2255 | if (!ret) |
2256 | return 0; | |
2257 | ||
34d52cb6 | 2258 | spin_lock(&ctl->tree_lock); |
5e71b5d5 | 2259 | |
34d52cb6 | 2260 | ctl->free_space -= bytes; |
5e71b5d5 | 2261 | if (entry->bytes == 0) { |
34d52cb6 | 2262 | ctl->free_extents--; |
4e69b598 JB |
2263 | if (entry->bitmap) { |
2264 | kfree(entry->bitmap); | |
34d52cb6 LZ |
2265 | ctl->total_bitmaps--; |
2266 | ctl->op->recalc_thresholds(ctl); | |
4e69b598 | 2267 | } |
dc89e982 | 2268 | kmem_cache_free(btrfs_free_space_cachep, entry); |
5e71b5d5 LZ |
2269 | } |
2270 | ||
34d52cb6 | 2271 | spin_unlock(&ctl->tree_lock); |
5e71b5d5 | 2272 | |
fa9c0d79 CM |
2273 | return ret; |
2274 | } | |
2275 | ||
96303081 JB |
2276 | static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group, |
2277 | struct btrfs_free_space *entry, | |
2278 | struct btrfs_free_cluster *cluster, | |
2279 | u64 offset, u64 bytes, u64 min_bytes) | |
2280 | { | |
34d52cb6 | 2281 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
96303081 JB |
2282 | unsigned long next_zero; |
2283 | unsigned long i; | |
2284 | unsigned long search_bits; | |
2285 | unsigned long total_bits; | |
2286 | unsigned long found_bits; | |
2287 | unsigned long start = 0; | |
2288 | unsigned long total_found = 0; | |
4e69b598 | 2289 | int ret; |
96303081 JB |
2290 | bool found = false; |
2291 | ||
2292 | i = offset_to_bit(entry->offset, block_group->sectorsize, | |
2293 | max_t(u64, offset, entry->offset)); | |
d0a365e8 JB |
2294 | search_bits = bytes_to_bits(bytes, block_group->sectorsize); |
2295 | total_bits = bytes_to_bits(min_bytes, block_group->sectorsize); | |
96303081 JB |
2296 | |
2297 | again: | |
2298 | found_bits = 0; | |
2299 | for (i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, i); | |
2300 | i < BITS_PER_BITMAP; | |
2301 | i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, i + 1)) { | |
2302 | next_zero = find_next_zero_bit(entry->bitmap, | |
2303 | BITS_PER_BITMAP, i); | |
2304 | if (next_zero - i >= search_bits) { | |
2305 | found_bits = next_zero - i; | |
2306 | break; | |
2307 | } | |
2308 | i = next_zero; | |
2309 | } | |
2310 | ||
2311 | if (!found_bits) | |
4e69b598 | 2312 | return -ENOSPC; |
96303081 JB |
2313 | |
2314 | if (!found) { | |
2315 | start = i; | |
2316 | found = true; | |
2317 | } | |
2318 | ||
2319 | total_found += found_bits; | |
2320 | ||
2321 | if (cluster->max_size < found_bits * block_group->sectorsize) | |
2322 | cluster->max_size = found_bits * block_group->sectorsize; | |
2323 | ||
2324 | if (total_found < total_bits) { | |
2325 | i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, next_zero); | |
2326 | if (i - start > total_bits * 2) { | |
2327 | total_found = 0; | |
2328 | cluster->max_size = 0; | |
2329 | found = false; | |
2330 | } | |
2331 | goto again; | |
2332 | } | |
2333 | ||
2334 | cluster->window_start = start * block_group->sectorsize + | |
2335 | entry->offset; | |
34d52cb6 | 2336 | rb_erase(&entry->offset_index, &ctl->free_space_offset); |
4e69b598 JB |
2337 | ret = tree_insert_offset(&cluster->root, entry->offset, |
2338 | &entry->offset_index, 1); | |
2339 | BUG_ON(ret); | |
96303081 JB |
2340 | |
2341 | return 0; | |
2342 | } | |
2343 | ||
4e69b598 JB |
2344 | /* |
2345 | * This searches the block group for just extents to fill the cluster with. | |
2346 | */ | |
3de85bb9 JB |
2347 | static noinline int |
2348 | setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group, | |
2349 | struct btrfs_free_cluster *cluster, | |
2350 | struct list_head *bitmaps, u64 offset, u64 bytes, | |
2351 | u64 min_bytes) | |
4e69b598 | 2352 | { |
34d52cb6 | 2353 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
4e69b598 JB |
2354 | struct btrfs_free_space *first = NULL; |
2355 | struct btrfs_free_space *entry = NULL; | |
2356 | struct btrfs_free_space *prev = NULL; | |
2357 | struct btrfs_free_space *last; | |
2358 | struct rb_node *node; | |
2359 | u64 window_start; | |
2360 | u64 window_free; | |
2361 | u64 max_extent; | |
2362 | u64 max_gap = 128 * 1024; | |
2363 | ||
34d52cb6 | 2364 | entry = tree_search_offset(ctl, offset, 0, 1); |
4e69b598 JB |
2365 | if (!entry) |
2366 | return -ENOSPC; | |
2367 | ||
2368 | /* | |
2369 | * We don't want bitmaps, so just move along until we find a normal | |
2370 | * extent entry. | |
2371 | */ | |
2372 | while (entry->bitmap) { | |
86d4a77b JB |
2373 | if (list_empty(&entry->list)) |
2374 | list_add_tail(&entry->list, bitmaps); | |
4e69b598 JB |
2375 | node = rb_next(&entry->offset_index); |
2376 | if (!node) | |
2377 | return -ENOSPC; | |
2378 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
2379 | } | |
2380 | ||
2381 | window_start = entry->offset; | |
2382 | window_free = entry->bytes; | |
2383 | max_extent = entry->bytes; | |
2384 | first = entry; | |
2385 | last = entry; | |
2386 | prev = entry; | |
2387 | ||
2388 | while (window_free <= min_bytes) { | |
2389 | node = rb_next(&entry->offset_index); | |
2390 | if (!node) | |
2391 | return -ENOSPC; | |
2392 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
2393 | ||
86d4a77b JB |
2394 | if (entry->bitmap) { |
2395 | if (list_empty(&entry->list)) | |
2396 | list_add_tail(&entry->list, bitmaps); | |
4e69b598 | 2397 | continue; |
86d4a77b JB |
2398 | } |
2399 | ||
4e69b598 JB |
2400 | /* |
2401 | * we haven't filled the empty size and the window is | |
2402 | * very large. reset and try again | |
2403 | */ | |
2404 | if (entry->offset - (prev->offset + prev->bytes) > max_gap || | |
2405 | entry->offset - window_start > (min_bytes * 2)) { | |
2406 | first = entry; | |
2407 | window_start = entry->offset; | |
2408 | window_free = entry->bytes; | |
2409 | last = entry; | |
2410 | max_extent = entry->bytes; | |
2411 | } else { | |
2412 | last = entry; | |
2413 | window_free += entry->bytes; | |
2414 | if (entry->bytes > max_extent) | |
2415 | max_extent = entry->bytes; | |
2416 | } | |
2417 | prev = entry; | |
2418 | } | |
2419 | ||
2420 | cluster->window_start = first->offset; | |
2421 | ||
2422 | node = &first->offset_index; | |
2423 | ||
2424 | /* | |
2425 | * now we've found our entries, pull them out of the free space | |
2426 | * cache and put them into the cluster rbtree | |
2427 | */ | |
2428 | do { | |
2429 | int ret; | |
2430 | ||
2431 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
2432 | node = rb_next(&entry->offset_index); | |
2433 | if (entry->bitmap) | |
2434 | continue; | |
2435 | ||
34d52cb6 | 2436 | rb_erase(&entry->offset_index, &ctl->free_space_offset); |
4e69b598 JB |
2437 | ret = tree_insert_offset(&cluster->root, entry->offset, |
2438 | &entry->offset_index, 0); | |
2439 | BUG_ON(ret); | |
2440 | } while (node && entry != last); | |
2441 | ||
2442 | cluster->max_size = max_extent; | |
2443 | ||
2444 | return 0; | |
2445 | } | |
2446 | ||
2447 | /* | |
2448 | * This specifically looks for bitmaps that may work in the cluster, we assume | |
2449 | * that we have already failed to find extents that will work. | |
2450 | */ | |
3de85bb9 JB |
2451 | static noinline int |
2452 | setup_cluster_bitmap(struct btrfs_block_group_cache *block_group, | |
2453 | struct btrfs_free_cluster *cluster, | |
2454 | struct list_head *bitmaps, u64 offset, u64 bytes, | |
2455 | u64 min_bytes) | |
4e69b598 | 2456 | { |
34d52cb6 | 2457 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
4e69b598 | 2458 | struct btrfs_free_space *entry; |
4e69b598 | 2459 | int ret = -ENOSPC; |
0f0fbf1d | 2460 | u64 bitmap_offset = offset_to_bitmap(ctl, offset); |
4e69b598 | 2461 | |
34d52cb6 | 2462 | if (ctl->total_bitmaps == 0) |
4e69b598 JB |
2463 | return -ENOSPC; |
2464 | ||
0f0fbf1d LZ |
2465 | /* |
2466 | * The bitmap that covers offset won't be in the list unless offset | |
2467 | * is just its start offset. | |
2468 | */ | |
2469 | entry = list_first_entry(bitmaps, struct btrfs_free_space, list); | |
2470 | if (entry->offset != bitmap_offset) { | |
2471 | entry = tree_search_offset(ctl, bitmap_offset, 1, 0); | |
2472 | if (entry && list_empty(&entry->list)) | |
2473 | list_add(&entry->list, bitmaps); | |
2474 | } | |
2475 | ||
86d4a77b JB |
2476 | list_for_each_entry(entry, bitmaps, list) { |
2477 | if (entry->bytes < min_bytes) | |
2478 | continue; | |
2479 | ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset, | |
2480 | bytes, min_bytes); | |
2481 | if (!ret) | |
2482 | return 0; | |
2483 | } | |
2484 | ||
2485 | /* | |
52621cb6 LZ |
2486 | * The bitmaps list has all the bitmaps that record free space |
2487 | * starting after offset, so no more search is required. | |
86d4a77b | 2488 | */ |
52621cb6 | 2489 | return -ENOSPC; |
4e69b598 JB |
2490 | } |
2491 | ||
fa9c0d79 CM |
2492 | /* |
2493 | * here we try to find a cluster of blocks in a block group. The goal | |
2494 | * is to find at least bytes free and up to empty_size + bytes free. | |
2495 | * We might not find them all in one contiguous area. | |
2496 | * | |
2497 | * returns zero and sets up cluster if things worked out, otherwise | |
2498 | * it returns -enospc | |
2499 | */ | |
2500 | int btrfs_find_space_cluster(struct btrfs_trans_handle *trans, | |
451d7585 | 2501 | struct btrfs_root *root, |
fa9c0d79 CM |
2502 | struct btrfs_block_group_cache *block_group, |
2503 | struct btrfs_free_cluster *cluster, | |
2504 | u64 offset, u64 bytes, u64 empty_size) | |
2505 | { | |
34d52cb6 | 2506 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
86d4a77b | 2507 | struct btrfs_free_space *entry, *tmp; |
52621cb6 | 2508 | LIST_HEAD(bitmaps); |
fa9c0d79 | 2509 | u64 min_bytes; |
fa9c0d79 CM |
2510 | int ret; |
2511 | ||
2512 | /* for metadata, allow allocates with more holes */ | |
451d7585 CM |
2513 | if (btrfs_test_opt(root, SSD_SPREAD)) { |
2514 | min_bytes = bytes + empty_size; | |
2515 | } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) { | |
fa9c0d79 CM |
2516 | /* |
2517 | * we want to do larger allocations when we are | |
2518 | * flushing out the delayed refs, it helps prevent | |
2519 | * making more work as we go along. | |
2520 | */ | |
2521 | if (trans->transaction->delayed_refs.flushing) | |
2522 | min_bytes = max(bytes, (bytes + empty_size) >> 1); | |
2523 | else | |
2524 | min_bytes = max(bytes, (bytes + empty_size) >> 4); | |
2525 | } else | |
2526 | min_bytes = max(bytes, (bytes + empty_size) >> 2); | |
2527 | ||
34d52cb6 | 2528 | spin_lock(&ctl->tree_lock); |
7d0d2e8e JB |
2529 | |
2530 | /* | |
2531 | * If we know we don't have enough space to make a cluster don't even | |
2532 | * bother doing all the work to try and find one. | |
2533 | */ | |
34d52cb6 LZ |
2534 | if (ctl->free_space < min_bytes) { |
2535 | spin_unlock(&ctl->tree_lock); | |
7d0d2e8e JB |
2536 | return -ENOSPC; |
2537 | } | |
2538 | ||
fa9c0d79 CM |
2539 | spin_lock(&cluster->lock); |
2540 | ||
2541 | /* someone already found a cluster, hooray */ | |
2542 | if (cluster->block_group) { | |
2543 | ret = 0; | |
2544 | goto out; | |
2545 | } | |
fa9c0d79 | 2546 | |
86d4a77b JB |
2547 | ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset, |
2548 | bytes, min_bytes); | |
4e69b598 | 2549 | if (ret) |
86d4a77b JB |
2550 | ret = setup_cluster_bitmap(block_group, cluster, &bitmaps, |
2551 | offset, bytes, min_bytes); | |
2552 | ||
2553 | /* Clear our temporary list */ | |
2554 | list_for_each_entry_safe(entry, tmp, &bitmaps, list) | |
2555 | list_del_init(&entry->list); | |
fa9c0d79 | 2556 | |
4e69b598 JB |
2557 | if (!ret) { |
2558 | atomic_inc(&block_group->count); | |
2559 | list_add_tail(&cluster->block_group_list, | |
2560 | &block_group->cluster_list); | |
2561 | cluster->block_group = block_group; | |
fa9c0d79 | 2562 | } |
fa9c0d79 CM |
2563 | out: |
2564 | spin_unlock(&cluster->lock); | |
34d52cb6 | 2565 | spin_unlock(&ctl->tree_lock); |
fa9c0d79 CM |
2566 | |
2567 | return ret; | |
2568 | } | |
2569 | ||
2570 | /* | |
2571 | * simple code to zero out a cluster | |
2572 | */ | |
2573 | void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster) | |
2574 | { | |
2575 | spin_lock_init(&cluster->lock); | |
2576 | spin_lock_init(&cluster->refill_lock); | |
6bef4d31 | 2577 | cluster->root = RB_ROOT; |
fa9c0d79 CM |
2578 | cluster->max_size = 0; |
2579 | INIT_LIST_HEAD(&cluster->block_group_list); | |
2580 | cluster->block_group = NULL; | |
2581 | } | |
2582 | ||
f7039b1d LD |
2583 | int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group, |
2584 | u64 *trimmed, u64 start, u64 end, u64 minlen) | |
2585 | { | |
34d52cb6 | 2586 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
f7039b1d LD |
2587 | struct btrfs_free_space *entry = NULL; |
2588 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
2589 | u64 bytes = 0; | |
2590 | u64 actually_trimmed; | |
2591 | int ret = 0; | |
2592 | ||
2593 | *trimmed = 0; | |
2594 | ||
2595 | while (start < end) { | |
34d52cb6 | 2596 | spin_lock(&ctl->tree_lock); |
f7039b1d | 2597 | |
34d52cb6 LZ |
2598 | if (ctl->free_space < minlen) { |
2599 | spin_unlock(&ctl->tree_lock); | |
f7039b1d LD |
2600 | break; |
2601 | } | |
2602 | ||
34d52cb6 | 2603 | entry = tree_search_offset(ctl, start, 0, 1); |
f7039b1d | 2604 | if (!entry) |
34d52cb6 LZ |
2605 | entry = tree_search_offset(ctl, |
2606 | offset_to_bitmap(ctl, start), | |
f7039b1d LD |
2607 | 1, 1); |
2608 | ||
2609 | if (!entry || entry->offset >= end) { | |
34d52cb6 | 2610 | spin_unlock(&ctl->tree_lock); |
f7039b1d LD |
2611 | break; |
2612 | } | |
2613 | ||
2614 | if (entry->bitmap) { | |
34d52cb6 | 2615 | ret = search_bitmap(ctl, entry, &start, &bytes); |
f7039b1d LD |
2616 | if (!ret) { |
2617 | if (start >= end) { | |
34d52cb6 | 2618 | spin_unlock(&ctl->tree_lock); |
f7039b1d LD |
2619 | break; |
2620 | } | |
2621 | bytes = min(bytes, end - start); | |
34d52cb6 | 2622 | bitmap_clear_bits(ctl, entry, start, bytes); |
f7039b1d | 2623 | if (entry->bytes == 0) |
34d52cb6 | 2624 | free_bitmap(ctl, entry); |
f7039b1d LD |
2625 | } else { |
2626 | start = entry->offset + BITS_PER_BITMAP * | |
2627 | block_group->sectorsize; | |
34d52cb6 | 2628 | spin_unlock(&ctl->tree_lock); |
f7039b1d LD |
2629 | ret = 0; |
2630 | continue; | |
2631 | } | |
2632 | } else { | |
2633 | start = entry->offset; | |
2634 | bytes = min(entry->bytes, end - start); | |
34d52cb6 | 2635 | unlink_free_space(ctl, entry); |
f789b684 | 2636 | kmem_cache_free(btrfs_free_space_cachep, entry); |
f7039b1d LD |
2637 | } |
2638 | ||
34d52cb6 | 2639 | spin_unlock(&ctl->tree_lock); |
f7039b1d LD |
2640 | |
2641 | if (bytes >= minlen) { | |
fb25e914 JB |
2642 | struct btrfs_space_info *space_info; |
2643 | int update = 0; | |
2644 | ||
2645 | space_info = block_group->space_info; | |
2646 | spin_lock(&space_info->lock); | |
2647 | spin_lock(&block_group->lock); | |
2648 | if (!block_group->ro) { | |
2649 | block_group->reserved += bytes; | |
2650 | space_info->bytes_reserved += bytes; | |
2651 | update = 1; | |
2652 | } | |
2653 | spin_unlock(&block_group->lock); | |
2654 | spin_unlock(&space_info->lock); | |
f7039b1d LD |
2655 | |
2656 | ret = btrfs_error_discard_extent(fs_info->extent_root, | |
2657 | start, | |
2658 | bytes, | |
2659 | &actually_trimmed); | |
2660 | ||
34d52cb6 | 2661 | btrfs_add_free_space(block_group, start, bytes); |
fb25e914 JB |
2662 | if (update) { |
2663 | spin_lock(&space_info->lock); | |
2664 | spin_lock(&block_group->lock); | |
2665 | if (block_group->ro) | |
2666 | space_info->bytes_readonly += bytes; | |
2667 | block_group->reserved -= bytes; | |
2668 | space_info->bytes_reserved -= bytes; | |
2669 | spin_unlock(&space_info->lock); | |
2670 | spin_unlock(&block_group->lock); | |
2671 | } | |
f7039b1d LD |
2672 | |
2673 | if (ret) | |
2674 | break; | |
2675 | *trimmed += actually_trimmed; | |
2676 | } | |
2677 | start += bytes; | |
2678 | bytes = 0; | |
2679 | ||
2680 | if (fatal_signal_pending(current)) { | |
2681 | ret = -ERESTARTSYS; | |
2682 | break; | |
2683 | } | |
2684 | ||
2685 | cond_resched(); | |
2686 | } | |
2687 | ||
2688 | return ret; | |
2689 | } | |
581bb050 LZ |
2690 | |
2691 | /* | |
2692 | * Find the left-most item in the cache tree, and then return the | |
2693 | * smallest inode number in the item. | |
2694 | * | |
2695 | * Note: the returned inode number may not be the smallest one in | |
2696 | * the tree, if the left-most item is a bitmap. | |
2697 | */ | |
2698 | u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root) | |
2699 | { | |
2700 | struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl; | |
2701 | struct btrfs_free_space *entry = NULL; | |
2702 | u64 ino = 0; | |
2703 | ||
2704 | spin_lock(&ctl->tree_lock); | |
2705 | ||
2706 | if (RB_EMPTY_ROOT(&ctl->free_space_offset)) | |
2707 | goto out; | |
2708 | ||
2709 | entry = rb_entry(rb_first(&ctl->free_space_offset), | |
2710 | struct btrfs_free_space, offset_index); | |
2711 | ||
2712 | if (!entry->bitmap) { | |
2713 | ino = entry->offset; | |
2714 | ||
2715 | unlink_free_space(ctl, entry); | |
2716 | entry->offset++; | |
2717 | entry->bytes--; | |
2718 | if (!entry->bytes) | |
2719 | kmem_cache_free(btrfs_free_space_cachep, entry); | |
2720 | else | |
2721 | link_free_space(ctl, entry); | |
2722 | } else { | |
2723 | u64 offset = 0; | |
2724 | u64 count = 1; | |
2725 | int ret; | |
2726 | ||
2727 | ret = search_bitmap(ctl, entry, &offset, &count); | |
2728 | BUG_ON(ret); | |
2729 | ||
2730 | ino = offset; | |
2731 | bitmap_clear_bits(ctl, entry, offset, 1); | |
2732 | if (entry->bytes == 0) | |
2733 | free_bitmap(ctl, entry); | |
2734 | } | |
2735 | out: | |
2736 | spin_unlock(&ctl->tree_lock); | |
2737 | ||
2738 | return ino; | |
2739 | } | |
82d5902d LZ |
2740 | |
2741 | struct inode *lookup_free_ino_inode(struct btrfs_root *root, | |
2742 | struct btrfs_path *path) | |
2743 | { | |
2744 | struct inode *inode = NULL; | |
2745 | ||
2746 | spin_lock(&root->cache_lock); | |
2747 | if (root->cache_inode) | |
2748 | inode = igrab(root->cache_inode); | |
2749 | spin_unlock(&root->cache_lock); | |
2750 | if (inode) | |
2751 | return inode; | |
2752 | ||
2753 | inode = __lookup_free_space_inode(root, path, 0); | |
2754 | if (IS_ERR(inode)) | |
2755 | return inode; | |
2756 | ||
2757 | spin_lock(&root->cache_lock); | |
7841cb28 | 2758 | if (!btrfs_fs_closing(root->fs_info)) |
82d5902d LZ |
2759 | root->cache_inode = igrab(inode); |
2760 | spin_unlock(&root->cache_lock); | |
2761 | ||
2762 | return inode; | |
2763 | } | |
2764 | ||
2765 | int create_free_ino_inode(struct btrfs_root *root, | |
2766 | struct btrfs_trans_handle *trans, | |
2767 | struct btrfs_path *path) | |
2768 | { | |
2769 | return __create_free_space_inode(root, trans, path, | |
2770 | BTRFS_FREE_INO_OBJECTID, 0); | |
2771 | } | |
2772 | ||
2773 | int load_free_ino_cache(struct btrfs_fs_info *fs_info, struct btrfs_root *root) | |
2774 | { | |
2775 | struct btrfs_free_space_ctl *ctl = root->free_ino_ctl; | |
2776 | struct btrfs_path *path; | |
2777 | struct inode *inode; | |
2778 | int ret = 0; | |
2779 | u64 root_gen = btrfs_root_generation(&root->root_item); | |
2780 | ||
4b9465cb CM |
2781 | if (!btrfs_test_opt(root, INODE_MAP_CACHE)) |
2782 | return 0; | |
2783 | ||
82d5902d LZ |
2784 | /* |
2785 | * If we're unmounting then just return, since this does a search on the | |
2786 | * normal root and not the commit root and we could deadlock. | |
2787 | */ | |
7841cb28 | 2788 | if (btrfs_fs_closing(fs_info)) |
82d5902d LZ |
2789 | return 0; |
2790 | ||
2791 | path = btrfs_alloc_path(); | |
2792 | if (!path) | |
2793 | return 0; | |
2794 | ||
2795 | inode = lookup_free_ino_inode(root, path); | |
2796 | if (IS_ERR(inode)) | |
2797 | goto out; | |
2798 | ||
2799 | if (root_gen != BTRFS_I(inode)->generation) | |
2800 | goto out_put; | |
2801 | ||
2802 | ret = __load_free_space_cache(root, inode, ctl, path, 0); | |
2803 | ||
2804 | if (ret < 0) | |
2805 | printk(KERN_ERR "btrfs: failed to load free ino cache for " | |
2806 | "root %llu\n", root->root_key.objectid); | |
2807 | out_put: | |
2808 | iput(inode); | |
2809 | out: | |
2810 | btrfs_free_path(path); | |
2811 | return ret; | |
2812 | } | |
2813 | ||
2814 | int btrfs_write_out_ino_cache(struct btrfs_root *root, | |
2815 | struct btrfs_trans_handle *trans, | |
2816 | struct btrfs_path *path) | |
2817 | { | |
2818 | struct btrfs_free_space_ctl *ctl = root->free_ino_ctl; | |
2819 | struct inode *inode; | |
2820 | int ret; | |
2821 | ||
4b9465cb CM |
2822 | if (!btrfs_test_opt(root, INODE_MAP_CACHE)) |
2823 | return 0; | |
2824 | ||
82d5902d LZ |
2825 | inode = lookup_free_ino_inode(root, path); |
2826 | if (IS_ERR(inode)) | |
2827 | return 0; | |
2828 | ||
2829 | ret = __btrfs_write_out_cache(root, inode, ctl, NULL, trans, path, 0); | |
c09544e0 JB |
2830 | if (ret) { |
2831 | btrfs_delalloc_release_metadata(inode, inode->i_size); | |
2832 | #ifdef DEBUG | |
82d5902d LZ |
2833 | printk(KERN_ERR "btrfs: failed to write free ino cache " |
2834 | "for root %llu\n", root->root_key.objectid); | |
c09544e0 JB |
2835 | #endif |
2836 | } | |
82d5902d LZ |
2837 | |
2838 | iput(inode); | |
2839 | return ret; | |
2840 | } |