Btrfs: fix unblocked autodefraggers when remount
[deliverable/linux.git] / fs / btrfs / delayed-ref.c
CommitLineData
56bec294
CM
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>
5a0e3ad6 20#include <linux/slab.h>
56bec294 21#include <linux/sort.h>
56bec294
CM
22#include "ctree.h"
23#include "delayed-ref.h"
24#include "transaction.h"
25
78a6184a
MX
26struct kmem_cache *btrfs_delayed_ref_head_cachep;
27struct kmem_cache *btrfs_delayed_tree_ref_cachep;
28struct kmem_cache *btrfs_delayed_data_ref_cachep;
29struct kmem_cache *btrfs_delayed_extent_op_cachep;
56bec294
CM
30/*
31 * delayed back reference update tracking. For subvolume trees
32 * we queue up extent allocations and backref maintenance for
33 * delayed processing. This avoids deep call chains where we
34 * add extents in the middle of btrfs_search_slot, and it allows
35 * us to buffer up frequently modified backrefs in an rb tree instead
36 * of hammering updates on the extent allocation tree.
56bec294
CM
37 */
38
39/*
5d4f98a2
YZ
40 * compare two delayed tree backrefs with same bytenr and type
41 */
42static int comp_tree_refs(struct btrfs_delayed_tree_ref *ref2,
41b0fc42 43 struct btrfs_delayed_tree_ref *ref1, int type)
5d4f98a2 44{
41b0fc42
JB
45 if (type == BTRFS_TREE_BLOCK_REF_KEY) {
46 if (ref1->root < ref2->root)
47 return -1;
48 if (ref1->root > ref2->root)
49 return 1;
50 } else {
51 if (ref1->parent < ref2->parent)
52 return -1;
53 if (ref1->parent > ref2->parent)
54 return 1;
55 }
5d4f98a2
YZ
56 return 0;
57}
58
59/*
60 * compare two delayed data backrefs with same bytenr and type
56bec294 61 */
5d4f98a2
YZ
62static int comp_data_refs(struct btrfs_delayed_data_ref *ref2,
63 struct btrfs_delayed_data_ref *ref1)
56bec294 64{
5d4f98a2
YZ
65 if (ref1->node.type == BTRFS_EXTENT_DATA_REF_KEY) {
66 if (ref1->root < ref2->root)
67 return -1;
68 if (ref1->root > ref2->root)
69 return 1;
70 if (ref1->objectid < ref2->objectid)
71 return -1;
72 if (ref1->objectid > ref2->objectid)
73 return 1;
74 if (ref1->offset < ref2->offset)
75 return -1;
76 if (ref1->offset > ref2->offset)
77 return 1;
78 } else {
79 if (ref1->parent < ref2->parent)
80 return -1;
81 if (ref1->parent > ref2->parent)
82 return 1;
83 }
84 return 0;
85}
86
87/*
88 * entries in the rb tree are ordered by the byte number of the extent,
89 * type of the delayed backrefs and content of delayed backrefs.
90 */
91static int comp_entry(struct btrfs_delayed_ref_node *ref2,
ae1e206b
JB
92 struct btrfs_delayed_ref_node *ref1,
93 bool compare_seq)
5d4f98a2
YZ
94{
95 if (ref1->bytenr < ref2->bytenr)
56bec294 96 return -1;
5d4f98a2 97 if (ref1->bytenr > ref2->bytenr)
56bec294 98 return 1;
5d4f98a2
YZ
99 if (ref1->is_head && ref2->is_head)
100 return 0;
101 if (ref2->is_head)
56bec294 102 return -1;
5d4f98a2 103 if (ref1->is_head)
56bec294 104 return 1;
5d4f98a2
YZ
105 if (ref1->type < ref2->type)
106 return -1;
107 if (ref1->type > ref2->type)
108 return 1;
00f04b88 109 /* merging of sequenced refs is not allowed */
ae1e206b
JB
110 if (compare_seq) {
111 if (ref1->seq < ref2->seq)
112 return -1;
113 if (ref1->seq > ref2->seq)
114 return 1;
115 }
5d4f98a2
YZ
116 if (ref1->type == BTRFS_TREE_BLOCK_REF_KEY ||
117 ref1->type == BTRFS_SHARED_BLOCK_REF_KEY) {
118 return comp_tree_refs(btrfs_delayed_node_to_tree_ref(ref2),
41b0fc42
JB
119 btrfs_delayed_node_to_tree_ref(ref1),
120 ref1->type);
5d4f98a2
YZ
121 } else if (ref1->type == BTRFS_EXTENT_DATA_REF_KEY ||
122 ref1->type == BTRFS_SHARED_DATA_REF_KEY) {
123 return comp_data_refs(btrfs_delayed_node_to_data_ref(ref2),
124 btrfs_delayed_node_to_data_ref(ref1));
125 }
126 BUG();
56bec294
CM
127 return 0;
128}
129
130/*
131 * insert a new ref into the rbtree. This returns any existing refs
132 * for the same (bytenr,parent) tuple, or NULL if the new node was properly
133 * inserted.
134 */
135static struct btrfs_delayed_ref_node *tree_insert(struct rb_root *root,
56bec294
CM
136 struct rb_node *node)
137{
138 struct rb_node **p = &root->rb_node;
139 struct rb_node *parent_node = NULL;
140 struct btrfs_delayed_ref_node *entry;
5d4f98a2 141 struct btrfs_delayed_ref_node *ins;
56bec294
CM
142 int cmp;
143
5d4f98a2 144 ins = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
56bec294
CM
145 while (*p) {
146 parent_node = *p;
147 entry = rb_entry(parent_node, struct btrfs_delayed_ref_node,
148 rb_node);
149
ae1e206b 150 cmp = comp_entry(entry, ins, 1);
56bec294
CM
151 if (cmp < 0)
152 p = &(*p)->rb_left;
153 else if (cmp > 0)
154 p = &(*p)->rb_right;
155 else
156 return entry;
157 }
158
56bec294
CM
159 rb_link_node(node, parent_node, p);
160 rb_insert_color(node, root);
161 return NULL;
162}
163
164/*
5d4f98a2 165 * find an head entry based on bytenr. This returns the delayed ref
d1270cd9
AJ
166 * head if it was able to find one, or NULL if nothing was in that spot.
167 * If return_bigger is given, the next bigger entry is returned if no exact
168 * match is found.
56bec294 169 */
5d4f98a2
YZ
170static struct btrfs_delayed_ref_node *find_ref_head(struct rb_root *root,
171 u64 bytenr,
d1270cd9
AJ
172 struct btrfs_delayed_ref_node **last,
173 int return_bigger)
56bec294 174{
d1270cd9 175 struct rb_node *n;
56bec294 176 struct btrfs_delayed_ref_node *entry;
d1270cd9 177 int cmp = 0;
56bec294 178
d1270cd9
AJ
179again:
180 n = root->rb_node;
181 entry = NULL;
56bec294
CM
182 while (n) {
183 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
184 WARN_ON(!entry->in_tree);
c3e69d58
CM
185 if (last)
186 *last = entry;
56bec294 187
5d4f98a2
YZ
188 if (bytenr < entry->bytenr)
189 cmp = -1;
190 else if (bytenr > entry->bytenr)
191 cmp = 1;
192 else if (!btrfs_delayed_ref_is_head(entry))
193 cmp = 1;
194 else
195 cmp = 0;
196
56bec294
CM
197 if (cmp < 0)
198 n = n->rb_left;
199 else if (cmp > 0)
200 n = n->rb_right;
201 else
202 return entry;
203 }
d1270cd9
AJ
204 if (entry && return_bigger) {
205 if (cmp > 0) {
206 n = rb_next(&entry->rb_node);
207 if (!n)
208 n = rb_first(root);
209 entry = rb_entry(n, struct btrfs_delayed_ref_node,
210 rb_node);
211 bytenr = entry->bytenr;
212 return_bigger = 0;
213 goto again;
214 }
215 return entry;
216 }
56bec294
CM
217 return NULL;
218}
219
c3e69d58
CM
220int btrfs_delayed_ref_lock(struct btrfs_trans_handle *trans,
221 struct btrfs_delayed_ref_head *head)
56bec294 222{
c3e69d58
CM
223 struct btrfs_delayed_ref_root *delayed_refs;
224
225 delayed_refs = &trans->transaction->delayed_refs;
226 assert_spin_locked(&delayed_refs->lock);
227 if (mutex_trylock(&head->mutex))
228 return 0;
229
230 atomic_inc(&head->node.refs);
231 spin_unlock(&delayed_refs->lock);
232
233 mutex_lock(&head->mutex);
234 spin_lock(&delayed_refs->lock);
235 if (!head->node.in_tree) {
236 mutex_unlock(&head->mutex);
237 btrfs_put_delayed_ref(&head->node);
238 return -EAGAIN;
239 }
240 btrfs_put_delayed_ref(&head->node);
241 return 0;
242}
243
ae1e206b
JB
244static void inline drop_delayed_ref(struct btrfs_trans_handle *trans,
245 struct btrfs_delayed_ref_root *delayed_refs,
246 struct btrfs_delayed_ref_node *ref)
247{
248 rb_erase(&ref->rb_node, &delayed_refs->root);
249 ref->in_tree = 0;
250 btrfs_put_delayed_ref(ref);
251 delayed_refs->num_entries--;
252 if (trans->delayed_ref_updates)
253 trans->delayed_ref_updates--;
254}
255
256static int merge_ref(struct btrfs_trans_handle *trans,
257 struct btrfs_delayed_ref_root *delayed_refs,
258 struct btrfs_delayed_ref_node *ref, u64 seq)
259{
260 struct rb_node *node;
261 int merged = 0;
262 int mod = 0;
263 int done = 0;
264
265 node = rb_prev(&ref->rb_node);
266 while (node) {
267 struct btrfs_delayed_ref_node *next;
268
269 next = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
270 node = rb_prev(node);
271 if (next->bytenr != ref->bytenr)
272 break;
273 if (seq && next->seq >= seq)
274 break;
275 if (comp_entry(ref, next, 0))
276 continue;
277
278 if (ref->action == next->action) {
279 mod = next->ref_mod;
280 } else {
281 if (ref->ref_mod < next->ref_mod) {
282 struct btrfs_delayed_ref_node *tmp;
283
284 tmp = ref;
285 ref = next;
286 next = tmp;
287 done = 1;
288 }
289 mod = -next->ref_mod;
290 }
291
292 merged++;
293 drop_delayed_ref(trans, delayed_refs, next);
294 ref->ref_mod += mod;
295 if (ref->ref_mod == 0) {
296 drop_delayed_ref(trans, delayed_refs, ref);
297 break;
298 } else {
299 /*
300 * You can't have multiples of the same ref on a tree
301 * block.
302 */
303 WARN_ON(ref->type == BTRFS_TREE_BLOCK_REF_KEY ||
304 ref->type == BTRFS_SHARED_BLOCK_REF_KEY);
305 }
306
307 if (done)
308 break;
309 node = rb_prev(&ref->rb_node);
310 }
311
312 return merged;
313}
314
315void btrfs_merge_delayed_refs(struct btrfs_trans_handle *trans,
316 struct btrfs_fs_info *fs_info,
317 struct btrfs_delayed_ref_root *delayed_refs,
318 struct btrfs_delayed_ref_head *head)
319{
320 struct rb_node *node;
321 u64 seq = 0;
322
323 spin_lock(&fs_info->tree_mod_seq_lock);
324 if (!list_empty(&fs_info->tree_mod_seq_list)) {
325 struct seq_list *elem;
326
327 elem = list_first_entry(&fs_info->tree_mod_seq_list,
328 struct seq_list, list);
329 seq = elem->seq;
330 }
331 spin_unlock(&fs_info->tree_mod_seq_lock);
332
333 node = rb_prev(&head->node.rb_node);
334 while (node) {
335 struct btrfs_delayed_ref_node *ref;
336
337 ref = rb_entry(node, struct btrfs_delayed_ref_node,
338 rb_node);
339 if (ref->bytenr != head->node.bytenr)
340 break;
341
342 /* We can't merge refs that are outside of our seq count */
343 if (seq && ref->seq >= seq)
344 break;
345 if (merge_ref(trans, delayed_refs, ref, seq))
346 node = rb_prev(&head->node.rb_node);
347 else
348 node = rb_prev(node);
349 }
350}
351
097b8a7c
JS
352int btrfs_check_delayed_seq(struct btrfs_fs_info *fs_info,
353 struct btrfs_delayed_ref_root *delayed_refs,
00f04b88
AJ
354 u64 seq)
355{
356 struct seq_list *elem;
097b8a7c
JS
357 int ret = 0;
358
359 spin_lock(&fs_info->tree_mod_seq_lock);
360 if (!list_empty(&fs_info->tree_mod_seq_list)) {
361 elem = list_first_entry(&fs_info->tree_mod_seq_list,
362 struct seq_list, list);
363 if (seq >= elem->seq) {
364 pr_debug("holding back delayed_ref %llu, lowest is "
365 "%llu (%p)\n", seq, elem->seq, delayed_refs);
366 ret = 1;
367 }
00f04b88 368 }
097b8a7c
JS
369
370 spin_unlock(&fs_info->tree_mod_seq_lock);
371 return ret;
00f04b88
AJ
372}
373
c3e69d58
CM
374int btrfs_find_ref_cluster(struct btrfs_trans_handle *trans,
375 struct list_head *cluster, u64 start)
376{
377 int count = 0;
378 struct btrfs_delayed_ref_root *delayed_refs;
56bec294 379 struct rb_node *node;
c3e69d58 380 struct btrfs_delayed_ref_node *ref;
56bec294 381 struct btrfs_delayed_ref_head *head;
56bec294 382
c3e69d58
CM
383 delayed_refs = &trans->transaction->delayed_refs;
384 if (start == 0) {
385 node = rb_first(&delayed_refs->root);
386 } else {
387 ref = NULL;
d1270cd9 388 find_ref_head(&delayed_refs->root, start + 1, &ref, 1);
c3e69d58 389 if (ref) {
c3e69d58
CM
390 node = &ref->rb_node;
391 } else
392 node = rb_first(&delayed_refs->root);
393 }
394again:
395 while (node && count < 32) {
396 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
56bec294
CM
397 if (btrfs_delayed_ref_is_head(ref)) {
398 head = btrfs_delayed_node_to_head(ref);
c3e69d58
CM
399 if (list_empty(&head->cluster)) {
400 list_add_tail(&head->cluster, cluster);
401 delayed_refs->run_delayed_start =
402 head->node.bytenr;
403 count++;
404
405 WARN_ON(delayed_refs->num_heads_ready == 0);
406 delayed_refs->num_heads_ready--;
407 } else if (count) {
408 /* the goal of the clustering is to find extents
409 * that are likely to end up in the same extent
410 * leaf on disk. So, we don't want them spread
411 * all over the tree. Stop now if we've hit
412 * a head that was already in use
413 */
56bec294
CM
414 break;
415 }
416 }
c3e69d58 417 node = rb_next(node);
56bec294 418 }
c3e69d58
CM
419 if (count) {
420 return 0;
421 } else if (start) {
422 /*
423 * we've gone to the end of the rbtree without finding any
424 * clusters. start from the beginning and try again
425 */
426 start = 0;
427 node = rb_first(&delayed_refs->root);
428 goto again;
429 }
430 return 1;
56bec294
CM
431}
432
093486c4
MX
433void btrfs_release_ref_cluster(struct list_head *cluster)
434{
435 struct list_head *pos, *q;
436
437 list_for_each_safe(pos, q, cluster)
438 list_del_init(pos);
439}
440
56bec294
CM
441/*
442 * helper function to update an extent delayed ref in the
443 * rbtree. existing and update must both have the same
444 * bytenr and parent
445 *
446 * This may free existing if the update cancels out whatever
447 * operation it was doing.
448 */
449static noinline void
450update_existing_ref(struct btrfs_trans_handle *trans,
451 struct btrfs_delayed_ref_root *delayed_refs,
452 struct btrfs_delayed_ref_node *existing,
453 struct btrfs_delayed_ref_node *update)
454{
5d4f98a2 455 if (update->action != existing->action) {
56bec294
CM
456 /*
457 * this is effectively undoing either an add or a
458 * drop. We decrement the ref_mod, and if it goes
459 * down to zero we just delete the entry without
460 * every changing the extent allocation tree.
461 */
462 existing->ref_mod--;
ae1e206b
JB
463 if (existing->ref_mod == 0)
464 drop_delayed_ref(trans, delayed_refs, existing);
465 else
5d4f98a2
YZ
466 WARN_ON(existing->type == BTRFS_TREE_BLOCK_REF_KEY ||
467 existing->type == BTRFS_SHARED_BLOCK_REF_KEY);
56bec294 468 } else {
5d4f98a2
YZ
469 WARN_ON(existing->type == BTRFS_TREE_BLOCK_REF_KEY ||
470 existing->type == BTRFS_SHARED_BLOCK_REF_KEY);
56bec294
CM
471 /*
472 * the action on the existing ref matches
473 * the action on the ref we're trying to add.
474 * Bump the ref_mod by one so the backref that
475 * is eventually added/removed has the correct
476 * reference count
477 */
478 existing->ref_mod += update->ref_mod;
479 }
480}
481
482/*
483 * helper function to update the accounting in the head ref
484 * existing and update must have the same bytenr
485 */
486static noinline void
487update_existing_head_ref(struct btrfs_delayed_ref_node *existing,
488 struct btrfs_delayed_ref_node *update)
489{
490 struct btrfs_delayed_ref_head *existing_ref;
491 struct btrfs_delayed_ref_head *ref;
492
493 existing_ref = btrfs_delayed_node_to_head(existing);
494 ref = btrfs_delayed_node_to_head(update);
5d4f98a2 495 BUG_ON(existing_ref->is_data != ref->is_data);
56bec294
CM
496
497 if (ref->must_insert_reserved) {
498 /* if the extent was freed and then
499 * reallocated before the delayed ref
500 * entries were processed, we can end up
501 * with an existing head ref without
502 * the must_insert_reserved flag set.
503 * Set it again here
504 */
505 existing_ref->must_insert_reserved = ref->must_insert_reserved;
506
507 /*
508 * update the num_bytes so we make sure the accounting
509 * is done correctly
510 */
511 existing->num_bytes = update->num_bytes;
512
513 }
514
5d4f98a2
YZ
515 if (ref->extent_op) {
516 if (!existing_ref->extent_op) {
517 existing_ref->extent_op = ref->extent_op;
518 } else {
519 if (ref->extent_op->update_key) {
520 memcpy(&existing_ref->extent_op->key,
521 &ref->extent_op->key,
522 sizeof(ref->extent_op->key));
523 existing_ref->extent_op->update_key = 1;
524 }
525 if (ref->extent_op->update_flags) {
526 existing_ref->extent_op->flags_to_set |=
527 ref->extent_op->flags_to_set;
528 existing_ref->extent_op->update_flags = 1;
529 }
78a6184a 530 btrfs_free_delayed_extent_op(ref->extent_op);
5d4f98a2
YZ
531 }
532 }
56bec294
CM
533 /*
534 * update the reference mod on the head to reflect this new operation
535 */
536 existing->ref_mod += update->ref_mod;
537}
538
539/*
5d4f98a2 540 * helper function to actually insert a head node into the rbtree.
56bec294 541 * this does all the dirty work in terms of maintaining the correct
5d4f98a2 542 * overall modification count.
56bec294 543 */
143bede5 544static noinline void add_delayed_ref_head(struct btrfs_fs_info *fs_info,
66d7e7f0 545 struct btrfs_trans_handle *trans,
5d4f98a2
YZ
546 struct btrfs_delayed_ref_node *ref,
547 u64 bytenr, u64 num_bytes,
548 int action, int is_data)
56bec294
CM
549{
550 struct btrfs_delayed_ref_node *existing;
c3e69d58 551 struct btrfs_delayed_ref_head *head_ref = NULL;
56bec294
CM
552 struct btrfs_delayed_ref_root *delayed_refs;
553 int count_mod = 1;
554 int must_insert_reserved = 0;
555
556 /*
557 * the head node stores the sum of all the mods, so dropping a ref
558 * should drop the sum in the head node by one.
559 */
5d4f98a2
YZ
560 if (action == BTRFS_UPDATE_DELAYED_HEAD)
561 count_mod = 0;
562 else if (action == BTRFS_DROP_DELAYED_REF)
563 count_mod = -1;
56bec294
CM
564
565 /*
566 * BTRFS_ADD_DELAYED_EXTENT means that we need to update
567 * the reserved accounting when the extent is finally added, or
568 * if a later modification deletes the delayed ref without ever
569 * inserting the extent into the extent allocation tree.
570 * ref->must_insert_reserved is the flag used to record
571 * that accounting mods are required.
572 *
573 * Once we record must_insert_reserved, switch the action to
574 * BTRFS_ADD_DELAYED_REF because other special casing is not required.
575 */
5d4f98a2 576 if (action == BTRFS_ADD_DELAYED_EXTENT)
56bec294 577 must_insert_reserved = 1;
5d4f98a2 578 else
56bec294 579 must_insert_reserved = 0;
56bec294
CM
580
581 delayed_refs = &trans->transaction->delayed_refs;
582
583 /* first set the basic ref node struct up */
584 atomic_set(&ref->refs, 1);
585 ref->bytenr = bytenr;
5d4f98a2 586 ref->num_bytes = num_bytes;
56bec294 587 ref->ref_mod = count_mod;
5d4f98a2
YZ
588 ref->type = 0;
589 ref->action = 0;
590 ref->is_head = 1;
56bec294 591 ref->in_tree = 1;
00f04b88 592 ref->seq = 0;
5d4f98a2
YZ
593
594 head_ref = btrfs_delayed_node_to_head(ref);
595 head_ref->must_insert_reserved = must_insert_reserved;
596 head_ref->is_data = is_data;
597
598 INIT_LIST_HEAD(&head_ref->cluster);
599 mutex_init(&head_ref->mutex);
600
1abe9b8a 601 trace_btrfs_delayed_ref_head(ref, head_ref, action);
602
5d4f98a2
YZ
603 existing = tree_insert(&delayed_refs->root, &ref->rb_node);
604
605 if (existing) {
606 update_existing_head_ref(existing, ref);
607 /*
608 * we've updated the existing ref, free the newly
609 * allocated ref
610 */
78a6184a 611 kmem_cache_free(btrfs_delayed_ref_head_cachep, head_ref);
5d4f98a2
YZ
612 } else {
613 delayed_refs->num_heads++;
614 delayed_refs->num_heads_ready++;
615 delayed_refs->num_entries++;
616 trans->delayed_ref_updates++;
617 }
5d4f98a2
YZ
618}
619
620/*
621 * helper to insert a delayed tree ref into the rbtree.
622 */
143bede5 623static noinline void add_delayed_tree_ref(struct btrfs_fs_info *fs_info,
66d7e7f0 624 struct btrfs_trans_handle *trans,
5d4f98a2
YZ
625 struct btrfs_delayed_ref_node *ref,
626 u64 bytenr, u64 num_bytes, u64 parent,
66d7e7f0
AJ
627 u64 ref_root, int level, int action,
628 int for_cow)
5d4f98a2
YZ
629{
630 struct btrfs_delayed_ref_node *existing;
631 struct btrfs_delayed_tree_ref *full_ref;
632 struct btrfs_delayed_ref_root *delayed_refs;
00f04b88 633 u64 seq = 0;
5d4f98a2
YZ
634
635 if (action == BTRFS_ADD_DELAYED_EXTENT)
636 action = BTRFS_ADD_DELAYED_REF;
637
638 delayed_refs = &trans->transaction->delayed_refs;
639
640 /* first set the basic ref node struct up */
641 atomic_set(&ref->refs, 1);
642 ref->bytenr = bytenr;
56bec294 643 ref->num_bytes = num_bytes;
5d4f98a2
YZ
644 ref->ref_mod = 1;
645 ref->action = action;
646 ref->is_head = 0;
647 ref->in_tree = 1;
56bec294 648
546adb0d
JS
649 if (need_ref_seq(for_cow, ref_root))
650 seq = btrfs_get_tree_mod_seq(fs_info, &trans->delayed_ref_elem);
00f04b88
AJ
651 ref->seq = seq;
652
5d4f98a2 653 full_ref = btrfs_delayed_node_to_tree_ref(ref);
eebe063b
AJ
654 full_ref->parent = parent;
655 full_ref->root = ref_root;
656 if (parent)
5d4f98a2 657 ref->type = BTRFS_SHARED_BLOCK_REF_KEY;
eebe063b 658 else
5d4f98a2 659 ref->type = BTRFS_TREE_BLOCK_REF_KEY;
5d4f98a2 660 full_ref->level = level;
56bec294 661
1abe9b8a 662 trace_btrfs_delayed_tree_ref(ref, full_ref, action);
663
5d4f98a2 664 existing = tree_insert(&delayed_refs->root, &ref->rb_node);
56bec294
CM
665
666 if (existing) {
5d4f98a2
YZ
667 update_existing_ref(trans, delayed_refs, existing, ref);
668 /*
669 * we've updated the existing ref, free the newly
670 * allocated ref
671 */
78a6184a 672 kmem_cache_free(btrfs_delayed_tree_ref_cachep, full_ref);
5d4f98a2
YZ
673 } else {
674 delayed_refs->num_entries++;
675 trans->delayed_ref_updates++;
676 }
5d4f98a2
YZ
677}
678
679/*
680 * helper to insert a delayed data ref into the rbtree.
681 */
143bede5 682static noinline void add_delayed_data_ref(struct btrfs_fs_info *fs_info,
66d7e7f0 683 struct btrfs_trans_handle *trans,
5d4f98a2
YZ
684 struct btrfs_delayed_ref_node *ref,
685 u64 bytenr, u64 num_bytes, u64 parent,
686 u64 ref_root, u64 owner, u64 offset,
66d7e7f0 687 int action, int for_cow)
5d4f98a2
YZ
688{
689 struct btrfs_delayed_ref_node *existing;
690 struct btrfs_delayed_data_ref *full_ref;
691 struct btrfs_delayed_ref_root *delayed_refs;
00f04b88 692 u64 seq = 0;
5d4f98a2
YZ
693
694 if (action == BTRFS_ADD_DELAYED_EXTENT)
695 action = BTRFS_ADD_DELAYED_REF;
696
697 delayed_refs = &trans->transaction->delayed_refs;
698
699 /* first set the basic ref node struct up */
700 atomic_set(&ref->refs, 1);
701 ref->bytenr = bytenr;
702 ref->num_bytes = num_bytes;
703 ref->ref_mod = 1;
704 ref->action = action;
705 ref->is_head = 0;
706 ref->in_tree = 1;
707
546adb0d
JS
708 if (need_ref_seq(for_cow, ref_root))
709 seq = btrfs_get_tree_mod_seq(fs_info, &trans->delayed_ref_elem);
00f04b88
AJ
710 ref->seq = seq;
711
5d4f98a2 712 full_ref = btrfs_delayed_node_to_data_ref(ref);
eebe063b
AJ
713 full_ref->parent = parent;
714 full_ref->root = ref_root;
715 if (parent)
5d4f98a2 716 ref->type = BTRFS_SHARED_DATA_REF_KEY;
eebe063b 717 else
5d4f98a2 718 ref->type = BTRFS_EXTENT_DATA_REF_KEY;
66d7e7f0 719
5d4f98a2
YZ
720 full_ref->objectid = owner;
721 full_ref->offset = offset;
56bec294 722
1abe9b8a 723 trace_btrfs_delayed_data_ref(ref, full_ref, action);
724
5d4f98a2
YZ
725 existing = tree_insert(&delayed_refs->root, &ref->rb_node);
726
727 if (existing) {
728 update_existing_ref(trans, delayed_refs, existing, ref);
56bec294
CM
729 /*
730 * we've updated the existing ref, free the newly
731 * allocated ref
732 */
78a6184a 733 kmem_cache_free(btrfs_delayed_data_ref_cachep, full_ref);
56bec294
CM
734 } else {
735 delayed_refs->num_entries++;
736 trans->delayed_ref_updates++;
737 }
56bec294
CM
738}
739
740/*
5d4f98a2 741 * add a delayed tree ref. This does all of the accounting required
56bec294
CM
742 * to make sure the delayed ref is eventually processed before this
743 * transaction commits.
744 */
66d7e7f0
AJ
745int btrfs_add_delayed_tree_ref(struct btrfs_fs_info *fs_info,
746 struct btrfs_trans_handle *trans,
5d4f98a2
YZ
747 u64 bytenr, u64 num_bytes, u64 parent,
748 u64 ref_root, int level, int action,
66d7e7f0
AJ
749 struct btrfs_delayed_extent_op *extent_op,
750 int for_cow)
56bec294 751{
5d4f98a2 752 struct btrfs_delayed_tree_ref *ref;
56bec294
CM
753 struct btrfs_delayed_ref_head *head_ref;
754 struct btrfs_delayed_ref_root *delayed_refs;
56bec294 755
5d4f98a2 756 BUG_ON(extent_op && extent_op->is_data);
78a6184a 757 ref = kmem_cache_alloc(btrfs_delayed_tree_ref_cachep, GFP_NOFS);
56bec294
CM
758 if (!ref)
759 return -ENOMEM;
760
78a6184a 761 head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS);
5d4f98a2 762 if (!head_ref) {
78a6184a 763 kmem_cache_free(btrfs_delayed_tree_ref_cachep, ref);
5d4f98a2
YZ
764 return -ENOMEM;
765 }
766
767 head_ref->extent_op = extent_op;
768
769 delayed_refs = &trans->transaction->delayed_refs;
770 spin_lock(&delayed_refs->lock);
771
56bec294 772 /*
5d4f98a2
YZ
773 * insert both the head node and the new ref without dropping
774 * the spin lock
56bec294 775 */
143bede5 776 add_delayed_ref_head(fs_info, trans, &head_ref->node, bytenr,
66d7e7f0 777 num_bytes, action, 0);
5d4f98a2 778
143bede5 779 add_delayed_tree_ref(fs_info, trans, &ref->node, bytenr,
66d7e7f0
AJ
780 num_bytes, parent, ref_root, level, action,
781 for_cow);
5d4f98a2 782 spin_unlock(&delayed_refs->lock);
546adb0d
JS
783 if (need_ref_seq(for_cow, ref_root))
784 btrfs_qgroup_record_ref(trans, &ref->node, extent_op);
95a06077 785
5d4f98a2
YZ
786 return 0;
787}
788
789/*
790 * add a delayed data ref. it's similar to btrfs_add_delayed_tree_ref.
791 */
66d7e7f0
AJ
792int btrfs_add_delayed_data_ref(struct btrfs_fs_info *fs_info,
793 struct btrfs_trans_handle *trans,
5d4f98a2
YZ
794 u64 bytenr, u64 num_bytes,
795 u64 parent, u64 ref_root,
796 u64 owner, u64 offset, int action,
66d7e7f0
AJ
797 struct btrfs_delayed_extent_op *extent_op,
798 int for_cow)
5d4f98a2
YZ
799{
800 struct btrfs_delayed_data_ref *ref;
801 struct btrfs_delayed_ref_head *head_ref;
802 struct btrfs_delayed_ref_root *delayed_refs;
5d4f98a2
YZ
803
804 BUG_ON(extent_op && !extent_op->is_data);
78a6184a 805 ref = kmem_cache_alloc(btrfs_delayed_data_ref_cachep, GFP_NOFS);
5d4f98a2
YZ
806 if (!ref)
807 return -ENOMEM;
56bec294 808
78a6184a 809 head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS);
56bec294 810 if (!head_ref) {
78a6184a 811 kmem_cache_free(btrfs_delayed_data_ref_cachep, ref);
56bec294
CM
812 return -ENOMEM;
813 }
5d4f98a2
YZ
814
815 head_ref->extent_op = extent_op;
816
56bec294
CM
817 delayed_refs = &trans->transaction->delayed_refs;
818 spin_lock(&delayed_refs->lock);
819
820 /*
821 * insert both the head node and the new ref without dropping
822 * the spin lock
823 */
143bede5 824 add_delayed_ref_head(fs_info, trans, &head_ref->node, bytenr,
66d7e7f0 825 num_bytes, action, 1);
56bec294 826
143bede5 827 add_delayed_data_ref(fs_info, trans, &ref->node, bytenr,
66d7e7f0
AJ
828 num_bytes, parent, ref_root, owner, offset,
829 action, for_cow);
5d4f98a2 830 spin_unlock(&delayed_refs->lock);
546adb0d
JS
831 if (need_ref_seq(for_cow, ref_root))
832 btrfs_qgroup_record_ref(trans, &ref->node, extent_op);
95a06077 833
5d4f98a2
YZ
834 return 0;
835}
836
66d7e7f0
AJ
837int btrfs_add_delayed_extent_op(struct btrfs_fs_info *fs_info,
838 struct btrfs_trans_handle *trans,
5d4f98a2
YZ
839 u64 bytenr, u64 num_bytes,
840 struct btrfs_delayed_extent_op *extent_op)
841{
842 struct btrfs_delayed_ref_head *head_ref;
843 struct btrfs_delayed_ref_root *delayed_refs;
5d4f98a2 844
78a6184a 845 head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS);
5d4f98a2
YZ
846 if (!head_ref)
847 return -ENOMEM;
848
849 head_ref->extent_op = extent_op;
850
851 delayed_refs = &trans->transaction->delayed_refs;
852 spin_lock(&delayed_refs->lock);
853
143bede5 854 add_delayed_ref_head(fs_info, trans, &head_ref->node, bytenr,
5d4f98a2
YZ
855 num_bytes, BTRFS_UPDATE_DELAYED_HEAD,
856 extent_op->is_data);
5d4f98a2 857
56bec294
CM
858 spin_unlock(&delayed_refs->lock);
859 return 0;
860}
861
1887be66
CM
862/*
863 * this does a simple search for the head node for a given extent.
864 * It must be called with the delayed ref spinlock held, and it returns
865 * the head node if any where found, or NULL if not.
866 */
867struct btrfs_delayed_ref_head *
868btrfs_find_delayed_ref_head(struct btrfs_trans_handle *trans, u64 bytenr)
869{
870 struct btrfs_delayed_ref_node *ref;
871 struct btrfs_delayed_ref_root *delayed_refs;
872
873 delayed_refs = &trans->transaction->delayed_refs;
d1270cd9 874 ref = find_ref_head(&delayed_refs->root, bytenr, NULL, 0);
1887be66
CM
875 if (ref)
876 return btrfs_delayed_node_to_head(ref);
877 return NULL;
878}
78a6184a
MX
879
880void btrfs_delayed_ref_exit(void)
881{
882 if (btrfs_delayed_ref_head_cachep)
883 kmem_cache_destroy(btrfs_delayed_ref_head_cachep);
884 if (btrfs_delayed_tree_ref_cachep)
885 kmem_cache_destroy(btrfs_delayed_tree_ref_cachep);
886 if (btrfs_delayed_data_ref_cachep)
887 kmem_cache_destroy(btrfs_delayed_data_ref_cachep);
888 if (btrfs_delayed_extent_op_cachep)
889 kmem_cache_destroy(btrfs_delayed_extent_op_cachep);
890}
891
892int btrfs_delayed_ref_init(void)
893{
894 btrfs_delayed_ref_head_cachep = kmem_cache_create(
895 "btrfs_delayed_ref_head",
896 sizeof(struct btrfs_delayed_ref_head), 0,
897 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
898 if (!btrfs_delayed_ref_head_cachep)
899 goto fail;
900
901 btrfs_delayed_tree_ref_cachep = kmem_cache_create(
902 "btrfs_delayed_tree_ref",
903 sizeof(struct btrfs_delayed_tree_ref), 0,
904 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
905 if (!btrfs_delayed_tree_ref_cachep)
906 goto fail;
907
908 btrfs_delayed_data_ref_cachep = kmem_cache_create(
909 "btrfs_delayed_data_ref",
910 sizeof(struct btrfs_delayed_data_ref), 0,
911 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
912 if (!btrfs_delayed_data_ref_cachep)
913 goto fail;
914
915 btrfs_delayed_extent_op_cachep = kmem_cache_create(
916 "btrfs_delayed_extent_op",
917 sizeof(struct btrfs_delayed_extent_op), 0,
918 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
919 if (!btrfs_delayed_extent_op_cachep)
920 goto fail;
921
922 return 0;
923fail:
924 btrfs_delayed_ref_exit();
925 return -ENOMEM;
926}
This page took 0.250916 seconds and 5 git commands to generate.