xref: /openbmc/linux/fs/btrfs/delayed-ref.c (revision 8781e5df)
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 
15 struct kmem_cache *btrfs_delayed_ref_head_cachep;
16 struct kmem_cache *btrfs_delayed_tree_ref_cachep;
17 struct kmem_cache *btrfs_delayed_data_ref_cachep;
18 struct 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 
28 bool 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 
53 int 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  */
79 void 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  */
99 void 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  */
126 void 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  */
173 int 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  */
204 static 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  */
224 static 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 
249 static 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 */
278 static 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 
310 static 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 
341 static 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  */
361 static 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 
394 int 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 
415 static 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 
430 static 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 
477 void 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 
505 again:
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 
516 int 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 
538 struct 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 
543 again:
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 
575 void 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  */
595 static 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;
638 inserted:
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  */
650 static 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 
731 static 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  */
798 static noinline struct btrfs_delayed_ref_head *
799 add_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  */
880 static 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  */
911 int 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  */
1008 int 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 
1101 int 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  */
1137 struct btrfs_delayed_ref_head *
1138 btrfs_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 
1145 void __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 
1153 int __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;
1184 fail:
1185 	btrfs_delayed_ref_exit();
1186 	return -ENOMEM;
1187 }
1188