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 
caching_kthread(void * data)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 	if (!btrfs_test_opt(root, INODE_MAP_CACHE))
42 		return 0;
43 
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 		if (btrfs_fs_closing(fs_info))
66 			goto out;
67 
68 		leaf = path->nodes[0];
69 		slot = path->slots[0];
70 		if (slot >= btrfs_header_nritems(leaf)) {
71 			ret = btrfs_next_leaf(root, path);
72 			if (ret < 0)
73 				goto out;
74 			else if (ret > 0)
75 				break;
76 
77 			if (need_resched() ||
78 			    btrfs_transaction_in_commit(fs_info)) {
79 				leaf = path->nodes[0];
80 
81 				if (btrfs_header_nritems(leaf) == 0) {
82 					WARN_ON(1);
83 					break;
84 				}
85 
86 				/*
87 				 * Save the key so we can advances forward
88 				 * in the next search.
89 				 */
90 				btrfs_item_key_to_cpu(leaf, &key, 0);
91 				btrfs_release_path(path);
92 				root->cache_progress = last;
93 				mutex_unlock(&root->fs_commit_mutex);
94 				schedule_timeout(1);
95 				goto again;
96 			} else
97 				continue;
98 		}
99 
100 		btrfs_item_key_to_cpu(leaf, &key, slot);
101 
102 		if (key.type != BTRFS_INODE_ITEM_KEY)
103 			goto next;
104 
105 		if (key.objectid >= root->highest_objectid)
106 			break;
107 
108 		if (last != (u64)-1 && last + 1 != key.objectid) {
109 			__btrfs_add_free_space(ctl, last + 1,
110 					       key.objectid - last - 1);
111 			wake_up(&root->cache_wait);
112 		}
113 
114 		last = key.objectid;
115 next:
116 		path->slots[0]++;
117 	}
118 
119 	if (last < root->highest_objectid - 1) {
120 		__btrfs_add_free_space(ctl, last + 1,
121 				       root->highest_objectid - last - 1);
122 	}
123 
124 	spin_lock(&root->cache_lock);
125 	root->cached = BTRFS_CACHE_FINISHED;
126 	spin_unlock(&root->cache_lock);
127 
128 	root->cache_progress = (u64)-1;
129 	btrfs_unpin_free_ino(root);
130 out:
131 	wake_up(&root->cache_wait);
132 	mutex_unlock(&root->fs_commit_mutex);
133 
134 	btrfs_free_path(path);
135 
136 	return ret;
137 }
138 
start_caching(struct btrfs_root * root)139 static void start_caching(struct btrfs_root *root)
140 {
141 	struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
142 	struct task_struct *tsk;
143 	int ret;
144 	u64 objectid;
145 
146 	if (!btrfs_test_opt(root, INODE_MAP_CACHE))
147 		return;
148 
149 	spin_lock(&root->cache_lock);
150 	if (root->cached != BTRFS_CACHE_NO) {
151 		spin_unlock(&root->cache_lock);
152 		return;
153 	}
154 
155 	root->cached = BTRFS_CACHE_STARTED;
156 	spin_unlock(&root->cache_lock);
157 
158 	ret = load_free_ino_cache(root->fs_info, root);
159 	if (ret == 1) {
160 		spin_lock(&root->cache_lock);
161 		root->cached = BTRFS_CACHE_FINISHED;
162 		spin_unlock(&root->cache_lock);
163 		return;
164 	}
165 
166 	/*
167 	 * It can be quite time-consuming to fill the cache by searching
168 	 * through the extent tree, and this can keep ino allocation path
169 	 * waiting. Therefore at start we quickly find out the highest
170 	 * inode number and we know we can use inode numbers which fall in
171 	 * [highest_ino + 1, BTRFS_LAST_FREE_OBJECTID].
172 	 */
173 	ret = btrfs_find_free_objectid(root, &objectid);
174 	if (!ret && objectid <= BTRFS_LAST_FREE_OBJECTID) {
175 		__btrfs_add_free_space(ctl, objectid,
176 				       BTRFS_LAST_FREE_OBJECTID - objectid + 1);
177 	}
178 
179 	tsk = kthread_run(caching_kthread, root, "btrfs-ino-cache-%llu\n",
180 			  root->root_key.objectid);
181 	BUG_ON(IS_ERR(tsk)); /* -ENOMEM */
182 }
183 
btrfs_find_free_ino(struct btrfs_root * root,u64 * objectid)184 int btrfs_find_free_ino(struct btrfs_root *root, u64 *objectid)
185 {
186 	if (!btrfs_test_opt(root, INODE_MAP_CACHE))
187 		return btrfs_find_free_objectid(root, objectid);
188 
189 again:
190 	*objectid = btrfs_find_ino_for_alloc(root);
191 
192 	if (*objectid != 0)
193 		return 0;
194 
195 	start_caching(root);
196 
197 	wait_event(root->cache_wait,
198 		   root->cached == BTRFS_CACHE_FINISHED ||
199 		   root->free_ino_ctl->free_space > 0);
200 
201 	if (root->cached == BTRFS_CACHE_FINISHED &&
202 	    root->free_ino_ctl->free_space == 0)
203 		return -ENOSPC;
204 	else
205 		goto again;
206 }
207 
btrfs_return_ino(struct btrfs_root * root,u64 objectid)208 void btrfs_return_ino(struct btrfs_root *root, u64 objectid)
209 {
210 	struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
211 	struct btrfs_free_space_ctl *pinned = root->free_ino_pinned;
212 
213 	if (!btrfs_test_opt(root, INODE_MAP_CACHE))
214 		return;
215 
216 again:
217 	if (root->cached == BTRFS_CACHE_FINISHED) {
218 		__btrfs_add_free_space(ctl, objectid, 1);
219 	} else {
220 		/*
221 		 * If we are in the process of caching free ino chunks,
222 		 * to avoid adding the same inode number to the free_ino
223 		 * tree twice due to cross transaction, we'll leave it
224 		 * in the pinned tree until a transaction is committed
225 		 * or the caching work is done.
226 		 */
227 
228 		mutex_lock(&root->fs_commit_mutex);
229 		spin_lock(&root->cache_lock);
230 		if (root->cached == BTRFS_CACHE_FINISHED) {
231 			spin_unlock(&root->cache_lock);
232 			mutex_unlock(&root->fs_commit_mutex);
233 			goto again;
234 		}
235 		spin_unlock(&root->cache_lock);
236 
237 		start_caching(root);
238 
239 		if (objectid <= root->cache_progress ||
240 		    objectid > root->highest_objectid)
241 			__btrfs_add_free_space(ctl, objectid, 1);
242 		else
243 			__btrfs_add_free_space(pinned, objectid, 1);
244 
245 		mutex_unlock(&root->fs_commit_mutex);
246 	}
247 }
248 
249 /*
250  * When a transaction is committed, we'll move those inode numbers which
251  * are smaller than root->cache_progress from pinned tree to free_ino tree,
252  * and others will just be dropped, because the commit root we were
253  * searching has changed.
254  *
255  * Must be called with root->fs_commit_mutex held
256  */
btrfs_unpin_free_ino(struct btrfs_root * root)257 void btrfs_unpin_free_ino(struct btrfs_root *root)
258 {
259 	struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
260 	struct rb_root *rbroot = &root->free_ino_pinned->free_space_offset;
261 	struct btrfs_free_space *info;
262 	struct rb_node *n;
263 	u64 count;
264 
265 	if (!btrfs_test_opt(root, INODE_MAP_CACHE))
266 		return;
267 
268 	while (1) {
269 		n = rb_first(rbroot);
270 		if (!n)
271 			break;
272 
273 		info = rb_entry(n, struct btrfs_free_space, offset_index);
274 		BUG_ON(info->bitmap); /* Logic error */
275 
276 		if (info->offset > root->cache_progress)
277 			goto free;
278 		else if (info->offset + info->bytes > root->cache_progress)
279 			count = root->cache_progress - info->offset + 1;
280 		else
281 			count = info->bytes;
282 
283 		__btrfs_add_free_space(ctl, info->offset, count);
284 free:
285 		rb_erase(&info->offset_index, rbroot);
286 		kfree(info);
287 	}
288 }
289 
290 #define INIT_THRESHOLD	(((1024 * 32) / 2) / sizeof(struct btrfs_free_space))
291 #define INODES_PER_BITMAP (PAGE_CACHE_SIZE * 8)
292 
293 /*
294  * The goal is to keep the memory used by the free_ino tree won't
295  * exceed the memory if we use bitmaps only.
296  */
recalculate_thresholds(struct btrfs_free_space_ctl * ctl)297 static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
298 {
299 	struct btrfs_free_space *info;
300 	struct rb_node *n;
301 	int max_ino;
302 	int max_bitmaps;
303 
304 	n = rb_last(&ctl->free_space_offset);
305 	if (!n) {
306 		ctl->extents_thresh = INIT_THRESHOLD;
307 		return;
308 	}
309 	info = rb_entry(n, struct btrfs_free_space, offset_index);
310 
311 	/*
312 	 * Find the maximum inode number in the filesystem. Note we
313 	 * ignore the fact that this can be a bitmap, because we are
314 	 * not doing precise calculation.
315 	 */
316 	max_ino = info->bytes - 1;
317 
318 	max_bitmaps = ALIGN(max_ino, INODES_PER_BITMAP) / INODES_PER_BITMAP;
319 	if (max_bitmaps <= ctl->total_bitmaps) {
320 		ctl->extents_thresh = 0;
321 		return;
322 	}
323 
324 	ctl->extents_thresh = (max_bitmaps - ctl->total_bitmaps) *
325 				PAGE_CACHE_SIZE / sizeof(*info);
326 }
327 
328 /*
329  * We don't fall back to bitmap, if we are below the extents threshold
330  * or this chunk of inode numbers is a big one.
331  */
use_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info)332 static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
333 		       struct btrfs_free_space *info)
334 {
335 	if (ctl->free_extents < ctl->extents_thresh ||
336 	    info->bytes > INODES_PER_BITMAP / 10)
337 		return false;
338 
339 	return true;
340 }
341 
342 static struct btrfs_free_space_op free_ino_op = {
343 	.recalc_thresholds	= recalculate_thresholds,
344 	.use_bitmap		= use_bitmap,
345 };
346 
pinned_recalc_thresholds(struct btrfs_free_space_ctl * ctl)347 static void pinned_recalc_thresholds(struct btrfs_free_space_ctl *ctl)
348 {
349 }
350 
pinned_use_bitmap(struct btrfs_free_space_ctl * ctl,struct btrfs_free_space * info)351 static bool pinned_use_bitmap(struct btrfs_free_space_ctl *ctl,
352 			      struct btrfs_free_space *info)
353 {
354 	/*
355 	 * We always use extents for two reasons:
356 	 *
357 	 * - The pinned tree is only used during the process of caching
358 	 *   work.
359 	 * - Make code simpler. See btrfs_unpin_free_ino().
360 	 */
361 	return false;
362 }
363 
364 static struct btrfs_free_space_op pinned_free_ino_op = {
365 	.recalc_thresholds	= pinned_recalc_thresholds,
366 	.use_bitmap		= pinned_use_bitmap,
367 };
368 
btrfs_init_free_ino_ctl(struct btrfs_root * root)369 void btrfs_init_free_ino_ctl(struct btrfs_root *root)
370 {
371 	struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
372 	struct btrfs_free_space_ctl *pinned = root->free_ino_pinned;
373 
374 	spin_lock_init(&ctl->tree_lock);
375 	ctl->unit = 1;
376 	ctl->start = 0;
377 	ctl->private = NULL;
378 	ctl->op = &free_ino_op;
379 
380 	/*
381 	 * Initially we allow to use 16K of ram to cache chunks of
382 	 * inode numbers before we resort to bitmaps. This is somewhat
383 	 * arbitrary, but it will be adjusted in runtime.
384 	 */
385 	ctl->extents_thresh = INIT_THRESHOLD;
386 
387 	spin_lock_init(&pinned->tree_lock);
388 	pinned->unit = 1;
389 	pinned->start = 0;
390 	pinned->private = NULL;
391 	pinned->extents_thresh = 0;
392 	pinned->op = &pinned_free_ino_op;
393 }
394 
btrfs_save_ino_cache(struct btrfs_root * root,struct btrfs_trans_handle * trans)395 int btrfs_save_ino_cache(struct btrfs_root *root,
396 			 struct btrfs_trans_handle *trans)
397 {
398 	struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
399 	struct btrfs_path *path;
400 	struct inode *inode;
401 	struct btrfs_block_rsv *rsv;
402 	u64 num_bytes;
403 	u64 alloc_hint = 0;
404 	int ret;
405 	int prealloc;
406 	bool retry = false;
407 
408 	/* only fs tree and subvol/snap needs ino cache */
409 	if (root->root_key.objectid != BTRFS_FS_TREE_OBJECTID &&
410 	    (root->root_key.objectid < BTRFS_FIRST_FREE_OBJECTID ||
411 	     root->root_key.objectid > BTRFS_LAST_FREE_OBJECTID))
412 		return 0;
413 
414 	/* Don't save inode cache if we are deleting this root */
415 	if (btrfs_root_refs(&root->root_item) == 0 &&
416 	    root != root->fs_info->tree_root)
417 		return 0;
418 
419 	if (!btrfs_test_opt(root, INODE_MAP_CACHE))
420 		return 0;
421 
422 	path = btrfs_alloc_path();
423 	if (!path)
424 		return -ENOMEM;
425 
426 	rsv = trans->block_rsv;
427 	trans->block_rsv = &root->fs_info->trans_block_rsv;
428 
429 	num_bytes = trans->bytes_reserved;
430 	/*
431 	 * 1 item for inode item insertion if need
432 	 * 3 items for inode item update (in the worst case)
433 	 * 1 item for free space object
434 	 * 3 items for pre-allocation
435 	 */
436 	trans->bytes_reserved = btrfs_calc_trans_metadata_size(root, 8);
437 	ret = btrfs_block_rsv_add_noflush(root, trans->block_rsv,
438 					  trans->bytes_reserved);
439 	if (ret)
440 		goto out;
441 	trace_btrfs_space_reservation(root->fs_info, "ino_cache",
442 				      trans->transid, trans->bytes_reserved, 1);
443 again:
444 	inode = lookup_free_ino_inode(root, path);
445 	if (IS_ERR(inode) && (PTR_ERR(inode) != -ENOENT || retry)) {
446 		ret = PTR_ERR(inode);
447 		goto out_release;
448 	}
449 
450 	if (IS_ERR(inode)) {
451 		BUG_ON(retry); /* Logic error */
452 		retry = true;
453 
454 		ret = create_free_ino_inode(root, trans, path);
455 		if (ret)
456 			goto out_release;
457 		goto again;
458 	}
459 
460 	BTRFS_I(inode)->generation = 0;
461 	ret = btrfs_update_inode(trans, root, inode);
462 	if (ret) {
463 		btrfs_abort_transaction(trans, root, ret);
464 		goto out_put;
465 	}
466 
467 	if (i_size_read(inode) > 0) {
468 		ret = btrfs_truncate_free_space_cache(root, trans, path, inode);
469 		if (ret) {
470 			btrfs_abort_transaction(trans, root, ret);
471 			goto out_put;
472 		}
473 	}
474 
475 	spin_lock(&root->cache_lock);
476 	if (root->cached != BTRFS_CACHE_FINISHED) {
477 		ret = -1;
478 		spin_unlock(&root->cache_lock);
479 		goto out_put;
480 	}
481 	spin_unlock(&root->cache_lock);
482 
483 	spin_lock(&ctl->tree_lock);
484 	prealloc = sizeof(struct btrfs_free_space) * ctl->free_extents;
485 	prealloc = ALIGN(prealloc, PAGE_CACHE_SIZE);
486 	prealloc += ctl->total_bitmaps * PAGE_CACHE_SIZE;
487 	spin_unlock(&ctl->tree_lock);
488 
489 	/* Just to make sure we have enough space */
490 	prealloc += 8 * PAGE_CACHE_SIZE;
491 
492 	ret = btrfs_delalloc_reserve_space(inode, prealloc);
493 	if (ret)
494 		goto out_put;
495 
496 	ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, prealloc,
497 					      prealloc, prealloc, &alloc_hint);
498 	if (ret) {
499 		btrfs_delalloc_release_space(inode, prealloc);
500 		goto out_put;
501 	}
502 	btrfs_free_reserved_data_space(inode, prealloc);
503 
504 	ret = btrfs_write_out_ino_cache(root, trans, path);
505 out_put:
506 	iput(inode);
507 out_release:
508 	trace_btrfs_space_reservation(root->fs_info, "ino_cache",
509 				      trans->transid, trans->bytes_reserved, 0);
510 	btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
511 out:
512 	trans->block_rsv = rsv;
513 	trans->bytes_reserved = num_bytes;
514 
515 	btrfs_free_path(path);
516 	return ret;
517 }
518 
btrfs_find_highest_objectid(struct btrfs_root * root,u64 * objectid)519 static int btrfs_find_highest_objectid(struct btrfs_root *root, u64 *objectid)
520 {
521 	struct btrfs_path *path;
522 	int ret;
523 	struct extent_buffer *l;
524 	struct btrfs_key search_key;
525 	struct btrfs_key found_key;
526 	int slot;
527 
528 	path = btrfs_alloc_path();
529 	if (!path)
530 		return -ENOMEM;
531 
532 	search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
533 	search_key.type = -1;
534 	search_key.offset = (u64)-1;
535 	ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
536 	if (ret < 0)
537 		goto error;
538 	BUG_ON(ret == 0); /* Corruption */
539 	if (path->slots[0] > 0) {
540 		slot = path->slots[0] - 1;
541 		l = path->nodes[0];
542 		btrfs_item_key_to_cpu(l, &found_key, slot);
543 		*objectid = max_t(u64, found_key.objectid,
544 				  BTRFS_FIRST_FREE_OBJECTID - 1);
545 	} else {
546 		*objectid = BTRFS_FIRST_FREE_OBJECTID - 1;
547 	}
548 	ret = 0;
549 error:
550 	btrfs_free_path(path);
551 	return ret;
552 }
553 
btrfs_find_free_objectid(struct btrfs_root * root,u64 * objectid)554 int btrfs_find_free_objectid(struct btrfs_root *root, u64 *objectid)
555 {
556 	int ret;
557 	mutex_lock(&root->objectid_mutex);
558 
559 	if (unlikely(root->highest_objectid < BTRFS_FIRST_FREE_OBJECTID)) {
560 		ret = btrfs_find_highest_objectid(root,
561 						  &root->highest_objectid);
562 		if (ret)
563 			goto out;
564 	}
565 
566 	if (unlikely(root->highest_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
567 		ret = -ENOSPC;
568 		goto out;
569 	}
570 
571 	*objectid = ++root->highest_objectid;
572 	ret = 0;
573 out:
574 	mutex_unlock(&root->objectid_mutex);
575 	return ret;
576 }
577