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
26struct kmem_cache *btrfs_delayed_ref_head_cachep;
27struct kmem_cache *btrfs_delayed_tree_ref_cachep;
28struct kmem_cache *btrfs_delayed_data_ref_cachep;
29struct 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 */
42static 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 */
62static 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 */
91static 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 */
139static 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 */
169static 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 */
204static struct btrfs_delayed_ref_head *
205find_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
237int 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
261static 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
280static 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
331void 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
373int 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
397struct btrfs_delayed_ref_head *
398btrfs_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
407again:
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 */
453static noinline void
454update_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 */
491static noinline void
492update_existing_head_ref(struct btrfs_delayed_ref_root *delayed_refs,
493			 struct btrfs_delayed_ref_node *existing,
494			 struct btrfs_delayed_ref_node *update)
495{
496	struct btrfs_delayed_ref_head *existing_ref;
497	struct btrfs_delayed_ref_head *ref;
498	int old_ref_mod;
499
500	existing_ref = btrfs_delayed_node_to_head(existing);
501	ref = btrfs_delayed_node_to_head(update);
502	BUG_ON(existing_ref->is_data != ref->is_data);
503
504	spin_lock(&existing_ref->lock);
505	if (ref->must_insert_reserved) {
506		/* if the extent was freed and then
507		 * reallocated before the delayed ref
508		 * entries were processed, we can end up
509		 * with an existing head ref without
510		 * the must_insert_reserved flag set.
511		 * Set it again here
512		 */
513		existing_ref->must_insert_reserved = ref->must_insert_reserved;
514
515		/*
516		 * update the num_bytes so we make sure the accounting
517		 * is done correctly
518		 */
519		existing->num_bytes = update->num_bytes;
520
521	}
522
523	if (ref->extent_op) {
524		if (!existing_ref->extent_op) {
525			existing_ref->extent_op = ref->extent_op;
526		} else {
527			if (ref->extent_op->update_key) {
528				memcpy(&existing_ref->extent_op->key,
529				       &ref->extent_op->key,
530				       sizeof(ref->extent_op->key));
531				existing_ref->extent_op->update_key = 1;
532			}
533			if (ref->extent_op->update_flags) {
534				existing_ref->extent_op->flags_to_set |=
535					ref->extent_op->flags_to_set;
536				existing_ref->extent_op->update_flags = 1;
537			}
538			btrfs_free_delayed_extent_op(ref->extent_op);
539		}
540	}
541	/*
542	 * update the reference mod on the head to reflect this new operation,
543	 * only need the lock for this case cause we could be processing it
544	 * currently, for refs we just added we know we're a-ok.
545	 */
546	old_ref_mod = existing_ref->total_ref_mod;
547	existing->ref_mod += update->ref_mod;
548	existing_ref->total_ref_mod += update->ref_mod;
549
550	/*
551	 * If we are going to from a positive ref mod to a negative or vice
552	 * versa we need to make sure to adjust pending_csums accordingly.
553	 */
554	if (existing_ref->is_data) {
555		if (existing_ref->total_ref_mod >= 0 && old_ref_mod < 0)
556			delayed_refs->pending_csums -= existing->num_bytes;
557		if (existing_ref->total_ref_mod < 0 && old_ref_mod >= 0)
558			delayed_refs->pending_csums += existing->num_bytes;
559	}
560	spin_unlock(&existing_ref->lock);
561}
562
563/*
564 * helper function to actually insert a head node into the rbtree.
565 * this does all the dirty work in terms of maintaining the correct
566 * overall modification count.
567 */
568static noinline struct btrfs_delayed_ref_head *
569add_delayed_ref_head(struct btrfs_fs_info *fs_info,
570		     struct btrfs_trans_handle *trans,
571		     struct btrfs_delayed_ref_node *ref, u64 bytenr,
572		     u64 num_bytes, int action, int is_data)
573{
574	struct btrfs_delayed_ref_head *existing;
575	struct btrfs_delayed_ref_head *head_ref = NULL;
576	struct btrfs_delayed_ref_root *delayed_refs;
577	int count_mod = 1;
578	int must_insert_reserved = 0;
579
580	/*
581	 * the head node stores the sum of all the mods, so dropping a ref
582	 * should drop the sum in the head node by one.
583	 */
584	if (action == BTRFS_UPDATE_DELAYED_HEAD)
585		count_mod = 0;
586	else if (action == BTRFS_DROP_DELAYED_REF)
587		count_mod = -1;
588
589	/*
590	 * BTRFS_ADD_DELAYED_EXTENT means that we need to update
591	 * the reserved accounting when the extent is finally added, or
592	 * if a later modification deletes the delayed ref without ever
593	 * inserting the extent into the extent allocation tree.
594	 * ref->must_insert_reserved is the flag used to record
595	 * that accounting mods are required.
596	 *
597	 * Once we record must_insert_reserved, switch the action to
598	 * BTRFS_ADD_DELAYED_REF because other special casing is not required.
599	 */
600	if (action == BTRFS_ADD_DELAYED_EXTENT)
601		must_insert_reserved = 1;
602	else
603		must_insert_reserved = 0;
604
605	delayed_refs = &trans->transaction->delayed_refs;
606
607	/* first set the basic ref node struct up */
608	atomic_set(&ref->refs, 1);
609	ref->bytenr = bytenr;
610	ref->num_bytes = num_bytes;
611	ref->ref_mod = count_mod;
612	ref->type  = 0;
613	ref->action  = 0;
614	ref->is_head = 1;
615	ref->in_tree = 1;
616	ref->seq = 0;
617
618	head_ref = btrfs_delayed_node_to_head(ref);
619	head_ref->must_insert_reserved = must_insert_reserved;
620	head_ref->is_data = is_data;
621	head_ref->ref_root = RB_ROOT;
622	head_ref->processing = 0;
623	head_ref->total_ref_mod = count_mod;
624
625	spin_lock_init(&head_ref->lock);
626	mutex_init(&head_ref->mutex);
627
628	trace_add_delayed_ref_head(ref, head_ref, action);
629
630	existing = htree_insert(&delayed_refs->href_root,
631				&head_ref->href_node);
632	if (existing) {
633		update_existing_head_ref(delayed_refs, &existing->node, ref);
634		/*
635		 * we've updated the existing ref, free the newly
636		 * allocated ref
637		 */
638		kmem_cache_free(btrfs_delayed_ref_head_cachep, head_ref);
639		head_ref = existing;
640	} else {
641		if (is_data && count_mod < 0)
642			delayed_refs->pending_csums += num_bytes;
643		delayed_refs->num_heads++;
644		delayed_refs->num_heads_ready++;
645		atomic_inc(&delayed_refs->num_entries);
646		trans->delayed_ref_updates++;
647	}
648	return head_ref;
649}
650
651/*
652 * helper to insert a delayed tree ref into the rbtree.
653 */
654static noinline void
655add_delayed_tree_ref(struct btrfs_fs_info *fs_info,
656		     struct btrfs_trans_handle *trans,
657		     struct btrfs_delayed_ref_head *head_ref,
658		     struct btrfs_delayed_ref_node *ref, u64 bytenr,
659		     u64 num_bytes, u64 parent, u64 ref_root, int level,
660		     int action, int no_quota)
661{
662	struct btrfs_delayed_ref_node *existing;
663	struct btrfs_delayed_tree_ref *full_ref;
664	struct btrfs_delayed_ref_root *delayed_refs;
665	u64 seq = 0;
666
667	if (action == BTRFS_ADD_DELAYED_EXTENT)
668		action = BTRFS_ADD_DELAYED_REF;
669
670	if (is_fstree(ref_root))
671		seq = atomic64_read(&fs_info->tree_mod_seq);
672	delayed_refs = &trans->transaction->delayed_refs;
673
674	/* first set the basic ref node struct up */
675	atomic_set(&ref->refs, 1);
676	ref->bytenr = bytenr;
677	ref->num_bytes = num_bytes;
678	ref->ref_mod = 1;
679	ref->action = action;
680	ref->is_head = 0;
681	ref->in_tree = 1;
682	ref->no_quota = no_quota;
683	ref->seq = seq;
684
685	full_ref = btrfs_delayed_node_to_tree_ref(ref);
686	full_ref->parent = parent;
687	full_ref->root = ref_root;
688	if (parent)
689		ref->type = BTRFS_SHARED_BLOCK_REF_KEY;
690	else
691		ref->type = BTRFS_TREE_BLOCK_REF_KEY;
692	full_ref->level = level;
693
694	trace_add_delayed_tree_ref(ref, full_ref, action);
695
696	spin_lock(&head_ref->lock);
697	existing = tree_insert(&head_ref->ref_root, &ref->rb_node);
698	if (existing) {
699		update_existing_ref(trans, delayed_refs, head_ref, existing,
700				    ref);
701		/*
702		 * we've updated the existing ref, free the newly
703		 * allocated ref
704		 */
705		kmem_cache_free(btrfs_delayed_tree_ref_cachep, full_ref);
706	} else {
707		atomic_inc(&delayed_refs->num_entries);
708		trans->delayed_ref_updates++;
709	}
710	spin_unlock(&head_ref->lock);
711}
712
713/*
714 * helper to insert a delayed data ref into the rbtree.
715 */
716static noinline void
717add_delayed_data_ref(struct btrfs_fs_info *fs_info,
718		     struct btrfs_trans_handle *trans,
719		     struct btrfs_delayed_ref_head *head_ref,
720		     struct btrfs_delayed_ref_node *ref, u64 bytenr,
721		     u64 num_bytes, u64 parent, u64 ref_root, u64 owner,
722		     u64 offset, int action, int no_quota)
723{
724	struct btrfs_delayed_ref_node *existing;
725	struct btrfs_delayed_data_ref *full_ref;
726	struct btrfs_delayed_ref_root *delayed_refs;
727	u64 seq = 0;
728
729	if (action == BTRFS_ADD_DELAYED_EXTENT)
730		action = BTRFS_ADD_DELAYED_REF;
731
732	delayed_refs = &trans->transaction->delayed_refs;
733
734	if (is_fstree(ref_root))
735		seq = atomic64_read(&fs_info->tree_mod_seq);
736
737	/* first set the basic ref node struct up */
738	atomic_set(&ref->refs, 1);
739	ref->bytenr = bytenr;
740	ref->num_bytes = num_bytes;
741	ref->ref_mod = 1;
742	ref->action = action;
743	ref->is_head = 0;
744	ref->in_tree = 1;
745	ref->no_quota = no_quota;
746	ref->seq = seq;
747
748	full_ref = btrfs_delayed_node_to_data_ref(ref);
749	full_ref->parent = parent;
750	full_ref->root = ref_root;
751	if (parent)
752		ref->type = BTRFS_SHARED_DATA_REF_KEY;
753	else
754		ref->type = BTRFS_EXTENT_DATA_REF_KEY;
755
756	full_ref->objectid = owner;
757	full_ref->offset = offset;
758
759	trace_add_delayed_data_ref(ref, full_ref, action);
760
761	spin_lock(&head_ref->lock);
762	existing = tree_insert(&head_ref->ref_root, &ref->rb_node);
763	if (existing) {
764		update_existing_ref(trans, delayed_refs, head_ref, existing,
765				    ref);
766		/*
767		 * we've updated the existing ref, free the newly
768		 * allocated ref
769		 */
770		kmem_cache_free(btrfs_delayed_data_ref_cachep, full_ref);
771	} else {
772		atomic_inc(&delayed_refs->num_entries);
773		trans->delayed_ref_updates++;
774	}
775	spin_unlock(&head_ref->lock);
776}
777
778/*
779 * add a delayed tree ref.  This does all of the accounting required
780 * to make sure the delayed ref is eventually processed before this
781 * transaction commits.
782 */
783int btrfs_add_delayed_tree_ref(struct btrfs_fs_info *fs_info,
784			       struct btrfs_trans_handle *trans,
785			       u64 bytenr, u64 num_bytes, u64 parent,
786			       u64 ref_root,  int level, int action,
787			       struct btrfs_delayed_extent_op *extent_op,
788			       int no_quota)
789{
790	struct btrfs_delayed_tree_ref *ref;
791	struct btrfs_delayed_ref_head *head_ref;
792	struct btrfs_delayed_ref_root *delayed_refs;
793
794	if (!is_fstree(ref_root) || !fs_info->quota_enabled)
795		no_quota = 0;
796
797	BUG_ON(extent_op && extent_op->is_data);
798	ref = kmem_cache_alloc(btrfs_delayed_tree_ref_cachep, GFP_NOFS);
799	if (!ref)
800		return -ENOMEM;
801
802	head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS);
803	if (!head_ref) {
804		kmem_cache_free(btrfs_delayed_tree_ref_cachep, ref);
805		return -ENOMEM;
806	}
807
808	head_ref->extent_op = extent_op;
809
810	delayed_refs = &trans->transaction->delayed_refs;
811	spin_lock(&delayed_refs->lock);
812
813	/*
814	 * insert both the head node and the new ref without dropping
815	 * the spin lock
816	 */
817	head_ref = add_delayed_ref_head(fs_info, trans, &head_ref->node,
818					bytenr, num_bytes, action, 0);
819
820	add_delayed_tree_ref(fs_info, trans, head_ref, &ref->node, bytenr,
821				   num_bytes, parent, ref_root, level, action,
822				   no_quota);
823	spin_unlock(&delayed_refs->lock);
824
825	return 0;
826}
827
828/*
829 * add a delayed data ref. it's similar to btrfs_add_delayed_tree_ref.
830 */
831int btrfs_add_delayed_data_ref(struct btrfs_fs_info *fs_info,
832			       struct btrfs_trans_handle *trans,
833			       u64 bytenr, u64 num_bytes,
834			       u64 parent, u64 ref_root,
835			       u64 owner, u64 offset, int action,
836			       struct btrfs_delayed_extent_op *extent_op,
837			       int no_quota)
838{
839	struct btrfs_delayed_data_ref *ref;
840	struct btrfs_delayed_ref_head *head_ref;
841	struct btrfs_delayed_ref_root *delayed_refs;
842
843	if (!is_fstree(ref_root) || !fs_info->quota_enabled)
844		no_quota = 0;
845
846	BUG_ON(extent_op && !extent_op->is_data);
847	ref = kmem_cache_alloc(btrfs_delayed_data_ref_cachep, GFP_NOFS);
848	if (!ref)
849		return -ENOMEM;
850
851	head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS);
852	if (!head_ref) {
853		kmem_cache_free(btrfs_delayed_data_ref_cachep, ref);
854		return -ENOMEM;
855	}
856
857	head_ref->extent_op = extent_op;
858
859	delayed_refs = &trans->transaction->delayed_refs;
860	spin_lock(&delayed_refs->lock);
861
862	/*
863	 * insert both the head node and the new ref without dropping
864	 * the spin lock
865	 */
866	head_ref = add_delayed_ref_head(fs_info, trans, &head_ref->node,
867					bytenr, num_bytes, action, 1);
868
869	add_delayed_data_ref(fs_info, trans, head_ref, &ref->node, bytenr,
870				   num_bytes, parent, ref_root, owner, offset,
871				   action, no_quota);
872	spin_unlock(&delayed_refs->lock);
873
874	return 0;
875}
876
877int btrfs_add_delayed_extent_op(struct btrfs_fs_info *fs_info,
878				struct btrfs_trans_handle *trans,
879				u64 bytenr, u64 num_bytes,
880				struct btrfs_delayed_extent_op *extent_op)
881{
882	struct btrfs_delayed_ref_head *head_ref;
883	struct btrfs_delayed_ref_root *delayed_refs;
884
885	head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS);
886	if (!head_ref)
887		return -ENOMEM;
888
889	head_ref->extent_op = extent_op;
890
891	delayed_refs = &trans->transaction->delayed_refs;
892	spin_lock(&delayed_refs->lock);
893
894	add_delayed_ref_head(fs_info, trans, &head_ref->node, bytenr,
895				   num_bytes, BTRFS_UPDATE_DELAYED_HEAD,
896				   extent_op->is_data);
897
898	spin_unlock(&delayed_refs->lock);
899	return 0;
900}
901
902/*
903 * this does a simple search for the head node for a given extent.
904 * It must be called with the delayed ref spinlock held, and it returns
905 * the head node if any where found, or NULL if not.
906 */
907struct btrfs_delayed_ref_head *
908btrfs_find_delayed_ref_head(struct btrfs_trans_handle *trans, u64 bytenr)
909{
910	struct btrfs_delayed_ref_root *delayed_refs;
911
912	delayed_refs = &trans->transaction->delayed_refs;
913	return find_ref_head(&delayed_refs->href_root, bytenr, 0);
914}
915
916void btrfs_delayed_ref_exit(void)
917{
918	if (btrfs_delayed_ref_head_cachep)
919		kmem_cache_destroy(btrfs_delayed_ref_head_cachep);
920	if (btrfs_delayed_tree_ref_cachep)
921		kmem_cache_destroy(btrfs_delayed_tree_ref_cachep);
922	if (btrfs_delayed_data_ref_cachep)
923		kmem_cache_destroy(btrfs_delayed_data_ref_cachep);
924	if (btrfs_delayed_extent_op_cachep)
925		kmem_cache_destroy(btrfs_delayed_extent_op_cachep);
926}
927
928int btrfs_delayed_ref_init(void)
929{
930	btrfs_delayed_ref_head_cachep = kmem_cache_create(
931				"btrfs_delayed_ref_head",
932				sizeof(struct btrfs_delayed_ref_head), 0,
933				SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
934	if (!btrfs_delayed_ref_head_cachep)
935		goto fail;
936
937	btrfs_delayed_tree_ref_cachep = kmem_cache_create(
938				"btrfs_delayed_tree_ref",
939				sizeof(struct btrfs_delayed_tree_ref), 0,
940				SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
941	if (!btrfs_delayed_tree_ref_cachep)
942		goto fail;
943
944	btrfs_delayed_data_ref_cachep = kmem_cache_create(
945				"btrfs_delayed_data_ref",
946				sizeof(struct btrfs_delayed_data_ref), 0,
947				SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
948	if (!btrfs_delayed_data_ref_cachep)
949		goto fail;
950
951	btrfs_delayed_extent_op_cachep = kmem_cache_create(
952				"btrfs_delayed_extent_op",
953				sizeof(struct btrfs_delayed_extent_op), 0,
954				SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
955	if (!btrfs_delayed_extent_op_cachep)
956		goto fail;
957
958	return 0;
959fail:
960	btrfs_delayed_ref_exit();
961	return -ENOMEM;
962}
963