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