Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 2009 Oracle. All rights reserved.
4 */
5
6#include <linux/sched.h>
7#include <linux/slab.h>
8#include <linux/sort.h>
9#include "ctree.h"
10#include "delayed-ref.h"
11#include "transaction.h"
12#include "qgroup.h"
13#include "space-info.h"
14
15struct kmem_cache *btrfs_delayed_ref_head_cachep;
16struct kmem_cache *btrfs_delayed_tree_ref_cachep;
17struct kmem_cache *btrfs_delayed_data_ref_cachep;
18struct kmem_cache *btrfs_delayed_extent_op_cachep;
19/*
20 * delayed back reference update tracking. For subvolume trees
21 * we queue up extent allocations and backref maintenance for
22 * delayed processing. This avoids deep call chains where we
23 * add extents in the middle of btrfs_search_slot, and it allows
24 * us to buffer up frequently modified backrefs in an rb tree instead
25 * of hammering updates on the extent allocation tree.
26 */
27
28bool btrfs_check_space_for_delayed_refs(struct btrfs_fs_info *fs_info)
29{
30 struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv;
31 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
32 bool ret = false;
33 u64 reserved;
34
35 spin_lock(&global_rsv->lock);
36 reserved = global_rsv->reserved;
37 spin_unlock(&global_rsv->lock);
38
39 /*
40 * Since the global reserve is just kind of magic we don't really want
41 * to rely on it to save our bacon, so if our size is more than the
42 * delayed_refs_rsv and the global rsv then it's time to think about
43 * bailing.
44 */
45 spin_lock(&delayed_refs_rsv->lock);
46 reserved += delayed_refs_rsv->reserved;
47 if (delayed_refs_rsv->size >= reserved)
48 ret = true;
49 spin_unlock(&delayed_refs_rsv->lock);
50 return ret;
51}
52
53int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans)
54{
55 u64 num_entries =
56 atomic_read(&trans->transaction->delayed_refs.num_entries);
57 u64 avg_runtime;
58 u64 val;
59
60 smp_mb();
61 avg_runtime = trans->fs_info->avg_delayed_ref_runtime;
62 val = num_entries * avg_runtime;
63 if (val >= NSEC_PER_SEC)
64 return 1;
65 if (val >= NSEC_PER_SEC / 2)
66 return 2;
67
68 return btrfs_check_space_for_delayed_refs(trans->fs_info);
69}
70
71/**
72 * btrfs_delayed_refs_rsv_release - release a ref head's reservation.
73 * @fs_info - the fs_info for our fs.
74 * @nr - the number of items to drop.
75 *
76 * This drops the delayed ref head's count from the delayed refs rsv and frees
77 * any excess reservation we had.
78 */
79void btrfs_delayed_refs_rsv_release(struct btrfs_fs_info *fs_info, int nr)
80{
81 struct btrfs_block_rsv *block_rsv = &fs_info->delayed_refs_rsv;
82 u64 num_bytes = btrfs_calc_insert_metadata_size(fs_info, nr);
83 u64 released = 0;
84
85 released = __btrfs_block_rsv_release(fs_info, block_rsv, num_bytes,
86 NULL);
87 if (released)
88 trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv",
89 0, released, 0);
90}
91
92/*
93 * btrfs_update_delayed_refs_rsv - adjust the size of the delayed refs rsv
94 * @trans - the trans that may have generated delayed refs
95 *
96 * This is to be called anytime we may have adjusted trans->delayed_ref_updates,
97 * it'll calculate the additional size and add it to the delayed_refs_rsv.
98 */
99void btrfs_update_delayed_refs_rsv(struct btrfs_trans_handle *trans)
100{
101 struct btrfs_fs_info *fs_info = trans->fs_info;
102 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_refs_rsv;
103 u64 num_bytes;
104
105 if (!trans->delayed_ref_updates)
106 return;
107
108 num_bytes = btrfs_calc_insert_metadata_size(fs_info,
109 trans->delayed_ref_updates);
110 spin_lock(&delayed_rsv->lock);
111 delayed_rsv->size += num_bytes;
112 delayed_rsv->full = 0;
113 spin_unlock(&delayed_rsv->lock);
114 trans->delayed_ref_updates = 0;
115}
116
117/**
118 * btrfs_migrate_to_delayed_refs_rsv - transfer bytes to our delayed refs rsv.
119 * @fs_info - the fs info for our fs.
120 * @src - the source block rsv to transfer from.
121 * @num_bytes - the number of bytes to transfer.
122 *
123 * This transfers up to the num_bytes amount from the src rsv to the
124 * delayed_refs_rsv. Any extra bytes are returned to the space info.
125 */
126void btrfs_migrate_to_delayed_refs_rsv(struct btrfs_fs_info *fs_info,
127 struct btrfs_block_rsv *src,
128 u64 num_bytes)
129{
130 struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv;
131 u64 to_free = 0;
132
133 spin_lock(&src->lock);
134 src->reserved -= num_bytes;
135 src->size -= num_bytes;
136 spin_unlock(&src->lock);
137
138 spin_lock(&delayed_refs_rsv->lock);
139 if (delayed_refs_rsv->size > delayed_refs_rsv->reserved) {
140 u64 delta = delayed_refs_rsv->size -
141 delayed_refs_rsv->reserved;
142 if (num_bytes > delta) {
143 to_free = num_bytes - delta;
144 num_bytes = delta;
145 }
146 } else {
147 to_free = num_bytes;
148 num_bytes = 0;
149 }
150
151 if (num_bytes)
152 delayed_refs_rsv->reserved += num_bytes;
153 if (delayed_refs_rsv->reserved >= delayed_refs_rsv->size)
154 delayed_refs_rsv->full = 1;
155 spin_unlock(&delayed_refs_rsv->lock);
156
157 if (num_bytes)
158 trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv",
159 0, num_bytes, 1);
160 if (to_free)
161 btrfs_space_info_free_bytes_may_use(fs_info,
162 delayed_refs_rsv->space_info, to_free);
163}
164
165/**
166 * btrfs_delayed_refs_rsv_refill - refill based on our delayed refs usage.
167 * @fs_info - the fs_info for our fs.
168 * @flush - control how we can flush for this reservation.
169 *
170 * This will refill the delayed block_rsv up to 1 items size worth of space and
171 * will return -ENOSPC if we can't make the reservation.
172 */
173int btrfs_delayed_refs_rsv_refill(struct btrfs_fs_info *fs_info,
174 enum btrfs_reserve_flush_enum flush)
175{
176 struct btrfs_block_rsv *block_rsv = &fs_info->delayed_refs_rsv;
177 u64 limit = btrfs_calc_insert_metadata_size(fs_info, 1);
178 u64 num_bytes = 0;
179 int ret = -ENOSPC;
180
181 spin_lock(&block_rsv->lock);
182 if (block_rsv->reserved < block_rsv->size) {
183 num_bytes = block_rsv->size - block_rsv->reserved;
184 num_bytes = min(num_bytes, limit);
185 }
186 spin_unlock(&block_rsv->lock);
187
188 if (!num_bytes)
189 return 0;
190
191 ret = btrfs_reserve_metadata_bytes(fs_info->extent_root, block_rsv,
192 num_bytes, flush);
193 if (ret)
194 return ret;
195 btrfs_block_rsv_add_bytes(block_rsv, num_bytes, 0);
196 trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv",
197 0, num_bytes, 1);
198 return 0;
199}
200
201/*
202 * compare two delayed tree backrefs with same bytenr and type
203 */
204static int comp_tree_refs(struct btrfs_delayed_tree_ref *ref1,
205 struct btrfs_delayed_tree_ref *ref2)
206{
207 if (ref1->node.type == BTRFS_TREE_BLOCK_REF_KEY) {
208 if (ref1->root < ref2->root)
209 return -1;
210 if (ref1->root > ref2->root)
211 return 1;
212 } else {
213 if (ref1->parent < ref2->parent)
214 return -1;
215 if (ref1->parent > ref2->parent)
216 return 1;
217 }
218 return 0;
219}
220
221/*
222 * compare two delayed data backrefs with same bytenr and type
223 */
224static int comp_data_refs(struct btrfs_delayed_data_ref *ref1,
225 struct btrfs_delayed_data_ref *ref2)
226{
227 if (ref1->node.type == BTRFS_EXTENT_DATA_REF_KEY) {
228 if (ref1->root < ref2->root)
229 return -1;
230 if (ref1->root > ref2->root)
231 return 1;
232 if (ref1->objectid < ref2->objectid)
233 return -1;
234 if (ref1->objectid > ref2->objectid)
235 return 1;
236 if (ref1->offset < ref2->offset)
237 return -1;
238 if (ref1->offset > ref2->offset)
239 return 1;
240 } else {
241 if (ref1->parent < ref2->parent)
242 return -1;
243 if (ref1->parent > ref2->parent)
244 return 1;
245 }
246 return 0;
247}
248
249static int comp_refs(struct btrfs_delayed_ref_node *ref1,
250 struct btrfs_delayed_ref_node *ref2,
251 bool check_seq)
252{
253 int ret = 0;
254
255 if (ref1->type < ref2->type)
256 return -1;
257 if (ref1->type > ref2->type)
258 return 1;
259 if (ref1->type == BTRFS_TREE_BLOCK_REF_KEY ||
260 ref1->type == BTRFS_SHARED_BLOCK_REF_KEY)
261 ret = comp_tree_refs(btrfs_delayed_node_to_tree_ref(ref1),
262 btrfs_delayed_node_to_tree_ref(ref2));
263 else
264 ret = comp_data_refs(btrfs_delayed_node_to_data_ref(ref1),
265 btrfs_delayed_node_to_data_ref(ref2));
266 if (ret)
267 return ret;
268 if (check_seq) {
269 if (ref1->seq < ref2->seq)
270 return -1;
271 if (ref1->seq > ref2->seq)
272 return 1;
273 }
274 return 0;
275}
276
277/* insert a new ref to head ref rbtree */
278static struct btrfs_delayed_ref_head *htree_insert(struct rb_root_cached *root,
279 struct rb_node *node)
280{
281 struct rb_node **p = &root->rb_root.rb_node;
282 struct rb_node *parent_node = NULL;
283 struct btrfs_delayed_ref_head *entry;
284 struct btrfs_delayed_ref_head *ins;
285 u64 bytenr;
286 bool leftmost = true;
287
288 ins = rb_entry(node, struct btrfs_delayed_ref_head, href_node);
289 bytenr = ins->bytenr;
290 while (*p) {
291 parent_node = *p;
292 entry = rb_entry(parent_node, struct btrfs_delayed_ref_head,
293 href_node);
294
295 if (bytenr < entry->bytenr) {
296 p = &(*p)->rb_left;
297 } else if (bytenr > entry->bytenr) {
298 p = &(*p)->rb_right;
299 leftmost = false;
300 } else {
301 return entry;
302 }
303 }
304
305 rb_link_node(node, parent_node, p);
306 rb_insert_color_cached(node, root, leftmost);
307 return NULL;
308}
309
310static struct btrfs_delayed_ref_node* tree_insert(struct rb_root_cached *root,
311 struct btrfs_delayed_ref_node *ins)
312{
313 struct rb_node **p = &root->rb_root.rb_node;
314 struct rb_node *node = &ins->ref_node;
315 struct rb_node *parent_node = NULL;
316 struct btrfs_delayed_ref_node *entry;
317 bool leftmost = true;
318
319 while (*p) {
320 int comp;
321
322 parent_node = *p;
323 entry = rb_entry(parent_node, struct btrfs_delayed_ref_node,
324 ref_node);
325 comp = comp_refs(ins, entry, true);
326 if (comp < 0) {
327 p = &(*p)->rb_left;
328 } else if (comp > 0) {
329 p = &(*p)->rb_right;
330 leftmost = false;
331 } else {
332 return entry;
333 }
334 }
335
336 rb_link_node(node, parent_node, p);
337 rb_insert_color_cached(node, root, leftmost);
338 return NULL;
339}
340
341static struct btrfs_delayed_ref_head *find_first_ref_head(
342 struct btrfs_delayed_ref_root *dr)
343{
344 struct rb_node *n;
345 struct btrfs_delayed_ref_head *entry;
346
347 n = rb_first_cached(&dr->href_root);
348 if (!n)
349 return NULL;
350
351 entry = rb_entry(n, struct btrfs_delayed_ref_head, href_node);
352
353 return entry;
354}
355
356/*
357 * Find a head entry based on bytenr. This returns the delayed ref head if it
358 * was able to find one, or NULL if nothing was in that spot. If return_bigger
359 * is given, the next bigger entry is returned if no exact match is found.
360 */
361static struct btrfs_delayed_ref_head *find_ref_head(
362 struct btrfs_delayed_ref_root *dr, u64 bytenr,
363 bool return_bigger)
364{
365 struct rb_root *root = &dr->href_root.rb_root;
366 struct rb_node *n;
367 struct btrfs_delayed_ref_head *entry;
368
369 n = root->rb_node;
370 entry = NULL;
371 while (n) {
372 entry = rb_entry(n, struct btrfs_delayed_ref_head, href_node);
373
374 if (bytenr < entry->bytenr)
375 n = n->rb_left;
376 else if (bytenr > entry->bytenr)
377 n = n->rb_right;
378 else
379 return entry;
380 }
381 if (entry && return_bigger) {
382 if (bytenr > entry->bytenr) {
383 n = rb_next(&entry->href_node);
384 if (!n)
385 return NULL;
386 entry = rb_entry(n, struct btrfs_delayed_ref_head,
387 href_node);
388 }
389 return entry;
390 }
391 return NULL;
392}
393
394int btrfs_delayed_ref_lock(struct btrfs_delayed_ref_root *delayed_refs,
395 struct btrfs_delayed_ref_head *head)
396{
397 lockdep_assert_held(&delayed_refs->lock);
398 if (mutex_trylock(&head->mutex))
399 return 0;
400
401 refcount_inc(&head->refs);
402 spin_unlock(&delayed_refs->lock);
403
404 mutex_lock(&head->mutex);
405 spin_lock(&delayed_refs->lock);
406 if (RB_EMPTY_NODE(&head->href_node)) {
407 mutex_unlock(&head->mutex);
408 btrfs_put_delayed_ref_head(head);
409 return -EAGAIN;
410 }
411 btrfs_put_delayed_ref_head(head);
412 return 0;
413}
414
415static inline void drop_delayed_ref(struct btrfs_trans_handle *trans,
416 struct btrfs_delayed_ref_root *delayed_refs,
417 struct btrfs_delayed_ref_head *head,
418 struct btrfs_delayed_ref_node *ref)
419{
420 lockdep_assert_held(&head->lock);
421 rb_erase_cached(&ref->ref_node, &head->ref_tree);
422 RB_CLEAR_NODE(&ref->ref_node);
423 if (!list_empty(&ref->add_list))
424 list_del(&ref->add_list);
425 ref->in_tree = 0;
426 btrfs_put_delayed_ref(ref);
427 atomic_dec(&delayed_refs->num_entries);
428}
429
430static bool merge_ref(struct btrfs_trans_handle *trans,
431 struct btrfs_delayed_ref_root *delayed_refs,
432 struct btrfs_delayed_ref_head *head,
433 struct btrfs_delayed_ref_node *ref,
434 u64 seq)
435{
436 struct btrfs_delayed_ref_node *next;
437 struct rb_node *node = rb_next(&ref->ref_node);
438 bool done = false;
439
440 while (!done && node) {
441 int mod;
442
443 next = rb_entry(node, struct btrfs_delayed_ref_node, ref_node);
444 node = rb_next(node);
445 if (seq && next->seq >= seq)
446 break;
447 if (comp_refs(ref, next, false))
448 break;
449
450 if (ref->action == next->action) {
451 mod = next->ref_mod;
452 } else {
453 if (ref->ref_mod < next->ref_mod) {
454 swap(ref, next);
455 done = true;
456 }
457 mod = -next->ref_mod;
458 }
459
460 drop_delayed_ref(trans, delayed_refs, head, next);
461 ref->ref_mod += mod;
462 if (ref->ref_mod == 0) {
463 drop_delayed_ref(trans, delayed_refs, head, ref);
464 done = true;
465 } else {
466 /*
467 * Can't have multiples of the same ref on a tree block.
468 */
469 WARN_ON(ref->type == BTRFS_TREE_BLOCK_REF_KEY ||
470 ref->type == BTRFS_SHARED_BLOCK_REF_KEY);
471 }
472 }
473
474 return done;
475}
476
477void btrfs_merge_delayed_refs(struct btrfs_trans_handle *trans,
478 struct btrfs_delayed_ref_root *delayed_refs,
479 struct btrfs_delayed_ref_head *head)
480{
481 struct btrfs_fs_info *fs_info = trans->fs_info;
482 struct btrfs_delayed_ref_node *ref;
483 struct rb_node *node;
484 u64 seq = 0;
485
486 lockdep_assert_held(&head->lock);
487
488 if (RB_EMPTY_ROOT(&head->ref_tree.rb_root))
489 return;
490
491 /* We don't have too many refs to merge for data. */
492 if (head->is_data)
493 return;
494
495 spin_lock(&fs_info->tree_mod_seq_lock);
496 if (!list_empty(&fs_info->tree_mod_seq_list)) {
497 struct seq_list *elem;
498
499 elem = list_first_entry(&fs_info->tree_mod_seq_list,
500 struct seq_list, list);
501 seq = elem->seq;
502 }
503 spin_unlock(&fs_info->tree_mod_seq_lock);
504
505again:
506 for (node = rb_first_cached(&head->ref_tree); node;
507 node = rb_next(node)) {
508 ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node);
509 if (seq && ref->seq >= seq)
510 continue;
511 if (merge_ref(trans, delayed_refs, head, ref, seq))
512 goto again;
513 }
514}
515
516int btrfs_check_delayed_seq(struct btrfs_fs_info *fs_info, u64 seq)
517{
518 struct seq_list *elem;
519 int ret = 0;
520
521 spin_lock(&fs_info->tree_mod_seq_lock);
522 if (!list_empty(&fs_info->tree_mod_seq_list)) {
523 elem = list_first_entry(&fs_info->tree_mod_seq_list,
524 struct seq_list, list);
525 if (seq >= elem->seq) {
526 btrfs_debug(fs_info,
527 "holding back delayed_ref %#x.%x, lowest is %#x.%x",
528 (u32)(seq >> 32), (u32)seq,
529 (u32)(elem->seq >> 32), (u32)elem->seq);
530 ret = 1;
531 }
532 }
533
534 spin_unlock(&fs_info->tree_mod_seq_lock);
535 return ret;
536}
537
538struct btrfs_delayed_ref_head *btrfs_select_ref_head(
539 struct btrfs_delayed_ref_root *delayed_refs)
540{
541 struct btrfs_delayed_ref_head *head;
542
543again:
544 head = find_ref_head(delayed_refs, delayed_refs->run_delayed_start,
545 true);
546 if (!head && delayed_refs->run_delayed_start != 0) {
547 delayed_refs->run_delayed_start = 0;
548 head = find_first_ref_head(delayed_refs);
549 }
550 if (!head)
551 return NULL;
552
553 while (head->processing) {
554 struct rb_node *node;
555
556 node = rb_next(&head->href_node);
557 if (!node) {
558 if (delayed_refs->run_delayed_start == 0)
559 return NULL;
560 delayed_refs->run_delayed_start = 0;
561 goto again;
562 }
563 head = rb_entry(node, struct btrfs_delayed_ref_head,
564 href_node);
565 }
566
567 head->processing = 1;
568 WARN_ON(delayed_refs->num_heads_ready == 0);
569 delayed_refs->num_heads_ready--;
570 delayed_refs->run_delayed_start = head->bytenr +
571 head->num_bytes;
572 return head;
573}
574
575void btrfs_delete_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
576 struct btrfs_delayed_ref_head *head)
577{
578 lockdep_assert_held(&delayed_refs->lock);
579 lockdep_assert_held(&head->lock);
580
581 rb_erase_cached(&head->href_node, &delayed_refs->href_root);
582 RB_CLEAR_NODE(&head->href_node);
583 atomic_dec(&delayed_refs->num_entries);
584 delayed_refs->num_heads--;
585 if (head->processing == 0)
586 delayed_refs->num_heads_ready--;
587}
588
589/*
590 * Helper to insert the ref_node to the tail or merge with tail.
591 *
592 * Return 0 for insert.
593 * Return >0 for merge.
594 */
595static int insert_delayed_ref(struct btrfs_trans_handle *trans,
596 struct btrfs_delayed_ref_root *root,
597 struct btrfs_delayed_ref_head *href,
598 struct btrfs_delayed_ref_node *ref)
599{
600 struct btrfs_delayed_ref_node *exist;
601 int mod;
602 int ret = 0;
603
604 spin_lock(&href->lock);
605 exist = tree_insert(&href->ref_tree, ref);
606 if (!exist)
607 goto inserted;
608
609 /* Now we are sure we can merge */
610 ret = 1;
611 if (exist->action == ref->action) {
612 mod = ref->ref_mod;
613 } else {
614 /* Need to change action */
615 if (exist->ref_mod < ref->ref_mod) {
616 exist->action = ref->action;
617 mod = -exist->ref_mod;
618 exist->ref_mod = ref->ref_mod;
619 if (ref->action == BTRFS_ADD_DELAYED_REF)
620 list_add_tail(&exist->add_list,
621 &href->ref_add_list);
622 else if (ref->action == BTRFS_DROP_DELAYED_REF) {
623 ASSERT(!list_empty(&exist->add_list));
624 list_del(&exist->add_list);
625 } else {
626 ASSERT(0);
627 }
628 } else
629 mod = -ref->ref_mod;
630 }
631 exist->ref_mod += mod;
632
633 /* remove existing tail if its ref_mod is zero */
634 if (exist->ref_mod == 0)
635 drop_delayed_ref(trans, root, href, exist);
636 spin_unlock(&href->lock);
637 return ret;
638inserted:
639 if (ref->action == BTRFS_ADD_DELAYED_REF)
640 list_add_tail(&ref->add_list, &href->ref_add_list);
641 atomic_inc(&root->num_entries);
642 spin_unlock(&href->lock);
643 return ret;
644}
645
646/*
647 * helper function to update the accounting in the head ref
648 * existing and update must have the same bytenr
649 */
650static noinline void update_existing_head_ref(struct btrfs_trans_handle *trans,
651 struct btrfs_delayed_ref_head *existing,
652 struct btrfs_delayed_ref_head *update,
653 int *old_ref_mod_ret)
654{
655 struct btrfs_delayed_ref_root *delayed_refs =
656 &trans->transaction->delayed_refs;
657 struct btrfs_fs_info *fs_info = trans->fs_info;
658 int old_ref_mod;
659
660 BUG_ON(existing->is_data != update->is_data);
661
662 spin_lock(&existing->lock);
663 if (update->must_insert_reserved) {
664 /* if the extent was freed and then
665 * reallocated before the delayed ref
666 * entries were processed, we can end up
667 * with an existing head ref without
668 * the must_insert_reserved flag set.
669 * Set it again here
670 */
671 existing->must_insert_reserved = update->must_insert_reserved;
672
673 /*
674 * update the num_bytes so we make sure the accounting
675 * is done correctly
676 */
677 existing->num_bytes = update->num_bytes;
678
679 }
680
681 if (update->extent_op) {
682 if (!existing->extent_op) {
683 existing->extent_op = update->extent_op;
684 } else {
685 if (update->extent_op->update_key) {
686 memcpy(&existing->extent_op->key,
687 &update->extent_op->key,
688 sizeof(update->extent_op->key));
689 existing->extent_op->update_key = true;
690 }
691 if (update->extent_op->update_flags) {
692 existing->extent_op->flags_to_set |=
693 update->extent_op->flags_to_set;
694 existing->extent_op->update_flags = true;
695 }
696 btrfs_free_delayed_extent_op(update->extent_op);
697 }
698 }
699 /*
700 * update the reference mod on the head to reflect this new operation,
701 * only need the lock for this case cause we could be processing it
702 * currently, for refs we just added we know we're a-ok.
703 */
704 old_ref_mod = existing->total_ref_mod;
705 if (old_ref_mod_ret)
706 *old_ref_mod_ret = old_ref_mod;
707 existing->ref_mod += update->ref_mod;
708 existing->total_ref_mod += update->ref_mod;
709
710 /*
711 * If we are going to from a positive ref mod to a negative or vice
712 * versa we need to make sure to adjust pending_csums accordingly.
713 */
714 if (existing->is_data) {
715 u64 csum_leaves =
716 btrfs_csum_bytes_to_leaves(fs_info,
717 existing->num_bytes);
718
719 if (existing->total_ref_mod >= 0 && old_ref_mod < 0) {
720 delayed_refs->pending_csums -= existing->num_bytes;
721 btrfs_delayed_refs_rsv_release(fs_info, csum_leaves);
722 }
723 if (existing->total_ref_mod < 0 && old_ref_mod >= 0) {
724 delayed_refs->pending_csums += existing->num_bytes;
725 trans->delayed_ref_updates += csum_leaves;
726 }
727 }
728 spin_unlock(&existing->lock);
729}
730
731static void init_delayed_ref_head(struct btrfs_delayed_ref_head *head_ref,
732 struct btrfs_qgroup_extent_record *qrecord,
733 u64 bytenr, u64 num_bytes, u64 ref_root,
734 u64 reserved, int action, bool is_data,
735 bool is_system)
736{
737 int count_mod = 1;
738 int must_insert_reserved = 0;
739
740 /* If reserved is provided, it must be a data extent. */
741 BUG_ON(!is_data && reserved);
742
743 /*
744 * The head node stores the sum of all the mods, so dropping a ref
745 * should drop the sum in the head node by one.
746 */
747 if (action == BTRFS_UPDATE_DELAYED_HEAD)
748 count_mod = 0;
749 else if (action == BTRFS_DROP_DELAYED_REF)
750 count_mod = -1;
751
752 /*
753 * BTRFS_ADD_DELAYED_EXTENT means that we need to update the reserved
754 * accounting when the extent is finally added, or if a later
755 * modification deletes the delayed ref without ever inserting the
756 * extent into the extent allocation tree. ref->must_insert_reserved
757 * is the flag used to record that accounting mods are required.
758 *
759 * Once we record must_insert_reserved, switch the action to
760 * BTRFS_ADD_DELAYED_REF because other special casing is not required.
761 */
762 if (action == BTRFS_ADD_DELAYED_EXTENT)
763 must_insert_reserved = 1;
764 else
765 must_insert_reserved = 0;
766
767 refcount_set(&head_ref->refs, 1);
768 head_ref->bytenr = bytenr;
769 head_ref->num_bytes = num_bytes;
770 head_ref->ref_mod = count_mod;
771 head_ref->must_insert_reserved = must_insert_reserved;
772 head_ref->is_data = is_data;
773 head_ref->is_system = is_system;
774 head_ref->ref_tree = RB_ROOT_CACHED;
775 INIT_LIST_HEAD(&head_ref->ref_add_list);
776 RB_CLEAR_NODE(&head_ref->href_node);
777 head_ref->processing = 0;
778 head_ref->total_ref_mod = count_mod;
779 spin_lock_init(&head_ref->lock);
780 mutex_init(&head_ref->mutex);
781
782 if (qrecord) {
783 if (ref_root && reserved) {
784 qrecord->data_rsv = reserved;
785 qrecord->data_rsv_refroot = ref_root;
786 }
787 qrecord->bytenr = bytenr;
788 qrecord->num_bytes = num_bytes;
789 qrecord->old_roots = NULL;
790 }
791}
792
793/*
794 * helper function to actually insert a head node into the rbtree.
795 * this does all the dirty work in terms of maintaining the correct
796 * overall modification count.
797 */
798static noinline struct btrfs_delayed_ref_head *
799add_delayed_ref_head(struct btrfs_trans_handle *trans,
800 struct btrfs_delayed_ref_head *head_ref,
801 struct btrfs_qgroup_extent_record *qrecord,
802 int action, int *qrecord_inserted_ret,
803 int *old_ref_mod, int *new_ref_mod)
804{
805 struct btrfs_delayed_ref_head *existing;
806 struct btrfs_delayed_ref_root *delayed_refs;
807 int qrecord_inserted = 0;
808
809 delayed_refs = &trans->transaction->delayed_refs;
810
811 /* Record qgroup extent info if provided */
812 if (qrecord) {
813 if (btrfs_qgroup_trace_extent_nolock(trans->fs_info,
814 delayed_refs, qrecord))
815 kfree(qrecord);
816 else
817 qrecord_inserted = 1;
818 }
819
820 trace_add_delayed_ref_head(trans->fs_info, head_ref, action);
821
822 existing = htree_insert(&delayed_refs->href_root,
823 &head_ref->href_node);
824 if (existing) {
825 update_existing_head_ref(trans, existing, head_ref,
826 old_ref_mod);
827 /*
828 * we've updated the existing ref, free the newly
829 * allocated ref
830 */
831 kmem_cache_free(btrfs_delayed_ref_head_cachep, head_ref);
832 head_ref = existing;
833 } else {
834 if (old_ref_mod)
835 *old_ref_mod = 0;
836 if (head_ref->is_data && head_ref->ref_mod < 0) {
837 delayed_refs->pending_csums += head_ref->num_bytes;
838 trans->delayed_ref_updates +=
839 btrfs_csum_bytes_to_leaves(trans->fs_info,
840 head_ref->num_bytes);
841 }
842 delayed_refs->num_heads++;
843 delayed_refs->num_heads_ready++;
844 atomic_inc(&delayed_refs->num_entries);
845 trans->delayed_ref_updates++;
846 }
847 if (qrecord_inserted_ret)
848 *qrecord_inserted_ret = qrecord_inserted;
849 if (new_ref_mod)
850 *new_ref_mod = head_ref->total_ref_mod;
851
852 return head_ref;
853}
854
855/*
856 * init_delayed_ref_common - Initialize the structure which represents a
857 * modification to a an extent.
858 *
859 * @fs_info: Internal to the mounted filesystem mount structure.
860 *
861 * @ref: The structure which is going to be initialized.
862 *
863 * @bytenr: The logical address of the extent for which a modification is
864 * going to be recorded.
865 *
866 * @num_bytes: Size of the extent whose modification is being recorded.
867 *
868 * @ref_root: The id of the root where this modification has originated, this
869 * can be either one of the well-known metadata trees or the
870 * subvolume id which references this extent.
871 *
872 * @action: Can be one of BTRFS_ADD_DELAYED_REF/BTRFS_DROP_DELAYED_REF or
873 * BTRFS_ADD_DELAYED_EXTENT
874 *
875 * @ref_type: Holds the type of the extent which is being recorded, can be
876 * one of BTRFS_SHARED_BLOCK_REF_KEY/BTRFS_TREE_BLOCK_REF_KEY
877 * when recording a metadata extent or BTRFS_SHARED_DATA_REF_KEY/
878 * BTRFS_EXTENT_DATA_REF_KEY when recording data extent
879 */
880static void init_delayed_ref_common(struct btrfs_fs_info *fs_info,
881 struct btrfs_delayed_ref_node *ref,
882 u64 bytenr, u64 num_bytes, u64 ref_root,
883 int action, u8 ref_type)
884{
885 u64 seq = 0;
886
887 if (action == BTRFS_ADD_DELAYED_EXTENT)
888 action = BTRFS_ADD_DELAYED_REF;
889
890 if (is_fstree(ref_root))
891 seq = atomic64_read(&fs_info->tree_mod_seq);
892
893 refcount_set(&ref->refs, 1);
894 ref->bytenr = bytenr;
895 ref->num_bytes = num_bytes;
896 ref->ref_mod = 1;
897 ref->action = action;
898 ref->is_head = 0;
899 ref->in_tree = 1;
900 ref->seq = seq;
901 ref->type = ref_type;
902 RB_CLEAR_NODE(&ref->ref_node);
903 INIT_LIST_HEAD(&ref->add_list);
904}
905
906/*
907 * add a delayed tree ref. This does all of the accounting required
908 * to make sure the delayed ref is eventually processed before this
909 * transaction commits.
910 */
911int btrfs_add_delayed_tree_ref(struct btrfs_trans_handle *trans,
912 struct btrfs_ref *generic_ref,
913 struct btrfs_delayed_extent_op *extent_op,
914 int *old_ref_mod, int *new_ref_mod)
915{
916 struct btrfs_fs_info *fs_info = trans->fs_info;
917 struct btrfs_delayed_tree_ref *ref;
918 struct btrfs_delayed_ref_head *head_ref;
919 struct btrfs_delayed_ref_root *delayed_refs;
920 struct btrfs_qgroup_extent_record *record = NULL;
921 int qrecord_inserted;
922 bool is_system;
923 int action = generic_ref->action;
924 int level = generic_ref->tree_ref.level;
925 int ret;
926 u64 bytenr = generic_ref->bytenr;
927 u64 num_bytes = generic_ref->len;
928 u64 parent = generic_ref->parent;
929 u8 ref_type;
930
931 is_system = (generic_ref->real_root == BTRFS_CHUNK_TREE_OBJECTID);
932
933 ASSERT(generic_ref->type == BTRFS_REF_METADATA && generic_ref->action);
934 BUG_ON(extent_op && extent_op->is_data);
935 ref = kmem_cache_alloc(btrfs_delayed_tree_ref_cachep, GFP_NOFS);
936 if (!ref)
937 return -ENOMEM;
938
939 head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS);
940 if (!head_ref) {
941 kmem_cache_free(btrfs_delayed_tree_ref_cachep, ref);
942 return -ENOMEM;
943 }
944
945 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) &&
946 is_fstree(generic_ref->real_root) &&
947 is_fstree(generic_ref->tree_ref.root) &&
948 !generic_ref->skip_qgroup) {
949 record = kzalloc(sizeof(*record), GFP_NOFS);
950 if (!record) {
951 kmem_cache_free(btrfs_delayed_tree_ref_cachep, ref);
952 kmem_cache_free(btrfs_delayed_ref_head_cachep, head_ref);
953 return -ENOMEM;
954 }
955 }
956
957 if (parent)
958 ref_type = BTRFS_SHARED_BLOCK_REF_KEY;
959 else
960 ref_type = BTRFS_TREE_BLOCK_REF_KEY;
961
962 init_delayed_ref_common(fs_info, &ref->node, bytenr, num_bytes,
963 generic_ref->tree_ref.root, action, ref_type);
964 ref->root = generic_ref->tree_ref.root;
965 ref->parent = parent;
966 ref->level = level;
967
968 init_delayed_ref_head(head_ref, record, bytenr, num_bytes,
969 generic_ref->tree_ref.root, 0, action, false,
970 is_system);
971 head_ref->extent_op = extent_op;
972
973 delayed_refs = &trans->transaction->delayed_refs;
974 spin_lock(&delayed_refs->lock);
975
976 /*
977 * insert both the head node and the new ref without dropping
978 * the spin lock
979 */
980 head_ref = add_delayed_ref_head(trans, head_ref, record,
981 action, &qrecord_inserted,
982 old_ref_mod, new_ref_mod);
983
984 ret = insert_delayed_ref(trans, delayed_refs, head_ref, &ref->node);
985 spin_unlock(&delayed_refs->lock);
986
987 /*
988 * Need to update the delayed_refs_rsv with any changes we may have
989 * made.
990 */
991 btrfs_update_delayed_refs_rsv(trans);
992
993 trace_add_delayed_tree_ref(fs_info, &ref->node, ref,
994 action == BTRFS_ADD_DELAYED_EXTENT ?
995 BTRFS_ADD_DELAYED_REF : action);
996 if (ret > 0)
997 kmem_cache_free(btrfs_delayed_tree_ref_cachep, ref);
998
999 if (qrecord_inserted)
1000 btrfs_qgroup_trace_extent_post(fs_info, record);
1001
1002 return 0;
1003}
1004
1005/*
1006 * add a delayed data ref. it's similar to btrfs_add_delayed_tree_ref.
1007 */
1008int btrfs_add_delayed_data_ref(struct btrfs_trans_handle *trans,
1009 struct btrfs_ref *generic_ref,
1010 u64 reserved, int *old_ref_mod,
1011 int *new_ref_mod)
1012{
1013 struct btrfs_fs_info *fs_info = trans->fs_info;
1014 struct btrfs_delayed_data_ref *ref;
1015 struct btrfs_delayed_ref_head *head_ref;
1016 struct btrfs_delayed_ref_root *delayed_refs;
1017 struct btrfs_qgroup_extent_record *record = NULL;
1018 int qrecord_inserted;
1019 int action = generic_ref->action;
1020 int ret;
1021 u64 bytenr = generic_ref->bytenr;
1022 u64 num_bytes = generic_ref->len;
1023 u64 parent = generic_ref->parent;
1024 u64 ref_root = generic_ref->data_ref.ref_root;
1025 u64 owner = generic_ref->data_ref.ino;
1026 u64 offset = generic_ref->data_ref.offset;
1027 u8 ref_type;
1028
1029 ASSERT(generic_ref->type == BTRFS_REF_DATA && action);
1030 ref = kmem_cache_alloc(btrfs_delayed_data_ref_cachep, GFP_NOFS);
1031 if (!ref)
1032 return -ENOMEM;
1033
1034 if (parent)
1035 ref_type = BTRFS_SHARED_DATA_REF_KEY;
1036 else
1037 ref_type = BTRFS_EXTENT_DATA_REF_KEY;
1038 init_delayed_ref_common(fs_info, &ref->node, bytenr, num_bytes,
1039 ref_root, action, ref_type);
1040 ref->root = ref_root;
1041 ref->parent = parent;
1042 ref->objectid = owner;
1043 ref->offset = offset;
1044
1045
1046 head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS);
1047 if (!head_ref) {
1048 kmem_cache_free(btrfs_delayed_data_ref_cachep, ref);
1049 return -ENOMEM;
1050 }
1051
1052 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) &&
1053 is_fstree(ref_root) &&
1054 is_fstree(generic_ref->real_root) &&
1055 !generic_ref->skip_qgroup) {
1056 record = kzalloc(sizeof(*record), GFP_NOFS);
1057 if (!record) {
1058 kmem_cache_free(btrfs_delayed_data_ref_cachep, ref);
1059 kmem_cache_free(btrfs_delayed_ref_head_cachep,
1060 head_ref);
1061 return -ENOMEM;
1062 }
1063 }
1064
1065 init_delayed_ref_head(head_ref, record, bytenr, num_bytes, ref_root,
1066 reserved, action, true, false);
1067 head_ref->extent_op = NULL;
1068
1069 delayed_refs = &trans->transaction->delayed_refs;
1070 spin_lock(&delayed_refs->lock);
1071
1072 /*
1073 * insert both the head node and the new ref without dropping
1074 * the spin lock
1075 */
1076 head_ref = add_delayed_ref_head(trans, head_ref, record,
1077 action, &qrecord_inserted,
1078 old_ref_mod, new_ref_mod);
1079
1080 ret = insert_delayed_ref(trans, delayed_refs, head_ref, &ref->node);
1081 spin_unlock(&delayed_refs->lock);
1082
1083 /*
1084 * Need to update the delayed_refs_rsv with any changes we may have
1085 * made.
1086 */
1087 btrfs_update_delayed_refs_rsv(trans);
1088
1089 trace_add_delayed_data_ref(trans->fs_info, &ref->node, ref,
1090 action == BTRFS_ADD_DELAYED_EXTENT ?
1091 BTRFS_ADD_DELAYED_REF : action);
1092 if (ret > 0)
1093 kmem_cache_free(btrfs_delayed_data_ref_cachep, ref);
1094
1095
1096 if (qrecord_inserted)
1097 return btrfs_qgroup_trace_extent_post(fs_info, record);
1098 return 0;
1099}
1100
1101int btrfs_add_delayed_extent_op(struct btrfs_trans_handle *trans,
1102 u64 bytenr, u64 num_bytes,
1103 struct btrfs_delayed_extent_op *extent_op)
1104{
1105 struct btrfs_delayed_ref_head *head_ref;
1106 struct btrfs_delayed_ref_root *delayed_refs;
1107
1108 head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS);
1109 if (!head_ref)
1110 return -ENOMEM;
1111
1112 init_delayed_ref_head(head_ref, NULL, bytenr, num_bytes, 0, 0,
1113 BTRFS_UPDATE_DELAYED_HEAD, extent_op->is_data,
1114 false);
1115 head_ref->extent_op = extent_op;
1116
1117 delayed_refs = &trans->transaction->delayed_refs;
1118 spin_lock(&delayed_refs->lock);
1119
1120 add_delayed_ref_head(trans, head_ref, NULL, BTRFS_UPDATE_DELAYED_HEAD,
1121 NULL, NULL, NULL);
1122
1123 spin_unlock(&delayed_refs->lock);
1124
1125 /*
1126 * Need to update the delayed_refs_rsv with any changes we may have
1127 * made.
1128 */
1129 btrfs_update_delayed_refs_rsv(trans);
1130 return 0;
1131}
1132
1133/*
1134 * This does a simple search for the head node for a given extent. Returns the
1135 * head node if found, or NULL if not.
1136 */
1137struct btrfs_delayed_ref_head *
1138btrfs_find_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs, u64 bytenr)
1139{
1140 lockdep_assert_held(&delayed_refs->lock);
1141
1142 return find_ref_head(delayed_refs, bytenr, false);
1143}
1144
1145void __cold btrfs_delayed_ref_exit(void)
1146{
1147 kmem_cache_destroy(btrfs_delayed_ref_head_cachep);
1148 kmem_cache_destroy(btrfs_delayed_tree_ref_cachep);
1149 kmem_cache_destroy(btrfs_delayed_data_ref_cachep);
1150 kmem_cache_destroy(btrfs_delayed_extent_op_cachep);
1151}
1152
1153int __init btrfs_delayed_ref_init(void)
1154{
1155 btrfs_delayed_ref_head_cachep = kmem_cache_create(
1156 "btrfs_delayed_ref_head",
1157 sizeof(struct btrfs_delayed_ref_head), 0,
1158 SLAB_MEM_SPREAD, NULL);
1159 if (!btrfs_delayed_ref_head_cachep)
1160 goto fail;
1161
1162 btrfs_delayed_tree_ref_cachep = kmem_cache_create(
1163 "btrfs_delayed_tree_ref",
1164 sizeof(struct btrfs_delayed_tree_ref), 0,
1165 SLAB_MEM_SPREAD, NULL);
1166 if (!btrfs_delayed_tree_ref_cachep)
1167 goto fail;
1168
1169 btrfs_delayed_data_ref_cachep = kmem_cache_create(
1170 "btrfs_delayed_data_ref",
1171 sizeof(struct btrfs_delayed_data_ref), 0,
1172 SLAB_MEM_SPREAD, NULL);
1173 if (!btrfs_delayed_data_ref_cachep)
1174 goto fail;
1175
1176 btrfs_delayed_extent_op_cachep = kmem_cache_create(
1177 "btrfs_delayed_extent_op",
1178 sizeof(struct btrfs_delayed_extent_op), 0,
1179 SLAB_MEM_SPREAD, NULL);
1180 if (!btrfs_delayed_extent_op_cachep)
1181 goto fail;
1182
1183 return 0;
1184fail:
1185 btrfs_delayed_ref_exit();
1186 return -ENOMEM;
1187}
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#include "qgroup.h"
26
27struct kmem_cache *btrfs_delayed_ref_head_cachep;
28struct kmem_cache *btrfs_delayed_tree_ref_cachep;
29struct kmem_cache *btrfs_delayed_data_ref_cachep;
30struct kmem_cache *btrfs_delayed_extent_op_cachep;
31/*
32 * delayed back reference update tracking. For subvolume trees
33 * we queue up extent allocations and backref maintenance for
34 * delayed processing. This avoids deep call chains where we
35 * add extents in the middle of btrfs_search_slot, and it allows
36 * us to buffer up frequently modified backrefs in an rb tree instead
37 * of hammering updates on the extent allocation tree.
38 */
39
40/*
41 * compare two delayed tree backrefs with same bytenr and type
42 */
43static int comp_tree_refs(struct btrfs_delayed_tree_ref *ref2,
44 struct btrfs_delayed_tree_ref *ref1, int type)
45{
46 if (type == BTRFS_TREE_BLOCK_REF_KEY) {
47 if (ref1->root < ref2->root)
48 return -1;
49 if (ref1->root > ref2->root)
50 return 1;
51 } else {
52 if (ref1->parent < ref2->parent)
53 return -1;
54 if (ref1->parent > ref2->parent)
55 return 1;
56 }
57 return 0;
58}
59
60/*
61 * compare two delayed data backrefs with same bytenr and type
62 */
63static int comp_data_refs(struct btrfs_delayed_data_ref *ref2,
64 struct btrfs_delayed_data_ref *ref1)
65{
66 if (ref1->node.type == BTRFS_EXTENT_DATA_REF_KEY) {
67 if (ref1->root < ref2->root)
68 return -1;
69 if (ref1->root > ref2->root)
70 return 1;
71 if (ref1->objectid < ref2->objectid)
72 return -1;
73 if (ref1->objectid > ref2->objectid)
74 return 1;
75 if (ref1->offset < ref2->offset)
76 return -1;
77 if (ref1->offset > ref2->offset)
78 return 1;
79 } else {
80 if (ref1->parent < ref2->parent)
81 return -1;
82 if (ref1->parent > ref2->parent)
83 return 1;
84 }
85 return 0;
86}
87
88/* insert a new ref to head ref rbtree */
89static struct btrfs_delayed_ref_head *htree_insert(struct rb_root *root,
90 struct rb_node *node)
91{
92 struct rb_node **p = &root->rb_node;
93 struct rb_node *parent_node = NULL;
94 struct btrfs_delayed_ref_head *entry;
95 struct btrfs_delayed_ref_head *ins;
96 u64 bytenr;
97
98 ins = rb_entry(node, struct btrfs_delayed_ref_head, href_node);
99 bytenr = ins->node.bytenr;
100 while (*p) {
101 parent_node = *p;
102 entry = rb_entry(parent_node, struct btrfs_delayed_ref_head,
103 href_node);
104
105 if (bytenr < entry->node.bytenr)
106 p = &(*p)->rb_left;
107 else if (bytenr > entry->node.bytenr)
108 p = &(*p)->rb_right;
109 else
110 return entry;
111 }
112
113 rb_link_node(node, parent_node, p);
114 rb_insert_color(node, root);
115 return NULL;
116}
117
118/*
119 * find an head entry based on bytenr. This returns the delayed ref
120 * head if it was able to find one, or NULL if nothing was in that spot.
121 * If return_bigger is given, the next bigger entry is returned if no exact
122 * match is found.
123 */
124static struct btrfs_delayed_ref_head *
125find_ref_head(struct rb_root *root, u64 bytenr,
126 int return_bigger)
127{
128 struct rb_node *n;
129 struct btrfs_delayed_ref_head *entry;
130
131 n = root->rb_node;
132 entry = NULL;
133 while (n) {
134 entry = rb_entry(n, struct btrfs_delayed_ref_head, href_node);
135
136 if (bytenr < entry->node.bytenr)
137 n = n->rb_left;
138 else if (bytenr > entry->node.bytenr)
139 n = n->rb_right;
140 else
141 return entry;
142 }
143 if (entry && return_bigger) {
144 if (bytenr > entry->node.bytenr) {
145 n = rb_next(&entry->href_node);
146 if (!n)
147 n = rb_first(root);
148 entry = rb_entry(n, struct btrfs_delayed_ref_head,
149 href_node);
150 return entry;
151 }
152 return entry;
153 }
154 return NULL;
155}
156
157int btrfs_delayed_ref_lock(struct btrfs_trans_handle *trans,
158 struct btrfs_delayed_ref_head *head)
159{
160 struct btrfs_delayed_ref_root *delayed_refs;
161
162 delayed_refs = &trans->transaction->delayed_refs;
163 assert_spin_locked(&delayed_refs->lock);
164 if (mutex_trylock(&head->mutex))
165 return 0;
166
167 atomic_inc(&head->node.refs);
168 spin_unlock(&delayed_refs->lock);
169
170 mutex_lock(&head->mutex);
171 spin_lock(&delayed_refs->lock);
172 if (!head->node.in_tree) {
173 mutex_unlock(&head->mutex);
174 btrfs_put_delayed_ref(&head->node);
175 return -EAGAIN;
176 }
177 btrfs_put_delayed_ref(&head->node);
178 return 0;
179}
180
181static inline void drop_delayed_ref(struct btrfs_trans_handle *trans,
182 struct btrfs_delayed_ref_root *delayed_refs,
183 struct btrfs_delayed_ref_head *head,
184 struct btrfs_delayed_ref_node *ref)
185{
186 if (btrfs_delayed_ref_is_head(ref)) {
187 head = btrfs_delayed_node_to_head(ref);
188 rb_erase(&head->href_node, &delayed_refs->href_root);
189 } else {
190 assert_spin_locked(&head->lock);
191 list_del(&ref->list);
192 }
193 ref->in_tree = 0;
194 btrfs_put_delayed_ref(ref);
195 atomic_dec(&delayed_refs->num_entries);
196 if (trans->delayed_ref_updates)
197 trans->delayed_ref_updates--;
198}
199
200static bool merge_ref(struct btrfs_trans_handle *trans,
201 struct btrfs_delayed_ref_root *delayed_refs,
202 struct btrfs_delayed_ref_head *head,
203 struct btrfs_delayed_ref_node *ref,
204 u64 seq)
205{
206 struct btrfs_delayed_ref_node *next;
207 bool done = false;
208
209 next = list_first_entry(&head->ref_list, struct btrfs_delayed_ref_node,
210 list);
211 while (!done && &next->list != &head->ref_list) {
212 int mod;
213 struct btrfs_delayed_ref_node *next2;
214
215 next2 = list_next_entry(next, list);
216
217 if (next == ref)
218 goto next;
219
220 if (seq && next->seq >= seq)
221 goto next;
222
223 if (next->type != ref->type)
224 goto next;
225
226 if ((ref->type == BTRFS_TREE_BLOCK_REF_KEY ||
227 ref->type == BTRFS_SHARED_BLOCK_REF_KEY) &&
228 comp_tree_refs(btrfs_delayed_node_to_tree_ref(ref),
229 btrfs_delayed_node_to_tree_ref(next),
230 ref->type))
231 goto next;
232 if ((ref->type == BTRFS_EXTENT_DATA_REF_KEY ||
233 ref->type == BTRFS_SHARED_DATA_REF_KEY) &&
234 comp_data_refs(btrfs_delayed_node_to_data_ref(ref),
235 btrfs_delayed_node_to_data_ref(next)))
236 goto next;
237
238 if (ref->action == next->action) {
239 mod = next->ref_mod;
240 } else {
241 if (ref->ref_mod < next->ref_mod) {
242 swap(ref, next);
243 done = true;
244 }
245 mod = -next->ref_mod;
246 }
247
248 drop_delayed_ref(trans, delayed_refs, head, next);
249 ref->ref_mod += mod;
250 if (ref->ref_mod == 0) {
251 drop_delayed_ref(trans, delayed_refs, head, ref);
252 done = true;
253 } else {
254 /*
255 * Can't have multiples of the same ref on a tree block.
256 */
257 WARN_ON(ref->type == BTRFS_TREE_BLOCK_REF_KEY ||
258 ref->type == BTRFS_SHARED_BLOCK_REF_KEY);
259 }
260next:
261 next = next2;
262 }
263
264 return done;
265}
266
267void btrfs_merge_delayed_refs(struct btrfs_trans_handle *trans,
268 struct btrfs_fs_info *fs_info,
269 struct btrfs_delayed_ref_root *delayed_refs,
270 struct btrfs_delayed_ref_head *head)
271{
272 struct btrfs_delayed_ref_node *ref;
273 u64 seq = 0;
274
275 assert_spin_locked(&head->lock);
276
277 if (list_empty(&head->ref_list))
278 return;
279
280 /* We don't have too many refs to merge for data. */
281 if (head->is_data)
282 return;
283
284 spin_lock(&fs_info->tree_mod_seq_lock);
285 if (!list_empty(&fs_info->tree_mod_seq_list)) {
286 struct seq_list *elem;
287
288 elem = list_first_entry(&fs_info->tree_mod_seq_list,
289 struct seq_list, list);
290 seq = elem->seq;
291 }
292 spin_unlock(&fs_info->tree_mod_seq_lock);
293
294 ref = list_first_entry(&head->ref_list, struct btrfs_delayed_ref_node,
295 list);
296 while (&ref->list != &head->ref_list) {
297 if (seq && ref->seq >= seq)
298 goto next;
299
300 if (merge_ref(trans, delayed_refs, head, ref, seq)) {
301 if (list_empty(&head->ref_list))
302 break;
303 ref = list_first_entry(&head->ref_list,
304 struct btrfs_delayed_ref_node,
305 list);
306 continue;
307 }
308next:
309 ref = list_next_entry(ref, list);
310 }
311}
312
313int btrfs_check_delayed_seq(struct btrfs_fs_info *fs_info,
314 struct btrfs_delayed_ref_root *delayed_refs,
315 u64 seq)
316{
317 struct seq_list *elem;
318 int ret = 0;
319
320 spin_lock(&fs_info->tree_mod_seq_lock);
321 if (!list_empty(&fs_info->tree_mod_seq_list)) {
322 elem = list_first_entry(&fs_info->tree_mod_seq_list,
323 struct seq_list, list);
324 if (seq >= elem->seq) {
325 pr_debug("holding back delayed_ref %#x.%x, lowest is %#x.%x (%p)\n",
326 (u32)(seq >> 32), (u32)seq,
327 (u32)(elem->seq >> 32), (u32)elem->seq,
328 delayed_refs);
329 ret = 1;
330 }
331 }
332
333 spin_unlock(&fs_info->tree_mod_seq_lock);
334 return ret;
335}
336
337struct btrfs_delayed_ref_head *
338btrfs_select_ref_head(struct btrfs_trans_handle *trans)
339{
340 struct btrfs_delayed_ref_root *delayed_refs;
341 struct btrfs_delayed_ref_head *head;
342 u64 start;
343 bool loop = false;
344
345 delayed_refs = &trans->transaction->delayed_refs;
346
347again:
348 start = delayed_refs->run_delayed_start;
349 head = find_ref_head(&delayed_refs->href_root, start, 1);
350 if (!head && !loop) {
351 delayed_refs->run_delayed_start = 0;
352 start = 0;
353 loop = true;
354 head = find_ref_head(&delayed_refs->href_root, start, 1);
355 if (!head)
356 return NULL;
357 } else if (!head && loop) {
358 return NULL;
359 }
360
361 while (head->processing) {
362 struct rb_node *node;
363
364 node = rb_next(&head->href_node);
365 if (!node) {
366 if (loop)
367 return NULL;
368 delayed_refs->run_delayed_start = 0;
369 start = 0;
370 loop = true;
371 goto again;
372 }
373 head = rb_entry(node, struct btrfs_delayed_ref_head,
374 href_node);
375 }
376
377 head->processing = 1;
378 WARN_ON(delayed_refs->num_heads_ready == 0);
379 delayed_refs->num_heads_ready--;
380 delayed_refs->run_delayed_start = head->node.bytenr +
381 head->node.num_bytes;
382 return head;
383}
384
385/*
386 * Helper to insert the ref_node to the tail or merge with tail.
387 *
388 * Return 0 for insert.
389 * Return >0 for merge.
390 */
391static int
392add_delayed_ref_tail_merge(struct btrfs_trans_handle *trans,
393 struct btrfs_delayed_ref_root *root,
394 struct btrfs_delayed_ref_head *href,
395 struct btrfs_delayed_ref_node *ref)
396{
397 struct btrfs_delayed_ref_node *exist;
398 int mod;
399 int ret = 0;
400
401 spin_lock(&href->lock);
402 /* Check whether we can merge the tail node with ref */
403 if (list_empty(&href->ref_list))
404 goto add_tail;
405 exist = list_entry(href->ref_list.prev, struct btrfs_delayed_ref_node,
406 list);
407 /* No need to compare bytenr nor is_head */
408 if (exist->type != ref->type || exist->seq != ref->seq)
409 goto add_tail;
410
411 if ((exist->type == BTRFS_TREE_BLOCK_REF_KEY ||
412 exist->type == BTRFS_SHARED_BLOCK_REF_KEY) &&
413 comp_tree_refs(btrfs_delayed_node_to_tree_ref(exist),
414 btrfs_delayed_node_to_tree_ref(ref),
415 ref->type))
416 goto add_tail;
417 if ((exist->type == BTRFS_EXTENT_DATA_REF_KEY ||
418 exist->type == BTRFS_SHARED_DATA_REF_KEY) &&
419 comp_data_refs(btrfs_delayed_node_to_data_ref(exist),
420 btrfs_delayed_node_to_data_ref(ref)))
421 goto add_tail;
422
423 /* Now we are sure we can merge */
424 ret = 1;
425 if (exist->action == ref->action) {
426 mod = ref->ref_mod;
427 } else {
428 /* Need to change action */
429 if (exist->ref_mod < ref->ref_mod) {
430 exist->action = ref->action;
431 mod = -exist->ref_mod;
432 exist->ref_mod = ref->ref_mod;
433 } else
434 mod = -ref->ref_mod;
435 }
436 exist->ref_mod += mod;
437
438 /* remove existing tail if its ref_mod is zero */
439 if (exist->ref_mod == 0)
440 drop_delayed_ref(trans, root, href, exist);
441 spin_unlock(&href->lock);
442 return ret;
443
444add_tail:
445 list_add_tail(&ref->list, &href->ref_list);
446 atomic_inc(&root->num_entries);
447 trans->delayed_ref_updates++;
448 spin_unlock(&href->lock);
449 return ret;
450}
451
452/*
453 * helper function to update the accounting in the head ref
454 * existing and update must have the same bytenr
455 */
456static noinline void
457update_existing_head_ref(struct btrfs_delayed_ref_root *delayed_refs,
458 struct btrfs_delayed_ref_node *existing,
459 struct btrfs_delayed_ref_node *update)
460{
461 struct btrfs_delayed_ref_head *existing_ref;
462 struct btrfs_delayed_ref_head *ref;
463 int old_ref_mod;
464
465 existing_ref = btrfs_delayed_node_to_head(existing);
466 ref = btrfs_delayed_node_to_head(update);
467 BUG_ON(existing_ref->is_data != ref->is_data);
468
469 spin_lock(&existing_ref->lock);
470 if (ref->must_insert_reserved) {
471 /* if the extent was freed and then
472 * reallocated before the delayed ref
473 * entries were processed, we can end up
474 * with an existing head ref without
475 * the must_insert_reserved flag set.
476 * Set it again here
477 */
478 existing_ref->must_insert_reserved = ref->must_insert_reserved;
479
480 /*
481 * update the num_bytes so we make sure the accounting
482 * is done correctly
483 */
484 existing->num_bytes = update->num_bytes;
485
486 }
487
488 if (ref->extent_op) {
489 if (!existing_ref->extent_op) {
490 existing_ref->extent_op = ref->extent_op;
491 } else {
492 if (ref->extent_op->update_key) {
493 memcpy(&existing_ref->extent_op->key,
494 &ref->extent_op->key,
495 sizeof(ref->extent_op->key));
496 existing_ref->extent_op->update_key = true;
497 }
498 if (ref->extent_op->update_flags) {
499 existing_ref->extent_op->flags_to_set |=
500 ref->extent_op->flags_to_set;
501 existing_ref->extent_op->update_flags = true;
502 }
503 btrfs_free_delayed_extent_op(ref->extent_op);
504 }
505 }
506 /*
507 * update the reference mod on the head to reflect this new operation,
508 * only need the lock for this case cause we could be processing it
509 * currently, for refs we just added we know we're a-ok.
510 */
511 old_ref_mod = existing_ref->total_ref_mod;
512 existing->ref_mod += update->ref_mod;
513 existing_ref->total_ref_mod += update->ref_mod;
514
515 /*
516 * If we are going to from a positive ref mod to a negative or vice
517 * versa we need to make sure to adjust pending_csums accordingly.
518 */
519 if (existing_ref->is_data) {
520 if (existing_ref->total_ref_mod >= 0 && old_ref_mod < 0)
521 delayed_refs->pending_csums -= existing->num_bytes;
522 if (existing_ref->total_ref_mod < 0 && old_ref_mod >= 0)
523 delayed_refs->pending_csums += existing->num_bytes;
524 }
525 spin_unlock(&existing_ref->lock);
526}
527
528/*
529 * helper function to actually insert a head node into the rbtree.
530 * this does all the dirty work in terms of maintaining the correct
531 * overall modification count.
532 */
533static noinline struct btrfs_delayed_ref_head *
534add_delayed_ref_head(struct btrfs_fs_info *fs_info,
535 struct btrfs_trans_handle *trans,
536 struct btrfs_delayed_ref_node *ref,
537 struct btrfs_qgroup_extent_record *qrecord,
538 u64 bytenr, u64 num_bytes, u64 ref_root, u64 reserved,
539 int action, int is_data)
540{
541 struct btrfs_delayed_ref_head *existing;
542 struct btrfs_delayed_ref_head *head_ref = NULL;
543 struct btrfs_delayed_ref_root *delayed_refs;
544 struct btrfs_qgroup_extent_record *qexisting;
545 int count_mod = 1;
546 int must_insert_reserved = 0;
547
548 /* If reserved is provided, it must be a data extent. */
549 BUG_ON(!is_data && reserved);
550
551 /*
552 * the head node stores the sum of all the mods, so dropping a ref
553 * should drop the sum in the head node by one.
554 */
555 if (action == BTRFS_UPDATE_DELAYED_HEAD)
556 count_mod = 0;
557 else if (action == BTRFS_DROP_DELAYED_REF)
558 count_mod = -1;
559
560 /*
561 * BTRFS_ADD_DELAYED_EXTENT means that we need to update
562 * the reserved accounting when the extent is finally added, or
563 * if a later modification deletes the delayed ref without ever
564 * inserting the extent into the extent allocation tree.
565 * ref->must_insert_reserved is the flag used to record
566 * that accounting mods are required.
567 *
568 * Once we record must_insert_reserved, switch the action to
569 * BTRFS_ADD_DELAYED_REF because other special casing is not required.
570 */
571 if (action == BTRFS_ADD_DELAYED_EXTENT)
572 must_insert_reserved = 1;
573 else
574 must_insert_reserved = 0;
575
576 delayed_refs = &trans->transaction->delayed_refs;
577
578 /* first set the basic ref node struct up */
579 atomic_set(&ref->refs, 1);
580 ref->bytenr = bytenr;
581 ref->num_bytes = num_bytes;
582 ref->ref_mod = count_mod;
583 ref->type = 0;
584 ref->action = 0;
585 ref->is_head = 1;
586 ref->in_tree = 1;
587 ref->seq = 0;
588
589 head_ref = btrfs_delayed_node_to_head(ref);
590 head_ref->must_insert_reserved = must_insert_reserved;
591 head_ref->is_data = is_data;
592 INIT_LIST_HEAD(&head_ref->ref_list);
593 head_ref->processing = 0;
594 head_ref->total_ref_mod = count_mod;
595 head_ref->qgroup_reserved = 0;
596 head_ref->qgroup_ref_root = 0;
597
598 /* Record qgroup extent info if provided */
599 if (qrecord) {
600 if (ref_root && reserved) {
601 head_ref->qgroup_ref_root = ref_root;
602 head_ref->qgroup_reserved = reserved;
603 }
604
605 qrecord->bytenr = bytenr;
606 qrecord->num_bytes = num_bytes;
607 qrecord->old_roots = NULL;
608
609 qexisting = btrfs_qgroup_insert_dirty_extent(delayed_refs,
610 qrecord);
611 if (qexisting)
612 kfree(qrecord);
613 }
614
615 spin_lock_init(&head_ref->lock);
616 mutex_init(&head_ref->mutex);
617
618 trace_add_delayed_ref_head(ref, head_ref, action);
619
620 existing = htree_insert(&delayed_refs->href_root,
621 &head_ref->href_node);
622 if (existing) {
623 WARN_ON(ref_root && reserved && existing->qgroup_ref_root
624 && existing->qgroup_reserved);
625 update_existing_head_ref(delayed_refs, &existing->node, ref);
626 /*
627 * we've updated the existing ref, free the newly
628 * allocated ref
629 */
630 kmem_cache_free(btrfs_delayed_ref_head_cachep, head_ref);
631 head_ref = existing;
632 } else {
633 if (is_data && count_mod < 0)
634 delayed_refs->pending_csums += num_bytes;
635 delayed_refs->num_heads++;
636 delayed_refs->num_heads_ready++;
637 atomic_inc(&delayed_refs->num_entries);
638 trans->delayed_ref_updates++;
639 }
640 return head_ref;
641}
642
643/*
644 * helper to insert a delayed tree ref into the rbtree.
645 */
646static noinline void
647add_delayed_tree_ref(struct btrfs_fs_info *fs_info,
648 struct btrfs_trans_handle *trans,
649 struct btrfs_delayed_ref_head *head_ref,
650 struct btrfs_delayed_ref_node *ref, u64 bytenr,
651 u64 num_bytes, u64 parent, u64 ref_root, int level,
652 int action)
653{
654 struct btrfs_delayed_tree_ref *full_ref;
655 struct btrfs_delayed_ref_root *delayed_refs;
656 u64 seq = 0;
657 int ret;
658
659 if (action == BTRFS_ADD_DELAYED_EXTENT)
660 action = BTRFS_ADD_DELAYED_REF;
661
662 if (is_fstree(ref_root))
663 seq = atomic64_read(&fs_info->tree_mod_seq);
664 delayed_refs = &trans->transaction->delayed_refs;
665
666 /* first set the basic ref node struct up */
667 atomic_set(&ref->refs, 1);
668 ref->bytenr = bytenr;
669 ref->num_bytes = num_bytes;
670 ref->ref_mod = 1;
671 ref->action = action;
672 ref->is_head = 0;
673 ref->in_tree = 1;
674 ref->seq = seq;
675
676 full_ref = btrfs_delayed_node_to_tree_ref(ref);
677 full_ref->parent = parent;
678 full_ref->root = ref_root;
679 if (parent)
680 ref->type = BTRFS_SHARED_BLOCK_REF_KEY;
681 else
682 ref->type = BTRFS_TREE_BLOCK_REF_KEY;
683 full_ref->level = level;
684
685 trace_add_delayed_tree_ref(ref, full_ref, action);
686
687 ret = add_delayed_ref_tail_merge(trans, delayed_refs, head_ref, ref);
688
689 /*
690 * XXX: memory should be freed at the same level allocated.
691 * But bad practice is anywhere... Follow it now. Need cleanup.
692 */
693 if (ret > 0)
694 kmem_cache_free(btrfs_delayed_tree_ref_cachep, full_ref);
695}
696
697/*
698 * helper to insert a delayed data ref into the rbtree.
699 */
700static noinline void
701add_delayed_data_ref(struct btrfs_fs_info *fs_info,
702 struct btrfs_trans_handle *trans,
703 struct btrfs_delayed_ref_head *head_ref,
704 struct btrfs_delayed_ref_node *ref, u64 bytenr,
705 u64 num_bytes, u64 parent, u64 ref_root, u64 owner,
706 u64 offset, int action)
707{
708 struct btrfs_delayed_data_ref *full_ref;
709 struct btrfs_delayed_ref_root *delayed_refs;
710 u64 seq = 0;
711 int ret;
712
713 if (action == BTRFS_ADD_DELAYED_EXTENT)
714 action = BTRFS_ADD_DELAYED_REF;
715
716 delayed_refs = &trans->transaction->delayed_refs;
717
718 if (is_fstree(ref_root))
719 seq = atomic64_read(&fs_info->tree_mod_seq);
720
721 /* first set the basic ref node struct up */
722 atomic_set(&ref->refs, 1);
723 ref->bytenr = bytenr;
724 ref->num_bytes = num_bytes;
725 ref->ref_mod = 1;
726 ref->action = action;
727 ref->is_head = 0;
728 ref->in_tree = 1;
729 ref->seq = seq;
730
731 full_ref = btrfs_delayed_node_to_data_ref(ref);
732 full_ref->parent = parent;
733 full_ref->root = ref_root;
734 if (parent)
735 ref->type = BTRFS_SHARED_DATA_REF_KEY;
736 else
737 ref->type = BTRFS_EXTENT_DATA_REF_KEY;
738
739 full_ref->objectid = owner;
740 full_ref->offset = offset;
741
742 trace_add_delayed_data_ref(ref, full_ref, action);
743
744 ret = add_delayed_ref_tail_merge(trans, delayed_refs, head_ref, ref);
745
746 if (ret > 0)
747 kmem_cache_free(btrfs_delayed_data_ref_cachep, full_ref);
748}
749
750/*
751 * add a delayed tree ref. This does all of the accounting required
752 * to make sure the delayed ref is eventually processed before this
753 * transaction commits.
754 */
755int btrfs_add_delayed_tree_ref(struct btrfs_fs_info *fs_info,
756 struct btrfs_trans_handle *trans,
757 u64 bytenr, u64 num_bytes, u64 parent,
758 u64 ref_root, int level, int action,
759 struct btrfs_delayed_extent_op *extent_op)
760{
761 struct btrfs_delayed_tree_ref *ref;
762 struct btrfs_delayed_ref_head *head_ref;
763 struct btrfs_delayed_ref_root *delayed_refs;
764 struct btrfs_qgroup_extent_record *record = NULL;
765
766 BUG_ON(extent_op && extent_op->is_data);
767 ref = kmem_cache_alloc(btrfs_delayed_tree_ref_cachep, GFP_NOFS);
768 if (!ref)
769 return -ENOMEM;
770
771 head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS);
772 if (!head_ref)
773 goto free_ref;
774
775 if (fs_info->quota_enabled && is_fstree(ref_root)) {
776 record = kmalloc(sizeof(*record), GFP_NOFS);
777 if (!record)
778 goto free_head_ref;
779 }
780
781 head_ref->extent_op = extent_op;
782
783 delayed_refs = &trans->transaction->delayed_refs;
784 spin_lock(&delayed_refs->lock);
785
786 /*
787 * insert both the head node and the new ref without dropping
788 * the spin lock
789 */
790 head_ref = add_delayed_ref_head(fs_info, trans, &head_ref->node, record,
791 bytenr, num_bytes, 0, 0, action, 0);
792
793 add_delayed_tree_ref(fs_info, trans, head_ref, &ref->node, bytenr,
794 num_bytes, parent, ref_root, level, action);
795 spin_unlock(&delayed_refs->lock);
796
797 return 0;
798
799free_head_ref:
800 kmem_cache_free(btrfs_delayed_ref_head_cachep, head_ref);
801free_ref:
802 kmem_cache_free(btrfs_delayed_tree_ref_cachep, ref);
803
804 return -ENOMEM;
805}
806
807/*
808 * add a delayed data ref. it's similar to btrfs_add_delayed_tree_ref.
809 */
810int btrfs_add_delayed_data_ref(struct btrfs_fs_info *fs_info,
811 struct btrfs_trans_handle *trans,
812 u64 bytenr, u64 num_bytes,
813 u64 parent, u64 ref_root,
814 u64 owner, u64 offset, u64 reserved, int action,
815 struct btrfs_delayed_extent_op *extent_op)
816{
817 struct btrfs_delayed_data_ref *ref;
818 struct btrfs_delayed_ref_head *head_ref;
819 struct btrfs_delayed_ref_root *delayed_refs;
820 struct btrfs_qgroup_extent_record *record = NULL;
821
822 BUG_ON(extent_op && !extent_op->is_data);
823 ref = kmem_cache_alloc(btrfs_delayed_data_ref_cachep, GFP_NOFS);
824 if (!ref)
825 return -ENOMEM;
826
827 head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS);
828 if (!head_ref) {
829 kmem_cache_free(btrfs_delayed_data_ref_cachep, ref);
830 return -ENOMEM;
831 }
832
833 if (fs_info->quota_enabled && is_fstree(ref_root)) {
834 record = kmalloc(sizeof(*record), GFP_NOFS);
835 if (!record) {
836 kmem_cache_free(btrfs_delayed_data_ref_cachep, ref);
837 kmem_cache_free(btrfs_delayed_ref_head_cachep,
838 head_ref);
839 return -ENOMEM;
840 }
841 }
842
843 head_ref->extent_op = extent_op;
844
845 delayed_refs = &trans->transaction->delayed_refs;
846 spin_lock(&delayed_refs->lock);
847
848 /*
849 * insert both the head node and the new ref without dropping
850 * the spin lock
851 */
852 head_ref = add_delayed_ref_head(fs_info, trans, &head_ref->node, record,
853 bytenr, num_bytes, ref_root, reserved,
854 action, 1);
855
856 add_delayed_data_ref(fs_info, trans, head_ref, &ref->node, bytenr,
857 num_bytes, parent, ref_root, owner, offset,
858 action);
859 spin_unlock(&delayed_refs->lock);
860
861 return 0;
862}
863
864int btrfs_add_delayed_qgroup_reserve(struct btrfs_fs_info *fs_info,
865 struct btrfs_trans_handle *trans,
866 u64 ref_root, u64 bytenr, u64 num_bytes)
867{
868 struct btrfs_delayed_ref_root *delayed_refs;
869 struct btrfs_delayed_ref_head *ref_head;
870 int ret = 0;
871
872 if (!fs_info->quota_enabled || !is_fstree(ref_root))
873 return 0;
874
875 delayed_refs = &trans->transaction->delayed_refs;
876
877 spin_lock(&delayed_refs->lock);
878 ref_head = find_ref_head(&delayed_refs->href_root, bytenr, 0);
879 if (!ref_head) {
880 ret = -ENOENT;
881 goto out;
882 }
883 WARN_ON(ref_head->qgroup_reserved || ref_head->qgroup_ref_root);
884 ref_head->qgroup_ref_root = ref_root;
885 ref_head->qgroup_reserved = num_bytes;
886out:
887 spin_unlock(&delayed_refs->lock);
888 return ret;
889}
890
891int btrfs_add_delayed_extent_op(struct btrfs_fs_info *fs_info,
892 struct btrfs_trans_handle *trans,
893 u64 bytenr, u64 num_bytes,
894 struct btrfs_delayed_extent_op *extent_op)
895{
896 struct btrfs_delayed_ref_head *head_ref;
897 struct btrfs_delayed_ref_root *delayed_refs;
898
899 head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS);
900 if (!head_ref)
901 return -ENOMEM;
902
903 head_ref->extent_op = extent_op;
904
905 delayed_refs = &trans->transaction->delayed_refs;
906 spin_lock(&delayed_refs->lock);
907
908 add_delayed_ref_head(fs_info, trans, &head_ref->node, NULL, bytenr,
909 num_bytes, 0, 0, BTRFS_UPDATE_DELAYED_HEAD,
910 extent_op->is_data);
911
912 spin_unlock(&delayed_refs->lock);
913 return 0;
914}
915
916/*
917 * this does a simple search for the head node for a given extent.
918 * It must be called with the delayed ref spinlock held, and it returns
919 * the head node if any where found, or NULL if not.
920 */
921struct btrfs_delayed_ref_head *
922btrfs_find_delayed_ref_head(struct btrfs_trans_handle *trans, u64 bytenr)
923{
924 struct btrfs_delayed_ref_root *delayed_refs;
925
926 delayed_refs = &trans->transaction->delayed_refs;
927 return find_ref_head(&delayed_refs->href_root, bytenr, 0);
928}
929
930void btrfs_delayed_ref_exit(void)
931{
932 kmem_cache_destroy(btrfs_delayed_ref_head_cachep);
933 kmem_cache_destroy(btrfs_delayed_tree_ref_cachep);
934 kmem_cache_destroy(btrfs_delayed_data_ref_cachep);
935 kmem_cache_destroy(btrfs_delayed_extent_op_cachep);
936}
937
938int btrfs_delayed_ref_init(void)
939{
940 btrfs_delayed_ref_head_cachep = kmem_cache_create(
941 "btrfs_delayed_ref_head",
942 sizeof(struct btrfs_delayed_ref_head), 0,
943 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
944 if (!btrfs_delayed_ref_head_cachep)
945 goto fail;
946
947 btrfs_delayed_tree_ref_cachep = kmem_cache_create(
948 "btrfs_delayed_tree_ref",
949 sizeof(struct btrfs_delayed_tree_ref), 0,
950 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
951 if (!btrfs_delayed_tree_ref_cachep)
952 goto fail;
953
954 btrfs_delayed_data_ref_cachep = kmem_cache_create(
955 "btrfs_delayed_data_ref",
956 sizeof(struct btrfs_delayed_data_ref), 0,
957 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
958 if (!btrfs_delayed_data_ref_cachep)
959 goto fail;
960
961 btrfs_delayed_extent_op_cachep = kmem_cache_create(
962 "btrfs_delayed_extent_op",
963 sizeof(struct btrfs_delayed_extent_op), 0,
964 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
965 if (!btrfs_delayed_extent_op_cachep)
966 goto fail;
967
968 return 0;
969fail:
970 btrfs_delayed_ref_exit();
971 return -ENOMEM;
972}