Commit | Line | Data |
---|---|---|
0f9dd46c JB |
1 | /* |
2 | * Copyright (C) 2008 Red Hat. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
96303081 | 19 | #include <linux/pagemap.h> |
0f9dd46c | 20 | #include <linux/sched.h> |
5a0e3ad6 | 21 | #include <linux/slab.h> |
96303081 | 22 | #include <linux/math64.h> |
6ab60601 | 23 | #include <linux/ratelimit.h> |
0f9dd46c | 24 | #include "ctree.h" |
fa9c0d79 CM |
25 | #include "free-space-cache.h" |
26 | #include "transaction.h" | |
0af3d00b | 27 | #include "disk-io.h" |
43be2146 | 28 | #include "extent_io.h" |
581bb050 | 29 | #include "inode-map.h" |
fa9c0d79 | 30 | |
96303081 JB |
31 | #define BITS_PER_BITMAP (PAGE_CACHE_SIZE * 8) |
32 | #define MAX_CACHE_BYTES_PER_GIG (32 * 1024) | |
0f9dd46c | 33 | |
34d52cb6 | 34 | static int link_free_space(struct btrfs_free_space_ctl *ctl, |
0cb59c99 | 35 | struct btrfs_free_space *info); |
cd023e7b JB |
36 | static void unlink_free_space(struct btrfs_free_space_ctl *ctl, |
37 | struct btrfs_free_space *info); | |
0cb59c99 | 38 | |
0414efae LZ |
39 | static struct inode *__lookup_free_space_inode(struct btrfs_root *root, |
40 | struct btrfs_path *path, | |
41 | u64 offset) | |
0af3d00b JB |
42 | { |
43 | struct btrfs_key key; | |
44 | struct btrfs_key location; | |
45 | struct btrfs_disk_key disk_key; | |
46 | struct btrfs_free_space_header *header; | |
47 | struct extent_buffer *leaf; | |
48 | struct inode *inode = NULL; | |
49 | int ret; | |
50 | ||
0af3d00b | 51 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; |
0414efae | 52 | key.offset = offset; |
0af3d00b JB |
53 | key.type = 0; |
54 | ||
55 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
56 | if (ret < 0) | |
57 | return ERR_PTR(ret); | |
58 | if (ret > 0) { | |
b3b4aa74 | 59 | btrfs_release_path(path); |
0af3d00b JB |
60 | return ERR_PTR(-ENOENT); |
61 | } | |
62 | ||
63 | leaf = path->nodes[0]; | |
64 | header = btrfs_item_ptr(leaf, path->slots[0], | |
65 | struct btrfs_free_space_header); | |
66 | btrfs_free_space_key(leaf, header, &disk_key); | |
67 | btrfs_disk_key_to_cpu(&location, &disk_key); | |
b3b4aa74 | 68 | btrfs_release_path(path); |
0af3d00b JB |
69 | |
70 | inode = btrfs_iget(root->fs_info->sb, &location, root, NULL); | |
71 | if (!inode) | |
72 | return ERR_PTR(-ENOENT); | |
73 | if (IS_ERR(inode)) | |
74 | return inode; | |
75 | if (is_bad_inode(inode)) { | |
76 | iput(inode); | |
77 | return ERR_PTR(-ENOENT); | |
78 | } | |
79 | ||
adae52b9 MX |
80 | inode->i_mapping->flags &= ~__GFP_FS; |
81 | ||
0414efae LZ |
82 | return inode; |
83 | } | |
84 | ||
85 | struct inode *lookup_free_space_inode(struct btrfs_root *root, | |
86 | struct btrfs_block_group_cache | |
87 | *block_group, struct btrfs_path *path) | |
88 | { | |
89 | struct inode *inode = NULL; | |
5b0e95bf | 90 | u32 flags = BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW; |
0414efae LZ |
91 | |
92 | spin_lock(&block_group->lock); | |
93 | if (block_group->inode) | |
94 | inode = igrab(block_group->inode); | |
95 | spin_unlock(&block_group->lock); | |
96 | if (inode) | |
97 | return inode; | |
98 | ||
99 | inode = __lookup_free_space_inode(root, path, | |
100 | block_group->key.objectid); | |
101 | if (IS_ERR(inode)) | |
102 | return inode; | |
103 | ||
0af3d00b | 104 | spin_lock(&block_group->lock); |
5b0e95bf | 105 | if (!((BTRFS_I(inode)->flags & flags) == flags)) { |
2f356126 | 106 | printk(KERN_INFO "Old style space inode found, converting.\n"); |
5b0e95bf JB |
107 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM | |
108 | BTRFS_INODE_NODATACOW; | |
2f356126 JB |
109 | block_group->disk_cache_state = BTRFS_DC_CLEAR; |
110 | } | |
111 | ||
300e4f8a | 112 | if (!block_group->iref) { |
0af3d00b JB |
113 | block_group->inode = igrab(inode); |
114 | block_group->iref = 1; | |
115 | } | |
116 | spin_unlock(&block_group->lock); | |
117 | ||
118 | return inode; | |
119 | } | |
120 | ||
0414efae LZ |
121 | int __create_free_space_inode(struct btrfs_root *root, |
122 | struct btrfs_trans_handle *trans, | |
123 | struct btrfs_path *path, u64 ino, u64 offset) | |
0af3d00b JB |
124 | { |
125 | struct btrfs_key key; | |
126 | struct btrfs_disk_key disk_key; | |
127 | struct btrfs_free_space_header *header; | |
128 | struct btrfs_inode_item *inode_item; | |
129 | struct extent_buffer *leaf; | |
5b0e95bf | 130 | u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC; |
0af3d00b JB |
131 | int ret; |
132 | ||
0414efae | 133 | ret = btrfs_insert_empty_inode(trans, root, path, ino); |
0af3d00b JB |
134 | if (ret) |
135 | return ret; | |
136 | ||
5b0e95bf JB |
137 | /* We inline crc's for the free disk space cache */ |
138 | if (ino != BTRFS_FREE_INO_OBJECTID) | |
139 | flags |= BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW; | |
140 | ||
0af3d00b JB |
141 | leaf = path->nodes[0]; |
142 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
143 | struct btrfs_inode_item); | |
144 | btrfs_item_key(leaf, &disk_key, path->slots[0]); | |
145 | memset_extent_buffer(leaf, 0, (unsigned long)inode_item, | |
146 | sizeof(*inode_item)); | |
147 | btrfs_set_inode_generation(leaf, inode_item, trans->transid); | |
148 | btrfs_set_inode_size(leaf, inode_item, 0); | |
149 | btrfs_set_inode_nbytes(leaf, inode_item, 0); | |
150 | btrfs_set_inode_uid(leaf, inode_item, 0); | |
151 | btrfs_set_inode_gid(leaf, inode_item, 0); | |
152 | btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600); | |
5b0e95bf | 153 | btrfs_set_inode_flags(leaf, inode_item, flags); |
0af3d00b JB |
154 | btrfs_set_inode_nlink(leaf, inode_item, 1); |
155 | btrfs_set_inode_transid(leaf, inode_item, trans->transid); | |
0414efae | 156 | btrfs_set_inode_block_group(leaf, inode_item, offset); |
0af3d00b | 157 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 158 | btrfs_release_path(path); |
0af3d00b JB |
159 | |
160 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
0414efae | 161 | key.offset = offset; |
0af3d00b JB |
162 | key.type = 0; |
163 | ||
164 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
165 | sizeof(struct btrfs_free_space_header)); | |
166 | if (ret < 0) { | |
b3b4aa74 | 167 | btrfs_release_path(path); |
0af3d00b JB |
168 | return ret; |
169 | } | |
170 | leaf = path->nodes[0]; | |
171 | header = btrfs_item_ptr(leaf, path->slots[0], | |
172 | struct btrfs_free_space_header); | |
173 | memset_extent_buffer(leaf, 0, (unsigned long)header, sizeof(*header)); | |
174 | btrfs_set_free_space_key(leaf, header, &disk_key); | |
175 | btrfs_mark_buffer_dirty(leaf); | |
b3b4aa74 | 176 | btrfs_release_path(path); |
0af3d00b JB |
177 | |
178 | return 0; | |
179 | } | |
180 | ||
0414efae LZ |
181 | int create_free_space_inode(struct btrfs_root *root, |
182 | struct btrfs_trans_handle *trans, | |
183 | struct btrfs_block_group_cache *block_group, | |
184 | struct btrfs_path *path) | |
185 | { | |
186 | int ret; | |
187 | u64 ino; | |
188 | ||
189 | ret = btrfs_find_free_objectid(root, &ino); | |
190 | if (ret < 0) | |
191 | return ret; | |
192 | ||
193 | return __create_free_space_inode(root, trans, path, ino, | |
194 | block_group->key.objectid); | |
195 | } | |
196 | ||
0af3d00b JB |
197 | int btrfs_truncate_free_space_cache(struct btrfs_root *root, |
198 | struct btrfs_trans_handle *trans, | |
199 | struct btrfs_path *path, | |
200 | struct inode *inode) | |
201 | { | |
65450aa6 | 202 | struct btrfs_block_rsv *rsv; |
c8174313 | 203 | u64 needed_bytes; |
0af3d00b JB |
204 | loff_t oldsize; |
205 | int ret = 0; | |
206 | ||
65450aa6 | 207 | rsv = trans->block_rsv; |
c8174313 JB |
208 | trans->block_rsv = &root->fs_info->global_block_rsv; |
209 | ||
210 | /* 1 for slack space, 1 for updating the inode */ | |
211 | needed_bytes = btrfs_calc_trunc_metadata_size(root, 1) + | |
212 | btrfs_calc_trans_metadata_size(root, 1); | |
213 | ||
214 | spin_lock(&trans->block_rsv->lock); | |
215 | if (trans->block_rsv->reserved < needed_bytes) { | |
216 | spin_unlock(&trans->block_rsv->lock); | |
217 | trans->block_rsv = rsv; | |
218 | return -ENOSPC; | |
219 | } | |
220 | spin_unlock(&trans->block_rsv->lock); | |
0af3d00b JB |
221 | |
222 | oldsize = i_size_read(inode); | |
223 | btrfs_i_size_write(inode, 0); | |
224 | truncate_pagecache(inode, oldsize, 0); | |
225 | ||
226 | /* | |
227 | * We don't need an orphan item because truncating the free space cache | |
228 | * will never be split across transactions. | |
229 | */ | |
230 | ret = btrfs_truncate_inode_items(trans, root, inode, | |
231 | 0, BTRFS_EXTENT_DATA_KEY); | |
65450aa6 | 232 | |
0af3d00b | 233 | if (ret) { |
c8174313 | 234 | trans->block_rsv = rsv; |
79787eaa | 235 | btrfs_abort_transaction(trans, root, ret); |
0af3d00b JB |
236 | return ret; |
237 | } | |
238 | ||
82d5902d | 239 | ret = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
240 | if (ret) |
241 | btrfs_abort_transaction(trans, root, ret); | |
c8174313 JB |
242 | trans->block_rsv = rsv; |
243 | ||
82d5902d | 244 | return ret; |
0af3d00b JB |
245 | } |
246 | ||
9d66e233 JB |
247 | static int readahead_cache(struct inode *inode) |
248 | { | |
249 | struct file_ra_state *ra; | |
250 | unsigned long last_index; | |
251 | ||
252 | ra = kzalloc(sizeof(*ra), GFP_NOFS); | |
253 | if (!ra) | |
254 | return -ENOMEM; | |
255 | ||
256 | file_ra_state_init(ra, inode->i_mapping); | |
257 | last_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT; | |
258 | ||
259 | page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index); | |
260 | ||
261 | kfree(ra); | |
262 | ||
263 | return 0; | |
264 | } | |
265 | ||
a67509c3 JB |
266 | struct io_ctl { |
267 | void *cur, *orig; | |
268 | struct page *page; | |
269 | struct page **pages; | |
270 | struct btrfs_root *root; | |
271 | unsigned long size; | |
272 | int index; | |
273 | int num_pages; | |
5b0e95bf | 274 | unsigned check_crcs:1; |
a67509c3 JB |
275 | }; |
276 | ||
277 | static int io_ctl_init(struct io_ctl *io_ctl, struct inode *inode, | |
278 | struct btrfs_root *root) | |
279 | { | |
280 | memset(io_ctl, 0, sizeof(struct io_ctl)); | |
281 | io_ctl->num_pages = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> | |
282 | PAGE_CACHE_SHIFT; | |
283 | io_ctl->pages = kzalloc(sizeof(struct page *) * io_ctl->num_pages, | |
284 | GFP_NOFS); | |
285 | if (!io_ctl->pages) | |
286 | return -ENOMEM; | |
287 | io_ctl->root = root; | |
5b0e95bf JB |
288 | if (btrfs_ino(inode) != BTRFS_FREE_INO_OBJECTID) |
289 | io_ctl->check_crcs = 1; | |
a67509c3 JB |
290 | return 0; |
291 | } | |
292 | ||
293 | static void io_ctl_free(struct io_ctl *io_ctl) | |
294 | { | |
295 | kfree(io_ctl->pages); | |
296 | } | |
297 | ||
298 | static void io_ctl_unmap_page(struct io_ctl *io_ctl) | |
299 | { | |
300 | if (io_ctl->cur) { | |
301 | kunmap(io_ctl->page); | |
302 | io_ctl->cur = NULL; | |
303 | io_ctl->orig = NULL; | |
304 | } | |
305 | } | |
306 | ||
307 | static void io_ctl_map_page(struct io_ctl *io_ctl, int clear) | |
308 | { | |
309 | WARN_ON(io_ctl->cur); | |
310 | BUG_ON(io_ctl->index >= io_ctl->num_pages); | |
311 | io_ctl->page = io_ctl->pages[io_ctl->index++]; | |
312 | io_ctl->cur = kmap(io_ctl->page); | |
313 | io_ctl->orig = io_ctl->cur; | |
314 | io_ctl->size = PAGE_CACHE_SIZE; | |
315 | if (clear) | |
316 | memset(io_ctl->cur, 0, PAGE_CACHE_SIZE); | |
317 | } | |
318 | ||
319 | static void io_ctl_drop_pages(struct io_ctl *io_ctl) | |
320 | { | |
321 | int i; | |
322 | ||
323 | io_ctl_unmap_page(io_ctl); | |
324 | ||
325 | for (i = 0; i < io_ctl->num_pages; i++) { | |
a1ee5a45 LZ |
326 | if (io_ctl->pages[i]) { |
327 | ClearPageChecked(io_ctl->pages[i]); | |
328 | unlock_page(io_ctl->pages[i]); | |
329 | page_cache_release(io_ctl->pages[i]); | |
330 | } | |
a67509c3 JB |
331 | } |
332 | } | |
333 | ||
334 | static int io_ctl_prepare_pages(struct io_ctl *io_ctl, struct inode *inode, | |
335 | int uptodate) | |
336 | { | |
337 | struct page *page; | |
338 | gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); | |
339 | int i; | |
340 | ||
341 | for (i = 0; i < io_ctl->num_pages; i++) { | |
342 | page = find_or_create_page(inode->i_mapping, i, mask); | |
343 | if (!page) { | |
344 | io_ctl_drop_pages(io_ctl); | |
345 | return -ENOMEM; | |
346 | } | |
347 | io_ctl->pages[i] = page; | |
348 | if (uptodate && !PageUptodate(page)) { | |
349 | btrfs_readpage(NULL, page); | |
350 | lock_page(page); | |
351 | if (!PageUptodate(page)) { | |
352 | printk(KERN_ERR "btrfs: error reading free " | |
353 | "space cache\n"); | |
354 | io_ctl_drop_pages(io_ctl); | |
355 | return -EIO; | |
356 | } | |
357 | } | |
358 | } | |
359 | ||
f7d61dcd JB |
360 | for (i = 0; i < io_ctl->num_pages; i++) { |
361 | clear_page_dirty_for_io(io_ctl->pages[i]); | |
362 | set_page_extent_mapped(io_ctl->pages[i]); | |
363 | } | |
364 | ||
a67509c3 JB |
365 | return 0; |
366 | } | |
367 | ||
368 | static void io_ctl_set_generation(struct io_ctl *io_ctl, u64 generation) | |
369 | { | |
370 | u64 *val; | |
371 | ||
372 | io_ctl_map_page(io_ctl, 1); | |
373 | ||
374 | /* | |
5b0e95bf JB |
375 | * Skip the csum areas. If we don't check crcs then we just have a |
376 | * 64bit chunk at the front of the first page. | |
a67509c3 | 377 | */ |
5b0e95bf JB |
378 | if (io_ctl->check_crcs) { |
379 | io_ctl->cur += (sizeof(u32) * io_ctl->num_pages); | |
380 | io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages); | |
381 | } else { | |
382 | io_ctl->cur += sizeof(u64); | |
383 | io_ctl->size -= sizeof(u64) * 2; | |
384 | } | |
a67509c3 JB |
385 | |
386 | val = io_ctl->cur; | |
387 | *val = cpu_to_le64(generation); | |
388 | io_ctl->cur += sizeof(u64); | |
a67509c3 JB |
389 | } |
390 | ||
391 | static int io_ctl_check_generation(struct io_ctl *io_ctl, u64 generation) | |
392 | { | |
393 | u64 *gen; | |
394 | ||
5b0e95bf JB |
395 | /* |
396 | * Skip the crc area. If we don't check crcs then we just have a 64bit | |
397 | * chunk at the front of the first page. | |
398 | */ | |
399 | if (io_ctl->check_crcs) { | |
400 | io_ctl->cur += sizeof(u32) * io_ctl->num_pages; | |
401 | io_ctl->size -= sizeof(u64) + | |
402 | (sizeof(u32) * io_ctl->num_pages); | |
403 | } else { | |
404 | io_ctl->cur += sizeof(u64); | |
405 | io_ctl->size -= sizeof(u64) * 2; | |
406 | } | |
a67509c3 | 407 | |
a67509c3 JB |
408 | gen = io_ctl->cur; |
409 | if (le64_to_cpu(*gen) != generation) { | |
410 | printk_ratelimited(KERN_ERR "btrfs: space cache generation " | |
411 | "(%Lu) does not match inode (%Lu)\n", *gen, | |
412 | generation); | |
413 | io_ctl_unmap_page(io_ctl); | |
414 | return -EIO; | |
415 | } | |
416 | io_ctl->cur += sizeof(u64); | |
5b0e95bf JB |
417 | return 0; |
418 | } | |
419 | ||
420 | static void io_ctl_set_crc(struct io_ctl *io_ctl, int index) | |
421 | { | |
422 | u32 *tmp; | |
423 | u32 crc = ~(u32)0; | |
424 | unsigned offset = 0; | |
425 | ||
426 | if (!io_ctl->check_crcs) { | |
427 | io_ctl_unmap_page(io_ctl); | |
428 | return; | |
429 | } | |
430 | ||
431 | if (index == 0) | |
cb54f257 | 432 | offset = sizeof(u32) * io_ctl->num_pages; |
5b0e95bf JB |
433 | |
434 | crc = btrfs_csum_data(io_ctl->root, io_ctl->orig + offset, crc, | |
435 | PAGE_CACHE_SIZE - offset); | |
436 | btrfs_csum_final(crc, (char *)&crc); | |
437 | io_ctl_unmap_page(io_ctl); | |
438 | tmp = kmap(io_ctl->pages[0]); | |
439 | tmp += index; | |
440 | *tmp = crc; | |
441 | kunmap(io_ctl->pages[0]); | |
442 | } | |
443 | ||
444 | static int io_ctl_check_crc(struct io_ctl *io_ctl, int index) | |
445 | { | |
446 | u32 *tmp, val; | |
447 | u32 crc = ~(u32)0; | |
448 | unsigned offset = 0; | |
449 | ||
450 | if (!io_ctl->check_crcs) { | |
451 | io_ctl_map_page(io_ctl, 0); | |
452 | return 0; | |
453 | } | |
454 | ||
455 | if (index == 0) | |
456 | offset = sizeof(u32) * io_ctl->num_pages; | |
457 | ||
458 | tmp = kmap(io_ctl->pages[0]); | |
459 | tmp += index; | |
460 | val = *tmp; | |
461 | kunmap(io_ctl->pages[0]); | |
462 | ||
463 | io_ctl_map_page(io_ctl, 0); | |
464 | crc = btrfs_csum_data(io_ctl->root, io_ctl->orig + offset, crc, | |
465 | PAGE_CACHE_SIZE - offset); | |
466 | btrfs_csum_final(crc, (char *)&crc); | |
467 | if (val != crc) { | |
468 | printk_ratelimited(KERN_ERR "btrfs: csum mismatch on free " | |
469 | "space cache\n"); | |
470 | io_ctl_unmap_page(io_ctl); | |
471 | return -EIO; | |
472 | } | |
473 | ||
a67509c3 JB |
474 | return 0; |
475 | } | |
476 | ||
477 | static int io_ctl_add_entry(struct io_ctl *io_ctl, u64 offset, u64 bytes, | |
478 | void *bitmap) | |
479 | { | |
480 | struct btrfs_free_space_entry *entry; | |
481 | ||
482 | if (!io_ctl->cur) | |
483 | return -ENOSPC; | |
484 | ||
485 | entry = io_ctl->cur; | |
486 | entry->offset = cpu_to_le64(offset); | |
487 | entry->bytes = cpu_to_le64(bytes); | |
488 | entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP : | |
489 | BTRFS_FREE_SPACE_EXTENT; | |
490 | io_ctl->cur += sizeof(struct btrfs_free_space_entry); | |
491 | io_ctl->size -= sizeof(struct btrfs_free_space_entry); | |
492 | ||
493 | if (io_ctl->size >= sizeof(struct btrfs_free_space_entry)) | |
494 | return 0; | |
495 | ||
5b0e95bf | 496 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); |
a67509c3 JB |
497 | |
498 | /* No more pages to map */ | |
499 | if (io_ctl->index >= io_ctl->num_pages) | |
500 | return 0; | |
501 | ||
502 | /* map the next page */ | |
503 | io_ctl_map_page(io_ctl, 1); | |
504 | return 0; | |
505 | } | |
506 | ||
507 | static int io_ctl_add_bitmap(struct io_ctl *io_ctl, void *bitmap) | |
508 | { | |
509 | if (!io_ctl->cur) | |
510 | return -ENOSPC; | |
511 | ||
512 | /* | |
513 | * If we aren't at the start of the current page, unmap this one and | |
514 | * map the next one if there is any left. | |
515 | */ | |
516 | if (io_ctl->cur != io_ctl->orig) { | |
5b0e95bf | 517 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); |
a67509c3 JB |
518 | if (io_ctl->index >= io_ctl->num_pages) |
519 | return -ENOSPC; | |
520 | io_ctl_map_page(io_ctl, 0); | |
521 | } | |
522 | ||
523 | memcpy(io_ctl->cur, bitmap, PAGE_CACHE_SIZE); | |
5b0e95bf | 524 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); |
a67509c3 JB |
525 | if (io_ctl->index < io_ctl->num_pages) |
526 | io_ctl_map_page(io_ctl, 0); | |
527 | return 0; | |
528 | } | |
529 | ||
530 | static void io_ctl_zero_remaining_pages(struct io_ctl *io_ctl) | |
531 | { | |
5b0e95bf JB |
532 | /* |
533 | * If we're not on the boundary we know we've modified the page and we | |
534 | * need to crc the page. | |
535 | */ | |
536 | if (io_ctl->cur != io_ctl->orig) | |
537 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); | |
538 | else | |
539 | io_ctl_unmap_page(io_ctl); | |
a67509c3 JB |
540 | |
541 | while (io_ctl->index < io_ctl->num_pages) { | |
542 | io_ctl_map_page(io_ctl, 1); | |
5b0e95bf | 543 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); |
a67509c3 JB |
544 | } |
545 | } | |
546 | ||
5b0e95bf JB |
547 | static int io_ctl_read_entry(struct io_ctl *io_ctl, |
548 | struct btrfs_free_space *entry, u8 *type) | |
a67509c3 JB |
549 | { |
550 | struct btrfs_free_space_entry *e; | |
2f120c05 JB |
551 | int ret; |
552 | ||
553 | if (!io_ctl->cur) { | |
554 | ret = io_ctl_check_crc(io_ctl, io_ctl->index); | |
555 | if (ret) | |
556 | return ret; | |
557 | } | |
a67509c3 JB |
558 | |
559 | e = io_ctl->cur; | |
560 | entry->offset = le64_to_cpu(e->offset); | |
561 | entry->bytes = le64_to_cpu(e->bytes); | |
5b0e95bf | 562 | *type = e->type; |
a67509c3 JB |
563 | io_ctl->cur += sizeof(struct btrfs_free_space_entry); |
564 | io_ctl->size -= sizeof(struct btrfs_free_space_entry); | |
565 | ||
566 | if (io_ctl->size >= sizeof(struct btrfs_free_space_entry)) | |
5b0e95bf | 567 | return 0; |
a67509c3 JB |
568 | |
569 | io_ctl_unmap_page(io_ctl); | |
570 | ||
2f120c05 | 571 | return 0; |
a67509c3 JB |
572 | } |
573 | ||
5b0e95bf JB |
574 | static int io_ctl_read_bitmap(struct io_ctl *io_ctl, |
575 | struct btrfs_free_space *entry) | |
a67509c3 | 576 | { |
5b0e95bf JB |
577 | int ret; |
578 | ||
5b0e95bf JB |
579 | ret = io_ctl_check_crc(io_ctl, io_ctl->index); |
580 | if (ret) | |
581 | return ret; | |
582 | ||
a67509c3 JB |
583 | memcpy(entry->bitmap, io_ctl->cur, PAGE_CACHE_SIZE); |
584 | io_ctl_unmap_page(io_ctl); | |
5b0e95bf JB |
585 | |
586 | return 0; | |
a67509c3 JB |
587 | } |
588 | ||
cd023e7b JB |
589 | /* |
590 | * Since we attach pinned extents after the fact we can have contiguous sections | |
591 | * of free space that are split up in entries. This poses a problem with the | |
592 | * tree logging stuff since it could have allocated across what appears to be 2 | |
593 | * entries since we would have merged the entries when adding the pinned extents | |
594 | * back to the free space cache. So run through the space cache that we just | |
595 | * loaded and merge contiguous entries. This will make the log replay stuff not | |
596 | * blow up and it will make for nicer allocator behavior. | |
597 | */ | |
598 | static void merge_space_tree(struct btrfs_free_space_ctl *ctl) | |
599 | { | |
600 | struct btrfs_free_space *e, *prev = NULL; | |
601 | struct rb_node *n; | |
602 | ||
603 | again: | |
604 | spin_lock(&ctl->tree_lock); | |
605 | for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) { | |
606 | e = rb_entry(n, struct btrfs_free_space, offset_index); | |
607 | if (!prev) | |
608 | goto next; | |
609 | if (e->bitmap || prev->bitmap) | |
610 | goto next; | |
611 | if (prev->offset + prev->bytes == e->offset) { | |
612 | unlink_free_space(ctl, prev); | |
613 | unlink_free_space(ctl, e); | |
614 | prev->bytes += e->bytes; | |
615 | kmem_cache_free(btrfs_free_space_cachep, e); | |
616 | link_free_space(ctl, prev); | |
617 | prev = NULL; | |
618 | spin_unlock(&ctl->tree_lock); | |
619 | goto again; | |
620 | } | |
621 | next: | |
622 | prev = e; | |
623 | } | |
624 | spin_unlock(&ctl->tree_lock); | |
625 | } | |
626 | ||
0414efae LZ |
627 | int __load_free_space_cache(struct btrfs_root *root, struct inode *inode, |
628 | struct btrfs_free_space_ctl *ctl, | |
629 | struct btrfs_path *path, u64 offset) | |
9d66e233 | 630 | { |
9d66e233 JB |
631 | struct btrfs_free_space_header *header; |
632 | struct extent_buffer *leaf; | |
a67509c3 | 633 | struct io_ctl io_ctl; |
9d66e233 | 634 | struct btrfs_key key; |
a67509c3 | 635 | struct btrfs_free_space *e, *n; |
9d66e233 JB |
636 | struct list_head bitmaps; |
637 | u64 num_entries; | |
638 | u64 num_bitmaps; | |
639 | u64 generation; | |
a67509c3 | 640 | u8 type; |
f6a39829 | 641 | int ret = 0; |
9d66e233 JB |
642 | |
643 | INIT_LIST_HEAD(&bitmaps); | |
644 | ||
9d66e233 | 645 | /* Nothing in the space cache, goodbye */ |
0414efae | 646 | if (!i_size_read(inode)) |
a67509c3 | 647 | return 0; |
9d66e233 JB |
648 | |
649 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
0414efae | 650 | key.offset = offset; |
9d66e233 JB |
651 | key.type = 0; |
652 | ||
653 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
0414efae | 654 | if (ret < 0) |
a67509c3 | 655 | return 0; |
0414efae | 656 | else if (ret > 0) { |
945d8962 | 657 | btrfs_release_path(path); |
a67509c3 | 658 | return 0; |
9d66e233 JB |
659 | } |
660 | ||
0414efae LZ |
661 | ret = -1; |
662 | ||
9d66e233 JB |
663 | leaf = path->nodes[0]; |
664 | header = btrfs_item_ptr(leaf, path->slots[0], | |
665 | struct btrfs_free_space_header); | |
666 | num_entries = btrfs_free_space_entries(leaf, header); | |
667 | num_bitmaps = btrfs_free_space_bitmaps(leaf, header); | |
668 | generation = btrfs_free_space_generation(leaf, header); | |
945d8962 | 669 | btrfs_release_path(path); |
9d66e233 JB |
670 | |
671 | if (BTRFS_I(inode)->generation != generation) { | |
672 | printk(KERN_ERR "btrfs: free space inode generation (%llu) did" | |
0414efae | 673 | " not match free space cache generation (%llu)\n", |
9d66e233 | 674 | (unsigned long long)BTRFS_I(inode)->generation, |
0414efae | 675 | (unsigned long long)generation); |
a67509c3 | 676 | return 0; |
9d66e233 JB |
677 | } |
678 | ||
679 | if (!num_entries) | |
a67509c3 | 680 | return 0; |
9d66e233 | 681 | |
706efc66 LZ |
682 | ret = io_ctl_init(&io_ctl, inode, root); |
683 | if (ret) | |
684 | return ret; | |
685 | ||
9d66e233 | 686 | ret = readahead_cache(inode); |
0414efae | 687 | if (ret) |
9d66e233 | 688 | goto out; |
9d66e233 | 689 | |
a67509c3 JB |
690 | ret = io_ctl_prepare_pages(&io_ctl, inode, 1); |
691 | if (ret) | |
692 | goto out; | |
9d66e233 | 693 | |
5b0e95bf JB |
694 | ret = io_ctl_check_crc(&io_ctl, 0); |
695 | if (ret) | |
696 | goto free_cache; | |
697 | ||
a67509c3 JB |
698 | ret = io_ctl_check_generation(&io_ctl, generation); |
699 | if (ret) | |
700 | goto free_cache; | |
9d66e233 | 701 | |
a67509c3 JB |
702 | while (num_entries) { |
703 | e = kmem_cache_zalloc(btrfs_free_space_cachep, | |
704 | GFP_NOFS); | |
705 | if (!e) | |
9d66e233 | 706 | goto free_cache; |
9d66e233 | 707 | |
5b0e95bf JB |
708 | ret = io_ctl_read_entry(&io_ctl, e, &type); |
709 | if (ret) { | |
710 | kmem_cache_free(btrfs_free_space_cachep, e); | |
711 | goto free_cache; | |
712 | } | |
713 | ||
a67509c3 JB |
714 | if (!e->bytes) { |
715 | kmem_cache_free(btrfs_free_space_cachep, e); | |
716 | goto free_cache; | |
9d66e233 | 717 | } |
a67509c3 JB |
718 | |
719 | if (type == BTRFS_FREE_SPACE_EXTENT) { | |
720 | spin_lock(&ctl->tree_lock); | |
721 | ret = link_free_space(ctl, e); | |
722 | spin_unlock(&ctl->tree_lock); | |
723 | if (ret) { | |
724 | printk(KERN_ERR "Duplicate entries in " | |
725 | "free space cache, dumping\n"); | |
726 | kmem_cache_free(btrfs_free_space_cachep, e); | |
9d66e233 JB |
727 | goto free_cache; |
728 | } | |
a67509c3 JB |
729 | } else { |
730 | BUG_ON(!num_bitmaps); | |
731 | num_bitmaps--; | |
732 | e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS); | |
733 | if (!e->bitmap) { | |
734 | kmem_cache_free( | |
735 | btrfs_free_space_cachep, e); | |
9d66e233 JB |
736 | goto free_cache; |
737 | } | |
a67509c3 JB |
738 | spin_lock(&ctl->tree_lock); |
739 | ret = link_free_space(ctl, e); | |
740 | ctl->total_bitmaps++; | |
741 | ctl->op->recalc_thresholds(ctl); | |
742 | spin_unlock(&ctl->tree_lock); | |
743 | if (ret) { | |
744 | printk(KERN_ERR "Duplicate entries in " | |
745 | "free space cache, dumping\n"); | |
dc89e982 | 746 | kmem_cache_free(btrfs_free_space_cachep, e); |
9d66e233 JB |
747 | goto free_cache; |
748 | } | |
a67509c3 | 749 | list_add_tail(&e->list, &bitmaps); |
9d66e233 JB |
750 | } |
751 | ||
a67509c3 JB |
752 | num_entries--; |
753 | } | |
9d66e233 | 754 | |
2f120c05 JB |
755 | io_ctl_unmap_page(&io_ctl); |
756 | ||
a67509c3 JB |
757 | /* |
758 | * We add the bitmaps at the end of the entries in order that | |
759 | * the bitmap entries are added to the cache. | |
760 | */ | |
761 | list_for_each_entry_safe(e, n, &bitmaps, list) { | |
9d66e233 | 762 | list_del_init(&e->list); |
5b0e95bf JB |
763 | ret = io_ctl_read_bitmap(&io_ctl, e); |
764 | if (ret) | |
765 | goto free_cache; | |
9d66e233 JB |
766 | } |
767 | ||
a67509c3 | 768 | io_ctl_drop_pages(&io_ctl); |
cd023e7b | 769 | merge_space_tree(ctl); |
9d66e233 JB |
770 | ret = 1; |
771 | out: | |
a67509c3 | 772 | io_ctl_free(&io_ctl); |
9d66e233 | 773 | return ret; |
9d66e233 | 774 | free_cache: |
a67509c3 | 775 | io_ctl_drop_pages(&io_ctl); |
0414efae | 776 | __btrfs_remove_free_space_cache(ctl); |
9d66e233 JB |
777 | goto out; |
778 | } | |
779 | ||
0414efae LZ |
780 | int load_free_space_cache(struct btrfs_fs_info *fs_info, |
781 | struct btrfs_block_group_cache *block_group) | |
0cb59c99 | 782 | { |
34d52cb6 | 783 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0414efae LZ |
784 | struct btrfs_root *root = fs_info->tree_root; |
785 | struct inode *inode; | |
786 | struct btrfs_path *path; | |
5b0e95bf | 787 | int ret = 0; |
0414efae LZ |
788 | bool matched; |
789 | u64 used = btrfs_block_group_used(&block_group->item); | |
790 | ||
0414efae LZ |
791 | /* |
792 | * If this block group has been marked to be cleared for one reason or | |
793 | * another then we can't trust the on disk cache, so just return. | |
794 | */ | |
9d66e233 | 795 | spin_lock(&block_group->lock); |
0414efae LZ |
796 | if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) { |
797 | spin_unlock(&block_group->lock); | |
798 | return 0; | |
799 | } | |
9d66e233 | 800 | spin_unlock(&block_group->lock); |
0414efae LZ |
801 | |
802 | path = btrfs_alloc_path(); | |
803 | if (!path) | |
804 | return 0; | |
d53ba474 JB |
805 | path->search_commit_root = 1; |
806 | path->skip_locking = 1; | |
0414efae LZ |
807 | |
808 | inode = lookup_free_space_inode(root, block_group, path); | |
809 | if (IS_ERR(inode)) { | |
810 | btrfs_free_path(path); | |
811 | return 0; | |
812 | } | |
813 | ||
5b0e95bf JB |
814 | /* We may have converted the inode and made the cache invalid. */ |
815 | spin_lock(&block_group->lock); | |
816 | if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) { | |
817 | spin_unlock(&block_group->lock); | |
a7e221e9 | 818 | btrfs_free_path(path); |
5b0e95bf JB |
819 | goto out; |
820 | } | |
821 | spin_unlock(&block_group->lock); | |
822 | ||
0414efae LZ |
823 | ret = __load_free_space_cache(fs_info->tree_root, inode, ctl, |
824 | path, block_group->key.objectid); | |
825 | btrfs_free_path(path); | |
826 | if (ret <= 0) | |
827 | goto out; | |
828 | ||
829 | spin_lock(&ctl->tree_lock); | |
830 | matched = (ctl->free_space == (block_group->key.offset - used - | |
831 | block_group->bytes_super)); | |
832 | spin_unlock(&ctl->tree_lock); | |
833 | ||
834 | if (!matched) { | |
835 | __btrfs_remove_free_space_cache(ctl); | |
836 | printk(KERN_ERR "block group %llu has an wrong amount of free " | |
837 | "space\n", block_group->key.objectid); | |
838 | ret = -1; | |
839 | } | |
840 | out: | |
841 | if (ret < 0) { | |
842 | /* This cache is bogus, make sure it gets cleared */ | |
843 | spin_lock(&block_group->lock); | |
844 | block_group->disk_cache_state = BTRFS_DC_CLEAR; | |
845 | spin_unlock(&block_group->lock); | |
82d5902d | 846 | ret = 0; |
0414efae LZ |
847 | |
848 | printk(KERN_ERR "btrfs: failed to load free space cache " | |
849 | "for block group %llu\n", block_group->key.objectid); | |
850 | } | |
851 | ||
852 | iput(inode); | |
853 | return ret; | |
9d66e233 JB |
854 | } |
855 | ||
c09544e0 JB |
856 | /** |
857 | * __btrfs_write_out_cache - write out cached info to an inode | |
858 | * @root - the root the inode belongs to | |
859 | * @ctl - the free space cache we are going to write out | |
860 | * @block_group - the block_group for this cache if it belongs to a block_group | |
861 | * @trans - the trans handle | |
862 | * @path - the path to use | |
863 | * @offset - the offset for the key we'll insert | |
864 | * | |
865 | * This function writes out a free space cache struct to disk for quick recovery | |
866 | * on mount. This will return 0 if it was successfull in writing the cache out, | |
867 | * and -1 if it was not. | |
868 | */ | |
0414efae LZ |
869 | int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode, |
870 | struct btrfs_free_space_ctl *ctl, | |
871 | struct btrfs_block_group_cache *block_group, | |
872 | struct btrfs_trans_handle *trans, | |
873 | struct btrfs_path *path, u64 offset) | |
0cb59c99 JB |
874 | { |
875 | struct btrfs_free_space_header *header; | |
876 | struct extent_buffer *leaf; | |
0cb59c99 JB |
877 | struct rb_node *node; |
878 | struct list_head *pos, *n; | |
0cb59c99 | 879 | struct extent_state *cached_state = NULL; |
43be2146 JB |
880 | struct btrfs_free_cluster *cluster = NULL; |
881 | struct extent_io_tree *unpin = NULL; | |
a67509c3 | 882 | struct io_ctl io_ctl; |
0cb59c99 JB |
883 | struct list_head bitmap_list; |
884 | struct btrfs_key key; | |
db804f23 | 885 | u64 start, extent_start, extent_end, len; |
0cb59c99 JB |
886 | int entries = 0; |
887 | int bitmaps = 0; | |
c09544e0 JB |
888 | int ret; |
889 | int err = -1; | |
0cb59c99 | 890 | |
0cb59c99 JB |
891 | INIT_LIST_HEAD(&bitmap_list); |
892 | ||
0414efae LZ |
893 | if (!i_size_read(inode)) |
894 | return -1; | |
2b20982e | 895 | |
706efc66 LZ |
896 | ret = io_ctl_init(&io_ctl, inode, root); |
897 | if (ret) | |
898 | return -1; | |
be1a12a0 | 899 | |
43be2146 | 900 | /* Get the cluster for this block_group if it exists */ |
0414efae | 901 | if (block_group && !list_empty(&block_group->cluster_list)) |
43be2146 JB |
902 | cluster = list_entry(block_group->cluster_list.next, |
903 | struct btrfs_free_cluster, | |
904 | block_group_list); | |
905 | ||
a67509c3 JB |
906 | /* Lock all pages first so we can lock the extent safely. */ |
907 | io_ctl_prepare_pages(&io_ctl, inode, 0); | |
0cb59c99 | 908 | |
0cb59c99 | 909 | lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1, |
d0082371 | 910 | 0, &cached_state); |
0cb59c99 | 911 | |
f75b130e JB |
912 | node = rb_first(&ctl->free_space_offset); |
913 | if (!node && cluster) { | |
914 | node = rb_first(&cluster->root); | |
915 | cluster = NULL; | |
916 | } | |
917 | ||
5b0e95bf JB |
918 | /* Make sure we can fit our crcs into the first page */ |
919 | if (io_ctl.check_crcs && | |
920 | (io_ctl.num_pages * sizeof(u32)) >= PAGE_CACHE_SIZE) { | |
921 | WARN_ON(1); | |
922 | goto out_nospc; | |
923 | } | |
924 | ||
a67509c3 | 925 | io_ctl_set_generation(&io_ctl, trans->transid); |
43be2146 | 926 | |
a67509c3 JB |
927 | /* Write out the extent entries */ |
928 | while (node) { | |
929 | struct btrfs_free_space *e; | |
0cb59c99 | 930 | |
a67509c3 JB |
931 | e = rb_entry(node, struct btrfs_free_space, offset_index); |
932 | entries++; | |
0cb59c99 | 933 | |
a67509c3 JB |
934 | ret = io_ctl_add_entry(&io_ctl, e->offset, e->bytes, |
935 | e->bitmap); | |
936 | if (ret) | |
937 | goto out_nospc; | |
2f356126 | 938 | |
a67509c3 JB |
939 | if (e->bitmap) { |
940 | list_add_tail(&e->list, &bitmap_list); | |
941 | bitmaps++; | |
2f356126 | 942 | } |
a67509c3 JB |
943 | node = rb_next(node); |
944 | if (!node && cluster) { | |
945 | node = rb_first(&cluster->root); | |
946 | cluster = NULL; | |
43be2146 | 947 | } |
a67509c3 | 948 | } |
43be2146 | 949 | |
a67509c3 JB |
950 | /* |
951 | * We want to add any pinned extents to our free space cache | |
952 | * so we don't leak the space | |
953 | */ | |
db804f23 LZ |
954 | |
955 | /* | |
956 | * We shouldn't have switched the pinned extents yet so this is the | |
957 | * right one | |
958 | */ | |
959 | unpin = root->fs_info->pinned_extents; | |
960 | ||
961 | if (block_group) | |
962 | start = block_group->key.objectid; | |
963 | ||
a67509c3 JB |
964 | while (block_group && (start < block_group->key.objectid + |
965 | block_group->key.offset)) { | |
db804f23 LZ |
966 | ret = find_first_extent_bit(unpin, start, |
967 | &extent_start, &extent_end, | |
a67509c3 JB |
968 | EXTENT_DIRTY); |
969 | if (ret) { | |
970 | ret = 0; | |
971 | break; | |
0cb59c99 | 972 | } |
0cb59c99 | 973 | |
a67509c3 | 974 | /* This pinned extent is out of our range */ |
db804f23 | 975 | if (extent_start >= block_group->key.objectid + |
a67509c3 JB |
976 | block_group->key.offset) |
977 | break; | |
2f356126 | 978 | |
db804f23 LZ |
979 | extent_start = max(extent_start, start); |
980 | extent_end = min(block_group->key.objectid + | |
981 | block_group->key.offset, extent_end + 1); | |
982 | len = extent_end - extent_start; | |
0cb59c99 | 983 | |
a67509c3 | 984 | entries++; |
db804f23 | 985 | ret = io_ctl_add_entry(&io_ctl, extent_start, len, NULL); |
a67509c3 JB |
986 | if (ret) |
987 | goto out_nospc; | |
0cb59c99 | 988 | |
db804f23 | 989 | start = extent_end; |
a67509c3 | 990 | } |
0cb59c99 JB |
991 | |
992 | /* Write out the bitmaps */ | |
993 | list_for_each_safe(pos, n, &bitmap_list) { | |
0cb59c99 JB |
994 | struct btrfs_free_space *entry = |
995 | list_entry(pos, struct btrfs_free_space, list); | |
996 | ||
a67509c3 JB |
997 | ret = io_ctl_add_bitmap(&io_ctl, entry->bitmap); |
998 | if (ret) | |
999 | goto out_nospc; | |
0cb59c99 | 1000 | list_del_init(&entry->list); |
be1a12a0 JB |
1001 | } |
1002 | ||
0cb59c99 | 1003 | /* Zero out the rest of the pages just to make sure */ |
a67509c3 | 1004 | io_ctl_zero_remaining_pages(&io_ctl); |
0cb59c99 | 1005 | |
a67509c3 JB |
1006 | ret = btrfs_dirty_pages(root, inode, io_ctl.pages, io_ctl.num_pages, |
1007 | 0, i_size_read(inode), &cached_state); | |
1008 | io_ctl_drop_pages(&io_ctl); | |
0cb59c99 JB |
1009 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0, |
1010 | i_size_read(inode) - 1, &cached_state, GFP_NOFS); | |
1011 | ||
c09544e0 | 1012 | if (ret) |
2f356126 | 1013 | goto out; |
be1a12a0 | 1014 | |
be1a12a0 | 1015 | |
5fd02043 | 1016 | btrfs_wait_ordered_range(inode, 0, (u64)-1); |
0cb59c99 JB |
1017 | |
1018 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
0414efae | 1019 | key.offset = offset; |
0cb59c99 JB |
1020 | key.type = 0; |
1021 | ||
a9b5fcdd | 1022 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
0cb59c99 | 1023 | if (ret < 0) { |
a67509c3 | 1024 | clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1, |
5b0e95bf JB |
1025 | EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL, |
1026 | GFP_NOFS); | |
2f356126 | 1027 | goto out; |
0cb59c99 JB |
1028 | } |
1029 | leaf = path->nodes[0]; | |
1030 | if (ret > 0) { | |
1031 | struct btrfs_key found_key; | |
1032 | BUG_ON(!path->slots[0]); | |
1033 | path->slots[0]--; | |
1034 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1035 | if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID || | |
0414efae | 1036 | found_key.offset != offset) { |
a67509c3 JB |
1037 | clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, |
1038 | inode->i_size - 1, | |
5b0e95bf JB |
1039 | EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, |
1040 | NULL, GFP_NOFS); | |
b3b4aa74 | 1041 | btrfs_release_path(path); |
2f356126 | 1042 | goto out; |
0cb59c99 JB |
1043 | } |
1044 | } | |
549b4fdb JB |
1045 | |
1046 | BTRFS_I(inode)->generation = trans->transid; | |
0cb59c99 JB |
1047 | header = btrfs_item_ptr(leaf, path->slots[0], |
1048 | struct btrfs_free_space_header); | |
1049 | btrfs_set_free_space_entries(leaf, header, entries); | |
1050 | btrfs_set_free_space_bitmaps(leaf, header, bitmaps); | |
1051 | btrfs_set_free_space_generation(leaf, header, trans->transid); | |
1052 | btrfs_mark_buffer_dirty(leaf); | |
b3b4aa74 | 1053 | btrfs_release_path(path); |
0cb59c99 | 1054 | |
c09544e0 | 1055 | err = 0; |
2f356126 | 1056 | out: |
a67509c3 | 1057 | io_ctl_free(&io_ctl); |
c09544e0 | 1058 | if (err) { |
a67509c3 | 1059 | invalidate_inode_pages2(inode->i_mapping); |
0cb59c99 JB |
1060 | BTRFS_I(inode)->generation = 0; |
1061 | } | |
0cb59c99 | 1062 | btrfs_update_inode(trans, root, inode); |
c09544e0 | 1063 | return err; |
a67509c3 JB |
1064 | |
1065 | out_nospc: | |
1066 | list_for_each_safe(pos, n, &bitmap_list) { | |
1067 | struct btrfs_free_space *entry = | |
1068 | list_entry(pos, struct btrfs_free_space, list); | |
1069 | list_del_init(&entry->list); | |
1070 | } | |
1071 | io_ctl_drop_pages(&io_ctl); | |
1072 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0, | |
1073 | i_size_read(inode) - 1, &cached_state, GFP_NOFS); | |
1074 | goto out; | |
0414efae LZ |
1075 | } |
1076 | ||
1077 | int btrfs_write_out_cache(struct btrfs_root *root, | |
1078 | struct btrfs_trans_handle *trans, | |
1079 | struct btrfs_block_group_cache *block_group, | |
1080 | struct btrfs_path *path) | |
1081 | { | |
1082 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
1083 | struct inode *inode; | |
1084 | int ret = 0; | |
1085 | ||
1086 | root = root->fs_info->tree_root; | |
1087 | ||
1088 | spin_lock(&block_group->lock); | |
1089 | if (block_group->disk_cache_state < BTRFS_DC_SETUP) { | |
1090 | spin_unlock(&block_group->lock); | |
1091 | return 0; | |
1092 | } | |
1093 | spin_unlock(&block_group->lock); | |
1094 | ||
1095 | inode = lookup_free_space_inode(root, block_group, path); | |
1096 | if (IS_ERR(inode)) | |
1097 | return 0; | |
1098 | ||
1099 | ret = __btrfs_write_out_cache(root, inode, ctl, block_group, trans, | |
1100 | path, block_group->key.objectid); | |
c09544e0 | 1101 | if (ret) { |
0414efae LZ |
1102 | spin_lock(&block_group->lock); |
1103 | block_group->disk_cache_state = BTRFS_DC_ERROR; | |
1104 | spin_unlock(&block_group->lock); | |
82d5902d | 1105 | ret = 0; |
c09544e0 | 1106 | #ifdef DEBUG |
934e7d44 | 1107 | printk(KERN_ERR "btrfs: failed to write free space cache " |
0414efae | 1108 | "for block group %llu\n", block_group->key.objectid); |
c09544e0 | 1109 | #endif |
0414efae LZ |
1110 | } |
1111 | ||
0cb59c99 JB |
1112 | iput(inode); |
1113 | return ret; | |
1114 | } | |
1115 | ||
34d52cb6 | 1116 | static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit, |
96303081 | 1117 | u64 offset) |
0f9dd46c | 1118 | { |
96303081 JB |
1119 | BUG_ON(offset < bitmap_start); |
1120 | offset -= bitmap_start; | |
34d52cb6 | 1121 | return (unsigned long)(div_u64(offset, unit)); |
96303081 | 1122 | } |
0f9dd46c | 1123 | |
34d52cb6 | 1124 | static inline unsigned long bytes_to_bits(u64 bytes, u32 unit) |
96303081 | 1125 | { |
34d52cb6 | 1126 | return (unsigned long)(div_u64(bytes, unit)); |
96303081 | 1127 | } |
0f9dd46c | 1128 | |
34d52cb6 | 1129 | static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1130 | u64 offset) |
1131 | { | |
1132 | u64 bitmap_start; | |
1133 | u64 bytes_per_bitmap; | |
0f9dd46c | 1134 | |
34d52cb6 LZ |
1135 | bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit; |
1136 | bitmap_start = offset - ctl->start; | |
96303081 JB |
1137 | bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap); |
1138 | bitmap_start *= bytes_per_bitmap; | |
34d52cb6 | 1139 | bitmap_start += ctl->start; |
0f9dd46c | 1140 | |
96303081 | 1141 | return bitmap_start; |
0f9dd46c JB |
1142 | } |
1143 | ||
96303081 JB |
1144 | static int tree_insert_offset(struct rb_root *root, u64 offset, |
1145 | struct rb_node *node, int bitmap) | |
0f9dd46c JB |
1146 | { |
1147 | struct rb_node **p = &root->rb_node; | |
1148 | struct rb_node *parent = NULL; | |
1149 | struct btrfs_free_space *info; | |
1150 | ||
1151 | while (*p) { | |
1152 | parent = *p; | |
96303081 | 1153 | info = rb_entry(parent, struct btrfs_free_space, offset_index); |
0f9dd46c | 1154 | |
96303081 | 1155 | if (offset < info->offset) { |
0f9dd46c | 1156 | p = &(*p)->rb_left; |
96303081 | 1157 | } else if (offset > info->offset) { |
0f9dd46c | 1158 | p = &(*p)->rb_right; |
96303081 JB |
1159 | } else { |
1160 | /* | |
1161 | * we could have a bitmap entry and an extent entry | |
1162 | * share the same offset. If this is the case, we want | |
1163 | * the extent entry to always be found first if we do a | |
1164 | * linear search through the tree, since we want to have | |
1165 | * the quickest allocation time, and allocating from an | |
1166 | * extent is faster than allocating from a bitmap. So | |
1167 | * if we're inserting a bitmap and we find an entry at | |
1168 | * this offset, we want to go right, or after this entry | |
1169 | * logically. If we are inserting an extent and we've | |
1170 | * found a bitmap, we want to go left, or before | |
1171 | * logically. | |
1172 | */ | |
1173 | if (bitmap) { | |
207dde82 JB |
1174 | if (info->bitmap) { |
1175 | WARN_ON_ONCE(1); | |
1176 | return -EEXIST; | |
1177 | } | |
96303081 JB |
1178 | p = &(*p)->rb_right; |
1179 | } else { | |
207dde82 JB |
1180 | if (!info->bitmap) { |
1181 | WARN_ON_ONCE(1); | |
1182 | return -EEXIST; | |
1183 | } | |
96303081 JB |
1184 | p = &(*p)->rb_left; |
1185 | } | |
1186 | } | |
0f9dd46c JB |
1187 | } |
1188 | ||
1189 | rb_link_node(node, parent, p); | |
1190 | rb_insert_color(node, root); | |
1191 | ||
1192 | return 0; | |
1193 | } | |
1194 | ||
1195 | /* | |
70cb0743 JB |
1196 | * searches the tree for the given offset. |
1197 | * | |
96303081 JB |
1198 | * fuzzy - If this is set, then we are trying to make an allocation, and we just |
1199 | * want a section that has at least bytes size and comes at or after the given | |
1200 | * offset. | |
0f9dd46c | 1201 | */ |
96303081 | 1202 | static struct btrfs_free_space * |
34d52cb6 | 1203 | tree_search_offset(struct btrfs_free_space_ctl *ctl, |
96303081 | 1204 | u64 offset, int bitmap_only, int fuzzy) |
0f9dd46c | 1205 | { |
34d52cb6 | 1206 | struct rb_node *n = ctl->free_space_offset.rb_node; |
96303081 JB |
1207 | struct btrfs_free_space *entry, *prev = NULL; |
1208 | ||
1209 | /* find entry that is closest to the 'offset' */ | |
1210 | while (1) { | |
1211 | if (!n) { | |
1212 | entry = NULL; | |
1213 | break; | |
1214 | } | |
0f9dd46c | 1215 | |
0f9dd46c | 1216 | entry = rb_entry(n, struct btrfs_free_space, offset_index); |
96303081 | 1217 | prev = entry; |
0f9dd46c | 1218 | |
96303081 | 1219 | if (offset < entry->offset) |
0f9dd46c | 1220 | n = n->rb_left; |
96303081 | 1221 | else if (offset > entry->offset) |
0f9dd46c | 1222 | n = n->rb_right; |
96303081 | 1223 | else |
0f9dd46c | 1224 | break; |
0f9dd46c JB |
1225 | } |
1226 | ||
96303081 JB |
1227 | if (bitmap_only) { |
1228 | if (!entry) | |
1229 | return NULL; | |
1230 | if (entry->bitmap) | |
1231 | return entry; | |
0f9dd46c | 1232 | |
96303081 JB |
1233 | /* |
1234 | * bitmap entry and extent entry may share same offset, | |
1235 | * in that case, bitmap entry comes after extent entry. | |
1236 | */ | |
1237 | n = rb_next(n); | |
1238 | if (!n) | |
1239 | return NULL; | |
1240 | entry = rb_entry(n, struct btrfs_free_space, offset_index); | |
1241 | if (entry->offset != offset) | |
1242 | return NULL; | |
0f9dd46c | 1243 | |
96303081 JB |
1244 | WARN_ON(!entry->bitmap); |
1245 | return entry; | |
1246 | } else if (entry) { | |
1247 | if (entry->bitmap) { | |
0f9dd46c | 1248 | /* |
96303081 JB |
1249 | * if previous extent entry covers the offset, |
1250 | * we should return it instead of the bitmap entry | |
0f9dd46c | 1251 | */ |
96303081 JB |
1252 | n = &entry->offset_index; |
1253 | while (1) { | |
1254 | n = rb_prev(n); | |
1255 | if (!n) | |
1256 | break; | |
1257 | prev = rb_entry(n, struct btrfs_free_space, | |
1258 | offset_index); | |
1259 | if (!prev->bitmap) { | |
1260 | if (prev->offset + prev->bytes > offset) | |
1261 | entry = prev; | |
1262 | break; | |
1263 | } | |
0f9dd46c | 1264 | } |
96303081 JB |
1265 | } |
1266 | return entry; | |
1267 | } | |
1268 | ||
1269 | if (!prev) | |
1270 | return NULL; | |
1271 | ||
1272 | /* find last entry before the 'offset' */ | |
1273 | entry = prev; | |
1274 | if (entry->offset > offset) { | |
1275 | n = rb_prev(&entry->offset_index); | |
1276 | if (n) { | |
1277 | entry = rb_entry(n, struct btrfs_free_space, | |
1278 | offset_index); | |
1279 | BUG_ON(entry->offset > offset); | |
0f9dd46c | 1280 | } else { |
96303081 JB |
1281 | if (fuzzy) |
1282 | return entry; | |
1283 | else | |
1284 | return NULL; | |
0f9dd46c JB |
1285 | } |
1286 | } | |
1287 | ||
96303081 JB |
1288 | if (entry->bitmap) { |
1289 | n = &entry->offset_index; | |
1290 | while (1) { | |
1291 | n = rb_prev(n); | |
1292 | if (!n) | |
1293 | break; | |
1294 | prev = rb_entry(n, struct btrfs_free_space, | |
1295 | offset_index); | |
1296 | if (!prev->bitmap) { | |
1297 | if (prev->offset + prev->bytes > offset) | |
1298 | return prev; | |
1299 | break; | |
1300 | } | |
1301 | } | |
34d52cb6 | 1302 | if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset) |
96303081 JB |
1303 | return entry; |
1304 | } else if (entry->offset + entry->bytes > offset) | |
1305 | return entry; | |
1306 | ||
1307 | if (!fuzzy) | |
1308 | return NULL; | |
1309 | ||
1310 | while (1) { | |
1311 | if (entry->bitmap) { | |
1312 | if (entry->offset + BITS_PER_BITMAP * | |
34d52cb6 | 1313 | ctl->unit > offset) |
96303081 JB |
1314 | break; |
1315 | } else { | |
1316 | if (entry->offset + entry->bytes > offset) | |
1317 | break; | |
1318 | } | |
1319 | ||
1320 | n = rb_next(&entry->offset_index); | |
1321 | if (!n) | |
1322 | return NULL; | |
1323 | entry = rb_entry(n, struct btrfs_free_space, offset_index); | |
1324 | } | |
1325 | return entry; | |
0f9dd46c JB |
1326 | } |
1327 | ||
f333adb5 | 1328 | static inline void |
34d52cb6 | 1329 | __unlink_free_space(struct btrfs_free_space_ctl *ctl, |
f333adb5 | 1330 | struct btrfs_free_space *info) |
0f9dd46c | 1331 | { |
34d52cb6 LZ |
1332 | rb_erase(&info->offset_index, &ctl->free_space_offset); |
1333 | ctl->free_extents--; | |
f333adb5 LZ |
1334 | } |
1335 | ||
34d52cb6 | 1336 | static void unlink_free_space(struct btrfs_free_space_ctl *ctl, |
f333adb5 LZ |
1337 | struct btrfs_free_space *info) |
1338 | { | |
34d52cb6 LZ |
1339 | __unlink_free_space(ctl, info); |
1340 | ctl->free_space -= info->bytes; | |
0f9dd46c JB |
1341 | } |
1342 | ||
34d52cb6 | 1343 | static int link_free_space(struct btrfs_free_space_ctl *ctl, |
0f9dd46c JB |
1344 | struct btrfs_free_space *info) |
1345 | { | |
1346 | int ret = 0; | |
1347 | ||
96303081 | 1348 | BUG_ON(!info->bitmap && !info->bytes); |
34d52cb6 | 1349 | ret = tree_insert_offset(&ctl->free_space_offset, info->offset, |
96303081 | 1350 | &info->offset_index, (info->bitmap != NULL)); |
0f9dd46c JB |
1351 | if (ret) |
1352 | return ret; | |
1353 | ||
34d52cb6 LZ |
1354 | ctl->free_space += info->bytes; |
1355 | ctl->free_extents++; | |
96303081 JB |
1356 | return ret; |
1357 | } | |
1358 | ||
34d52cb6 | 1359 | static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl) |
96303081 | 1360 | { |
34d52cb6 | 1361 | struct btrfs_block_group_cache *block_group = ctl->private; |
25891f79 JB |
1362 | u64 max_bytes; |
1363 | u64 bitmap_bytes; | |
1364 | u64 extent_bytes; | |
8eb2d829 | 1365 | u64 size = block_group->key.offset; |
34d52cb6 LZ |
1366 | u64 bytes_per_bg = BITS_PER_BITMAP * block_group->sectorsize; |
1367 | int max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg); | |
1368 | ||
1369 | BUG_ON(ctl->total_bitmaps > max_bitmaps); | |
96303081 JB |
1370 | |
1371 | /* | |
1372 | * The goal is to keep the total amount of memory used per 1gb of space | |
1373 | * at or below 32k, so we need to adjust how much memory we allow to be | |
1374 | * used by extent based free space tracking | |
1375 | */ | |
8eb2d829 LZ |
1376 | if (size < 1024 * 1024 * 1024) |
1377 | max_bytes = MAX_CACHE_BYTES_PER_GIG; | |
1378 | else | |
1379 | max_bytes = MAX_CACHE_BYTES_PER_GIG * | |
1380 | div64_u64(size, 1024 * 1024 * 1024); | |
96303081 | 1381 | |
25891f79 JB |
1382 | /* |
1383 | * we want to account for 1 more bitmap than what we have so we can make | |
1384 | * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as | |
1385 | * we add more bitmaps. | |
1386 | */ | |
34d52cb6 | 1387 | bitmap_bytes = (ctl->total_bitmaps + 1) * PAGE_CACHE_SIZE; |
96303081 | 1388 | |
25891f79 | 1389 | if (bitmap_bytes >= max_bytes) { |
34d52cb6 | 1390 | ctl->extents_thresh = 0; |
25891f79 JB |
1391 | return; |
1392 | } | |
96303081 | 1393 | |
25891f79 JB |
1394 | /* |
1395 | * we want the extent entry threshold to always be at most 1/2 the maxw | |
1396 | * bytes we can have, or whatever is less than that. | |
1397 | */ | |
1398 | extent_bytes = max_bytes - bitmap_bytes; | |
1399 | extent_bytes = min_t(u64, extent_bytes, div64_u64(max_bytes, 2)); | |
96303081 | 1400 | |
34d52cb6 | 1401 | ctl->extents_thresh = |
25891f79 | 1402 | div64_u64(extent_bytes, (sizeof(struct btrfs_free_space))); |
96303081 JB |
1403 | } |
1404 | ||
bb3ac5a4 MX |
1405 | static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl, |
1406 | struct btrfs_free_space *info, | |
1407 | u64 offset, u64 bytes) | |
96303081 | 1408 | { |
f38b6e75 | 1409 | unsigned long start, count; |
96303081 | 1410 | |
34d52cb6 LZ |
1411 | start = offset_to_bit(info->offset, ctl->unit, offset); |
1412 | count = bytes_to_bits(bytes, ctl->unit); | |
f38b6e75 | 1413 | BUG_ON(start + count > BITS_PER_BITMAP); |
96303081 | 1414 | |
f38b6e75 | 1415 | bitmap_clear(info->bitmap, start, count); |
96303081 JB |
1416 | |
1417 | info->bytes -= bytes; | |
bb3ac5a4 MX |
1418 | } |
1419 | ||
1420 | static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl, | |
1421 | struct btrfs_free_space *info, u64 offset, | |
1422 | u64 bytes) | |
1423 | { | |
1424 | __bitmap_clear_bits(ctl, info, offset, bytes); | |
34d52cb6 | 1425 | ctl->free_space -= bytes; |
96303081 JB |
1426 | } |
1427 | ||
34d52cb6 | 1428 | static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl, |
817d52f8 JB |
1429 | struct btrfs_free_space *info, u64 offset, |
1430 | u64 bytes) | |
96303081 | 1431 | { |
f38b6e75 | 1432 | unsigned long start, count; |
96303081 | 1433 | |
34d52cb6 LZ |
1434 | start = offset_to_bit(info->offset, ctl->unit, offset); |
1435 | count = bytes_to_bits(bytes, ctl->unit); | |
f38b6e75 | 1436 | BUG_ON(start + count > BITS_PER_BITMAP); |
96303081 | 1437 | |
f38b6e75 | 1438 | bitmap_set(info->bitmap, start, count); |
96303081 JB |
1439 | |
1440 | info->bytes += bytes; | |
34d52cb6 | 1441 | ctl->free_space += bytes; |
96303081 JB |
1442 | } |
1443 | ||
34d52cb6 | 1444 | static int search_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1445 | struct btrfs_free_space *bitmap_info, u64 *offset, |
1446 | u64 *bytes) | |
1447 | { | |
1448 | unsigned long found_bits = 0; | |
1449 | unsigned long bits, i; | |
1450 | unsigned long next_zero; | |
1451 | ||
34d52cb6 | 1452 | i = offset_to_bit(bitmap_info->offset, ctl->unit, |
96303081 | 1453 | max_t(u64, *offset, bitmap_info->offset)); |
34d52cb6 | 1454 | bits = bytes_to_bits(*bytes, ctl->unit); |
96303081 JB |
1455 | |
1456 | for (i = find_next_bit(bitmap_info->bitmap, BITS_PER_BITMAP, i); | |
1457 | i < BITS_PER_BITMAP; | |
1458 | i = find_next_bit(bitmap_info->bitmap, BITS_PER_BITMAP, i + 1)) { | |
1459 | next_zero = find_next_zero_bit(bitmap_info->bitmap, | |
1460 | BITS_PER_BITMAP, i); | |
1461 | if ((next_zero - i) >= bits) { | |
1462 | found_bits = next_zero - i; | |
1463 | break; | |
1464 | } | |
1465 | i = next_zero; | |
1466 | } | |
1467 | ||
1468 | if (found_bits) { | |
34d52cb6 LZ |
1469 | *offset = (u64)(i * ctl->unit) + bitmap_info->offset; |
1470 | *bytes = (u64)(found_bits) * ctl->unit; | |
96303081 JB |
1471 | return 0; |
1472 | } | |
1473 | ||
1474 | return -1; | |
1475 | } | |
1476 | ||
34d52cb6 LZ |
1477 | static struct btrfs_free_space * |
1478 | find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes) | |
96303081 JB |
1479 | { |
1480 | struct btrfs_free_space *entry; | |
1481 | struct rb_node *node; | |
1482 | int ret; | |
1483 | ||
34d52cb6 | 1484 | if (!ctl->free_space_offset.rb_node) |
96303081 JB |
1485 | return NULL; |
1486 | ||
34d52cb6 | 1487 | entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1); |
96303081 JB |
1488 | if (!entry) |
1489 | return NULL; | |
1490 | ||
1491 | for (node = &entry->offset_index; node; node = rb_next(node)) { | |
1492 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
1493 | if (entry->bytes < *bytes) | |
1494 | continue; | |
1495 | ||
1496 | if (entry->bitmap) { | |
34d52cb6 | 1497 | ret = search_bitmap(ctl, entry, offset, bytes); |
96303081 JB |
1498 | if (!ret) |
1499 | return entry; | |
1500 | continue; | |
1501 | } | |
1502 | ||
1503 | *offset = entry->offset; | |
1504 | *bytes = entry->bytes; | |
1505 | return entry; | |
1506 | } | |
1507 | ||
1508 | return NULL; | |
1509 | } | |
1510 | ||
34d52cb6 | 1511 | static void add_new_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1512 | struct btrfs_free_space *info, u64 offset) |
1513 | { | |
34d52cb6 | 1514 | info->offset = offset_to_bitmap(ctl, offset); |
f019f426 | 1515 | info->bytes = 0; |
f2d0f676 | 1516 | INIT_LIST_HEAD(&info->list); |
34d52cb6 LZ |
1517 | link_free_space(ctl, info); |
1518 | ctl->total_bitmaps++; | |
96303081 | 1519 | |
34d52cb6 | 1520 | ctl->op->recalc_thresholds(ctl); |
96303081 JB |
1521 | } |
1522 | ||
34d52cb6 | 1523 | static void free_bitmap(struct btrfs_free_space_ctl *ctl, |
edf6e2d1 LZ |
1524 | struct btrfs_free_space *bitmap_info) |
1525 | { | |
34d52cb6 | 1526 | unlink_free_space(ctl, bitmap_info); |
edf6e2d1 | 1527 | kfree(bitmap_info->bitmap); |
dc89e982 | 1528 | kmem_cache_free(btrfs_free_space_cachep, bitmap_info); |
34d52cb6 LZ |
1529 | ctl->total_bitmaps--; |
1530 | ctl->op->recalc_thresholds(ctl); | |
edf6e2d1 LZ |
1531 | } |
1532 | ||
34d52cb6 | 1533 | static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1534 | struct btrfs_free_space *bitmap_info, |
1535 | u64 *offset, u64 *bytes) | |
1536 | { | |
1537 | u64 end; | |
6606bb97 JB |
1538 | u64 search_start, search_bytes; |
1539 | int ret; | |
96303081 JB |
1540 | |
1541 | again: | |
34d52cb6 | 1542 | end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1; |
96303081 | 1543 | |
6606bb97 JB |
1544 | /* |
1545 | * XXX - this can go away after a few releases. | |
1546 | * | |
1547 | * since the only user of btrfs_remove_free_space is the tree logging | |
1548 | * stuff, and the only way to test that is under crash conditions, we | |
1549 | * want to have this debug stuff here just in case somethings not | |
1550 | * working. Search the bitmap for the space we are trying to use to | |
1551 | * make sure its actually there. If its not there then we need to stop | |
1552 | * because something has gone wrong. | |
1553 | */ | |
1554 | search_start = *offset; | |
1555 | search_bytes = *bytes; | |
13dbc089 | 1556 | search_bytes = min(search_bytes, end - search_start + 1); |
34d52cb6 | 1557 | ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes); |
6606bb97 JB |
1558 | BUG_ON(ret < 0 || search_start != *offset); |
1559 | ||
96303081 | 1560 | if (*offset > bitmap_info->offset && *offset + *bytes > end) { |
34d52cb6 | 1561 | bitmap_clear_bits(ctl, bitmap_info, *offset, end - *offset + 1); |
96303081 JB |
1562 | *bytes -= end - *offset + 1; |
1563 | *offset = end + 1; | |
1564 | } else if (*offset >= bitmap_info->offset && *offset + *bytes <= end) { | |
34d52cb6 | 1565 | bitmap_clear_bits(ctl, bitmap_info, *offset, *bytes); |
96303081 JB |
1566 | *bytes = 0; |
1567 | } | |
1568 | ||
1569 | if (*bytes) { | |
6606bb97 | 1570 | struct rb_node *next = rb_next(&bitmap_info->offset_index); |
edf6e2d1 | 1571 | if (!bitmap_info->bytes) |
34d52cb6 | 1572 | free_bitmap(ctl, bitmap_info); |
96303081 | 1573 | |
6606bb97 JB |
1574 | /* |
1575 | * no entry after this bitmap, but we still have bytes to | |
1576 | * remove, so something has gone wrong. | |
1577 | */ | |
1578 | if (!next) | |
96303081 JB |
1579 | return -EINVAL; |
1580 | ||
6606bb97 JB |
1581 | bitmap_info = rb_entry(next, struct btrfs_free_space, |
1582 | offset_index); | |
1583 | ||
1584 | /* | |
1585 | * if the next entry isn't a bitmap we need to return to let the | |
1586 | * extent stuff do its work. | |
1587 | */ | |
96303081 JB |
1588 | if (!bitmap_info->bitmap) |
1589 | return -EAGAIN; | |
1590 | ||
6606bb97 JB |
1591 | /* |
1592 | * Ok the next item is a bitmap, but it may not actually hold | |
1593 | * the information for the rest of this free space stuff, so | |
1594 | * look for it, and if we don't find it return so we can try | |
1595 | * everything over again. | |
1596 | */ | |
1597 | search_start = *offset; | |
1598 | search_bytes = *bytes; | |
34d52cb6 | 1599 | ret = search_bitmap(ctl, bitmap_info, &search_start, |
6606bb97 JB |
1600 | &search_bytes); |
1601 | if (ret < 0 || search_start != *offset) | |
1602 | return -EAGAIN; | |
1603 | ||
96303081 | 1604 | goto again; |
edf6e2d1 | 1605 | } else if (!bitmap_info->bytes) |
34d52cb6 | 1606 | free_bitmap(ctl, bitmap_info); |
96303081 JB |
1607 | |
1608 | return 0; | |
1609 | } | |
1610 | ||
2cdc342c JB |
1611 | static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl, |
1612 | struct btrfs_free_space *info, u64 offset, | |
1613 | u64 bytes) | |
1614 | { | |
1615 | u64 bytes_to_set = 0; | |
1616 | u64 end; | |
1617 | ||
1618 | end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit); | |
1619 | ||
1620 | bytes_to_set = min(end - offset, bytes); | |
1621 | ||
1622 | bitmap_set_bits(ctl, info, offset, bytes_to_set); | |
1623 | ||
1624 | return bytes_to_set; | |
1625 | ||
1626 | } | |
1627 | ||
34d52cb6 LZ |
1628 | static bool use_bitmap(struct btrfs_free_space_ctl *ctl, |
1629 | struct btrfs_free_space *info) | |
96303081 | 1630 | { |
34d52cb6 | 1631 | struct btrfs_block_group_cache *block_group = ctl->private; |
96303081 JB |
1632 | |
1633 | /* | |
1634 | * If we are below the extents threshold then we can add this as an | |
1635 | * extent, and don't have to deal with the bitmap | |
1636 | */ | |
34d52cb6 | 1637 | if (ctl->free_extents < ctl->extents_thresh) { |
32cb0840 JB |
1638 | /* |
1639 | * If this block group has some small extents we don't want to | |
1640 | * use up all of our free slots in the cache with them, we want | |
1641 | * to reserve them to larger extents, however if we have plent | |
1642 | * of cache left then go ahead an dadd them, no sense in adding | |
1643 | * the overhead of a bitmap if we don't have to. | |
1644 | */ | |
1645 | if (info->bytes <= block_group->sectorsize * 4) { | |
34d52cb6 LZ |
1646 | if (ctl->free_extents * 2 <= ctl->extents_thresh) |
1647 | return false; | |
32cb0840 | 1648 | } else { |
34d52cb6 | 1649 | return false; |
32cb0840 JB |
1650 | } |
1651 | } | |
96303081 JB |
1652 | |
1653 | /* | |
1654 | * some block groups are so tiny they can't be enveloped by a bitmap, so | |
1655 | * don't even bother to create a bitmap for this | |
1656 | */ | |
1657 | if (BITS_PER_BITMAP * block_group->sectorsize > | |
1658 | block_group->key.offset) | |
34d52cb6 LZ |
1659 | return false; |
1660 | ||
1661 | return true; | |
1662 | } | |
1663 | ||
2cdc342c JB |
1664 | static struct btrfs_free_space_op free_space_op = { |
1665 | .recalc_thresholds = recalculate_thresholds, | |
1666 | .use_bitmap = use_bitmap, | |
1667 | }; | |
1668 | ||
34d52cb6 LZ |
1669 | static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl, |
1670 | struct btrfs_free_space *info) | |
1671 | { | |
1672 | struct btrfs_free_space *bitmap_info; | |
2cdc342c | 1673 | struct btrfs_block_group_cache *block_group = NULL; |
34d52cb6 | 1674 | int added = 0; |
2cdc342c | 1675 | u64 bytes, offset, bytes_added; |
34d52cb6 | 1676 | int ret; |
96303081 JB |
1677 | |
1678 | bytes = info->bytes; | |
1679 | offset = info->offset; | |
1680 | ||
34d52cb6 LZ |
1681 | if (!ctl->op->use_bitmap(ctl, info)) |
1682 | return 0; | |
1683 | ||
2cdc342c JB |
1684 | if (ctl->op == &free_space_op) |
1685 | block_group = ctl->private; | |
38e87880 | 1686 | again: |
2cdc342c JB |
1687 | /* |
1688 | * Since we link bitmaps right into the cluster we need to see if we | |
1689 | * have a cluster here, and if so and it has our bitmap we need to add | |
1690 | * the free space to that bitmap. | |
1691 | */ | |
1692 | if (block_group && !list_empty(&block_group->cluster_list)) { | |
1693 | struct btrfs_free_cluster *cluster; | |
1694 | struct rb_node *node; | |
1695 | struct btrfs_free_space *entry; | |
1696 | ||
1697 | cluster = list_entry(block_group->cluster_list.next, | |
1698 | struct btrfs_free_cluster, | |
1699 | block_group_list); | |
1700 | spin_lock(&cluster->lock); | |
1701 | node = rb_first(&cluster->root); | |
1702 | if (!node) { | |
1703 | spin_unlock(&cluster->lock); | |
38e87880 | 1704 | goto no_cluster_bitmap; |
2cdc342c JB |
1705 | } |
1706 | ||
1707 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
1708 | if (!entry->bitmap) { | |
1709 | spin_unlock(&cluster->lock); | |
38e87880 | 1710 | goto no_cluster_bitmap; |
2cdc342c JB |
1711 | } |
1712 | ||
1713 | if (entry->offset == offset_to_bitmap(ctl, offset)) { | |
1714 | bytes_added = add_bytes_to_bitmap(ctl, entry, | |
1715 | offset, bytes); | |
1716 | bytes -= bytes_added; | |
1717 | offset += bytes_added; | |
1718 | } | |
1719 | spin_unlock(&cluster->lock); | |
1720 | if (!bytes) { | |
1721 | ret = 1; | |
1722 | goto out; | |
1723 | } | |
1724 | } | |
38e87880 CM |
1725 | |
1726 | no_cluster_bitmap: | |
34d52cb6 | 1727 | bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), |
96303081 JB |
1728 | 1, 0); |
1729 | if (!bitmap_info) { | |
1730 | BUG_ON(added); | |
1731 | goto new_bitmap; | |
1732 | } | |
1733 | ||
2cdc342c JB |
1734 | bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes); |
1735 | bytes -= bytes_added; | |
1736 | offset += bytes_added; | |
1737 | added = 0; | |
96303081 JB |
1738 | |
1739 | if (!bytes) { | |
1740 | ret = 1; | |
1741 | goto out; | |
1742 | } else | |
1743 | goto again; | |
1744 | ||
1745 | new_bitmap: | |
1746 | if (info && info->bitmap) { | |
34d52cb6 | 1747 | add_new_bitmap(ctl, info, offset); |
96303081 JB |
1748 | added = 1; |
1749 | info = NULL; | |
1750 | goto again; | |
1751 | } else { | |
34d52cb6 | 1752 | spin_unlock(&ctl->tree_lock); |
96303081 JB |
1753 | |
1754 | /* no pre-allocated info, allocate a new one */ | |
1755 | if (!info) { | |
dc89e982 JB |
1756 | info = kmem_cache_zalloc(btrfs_free_space_cachep, |
1757 | GFP_NOFS); | |
96303081 | 1758 | if (!info) { |
34d52cb6 | 1759 | spin_lock(&ctl->tree_lock); |
96303081 JB |
1760 | ret = -ENOMEM; |
1761 | goto out; | |
1762 | } | |
1763 | } | |
1764 | ||
1765 | /* allocate the bitmap */ | |
1766 | info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS); | |
34d52cb6 | 1767 | spin_lock(&ctl->tree_lock); |
96303081 JB |
1768 | if (!info->bitmap) { |
1769 | ret = -ENOMEM; | |
1770 | goto out; | |
1771 | } | |
1772 | goto again; | |
1773 | } | |
1774 | ||
1775 | out: | |
1776 | if (info) { | |
1777 | if (info->bitmap) | |
1778 | kfree(info->bitmap); | |
dc89e982 | 1779 | kmem_cache_free(btrfs_free_space_cachep, info); |
96303081 | 1780 | } |
0f9dd46c JB |
1781 | |
1782 | return ret; | |
1783 | } | |
1784 | ||
945d8962 | 1785 | static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl, |
f333adb5 | 1786 | struct btrfs_free_space *info, bool update_stat) |
0f9dd46c | 1787 | { |
120d66ee LZ |
1788 | struct btrfs_free_space *left_info; |
1789 | struct btrfs_free_space *right_info; | |
1790 | bool merged = false; | |
1791 | u64 offset = info->offset; | |
1792 | u64 bytes = info->bytes; | |
6226cb0a | 1793 | |
0f9dd46c JB |
1794 | /* |
1795 | * first we want to see if there is free space adjacent to the range we | |
1796 | * are adding, if there is remove that struct and add a new one to | |
1797 | * cover the entire range | |
1798 | */ | |
34d52cb6 | 1799 | right_info = tree_search_offset(ctl, offset + bytes, 0, 0); |
96303081 JB |
1800 | if (right_info && rb_prev(&right_info->offset_index)) |
1801 | left_info = rb_entry(rb_prev(&right_info->offset_index), | |
1802 | struct btrfs_free_space, offset_index); | |
1803 | else | |
34d52cb6 | 1804 | left_info = tree_search_offset(ctl, offset - 1, 0, 0); |
0f9dd46c | 1805 | |
96303081 | 1806 | if (right_info && !right_info->bitmap) { |
f333adb5 | 1807 | if (update_stat) |
34d52cb6 | 1808 | unlink_free_space(ctl, right_info); |
f333adb5 | 1809 | else |
34d52cb6 | 1810 | __unlink_free_space(ctl, right_info); |
6226cb0a | 1811 | info->bytes += right_info->bytes; |
dc89e982 | 1812 | kmem_cache_free(btrfs_free_space_cachep, right_info); |
120d66ee | 1813 | merged = true; |
0f9dd46c JB |
1814 | } |
1815 | ||
96303081 JB |
1816 | if (left_info && !left_info->bitmap && |
1817 | left_info->offset + left_info->bytes == offset) { | |
f333adb5 | 1818 | if (update_stat) |
34d52cb6 | 1819 | unlink_free_space(ctl, left_info); |
f333adb5 | 1820 | else |
34d52cb6 | 1821 | __unlink_free_space(ctl, left_info); |
6226cb0a JB |
1822 | info->offset = left_info->offset; |
1823 | info->bytes += left_info->bytes; | |
dc89e982 | 1824 | kmem_cache_free(btrfs_free_space_cachep, left_info); |
120d66ee | 1825 | merged = true; |
0f9dd46c JB |
1826 | } |
1827 | ||
120d66ee LZ |
1828 | return merged; |
1829 | } | |
1830 | ||
581bb050 LZ |
1831 | int __btrfs_add_free_space(struct btrfs_free_space_ctl *ctl, |
1832 | u64 offset, u64 bytes) | |
120d66ee LZ |
1833 | { |
1834 | struct btrfs_free_space *info; | |
1835 | int ret = 0; | |
1836 | ||
dc89e982 | 1837 | info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS); |
120d66ee LZ |
1838 | if (!info) |
1839 | return -ENOMEM; | |
1840 | ||
1841 | info->offset = offset; | |
1842 | info->bytes = bytes; | |
1843 | ||
34d52cb6 | 1844 | spin_lock(&ctl->tree_lock); |
120d66ee | 1845 | |
34d52cb6 | 1846 | if (try_merge_free_space(ctl, info, true)) |
120d66ee LZ |
1847 | goto link; |
1848 | ||
1849 | /* | |
1850 | * There was no extent directly to the left or right of this new | |
1851 | * extent then we know we're going to have to allocate a new extent, so | |
1852 | * before we do that see if we need to drop this into a bitmap | |
1853 | */ | |
34d52cb6 | 1854 | ret = insert_into_bitmap(ctl, info); |
120d66ee LZ |
1855 | if (ret < 0) { |
1856 | goto out; | |
1857 | } else if (ret) { | |
1858 | ret = 0; | |
1859 | goto out; | |
1860 | } | |
1861 | link: | |
34d52cb6 | 1862 | ret = link_free_space(ctl, info); |
0f9dd46c | 1863 | if (ret) |
dc89e982 | 1864 | kmem_cache_free(btrfs_free_space_cachep, info); |
96303081 | 1865 | out: |
34d52cb6 | 1866 | spin_unlock(&ctl->tree_lock); |
6226cb0a | 1867 | |
0f9dd46c | 1868 | if (ret) { |
96303081 | 1869 | printk(KERN_CRIT "btrfs: unable to add free space :%d\n", ret); |
c293498b | 1870 | BUG_ON(ret == -EEXIST); |
0f9dd46c JB |
1871 | } |
1872 | ||
0f9dd46c JB |
1873 | return ret; |
1874 | } | |
1875 | ||
6226cb0a JB |
1876 | int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group, |
1877 | u64 offset, u64 bytes) | |
0f9dd46c | 1878 | { |
34d52cb6 | 1879 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0f9dd46c | 1880 | struct btrfs_free_space *info; |
96303081 | 1881 | struct btrfs_free_space *next_info = NULL; |
0f9dd46c JB |
1882 | int ret = 0; |
1883 | ||
34d52cb6 | 1884 | spin_lock(&ctl->tree_lock); |
6226cb0a | 1885 | |
96303081 | 1886 | again: |
34d52cb6 | 1887 | info = tree_search_offset(ctl, offset, 0, 0); |
96303081 | 1888 | if (!info) { |
6606bb97 JB |
1889 | /* |
1890 | * oops didn't find an extent that matched the space we wanted | |
1891 | * to remove, look for a bitmap instead | |
1892 | */ | |
34d52cb6 | 1893 | info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), |
6606bb97 JB |
1894 | 1, 0); |
1895 | if (!info) { | |
24a70313 CM |
1896 | /* the tree logging code might be calling us before we |
1897 | * have fully loaded the free space rbtree for this | |
1898 | * block group. So it is possible the entry won't | |
1899 | * be in the rbtree yet at all. The caching code | |
1900 | * will make sure not to put it in the rbtree if | |
1901 | * the logging code has pinned it. | |
1902 | */ | |
6606bb97 JB |
1903 | goto out_lock; |
1904 | } | |
96303081 JB |
1905 | } |
1906 | ||
1907 | if (info->bytes < bytes && rb_next(&info->offset_index)) { | |
1908 | u64 end; | |
1909 | next_info = rb_entry(rb_next(&info->offset_index), | |
1910 | struct btrfs_free_space, | |
1911 | offset_index); | |
1912 | ||
1913 | if (next_info->bitmap) | |
34d52cb6 LZ |
1914 | end = next_info->offset + |
1915 | BITS_PER_BITMAP * ctl->unit - 1; | |
96303081 JB |
1916 | else |
1917 | end = next_info->offset + next_info->bytes; | |
1918 | ||
1919 | if (next_info->bytes < bytes || | |
1920 | next_info->offset > offset || offset > end) { | |
1921 | printk(KERN_CRIT "Found free space at %llu, size %llu," | |
1922 | " trying to use %llu\n", | |
1923 | (unsigned long long)info->offset, | |
1924 | (unsigned long long)info->bytes, | |
1925 | (unsigned long long)bytes); | |
0f9dd46c JB |
1926 | WARN_ON(1); |
1927 | ret = -EINVAL; | |
96303081 | 1928 | goto out_lock; |
0f9dd46c | 1929 | } |
0f9dd46c | 1930 | |
96303081 JB |
1931 | info = next_info; |
1932 | } | |
1933 | ||
1934 | if (info->bytes == bytes) { | |
34d52cb6 | 1935 | unlink_free_space(ctl, info); |
96303081 JB |
1936 | if (info->bitmap) { |
1937 | kfree(info->bitmap); | |
34d52cb6 | 1938 | ctl->total_bitmaps--; |
0f9dd46c | 1939 | } |
dc89e982 | 1940 | kmem_cache_free(btrfs_free_space_cachep, info); |
1eae31e9 | 1941 | ret = 0; |
96303081 JB |
1942 | goto out_lock; |
1943 | } | |
0f9dd46c | 1944 | |
96303081 | 1945 | if (!info->bitmap && info->offset == offset) { |
34d52cb6 | 1946 | unlink_free_space(ctl, info); |
0f9dd46c JB |
1947 | info->offset += bytes; |
1948 | info->bytes -= bytes; | |
1eae31e9 CM |
1949 | ret = link_free_space(ctl, info); |
1950 | WARN_ON(ret); | |
96303081 JB |
1951 | goto out_lock; |
1952 | } | |
0f9dd46c | 1953 | |
96303081 JB |
1954 | if (!info->bitmap && info->offset <= offset && |
1955 | info->offset + info->bytes >= offset + bytes) { | |
9b49c9b9 CM |
1956 | u64 old_start = info->offset; |
1957 | /* | |
1958 | * we're freeing space in the middle of the info, | |
1959 | * this can happen during tree log replay | |
1960 | * | |
1961 | * first unlink the old info and then | |
1962 | * insert it again after the hole we're creating | |
1963 | */ | |
34d52cb6 | 1964 | unlink_free_space(ctl, info); |
9b49c9b9 CM |
1965 | if (offset + bytes < info->offset + info->bytes) { |
1966 | u64 old_end = info->offset + info->bytes; | |
1967 | ||
1968 | info->offset = offset + bytes; | |
1969 | info->bytes = old_end - info->offset; | |
34d52cb6 | 1970 | ret = link_free_space(ctl, info); |
96303081 JB |
1971 | WARN_ON(ret); |
1972 | if (ret) | |
1973 | goto out_lock; | |
9b49c9b9 CM |
1974 | } else { |
1975 | /* the hole we're creating ends at the end | |
1976 | * of the info struct, just free the info | |
1977 | */ | |
dc89e982 | 1978 | kmem_cache_free(btrfs_free_space_cachep, info); |
9b49c9b9 | 1979 | } |
34d52cb6 | 1980 | spin_unlock(&ctl->tree_lock); |
96303081 JB |
1981 | |
1982 | /* step two, insert a new info struct to cover | |
1983 | * anything before the hole | |
9b49c9b9 | 1984 | */ |
6226cb0a JB |
1985 | ret = btrfs_add_free_space(block_group, old_start, |
1986 | offset - old_start); | |
79787eaa | 1987 | WARN_ON(ret); /* -ENOMEM */ |
96303081 | 1988 | goto out; |
0f9dd46c | 1989 | } |
96303081 | 1990 | |
34d52cb6 | 1991 | ret = remove_from_bitmap(ctl, info, &offset, &bytes); |
96303081 JB |
1992 | if (ret == -EAGAIN) |
1993 | goto again; | |
79787eaa | 1994 | BUG_ON(ret); /* logic error */ |
96303081 | 1995 | out_lock: |
34d52cb6 | 1996 | spin_unlock(&ctl->tree_lock); |
0f9dd46c | 1997 | out: |
25179201 JB |
1998 | return ret; |
1999 | } | |
2000 | ||
0f9dd46c JB |
2001 | void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group, |
2002 | u64 bytes) | |
2003 | { | |
34d52cb6 | 2004 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0f9dd46c JB |
2005 | struct btrfs_free_space *info; |
2006 | struct rb_node *n; | |
2007 | int count = 0; | |
2008 | ||
34d52cb6 | 2009 | for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) { |
0f9dd46c JB |
2010 | info = rb_entry(n, struct btrfs_free_space, offset_index); |
2011 | if (info->bytes >= bytes) | |
2012 | count++; | |
96303081 | 2013 | printk(KERN_CRIT "entry offset %llu, bytes %llu, bitmap %s\n", |
21380931 | 2014 | (unsigned long long)info->offset, |
96303081 JB |
2015 | (unsigned long long)info->bytes, |
2016 | (info->bitmap) ? "yes" : "no"); | |
0f9dd46c | 2017 | } |
96303081 JB |
2018 | printk(KERN_INFO "block group has cluster?: %s\n", |
2019 | list_empty(&block_group->cluster_list) ? "no" : "yes"); | |
0f9dd46c JB |
2020 | printk(KERN_INFO "%d blocks of free space at or bigger than bytes is" |
2021 | "\n", count); | |
2022 | } | |
2023 | ||
34d52cb6 | 2024 | void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group) |
0f9dd46c | 2025 | { |
34d52cb6 | 2026 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0f9dd46c | 2027 | |
34d52cb6 LZ |
2028 | spin_lock_init(&ctl->tree_lock); |
2029 | ctl->unit = block_group->sectorsize; | |
2030 | ctl->start = block_group->key.objectid; | |
2031 | ctl->private = block_group; | |
2032 | ctl->op = &free_space_op; | |
0f9dd46c | 2033 | |
34d52cb6 LZ |
2034 | /* |
2035 | * we only want to have 32k of ram per block group for keeping | |
2036 | * track of free space, and if we pass 1/2 of that we want to | |
2037 | * start converting things over to using bitmaps | |
2038 | */ | |
2039 | ctl->extents_thresh = ((1024 * 32) / 2) / | |
2040 | sizeof(struct btrfs_free_space); | |
0f9dd46c JB |
2041 | } |
2042 | ||
fa9c0d79 CM |
2043 | /* |
2044 | * for a given cluster, put all of its extents back into the free | |
2045 | * space cache. If the block group passed doesn't match the block group | |
2046 | * pointed to by the cluster, someone else raced in and freed the | |
2047 | * cluster already. In that case, we just return without changing anything | |
2048 | */ | |
2049 | static int | |
2050 | __btrfs_return_cluster_to_free_space( | |
2051 | struct btrfs_block_group_cache *block_group, | |
2052 | struct btrfs_free_cluster *cluster) | |
2053 | { | |
34d52cb6 | 2054 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
fa9c0d79 CM |
2055 | struct btrfs_free_space *entry; |
2056 | struct rb_node *node; | |
2057 | ||
2058 | spin_lock(&cluster->lock); | |
2059 | if (cluster->block_group != block_group) | |
2060 | goto out; | |
2061 | ||
96303081 | 2062 | cluster->block_group = NULL; |
fa9c0d79 | 2063 | cluster->window_start = 0; |
96303081 | 2064 | list_del_init(&cluster->block_group_list); |
96303081 | 2065 | |
fa9c0d79 | 2066 | node = rb_first(&cluster->root); |
96303081 | 2067 | while (node) { |
4e69b598 JB |
2068 | bool bitmap; |
2069 | ||
fa9c0d79 CM |
2070 | entry = rb_entry(node, struct btrfs_free_space, offset_index); |
2071 | node = rb_next(&entry->offset_index); | |
2072 | rb_erase(&entry->offset_index, &cluster->root); | |
4e69b598 JB |
2073 | |
2074 | bitmap = (entry->bitmap != NULL); | |
2075 | if (!bitmap) | |
34d52cb6 LZ |
2076 | try_merge_free_space(ctl, entry, false); |
2077 | tree_insert_offset(&ctl->free_space_offset, | |
4e69b598 | 2078 | entry->offset, &entry->offset_index, bitmap); |
fa9c0d79 | 2079 | } |
6bef4d31 | 2080 | cluster->root = RB_ROOT; |
96303081 | 2081 | |
fa9c0d79 CM |
2082 | out: |
2083 | spin_unlock(&cluster->lock); | |
96303081 | 2084 | btrfs_put_block_group(block_group); |
fa9c0d79 CM |
2085 | return 0; |
2086 | } | |
2087 | ||
09655373 | 2088 | void __btrfs_remove_free_space_cache_locked(struct btrfs_free_space_ctl *ctl) |
0f9dd46c JB |
2089 | { |
2090 | struct btrfs_free_space *info; | |
2091 | struct rb_node *node; | |
581bb050 | 2092 | |
581bb050 LZ |
2093 | while ((node = rb_last(&ctl->free_space_offset)) != NULL) { |
2094 | info = rb_entry(node, struct btrfs_free_space, offset_index); | |
9b90f513 JB |
2095 | if (!info->bitmap) { |
2096 | unlink_free_space(ctl, info); | |
2097 | kmem_cache_free(btrfs_free_space_cachep, info); | |
2098 | } else { | |
2099 | free_bitmap(ctl, info); | |
2100 | } | |
581bb050 LZ |
2101 | if (need_resched()) { |
2102 | spin_unlock(&ctl->tree_lock); | |
2103 | cond_resched(); | |
2104 | spin_lock(&ctl->tree_lock); | |
2105 | } | |
2106 | } | |
09655373 CM |
2107 | } |
2108 | ||
2109 | void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl) | |
2110 | { | |
2111 | spin_lock(&ctl->tree_lock); | |
2112 | __btrfs_remove_free_space_cache_locked(ctl); | |
581bb050 LZ |
2113 | spin_unlock(&ctl->tree_lock); |
2114 | } | |
2115 | ||
2116 | void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group) | |
2117 | { | |
2118 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
fa9c0d79 | 2119 | struct btrfs_free_cluster *cluster; |
96303081 | 2120 | struct list_head *head; |
0f9dd46c | 2121 | |
34d52cb6 | 2122 | spin_lock(&ctl->tree_lock); |
96303081 JB |
2123 | while ((head = block_group->cluster_list.next) != |
2124 | &block_group->cluster_list) { | |
2125 | cluster = list_entry(head, struct btrfs_free_cluster, | |
2126 | block_group_list); | |
fa9c0d79 CM |
2127 | |
2128 | WARN_ON(cluster->block_group != block_group); | |
2129 | __btrfs_return_cluster_to_free_space(block_group, cluster); | |
96303081 | 2130 | if (need_resched()) { |
34d52cb6 | 2131 | spin_unlock(&ctl->tree_lock); |
96303081 | 2132 | cond_resched(); |
34d52cb6 | 2133 | spin_lock(&ctl->tree_lock); |
96303081 | 2134 | } |
fa9c0d79 | 2135 | } |
09655373 | 2136 | __btrfs_remove_free_space_cache_locked(ctl); |
34d52cb6 | 2137 | spin_unlock(&ctl->tree_lock); |
fa9c0d79 | 2138 | |
0f9dd46c JB |
2139 | } |
2140 | ||
6226cb0a JB |
2141 | u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group, |
2142 | u64 offset, u64 bytes, u64 empty_size) | |
0f9dd46c | 2143 | { |
34d52cb6 | 2144 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
6226cb0a | 2145 | struct btrfs_free_space *entry = NULL; |
96303081 | 2146 | u64 bytes_search = bytes + empty_size; |
6226cb0a | 2147 | u64 ret = 0; |
0f9dd46c | 2148 | |
34d52cb6 LZ |
2149 | spin_lock(&ctl->tree_lock); |
2150 | entry = find_free_space(ctl, &offset, &bytes_search); | |
6226cb0a | 2151 | if (!entry) |
96303081 JB |
2152 | goto out; |
2153 | ||
2154 | ret = offset; | |
2155 | if (entry->bitmap) { | |
34d52cb6 | 2156 | bitmap_clear_bits(ctl, entry, offset, bytes); |
edf6e2d1 | 2157 | if (!entry->bytes) |
34d52cb6 | 2158 | free_bitmap(ctl, entry); |
96303081 | 2159 | } else { |
34d52cb6 | 2160 | unlink_free_space(ctl, entry); |
6226cb0a JB |
2161 | entry->offset += bytes; |
2162 | entry->bytes -= bytes; | |
6226cb0a | 2163 | if (!entry->bytes) |
dc89e982 | 2164 | kmem_cache_free(btrfs_free_space_cachep, entry); |
6226cb0a | 2165 | else |
34d52cb6 | 2166 | link_free_space(ctl, entry); |
6226cb0a | 2167 | } |
0f9dd46c | 2168 | |
96303081 | 2169 | out: |
34d52cb6 | 2170 | spin_unlock(&ctl->tree_lock); |
817d52f8 | 2171 | |
0f9dd46c JB |
2172 | return ret; |
2173 | } | |
fa9c0d79 CM |
2174 | |
2175 | /* | |
2176 | * given a cluster, put all of its extents back into the free space | |
2177 | * cache. If a block group is passed, this function will only free | |
2178 | * a cluster that belongs to the passed block group. | |
2179 | * | |
2180 | * Otherwise, it'll get a reference on the block group pointed to by the | |
2181 | * cluster and remove the cluster from it. | |
2182 | */ | |
2183 | int btrfs_return_cluster_to_free_space( | |
2184 | struct btrfs_block_group_cache *block_group, | |
2185 | struct btrfs_free_cluster *cluster) | |
2186 | { | |
34d52cb6 | 2187 | struct btrfs_free_space_ctl *ctl; |
fa9c0d79 CM |
2188 | int ret; |
2189 | ||
2190 | /* first, get a safe pointer to the block group */ | |
2191 | spin_lock(&cluster->lock); | |
2192 | if (!block_group) { | |
2193 | block_group = cluster->block_group; | |
2194 | if (!block_group) { | |
2195 | spin_unlock(&cluster->lock); | |
2196 | return 0; | |
2197 | } | |
2198 | } else if (cluster->block_group != block_group) { | |
2199 | /* someone else has already freed it don't redo their work */ | |
2200 | spin_unlock(&cluster->lock); | |
2201 | return 0; | |
2202 | } | |
2203 | atomic_inc(&block_group->count); | |
2204 | spin_unlock(&cluster->lock); | |
2205 | ||
34d52cb6 LZ |
2206 | ctl = block_group->free_space_ctl; |
2207 | ||
fa9c0d79 | 2208 | /* now return any extents the cluster had on it */ |
34d52cb6 | 2209 | spin_lock(&ctl->tree_lock); |
fa9c0d79 | 2210 | ret = __btrfs_return_cluster_to_free_space(block_group, cluster); |
34d52cb6 | 2211 | spin_unlock(&ctl->tree_lock); |
fa9c0d79 CM |
2212 | |
2213 | /* finally drop our ref */ | |
2214 | btrfs_put_block_group(block_group); | |
2215 | return ret; | |
2216 | } | |
2217 | ||
96303081 JB |
2218 | static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group, |
2219 | struct btrfs_free_cluster *cluster, | |
4e69b598 | 2220 | struct btrfs_free_space *entry, |
96303081 JB |
2221 | u64 bytes, u64 min_start) |
2222 | { | |
34d52cb6 | 2223 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
96303081 JB |
2224 | int err; |
2225 | u64 search_start = cluster->window_start; | |
2226 | u64 search_bytes = bytes; | |
2227 | u64 ret = 0; | |
2228 | ||
96303081 JB |
2229 | search_start = min_start; |
2230 | search_bytes = bytes; | |
2231 | ||
34d52cb6 | 2232 | err = search_bitmap(ctl, entry, &search_start, &search_bytes); |
96303081 | 2233 | if (err) |
4e69b598 | 2234 | return 0; |
96303081 JB |
2235 | |
2236 | ret = search_start; | |
bb3ac5a4 | 2237 | __bitmap_clear_bits(ctl, entry, ret, bytes); |
96303081 JB |
2238 | |
2239 | return ret; | |
2240 | } | |
2241 | ||
fa9c0d79 CM |
2242 | /* |
2243 | * given a cluster, try to allocate 'bytes' from it, returns 0 | |
2244 | * if it couldn't find anything suitably large, or a logical disk offset | |
2245 | * if things worked out | |
2246 | */ | |
2247 | u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group, | |
2248 | struct btrfs_free_cluster *cluster, u64 bytes, | |
2249 | u64 min_start) | |
2250 | { | |
34d52cb6 | 2251 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
fa9c0d79 CM |
2252 | struct btrfs_free_space *entry = NULL; |
2253 | struct rb_node *node; | |
2254 | u64 ret = 0; | |
2255 | ||
2256 | spin_lock(&cluster->lock); | |
2257 | if (bytes > cluster->max_size) | |
2258 | goto out; | |
2259 | ||
2260 | if (cluster->block_group != block_group) | |
2261 | goto out; | |
2262 | ||
2263 | node = rb_first(&cluster->root); | |
2264 | if (!node) | |
2265 | goto out; | |
2266 | ||
2267 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
fa9c0d79 | 2268 | while(1) { |
4e69b598 JB |
2269 | if (entry->bytes < bytes || |
2270 | (!entry->bitmap && entry->offset < min_start)) { | |
fa9c0d79 CM |
2271 | node = rb_next(&entry->offset_index); |
2272 | if (!node) | |
2273 | break; | |
2274 | entry = rb_entry(node, struct btrfs_free_space, | |
2275 | offset_index); | |
2276 | continue; | |
2277 | } | |
fa9c0d79 | 2278 | |
4e69b598 JB |
2279 | if (entry->bitmap) { |
2280 | ret = btrfs_alloc_from_bitmap(block_group, | |
2281 | cluster, entry, bytes, | |
0b4a9d24 | 2282 | cluster->window_start); |
4e69b598 | 2283 | if (ret == 0) { |
4e69b598 JB |
2284 | node = rb_next(&entry->offset_index); |
2285 | if (!node) | |
2286 | break; | |
2287 | entry = rb_entry(node, struct btrfs_free_space, | |
2288 | offset_index); | |
2289 | continue; | |
2290 | } | |
9b230628 | 2291 | cluster->window_start += bytes; |
4e69b598 | 2292 | } else { |
4e69b598 JB |
2293 | ret = entry->offset; |
2294 | ||
2295 | entry->offset += bytes; | |
2296 | entry->bytes -= bytes; | |
2297 | } | |
fa9c0d79 | 2298 | |
5e71b5d5 | 2299 | if (entry->bytes == 0) |
fa9c0d79 | 2300 | rb_erase(&entry->offset_index, &cluster->root); |
fa9c0d79 CM |
2301 | break; |
2302 | } | |
2303 | out: | |
2304 | spin_unlock(&cluster->lock); | |
96303081 | 2305 | |
5e71b5d5 LZ |
2306 | if (!ret) |
2307 | return 0; | |
2308 | ||
34d52cb6 | 2309 | spin_lock(&ctl->tree_lock); |
5e71b5d5 | 2310 | |
34d52cb6 | 2311 | ctl->free_space -= bytes; |
5e71b5d5 | 2312 | if (entry->bytes == 0) { |
34d52cb6 | 2313 | ctl->free_extents--; |
4e69b598 JB |
2314 | if (entry->bitmap) { |
2315 | kfree(entry->bitmap); | |
34d52cb6 LZ |
2316 | ctl->total_bitmaps--; |
2317 | ctl->op->recalc_thresholds(ctl); | |
4e69b598 | 2318 | } |
dc89e982 | 2319 | kmem_cache_free(btrfs_free_space_cachep, entry); |
5e71b5d5 LZ |
2320 | } |
2321 | ||
34d52cb6 | 2322 | spin_unlock(&ctl->tree_lock); |
5e71b5d5 | 2323 | |
fa9c0d79 CM |
2324 | return ret; |
2325 | } | |
2326 | ||
96303081 JB |
2327 | static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group, |
2328 | struct btrfs_free_space *entry, | |
2329 | struct btrfs_free_cluster *cluster, | |
1bb91902 AO |
2330 | u64 offset, u64 bytes, |
2331 | u64 cont1_bytes, u64 min_bytes) | |
96303081 | 2332 | { |
34d52cb6 | 2333 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
96303081 JB |
2334 | unsigned long next_zero; |
2335 | unsigned long i; | |
1bb91902 AO |
2336 | unsigned long want_bits; |
2337 | unsigned long min_bits; | |
96303081 JB |
2338 | unsigned long found_bits; |
2339 | unsigned long start = 0; | |
2340 | unsigned long total_found = 0; | |
4e69b598 | 2341 | int ret; |
96303081 JB |
2342 | |
2343 | i = offset_to_bit(entry->offset, block_group->sectorsize, | |
2344 | max_t(u64, offset, entry->offset)); | |
1bb91902 AO |
2345 | want_bits = bytes_to_bits(bytes, block_group->sectorsize); |
2346 | min_bits = bytes_to_bits(min_bytes, block_group->sectorsize); | |
96303081 JB |
2347 | |
2348 | again: | |
2349 | found_bits = 0; | |
2350 | for (i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, i); | |
2351 | i < BITS_PER_BITMAP; | |
2352 | i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, i + 1)) { | |
2353 | next_zero = find_next_zero_bit(entry->bitmap, | |
2354 | BITS_PER_BITMAP, i); | |
1bb91902 | 2355 | if (next_zero - i >= min_bits) { |
96303081 JB |
2356 | found_bits = next_zero - i; |
2357 | break; | |
2358 | } | |
2359 | i = next_zero; | |
2360 | } | |
2361 | ||
2362 | if (!found_bits) | |
4e69b598 | 2363 | return -ENOSPC; |
96303081 | 2364 | |
1bb91902 | 2365 | if (!total_found) { |
96303081 | 2366 | start = i; |
b78d09bc | 2367 | cluster->max_size = 0; |
96303081 JB |
2368 | } |
2369 | ||
2370 | total_found += found_bits; | |
2371 | ||
2372 | if (cluster->max_size < found_bits * block_group->sectorsize) | |
2373 | cluster->max_size = found_bits * block_group->sectorsize; | |
2374 | ||
1bb91902 AO |
2375 | if (total_found < want_bits || cluster->max_size < cont1_bytes) { |
2376 | i = next_zero + 1; | |
96303081 JB |
2377 | goto again; |
2378 | } | |
2379 | ||
2380 | cluster->window_start = start * block_group->sectorsize + | |
2381 | entry->offset; | |
34d52cb6 | 2382 | rb_erase(&entry->offset_index, &ctl->free_space_offset); |
4e69b598 JB |
2383 | ret = tree_insert_offset(&cluster->root, entry->offset, |
2384 | &entry->offset_index, 1); | |
79787eaa | 2385 | BUG_ON(ret); /* -EEXIST; Logic error */ |
96303081 | 2386 | |
3f7de037 JB |
2387 | trace_btrfs_setup_cluster(block_group, cluster, |
2388 | total_found * block_group->sectorsize, 1); | |
96303081 JB |
2389 | return 0; |
2390 | } | |
2391 | ||
4e69b598 JB |
2392 | /* |
2393 | * This searches the block group for just extents to fill the cluster with. | |
1bb91902 AO |
2394 | * Try to find a cluster with at least bytes total bytes, at least one |
2395 | * extent of cont1_bytes, and other clusters of at least min_bytes. | |
4e69b598 | 2396 | */ |
3de85bb9 JB |
2397 | static noinline int |
2398 | setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group, | |
2399 | struct btrfs_free_cluster *cluster, | |
2400 | struct list_head *bitmaps, u64 offset, u64 bytes, | |
1bb91902 | 2401 | u64 cont1_bytes, u64 min_bytes) |
4e69b598 | 2402 | { |
34d52cb6 | 2403 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
4e69b598 JB |
2404 | struct btrfs_free_space *first = NULL; |
2405 | struct btrfs_free_space *entry = NULL; | |
4e69b598 JB |
2406 | struct btrfs_free_space *last; |
2407 | struct rb_node *node; | |
2408 | u64 window_start; | |
2409 | u64 window_free; | |
2410 | u64 max_extent; | |
3f7de037 | 2411 | u64 total_size = 0; |
4e69b598 | 2412 | |
34d52cb6 | 2413 | entry = tree_search_offset(ctl, offset, 0, 1); |
4e69b598 JB |
2414 | if (!entry) |
2415 | return -ENOSPC; | |
2416 | ||
2417 | /* | |
2418 | * We don't want bitmaps, so just move along until we find a normal | |
2419 | * extent entry. | |
2420 | */ | |
1bb91902 AO |
2421 | while (entry->bitmap || entry->bytes < min_bytes) { |
2422 | if (entry->bitmap && list_empty(&entry->list)) | |
86d4a77b | 2423 | list_add_tail(&entry->list, bitmaps); |
4e69b598 JB |
2424 | node = rb_next(&entry->offset_index); |
2425 | if (!node) | |
2426 | return -ENOSPC; | |
2427 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
2428 | } | |
2429 | ||
2430 | window_start = entry->offset; | |
2431 | window_free = entry->bytes; | |
2432 | max_extent = entry->bytes; | |
2433 | first = entry; | |
2434 | last = entry; | |
4e69b598 | 2435 | |
1bb91902 AO |
2436 | for (node = rb_next(&entry->offset_index); node; |
2437 | node = rb_next(&entry->offset_index)) { | |
4e69b598 JB |
2438 | entry = rb_entry(node, struct btrfs_free_space, offset_index); |
2439 | ||
86d4a77b JB |
2440 | if (entry->bitmap) { |
2441 | if (list_empty(&entry->list)) | |
2442 | list_add_tail(&entry->list, bitmaps); | |
4e69b598 | 2443 | continue; |
86d4a77b JB |
2444 | } |
2445 | ||
1bb91902 AO |
2446 | if (entry->bytes < min_bytes) |
2447 | continue; | |
2448 | ||
2449 | last = entry; | |
2450 | window_free += entry->bytes; | |
2451 | if (entry->bytes > max_extent) | |
4e69b598 | 2452 | max_extent = entry->bytes; |
4e69b598 JB |
2453 | } |
2454 | ||
1bb91902 AO |
2455 | if (window_free < bytes || max_extent < cont1_bytes) |
2456 | return -ENOSPC; | |
2457 | ||
4e69b598 JB |
2458 | cluster->window_start = first->offset; |
2459 | ||
2460 | node = &first->offset_index; | |
2461 | ||
2462 | /* | |
2463 | * now we've found our entries, pull them out of the free space | |
2464 | * cache and put them into the cluster rbtree | |
2465 | */ | |
2466 | do { | |
2467 | int ret; | |
2468 | ||
2469 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
2470 | node = rb_next(&entry->offset_index); | |
1bb91902 | 2471 | if (entry->bitmap || entry->bytes < min_bytes) |
4e69b598 JB |
2472 | continue; |
2473 | ||
34d52cb6 | 2474 | rb_erase(&entry->offset_index, &ctl->free_space_offset); |
4e69b598 JB |
2475 | ret = tree_insert_offset(&cluster->root, entry->offset, |
2476 | &entry->offset_index, 0); | |
3f7de037 | 2477 | total_size += entry->bytes; |
79787eaa | 2478 | BUG_ON(ret); /* -EEXIST; Logic error */ |
4e69b598 JB |
2479 | } while (node && entry != last); |
2480 | ||
2481 | cluster->max_size = max_extent; | |
3f7de037 | 2482 | trace_btrfs_setup_cluster(block_group, cluster, total_size, 0); |
4e69b598 JB |
2483 | return 0; |
2484 | } | |
2485 | ||
2486 | /* | |
2487 | * This specifically looks for bitmaps that may work in the cluster, we assume | |
2488 | * that we have already failed to find extents that will work. | |
2489 | */ | |
3de85bb9 JB |
2490 | static noinline int |
2491 | setup_cluster_bitmap(struct btrfs_block_group_cache *block_group, | |
2492 | struct btrfs_free_cluster *cluster, | |
2493 | struct list_head *bitmaps, u64 offset, u64 bytes, | |
1bb91902 | 2494 | u64 cont1_bytes, u64 min_bytes) |
4e69b598 | 2495 | { |
34d52cb6 | 2496 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
4e69b598 | 2497 | struct btrfs_free_space *entry; |
4e69b598 | 2498 | int ret = -ENOSPC; |
0f0fbf1d | 2499 | u64 bitmap_offset = offset_to_bitmap(ctl, offset); |
4e69b598 | 2500 | |
34d52cb6 | 2501 | if (ctl->total_bitmaps == 0) |
4e69b598 JB |
2502 | return -ENOSPC; |
2503 | ||
0f0fbf1d LZ |
2504 | /* |
2505 | * The bitmap that covers offset won't be in the list unless offset | |
2506 | * is just its start offset. | |
2507 | */ | |
2508 | entry = list_first_entry(bitmaps, struct btrfs_free_space, list); | |
2509 | if (entry->offset != bitmap_offset) { | |
2510 | entry = tree_search_offset(ctl, bitmap_offset, 1, 0); | |
2511 | if (entry && list_empty(&entry->list)) | |
2512 | list_add(&entry->list, bitmaps); | |
2513 | } | |
2514 | ||
86d4a77b | 2515 | list_for_each_entry(entry, bitmaps, list) { |
357b9784 | 2516 | if (entry->bytes < bytes) |
86d4a77b JB |
2517 | continue; |
2518 | ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset, | |
1bb91902 | 2519 | bytes, cont1_bytes, min_bytes); |
86d4a77b JB |
2520 | if (!ret) |
2521 | return 0; | |
2522 | } | |
2523 | ||
2524 | /* | |
52621cb6 LZ |
2525 | * The bitmaps list has all the bitmaps that record free space |
2526 | * starting after offset, so no more search is required. | |
86d4a77b | 2527 | */ |
52621cb6 | 2528 | return -ENOSPC; |
4e69b598 JB |
2529 | } |
2530 | ||
fa9c0d79 CM |
2531 | /* |
2532 | * here we try to find a cluster of blocks in a block group. The goal | |
1bb91902 | 2533 | * is to find at least bytes+empty_size. |
fa9c0d79 CM |
2534 | * We might not find them all in one contiguous area. |
2535 | * | |
2536 | * returns zero and sets up cluster if things worked out, otherwise | |
2537 | * it returns -enospc | |
2538 | */ | |
2539 | int btrfs_find_space_cluster(struct btrfs_trans_handle *trans, | |
451d7585 | 2540 | struct btrfs_root *root, |
fa9c0d79 CM |
2541 | struct btrfs_block_group_cache *block_group, |
2542 | struct btrfs_free_cluster *cluster, | |
2543 | u64 offset, u64 bytes, u64 empty_size) | |
2544 | { | |
34d52cb6 | 2545 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
86d4a77b | 2546 | struct btrfs_free_space *entry, *tmp; |
52621cb6 | 2547 | LIST_HEAD(bitmaps); |
fa9c0d79 | 2548 | u64 min_bytes; |
1bb91902 | 2549 | u64 cont1_bytes; |
fa9c0d79 CM |
2550 | int ret; |
2551 | ||
1bb91902 AO |
2552 | /* |
2553 | * Choose the minimum extent size we'll require for this | |
2554 | * cluster. For SSD_SPREAD, don't allow any fragmentation. | |
2555 | * For metadata, allow allocates with smaller extents. For | |
2556 | * data, keep it dense. | |
2557 | */ | |
451d7585 | 2558 | if (btrfs_test_opt(root, SSD_SPREAD)) { |
1bb91902 | 2559 | cont1_bytes = min_bytes = bytes + empty_size; |
451d7585 | 2560 | } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) { |
1bb91902 AO |
2561 | cont1_bytes = bytes; |
2562 | min_bytes = block_group->sectorsize; | |
2563 | } else { | |
2564 | cont1_bytes = max(bytes, (bytes + empty_size) >> 2); | |
2565 | min_bytes = block_group->sectorsize; | |
2566 | } | |
fa9c0d79 | 2567 | |
34d52cb6 | 2568 | spin_lock(&ctl->tree_lock); |
7d0d2e8e JB |
2569 | |
2570 | /* | |
2571 | * If we know we don't have enough space to make a cluster don't even | |
2572 | * bother doing all the work to try and find one. | |
2573 | */ | |
1bb91902 | 2574 | if (ctl->free_space < bytes) { |
34d52cb6 | 2575 | spin_unlock(&ctl->tree_lock); |
7d0d2e8e JB |
2576 | return -ENOSPC; |
2577 | } | |
2578 | ||
fa9c0d79 CM |
2579 | spin_lock(&cluster->lock); |
2580 | ||
2581 | /* someone already found a cluster, hooray */ | |
2582 | if (cluster->block_group) { | |
2583 | ret = 0; | |
2584 | goto out; | |
2585 | } | |
fa9c0d79 | 2586 | |
3f7de037 JB |
2587 | trace_btrfs_find_cluster(block_group, offset, bytes, empty_size, |
2588 | min_bytes); | |
2589 | ||
2590 | INIT_LIST_HEAD(&bitmaps); | |
86d4a77b | 2591 | ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset, |
1bb91902 AO |
2592 | bytes + empty_size, |
2593 | cont1_bytes, min_bytes); | |
4e69b598 | 2594 | if (ret) |
86d4a77b | 2595 | ret = setup_cluster_bitmap(block_group, cluster, &bitmaps, |
1bb91902 AO |
2596 | offset, bytes + empty_size, |
2597 | cont1_bytes, min_bytes); | |
86d4a77b JB |
2598 | |
2599 | /* Clear our temporary list */ | |
2600 | list_for_each_entry_safe(entry, tmp, &bitmaps, list) | |
2601 | list_del_init(&entry->list); | |
fa9c0d79 | 2602 | |
4e69b598 JB |
2603 | if (!ret) { |
2604 | atomic_inc(&block_group->count); | |
2605 | list_add_tail(&cluster->block_group_list, | |
2606 | &block_group->cluster_list); | |
2607 | cluster->block_group = block_group; | |
3f7de037 JB |
2608 | } else { |
2609 | trace_btrfs_failed_cluster_setup(block_group); | |
fa9c0d79 | 2610 | } |
fa9c0d79 CM |
2611 | out: |
2612 | spin_unlock(&cluster->lock); | |
34d52cb6 | 2613 | spin_unlock(&ctl->tree_lock); |
fa9c0d79 CM |
2614 | |
2615 | return ret; | |
2616 | } | |
2617 | ||
2618 | /* | |
2619 | * simple code to zero out a cluster | |
2620 | */ | |
2621 | void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster) | |
2622 | { | |
2623 | spin_lock_init(&cluster->lock); | |
2624 | spin_lock_init(&cluster->refill_lock); | |
6bef4d31 | 2625 | cluster->root = RB_ROOT; |
fa9c0d79 CM |
2626 | cluster->max_size = 0; |
2627 | INIT_LIST_HEAD(&cluster->block_group_list); | |
2628 | cluster->block_group = NULL; | |
2629 | } | |
2630 | ||
7fe1e641 LZ |
2631 | static int do_trimming(struct btrfs_block_group_cache *block_group, |
2632 | u64 *total_trimmed, u64 start, u64 bytes, | |
2633 | u64 reserved_start, u64 reserved_bytes) | |
f7039b1d | 2634 | { |
7fe1e641 | 2635 | struct btrfs_space_info *space_info = block_group->space_info; |
f7039b1d | 2636 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
7fe1e641 LZ |
2637 | int ret; |
2638 | int update = 0; | |
2639 | u64 trimmed = 0; | |
f7039b1d | 2640 | |
7fe1e641 LZ |
2641 | spin_lock(&space_info->lock); |
2642 | spin_lock(&block_group->lock); | |
2643 | if (!block_group->ro) { | |
2644 | block_group->reserved += reserved_bytes; | |
2645 | space_info->bytes_reserved += reserved_bytes; | |
2646 | update = 1; | |
2647 | } | |
2648 | spin_unlock(&block_group->lock); | |
2649 | spin_unlock(&space_info->lock); | |
2650 | ||
2651 | ret = btrfs_error_discard_extent(fs_info->extent_root, | |
2652 | start, bytes, &trimmed); | |
2653 | if (!ret) | |
2654 | *total_trimmed += trimmed; | |
2655 | ||
2656 | btrfs_add_free_space(block_group, reserved_start, reserved_bytes); | |
2657 | ||
2658 | if (update) { | |
2659 | spin_lock(&space_info->lock); | |
2660 | spin_lock(&block_group->lock); | |
2661 | if (block_group->ro) | |
2662 | space_info->bytes_readonly += reserved_bytes; | |
2663 | block_group->reserved -= reserved_bytes; | |
2664 | space_info->bytes_reserved -= reserved_bytes; | |
2665 | spin_unlock(&space_info->lock); | |
2666 | spin_unlock(&block_group->lock); | |
2667 | } | |
2668 | ||
2669 | return ret; | |
2670 | } | |
2671 | ||
2672 | static int trim_no_bitmap(struct btrfs_block_group_cache *block_group, | |
2673 | u64 *total_trimmed, u64 start, u64 end, u64 minlen) | |
2674 | { | |
2675 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
2676 | struct btrfs_free_space *entry; | |
2677 | struct rb_node *node; | |
2678 | int ret = 0; | |
2679 | u64 extent_start; | |
2680 | u64 extent_bytes; | |
2681 | u64 bytes; | |
f7039b1d LD |
2682 | |
2683 | while (start < end) { | |
34d52cb6 | 2684 | spin_lock(&ctl->tree_lock); |
f7039b1d | 2685 | |
34d52cb6 LZ |
2686 | if (ctl->free_space < minlen) { |
2687 | spin_unlock(&ctl->tree_lock); | |
f7039b1d LD |
2688 | break; |
2689 | } | |
2690 | ||
34d52cb6 | 2691 | entry = tree_search_offset(ctl, start, 0, 1); |
7fe1e641 | 2692 | if (!entry) { |
34d52cb6 | 2693 | spin_unlock(&ctl->tree_lock); |
f7039b1d LD |
2694 | break; |
2695 | } | |
2696 | ||
7fe1e641 LZ |
2697 | /* skip bitmaps */ |
2698 | while (entry->bitmap) { | |
2699 | node = rb_next(&entry->offset_index); | |
2700 | if (!node) { | |
34d52cb6 | 2701 | spin_unlock(&ctl->tree_lock); |
7fe1e641 | 2702 | goto out; |
f7039b1d | 2703 | } |
7fe1e641 LZ |
2704 | entry = rb_entry(node, struct btrfs_free_space, |
2705 | offset_index); | |
f7039b1d LD |
2706 | } |
2707 | ||
7fe1e641 LZ |
2708 | if (entry->offset >= end) { |
2709 | spin_unlock(&ctl->tree_lock); | |
2710 | break; | |
f7039b1d LD |
2711 | } |
2712 | ||
7fe1e641 LZ |
2713 | extent_start = entry->offset; |
2714 | extent_bytes = entry->bytes; | |
2715 | start = max(start, extent_start); | |
2716 | bytes = min(extent_start + extent_bytes, end) - start; | |
2717 | if (bytes < minlen) { | |
2718 | spin_unlock(&ctl->tree_lock); | |
2719 | goto next; | |
f7039b1d LD |
2720 | } |
2721 | ||
7fe1e641 LZ |
2722 | unlink_free_space(ctl, entry); |
2723 | kmem_cache_free(btrfs_free_space_cachep, entry); | |
2724 | ||
34d52cb6 | 2725 | spin_unlock(&ctl->tree_lock); |
f7039b1d | 2726 | |
7fe1e641 LZ |
2727 | ret = do_trimming(block_group, total_trimmed, start, bytes, |
2728 | extent_start, extent_bytes); | |
2729 | if (ret) | |
2730 | break; | |
2731 | next: | |
2732 | start += bytes; | |
f7039b1d | 2733 | |
7fe1e641 LZ |
2734 | if (fatal_signal_pending(current)) { |
2735 | ret = -ERESTARTSYS; | |
2736 | break; | |
2737 | } | |
2738 | ||
2739 | cond_resched(); | |
2740 | } | |
2741 | out: | |
2742 | return ret; | |
2743 | } | |
2744 | ||
2745 | static int trim_bitmaps(struct btrfs_block_group_cache *block_group, | |
2746 | u64 *total_trimmed, u64 start, u64 end, u64 minlen) | |
2747 | { | |
2748 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
2749 | struct btrfs_free_space *entry; | |
2750 | int ret = 0; | |
2751 | int ret2; | |
2752 | u64 bytes; | |
2753 | u64 offset = offset_to_bitmap(ctl, start); | |
2754 | ||
2755 | while (offset < end) { | |
2756 | bool next_bitmap = false; | |
2757 | ||
2758 | spin_lock(&ctl->tree_lock); | |
2759 | ||
2760 | if (ctl->free_space < minlen) { | |
2761 | spin_unlock(&ctl->tree_lock); | |
2762 | break; | |
2763 | } | |
2764 | ||
2765 | entry = tree_search_offset(ctl, offset, 1, 0); | |
2766 | if (!entry) { | |
2767 | spin_unlock(&ctl->tree_lock); | |
2768 | next_bitmap = true; | |
2769 | goto next; | |
2770 | } | |
2771 | ||
2772 | bytes = minlen; | |
2773 | ret2 = search_bitmap(ctl, entry, &start, &bytes); | |
2774 | if (ret2 || start >= end) { | |
2775 | spin_unlock(&ctl->tree_lock); | |
2776 | next_bitmap = true; | |
2777 | goto next; | |
2778 | } | |
2779 | ||
2780 | bytes = min(bytes, end - start); | |
2781 | if (bytes < minlen) { | |
2782 | spin_unlock(&ctl->tree_lock); | |
2783 | goto next; | |
2784 | } | |
2785 | ||
2786 | bitmap_clear_bits(ctl, entry, start, bytes); | |
2787 | if (entry->bytes == 0) | |
2788 | free_bitmap(ctl, entry); | |
2789 | ||
2790 | spin_unlock(&ctl->tree_lock); | |
2791 | ||
2792 | ret = do_trimming(block_group, total_trimmed, start, bytes, | |
2793 | start, bytes); | |
2794 | if (ret) | |
2795 | break; | |
2796 | next: | |
2797 | if (next_bitmap) { | |
2798 | offset += BITS_PER_BITMAP * ctl->unit; | |
2799 | } else { | |
2800 | start += bytes; | |
2801 | if (start >= offset + BITS_PER_BITMAP * ctl->unit) | |
2802 | offset += BITS_PER_BITMAP * ctl->unit; | |
f7039b1d | 2803 | } |
f7039b1d LD |
2804 | |
2805 | if (fatal_signal_pending(current)) { | |
2806 | ret = -ERESTARTSYS; | |
2807 | break; | |
2808 | } | |
2809 | ||
2810 | cond_resched(); | |
2811 | } | |
2812 | ||
2813 | return ret; | |
2814 | } | |
581bb050 | 2815 | |
7fe1e641 LZ |
2816 | int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group, |
2817 | u64 *trimmed, u64 start, u64 end, u64 minlen) | |
2818 | { | |
2819 | int ret; | |
2820 | ||
2821 | *trimmed = 0; | |
2822 | ||
2823 | ret = trim_no_bitmap(block_group, trimmed, start, end, minlen); | |
2824 | if (ret) | |
2825 | return ret; | |
2826 | ||
2827 | ret = trim_bitmaps(block_group, trimmed, start, end, minlen); | |
2828 | ||
2829 | return ret; | |
2830 | } | |
2831 | ||
581bb050 LZ |
2832 | /* |
2833 | * Find the left-most item in the cache tree, and then return the | |
2834 | * smallest inode number in the item. | |
2835 | * | |
2836 | * Note: the returned inode number may not be the smallest one in | |
2837 | * the tree, if the left-most item is a bitmap. | |
2838 | */ | |
2839 | u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root) | |
2840 | { | |
2841 | struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl; | |
2842 | struct btrfs_free_space *entry = NULL; | |
2843 | u64 ino = 0; | |
2844 | ||
2845 | spin_lock(&ctl->tree_lock); | |
2846 | ||
2847 | if (RB_EMPTY_ROOT(&ctl->free_space_offset)) | |
2848 | goto out; | |
2849 | ||
2850 | entry = rb_entry(rb_first(&ctl->free_space_offset), | |
2851 | struct btrfs_free_space, offset_index); | |
2852 | ||
2853 | if (!entry->bitmap) { | |
2854 | ino = entry->offset; | |
2855 | ||
2856 | unlink_free_space(ctl, entry); | |
2857 | entry->offset++; | |
2858 | entry->bytes--; | |
2859 | if (!entry->bytes) | |
2860 | kmem_cache_free(btrfs_free_space_cachep, entry); | |
2861 | else | |
2862 | link_free_space(ctl, entry); | |
2863 | } else { | |
2864 | u64 offset = 0; | |
2865 | u64 count = 1; | |
2866 | int ret; | |
2867 | ||
2868 | ret = search_bitmap(ctl, entry, &offset, &count); | |
79787eaa | 2869 | /* Logic error; Should be empty if it can't find anything */ |
581bb050 LZ |
2870 | BUG_ON(ret); |
2871 | ||
2872 | ino = offset; | |
2873 | bitmap_clear_bits(ctl, entry, offset, 1); | |
2874 | if (entry->bytes == 0) | |
2875 | free_bitmap(ctl, entry); | |
2876 | } | |
2877 | out: | |
2878 | spin_unlock(&ctl->tree_lock); | |
2879 | ||
2880 | return ino; | |
2881 | } | |
82d5902d LZ |
2882 | |
2883 | struct inode *lookup_free_ino_inode(struct btrfs_root *root, | |
2884 | struct btrfs_path *path) | |
2885 | { | |
2886 | struct inode *inode = NULL; | |
2887 | ||
2888 | spin_lock(&root->cache_lock); | |
2889 | if (root->cache_inode) | |
2890 | inode = igrab(root->cache_inode); | |
2891 | spin_unlock(&root->cache_lock); | |
2892 | if (inode) | |
2893 | return inode; | |
2894 | ||
2895 | inode = __lookup_free_space_inode(root, path, 0); | |
2896 | if (IS_ERR(inode)) | |
2897 | return inode; | |
2898 | ||
2899 | spin_lock(&root->cache_lock); | |
7841cb28 | 2900 | if (!btrfs_fs_closing(root->fs_info)) |
82d5902d LZ |
2901 | root->cache_inode = igrab(inode); |
2902 | spin_unlock(&root->cache_lock); | |
2903 | ||
2904 | return inode; | |
2905 | } | |
2906 | ||
2907 | int create_free_ino_inode(struct btrfs_root *root, | |
2908 | struct btrfs_trans_handle *trans, | |
2909 | struct btrfs_path *path) | |
2910 | { | |
2911 | return __create_free_space_inode(root, trans, path, | |
2912 | BTRFS_FREE_INO_OBJECTID, 0); | |
2913 | } | |
2914 | ||
2915 | int load_free_ino_cache(struct btrfs_fs_info *fs_info, struct btrfs_root *root) | |
2916 | { | |
2917 | struct btrfs_free_space_ctl *ctl = root->free_ino_ctl; | |
2918 | struct btrfs_path *path; | |
2919 | struct inode *inode; | |
2920 | int ret = 0; | |
2921 | u64 root_gen = btrfs_root_generation(&root->root_item); | |
2922 | ||
4b9465cb CM |
2923 | if (!btrfs_test_opt(root, INODE_MAP_CACHE)) |
2924 | return 0; | |
2925 | ||
82d5902d LZ |
2926 | /* |
2927 | * If we're unmounting then just return, since this does a search on the | |
2928 | * normal root and not the commit root and we could deadlock. | |
2929 | */ | |
7841cb28 | 2930 | if (btrfs_fs_closing(fs_info)) |
82d5902d LZ |
2931 | return 0; |
2932 | ||
2933 | path = btrfs_alloc_path(); | |
2934 | if (!path) | |
2935 | return 0; | |
2936 | ||
2937 | inode = lookup_free_ino_inode(root, path); | |
2938 | if (IS_ERR(inode)) | |
2939 | goto out; | |
2940 | ||
2941 | if (root_gen != BTRFS_I(inode)->generation) | |
2942 | goto out_put; | |
2943 | ||
2944 | ret = __load_free_space_cache(root, inode, ctl, path, 0); | |
2945 | ||
2946 | if (ret < 0) | |
2947 | printk(KERN_ERR "btrfs: failed to load free ino cache for " | |
2948 | "root %llu\n", root->root_key.objectid); | |
2949 | out_put: | |
2950 | iput(inode); | |
2951 | out: | |
2952 | btrfs_free_path(path); | |
2953 | return ret; | |
2954 | } | |
2955 | ||
2956 | int btrfs_write_out_ino_cache(struct btrfs_root *root, | |
2957 | struct btrfs_trans_handle *trans, | |
2958 | struct btrfs_path *path) | |
2959 | { | |
2960 | struct btrfs_free_space_ctl *ctl = root->free_ino_ctl; | |
2961 | struct inode *inode; | |
2962 | int ret; | |
2963 | ||
4b9465cb CM |
2964 | if (!btrfs_test_opt(root, INODE_MAP_CACHE)) |
2965 | return 0; | |
2966 | ||
82d5902d LZ |
2967 | inode = lookup_free_ino_inode(root, path); |
2968 | if (IS_ERR(inode)) | |
2969 | return 0; | |
2970 | ||
2971 | ret = __btrfs_write_out_cache(root, inode, ctl, NULL, trans, path, 0); | |
c09544e0 JB |
2972 | if (ret) { |
2973 | btrfs_delalloc_release_metadata(inode, inode->i_size); | |
2974 | #ifdef DEBUG | |
82d5902d LZ |
2975 | printk(KERN_ERR "btrfs: failed to write free ino cache " |
2976 | "for root %llu\n", root->root_key.objectid); | |
c09544e0 JB |
2977 | #endif |
2978 | } | |
82d5902d LZ |
2979 | |
2980 | iput(inode); | |
2981 | return ret; | |
2982 | } |