]>
Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. 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 | ||
581bb050 LZ |
19 | #include <linux/delay.h> |
20 | #include <linux/kthread.h> | |
21 | #include <linux/pagemap.h> | |
22 | ||
9f5fae2f CM |
23 | #include "ctree.h" |
24 | #include "disk-io.h" | |
581bb050 LZ |
25 | #include "free-space-cache.h" |
26 | #include "inode-map.h" | |
9f5fae2f CM |
27 | #include "transaction.h" |
28 | ||
581bb050 LZ |
29 | static int caching_kthread(void *data) |
30 | { | |
31 | struct btrfs_root *root = data; | |
32 | struct btrfs_fs_info *fs_info = root->fs_info; | |
33 | struct btrfs_free_space_ctl *ctl = root->free_ino_ctl; | |
34 | struct btrfs_key key; | |
35 | struct btrfs_path *path; | |
36 | struct extent_buffer *leaf; | |
37 | u64 last = (u64)-1; | |
38 | int slot; | |
39 | int ret; | |
40 | ||
4b9465cb CM |
41 | if (!btrfs_test_opt(root, INODE_MAP_CACHE)) |
42 | return 0; | |
43 | ||
581bb050 LZ |
44 | path = btrfs_alloc_path(); |
45 | if (!path) | |
46 | return -ENOMEM; | |
47 | ||
48 | /* Since the commit root is read-only, we can safely skip locking. */ | |
49 | path->skip_locking = 1; | |
50 | path->search_commit_root = 1; | |
51 | path->reada = 2; | |
52 | ||
53 | key.objectid = BTRFS_FIRST_FREE_OBJECTID; | |
54 | key.offset = 0; | |
55 | key.type = BTRFS_INODE_ITEM_KEY; | |
56 | again: | |
57 | /* need to make sure the commit_root doesn't disappear */ | |
58 | mutex_lock(&root->fs_commit_mutex); | |
59 | ||
60 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
61 | if (ret < 0) | |
62 | goto out; | |
63 | ||
64 | while (1) { | |
65 | smp_mb(); | |
a47d6b70 | 66 | if (fs_info->closing) |
581bb050 LZ |
67 | goto out; |
68 | ||
69 | leaf = path->nodes[0]; | |
70 | slot = path->slots[0]; | |
a47d6b70 | 71 | if (slot >= btrfs_header_nritems(leaf)) { |
581bb050 LZ |
72 | ret = btrfs_next_leaf(root, path); |
73 | if (ret < 0) | |
74 | goto out; | |
75 | else if (ret > 0) | |
76 | break; | |
77 | ||
78 | if (need_resched() || | |
79 | btrfs_transaction_in_commit(fs_info)) { | |
80 | leaf = path->nodes[0]; | |
81 | ||
82 | if (btrfs_header_nritems(leaf) == 0) { | |
83 | WARN_ON(1); | |
84 | break; | |
85 | } | |
86 | ||
87 | /* | |
88 | * Save the key so we can advances forward | |
89 | * in the next search. | |
90 | */ | |
91 | btrfs_item_key_to_cpu(leaf, &key, 0); | |
945d8962 | 92 | btrfs_release_path(path); |
581bb050 LZ |
93 | root->cache_progress = last; |
94 | mutex_unlock(&root->fs_commit_mutex); | |
95 | schedule_timeout(1); | |
96 | goto again; | |
97 | } else | |
98 | continue; | |
99 | } | |
100 | ||
101 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
102 | ||
103 | if (key.type != BTRFS_INODE_ITEM_KEY) | |
104 | goto next; | |
105 | ||
a47d6b70 | 106 | if (key.objectid >= root->highest_objectid) |
581bb050 LZ |
107 | break; |
108 | ||
109 | if (last != (u64)-1 && last + 1 != key.objectid) { | |
110 | __btrfs_add_free_space(ctl, last + 1, | |
111 | key.objectid - last - 1); | |
112 | wake_up(&root->cache_wait); | |
113 | } | |
114 | ||
115 | last = key.objectid; | |
116 | next: | |
117 | path->slots[0]++; | |
118 | } | |
119 | ||
a47d6b70 | 120 | if (last < root->highest_objectid - 1) { |
581bb050 | 121 | __btrfs_add_free_space(ctl, last + 1, |
a47d6b70 | 122 | root->highest_objectid - last - 1); |
581bb050 LZ |
123 | } |
124 | ||
125 | spin_lock(&root->cache_lock); | |
126 | root->cached = BTRFS_CACHE_FINISHED; | |
127 | spin_unlock(&root->cache_lock); | |
128 | ||
129 | root->cache_progress = (u64)-1; | |
130 | btrfs_unpin_free_ino(root); | |
131 | out: | |
132 | wake_up(&root->cache_wait); | |
133 | mutex_unlock(&root->fs_commit_mutex); | |
134 | ||
135 | btrfs_free_path(path); | |
136 | ||
137 | return ret; | |
138 | } | |
139 | ||
140 | static void start_caching(struct btrfs_root *root) | |
141 | { | |
a47d6b70 | 142 | struct btrfs_free_space_ctl *ctl = root->free_ino_ctl; |
581bb050 | 143 | struct task_struct *tsk; |
82d5902d | 144 | int ret; |
a47d6b70 | 145 | u64 objectid; |
581bb050 | 146 | |
4b9465cb CM |
147 | if (!btrfs_test_opt(root, INODE_MAP_CACHE)) |
148 | return; | |
149 | ||
581bb050 LZ |
150 | spin_lock(&root->cache_lock); |
151 | if (root->cached != BTRFS_CACHE_NO) { | |
152 | spin_unlock(&root->cache_lock); | |
153 | return; | |
154 | } | |
155 | ||
156 | root->cached = BTRFS_CACHE_STARTED; | |
157 | spin_unlock(&root->cache_lock); | |
158 | ||
82d5902d LZ |
159 | ret = load_free_ino_cache(root->fs_info, root); |
160 | if (ret == 1) { | |
161 | spin_lock(&root->cache_lock); | |
162 | root->cached = BTRFS_CACHE_FINISHED; | |
163 | spin_unlock(&root->cache_lock); | |
164 | return; | |
165 | } | |
166 | ||
a47d6b70 LZ |
167 | /* |
168 | * It can be quite time-consuming to fill the cache by searching | |
169 | * through the extent tree, and this can keep ino allocation path | |
170 | * waiting. Therefore at start we quickly find out the highest | |
171 | * inode number and we know we can use inode numbers which fall in | |
172 | * [highest_ino + 1, BTRFS_LAST_FREE_OBJECTID]. | |
173 | */ | |
174 | ret = btrfs_find_free_objectid(root, &objectid); | |
175 | if (!ret && objectid <= BTRFS_LAST_FREE_OBJECTID) { | |
176 | __btrfs_add_free_space(ctl, objectid, | |
177 | BTRFS_LAST_FREE_OBJECTID - objectid + 1); | |
178 | } | |
179 | ||
581bb050 LZ |
180 | tsk = kthread_run(caching_kthread, root, "btrfs-ino-cache-%llu\n", |
181 | root->root_key.objectid); | |
182 | BUG_ON(IS_ERR(tsk)); | |
183 | } | |
184 | ||
185 | int btrfs_find_free_ino(struct btrfs_root *root, u64 *objectid) | |
186 | { | |
4b9465cb CM |
187 | if (!btrfs_test_opt(root, INODE_MAP_CACHE)) |
188 | return btrfs_find_free_objectid(root, objectid); | |
189 | ||
581bb050 LZ |
190 | again: |
191 | *objectid = btrfs_find_ino_for_alloc(root); | |
192 | ||
193 | if (*objectid != 0) | |
194 | return 0; | |
195 | ||
196 | start_caching(root); | |
197 | ||
198 | wait_event(root->cache_wait, | |
199 | root->cached == BTRFS_CACHE_FINISHED || | |
200 | root->free_ino_ctl->free_space > 0); | |
201 | ||
202 | if (root->cached == BTRFS_CACHE_FINISHED && | |
203 | root->free_ino_ctl->free_space == 0) | |
204 | return -ENOSPC; | |
205 | else | |
206 | goto again; | |
207 | } | |
208 | ||
209 | void btrfs_return_ino(struct btrfs_root *root, u64 objectid) | |
210 | { | |
211 | struct btrfs_free_space_ctl *ctl = root->free_ino_ctl; | |
212 | struct btrfs_free_space_ctl *pinned = root->free_ino_pinned; | |
4b9465cb CM |
213 | |
214 | if (!btrfs_test_opt(root, INODE_MAP_CACHE)) | |
215 | return; | |
216 | ||
581bb050 LZ |
217 | again: |
218 | if (root->cached == BTRFS_CACHE_FINISHED) { | |
219 | __btrfs_add_free_space(ctl, objectid, 1); | |
220 | } else { | |
221 | /* | |
222 | * If we are in the process of caching free ino chunks, | |
223 | * to avoid adding the same inode number to the free_ino | |
224 | * tree twice due to cross transaction, we'll leave it | |
225 | * in the pinned tree until a transaction is committed | |
226 | * or the caching work is done. | |
227 | */ | |
228 | ||
229 | mutex_lock(&root->fs_commit_mutex); | |
230 | spin_lock(&root->cache_lock); | |
231 | if (root->cached == BTRFS_CACHE_FINISHED) { | |
232 | spin_unlock(&root->cache_lock); | |
233 | mutex_unlock(&root->fs_commit_mutex); | |
234 | goto again; | |
235 | } | |
236 | spin_unlock(&root->cache_lock); | |
237 | ||
238 | start_caching(root); | |
239 | ||
a47d6b70 LZ |
240 | if (objectid <= root->cache_progress || |
241 | objectid > root->highest_objectid) | |
581bb050 LZ |
242 | __btrfs_add_free_space(ctl, objectid, 1); |
243 | else | |
244 | __btrfs_add_free_space(pinned, objectid, 1); | |
245 | ||
246 | mutex_unlock(&root->fs_commit_mutex); | |
247 | } | |
248 | } | |
249 | ||
250 | /* | |
251 | * When a transaction is committed, we'll move those inode numbers which | |
252 | * are smaller than root->cache_progress from pinned tree to free_ino tree, | |
253 | * and others will just be dropped, because the commit root we were | |
254 | * searching has changed. | |
255 | * | |
256 | * Must be called with root->fs_commit_mutex held | |
257 | */ | |
258 | void btrfs_unpin_free_ino(struct btrfs_root *root) | |
259 | { | |
260 | struct btrfs_free_space_ctl *ctl = root->free_ino_ctl; | |
261 | struct rb_root *rbroot = &root->free_ino_pinned->free_space_offset; | |
262 | struct btrfs_free_space *info; | |
263 | struct rb_node *n; | |
264 | u64 count; | |
265 | ||
4b9465cb CM |
266 | if (!btrfs_test_opt(root, INODE_MAP_CACHE)) |
267 | return; | |
268 | ||
581bb050 LZ |
269 | while (1) { |
270 | n = rb_first(rbroot); | |
271 | if (!n) | |
272 | break; | |
273 | ||
274 | info = rb_entry(n, struct btrfs_free_space, offset_index); | |
275 | BUG_ON(info->bitmap); | |
276 | ||
277 | if (info->offset > root->cache_progress) | |
278 | goto free; | |
279 | else if (info->offset + info->bytes > root->cache_progress) | |
280 | count = root->cache_progress - info->offset + 1; | |
281 | else | |
282 | count = info->bytes; | |
283 | ||
284 | __btrfs_add_free_space(ctl, info->offset, count); | |
285 | free: | |
286 | rb_erase(&info->offset_index, rbroot); | |
287 | kfree(info); | |
288 | } | |
289 | } | |
290 | ||
291 | #define INIT_THRESHOLD (((1024 * 32) / 2) / sizeof(struct btrfs_free_space)) | |
292 | #define INODES_PER_BITMAP (PAGE_CACHE_SIZE * 8) | |
293 | ||
294 | /* | |
295 | * The goal is to keep the memory used by the free_ino tree won't | |
296 | * exceed the memory if we use bitmaps only. | |
297 | */ | |
298 | static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl) | |
299 | { | |
300 | struct btrfs_free_space *info; | |
301 | struct rb_node *n; | |
302 | int max_ino; | |
303 | int max_bitmaps; | |
304 | ||
305 | n = rb_last(&ctl->free_space_offset); | |
306 | if (!n) { | |
307 | ctl->extents_thresh = INIT_THRESHOLD; | |
308 | return; | |
309 | } | |
310 | info = rb_entry(n, struct btrfs_free_space, offset_index); | |
311 | ||
312 | /* | |
313 | * Find the maximum inode number in the filesystem. Note we | |
314 | * ignore the fact that this can be a bitmap, because we are | |
315 | * not doing precise calculation. | |
316 | */ | |
317 | max_ino = info->bytes - 1; | |
318 | ||
319 | max_bitmaps = ALIGN(max_ino, INODES_PER_BITMAP) / INODES_PER_BITMAP; | |
320 | if (max_bitmaps <= ctl->total_bitmaps) { | |
321 | ctl->extents_thresh = 0; | |
322 | return; | |
323 | } | |
324 | ||
325 | ctl->extents_thresh = (max_bitmaps - ctl->total_bitmaps) * | |
326 | PAGE_CACHE_SIZE / sizeof(*info); | |
327 | } | |
328 | ||
329 | /* | |
330 | * We don't fall back to bitmap, if we are below the extents threshold | |
331 | * or this chunk of inode numbers is a big one. | |
332 | */ | |
333 | static bool use_bitmap(struct btrfs_free_space_ctl *ctl, | |
334 | struct btrfs_free_space *info) | |
335 | { | |
336 | if (ctl->free_extents < ctl->extents_thresh || | |
337 | info->bytes > INODES_PER_BITMAP / 10) | |
338 | return false; | |
339 | ||
340 | return true; | |
341 | } | |
342 | ||
343 | static struct btrfs_free_space_op free_ino_op = { | |
344 | .recalc_thresholds = recalculate_thresholds, | |
345 | .use_bitmap = use_bitmap, | |
346 | }; | |
347 | ||
348 | static void pinned_recalc_thresholds(struct btrfs_free_space_ctl *ctl) | |
349 | { | |
350 | } | |
351 | ||
352 | static bool pinned_use_bitmap(struct btrfs_free_space_ctl *ctl, | |
353 | struct btrfs_free_space *info) | |
354 | { | |
355 | /* | |
356 | * We always use extents for two reasons: | |
357 | * | |
358 | * - The pinned tree is only used during the process of caching | |
359 | * work. | |
360 | * - Make code simpler. See btrfs_unpin_free_ino(). | |
361 | */ | |
362 | return false; | |
363 | } | |
364 | ||
365 | static struct btrfs_free_space_op pinned_free_ino_op = { | |
366 | .recalc_thresholds = pinned_recalc_thresholds, | |
367 | .use_bitmap = pinned_use_bitmap, | |
368 | }; | |
369 | ||
370 | void btrfs_init_free_ino_ctl(struct btrfs_root *root) | |
371 | { | |
372 | struct btrfs_free_space_ctl *ctl = root->free_ino_ctl; | |
373 | struct btrfs_free_space_ctl *pinned = root->free_ino_pinned; | |
374 | ||
375 | spin_lock_init(&ctl->tree_lock); | |
376 | ctl->unit = 1; | |
377 | ctl->start = 0; | |
378 | ctl->private = NULL; | |
379 | ctl->op = &free_ino_op; | |
380 | ||
381 | /* | |
382 | * Initially we allow to use 16K of ram to cache chunks of | |
383 | * inode numbers before we resort to bitmaps. This is somewhat | |
384 | * arbitrary, but it will be adjusted in runtime. | |
385 | */ | |
386 | ctl->extents_thresh = INIT_THRESHOLD; | |
387 | ||
388 | spin_lock_init(&pinned->tree_lock); | |
389 | pinned->unit = 1; | |
390 | pinned->start = 0; | |
391 | pinned->private = NULL; | |
392 | pinned->extents_thresh = 0; | |
393 | pinned->op = &pinned_free_ino_op; | |
394 | } | |
395 | ||
82d5902d LZ |
396 | int btrfs_save_ino_cache(struct btrfs_root *root, |
397 | struct btrfs_trans_handle *trans) | |
398 | { | |
399 | struct btrfs_free_space_ctl *ctl = root->free_ino_ctl; | |
400 | struct btrfs_path *path; | |
401 | struct inode *inode; | |
402 | u64 alloc_hint = 0; | |
403 | int ret; | |
404 | int prealloc; | |
405 | bool retry = false; | |
406 | ||
ca456ae2 | 407 | /* only fs tree and subvol/snap needs ino cache */ |
408 | if (root->root_key.objectid != BTRFS_FS_TREE_OBJECTID && | |
409 | (root->root_key.objectid < BTRFS_FIRST_FREE_OBJECTID || | |
410 | root->root_key.objectid > BTRFS_LAST_FREE_OBJECTID)) | |
411 | return 0; | |
412 | ||
d132a538 JB |
413 | /* Don't save inode cache if we are deleting this root */ |
414 | if (btrfs_root_refs(&root->root_item) == 0 && | |
415 | root != root->fs_info->tree_root) | |
416 | return 0; | |
417 | ||
4b9465cb CM |
418 | if (!btrfs_test_opt(root, INODE_MAP_CACHE)) |
419 | return 0; | |
420 | ||
82d5902d LZ |
421 | path = btrfs_alloc_path(); |
422 | if (!path) | |
423 | return -ENOMEM; | |
4b9465cb | 424 | |
82d5902d LZ |
425 | again: |
426 | inode = lookup_free_ino_inode(root, path); | |
427 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { | |
428 | ret = PTR_ERR(inode); | |
429 | goto out; | |
430 | } | |
431 | ||
432 | if (IS_ERR(inode)) { | |
433 | BUG_ON(retry); | |
434 | retry = true; | |
435 | ||
436 | ret = create_free_ino_inode(root, trans, path); | |
437 | if (ret) | |
438 | goto out; | |
439 | goto again; | |
440 | } | |
441 | ||
442 | BTRFS_I(inode)->generation = 0; | |
443 | ret = btrfs_update_inode(trans, root, inode); | |
444 | WARN_ON(ret); | |
445 | ||
446 | if (i_size_read(inode) > 0) { | |
447 | ret = btrfs_truncate_free_space_cache(root, trans, path, inode); | |
448 | if (ret) | |
449 | goto out_put; | |
450 | } | |
451 | ||
452 | spin_lock(&root->cache_lock); | |
453 | if (root->cached != BTRFS_CACHE_FINISHED) { | |
454 | ret = -1; | |
455 | spin_unlock(&root->cache_lock); | |
456 | goto out_put; | |
457 | } | |
458 | spin_unlock(&root->cache_lock); | |
459 | ||
460 | spin_lock(&ctl->tree_lock); | |
461 | prealloc = sizeof(struct btrfs_free_space) * ctl->free_extents; | |
462 | prealloc = ALIGN(prealloc, PAGE_CACHE_SIZE); | |
463 | prealloc += ctl->total_bitmaps * PAGE_CACHE_SIZE; | |
464 | spin_unlock(&ctl->tree_lock); | |
465 | ||
466 | /* Just to make sure we have enough space */ | |
467 | prealloc += 8 * PAGE_CACHE_SIZE; | |
468 | ||
469 | ret = btrfs_check_data_free_space(inode, prealloc); | |
470 | if (ret) | |
471 | goto out_put; | |
472 | ||
473 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, prealloc, | |
474 | prealloc, prealloc, &alloc_hint); | |
475 | if (ret) | |
476 | goto out_put; | |
477 | btrfs_free_reserved_data_space(inode, prealloc); | |
478 | ||
479 | out_put: | |
480 | iput(inode); | |
481 | out: | |
482 | if (ret == 0) | |
483 | ret = btrfs_write_out_ino_cache(root, trans, path); | |
484 | ||
485 | btrfs_free_path(path); | |
486 | return ret; | |
487 | } | |
488 | ||
581bb050 | 489 | static int btrfs_find_highest_objectid(struct btrfs_root *root, u64 *objectid) |
5be6f7f1 CM |
490 | { |
491 | struct btrfs_path *path; | |
492 | int ret; | |
5f39d397 | 493 | struct extent_buffer *l; |
5be6f7f1 | 494 | struct btrfs_key search_key; |
5f39d397 | 495 | struct btrfs_key found_key; |
5be6f7f1 CM |
496 | int slot; |
497 | ||
498 | path = btrfs_alloc_path(); | |
db5b493a TI |
499 | if (!path) |
500 | return -ENOMEM; | |
5be6f7f1 | 501 | |
6527cdbe ZY |
502 | search_key.objectid = BTRFS_LAST_FREE_OBJECTID; |
503 | search_key.type = -1; | |
5be6f7f1 CM |
504 | search_key.offset = (u64)-1; |
505 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
506 | if (ret < 0) | |
507 | goto error; | |
508 | BUG_ON(ret == 0); | |
509 | if (path->slots[0] > 0) { | |
510 | slot = path->slots[0] - 1; | |
5f39d397 CM |
511 | l = path->nodes[0]; |
512 | btrfs_item_key_to_cpu(l, &found_key, slot); | |
13a8a7c8 YZ |
513 | *objectid = max_t(u64, found_key.objectid, |
514 | BTRFS_FIRST_FREE_OBJECTID - 1); | |
5be6f7f1 | 515 | } else { |
13a8a7c8 | 516 | *objectid = BTRFS_FIRST_FREE_OBJECTID - 1; |
5be6f7f1 CM |
517 | } |
518 | ret = 0; | |
519 | error: | |
520 | btrfs_free_path(path); | |
521 | return ret; | |
522 | } | |
523 | ||
581bb050 | 524 | int btrfs_find_free_objectid(struct btrfs_root *root, u64 *objectid) |
9f5fae2f | 525 | { |
9f5fae2f | 526 | int ret; |
a2135011 | 527 | mutex_lock(&root->objectid_mutex); |
9f5fae2f | 528 | |
13a8a7c8 | 529 | if (unlikely(root->highest_objectid < BTRFS_FIRST_FREE_OBJECTID)) { |
581bb050 LZ |
530 | ret = btrfs_find_highest_objectid(root, |
531 | &root->highest_objectid); | |
13a8a7c8 YZ |
532 | if (ret) |
533 | goto out; | |
534 | } | |
5b21f2ed | 535 | |
13a8a7c8 YZ |
536 | if (unlikely(root->highest_objectid >= BTRFS_LAST_FREE_OBJECTID)) { |
537 | ret = -ENOSPC; | |
538 | goto out; | |
9f5fae2f | 539 | } |
13a8a7c8 YZ |
540 | |
541 | *objectid = ++root->highest_objectid; | |
542 | ret = 0; | |
543 | out: | |
a2135011 | 544 | mutex_unlock(&root->objectid_mutex); |
9f5fae2f CM |
545 | return ret; |
546 | } |