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++) { | |
a1ee5a45 LZ |
322 | if (io_ctl->pages[i]) { |
323 | ClearPageChecked(io_ctl->pages[i]); | |
324 | unlock_page(io_ctl->pages[i]); | |
325 | page_cache_release(io_ctl->pages[i]); | |
326 | } | |
a67509c3 JB |
327 | } |
328 | } | |
329 | ||
330 | static int io_ctl_prepare_pages(struct io_ctl *io_ctl, struct inode *inode, | |
331 | int uptodate) | |
332 | { | |
333 | struct page *page; | |
334 | gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); | |
335 | int i; | |
336 | ||
337 | for (i = 0; i < io_ctl->num_pages; i++) { | |
338 | page = find_or_create_page(inode->i_mapping, i, mask); | |
339 | if (!page) { | |
340 | io_ctl_drop_pages(io_ctl); | |
341 | return -ENOMEM; | |
342 | } | |
343 | io_ctl->pages[i] = page; | |
344 | if (uptodate && !PageUptodate(page)) { | |
345 | btrfs_readpage(NULL, page); | |
346 | lock_page(page); | |
347 | if (!PageUptodate(page)) { | |
348 | printk(KERN_ERR "btrfs: error reading free " | |
349 | "space cache\n"); | |
350 | io_ctl_drop_pages(io_ctl); | |
351 | return -EIO; | |
352 | } | |
353 | } | |
354 | } | |
355 | ||
f7d61dcd JB |
356 | for (i = 0; i < io_ctl->num_pages; i++) { |
357 | clear_page_dirty_for_io(io_ctl->pages[i]); | |
358 | set_page_extent_mapped(io_ctl->pages[i]); | |
359 | } | |
360 | ||
a67509c3 JB |
361 | return 0; |
362 | } | |
363 | ||
364 | static void io_ctl_set_generation(struct io_ctl *io_ctl, u64 generation) | |
365 | { | |
366 | u64 *val; | |
367 | ||
368 | io_ctl_map_page(io_ctl, 1); | |
369 | ||
370 | /* | |
5b0e95bf JB |
371 | * Skip the csum areas. If we don't check crcs then we just have a |
372 | * 64bit chunk at the front of the first page. | |
a67509c3 | 373 | */ |
5b0e95bf JB |
374 | if (io_ctl->check_crcs) { |
375 | io_ctl->cur += (sizeof(u32) * io_ctl->num_pages); | |
376 | io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages); | |
377 | } else { | |
378 | io_ctl->cur += sizeof(u64); | |
379 | io_ctl->size -= sizeof(u64) * 2; | |
380 | } | |
a67509c3 JB |
381 | |
382 | val = io_ctl->cur; | |
383 | *val = cpu_to_le64(generation); | |
384 | io_ctl->cur += sizeof(u64); | |
a67509c3 JB |
385 | } |
386 | ||
387 | static int io_ctl_check_generation(struct io_ctl *io_ctl, u64 generation) | |
388 | { | |
389 | u64 *gen; | |
390 | ||
5b0e95bf JB |
391 | /* |
392 | * Skip the crc area. If we don't check crcs then we just have a 64bit | |
393 | * chunk at the front of the first page. | |
394 | */ | |
395 | if (io_ctl->check_crcs) { | |
396 | io_ctl->cur += sizeof(u32) * io_ctl->num_pages; | |
397 | io_ctl->size -= sizeof(u64) + | |
398 | (sizeof(u32) * io_ctl->num_pages); | |
399 | } else { | |
400 | io_ctl->cur += sizeof(u64); | |
401 | io_ctl->size -= sizeof(u64) * 2; | |
402 | } | |
a67509c3 | 403 | |
a67509c3 JB |
404 | gen = io_ctl->cur; |
405 | if (le64_to_cpu(*gen) != generation) { | |
406 | printk_ratelimited(KERN_ERR "btrfs: space cache generation " | |
407 | "(%Lu) does not match inode (%Lu)\n", *gen, | |
408 | generation); | |
409 | io_ctl_unmap_page(io_ctl); | |
410 | return -EIO; | |
411 | } | |
412 | io_ctl->cur += sizeof(u64); | |
5b0e95bf JB |
413 | return 0; |
414 | } | |
415 | ||
416 | static void io_ctl_set_crc(struct io_ctl *io_ctl, int index) | |
417 | { | |
418 | u32 *tmp; | |
419 | u32 crc = ~(u32)0; | |
420 | unsigned offset = 0; | |
421 | ||
422 | if (!io_ctl->check_crcs) { | |
423 | io_ctl_unmap_page(io_ctl); | |
424 | return; | |
425 | } | |
426 | ||
427 | if (index == 0) | |
428 | offset = sizeof(u32) * io_ctl->num_pages;; | |
429 | ||
430 | crc = btrfs_csum_data(io_ctl->root, io_ctl->orig + offset, crc, | |
431 | PAGE_CACHE_SIZE - offset); | |
432 | btrfs_csum_final(crc, (char *)&crc); | |
433 | io_ctl_unmap_page(io_ctl); | |
434 | tmp = kmap(io_ctl->pages[0]); | |
435 | tmp += index; | |
436 | *tmp = crc; | |
437 | kunmap(io_ctl->pages[0]); | |
438 | } | |
439 | ||
440 | static int io_ctl_check_crc(struct io_ctl *io_ctl, int index) | |
441 | { | |
442 | u32 *tmp, val; | |
443 | u32 crc = ~(u32)0; | |
444 | unsigned offset = 0; | |
445 | ||
446 | if (!io_ctl->check_crcs) { | |
447 | io_ctl_map_page(io_ctl, 0); | |
448 | return 0; | |
449 | } | |
450 | ||
451 | if (index == 0) | |
452 | offset = sizeof(u32) * io_ctl->num_pages; | |
453 | ||
454 | tmp = kmap(io_ctl->pages[0]); | |
455 | tmp += index; | |
456 | val = *tmp; | |
457 | kunmap(io_ctl->pages[0]); | |
458 | ||
459 | io_ctl_map_page(io_ctl, 0); | |
460 | crc = btrfs_csum_data(io_ctl->root, io_ctl->orig + offset, crc, | |
461 | PAGE_CACHE_SIZE - offset); | |
462 | btrfs_csum_final(crc, (char *)&crc); | |
463 | if (val != crc) { | |
464 | printk_ratelimited(KERN_ERR "btrfs: csum mismatch on free " | |
465 | "space cache\n"); | |
466 | io_ctl_unmap_page(io_ctl); | |
467 | return -EIO; | |
468 | } | |
469 | ||
a67509c3 JB |
470 | return 0; |
471 | } | |
472 | ||
473 | static int io_ctl_add_entry(struct io_ctl *io_ctl, u64 offset, u64 bytes, | |
474 | void *bitmap) | |
475 | { | |
476 | struct btrfs_free_space_entry *entry; | |
477 | ||
478 | if (!io_ctl->cur) | |
479 | return -ENOSPC; | |
480 | ||
481 | entry = io_ctl->cur; | |
482 | entry->offset = cpu_to_le64(offset); | |
483 | entry->bytes = cpu_to_le64(bytes); | |
484 | entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP : | |
485 | BTRFS_FREE_SPACE_EXTENT; | |
486 | io_ctl->cur += sizeof(struct btrfs_free_space_entry); | |
487 | io_ctl->size -= sizeof(struct btrfs_free_space_entry); | |
488 | ||
489 | if (io_ctl->size >= sizeof(struct btrfs_free_space_entry)) | |
490 | return 0; | |
491 | ||
5b0e95bf | 492 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); |
a67509c3 JB |
493 | |
494 | /* No more pages to map */ | |
495 | if (io_ctl->index >= io_ctl->num_pages) | |
496 | return 0; | |
497 | ||
498 | /* map the next page */ | |
499 | io_ctl_map_page(io_ctl, 1); | |
500 | return 0; | |
501 | } | |
502 | ||
503 | static int io_ctl_add_bitmap(struct io_ctl *io_ctl, void *bitmap) | |
504 | { | |
505 | if (!io_ctl->cur) | |
506 | return -ENOSPC; | |
507 | ||
508 | /* | |
509 | * If we aren't at the start of the current page, unmap this one and | |
510 | * map the next one if there is any left. | |
511 | */ | |
512 | if (io_ctl->cur != io_ctl->orig) { | |
5b0e95bf | 513 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); |
a67509c3 JB |
514 | if (io_ctl->index >= io_ctl->num_pages) |
515 | return -ENOSPC; | |
516 | io_ctl_map_page(io_ctl, 0); | |
517 | } | |
518 | ||
519 | memcpy(io_ctl->cur, bitmap, PAGE_CACHE_SIZE); | |
5b0e95bf | 520 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); |
a67509c3 JB |
521 | if (io_ctl->index < io_ctl->num_pages) |
522 | io_ctl_map_page(io_ctl, 0); | |
523 | return 0; | |
524 | } | |
525 | ||
526 | static void io_ctl_zero_remaining_pages(struct io_ctl *io_ctl) | |
527 | { | |
5b0e95bf JB |
528 | /* |
529 | * If we're not on the boundary we know we've modified the page and we | |
530 | * need to crc the page. | |
531 | */ | |
532 | if (io_ctl->cur != io_ctl->orig) | |
533 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); | |
534 | else | |
535 | io_ctl_unmap_page(io_ctl); | |
a67509c3 JB |
536 | |
537 | while (io_ctl->index < io_ctl->num_pages) { | |
538 | io_ctl_map_page(io_ctl, 1); | |
5b0e95bf | 539 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); |
a67509c3 JB |
540 | } |
541 | } | |
542 | ||
5b0e95bf JB |
543 | static int io_ctl_read_entry(struct io_ctl *io_ctl, |
544 | struct btrfs_free_space *entry, u8 *type) | |
a67509c3 JB |
545 | { |
546 | struct btrfs_free_space_entry *e; | |
2f120c05 JB |
547 | int ret; |
548 | ||
549 | if (!io_ctl->cur) { | |
550 | ret = io_ctl_check_crc(io_ctl, io_ctl->index); | |
551 | if (ret) | |
552 | return ret; | |
553 | } | |
a67509c3 JB |
554 | |
555 | e = io_ctl->cur; | |
556 | entry->offset = le64_to_cpu(e->offset); | |
557 | entry->bytes = le64_to_cpu(e->bytes); | |
5b0e95bf | 558 | *type = e->type; |
a67509c3 JB |
559 | io_ctl->cur += sizeof(struct btrfs_free_space_entry); |
560 | io_ctl->size -= sizeof(struct btrfs_free_space_entry); | |
561 | ||
562 | if (io_ctl->size >= sizeof(struct btrfs_free_space_entry)) | |
5b0e95bf | 563 | return 0; |
a67509c3 JB |
564 | |
565 | io_ctl_unmap_page(io_ctl); | |
566 | ||
2f120c05 | 567 | return 0; |
a67509c3 JB |
568 | } |
569 | ||
5b0e95bf JB |
570 | static int io_ctl_read_bitmap(struct io_ctl *io_ctl, |
571 | struct btrfs_free_space *entry) | |
a67509c3 | 572 | { |
5b0e95bf JB |
573 | int ret; |
574 | ||
5b0e95bf JB |
575 | ret = io_ctl_check_crc(io_ctl, io_ctl->index); |
576 | if (ret) | |
577 | return ret; | |
578 | ||
a67509c3 JB |
579 | memcpy(entry->bitmap, io_ctl->cur, PAGE_CACHE_SIZE); |
580 | io_ctl_unmap_page(io_ctl); | |
5b0e95bf JB |
581 | |
582 | return 0; | |
a67509c3 JB |
583 | } |
584 | ||
0414efae LZ |
585 | int __load_free_space_cache(struct btrfs_root *root, struct inode *inode, |
586 | struct btrfs_free_space_ctl *ctl, | |
587 | struct btrfs_path *path, u64 offset) | |
9d66e233 | 588 | { |
9d66e233 JB |
589 | struct btrfs_free_space_header *header; |
590 | struct extent_buffer *leaf; | |
a67509c3 | 591 | struct io_ctl io_ctl; |
9d66e233 | 592 | struct btrfs_key key; |
a67509c3 | 593 | struct btrfs_free_space *e, *n; |
9d66e233 JB |
594 | struct list_head bitmaps; |
595 | u64 num_entries; | |
596 | u64 num_bitmaps; | |
597 | u64 generation; | |
a67509c3 | 598 | u8 type; |
f6a39829 | 599 | int ret = 0; |
9d66e233 JB |
600 | |
601 | INIT_LIST_HEAD(&bitmaps); | |
602 | ||
9d66e233 | 603 | /* Nothing in the space cache, goodbye */ |
0414efae | 604 | if (!i_size_read(inode)) |
a67509c3 | 605 | return 0; |
9d66e233 JB |
606 | |
607 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
0414efae | 608 | key.offset = offset; |
9d66e233 JB |
609 | key.type = 0; |
610 | ||
611 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
0414efae | 612 | if (ret < 0) |
a67509c3 | 613 | return 0; |
0414efae | 614 | else if (ret > 0) { |
945d8962 | 615 | btrfs_release_path(path); |
a67509c3 | 616 | return 0; |
9d66e233 JB |
617 | } |
618 | ||
0414efae LZ |
619 | ret = -1; |
620 | ||
9d66e233 JB |
621 | leaf = path->nodes[0]; |
622 | header = btrfs_item_ptr(leaf, path->slots[0], | |
623 | struct btrfs_free_space_header); | |
624 | num_entries = btrfs_free_space_entries(leaf, header); | |
625 | num_bitmaps = btrfs_free_space_bitmaps(leaf, header); | |
626 | generation = btrfs_free_space_generation(leaf, header); | |
945d8962 | 627 | btrfs_release_path(path); |
9d66e233 JB |
628 | |
629 | if (BTRFS_I(inode)->generation != generation) { | |
630 | printk(KERN_ERR "btrfs: free space inode generation (%llu) did" | |
0414efae | 631 | " not match free space cache generation (%llu)\n", |
9d66e233 | 632 | (unsigned long long)BTRFS_I(inode)->generation, |
0414efae | 633 | (unsigned long long)generation); |
a67509c3 | 634 | return 0; |
9d66e233 JB |
635 | } |
636 | ||
637 | if (!num_entries) | |
a67509c3 | 638 | return 0; |
9d66e233 | 639 | |
a67509c3 | 640 | io_ctl_init(&io_ctl, inode, root); |
9d66e233 | 641 | ret = readahead_cache(inode); |
0414efae | 642 | if (ret) |
9d66e233 | 643 | goto out; |
9d66e233 | 644 | |
a67509c3 JB |
645 | ret = io_ctl_prepare_pages(&io_ctl, inode, 1); |
646 | if (ret) | |
647 | goto out; | |
9d66e233 | 648 | |
5b0e95bf JB |
649 | ret = io_ctl_check_crc(&io_ctl, 0); |
650 | if (ret) | |
651 | goto free_cache; | |
652 | ||
a67509c3 JB |
653 | ret = io_ctl_check_generation(&io_ctl, generation); |
654 | if (ret) | |
655 | goto free_cache; | |
9d66e233 | 656 | |
a67509c3 JB |
657 | while (num_entries) { |
658 | e = kmem_cache_zalloc(btrfs_free_space_cachep, | |
659 | GFP_NOFS); | |
660 | if (!e) | |
9d66e233 | 661 | goto free_cache; |
9d66e233 | 662 | |
5b0e95bf JB |
663 | ret = io_ctl_read_entry(&io_ctl, e, &type); |
664 | if (ret) { | |
665 | kmem_cache_free(btrfs_free_space_cachep, e); | |
666 | goto free_cache; | |
667 | } | |
668 | ||
a67509c3 JB |
669 | if (!e->bytes) { |
670 | kmem_cache_free(btrfs_free_space_cachep, e); | |
671 | goto free_cache; | |
9d66e233 | 672 | } |
a67509c3 JB |
673 | |
674 | if (type == BTRFS_FREE_SPACE_EXTENT) { | |
675 | spin_lock(&ctl->tree_lock); | |
676 | ret = link_free_space(ctl, e); | |
677 | spin_unlock(&ctl->tree_lock); | |
678 | if (ret) { | |
679 | printk(KERN_ERR "Duplicate entries in " | |
680 | "free space cache, dumping\n"); | |
681 | kmem_cache_free(btrfs_free_space_cachep, e); | |
9d66e233 JB |
682 | goto free_cache; |
683 | } | |
a67509c3 JB |
684 | } else { |
685 | BUG_ON(!num_bitmaps); | |
686 | num_bitmaps--; | |
687 | e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS); | |
688 | if (!e->bitmap) { | |
689 | kmem_cache_free( | |
690 | btrfs_free_space_cachep, e); | |
9d66e233 JB |
691 | goto free_cache; |
692 | } | |
a67509c3 JB |
693 | spin_lock(&ctl->tree_lock); |
694 | ret = link_free_space(ctl, e); | |
695 | ctl->total_bitmaps++; | |
696 | ctl->op->recalc_thresholds(ctl); | |
697 | spin_unlock(&ctl->tree_lock); | |
698 | if (ret) { | |
699 | printk(KERN_ERR "Duplicate entries in " | |
700 | "free space cache, dumping\n"); | |
dc89e982 | 701 | kmem_cache_free(btrfs_free_space_cachep, e); |
9d66e233 JB |
702 | goto free_cache; |
703 | } | |
a67509c3 | 704 | list_add_tail(&e->list, &bitmaps); |
9d66e233 JB |
705 | } |
706 | ||
a67509c3 JB |
707 | num_entries--; |
708 | } | |
9d66e233 | 709 | |
2f120c05 JB |
710 | io_ctl_unmap_page(&io_ctl); |
711 | ||
a67509c3 JB |
712 | /* |
713 | * We add the bitmaps at the end of the entries in order that | |
714 | * the bitmap entries are added to the cache. | |
715 | */ | |
716 | list_for_each_entry_safe(e, n, &bitmaps, list) { | |
9d66e233 | 717 | list_del_init(&e->list); |
5b0e95bf JB |
718 | ret = io_ctl_read_bitmap(&io_ctl, e); |
719 | if (ret) | |
720 | goto free_cache; | |
9d66e233 JB |
721 | } |
722 | ||
a67509c3 | 723 | io_ctl_drop_pages(&io_ctl); |
9d66e233 JB |
724 | ret = 1; |
725 | out: | |
a67509c3 | 726 | io_ctl_free(&io_ctl); |
9d66e233 | 727 | return ret; |
9d66e233 | 728 | free_cache: |
a67509c3 | 729 | io_ctl_drop_pages(&io_ctl); |
0414efae | 730 | __btrfs_remove_free_space_cache(ctl); |
9d66e233 JB |
731 | goto out; |
732 | } | |
733 | ||
0414efae LZ |
734 | int load_free_space_cache(struct btrfs_fs_info *fs_info, |
735 | struct btrfs_block_group_cache *block_group) | |
0cb59c99 | 736 | { |
34d52cb6 | 737 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0414efae LZ |
738 | struct btrfs_root *root = fs_info->tree_root; |
739 | struct inode *inode; | |
740 | struct btrfs_path *path; | |
5b0e95bf | 741 | int ret = 0; |
0414efae LZ |
742 | bool matched; |
743 | u64 used = btrfs_block_group_used(&block_group->item); | |
744 | ||
745 | /* | |
746 | * If we're unmounting then just return, since this does a search on the | |
747 | * normal root and not the commit root and we could deadlock. | |
748 | */ | |
7841cb28 | 749 | if (btrfs_fs_closing(fs_info)) |
0414efae LZ |
750 | return 0; |
751 | ||
752 | /* | |
753 | * If this block group has been marked to be cleared for one reason or | |
754 | * another then we can't trust the on disk cache, so just return. | |
755 | */ | |
9d66e233 | 756 | spin_lock(&block_group->lock); |
0414efae LZ |
757 | if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) { |
758 | spin_unlock(&block_group->lock); | |
759 | return 0; | |
760 | } | |
9d66e233 | 761 | spin_unlock(&block_group->lock); |
0414efae LZ |
762 | |
763 | path = btrfs_alloc_path(); | |
764 | if (!path) | |
765 | return 0; | |
766 | ||
767 | inode = lookup_free_space_inode(root, block_group, path); | |
768 | if (IS_ERR(inode)) { | |
769 | btrfs_free_path(path); | |
770 | return 0; | |
771 | } | |
772 | ||
5b0e95bf JB |
773 | /* We may have converted the inode and made the cache invalid. */ |
774 | spin_lock(&block_group->lock); | |
775 | if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) { | |
776 | spin_unlock(&block_group->lock); | |
777 | goto out; | |
778 | } | |
779 | spin_unlock(&block_group->lock); | |
780 | ||
0414efae LZ |
781 | ret = __load_free_space_cache(fs_info->tree_root, inode, ctl, |
782 | path, block_group->key.objectid); | |
783 | btrfs_free_path(path); | |
784 | if (ret <= 0) | |
785 | goto out; | |
786 | ||
787 | spin_lock(&ctl->tree_lock); | |
788 | matched = (ctl->free_space == (block_group->key.offset - used - | |
789 | block_group->bytes_super)); | |
790 | spin_unlock(&ctl->tree_lock); | |
791 | ||
792 | if (!matched) { | |
793 | __btrfs_remove_free_space_cache(ctl); | |
794 | printk(KERN_ERR "block group %llu has an wrong amount of free " | |
795 | "space\n", block_group->key.objectid); | |
796 | ret = -1; | |
797 | } | |
798 | out: | |
799 | if (ret < 0) { | |
800 | /* This cache is bogus, make sure it gets cleared */ | |
801 | spin_lock(&block_group->lock); | |
802 | block_group->disk_cache_state = BTRFS_DC_CLEAR; | |
803 | spin_unlock(&block_group->lock); | |
82d5902d | 804 | ret = 0; |
0414efae LZ |
805 | |
806 | printk(KERN_ERR "btrfs: failed to load free space cache " | |
807 | "for block group %llu\n", block_group->key.objectid); | |
808 | } | |
809 | ||
810 | iput(inode); | |
811 | return ret; | |
9d66e233 JB |
812 | } |
813 | ||
c09544e0 JB |
814 | /** |
815 | * __btrfs_write_out_cache - write out cached info to an inode | |
816 | * @root - the root the inode belongs to | |
817 | * @ctl - the free space cache we are going to write out | |
818 | * @block_group - the block_group for this cache if it belongs to a block_group | |
819 | * @trans - the trans handle | |
820 | * @path - the path to use | |
821 | * @offset - the offset for the key we'll insert | |
822 | * | |
823 | * This function writes out a free space cache struct to disk for quick recovery | |
824 | * on mount. This will return 0 if it was successfull in writing the cache out, | |
825 | * and -1 if it was not. | |
826 | */ | |
0414efae LZ |
827 | int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode, |
828 | struct btrfs_free_space_ctl *ctl, | |
829 | struct btrfs_block_group_cache *block_group, | |
830 | struct btrfs_trans_handle *trans, | |
831 | struct btrfs_path *path, u64 offset) | |
0cb59c99 JB |
832 | { |
833 | struct btrfs_free_space_header *header; | |
834 | struct extent_buffer *leaf; | |
0cb59c99 JB |
835 | struct rb_node *node; |
836 | struct list_head *pos, *n; | |
0cb59c99 | 837 | struct extent_state *cached_state = NULL; |
43be2146 JB |
838 | struct btrfs_free_cluster *cluster = NULL; |
839 | struct extent_io_tree *unpin = NULL; | |
a67509c3 | 840 | struct io_ctl io_ctl; |
0cb59c99 JB |
841 | struct list_head bitmap_list; |
842 | struct btrfs_key key; | |
db804f23 | 843 | u64 start, extent_start, extent_end, len; |
0cb59c99 JB |
844 | int entries = 0; |
845 | int bitmaps = 0; | |
c09544e0 JB |
846 | int ret; |
847 | int err = -1; | |
0cb59c99 | 848 | |
0cb59c99 JB |
849 | INIT_LIST_HEAD(&bitmap_list); |
850 | ||
0414efae LZ |
851 | if (!i_size_read(inode)) |
852 | return -1; | |
2b20982e | 853 | |
a67509c3 | 854 | io_ctl_init(&io_ctl, inode, root); |
be1a12a0 | 855 | |
43be2146 | 856 | /* Get the cluster for this block_group if it exists */ |
0414efae | 857 | if (block_group && !list_empty(&block_group->cluster_list)) |
43be2146 JB |
858 | cluster = list_entry(block_group->cluster_list.next, |
859 | struct btrfs_free_cluster, | |
860 | block_group_list); | |
861 | ||
a67509c3 JB |
862 | /* Lock all pages first so we can lock the extent safely. */ |
863 | io_ctl_prepare_pages(&io_ctl, inode, 0); | |
0cb59c99 | 864 | |
0cb59c99 JB |
865 | lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1, |
866 | 0, &cached_state, GFP_NOFS); | |
867 | ||
f75b130e JB |
868 | node = rb_first(&ctl->free_space_offset); |
869 | if (!node && cluster) { | |
870 | node = rb_first(&cluster->root); | |
871 | cluster = NULL; | |
872 | } | |
873 | ||
5b0e95bf JB |
874 | /* Make sure we can fit our crcs into the first page */ |
875 | if (io_ctl.check_crcs && | |
876 | (io_ctl.num_pages * sizeof(u32)) >= PAGE_CACHE_SIZE) { | |
877 | WARN_ON(1); | |
878 | goto out_nospc; | |
879 | } | |
880 | ||
a67509c3 | 881 | io_ctl_set_generation(&io_ctl, trans->transid); |
43be2146 | 882 | |
a67509c3 JB |
883 | /* Write out the extent entries */ |
884 | while (node) { | |
885 | struct btrfs_free_space *e; | |
0cb59c99 | 886 | |
a67509c3 JB |
887 | e = rb_entry(node, struct btrfs_free_space, offset_index); |
888 | entries++; | |
0cb59c99 | 889 | |
a67509c3 JB |
890 | ret = io_ctl_add_entry(&io_ctl, e->offset, e->bytes, |
891 | e->bitmap); | |
892 | if (ret) | |
893 | goto out_nospc; | |
2f356126 | 894 | |
a67509c3 JB |
895 | if (e->bitmap) { |
896 | list_add_tail(&e->list, &bitmap_list); | |
897 | bitmaps++; | |
2f356126 | 898 | } |
a67509c3 JB |
899 | node = rb_next(node); |
900 | if (!node && cluster) { | |
901 | node = rb_first(&cluster->root); | |
902 | cluster = NULL; | |
43be2146 | 903 | } |
a67509c3 | 904 | } |
43be2146 | 905 | |
a67509c3 JB |
906 | /* |
907 | * We want to add any pinned extents to our free space cache | |
908 | * so we don't leak the space | |
909 | */ | |
db804f23 LZ |
910 | |
911 | /* | |
912 | * We shouldn't have switched the pinned extents yet so this is the | |
913 | * right one | |
914 | */ | |
915 | unpin = root->fs_info->pinned_extents; | |
916 | ||
917 | if (block_group) | |
918 | start = block_group->key.objectid; | |
919 | ||
a67509c3 JB |
920 | while (block_group && (start < block_group->key.objectid + |
921 | block_group->key.offset)) { | |
db804f23 LZ |
922 | ret = find_first_extent_bit(unpin, start, |
923 | &extent_start, &extent_end, | |
a67509c3 JB |
924 | EXTENT_DIRTY); |
925 | if (ret) { | |
926 | ret = 0; | |
927 | break; | |
0cb59c99 | 928 | } |
0cb59c99 | 929 | |
a67509c3 | 930 | /* This pinned extent is out of our range */ |
db804f23 | 931 | if (extent_start >= block_group->key.objectid + |
a67509c3 JB |
932 | block_group->key.offset) |
933 | break; | |
2f356126 | 934 | |
db804f23 LZ |
935 | extent_start = max(extent_start, start); |
936 | extent_end = min(block_group->key.objectid + | |
937 | block_group->key.offset, extent_end + 1); | |
938 | len = extent_end - extent_start; | |
0cb59c99 | 939 | |
a67509c3 | 940 | entries++; |
db804f23 | 941 | ret = io_ctl_add_entry(&io_ctl, extent_start, len, NULL); |
a67509c3 JB |
942 | if (ret) |
943 | goto out_nospc; | |
0cb59c99 | 944 | |
db804f23 | 945 | start = extent_end; |
a67509c3 | 946 | } |
0cb59c99 JB |
947 | |
948 | /* Write out the bitmaps */ | |
949 | list_for_each_safe(pos, n, &bitmap_list) { | |
0cb59c99 JB |
950 | struct btrfs_free_space *entry = |
951 | list_entry(pos, struct btrfs_free_space, list); | |
952 | ||
a67509c3 JB |
953 | ret = io_ctl_add_bitmap(&io_ctl, entry->bitmap); |
954 | if (ret) | |
955 | goto out_nospc; | |
0cb59c99 | 956 | list_del_init(&entry->list); |
be1a12a0 JB |
957 | } |
958 | ||
0cb59c99 | 959 | /* Zero out the rest of the pages just to make sure */ |
a67509c3 | 960 | io_ctl_zero_remaining_pages(&io_ctl); |
0cb59c99 | 961 | |
a67509c3 JB |
962 | ret = btrfs_dirty_pages(root, inode, io_ctl.pages, io_ctl.num_pages, |
963 | 0, i_size_read(inode), &cached_state); | |
964 | io_ctl_drop_pages(&io_ctl); | |
0cb59c99 JB |
965 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0, |
966 | i_size_read(inode) - 1, &cached_state, GFP_NOFS); | |
967 | ||
c09544e0 | 968 | if (ret) |
2f356126 | 969 | goto out; |
be1a12a0 | 970 | |
be1a12a0 | 971 | |
549b4fdb JB |
972 | ret = filemap_write_and_wait(inode->i_mapping); |
973 | if (ret) | |
974 | goto out; | |
0cb59c99 JB |
975 | |
976 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
0414efae | 977 | key.offset = offset; |
0cb59c99 JB |
978 | key.type = 0; |
979 | ||
a9b5fcdd | 980 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
0cb59c99 | 981 | if (ret < 0) { |
a67509c3 | 982 | clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1, |
5b0e95bf JB |
983 | EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL, |
984 | GFP_NOFS); | |
2f356126 | 985 | goto out; |
0cb59c99 JB |
986 | } |
987 | leaf = path->nodes[0]; | |
988 | if (ret > 0) { | |
989 | struct btrfs_key found_key; | |
990 | BUG_ON(!path->slots[0]); | |
991 | path->slots[0]--; | |
992 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
993 | if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID || | |
0414efae | 994 | found_key.offset != offset) { |
a67509c3 JB |
995 | clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, |
996 | inode->i_size - 1, | |
5b0e95bf JB |
997 | EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, |
998 | NULL, GFP_NOFS); | |
b3b4aa74 | 999 | btrfs_release_path(path); |
2f356126 | 1000 | goto out; |
0cb59c99 JB |
1001 | } |
1002 | } | |
549b4fdb JB |
1003 | |
1004 | BTRFS_I(inode)->generation = trans->transid; | |
0cb59c99 JB |
1005 | header = btrfs_item_ptr(leaf, path->slots[0], |
1006 | struct btrfs_free_space_header); | |
1007 | btrfs_set_free_space_entries(leaf, header, entries); | |
1008 | btrfs_set_free_space_bitmaps(leaf, header, bitmaps); | |
1009 | btrfs_set_free_space_generation(leaf, header, trans->transid); | |
1010 | btrfs_mark_buffer_dirty(leaf); | |
b3b4aa74 | 1011 | btrfs_release_path(path); |
0cb59c99 | 1012 | |
c09544e0 | 1013 | err = 0; |
2f356126 | 1014 | out: |
a67509c3 | 1015 | io_ctl_free(&io_ctl); |
c09544e0 | 1016 | if (err) { |
a67509c3 | 1017 | invalidate_inode_pages2(inode->i_mapping); |
0cb59c99 JB |
1018 | BTRFS_I(inode)->generation = 0; |
1019 | } | |
0cb59c99 | 1020 | btrfs_update_inode(trans, root, inode); |
c09544e0 | 1021 | return err; |
a67509c3 JB |
1022 | |
1023 | out_nospc: | |
1024 | list_for_each_safe(pos, n, &bitmap_list) { | |
1025 | struct btrfs_free_space *entry = | |
1026 | list_entry(pos, struct btrfs_free_space, list); | |
1027 | list_del_init(&entry->list); | |
1028 | } | |
1029 | io_ctl_drop_pages(&io_ctl); | |
1030 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0, | |
1031 | i_size_read(inode) - 1, &cached_state, GFP_NOFS); | |
1032 | goto out; | |
0414efae LZ |
1033 | } |
1034 | ||
1035 | int btrfs_write_out_cache(struct btrfs_root *root, | |
1036 | struct btrfs_trans_handle *trans, | |
1037 | struct btrfs_block_group_cache *block_group, | |
1038 | struct btrfs_path *path) | |
1039 | { | |
1040 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
1041 | struct inode *inode; | |
1042 | int ret = 0; | |
1043 | ||
1044 | root = root->fs_info->tree_root; | |
1045 | ||
1046 | spin_lock(&block_group->lock); | |
1047 | if (block_group->disk_cache_state < BTRFS_DC_SETUP) { | |
1048 | spin_unlock(&block_group->lock); | |
1049 | return 0; | |
1050 | } | |
1051 | spin_unlock(&block_group->lock); | |
1052 | ||
1053 | inode = lookup_free_space_inode(root, block_group, path); | |
1054 | if (IS_ERR(inode)) | |
1055 | return 0; | |
1056 | ||
1057 | ret = __btrfs_write_out_cache(root, inode, ctl, block_group, trans, | |
1058 | path, block_group->key.objectid); | |
c09544e0 | 1059 | if (ret) { |
0414efae LZ |
1060 | spin_lock(&block_group->lock); |
1061 | block_group->disk_cache_state = BTRFS_DC_ERROR; | |
1062 | spin_unlock(&block_group->lock); | |
82d5902d | 1063 | ret = 0; |
c09544e0 | 1064 | #ifdef DEBUG |
0414efae LZ |
1065 | printk(KERN_ERR "btrfs: failed to write free space cace " |
1066 | "for block group %llu\n", block_group->key.objectid); | |
c09544e0 | 1067 | #endif |
0414efae LZ |
1068 | } |
1069 | ||
0cb59c99 JB |
1070 | iput(inode); |
1071 | return ret; | |
1072 | } | |
1073 | ||
34d52cb6 | 1074 | static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit, |
96303081 | 1075 | u64 offset) |
0f9dd46c | 1076 | { |
96303081 JB |
1077 | BUG_ON(offset < bitmap_start); |
1078 | offset -= bitmap_start; | |
34d52cb6 | 1079 | return (unsigned long)(div_u64(offset, unit)); |
96303081 | 1080 | } |
0f9dd46c | 1081 | |
34d52cb6 | 1082 | static inline unsigned long bytes_to_bits(u64 bytes, u32 unit) |
96303081 | 1083 | { |
34d52cb6 | 1084 | return (unsigned long)(div_u64(bytes, unit)); |
96303081 | 1085 | } |
0f9dd46c | 1086 | |
34d52cb6 | 1087 | static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1088 | u64 offset) |
1089 | { | |
1090 | u64 bitmap_start; | |
1091 | u64 bytes_per_bitmap; | |
0f9dd46c | 1092 | |
34d52cb6 LZ |
1093 | bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit; |
1094 | bitmap_start = offset - ctl->start; | |
96303081 JB |
1095 | bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap); |
1096 | bitmap_start *= bytes_per_bitmap; | |
34d52cb6 | 1097 | bitmap_start += ctl->start; |
0f9dd46c | 1098 | |
96303081 | 1099 | return bitmap_start; |
0f9dd46c JB |
1100 | } |
1101 | ||
96303081 JB |
1102 | static int tree_insert_offset(struct rb_root *root, u64 offset, |
1103 | struct rb_node *node, int bitmap) | |
0f9dd46c JB |
1104 | { |
1105 | struct rb_node **p = &root->rb_node; | |
1106 | struct rb_node *parent = NULL; | |
1107 | struct btrfs_free_space *info; | |
1108 | ||
1109 | while (*p) { | |
1110 | parent = *p; | |
96303081 | 1111 | info = rb_entry(parent, struct btrfs_free_space, offset_index); |
0f9dd46c | 1112 | |
96303081 | 1113 | if (offset < info->offset) { |
0f9dd46c | 1114 | p = &(*p)->rb_left; |
96303081 | 1115 | } else if (offset > info->offset) { |
0f9dd46c | 1116 | p = &(*p)->rb_right; |
96303081 JB |
1117 | } else { |
1118 | /* | |
1119 | * we could have a bitmap entry and an extent entry | |
1120 | * share the same offset. If this is the case, we want | |
1121 | * the extent entry to always be found first if we do a | |
1122 | * linear search through the tree, since we want to have | |
1123 | * the quickest allocation time, and allocating from an | |
1124 | * extent is faster than allocating from a bitmap. So | |
1125 | * if we're inserting a bitmap and we find an entry at | |
1126 | * this offset, we want to go right, or after this entry | |
1127 | * logically. If we are inserting an extent and we've | |
1128 | * found a bitmap, we want to go left, or before | |
1129 | * logically. | |
1130 | */ | |
1131 | if (bitmap) { | |
207dde82 JB |
1132 | if (info->bitmap) { |
1133 | WARN_ON_ONCE(1); | |
1134 | return -EEXIST; | |
1135 | } | |
96303081 JB |
1136 | p = &(*p)->rb_right; |
1137 | } else { | |
207dde82 JB |
1138 | if (!info->bitmap) { |
1139 | WARN_ON_ONCE(1); | |
1140 | return -EEXIST; | |
1141 | } | |
96303081 JB |
1142 | p = &(*p)->rb_left; |
1143 | } | |
1144 | } | |
0f9dd46c JB |
1145 | } |
1146 | ||
1147 | rb_link_node(node, parent, p); | |
1148 | rb_insert_color(node, root); | |
1149 | ||
1150 | return 0; | |
1151 | } | |
1152 | ||
1153 | /* | |
70cb0743 JB |
1154 | * searches the tree for the given offset. |
1155 | * | |
96303081 JB |
1156 | * fuzzy - If this is set, then we are trying to make an allocation, and we just |
1157 | * want a section that has at least bytes size and comes at or after the given | |
1158 | * offset. | |
0f9dd46c | 1159 | */ |
96303081 | 1160 | static struct btrfs_free_space * |
34d52cb6 | 1161 | tree_search_offset(struct btrfs_free_space_ctl *ctl, |
96303081 | 1162 | u64 offset, int bitmap_only, int fuzzy) |
0f9dd46c | 1163 | { |
34d52cb6 | 1164 | struct rb_node *n = ctl->free_space_offset.rb_node; |
96303081 JB |
1165 | struct btrfs_free_space *entry, *prev = NULL; |
1166 | ||
1167 | /* find entry that is closest to the 'offset' */ | |
1168 | while (1) { | |
1169 | if (!n) { | |
1170 | entry = NULL; | |
1171 | break; | |
1172 | } | |
0f9dd46c | 1173 | |
0f9dd46c | 1174 | entry = rb_entry(n, struct btrfs_free_space, offset_index); |
96303081 | 1175 | prev = entry; |
0f9dd46c | 1176 | |
96303081 | 1177 | if (offset < entry->offset) |
0f9dd46c | 1178 | n = n->rb_left; |
96303081 | 1179 | else if (offset > entry->offset) |
0f9dd46c | 1180 | n = n->rb_right; |
96303081 | 1181 | else |
0f9dd46c | 1182 | break; |
0f9dd46c JB |
1183 | } |
1184 | ||
96303081 JB |
1185 | if (bitmap_only) { |
1186 | if (!entry) | |
1187 | return NULL; | |
1188 | if (entry->bitmap) | |
1189 | return entry; | |
0f9dd46c | 1190 | |
96303081 JB |
1191 | /* |
1192 | * bitmap entry and extent entry may share same offset, | |
1193 | * in that case, bitmap entry comes after extent entry. | |
1194 | */ | |
1195 | n = rb_next(n); | |
1196 | if (!n) | |
1197 | return NULL; | |
1198 | entry = rb_entry(n, struct btrfs_free_space, offset_index); | |
1199 | if (entry->offset != offset) | |
1200 | return NULL; | |
0f9dd46c | 1201 | |
96303081 JB |
1202 | WARN_ON(!entry->bitmap); |
1203 | return entry; | |
1204 | } else if (entry) { | |
1205 | if (entry->bitmap) { | |
0f9dd46c | 1206 | /* |
96303081 JB |
1207 | * if previous extent entry covers the offset, |
1208 | * we should return it instead of the bitmap entry | |
0f9dd46c | 1209 | */ |
96303081 JB |
1210 | n = &entry->offset_index; |
1211 | while (1) { | |
1212 | n = rb_prev(n); | |
1213 | if (!n) | |
1214 | break; | |
1215 | prev = rb_entry(n, struct btrfs_free_space, | |
1216 | offset_index); | |
1217 | if (!prev->bitmap) { | |
1218 | if (prev->offset + prev->bytes > offset) | |
1219 | entry = prev; | |
1220 | break; | |
1221 | } | |
0f9dd46c | 1222 | } |
96303081 JB |
1223 | } |
1224 | return entry; | |
1225 | } | |
1226 | ||
1227 | if (!prev) | |
1228 | return NULL; | |
1229 | ||
1230 | /* find last entry before the 'offset' */ | |
1231 | entry = prev; | |
1232 | if (entry->offset > offset) { | |
1233 | n = rb_prev(&entry->offset_index); | |
1234 | if (n) { | |
1235 | entry = rb_entry(n, struct btrfs_free_space, | |
1236 | offset_index); | |
1237 | BUG_ON(entry->offset > offset); | |
0f9dd46c | 1238 | } else { |
96303081 JB |
1239 | if (fuzzy) |
1240 | return entry; | |
1241 | else | |
1242 | return NULL; | |
0f9dd46c JB |
1243 | } |
1244 | } | |
1245 | ||
96303081 JB |
1246 | if (entry->bitmap) { |
1247 | n = &entry->offset_index; | |
1248 | while (1) { | |
1249 | n = rb_prev(n); | |
1250 | if (!n) | |
1251 | break; | |
1252 | prev = rb_entry(n, struct btrfs_free_space, | |
1253 | offset_index); | |
1254 | if (!prev->bitmap) { | |
1255 | if (prev->offset + prev->bytes > offset) | |
1256 | return prev; | |
1257 | break; | |
1258 | } | |
1259 | } | |
34d52cb6 | 1260 | if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset) |
96303081 JB |
1261 | return entry; |
1262 | } else if (entry->offset + entry->bytes > offset) | |
1263 | return entry; | |
1264 | ||
1265 | if (!fuzzy) | |
1266 | return NULL; | |
1267 | ||
1268 | while (1) { | |
1269 | if (entry->bitmap) { | |
1270 | if (entry->offset + BITS_PER_BITMAP * | |
34d52cb6 | 1271 | ctl->unit > offset) |
96303081 JB |
1272 | break; |
1273 | } else { | |
1274 | if (entry->offset + entry->bytes > offset) | |
1275 | break; | |
1276 | } | |
1277 | ||
1278 | n = rb_next(&entry->offset_index); | |
1279 | if (!n) | |
1280 | return NULL; | |
1281 | entry = rb_entry(n, struct btrfs_free_space, offset_index); | |
1282 | } | |
1283 | return entry; | |
0f9dd46c JB |
1284 | } |
1285 | ||
f333adb5 | 1286 | static inline void |
34d52cb6 | 1287 | __unlink_free_space(struct btrfs_free_space_ctl *ctl, |
f333adb5 | 1288 | struct btrfs_free_space *info) |
0f9dd46c | 1289 | { |
34d52cb6 LZ |
1290 | rb_erase(&info->offset_index, &ctl->free_space_offset); |
1291 | ctl->free_extents--; | |
f333adb5 LZ |
1292 | } |
1293 | ||
34d52cb6 | 1294 | static void unlink_free_space(struct btrfs_free_space_ctl *ctl, |
f333adb5 LZ |
1295 | struct btrfs_free_space *info) |
1296 | { | |
34d52cb6 LZ |
1297 | __unlink_free_space(ctl, info); |
1298 | ctl->free_space -= info->bytes; | |
0f9dd46c JB |
1299 | } |
1300 | ||
34d52cb6 | 1301 | static int link_free_space(struct btrfs_free_space_ctl *ctl, |
0f9dd46c JB |
1302 | struct btrfs_free_space *info) |
1303 | { | |
1304 | int ret = 0; | |
1305 | ||
96303081 | 1306 | BUG_ON(!info->bitmap && !info->bytes); |
34d52cb6 | 1307 | ret = tree_insert_offset(&ctl->free_space_offset, info->offset, |
96303081 | 1308 | &info->offset_index, (info->bitmap != NULL)); |
0f9dd46c JB |
1309 | if (ret) |
1310 | return ret; | |
1311 | ||
34d52cb6 LZ |
1312 | ctl->free_space += info->bytes; |
1313 | ctl->free_extents++; | |
96303081 JB |
1314 | return ret; |
1315 | } | |
1316 | ||
34d52cb6 | 1317 | static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl) |
96303081 | 1318 | { |
34d52cb6 | 1319 | struct btrfs_block_group_cache *block_group = ctl->private; |
25891f79 JB |
1320 | u64 max_bytes; |
1321 | u64 bitmap_bytes; | |
1322 | u64 extent_bytes; | |
8eb2d829 | 1323 | u64 size = block_group->key.offset; |
34d52cb6 LZ |
1324 | u64 bytes_per_bg = BITS_PER_BITMAP * block_group->sectorsize; |
1325 | int max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg); | |
1326 | ||
1327 | BUG_ON(ctl->total_bitmaps > max_bitmaps); | |
96303081 JB |
1328 | |
1329 | /* | |
1330 | * The goal is to keep the total amount of memory used per 1gb of space | |
1331 | * at or below 32k, so we need to adjust how much memory we allow to be | |
1332 | * used by extent based free space tracking | |
1333 | */ | |
8eb2d829 LZ |
1334 | if (size < 1024 * 1024 * 1024) |
1335 | max_bytes = MAX_CACHE_BYTES_PER_GIG; | |
1336 | else | |
1337 | max_bytes = MAX_CACHE_BYTES_PER_GIG * | |
1338 | div64_u64(size, 1024 * 1024 * 1024); | |
96303081 | 1339 | |
25891f79 JB |
1340 | /* |
1341 | * we want to account for 1 more bitmap than what we have so we can make | |
1342 | * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as | |
1343 | * we add more bitmaps. | |
1344 | */ | |
34d52cb6 | 1345 | bitmap_bytes = (ctl->total_bitmaps + 1) * PAGE_CACHE_SIZE; |
96303081 | 1346 | |
25891f79 | 1347 | if (bitmap_bytes >= max_bytes) { |
34d52cb6 | 1348 | ctl->extents_thresh = 0; |
25891f79 JB |
1349 | return; |
1350 | } | |
96303081 | 1351 | |
25891f79 JB |
1352 | /* |
1353 | * we want the extent entry threshold to always be at most 1/2 the maxw | |
1354 | * bytes we can have, or whatever is less than that. | |
1355 | */ | |
1356 | extent_bytes = max_bytes - bitmap_bytes; | |
1357 | extent_bytes = min_t(u64, extent_bytes, div64_u64(max_bytes, 2)); | |
96303081 | 1358 | |
34d52cb6 | 1359 | ctl->extents_thresh = |
25891f79 | 1360 | div64_u64(extent_bytes, (sizeof(struct btrfs_free_space))); |
96303081 JB |
1361 | } |
1362 | ||
bb3ac5a4 MX |
1363 | static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl, |
1364 | struct btrfs_free_space *info, | |
1365 | u64 offset, u64 bytes) | |
96303081 | 1366 | { |
f38b6e75 | 1367 | unsigned long start, count; |
96303081 | 1368 | |
34d52cb6 LZ |
1369 | start = offset_to_bit(info->offset, ctl->unit, offset); |
1370 | count = bytes_to_bits(bytes, ctl->unit); | |
f38b6e75 | 1371 | BUG_ON(start + count > BITS_PER_BITMAP); |
96303081 | 1372 | |
f38b6e75 | 1373 | bitmap_clear(info->bitmap, start, count); |
96303081 JB |
1374 | |
1375 | info->bytes -= bytes; | |
bb3ac5a4 MX |
1376 | } |
1377 | ||
1378 | static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl, | |
1379 | struct btrfs_free_space *info, u64 offset, | |
1380 | u64 bytes) | |
1381 | { | |
1382 | __bitmap_clear_bits(ctl, info, offset, bytes); | |
34d52cb6 | 1383 | ctl->free_space -= bytes; |
96303081 JB |
1384 | } |
1385 | ||
34d52cb6 | 1386 | static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl, |
817d52f8 JB |
1387 | struct btrfs_free_space *info, u64 offset, |
1388 | u64 bytes) | |
96303081 | 1389 | { |
f38b6e75 | 1390 | unsigned long start, count; |
96303081 | 1391 | |
34d52cb6 LZ |
1392 | start = offset_to_bit(info->offset, ctl->unit, offset); |
1393 | count = bytes_to_bits(bytes, ctl->unit); | |
f38b6e75 | 1394 | BUG_ON(start + count > BITS_PER_BITMAP); |
96303081 | 1395 | |
f38b6e75 | 1396 | bitmap_set(info->bitmap, start, count); |
96303081 JB |
1397 | |
1398 | info->bytes += bytes; | |
34d52cb6 | 1399 | ctl->free_space += bytes; |
96303081 JB |
1400 | } |
1401 | ||
34d52cb6 | 1402 | static int search_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1403 | struct btrfs_free_space *bitmap_info, u64 *offset, |
1404 | u64 *bytes) | |
1405 | { | |
1406 | unsigned long found_bits = 0; | |
1407 | unsigned long bits, i; | |
1408 | unsigned long next_zero; | |
1409 | ||
34d52cb6 | 1410 | i = offset_to_bit(bitmap_info->offset, ctl->unit, |
96303081 | 1411 | max_t(u64, *offset, bitmap_info->offset)); |
34d52cb6 | 1412 | bits = bytes_to_bits(*bytes, ctl->unit); |
96303081 JB |
1413 | |
1414 | for (i = find_next_bit(bitmap_info->bitmap, BITS_PER_BITMAP, i); | |
1415 | i < BITS_PER_BITMAP; | |
1416 | i = find_next_bit(bitmap_info->bitmap, BITS_PER_BITMAP, i + 1)) { | |
1417 | next_zero = find_next_zero_bit(bitmap_info->bitmap, | |
1418 | BITS_PER_BITMAP, i); | |
1419 | if ((next_zero - i) >= bits) { | |
1420 | found_bits = next_zero - i; | |
1421 | break; | |
1422 | } | |
1423 | i = next_zero; | |
1424 | } | |
1425 | ||
1426 | if (found_bits) { | |
34d52cb6 LZ |
1427 | *offset = (u64)(i * ctl->unit) + bitmap_info->offset; |
1428 | *bytes = (u64)(found_bits) * ctl->unit; | |
96303081 JB |
1429 | return 0; |
1430 | } | |
1431 | ||
1432 | return -1; | |
1433 | } | |
1434 | ||
34d52cb6 LZ |
1435 | static struct btrfs_free_space * |
1436 | find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes) | |
96303081 JB |
1437 | { |
1438 | struct btrfs_free_space *entry; | |
1439 | struct rb_node *node; | |
1440 | int ret; | |
1441 | ||
34d52cb6 | 1442 | if (!ctl->free_space_offset.rb_node) |
96303081 JB |
1443 | return NULL; |
1444 | ||
34d52cb6 | 1445 | entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1); |
96303081 JB |
1446 | if (!entry) |
1447 | return NULL; | |
1448 | ||
1449 | for (node = &entry->offset_index; node; node = rb_next(node)) { | |
1450 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
1451 | if (entry->bytes < *bytes) | |
1452 | continue; | |
1453 | ||
1454 | if (entry->bitmap) { | |
34d52cb6 | 1455 | ret = search_bitmap(ctl, entry, offset, bytes); |
96303081 JB |
1456 | if (!ret) |
1457 | return entry; | |
1458 | continue; | |
1459 | } | |
1460 | ||
1461 | *offset = entry->offset; | |
1462 | *bytes = entry->bytes; | |
1463 | return entry; | |
1464 | } | |
1465 | ||
1466 | return NULL; | |
1467 | } | |
1468 | ||
34d52cb6 | 1469 | static void add_new_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1470 | struct btrfs_free_space *info, u64 offset) |
1471 | { | |
34d52cb6 | 1472 | info->offset = offset_to_bitmap(ctl, offset); |
f019f426 | 1473 | info->bytes = 0; |
f2d0f676 | 1474 | INIT_LIST_HEAD(&info->list); |
34d52cb6 LZ |
1475 | link_free_space(ctl, info); |
1476 | ctl->total_bitmaps++; | |
96303081 | 1477 | |
34d52cb6 | 1478 | ctl->op->recalc_thresholds(ctl); |
96303081 JB |
1479 | } |
1480 | ||
34d52cb6 | 1481 | static void free_bitmap(struct btrfs_free_space_ctl *ctl, |
edf6e2d1 LZ |
1482 | struct btrfs_free_space *bitmap_info) |
1483 | { | |
34d52cb6 | 1484 | unlink_free_space(ctl, bitmap_info); |
edf6e2d1 | 1485 | kfree(bitmap_info->bitmap); |
dc89e982 | 1486 | kmem_cache_free(btrfs_free_space_cachep, bitmap_info); |
34d52cb6 LZ |
1487 | ctl->total_bitmaps--; |
1488 | ctl->op->recalc_thresholds(ctl); | |
edf6e2d1 LZ |
1489 | } |
1490 | ||
34d52cb6 | 1491 | static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1492 | struct btrfs_free_space *bitmap_info, |
1493 | u64 *offset, u64 *bytes) | |
1494 | { | |
1495 | u64 end; | |
6606bb97 JB |
1496 | u64 search_start, search_bytes; |
1497 | int ret; | |
96303081 JB |
1498 | |
1499 | again: | |
34d52cb6 | 1500 | end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1; |
96303081 | 1501 | |
6606bb97 JB |
1502 | /* |
1503 | * XXX - this can go away after a few releases. | |
1504 | * | |
1505 | * since the only user of btrfs_remove_free_space is the tree logging | |
1506 | * stuff, and the only way to test that is under crash conditions, we | |
1507 | * want to have this debug stuff here just in case somethings not | |
1508 | * working. Search the bitmap for the space we are trying to use to | |
1509 | * make sure its actually there. If its not there then we need to stop | |
1510 | * because something has gone wrong. | |
1511 | */ | |
1512 | search_start = *offset; | |
1513 | search_bytes = *bytes; | |
13dbc089 | 1514 | search_bytes = min(search_bytes, end - search_start + 1); |
34d52cb6 | 1515 | ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes); |
6606bb97 JB |
1516 | BUG_ON(ret < 0 || search_start != *offset); |
1517 | ||
96303081 | 1518 | if (*offset > bitmap_info->offset && *offset + *bytes > end) { |
34d52cb6 | 1519 | bitmap_clear_bits(ctl, bitmap_info, *offset, end - *offset + 1); |
96303081 JB |
1520 | *bytes -= end - *offset + 1; |
1521 | *offset = end + 1; | |
1522 | } else if (*offset >= bitmap_info->offset && *offset + *bytes <= end) { | |
34d52cb6 | 1523 | bitmap_clear_bits(ctl, bitmap_info, *offset, *bytes); |
96303081 JB |
1524 | *bytes = 0; |
1525 | } | |
1526 | ||
1527 | if (*bytes) { | |
6606bb97 | 1528 | struct rb_node *next = rb_next(&bitmap_info->offset_index); |
edf6e2d1 | 1529 | if (!bitmap_info->bytes) |
34d52cb6 | 1530 | free_bitmap(ctl, bitmap_info); |
96303081 | 1531 | |
6606bb97 JB |
1532 | /* |
1533 | * no entry after this bitmap, but we still have bytes to | |
1534 | * remove, so something has gone wrong. | |
1535 | */ | |
1536 | if (!next) | |
96303081 JB |
1537 | return -EINVAL; |
1538 | ||
6606bb97 JB |
1539 | bitmap_info = rb_entry(next, struct btrfs_free_space, |
1540 | offset_index); | |
1541 | ||
1542 | /* | |
1543 | * if the next entry isn't a bitmap we need to return to let the | |
1544 | * extent stuff do its work. | |
1545 | */ | |
96303081 JB |
1546 | if (!bitmap_info->bitmap) |
1547 | return -EAGAIN; | |
1548 | ||
6606bb97 JB |
1549 | /* |
1550 | * Ok the next item is a bitmap, but it may not actually hold | |
1551 | * the information for the rest of this free space stuff, so | |
1552 | * look for it, and if we don't find it return so we can try | |
1553 | * everything over again. | |
1554 | */ | |
1555 | search_start = *offset; | |
1556 | search_bytes = *bytes; | |
34d52cb6 | 1557 | ret = search_bitmap(ctl, bitmap_info, &search_start, |
6606bb97 JB |
1558 | &search_bytes); |
1559 | if (ret < 0 || search_start != *offset) | |
1560 | return -EAGAIN; | |
1561 | ||
96303081 | 1562 | goto again; |
edf6e2d1 | 1563 | } else if (!bitmap_info->bytes) |
34d52cb6 | 1564 | free_bitmap(ctl, bitmap_info); |
96303081 JB |
1565 | |
1566 | return 0; | |
1567 | } | |
1568 | ||
2cdc342c JB |
1569 | static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl, |
1570 | struct btrfs_free_space *info, u64 offset, | |
1571 | u64 bytes) | |
1572 | { | |
1573 | u64 bytes_to_set = 0; | |
1574 | u64 end; | |
1575 | ||
1576 | end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit); | |
1577 | ||
1578 | bytes_to_set = min(end - offset, bytes); | |
1579 | ||
1580 | bitmap_set_bits(ctl, info, offset, bytes_to_set); | |
1581 | ||
1582 | return bytes_to_set; | |
1583 | ||
1584 | } | |
1585 | ||
34d52cb6 LZ |
1586 | static bool use_bitmap(struct btrfs_free_space_ctl *ctl, |
1587 | struct btrfs_free_space *info) | |
96303081 | 1588 | { |
34d52cb6 | 1589 | struct btrfs_block_group_cache *block_group = ctl->private; |
96303081 JB |
1590 | |
1591 | /* | |
1592 | * If we are below the extents threshold then we can add this as an | |
1593 | * extent, and don't have to deal with the bitmap | |
1594 | */ | |
34d52cb6 | 1595 | if (ctl->free_extents < ctl->extents_thresh) { |
32cb0840 JB |
1596 | /* |
1597 | * If this block group has some small extents we don't want to | |
1598 | * use up all of our free slots in the cache with them, we want | |
1599 | * to reserve them to larger extents, however if we have plent | |
1600 | * of cache left then go ahead an dadd them, no sense in adding | |
1601 | * the overhead of a bitmap if we don't have to. | |
1602 | */ | |
1603 | if (info->bytes <= block_group->sectorsize * 4) { | |
34d52cb6 LZ |
1604 | if (ctl->free_extents * 2 <= ctl->extents_thresh) |
1605 | return false; | |
32cb0840 | 1606 | } else { |
34d52cb6 | 1607 | return false; |
32cb0840 JB |
1608 | } |
1609 | } | |
96303081 JB |
1610 | |
1611 | /* | |
1612 | * some block groups are so tiny they can't be enveloped by a bitmap, so | |
1613 | * don't even bother to create a bitmap for this | |
1614 | */ | |
1615 | if (BITS_PER_BITMAP * block_group->sectorsize > | |
1616 | block_group->key.offset) | |
34d52cb6 LZ |
1617 | return false; |
1618 | ||
1619 | return true; | |
1620 | } | |
1621 | ||
2cdc342c JB |
1622 | static struct btrfs_free_space_op free_space_op = { |
1623 | .recalc_thresholds = recalculate_thresholds, | |
1624 | .use_bitmap = use_bitmap, | |
1625 | }; | |
1626 | ||
34d52cb6 LZ |
1627 | static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl, |
1628 | struct btrfs_free_space *info) | |
1629 | { | |
1630 | struct btrfs_free_space *bitmap_info; | |
2cdc342c | 1631 | struct btrfs_block_group_cache *block_group = NULL; |
34d52cb6 | 1632 | int added = 0; |
2cdc342c | 1633 | u64 bytes, offset, bytes_added; |
34d52cb6 | 1634 | int ret; |
96303081 JB |
1635 | |
1636 | bytes = info->bytes; | |
1637 | offset = info->offset; | |
1638 | ||
34d52cb6 LZ |
1639 | if (!ctl->op->use_bitmap(ctl, info)) |
1640 | return 0; | |
1641 | ||
2cdc342c JB |
1642 | if (ctl->op == &free_space_op) |
1643 | block_group = ctl->private; | |
38e87880 | 1644 | again: |
2cdc342c JB |
1645 | /* |
1646 | * Since we link bitmaps right into the cluster we need to see if we | |
1647 | * have a cluster here, and if so and it has our bitmap we need to add | |
1648 | * the free space to that bitmap. | |
1649 | */ | |
1650 | if (block_group && !list_empty(&block_group->cluster_list)) { | |
1651 | struct btrfs_free_cluster *cluster; | |
1652 | struct rb_node *node; | |
1653 | struct btrfs_free_space *entry; | |
1654 | ||
1655 | cluster = list_entry(block_group->cluster_list.next, | |
1656 | struct btrfs_free_cluster, | |
1657 | block_group_list); | |
1658 | spin_lock(&cluster->lock); | |
1659 | node = rb_first(&cluster->root); | |
1660 | if (!node) { | |
1661 | spin_unlock(&cluster->lock); | |
38e87880 | 1662 | goto no_cluster_bitmap; |
2cdc342c JB |
1663 | } |
1664 | ||
1665 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
1666 | if (!entry->bitmap) { | |
1667 | spin_unlock(&cluster->lock); | |
38e87880 | 1668 | goto no_cluster_bitmap; |
2cdc342c JB |
1669 | } |
1670 | ||
1671 | if (entry->offset == offset_to_bitmap(ctl, offset)) { | |
1672 | bytes_added = add_bytes_to_bitmap(ctl, entry, | |
1673 | offset, bytes); | |
1674 | bytes -= bytes_added; | |
1675 | offset += bytes_added; | |
1676 | } | |
1677 | spin_unlock(&cluster->lock); | |
1678 | if (!bytes) { | |
1679 | ret = 1; | |
1680 | goto out; | |
1681 | } | |
1682 | } | |
38e87880 CM |
1683 | |
1684 | no_cluster_bitmap: | |
34d52cb6 | 1685 | bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), |
96303081 JB |
1686 | 1, 0); |
1687 | if (!bitmap_info) { | |
1688 | BUG_ON(added); | |
1689 | goto new_bitmap; | |
1690 | } | |
1691 | ||
2cdc342c JB |
1692 | bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes); |
1693 | bytes -= bytes_added; | |
1694 | offset += bytes_added; | |
1695 | added = 0; | |
96303081 JB |
1696 | |
1697 | if (!bytes) { | |
1698 | ret = 1; | |
1699 | goto out; | |
1700 | } else | |
1701 | goto again; | |
1702 | ||
1703 | new_bitmap: | |
1704 | if (info && info->bitmap) { | |
34d52cb6 | 1705 | add_new_bitmap(ctl, info, offset); |
96303081 JB |
1706 | added = 1; |
1707 | info = NULL; | |
1708 | goto again; | |
1709 | } else { | |
34d52cb6 | 1710 | spin_unlock(&ctl->tree_lock); |
96303081 JB |
1711 | |
1712 | /* no pre-allocated info, allocate a new one */ | |
1713 | if (!info) { | |
dc89e982 JB |
1714 | info = kmem_cache_zalloc(btrfs_free_space_cachep, |
1715 | GFP_NOFS); | |
96303081 | 1716 | if (!info) { |
34d52cb6 | 1717 | spin_lock(&ctl->tree_lock); |
96303081 JB |
1718 | ret = -ENOMEM; |
1719 | goto out; | |
1720 | } | |
1721 | } | |
1722 | ||
1723 | /* allocate the bitmap */ | |
1724 | info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS); | |
34d52cb6 | 1725 | spin_lock(&ctl->tree_lock); |
96303081 JB |
1726 | if (!info->bitmap) { |
1727 | ret = -ENOMEM; | |
1728 | goto out; | |
1729 | } | |
1730 | goto again; | |
1731 | } | |
1732 | ||
1733 | out: | |
1734 | if (info) { | |
1735 | if (info->bitmap) | |
1736 | kfree(info->bitmap); | |
dc89e982 | 1737 | kmem_cache_free(btrfs_free_space_cachep, info); |
96303081 | 1738 | } |
0f9dd46c JB |
1739 | |
1740 | return ret; | |
1741 | } | |
1742 | ||
945d8962 | 1743 | static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl, |
f333adb5 | 1744 | struct btrfs_free_space *info, bool update_stat) |
0f9dd46c | 1745 | { |
120d66ee LZ |
1746 | struct btrfs_free_space *left_info; |
1747 | struct btrfs_free_space *right_info; | |
1748 | bool merged = false; | |
1749 | u64 offset = info->offset; | |
1750 | u64 bytes = info->bytes; | |
6226cb0a | 1751 | |
0f9dd46c JB |
1752 | /* |
1753 | * first we want to see if there is free space adjacent to the range we | |
1754 | * are adding, if there is remove that struct and add a new one to | |
1755 | * cover the entire range | |
1756 | */ | |
34d52cb6 | 1757 | right_info = tree_search_offset(ctl, offset + bytes, 0, 0); |
96303081 JB |
1758 | if (right_info && rb_prev(&right_info->offset_index)) |
1759 | left_info = rb_entry(rb_prev(&right_info->offset_index), | |
1760 | struct btrfs_free_space, offset_index); | |
1761 | else | |
34d52cb6 | 1762 | left_info = tree_search_offset(ctl, offset - 1, 0, 0); |
0f9dd46c | 1763 | |
96303081 | 1764 | if (right_info && !right_info->bitmap) { |
f333adb5 | 1765 | if (update_stat) |
34d52cb6 | 1766 | unlink_free_space(ctl, right_info); |
f333adb5 | 1767 | else |
34d52cb6 | 1768 | __unlink_free_space(ctl, right_info); |
6226cb0a | 1769 | info->bytes += right_info->bytes; |
dc89e982 | 1770 | kmem_cache_free(btrfs_free_space_cachep, right_info); |
120d66ee | 1771 | merged = true; |
0f9dd46c JB |
1772 | } |
1773 | ||
96303081 JB |
1774 | if (left_info && !left_info->bitmap && |
1775 | left_info->offset + left_info->bytes == offset) { | |
f333adb5 | 1776 | if (update_stat) |
34d52cb6 | 1777 | unlink_free_space(ctl, left_info); |
f333adb5 | 1778 | else |
34d52cb6 | 1779 | __unlink_free_space(ctl, left_info); |
6226cb0a JB |
1780 | info->offset = left_info->offset; |
1781 | info->bytes += left_info->bytes; | |
dc89e982 | 1782 | kmem_cache_free(btrfs_free_space_cachep, left_info); |
120d66ee | 1783 | merged = true; |
0f9dd46c JB |
1784 | } |
1785 | ||
120d66ee LZ |
1786 | return merged; |
1787 | } | |
1788 | ||
581bb050 LZ |
1789 | int __btrfs_add_free_space(struct btrfs_free_space_ctl *ctl, |
1790 | u64 offset, u64 bytes) | |
120d66ee LZ |
1791 | { |
1792 | struct btrfs_free_space *info; | |
1793 | int ret = 0; | |
1794 | ||
dc89e982 | 1795 | info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS); |
120d66ee LZ |
1796 | if (!info) |
1797 | return -ENOMEM; | |
1798 | ||
1799 | info->offset = offset; | |
1800 | info->bytes = bytes; | |
1801 | ||
34d52cb6 | 1802 | spin_lock(&ctl->tree_lock); |
120d66ee | 1803 | |
34d52cb6 | 1804 | if (try_merge_free_space(ctl, info, true)) |
120d66ee LZ |
1805 | goto link; |
1806 | ||
1807 | /* | |
1808 | * There was no extent directly to the left or right of this new | |
1809 | * extent then we know we're going to have to allocate a new extent, so | |
1810 | * before we do that see if we need to drop this into a bitmap | |
1811 | */ | |
34d52cb6 | 1812 | ret = insert_into_bitmap(ctl, info); |
120d66ee LZ |
1813 | if (ret < 0) { |
1814 | goto out; | |
1815 | } else if (ret) { | |
1816 | ret = 0; | |
1817 | goto out; | |
1818 | } | |
1819 | link: | |
34d52cb6 | 1820 | ret = link_free_space(ctl, info); |
0f9dd46c | 1821 | if (ret) |
dc89e982 | 1822 | kmem_cache_free(btrfs_free_space_cachep, info); |
96303081 | 1823 | out: |
34d52cb6 | 1824 | spin_unlock(&ctl->tree_lock); |
6226cb0a | 1825 | |
0f9dd46c | 1826 | if (ret) { |
96303081 | 1827 | printk(KERN_CRIT "btrfs: unable to add free space :%d\n", ret); |
c293498b | 1828 | BUG_ON(ret == -EEXIST); |
0f9dd46c JB |
1829 | } |
1830 | ||
0f9dd46c JB |
1831 | return ret; |
1832 | } | |
1833 | ||
6226cb0a JB |
1834 | int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group, |
1835 | u64 offset, u64 bytes) | |
0f9dd46c | 1836 | { |
34d52cb6 | 1837 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0f9dd46c | 1838 | struct btrfs_free_space *info; |
96303081 | 1839 | struct btrfs_free_space *next_info = NULL; |
0f9dd46c JB |
1840 | int ret = 0; |
1841 | ||
34d52cb6 | 1842 | spin_lock(&ctl->tree_lock); |
6226cb0a | 1843 | |
96303081 | 1844 | again: |
34d52cb6 | 1845 | info = tree_search_offset(ctl, offset, 0, 0); |
96303081 | 1846 | if (!info) { |
6606bb97 JB |
1847 | /* |
1848 | * oops didn't find an extent that matched the space we wanted | |
1849 | * to remove, look for a bitmap instead | |
1850 | */ | |
34d52cb6 | 1851 | info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), |
6606bb97 JB |
1852 | 1, 0); |
1853 | if (!info) { | |
24a70313 CM |
1854 | /* the tree logging code might be calling us before we |
1855 | * have fully loaded the free space rbtree for this | |
1856 | * block group. So it is possible the entry won't | |
1857 | * be in the rbtree yet at all. The caching code | |
1858 | * will make sure not to put it in the rbtree if | |
1859 | * the logging code has pinned it. | |
1860 | */ | |
6606bb97 JB |
1861 | goto out_lock; |
1862 | } | |
96303081 JB |
1863 | } |
1864 | ||
1865 | if (info->bytes < bytes && rb_next(&info->offset_index)) { | |
1866 | u64 end; | |
1867 | next_info = rb_entry(rb_next(&info->offset_index), | |
1868 | struct btrfs_free_space, | |
1869 | offset_index); | |
1870 | ||
1871 | if (next_info->bitmap) | |
34d52cb6 LZ |
1872 | end = next_info->offset + |
1873 | BITS_PER_BITMAP * ctl->unit - 1; | |
96303081 JB |
1874 | else |
1875 | end = next_info->offset + next_info->bytes; | |
1876 | ||
1877 | if (next_info->bytes < bytes || | |
1878 | next_info->offset > offset || offset > end) { | |
1879 | printk(KERN_CRIT "Found free space at %llu, size %llu," | |
1880 | " trying to use %llu\n", | |
1881 | (unsigned long long)info->offset, | |
1882 | (unsigned long long)info->bytes, | |
1883 | (unsigned long long)bytes); | |
0f9dd46c JB |
1884 | WARN_ON(1); |
1885 | ret = -EINVAL; | |
96303081 | 1886 | goto out_lock; |
0f9dd46c | 1887 | } |
0f9dd46c | 1888 | |
96303081 JB |
1889 | info = next_info; |
1890 | } | |
1891 | ||
1892 | if (info->bytes == bytes) { | |
34d52cb6 | 1893 | unlink_free_space(ctl, info); |
96303081 JB |
1894 | if (info->bitmap) { |
1895 | kfree(info->bitmap); | |
34d52cb6 | 1896 | ctl->total_bitmaps--; |
0f9dd46c | 1897 | } |
dc89e982 | 1898 | kmem_cache_free(btrfs_free_space_cachep, info); |
1eae31e9 | 1899 | ret = 0; |
96303081 JB |
1900 | goto out_lock; |
1901 | } | |
0f9dd46c | 1902 | |
96303081 | 1903 | if (!info->bitmap && info->offset == offset) { |
34d52cb6 | 1904 | unlink_free_space(ctl, info); |
0f9dd46c JB |
1905 | info->offset += bytes; |
1906 | info->bytes -= bytes; | |
1eae31e9 CM |
1907 | ret = link_free_space(ctl, info); |
1908 | WARN_ON(ret); | |
96303081 JB |
1909 | goto out_lock; |
1910 | } | |
0f9dd46c | 1911 | |
96303081 JB |
1912 | if (!info->bitmap && info->offset <= offset && |
1913 | info->offset + info->bytes >= offset + bytes) { | |
9b49c9b9 CM |
1914 | u64 old_start = info->offset; |
1915 | /* | |
1916 | * we're freeing space in the middle of the info, | |
1917 | * this can happen during tree log replay | |
1918 | * | |
1919 | * first unlink the old info and then | |
1920 | * insert it again after the hole we're creating | |
1921 | */ | |
34d52cb6 | 1922 | unlink_free_space(ctl, info); |
9b49c9b9 CM |
1923 | if (offset + bytes < info->offset + info->bytes) { |
1924 | u64 old_end = info->offset + info->bytes; | |
1925 | ||
1926 | info->offset = offset + bytes; | |
1927 | info->bytes = old_end - info->offset; | |
34d52cb6 | 1928 | ret = link_free_space(ctl, info); |
96303081 JB |
1929 | WARN_ON(ret); |
1930 | if (ret) | |
1931 | goto out_lock; | |
9b49c9b9 CM |
1932 | } else { |
1933 | /* the hole we're creating ends at the end | |
1934 | * of the info struct, just free the info | |
1935 | */ | |
dc89e982 | 1936 | kmem_cache_free(btrfs_free_space_cachep, info); |
9b49c9b9 | 1937 | } |
34d52cb6 | 1938 | spin_unlock(&ctl->tree_lock); |
96303081 JB |
1939 | |
1940 | /* step two, insert a new info struct to cover | |
1941 | * anything before the hole | |
9b49c9b9 | 1942 | */ |
6226cb0a JB |
1943 | ret = btrfs_add_free_space(block_group, old_start, |
1944 | offset - old_start); | |
96303081 JB |
1945 | WARN_ON(ret); |
1946 | goto out; | |
0f9dd46c | 1947 | } |
96303081 | 1948 | |
34d52cb6 | 1949 | ret = remove_from_bitmap(ctl, info, &offset, &bytes); |
96303081 JB |
1950 | if (ret == -EAGAIN) |
1951 | goto again; | |
1952 | BUG_ON(ret); | |
1953 | out_lock: | |
34d52cb6 | 1954 | spin_unlock(&ctl->tree_lock); |
0f9dd46c | 1955 | out: |
25179201 JB |
1956 | return ret; |
1957 | } | |
1958 | ||
0f9dd46c JB |
1959 | void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group, |
1960 | u64 bytes) | |
1961 | { | |
34d52cb6 | 1962 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0f9dd46c JB |
1963 | struct btrfs_free_space *info; |
1964 | struct rb_node *n; | |
1965 | int count = 0; | |
1966 | ||
34d52cb6 | 1967 | for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) { |
0f9dd46c JB |
1968 | info = rb_entry(n, struct btrfs_free_space, offset_index); |
1969 | if (info->bytes >= bytes) | |
1970 | count++; | |
96303081 | 1971 | printk(KERN_CRIT "entry offset %llu, bytes %llu, bitmap %s\n", |
21380931 | 1972 | (unsigned long long)info->offset, |
96303081 JB |
1973 | (unsigned long long)info->bytes, |
1974 | (info->bitmap) ? "yes" : "no"); | |
0f9dd46c | 1975 | } |
96303081 JB |
1976 | printk(KERN_INFO "block group has cluster?: %s\n", |
1977 | list_empty(&block_group->cluster_list) ? "no" : "yes"); | |
0f9dd46c JB |
1978 | printk(KERN_INFO "%d blocks of free space at or bigger than bytes is" |
1979 | "\n", count); | |
1980 | } | |
1981 | ||
34d52cb6 | 1982 | void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group) |
0f9dd46c | 1983 | { |
34d52cb6 | 1984 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0f9dd46c | 1985 | |
34d52cb6 LZ |
1986 | spin_lock_init(&ctl->tree_lock); |
1987 | ctl->unit = block_group->sectorsize; | |
1988 | ctl->start = block_group->key.objectid; | |
1989 | ctl->private = block_group; | |
1990 | ctl->op = &free_space_op; | |
0f9dd46c | 1991 | |
34d52cb6 LZ |
1992 | /* |
1993 | * we only want to have 32k of ram per block group for keeping | |
1994 | * track of free space, and if we pass 1/2 of that we want to | |
1995 | * start converting things over to using bitmaps | |
1996 | */ | |
1997 | ctl->extents_thresh = ((1024 * 32) / 2) / | |
1998 | sizeof(struct btrfs_free_space); | |
0f9dd46c JB |
1999 | } |
2000 | ||
fa9c0d79 CM |
2001 | /* |
2002 | * for a given cluster, put all of its extents back into the free | |
2003 | * space cache. If the block group passed doesn't match the block group | |
2004 | * pointed to by the cluster, someone else raced in and freed the | |
2005 | * cluster already. In that case, we just return without changing anything | |
2006 | */ | |
2007 | static int | |
2008 | __btrfs_return_cluster_to_free_space( | |
2009 | struct btrfs_block_group_cache *block_group, | |
2010 | struct btrfs_free_cluster *cluster) | |
2011 | { | |
34d52cb6 | 2012 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
fa9c0d79 CM |
2013 | struct btrfs_free_space *entry; |
2014 | struct rb_node *node; | |
2015 | ||
2016 | spin_lock(&cluster->lock); | |
2017 | if (cluster->block_group != block_group) | |
2018 | goto out; | |
2019 | ||
96303081 | 2020 | cluster->block_group = NULL; |
fa9c0d79 | 2021 | cluster->window_start = 0; |
96303081 | 2022 | list_del_init(&cluster->block_group_list); |
96303081 | 2023 | |
fa9c0d79 | 2024 | node = rb_first(&cluster->root); |
96303081 | 2025 | while (node) { |
4e69b598 JB |
2026 | bool bitmap; |
2027 | ||
fa9c0d79 CM |
2028 | entry = rb_entry(node, struct btrfs_free_space, offset_index); |
2029 | node = rb_next(&entry->offset_index); | |
2030 | rb_erase(&entry->offset_index, &cluster->root); | |
4e69b598 JB |
2031 | |
2032 | bitmap = (entry->bitmap != NULL); | |
2033 | if (!bitmap) | |
34d52cb6 LZ |
2034 | try_merge_free_space(ctl, entry, false); |
2035 | tree_insert_offset(&ctl->free_space_offset, | |
4e69b598 | 2036 | entry->offset, &entry->offset_index, bitmap); |
fa9c0d79 | 2037 | } |
6bef4d31 | 2038 | cluster->root = RB_ROOT; |
96303081 | 2039 | |
fa9c0d79 CM |
2040 | out: |
2041 | spin_unlock(&cluster->lock); | |
96303081 | 2042 | btrfs_put_block_group(block_group); |
fa9c0d79 CM |
2043 | return 0; |
2044 | } | |
2045 | ||
09655373 | 2046 | void __btrfs_remove_free_space_cache_locked(struct btrfs_free_space_ctl *ctl) |
0f9dd46c JB |
2047 | { |
2048 | struct btrfs_free_space *info; | |
2049 | struct rb_node *node; | |
581bb050 | 2050 | |
581bb050 LZ |
2051 | while ((node = rb_last(&ctl->free_space_offset)) != NULL) { |
2052 | info = rb_entry(node, struct btrfs_free_space, offset_index); | |
9b90f513 JB |
2053 | if (!info->bitmap) { |
2054 | unlink_free_space(ctl, info); | |
2055 | kmem_cache_free(btrfs_free_space_cachep, info); | |
2056 | } else { | |
2057 | free_bitmap(ctl, info); | |
2058 | } | |
581bb050 LZ |
2059 | if (need_resched()) { |
2060 | spin_unlock(&ctl->tree_lock); | |
2061 | cond_resched(); | |
2062 | spin_lock(&ctl->tree_lock); | |
2063 | } | |
2064 | } | |
09655373 CM |
2065 | } |
2066 | ||
2067 | void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl) | |
2068 | { | |
2069 | spin_lock(&ctl->tree_lock); | |
2070 | __btrfs_remove_free_space_cache_locked(ctl); | |
581bb050 LZ |
2071 | spin_unlock(&ctl->tree_lock); |
2072 | } | |
2073 | ||
2074 | void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group) | |
2075 | { | |
2076 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
fa9c0d79 | 2077 | struct btrfs_free_cluster *cluster; |
96303081 | 2078 | struct list_head *head; |
0f9dd46c | 2079 | |
34d52cb6 | 2080 | spin_lock(&ctl->tree_lock); |
96303081 JB |
2081 | while ((head = block_group->cluster_list.next) != |
2082 | &block_group->cluster_list) { | |
2083 | cluster = list_entry(head, struct btrfs_free_cluster, | |
2084 | block_group_list); | |
fa9c0d79 CM |
2085 | |
2086 | WARN_ON(cluster->block_group != block_group); | |
2087 | __btrfs_return_cluster_to_free_space(block_group, cluster); | |
96303081 | 2088 | if (need_resched()) { |
34d52cb6 | 2089 | spin_unlock(&ctl->tree_lock); |
96303081 | 2090 | cond_resched(); |
34d52cb6 | 2091 | spin_lock(&ctl->tree_lock); |
96303081 | 2092 | } |
fa9c0d79 | 2093 | } |
09655373 | 2094 | __btrfs_remove_free_space_cache_locked(ctl); |
34d52cb6 | 2095 | spin_unlock(&ctl->tree_lock); |
fa9c0d79 | 2096 | |
0f9dd46c JB |
2097 | } |
2098 | ||
6226cb0a JB |
2099 | u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group, |
2100 | u64 offset, u64 bytes, u64 empty_size) | |
0f9dd46c | 2101 | { |
34d52cb6 | 2102 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
6226cb0a | 2103 | struct btrfs_free_space *entry = NULL; |
96303081 | 2104 | u64 bytes_search = bytes + empty_size; |
6226cb0a | 2105 | u64 ret = 0; |
0f9dd46c | 2106 | |
34d52cb6 LZ |
2107 | spin_lock(&ctl->tree_lock); |
2108 | entry = find_free_space(ctl, &offset, &bytes_search); | |
6226cb0a | 2109 | if (!entry) |
96303081 JB |
2110 | goto out; |
2111 | ||
2112 | ret = offset; | |
2113 | if (entry->bitmap) { | |
34d52cb6 | 2114 | bitmap_clear_bits(ctl, entry, offset, bytes); |
edf6e2d1 | 2115 | if (!entry->bytes) |
34d52cb6 | 2116 | free_bitmap(ctl, entry); |
96303081 | 2117 | } else { |
34d52cb6 | 2118 | unlink_free_space(ctl, entry); |
6226cb0a JB |
2119 | entry->offset += bytes; |
2120 | entry->bytes -= bytes; | |
6226cb0a | 2121 | if (!entry->bytes) |
dc89e982 | 2122 | kmem_cache_free(btrfs_free_space_cachep, entry); |
6226cb0a | 2123 | else |
34d52cb6 | 2124 | link_free_space(ctl, entry); |
6226cb0a | 2125 | } |
0f9dd46c | 2126 | |
96303081 | 2127 | out: |
34d52cb6 | 2128 | spin_unlock(&ctl->tree_lock); |
817d52f8 | 2129 | |
0f9dd46c JB |
2130 | return ret; |
2131 | } | |
fa9c0d79 CM |
2132 | |
2133 | /* | |
2134 | * given a cluster, put all of its extents back into the free space | |
2135 | * cache. If a block group is passed, this function will only free | |
2136 | * a cluster that belongs to the passed block group. | |
2137 | * | |
2138 | * Otherwise, it'll get a reference on the block group pointed to by the | |
2139 | * cluster and remove the cluster from it. | |
2140 | */ | |
2141 | int btrfs_return_cluster_to_free_space( | |
2142 | struct btrfs_block_group_cache *block_group, | |
2143 | struct btrfs_free_cluster *cluster) | |
2144 | { | |
34d52cb6 | 2145 | struct btrfs_free_space_ctl *ctl; |
fa9c0d79 CM |
2146 | int ret; |
2147 | ||
2148 | /* first, get a safe pointer to the block group */ | |
2149 | spin_lock(&cluster->lock); | |
2150 | if (!block_group) { | |
2151 | block_group = cluster->block_group; | |
2152 | if (!block_group) { | |
2153 | spin_unlock(&cluster->lock); | |
2154 | return 0; | |
2155 | } | |
2156 | } else if (cluster->block_group != block_group) { | |
2157 | /* someone else has already freed it don't redo their work */ | |
2158 | spin_unlock(&cluster->lock); | |
2159 | return 0; | |
2160 | } | |
2161 | atomic_inc(&block_group->count); | |
2162 | spin_unlock(&cluster->lock); | |
2163 | ||
34d52cb6 LZ |
2164 | ctl = block_group->free_space_ctl; |
2165 | ||
fa9c0d79 | 2166 | /* now return any extents the cluster had on it */ |
34d52cb6 | 2167 | spin_lock(&ctl->tree_lock); |
fa9c0d79 | 2168 | ret = __btrfs_return_cluster_to_free_space(block_group, cluster); |
34d52cb6 | 2169 | spin_unlock(&ctl->tree_lock); |
fa9c0d79 CM |
2170 | |
2171 | /* finally drop our ref */ | |
2172 | btrfs_put_block_group(block_group); | |
2173 | return ret; | |
2174 | } | |
2175 | ||
96303081 JB |
2176 | static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group, |
2177 | struct btrfs_free_cluster *cluster, | |
4e69b598 | 2178 | struct btrfs_free_space *entry, |
96303081 JB |
2179 | u64 bytes, u64 min_start) |
2180 | { | |
34d52cb6 | 2181 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
96303081 JB |
2182 | int err; |
2183 | u64 search_start = cluster->window_start; | |
2184 | u64 search_bytes = bytes; | |
2185 | u64 ret = 0; | |
2186 | ||
96303081 JB |
2187 | search_start = min_start; |
2188 | search_bytes = bytes; | |
2189 | ||
34d52cb6 | 2190 | err = search_bitmap(ctl, entry, &search_start, &search_bytes); |
96303081 | 2191 | if (err) |
4e69b598 | 2192 | return 0; |
96303081 JB |
2193 | |
2194 | ret = search_start; | |
bb3ac5a4 | 2195 | __bitmap_clear_bits(ctl, entry, ret, bytes); |
96303081 JB |
2196 | |
2197 | return ret; | |
2198 | } | |
2199 | ||
fa9c0d79 CM |
2200 | /* |
2201 | * given a cluster, try to allocate 'bytes' from it, returns 0 | |
2202 | * if it couldn't find anything suitably large, or a logical disk offset | |
2203 | * if things worked out | |
2204 | */ | |
2205 | u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group, | |
2206 | struct btrfs_free_cluster *cluster, u64 bytes, | |
2207 | u64 min_start) | |
2208 | { | |
34d52cb6 | 2209 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
fa9c0d79 CM |
2210 | struct btrfs_free_space *entry = NULL; |
2211 | struct rb_node *node; | |
2212 | u64 ret = 0; | |
2213 | ||
2214 | spin_lock(&cluster->lock); | |
2215 | if (bytes > cluster->max_size) | |
2216 | goto out; | |
2217 | ||
2218 | if (cluster->block_group != block_group) | |
2219 | goto out; | |
2220 | ||
2221 | node = rb_first(&cluster->root); | |
2222 | if (!node) | |
2223 | goto out; | |
2224 | ||
2225 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
fa9c0d79 | 2226 | while(1) { |
4e69b598 JB |
2227 | if (entry->bytes < bytes || |
2228 | (!entry->bitmap && entry->offset < min_start)) { | |
fa9c0d79 CM |
2229 | node = rb_next(&entry->offset_index); |
2230 | if (!node) | |
2231 | break; | |
2232 | entry = rb_entry(node, struct btrfs_free_space, | |
2233 | offset_index); | |
2234 | continue; | |
2235 | } | |
fa9c0d79 | 2236 | |
4e69b598 JB |
2237 | if (entry->bitmap) { |
2238 | ret = btrfs_alloc_from_bitmap(block_group, | |
2239 | cluster, entry, bytes, | |
2240 | min_start); | |
2241 | if (ret == 0) { | |
4e69b598 JB |
2242 | node = rb_next(&entry->offset_index); |
2243 | if (!node) | |
2244 | break; | |
2245 | entry = rb_entry(node, struct btrfs_free_space, | |
2246 | offset_index); | |
2247 | continue; | |
2248 | } | |
2249 | } else { | |
4e69b598 JB |
2250 | ret = entry->offset; |
2251 | ||
2252 | entry->offset += bytes; | |
2253 | entry->bytes -= bytes; | |
2254 | } | |
fa9c0d79 | 2255 | |
5e71b5d5 | 2256 | if (entry->bytes == 0) |
fa9c0d79 | 2257 | rb_erase(&entry->offset_index, &cluster->root); |
fa9c0d79 CM |
2258 | break; |
2259 | } | |
2260 | out: | |
2261 | spin_unlock(&cluster->lock); | |
96303081 | 2262 | |
5e71b5d5 LZ |
2263 | if (!ret) |
2264 | return 0; | |
2265 | ||
34d52cb6 | 2266 | spin_lock(&ctl->tree_lock); |
5e71b5d5 | 2267 | |
34d52cb6 | 2268 | ctl->free_space -= bytes; |
5e71b5d5 | 2269 | if (entry->bytes == 0) { |
34d52cb6 | 2270 | ctl->free_extents--; |
4e69b598 JB |
2271 | if (entry->bitmap) { |
2272 | kfree(entry->bitmap); | |
34d52cb6 LZ |
2273 | ctl->total_bitmaps--; |
2274 | ctl->op->recalc_thresholds(ctl); | |
4e69b598 | 2275 | } |
dc89e982 | 2276 | kmem_cache_free(btrfs_free_space_cachep, entry); |
5e71b5d5 LZ |
2277 | } |
2278 | ||
34d52cb6 | 2279 | spin_unlock(&ctl->tree_lock); |
5e71b5d5 | 2280 | |
fa9c0d79 CM |
2281 | return ret; |
2282 | } | |
2283 | ||
96303081 JB |
2284 | static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group, |
2285 | struct btrfs_free_space *entry, | |
2286 | struct btrfs_free_cluster *cluster, | |
2287 | u64 offset, u64 bytes, u64 min_bytes) | |
2288 | { | |
34d52cb6 | 2289 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
96303081 JB |
2290 | unsigned long next_zero; |
2291 | unsigned long i; | |
2292 | unsigned long search_bits; | |
2293 | unsigned long total_bits; | |
2294 | unsigned long found_bits; | |
2295 | unsigned long start = 0; | |
2296 | unsigned long total_found = 0; | |
4e69b598 | 2297 | int ret; |
96303081 JB |
2298 | bool found = false; |
2299 | ||
2300 | i = offset_to_bit(entry->offset, block_group->sectorsize, | |
2301 | max_t(u64, offset, entry->offset)); | |
d0a365e8 JB |
2302 | search_bits = bytes_to_bits(bytes, block_group->sectorsize); |
2303 | total_bits = bytes_to_bits(min_bytes, block_group->sectorsize); | |
96303081 JB |
2304 | |
2305 | again: | |
2306 | found_bits = 0; | |
2307 | for (i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, i); | |
2308 | i < BITS_PER_BITMAP; | |
2309 | i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, i + 1)) { | |
2310 | next_zero = find_next_zero_bit(entry->bitmap, | |
2311 | BITS_PER_BITMAP, i); | |
2312 | if (next_zero - i >= search_bits) { | |
2313 | found_bits = next_zero - i; | |
2314 | break; | |
2315 | } | |
2316 | i = next_zero; | |
2317 | } | |
2318 | ||
2319 | if (!found_bits) | |
4e69b598 | 2320 | return -ENOSPC; |
96303081 JB |
2321 | |
2322 | if (!found) { | |
2323 | start = i; | |
b78d09bc | 2324 | cluster->max_size = 0; |
96303081 JB |
2325 | found = true; |
2326 | } | |
2327 | ||
2328 | total_found += found_bits; | |
2329 | ||
2330 | if (cluster->max_size < found_bits * block_group->sectorsize) | |
2331 | cluster->max_size = found_bits * block_group->sectorsize; | |
2332 | ||
2333 | if (total_found < total_bits) { | |
2334 | i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, next_zero); | |
2335 | if (i - start > total_bits * 2) { | |
2336 | total_found = 0; | |
2337 | cluster->max_size = 0; | |
2338 | found = false; | |
2339 | } | |
2340 | goto again; | |
2341 | } | |
2342 | ||
2343 | cluster->window_start = start * block_group->sectorsize + | |
2344 | entry->offset; | |
34d52cb6 | 2345 | rb_erase(&entry->offset_index, &ctl->free_space_offset); |
4e69b598 JB |
2346 | ret = tree_insert_offset(&cluster->root, entry->offset, |
2347 | &entry->offset_index, 1); | |
2348 | BUG_ON(ret); | |
96303081 JB |
2349 | |
2350 | return 0; | |
2351 | } | |
2352 | ||
4e69b598 JB |
2353 | /* |
2354 | * This searches the block group for just extents to fill the cluster with. | |
2355 | */ | |
3de85bb9 JB |
2356 | static noinline int |
2357 | setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group, | |
2358 | struct btrfs_free_cluster *cluster, | |
2359 | struct list_head *bitmaps, u64 offset, u64 bytes, | |
2360 | u64 min_bytes) | |
4e69b598 | 2361 | { |
34d52cb6 | 2362 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
4e69b598 JB |
2363 | struct btrfs_free_space *first = NULL; |
2364 | struct btrfs_free_space *entry = NULL; | |
2365 | struct btrfs_free_space *prev = NULL; | |
2366 | struct btrfs_free_space *last; | |
2367 | struct rb_node *node; | |
2368 | u64 window_start; | |
2369 | u64 window_free; | |
2370 | u64 max_extent; | |
2371 | u64 max_gap = 128 * 1024; | |
2372 | ||
34d52cb6 | 2373 | entry = tree_search_offset(ctl, offset, 0, 1); |
4e69b598 JB |
2374 | if (!entry) |
2375 | return -ENOSPC; | |
2376 | ||
2377 | /* | |
2378 | * We don't want bitmaps, so just move along until we find a normal | |
2379 | * extent entry. | |
2380 | */ | |
2381 | while (entry->bitmap) { | |
86d4a77b JB |
2382 | if (list_empty(&entry->list)) |
2383 | list_add_tail(&entry->list, bitmaps); | |
4e69b598 JB |
2384 | node = rb_next(&entry->offset_index); |
2385 | if (!node) | |
2386 | return -ENOSPC; | |
2387 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
2388 | } | |
2389 | ||
2390 | window_start = entry->offset; | |
2391 | window_free = entry->bytes; | |
2392 | max_extent = entry->bytes; | |
2393 | first = entry; | |
2394 | last = entry; | |
2395 | prev = entry; | |
2396 | ||
2397 | while (window_free <= min_bytes) { | |
2398 | node = rb_next(&entry->offset_index); | |
2399 | if (!node) | |
2400 | return -ENOSPC; | |
2401 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
2402 | ||
86d4a77b JB |
2403 | if (entry->bitmap) { |
2404 | if (list_empty(&entry->list)) | |
2405 | list_add_tail(&entry->list, bitmaps); | |
4e69b598 | 2406 | continue; |
86d4a77b JB |
2407 | } |
2408 | ||
4e69b598 JB |
2409 | /* |
2410 | * we haven't filled the empty size and the window is | |
2411 | * very large. reset and try again | |
2412 | */ | |
2413 | if (entry->offset - (prev->offset + prev->bytes) > max_gap || | |
2414 | entry->offset - window_start > (min_bytes * 2)) { | |
2415 | first = entry; | |
2416 | window_start = entry->offset; | |
2417 | window_free = entry->bytes; | |
2418 | last = entry; | |
2419 | max_extent = entry->bytes; | |
2420 | } else { | |
2421 | last = entry; | |
2422 | window_free += entry->bytes; | |
2423 | if (entry->bytes > max_extent) | |
2424 | max_extent = entry->bytes; | |
2425 | } | |
2426 | prev = entry; | |
2427 | } | |
2428 | ||
2429 | cluster->window_start = first->offset; | |
2430 | ||
2431 | node = &first->offset_index; | |
2432 | ||
2433 | /* | |
2434 | * now we've found our entries, pull them out of the free space | |
2435 | * cache and put them into the cluster rbtree | |
2436 | */ | |
2437 | do { | |
2438 | int ret; | |
2439 | ||
2440 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
2441 | node = rb_next(&entry->offset_index); | |
2442 | if (entry->bitmap) | |
2443 | continue; | |
2444 | ||
34d52cb6 | 2445 | rb_erase(&entry->offset_index, &ctl->free_space_offset); |
4e69b598 JB |
2446 | ret = tree_insert_offset(&cluster->root, entry->offset, |
2447 | &entry->offset_index, 0); | |
2448 | BUG_ON(ret); | |
2449 | } while (node && entry != last); | |
2450 | ||
2451 | cluster->max_size = max_extent; | |
2452 | ||
2453 | return 0; | |
2454 | } | |
2455 | ||
2456 | /* | |
2457 | * This specifically looks for bitmaps that may work in the cluster, we assume | |
2458 | * that we have already failed to find extents that will work. | |
2459 | */ | |
3de85bb9 JB |
2460 | static noinline int |
2461 | setup_cluster_bitmap(struct btrfs_block_group_cache *block_group, | |
2462 | struct btrfs_free_cluster *cluster, | |
2463 | struct list_head *bitmaps, u64 offset, u64 bytes, | |
2464 | u64 min_bytes) | |
4e69b598 | 2465 | { |
34d52cb6 | 2466 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
4e69b598 | 2467 | struct btrfs_free_space *entry; |
4e69b598 | 2468 | int ret = -ENOSPC; |
0f0fbf1d | 2469 | u64 bitmap_offset = offset_to_bitmap(ctl, offset); |
4e69b598 | 2470 | |
34d52cb6 | 2471 | if (ctl->total_bitmaps == 0) |
4e69b598 JB |
2472 | return -ENOSPC; |
2473 | ||
0f0fbf1d LZ |
2474 | /* |
2475 | * The bitmap that covers offset won't be in the list unless offset | |
2476 | * is just its start offset. | |
2477 | */ | |
2478 | entry = list_first_entry(bitmaps, struct btrfs_free_space, list); | |
2479 | if (entry->offset != bitmap_offset) { | |
2480 | entry = tree_search_offset(ctl, bitmap_offset, 1, 0); | |
2481 | if (entry && list_empty(&entry->list)) | |
2482 | list_add(&entry->list, bitmaps); | |
2483 | } | |
2484 | ||
86d4a77b JB |
2485 | list_for_each_entry(entry, bitmaps, list) { |
2486 | if (entry->bytes < min_bytes) | |
2487 | continue; | |
2488 | ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset, | |
2489 | bytes, min_bytes); | |
2490 | if (!ret) | |
2491 | return 0; | |
2492 | } | |
2493 | ||
2494 | /* | |
52621cb6 LZ |
2495 | * The bitmaps list has all the bitmaps that record free space |
2496 | * starting after offset, so no more search is required. | |
86d4a77b | 2497 | */ |
52621cb6 | 2498 | return -ENOSPC; |
4e69b598 JB |
2499 | } |
2500 | ||
fa9c0d79 CM |
2501 | /* |
2502 | * here we try to find a cluster of blocks in a block group. The goal | |
2503 | * is to find at least bytes free and up to empty_size + bytes free. | |
2504 | * We might not find them all in one contiguous area. | |
2505 | * | |
2506 | * returns zero and sets up cluster if things worked out, otherwise | |
2507 | * it returns -enospc | |
2508 | */ | |
2509 | int btrfs_find_space_cluster(struct btrfs_trans_handle *trans, | |
451d7585 | 2510 | struct btrfs_root *root, |
fa9c0d79 CM |
2511 | struct btrfs_block_group_cache *block_group, |
2512 | struct btrfs_free_cluster *cluster, | |
2513 | u64 offset, u64 bytes, u64 empty_size) | |
2514 | { | |
34d52cb6 | 2515 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
86d4a77b | 2516 | struct btrfs_free_space *entry, *tmp; |
52621cb6 | 2517 | LIST_HEAD(bitmaps); |
fa9c0d79 | 2518 | u64 min_bytes; |
fa9c0d79 CM |
2519 | int ret; |
2520 | ||
2521 | /* for metadata, allow allocates with more holes */ | |
451d7585 CM |
2522 | if (btrfs_test_opt(root, SSD_SPREAD)) { |
2523 | min_bytes = bytes + empty_size; | |
2524 | } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) { | |
fa9c0d79 CM |
2525 | /* |
2526 | * we want to do larger allocations when we are | |
2527 | * flushing out the delayed refs, it helps prevent | |
2528 | * making more work as we go along. | |
2529 | */ | |
2530 | if (trans->transaction->delayed_refs.flushing) | |
2531 | min_bytes = max(bytes, (bytes + empty_size) >> 1); | |
2532 | else | |
2533 | min_bytes = max(bytes, (bytes + empty_size) >> 4); | |
2534 | } else | |
2535 | min_bytes = max(bytes, (bytes + empty_size) >> 2); | |
2536 | ||
34d52cb6 | 2537 | spin_lock(&ctl->tree_lock); |
7d0d2e8e JB |
2538 | |
2539 | /* | |
2540 | * If we know we don't have enough space to make a cluster don't even | |
2541 | * bother doing all the work to try and find one. | |
2542 | */ | |
34d52cb6 LZ |
2543 | if (ctl->free_space < min_bytes) { |
2544 | spin_unlock(&ctl->tree_lock); | |
7d0d2e8e JB |
2545 | return -ENOSPC; |
2546 | } | |
2547 | ||
fa9c0d79 CM |
2548 | spin_lock(&cluster->lock); |
2549 | ||
2550 | /* someone already found a cluster, hooray */ | |
2551 | if (cluster->block_group) { | |
2552 | ret = 0; | |
2553 | goto out; | |
2554 | } | |
fa9c0d79 | 2555 | |
86d4a77b JB |
2556 | ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset, |
2557 | bytes, min_bytes); | |
4e69b598 | 2558 | if (ret) |
86d4a77b JB |
2559 | ret = setup_cluster_bitmap(block_group, cluster, &bitmaps, |
2560 | offset, bytes, min_bytes); | |
2561 | ||
2562 | /* Clear our temporary list */ | |
2563 | list_for_each_entry_safe(entry, tmp, &bitmaps, list) | |
2564 | list_del_init(&entry->list); | |
fa9c0d79 | 2565 | |
4e69b598 JB |
2566 | if (!ret) { |
2567 | atomic_inc(&block_group->count); | |
2568 | list_add_tail(&cluster->block_group_list, | |
2569 | &block_group->cluster_list); | |
2570 | cluster->block_group = block_group; | |
fa9c0d79 | 2571 | } |
fa9c0d79 CM |
2572 | out: |
2573 | spin_unlock(&cluster->lock); | |
34d52cb6 | 2574 | spin_unlock(&ctl->tree_lock); |
fa9c0d79 CM |
2575 | |
2576 | return ret; | |
2577 | } | |
2578 | ||
2579 | /* | |
2580 | * simple code to zero out a cluster | |
2581 | */ | |
2582 | void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster) | |
2583 | { | |
2584 | spin_lock_init(&cluster->lock); | |
2585 | spin_lock_init(&cluster->refill_lock); | |
6bef4d31 | 2586 | cluster->root = RB_ROOT; |
fa9c0d79 CM |
2587 | cluster->max_size = 0; |
2588 | INIT_LIST_HEAD(&cluster->block_group_list); | |
2589 | cluster->block_group = NULL; | |
2590 | } | |
2591 | ||
f7039b1d LD |
2592 | int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group, |
2593 | u64 *trimmed, u64 start, u64 end, u64 minlen) | |
2594 | { | |
34d52cb6 | 2595 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
f7039b1d LD |
2596 | struct btrfs_free_space *entry = NULL; |
2597 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
2598 | u64 bytes = 0; | |
2599 | u64 actually_trimmed; | |
2600 | int ret = 0; | |
2601 | ||
2602 | *trimmed = 0; | |
2603 | ||
2604 | while (start < end) { | |
34d52cb6 | 2605 | spin_lock(&ctl->tree_lock); |
f7039b1d | 2606 | |
34d52cb6 LZ |
2607 | if (ctl->free_space < minlen) { |
2608 | spin_unlock(&ctl->tree_lock); | |
f7039b1d LD |
2609 | break; |
2610 | } | |
2611 | ||
34d52cb6 | 2612 | entry = tree_search_offset(ctl, start, 0, 1); |
f7039b1d | 2613 | if (!entry) |
34d52cb6 LZ |
2614 | entry = tree_search_offset(ctl, |
2615 | offset_to_bitmap(ctl, start), | |
f7039b1d LD |
2616 | 1, 1); |
2617 | ||
2618 | if (!entry || entry->offset >= end) { | |
34d52cb6 | 2619 | spin_unlock(&ctl->tree_lock); |
f7039b1d LD |
2620 | break; |
2621 | } | |
2622 | ||
2623 | if (entry->bitmap) { | |
34d52cb6 | 2624 | ret = search_bitmap(ctl, entry, &start, &bytes); |
f7039b1d LD |
2625 | if (!ret) { |
2626 | if (start >= end) { | |
34d52cb6 | 2627 | spin_unlock(&ctl->tree_lock); |
f7039b1d LD |
2628 | break; |
2629 | } | |
2630 | bytes = min(bytes, end - start); | |
34d52cb6 | 2631 | bitmap_clear_bits(ctl, entry, start, bytes); |
f7039b1d | 2632 | if (entry->bytes == 0) |
34d52cb6 | 2633 | free_bitmap(ctl, entry); |
f7039b1d LD |
2634 | } else { |
2635 | start = entry->offset + BITS_PER_BITMAP * | |
2636 | block_group->sectorsize; | |
34d52cb6 | 2637 | spin_unlock(&ctl->tree_lock); |
f7039b1d LD |
2638 | ret = 0; |
2639 | continue; | |
2640 | } | |
2641 | } else { | |
2642 | start = entry->offset; | |
2643 | bytes = min(entry->bytes, end - start); | |
34d52cb6 | 2644 | unlink_free_space(ctl, entry); |
f789b684 | 2645 | kmem_cache_free(btrfs_free_space_cachep, entry); |
f7039b1d LD |
2646 | } |
2647 | ||
34d52cb6 | 2648 | spin_unlock(&ctl->tree_lock); |
f7039b1d LD |
2649 | |
2650 | if (bytes >= minlen) { | |
fb25e914 JB |
2651 | struct btrfs_space_info *space_info; |
2652 | int update = 0; | |
2653 | ||
2654 | space_info = block_group->space_info; | |
2655 | spin_lock(&space_info->lock); | |
2656 | spin_lock(&block_group->lock); | |
2657 | if (!block_group->ro) { | |
2658 | block_group->reserved += bytes; | |
2659 | space_info->bytes_reserved += bytes; | |
2660 | update = 1; | |
2661 | } | |
2662 | spin_unlock(&block_group->lock); | |
2663 | spin_unlock(&space_info->lock); | |
f7039b1d LD |
2664 | |
2665 | ret = btrfs_error_discard_extent(fs_info->extent_root, | |
2666 | start, | |
2667 | bytes, | |
2668 | &actually_trimmed); | |
2669 | ||
34d52cb6 | 2670 | btrfs_add_free_space(block_group, start, bytes); |
fb25e914 JB |
2671 | if (update) { |
2672 | spin_lock(&space_info->lock); | |
2673 | spin_lock(&block_group->lock); | |
2674 | if (block_group->ro) | |
2675 | space_info->bytes_readonly += bytes; | |
2676 | block_group->reserved -= bytes; | |
2677 | space_info->bytes_reserved -= bytes; | |
2678 | spin_unlock(&space_info->lock); | |
2679 | spin_unlock(&block_group->lock); | |
2680 | } | |
f7039b1d LD |
2681 | |
2682 | if (ret) | |
2683 | break; | |
2684 | *trimmed += actually_trimmed; | |
2685 | } | |
2686 | start += bytes; | |
2687 | bytes = 0; | |
2688 | ||
2689 | if (fatal_signal_pending(current)) { | |
2690 | ret = -ERESTARTSYS; | |
2691 | break; | |
2692 | } | |
2693 | ||
2694 | cond_resched(); | |
2695 | } | |
2696 | ||
2697 | return ret; | |
2698 | } | |
581bb050 LZ |
2699 | |
2700 | /* | |
2701 | * Find the left-most item in the cache tree, and then return the | |
2702 | * smallest inode number in the item. | |
2703 | * | |
2704 | * Note: the returned inode number may not be the smallest one in | |
2705 | * the tree, if the left-most item is a bitmap. | |
2706 | */ | |
2707 | u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root) | |
2708 | { | |
2709 | struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl; | |
2710 | struct btrfs_free_space *entry = NULL; | |
2711 | u64 ino = 0; | |
2712 | ||
2713 | spin_lock(&ctl->tree_lock); | |
2714 | ||
2715 | if (RB_EMPTY_ROOT(&ctl->free_space_offset)) | |
2716 | goto out; | |
2717 | ||
2718 | entry = rb_entry(rb_first(&ctl->free_space_offset), | |
2719 | struct btrfs_free_space, offset_index); | |
2720 | ||
2721 | if (!entry->bitmap) { | |
2722 | ino = entry->offset; | |
2723 | ||
2724 | unlink_free_space(ctl, entry); | |
2725 | entry->offset++; | |
2726 | entry->bytes--; | |
2727 | if (!entry->bytes) | |
2728 | kmem_cache_free(btrfs_free_space_cachep, entry); | |
2729 | else | |
2730 | link_free_space(ctl, entry); | |
2731 | } else { | |
2732 | u64 offset = 0; | |
2733 | u64 count = 1; | |
2734 | int ret; | |
2735 | ||
2736 | ret = search_bitmap(ctl, entry, &offset, &count); | |
2737 | BUG_ON(ret); | |
2738 | ||
2739 | ino = offset; | |
2740 | bitmap_clear_bits(ctl, entry, offset, 1); | |
2741 | if (entry->bytes == 0) | |
2742 | free_bitmap(ctl, entry); | |
2743 | } | |
2744 | out: | |
2745 | spin_unlock(&ctl->tree_lock); | |
2746 | ||
2747 | return ino; | |
2748 | } | |
82d5902d LZ |
2749 | |
2750 | struct inode *lookup_free_ino_inode(struct btrfs_root *root, | |
2751 | struct btrfs_path *path) | |
2752 | { | |
2753 | struct inode *inode = NULL; | |
2754 | ||
2755 | spin_lock(&root->cache_lock); | |
2756 | if (root->cache_inode) | |
2757 | inode = igrab(root->cache_inode); | |
2758 | spin_unlock(&root->cache_lock); | |
2759 | if (inode) | |
2760 | return inode; | |
2761 | ||
2762 | inode = __lookup_free_space_inode(root, path, 0); | |
2763 | if (IS_ERR(inode)) | |
2764 | return inode; | |
2765 | ||
2766 | spin_lock(&root->cache_lock); | |
7841cb28 | 2767 | if (!btrfs_fs_closing(root->fs_info)) |
82d5902d LZ |
2768 | root->cache_inode = igrab(inode); |
2769 | spin_unlock(&root->cache_lock); | |
2770 | ||
2771 | return inode; | |
2772 | } | |
2773 | ||
2774 | int create_free_ino_inode(struct btrfs_root *root, | |
2775 | struct btrfs_trans_handle *trans, | |
2776 | struct btrfs_path *path) | |
2777 | { | |
2778 | return __create_free_space_inode(root, trans, path, | |
2779 | BTRFS_FREE_INO_OBJECTID, 0); | |
2780 | } | |
2781 | ||
2782 | int load_free_ino_cache(struct btrfs_fs_info *fs_info, struct btrfs_root *root) | |
2783 | { | |
2784 | struct btrfs_free_space_ctl *ctl = root->free_ino_ctl; | |
2785 | struct btrfs_path *path; | |
2786 | struct inode *inode; | |
2787 | int ret = 0; | |
2788 | u64 root_gen = btrfs_root_generation(&root->root_item); | |
2789 | ||
4b9465cb CM |
2790 | if (!btrfs_test_opt(root, INODE_MAP_CACHE)) |
2791 | return 0; | |
2792 | ||
82d5902d LZ |
2793 | /* |
2794 | * If we're unmounting then just return, since this does a search on the | |
2795 | * normal root and not the commit root and we could deadlock. | |
2796 | */ | |
7841cb28 | 2797 | if (btrfs_fs_closing(fs_info)) |
82d5902d LZ |
2798 | return 0; |
2799 | ||
2800 | path = btrfs_alloc_path(); | |
2801 | if (!path) | |
2802 | return 0; | |
2803 | ||
2804 | inode = lookup_free_ino_inode(root, path); | |
2805 | if (IS_ERR(inode)) | |
2806 | goto out; | |
2807 | ||
2808 | if (root_gen != BTRFS_I(inode)->generation) | |
2809 | goto out_put; | |
2810 | ||
2811 | ret = __load_free_space_cache(root, inode, ctl, path, 0); | |
2812 | ||
2813 | if (ret < 0) | |
2814 | printk(KERN_ERR "btrfs: failed to load free ino cache for " | |
2815 | "root %llu\n", root->root_key.objectid); | |
2816 | out_put: | |
2817 | iput(inode); | |
2818 | out: | |
2819 | btrfs_free_path(path); | |
2820 | return ret; | |
2821 | } | |
2822 | ||
2823 | int btrfs_write_out_ino_cache(struct btrfs_root *root, | |
2824 | struct btrfs_trans_handle *trans, | |
2825 | struct btrfs_path *path) | |
2826 | { | |
2827 | struct btrfs_free_space_ctl *ctl = root->free_ino_ctl; | |
2828 | struct inode *inode; | |
2829 | int ret; | |
2830 | ||
4b9465cb CM |
2831 | if (!btrfs_test_opt(root, INODE_MAP_CACHE)) |
2832 | return 0; | |
2833 | ||
82d5902d LZ |
2834 | inode = lookup_free_ino_inode(root, path); |
2835 | if (IS_ERR(inode)) | |
2836 | return 0; | |
2837 | ||
2838 | ret = __btrfs_write_out_cache(root, inode, ctl, NULL, trans, path, 0); | |
c09544e0 JB |
2839 | if (ret) { |
2840 | btrfs_delalloc_release_metadata(inode, inode->i_size); | |
2841 | #ifdef DEBUG | |
82d5902d LZ |
2842 | printk(KERN_ERR "btrfs: failed to write free ino cache " |
2843 | "for root %llu\n", root->root_key.objectid); | |
c09544e0 JB |
2844 | #endif |
2845 | } | |
82d5902d LZ |
2846 | |
2847 | iput(inode); | |
2848 | return ret; | |
2849 | } |