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" |
04216820 | 30 | #include "volumes.h" |
fa9c0d79 | 31 | |
96303081 JB |
32 | #define BITS_PER_BITMAP (PAGE_CACHE_SIZE * 8) |
33 | #define MAX_CACHE_BYTES_PER_GIG (32 * 1024) | |
0f9dd46c | 34 | |
34d52cb6 | 35 | static int link_free_space(struct btrfs_free_space_ctl *ctl, |
0cb59c99 | 36 | struct btrfs_free_space *info); |
cd023e7b JB |
37 | static void unlink_free_space(struct btrfs_free_space_ctl *ctl, |
38 | struct btrfs_free_space *info); | |
0cb59c99 | 39 | |
0414efae LZ |
40 | static struct inode *__lookup_free_space_inode(struct btrfs_root *root, |
41 | struct btrfs_path *path, | |
42 | u64 offset) | |
0af3d00b JB |
43 | { |
44 | struct btrfs_key key; | |
45 | struct btrfs_key location; | |
46 | struct btrfs_disk_key disk_key; | |
47 | struct btrfs_free_space_header *header; | |
48 | struct extent_buffer *leaf; | |
49 | struct inode *inode = NULL; | |
50 | int ret; | |
51 | ||
0af3d00b | 52 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; |
0414efae | 53 | key.offset = offset; |
0af3d00b JB |
54 | key.type = 0; |
55 | ||
56 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
57 | if (ret < 0) | |
58 | return ERR_PTR(ret); | |
59 | if (ret > 0) { | |
b3b4aa74 | 60 | btrfs_release_path(path); |
0af3d00b JB |
61 | return ERR_PTR(-ENOENT); |
62 | } | |
63 | ||
64 | leaf = path->nodes[0]; | |
65 | header = btrfs_item_ptr(leaf, path->slots[0], | |
66 | struct btrfs_free_space_header); | |
67 | btrfs_free_space_key(leaf, header, &disk_key); | |
68 | btrfs_disk_key_to_cpu(&location, &disk_key); | |
b3b4aa74 | 69 | btrfs_release_path(path); |
0af3d00b JB |
70 | |
71 | inode = btrfs_iget(root->fs_info->sb, &location, root, NULL); | |
72 | if (!inode) | |
73 | return ERR_PTR(-ENOENT); | |
74 | if (IS_ERR(inode)) | |
75 | return inode; | |
76 | if (is_bad_inode(inode)) { | |
77 | iput(inode); | |
78 | return ERR_PTR(-ENOENT); | |
79 | } | |
80 | ||
528c0327 AV |
81 | mapping_set_gfp_mask(inode->i_mapping, |
82 | mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS); | |
adae52b9 | 83 | |
0414efae LZ |
84 | return inode; |
85 | } | |
86 | ||
87 | struct inode *lookup_free_space_inode(struct btrfs_root *root, | |
88 | struct btrfs_block_group_cache | |
89 | *block_group, struct btrfs_path *path) | |
90 | { | |
91 | struct inode *inode = NULL; | |
5b0e95bf | 92 | u32 flags = BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW; |
0414efae LZ |
93 | |
94 | spin_lock(&block_group->lock); | |
95 | if (block_group->inode) | |
96 | inode = igrab(block_group->inode); | |
97 | spin_unlock(&block_group->lock); | |
98 | if (inode) | |
99 | return inode; | |
100 | ||
101 | inode = __lookup_free_space_inode(root, path, | |
102 | block_group->key.objectid); | |
103 | if (IS_ERR(inode)) | |
104 | return inode; | |
105 | ||
0af3d00b | 106 | spin_lock(&block_group->lock); |
5b0e95bf | 107 | if (!((BTRFS_I(inode)->flags & flags) == flags)) { |
c2cf52eb SK |
108 | btrfs_info(root->fs_info, |
109 | "Old style space inode found, converting."); | |
5b0e95bf JB |
110 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM | |
111 | BTRFS_INODE_NODATACOW; | |
2f356126 JB |
112 | block_group->disk_cache_state = BTRFS_DC_CLEAR; |
113 | } | |
114 | ||
300e4f8a | 115 | if (!block_group->iref) { |
0af3d00b JB |
116 | block_group->inode = igrab(inode); |
117 | block_group->iref = 1; | |
118 | } | |
119 | spin_unlock(&block_group->lock); | |
120 | ||
121 | return inode; | |
122 | } | |
123 | ||
48a3b636 ES |
124 | static int __create_free_space_inode(struct btrfs_root *root, |
125 | struct btrfs_trans_handle *trans, | |
126 | struct btrfs_path *path, | |
127 | u64 ino, u64 offset) | |
0af3d00b JB |
128 | { |
129 | struct btrfs_key key; | |
130 | struct btrfs_disk_key disk_key; | |
131 | struct btrfs_free_space_header *header; | |
132 | struct btrfs_inode_item *inode_item; | |
133 | struct extent_buffer *leaf; | |
5b0e95bf | 134 | u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC; |
0af3d00b JB |
135 | int ret; |
136 | ||
0414efae | 137 | ret = btrfs_insert_empty_inode(trans, root, path, ino); |
0af3d00b JB |
138 | if (ret) |
139 | return ret; | |
140 | ||
5b0e95bf JB |
141 | /* We inline crc's for the free disk space cache */ |
142 | if (ino != BTRFS_FREE_INO_OBJECTID) | |
143 | flags |= BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW; | |
144 | ||
0af3d00b JB |
145 | leaf = path->nodes[0]; |
146 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
147 | struct btrfs_inode_item); | |
148 | btrfs_item_key(leaf, &disk_key, path->slots[0]); | |
149 | memset_extent_buffer(leaf, 0, (unsigned long)inode_item, | |
150 | sizeof(*inode_item)); | |
151 | btrfs_set_inode_generation(leaf, inode_item, trans->transid); | |
152 | btrfs_set_inode_size(leaf, inode_item, 0); | |
153 | btrfs_set_inode_nbytes(leaf, inode_item, 0); | |
154 | btrfs_set_inode_uid(leaf, inode_item, 0); | |
155 | btrfs_set_inode_gid(leaf, inode_item, 0); | |
156 | btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600); | |
5b0e95bf | 157 | btrfs_set_inode_flags(leaf, inode_item, flags); |
0af3d00b JB |
158 | btrfs_set_inode_nlink(leaf, inode_item, 1); |
159 | btrfs_set_inode_transid(leaf, inode_item, trans->transid); | |
0414efae | 160 | btrfs_set_inode_block_group(leaf, inode_item, offset); |
0af3d00b | 161 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 162 | btrfs_release_path(path); |
0af3d00b JB |
163 | |
164 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
0414efae | 165 | key.offset = offset; |
0af3d00b JB |
166 | key.type = 0; |
167 | ||
168 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
169 | sizeof(struct btrfs_free_space_header)); | |
170 | if (ret < 0) { | |
b3b4aa74 | 171 | btrfs_release_path(path); |
0af3d00b JB |
172 | return ret; |
173 | } | |
174 | leaf = path->nodes[0]; | |
175 | header = btrfs_item_ptr(leaf, path->slots[0], | |
176 | struct btrfs_free_space_header); | |
177 | memset_extent_buffer(leaf, 0, (unsigned long)header, sizeof(*header)); | |
178 | btrfs_set_free_space_key(leaf, header, &disk_key); | |
179 | btrfs_mark_buffer_dirty(leaf); | |
b3b4aa74 | 180 | btrfs_release_path(path); |
0af3d00b JB |
181 | |
182 | return 0; | |
183 | } | |
184 | ||
0414efae LZ |
185 | int create_free_space_inode(struct btrfs_root *root, |
186 | struct btrfs_trans_handle *trans, | |
187 | struct btrfs_block_group_cache *block_group, | |
188 | struct btrfs_path *path) | |
189 | { | |
190 | int ret; | |
191 | u64 ino; | |
192 | ||
193 | ret = btrfs_find_free_objectid(root, &ino); | |
194 | if (ret < 0) | |
195 | return ret; | |
196 | ||
197 | return __create_free_space_inode(root, trans, path, ino, | |
198 | block_group->key.objectid); | |
199 | } | |
200 | ||
7b61cd92 MX |
201 | int btrfs_check_trunc_cache_free_space(struct btrfs_root *root, |
202 | struct btrfs_block_rsv *rsv) | |
0af3d00b | 203 | { |
c8174313 | 204 | u64 needed_bytes; |
7b61cd92 | 205 | int ret; |
c8174313 JB |
206 | |
207 | /* 1 for slack space, 1 for updating the inode */ | |
208 | needed_bytes = btrfs_calc_trunc_metadata_size(root, 1) + | |
209 | btrfs_calc_trans_metadata_size(root, 1); | |
210 | ||
7b61cd92 MX |
211 | spin_lock(&rsv->lock); |
212 | if (rsv->reserved < needed_bytes) | |
213 | ret = -ENOSPC; | |
214 | else | |
215 | ret = 0; | |
216 | spin_unlock(&rsv->lock); | |
4b286cd1 | 217 | return ret; |
7b61cd92 MX |
218 | } |
219 | ||
220 | int btrfs_truncate_free_space_cache(struct btrfs_root *root, | |
221 | struct btrfs_trans_handle *trans, | |
7b61cd92 MX |
222 | struct inode *inode) |
223 | { | |
7b61cd92 | 224 | int ret = 0; |
0af3d00b | 225 | |
0af3d00b | 226 | btrfs_i_size_write(inode, 0); |
7caef267 | 227 | truncate_pagecache(inode, 0); |
0af3d00b JB |
228 | |
229 | /* | |
230 | * We don't need an orphan item because truncating the free space cache | |
231 | * will never be split across transactions. | |
232 | */ | |
233 | ret = btrfs_truncate_inode_items(trans, root, inode, | |
234 | 0, BTRFS_EXTENT_DATA_KEY); | |
235 | if (ret) { | |
79787eaa | 236 | btrfs_abort_transaction(trans, root, ret); |
0af3d00b JB |
237 | return ret; |
238 | } | |
239 | ||
82d5902d | 240 | ret = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
241 | if (ret) |
242 | btrfs_abort_transaction(trans, root, ret); | |
c8174313 | 243 | |
82d5902d | 244 | return ret; |
0af3d00b JB |
245 | } |
246 | ||
9d66e233 JB |
247 | static int readahead_cache(struct inode *inode) |
248 | { | |
249 | struct file_ra_state *ra; | |
250 | unsigned long last_index; | |
251 | ||
252 | ra = kzalloc(sizeof(*ra), GFP_NOFS); | |
253 | if (!ra) | |
254 | return -ENOMEM; | |
255 | ||
256 | file_ra_state_init(ra, inode->i_mapping); | |
257 | last_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT; | |
258 | ||
259 | page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index); | |
260 | ||
261 | kfree(ra); | |
262 | ||
263 | return 0; | |
264 | } | |
265 | ||
a67509c3 JB |
266 | struct io_ctl { |
267 | void *cur, *orig; | |
268 | struct page *page; | |
269 | struct page **pages; | |
270 | struct btrfs_root *root; | |
271 | unsigned long size; | |
272 | int index; | |
273 | int num_pages; | |
5b0e95bf | 274 | unsigned check_crcs:1; |
a67509c3 JB |
275 | }; |
276 | ||
277 | static int io_ctl_init(struct io_ctl *io_ctl, struct inode *inode, | |
5349d6c3 | 278 | struct btrfs_root *root, int write) |
a67509c3 | 279 | { |
5349d6c3 MX |
280 | int num_pages; |
281 | int check_crcs = 0; | |
282 | ||
ed6078f7 | 283 | num_pages = DIV_ROUND_UP(i_size_read(inode), PAGE_CACHE_SIZE); |
5349d6c3 MX |
284 | |
285 | if (btrfs_ino(inode) != BTRFS_FREE_INO_OBJECTID) | |
286 | check_crcs = 1; | |
287 | ||
288 | /* Make sure we can fit our crcs into the first page */ | |
289 | if (write && check_crcs && | |
290 | (num_pages * sizeof(u32)) >= PAGE_CACHE_SIZE) | |
291 | return -ENOSPC; | |
292 | ||
a67509c3 | 293 | memset(io_ctl, 0, sizeof(struct io_ctl)); |
5349d6c3 MX |
294 | |
295 | io_ctl->pages = kzalloc(sizeof(struct page *) * num_pages, GFP_NOFS); | |
a67509c3 JB |
296 | if (!io_ctl->pages) |
297 | return -ENOMEM; | |
5349d6c3 MX |
298 | |
299 | io_ctl->num_pages = num_pages; | |
a67509c3 | 300 | io_ctl->root = root; |
5349d6c3 MX |
301 | io_ctl->check_crcs = check_crcs; |
302 | ||
a67509c3 JB |
303 | return 0; |
304 | } | |
305 | ||
306 | static void io_ctl_free(struct io_ctl *io_ctl) | |
307 | { | |
308 | kfree(io_ctl->pages); | |
309 | } | |
310 | ||
311 | static void io_ctl_unmap_page(struct io_ctl *io_ctl) | |
312 | { | |
313 | if (io_ctl->cur) { | |
314 | kunmap(io_ctl->page); | |
315 | io_ctl->cur = NULL; | |
316 | io_ctl->orig = NULL; | |
317 | } | |
318 | } | |
319 | ||
320 | static void io_ctl_map_page(struct io_ctl *io_ctl, int clear) | |
321 | { | |
b12d6869 | 322 | ASSERT(io_ctl->index < io_ctl->num_pages); |
a67509c3 JB |
323 | io_ctl->page = io_ctl->pages[io_ctl->index++]; |
324 | io_ctl->cur = kmap(io_ctl->page); | |
325 | io_ctl->orig = io_ctl->cur; | |
326 | io_ctl->size = PAGE_CACHE_SIZE; | |
327 | if (clear) | |
328 | memset(io_ctl->cur, 0, PAGE_CACHE_SIZE); | |
329 | } | |
330 | ||
331 | static void io_ctl_drop_pages(struct io_ctl *io_ctl) | |
332 | { | |
333 | int i; | |
334 | ||
335 | io_ctl_unmap_page(io_ctl); | |
336 | ||
337 | for (i = 0; i < io_ctl->num_pages; i++) { | |
a1ee5a45 LZ |
338 | if (io_ctl->pages[i]) { |
339 | ClearPageChecked(io_ctl->pages[i]); | |
340 | unlock_page(io_ctl->pages[i]); | |
341 | page_cache_release(io_ctl->pages[i]); | |
342 | } | |
a67509c3 JB |
343 | } |
344 | } | |
345 | ||
346 | static int io_ctl_prepare_pages(struct io_ctl *io_ctl, struct inode *inode, | |
347 | int uptodate) | |
348 | { | |
349 | struct page *page; | |
350 | gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); | |
351 | int i; | |
352 | ||
353 | for (i = 0; i < io_ctl->num_pages; i++) { | |
354 | page = find_or_create_page(inode->i_mapping, i, mask); | |
355 | if (!page) { | |
356 | io_ctl_drop_pages(io_ctl); | |
357 | return -ENOMEM; | |
358 | } | |
359 | io_ctl->pages[i] = page; | |
360 | if (uptodate && !PageUptodate(page)) { | |
361 | btrfs_readpage(NULL, page); | |
362 | lock_page(page); | |
363 | if (!PageUptodate(page)) { | |
efe120a0 FH |
364 | btrfs_err(BTRFS_I(inode)->root->fs_info, |
365 | "error reading free space cache"); | |
a67509c3 JB |
366 | io_ctl_drop_pages(io_ctl); |
367 | return -EIO; | |
368 | } | |
369 | } | |
370 | } | |
371 | ||
f7d61dcd JB |
372 | for (i = 0; i < io_ctl->num_pages; i++) { |
373 | clear_page_dirty_for_io(io_ctl->pages[i]); | |
374 | set_page_extent_mapped(io_ctl->pages[i]); | |
375 | } | |
376 | ||
a67509c3 JB |
377 | return 0; |
378 | } | |
379 | ||
380 | static void io_ctl_set_generation(struct io_ctl *io_ctl, u64 generation) | |
381 | { | |
528c0327 | 382 | __le64 *val; |
a67509c3 JB |
383 | |
384 | io_ctl_map_page(io_ctl, 1); | |
385 | ||
386 | /* | |
5b0e95bf JB |
387 | * Skip the csum areas. If we don't check crcs then we just have a |
388 | * 64bit chunk at the front of the first page. | |
a67509c3 | 389 | */ |
5b0e95bf JB |
390 | if (io_ctl->check_crcs) { |
391 | io_ctl->cur += (sizeof(u32) * io_ctl->num_pages); | |
392 | io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages); | |
393 | } else { | |
394 | io_ctl->cur += sizeof(u64); | |
395 | io_ctl->size -= sizeof(u64) * 2; | |
396 | } | |
a67509c3 JB |
397 | |
398 | val = io_ctl->cur; | |
399 | *val = cpu_to_le64(generation); | |
400 | io_ctl->cur += sizeof(u64); | |
a67509c3 JB |
401 | } |
402 | ||
403 | static int io_ctl_check_generation(struct io_ctl *io_ctl, u64 generation) | |
404 | { | |
528c0327 | 405 | __le64 *gen; |
a67509c3 | 406 | |
5b0e95bf JB |
407 | /* |
408 | * Skip the crc area. If we don't check crcs then we just have a 64bit | |
409 | * chunk at the front of the first page. | |
410 | */ | |
411 | if (io_ctl->check_crcs) { | |
412 | io_ctl->cur += sizeof(u32) * io_ctl->num_pages; | |
413 | io_ctl->size -= sizeof(u64) + | |
414 | (sizeof(u32) * io_ctl->num_pages); | |
415 | } else { | |
416 | io_ctl->cur += sizeof(u64); | |
417 | io_ctl->size -= sizeof(u64) * 2; | |
418 | } | |
a67509c3 | 419 | |
a67509c3 JB |
420 | gen = io_ctl->cur; |
421 | if (le64_to_cpu(*gen) != generation) { | |
efe120a0 | 422 | printk_ratelimited(KERN_ERR "BTRFS: space cache generation " |
a67509c3 JB |
423 | "(%Lu) does not match inode (%Lu)\n", *gen, |
424 | generation); | |
425 | io_ctl_unmap_page(io_ctl); | |
426 | return -EIO; | |
427 | } | |
428 | io_ctl->cur += sizeof(u64); | |
5b0e95bf JB |
429 | return 0; |
430 | } | |
431 | ||
432 | static void io_ctl_set_crc(struct io_ctl *io_ctl, int index) | |
433 | { | |
434 | u32 *tmp; | |
435 | u32 crc = ~(u32)0; | |
436 | unsigned offset = 0; | |
437 | ||
438 | if (!io_ctl->check_crcs) { | |
439 | io_ctl_unmap_page(io_ctl); | |
440 | return; | |
441 | } | |
442 | ||
443 | if (index == 0) | |
cb54f257 | 444 | offset = sizeof(u32) * io_ctl->num_pages; |
5b0e95bf | 445 | |
b0496686 | 446 | crc = btrfs_csum_data(io_ctl->orig + offset, crc, |
5b0e95bf JB |
447 | PAGE_CACHE_SIZE - offset); |
448 | btrfs_csum_final(crc, (char *)&crc); | |
449 | io_ctl_unmap_page(io_ctl); | |
450 | tmp = kmap(io_ctl->pages[0]); | |
451 | tmp += index; | |
452 | *tmp = crc; | |
453 | kunmap(io_ctl->pages[0]); | |
454 | } | |
455 | ||
456 | static int io_ctl_check_crc(struct io_ctl *io_ctl, int index) | |
457 | { | |
458 | u32 *tmp, val; | |
459 | u32 crc = ~(u32)0; | |
460 | unsigned offset = 0; | |
461 | ||
462 | if (!io_ctl->check_crcs) { | |
463 | io_ctl_map_page(io_ctl, 0); | |
464 | return 0; | |
465 | } | |
466 | ||
467 | if (index == 0) | |
468 | offset = sizeof(u32) * io_ctl->num_pages; | |
469 | ||
470 | tmp = kmap(io_ctl->pages[0]); | |
471 | tmp += index; | |
472 | val = *tmp; | |
473 | kunmap(io_ctl->pages[0]); | |
474 | ||
475 | io_ctl_map_page(io_ctl, 0); | |
b0496686 | 476 | crc = btrfs_csum_data(io_ctl->orig + offset, crc, |
5b0e95bf JB |
477 | PAGE_CACHE_SIZE - offset); |
478 | btrfs_csum_final(crc, (char *)&crc); | |
479 | if (val != crc) { | |
efe120a0 | 480 | printk_ratelimited(KERN_ERR "BTRFS: csum mismatch on free " |
5b0e95bf JB |
481 | "space cache\n"); |
482 | io_ctl_unmap_page(io_ctl); | |
483 | return -EIO; | |
484 | } | |
485 | ||
a67509c3 JB |
486 | return 0; |
487 | } | |
488 | ||
489 | static int io_ctl_add_entry(struct io_ctl *io_ctl, u64 offset, u64 bytes, | |
490 | void *bitmap) | |
491 | { | |
492 | struct btrfs_free_space_entry *entry; | |
493 | ||
494 | if (!io_ctl->cur) | |
495 | return -ENOSPC; | |
496 | ||
497 | entry = io_ctl->cur; | |
498 | entry->offset = cpu_to_le64(offset); | |
499 | entry->bytes = cpu_to_le64(bytes); | |
500 | entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP : | |
501 | BTRFS_FREE_SPACE_EXTENT; | |
502 | io_ctl->cur += sizeof(struct btrfs_free_space_entry); | |
503 | io_ctl->size -= sizeof(struct btrfs_free_space_entry); | |
504 | ||
505 | if (io_ctl->size >= sizeof(struct btrfs_free_space_entry)) | |
506 | return 0; | |
507 | ||
5b0e95bf | 508 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); |
a67509c3 JB |
509 | |
510 | /* No more pages to map */ | |
511 | if (io_ctl->index >= io_ctl->num_pages) | |
512 | return 0; | |
513 | ||
514 | /* map the next page */ | |
515 | io_ctl_map_page(io_ctl, 1); | |
516 | return 0; | |
517 | } | |
518 | ||
519 | static int io_ctl_add_bitmap(struct io_ctl *io_ctl, void *bitmap) | |
520 | { | |
521 | if (!io_ctl->cur) | |
522 | return -ENOSPC; | |
523 | ||
524 | /* | |
525 | * If we aren't at the start of the current page, unmap this one and | |
526 | * map the next one if there is any left. | |
527 | */ | |
528 | if (io_ctl->cur != io_ctl->orig) { | |
5b0e95bf | 529 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); |
a67509c3 JB |
530 | if (io_ctl->index >= io_ctl->num_pages) |
531 | return -ENOSPC; | |
532 | io_ctl_map_page(io_ctl, 0); | |
533 | } | |
534 | ||
535 | memcpy(io_ctl->cur, bitmap, PAGE_CACHE_SIZE); | |
5b0e95bf | 536 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); |
a67509c3 JB |
537 | if (io_ctl->index < io_ctl->num_pages) |
538 | io_ctl_map_page(io_ctl, 0); | |
539 | return 0; | |
540 | } | |
541 | ||
542 | static void io_ctl_zero_remaining_pages(struct io_ctl *io_ctl) | |
543 | { | |
5b0e95bf JB |
544 | /* |
545 | * If we're not on the boundary we know we've modified the page and we | |
546 | * need to crc the page. | |
547 | */ | |
548 | if (io_ctl->cur != io_ctl->orig) | |
549 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); | |
550 | else | |
551 | io_ctl_unmap_page(io_ctl); | |
a67509c3 JB |
552 | |
553 | while (io_ctl->index < io_ctl->num_pages) { | |
554 | io_ctl_map_page(io_ctl, 1); | |
5b0e95bf | 555 | io_ctl_set_crc(io_ctl, io_ctl->index - 1); |
a67509c3 JB |
556 | } |
557 | } | |
558 | ||
5b0e95bf JB |
559 | static int io_ctl_read_entry(struct io_ctl *io_ctl, |
560 | struct btrfs_free_space *entry, u8 *type) | |
a67509c3 JB |
561 | { |
562 | struct btrfs_free_space_entry *e; | |
2f120c05 JB |
563 | int ret; |
564 | ||
565 | if (!io_ctl->cur) { | |
566 | ret = io_ctl_check_crc(io_ctl, io_ctl->index); | |
567 | if (ret) | |
568 | return ret; | |
569 | } | |
a67509c3 JB |
570 | |
571 | e = io_ctl->cur; | |
572 | entry->offset = le64_to_cpu(e->offset); | |
573 | entry->bytes = le64_to_cpu(e->bytes); | |
5b0e95bf | 574 | *type = e->type; |
a67509c3 JB |
575 | io_ctl->cur += sizeof(struct btrfs_free_space_entry); |
576 | io_ctl->size -= sizeof(struct btrfs_free_space_entry); | |
577 | ||
578 | if (io_ctl->size >= sizeof(struct btrfs_free_space_entry)) | |
5b0e95bf | 579 | return 0; |
a67509c3 JB |
580 | |
581 | io_ctl_unmap_page(io_ctl); | |
582 | ||
2f120c05 | 583 | return 0; |
a67509c3 JB |
584 | } |
585 | ||
5b0e95bf JB |
586 | static int io_ctl_read_bitmap(struct io_ctl *io_ctl, |
587 | struct btrfs_free_space *entry) | |
a67509c3 | 588 | { |
5b0e95bf JB |
589 | int ret; |
590 | ||
5b0e95bf JB |
591 | ret = io_ctl_check_crc(io_ctl, io_ctl->index); |
592 | if (ret) | |
593 | return ret; | |
594 | ||
a67509c3 JB |
595 | memcpy(entry->bitmap, io_ctl->cur, PAGE_CACHE_SIZE); |
596 | io_ctl_unmap_page(io_ctl); | |
5b0e95bf JB |
597 | |
598 | return 0; | |
a67509c3 JB |
599 | } |
600 | ||
cd023e7b JB |
601 | /* |
602 | * Since we attach pinned extents after the fact we can have contiguous sections | |
603 | * of free space that are split up in entries. This poses a problem with the | |
604 | * tree logging stuff since it could have allocated across what appears to be 2 | |
605 | * entries since we would have merged the entries when adding the pinned extents | |
606 | * back to the free space cache. So run through the space cache that we just | |
607 | * loaded and merge contiguous entries. This will make the log replay stuff not | |
608 | * blow up and it will make for nicer allocator behavior. | |
609 | */ | |
610 | static void merge_space_tree(struct btrfs_free_space_ctl *ctl) | |
611 | { | |
612 | struct btrfs_free_space *e, *prev = NULL; | |
613 | struct rb_node *n; | |
614 | ||
615 | again: | |
616 | spin_lock(&ctl->tree_lock); | |
617 | for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) { | |
618 | e = rb_entry(n, struct btrfs_free_space, offset_index); | |
619 | if (!prev) | |
620 | goto next; | |
621 | if (e->bitmap || prev->bitmap) | |
622 | goto next; | |
623 | if (prev->offset + prev->bytes == e->offset) { | |
624 | unlink_free_space(ctl, prev); | |
625 | unlink_free_space(ctl, e); | |
626 | prev->bytes += e->bytes; | |
627 | kmem_cache_free(btrfs_free_space_cachep, e); | |
628 | link_free_space(ctl, prev); | |
629 | prev = NULL; | |
630 | spin_unlock(&ctl->tree_lock); | |
631 | goto again; | |
632 | } | |
633 | next: | |
634 | prev = e; | |
635 | } | |
636 | spin_unlock(&ctl->tree_lock); | |
637 | } | |
638 | ||
48a3b636 ES |
639 | static int __load_free_space_cache(struct btrfs_root *root, struct inode *inode, |
640 | struct btrfs_free_space_ctl *ctl, | |
641 | struct btrfs_path *path, u64 offset) | |
9d66e233 | 642 | { |
9d66e233 JB |
643 | struct btrfs_free_space_header *header; |
644 | struct extent_buffer *leaf; | |
a67509c3 | 645 | struct io_ctl io_ctl; |
9d66e233 | 646 | struct btrfs_key key; |
a67509c3 | 647 | struct btrfs_free_space *e, *n; |
9d66e233 JB |
648 | struct list_head bitmaps; |
649 | u64 num_entries; | |
650 | u64 num_bitmaps; | |
651 | u64 generation; | |
a67509c3 | 652 | u8 type; |
f6a39829 | 653 | int ret = 0; |
9d66e233 JB |
654 | |
655 | INIT_LIST_HEAD(&bitmaps); | |
656 | ||
9d66e233 | 657 | /* Nothing in the space cache, goodbye */ |
0414efae | 658 | if (!i_size_read(inode)) |
a67509c3 | 659 | return 0; |
9d66e233 JB |
660 | |
661 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
0414efae | 662 | key.offset = offset; |
9d66e233 JB |
663 | key.type = 0; |
664 | ||
665 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
0414efae | 666 | if (ret < 0) |
a67509c3 | 667 | return 0; |
0414efae | 668 | else if (ret > 0) { |
945d8962 | 669 | btrfs_release_path(path); |
a67509c3 | 670 | return 0; |
9d66e233 JB |
671 | } |
672 | ||
0414efae LZ |
673 | ret = -1; |
674 | ||
9d66e233 JB |
675 | leaf = path->nodes[0]; |
676 | header = btrfs_item_ptr(leaf, path->slots[0], | |
677 | struct btrfs_free_space_header); | |
678 | num_entries = btrfs_free_space_entries(leaf, header); | |
679 | num_bitmaps = btrfs_free_space_bitmaps(leaf, header); | |
680 | generation = btrfs_free_space_generation(leaf, header); | |
945d8962 | 681 | btrfs_release_path(path); |
9d66e233 | 682 | |
e570fd27 MX |
683 | if (!BTRFS_I(inode)->generation) { |
684 | btrfs_info(root->fs_info, | |
685 | "The free space cache file (%llu) is invalid. skip it\n", | |
686 | offset); | |
687 | return 0; | |
688 | } | |
689 | ||
9d66e233 | 690 | if (BTRFS_I(inode)->generation != generation) { |
c2cf52eb SK |
691 | btrfs_err(root->fs_info, |
692 | "free space inode generation (%llu) " | |
693 | "did not match free space cache generation (%llu)", | |
c1c9ff7c | 694 | BTRFS_I(inode)->generation, generation); |
a67509c3 | 695 | return 0; |
9d66e233 JB |
696 | } |
697 | ||
698 | if (!num_entries) | |
a67509c3 | 699 | return 0; |
9d66e233 | 700 | |
5349d6c3 | 701 | ret = io_ctl_init(&io_ctl, inode, root, 0); |
706efc66 LZ |
702 | if (ret) |
703 | return ret; | |
704 | ||
9d66e233 | 705 | ret = readahead_cache(inode); |
0414efae | 706 | if (ret) |
9d66e233 | 707 | goto out; |
9d66e233 | 708 | |
a67509c3 JB |
709 | ret = io_ctl_prepare_pages(&io_ctl, inode, 1); |
710 | if (ret) | |
711 | goto out; | |
9d66e233 | 712 | |
5b0e95bf JB |
713 | ret = io_ctl_check_crc(&io_ctl, 0); |
714 | if (ret) | |
715 | goto free_cache; | |
716 | ||
a67509c3 JB |
717 | ret = io_ctl_check_generation(&io_ctl, generation); |
718 | if (ret) | |
719 | goto free_cache; | |
9d66e233 | 720 | |
a67509c3 JB |
721 | while (num_entries) { |
722 | e = kmem_cache_zalloc(btrfs_free_space_cachep, | |
723 | GFP_NOFS); | |
724 | if (!e) | |
9d66e233 | 725 | goto free_cache; |
9d66e233 | 726 | |
5b0e95bf JB |
727 | ret = io_ctl_read_entry(&io_ctl, e, &type); |
728 | if (ret) { | |
729 | kmem_cache_free(btrfs_free_space_cachep, e); | |
730 | goto free_cache; | |
731 | } | |
732 | ||
a67509c3 JB |
733 | if (!e->bytes) { |
734 | kmem_cache_free(btrfs_free_space_cachep, e); | |
735 | goto free_cache; | |
9d66e233 | 736 | } |
a67509c3 JB |
737 | |
738 | if (type == BTRFS_FREE_SPACE_EXTENT) { | |
739 | spin_lock(&ctl->tree_lock); | |
740 | ret = link_free_space(ctl, e); | |
741 | spin_unlock(&ctl->tree_lock); | |
742 | if (ret) { | |
c2cf52eb SK |
743 | btrfs_err(root->fs_info, |
744 | "Duplicate entries in free space cache, dumping"); | |
a67509c3 | 745 | kmem_cache_free(btrfs_free_space_cachep, e); |
9d66e233 JB |
746 | goto free_cache; |
747 | } | |
a67509c3 | 748 | } else { |
b12d6869 | 749 | ASSERT(num_bitmaps); |
a67509c3 JB |
750 | num_bitmaps--; |
751 | e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS); | |
752 | if (!e->bitmap) { | |
753 | kmem_cache_free( | |
754 | btrfs_free_space_cachep, e); | |
9d66e233 JB |
755 | goto free_cache; |
756 | } | |
a67509c3 JB |
757 | spin_lock(&ctl->tree_lock); |
758 | ret = link_free_space(ctl, e); | |
759 | ctl->total_bitmaps++; | |
760 | ctl->op->recalc_thresholds(ctl); | |
761 | spin_unlock(&ctl->tree_lock); | |
762 | if (ret) { | |
c2cf52eb SK |
763 | btrfs_err(root->fs_info, |
764 | "Duplicate entries in free space cache, dumping"); | |
dc89e982 | 765 | kmem_cache_free(btrfs_free_space_cachep, e); |
9d66e233 JB |
766 | goto free_cache; |
767 | } | |
a67509c3 | 768 | list_add_tail(&e->list, &bitmaps); |
9d66e233 JB |
769 | } |
770 | ||
a67509c3 JB |
771 | num_entries--; |
772 | } | |
9d66e233 | 773 | |
2f120c05 JB |
774 | io_ctl_unmap_page(&io_ctl); |
775 | ||
a67509c3 JB |
776 | /* |
777 | * We add the bitmaps at the end of the entries in order that | |
778 | * the bitmap entries are added to the cache. | |
779 | */ | |
780 | list_for_each_entry_safe(e, n, &bitmaps, list) { | |
9d66e233 | 781 | list_del_init(&e->list); |
5b0e95bf JB |
782 | ret = io_ctl_read_bitmap(&io_ctl, e); |
783 | if (ret) | |
784 | goto free_cache; | |
9d66e233 JB |
785 | } |
786 | ||
a67509c3 | 787 | io_ctl_drop_pages(&io_ctl); |
cd023e7b | 788 | merge_space_tree(ctl); |
9d66e233 JB |
789 | ret = 1; |
790 | out: | |
a67509c3 | 791 | io_ctl_free(&io_ctl); |
9d66e233 | 792 | return ret; |
9d66e233 | 793 | free_cache: |
a67509c3 | 794 | io_ctl_drop_pages(&io_ctl); |
0414efae | 795 | __btrfs_remove_free_space_cache(ctl); |
9d66e233 JB |
796 | goto out; |
797 | } | |
798 | ||
0414efae LZ |
799 | int load_free_space_cache(struct btrfs_fs_info *fs_info, |
800 | struct btrfs_block_group_cache *block_group) | |
0cb59c99 | 801 | { |
34d52cb6 | 802 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0414efae LZ |
803 | struct btrfs_root *root = fs_info->tree_root; |
804 | struct inode *inode; | |
805 | struct btrfs_path *path; | |
5b0e95bf | 806 | int ret = 0; |
0414efae LZ |
807 | bool matched; |
808 | u64 used = btrfs_block_group_used(&block_group->item); | |
809 | ||
0414efae LZ |
810 | /* |
811 | * If this block group has been marked to be cleared for one reason or | |
812 | * another then we can't trust the on disk cache, so just return. | |
813 | */ | |
9d66e233 | 814 | spin_lock(&block_group->lock); |
0414efae LZ |
815 | if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) { |
816 | spin_unlock(&block_group->lock); | |
817 | return 0; | |
818 | } | |
9d66e233 | 819 | spin_unlock(&block_group->lock); |
0414efae LZ |
820 | |
821 | path = btrfs_alloc_path(); | |
822 | if (!path) | |
823 | return 0; | |
d53ba474 JB |
824 | path->search_commit_root = 1; |
825 | path->skip_locking = 1; | |
0414efae LZ |
826 | |
827 | inode = lookup_free_space_inode(root, block_group, path); | |
828 | if (IS_ERR(inode)) { | |
829 | btrfs_free_path(path); | |
830 | return 0; | |
831 | } | |
832 | ||
5b0e95bf JB |
833 | /* We may have converted the inode and made the cache invalid. */ |
834 | spin_lock(&block_group->lock); | |
835 | if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) { | |
836 | spin_unlock(&block_group->lock); | |
a7e221e9 | 837 | btrfs_free_path(path); |
5b0e95bf JB |
838 | goto out; |
839 | } | |
840 | spin_unlock(&block_group->lock); | |
841 | ||
0414efae LZ |
842 | ret = __load_free_space_cache(fs_info->tree_root, inode, ctl, |
843 | path, block_group->key.objectid); | |
844 | btrfs_free_path(path); | |
845 | if (ret <= 0) | |
846 | goto out; | |
847 | ||
848 | spin_lock(&ctl->tree_lock); | |
849 | matched = (ctl->free_space == (block_group->key.offset - used - | |
850 | block_group->bytes_super)); | |
851 | spin_unlock(&ctl->tree_lock); | |
852 | ||
853 | if (!matched) { | |
854 | __btrfs_remove_free_space_cache(ctl); | |
32d6b47f | 855 | btrfs_warn(fs_info, "block group %llu has wrong amount of free space", |
c2cf52eb | 856 | block_group->key.objectid); |
0414efae LZ |
857 | ret = -1; |
858 | } | |
859 | out: | |
860 | if (ret < 0) { | |
861 | /* This cache is bogus, make sure it gets cleared */ | |
862 | spin_lock(&block_group->lock); | |
863 | block_group->disk_cache_state = BTRFS_DC_CLEAR; | |
864 | spin_unlock(&block_group->lock); | |
82d5902d | 865 | ret = 0; |
0414efae | 866 | |
32d6b47f | 867 | btrfs_warn(fs_info, "failed to load free space cache for block group %llu, rebuild it now", |
c2cf52eb | 868 | block_group->key.objectid); |
0414efae LZ |
869 | } |
870 | ||
871 | iput(inode); | |
872 | return ret; | |
9d66e233 JB |
873 | } |
874 | ||
d4452bc5 CM |
875 | static noinline_for_stack |
876 | int write_cache_extent_entries(struct io_ctl *io_ctl, | |
877 | struct btrfs_free_space_ctl *ctl, | |
878 | struct btrfs_block_group_cache *block_group, | |
879 | int *entries, int *bitmaps, | |
880 | struct list_head *bitmap_list) | |
0cb59c99 | 881 | { |
c09544e0 | 882 | int ret; |
d4452bc5 CM |
883 | struct btrfs_free_cluster *cluster = NULL; |
884 | struct rb_node *node = rb_first(&ctl->free_space_offset); | |
be1a12a0 | 885 | |
43be2146 | 886 | /* Get the cluster for this block_group if it exists */ |
d4452bc5 | 887 | if (block_group && !list_empty(&block_group->cluster_list)) { |
43be2146 JB |
888 | cluster = list_entry(block_group->cluster_list.next, |
889 | struct btrfs_free_cluster, | |
890 | block_group_list); | |
d4452bc5 | 891 | } |
43be2146 | 892 | |
f75b130e JB |
893 | if (!node && cluster) { |
894 | node = rb_first(&cluster->root); | |
895 | cluster = NULL; | |
896 | } | |
897 | ||
a67509c3 JB |
898 | /* Write out the extent entries */ |
899 | while (node) { | |
900 | struct btrfs_free_space *e; | |
0cb59c99 | 901 | |
a67509c3 | 902 | e = rb_entry(node, struct btrfs_free_space, offset_index); |
d4452bc5 | 903 | *entries += 1; |
0cb59c99 | 904 | |
d4452bc5 | 905 | ret = io_ctl_add_entry(io_ctl, e->offset, e->bytes, |
a67509c3 JB |
906 | e->bitmap); |
907 | if (ret) | |
d4452bc5 | 908 | goto fail; |
2f356126 | 909 | |
a67509c3 | 910 | if (e->bitmap) { |
d4452bc5 CM |
911 | list_add_tail(&e->list, bitmap_list); |
912 | *bitmaps += 1; | |
2f356126 | 913 | } |
a67509c3 JB |
914 | node = rb_next(node); |
915 | if (!node && cluster) { | |
916 | node = rb_first(&cluster->root); | |
917 | cluster = NULL; | |
43be2146 | 918 | } |
a67509c3 | 919 | } |
d4452bc5 CM |
920 | return 0; |
921 | fail: | |
922 | return -ENOSPC; | |
923 | } | |
924 | ||
925 | static noinline_for_stack int | |
926 | update_cache_item(struct btrfs_trans_handle *trans, | |
927 | struct btrfs_root *root, | |
928 | struct inode *inode, | |
929 | struct btrfs_path *path, u64 offset, | |
930 | int entries, int bitmaps) | |
931 | { | |
932 | struct btrfs_key key; | |
933 | struct btrfs_free_space_header *header; | |
934 | struct extent_buffer *leaf; | |
935 | int ret; | |
936 | ||
937 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
938 | key.offset = offset; | |
939 | key.type = 0; | |
940 | ||
941 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
942 | if (ret < 0) { | |
943 | clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1, | |
944 | EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL, | |
945 | GFP_NOFS); | |
946 | goto fail; | |
947 | } | |
948 | leaf = path->nodes[0]; | |
949 | if (ret > 0) { | |
950 | struct btrfs_key found_key; | |
951 | ASSERT(path->slots[0]); | |
952 | path->slots[0]--; | |
953 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
954 | if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID || | |
955 | found_key.offset != offset) { | |
956 | clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, | |
957 | inode->i_size - 1, | |
958 | EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, | |
959 | NULL, GFP_NOFS); | |
960 | btrfs_release_path(path); | |
961 | goto fail; | |
962 | } | |
963 | } | |
964 | ||
965 | BTRFS_I(inode)->generation = trans->transid; | |
966 | header = btrfs_item_ptr(leaf, path->slots[0], | |
967 | struct btrfs_free_space_header); | |
968 | btrfs_set_free_space_entries(leaf, header, entries); | |
969 | btrfs_set_free_space_bitmaps(leaf, header, bitmaps); | |
970 | btrfs_set_free_space_generation(leaf, header, trans->transid); | |
971 | btrfs_mark_buffer_dirty(leaf); | |
972 | btrfs_release_path(path); | |
973 | ||
974 | return 0; | |
975 | ||
976 | fail: | |
977 | return -1; | |
978 | } | |
979 | ||
980 | static noinline_for_stack int | |
5349d6c3 MX |
981 | write_pinned_extent_entries(struct btrfs_root *root, |
982 | struct btrfs_block_group_cache *block_group, | |
983 | struct io_ctl *io_ctl, | |
984 | int *entries) | |
d4452bc5 CM |
985 | { |
986 | u64 start, extent_start, extent_end, len; | |
d4452bc5 CM |
987 | struct extent_io_tree *unpin = NULL; |
988 | int ret; | |
43be2146 | 989 | |
5349d6c3 MX |
990 | if (!block_group) |
991 | return 0; | |
992 | ||
a67509c3 JB |
993 | /* |
994 | * We want to add any pinned extents to our free space cache | |
995 | * so we don't leak the space | |
d4452bc5 | 996 | * |
db804f23 LZ |
997 | * We shouldn't have switched the pinned extents yet so this is the |
998 | * right one | |
999 | */ | |
1000 | unpin = root->fs_info->pinned_extents; | |
1001 | ||
5349d6c3 | 1002 | start = block_group->key.objectid; |
db804f23 | 1003 | |
5349d6c3 | 1004 | while (start < block_group->key.objectid + block_group->key.offset) { |
db804f23 LZ |
1005 | ret = find_first_extent_bit(unpin, start, |
1006 | &extent_start, &extent_end, | |
e6138876 | 1007 | EXTENT_DIRTY, NULL); |
5349d6c3 MX |
1008 | if (ret) |
1009 | return 0; | |
0cb59c99 | 1010 | |
a67509c3 | 1011 | /* This pinned extent is out of our range */ |
db804f23 | 1012 | if (extent_start >= block_group->key.objectid + |
a67509c3 | 1013 | block_group->key.offset) |
5349d6c3 | 1014 | return 0; |
2f356126 | 1015 | |
db804f23 LZ |
1016 | extent_start = max(extent_start, start); |
1017 | extent_end = min(block_group->key.objectid + | |
1018 | block_group->key.offset, extent_end + 1); | |
1019 | len = extent_end - extent_start; | |
0cb59c99 | 1020 | |
d4452bc5 CM |
1021 | *entries += 1; |
1022 | ret = io_ctl_add_entry(io_ctl, extent_start, len, NULL); | |
a67509c3 | 1023 | if (ret) |
5349d6c3 | 1024 | return -ENOSPC; |
0cb59c99 | 1025 | |
db804f23 | 1026 | start = extent_end; |
a67509c3 | 1027 | } |
0cb59c99 | 1028 | |
5349d6c3 MX |
1029 | return 0; |
1030 | } | |
1031 | ||
1032 | static noinline_for_stack int | |
1033 | write_bitmap_entries(struct io_ctl *io_ctl, struct list_head *bitmap_list) | |
1034 | { | |
1035 | struct list_head *pos, *n; | |
1036 | int ret; | |
1037 | ||
0cb59c99 | 1038 | /* Write out the bitmaps */ |
d4452bc5 | 1039 | list_for_each_safe(pos, n, bitmap_list) { |
0cb59c99 JB |
1040 | struct btrfs_free_space *entry = |
1041 | list_entry(pos, struct btrfs_free_space, list); | |
1042 | ||
d4452bc5 | 1043 | ret = io_ctl_add_bitmap(io_ctl, entry->bitmap); |
a67509c3 | 1044 | if (ret) |
5349d6c3 | 1045 | return -ENOSPC; |
0cb59c99 | 1046 | list_del_init(&entry->list); |
be1a12a0 JB |
1047 | } |
1048 | ||
5349d6c3 MX |
1049 | return 0; |
1050 | } | |
0cb59c99 | 1051 | |
5349d6c3 MX |
1052 | static int flush_dirty_cache(struct inode *inode) |
1053 | { | |
1054 | int ret; | |
be1a12a0 | 1055 | |
0ef8b726 | 1056 | ret = btrfs_wait_ordered_range(inode, 0, (u64)-1); |
5349d6c3 | 1057 | if (ret) |
0ef8b726 JB |
1058 | clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1, |
1059 | EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL, | |
1060 | GFP_NOFS); | |
0cb59c99 | 1061 | |
5349d6c3 | 1062 | return ret; |
d4452bc5 CM |
1063 | } |
1064 | ||
1065 | static void noinline_for_stack | |
1066 | cleanup_write_cache_enospc(struct inode *inode, | |
1067 | struct io_ctl *io_ctl, | |
1068 | struct extent_state **cached_state, | |
1069 | struct list_head *bitmap_list) | |
1070 | { | |
1071 | struct list_head *pos, *n; | |
5349d6c3 | 1072 | |
d4452bc5 CM |
1073 | list_for_each_safe(pos, n, bitmap_list) { |
1074 | struct btrfs_free_space *entry = | |
1075 | list_entry(pos, struct btrfs_free_space, list); | |
1076 | list_del_init(&entry->list); | |
0cb59c99 | 1077 | } |
d4452bc5 CM |
1078 | io_ctl_drop_pages(io_ctl); |
1079 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0, | |
1080 | i_size_read(inode) - 1, cached_state, | |
1081 | GFP_NOFS); | |
1082 | } | |
549b4fdb | 1083 | |
d4452bc5 CM |
1084 | /** |
1085 | * __btrfs_write_out_cache - write out cached info to an inode | |
1086 | * @root - the root the inode belongs to | |
1087 | * @ctl - the free space cache we are going to write out | |
1088 | * @block_group - the block_group for this cache if it belongs to a block_group | |
1089 | * @trans - the trans handle | |
1090 | * @path - the path to use | |
1091 | * @offset - the offset for the key we'll insert | |
1092 | * | |
1093 | * This function writes out a free space cache struct to disk for quick recovery | |
1094 | * on mount. This will return 0 if it was successfull in writing the cache out, | |
1095 | * and -1 if it was not. | |
1096 | */ | |
1097 | static int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode, | |
1098 | struct btrfs_free_space_ctl *ctl, | |
1099 | struct btrfs_block_group_cache *block_group, | |
1100 | struct btrfs_trans_handle *trans, | |
1101 | struct btrfs_path *path, u64 offset) | |
1102 | { | |
1103 | struct extent_state *cached_state = NULL; | |
1104 | struct io_ctl io_ctl; | |
5349d6c3 | 1105 | LIST_HEAD(bitmap_list); |
d4452bc5 CM |
1106 | int entries = 0; |
1107 | int bitmaps = 0; | |
1108 | int ret; | |
d4452bc5 CM |
1109 | |
1110 | if (!i_size_read(inode)) | |
1111 | return -1; | |
1112 | ||
5349d6c3 | 1113 | ret = io_ctl_init(&io_ctl, inode, root, 1); |
d4452bc5 CM |
1114 | if (ret) |
1115 | return -1; | |
1116 | ||
e570fd27 MX |
1117 | if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA)) { |
1118 | down_write(&block_group->data_rwsem); | |
1119 | spin_lock(&block_group->lock); | |
1120 | if (block_group->delalloc_bytes) { | |
1121 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
1122 | spin_unlock(&block_group->lock); | |
1123 | up_write(&block_group->data_rwsem); | |
1124 | BTRFS_I(inode)->generation = 0; | |
1125 | ret = 0; | |
1126 | goto out; | |
1127 | } | |
1128 | spin_unlock(&block_group->lock); | |
1129 | } | |
1130 | ||
d4452bc5 CM |
1131 | /* Lock all pages first so we can lock the extent safely. */ |
1132 | io_ctl_prepare_pages(&io_ctl, inode, 0); | |
1133 | ||
1134 | lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1, | |
1135 | 0, &cached_state); | |
1136 | ||
d4452bc5 CM |
1137 | io_ctl_set_generation(&io_ctl, trans->transid); |
1138 | ||
5349d6c3 | 1139 | /* Write out the extent entries in the free space cache */ |
d4452bc5 CM |
1140 | ret = write_cache_extent_entries(&io_ctl, ctl, |
1141 | block_group, &entries, &bitmaps, | |
1142 | &bitmap_list); | |
1143 | if (ret) | |
1144 | goto out_nospc; | |
1145 | ||
5349d6c3 MX |
1146 | /* |
1147 | * Some spaces that are freed in the current transaction are pinned, | |
1148 | * they will be added into free space cache after the transaction is | |
1149 | * committed, we shouldn't lose them. | |
1150 | */ | |
1151 | ret = write_pinned_extent_entries(root, block_group, &io_ctl, &entries); | |
1152 | if (ret) | |
1153 | goto out_nospc; | |
1154 | ||
1155 | /* At last, we write out all the bitmaps. */ | |
1156 | ret = write_bitmap_entries(&io_ctl, &bitmap_list); | |
1157 | if (ret) | |
1158 | goto out_nospc; | |
1159 | ||
1160 | /* Zero out the rest of the pages just to make sure */ | |
1161 | io_ctl_zero_remaining_pages(&io_ctl); | |
d4452bc5 | 1162 | |
5349d6c3 MX |
1163 | /* Everything is written out, now we dirty the pages in the file. */ |
1164 | ret = btrfs_dirty_pages(root, inode, io_ctl.pages, io_ctl.num_pages, | |
1165 | 0, i_size_read(inode), &cached_state); | |
1166 | if (ret) | |
d4452bc5 | 1167 | goto out_nospc; |
5349d6c3 | 1168 | |
e570fd27 MX |
1169 | if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA)) |
1170 | up_write(&block_group->data_rwsem); | |
5349d6c3 MX |
1171 | /* |
1172 | * Release the pages and unlock the extent, we will flush | |
1173 | * them out later | |
1174 | */ | |
1175 | io_ctl_drop_pages(&io_ctl); | |
1176 | ||
1177 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0, | |
1178 | i_size_read(inode) - 1, &cached_state, GFP_NOFS); | |
1179 | ||
1180 | /* Flush the dirty pages in the cache file. */ | |
1181 | ret = flush_dirty_cache(inode); | |
1182 | if (ret) | |
d4452bc5 CM |
1183 | goto out; |
1184 | ||
5349d6c3 MX |
1185 | /* Update the cache item to tell everyone this cache file is valid. */ |
1186 | ret = update_cache_item(trans, root, inode, path, offset, | |
d4452bc5 | 1187 | entries, bitmaps); |
2f356126 | 1188 | out: |
a67509c3 | 1189 | io_ctl_free(&io_ctl); |
5349d6c3 | 1190 | if (ret) { |
a67509c3 | 1191 | invalidate_inode_pages2(inode->i_mapping); |
0cb59c99 JB |
1192 | BTRFS_I(inode)->generation = 0; |
1193 | } | |
0cb59c99 | 1194 | btrfs_update_inode(trans, root, inode); |
5349d6c3 | 1195 | return ret; |
a67509c3 JB |
1196 | |
1197 | out_nospc: | |
d4452bc5 | 1198 | cleanup_write_cache_enospc(inode, &io_ctl, &cached_state, &bitmap_list); |
e570fd27 MX |
1199 | |
1200 | if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA)) | |
1201 | up_write(&block_group->data_rwsem); | |
1202 | ||
a67509c3 | 1203 | goto out; |
0414efae LZ |
1204 | } |
1205 | ||
1206 | int btrfs_write_out_cache(struct btrfs_root *root, | |
1207 | struct btrfs_trans_handle *trans, | |
1208 | struct btrfs_block_group_cache *block_group, | |
1209 | struct btrfs_path *path) | |
1210 | { | |
1211 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
1212 | struct inode *inode; | |
1213 | int ret = 0; | |
1214 | ||
1215 | root = root->fs_info->tree_root; | |
1216 | ||
1217 | spin_lock(&block_group->lock); | |
1218 | if (block_group->disk_cache_state < BTRFS_DC_SETUP) { | |
1219 | spin_unlock(&block_group->lock); | |
1220 | return 0; | |
1221 | } | |
e570fd27 MX |
1222 | |
1223 | if (block_group->delalloc_bytes) { | |
1224 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
1225 | spin_unlock(&block_group->lock); | |
1226 | return 0; | |
1227 | } | |
0414efae LZ |
1228 | spin_unlock(&block_group->lock); |
1229 | ||
1230 | inode = lookup_free_space_inode(root, block_group, path); | |
1231 | if (IS_ERR(inode)) | |
1232 | return 0; | |
1233 | ||
1234 | ret = __btrfs_write_out_cache(root, inode, ctl, block_group, trans, | |
1235 | path, block_group->key.objectid); | |
c09544e0 | 1236 | if (ret) { |
0414efae LZ |
1237 | spin_lock(&block_group->lock); |
1238 | block_group->disk_cache_state = BTRFS_DC_ERROR; | |
1239 | spin_unlock(&block_group->lock); | |
82d5902d | 1240 | ret = 0; |
c09544e0 | 1241 | #ifdef DEBUG |
c2cf52eb SK |
1242 | btrfs_err(root->fs_info, |
1243 | "failed to write free space cache for block group %llu", | |
1244 | block_group->key.objectid); | |
c09544e0 | 1245 | #endif |
0414efae LZ |
1246 | } |
1247 | ||
0cb59c99 JB |
1248 | iput(inode); |
1249 | return ret; | |
1250 | } | |
1251 | ||
34d52cb6 | 1252 | static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit, |
96303081 | 1253 | u64 offset) |
0f9dd46c | 1254 | { |
b12d6869 | 1255 | ASSERT(offset >= bitmap_start); |
96303081 | 1256 | offset -= bitmap_start; |
34d52cb6 | 1257 | return (unsigned long)(div_u64(offset, unit)); |
96303081 | 1258 | } |
0f9dd46c | 1259 | |
34d52cb6 | 1260 | static inline unsigned long bytes_to_bits(u64 bytes, u32 unit) |
96303081 | 1261 | { |
34d52cb6 | 1262 | return (unsigned long)(div_u64(bytes, unit)); |
96303081 | 1263 | } |
0f9dd46c | 1264 | |
34d52cb6 | 1265 | static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1266 | u64 offset) |
1267 | { | |
1268 | u64 bitmap_start; | |
1269 | u64 bytes_per_bitmap; | |
0f9dd46c | 1270 | |
34d52cb6 LZ |
1271 | bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit; |
1272 | bitmap_start = offset - ctl->start; | |
96303081 JB |
1273 | bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap); |
1274 | bitmap_start *= bytes_per_bitmap; | |
34d52cb6 | 1275 | bitmap_start += ctl->start; |
0f9dd46c | 1276 | |
96303081 | 1277 | return bitmap_start; |
0f9dd46c JB |
1278 | } |
1279 | ||
96303081 JB |
1280 | static int tree_insert_offset(struct rb_root *root, u64 offset, |
1281 | struct rb_node *node, int bitmap) | |
0f9dd46c JB |
1282 | { |
1283 | struct rb_node **p = &root->rb_node; | |
1284 | struct rb_node *parent = NULL; | |
1285 | struct btrfs_free_space *info; | |
1286 | ||
1287 | while (*p) { | |
1288 | parent = *p; | |
96303081 | 1289 | info = rb_entry(parent, struct btrfs_free_space, offset_index); |
0f9dd46c | 1290 | |
96303081 | 1291 | if (offset < info->offset) { |
0f9dd46c | 1292 | p = &(*p)->rb_left; |
96303081 | 1293 | } else if (offset > info->offset) { |
0f9dd46c | 1294 | p = &(*p)->rb_right; |
96303081 JB |
1295 | } else { |
1296 | /* | |
1297 | * we could have a bitmap entry and an extent entry | |
1298 | * share the same offset. If this is the case, we want | |
1299 | * the extent entry to always be found first if we do a | |
1300 | * linear search through the tree, since we want to have | |
1301 | * the quickest allocation time, and allocating from an | |
1302 | * extent is faster than allocating from a bitmap. So | |
1303 | * if we're inserting a bitmap and we find an entry at | |
1304 | * this offset, we want to go right, or after this entry | |
1305 | * logically. If we are inserting an extent and we've | |
1306 | * found a bitmap, we want to go left, or before | |
1307 | * logically. | |
1308 | */ | |
1309 | if (bitmap) { | |
207dde82 JB |
1310 | if (info->bitmap) { |
1311 | WARN_ON_ONCE(1); | |
1312 | return -EEXIST; | |
1313 | } | |
96303081 JB |
1314 | p = &(*p)->rb_right; |
1315 | } else { | |
207dde82 JB |
1316 | if (!info->bitmap) { |
1317 | WARN_ON_ONCE(1); | |
1318 | return -EEXIST; | |
1319 | } | |
96303081 JB |
1320 | p = &(*p)->rb_left; |
1321 | } | |
1322 | } | |
0f9dd46c JB |
1323 | } |
1324 | ||
1325 | rb_link_node(node, parent, p); | |
1326 | rb_insert_color(node, root); | |
1327 | ||
1328 | return 0; | |
1329 | } | |
1330 | ||
1331 | /* | |
70cb0743 JB |
1332 | * searches the tree for the given offset. |
1333 | * | |
96303081 JB |
1334 | * fuzzy - If this is set, then we are trying to make an allocation, and we just |
1335 | * want a section that has at least bytes size and comes at or after the given | |
1336 | * offset. | |
0f9dd46c | 1337 | */ |
96303081 | 1338 | static struct btrfs_free_space * |
34d52cb6 | 1339 | tree_search_offset(struct btrfs_free_space_ctl *ctl, |
96303081 | 1340 | u64 offset, int bitmap_only, int fuzzy) |
0f9dd46c | 1341 | { |
34d52cb6 | 1342 | struct rb_node *n = ctl->free_space_offset.rb_node; |
96303081 JB |
1343 | struct btrfs_free_space *entry, *prev = NULL; |
1344 | ||
1345 | /* find entry that is closest to the 'offset' */ | |
1346 | while (1) { | |
1347 | if (!n) { | |
1348 | entry = NULL; | |
1349 | break; | |
1350 | } | |
0f9dd46c | 1351 | |
0f9dd46c | 1352 | entry = rb_entry(n, struct btrfs_free_space, offset_index); |
96303081 | 1353 | prev = entry; |
0f9dd46c | 1354 | |
96303081 | 1355 | if (offset < entry->offset) |
0f9dd46c | 1356 | n = n->rb_left; |
96303081 | 1357 | else if (offset > entry->offset) |
0f9dd46c | 1358 | n = n->rb_right; |
96303081 | 1359 | else |
0f9dd46c | 1360 | break; |
0f9dd46c JB |
1361 | } |
1362 | ||
96303081 JB |
1363 | if (bitmap_only) { |
1364 | if (!entry) | |
1365 | return NULL; | |
1366 | if (entry->bitmap) | |
1367 | return entry; | |
0f9dd46c | 1368 | |
96303081 JB |
1369 | /* |
1370 | * bitmap entry and extent entry may share same offset, | |
1371 | * in that case, bitmap entry comes after extent entry. | |
1372 | */ | |
1373 | n = rb_next(n); | |
1374 | if (!n) | |
1375 | return NULL; | |
1376 | entry = rb_entry(n, struct btrfs_free_space, offset_index); | |
1377 | if (entry->offset != offset) | |
1378 | return NULL; | |
0f9dd46c | 1379 | |
96303081 JB |
1380 | WARN_ON(!entry->bitmap); |
1381 | return entry; | |
1382 | } else if (entry) { | |
1383 | if (entry->bitmap) { | |
0f9dd46c | 1384 | /* |
96303081 JB |
1385 | * if previous extent entry covers the offset, |
1386 | * we should return it instead of the bitmap entry | |
0f9dd46c | 1387 | */ |
de6c4115 MX |
1388 | n = rb_prev(&entry->offset_index); |
1389 | if (n) { | |
96303081 JB |
1390 | prev = rb_entry(n, struct btrfs_free_space, |
1391 | offset_index); | |
de6c4115 MX |
1392 | if (!prev->bitmap && |
1393 | prev->offset + prev->bytes > offset) | |
1394 | entry = prev; | |
0f9dd46c | 1395 | } |
96303081 JB |
1396 | } |
1397 | return entry; | |
1398 | } | |
1399 | ||
1400 | if (!prev) | |
1401 | return NULL; | |
1402 | ||
1403 | /* find last entry before the 'offset' */ | |
1404 | entry = prev; | |
1405 | if (entry->offset > offset) { | |
1406 | n = rb_prev(&entry->offset_index); | |
1407 | if (n) { | |
1408 | entry = rb_entry(n, struct btrfs_free_space, | |
1409 | offset_index); | |
b12d6869 | 1410 | ASSERT(entry->offset <= offset); |
0f9dd46c | 1411 | } else { |
96303081 JB |
1412 | if (fuzzy) |
1413 | return entry; | |
1414 | else | |
1415 | return NULL; | |
0f9dd46c JB |
1416 | } |
1417 | } | |
1418 | ||
96303081 | 1419 | if (entry->bitmap) { |
de6c4115 MX |
1420 | n = rb_prev(&entry->offset_index); |
1421 | if (n) { | |
96303081 JB |
1422 | prev = rb_entry(n, struct btrfs_free_space, |
1423 | offset_index); | |
de6c4115 MX |
1424 | if (!prev->bitmap && |
1425 | prev->offset + prev->bytes > offset) | |
1426 | return prev; | |
96303081 | 1427 | } |
34d52cb6 | 1428 | if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset) |
96303081 JB |
1429 | return entry; |
1430 | } else if (entry->offset + entry->bytes > offset) | |
1431 | return entry; | |
1432 | ||
1433 | if (!fuzzy) | |
1434 | return NULL; | |
1435 | ||
1436 | while (1) { | |
1437 | if (entry->bitmap) { | |
1438 | if (entry->offset + BITS_PER_BITMAP * | |
34d52cb6 | 1439 | ctl->unit > offset) |
96303081 JB |
1440 | break; |
1441 | } else { | |
1442 | if (entry->offset + entry->bytes > offset) | |
1443 | break; | |
1444 | } | |
1445 | ||
1446 | n = rb_next(&entry->offset_index); | |
1447 | if (!n) | |
1448 | return NULL; | |
1449 | entry = rb_entry(n, struct btrfs_free_space, offset_index); | |
1450 | } | |
1451 | return entry; | |
0f9dd46c JB |
1452 | } |
1453 | ||
f333adb5 | 1454 | static inline void |
34d52cb6 | 1455 | __unlink_free_space(struct btrfs_free_space_ctl *ctl, |
f333adb5 | 1456 | struct btrfs_free_space *info) |
0f9dd46c | 1457 | { |
34d52cb6 LZ |
1458 | rb_erase(&info->offset_index, &ctl->free_space_offset); |
1459 | ctl->free_extents--; | |
f333adb5 LZ |
1460 | } |
1461 | ||
34d52cb6 | 1462 | static void unlink_free_space(struct btrfs_free_space_ctl *ctl, |
f333adb5 LZ |
1463 | struct btrfs_free_space *info) |
1464 | { | |
34d52cb6 LZ |
1465 | __unlink_free_space(ctl, info); |
1466 | ctl->free_space -= info->bytes; | |
0f9dd46c JB |
1467 | } |
1468 | ||
34d52cb6 | 1469 | static int link_free_space(struct btrfs_free_space_ctl *ctl, |
0f9dd46c JB |
1470 | struct btrfs_free_space *info) |
1471 | { | |
1472 | int ret = 0; | |
1473 | ||
b12d6869 | 1474 | ASSERT(info->bytes || info->bitmap); |
34d52cb6 | 1475 | ret = tree_insert_offset(&ctl->free_space_offset, info->offset, |
96303081 | 1476 | &info->offset_index, (info->bitmap != NULL)); |
0f9dd46c JB |
1477 | if (ret) |
1478 | return ret; | |
1479 | ||
34d52cb6 LZ |
1480 | ctl->free_space += info->bytes; |
1481 | ctl->free_extents++; | |
96303081 JB |
1482 | return ret; |
1483 | } | |
1484 | ||
34d52cb6 | 1485 | static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl) |
96303081 | 1486 | { |
34d52cb6 | 1487 | struct btrfs_block_group_cache *block_group = ctl->private; |
25891f79 JB |
1488 | u64 max_bytes; |
1489 | u64 bitmap_bytes; | |
1490 | u64 extent_bytes; | |
8eb2d829 | 1491 | u64 size = block_group->key.offset; |
96009762 | 1492 | u64 bytes_per_bg = BITS_PER_BITMAP * ctl->unit; |
34d52cb6 LZ |
1493 | int max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg); |
1494 | ||
dde5740f JB |
1495 | max_bitmaps = max(max_bitmaps, 1); |
1496 | ||
b12d6869 | 1497 | ASSERT(ctl->total_bitmaps <= max_bitmaps); |
96303081 JB |
1498 | |
1499 | /* | |
1500 | * The goal is to keep the total amount of memory used per 1gb of space | |
1501 | * at or below 32k, so we need to adjust how much memory we allow to be | |
1502 | * used by extent based free space tracking | |
1503 | */ | |
8eb2d829 LZ |
1504 | if (size < 1024 * 1024 * 1024) |
1505 | max_bytes = MAX_CACHE_BYTES_PER_GIG; | |
1506 | else | |
1507 | max_bytes = MAX_CACHE_BYTES_PER_GIG * | |
1508 | div64_u64(size, 1024 * 1024 * 1024); | |
96303081 | 1509 | |
25891f79 JB |
1510 | /* |
1511 | * we want to account for 1 more bitmap than what we have so we can make | |
1512 | * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as | |
1513 | * we add more bitmaps. | |
1514 | */ | |
34d52cb6 | 1515 | bitmap_bytes = (ctl->total_bitmaps + 1) * PAGE_CACHE_SIZE; |
96303081 | 1516 | |
25891f79 | 1517 | if (bitmap_bytes >= max_bytes) { |
34d52cb6 | 1518 | ctl->extents_thresh = 0; |
25891f79 JB |
1519 | return; |
1520 | } | |
96303081 | 1521 | |
25891f79 JB |
1522 | /* |
1523 | * we want the extent entry threshold to always be at most 1/2 the maxw | |
1524 | * bytes we can have, or whatever is less than that. | |
1525 | */ | |
1526 | extent_bytes = max_bytes - bitmap_bytes; | |
1527 | extent_bytes = min_t(u64, extent_bytes, div64_u64(max_bytes, 2)); | |
96303081 | 1528 | |
34d52cb6 | 1529 | ctl->extents_thresh = |
25891f79 | 1530 | div64_u64(extent_bytes, (sizeof(struct btrfs_free_space))); |
96303081 JB |
1531 | } |
1532 | ||
bb3ac5a4 MX |
1533 | static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl, |
1534 | struct btrfs_free_space *info, | |
1535 | u64 offset, u64 bytes) | |
96303081 | 1536 | { |
f38b6e75 | 1537 | unsigned long start, count; |
96303081 | 1538 | |
34d52cb6 LZ |
1539 | start = offset_to_bit(info->offset, ctl->unit, offset); |
1540 | count = bytes_to_bits(bytes, ctl->unit); | |
b12d6869 | 1541 | ASSERT(start + count <= BITS_PER_BITMAP); |
96303081 | 1542 | |
f38b6e75 | 1543 | bitmap_clear(info->bitmap, start, count); |
96303081 JB |
1544 | |
1545 | info->bytes -= bytes; | |
bb3ac5a4 MX |
1546 | } |
1547 | ||
1548 | static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl, | |
1549 | struct btrfs_free_space *info, u64 offset, | |
1550 | u64 bytes) | |
1551 | { | |
1552 | __bitmap_clear_bits(ctl, info, offset, bytes); | |
34d52cb6 | 1553 | ctl->free_space -= bytes; |
96303081 JB |
1554 | } |
1555 | ||
34d52cb6 | 1556 | static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl, |
817d52f8 JB |
1557 | struct btrfs_free_space *info, u64 offset, |
1558 | u64 bytes) | |
96303081 | 1559 | { |
f38b6e75 | 1560 | unsigned long start, count; |
96303081 | 1561 | |
34d52cb6 LZ |
1562 | start = offset_to_bit(info->offset, ctl->unit, offset); |
1563 | count = bytes_to_bits(bytes, ctl->unit); | |
b12d6869 | 1564 | ASSERT(start + count <= BITS_PER_BITMAP); |
96303081 | 1565 | |
f38b6e75 | 1566 | bitmap_set(info->bitmap, start, count); |
96303081 JB |
1567 | |
1568 | info->bytes += bytes; | |
34d52cb6 | 1569 | ctl->free_space += bytes; |
96303081 JB |
1570 | } |
1571 | ||
a4820398 MX |
1572 | /* |
1573 | * If we can not find suitable extent, we will use bytes to record | |
1574 | * the size of the max extent. | |
1575 | */ | |
34d52cb6 | 1576 | static int search_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1577 | struct btrfs_free_space *bitmap_info, u64 *offset, |
1578 | u64 *bytes) | |
1579 | { | |
1580 | unsigned long found_bits = 0; | |
a4820398 | 1581 | unsigned long max_bits = 0; |
96303081 JB |
1582 | unsigned long bits, i; |
1583 | unsigned long next_zero; | |
a4820398 | 1584 | unsigned long extent_bits; |
96303081 | 1585 | |
34d52cb6 | 1586 | i = offset_to_bit(bitmap_info->offset, ctl->unit, |
96303081 | 1587 | max_t(u64, *offset, bitmap_info->offset)); |
34d52cb6 | 1588 | bits = bytes_to_bits(*bytes, ctl->unit); |
96303081 | 1589 | |
ebb3dad4 | 1590 | for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) { |
96303081 JB |
1591 | next_zero = find_next_zero_bit(bitmap_info->bitmap, |
1592 | BITS_PER_BITMAP, i); | |
a4820398 MX |
1593 | extent_bits = next_zero - i; |
1594 | if (extent_bits >= bits) { | |
1595 | found_bits = extent_bits; | |
96303081 | 1596 | break; |
a4820398 MX |
1597 | } else if (extent_bits > max_bits) { |
1598 | max_bits = extent_bits; | |
96303081 JB |
1599 | } |
1600 | i = next_zero; | |
1601 | } | |
1602 | ||
1603 | if (found_bits) { | |
34d52cb6 LZ |
1604 | *offset = (u64)(i * ctl->unit) + bitmap_info->offset; |
1605 | *bytes = (u64)(found_bits) * ctl->unit; | |
96303081 JB |
1606 | return 0; |
1607 | } | |
1608 | ||
a4820398 | 1609 | *bytes = (u64)(max_bits) * ctl->unit; |
96303081 JB |
1610 | return -1; |
1611 | } | |
1612 | ||
a4820398 | 1613 | /* Cache the size of the max extent in bytes */ |
34d52cb6 | 1614 | static struct btrfs_free_space * |
53b381b3 | 1615 | find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes, |
a4820398 | 1616 | unsigned long align, u64 *max_extent_size) |
96303081 JB |
1617 | { |
1618 | struct btrfs_free_space *entry; | |
1619 | struct rb_node *node; | |
53b381b3 DW |
1620 | u64 tmp; |
1621 | u64 align_off; | |
96303081 JB |
1622 | int ret; |
1623 | ||
34d52cb6 | 1624 | if (!ctl->free_space_offset.rb_node) |
a4820398 | 1625 | goto out; |
96303081 | 1626 | |
34d52cb6 | 1627 | entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1); |
96303081 | 1628 | if (!entry) |
a4820398 | 1629 | goto out; |
96303081 JB |
1630 | |
1631 | for (node = &entry->offset_index; node; node = rb_next(node)) { | |
1632 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
a4820398 MX |
1633 | if (entry->bytes < *bytes) { |
1634 | if (entry->bytes > *max_extent_size) | |
1635 | *max_extent_size = entry->bytes; | |
96303081 | 1636 | continue; |
a4820398 | 1637 | } |
96303081 | 1638 | |
53b381b3 DW |
1639 | /* make sure the space returned is big enough |
1640 | * to match our requested alignment | |
1641 | */ | |
1642 | if (*bytes >= align) { | |
a4820398 | 1643 | tmp = entry->offset - ctl->start + align - 1; |
53b381b3 DW |
1644 | do_div(tmp, align); |
1645 | tmp = tmp * align + ctl->start; | |
1646 | align_off = tmp - entry->offset; | |
1647 | } else { | |
1648 | align_off = 0; | |
1649 | tmp = entry->offset; | |
1650 | } | |
1651 | ||
a4820398 MX |
1652 | if (entry->bytes < *bytes + align_off) { |
1653 | if (entry->bytes > *max_extent_size) | |
1654 | *max_extent_size = entry->bytes; | |
53b381b3 | 1655 | continue; |
a4820398 | 1656 | } |
53b381b3 | 1657 | |
96303081 | 1658 | if (entry->bitmap) { |
a4820398 MX |
1659 | u64 size = *bytes; |
1660 | ||
1661 | ret = search_bitmap(ctl, entry, &tmp, &size); | |
53b381b3 DW |
1662 | if (!ret) { |
1663 | *offset = tmp; | |
a4820398 | 1664 | *bytes = size; |
96303081 | 1665 | return entry; |
a4820398 MX |
1666 | } else if (size > *max_extent_size) { |
1667 | *max_extent_size = size; | |
53b381b3 | 1668 | } |
96303081 JB |
1669 | continue; |
1670 | } | |
1671 | ||
53b381b3 DW |
1672 | *offset = tmp; |
1673 | *bytes = entry->bytes - align_off; | |
96303081 JB |
1674 | return entry; |
1675 | } | |
a4820398 | 1676 | out: |
96303081 JB |
1677 | return NULL; |
1678 | } | |
1679 | ||
34d52cb6 | 1680 | static void add_new_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1681 | struct btrfs_free_space *info, u64 offset) |
1682 | { | |
34d52cb6 | 1683 | info->offset = offset_to_bitmap(ctl, offset); |
f019f426 | 1684 | info->bytes = 0; |
f2d0f676 | 1685 | INIT_LIST_HEAD(&info->list); |
34d52cb6 LZ |
1686 | link_free_space(ctl, info); |
1687 | ctl->total_bitmaps++; | |
96303081 | 1688 | |
34d52cb6 | 1689 | ctl->op->recalc_thresholds(ctl); |
96303081 JB |
1690 | } |
1691 | ||
34d52cb6 | 1692 | static void free_bitmap(struct btrfs_free_space_ctl *ctl, |
edf6e2d1 LZ |
1693 | struct btrfs_free_space *bitmap_info) |
1694 | { | |
34d52cb6 | 1695 | unlink_free_space(ctl, bitmap_info); |
edf6e2d1 | 1696 | kfree(bitmap_info->bitmap); |
dc89e982 | 1697 | kmem_cache_free(btrfs_free_space_cachep, bitmap_info); |
34d52cb6 LZ |
1698 | ctl->total_bitmaps--; |
1699 | ctl->op->recalc_thresholds(ctl); | |
edf6e2d1 LZ |
1700 | } |
1701 | ||
34d52cb6 | 1702 | static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl, |
96303081 JB |
1703 | struct btrfs_free_space *bitmap_info, |
1704 | u64 *offset, u64 *bytes) | |
1705 | { | |
1706 | u64 end; | |
6606bb97 JB |
1707 | u64 search_start, search_bytes; |
1708 | int ret; | |
96303081 JB |
1709 | |
1710 | again: | |
34d52cb6 | 1711 | end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1; |
96303081 | 1712 | |
6606bb97 | 1713 | /* |
bdb7d303 JB |
1714 | * We need to search for bits in this bitmap. We could only cover some |
1715 | * of the extent in this bitmap thanks to how we add space, so we need | |
1716 | * to search for as much as it as we can and clear that amount, and then | |
1717 | * go searching for the next bit. | |
6606bb97 JB |
1718 | */ |
1719 | search_start = *offset; | |
bdb7d303 | 1720 | search_bytes = ctl->unit; |
13dbc089 | 1721 | search_bytes = min(search_bytes, end - search_start + 1); |
34d52cb6 | 1722 | ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes); |
b50c6e25 JB |
1723 | if (ret < 0 || search_start != *offset) |
1724 | return -EINVAL; | |
6606bb97 | 1725 | |
bdb7d303 JB |
1726 | /* We may have found more bits than what we need */ |
1727 | search_bytes = min(search_bytes, *bytes); | |
1728 | ||
1729 | /* Cannot clear past the end of the bitmap */ | |
1730 | search_bytes = min(search_bytes, end - search_start + 1); | |
1731 | ||
1732 | bitmap_clear_bits(ctl, bitmap_info, search_start, search_bytes); | |
1733 | *offset += search_bytes; | |
1734 | *bytes -= search_bytes; | |
96303081 JB |
1735 | |
1736 | if (*bytes) { | |
6606bb97 | 1737 | struct rb_node *next = rb_next(&bitmap_info->offset_index); |
edf6e2d1 | 1738 | if (!bitmap_info->bytes) |
34d52cb6 | 1739 | free_bitmap(ctl, bitmap_info); |
96303081 | 1740 | |
6606bb97 JB |
1741 | /* |
1742 | * no entry after this bitmap, but we still have bytes to | |
1743 | * remove, so something has gone wrong. | |
1744 | */ | |
1745 | if (!next) | |
96303081 JB |
1746 | return -EINVAL; |
1747 | ||
6606bb97 JB |
1748 | bitmap_info = rb_entry(next, struct btrfs_free_space, |
1749 | offset_index); | |
1750 | ||
1751 | /* | |
1752 | * if the next entry isn't a bitmap we need to return to let the | |
1753 | * extent stuff do its work. | |
1754 | */ | |
96303081 JB |
1755 | if (!bitmap_info->bitmap) |
1756 | return -EAGAIN; | |
1757 | ||
6606bb97 JB |
1758 | /* |
1759 | * Ok the next item is a bitmap, but it may not actually hold | |
1760 | * the information for the rest of this free space stuff, so | |
1761 | * look for it, and if we don't find it return so we can try | |
1762 | * everything over again. | |
1763 | */ | |
1764 | search_start = *offset; | |
bdb7d303 | 1765 | search_bytes = ctl->unit; |
34d52cb6 | 1766 | ret = search_bitmap(ctl, bitmap_info, &search_start, |
6606bb97 JB |
1767 | &search_bytes); |
1768 | if (ret < 0 || search_start != *offset) | |
1769 | return -EAGAIN; | |
1770 | ||
96303081 | 1771 | goto again; |
edf6e2d1 | 1772 | } else if (!bitmap_info->bytes) |
34d52cb6 | 1773 | free_bitmap(ctl, bitmap_info); |
96303081 JB |
1774 | |
1775 | return 0; | |
1776 | } | |
1777 | ||
2cdc342c JB |
1778 | static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl, |
1779 | struct btrfs_free_space *info, u64 offset, | |
1780 | u64 bytes) | |
1781 | { | |
1782 | u64 bytes_to_set = 0; | |
1783 | u64 end; | |
1784 | ||
1785 | end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit); | |
1786 | ||
1787 | bytes_to_set = min(end - offset, bytes); | |
1788 | ||
1789 | bitmap_set_bits(ctl, info, offset, bytes_to_set); | |
1790 | ||
1791 | return bytes_to_set; | |
1792 | ||
1793 | } | |
1794 | ||
34d52cb6 LZ |
1795 | static bool use_bitmap(struct btrfs_free_space_ctl *ctl, |
1796 | struct btrfs_free_space *info) | |
96303081 | 1797 | { |
34d52cb6 | 1798 | struct btrfs_block_group_cache *block_group = ctl->private; |
96303081 JB |
1799 | |
1800 | /* | |
1801 | * If we are below the extents threshold then we can add this as an | |
1802 | * extent, and don't have to deal with the bitmap | |
1803 | */ | |
34d52cb6 | 1804 | if (ctl->free_extents < ctl->extents_thresh) { |
32cb0840 JB |
1805 | /* |
1806 | * If this block group has some small extents we don't want to | |
1807 | * use up all of our free slots in the cache with them, we want | |
1808 | * to reserve them to larger extents, however if we have plent | |
1809 | * of cache left then go ahead an dadd them, no sense in adding | |
1810 | * the overhead of a bitmap if we don't have to. | |
1811 | */ | |
1812 | if (info->bytes <= block_group->sectorsize * 4) { | |
34d52cb6 LZ |
1813 | if (ctl->free_extents * 2 <= ctl->extents_thresh) |
1814 | return false; | |
32cb0840 | 1815 | } else { |
34d52cb6 | 1816 | return false; |
32cb0840 JB |
1817 | } |
1818 | } | |
96303081 JB |
1819 | |
1820 | /* | |
dde5740f JB |
1821 | * The original block groups from mkfs can be really small, like 8 |
1822 | * megabytes, so don't bother with a bitmap for those entries. However | |
1823 | * some block groups can be smaller than what a bitmap would cover but | |
1824 | * are still large enough that they could overflow the 32k memory limit, | |
1825 | * so allow those block groups to still be allowed to have a bitmap | |
1826 | * entry. | |
96303081 | 1827 | */ |
dde5740f | 1828 | if (((BITS_PER_BITMAP * ctl->unit) >> 1) > block_group->key.offset) |
34d52cb6 LZ |
1829 | return false; |
1830 | ||
1831 | return true; | |
1832 | } | |
1833 | ||
2cdc342c JB |
1834 | static struct btrfs_free_space_op free_space_op = { |
1835 | .recalc_thresholds = recalculate_thresholds, | |
1836 | .use_bitmap = use_bitmap, | |
1837 | }; | |
1838 | ||
34d52cb6 LZ |
1839 | static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl, |
1840 | struct btrfs_free_space *info) | |
1841 | { | |
1842 | struct btrfs_free_space *bitmap_info; | |
2cdc342c | 1843 | struct btrfs_block_group_cache *block_group = NULL; |
34d52cb6 | 1844 | int added = 0; |
2cdc342c | 1845 | u64 bytes, offset, bytes_added; |
34d52cb6 | 1846 | int ret; |
96303081 JB |
1847 | |
1848 | bytes = info->bytes; | |
1849 | offset = info->offset; | |
1850 | ||
34d52cb6 LZ |
1851 | if (!ctl->op->use_bitmap(ctl, info)) |
1852 | return 0; | |
1853 | ||
2cdc342c JB |
1854 | if (ctl->op == &free_space_op) |
1855 | block_group = ctl->private; | |
38e87880 | 1856 | again: |
2cdc342c JB |
1857 | /* |
1858 | * Since we link bitmaps right into the cluster we need to see if we | |
1859 | * have a cluster here, and if so and it has our bitmap we need to add | |
1860 | * the free space to that bitmap. | |
1861 | */ | |
1862 | if (block_group && !list_empty(&block_group->cluster_list)) { | |
1863 | struct btrfs_free_cluster *cluster; | |
1864 | struct rb_node *node; | |
1865 | struct btrfs_free_space *entry; | |
1866 | ||
1867 | cluster = list_entry(block_group->cluster_list.next, | |
1868 | struct btrfs_free_cluster, | |
1869 | block_group_list); | |
1870 | spin_lock(&cluster->lock); | |
1871 | node = rb_first(&cluster->root); | |
1872 | if (!node) { | |
1873 | spin_unlock(&cluster->lock); | |
38e87880 | 1874 | goto no_cluster_bitmap; |
2cdc342c JB |
1875 | } |
1876 | ||
1877 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
1878 | if (!entry->bitmap) { | |
1879 | spin_unlock(&cluster->lock); | |
38e87880 | 1880 | goto no_cluster_bitmap; |
2cdc342c JB |
1881 | } |
1882 | ||
1883 | if (entry->offset == offset_to_bitmap(ctl, offset)) { | |
1884 | bytes_added = add_bytes_to_bitmap(ctl, entry, | |
1885 | offset, bytes); | |
1886 | bytes -= bytes_added; | |
1887 | offset += bytes_added; | |
1888 | } | |
1889 | spin_unlock(&cluster->lock); | |
1890 | if (!bytes) { | |
1891 | ret = 1; | |
1892 | goto out; | |
1893 | } | |
1894 | } | |
38e87880 CM |
1895 | |
1896 | no_cluster_bitmap: | |
34d52cb6 | 1897 | bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), |
96303081 JB |
1898 | 1, 0); |
1899 | if (!bitmap_info) { | |
b12d6869 | 1900 | ASSERT(added == 0); |
96303081 JB |
1901 | goto new_bitmap; |
1902 | } | |
1903 | ||
2cdc342c JB |
1904 | bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes); |
1905 | bytes -= bytes_added; | |
1906 | offset += bytes_added; | |
1907 | added = 0; | |
96303081 JB |
1908 | |
1909 | if (!bytes) { | |
1910 | ret = 1; | |
1911 | goto out; | |
1912 | } else | |
1913 | goto again; | |
1914 | ||
1915 | new_bitmap: | |
1916 | if (info && info->bitmap) { | |
34d52cb6 | 1917 | add_new_bitmap(ctl, info, offset); |
96303081 JB |
1918 | added = 1; |
1919 | info = NULL; | |
1920 | goto again; | |
1921 | } else { | |
34d52cb6 | 1922 | spin_unlock(&ctl->tree_lock); |
96303081 JB |
1923 | |
1924 | /* no pre-allocated info, allocate a new one */ | |
1925 | if (!info) { | |
dc89e982 JB |
1926 | info = kmem_cache_zalloc(btrfs_free_space_cachep, |
1927 | GFP_NOFS); | |
96303081 | 1928 | if (!info) { |
34d52cb6 | 1929 | spin_lock(&ctl->tree_lock); |
96303081 JB |
1930 | ret = -ENOMEM; |
1931 | goto out; | |
1932 | } | |
1933 | } | |
1934 | ||
1935 | /* allocate the bitmap */ | |
1936 | info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS); | |
34d52cb6 | 1937 | spin_lock(&ctl->tree_lock); |
96303081 JB |
1938 | if (!info->bitmap) { |
1939 | ret = -ENOMEM; | |
1940 | goto out; | |
1941 | } | |
1942 | goto again; | |
1943 | } | |
1944 | ||
1945 | out: | |
1946 | if (info) { | |
1947 | if (info->bitmap) | |
1948 | kfree(info->bitmap); | |
dc89e982 | 1949 | kmem_cache_free(btrfs_free_space_cachep, info); |
96303081 | 1950 | } |
0f9dd46c JB |
1951 | |
1952 | return ret; | |
1953 | } | |
1954 | ||
945d8962 | 1955 | static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl, |
f333adb5 | 1956 | struct btrfs_free_space *info, bool update_stat) |
0f9dd46c | 1957 | { |
120d66ee LZ |
1958 | struct btrfs_free_space *left_info; |
1959 | struct btrfs_free_space *right_info; | |
1960 | bool merged = false; | |
1961 | u64 offset = info->offset; | |
1962 | u64 bytes = info->bytes; | |
6226cb0a | 1963 | |
0f9dd46c JB |
1964 | /* |
1965 | * first we want to see if there is free space adjacent to the range we | |
1966 | * are adding, if there is remove that struct and add a new one to | |
1967 | * cover the entire range | |
1968 | */ | |
34d52cb6 | 1969 | right_info = tree_search_offset(ctl, offset + bytes, 0, 0); |
96303081 JB |
1970 | if (right_info && rb_prev(&right_info->offset_index)) |
1971 | left_info = rb_entry(rb_prev(&right_info->offset_index), | |
1972 | struct btrfs_free_space, offset_index); | |
1973 | else | |
34d52cb6 | 1974 | left_info = tree_search_offset(ctl, offset - 1, 0, 0); |
0f9dd46c | 1975 | |
96303081 | 1976 | if (right_info && !right_info->bitmap) { |
f333adb5 | 1977 | if (update_stat) |
34d52cb6 | 1978 | unlink_free_space(ctl, right_info); |
f333adb5 | 1979 | else |
34d52cb6 | 1980 | __unlink_free_space(ctl, right_info); |
6226cb0a | 1981 | info->bytes += right_info->bytes; |
dc89e982 | 1982 | kmem_cache_free(btrfs_free_space_cachep, right_info); |
120d66ee | 1983 | merged = true; |
0f9dd46c JB |
1984 | } |
1985 | ||
96303081 JB |
1986 | if (left_info && !left_info->bitmap && |
1987 | left_info->offset + left_info->bytes == offset) { | |
f333adb5 | 1988 | if (update_stat) |
34d52cb6 | 1989 | unlink_free_space(ctl, left_info); |
f333adb5 | 1990 | else |
34d52cb6 | 1991 | __unlink_free_space(ctl, left_info); |
6226cb0a JB |
1992 | info->offset = left_info->offset; |
1993 | info->bytes += left_info->bytes; | |
dc89e982 | 1994 | kmem_cache_free(btrfs_free_space_cachep, left_info); |
120d66ee | 1995 | merged = true; |
0f9dd46c JB |
1996 | } |
1997 | ||
120d66ee LZ |
1998 | return merged; |
1999 | } | |
2000 | ||
20005523 FM |
2001 | static bool steal_from_bitmap_to_end(struct btrfs_free_space_ctl *ctl, |
2002 | struct btrfs_free_space *info, | |
2003 | bool update_stat) | |
2004 | { | |
2005 | struct btrfs_free_space *bitmap; | |
2006 | unsigned long i; | |
2007 | unsigned long j; | |
2008 | const u64 end = info->offset + info->bytes; | |
2009 | const u64 bitmap_offset = offset_to_bitmap(ctl, end); | |
2010 | u64 bytes; | |
2011 | ||
2012 | bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0); | |
2013 | if (!bitmap) | |
2014 | return false; | |
2015 | ||
2016 | i = offset_to_bit(bitmap->offset, ctl->unit, end); | |
2017 | j = find_next_zero_bit(bitmap->bitmap, BITS_PER_BITMAP, i); | |
2018 | if (j == i) | |
2019 | return false; | |
2020 | bytes = (j - i) * ctl->unit; | |
2021 | info->bytes += bytes; | |
2022 | ||
2023 | if (update_stat) | |
2024 | bitmap_clear_bits(ctl, bitmap, end, bytes); | |
2025 | else | |
2026 | __bitmap_clear_bits(ctl, bitmap, end, bytes); | |
2027 | ||
2028 | if (!bitmap->bytes) | |
2029 | free_bitmap(ctl, bitmap); | |
2030 | ||
2031 | return true; | |
2032 | } | |
2033 | ||
2034 | static bool steal_from_bitmap_to_front(struct btrfs_free_space_ctl *ctl, | |
2035 | struct btrfs_free_space *info, | |
2036 | bool update_stat) | |
2037 | { | |
2038 | struct btrfs_free_space *bitmap; | |
2039 | u64 bitmap_offset; | |
2040 | unsigned long i; | |
2041 | unsigned long j; | |
2042 | unsigned long prev_j; | |
2043 | u64 bytes; | |
2044 | ||
2045 | bitmap_offset = offset_to_bitmap(ctl, info->offset); | |
2046 | /* If we're on a boundary, try the previous logical bitmap. */ | |
2047 | if (bitmap_offset == info->offset) { | |
2048 | if (info->offset == 0) | |
2049 | return false; | |
2050 | bitmap_offset = offset_to_bitmap(ctl, info->offset - 1); | |
2051 | } | |
2052 | ||
2053 | bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0); | |
2054 | if (!bitmap) | |
2055 | return false; | |
2056 | ||
2057 | i = offset_to_bit(bitmap->offset, ctl->unit, info->offset) - 1; | |
2058 | j = 0; | |
2059 | prev_j = (unsigned long)-1; | |
2060 | for_each_clear_bit_from(j, bitmap->bitmap, BITS_PER_BITMAP) { | |
2061 | if (j > i) | |
2062 | break; | |
2063 | prev_j = j; | |
2064 | } | |
2065 | if (prev_j == i) | |
2066 | return false; | |
2067 | ||
2068 | if (prev_j == (unsigned long)-1) | |
2069 | bytes = (i + 1) * ctl->unit; | |
2070 | else | |
2071 | bytes = (i - prev_j) * ctl->unit; | |
2072 | ||
2073 | info->offset -= bytes; | |
2074 | info->bytes += bytes; | |
2075 | ||
2076 | if (update_stat) | |
2077 | bitmap_clear_bits(ctl, bitmap, info->offset, bytes); | |
2078 | else | |
2079 | __bitmap_clear_bits(ctl, bitmap, info->offset, bytes); | |
2080 | ||
2081 | if (!bitmap->bytes) | |
2082 | free_bitmap(ctl, bitmap); | |
2083 | ||
2084 | return true; | |
2085 | } | |
2086 | ||
2087 | /* | |
2088 | * We prefer always to allocate from extent entries, both for clustered and | |
2089 | * non-clustered allocation requests. So when attempting to add a new extent | |
2090 | * entry, try to see if there's adjacent free space in bitmap entries, and if | |
2091 | * there is, migrate that space from the bitmaps to the extent. | |
2092 | * Like this we get better chances of satisfying space allocation requests | |
2093 | * because we attempt to satisfy them based on a single cache entry, and never | |
2094 | * on 2 or more entries - even if the entries represent a contiguous free space | |
2095 | * region (e.g. 1 extent entry + 1 bitmap entry starting where the extent entry | |
2096 | * ends). | |
2097 | */ | |
2098 | static void steal_from_bitmap(struct btrfs_free_space_ctl *ctl, | |
2099 | struct btrfs_free_space *info, | |
2100 | bool update_stat) | |
2101 | { | |
2102 | /* | |
2103 | * Only work with disconnected entries, as we can change their offset, | |
2104 | * and must be extent entries. | |
2105 | */ | |
2106 | ASSERT(!info->bitmap); | |
2107 | ASSERT(RB_EMPTY_NODE(&info->offset_index)); | |
2108 | ||
2109 | if (ctl->total_bitmaps > 0) { | |
2110 | bool stole_end; | |
2111 | bool stole_front = false; | |
2112 | ||
2113 | stole_end = steal_from_bitmap_to_end(ctl, info, update_stat); | |
2114 | if (ctl->total_bitmaps > 0) | |
2115 | stole_front = steal_from_bitmap_to_front(ctl, info, | |
2116 | update_stat); | |
2117 | ||
2118 | if (stole_end || stole_front) | |
2119 | try_merge_free_space(ctl, info, update_stat); | |
2120 | } | |
2121 | } | |
2122 | ||
581bb050 LZ |
2123 | int __btrfs_add_free_space(struct btrfs_free_space_ctl *ctl, |
2124 | u64 offset, u64 bytes) | |
120d66ee LZ |
2125 | { |
2126 | struct btrfs_free_space *info; | |
2127 | int ret = 0; | |
2128 | ||
dc89e982 | 2129 | info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS); |
120d66ee LZ |
2130 | if (!info) |
2131 | return -ENOMEM; | |
2132 | ||
2133 | info->offset = offset; | |
2134 | info->bytes = bytes; | |
20005523 | 2135 | RB_CLEAR_NODE(&info->offset_index); |
120d66ee | 2136 | |
34d52cb6 | 2137 | spin_lock(&ctl->tree_lock); |
120d66ee | 2138 | |
34d52cb6 | 2139 | if (try_merge_free_space(ctl, info, true)) |
120d66ee LZ |
2140 | goto link; |
2141 | ||
2142 | /* | |
2143 | * There was no extent directly to the left or right of this new | |
2144 | * extent then we know we're going to have to allocate a new extent, so | |
2145 | * before we do that see if we need to drop this into a bitmap | |
2146 | */ | |
34d52cb6 | 2147 | ret = insert_into_bitmap(ctl, info); |
120d66ee LZ |
2148 | if (ret < 0) { |
2149 | goto out; | |
2150 | } else if (ret) { | |
2151 | ret = 0; | |
2152 | goto out; | |
2153 | } | |
2154 | link: | |
20005523 FM |
2155 | /* |
2156 | * Only steal free space from adjacent bitmaps if we're sure we're not | |
2157 | * going to add the new free space to existing bitmap entries - because | |
2158 | * that would mean unnecessary work that would be reverted. Therefore | |
2159 | * attempt to steal space from bitmaps if we're adding an extent entry. | |
2160 | */ | |
2161 | steal_from_bitmap(ctl, info, true); | |
2162 | ||
34d52cb6 | 2163 | ret = link_free_space(ctl, info); |
0f9dd46c | 2164 | if (ret) |
dc89e982 | 2165 | kmem_cache_free(btrfs_free_space_cachep, info); |
96303081 | 2166 | out: |
34d52cb6 | 2167 | spin_unlock(&ctl->tree_lock); |
6226cb0a | 2168 | |
0f9dd46c | 2169 | if (ret) { |
efe120a0 | 2170 | printk(KERN_CRIT "BTRFS: unable to add free space :%d\n", ret); |
b12d6869 | 2171 | ASSERT(ret != -EEXIST); |
0f9dd46c JB |
2172 | } |
2173 | ||
0f9dd46c JB |
2174 | return ret; |
2175 | } | |
2176 | ||
6226cb0a JB |
2177 | int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group, |
2178 | u64 offset, u64 bytes) | |
0f9dd46c | 2179 | { |
34d52cb6 | 2180 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0f9dd46c | 2181 | struct btrfs_free_space *info; |
b0175117 JB |
2182 | int ret; |
2183 | bool re_search = false; | |
0f9dd46c | 2184 | |
34d52cb6 | 2185 | spin_lock(&ctl->tree_lock); |
6226cb0a | 2186 | |
96303081 | 2187 | again: |
b0175117 | 2188 | ret = 0; |
bdb7d303 JB |
2189 | if (!bytes) |
2190 | goto out_lock; | |
2191 | ||
34d52cb6 | 2192 | info = tree_search_offset(ctl, offset, 0, 0); |
96303081 | 2193 | if (!info) { |
6606bb97 JB |
2194 | /* |
2195 | * oops didn't find an extent that matched the space we wanted | |
2196 | * to remove, look for a bitmap instead | |
2197 | */ | |
34d52cb6 | 2198 | info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), |
6606bb97 JB |
2199 | 1, 0); |
2200 | if (!info) { | |
b0175117 JB |
2201 | /* |
2202 | * If we found a partial bit of our free space in a | |
2203 | * bitmap but then couldn't find the other part this may | |
2204 | * be a problem, so WARN about it. | |
24a70313 | 2205 | */ |
b0175117 | 2206 | WARN_ON(re_search); |
6606bb97 JB |
2207 | goto out_lock; |
2208 | } | |
96303081 JB |
2209 | } |
2210 | ||
b0175117 | 2211 | re_search = false; |
bdb7d303 | 2212 | if (!info->bitmap) { |
34d52cb6 | 2213 | unlink_free_space(ctl, info); |
bdb7d303 JB |
2214 | if (offset == info->offset) { |
2215 | u64 to_free = min(bytes, info->bytes); | |
2216 | ||
2217 | info->bytes -= to_free; | |
2218 | info->offset += to_free; | |
2219 | if (info->bytes) { | |
2220 | ret = link_free_space(ctl, info); | |
2221 | WARN_ON(ret); | |
2222 | } else { | |
2223 | kmem_cache_free(btrfs_free_space_cachep, info); | |
2224 | } | |
0f9dd46c | 2225 | |
bdb7d303 JB |
2226 | offset += to_free; |
2227 | bytes -= to_free; | |
2228 | goto again; | |
2229 | } else { | |
2230 | u64 old_end = info->bytes + info->offset; | |
9b49c9b9 | 2231 | |
bdb7d303 | 2232 | info->bytes = offset - info->offset; |
34d52cb6 | 2233 | ret = link_free_space(ctl, info); |
96303081 JB |
2234 | WARN_ON(ret); |
2235 | if (ret) | |
2236 | goto out_lock; | |
96303081 | 2237 | |
bdb7d303 JB |
2238 | /* Not enough bytes in this entry to satisfy us */ |
2239 | if (old_end < offset + bytes) { | |
2240 | bytes -= old_end - offset; | |
2241 | offset = old_end; | |
2242 | goto again; | |
2243 | } else if (old_end == offset + bytes) { | |
2244 | /* all done */ | |
2245 | goto out_lock; | |
2246 | } | |
2247 | spin_unlock(&ctl->tree_lock); | |
2248 | ||
2249 | ret = btrfs_add_free_space(block_group, offset + bytes, | |
2250 | old_end - (offset + bytes)); | |
2251 | WARN_ON(ret); | |
2252 | goto out; | |
2253 | } | |
0f9dd46c | 2254 | } |
96303081 | 2255 | |
34d52cb6 | 2256 | ret = remove_from_bitmap(ctl, info, &offset, &bytes); |
b0175117 JB |
2257 | if (ret == -EAGAIN) { |
2258 | re_search = true; | |
96303081 | 2259 | goto again; |
b0175117 | 2260 | } |
96303081 | 2261 | out_lock: |
34d52cb6 | 2262 | spin_unlock(&ctl->tree_lock); |
0f9dd46c | 2263 | out: |
25179201 JB |
2264 | return ret; |
2265 | } | |
2266 | ||
0f9dd46c JB |
2267 | void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group, |
2268 | u64 bytes) | |
2269 | { | |
34d52cb6 | 2270 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0f9dd46c JB |
2271 | struct btrfs_free_space *info; |
2272 | struct rb_node *n; | |
2273 | int count = 0; | |
2274 | ||
34d52cb6 | 2275 | for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) { |
0f9dd46c | 2276 | info = rb_entry(n, struct btrfs_free_space, offset_index); |
f6175efa | 2277 | if (info->bytes >= bytes && !block_group->ro) |
0f9dd46c | 2278 | count++; |
efe120a0 FH |
2279 | btrfs_crit(block_group->fs_info, |
2280 | "entry offset %llu, bytes %llu, bitmap %s", | |
2281 | info->offset, info->bytes, | |
96303081 | 2282 | (info->bitmap) ? "yes" : "no"); |
0f9dd46c | 2283 | } |
efe120a0 | 2284 | btrfs_info(block_group->fs_info, "block group has cluster?: %s", |
96303081 | 2285 | list_empty(&block_group->cluster_list) ? "no" : "yes"); |
efe120a0 FH |
2286 | btrfs_info(block_group->fs_info, |
2287 | "%d blocks of free space at or bigger than bytes is", count); | |
0f9dd46c JB |
2288 | } |
2289 | ||
34d52cb6 | 2290 | void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group) |
0f9dd46c | 2291 | { |
34d52cb6 | 2292 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
0f9dd46c | 2293 | |
34d52cb6 LZ |
2294 | spin_lock_init(&ctl->tree_lock); |
2295 | ctl->unit = block_group->sectorsize; | |
2296 | ctl->start = block_group->key.objectid; | |
2297 | ctl->private = block_group; | |
2298 | ctl->op = &free_space_op; | |
0f9dd46c | 2299 | |
34d52cb6 LZ |
2300 | /* |
2301 | * we only want to have 32k of ram per block group for keeping | |
2302 | * track of free space, and if we pass 1/2 of that we want to | |
2303 | * start converting things over to using bitmaps | |
2304 | */ | |
2305 | ctl->extents_thresh = ((1024 * 32) / 2) / | |
2306 | sizeof(struct btrfs_free_space); | |
0f9dd46c JB |
2307 | } |
2308 | ||
fa9c0d79 CM |
2309 | /* |
2310 | * for a given cluster, put all of its extents back into the free | |
2311 | * space cache. If the block group passed doesn't match the block group | |
2312 | * pointed to by the cluster, someone else raced in and freed the | |
2313 | * cluster already. In that case, we just return without changing anything | |
2314 | */ | |
2315 | static int | |
2316 | __btrfs_return_cluster_to_free_space( | |
2317 | struct btrfs_block_group_cache *block_group, | |
2318 | struct btrfs_free_cluster *cluster) | |
2319 | { | |
34d52cb6 | 2320 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
fa9c0d79 CM |
2321 | struct btrfs_free_space *entry; |
2322 | struct rb_node *node; | |
2323 | ||
2324 | spin_lock(&cluster->lock); | |
2325 | if (cluster->block_group != block_group) | |
2326 | goto out; | |
2327 | ||
96303081 | 2328 | cluster->block_group = NULL; |
fa9c0d79 | 2329 | cluster->window_start = 0; |
96303081 | 2330 | list_del_init(&cluster->block_group_list); |
96303081 | 2331 | |
fa9c0d79 | 2332 | node = rb_first(&cluster->root); |
96303081 | 2333 | while (node) { |
4e69b598 JB |
2334 | bool bitmap; |
2335 | ||
fa9c0d79 CM |
2336 | entry = rb_entry(node, struct btrfs_free_space, offset_index); |
2337 | node = rb_next(&entry->offset_index); | |
2338 | rb_erase(&entry->offset_index, &cluster->root); | |
20005523 | 2339 | RB_CLEAR_NODE(&entry->offset_index); |
4e69b598 JB |
2340 | |
2341 | bitmap = (entry->bitmap != NULL); | |
20005523 | 2342 | if (!bitmap) { |
34d52cb6 | 2343 | try_merge_free_space(ctl, entry, false); |
20005523 FM |
2344 | steal_from_bitmap(ctl, entry, false); |
2345 | } | |
34d52cb6 | 2346 | tree_insert_offset(&ctl->free_space_offset, |
4e69b598 | 2347 | entry->offset, &entry->offset_index, bitmap); |
fa9c0d79 | 2348 | } |
6bef4d31 | 2349 | cluster->root = RB_ROOT; |
96303081 | 2350 | |
fa9c0d79 CM |
2351 | out: |
2352 | spin_unlock(&cluster->lock); | |
96303081 | 2353 | btrfs_put_block_group(block_group); |
fa9c0d79 CM |
2354 | return 0; |
2355 | } | |
2356 | ||
48a3b636 ES |
2357 | static void __btrfs_remove_free_space_cache_locked( |
2358 | struct btrfs_free_space_ctl *ctl) | |
0f9dd46c JB |
2359 | { |
2360 | struct btrfs_free_space *info; | |
2361 | struct rb_node *node; | |
581bb050 | 2362 | |
581bb050 LZ |
2363 | while ((node = rb_last(&ctl->free_space_offset)) != NULL) { |
2364 | info = rb_entry(node, struct btrfs_free_space, offset_index); | |
9b90f513 JB |
2365 | if (!info->bitmap) { |
2366 | unlink_free_space(ctl, info); | |
2367 | kmem_cache_free(btrfs_free_space_cachep, info); | |
2368 | } else { | |
2369 | free_bitmap(ctl, info); | |
2370 | } | |
581bb050 LZ |
2371 | if (need_resched()) { |
2372 | spin_unlock(&ctl->tree_lock); | |
2373 | cond_resched(); | |
2374 | spin_lock(&ctl->tree_lock); | |
2375 | } | |
2376 | } | |
09655373 CM |
2377 | } |
2378 | ||
2379 | void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl) | |
2380 | { | |
2381 | spin_lock(&ctl->tree_lock); | |
2382 | __btrfs_remove_free_space_cache_locked(ctl); | |
581bb050 LZ |
2383 | spin_unlock(&ctl->tree_lock); |
2384 | } | |
2385 | ||
2386 | void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group) | |
2387 | { | |
2388 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
fa9c0d79 | 2389 | struct btrfs_free_cluster *cluster; |
96303081 | 2390 | struct list_head *head; |
0f9dd46c | 2391 | |
34d52cb6 | 2392 | spin_lock(&ctl->tree_lock); |
96303081 JB |
2393 | while ((head = block_group->cluster_list.next) != |
2394 | &block_group->cluster_list) { | |
2395 | cluster = list_entry(head, struct btrfs_free_cluster, | |
2396 | block_group_list); | |
fa9c0d79 CM |
2397 | |
2398 | WARN_ON(cluster->block_group != block_group); | |
2399 | __btrfs_return_cluster_to_free_space(block_group, cluster); | |
96303081 | 2400 | if (need_resched()) { |
34d52cb6 | 2401 | spin_unlock(&ctl->tree_lock); |
96303081 | 2402 | cond_resched(); |
34d52cb6 | 2403 | spin_lock(&ctl->tree_lock); |
96303081 | 2404 | } |
fa9c0d79 | 2405 | } |
09655373 | 2406 | __btrfs_remove_free_space_cache_locked(ctl); |
34d52cb6 | 2407 | spin_unlock(&ctl->tree_lock); |
fa9c0d79 | 2408 | |
0f9dd46c JB |
2409 | } |
2410 | ||
6226cb0a | 2411 | u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group, |
a4820398 MX |
2412 | u64 offset, u64 bytes, u64 empty_size, |
2413 | u64 *max_extent_size) | |
0f9dd46c | 2414 | { |
34d52cb6 | 2415 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
6226cb0a | 2416 | struct btrfs_free_space *entry = NULL; |
96303081 | 2417 | u64 bytes_search = bytes + empty_size; |
6226cb0a | 2418 | u64 ret = 0; |
53b381b3 DW |
2419 | u64 align_gap = 0; |
2420 | u64 align_gap_len = 0; | |
0f9dd46c | 2421 | |
34d52cb6 | 2422 | spin_lock(&ctl->tree_lock); |
53b381b3 | 2423 | entry = find_free_space(ctl, &offset, &bytes_search, |
a4820398 | 2424 | block_group->full_stripe_len, max_extent_size); |
6226cb0a | 2425 | if (!entry) |
96303081 JB |
2426 | goto out; |
2427 | ||
2428 | ret = offset; | |
2429 | if (entry->bitmap) { | |
34d52cb6 | 2430 | bitmap_clear_bits(ctl, entry, offset, bytes); |
edf6e2d1 | 2431 | if (!entry->bytes) |
34d52cb6 | 2432 | free_bitmap(ctl, entry); |
96303081 | 2433 | } else { |
34d52cb6 | 2434 | unlink_free_space(ctl, entry); |
53b381b3 DW |
2435 | align_gap_len = offset - entry->offset; |
2436 | align_gap = entry->offset; | |
2437 | ||
2438 | entry->offset = offset + bytes; | |
2439 | WARN_ON(entry->bytes < bytes + align_gap_len); | |
2440 | ||
2441 | entry->bytes -= bytes + align_gap_len; | |
6226cb0a | 2442 | if (!entry->bytes) |
dc89e982 | 2443 | kmem_cache_free(btrfs_free_space_cachep, entry); |
6226cb0a | 2444 | else |
34d52cb6 | 2445 | link_free_space(ctl, entry); |
6226cb0a | 2446 | } |
96303081 | 2447 | out: |
34d52cb6 | 2448 | spin_unlock(&ctl->tree_lock); |
817d52f8 | 2449 | |
53b381b3 DW |
2450 | if (align_gap_len) |
2451 | __btrfs_add_free_space(ctl, align_gap, align_gap_len); | |
0f9dd46c JB |
2452 | return ret; |
2453 | } | |
fa9c0d79 CM |
2454 | |
2455 | /* | |
2456 | * given a cluster, put all of its extents back into the free space | |
2457 | * cache. If a block group is passed, this function will only free | |
2458 | * a cluster that belongs to the passed block group. | |
2459 | * | |
2460 | * Otherwise, it'll get a reference on the block group pointed to by the | |
2461 | * cluster and remove the cluster from it. | |
2462 | */ | |
2463 | int btrfs_return_cluster_to_free_space( | |
2464 | struct btrfs_block_group_cache *block_group, | |
2465 | struct btrfs_free_cluster *cluster) | |
2466 | { | |
34d52cb6 | 2467 | struct btrfs_free_space_ctl *ctl; |
fa9c0d79 CM |
2468 | int ret; |
2469 | ||
2470 | /* first, get a safe pointer to the block group */ | |
2471 | spin_lock(&cluster->lock); | |
2472 | if (!block_group) { | |
2473 | block_group = cluster->block_group; | |
2474 | if (!block_group) { | |
2475 | spin_unlock(&cluster->lock); | |
2476 | return 0; | |
2477 | } | |
2478 | } else if (cluster->block_group != block_group) { | |
2479 | /* someone else has already freed it don't redo their work */ | |
2480 | spin_unlock(&cluster->lock); | |
2481 | return 0; | |
2482 | } | |
2483 | atomic_inc(&block_group->count); | |
2484 | spin_unlock(&cluster->lock); | |
2485 | ||
34d52cb6 LZ |
2486 | ctl = block_group->free_space_ctl; |
2487 | ||
fa9c0d79 | 2488 | /* now return any extents the cluster had on it */ |
34d52cb6 | 2489 | spin_lock(&ctl->tree_lock); |
fa9c0d79 | 2490 | ret = __btrfs_return_cluster_to_free_space(block_group, cluster); |
34d52cb6 | 2491 | spin_unlock(&ctl->tree_lock); |
fa9c0d79 CM |
2492 | |
2493 | /* finally drop our ref */ | |
2494 | btrfs_put_block_group(block_group); | |
2495 | return ret; | |
2496 | } | |
2497 | ||
96303081 JB |
2498 | static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group, |
2499 | struct btrfs_free_cluster *cluster, | |
4e69b598 | 2500 | struct btrfs_free_space *entry, |
a4820398 MX |
2501 | u64 bytes, u64 min_start, |
2502 | u64 *max_extent_size) | |
96303081 | 2503 | { |
34d52cb6 | 2504 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
96303081 JB |
2505 | int err; |
2506 | u64 search_start = cluster->window_start; | |
2507 | u64 search_bytes = bytes; | |
2508 | u64 ret = 0; | |
2509 | ||
96303081 JB |
2510 | search_start = min_start; |
2511 | search_bytes = bytes; | |
2512 | ||
34d52cb6 | 2513 | err = search_bitmap(ctl, entry, &search_start, &search_bytes); |
a4820398 MX |
2514 | if (err) { |
2515 | if (search_bytes > *max_extent_size) | |
2516 | *max_extent_size = search_bytes; | |
4e69b598 | 2517 | return 0; |
a4820398 | 2518 | } |
96303081 JB |
2519 | |
2520 | ret = search_start; | |
bb3ac5a4 | 2521 | __bitmap_clear_bits(ctl, entry, ret, bytes); |
96303081 JB |
2522 | |
2523 | return ret; | |
2524 | } | |
2525 | ||
fa9c0d79 CM |
2526 | /* |
2527 | * given a cluster, try to allocate 'bytes' from it, returns 0 | |
2528 | * if it couldn't find anything suitably large, or a logical disk offset | |
2529 | * if things worked out | |
2530 | */ | |
2531 | u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group, | |
2532 | struct btrfs_free_cluster *cluster, u64 bytes, | |
a4820398 | 2533 | u64 min_start, u64 *max_extent_size) |
fa9c0d79 | 2534 | { |
34d52cb6 | 2535 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
fa9c0d79 CM |
2536 | struct btrfs_free_space *entry = NULL; |
2537 | struct rb_node *node; | |
2538 | u64 ret = 0; | |
2539 | ||
2540 | spin_lock(&cluster->lock); | |
2541 | if (bytes > cluster->max_size) | |
2542 | goto out; | |
2543 | ||
2544 | if (cluster->block_group != block_group) | |
2545 | goto out; | |
2546 | ||
2547 | node = rb_first(&cluster->root); | |
2548 | if (!node) | |
2549 | goto out; | |
2550 | ||
2551 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
67871254 | 2552 | while (1) { |
a4820398 MX |
2553 | if (entry->bytes < bytes && entry->bytes > *max_extent_size) |
2554 | *max_extent_size = entry->bytes; | |
2555 | ||
4e69b598 JB |
2556 | if (entry->bytes < bytes || |
2557 | (!entry->bitmap && entry->offset < min_start)) { | |
fa9c0d79 CM |
2558 | node = rb_next(&entry->offset_index); |
2559 | if (!node) | |
2560 | break; | |
2561 | entry = rb_entry(node, struct btrfs_free_space, | |
2562 | offset_index); | |
2563 | continue; | |
2564 | } | |
fa9c0d79 | 2565 | |
4e69b598 JB |
2566 | if (entry->bitmap) { |
2567 | ret = btrfs_alloc_from_bitmap(block_group, | |
2568 | cluster, entry, bytes, | |
a4820398 MX |
2569 | cluster->window_start, |
2570 | max_extent_size); | |
4e69b598 | 2571 | if (ret == 0) { |
4e69b598 JB |
2572 | node = rb_next(&entry->offset_index); |
2573 | if (!node) | |
2574 | break; | |
2575 | entry = rb_entry(node, struct btrfs_free_space, | |
2576 | offset_index); | |
2577 | continue; | |
2578 | } | |
9b230628 | 2579 | cluster->window_start += bytes; |
4e69b598 | 2580 | } else { |
4e69b598 JB |
2581 | ret = entry->offset; |
2582 | ||
2583 | entry->offset += bytes; | |
2584 | entry->bytes -= bytes; | |
2585 | } | |
fa9c0d79 | 2586 | |
5e71b5d5 | 2587 | if (entry->bytes == 0) |
fa9c0d79 | 2588 | rb_erase(&entry->offset_index, &cluster->root); |
fa9c0d79 CM |
2589 | break; |
2590 | } | |
2591 | out: | |
2592 | spin_unlock(&cluster->lock); | |
96303081 | 2593 | |
5e71b5d5 LZ |
2594 | if (!ret) |
2595 | return 0; | |
2596 | ||
34d52cb6 | 2597 | spin_lock(&ctl->tree_lock); |
5e71b5d5 | 2598 | |
34d52cb6 | 2599 | ctl->free_space -= bytes; |
5e71b5d5 | 2600 | if (entry->bytes == 0) { |
34d52cb6 | 2601 | ctl->free_extents--; |
4e69b598 JB |
2602 | if (entry->bitmap) { |
2603 | kfree(entry->bitmap); | |
34d52cb6 LZ |
2604 | ctl->total_bitmaps--; |
2605 | ctl->op->recalc_thresholds(ctl); | |
4e69b598 | 2606 | } |
dc89e982 | 2607 | kmem_cache_free(btrfs_free_space_cachep, entry); |
5e71b5d5 LZ |
2608 | } |
2609 | ||
34d52cb6 | 2610 | spin_unlock(&ctl->tree_lock); |
5e71b5d5 | 2611 | |
fa9c0d79 CM |
2612 | return ret; |
2613 | } | |
2614 | ||
96303081 JB |
2615 | static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group, |
2616 | struct btrfs_free_space *entry, | |
2617 | struct btrfs_free_cluster *cluster, | |
1bb91902 AO |
2618 | u64 offset, u64 bytes, |
2619 | u64 cont1_bytes, u64 min_bytes) | |
96303081 | 2620 | { |
34d52cb6 | 2621 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
96303081 JB |
2622 | unsigned long next_zero; |
2623 | unsigned long i; | |
1bb91902 AO |
2624 | unsigned long want_bits; |
2625 | unsigned long min_bits; | |
96303081 JB |
2626 | unsigned long found_bits; |
2627 | unsigned long start = 0; | |
2628 | unsigned long total_found = 0; | |
4e69b598 | 2629 | int ret; |
96303081 | 2630 | |
96009762 | 2631 | i = offset_to_bit(entry->offset, ctl->unit, |
96303081 | 2632 | max_t(u64, offset, entry->offset)); |
96009762 WSH |
2633 | want_bits = bytes_to_bits(bytes, ctl->unit); |
2634 | min_bits = bytes_to_bits(min_bytes, ctl->unit); | |
96303081 JB |
2635 | |
2636 | again: | |
2637 | found_bits = 0; | |
ebb3dad4 | 2638 | for_each_set_bit_from(i, entry->bitmap, BITS_PER_BITMAP) { |
96303081 JB |
2639 | next_zero = find_next_zero_bit(entry->bitmap, |
2640 | BITS_PER_BITMAP, i); | |
1bb91902 | 2641 | if (next_zero - i >= min_bits) { |
96303081 JB |
2642 | found_bits = next_zero - i; |
2643 | break; | |
2644 | } | |
2645 | i = next_zero; | |
2646 | } | |
2647 | ||
2648 | if (!found_bits) | |
4e69b598 | 2649 | return -ENOSPC; |
96303081 | 2650 | |
1bb91902 | 2651 | if (!total_found) { |
96303081 | 2652 | start = i; |
b78d09bc | 2653 | cluster->max_size = 0; |
96303081 JB |
2654 | } |
2655 | ||
2656 | total_found += found_bits; | |
2657 | ||
96009762 WSH |
2658 | if (cluster->max_size < found_bits * ctl->unit) |
2659 | cluster->max_size = found_bits * ctl->unit; | |
96303081 | 2660 | |
1bb91902 AO |
2661 | if (total_found < want_bits || cluster->max_size < cont1_bytes) { |
2662 | i = next_zero + 1; | |
96303081 JB |
2663 | goto again; |
2664 | } | |
2665 | ||
96009762 | 2666 | cluster->window_start = start * ctl->unit + entry->offset; |
34d52cb6 | 2667 | rb_erase(&entry->offset_index, &ctl->free_space_offset); |
4e69b598 JB |
2668 | ret = tree_insert_offset(&cluster->root, entry->offset, |
2669 | &entry->offset_index, 1); | |
b12d6869 | 2670 | ASSERT(!ret); /* -EEXIST; Logic error */ |
96303081 | 2671 | |
3f7de037 | 2672 | trace_btrfs_setup_cluster(block_group, cluster, |
96009762 | 2673 | total_found * ctl->unit, 1); |
96303081 JB |
2674 | return 0; |
2675 | } | |
2676 | ||
4e69b598 JB |
2677 | /* |
2678 | * This searches the block group for just extents to fill the cluster with. | |
1bb91902 AO |
2679 | * Try to find a cluster with at least bytes total bytes, at least one |
2680 | * extent of cont1_bytes, and other clusters of at least min_bytes. | |
4e69b598 | 2681 | */ |
3de85bb9 JB |
2682 | static noinline int |
2683 | setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group, | |
2684 | struct btrfs_free_cluster *cluster, | |
2685 | struct list_head *bitmaps, u64 offset, u64 bytes, | |
1bb91902 | 2686 | u64 cont1_bytes, u64 min_bytes) |
4e69b598 | 2687 | { |
34d52cb6 | 2688 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
4e69b598 JB |
2689 | struct btrfs_free_space *first = NULL; |
2690 | struct btrfs_free_space *entry = NULL; | |
4e69b598 JB |
2691 | struct btrfs_free_space *last; |
2692 | struct rb_node *node; | |
4e69b598 JB |
2693 | u64 window_free; |
2694 | u64 max_extent; | |
3f7de037 | 2695 | u64 total_size = 0; |
4e69b598 | 2696 | |
34d52cb6 | 2697 | entry = tree_search_offset(ctl, offset, 0, 1); |
4e69b598 JB |
2698 | if (!entry) |
2699 | return -ENOSPC; | |
2700 | ||
2701 | /* | |
2702 | * We don't want bitmaps, so just move along until we find a normal | |
2703 | * extent entry. | |
2704 | */ | |
1bb91902 AO |
2705 | while (entry->bitmap || entry->bytes < min_bytes) { |
2706 | if (entry->bitmap && list_empty(&entry->list)) | |
86d4a77b | 2707 | list_add_tail(&entry->list, bitmaps); |
4e69b598 JB |
2708 | node = rb_next(&entry->offset_index); |
2709 | if (!node) | |
2710 | return -ENOSPC; | |
2711 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
2712 | } | |
2713 | ||
4e69b598 JB |
2714 | window_free = entry->bytes; |
2715 | max_extent = entry->bytes; | |
2716 | first = entry; | |
2717 | last = entry; | |
4e69b598 | 2718 | |
1bb91902 AO |
2719 | for (node = rb_next(&entry->offset_index); node; |
2720 | node = rb_next(&entry->offset_index)) { | |
4e69b598 JB |
2721 | entry = rb_entry(node, struct btrfs_free_space, offset_index); |
2722 | ||
86d4a77b JB |
2723 | if (entry->bitmap) { |
2724 | if (list_empty(&entry->list)) | |
2725 | list_add_tail(&entry->list, bitmaps); | |
4e69b598 | 2726 | continue; |
86d4a77b JB |
2727 | } |
2728 | ||
1bb91902 AO |
2729 | if (entry->bytes < min_bytes) |
2730 | continue; | |
2731 | ||
2732 | last = entry; | |
2733 | window_free += entry->bytes; | |
2734 | if (entry->bytes > max_extent) | |
4e69b598 | 2735 | max_extent = entry->bytes; |
4e69b598 JB |
2736 | } |
2737 | ||
1bb91902 AO |
2738 | if (window_free < bytes || max_extent < cont1_bytes) |
2739 | return -ENOSPC; | |
2740 | ||
4e69b598 JB |
2741 | cluster->window_start = first->offset; |
2742 | ||
2743 | node = &first->offset_index; | |
2744 | ||
2745 | /* | |
2746 | * now we've found our entries, pull them out of the free space | |
2747 | * cache and put them into the cluster rbtree | |
2748 | */ | |
2749 | do { | |
2750 | int ret; | |
2751 | ||
2752 | entry = rb_entry(node, struct btrfs_free_space, offset_index); | |
2753 | node = rb_next(&entry->offset_index); | |
1bb91902 | 2754 | if (entry->bitmap || entry->bytes < min_bytes) |
4e69b598 JB |
2755 | continue; |
2756 | ||
34d52cb6 | 2757 | rb_erase(&entry->offset_index, &ctl->free_space_offset); |
4e69b598 JB |
2758 | ret = tree_insert_offset(&cluster->root, entry->offset, |
2759 | &entry->offset_index, 0); | |
3f7de037 | 2760 | total_size += entry->bytes; |
b12d6869 | 2761 | ASSERT(!ret); /* -EEXIST; Logic error */ |
4e69b598 JB |
2762 | } while (node && entry != last); |
2763 | ||
2764 | cluster->max_size = max_extent; | |
3f7de037 | 2765 | trace_btrfs_setup_cluster(block_group, cluster, total_size, 0); |
4e69b598 JB |
2766 | return 0; |
2767 | } | |
2768 | ||
2769 | /* | |
2770 | * This specifically looks for bitmaps that may work in the cluster, we assume | |
2771 | * that we have already failed to find extents that will work. | |
2772 | */ | |
3de85bb9 JB |
2773 | static noinline int |
2774 | setup_cluster_bitmap(struct btrfs_block_group_cache *block_group, | |
2775 | struct btrfs_free_cluster *cluster, | |
2776 | struct list_head *bitmaps, u64 offset, u64 bytes, | |
1bb91902 | 2777 | u64 cont1_bytes, u64 min_bytes) |
4e69b598 | 2778 | { |
34d52cb6 | 2779 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
4e69b598 | 2780 | struct btrfs_free_space *entry; |
4e69b598 | 2781 | int ret = -ENOSPC; |
0f0fbf1d | 2782 | u64 bitmap_offset = offset_to_bitmap(ctl, offset); |
4e69b598 | 2783 | |
34d52cb6 | 2784 | if (ctl->total_bitmaps == 0) |
4e69b598 JB |
2785 | return -ENOSPC; |
2786 | ||
0f0fbf1d LZ |
2787 | /* |
2788 | * The bitmap that covers offset won't be in the list unless offset | |
2789 | * is just its start offset. | |
2790 | */ | |
2791 | entry = list_first_entry(bitmaps, struct btrfs_free_space, list); | |
2792 | if (entry->offset != bitmap_offset) { | |
2793 | entry = tree_search_offset(ctl, bitmap_offset, 1, 0); | |
2794 | if (entry && list_empty(&entry->list)) | |
2795 | list_add(&entry->list, bitmaps); | |
2796 | } | |
2797 | ||
86d4a77b | 2798 | list_for_each_entry(entry, bitmaps, list) { |
357b9784 | 2799 | if (entry->bytes < bytes) |
86d4a77b JB |
2800 | continue; |
2801 | ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset, | |
1bb91902 | 2802 | bytes, cont1_bytes, min_bytes); |
86d4a77b JB |
2803 | if (!ret) |
2804 | return 0; | |
2805 | } | |
2806 | ||
2807 | /* | |
52621cb6 LZ |
2808 | * The bitmaps list has all the bitmaps that record free space |
2809 | * starting after offset, so no more search is required. | |
86d4a77b | 2810 | */ |
52621cb6 | 2811 | return -ENOSPC; |
4e69b598 JB |
2812 | } |
2813 | ||
fa9c0d79 CM |
2814 | /* |
2815 | * here we try to find a cluster of blocks in a block group. The goal | |
1bb91902 | 2816 | * is to find at least bytes+empty_size. |
fa9c0d79 CM |
2817 | * We might not find them all in one contiguous area. |
2818 | * | |
2819 | * returns zero and sets up cluster if things worked out, otherwise | |
2820 | * it returns -enospc | |
2821 | */ | |
00361589 | 2822 | int btrfs_find_space_cluster(struct btrfs_root *root, |
fa9c0d79 CM |
2823 | struct btrfs_block_group_cache *block_group, |
2824 | struct btrfs_free_cluster *cluster, | |
2825 | u64 offset, u64 bytes, u64 empty_size) | |
2826 | { | |
34d52cb6 | 2827 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; |
86d4a77b | 2828 | struct btrfs_free_space *entry, *tmp; |
52621cb6 | 2829 | LIST_HEAD(bitmaps); |
fa9c0d79 | 2830 | u64 min_bytes; |
1bb91902 | 2831 | u64 cont1_bytes; |
fa9c0d79 CM |
2832 | int ret; |
2833 | ||
1bb91902 AO |
2834 | /* |
2835 | * Choose the minimum extent size we'll require for this | |
2836 | * cluster. For SSD_SPREAD, don't allow any fragmentation. | |
2837 | * For metadata, allow allocates with smaller extents. For | |
2838 | * data, keep it dense. | |
2839 | */ | |
451d7585 | 2840 | if (btrfs_test_opt(root, SSD_SPREAD)) { |
1bb91902 | 2841 | cont1_bytes = min_bytes = bytes + empty_size; |
451d7585 | 2842 | } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) { |
1bb91902 AO |
2843 | cont1_bytes = bytes; |
2844 | min_bytes = block_group->sectorsize; | |
2845 | } else { | |
2846 | cont1_bytes = max(bytes, (bytes + empty_size) >> 2); | |
2847 | min_bytes = block_group->sectorsize; | |
2848 | } | |
fa9c0d79 | 2849 | |
34d52cb6 | 2850 | spin_lock(&ctl->tree_lock); |
7d0d2e8e JB |
2851 | |
2852 | /* | |
2853 | * If we know we don't have enough space to make a cluster don't even | |
2854 | * bother doing all the work to try and find one. | |
2855 | */ | |
1bb91902 | 2856 | if (ctl->free_space < bytes) { |
34d52cb6 | 2857 | spin_unlock(&ctl->tree_lock); |
7d0d2e8e JB |
2858 | return -ENOSPC; |
2859 | } | |
2860 | ||
fa9c0d79 CM |
2861 | spin_lock(&cluster->lock); |
2862 | ||
2863 | /* someone already found a cluster, hooray */ | |
2864 | if (cluster->block_group) { | |
2865 | ret = 0; | |
2866 | goto out; | |
2867 | } | |
fa9c0d79 | 2868 | |
3f7de037 JB |
2869 | trace_btrfs_find_cluster(block_group, offset, bytes, empty_size, |
2870 | min_bytes); | |
2871 | ||
2872 | INIT_LIST_HEAD(&bitmaps); | |
86d4a77b | 2873 | ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset, |
1bb91902 AO |
2874 | bytes + empty_size, |
2875 | cont1_bytes, min_bytes); | |
4e69b598 | 2876 | if (ret) |
86d4a77b | 2877 | ret = setup_cluster_bitmap(block_group, cluster, &bitmaps, |
1bb91902 AO |
2878 | offset, bytes + empty_size, |
2879 | cont1_bytes, min_bytes); | |
86d4a77b JB |
2880 | |
2881 | /* Clear our temporary list */ | |
2882 | list_for_each_entry_safe(entry, tmp, &bitmaps, list) | |
2883 | list_del_init(&entry->list); | |
fa9c0d79 | 2884 | |
4e69b598 JB |
2885 | if (!ret) { |
2886 | atomic_inc(&block_group->count); | |
2887 | list_add_tail(&cluster->block_group_list, | |
2888 | &block_group->cluster_list); | |
2889 | cluster->block_group = block_group; | |
3f7de037 JB |
2890 | } else { |
2891 | trace_btrfs_failed_cluster_setup(block_group); | |
fa9c0d79 | 2892 | } |
fa9c0d79 CM |
2893 | out: |
2894 | spin_unlock(&cluster->lock); | |
34d52cb6 | 2895 | spin_unlock(&ctl->tree_lock); |
fa9c0d79 CM |
2896 | |
2897 | return ret; | |
2898 | } | |
2899 | ||
2900 | /* | |
2901 | * simple code to zero out a cluster | |
2902 | */ | |
2903 | void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster) | |
2904 | { | |
2905 | spin_lock_init(&cluster->lock); | |
2906 | spin_lock_init(&cluster->refill_lock); | |
6bef4d31 | 2907 | cluster->root = RB_ROOT; |
fa9c0d79 CM |
2908 | cluster->max_size = 0; |
2909 | INIT_LIST_HEAD(&cluster->block_group_list); | |
2910 | cluster->block_group = NULL; | |
2911 | } | |
2912 | ||
7fe1e641 LZ |
2913 | static int do_trimming(struct btrfs_block_group_cache *block_group, |
2914 | u64 *total_trimmed, u64 start, u64 bytes, | |
2915 | u64 reserved_start, u64 reserved_bytes) | |
f7039b1d | 2916 | { |
7fe1e641 | 2917 | struct btrfs_space_info *space_info = block_group->space_info; |
f7039b1d | 2918 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
7fe1e641 LZ |
2919 | int ret; |
2920 | int update = 0; | |
2921 | u64 trimmed = 0; | |
f7039b1d | 2922 | |
7fe1e641 LZ |
2923 | spin_lock(&space_info->lock); |
2924 | spin_lock(&block_group->lock); | |
2925 | if (!block_group->ro) { | |
2926 | block_group->reserved += reserved_bytes; | |
2927 | space_info->bytes_reserved += reserved_bytes; | |
2928 | update = 1; | |
2929 | } | |
2930 | spin_unlock(&block_group->lock); | |
2931 | spin_unlock(&space_info->lock); | |
2932 | ||
2933 | ret = btrfs_error_discard_extent(fs_info->extent_root, | |
2934 | start, bytes, &trimmed); | |
2935 | if (!ret) | |
2936 | *total_trimmed += trimmed; | |
2937 | ||
2938 | btrfs_add_free_space(block_group, reserved_start, reserved_bytes); | |
2939 | ||
2940 | if (update) { | |
2941 | spin_lock(&space_info->lock); | |
2942 | spin_lock(&block_group->lock); | |
2943 | if (block_group->ro) | |
2944 | space_info->bytes_readonly += reserved_bytes; | |
2945 | block_group->reserved -= reserved_bytes; | |
2946 | space_info->bytes_reserved -= reserved_bytes; | |
2947 | spin_unlock(&space_info->lock); | |
2948 | spin_unlock(&block_group->lock); | |
2949 | } | |
2950 | ||
2951 | return ret; | |
2952 | } | |
2953 | ||
2954 | static int trim_no_bitmap(struct btrfs_block_group_cache *block_group, | |
2955 | u64 *total_trimmed, u64 start, u64 end, u64 minlen) | |
2956 | { | |
2957 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
2958 | struct btrfs_free_space *entry; | |
2959 | struct rb_node *node; | |
2960 | int ret = 0; | |
2961 | u64 extent_start; | |
2962 | u64 extent_bytes; | |
2963 | u64 bytes; | |
f7039b1d LD |
2964 | |
2965 | while (start < end) { | |
34d52cb6 | 2966 | spin_lock(&ctl->tree_lock); |
f7039b1d | 2967 | |
34d52cb6 LZ |
2968 | if (ctl->free_space < minlen) { |
2969 | spin_unlock(&ctl->tree_lock); | |
f7039b1d LD |
2970 | break; |
2971 | } | |
2972 | ||
34d52cb6 | 2973 | entry = tree_search_offset(ctl, start, 0, 1); |
7fe1e641 | 2974 | if (!entry) { |
34d52cb6 | 2975 | spin_unlock(&ctl->tree_lock); |
f7039b1d LD |
2976 | break; |
2977 | } | |
2978 | ||
7fe1e641 LZ |
2979 | /* skip bitmaps */ |
2980 | while (entry->bitmap) { | |
2981 | node = rb_next(&entry->offset_index); | |
2982 | if (!node) { | |
34d52cb6 | 2983 | spin_unlock(&ctl->tree_lock); |
7fe1e641 | 2984 | goto out; |
f7039b1d | 2985 | } |
7fe1e641 LZ |
2986 | entry = rb_entry(node, struct btrfs_free_space, |
2987 | offset_index); | |
f7039b1d LD |
2988 | } |
2989 | ||
7fe1e641 LZ |
2990 | if (entry->offset >= end) { |
2991 | spin_unlock(&ctl->tree_lock); | |
2992 | break; | |
f7039b1d LD |
2993 | } |
2994 | ||
7fe1e641 LZ |
2995 | extent_start = entry->offset; |
2996 | extent_bytes = entry->bytes; | |
2997 | start = max(start, extent_start); | |
2998 | bytes = min(extent_start + extent_bytes, end) - start; | |
2999 | if (bytes < minlen) { | |
3000 | spin_unlock(&ctl->tree_lock); | |
3001 | goto next; | |
f7039b1d LD |
3002 | } |
3003 | ||
7fe1e641 LZ |
3004 | unlink_free_space(ctl, entry); |
3005 | kmem_cache_free(btrfs_free_space_cachep, entry); | |
3006 | ||
34d52cb6 | 3007 | spin_unlock(&ctl->tree_lock); |
f7039b1d | 3008 | |
7fe1e641 LZ |
3009 | ret = do_trimming(block_group, total_trimmed, start, bytes, |
3010 | extent_start, extent_bytes); | |
3011 | if (ret) | |
3012 | break; | |
3013 | next: | |
3014 | start += bytes; | |
f7039b1d | 3015 | |
7fe1e641 LZ |
3016 | if (fatal_signal_pending(current)) { |
3017 | ret = -ERESTARTSYS; | |
3018 | break; | |
3019 | } | |
3020 | ||
3021 | cond_resched(); | |
3022 | } | |
3023 | out: | |
3024 | return ret; | |
3025 | } | |
3026 | ||
3027 | static int trim_bitmaps(struct btrfs_block_group_cache *block_group, | |
3028 | u64 *total_trimmed, u64 start, u64 end, u64 minlen) | |
3029 | { | |
3030 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
3031 | struct btrfs_free_space *entry; | |
3032 | int ret = 0; | |
3033 | int ret2; | |
3034 | u64 bytes; | |
3035 | u64 offset = offset_to_bitmap(ctl, start); | |
3036 | ||
3037 | while (offset < end) { | |
3038 | bool next_bitmap = false; | |
3039 | ||
3040 | spin_lock(&ctl->tree_lock); | |
3041 | ||
3042 | if (ctl->free_space < minlen) { | |
3043 | spin_unlock(&ctl->tree_lock); | |
3044 | break; | |
3045 | } | |
3046 | ||
3047 | entry = tree_search_offset(ctl, offset, 1, 0); | |
3048 | if (!entry) { | |
3049 | spin_unlock(&ctl->tree_lock); | |
3050 | next_bitmap = true; | |
3051 | goto next; | |
3052 | } | |
3053 | ||
3054 | bytes = minlen; | |
3055 | ret2 = search_bitmap(ctl, entry, &start, &bytes); | |
3056 | if (ret2 || start >= end) { | |
3057 | spin_unlock(&ctl->tree_lock); | |
3058 | next_bitmap = true; | |
3059 | goto next; | |
3060 | } | |
3061 | ||
3062 | bytes = min(bytes, end - start); | |
3063 | if (bytes < minlen) { | |
3064 | spin_unlock(&ctl->tree_lock); | |
3065 | goto next; | |
3066 | } | |
3067 | ||
3068 | bitmap_clear_bits(ctl, entry, start, bytes); | |
3069 | if (entry->bytes == 0) | |
3070 | free_bitmap(ctl, entry); | |
3071 | ||
3072 | spin_unlock(&ctl->tree_lock); | |
3073 | ||
3074 | ret = do_trimming(block_group, total_trimmed, start, bytes, | |
3075 | start, bytes); | |
3076 | if (ret) | |
3077 | break; | |
3078 | next: | |
3079 | if (next_bitmap) { | |
3080 | offset += BITS_PER_BITMAP * ctl->unit; | |
3081 | } else { | |
3082 | start += bytes; | |
3083 | if (start >= offset + BITS_PER_BITMAP * ctl->unit) | |
3084 | offset += BITS_PER_BITMAP * ctl->unit; | |
f7039b1d | 3085 | } |
f7039b1d LD |
3086 | |
3087 | if (fatal_signal_pending(current)) { | |
3088 | ret = -ERESTARTSYS; | |
3089 | break; | |
3090 | } | |
3091 | ||
3092 | cond_resched(); | |
3093 | } | |
3094 | ||
3095 | return ret; | |
3096 | } | |
581bb050 | 3097 | |
7fe1e641 LZ |
3098 | int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group, |
3099 | u64 *trimmed, u64 start, u64 end, u64 minlen) | |
3100 | { | |
3101 | int ret; | |
3102 | ||
3103 | *trimmed = 0; | |
3104 | ||
04216820 FM |
3105 | spin_lock(&block_group->lock); |
3106 | if (block_group->removed) { | |
3107 | spin_unlock(&block_group->lock); | |
3108 | return 0; | |
3109 | } | |
3110 | atomic_inc(&block_group->trimming); | |
3111 | spin_unlock(&block_group->lock); | |
3112 | ||
7fe1e641 LZ |
3113 | ret = trim_no_bitmap(block_group, trimmed, start, end, minlen); |
3114 | if (ret) | |
04216820 | 3115 | goto out; |
7fe1e641 LZ |
3116 | |
3117 | ret = trim_bitmaps(block_group, trimmed, start, end, minlen); | |
04216820 FM |
3118 | out: |
3119 | spin_lock(&block_group->lock); | |
3120 | if (atomic_dec_and_test(&block_group->trimming) && | |
3121 | block_group->removed) { | |
3122 | struct extent_map_tree *em_tree; | |
3123 | struct extent_map *em; | |
3124 | ||
3125 | spin_unlock(&block_group->lock); | |
3126 | ||
3127 | em_tree = &block_group->fs_info->mapping_tree.map_tree; | |
3128 | write_lock(&em_tree->lock); | |
3129 | em = lookup_extent_mapping(em_tree, block_group->key.objectid, | |
3130 | 1); | |
3131 | BUG_ON(!em); /* logic error, can't happen */ | |
3132 | remove_extent_mapping(em_tree, em); | |
3133 | write_unlock(&em_tree->lock); | |
3134 | ||
3135 | lock_chunks(block_group->fs_info->chunk_root); | |
3136 | list_del_init(&em->list); | |
3137 | unlock_chunks(block_group->fs_info->chunk_root); | |
3138 | ||
3139 | /* once for us and once for the tree */ | |
3140 | free_extent_map(em); | |
3141 | free_extent_map(em); | |
3142 | } else { | |
3143 | spin_unlock(&block_group->lock); | |
3144 | } | |
7fe1e641 LZ |
3145 | |
3146 | return ret; | |
3147 | } | |
3148 | ||
581bb050 LZ |
3149 | /* |
3150 | * Find the left-most item in the cache tree, and then return the | |
3151 | * smallest inode number in the item. | |
3152 | * | |
3153 | * Note: the returned inode number may not be the smallest one in | |
3154 | * the tree, if the left-most item is a bitmap. | |
3155 | */ | |
3156 | u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root) | |
3157 | { | |
3158 | struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl; | |
3159 | struct btrfs_free_space *entry = NULL; | |
3160 | u64 ino = 0; | |
3161 | ||
3162 | spin_lock(&ctl->tree_lock); | |
3163 | ||
3164 | if (RB_EMPTY_ROOT(&ctl->free_space_offset)) | |
3165 | goto out; | |
3166 | ||
3167 | entry = rb_entry(rb_first(&ctl->free_space_offset), | |
3168 | struct btrfs_free_space, offset_index); | |
3169 | ||
3170 | if (!entry->bitmap) { | |
3171 | ino = entry->offset; | |
3172 | ||
3173 | unlink_free_space(ctl, entry); | |
3174 | entry->offset++; | |
3175 | entry->bytes--; | |
3176 | if (!entry->bytes) | |
3177 | kmem_cache_free(btrfs_free_space_cachep, entry); | |
3178 | else | |
3179 | link_free_space(ctl, entry); | |
3180 | } else { | |
3181 | u64 offset = 0; | |
3182 | u64 count = 1; | |
3183 | int ret; | |
3184 | ||
3185 | ret = search_bitmap(ctl, entry, &offset, &count); | |
79787eaa | 3186 | /* Logic error; Should be empty if it can't find anything */ |
b12d6869 | 3187 | ASSERT(!ret); |
581bb050 LZ |
3188 | |
3189 | ino = offset; | |
3190 | bitmap_clear_bits(ctl, entry, offset, 1); | |
3191 | if (entry->bytes == 0) | |
3192 | free_bitmap(ctl, entry); | |
3193 | } | |
3194 | out: | |
3195 | spin_unlock(&ctl->tree_lock); | |
3196 | ||
3197 | return ino; | |
3198 | } | |
82d5902d LZ |
3199 | |
3200 | struct inode *lookup_free_ino_inode(struct btrfs_root *root, | |
3201 | struct btrfs_path *path) | |
3202 | { | |
3203 | struct inode *inode = NULL; | |
3204 | ||
57cdc8db DS |
3205 | spin_lock(&root->ino_cache_lock); |
3206 | if (root->ino_cache_inode) | |
3207 | inode = igrab(root->ino_cache_inode); | |
3208 | spin_unlock(&root->ino_cache_lock); | |
82d5902d LZ |
3209 | if (inode) |
3210 | return inode; | |
3211 | ||
3212 | inode = __lookup_free_space_inode(root, path, 0); | |
3213 | if (IS_ERR(inode)) | |
3214 | return inode; | |
3215 | ||
57cdc8db | 3216 | spin_lock(&root->ino_cache_lock); |
7841cb28 | 3217 | if (!btrfs_fs_closing(root->fs_info)) |
57cdc8db DS |
3218 | root->ino_cache_inode = igrab(inode); |
3219 | spin_unlock(&root->ino_cache_lock); | |
82d5902d LZ |
3220 | |
3221 | return inode; | |
3222 | } | |
3223 | ||
3224 | int create_free_ino_inode(struct btrfs_root *root, | |
3225 | struct btrfs_trans_handle *trans, | |
3226 | struct btrfs_path *path) | |
3227 | { | |
3228 | return __create_free_space_inode(root, trans, path, | |
3229 | BTRFS_FREE_INO_OBJECTID, 0); | |
3230 | } | |
3231 | ||
3232 | int load_free_ino_cache(struct btrfs_fs_info *fs_info, struct btrfs_root *root) | |
3233 | { | |
3234 | struct btrfs_free_space_ctl *ctl = root->free_ino_ctl; | |
3235 | struct btrfs_path *path; | |
3236 | struct inode *inode; | |
3237 | int ret = 0; | |
3238 | u64 root_gen = btrfs_root_generation(&root->root_item); | |
3239 | ||
4b9465cb CM |
3240 | if (!btrfs_test_opt(root, INODE_MAP_CACHE)) |
3241 | return 0; | |
3242 | ||
82d5902d LZ |
3243 | /* |
3244 | * If we're unmounting then just return, since this does a search on the | |
3245 | * normal root and not the commit root and we could deadlock. | |
3246 | */ | |
7841cb28 | 3247 | if (btrfs_fs_closing(fs_info)) |
82d5902d LZ |
3248 | return 0; |
3249 | ||
3250 | path = btrfs_alloc_path(); | |
3251 | if (!path) | |
3252 | return 0; | |
3253 | ||
3254 | inode = lookup_free_ino_inode(root, path); | |
3255 | if (IS_ERR(inode)) | |
3256 | goto out; | |
3257 | ||
3258 | if (root_gen != BTRFS_I(inode)->generation) | |
3259 | goto out_put; | |
3260 | ||
3261 | ret = __load_free_space_cache(root, inode, ctl, path, 0); | |
3262 | ||
3263 | if (ret < 0) | |
c2cf52eb SK |
3264 | btrfs_err(fs_info, |
3265 | "failed to load free ino cache for root %llu", | |
3266 | root->root_key.objectid); | |
82d5902d LZ |
3267 | out_put: |
3268 | iput(inode); | |
3269 | out: | |
3270 | btrfs_free_path(path); | |
3271 | return ret; | |
3272 | } | |
3273 | ||
3274 | int btrfs_write_out_ino_cache(struct btrfs_root *root, | |
3275 | struct btrfs_trans_handle *trans, | |
53645a91 FDBM |
3276 | struct btrfs_path *path, |
3277 | struct inode *inode) | |
82d5902d LZ |
3278 | { |
3279 | struct btrfs_free_space_ctl *ctl = root->free_ino_ctl; | |
82d5902d LZ |
3280 | int ret; |
3281 | ||
4b9465cb CM |
3282 | if (!btrfs_test_opt(root, INODE_MAP_CACHE)) |
3283 | return 0; | |
3284 | ||
82d5902d | 3285 | ret = __btrfs_write_out_cache(root, inode, ctl, NULL, trans, path, 0); |
c09544e0 JB |
3286 | if (ret) { |
3287 | btrfs_delalloc_release_metadata(inode, inode->i_size); | |
3288 | #ifdef DEBUG | |
c2cf52eb SK |
3289 | btrfs_err(root->fs_info, |
3290 | "failed to write free ino cache for root %llu", | |
3291 | root->root_key.objectid); | |
c09544e0 JB |
3292 | #endif |
3293 | } | |
82d5902d | 3294 | |
82d5902d LZ |
3295 | return ret; |
3296 | } | |
74255aa0 JB |
3297 | |
3298 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS | |
dc11dd5d JB |
3299 | /* |
3300 | * Use this if you need to make a bitmap or extent entry specifically, it | |
3301 | * doesn't do any of the merging that add_free_space does, this acts a lot like | |
3302 | * how the free space cache loading stuff works, so you can get really weird | |
3303 | * configurations. | |
3304 | */ | |
3305 | int test_add_free_space_entry(struct btrfs_block_group_cache *cache, | |
3306 | u64 offset, u64 bytes, bool bitmap) | |
74255aa0 | 3307 | { |
dc11dd5d JB |
3308 | struct btrfs_free_space_ctl *ctl = cache->free_space_ctl; |
3309 | struct btrfs_free_space *info = NULL, *bitmap_info; | |
3310 | void *map = NULL; | |
3311 | u64 bytes_added; | |
3312 | int ret; | |
74255aa0 | 3313 | |
dc11dd5d JB |
3314 | again: |
3315 | if (!info) { | |
3316 | info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS); | |
3317 | if (!info) | |
3318 | return -ENOMEM; | |
74255aa0 JB |
3319 | } |
3320 | ||
dc11dd5d JB |
3321 | if (!bitmap) { |
3322 | spin_lock(&ctl->tree_lock); | |
3323 | info->offset = offset; | |
3324 | info->bytes = bytes; | |
3325 | ret = link_free_space(ctl, info); | |
3326 | spin_unlock(&ctl->tree_lock); | |
3327 | if (ret) | |
3328 | kmem_cache_free(btrfs_free_space_cachep, info); | |
3329 | return ret; | |
3330 | } | |
3331 | ||
3332 | if (!map) { | |
3333 | map = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS); | |
3334 | if (!map) { | |
3335 | kmem_cache_free(btrfs_free_space_cachep, info); | |
3336 | return -ENOMEM; | |
3337 | } | |
3338 | } | |
3339 | ||
3340 | spin_lock(&ctl->tree_lock); | |
3341 | bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), | |
3342 | 1, 0); | |
3343 | if (!bitmap_info) { | |
3344 | info->bitmap = map; | |
3345 | map = NULL; | |
3346 | add_new_bitmap(ctl, info, offset); | |
3347 | bitmap_info = info; | |
20005523 | 3348 | info = NULL; |
dc11dd5d | 3349 | } |
74255aa0 | 3350 | |
dc11dd5d JB |
3351 | bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes); |
3352 | bytes -= bytes_added; | |
3353 | offset += bytes_added; | |
3354 | spin_unlock(&ctl->tree_lock); | |
74255aa0 | 3355 | |
dc11dd5d JB |
3356 | if (bytes) |
3357 | goto again; | |
74255aa0 | 3358 | |
20005523 FM |
3359 | if (info) |
3360 | kmem_cache_free(btrfs_free_space_cachep, info); | |
dc11dd5d JB |
3361 | if (map) |
3362 | kfree(map); | |
3363 | return 0; | |
74255aa0 JB |
3364 | } |
3365 | ||
3366 | /* | |
3367 | * Checks to see if the given range is in the free space cache. This is really | |
3368 | * just used to check the absence of space, so if there is free space in the | |
3369 | * range at all we will return 1. | |
3370 | */ | |
dc11dd5d JB |
3371 | int test_check_exists(struct btrfs_block_group_cache *cache, |
3372 | u64 offset, u64 bytes) | |
74255aa0 JB |
3373 | { |
3374 | struct btrfs_free_space_ctl *ctl = cache->free_space_ctl; | |
3375 | struct btrfs_free_space *info; | |
3376 | int ret = 0; | |
3377 | ||
3378 | spin_lock(&ctl->tree_lock); | |
3379 | info = tree_search_offset(ctl, offset, 0, 0); | |
3380 | if (!info) { | |
3381 | info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), | |
3382 | 1, 0); | |
3383 | if (!info) | |
3384 | goto out; | |
3385 | } | |
3386 | ||
3387 | have_info: | |
3388 | if (info->bitmap) { | |
3389 | u64 bit_off, bit_bytes; | |
3390 | struct rb_node *n; | |
3391 | struct btrfs_free_space *tmp; | |
3392 | ||
3393 | bit_off = offset; | |
3394 | bit_bytes = ctl->unit; | |
3395 | ret = search_bitmap(ctl, info, &bit_off, &bit_bytes); | |
3396 | if (!ret) { | |
3397 | if (bit_off == offset) { | |
3398 | ret = 1; | |
3399 | goto out; | |
3400 | } else if (bit_off > offset && | |
3401 | offset + bytes > bit_off) { | |
3402 | ret = 1; | |
3403 | goto out; | |
3404 | } | |
3405 | } | |
3406 | ||
3407 | n = rb_prev(&info->offset_index); | |
3408 | while (n) { | |
3409 | tmp = rb_entry(n, struct btrfs_free_space, | |
3410 | offset_index); | |
3411 | if (tmp->offset + tmp->bytes < offset) | |
3412 | break; | |
3413 | if (offset + bytes < tmp->offset) { | |
3414 | n = rb_prev(&info->offset_index); | |
3415 | continue; | |
3416 | } | |
3417 | info = tmp; | |
3418 | goto have_info; | |
3419 | } | |
3420 | ||
3421 | n = rb_next(&info->offset_index); | |
3422 | while (n) { | |
3423 | tmp = rb_entry(n, struct btrfs_free_space, | |
3424 | offset_index); | |
3425 | if (offset + bytes < tmp->offset) | |
3426 | break; | |
3427 | if (tmp->offset + tmp->bytes < offset) { | |
3428 | n = rb_next(&info->offset_index); | |
3429 | continue; | |
3430 | } | |
3431 | info = tmp; | |
3432 | goto have_info; | |
3433 | } | |
3434 | ||
20005523 | 3435 | ret = 0; |
74255aa0 JB |
3436 | goto out; |
3437 | } | |
3438 | ||
3439 | if (info->offset == offset) { | |
3440 | ret = 1; | |
3441 | goto out; | |
3442 | } | |
3443 | ||
3444 | if (offset > info->offset && offset < info->offset + info->bytes) | |
3445 | ret = 1; | |
3446 | out: | |
3447 | spin_unlock(&ctl->tree_lock); | |
3448 | return ret; | |
3449 | } | |
dc11dd5d | 3450 | #endif /* CONFIG_BTRFS_FS_RUN_SANITY_TESTS */ |