Line data Source code
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 :
19 : #include <linux/delay.h>
20 : #include <linux/kthread.h>
21 : #include <linux/pagemap.h>
22 :
23 : #include "ctree.h"
24 : #include "disk-io.h"
25 : #include "free-space-cache.h"
26 : #include "inode-map.h"
27 : #include "transaction.h"
28 :
29 0 : static int caching_kthread(void *data)
30 : {
31 : struct btrfs_root *root = data;
32 0 : struct btrfs_fs_info *fs_info = root->fs_info;
33 0 : struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
34 : struct btrfs_key key;
35 : struct btrfs_path *path;
36 0 : struct extent_buffer *leaf;
37 : u64 last = (u64)-1;
38 : int slot;
39 : int ret;
40 :
41 0 : if (!btrfs_test_opt(root, INODE_MAP_CACHE))
42 : return 0;
43 :
44 0 : path = btrfs_alloc_path();
45 0 : if (!path)
46 : return -ENOMEM;
47 :
48 : /* Since the commit root is read-only, we can safely skip locking. */
49 0 : path->skip_locking = 1;
50 0 : path->search_commit_root = 1;
51 0 : path->reada = 2;
52 :
53 0 : key.objectid = BTRFS_FIRST_FREE_OBJECTID;
54 0 : key.offset = 0;
55 0 : key.type = BTRFS_INODE_ITEM_KEY;
56 : again:
57 : /* need to make sure the commit_root doesn't disappear */
58 0 : down_read(&fs_info->commit_root_sem);
59 :
60 0 : ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
61 0 : if (ret < 0)
62 : goto out;
63 :
64 : while (1) {
65 0 : if (btrfs_fs_closing(fs_info))
66 : goto out;
67 :
68 0 : leaf = path->nodes[0];
69 0 : slot = path->slots[0];
70 0 : if (slot >= btrfs_header_nritems(leaf)) {
71 0 : ret = btrfs_next_leaf(root, path);
72 0 : if (ret < 0)
73 : goto out;
74 0 : else if (ret > 0)
75 : break;
76 :
77 0 : if (need_resched() ||
78 0 : btrfs_transaction_in_commit(fs_info)) {
79 0 : leaf = path->nodes[0];
80 :
81 0 : if (WARN_ON(btrfs_header_nritems(leaf) == 0))
82 : break;
83 :
84 : /*
85 : * Save the key so we can advances forward
86 : * in the next search.
87 : */
88 0 : btrfs_item_key_to_cpu(leaf, &key, 0);
89 0 : btrfs_release_path(path);
90 0 : root->cache_progress = last;
91 0 : up_read(&fs_info->commit_root_sem);
92 0 : schedule_timeout(1);
93 0 : goto again;
94 : } else
95 0 : continue;
96 : }
97 :
98 0 : btrfs_item_key_to_cpu(leaf, &key, slot);
99 :
100 0 : if (key.type != BTRFS_INODE_ITEM_KEY)
101 : goto next;
102 :
103 0 : if (key.objectid >= root->highest_objectid)
104 : break;
105 :
106 0 : if (last != (u64)-1 && last + 1 != key.objectid) {
107 0 : __btrfs_add_free_space(ctl, last + 1,
108 0 : key.objectid - last - 1);
109 0 : wake_up(&root->cache_wait);
110 : }
111 :
112 0 : last = key.objectid;
113 : next:
114 0 : path->slots[0]++;
115 : }
116 :
117 0 : if (last < root->highest_objectid - 1) {
118 0 : __btrfs_add_free_space(ctl, last + 1,
119 0 : root->highest_objectid - last - 1);
120 : }
121 :
122 : spin_lock(&root->cache_lock);
123 0 : root->cached = BTRFS_CACHE_FINISHED;
124 : spin_unlock(&root->cache_lock);
125 :
126 0 : root->cache_progress = (u64)-1;
127 0 : btrfs_unpin_free_ino(root);
128 : out:
129 0 : wake_up(&root->cache_wait);
130 0 : up_read(&fs_info->commit_root_sem);
131 :
132 0 : btrfs_free_path(path);
133 :
134 0 : return ret;
135 : }
136 :
137 0 : static void start_caching(struct btrfs_root *root)
138 : {
139 0 : struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
140 : struct task_struct *tsk;
141 : int ret;
142 : u64 objectid;
143 :
144 0 : if (!btrfs_test_opt(root, INODE_MAP_CACHE))
145 0 : return;
146 :
147 : spin_lock(&root->cache_lock);
148 0 : if (root->cached != BTRFS_CACHE_NO) {
149 : spin_unlock(&root->cache_lock);
150 : return;
151 : }
152 :
153 0 : root->cached = BTRFS_CACHE_STARTED;
154 : spin_unlock(&root->cache_lock);
155 :
156 0 : ret = load_free_ino_cache(root->fs_info, root);
157 0 : if (ret == 1) {
158 : spin_lock(&root->cache_lock);
159 0 : root->cached = BTRFS_CACHE_FINISHED;
160 : spin_unlock(&root->cache_lock);
161 : return;
162 : }
163 :
164 : /*
165 : * It can be quite time-consuming to fill the cache by searching
166 : * through the extent tree, and this can keep ino allocation path
167 : * waiting. Therefore at start we quickly find out the highest
168 : * inode number and we know we can use inode numbers which fall in
169 : * [highest_ino + 1, BTRFS_LAST_FREE_OBJECTID].
170 : */
171 0 : ret = btrfs_find_free_objectid(root, &objectid);
172 0 : if (!ret && objectid <= BTRFS_LAST_FREE_OBJECTID) {
173 0 : __btrfs_add_free_space(ctl, objectid,
174 : BTRFS_LAST_FREE_OBJECTID - objectid + 1);
175 : }
176 :
177 0 : tsk = kthread_run(caching_kthread, root, "btrfs-ino-cache-%llu",
178 : root->root_key.objectid);
179 0 : if (IS_ERR(tsk)) {
180 0 : btrfs_warn(root->fs_info, "failed to start inode caching task");
181 0 : btrfs_clear_and_info(root, CHANGE_INODE_CACHE,
182 : "disabling inode map caching");
183 : }
184 : }
185 :
186 20423 : int btrfs_find_free_ino(struct btrfs_root *root, u64 *objectid)
187 : {
188 20423 : if (!btrfs_test_opt(root, INODE_MAP_CACHE))
189 20423 : return btrfs_find_free_objectid(root, objectid);
190 :
191 : again:
192 0 : *objectid = btrfs_find_ino_for_alloc(root);
193 :
194 0 : if (*objectid != 0)
195 : return 0;
196 :
197 0 : start_caching(root);
198 :
199 0 : wait_event(root->cache_wait,
200 : root->cached == BTRFS_CACHE_FINISHED ||
201 : root->free_ino_ctl->free_space > 0);
202 :
203 0 : if (root->cached == BTRFS_CACHE_FINISHED &&
204 0 : root->free_ino_ctl->free_space == 0)
205 : return -ENOSPC;
206 : else
207 : goto again;
208 : }
209 :
210 6866 : void btrfs_return_ino(struct btrfs_root *root, u64 objectid)
211 : {
212 6866 : struct btrfs_free_space_ctl *pinned = root->free_ino_pinned;
213 :
214 6866 : if (!btrfs_test_opt(root, INODE_MAP_CACHE))
215 6866 : return;
216 : again:
217 0 : if (root->cached == BTRFS_CACHE_FINISHED) {
218 0 : __btrfs_add_free_space(pinned, objectid, 1);
219 : } else {
220 0 : down_write(&root->fs_info->commit_root_sem);
221 : spin_lock(&root->cache_lock);
222 0 : if (root->cached == BTRFS_CACHE_FINISHED) {
223 : spin_unlock(&root->cache_lock);
224 0 : up_write(&root->fs_info->commit_root_sem);
225 0 : goto again;
226 : }
227 : spin_unlock(&root->cache_lock);
228 :
229 0 : start_caching(root);
230 :
231 0 : __btrfs_add_free_space(pinned, objectid, 1);
232 :
233 0 : up_write(&root->fs_info->commit_root_sem);
234 : }
235 : }
236 :
237 : /*
238 : * When a transaction is committed, we'll move those inode numbers which
239 : * are smaller than root->cache_progress from pinned tree to free_ino tree,
240 : * and others will just be dropped, because the commit root we were
241 : * searching has changed.
242 : *
243 : * Must be called with root->fs_info->commit_root_sem held
244 : */
245 1888 : void btrfs_unpin_free_ino(struct btrfs_root *root)
246 : {
247 1888 : struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
248 1888 : struct rb_root *rbroot = &root->free_ino_pinned->free_space_offset;
249 : struct btrfs_free_space *info;
250 : struct rb_node *n;
251 : u64 count;
252 :
253 1888 : if (!btrfs_test_opt(root, INODE_MAP_CACHE))
254 1888 : return;
255 :
256 : while (1) {
257 0 : n = rb_first(rbroot);
258 0 : if (!n)
259 : break;
260 :
261 : info = rb_entry(n, struct btrfs_free_space, offset_index);
262 0 : BUG_ON(info->bitmap); /* Logic error */
263 :
264 0 : if (info->offset > root->cache_progress)
265 : goto free;
266 0 : else if (info->offset + info->bytes > root->cache_progress)
267 0 : count = root->cache_progress - info->offset + 1;
268 : else
269 : count = info->bytes;
270 :
271 0 : __btrfs_add_free_space(ctl, info->offset, count);
272 : free:
273 0 : rb_erase(&info->offset_index, rbroot);
274 0 : kfree(info);
275 0 : }
276 : }
277 :
278 : #define INIT_THRESHOLD (((1024 * 32) / 2) / sizeof(struct btrfs_free_space))
279 : #define INODES_PER_BITMAP (PAGE_CACHE_SIZE * 8)
280 :
281 : /*
282 : * The goal is to keep the memory used by the free_ino tree won't
283 : * exceed the memory if we use bitmaps only.
284 : */
285 0 : static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
286 : {
287 : struct btrfs_free_space *info;
288 : struct rb_node *n;
289 : int max_ino;
290 : int max_bitmaps;
291 :
292 0 : n = rb_last(&ctl->free_space_offset);
293 0 : if (!n) {
294 0 : ctl->extents_thresh = INIT_THRESHOLD;
295 0 : return;
296 : }
297 : info = rb_entry(n, struct btrfs_free_space, offset_index);
298 :
299 : /*
300 : * Find the maximum inode number in the filesystem. Note we
301 : * ignore the fact that this can be a bitmap, because we are
302 : * not doing precise calculation.
303 : */
304 0 : max_ino = info->bytes - 1;
305 :
306 0 : max_bitmaps = ALIGN(max_ino, INODES_PER_BITMAP) / INODES_PER_BITMAP;
307 0 : if (max_bitmaps <= ctl->total_bitmaps) {
308 0 : ctl->extents_thresh = 0;
309 0 : return;
310 : }
311 :
312 0 : ctl->extents_thresh = (max_bitmaps - ctl->total_bitmaps) *
313 0 : PAGE_CACHE_SIZE / sizeof(*info);
314 : }
315 :
316 : /*
317 : * We don't fall back to bitmap, if we are below the extents threshold
318 : * or this chunk of inode numbers is a big one.
319 : */
320 0 : static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
321 : struct btrfs_free_space *info)
322 : {
323 0 : if (ctl->free_extents < ctl->extents_thresh ||
324 0 : info->bytes > INODES_PER_BITMAP / 10)
325 : return false;
326 :
327 0 : return true;
328 : }
329 :
330 : static struct btrfs_free_space_op free_ino_op = {
331 : .recalc_thresholds = recalculate_thresholds,
332 : .use_bitmap = use_bitmap,
333 : };
334 :
335 0 : static void pinned_recalc_thresholds(struct btrfs_free_space_ctl *ctl)
336 : {
337 0 : }
338 :
339 0 : static bool pinned_use_bitmap(struct btrfs_free_space_ctl *ctl,
340 : struct btrfs_free_space *info)
341 : {
342 : /*
343 : * We always use extents for two reasons:
344 : *
345 : * - The pinned tree is only used during the process of caching
346 : * work.
347 : * - Make code simpler. See btrfs_unpin_free_ino().
348 : */
349 0 : return false;
350 : }
351 :
352 : static struct btrfs_free_space_op pinned_free_ino_op = {
353 : .recalc_thresholds = pinned_recalc_thresholds,
354 : .use_bitmap = pinned_use_bitmap,
355 : };
356 :
357 688 : void btrfs_init_free_ino_ctl(struct btrfs_root *root)
358 : {
359 688 : struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
360 688 : struct btrfs_free_space_ctl *pinned = root->free_ino_pinned;
361 :
362 688 : spin_lock_init(&ctl->tree_lock);
363 688 : ctl->unit = 1;
364 688 : ctl->start = 0;
365 688 : ctl->private = NULL;
366 688 : ctl->op = &free_ino_op;
367 :
368 : /*
369 : * Initially we allow to use 16K of ram to cache chunks of
370 : * inode numbers before we resort to bitmaps. This is somewhat
371 : * arbitrary, but it will be adjusted in runtime.
372 : */
373 688 : ctl->extents_thresh = INIT_THRESHOLD;
374 :
375 688 : spin_lock_init(&pinned->tree_lock);
376 688 : pinned->unit = 1;
377 688 : pinned->start = 0;
378 688 : pinned->private = NULL;
379 688 : pinned->extents_thresh = 0;
380 688 : pinned->op = &pinned_free_ino_op;
381 688 : }
382 :
383 2548 : int btrfs_save_ino_cache(struct btrfs_root *root,
384 : struct btrfs_trans_handle *trans)
385 : {
386 2548 : struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
387 : struct btrfs_path *path;
388 0 : struct inode *inode;
389 : struct btrfs_block_rsv *rsv;
390 : u64 num_bytes;
391 2548 : u64 alloc_hint = 0;
392 : int ret;
393 : int prealloc;
394 : bool retry = false;
395 :
396 : /* only fs tree and subvol/snap needs ino cache */
397 2548 : if (root->root_key.objectid != BTRFS_FS_TREE_OBJECTID &&
398 : (root->root_key.objectid < BTRFS_FIRST_FREE_OBJECTID ||
399 : root->root_key.objectid > BTRFS_LAST_FREE_OBJECTID))
400 : return 0;
401 :
402 : /* Don't save inode cache if we are deleting this root */
403 2424 : if (btrfs_root_refs(&root->root_item) == 0)
404 : return 0;
405 :
406 1620 : if (!btrfs_test_opt(root, INODE_MAP_CACHE))
407 : return 0;
408 :
409 0 : path = btrfs_alloc_path();
410 0 : if (!path)
411 : return -ENOMEM;
412 :
413 0 : rsv = trans->block_rsv;
414 0 : trans->block_rsv = &root->fs_info->trans_block_rsv;
415 :
416 0 : num_bytes = trans->bytes_reserved;
417 : /*
418 : * 1 item for inode item insertion if need
419 : * 4 items for inode item update (in the worst case)
420 : * 1 items for slack space if we need do truncation
421 : * 1 item for free space object
422 : * 3 items for pre-allocation
423 : */
424 0 : trans->bytes_reserved = btrfs_calc_trans_metadata_size(root, 10);
425 0 : ret = btrfs_block_rsv_add(root, trans->block_rsv,
426 : trans->bytes_reserved,
427 : BTRFS_RESERVE_NO_FLUSH);
428 0 : if (ret)
429 : goto out;
430 0 : trace_btrfs_space_reservation(root->fs_info, "ino_cache",
431 : trans->transid, trans->bytes_reserved, 1);
432 : again:
433 0 : inode = lookup_free_ino_inode(root, path);
434 0 : if (IS_ERR(inode) && (PTR_ERR(inode) != -ENOENT || retry)) {
435 0 : ret = PTR_ERR(inode);
436 0 : goto out_release;
437 : }
438 :
439 0 : if (IS_ERR(inode)) {
440 0 : BUG_ON(retry); /* Logic error */
441 : retry = true;
442 :
443 0 : ret = create_free_ino_inode(root, trans, path);
444 0 : if (ret)
445 : goto out_release;
446 : goto again;
447 : }
448 :
449 0 : BTRFS_I(inode)->generation = 0;
450 0 : ret = btrfs_update_inode(trans, root, inode);
451 0 : if (ret) {
452 0 : btrfs_abort_transaction(trans, root, ret);
453 0 : goto out_put;
454 : }
455 :
456 0 : if (i_size_read(inode) > 0) {
457 0 : ret = btrfs_truncate_free_space_cache(root, trans, inode);
458 0 : if (ret) {
459 0 : if (ret != -ENOSPC)
460 0 : btrfs_abort_transaction(trans, root, ret);
461 : goto out_put;
462 : }
463 : }
464 :
465 : spin_lock(&root->cache_lock);
466 0 : if (root->cached != BTRFS_CACHE_FINISHED) {
467 : ret = -1;
468 : spin_unlock(&root->cache_lock);
469 : goto out_put;
470 : }
471 : spin_unlock(&root->cache_lock);
472 :
473 : spin_lock(&ctl->tree_lock);
474 0 : prealloc = sizeof(struct btrfs_free_space) * ctl->free_extents;
475 0 : prealloc = ALIGN(prealloc, PAGE_CACHE_SIZE);
476 0 : prealloc += ctl->total_bitmaps * PAGE_CACHE_SIZE;
477 : spin_unlock(&ctl->tree_lock);
478 :
479 : /* Just to make sure we have enough space */
480 0 : prealloc += 8 * PAGE_CACHE_SIZE;
481 :
482 0 : ret = btrfs_delalloc_reserve_space(inode, prealloc);
483 0 : if (ret)
484 : goto out_put;
485 :
486 0 : ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, prealloc,
487 : prealloc, prealloc, &alloc_hint);
488 0 : if (ret) {
489 0 : btrfs_delalloc_release_space(inode, prealloc);
490 0 : goto out_put;
491 : }
492 0 : btrfs_free_reserved_data_space(inode, prealloc);
493 :
494 0 : ret = btrfs_write_out_ino_cache(root, trans, path, inode);
495 : out_put:
496 0 : iput(inode);
497 : out_release:
498 0 : trace_btrfs_space_reservation(root->fs_info, "ino_cache",
499 : trans->transid, trans->bytes_reserved, 0);
500 0 : btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
501 : out:
502 0 : trans->block_rsv = rsv;
503 0 : trans->bytes_reserved = num_bytes;
504 :
505 0 : btrfs_free_path(path);
506 0 : return ret;
507 : }
508 :
509 238 : static int btrfs_find_highest_objectid(struct btrfs_root *root, u64 *objectid)
510 : {
511 : struct btrfs_path *path;
512 : int ret;
513 : struct extent_buffer *l;
514 : struct btrfs_key search_key;
515 : struct btrfs_key found_key;
516 : int slot;
517 :
518 238 : path = btrfs_alloc_path();
519 238 : if (!path)
520 : return -ENOMEM;
521 :
522 238 : search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
523 238 : search_key.type = -1;
524 238 : search_key.offset = (u64)-1;
525 238 : ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
526 238 : if (ret < 0)
527 : goto error;
528 238 : BUG_ON(ret == 0); /* Corruption */
529 238 : if (path->slots[0] > 0) {
530 238 : slot = path->slots[0] - 1;
531 238 : l = path->nodes[0];
532 238 : btrfs_item_key_to_cpu(l, &found_key, slot);
533 238 : *objectid = max_t(u64, found_key.objectid,
534 : BTRFS_FIRST_FREE_OBJECTID - 1);
535 : } else {
536 0 : *objectid = BTRFS_FIRST_FREE_OBJECTID - 1;
537 : }
538 : ret = 0;
539 : error:
540 238 : btrfs_free_path(path);
541 238 : return ret;
542 : }
543 :
544 20869 : int btrfs_find_free_objectid(struct btrfs_root *root, u64 *objectid)
545 : {
546 : int ret;
547 20869 : mutex_lock(&root->objectid_mutex);
548 :
549 20869 : if (unlikely(root->highest_objectid < BTRFS_FIRST_FREE_OBJECTID)) {
550 238 : ret = btrfs_find_highest_objectid(root,
551 : &root->highest_objectid);
552 238 : if (ret)
553 : goto out;
554 : }
555 :
556 20869 : if (unlikely(root->highest_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
557 : ret = -ENOSPC;
558 : goto out;
559 : }
560 :
561 20869 : *objectid = ++root->highest_objectid;
562 : ret = 0;
563 : out:
564 20869 : mutex_unlock(&root->objectid_mutex);
565 20869 : return ret;
566 : }
|
