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