xref: /openbmc/linux/fs/btrfs/delayed-ref.c (revision 609e478b)
1 /*
2  * Copyright (C) 2009 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/sched.h>
20 #include <linux/slab.h>
21 #include <linux/sort.h>
22 #include "ctree.h"
23 #include "delayed-ref.h"
24 #include "transaction.h"
25 
26 struct kmem_cache *btrfs_delayed_ref_head_cachep;
27 struct kmem_cache *btrfs_delayed_tree_ref_cachep;
28 struct kmem_cache *btrfs_delayed_data_ref_cachep;
29 struct kmem_cache *btrfs_delayed_extent_op_cachep;
30 /*
31  * delayed back reference update tracking.  For subvolume trees
32  * we queue up extent allocations and backref maintenance for
33  * delayed processing.   This avoids deep call chains where we
34  * add extents in the middle of btrfs_search_slot, and it allows
35  * us to buffer up frequently modified backrefs in an rb tree instead
36  * of hammering updates on the extent allocation tree.
37  */
38 
39 /*
40  * compare two delayed tree backrefs with same bytenr and type
41  */
42 static int comp_tree_refs(struct btrfs_delayed_tree_ref *ref2,
43 			  struct btrfs_delayed_tree_ref *ref1, int type)
44 {
45 	if (type == BTRFS_TREE_BLOCK_REF_KEY) {
46 		if (ref1->root < ref2->root)
47 			return -1;
48 		if (ref1->root > ref2->root)
49 			return 1;
50 	} else {
51 		if (ref1->parent < ref2->parent)
52 			return -1;
53 		if (ref1->parent > ref2->parent)
54 			return 1;
55 	}
56 	return 0;
57 }
58 
59 /*
60  * compare two delayed data backrefs with same bytenr and type
61  */
62 static int comp_data_refs(struct btrfs_delayed_data_ref *ref2,
63 			  struct btrfs_delayed_data_ref *ref1)
64 {
65 	if (ref1->node.type == BTRFS_EXTENT_DATA_REF_KEY) {
66 		if (ref1->root < ref2->root)
67 			return -1;
68 		if (ref1->root > ref2->root)
69 			return 1;
70 		if (ref1->objectid < ref2->objectid)
71 			return -1;
72 		if (ref1->objectid > ref2->objectid)
73 			return 1;
74 		if (ref1->offset < ref2->offset)
75 			return -1;
76 		if (ref1->offset > ref2->offset)
77 			return 1;
78 	} else {
79 		if (ref1->parent < ref2->parent)
80 			return -1;
81 		if (ref1->parent > ref2->parent)
82 			return 1;
83 	}
84 	return 0;
85 }
86 
87 /*
88  * entries in the rb tree are ordered by the byte number of the extent,
89  * type of the delayed backrefs and content of delayed backrefs.
90  */
91 static int comp_entry(struct btrfs_delayed_ref_node *ref2,
92 		      struct btrfs_delayed_ref_node *ref1,
93 		      bool compare_seq)
94 {
95 	if (ref1->bytenr < ref2->bytenr)
96 		return -1;
97 	if (ref1->bytenr > ref2->bytenr)
98 		return 1;
99 	if (ref1->is_head && ref2->is_head)
100 		return 0;
101 	if (ref2->is_head)
102 		return -1;
103 	if (ref1->is_head)
104 		return 1;
105 	if (ref1->type < ref2->type)
106 		return -1;
107 	if (ref1->type > ref2->type)
108 		return 1;
109 	if (ref1->no_quota > ref2->no_quota)
110 		return 1;
111 	if (ref1->no_quota < ref2->no_quota)
112 		return -1;
113 	/* merging of sequenced refs is not allowed */
114 	if (compare_seq) {
115 		if (ref1->seq < ref2->seq)
116 			return -1;
117 		if (ref1->seq > ref2->seq)
118 			return 1;
119 	}
120 	if (ref1->type == BTRFS_TREE_BLOCK_REF_KEY ||
121 	    ref1->type == BTRFS_SHARED_BLOCK_REF_KEY) {
122 		return comp_tree_refs(btrfs_delayed_node_to_tree_ref(ref2),
123 				      btrfs_delayed_node_to_tree_ref(ref1),
124 				      ref1->type);
125 	} else if (ref1->type == BTRFS_EXTENT_DATA_REF_KEY ||
126 		   ref1->type == BTRFS_SHARED_DATA_REF_KEY) {
127 		return comp_data_refs(btrfs_delayed_node_to_data_ref(ref2),
128 				      btrfs_delayed_node_to_data_ref(ref1));
129 	}
130 	BUG();
131 	return 0;
132 }
133 
134 /*
135  * insert a new ref into the rbtree.  This returns any existing refs
136  * for the same (bytenr,parent) tuple, or NULL if the new node was properly
137  * inserted.
138  */
139 static struct btrfs_delayed_ref_node *tree_insert(struct rb_root *root,
140 						  struct rb_node *node)
141 {
142 	struct rb_node **p = &root->rb_node;
143 	struct rb_node *parent_node = NULL;
144 	struct btrfs_delayed_ref_node *entry;
145 	struct btrfs_delayed_ref_node *ins;
146 	int cmp;
147 
148 	ins = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
149 	while (*p) {
150 		parent_node = *p;
151 		entry = rb_entry(parent_node, struct btrfs_delayed_ref_node,
152 				 rb_node);
153 
154 		cmp = comp_entry(entry, ins, 1);
155 		if (cmp < 0)
156 			p = &(*p)->rb_left;
157 		else if (cmp > 0)
158 			p = &(*p)->rb_right;
159 		else
160 			return entry;
161 	}
162 
163 	rb_link_node(node, parent_node, p);
164 	rb_insert_color(node, root);
165 	return NULL;
166 }
167 
168 /* insert a new ref to head ref rbtree */
169 static struct btrfs_delayed_ref_head *htree_insert(struct rb_root *root,
170 						   struct rb_node *node)
171 {
172 	struct rb_node **p = &root->rb_node;
173 	struct rb_node *parent_node = NULL;
174 	struct btrfs_delayed_ref_head *entry;
175 	struct btrfs_delayed_ref_head *ins;
176 	u64 bytenr;
177 
178 	ins = rb_entry(node, struct btrfs_delayed_ref_head, href_node);
179 	bytenr = ins->node.bytenr;
180 	while (*p) {
181 		parent_node = *p;
182 		entry = rb_entry(parent_node, struct btrfs_delayed_ref_head,
183 				 href_node);
184 
185 		if (bytenr < entry->node.bytenr)
186 			p = &(*p)->rb_left;
187 		else if (bytenr > entry->node.bytenr)
188 			p = &(*p)->rb_right;
189 		else
190 			return entry;
191 	}
192 
193 	rb_link_node(node, parent_node, p);
194 	rb_insert_color(node, root);
195 	return NULL;
196 }
197 
198 /*
199  * find an head entry based on bytenr. This returns the delayed ref
200  * head if it was able to find one, or NULL if nothing was in that spot.
201  * If return_bigger is given, the next bigger entry is returned if no exact
202  * match is found.
203  */
204 static struct btrfs_delayed_ref_head *
205 find_ref_head(struct rb_root *root, u64 bytenr,
206 	      int return_bigger)
207 {
208 	struct rb_node *n;
209 	struct btrfs_delayed_ref_head *entry;
210 
211 	n = root->rb_node;
212 	entry = NULL;
213 	while (n) {
214 		entry = rb_entry(n, struct btrfs_delayed_ref_head, href_node);
215 
216 		if (bytenr < entry->node.bytenr)
217 			n = n->rb_left;
218 		else if (bytenr > entry->node.bytenr)
219 			n = n->rb_right;
220 		else
221 			return entry;
222 	}
223 	if (entry && return_bigger) {
224 		if (bytenr > entry->node.bytenr) {
225 			n = rb_next(&entry->href_node);
226 			if (!n)
227 				n = rb_first(root);
228 			entry = rb_entry(n, struct btrfs_delayed_ref_head,
229 					 href_node);
230 			return entry;
231 		}
232 		return entry;
233 	}
234 	return NULL;
235 }
236 
237 int btrfs_delayed_ref_lock(struct btrfs_trans_handle *trans,
238 			   struct btrfs_delayed_ref_head *head)
239 {
240 	struct btrfs_delayed_ref_root *delayed_refs;
241 
242 	delayed_refs = &trans->transaction->delayed_refs;
243 	assert_spin_locked(&delayed_refs->lock);
244 	if (mutex_trylock(&head->mutex))
245 		return 0;
246 
247 	atomic_inc(&head->node.refs);
248 	spin_unlock(&delayed_refs->lock);
249 
250 	mutex_lock(&head->mutex);
251 	spin_lock(&delayed_refs->lock);
252 	if (!head->node.in_tree) {
253 		mutex_unlock(&head->mutex);
254 		btrfs_put_delayed_ref(&head->node);
255 		return -EAGAIN;
256 	}
257 	btrfs_put_delayed_ref(&head->node);
258 	return 0;
259 }
260 
261 static inline void drop_delayed_ref(struct btrfs_trans_handle *trans,
262 				    struct btrfs_delayed_ref_root *delayed_refs,
263 				    struct btrfs_delayed_ref_head *head,
264 				    struct btrfs_delayed_ref_node *ref)
265 {
266 	if (btrfs_delayed_ref_is_head(ref)) {
267 		head = btrfs_delayed_node_to_head(ref);
268 		rb_erase(&head->href_node, &delayed_refs->href_root);
269 	} else {
270 		assert_spin_locked(&head->lock);
271 		rb_erase(&ref->rb_node, &head->ref_root);
272 	}
273 	ref->in_tree = 0;
274 	btrfs_put_delayed_ref(ref);
275 	atomic_dec(&delayed_refs->num_entries);
276 	if (trans->delayed_ref_updates)
277 		trans->delayed_ref_updates--;
278 }
279 
280 static int merge_ref(struct btrfs_trans_handle *trans,
281 		     struct btrfs_delayed_ref_root *delayed_refs,
282 		     struct btrfs_delayed_ref_head *head,
283 		     struct btrfs_delayed_ref_node *ref, u64 seq)
284 {
285 	struct rb_node *node;
286 	int mod = 0;
287 	int done = 0;
288 
289 	node = rb_next(&ref->rb_node);
290 	while (!done && node) {
291 		struct btrfs_delayed_ref_node *next;
292 
293 		next = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
294 		node = rb_next(node);
295 		if (seq && next->seq >= seq)
296 			break;
297 		if (comp_entry(ref, next, 0))
298 			continue;
299 
300 		if (ref->action == next->action) {
301 			mod = next->ref_mod;
302 		} else {
303 			if (ref->ref_mod < next->ref_mod) {
304 				struct btrfs_delayed_ref_node *tmp;
305 
306 				tmp = ref;
307 				ref = next;
308 				next = tmp;
309 				done = 1;
310 			}
311 			mod = -next->ref_mod;
312 		}
313 
314 		drop_delayed_ref(trans, delayed_refs, head, next);
315 		ref->ref_mod += mod;
316 		if (ref->ref_mod == 0) {
317 			drop_delayed_ref(trans, delayed_refs, head, ref);
318 			done = 1;
319 		} else {
320 			/*
321 			 * You can't have multiples of the same ref on a tree
322 			 * block.
323 			 */
324 			WARN_ON(ref->type == BTRFS_TREE_BLOCK_REF_KEY ||
325 				ref->type == BTRFS_SHARED_BLOCK_REF_KEY);
326 		}
327 	}
328 	return done;
329 }
330 
331 void btrfs_merge_delayed_refs(struct btrfs_trans_handle *trans,
332 			      struct btrfs_fs_info *fs_info,
333 			      struct btrfs_delayed_ref_root *delayed_refs,
334 			      struct btrfs_delayed_ref_head *head)
335 {
336 	struct rb_node *node;
337 	u64 seq = 0;
338 
339 	assert_spin_locked(&head->lock);
340 	/*
341 	 * We don't have too much refs to merge in the case of delayed data
342 	 * refs.
343 	 */
344 	if (head->is_data)
345 		return;
346 
347 	spin_lock(&fs_info->tree_mod_seq_lock);
348 	if (!list_empty(&fs_info->tree_mod_seq_list)) {
349 		struct seq_list *elem;
350 
351 		elem = list_first_entry(&fs_info->tree_mod_seq_list,
352 					struct seq_list, list);
353 		seq = elem->seq;
354 	}
355 	spin_unlock(&fs_info->tree_mod_seq_lock);
356 
357 	node = rb_first(&head->ref_root);
358 	while (node) {
359 		struct btrfs_delayed_ref_node *ref;
360 
361 		ref = rb_entry(node, struct btrfs_delayed_ref_node,
362 			       rb_node);
363 		/* We can't merge refs that are outside of our seq count */
364 		if (seq && ref->seq >= seq)
365 			break;
366 		if (merge_ref(trans, delayed_refs, head, ref, seq))
367 			node = rb_first(&head->ref_root);
368 		else
369 			node = rb_next(&ref->rb_node);
370 	}
371 }
372 
373 int btrfs_check_delayed_seq(struct btrfs_fs_info *fs_info,
374 			    struct btrfs_delayed_ref_root *delayed_refs,
375 			    u64 seq)
376 {
377 	struct seq_list *elem;
378 	int ret = 0;
379 
380 	spin_lock(&fs_info->tree_mod_seq_lock);
381 	if (!list_empty(&fs_info->tree_mod_seq_list)) {
382 		elem = list_first_entry(&fs_info->tree_mod_seq_list,
383 					struct seq_list, list);
384 		if (seq >= elem->seq) {
385 			pr_debug("holding back delayed_ref %#x.%x, lowest is %#x.%x (%p)\n",
386 				 (u32)(seq >> 32), (u32)seq,
387 				 (u32)(elem->seq >> 32), (u32)elem->seq,
388 				 delayed_refs);
389 			ret = 1;
390 		}
391 	}
392 
393 	spin_unlock(&fs_info->tree_mod_seq_lock);
394 	return ret;
395 }
396 
397 struct btrfs_delayed_ref_head *
398 btrfs_select_ref_head(struct btrfs_trans_handle *trans)
399 {
400 	struct btrfs_delayed_ref_root *delayed_refs;
401 	struct btrfs_delayed_ref_head *head;
402 	u64 start;
403 	bool loop = false;
404 
405 	delayed_refs = &trans->transaction->delayed_refs;
406 
407 again:
408 	start = delayed_refs->run_delayed_start;
409 	head = find_ref_head(&delayed_refs->href_root, start, 1);
410 	if (!head && !loop) {
411 		delayed_refs->run_delayed_start = 0;
412 		start = 0;
413 		loop = true;
414 		head = find_ref_head(&delayed_refs->href_root, start, 1);
415 		if (!head)
416 			return NULL;
417 	} else if (!head && loop) {
418 		return NULL;
419 	}
420 
421 	while (head->processing) {
422 		struct rb_node *node;
423 
424 		node = rb_next(&head->href_node);
425 		if (!node) {
426 			if (loop)
427 				return NULL;
428 			delayed_refs->run_delayed_start = 0;
429 			start = 0;
430 			loop = true;
431 			goto again;
432 		}
433 		head = rb_entry(node, struct btrfs_delayed_ref_head,
434 				href_node);
435 	}
436 
437 	head->processing = 1;
438 	WARN_ON(delayed_refs->num_heads_ready == 0);
439 	delayed_refs->num_heads_ready--;
440 	delayed_refs->run_delayed_start = head->node.bytenr +
441 		head->node.num_bytes;
442 	return head;
443 }
444 
445 /*
446  * helper function to update an extent delayed ref in the
447  * rbtree.  existing and update must both have the same
448  * bytenr and parent
449  *
450  * This may free existing if the update cancels out whatever
451  * operation it was doing.
452  */
453 static noinline void
454 update_existing_ref(struct btrfs_trans_handle *trans,
455 		    struct btrfs_delayed_ref_root *delayed_refs,
456 		    struct btrfs_delayed_ref_head *head,
457 		    struct btrfs_delayed_ref_node *existing,
458 		    struct btrfs_delayed_ref_node *update)
459 {
460 	if (update->action != existing->action) {
461 		/*
462 		 * this is effectively undoing either an add or a
463 		 * drop.  We decrement the ref_mod, and if it goes
464 		 * down to zero we just delete the entry without
465 		 * every changing the extent allocation tree.
466 		 */
467 		existing->ref_mod--;
468 		if (existing->ref_mod == 0)
469 			drop_delayed_ref(trans, delayed_refs, head, existing);
470 		else
471 			WARN_ON(existing->type == BTRFS_TREE_BLOCK_REF_KEY ||
472 				existing->type == BTRFS_SHARED_BLOCK_REF_KEY);
473 	} else {
474 		WARN_ON(existing->type == BTRFS_TREE_BLOCK_REF_KEY ||
475 			existing->type == BTRFS_SHARED_BLOCK_REF_KEY);
476 		/*
477 		 * the action on the existing ref matches
478 		 * the action on the ref we're trying to add.
479 		 * Bump the ref_mod by one so the backref that
480 		 * is eventually added/removed has the correct
481 		 * reference count
482 		 */
483 		existing->ref_mod += update->ref_mod;
484 	}
485 }
486 
487 /*
488  * helper function to update the accounting in the head ref
489  * existing and update must have the same bytenr
490  */
491 static noinline void
492 update_existing_head_ref(struct btrfs_delayed_ref_node *existing,
493 			 struct btrfs_delayed_ref_node *update)
494 {
495 	struct btrfs_delayed_ref_head *existing_ref;
496 	struct btrfs_delayed_ref_head *ref;
497 
498 	existing_ref = btrfs_delayed_node_to_head(existing);
499 	ref = btrfs_delayed_node_to_head(update);
500 	BUG_ON(existing_ref->is_data != ref->is_data);
501 
502 	spin_lock(&existing_ref->lock);
503 	if (ref->must_insert_reserved) {
504 		/* if the extent was freed and then
505 		 * reallocated before the delayed ref
506 		 * entries were processed, we can end up
507 		 * with an existing head ref without
508 		 * the must_insert_reserved flag set.
509 		 * Set it again here
510 		 */
511 		existing_ref->must_insert_reserved = ref->must_insert_reserved;
512 
513 		/*
514 		 * update the num_bytes so we make sure the accounting
515 		 * is done correctly
516 		 */
517 		existing->num_bytes = update->num_bytes;
518 
519 	}
520 
521 	if (ref->extent_op) {
522 		if (!existing_ref->extent_op) {
523 			existing_ref->extent_op = ref->extent_op;
524 		} else {
525 			if (ref->extent_op->update_key) {
526 				memcpy(&existing_ref->extent_op->key,
527 				       &ref->extent_op->key,
528 				       sizeof(ref->extent_op->key));
529 				existing_ref->extent_op->update_key = 1;
530 			}
531 			if (ref->extent_op->update_flags) {
532 				existing_ref->extent_op->flags_to_set |=
533 					ref->extent_op->flags_to_set;
534 				existing_ref->extent_op->update_flags = 1;
535 			}
536 			btrfs_free_delayed_extent_op(ref->extent_op);
537 		}
538 	}
539 	/*
540 	 * update the reference mod on the head to reflect this new operation,
541 	 * only need the lock for this case cause we could be processing it
542 	 * currently, for refs we just added we know we're a-ok.
543 	 */
544 	existing->ref_mod += update->ref_mod;
545 	spin_unlock(&existing_ref->lock);
546 }
547 
548 /*
549  * helper function to actually insert a head node into the rbtree.
550  * this does all the dirty work in terms of maintaining the correct
551  * overall modification count.
552  */
553 static noinline struct btrfs_delayed_ref_head *
554 add_delayed_ref_head(struct btrfs_fs_info *fs_info,
555 		     struct btrfs_trans_handle *trans,
556 		     struct btrfs_delayed_ref_node *ref, u64 bytenr,
557 		     u64 num_bytes, int action, int is_data)
558 {
559 	struct btrfs_delayed_ref_head *existing;
560 	struct btrfs_delayed_ref_head *head_ref = NULL;
561 	struct btrfs_delayed_ref_root *delayed_refs;
562 	int count_mod = 1;
563 	int must_insert_reserved = 0;
564 
565 	/*
566 	 * the head node stores the sum of all the mods, so dropping a ref
567 	 * should drop the sum in the head node by one.
568 	 */
569 	if (action == BTRFS_UPDATE_DELAYED_HEAD)
570 		count_mod = 0;
571 	else if (action == BTRFS_DROP_DELAYED_REF)
572 		count_mod = -1;
573 
574 	/*
575 	 * BTRFS_ADD_DELAYED_EXTENT means that we need to update
576 	 * the reserved accounting when the extent is finally added, or
577 	 * if a later modification deletes the delayed ref without ever
578 	 * inserting the extent into the extent allocation tree.
579 	 * ref->must_insert_reserved is the flag used to record
580 	 * that accounting mods are required.
581 	 *
582 	 * Once we record must_insert_reserved, switch the action to
583 	 * BTRFS_ADD_DELAYED_REF because other special casing is not required.
584 	 */
585 	if (action == BTRFS_ADD_DELAYED_EXTENT)
586 		must_insert_reserved = 1;
587 	else
588 		must_insert_reserved = 0;
589 
590 	delayed_refs = &trans->transaction->delayed_refs;
591 
592 	/* first set the basic ref node struct up */
593 	atomic_set(&ref->refs, 1);
594 	ref->bytenr = bytenr;
595 	ref->num_bytes = num_bytes;
596 	ref->ref_mod = count_mod;
597 	ref->type  = 0;
598 	ref->action  = 0;
599 	ref->is_head = 1;
600 	ref->in_tree = 1;
601 	ref->seq = 0;
602 
603 	head_ref = btrfs_delayed_node_to_head(ref);
604 	head_ref->must_insert_reserved = must_insert_reserved;
605 	head_ref->is_data = is_data;
606 	head_ref->ref_root = RB_ROOT;
607 	head_ref->processing = 0;
608 
609 	spin_lock_init(&head_ref->lock);
610 	mutex_init(&head_ref->mutex);
611 
612 	trace_add_delayed_ref_head(ref, head_ref, action);
613 
614 	existing = htree_insert(&delayed_refs->href_root,
615 				&head_ref->href_node);
616 	if (existing) {
617 		update_existing_head_ref(&existing->node, ref);
618 		/*
619 		 * we've updated the existing ref, free the newly
620 		 * allocated ref
621 		 */
622 		kmem_cache_free(btrfs_delayed_ref_head_cachep, head_ref);
623 		head_ref = existing;
624 	} else {
625 		delayed_refs->num_heads++;
626 		delayed_refs->num_heads_ready++;
627 		atomic_inc(&delayed_refs->num_entries);
628 		trans->delayed_ref_updates++;
629 	}
630 	return head_ref;
631 }
632 
633 /*
634  * helper to insert a delayed tree ref into the rbtree.
635  */
636 static noinline void
637 add_delayed_tree_ref(struct btrfs_fs_info *fs_info,
638 		     struct btrfs_trans_handle *trans,
639 		     struct btrfs_delayed_ref_head *head_ref,
640 		     struct btrfs_delayed_ref_node *ref, u64 bytenr,
641 		     u64 num_bytes, u64 parent, u64 ref_root, int level,
642 		     int action, int no_quota)
643 {
644 	struct btrfs_delayed_ref_node *existing;
645 	struct btrfs_delayed_tree_ref *full_ref;
646 	struct btrfs_delayed_ref_root *delayed_refs;
647 	u64 seq = 0;
648 
649 	if (action == BTRFS_ADD_DELAYED_EXTENT)
650 		action = BTRFS_ADD_DELAYED_REF;
651 
652 	if (is_fstree(ref_root))
653 		seq = atomic64_read(&fs_info->tree_mod_seq);
654 	delayed_refs = &trans->transaction->delayed_refs;
655 
656 	/* first set the basic ref node struct up */
657 	atomic_set(&ref->refs, 1);
658 	ref->bytenr = bytenr;
659 	ref->num_bytes = num_bytes;
660 	ref->ref_mod = 1;
661 	ref->action = action;
662 	ref->is_head = 0;
663 	ref->in_tree = 1;
664 	ref->no_quota = no_quota;
665 	ref->seq = seq;
666 
667 	full_ref = btrfs_delayed_node_to_tree_ref(ref);
668 	full_ref->parent = parent;
669 	full_ref->root = ref_root;
670 	if (parent)
671 		ref->type = BTRFS_SHARED_BLOCK_REF_KEY;
672 	else
673 		ref->type = BTRFS_TREE_BLOCK_REF_KEY;
674 	full_ref->level = level;
675 
676 	trace_add_delayed_tree_ref(ref, full_ref, action);
677 
678 	spin_lock(&head_ref->lock);
679 	existing = tree_insert(&head_ref->ref_root, &ref->rb_node);
680 	if (existing) {
681 		update_existing_ref(trans, delayed_refs, head_ref, existing,
682 				    ref);
683 		/*
684 		 * we've updated the existing ref, free the newly
685 		 * allocated ref
686 		 */
687 		kmem_cache_free(btrfs_delayed_tree_ref_cachep, full_ref);
688 	} else {
689 		atomic_inc(&delayed_refs->num_entries);
690 		trans->delayed_ref_updates++;
691 	}
692 	spin_unlock(&head_ref->lock);
693 }
694 
695 /*
696  * helper to insert a delayed data ref into the rbtree.
697  */
698 static noinline void
699 add_delayed_data_ref(struct btrfs_fs_info *fs_info,
700 		     struct btrfs_trans_handle *trans,
701 		     struct btrfs_delayed_ref_head *head_ref,
702 		     struct btrfs_delayed_ref_node *ref, u64 bytenr,
703 		     u64 num_bytes, u64 parent, u64 ref_root, u64 owner,
704 		     u64 offset, int action, int no_quota)
705 {
706 	struct btrfs_delayed_ref_node *existing;
707 	struct btrfs_delayed_data_ref *full_ref;
708 	struct btrfs_delayed_ref_root *delayed_refs;
709 	u64 seq = 0;
710 
711 	if (action == BTRFS_ADD_DELAYED_EXTENT)
712 		action = BTRFS_ADD_DELAYED_REF;
713 
714 	delayed_refs = &trans->transaction->delayed_refs;
715 
716 	if (is_fstree(ref_root))
717 		seq = atomic64_read(&fs_info->tree_mod_seq);
718 
719 	/* first set the basic ref node struct up */
720 	atomic_set(&ref->refs, 1);
721 	ref->bytenr = bytenr;
722 	ref->num_bytes = num_bytes;
723 	ref->ref_mod = 1;
724 	ref->action = action;
725 	ref->is_head = 0;
726 	ref->in_tree = 1;
727 	ref->no_quota = no_quota;
728 	ref->seq = seq;
729 
730 	full_ref = btrfs_delayed_node_to_data_ref(ref);
731 	full_ref->parent = parent;
732 	full_ref->root = ref_root;
733 	if (parent)
734 		ref->type = BTRFS_SHARED_DATA_REF_KEY;
735 	else
736 		ref->type = BTRFS_EXTENT_DATA_REF_KEY;
737 
738 	full_ref->objectid = owner;
739 	full_ref->offset = offset;
740 
741 	trace_add_delayed_data_ref(ref, full_ref, action);
742 
743 	spin_lock(&head_ref->lock);
744 	existing = tree_insert(&head_ref->ref_root, &ref->rb_node);
745 	if (existing) {
746 		update_existing_ref(trans, delayed_refs, head_ref, existing,
747 				    ref);
748 		/*
749 		 * we've updated the existing ref, free the newly
750 		 * allocated ref
751 		 */
752 		kmem_cache_free(btrfs_delayed_data_ref_cachep, full_ref);
753 	} else {
754 		atomic_inc(&delayed_refs->num_entries);
755 		trans->delayed_ref_updates++;
756 	}
757 	spin_unlock(&head_ref->lock);
758 }
759 
760 /*
761  * add a delayed tree ref.  This does all of the accounting required
762  * to make sure the delayed ref is eventually processed before this
763  * transaction commits.
764  */
765 int btrfs_add_delayed_tree_ref(struct btrfs_fs_info *fs_info,
766 			       struct btrfs_trans_handle *trans,
767 			       u64 bytenr, u64 num_bytes, u64 parent,
768 			       u64 ref_root,  int level, int action,
769 			       struct btrfs_delayed_extent_op *extent_op,
770 			       int no_quota)
771 {
772 	struct btrfs_delayed_tree_ref *ref;
773 	struct btrfs_delayed_ref_head *head_ref;
774 	struct btrfs_delayed_ref_root *delayed_refs;
775 
776 	if (!is_fstree(ref_root) || !fs_info->quota_enabled)
777 		no_quota = 0;
778 
779 	BUG_ON(extent_op && extent_op->is_data);
780 	ref = kmem_cache_alloc(btrfs_delayed_tree_ref_cachep, GFP_NOFS);
781 	if (!ref)
782 		return -ENOMEM;
783 
784 	head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS);
785 	if (!head_ref) {
786 		kmem_cache_free(btrfs_delayed_tree_ref_cachep, ref);
787 		return -ENOMEM;
788 	}
789 
790 	head_ref->extent_op = extent_op;
791 
792 	delayed_refs = &trans->transaction->delayed_refs;
793 	spin_lock(&delayed_refs->lock);
794 
795 	/*
796 	 * insert both the head node and the new ref without dropping
797 	 * the spin lock
798 	 */
799 	head_ref = add_delayed_ref_head(fs_info, trans, &head_ref->node,
800 					bytenr, num_bytes, action, 0);
801 
802 	add_delayed_tree_ref(fs_info, trans, head_ref, &ref->node, bytenr,
803 				   num_bytes, parent, ref_root, level, action,
804 				   no_quota);
805 	spin_unlock(&delayed_refs->lock);
806 
807 	return 0;
808 }
809 
810 /*
811  * add a delayed data ref. it's similar to btrfs_add_delayed_tree_ref.
812  */
813 int btrfs_add_delayed_data_ref(struct btrfs_fs_info *fs_info,
814 			       struct btrfs_trans_handle *trans,
815 			       u64 bytenr, u64 num_bytes,
816 			       u64 parent, u64 ref_root,
817 			       u64 owner, u64 offset, int action,
818 			       struct btrfs_delayed_extent_op *extent_op,
819 			       int no_quota)
820 {
821 	struct btrfs_delayed_data_ref *ref;
822 	struct btrfs_delayed_ref_head *head_ref;
823 	struct btrfs_delayed_ref_root *delayed_refs;
824 
825 	if (!is_fstree(ref_root) || !fs_info->quota_enabled)
826 		no_quota = 0;
827 
828 	BUG_ON(extent_op && !extent_op->is_data);
829 	ref = kmem_cache_alloc(btrfs_delayed_data_ref_cachep, GFP_NOFS);
830 	if (!ref)
831 		return -ENOMEM;
832 
833 	head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS);
834 	if (!head_ref) {
835 		kmem_cache_free(btrfs_delayed_data_ref_cachep, ref);
836 		return -ENOMEM;
837 	}
838 
839 	head_ref->extent_op = extent_op;
840 
841 	delayed_refs = &trans->transaction->delayed_refs;
842 	spin_lock(&delayed_refs->lock);
843 
844 	/*
845 	 * insert both the head node and the new ref without dropping
846 	 * the spin lock
847 	 */
848 	head_ref = add_delayed_ref_head(fs_info, trans, &head_ref->node,
849 					bytenr, num_bytes, action, 1);
850 
851 	add_delayed_data_ref(fs_info, trans, head_ref, &ref->node, bytenr,
852 				   num_bytes, parent, ref_root, owner, offset,
853 				   action, no_quota);
854 	spin_unlock(&delayed_refs->lock);
855 
856 	return 0;
857 }
858 
859 int btrfs_add_delayed_extent_op(struct btrfs_fs_info *fs_info,
860 				struct btrfs_trans_handle *trans,
861 				u64 bytenr, u64 num_bytes,
862 				struct btrfs_delayed_extent_op *extent_op)
863 {
864 	struct btrfs_delayed_ref_head *head_ref;
865 	struct btrfs_delayed_ref_root *delayed_refs;
866 
867 	head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS);
868 	if (!head_ref)
869 		return -ENOMEM;
870 
871 	head_ref->extent_op = extent_op;
872 
873 	delayed_refs = &trans->transaction->delayed_refs;
874 	spin_lock(&delayed_refs->lock);
875 
876 	add_delayed_ref_head(fs_info, trans, &head_ref->node, bytenr,
877 				   num_bytes, BTRFS_UPDATE_DELAYED_HEAD,
878 				   extent_op->is_data);
879 
880 	spin_unlock(&delayed_refs->lock);
881 	return 0;
882 }
883 
884 /*
885  * this does a simple search for the head node for a given extent.
886  * It must be called with the delayed ref spinlock held, and it returns
887  * the head node if any where found, or NULL if not.
888  */
889 struct btrfs_delayed_ref_head *
890 btrfs_find_delayed_ref_head(struct btrfs_trans_handle *trans, u64 bytenr)
891 {
892 	struct btrfs_delayed_ref_root *delayed_refs;
893 
894 	delayed_refs = &trans->transaction->delayed_refs;
895 	return find_ref_head(&delayed_refs->href_root, bytenr, 0);
896 }
897 
898 void btrfs_delayed_ref_exit(void)
899 {
900 	if (btrfs_delayed_ref_head_cachep)
901 		kmem_cache_destroy(btrfs_delayed_ref_head_cachep);
902 	if (btrfs_delayed_tree_ref_cachep)
903 		kmem_cache_destroy(btrfs_delayed_tree_ref_cachep);
904 	if (btrfs_delayed_data_ref_cachep)
905 		kmem_cache_destroy(btrfs_delayed_data_ref_cachep);
906 	if (btrfs_delayed_extent_op_cachep)
907 		kmem_cache_destroy(btrfs_delayed_extent_op_cachep);
908 }
909 
910 int btrfs_delayed_ref_init(void)
911 {
912 	btrfs_delayed_ref_head_cachep = kmem_cache_create(
913 				"btrfs_delayed_ref_head",
914 				sizeof(struct btrfs_delayed_ref_head), 0,
915 				SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
916 	if (!btrfs_delayed_ref_head_cachep)
917 		goto fail;
918 
919 	btrfs_delayed_tree_ref_cachep = kmem_cache_create(
920 				"btrfs_delayed_tree_ref",
921 				sizeof(struct btrfs_delayed_tree_ref), 0,
922 				SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
923 	if (!btrfs_delayed_tree_ref_cachep)
924 		goto fail;
925 
926 	btrfs_delayed_data_ref_cachep = kmem_cache_create(
927 				"btrfs_delayed_data_ref",
928 				sizeof(struct btrfs_delayed_data_ref), 0,
929 				SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
930 	if (!btrfs_delayed_data_ref_cachep)
931 		goto fail;
932 
933 	btrfs_delayed_extent_op_cachep = kmem_cache_create(
934 				"btrfs_delayed_extent_op",
935 				sizeof(struct btrfs_delayed_extent_op), 0,
936 				SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
937 	if (!btrfs_delayed_extent_op_cachep)
938 		goto fail;
939 
940 	return 0;
941 fail:
942 	btrfs_delayed_ref_exit();
943 	return -ENOMEM;
944 }
945