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); | |
89 | btrfs_release_path(root, path); | |
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; | |
140 | ||
141 | spin_lock(&root->cache_lock); | |
142 | if (root->cached != BTRFS_CACHE_NO) { | |
143 | spin_unlock(&root->cache_lock); | |
144 | return; | |
145 | } | |
146 | ||
147 | root->cached = BTRFS_CACHE_STARTED; | |
148 | spin_unlock(&root->cache_lock); | |
149 | ||
150 | tsk = kthread_run(caching_kthread, root, "btrfs-ino-cache-%llu\n", | |
151 | root->root_key.objectid); | |
152 | BUG_ON(IS_ERR(tsk)); | |
153 | } | |
154 | ||
155 | int btrfs_find_free_ino(struct btrfs_root *root, u64 *objectid) | |
156 | { | |
157 | again: | |
158 | *objectid = btrfs_find_ino_for_alloc(root); | |
159 | ||
160 | if (*objectid != 0) | |
161 | return 0; | |
162 | ||
163 | start_caching(root); | |
164 | ||
165 | wait_event(root->cache_wait, | |
166 | root->cached == BTRFS_CACHE_FINISHED || | |
167 | root->free_ino_ctl->free_space > 0); | |
168 | ||
169 | if (root->cached == BTRFS_CACHE_FINISHED && | |
170 | root->free_ino_ctl->free_space == 0) | |
171 | return -ENOSPC; | |
172 | else | |
173 | goto again; | |
174 | } | |
175 | ||
176 | void btrfs_return_ino(struct btrfs_root *root, u64 objectid) | |
177 | { | |
178 | struct btrfs_free_space_ctl *ctl = root->free_ino_ctl; | |
179 | struct btrfs_free_space_ctl *pinned = root->free_ino_pinned; | |
180 | again: | |
181 | if (root->cached == BTRFS_CACHE_FINISHED) { | |
182 | __btrfs_add_free_space(ctl, objectid, 1); | |
183 | } else { | |
184 | /* | |
185 | * If we are in the process of caching free ino chunks, | |
186 | * to avoid adding the same inode number to the free_ino | |
187 | * tree twice due to cross transaction, we'll leave it | |
188 | * in the pinned tree until a transaction is committed | |
189 | * or the caching work is done. | |
190 | */ | |
191 | ||
192 | mutex_lock(&root->fs_commit_mutex); | |
193 | spin_lock(&root->cache_lock); | |
194 | if (root->cached == BTRFS_CACHE_FINISHED) { | |
195 | spin_unlock(&root->cache_lock); | |
196 | mutex_unlock(&root->fs_commit_mutex); | |
197 | goto again; | |
198 | } | |
199 | spin_unlock(&root->cache_lock); | |
200 | ||
201 | start_caching(root); | |
202 | ||
203 | if (objectid <= root->cache_progress) | |
204 | __btrfs_add_free_space(ctl, objectid, 1); | |
205 | else | |
206 | __btrfs_add_free_space(pinned, objectid, 1); | |
207 | ||
208 | mutex_unlock(&root->fs_commit_mutex); | |
209 | } | |
210 | } | |
211 | ||
212 | /* | |
213 | * When a transaction is committed, we'll move those inode numbers which | |
214 | * are smaller than root->cache_progress from pinned tree to free_ino tree, | |
215 | * and others will just be dropped, because the commit root we were | |
216 | * searching has changed. | |
217 | * | |
218 | * Must be called with root->fs_commit_mutex held | |
219 | */ | |
220 | void btrfs_unpin_free_ino(struct btrfs_root *root) | |
221 | { | |
222 | struct btrfs_free_space_ctl *ctl = root->free_ino_ctl; | |
223 | struct rb_root *rbroot = &root->free_ino_pinned->free_space_offset; | |
224 | struct btrfs_free_space *info; | |
225 | struct rb_node *n; | |
226 | u64 count; | |
227 | ||
228 | while (1) { | |
229 | n = rb_first(rbroot); | |
230 | if (!n) | |
231 | break; | |
232 | ||
233 | info = rb_entry(n, struct btrfs_free_space, offset_index); | |
234 | BUG_ON(info->bitmap); | |
235 | ||
236 | if (info->offset > root->cache_progress) | |
237 | goto free; | |
238 | else if (info->offset + info->bytes > root->cache_progress) | |
239 | count = root->cache_progress - info->offset + 1; | |
240 | else | |
241 | count = info->bytes; | |
242 | ||
243 | __btrfs_add_free_space(ctl, info->offset, count); | |
244 | free: | |
245 | rb_erase(&info->offset_index, rbroot); | |
246 | kfree(info); | |
247 | } | |
248 | } | |
249 | ||
250 | #define INIT_THRESHOLD (((1024 * 32) / 2) / sizeof(struct btrfs_free_space)) | |
251 | #define INODES_PER_BITMAP (PAGE_CACHE_SIZE * 8) | |
252 | ||
253 | /* | |
254 | * The goal is to keep the memory used by the free_ino tree won't | |
255 | * exceed the memory if we use bitmaps only. | |
256 | */ | |
257 | static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl) | |
258 | { | |
259 | struct btrfs_free_space *info; | |
260 | struct rb_node *n; | |
261 | int max_ino; | |
262 | int max_bitmaps; | |
263 | ||
264 | n = rb_last(&ctl->free_space_offset); | |
265 | if (!n) { | |
266 | ctl->extents_thresh = INIT_THRESHOLD; | |
267 | return; | |
268 | } | |
269 | info = rb_entry(n, struct btrfs_free_space, offset_index); | |
270 | ||
271 | /* | |
272 | * Find the maximum inode number in the filesystem. Note we | |
273 | * ignore the fact that this can be a bitmap, because we are | |
274 | * not doing precise calculation. | |
275 | */ | |
276 | max_ino = info->bytes - 1; | |
277 | ||
278 | max_bitmaps = ALIGN(max_ino, INODES_PER_BITMAP) / INODES_PER_BITMAP; | |
279 | if (max_bitmaps <= ctl->total_bitmaps) { | |
280 | ctl->extents_thresh = 0; | |
281 | return; | |
282 | } | |
283 | ||
284 | ctl->extents_thresh = (max_bitmaps - ctl->total_bitmaps) * | |
285 | PAGE_CACHE_SIZE / sizeof(*info); | |
286 | } | |
287 | ||
288 | /* | |
289 | * We don't fall back to bitmap, if we are below the extents threshold | |
290 | * or this chunk of inode numbers is a big one. | |
291 | */ | |
292 | static bool use_bitmap(struct btrfs_free_space_ctl *ctl, | |
293 | struct btrfs_free_space *info) | |
294 | { | |
295 | if (ctl->free_extents < ctl->extents_thresh || | |
296 | info->bytes > INODES_PER_BITMAP / 10) | |
297 | return false; | |
298 | ||
299 | return true; | |
300 | } | |
301 | ||
302 | static struct btrfs_free_space_op free_ino_op = { | |
303 | .recalc_thresholds = recalculate_thresholds, | |
304 | .use_bitmap = use_bitmap, | |
305 | }; | |
306 | ||
307 | static void pinned_recalc_thresholds(struct btrfs_free_space_ctl *ctl) | |
308 | { | |
309 | } | |
310 | ||
311 | static bool pinned_use_bitmap(struct btrfs_free_space_ctl *ctl, | |
312 | struct btrfs_free_space *info) | |
313 | { | |
314 | /* | |
315 | * We always use extents for two reasons: | |
316 | * | |
317 | * - The pinned tree is only used during the process of caching | |
318 | * work. | |
319 | * - Make code simpler. See btrfs_unpin_free_ino(). | |
320 | */ | |
321 | return false; | |
322 | } | |
323 | ||
324 | static struct btrfs_free_space_op pinned_free_ino_op = { | |
325 | .recalc_thresholds = pinned_recalc_thresholds, | |
326 | .use_bitmap = pinned_use_bitmap, | |
327 | }; | |
328 | ||
329 | void btrfs_init_free_ino_ctl(struct btrfs_root *root) | |
330 | { | |
331 | struct btrfs_free_space_ctl *ctl = root->free_ino_ctl; | |
332 | struct btrfs_free_space_ctl *pinned = root->free_ino_pinned; | |
333 | ||
334 | spin_lock_init(&ctl->tree_lock); | |
335 | ctl->unit = 1; | |
336 | ctl->start = 0; | |
337 | ctl->private = NULL; | |
338 | ctl->op = &free_ino_op; | |
339 | ||
340 | /* | |
341 | * Initially we allow to use 16K of ram to cache chunks of | |
342 | * inode numbers before we resort to bitmaps. This is somewhat | |
343 | * arbitrary, but it will be adjusted in runtime. | |
344 | */ | |
345 | ctl->extents_thresh = INIT_THRESHOLD; | |
346 | ||
347 | spin_lock_init(&pinned->tree_lock); | |
348 | pinned->unit = 1; | |
349 | pinned->start = 0; | |
350 | pinned->private = NULL; | |
351 | pinned->extents_thresh = 0; | |
352 | pinned->op = &pinned_free_ino_op; | |
353 | } | |
354 | ||
355 | static int btrfs_find_highest_objectid(struct btrfs_root *root, u64 *objectid) | |
5be6f7f1 CM |
356 | { |
357 | struct btrfs_path *path; | |
358 | int ret; | |
5f39d397 | 359 | struct extent_buffer *l; |
5be6f7f1 | 360 | struct btrfs_key search_key; |
5f39d397 | 361 | struct btrfs_key found_key; |
5be6f7f1 CM |
362 | int slot; |
363 | ||
364 | path = btrfs_alloc_path(); | |
db5b493a TI |
365 | if (!path) |
366 | return -ENOMEM; | |
5be6f7f1 | 367 | |
6527cdbe ZY |
368 | search_key.objectid = BTRFS_LAST_FREE_OBJECTID; |
369 | search_key.type = -1; | |
5be6f7f1 CM |
370 | search_key.offset = (u64)-1; |
371 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
372 | if (ret < 0) | |
373 | goto error; | |
374 | BUG_ON(ret == 0); | |
375 | if (path->slots[0] > 0) { | |
376 | slot = path->slots[0] - 1; | |
5f39d397 CM |
377 | l = path->nodes[0]; |
378 | btrfs_item_key_to_cpu(l, &found_key, slot); | |
13a8a7c8 YZ |
379 | *objectid = max_t(u64, found_key.objectid, |
380 | BTRFS_FIRST_FREE_OBJECTID - 1); | |
5be6f7f1 | 381 | } else { |
13a8a7c8 | 382 | *objectid = BTRFS_FIRST_FREE_OBJECTID - 1; |
5be6f7f1 CM |
383 | } |
384 | ret = 0; | |
385 | error: | |
386 | btrfs_free_path(path); | |
387 | return ret; | |
388 | } | |
389 | ||
581bb050 | 390 | int btrfs_find_free_objectid(struct btrfs_root *root, u64 *objectid) |
9f5fae2f | 391 | { |
9f5fae2f | 392 | int ret; |
a2135011 | 393 | mutex_lock(&root->objectid_mutex); |
9f5fae2f | 394 | |
13a8a7c8 | 395 | if (unlikely(root->highest_objectid < BTRFS_FIRST_FREE_OBJECTID)) { |
581bb050 LZ |
396 | ret = btrfs_find_highest_objectid(root, |
397 | &root->highest_objectid); | |
13a8a7c8 YZ |
398 | if (ret) |
399 | goto out; | |
400 | } | |
5b21f2ed | 401 | |
13a8a7c8 YZ |
402 | if (unlikely(root->highest_objectid >= BTRFS_LAST_FREE_OBJECTID)) { |
403 | ret = -ENOSPC; | |
404 | goto out; | |
9f5fae2f | 405 | } |
13a8a7c8 YZ |
406 | |
407 | *objectid = ++root->highest_objectid; | |
408 | ret = 0; | |
409 | out: | |
a2135011 | 410 | mutex_unlock(&root->objectid_mutex); |
9f5fae2f CM |
411 | return ret; |
412 | } |