2 * Copyright (C) 2009 Oracle. All rights reserved.
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.
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.
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.
19 #include <linux/sched.h>
20 #include <linux/slab.h>
21 #include <linux/sort.h>
23 #include "delayed-ref.h"
24 #include "transaction.h"
27 * delayed back reference update tracking. For subvolume trees
28 * we queue up extent allocations and backref maintenance for
29 * delayed processing. This avoids deep call chains where we
30 * add extents in the middle of btrfs_search_slot, and it allows
31 * us to buffer up frequently modified backrefs in an rb tree instead
32 * of hammering updates on the extent allocation tree.
36 * compare two delayed tree backrefs with same bytenr and type
38 static int comp_tree_refs(struct btrfs_delayed_tree_ref *ref2,
39 struct btrfs_delayed_tree_ref *ref1)
41 if (ref1->node.type == BTRFS_TREE_BLOCK_REF_KEY) {
42 if (ref1->root < ref2->root)
44 if (ref1->root > ref2->root)
47 if (ref1->parent < ref2->parent)
49 if (ref1->parent > ref2->parent)
56 * compare two delayed data backrefs with same bytenr and type
58 static int comp_data_refs(struct btrfs_delayed_data_ref *ref2,
59 struct btrfs_delayed_data_ref *ref1)
61 if (ref1->node.type == BTRFS_EXTENT_DATA_REF_KEY) {
62 if (ref1->root < ref2->root)
64 if (ref1->root > ref2->root)
66 if (ref1->objectid < ref2->objectid)
68 if (ref1->objectid > ref2->objectid)
70 if (ref1->offset < ref2->offset)
72 if (ref1->offset > ref2->offset)
75 if (ref1->parent < ref2->parent)
77 if (ref1->parent > ref2->parent)
84 * entries in the rb tree are ordered by the byte number of the extent,
85 * type of the delayed backrefs and content of delayed backrefs.
87 static int comp_entry(struct btrfs_delayed_ref_node *ref2,
88 struct btrfs_delayed_ref_node *ref1)
90 if (ref1->bytenr < ref2->bytenr)
92 if (ref1->bytenr > ref2->bytenr)
94 if (ref1->is_head && ref2->is_head)
100 if (ref1->type < ref2->type)
102 if (ref1->type > ref2->type)
104 if (ref1->type == BTRFS_TREE_BLOCK_REF_KEY ||
105 ref1->type == BTRFS_SHARED_BLOCK_REF_KEY) {
106 return comp_tree_refs(btrfs_delayed_node_to_tree_ref(ref2),
107 btrfs_delayed_node_to_tree_ref(ref1));
108 } else if (ref1->type == BTRFS_EXTENT_DATA_REF_KEY ||
109 ref1->type == BTRFS_SHARED_DATA_REF_KEY) {
110 return comp_data_refs(btrfs_delayed_node_to_data_ref(ref2),
111 btrfs_delayed_node_to_data_ref(ref1));
118 * insert a new ref into the rbtree. This returns any existing refs
119 * for the same (bytenr,parent) tuple, or NULL if the new node was properly
122 static struct btrfs_delayed_ref_node *tree_insert(struct rb_root *root,
123 struct rb_node *node)
125 struct rb_node **p = &root->rb_node;
126 struct rb_node *parent_node = NULL;
127 struct btrfs_delayed_ref_node *entry;
128 struct btrfs_delayed_ref_node *ins;
131 ins = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
134 entry = rb_entry(parent_node, struct btrfs_delayed_ref_node,
137 cmp = comp_entry(entry, ins);
146 rb_link_node(node, parent_node, p);
147 rb_insert_color(node, root);
152 * find an head entry based on bytenr. This returns the delayed ref
153 * head if it was able to find one, or NULL if nothing was in that spot
155 static struct btrfs_delayed_ref_node *find_ref_head(struct rb_root *root,
157 struct btrfs_delayed_ref_node **last)
159 struct rb_node *n = root->rb_node;
160 struct btrfs_delayed_ref_node *entry;
164 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
165 WARN_ON(!entry->in_tree);
169 if (bytenr < entry->bytenr)
171 else if (bytenr > entry->bytenr)
173 else if (!btrfs_delayed_ref_is_head(entry))
188 int btrfs_delayed_ref_lock(struct btrfs_trans_handle *trans,
189 struct btrfs_delayed_ref_head *head)
191 struct btrfs_delayed_ref_root *delayed_refs;
193 delayed_refs = &trans->transaction->delayed_refs;
194 assert_spin_locked(&delayed_refs->lock);
195 if (mutex_trylock(&head->mutex))
198 atomic_inc(&head->node.refs);
199 spin_unlock(&delayed_refs->lock);
201 mutex_lock(&head->mutex);
202 spin_lock(&delayed_refs->lock);
203 if (!head->node.in_tree) {
204 mutex_unlock(&head->mutex);
205 btrfs_put_delayed_ref(&head->node);
208 btrfs_put_delayed_ref(&head->node);
212 int btrfs_find_ref_cluster(struct btrfs_trans_handle *trans,
213 struct list_head *cluster, u64 start)
216 struct btrfs_delayed_ref_root *delayed_refs;
217 struct rb_node *node;
218 struct btrfs_delayed_ref_node *ref;
219 struct btrfs_delayed_ref_head *head;
221 delayed_refs = &trans->transaction->delayed_refs;
223 node = rb_first(&delayed_refs->root);
226 find_ref_head(&delayed_refs->root, start, &ref);
228 struct btrfs_delayed_ref_node *tmp;
230 node = rb_prev(&ref->rb_node);
233 struct btrfs_delayed_ref_node,
235 if (tmp->bytenr < start)
238 node = rb_prev(&ref->rb_node);
240 node = &ref->rb_node;
242 node = rb_first(&delayed_refs->root);
245 while (node && count < 32) {
246 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
247 if (btrfs_delayed_ref_is_head(ref)) {
248 head = btrfs_delayed_node_to_head(ref);
249 if (list_empty(&head->cluster)) {
250 list_add_tail(&head->cluster, cluster);
251 delayed_refs->run_delayed_start =
255 WARN_ON(delayed_refs->num_heads_ready == 0);
256 delayed_refs->num_heads_ready--;
258 /* the goal of the clustering is to find extents
259 * that are likely to end up in the same extent
260 * leaf on disk. So, we don't want them spread
261 * all over the tree. Stop now if we've hit
262 * a head that was already in use
267 node = rb_next(node);
273 * we've gone to the end of the rbtree without finding any
274 * clusters. start from the beginning and try again
277 node = rb_first(&delayed_refs->root);
284 * This checks to see if there are any delayed refs in the
285 * btree for a given bytenr. It returns one if it finds any
286 * and zero otherwise.
288 * If it only finds a head node, it returns 0.
290 * The idea is to use this when deciding if you can safely delete an
291 * extent from the extent allocation tree. There may be a pending
292 * ref in the rbtree that adds or removes references, so as long as this
293 * returns one you need to leave the BTRFS_EXTENT_ITEM in the extent
296 int btrfs_delayed_ref_pending(struct btrfs_trans_handle *trans, u64 bytenr)
298 struct btrfs_delayed_ref_node *ref;
299 struct btrfs_delayed_ref_root *delayed_refs;
300 struct rb_node *prev_node;
303 delayed_refs = &trans->transaction->delayed_refs;
304 spin_lock(&delayed_refs->lock);
306 ref = find_ref_head(&delayed_refs->root, bytenr, NULL);
308 prev_node = rb_prev(&ref->rb_node);
311 ref = rb_entry(prev_node, struct btrfs_delayed_ref_node,
313 if (ref->bytenr == bytenr)
317 spin_unlock(&delayed_refs->lock);
322 * helper function to update an extent delayed ref in the
323 * rbtree. existing and update must both have the same
326 * This may free existing if the update cancels out whatever
327 * operation it was doing.
330 update_existing_ref(struct btrfs_trans_handle *trans,
331 struct btrfs_delayed_ref_root *delayed_refs,
332 struct btrfs_delayed_ref_node *existing,
333 struct btrfs_delayed_ref_node *update)
335 if (update->action != existing->action) {
337 * this is effectively undoing either an add or a
338 * drop. We decrement the ref_mod, and if it goes
339 * down to zero we just delete the entry without
340 * every changing the extent allocation tree.
343 if (existing->ref_mod == 0) {
344 rb_erase(&existing->rb_node,
345 &delayed_refs->root);
346 existing->in_tree = 0;
347 btrfs_put_delayed_ref(existing);
348 delayed_refs->num_entries--;
349 if (trans->delayed_ref_updates)
350 trans->delayed_ref_updates--;
352 WARN_ON(existing->type == BTRFS_TREE_BLOCK_REF_KEY ||
353 existing->type == BTRFS_SHARED_BLOCK_REF_KEY);
356 WARN_ON(existing->type == BTRFS_TREE_BLOCK_REF_KEY ||
357 existing->type == BTRFS_SHARED_BLOCK_REF_KEY);
359 * the action on the existing ref matches
360 * the action on the ref we're trying to add.
361 * Bump the ref_mod by one so the backref that
362 * is eventually added/removed has the correct
365 existing->ref_mod += update->ref_mod;
370 * helper function to update the accounting in the head ref
371 * existing and update must have the same bytenr
374 update_existing_head_ref(struct btrfs_delayed_ref_node *existing,
375 struct btrfs_delayed_ref_node *update)
377 struct btrfs_delayed_ref_head *existing_ref;
378 struct btrfs_delayed_ref_head *ref;
380 existing_ref = btrfs_delayed_node_to_head(existing);
381 ref = btrfs_delayed_node_to_head(update);
382 BUG_ON(existing_ref->is_data != ref->is_data);
384 if (ref->must_insert_reserved) {
385 /* if the extent was freed and then
386 * reallocated before the delayed ref
387 * entries were processed, we can end up
388 * with an existing head ref without
389 * the must_insert_reserved flag set.
392 existing_ref->must_insert_reserved = ref->must_insert_reserved;
395 * update the num_bytes so we make sure the accounting
398 existing->num_bytes = update->num_bytes;
402 if (ref->extent_op) {
403 if (!existing_ref->extent_op) {
404 existing_ref->extent_op = ref->extent_op;
406 if (ref->extent_op->update_key) {
407 memcpy(&existing_ref->extent_op->key,
408 &ref->extent_op->key,
409 sizeof(ref->extent_op->key));
410 existing_ref->extent_op->update_key = 1;
412 if (ref->extent_op->update_flags) {
413 existing_ref->extent_op->flags_to_set |=
414 ref->extent_op->flags_to_set;
415 existing_ref->extent_op->update_flags = 1;
417 kfree(ref->extent_op);
421 * update the reference mod on the head to reflect this new operation
423 existing->ref_mod += update->ref_mod;
427 * helper function to actually insert a head node into the rbtree.
428 * this does all the dirty work in terms of maintaining the correct
429 * overall modification count.
431 static noinline int add_delayed_ref_head(struct btrfs_trans_handle *trans,
432 struct btrfs_delayed_ref_node *ref,
433 u64 bytenr, u64 num_bytes,
434 int action, int is_data)
436 struct btrfs_delayed_ref_node *existing;
437 struct btrfs_delayed_ref_head *head_ref = NULL;
438 struct btrfs_delayed_ref_root *delayed_refs;
440 int must_insert_reserved = 0;
443 * the head node stores the sum of all the mods, so dropping a ref
444 * should drop the sum in the head node by one.
446 if (action == BTRFS_UPDATE_DELAYED_HEAD)
448 else if (action == BTRFS_DROP_DELAYED_REF)
452 * BTRFS_ADD_DELAYED_EXTENT means that we need to update
453 * the reserved accounting when the extent is finally added, or
454 * if a later modification deletes the delayed ref without ever
455 * inserting the extent into the extent allocation tree.
456 * ref->must_insert_reserved is the flag used to record
457 * that accounting mods are required.
459 * Once we record must_insert_reserved, switch the action to
460 * BTRFS_ADD_DELAYED_REF because other special casing is not required.
462 if (action == BTRFS_ADD_DELAYED_EXTENT)
463 must_insert_reserved = 1;
465 must_insert_reserved = 0;
467 delayed_refs = &trans->transaction->delayed_refs;
469 /* first set the basic ref node struct up */
470 atomic_set(&ref->refs, 1);
471 ref->bytenr = bytenr;
472 ref->num_bytes = num_bytes;
473 ref->ref_mod = count_mod;
479 head_ref = btrfs_delayed_node_to_head(ref);
480 head_ref->must_insert_reserved = must_insert_reserved;
481 head_ref->is_data = is_data;
483 INIT_LIST_HEAD(&head_ref->cluster);
484 mutex_init(&head_ref->mutex);
486 existing = tree_insert(&delayed_refs->root, &ref->rb_node);
489 update_existing_head_ref(existing, ref);
491 * we've updated the existing ref, free the newly
496 delayed_refs->num_heads++;
497 delayed_refs->num_heads_ready++;
498 delayed_refs->num_entries++;
499 trans->delayed_ref_updates++;
505 * helper to insert a delayed tree ref into the rbtree.
507 static noinline int add_delayed_tree_ref(struct btrfs_trans_handle *trans,
508 struct btrfs_delayed_ref_node *ref,
509 u64 bytenr, u64 num_bytes, u64 parent,
510 u64 ref_root, int level, int action)
512 struct btrfs_delayed_ref_node *existing;
513 struct btrfs_delayed_tree_ref *full_ref;
514 struct btrfs_delayed_ref_root *delayed_refs;
516 if (action == BTRFS_ADD_DELAYED_EXTENT)
517 action = BTRFS_ADD_DELAYED_REF;
519 delayed_refs = &trans->transaction->delayed_refs;
521 /* first set the basic ref node struct up */
522 atomic_set(&ref->refs, 1);
523 ref->bytenr = bytenr;
524 ref->num_bytes = num_bytes;
526 ref->action = action;
530 full_ref = btrfs_delayed_node_to_tree_ref(ref);
532 full_ref->parent = parent;
533 ref->type = BTRFS_SHARED_BLOCK_REF_KEY;
535 full_ref->root = ref_root;
536 ref->type = BTRFS_TREE_BLOCK_REF_KEY;
538 full_ref->level = level;
540 existing = tree_insert(&delayed_refs->root, &ref->rb_node);
543 update_existing_ref(trans, delayed_refs, existing, ref);
545 * we've updated the existing ref, free the newly
550 delayed_refs->num_entries++;
551 trans->delayed_ref_updates++;
557 * helper to insert a delayed data ref into the rbtree.
559 static noinline int add_delayed_data_ref(struct btrfs_trans_handle *trans,
560 struct btrfs_delayed_ref_node *ref,
561 u64 bytenr, u64 num_bytes, u64 parent,
562 u64 ref_root, u64 owner, u64 offset,
565 struct btrfs_delayed_ref_node *existing;
566 struct btrfs_delayed_data_ref *full_ref;
567 struct btrfs_delayed_ref_root *delayed_refs;
569 if (action == BTRFS_ADD_DELAYED_EXTENT)
570 action = BTRFS_ADD_DELAYED_REF;
572 delayed_refs = &trans->transaction->delayed_refs;
574 /* first set the basic ref node struct up */
575 atomic_set(&ref->refs, 1);
576 ref->bytenr = bytenr;
577 ref->num_bytes = num_bytes;
579 ref->action = action;
583 full_ref = btrfs_delayed_node_to_data_ref(ref);
585 full_ref->parent = parent;
586 ref->type = BTRFS_SHARED_DATA_REF_KEY;
588 full_ref->root = ref_root;
589 ref->type = BTRFS_EXTENT_DATA_REF_KEY;
591 full_ref->objectid = owner;
592 full_ref->offset = offset;
594 existing = tree_insert(&delayed_refs->root, &ref->rb_node);
597 update_existing_ref(trans, delayed_refs, existing, ref);
599 * we've updated the existing ref, free the newly
604 delayed_refs->num_entries++;
605 trans->delayed_ref_updates++;
611 * add a delayed tree ref. This does all of the accounting required
612 * to make sure the delayed ref is eventually processed before this
613 * transaction commits.
615 int btrfs_add_delayed_tree_ref(struct btrfs_trans_handle *trans,
616 u64 bytenr, u64 num_bytes, u64 parent,
617 u64 ref_root, int level, int action,
618 struct btrfs_delayed_extent_op *extent_op)
620 struct btrfs_delayed_tree_ref *ref;
621 struct btrfs_delayed_ref_head *head_ref;
622 struct btrfs_delayed_ref_root *delayed_refs;
625 BUG_ON(extent_op && extent_op->is_data);
626 ref = kmalloc(sizeof(*ref), GFP_NOFS);
630 head_ref = kmalloc(sizeof(*head_ref), GFP_NOFS);
636 head_ref->extent_op = extent_op;
638 delayed_refs = &trans->transaction->delayed_refs;
639 spin_lock(&delayed_refs->lock);
642 * insert both the head node and the new ref without dropping
645 ret = add_delayed_ref_head(trans, &head_ref->node, bytenr, num_bytes,
649 ret = add_delayed_tree_ref(trans, &ref->node, bytenr, num_bytes,
650 parent, ref_root, level, action);
652 spin_unlock(&delayed_refs->lock);
657 * add a delayed data ref. it's similar to btrfs_add_delayed_tree_ref.
659 int btrfs_add_delayed_data_ref(struct btrfs_trans_handle *trans,
660 u64 bytenr, u64 num_bytes,
661 u64 parent, u64 ref_root,
662 u64 owner, u64 offset, int action,
663 struct btrfs_delayed_extent_op *extent_op)
665 struct btrfs_delayed_data_ref *ref;
666 struct btrfs_delayed_ref_head *head_ref;
667 struct btrfs_delayed_ref_root *delayed_refs;
670 BUG_ON(extent_op && !extent_op->is_data);
671 ref = kmalloc(sizeof(*ref), GFP_NOFS);
675 head_ref = kmalloc(sizeof(*head_ref), GFP_NOFS);
681 head_ref->extent_op = extent_op;
683 delayed_refs = &trans->transaction->delayed_refs;
684 spin_lock(&delayed_refs->lock);
687 * insert both the head node and the new ref without dropping
690 ret = add_delayed_ref_head(trans, &head_ref->node, bytenr, num_bytes,
694 ret = add_delayed_data_ref(trans, &ref->node, bytenr, num_bytes,
695 parent, ref_root, owner, offset, action);
697 spin_unlock(&delayed_refs->lock);
701 int btrfs_add_delayed_extent_op(struct btrfs_trans_handle *trans,
702 u64 bytenr, u64 num_bytes,
703 struct btrfs_delayed_extent_op *extent_op)
705 struct btrfs_delayed_ref_head *head_ref;
706 struct btrfs_delayed_ref_root *delayed_refs;
709 head_ref = kmalloc(sizeof(*head_ref), GFP_NOFS);
713 head_ref->extent_op = extent_op;
715 delayed_refs = &trans->transaction->delayed_refs;
716 spin_lock(&delayed_refs->lock);
718 ret = add_delayed_ref_head(trans, &head_ref->node, bytenr,
719 num_bytes, BTRFS_UPDATE_DELAYED_HEAD,
723 spin_unlock(&delayed_refs->lock);
728 * this does a simple search for the head node for a given extent.
729 * It must be called with the delayed ref spinlock held, and it returns
730 * the head node if any where found, or NULL if not.
732 struct btrfs_delayed_ref_head *
733 btrfs_find_delayed_ref_head(struct btrfs_trans_handle *trans, u64 bytenr)
735 struct btrfs_delayed_ref_node *ref;
736 struct btrfs_delayed_ref_root *delayed_refs;
738 delayed_refs = &trans->transaction->delayed_refs;
739 ref = find_ref_head(&delayed_refs->root, bytenr, NULL);
741 return btrfs_delayed_node_to_head(ref);
746 * add a delayed ref to the tree. This does all of the accounting required
747 * to make sure the delayed ref is eventually processed before this
748 * transaction commits.
750 * The main point of this call is to add and remove a backreference in a single
751 * shot, taking the lock only once, and only searching for the head node once.
753 * It is the same as doing a ref add and delete in two separate calls.
756 int btrfs_update_delayed_ref(struct btrfs_trans_handle *trans,
757 u64 bytenr, u64 num_bytes, u64 orig_parent,
758 u64 parent, u64 orig_ref_root, u64 ref_root,
759 u64 orig_ref_generation, u64 ref_generation,
760 u64 owner_objectid, int pin)
762 struct btrfs_delayed_ref *ref;
763 struct btrfs_delayed_ref *old_ref;
764 struct btrfs_delayed_ref_head *head_ref;
765 struct btrfs_delayed_ref_root *delayed_refs;
768 ref = kmalloc(sizeof(*ref), GFP_NOFS);
772 old_ref = kmalloc(sizeof(*old_ref), GFP_NOFS);
779 * the parent = 0 case comes from cases where we don't actually
780 * know the parent yet. It will get updated later via a add/drop
785 if (orig_parent == 0)
786 orig_parent = bytenr;
788 head_ref = kmalloc(sizeof(*head_ref), GFP_NOFS);
794 delayed_refs = &trans->transaction->delayed_refs;
795 spin_lock(&delayed_refs->lock);
798 * insert both the head node and the new ref without dropping
801 ret = __btrfs_add_delayed_ref(trans, &head_ref->node, bytenr, num_bytes,
803 BTRFS_UPDATE_DELAYED_HEAD, 0);
806 ret = __btrfs_add_delayed_ref(trans, &ref->node, bytenr, num_bytes,
807 parent, ref_root, ref_generation,
808 owner_objectid, BTRFS_ADD_DELAYED_REF, 0);
811 ret = __btrfs_add_delayed_ref(trans, &old_ref->node, bytenr, num_bytes,
812 orig_parent, orig_ref_root,
813 orig_ref_generation, owner_objectid,
814 BTRFS_DROP_DELAYED_REF, pin);
816 spin_unlock(&delayed_refs->lock);