Loading...
1/*
2 * Copyright (C) 2011 STRATO. 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 "ctree.h"
20#include "disk-io.h"
21#include "backref.h"
22#include "ulist.h"
23#include "transaction.h"
24#include "delayed-ref.h"
25#include "locking.h"
26
27struct extent_inode_elem {
28 u64 inum;
29 u64 offset;
30 struct extent_inode_elem *next;
31};
32
33static int check_extent_in_eb(struct btrfs_key *key, struct extent_buffer *eb,
34 struct btrfs_file_extent_item *fi,
35 u64 extent_item_pos,
36 struct extent_inode_elem **eie)
37{
38 u64 data_offset;
39 u64 data_len;
40 struct extent_inode_elem *e;
41
42 data_offset = btrfs_file_extent_offset(eb, fi);
43 data_len = btrfs_file_extent_num_bytes(eb, fi);
44
45 if (extent_item_pos < data_offset ||
46 extent_item_pos >= data_offset + data_len)
47 return 1;
48
49 e = kmalloc(sizeof(*e), GFP_NOFS);
50 if (!e)
51 return -ENOMEM;
52
53 e->next = *eie;
54 e->inum = key->objectid;
55 e->offset = key->offset + (extent_item_pos - data_offset);
56 *eie = e;
57
58 return 0;
59}
60
61static int find_extent_in_eb(struct extent_buffer *eb, u64 wanted_disk_byte,
62 u64 extent_item_pos,
63 struct extent_inode_elem **eie)
64{
65 u64 disk_byte;
66 struct btrfs_key key;
67 struct btrfs_file_extent_item *fi;
68 int slot;
69 int nritems;
70 int extent_type;
71 int ret;
72
73 /*
74 * from the shared data ref, we only have the leaf but we need
75 * the key. thus, we must look into all items and see that we
76 * find one (some) with a reference to our extent item.
77 */
78 nritems = btrfs_header_nritems(eb);
79 for (slot = 0; slot < nritems; ++slot) {
80 btrfs_item_key_to_cpu(eb, &key, slot);
81 if (key.type != BTRFS_EXTENT_DATA_KEY)
82 continue;
83 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
84 extent_type = btrfs_file_extent_type(eb, fi);
85 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
86 continue;
87 /* don't skip BTRFS_FILE_EXTENT_PREALLOC, we can handle that */
88 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
89 if (disk_byte != wanted_disk_byte)
90 continue;
91
92 ret = check_extent_in_eb(&key, eb, fi, extent_item_pos, eie);
93 if (ret < 0)
94 return ret;
95 }
96
97 return 0;
98}
99
100/*
101 * this structure records all encountered refs on the way up to the root
102 */
103struct __prelim_ref {
104 struct list_head list;
105 u64 root_id;
106 struct btrfs_key key_for_search;
107 int level;
108 int count;
109 struct extent_inode_elem *inode_list;
110 u64 parent;
111 u64 wanted_disk_byte;
112};
113
114/*
115 * the rules for all callers of this function are:
116 * - obtaining the parent is the goal
117 * - if you add a key, you must know that it is a correct key
118 * - if you cannot add the parent or a correct key, then we will look into the
119 * block later to set a correct key
120 *
121 * delayed refs
122 * ============
123 * backref type | shared | indirect | shared | indirect
124 * information | tree | tree | data | data
125 * --------------------+--------+----------+--------+----------
126 * parent logical | y | - | - | -
127 * key to resolve | - | y | y | y
128 * tree block logical | - | - | - | -
129 * root for resolving | y | y | y | y
130 *
131 * - column 1: we've the parent -> done
132 * - column 2, 3, 4: we use the key to find the parent
133 *
134 * on disk refs (inline or keyed)
135 * ==============================
136 * backref type | shared | indirect | shared | indirect
137 * information | tree | tree | data | data
138 * --------------------+--------+----------+--------+----------
139 * parent logical | y | - | y | -
140 * key to resolve | - | - | - | y
141 * tree block logical | y | y | y | y
142 * root for resolving | - | y | y | y
143 *
144 * - column 1, 3: we've the parent -> done
145 * - column 2: we take the first key from the block to find the parent
146 * (see __add_missing_keys)
147 * - column 4: we use the key to find the parent
148 *
149 * additional information that's available but not required to find the parent
150 * block might help in merging entries to gain some speed.
151 */
152
153static int __add_prelim_ref(struct list_head *head, u64 root_id,
154 struct btrfs_key *key, int level,
155 u64 parent, u64 wanted_disk_byte, int count)
156{
157 struct __prelim_ref *ref;
158
159 /* in case we're adding delayed refs, we're holding the refs spinlock */
160 ref = kmalloc(sizeof(*ref), GFP_ATOMIC);
161 if (!ref)
162 return -ENOMEM;
163
164 ref->root_id = root_id;
165 if (key)
166 ref->key_for_search = *key;
167 else
168 memset(&ref->key_for_search, 0, sizeof(ref->key_for_search));
169
170 ref->inode_list = NULL;
171 ref->level = level;
172 ref->count = count;
173 ref->parent = parent;
174 ref->wanted_disk_byte = wanted_disk_byte;
175 list_add_tail(&ref->list, head);
176
177 return 0;
178}
179
180static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path,
181 struct ulist *parents, int level,
182 struct btrfs_key *key_for_search, u64 time_seq,
183 u64 wanted_disk_byte,
184 const u64 *extent_item_pos)
185{
186 int ret = 0;
187 int slot;
188 struct extent_buffer *eb;
189 struct btrfs_key key;
190 struct btrfs_file_extent_item *fi;
191 struct extent_inode_elem *eie = NULL;
192 u64 disk_byte;
193
194 if (level != 0) {
195 eb = path->nodes[level];
196 ret = ulist_add(parents, eb->start, 0, GFP_NOFS);
197 if (ret < 0)
198 return ret;
199 return 0;
200 }
201
202 /*
203 * We normally enter this function with the path already pointing to
204 * the first item to check. But sometimes, we may enter it with
205 * slot==nritems. In that case, go to the next leaf before we continue.
206 */
207 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0]))
208 ret = btrfs_next_old_leaf(root, path, time_seq);
209
210 while (!ret) {
211 eb = path->nodes[0];
212 slot = path->slots[0];
213
214 btrfs_item_key_to_cpu(eb, &key, slot);
215
216 if (key.objectid != key_for_search->objectid ||
217 key.type != BTRFS_EXTENT_DATA_KEY)
218 break;
219
220 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
221 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
222
223 if (disk_byte == wanted_disk_byte) {
224 eie = NULL;
225 if (extent_item_pos) {
226 ret = check_extent_in_eb(&key, eb, fi,
227 *extent_item_pos,
228 &eie);
229 if (ret < 0)
230 break;
231 }
232 if (!ret) {
233 ret = ulist_add(parents, eb->start,
234 (unsigned long)eie, GFP_NOFS);
235 if (ret < 0)
236 break;
237 if (!extent_item_pos) {
238 ret = btrfs_next_old_leaf(root, path,
239 time_seq);
240 continue;
241 }
242 }
243 }
244 ret = btrfs_next_old_item(root, path, time_seq);
245 }
246
247 if (ret > 0)
248 ret = 0;
249 return ret;
250}
251
252/*
253 * resolve an indirect backref in the form (root_id, key, level)
254 * to a logical address
255 */
256static int __resolve_indirect_ref(struct btrfs_fs_info *fs_info,
257 int search_commit_root,
258 u64 time_seq,
259 struct __prelim_ref *ref,
260 struct ulist *parents,
261 const u64 *extent_item_pos)
262{
263 struct btrfs_path *path;
264 struct btrfs_root *root;
265 struct btrfs_key root_key;
266 struct extent_buffer *eb;
267 int ret = 0;
268 int root_level;
269 int level = ref->level;
270
271 path = btrfs_alloc_path();
272 if (!path)
273 return -ENOMEM;
274 path->search_commit_root = !!search_commit_root;
275
276 root_key.objectid = ref->root_id;
277 root_key.type = BTRFS_ROOT_ITEM_KEY;
278 root_key.offset = (u64)-1;
279 root = btrfs_read_fs_root_no_name(fs_info, &root_key);
280 if (IS_ERR(root)) {
281 ret = PTR_ERR(root);
282 goto out;
283 }
284
285 rcu_read_lock();
286 root_level = btrfs_header_level(root->node);
287 rcu_read_unlock();
288
289 if (root_level + 1 == level)
290 goto out;
291
292 path->lowest_level = level;
293 ret = btrfs_search_old_slot(root, &ref->key_for_search, path, time_seq);
294 pr_debug("search slot in root %llu (level %d, ref count %d) returned "
295 "%d for key (%llu %u %llu)\n",
296 (unsigned long long)ref->root_id, level, ref->count, ret,
297 (unsigned long long)ref->key_for_search.objectid,
298 ref->key_for_search.type,
299 (unsigned long long)ref->key_for_search.offset);
300 if (ret < 0)
301 goto out;
302
303 eb = path->nodes[level];
304 while (!eb) {
305 if (!level) {
306 WARN_ON(1);
307 ret = 1;
308 goto out;
309 }
310 level--;
311 eb = path->nodes[level];
312 }
313
314 ret = add_all_parents(root, path, parents, level, &ref->key_for_search,
315 time_seq, ref->wanted_disk_byte,
316 extent_item_pos);
317out:
318 btrfs_free_path(path);
319 return ret;
320}
321
322/*
323 * resolve all indirect backrefs from the list
324 */
325static int __resolve_indirect_refs(struct btrfs_fs_info *fs_info,
326 int search_commit_root, u64 time_seq,
327 struct list_head *head,
328 const u64 *extent_item_pos)
329{
330 int err;
331 int ret = 0;
332 struct __prelim_ref *ref;
333 struct __prelim_ref *ref_safe;
334 struct __prelim_ref *new_ref;
335 struct ulist *parents;
336 struct ulist_node *node;
337 struct ulist_iterator uiter;
338
339 parents = ulist_alloc(GFP_NOFS);
340 if (!parents)
341 return -ENOMEM;
342
343 /*
344 * _safe allows us to insert directly after the current item without
345 * iterating over the newly inserted items.
346 * we're also allowed to re-assign ref during iteration.
347 */
348 list_for_each_entry_safe(ref, ref_safe, head, list) {
349 if (ref->parent) /* already direct */
350 continue;
351 if (ref->count == 0)
352 continue;
353 err = __resolve_indirect_ref(fs_info, search_commit_root,
354 time_seq, ref, parents,
355 extent_item_pos);
356 if (err) {
357 if (ret == 0)
358 ret = err;
359 continue;
360 }
361
362 /* we put the first parent into the ref at hand */
363 ULIST_ITER_INIT(&uiter);
364 node = ulist_next(parents, &uiter);
365 ref->parent = node ? node->val : 0;
366 ref->inode_list =
367 node ? (struct extent_inode_elem *)node->aux : 0;
368
369 /* additional parents require new refs being added here */
370 while ((node = ulist_next(parents, &uiter))) {
371 new_ref = kmalloc(sizeof(*new_ref), GFP_NOFS);
372 if (!new_ref) {
373 ret = -ENOMEM;
374 break;
375 }
376 memcpy(new_ref, ref, sizeof(*ref));
377 new_ref->parent = node->val;
378 new_ref->inode_list =
379 (struct extent_inode_elem *)node->aux;
380 list_add(&new_ref->list, &ref->list);
381 }
382 ulist_reinit(parents);
383 }
384
385 ulist_free(parents);
386 return ret;
387}
388
389static inline int ref_for_same_block(struct __prelim_ref *ref1,
390 struct __prelim_ref *ref2)
391{
392 if (ref1->level != ref2->level)
393 return 0;
394 if (ref1->root_id != ref2->root_id)
395 return 0;
396 if (ref1->key_for_search.type != ref2->key_for_search.type)
397 return 0;
398 if (ref1->key_for_search.objectid != ref2->key_for_search.objectid)
399 return 0;
400 if (ref1->key_for_search.offset != ref2->key_for_search.offset)
401 return 0;
402 if (ref1->parent != ref2->parent)
403 return 0;
404
405 return 1;
406}
407
408/*
409 * read tree blocks and add keys where required.
410 */
411static int __add_missing_keys(struct btrfs_fs_info *fs_info,
412 struct list_head *head)
413{
414 struct list_head *pos;
415 struct extent_buffer *eb;
416
417 list_for_each(pos, head) {
418 struct __prelim_ref *ref;
419 ref = list_entry(pos, struct __prelim_ref, list);
420
421 if (ref->parent)
422 continue;
423 if (ref->key_for_search.type)
424 continue;
425 BUG_ON(!ref->wanted_disk_byte);
426 eb = read_tree_block(fs_info->tree_root, ref->wanted_disk_byte,
427 fs_info->tree_root->leafsize, 0);
428 BUG_ON(!eb);
429 btrfs_tree_read_lock(eb);
430 if (btrfs_header_level(eb) == 0)
431 btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
432 else
433 btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
434 btrfs_tree_read_unlock(eb);
435 free_extent_buffer(eb);
436 }
437 return 0;
438}
439
440/*
441 * merge two lists of backrefs and adjust counts accordingly
442 *
443 * mode = 1: merge identical keys, if key is set
444 * FIXME: if we add more keys in __add_prelim_ref, we can merge more here.
445 * additionally, we could even add a key range for the blocks we
446 * looked into to merge even more (-> replace unresolved refs by those
447 * having a parent).
448 * mode = 2: merge identical parents
449 */
450static int __merge_refs(struct list_head *head, int mode)
451{
452 struct list_head *pos1;
453
454 list_for_each(pos1, head) {
455 struct list_head *n2;
456 struct list_head *pos2;
457 struct __prelim_ref *ref1;
458
459 ref1 = list_entry(pos1, struct __prelim_ref, list);
460
461 for (pos2 = pos1->next, n2 = pos2->next; pos2 != head;
462 pos2 = n2, n2 = pos2->next) {
463 struct __prelim_ref *ref2;
464 struct __prelim_ref *xchg;
465
466 ref2 = list_entry(pos2, struct __prelim_ref, list);
467
468 if (mode == 1) {
469 if (!ref_for_same_block(ref1, ref2))
470 continue;
471 if (!ref1->parent && ref2->parent) {
472 xchg = ref1;
473 ref1 = ref2;
474 ref2 = xchg;
475 }
476 ref1->count += ref2->count;
477 } else {
478 if (ref1->parent != ref2->parent)
479 continue;
480 ref1->count += ref2->count;
481 }
482 list_del(&ref2->list);
483 kfree(ref2);
484 }
485
486 }
487 return 0;
488}
489
490/*
491 * add all currently queued delayed refs from this head whose seq nr is
492 * smaller or equal that seq to the list
493 */
494static int __add_delayed_refs(struct btrfs_delayed_ref_head *head, u64 seq,
495 struct list_head *prefs)
496{
497 struct btrfs_delayed_extent_op *extent_op = head->extent_op;
498 struct rb_node *n = &head->node.rb_node;
499 struct btrfs_key key;
500 struct btrfs_key op_key = {0};
501 int sgn;
502 int ret = 0;
503
504 if (extent_op && extent_op->update_key)
505 btrfs_disk_key_to_cpu(&op_key, &extent_op->key);
506
507 while ((n = rb_prev(n))) {
508 struct btrfs_delayed_ref_node *node;
509 node = rb_entry(n, struct btrfs_delayed_ref_node,
510 rb_node);
511 if (node->bytenr != head->node.bytenr)
512 break;
513 WARN_ON(node->is_head);
514
515 if (node->seq > seq)
516 continue;
517
518 switch (node->action) {
519 case BTRFS_ADD_DELAYED_EXTENT:
520 case BTRFS_UPDATE_DELAYED_HEAD:
521 WARN_ON(1);
522 continue;
523 case BTRFS_ADD_DELAYED_REF:
524 sgn = 1;
525 break;
526 case BTRFS_DROP_DELAYED_REF:
527 sgn = -1;
528 break;
529 default:
530 BUG_ON(1);
531 }
532 switch (node->type) {
533 case BTRFS_TREE_BLOCK_REF_KEY: {
534 struct btrfs_delayed_tree_ref *ref;
535
536 ref = btrfs_delayed_node_to_tree_ref(node);
537 ret = __add_prelim_ref(prefs, ref->root, &op_key,
538 ref->level + 1, 0, node->bytenr,
539 node->ref_mod * sgn);
540 break;
541 }
542 case BTRFS_SHARED_BLOCK_REF_KEY: {
543 struct btrfs_delayed_tree_ref *ref;
544
545 ref = btrfs_delayed_node_to_tree_ref(node);
546 ret = __add_prelim_ref(prefs, ref->root, NULL,
547 ref->level + 1, ref->parent,
548 node->bytenr,
549 node->ref_mod * sgn);
550 break;
551 }
552 case BTRFS_EXTENT_DATA_REF_KEY: {
553 struct btrfs_delayed_data_ref *ref;
554 ref = btrfs_delayed_node_to_data_ref(node);
555
556 key.objectid = ref->objectid;
557 key.type = BTRFS_EXTENT_DATA_KEY;
558 key.offset = ref->offset;
559 ret = __add_prelim_ref(prefs, ref->root, &key, 0, 0,
560 node->bytenr,
561 node->ref_mod * sgn);
562 break;
563 }
564 case BTRFS_SHARED_DATA_REF_KEY: {
565 struct btrfs_delayed_data_ref *ref;
566
567 ref = btrfs_delayed_node_to_data_ref(node);
568
569 key.objectid = ref->objectid;
570 key.type = BTRFS_EXTENT_DATA_KEY;
571 key.offset = ref->offset;
572 ret = __add_prelim_ref(prefs, ref->root, &key, 0,
573 ref->parent, node->bytenr,
574 node->ref_mod * sgn);
575 break;
576 }
577 default:
578 WARN_ON(1);
579 }
580 BUG_ON(ret);
581 }
582
583 return 0;
584}
585
586/*
587 * add all inline backrefs for bytenr to the list
588 */
589static int __add_inline_refs(struct btrfs_fs_info *fs_info,
590 struct btrfs_path *path, u64 bytenr,
591 int *info_level, struct list_head *prefs)
592{
593 int ret = 0;
594 int slot;
595 struct extent_buffer *leaf;
596 struct btrfs_key key;
597 unsigned long ptr;
598 unsigned long end;
599 struct btrfs_extent_item *ei;
600 u64 flags;
601 u64 item_size;
602
603 /*
604 * enumerate all inline refs
605 */
606 leaf = path->nodes[0];
607 slot = path->slots[0];
608
609 item_size = btrfs_item_size_nr(leaf, slot);
610 BUG_ON(item_size < sizeof(*ei));
611
612 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
613 flags = btrfs_extent_flags(leaf, ei);
614
615 ptr = (unsigned long)(ei + 1);
616 end = (unsigned long)ei + item_size;
617
618 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
619 struct btrfs_tree_block_info *info;
620
621 info = (struct btrfs_tree_block_info *)ptr;
622 *info_level = btrfs_tree_block_level(leaf, info);
623 ptr += sizeof(struct btrfs_tree_block_info);
624 BUG_ON(ptr > end);
625 } else {
626 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
627 }
628
629 while (ptr < end) {
630 struct btrfs_extent_inline_ref *iref;
631 u64 offset;
632 int type;
633
634 iref = (struct btrfs_extent_inline_ref *)ptr;
635 type = btrfs_extent_inline_ref_type(leaf, iref);
636 offset = btrfs_extent_inline_ref_offset(leaf, iref);
637
638 switch (type) {
639 case BTRFS_SHARED_BLOCK_REF_KEY:
640 ret = __add_prelim_ref(prefs, 0, NULL,
641 *info_level + 1, offset,
642 bytenr, 1);
643 break;
644 case BTRFS_SHARED_DATA_REF_KEY: {
645 struct btrfs_shared_data_ref *sdref;
646 int count;
647
648 sdref = (struct btrfs_shared_data_ref *)(iref + 1);
649 count = btrfs_shared_data_ref_count(leaf, sdref);
650 ret = __add_prelim_ref(prefs, 0, NULL, 0, offset,
651 bytenr, count);
652 break;
653 }
654 case BTRFS_TREE_BLOCK_REF_KEY:
655 ret = __add_prelim_ref(prefs, offset, NULL,
656 *info_level + 1, 0,
657 bytenr, 1);
658 break;
659 case BTRFS_EXTENT_DATA_REF_KEY: {
660 struct btrfs_extent_data_ref *dref;
661 int count;
662 u64 root;
663
664 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
665 count = btrfs_extent_data_ref_count(leaf, dref);
666 key.objectid = btrfs_extent_data_ref_objectid(leaf,
667 dref);
668 key.type = BTRFS_EXTENT_DATA_KEY;
669 key.offset = btrfs_extent_data_ref_offset(leaf, dref);
670 root = btrfs_extent_data_ref_root(leaf, dref);
671 ret = __add_prelim_ref(prefs, root, &key, 0, 0,
672 bytenr, count);
673 break;
674 }
675 default:
676 WARN_ON(1);
677 }
678 BUG_ON(ret);
679 ptr += btrfs_extent_inline_ref_size(type);
680 }
681
682 return 0;
683}
684
685/*
686 * add all non-inline backrefs for bytenr to the list
687 */
688static int __add_keyed_refs(struct btrfs_fs_info *fs_info,
689 struct btrfs_path *path, u64 bytenr,
690 int info_level, struct list_head *prefs)
691{
692 struct btrfs_root *extent_root = fs_info->extent_root;
693 int ret;
694 int slot;
695 struct extent_buffer *leaf;
696 struct btrfs_key key;
697
698 while (1) {
699 ret = btrfs_next_item(extent_root, path);
700 if (ret < 0)
701 break;
702 if (ret) {
703 ret = 0;
704 break;
705 }
706
707 slot = path->slots[0];
708 leaf = path->nodes[0];
709 btrfs_item_key_to_cpu(leaf, &key, slot);
710
711 if (key.objectid != bytenr)
712 break;
713 if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
714 continue;
715 if (key.type > BTRFS_SHARED_DATA_REF_KEY)
716 break;
717
718 switch (key.type) {
719 case BTRFS_SHARED_BLOCK_REF_KEY:
720 ret = __add_prelim_ref(prefs, 0, NULL,
721 info_level + 1, key.offset,
722 bytenr, 1);
723 break;
724 case BTRFS_SHARED_DATA_REF_KEY: {
725 struct btrfs_shared_data_ref *sdref;
726 int count;
727
728 sdref = btrfs_item_ptr(leaf, slot,
729 struct btrfs_shared_data_ref);
730 count = btrfs_shared_data_ref_count(leaf, sdref);
731 ret = __add_prelim_ref(prefs, 0, NULL, 0, key.offset,
732 bytenr, count);
733 break;
734 }
735 case BTRFS_TREE_BLOCK_REF_KEY:
736 ret = __add_prelim_ref(prefs, key.offset, NULL,
737 info_level + 1, 0,
738 bytenr, 1);
739 break;
740 case BTRFS_EXTENT_DATA_REF_KEY: {
741 struct btrfs_extent_data_ref *dref;
742 int count;
743 u64 root;
744
745 dref = btrfs_item_ptr(leaf, slot,
746 struct btrfs_extent_data_ref);
747 count = btrfs_extent_data_ref_count(leaf, dref);
748 key.objectid = btrfs_extent_data_ref_objectid(leaf,
749 dref);
750 key.type = BTRFS_EXTENT_DATA_KEY;
751 key.offset = btrfs_extent_data_ref_offset(leaf, dref);
752 root = btrfs_extent_data_ref_root(leaf, dref);
753 ret = __add_prelim_ref(prefs, root, &key, 0, 0,
754 bytenr, count);
755 break;
756 }
757 default:
758 WARN_ON(1);
759 }
760 BUG_ON(ret);
761 }
762
763 return ret;
764}
765
766/*
767 * this adds all existing backrefs (inline backrefs, backrefs and delayed
768 * refs) for the given bytenr to the refs list, merges duplicates and resolves
769 * indirect refs to their parent bytenr.
770 * When roots are found, they're added to the roots list
771 *
772 * FIXME some caching might speed things up
773 */
774static int find_parent_nodes(struct btrfs_trans_handle *trans,
775 struct btrfs_fs_info *fs_info, u64 bytenr,
776 u64 delayed_ref_seq, u64 time_seq,
777 struct ulist *refs, struct ulist *roots,
778 const u64 *extent_item_pos)
779{
780 struct btrfs_key key;
781 struct btrfs_path *path;
782 struct btrfs_delayed_ref_root *delayed_refs = NULL;
783 struct btrfs_delayed_ref_head *head;
784 int info_level = 0;
785 int ret;
786 int search_commit_root = (trans == BTRFS_BACKREF_SEARCH_COMMIT_ROOT);
787 struct list_head prefs_delayed;
788 struct list_head prefs;
789 struct __prelim_ref *ref;
790
791 INIT_LIST_HEAD(&prefs);
792 INIT_LIST_HEAD(&prefs_delayed);
793
794 key.objectid = bytenr;
795 key.type = BTRFS_EXTENT_ITEM_KEY;
796 key.offset = (u64)-1;
797
798 path = btrfs_alloc_path();
799 if (!path)
800 return -ENOMEM;
801 path->search_commit_root = !!search_commit_root;
802
803 /*
804 * grab both a lock on the path and a lock on the delayed ref head.
805 * We need both to get a consistent picture of how the refs look
806 * at a specified point in time
807 */
808again:
809 head = NULL;
810
811 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
812 if (ret < 0)
813 goto out;
814 BUG_ON(ret == 0);
815
816 if (trans != BTRFS_BACKREF_SEARCH_COMMIT_ROOT) {
817 /*
818 * look if there are updates for this ref queued and lock the
819 * head
820 */
821 delayed_refs = &trans->transaction->delayed_refs;
822 spin_lock(&delayed_refs->lock);
823 head = btrfs_find_delayed_ref_head(trans, bytenr);
824 if (head) {
825 if (!mutex_trylock(&head->mutex)) {
826 atomic_inc(&head->node.refs);
827 spin_unlock(&delayed_refs->lock);
828
829 btrfs_release_path(path);
830
831 /*
832 * Mutex was contended, block until it's
833 * released and try again
834 */
835 mutex_lock(&head->mutex);
836 mutex_unlock(&head->mutex);
837 btrfs_put_delayed_ref(&head->node);
838 goto again;
839 }
840 ret = __add_delayed_refs(head, delayed_ref_seq,
841 &prefs_delayed);
842 mutex_unlock(&head->mutex);
843 if (ret) {
844 spin_unlock(&delayed_refs->lock);
845 goto out;
846 }
847 }
848 spin_unlock(&delayed_refs->lock);
849 }
850
851 if (path->slots[0]) {
852 struct extent_buffer *leaf;
853 int slot;
854
855 path->slots[0]--;
856 leaf = path->nodes[0];
857 slot = path->slots[0];
858 btrfs_item_key_to_cpu(leaf, &key, slot);
859 if (key.objectid == bytenr &&
860 key.type == BTRFS_EXTENT_ITEM_KEY) {
861 ret = __add_inline_refs(fs_info, path, bytenr,
862 &info_level, &prefs);
863 if (ret)
864 goto out;
865 ret = __add_keyed_refs(fs_info, path, bytenr,
866 info_level, &prefs);
867 if (ret)
868 goto out;
869 }
870 }
871 btrfs_release_path(path);
872
873 list_splice_init(&prefs_delayed, &prefs);
874
875 ret = __add_missing_keys(fs_info, &prefs);
876 if (ret)
877 goto out;
878
879 ret = __merge_refs(&prefs, 1);
880 if (ret)
881 goto out;
882
883 ret = __resolve_indirect_refs(fs_info, search_commit_root, time_seq,
884 &prefs, extent_item_pos);
885 if (ret)
886 goto out;
887
888 ret = __merge_refs(&prefs, 2);
889 if (ret)
890 goto out;
891
892 while (!list_empty(&prefs)) {
893 ref = list_first_entry(&prefs, struct __prelim_ref, list);
894 list_del(&ref->list);
895 if (ref->count < 0)
896 WARN_ON(1);
897 if (ref->count && ref->root_id && ref->parent == 0) {
898 /* no parent == root of tree */
899 ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS);
900 BUG_ON(ret < 0);
901 }
902 if (ref->count && ref->parent) {
903 struct extent_inode_elem *eie = NULL;
904 if (extent_item_pos && !ref->inode_list) {
905 u32 bsz;
906 struct extent_buffer *eb;
907 bsz = btrfs_level_size(fs_info->extent_root,
908 info_level);
909 eb = read_tree_block(fs_info->extent_root,
910 ref->parent, bsz, 0);
911 BUG_ON(!eb);
912 ret = find_extent_in_eb(eb, bytenr,
913 *extent_item_pos, &eie);
914 ref->inode_list = eie;
915 free_extent_buffer(eb);
916 }
917 ret = ulist_add_merge(refs, ref->parent,
918 (unsigned long)ref->inode_list,
919 (unsigned long *)&eie, GFP_NOFS);
920 if (!ret && extent_item_pos) {
921 /*
922 * we've recorded that parent, so we must extend
923 * its inode list here
924 */
925 BUG_ON(!eie);
926 while (eie->next)
927 eie = eie->next;
928 eie->next = ref->inode_list;
929 }
930 BUG_ON(ret < 0);
931 }
932 kfree(ref);
933 }
934
935out:
936 btrfs_free_path(path);
937 while (!list_empty(&prefs)) {
938 ref = list_first_entry(&prefs, struct __prelim_ref, list);
939 list_del(&ref->list);
940 kfree(ref);
941 }
942 while (!list_empty(&prefs_delayed)) {
943 ref = list_first_entry(&prefs_delayed, struct __prelim_ref,
944 list);
945 list_del(&ref->list);
946 kfree(ref);
947 }
948
949 return ret;
950}
951
952static void free_leaf_list(struct ulist *blocks)
953{
954 struct ulist_node *node = NULL;
955 struct extent_inode_elem *eie;
956 struct extent_inode_elem *eie_next;
957 struct ulist_iterator uiter;
958
959 ULIST_ITER_INIT(&uiter);
960 while ((node = ulist_next(blocks, &uiter))) {
961 if (!node->aux)
962 continue;
963 eie = (struct extent_inode_elem *)node->aux;
964 for (; eie; eie = eie_next) {
965 eie_next = eie->next;
966 kfree(eie);
967 }
968 node->aux = 0;
969 }
970
971 ulist_free(blocks);
972}
973
974/*
975 * Finds all leafs with a reference to the specified combination of bytenr and
976 * offset. key_list_head will point to a list of corresponding keys (caller must
977 * free each list element). The leafs will be stored in the leafs ulist, which
978 * must be freed with ulist_free.
979 *
980 * returns 0 on success, <0 on error
981 */
982static int btrfs_find_all_leafs(struct btrfs_trans_handle *trans,
983 struct btrfs_fs_info *fs_info, u64 bytenr,
984 u64 delayed_ref_seq, u64 time_seq,
985 struct ulist **leafs,
986 const u64 *extent_item_pos)
987{
988 struct ulist *tmp;
989 int ret;
990
991 tmp = ulist_alloc(GFP_NOFS);
992 if (!tmp)
993 return -ENOMEM;
994 *leafs = ulist_alloc(GFP_NOFS);
995 if (!*leafs) {
996 ulist_free(tmp);
997 return -ENOMEM;
998 }
999
1000 ret = find_parent_nodes(trans, fs_info, bytenr, delayed_ref_seq,
1001 time_seq, *leafs, tmp, extent_item_pos);
1002 ulist_free(tmp);
1003
1004 if (ret < 0 && ret != -ENOENT) {
1005 free_leaf_list(*leafs);
1006 return ret;
1007 }
1008
1009 return 0;
1010}
1011
1012/*
1013 * walk all backrefs for a given extent to find all roots that reference this
1014 * extent. Walking a backref means finding all extents that reference this
1015 * extent and in turn walk the backrefs of those, too. Naturally this is a
1016 * recursive process, but here it is implemented in an iterative fashion: We
1017 * find all referencing extents for the extent in question and put them on a
1018 * list. In turn, we find all referencing extents for those, further appending
1019 * to the list. The way we iterate the list allows adding more elements after
1020 * the current while iterating. The process stops when we reach the end of the
1021 * list. Found roots are added to the roots list.
1022 *
1023 * returns 0 on success, < 0 on error.
1024 */
1025int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1026 struct btrfs_fs_info *fs_info, u64 bytenr,
1027 u64 delayed_ref_seq, u64 time_seq,
1028 struct ulist **roots)
1029{
1030 struct ulist *tmp;
1031 struct ulist_node *node = NULL;
1032 struct ulist_iterator uiter;
1033 int ret;
1034
1035 tmp = ulist_alloc(GFP_NOFS);
1036 if (!tmp)
1037 return -ENOMEM;
1038 *roots = ulist_alloc(GFP_NOFS);
1039 if (!*roots) {
1040 ulist_free(tmp);
1041 return -ENOMEM;
1042 }
1043
1044 ULIST_ITER_INIT(&uiter);
1045 while (1) {
1046 ret = find_parent_nodes(trans, fs_info, bytenr, delayed_ref_seq,
1047 time_seq, tmp, *roots, NULL);
1048 if (ret < 0 && ret != -ENOENT) {
1049 ulist_free(tmp);
1050 ulist_free(*roots);
1051 return ret;
1052 }
1053 node = ulist_next(tmp, &uiter);
1054 if (!node)
1055 break;
1056 bytenr = node->val;
1057 }
1058
1059 ulist_free(tmp);
1060 return 0;
1061}
1062
1063
1064static int __inode_info(u64 inum, u64 ioff, u8 key_type,
1065 struct btrfs_root *fs_root, struct btrfs_path *path,
1066 struct btrfs_key *found_key)
1067{
1068 int ret;
1069 struct btrfs_key key;
1070 struct extent_buffer *eb;
1071
1072 key.type = key_type;
1073 key.objectid = inum;
1074 key.offset = ioff;
1075
1076 ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
1077 if (ret < 0)
1078 return ret;
1079
1080 eb = path->nodes[0];
1081 if (ret && path->slots[0] >= btrfs_header_nritems(eb)) {
1082 ret = btrfs_next_leaf(fs_root, path);
1083 if (ret)
1084 return ret;
1085 eb = path->nodes[0];
1086 }
1087
1088 btrfs_item_key_to_cpu(eb, found_key, path->slots[0]);
1089 if (found_key->type != key.type || found_key->objectid != key.objectid)
1090 return 1;
1091
1092 return 0;
1093}
1094
1095/*
1096 * this makes the path point to (inum INODE_ITEM ioff)
1097 */
1098int inode_item_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
1099 struct btrfs_path *path)
1100{
1101 struct btrfs_key key;
1102 return __inode_info(inum, ioff, BTRFS_INODE_ITEM_KEY, fs_root, path,
1103 &key);
1104}
1105
1106static int inode_ref_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
1107 struct btrfs_path *path,
1108 struct btrfs_key *found_key)
1109{
1110 return __inode_info(inum, ioff, BTRFS_INODE_REF_KEY, fs_root, path,
1111 found_key);
1112}
1113
1114/*
1115 * this iterates to turn a btrfs_inode_ref into a full filesystem path. elements
1116 * of the path are separated by '/' and the path is guaranteed to be
1117 * 0-terminated. the path is only given within the current file system.
1118 * Therefore, it never starts with a '/'. the caller is responsible to provide
1119 * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
1120 * the start point of the resulting string is returned. this pointer is within
1121 * dest, normally.
1122 * in case the path buffer would overflow, the pointer is decremented further
1123 * as if output was written to the buffer, though no more output is actually
1124 * generated. that way, the caller can determine how much space would be
1125 * required for the path to fit into the buffer. in that case, the returned
1126 * value will be smaller than dest. callers must check this!
1127 */
1128static char *iref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
1129 struct btrfs_inode_ref *iref,
1130 struct extent_buffer *eb_in, u64 parent,
1131 char *dest, u32 size)
1132{
1133 u32 len;
1134 int slot;
1135 u64 next_inum;
1136 int ret;
1137 s64 bytes_left = size - 1;
1138 struct extent_buffer *eb = eb_in;
1139 struct btrfs_key found_key;
1140 int leave_spinning = path->leave_spinning;
1141
1142 if (bytes_left >= 0)
1143 dest[bytes_left] = '\0';
1144
1145 path->leave_spinning = 1;
1146 while (1) {
1147 len = btrfs_inode_ref_name_len(eb, iref);
1148 bytes_left -= len;
1149 if (bytes_left >= 0)
1150 read_extent_buffer(eb, dest + bytes_left,
1151 (unsigned long)(iref + 1), len);
1152 if (eb != eb_in) {
1153 btrfs_tree_read_unlock_blocking(eb);
1154 free_extent_buffer(eb);
1155 }
1156 ret = inode_ref_info(parent, 0, fs_root, path, &found_key);
1157 if (ret > 0)
1158 ret = -ENOENT;
1159 if (ret)
1160 break;
1161 next_inum = found_key.offset;
1162
1163 /* regular exit ahead */
1164 if (parent == next_inum)
1165 break;
1166
1167 slot = path->slots[0];
1168 eb = path->nodes[0];
1169 /* make sure we can use eb after releasing the path */
1170 if (eb != eb_in) {
1171 atomic_inc(&eb->refs);
1172 btrfs_tree_read_lock(eb);
1173 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1174 }
1175 btrfs_release_path(path);
1176
1177 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1178 parent = next_inum;
1179 --bytes_left;
1180 if (bytes_left >= 0)
1181 dest[bytes_left] = '/';
1182 }
1183
1184 btrfs_release_path(path);
1185 path->leave_spinning = leave_spinning;
1186
1187 if (ret)
1188 return ERR_PTR(ret);
1189
1190 return dest + bytes_left;
1191}
1192
1193/*
1194 * this makes the path point to (logical EXTENT_ITEM *)
1195 * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
1196 * tree blocks and <0 on error.
1197 */
1198int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
1199 struct btrfs_path *path, struct btrfs_key *found_key)
1200{
1201 int ret;
1202 u64 flags;
1203 u32 item_size;
1204 struct extent_buffer *eb;
1205 struct btrfs_extent_item *ei;
1206 struct btrfs_key key;
1207
1208 key.type = BTRFS_EXTENT_ITEM_KEY;
1209 key.objectid = logical;
1210 key.offset = (u64)-1;
1211
1212 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
1213 if (ret < 0)
1214 return ret;
1215 ret = btrfs_previous_item(fs_info->extent_root, path,
1216 0, BTRFS_EXTENT_ITEM_KEY);
1217 if (ret < 0)
1218 return ret;
1219
1220 btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]);
1221 if (found_key->type != BTRFS_EXTENT_ITEM_KEY ||
1222 found_key->objectid > logical ||
1223 found_key->objectid + found_key->offset <= logical) {
1224 pr_debug("logical %llu is not within any extent\n",
1225 (unsigned long long)logical);
1226 return -ENOENT;
1227 }
1228
1229 eb = path->nodes[0];
1230 item_size = btrfs_item_size_nr(eb, path->slots[0]);
1231 BUG_ON(item_size < sizeof(*ei));
1232
1233 ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
1234 flags = btrfs_extent_flags(eb, ei);
1235
1236 pr_debug("logical %llu is at position %llu within the extent (%llu "
1237 "EXTENT_ITEM %llu) flags %#llx size %u\n",
1238 (unsigned long long)logical,
1239 (unsigned long long)(logical - found_key->objectid),
1240 (unsigned long long)found_key->objectid,
1241 (unsigned long long)found_key->offset,
1242 (unsigned long long)flags, item_size);
1243 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1244 return BTRFS_EXTENT_FLAG_TREE_BLOCK;
1245 if (flags & BTRFS_EXTENT_FLAG_DATA)
1246 return BTRFS_EXTENT_FLAG_DATA;
1247
1248 return -EIO;
1249}
1250
1251/*
1252 * helper function to iterate extent inline refs. ptr must point to a 0 value
1253 * for the first call and may be modified. it is used to track state.
1254 * if more refs exist, 0 is returned and the next call to
1255 * __get_extent_inline_ref must pass the modified ptr parameter to get the
1256 * next ref. after the last ref was processed, 1 is returned.
1257 * returns <0 on error
1258 */
1259static int __get_extent_inline_ref(unsigned long *ptr, struct extent_buffer *eb,
1260 struct btrfs_extent_item *ei, u32 item_size,
1261 struct btrfs_extent_inline_ref **out_eiref,
1262 int *out_type)
1263{
1264 unsigned long end;
1265 u64 flags;
1266 struct btrfs_tree_block_info *info;
1267
1268 if (!*ptr) {
1269 /* first call */
1270 flags = btrfs_extent_flags(eb, ei);
1271 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1272 info = (struct btrfs_tree_block_info *)(ei + 1);
1273 *out_eiref =
1274 (struct btrfs_extent_inline_ref *)(info + 1);
1275 } else {
1276 *out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1);
1277 }
1278 *ptr = (unsigned long)*out_eiref;
1279 if ((void *)*ptr >= (void *)ei + item_size)
1280 return -ENOENT;
1281 }
1282
1283 end = (unsigned long)ei + item_size;
1284 *out_eiref = (struct btrfs_extent_inline_ref *)*ptr;
1285 *out_type = btrfs_extent_inline_ref_type(eb, *out_eiref);
1286
1287 *ptr += btrfs_extent_inline_ref_size(*out_type);
1288 WARN_ON(*ptr > end);
1289 if (*ptr == end)
1290 return 1; /* last */
1291
1292 return 0;
1293}
1294
1295/*
1296 * reads the tree block backref for an extent. tree level and root are returned
1297 * through out_level and out_root. ptr must point to a 0 value for the first
1298 * call and may be modified (see __get_extent_inline_ref comment).
1299 * returns 0 if data was provided, 1 if there was no more data to provide or
1300 * <0 on error.
1301 */
1302int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
1303 struct btrfs_extent_item *ei, u32 item_size,
1304 u64 *out_root, u8 *out_level)
1305{
1306 int ret;
1307 int type;
1308 struct btrfs_tree_block_info *info;
1309 struct btrfs_extent_inline_ref *eiref;
1310
1311 if (*ptr == (unsigned long)-1)
1312 return 1;
1313
1314 while (1) {
1315 ret = __get_extent_inline_ref(ptr, eb, ei, item_size,
1316 &eiref, &type);
1317 if (ret < 0)
1318 return ret;
1319
1320 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1321 type == BTRFS_SHARED_BLOCK_REF_KEY)
1322 break;
1323
1324 if (ret == 1)
1325 return 1;
1326 }
1327
1328 /* we can treat both ref types equally here */
1329 info = (struct btrfs_tree_block_info *)(ei + 1);
1330 *out_root = btrfs_extent_inline_ref_offset(eb, eiref);
1331 *out_level = btrfs_tree_block_level(eb, info);
1332
1333 if (ret == 1)
1334 *ptr = (unsigned long)-1;
1335
1336 return 0;
1337}
1338
1339static int iterate_leaf_refs(struct extent_inode_elem *inode_list,
1340 u64 root, u64 extent_item_objectid,
1341 iterate_extent_inodes_t *iterate, void *ctx)
1342{
1343 struct extent_inode_elem *eie;
1344 int ret = 0;
1345
1346 for (eie = inode_list; eie; eie = eie->next) {
1347 pr_debug("ref for %llu resolved, key (%llu EXTEND_DATA %llu), "
1348 "root %llu\n", extent_item_objectid,
1349 eie->inum, eie->offset, root);
1350 ret = iterate(eie->inum, eie->offset, root, ctx);
1351 if (ret) {
1352 pr_debug("stopping iteration for %llu due to ret=%d\n",
1353 extent_item_objectid, ret);
1354 break;
1355 }
1356 }
1357
1358 return ret;
1359}
1360
1361/*
1362 * calls iterate() for every inode that references the extent identified by
1363 * the given parameters.
1364 * when the iterator function returns a non-zero value, iteration stops.
1365 */
1366int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
1367 u64 extent_item_objectid, u64 extent_item_pos,
1368 int search_commit_root,
1369 iterate_extent_inodes_t *iterate, void *ctx)
1370{
1371 int ret;
1372 struct list_head data_refs = LIST_HEAD_INIT(data_refs);
1373 struct list_head shared_refs = LIST_HEAD_INIT(shared_refs);
1374 struct btrfs_trans_handle *trans;
1375 struct ulist *refs = NULL;
1376 struct ulist *roots = NULL;
1377 struct ulist_node *ref_node = NULL;
1378 struct ulist_node *root_node = NULL;
1379 struct seq_list seq_elem = {};
1380 struct seq_list tree_mod_seq_elem = {};
1381 struct ulist_iterator ref_uiter;
1382 struct ulist_iterator root_uiter;
1383 struct btrfs_delayed_ref_root *delayed_refs = NULL;
1384
1385 pr_debug("resolving all inodes for extent %llu\n",
1386 extent_item_objectid);
1387
1388 if (search_commit_root) {
1389 trans = BTRFS_BACKREF_SEARCH_COMMIT_ROOT;
1390 } else {
1391 trans = btrfs_join_transaction(fs_info->extent_root);
1392 if (IS_ERR(trans))
1393 return PTR_ERR(trans);
1394
1395 delayed_refs = &trans->transaction->delayed_refs;
1396 spin_lock(&delayed_refs->lock);
1397 btrfs_get_delayed_seq(delayed_refs, &seq_elem);
1398 spin_unlock(&delayed_refs->lock);
1399 btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1400 }
1401
1402 ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid,
1403 seq_elem.seq, tree_mod_seq_elem.seq, &refs,
1404 &extent_item_pos);
1405 if (ret)
1406 goto out;
1407
1408 ULIST_ITER_INIT(&ref_uiter);
1409 while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) {
1410 ret = btrfs_find_all_roots(trans, fs_info, ref_node->val,
1411 seq_elem.seq,
1412 tree_mod_seq_elem.seq, &roots);
1413 if (ret)
1414 break;
1415 ULIST_ITER_INIT(&root_uiter);
1416 while (!ret && (root_node = ulist_next(roots, &root_uiter))) {
1417 pr_debug("root %llu references leaf %llu, data list "
1418 "%#lx\n", root_node->val, ref_node->val,
1419 ref_node->aux);
1420 ret = iterate_leaf_refs(
1421 (struct extent_inode_elem *)ref_node->aux,
1422 root_node->val, extent_item_objectid,
1423 iterate, ctx);
1424 }
1425 ulist_free(roots);
1426 roots = NULL;
1427 }
1428
1429 free_leaf_list(refs);
1430 ulist_free(roots);
1431out:
1432 if (!search_commit_root) {
1433 btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1434 btrfs_put_delayed_seq(delayed_refs, &seq_elem);
1435 btrfs_end_transaction(trans, fs_info->extent_root);
1436 }
1437
1438 return ret;
1439}
1440
1441int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
1442 struct btrfs_path *path,
1443 iterate_extent_inodes_t *iterate, void *ctx)
1444{
1445 int ret;
1446 u64 extent_item_pos;
1447 struct btrfs_key found_key;
1448 int search_commit_root = path->search_commit_root;
1449
1450 ret = extent_from_logical(fs_info, logical, path,
1451 &found_key);
1452 btrfs_release_path(path);
1453 if (ret & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1454 ret = -EINVAL;
1455 if (ret < 0)
1456 return ret;
1457
1458 extent_item_pos = logical - found_key.objectid;
1459 ret = iterate_extent_inodes(fs_info, found_key.objectid,
1460 extent_item_pos, search_commit_root,
1461 iterate, ctx);
1462
1463 return ret;
1464}
1465
1466static int iterate_irefs(u64 inum, struct btrfs_root *fs_root,
1467 struct btrfs_path *path,
1468 iterate_irefs_t *iterate, void *ctx)
1469{
1470 int ret = 0;
1471 int slot;
1472 u32 cur;
1473 u32 len;
1474 u32 name_len;
1475 u64 parent = 0;
1476 int found = 0;
1477 struct extent_buffer *eb;
1478 struct btrfs_item *item;
1479 struct btrfs_inode_ref *iref;
1480 struct btrfs_key found_key;
1481
1482 while (!ret) {
1483 path->leave_spinning = 1;
1484 ret = inode_ref_info(inum, parent ? parent+1 : 0, fs_root, path,
1485 &found_key);
1486 if (ret < 0)
1487 break;
1488 if (ret) {
1489 ret = found ? 0 : -ENOENT;
1490 break;
1491 }
1492 ++found;
1493
1494 parent = found_key.offset;
1495 slot = path->slots[0];
1496 eb = path->nodes[0];
1497 /* make sure we can use eb after releasing the path */
1498 atomic_inc(&eb->refs);
1499 btrfs_tree_read_lock(eb);
1500 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1501 btrfs_release_path(path);
1502
1503 item = btrfs_item_nr(eb, slot);
1504 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1505
1506 for (cur = 0; cur < btrfs_item_size(eb, item); cur += len) {
1507 name_len = btrfs_inode_ref_name_len(eb, iref);
1508 /* path must be released before calling iterate()! */
1509 pr_debug("following ref at offset %u for inode %llu in "
1510 "tree %llu\n", cur,
1511 (unsigned long long)found_key.objectid,
1512 (unsigned long long)fs_root->objectid);
1513 ret = iterate(parent, iref, eb, ctx);
1514 if (ret)
1515 break;
1516 len = sizeof(*iref) + name_len;
1517 iref = (struct btrfs_inode_ref *)((char *)iref + len);
1518 }
1519 btrfs_tree_read_unlock_blocking(eb);
1520 free_extent_buffer(eb);
1521 }
1522
1523 btrfs_release_path(path);
1524
1525 return ret;
1526}
1527
1528/*
1529 * returns 0 if the path could be dumped (probably truncated)
1530 * returns <0 in case of an error
1531 */
1532static int inode_to_path(u64 inum, struct btrfs_inode_ref *iref,
1533 struct extent_buffer *eb, void *ctx)
1534{
1535 struct inode_fs_paths *ipath = ctx;
1536 char *fspath;
1537 char *fspath_min;
1538 int i = ipath->fspath->elem_cnt;
1539 const int s_ptr = sizeof(char *);
1540 u32 bytes_left;
1541
1542 bytes_left = ipath->fspath->bytes_left > s_ptr ?
1543 ipath->fspath->bytes_left - s_ptr : 0;
1544
1545 fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
1546 fspath = iref_to_path(ipath->fs_root, ipath->btrfs_path, iref, eb,
1547 inum, fspath_min, bytes_left);
1548 if (IS_ERR(fspath))
1549 return PTR_ERR(fspath);
1550
1551 if (fspath > fspath_min) {
1552 pr_debug("path resolved: %s\n", fspath);
1553 ipath->fspath->val[i] = (u64)(unsigned long)fspath;
1554 ++ipath->fspath->elem_cnt;
1555 ipath->fspath->bytes_left = fspath - fspath_min;
1556 } else {
1557 pr_debug("missed path, not enough space. missing bytes: %lu, "
1558 "constructed so far: %s\n",
1559 (unsigned long)(fspath_min - fspath), fspath_min);
1560 ++ipath->fspath->elem_missed;
1561 ipath->fspath->bytes_missing += fspath_min - fspath;
1562 ipath->fspath->bytes_left = 0;
1563 }
1564
1565 return 0;
1566}
1567
1568/*
1569 * this dumps all file system paths to the inode into the ipath struct, provided
1570 * is has been created large enough. each path is zero-terminated and accessed
1571 * from ipath->fspath->val[i].
1572 * when it returns, there are ipath->fspath->elem_cnt number of paths available
1573 * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the
1574 * number of missed paths in recored in ipath->fspath->elem_missed, otherwise,
1575 * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
1576 * have been needed to return all paths.
1577 */
1578int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
1579{
1580 return iterate_irefs(inum, ipath->fs_root, ipath->btrfs_path,
1581 inode_to_path, ipath);
1582}
1583
1584struct btrfs_data_container *init_data_container(u32 total_bytes)
1585{
1586 struct btrfs_data_container *data;
1587 size_t alloc_bytes;
1588
1589 alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
1590 data = kmalloc(alloc_bytes, GFP_NOFS);
1591 if (!data)
1592 return ERR_PTR(-ENOMEM);
1593
1594 if (total_bytes >= sizeof(*data)) {
1595 data->bytes_left = total_bytes - sizeof(*data);
1596 data->bytes_missing = 0;
1597 } else {
1598 data->bytes_missing = sizeof(*data) - total_bytes;
1599 data->bytes_left = 0;
1600 }
1601
1602 data->elem_cnt = 0;
1603 data->elem_missed = 0;
1604
1605 return data;
1606}
1607
1608/*
1609 * allocates space to return multiple file system paths for an inode.
1610 * total_bytes to allocate are passed, note that space usable for actual path
1611 * information will be total_bytes - sizeof(struct inode_fs_paths).
1612 * the returned pointer must be freed with free_ipath() in the end.
1613 */
1614struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
1615 struct btrfs_path *path)
1616{
1617 struct inode_fs_paths *ifp;
1618 struct btrfs_data_container *fspath;
1619
1620 fspath = init_data_container(total_bytes);
1621 if (IS_ERR(fspath))
1622 return (void *)fspath;
1623
1624 ifp = kmalloc(sizeof(*ifp), GFP_NOFS);
1625 if (!ifp) {
1626 kfree(fspath);
1627 return ERR_PTR(-ENOMEM);
1628 }
1629
1630 ifp->btrfs_path = path;
1631 ifp->fspath = fspath;
1632 ifp->fs_root = fs_root;
1633
1634 return ifp;
1635}
1636
1637void free_ipath(struct inode_fs_paths *ipath)
1638{
1639 if (!ipath)
1640 return;
1641 kfree(ipath->fspath);
1642 kfree(ipath);
1643}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 2011 STRATO. All rights reserved.
4 */
5
6#include <linux/mm.h>
7#include <linux/rbtree.h>
8#include <trace/events/btrfs.h>
9#include "ctree.h"
10#include "disk-io.h"
11#include "backref.h"
12#include "ulist.h"
13#include "transaction.h"
14#include "delayed-ref.h"
15#include "locking.h"
16#include "misc.h"
17#include "tree-mod-log.h"
18
19/* Just an arbitrary number so we can be sure this happened */
20#define BACKREF_FOUND_SHARED 6
21
22struct extent_inode_elem {
23 u64 inum;
24 u64 offset;
25 struct extent_inode_elem *next;
26};
27
28static int check_extent_in_eb(const struct btrfs_key *key,
29 const struct extent_buffer *eb,
30 const struct btrfs_file_extent_item *fi,
31 u64 extent_item_pos,
32 struct extent_inode_elem **eie,
33 bool ignore_offset)
34{
35 u64 offset = 0;
36 struct extent_inode_elem *e;
37
38 if (!ignore_offset &&
39 !btrfs_file_extent_compression(eb, fi) &&
40 !btrfs_file_extent_encryption(eb, fi) &&
41 !btrfs_file_extent_other_encoding(eb, fi)) {
42 u64 data_offset;
43 u64 data_len;
44
45 data_offset = btrfs_file_extent_offset(eb, fi);
46 data_len = btrfs_file_extent_num_bytes(eb, fi);
47
48 if (extent_item_pos < data_offset ||
49 extent_item_pos >= data_offset + data_len)
50 return 1;
51 offset = extent_item_pos - data_offset;
52 }
53
54 e = kmalloc(sizeof(*e), GFP_NOFS);
55 if (!e)
56 return -ENOMEM;
57
58 e->next = *eie;
59 e->inum = key->objectid;
60 e->offset = key->offset + offset;
61 *eie = e;
62
63 return 0;
64}
65
66static void free_inode_elem_list(struct extent_inode_elem *eie)
67{
68 struct extent_inode_elem *eie_next;
69
70 for (; eie; eie = eie_next) {
71 eie_next = eie->next;
72 kfree(eie);
73 }
74}
75
76static int find_extent_in_eb(const struct extent_buffer *eb,
77 u64 wanted_disk_byte, u64 extent_item_pos,
78 struct extent_inode_elem **eie,
79 bool ignore_offset)
80{
81 u64 disk_byte;
82 struct btrfs_key key;
83 struct btrfs_file_extent_item *fi;
84 int slot;
85 int nritems;
86 int extent_type;
87 int ret;
88
89 /*
90 * from the shared data ref, we only have the leaf but we need
91 * the key. thus, we must look into all items and see that we
92 * find one (some) with a reference to our extent item.
93 */
94 nritems = btrfs_header_nritems(eb);
95 for (slot = 0; slot < nritems; ++slot) {
96 btrfs_item_key_to_cpu(eb, &key, slot);
97 if (key.type != BTRFS_EXTENT_DATA_KEY)
98 continue;
99 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
100 extent_type = btrfs_file_extent_type(eb, fi);
101 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
102 continue;
103 /* don't skip BTRFS_FILE_EXTENT_PREALLOC, we can handle that */
104 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
105 if (disk_byte != wanted_disk_byte)
106 continue;
107
108 ret = check_extent_in_eb(&key, eb, fi, extent_item_pos, eie, ignore_offset);
109 if (ret < 0)
110 return ret;
111 }
112
113 return 0;
114}
115
116struct preftree {
117 struct rb_root_cached root;
118 unsigned int count;
119};
120
121#define PREFTREE_INIT { .root = RB_ROOT_CACHED, .count = 0 }
122
123struct preftrees {
124 struct preftree direct; /* BTRFS_SHARED_[DATA|BLOCK]_REF_KEY */
125 struct preftree indirect; /* BTRFS_[TREE_BLOCK|EXTENT_DATA]_REF_KEY */
126 struct preftree indirect_missing_keys;
127};
128
129/*
130 * Checks for a shared extent during backref search.
131 *
132 * The share_count tracks prelim_refs (direct and indirect) having a
133 * ref->count >0:
134 * - incremented when a ref->count transitions to >0
135 * - decremented when a ref->count transitions to <1
136 */
137struct share_check {
138 u64 root_objectid;
139 u64 inum;
140 int share_count;
141};
142
143static inline int extent_is_shared(struct share_check *sc)
144{
145 return (sc && sc->share_count > 1) ? BACKREF_FOUND_SHARED : 0;
146}
147
148static struct kmem_cache *btrfs_prelim_ref_cache;
149
150int __init btrfs_prelim_ref_init(void)
151{
152 btrfs_prelim_ref_cache = kmem_cache_create("btrfs_prelim_ref",
153 sizeof(struct prelim_ref),
154 0,
155 SLAB_MEM_SPREAD,
156 NULL);
157 if (!btrfs_prelim_ref_cache)
158 return -ENOMEM;
159 return 0;
160}
161
162void __cold btrfs_prelim_ref_exit(void)
163{
164 kmem_cache_destroy(btrfs_prelim_ref_cache);
165}
166
167static void free_pref(struct prelim_ref *ref)
168{
169 kmem_cache_free(btrfs_prelim_ref_cache, ref);
170}
171
172/*
173 * Return 0 when both refs are for the same block (and can be merged).
174 * A -1 return indicates ref1 is a 'lower' block than ref2, while 1
175 * indicates a 'higher' block.
176 */
177static int prelim_ref_compare(struct prelim_ref *ref1,
178 struct prelim_ref *ref2)
179{
180 if (ref1->level < ref2->level)
181 return -1;
182 if (ref1->level > ref2->level)
183 return 1;
184 if (ref1->root_id < ref2->root_id)
185 return -1;
186 if (ref1->root_id > ref2->root_id)
187 return 1;
188 if (ref1->key_for_search.type < ref2->key_for_search.type)
189 return -1;
190 if (ref1->key_for_search.type > ref2->key_for_search.type)
191 return 1;
192 if (ref1->key_for_search.objectid < ref2->key_for_search.objectid)
193 return -1;
194 if (ref1->key_for_search.objectid > ref2->key_for_search.objectid)
195 return 1;
196 if (ref1->key_for_search.offset < ref2->key_for_search.offset)
197 return -1;
198 if (ref1->key_for_search.offset > ref2->key_for_search.offset)
199 return 1;
200 if (ref1->parent < ref2->parent)
201 return -1;
202 if (ref1->parent > ref2->parent)
203 return 1;
204
205 return 0;
206}
207
208static void update_share_count(struct share_check *sc, int oldcount,
209 int newcount)
210{
211 if ((!sc) || (oldcount == 0 && newcount < 1))
212 return;
213
214 if (oldcount > 0 && newcount < 1)
215 sc->share_count--;
216 else if (oldcount < 1 && newcount > 0)
217 sc->share_count++;
218}
219
220/*
221 * Add @newref to the @root rbtree, merging identical refs.
222 *
223 * Callers should assume that newref has been freed after calling.
224 */
225static void prelim_ref_insert(const struct btrfs_fs_info *fs_info,
226 struct preftree *preftree,
227 struct prelim_ref *newref,
228 struct share_check *sc)
229{
230 struct rb_root_cached *root;
231 struct rb_node **p;
232 struct rb_node *parent = NULL;
233 struct prelim_ref *ref;
234 int result;
235 bool leftmost = true;
236
237 root = &preftree->root;
238 p = &root->rb_root.rb_node;
239
240 while (*p) {
241 parent = *p;
242 ref = rb_entry(parent, struct prelim_ref, rbnode);
243 result = prelim_ref_compare(ref, newref);
244 if (result < 0) {
245 p = &(*p)->rb_left;
246 } else if (result > 0) {
247 p = &(*p)->rb_right;
248 leftmost = false;
249 } else {
250 /* Identical refs, merge them and free @newref */
251 struct extent_inode_elem *eie = ref->inode_list;
252
253 while (eie && eie->next)
254 eie = eie->next;
255
256 if (!eie)
257 ref->inode_list = newref->inode_list;
258 else
259 eie->next = newref->inode_list;
260 trace_btrfs_prelim_ref_merge(fs_info, ref, newref,
261 preftree->count);
262 /*
263 * A delayed ref can have newref->count < 0.
264 * The ref->count is updated to follow any
265 * BTRFS_[ADD|DROP]_DELAYED_REF actions.
266 */
267 update_share_count(sc, ref->count,
268 ref->count + newref->count);
269 ref->count += newref->count;
270 free_pref(newref);
271 return;
272 }
273 }
274
275 update_share_count(sc, 0, newref->count);
276 preftree->count++;
277 trace_btrfs_prelim_ref_insert(fs_info, newref, NULL, preftree->count);
278 rb_link_node(&newref->rbnode, parent, p);
279 rb_insert_color_cached(&newref->rbnode, root, leftmost);
280}
281
282/*
283 * Release the entire tree. We don't care about internal consistency so
284 * just free everything and then reset the tree root.
285 */
286static void prelim_release(struct preftree *preftree)
287{
288 struct prelim_ref *ref, *next_ref;
289
290 rbtree_postorder_for_each_entry_safe(ref, next_ref,
291 &preftree->root.rb_root, rbnode)
292 free_pref(ref);
293
294 preftree->root = RB_ROOT_CACHED;
295 preftree->count = 0;
296}
297
298/*
299 * the rules for all callers of this function are:
300 * - obtaining the parent is the goal
301 * - if you add a key, you must know that it is a correct key
302 * - if you cannot add the parent or a correct key, then we will look into the
303 * block later to set a correct key
304 *
305 * delayed refs
306 * ============
307 * backref type | shared | indirect | shared | indirect
308 * information | tree | tree | data | data
309 * --------------------+--------+----------+--------+----------
310 * parent logical | y | - | - | -
311 * key to resolve | - | y | y | y
312 * tree block logical | - | - | - | -
313 * root for resolving | y | y | y | y
314 *
315 * - column 1: we've the parent -> done
316 * - column 2, 3, 4: we use the key to find the parent
317 *
318 * on disk refs (inline or keyed)
319 * ==============================
320 * backref type | shared | indirect | shared | indirect
321 * information | tree | tree | data | data
322 * --------------------+--------+----------+--------+----------
323 * parent logical | y | - | y | -
324 * key to resolve | - | - | - | y
325 * tree block logical | y | y | y | y
326 * root for resolving | - | y | y | y
327 *
328 * - column 1, 3: we've the parent -> done
329 * - column 2: we take the first key from the block to find the parent
330 * (see add_missing_keys)
331 * - column 4: we use the key to find the parent
332 *
333 * additional information that's available but not required to find the parent
334 * block might help in merging entries to gain some speed.
335 */
336static int add_prelim_ref(const struct btrfs_fs_info *fs_info,
337 struct preftree *preftree, u64 root_id,
338 const struct btrfs_key *key, int level, u64 parent,
339 u64 wanted_disk_byte, int count,
340 struct share_check *sc, gfp_t gfp_mask)
341{
342 struct prelim_ref *ref;
343
344 if (root_id == BTRFS_DATA_RELOC_TREE_OBJECTID)
345 return 0;
346
347 ref = kmem_cache_alloc(btrfs_prelim_ref_cache, gfp_mask);
348 if (!ref)
349 return -ENOMEM;
350
351 ref->root_id = root_id;
352 if (key)
353 ref->key_for_search = *key;
354 else
355 memset(&ref->key_for_search, 0, sizeof(ref->key_for_search));
356
357 ref->inode_list = NULL;
358 ref->level = level;
359 ref->count = count;
360 ref->parent = parent;
361 ref->wanted_disk_byte = wanted_disk_byte;
362 prelim_ref_insert(fs_info, preftree, ref, sc);
363 return extent_is_shared(sc);
364}
365
366/* direct refs use root == 0, key == NULL */
367static int add_direct_ref(const struct btrfs_fs_info *fs_info,
368 struct preftrees *preftrees, int level, u64 parent,
369 u64 wanted_disk_byte, int count,
370 struct share_check *sc, gfp_t gfp_mask)
371{
372 return add_prelim_ref(fs_info, &preftrees->direct, 0, NULL, level,
373 parent, wanted_disk_byte, count, sc, gfp_mask);
374}
375
376/* indirect refs use parent == 0 */
377static int add_indirect_ref(const struct btrfs_fs_info *fs_info,
378 struct preftrees *preftrees, u64 root_id,
379 const struct btrfs_key *key, int level,
380 u64 wanted_disk_byte, int count,
381 struct share_check *sc, gfp_t gfp_mask)
382{
383 struct preftree *tree = &preftrees->indirect;
384
385 if (!key)
386 tree = &preftrees->indirect_missing_keys;
387 return add_prelim_ref(fs_info, tree, root_id, key, level, 0,
388 wanted_disk_byte, count, sc, gfp_mask);
389}
390
391static int is_shared_data_backref(struct preftrees *preftrees, u64 bytenr)
392{
393 struct rb_node **p = &preftrees->direct.root.rb_root.rb_node;
394 struct rb_node *parent = NULL;
395 struct prelim_ref *ref = NULL;
396 struct prelim_ref target = {};
397 int result;
398
399 target.parent = bytenr;
400
401 while (*p) {
402 parent = *p;
403 ref = rb_entry(parent, struct prelim_ref, rbnode);
404 result = prelim_ref_compare(ref, &target);
405
406 if (result < 0)
407 p = &(*p)->rb_left;
408 else if (result > 0)
409 p = &(*p)->rb_right;
410 else
411 return 1;
412 }
413 return 0;
414}
415
416static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path,
417 struct ulist *parents,
418 struct preftrees *preftrees, struct prelim_ref *ref,
419 int level, u64 time_seq, const u64 *extent_item_pos,
420 bool ignore_offset)
421{
422 int ret = 0;
423 int slot;
424 struct extent_buffer *eb;
425 struct btrfs_key key;
426 struct btrfs_key *key_for_search = &ref->key_for_search;
427 struct btrfs_file_extent_item *fi;
428 struct extent_inode_elem *eie = NULL, *old = NULL;
429 u64 disk_byte;
430 u64 wanted_disk_byte = ref->wanted_disk_byte;
431 u64 count = 0;
432 u64 data_offset;
433
434 if (level != 0) {
435 eb = path->nodes[level];
436 ret = ulist_add(parents, eb->start, 0, GFP_NOFS);
437 if (ret < 0)
438 return ret;
439 return 0;
440 }
441
442 /*
443 * 1. We normally enter this function with the path already pointing to
444 * the first item to check. But sometimes, we may enter it with
445 * slot == nritems.
446 * 2. We are searching for normal backref but bytenr of this leaf
447 * matches shared data backref
448 * 3. The leaf owner is not equal to the root we are searching
449 *
450 * For these cases, go to the next leaf before we continue.
451 */
452 eb = path->nodes[0];
453 if (path->slots[0] >= btrfs_header_nritems(eb) ||
454 is_shared_data_backref(preftrees, eb->start) ||
455 ref->root_id != btrfs_header_owner(eb)) {
456 if (time_seq == BTRFS_SEQ_LAST)
457 ret = btrfs_next_leaf(root, path);
458 else
459 ret = btrfs_next_old_leaf(root, path, time_seq);
460 }
461
462 while (!ret && count < ref->count) {
463 eb = path->nodes[0];
464 slot = path->slots[0];
465
466 btrfs_item_key_to_cpu(eb, &key, slot);
467
468 if (key.objectid != key_for_search->objectid ||
469 key.type != BTRFS_EXTENT_DATA_KEY)
470 break;
471
472 /*
473 * We are searching for normal backref but bytenr of this leaf
474 * matches shared data backref, OR
475 * the leaf owner is not equal to the root we are searching for
476 */
477 if (slot == 0 &&
478 (is_shared_data_backref(preftrees, eb->start) ||
479 ref->root_id != btrfs_header_owner(eb))) {
480 if (time_seq == BTRFS_SEQ_LAST)
481 ret = btrfs_next_leaf(root, path);
482 else
483 ret = btrfs_next_old_leaf(root, path, time_seq);
484 continue;
485 }
486 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
487 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
488 data_offset = btrfs_file_extent_offset(eb, fi);
489
490 if (disk_byte == wanted_disk_byte) {
491 eie = NULL;
492 old = NULL;
493 if (ref->key_for_search.offset == key.offset - data_offset)
494 count++;
495 else
496 goto next;
497 if (extent_item_pos) {
498 ret = check_extent_in_eb(&key, eb, fi,
499 *extent_item_pos,
500 &eie, ignore_offset);
501 if (ret < 0)
502 break;
503 }
504 if (ret > 0)
505 goto next;
506 ret = ulist_add_merge_ptr(parents, eb->start,
507 eie, (void **)&old, GFP_NOFS);
508 if (ret < 0)
509 break;
510 if (!ret && extent_item_pos) {
511 while (old->next)
512 old = old->next;
513 old->next = eie;
514 }
515 eie = NULL;
516 }
517next:
518 if (time_seq == BTRFS_SEQ_LAST)
519 ret = btrfs_next_item(root, path);
520 else
521 ret = btrfs_next_old_item(root, path, time_seq);
522 }
523
524 if (ret > 0)
525 ret = 0;
526 else if (ret < 0)
527 free_inode_elem_list(eie);
528 return ret;
529}
530
531/*
532 * resolve an indirect backref in the form (root_id, key, level)
533 * to a logical address
534 */
535static int resolve_indirect_ref(struct btrfs_fs_info *fs_info,
536 struct btrfs_path *path, u64 time_seq,
537 struct preftrees *preftrees,
538 struct prelim_ref *ref, struct ulist *parents,
539 const u64 *extent_item_pos, bool ignore_offset)
540{
541 struct btrfs_root *root;
542 struct extent_buffer *eb;
543 int ret = 0;
544 int root_level;
545 int level = ref->level;
546 struct btrfs_key search_key = ref->key_for_search;
547
548 /*
549 * If we're search_commit_root we could possibly be holding locks on
550 * other tree nodes. This happens when qgroups does backref walks when
551 * adding new delayed refs. To deal with this we need to look in cache
552 * for the root, and if we don't find it then we need to search the
553 * tree_root's commit root, thus the btrfs_get_fs_root_commit_root usage
554 * here.
555 */
556 if (path->search_commit_root)
557 root = btrfs_get_fs_root_commit_root(fs_info, path, ref->root_id);
558 else
559 root = btrfs_get_fs_root(fs_info, ref->root_id, false);
560 if (IS_ERR(root)) {
561 ret = PTR_ERR(root);
562 goto out_free;
563 }
564
565 if (!path->search_commit_root &&
566 test_bit(BTRFS_ROOT_DELETING, &root->state)) {
567 ret = -ENOENT;
568 goto out;
569 }
570
571 if (btrfs_is_testing(fs_info)) {
572 ret = -ENOENT;
573 goto out;
574 }
575
576 if (path->search_commit_root)
577 root_level = btrfs_header_level(root->commit_root);
578 else if (time_seq == BTRFS_SEQ_LAST)
579 root_level = btrfs_header_level(root->node);
580 else
581 root_level = btrfs_old_root_level(root, time_seq);
582
583 if (root_level + 1 == level)
584 goto out;
585
586 /*
587 * We can often find data backrefs with an offset that is too large
588 * (>= LLONG_MAX, maximum allowed file offset) due to underflows when
589 * subtracting a file's offset with the data offset of its
590 * corresponding extent data item. This can happen for example in the
591 * clone ioctl.
592 *
593 * So if we detect such case we set the search key's offset to zero to
594 * make sure we will find the matching file extent item at
595 * add_all_parents(), otherwise we will miss it because the offset
596 * taken form the backref is much larger then the offset of the file
597 * extent item. This can make us scan a very large number of file
598 * extent items, but at least it will not make us miss any.
599 *
600 * This is an ugly workaround for a behaviour that should have never
601 * existed, but it does and a fix for the clone ioctl would touch a lot
602 * of places, cause backwards incompatibility and would not fix the
603 * problem for extents cloned with older kernels.
604 */
605 if (search_key.type == BTRFS_EXTENT_DATA_KEY &&
606 search_key.offset >= LLONG_MAX)
607 search_key.offset = 0;
608 path->lowest_level = level;
609 if (time_seq == BTRFS_SEQ_LAST)
610 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
611 else
612 ret = btrfs_search_old_slot(root, &search_key, path, time_seq);
613
614 btrfs_debug(fs_info,
615 "search slot in root %llu (level %d, ref count %d) returned %d for key (%llu %u %llu)",
616 ref->root_id, level, ref->count, ret,
617 ref->key_for_search.objectid, ref->key_for_search.type,
618 ref->key_for_search.offset);
619 if (ret < 0)
620 goto out;
621
622 eb = path->nodes[level];
623 while (!eb) {
624 if (WARN_ON(!level)) {
625 ret = 1;
626 goto out;
627 }
628 level--;
629 eb = path->nodes[level];
630 }
631
632 ret = add_all_parents(root, path, parents, preftrees, ref, level,
633 time_seq, extent_item_pos, ignore_offset);
634out:
635 btrfs_put_root(root);
636out_free:
637 path->lowest_level = 0;
638 btrfs_release_path(path);
639 return ret;
640}
641
642static struct extent_inode_elem *
643unode_aux_to_inode_list(struct ulist_node *node)
644{
645 if (!node)
646 return NULL;
647 return (struct extent_inode_elem *)(uintptr_t)node->aux;
648}
649
650/*
651 * We maintain three separate rbtrees: one for direct refs, one for
652 * indirect refs which have a key, and one for indirect refs which do not
653 * have a key. Each tree does merge on insertion.
654 *
655 * Once all of the references are located, we iterate over the tree of
656 * indirect refs with missing keys. An appropriate key is located and
657 * the ref is moved onto the tree for indirect refs. After all missing
658 * keys are thus located, we iterate over the indirect ref tree, resolve
659 * each reference, and then insert the resolved reference onto the
660 * direct tree (merging there too).
661 *
662 * New backrefs (i.e., for parent nodes) are added to the appropriate
663 * rbtree as they are encountered. The new backrefs are subsequently
664 * resolved as above.
665 */
666static int resolve_indirect_refs(struct btrfs_fs_info *fs_info,
667 struct btrfs_path *path, u64 time_seq,
668 struct preftrees *preftrees,
669 const u64 *extent_item_pos,
670 struct share_check *sc, bool ignore_offset)
671{
672 int err;
673 int ret = 0;
674 struct ulist *parents;
675 struct ulist_node *node;
676 struct ulist_iterator uiter;
677 struct rb_node *rnode;
678
679 parents = ulist_alloc(GFP_NOFS);
680 if (!parents)
681 return -ENOMEM;
682
683 /*
684 * We could trade memory usage for performance here by iterating
685 * the tree, allocating new refs for each insertion, and then
686 * freeing the entire indirect tree when we're done. In some test
687 * cases, the tree can grow quite large (~200k objects).
688 */
689 while ((rnode = rb_first_cached(&preftrees->indirect.root))) {
690 struct prelim_ref *ref;
691
692 ref = rb_entry(rnode, struct prelim_ref, rbnode);
693 if (WARN(ref->parent,
694 "BUG: direct ref found in indirect tree")) {
695 ret = -EINVAL;
696 goto out;
697 }
698
699 rb_erase_cached(&ref->rbnode, &preftrees->indirect.root);
700 preftrees->indirect.count--;
701
702 if (ref->count == 0) {
703 free_pref(ref);
704 continue;
705 }
706
707 if (sc && sc->root_objectid &&
708 ref->root_id != sc->root_objectid) {
709 free_pref(ref);
710 ret = BACKREF_FOUND_SHARED;
711 goto out;
712 }
713 err = resolve_indirect_ref(fs_info, path, time_seq, preftrees,
714 ref, parents, extent_item_pos,
715 ignore_offset);
716 /*
717 * we can only tolerate ENOENT,otherwise,we should catch error
718 * and return directly.
719 */
720 if (err == -ENOENT) {
721 prelim_ref_insert(fs_info, &preftrees->direct, ref,
722 NULL);
723 continue;
724 } else if (err) {
725 free_pref(ref);
726 ret = err;
727 goto out;
728 }
729
730 /* we put the first parent into the ref at hand */
731 ULIST_ITER_INIT(&uiter);
732 node = ulist_next(parents, &uiter);
733 ref->parent = node ? node->val : 0;
734 ref->inode_list = unode_aux_to_inode_list(node);
735
736 /* Add a prelim_ref(s) for any other parent(s). */
737 while ((node = ulist_next(parents, &uiter))) {
738 struct prelim_ref *new_ref;
739
740 new_ref = kmem_cache_alloc(btrfs_prelim_ref_cache,
741 GFP_NOFS);
742 if (!new_ref) {
743 free_pref(ref);
744 ret = -ENOMEM;
745 goto out;
746 }
747 memcpy(new_ref, ref, sizeof(*ref));
748 new_ref->parent = node->val;
749 new_ref->inode_list = unode_aux_to_inode_list(node);
750 prelim_ref_insert(fs_info, &preftrees->direct,
751 new_ref, NULL);
752 }
753
754 /*
755 * Now it's a direct ref, put it in the direct tree. We must
756 * do this last because the ref could be merged/freed here.
757 */
758 prelim_ref_insert(fs_info, &preftrees->direct, ref, NULL);
759
760 ulist_reinit(parents);
761 cond_resched();
762 }
763out:
764 ulist_free(parents);
765 return ret;
766}
767
768/*
769 * read tree blocks and add keys where required.
770 */
771static int add_missing_keys(struct btrfs_fs_info *fs_info,
772 struct preftrees *preftrees, bool lock)
773{
774 struct prelim_ref *ref;
775 struct extent_buffer *eb;
776 struct preftree *tree = &preftrees->indirect_missing_keys;
777 struct rb_node *node;
778
779 while ((node = rb_first_cached(&tree->root))) {
780 ref = rb_entry(node, struct prelim_ref, rbnode);
781 rb_erase_cached(node, &tree->root);
782
783 BUG_ON(ref->parent); /* should not be a direct ref */
784 BUG_ON(ref->key_for_search.type);
785 BUG_ON(!ref->wanted_disk_byte);
786
787 eb = read_tree_block(fs_info, ref->wanted_disk_byte,
788 ref->root_id, 0, ref->level - 1, NULL);
789 if (IS_ERR(eb)) {
790 free_pref(ref);
791 return PTR_ERR(eb);
792 } else if (!extent_buffer_uptodate(eb)) {
793 free_pref(ref);
794 free_extent_buffer(eb);
795 return -EIO;
796 }
797 if (lock)
798 btrfs_tree_read_lock(eb);
799 if (btrfs_header_level(eb) == 0)
800 btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
801 else
802 btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
803 if (lock)
804 btrfs_tree_read_unlock(eb);
805 free_extent_buffer(eb);
806 prelim_ref_insert(fs_info, &preftrees->indirect, ref, NULL);
807 cond_resched();
808 }
809 return 0;
810}
811
812/*
813 * add all currently queued delayed refs from this head whose seq nr is
814 * smaller or equal that seq to the list
815 */
816static int add_delayed_refs(const struct btrfs_fs_info *fs_info,
817 struct btrfs_delayed_ref_head *head, u64 seq,
818 struct preftrees *preftrees, struct share_check *sc)
819{
820 struct btrfs_delayed_ref_node *node;
821 struct btrfs_delayed_extent_op *extent_op = head->extent_op;
822 struct btrfs_key key;
823 struct btrfs_key tmp_op_key;
824 struct rb_node *n;
825 int count;
826 int ret = 0;
827
828 if (extent_op && extent_op->update_key)
829 btrfs_disk_key_to_cpu(&tmp_op_key, &extent_op->key);
830
831 spin_lock(&head->lock);
832 for (n = rb_first_cached(&head->ref_tree); n; n = rb_next(n)) {
833 node = rb_entry(n, struct btrfs_delayed_ref_node,
834 ref_node);
835 if (node->seq > seq)
836 continue;
837
838 switch (node->action) {
839 case BTRFS_ADD_DELAYED_EXTENT:
840 case BTRFS_UPDATE_DELAYED_HEAD:
841 WARN_ON(1);
842 continue;
843 case BTRFS_ADD_DELAYED_REF:
844 count = node->ref_mod;
845 break;
846 case BTRFS_DROP_DELAYED_REF:
847 count = node->ref_mod * -1;
848 break;
849 default:
850 BUG();
851 }
852 switch (node->type) {
853 case BTRFS_TREE_BLOCK_REF_KEY: {
854 /* NORMAL INDIRECT METADATA backref */
855 struct btrfs_delayed_tree_ref *ref;
856
857 ref = btrfs_delayed_node_to_tree_ref(node);
858 ret = add_indirect_ref(fs_info, preftrees, ref->root,
859 &tmp_op_key, ref->level + 1,
860 node->bytenr, count, sc,
861 GFP_ATOMIC);
862 break;
863 }
864 case BTRFS_SHARED_BLOCK_REF_KEY: {
865 /* SHARED DIRECT METADATA backref */
866 struct btrfs_delayed_tree_ref *ref;
867
868 ref = btrfs_delayed_node_to_tree_ref(node);
869
870 ret = add_direct_ref(fs_info, preftrees, ref->level + 1,
871 ref->parent, node->bytenr, count,
872 sc, GFP_ATOMIC);
873 break;
874 }
875 case BTRFS_EXTENT_DATA_REF_KEY: {
876 /* NORMAL INDIRECT DATA backref */
877 struct btrfs_delayed_data_ref *ref;
878 ref = btrfs_delayed_node_to_data_ref(node);
879
880 key.objectid = ref->objectid;
881 key.type = BTRFS_EXTENT_DATA_KEY;
882 key.offset = ref->offset;
883
884 /*
885 * Found a inum that doesn't match our known inum, we
886 * know it's shared.
887 */
888 if (sc && sc->inum && ref->objectid != sc->inum) {
889 ret = BACKREF_FOUND_SHARED;
890 goto out;
891 }
892
893 ret = add_indirect_ref(fs_info, preftrees, ref->root,
894 &key, 0, node->bytenr, count, sc,
895 GFP_ATOMIC);
896 break;
897 }
898 case BTRFS_SHARED_DATA_REF_KEY: {
899 /* SHARED DIRECT FULL backref */
900 struct btrfs_delayed_data_ref *ref;
901
902 ref = btrfs_delayed_node_to_data_ref(node);
903
904 ret = add_direct_ref(fs_info, preftrees, 0, ref->parent,
905 node->bytenr, count, sc,
906 GFP_ATOMIC);
907 break;
908 }
909 default:
910 WARN_ON(1);
911 }
912 /*
913 * We must ignore BACKREF_FOUND_SHARED until all delayed
914 * refs have been checked.
915 */
916 if (ret && (ret != BACKREF_FOUND_SHARED))
917 break;
918 }
919 if (!ret)
920 ret = extent_is_shared(sc);
921out:
922 spin_unlock(&head->lock);
923 return ret;
924}
925
926/*
927 * add all inline backrefs for bytenr to the list
928 *
929 * Returns 0 on success, <0 on error, or BACKREF_FOUND_SHARED.
930 */
931static int add_inline_refs(const struct btrfs_fs_info *fs_info,
932 struct btrfs_path *path, u64 bytenr,
933 int *info_level, struct preftrees *preftrees,
934 struct share_check *sc)
935{
936 int ret = 0;
937 int slot;
938 struct extent_buffer *leaf;
939 struct btrfs_key key;
940 struct btrfs_key found_key;
941 unsigned long ptr;
942 unsigned long end;
943 struct btrfs_extent_item *ei;
944 u64 flags;
945 u64 item_size;
946
947 /*
948 * enumerate all inline refs
949 */
950 leaf = path->nodes[0];
951 slot = path->slots[0];
952
953 item_size = btrfs_item_size_nr(leaf, slot);
954 BUG_ON(item_size < sizeof(*ei));
955
956 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
957 flags = btrfs_extent_flags(leaf, ei);
958 btrfs_item_key_to_cpu(leaf, &found_key, slot);
959
960 ptr = (unsigned long)(ei + 1);
961 end = (unsigned long)ei + item_size;
962
963 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
964 flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
965 struct btrfs_tree_block_info *info;
966
967 info = (struct btrfs_tree_block_info *)ptr;
968 *info_level = btrfs_tree_block_level(leaf, info);
969 ptr += sizeof(struct btrfs_tree_block_info);
970 BUG_ON(ptr > end);
971 } else if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
972 *info_level = found_key.offset;
973 } else {
974 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
975 }
976
977 while (ptr < end) {
978 struct btrfs_extent_inline_ref *iref;
979 u64 offset;
980 int type;
981
982 iref = (struct btrfs_extent_inline_ref *)ptr;
983 type = btrfs_get_extent_inline_ref_type(leaf, iref,
984 BTRFS_REF_TYPE_ANY);
985 if (type == BTRFS_REF_TYPE_INVALID)
986 return -EUCLEAN;
987
988 offset = btrfs_extent_inline_ref_offset(leaf, iref);
989
990 switch (type) {
991 case BTRFS_SHARED_BLOCK_REF_KEY:
992 ret = add_direct_ref(fs_info, preftrees,
993 *info_level + 1, offset,
994 bytenr, 1, NULL, GFP_NOFS);
995 break;
996 case BTRFS_SHARED_DATA_REF_KEY: {
997 struct btrfs_shared_data_ref *sdref;
998 int count;
999
1000 sdref = (struct btrfs_shared_data_ref *)(iref + 1);
1001 count = btrfs_shared_data_ref_count(leaf, sdref);
1002
1003 ret = add_direct_ref(fs_info, preftrees, 0, offset,
1004 bytenr, count, sc, GFP_NOFS);
1005 break;
1006 }
1007 case BTRFS_TREE_BLOCK_REF_KEY:
1008 ret = add_indirect_ref(fs_info, preftrees, offset,
1009 NULL, *info_level + 1,
1010 bytenr, 1, NULL, GFP_NOFS);
1011 break;
1012 case BTRFS_EXTENT_DATA_REF_KEY: {
1013 struct btrfs_extent_data_ref *dref;
1014 int count;
1015 u64 root;
1016
1017 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1018 count = btrfs_extent_data_ref_count(leaf, dref);
1019 key.objectid = btrfs_extent_data_ref_objectid(leaf,
1020 dref);
1021 key.type = BTRFS_EXTENT_DATA_KEY;
1022 key.offset = btrfs_extent_data_ref_offset(leaf, dref);
1023
1024 if (sc && sc->inum && key.objectid != sc->inum) {
1025 ret = BACKREF_FOUND_SHARED;
1026 break;
1027 }
1028
1029 root = btrfs_extent_data_ref_root(leaf, dref);
1030
1031 ret = add_indirect_ref(fs_info, preftrees, root,
1032 &key, 0, bytenr, count,
1033 sc, GFP_NOFS);
1034 break;
1035 }
1036 default:
1037 WARN_ON(1);
1038 }
1039 if (ret)
1040 return ret;
1041 ptr += btrfs_extent_inline_ref_size(type);
1042 }
1043
1044 return 0;
1045}
1046
1047/*
1048 * add all non-inline backrefs for bytenr to the list
1049 *
1050 * Returns 0 on success, <0 on error, or BACKREF_FOUND_SHARED.
1051 */
1052static int add_keyed_refs(struct btrfs_fs_info *fs_info,
1053 struct btrfs_path *path, u64 bytenr,
1054 int info_level, struct preftrees *preftrees,
1055 struct share_check *sc)
1056{
1057 struct btrfs_root *extent_root = fs_info->extent_root;
1058 int ret;
1059 int slot;
1060 struct extent_buffer *leaf;
1061 struct btrfs_key key;
1062
1063 while (1) {
1064 ret = btrfs_next_item(extent_root, path);
1065 if (ret < 0)
1066 break;
1067 if (ret) {
1068 ret = 0;
1069 break;
1070 }
1071
1072 slot = path->slots[0];
1073 leaf = path->nodes[0];
1074 btrfs_item_key_to_cpu(leaf, &key, slot);
1075
1076 if (key.objectid != bytenr)
1077 break;
1078 if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
1079 continue;
1080 if (key.type > BTRFS_SHARED_DATA_REF_KEY)
1081 break;
1082
1083 switch (key.type) {
1084 case BTRFS_SHARED_BLOCK_REF_KEY:
1085 /* SHARED DIRECT METADATA backref */
1086 ret = add_direct_ref(fs_info, preftrees,
1087 info_level + 1, key.offset,
1088 bytenr, 1, NULL, GFP_NOFS);
1089 break;
1090 case BTRFS_SHARED_DATA_REF_KEY: {
1091 /* SHARED DIRECT FULL backref */
1092 struct btrfs_shared_data_ref *sdref;
1093 int count;
1094
1095 sdref = btrfs_item_ptr(leaf, slot,
1096 struct btrfs_shared_data_ref);
1097 count = btrfs_shared_data_ref_count(leaf, sdref);
1098 ret = add_direct_ref(fs_info, preftrees, 0,
1099 key.offset, bytenr, count,
1100 sc, GFP_NOFS);
1101 break;
1102 }
1103 case BTRFS_TREE_BLOCK_REF_KEY:
1104 /* NORMAL INDIRECT METADATA backref */
1105 ret = add_indirect_ref(fs_info, preftrees, key.offset,
1106 NULL, info_level + 1, bytenr,
1107 1, NULL, GFP_NOFS);
1108 break;
1109 case BTRFS_EXTENT_DATA_REF_KEY: {
1110 /* NORMAL INDIRECT DATA backref */
1111 struct btrfs_extent_data_ref *dref;
1112 int count;
1113 u64 root;
1114
1115 dref = btrfs_item_ptr(leaf, slot,
1116 struct btrfs_extent_data_ref);
1117 count = btrfs_extent_data_ref_count(leaf, dref);
1118 key.objectid = btrfs_extent_data_ref_objectid(leaf,
1119 dref);
1120 key.type = BTRFS_EXTENT_DATA_KEY;
1121 key.offset = btrfs_extent_data_ref_offset(leaf, dref);
1122
1123 if (sc && sc->inum && key.objectid != sc->inum) {
1124 ret = BACKREF_FOUND_SHARED;
1125 break;
1126 }
1127
1128 root = btrfs_extent_data_ref_root(leaf, dref);
1129 ret = add_indirect_ref(fs_info, preftrees, root,
1130 &key, 0, bytenr, count,
1131 sc, GFP_NOFS);
1132 break;
1133 }
1134 default:
1135 WARN_ON(1);
1136 }
1137 if (ret)
1138 return ret;
1139
1140 }
1141
1142 return ret;
1143}
1144
1145/*
1146 * this adds all existing backrefs (inline backrefs, backrefs and delayed
1147 * refs) for the given bytenr to the refs list, merges duplicates and resolves
1148 * indirect refs to their parent bytenr.
1149 * When roots are found, they're added to the roots list
1150 *
1151 * If time_seq is set to BTRFS_SEQ_LAST, it will not search delayed_refs, and
1152 * behave much like trans == NULL case, the difference only lies in it will not
1153 * commit root.
1154 * The special case is for qgroup to search roots in commit_transaction().
1155 *
1156 * @sc - if !NULL, then immediately return BACKREF_FOUND_SHARED when a
1157 * shared extent is detected.
1158 *
1159 * Otherwise this returns 0 for success and <0 for an error.
1160 *
1161 * If ignore_offset is set to false, only extent refs whose offsets match
1162 * extent_item_pos are returned. If true, every extent ref is returned
1163 * and extent_item_pos is ignored.
1164 *
1165 * FIXME some caching might speed things up
1166 */
1167static int find_parent_nodes(struct btrfs_trans_handle *trans,
1168 struct btrfs_fs_info *fs_info, u64 bytenr,
1169 u64 time_seq, struct ulist *refs,
1170 struct ulist *roots, const u64 *extent_item_pos,
1171 struct share_check *sc, bool ignore_offset)
1172{
1173 struct btrfs_key key;
1174 struct btrfs_path *path;
1175 struct btrfs_delayed_ref_root *delayed_refs = NULL;
1176 struct btrfs_delayed_ref_head *head;
1177 int info_level = 0;
1178 int ret;
1179 struct prelim_ref *ref;
1180 struct rb_node *node;
1181 struct extent_inode_elem *eie = NULL;
1182 struct preftrees preftrees = {
1183 .direct = PREFTREE_INIT,
1184 .indirect = PREFTREE_INIT,
1185 .indirect_missing_keys = PREFTREE_INIT
1186 };
1187
1188 key.objectid = bytenr;
1189 key.offset = (u64)-1;
1190 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1191 key.type = BTRFS_METADATA_ITEM_KEY;
1192 else
1193 key.type = BTRFS_EXTENT_ITEM_KEY;
1194
1195 path = btrfs_alloc_path();
1196 if (!path)
1197 return -ENOMEM;
1198 if (!trans) {
1199 path->search_commit_root = 1;
1200 path->skip_locking = 1;
1201 }
1202
1203 if (time_seq == BTRFS_SEQ_LAST)
1204 path->skip_locking = 1;
1205
1206 /*
1207 * grab both a lock on the path and a lock on the delayed ref head.
1208 * We need both to get a consistent picture of how the refs look
1209 * at a specified point in time
1210 */
1211again:
1212 head = NULL;
1213
1214 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
1215 if (ret < 0)
1216 goto out;
1217 BUG_ON(ret == 0);
1218
1219#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1220 if (trans && likely(trans->type != __TRANS_DUMMY) &&
1221 time_seq != BTRFS_SEQ_LAST) {
1222#else
1223 if (trans && time_seq != BTRFS_SEQ_LAST) {
1224#endif
1225 /*
1226 * look if there are updates for this ref queued and lock the
1227 * head
1228 */
1229 delayed_refs = &trans->transaction->delayed_refs;
1230 spin_lock(&delayed_refs->lock);
1231 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
1232 if (head) {
1233 if (!mutex_trylock(&head->mutex)) {
1234 refcount_inc(&head->refs);
1235 spin_unlock(&delayed_refs->lock);
1236
1237 btrfs_release_path(path);
1238
1239 /*
1240 * Mutex was contended, block until it's
1241 * released and try again
1242 */
1243 mutex_lock(&head->mutex);
1244 mutex_unlock(&head->mutex);
1245 btrfs_put_delayed_ref_head(head);
1246 goto again;
1247 }
1248 spin_unlock(&delayed_refs->lock);
1249 ret = add_delayed_refs(fs_info, head, time_seq,
1250 &preftrees, sc);
1251 mutex_unlock(&head->mutex);
1252 if (ret)
1253 goto out;
1254 } else {
1255 spin_unlock(&delayed_refs->lock);
1256 }
1257 }
1258
1259 if (path->slots[0]) {
1260 struct extent_buffer *leaf;
1261 int slot;
1262
1263 path->slots[0]--;
1264 leaf = path->nodes[0];
1265 slot = path->slots[0];
1266 btrfs_item_key_to_cpu(leaf, &key, slot);
1267 if (key.objectid == bytenr &&
1268 (key.type == BTRFS_EXTENT_ITEM_KEY ||
1269 key.type == BTRFS_METADATA_ITEM_KEY)) {
1270 ret = add_inline_refs(fs_info, path, bytenr,
1271 &info_level, &preftrees, sc);
1272 if (ret)
1273 goto out;
1274 ret = add_keyed_refs(fs_info, path, bytenr, info_level,
1275 &preftrees, sc);
1276 if (ret)
1277 goto out;
1278 }
1279 }
1280
1281 btrfs_release_path(path);
1282
1283 ret = add_missing_keys(fs_info, &preftrees, path->skip_locking == 0);
1284 if (ret)
1285 goto out;
1286
1287 WARN_ON(!RB_EMPTY_ROOT(&preftrees.indirect_missing_keys.root.rb_root));
1288
1289 ret = resolve_indirect_refs(fs_info, path, time_seq, &preftrees,
1290 extent_item_pos, sc, ignore_offset);
1291 if (ret)
1292 goto out;
1293
1294 WARN_ON(!RB_EMPTY_ROOT(&preftrees.indirect.root.rb_root));
1295
1296 /*
1297 * This walks the tree of merged and resolved refs. Tree blocks are
1298 * read in as needed. Unique entries are added to the ulist, and
1299 * the list of found roots is updated.
1300 *
1301 * We release the entire tree in one go before returning.
1302 */
1303 node = rb_first_cached(&preftrees.direct.root);
1304 while (node) {
1305 ref = rb_entry(node, struct prelim_ref, rbnode);
1306 node = rb_next(&ref->rbnode);
1307 /*
1308 * ref->count < 0 can happen here if there are delayed
1309 * refs with a node->action of BTRFS_DROP_DELAYED_REF.
1310 * prelim_ref_insert() relies on this when merging
1311 * identical refs to keep the overall count correct.
1312 * prelim_ref_insert() will merge only those refs
1313 * which compare identically. Any refs having
1314 * e.g. different offsets would not be merged,
1315 * and would retain their original ref->count < 0.
1316 */
1317 if (roots && ref->count && ref->root_id && ref->parent == 0) {
1318 if (sc && sc->root_objectid &&
1319 ref->root_id != sc->root_objectid) {
1320 ret = BACKREF_FOUND_SHARED;
1321 goto out;
1322 }
1323
1324 /* no parent == root of tree */
1325 ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS);
1326 if (ret < 0)
1327 goto out;
1328 }
1329 if (ref->count && ref->parent) {
1330 if (extent_item_pos && !ref->inode_list &&
1331 ref->level == 0) {
1332 struct extent_buffer *eb;
1333
1334 eb = read_tree_block(fs_info, ref->parent, 0,
1335 0, ref->level, NULL);
1336 if (IS_ERR(eb)) {
1337 ret = PTR_ERR(eb);
1338 goto out;
1339 } else if (!extent_buffer_uptodate(eb)) {
1340 free_extent_buffer(eb);
1341 ret = -EIO;
1342 goto out;
1343 }
1344
1345 if (!path->skip_locking)
1346 btrfs_tree_read_lock(eb);
1347 ret = find_extent_in_eb(eb, bytenr,
1348 *extent_item_pos, &eie, ignore_offset);
1349 if (!path->skip_locking)
1350 btrfs_tree_read_unlock(eb);
1351 free_extent_buffer(eb);
1352 if (ret < 0)
1353 goto out;
1354 ref->inode_list = eie;
1355 }
1356 ret = ulist_add_merge_ptr(refs, ref->parent,
1357 ref->inode_list,
1358 (void **)&eie, GFP_NOFS);
1359 if (ret < 0)
1360 goto out;
1361 if (!ret && extent_item_pos) {
1362 /*
1363 * we've recorded that parent, so we must extend
1364 * its inode list here
1365 */
1366 BUG_ON(!eie);
1367 while (eie->next)
1368 eie = eie->next;
1369 eie->next = ref->inode_list;
1370 }
1371 eie = NULL;
1372 }
1373 cond_resched();
1374 }
1375
1376out:
1377 btrfs_free_path(path);
1378
1379 prelim_release(&preftrees.direct);
1380 prelim_release(&preftrees.indirect);
1381 prelim_release(&preftrees.indirect_missing_keys);
1382
1383 if (ret < 0)
1384 free_inode_elem_list(eie);
1385 return ret;
1386}
1387
1388static void free_leaf_list(struct ulist *blocks)
1389{
1390 struct ulist_node *node = NULL;
1391 struct extent_inode_elem *eie;
1392 struct ulist_iterator uiter;
1393
1394 ULIST_ITER_INIT(&uiter);
1395 while ((node = ulist_next(blocks, &uiter))) {
1396 if (!node->aux)
1397 continue;
1398 eie = unode_aux_to_inode_list(node);
1399 free_inode_elem_list(eie);
1400 node->aux = 0;
1401 }
1402
1403 ulist_free(blocks);
1404}
1405
1406/*
1407 * Finds all leafs with a reference to the specified combination of bytenr and
1408 * offset. key_list_head will point to a list of corresponding keys (caller must
1409 * free each list element). The leafs will be stored in the leafs ulist, which
1410 * must be freed with ulist_free.
1411 *
1412 * returns 0 on success, <0 on error
1413 */
1414int btrfs_find_all_leafs(struct btrfs_trans_handle *trans,
1415 struct btrfs_fs_info *fs_info, u64 bytenr,
1416 u64 time_seq, struct ulist **leafs,
1417 const u64 *extent_item_pos, bool ignore_offset)
1418{
1419 int ret;
1420
1421 *leafs = ulist_alloc(GFP_NOFS);
1422 if (!*leafs)
1423 return -ENOMEM;
1424
1425 ret = find_parent_nodes(trans, fs_info, bytenr, time_seq,
1426 *leafs, NULL, extent_item_pos, NULL, ignore_offset);
1427 if (ret < 0 && ret != -ENOENT) {
1428 free_leaf_list(*leafs);
1429 return ret;
1430 }
1431
1432 return 0;
1433}
1434
1435/*
1436 * walk all backrefs for a given extent to find all roots that reference this
1437 * extent. Walking a backref means finding all extents that reference this
1438 * extent and in turn walk the backrefs of those, too. Naturally this is a
1439 * recursive process, but here it is implemented in an iterative fashion: We
1440 * find all referencing extents for the extent in question and put them on a
1441 * list. In turn, we find all referencing extents for those, further appending
1442 * to the list. The way we iterate the list allows adding more elements after
1443 * the current while iterating. The process stops when we reach the end of the
1444 * list. Found roots are added to the roots list.
1445 *
1446 * returns 0 on success, < 0 on error.
1447 */
1448static int btrfs_find_all_roots_safe(struct btrfs_trans_handle *trans,
1449 struct btrfs_fs_info *fs_info, u64 bytenr,
1450 u64 time_seq, struct ulist **roots,
1451 bool ignore_offset)
1452{
1453 struct ulist *tmp;
1454 struct ulist_node *node = NULL;
1455 struct ulist_iterator uiter;
1456 int ret;
1457
1458 tmp = ulist_alloc(GFP_NOFS);
1459 if (!tmp)
1460 return -ENOMEM;
1461 *roots = ulist_alloc(GFP_NOFS);
1462 if (!*roots) {
1463 ulist_free(tmp);
1464 return -ENOMEM;
1465 }
1466
1467 ULIST_ITER_INIT(&uiter);
1468 while (1) {
1469 ret = find_parent_nodes(trans, fs_info, bytenr, time_seq,
1470 tmp, *roots, NULL, NULL, ignore_offset);
1471 if (ret < 0 && ret != -ENOENT) {
1472 ulist_free(tmp);
1473 ulist_free(*roots);
1474 *roots = NULL;
1475 return ret;
1476 }
1477 node = ulist_next(tmp, &uiter);
1478 if (!node)
1479 break;
1480 bytenr = node->val;
1481 cond_resched();
1482 }
1483
1484 ulist_free(tmp);
1485 return 0;
1486}
1487
1488int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1489 struct btrfs_fs_info *fs_info, u64 bytenr,
1490 u64 time_seq, struct ulist **roots,
1491 bool ignore_offset, bool skip_commit_root_sem)
1492{
1493 int ret;
1494
1495 if (!trans && !skip_commit_root_sem)
1496 down_read(&fs_info->commit_root_sem);
1497 ret = btrfs_find_all_roots_safe(trans, fs_info, bytenr,
1498 time_seq, roots, ignore_offset);
1499 if (!trans && !skip_commit_root_sem)
1500 up_read(&fs_info->commit_root_sem);
1501 return ret;
1502}
1503
1504/**
1505 * Check if an extent is shared or not
1506 *
1507 * @root: root inode belongs to
1508 * @inum: inode number of the inode whose extent we are checking
1509 * @bytenr: logical bytenr of the extent we are checking
1510 * @roots: list of roots this extent is shared among
1511 * @tmp: temporary list used for iteration
1512 *
1513 * btrfs_check_shared uses the backref walking code but will short
1514 * circuit as soon as it finds a root or inode that doesn't match the
1515 * one passed in. This provides a significant performance benefit for
1516 * callers (such as fiemap) which want to know whether the extent is
1517 * shared but do not need a ref count.
1518 *
1519 * This attempts to attach to the running transaction in order to account for
1520 * delayed refs, but continues on even when no running transaction exists.
1521 *
1522 * Return: 0 if extent is not shared, 1 if it is shared, < 0 on error.
1523 */
1524int btrfs_check_shared(struct btrfs_root *root, u64 inum, u64 bytenr,
1525 struct ulist *roots, struct ulist *tmp)
1526{
1527 struct btrfs_fs_info *fs_info = root->fs_info;
1528 struct btrfs_trans_handle *trans;
1529 struct ulist_iterator uiter;
1530 struct ulist_node *node;
1531 struct btrfs_seq_list elem = BTRFS_SEQ_LIST_INIT(elem);
1532 int ret = 0;
1533 struct share_check shared = {
1534 .root_objectid = root->root_key.objectid,
1535 .inum = inum,
1536 .share_count = 0,
1537 };
1538
1539 ulist_init(roots);
1540 ulist_init(tmp);
1541
1542 trans = btrfs_join_transaction_nostart(root);
1543 if (IS_ERR(trans)) {
1544 if (PTR_ERR(trans) != -ENOENT && PTR_ERR(trans) != -EROFS) {
1545 ret = PTR_ERR(trans);
1546 goto out;
1547 }
1548 trans = NULL;
1549 down_read(&fs_info->commit_root_sem);
1550 } else {
1551 btrfs_get_tree_mod_seq(fs_info, &elem);
1552 }
1553
1554 ULIST_ITER_INIT(&uiter);
1555 while (1) {
1556 ret = find_parent_nodes(trans, fs_info, bytenr, elem.seq, tmp,
1557 roots, NULL, &shared, false);
1558 if (ret == BACKREF_FOUND_SHARED) {
1559 /* this is the only condition under which we return 1 */
1560 ret = 1;
1561 break;
1562 }
1563 if (ret < 0 && ret != -ENOENT)
1564 break;
1565 ret = 0;
1566 node = ulist_next(tmp, &uiter);
1567 if (!node)
1568 break;
1569 bytenr = node->val;
1570 shared.share_count = 0;
1571 cond_resched();
1572 }
1573
1574 if (trans) {
1575 btrfs_put_tree_mod_seq(fs_info, &elem);
1576 btrfs_end_transaction(trans);
1577 } else {
1578 up_read(&fs_info->commit_root_sem);
1579 }
1580out:
1581 ulist_release(roots);
1582 ulist_release(tmp);
1583 return ret;
1584}
1585
1586int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,
1587 u64 start_off, struct btrfs_path *path,
1588 struct btrfs_inode_extref **ret_extref,
1589 u64 *found_off)
1590{
1591 int ret, slot;
1592 struct btrfs_key key;
1593 struct btrfs_key found_key;
1594 struct btrfs_inode_extref *extref;
1595 const struct extent_buffer *leaf;
1596 unsigned long ptr;
1597
1598 key.objectid = inode_objectid;
1599 key.type = BTRFS_INODE_EXTREF_KEY;
1600 key.offset = start_off;
1601
1602 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1603 if (ret < 0)
1604 return ret;
1605
1606 while (1) {
1607 leaf = path->nodes[0];
1608 slot = path->slots[0];
1609 if (slot >= btrfs_header_nritems(leaf)) {
1610 /*
1611 * If the item at offset is not found,
1612 * btrfs_search_slot will point us to the slot
1613 * where it should be inserted. In our case
1614 * that will be the slot directly before the
1615 * next INODE_REF_KEY_V2 item. In the case
1616 * that we're pointing to the last slot in a
1617 * leaf, we must move one leaf over.
1618 */
1619 ret = btrfs_next_leaf(root, path);
1620 if (ret) {
1621 if (ret >= 1)
1622 ret = -ENOENT;
1623 break;
1624 }
1625 continue;
1626 }
1627
1628 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1629
1630 /*
1631 * Check that we're still looking at an extended ref key for
1632 * this particular objectid. If we have different
1633 * objectid or type then there are no more to be found
1634 * in the tree and we can exit.
1635 */
1636 ret = -ENOENT;
1637 if (found_key.objectid != inode_objectid)
1638 break;
1639 if (found_key.type != BTRFS_INODE_EXTREF_KEY)
1640 break;
1641
1642 ret = 0;
1643 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1644 extref = (struct btrfs_inode_extref *)ptr;
1645 *ret_extref = extref;
1646 if (found_off)
1647 *found_off = found_key.offset;
1648 break;
1649 }
1650
1651 return ret;
1652}
1653
1654/*
1655 * this iterates to turn a name (from iref/extref) into a full filesystem path.
1656 * Elements of the path are separated by '/' and the path is guaranteed to be
1657 * 0-terminated. the path is only given within the current file system.
1658 * Therefore, it never starts with a '/'. the caller is responsible to provide
1659 * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
1660 * the start point of the resulting string is returned. this pointer is within
1661 * dest, normally.
1662 * in case the path buffer would overflow, the pointer is decremented further
1663 * as if output was written to the buffer, though no more output is actually
1664 * generated. that way, the caller can determine how much space would be
1665 * required for the path to fit into the buffer. in that case, the returned
1666 * value will be smaller than dest. callers must check this!
1667 */
1668char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
1669 u32 name_len, unsigned long name_off,
1670 struct extent_buffer *eb_in, u64 parent,
1671 char *dest, u32 size)
1672{
1673 int slot;
1674 u64 next_inum;
1675 int ret;
1676 s64 bytes_left = ((s64)size) - 1;
1677 struct extent_buffer *eb = eb_in;
1678 struct btrfs_key found_key;
1679 struct btrfs_inode_ref *iref;
1680
1681 if (bytes_left >= 0)
1682 dest[bytes_left] = '\0';
1683
1684 while (1) {
1685 bytes_left -= name_len;
1686 if (bytes_left >= 0)
1687 read_extent_buffer(eb, dest + bytes_left,
1688 name_off, name_len);
1689 if (eb != eb_in) {
1690 if (!path->skip_locking)
1691 btrfs_tree_read_unlock(eb);
1692 free_extent_buffer(eb);
1693 }
1694 ret = btrfs_find_item(fs_root, path, parent, 0,
1695 BTRFS_INODE_REF_KEY, &found_key);
1696 if (ret > 0)
1697 ret = -ENOENT;
1698 if (ret)
1699 break;
1700
1701 next_inum = found_key.offset;
1702
1703 /* regular exit ahead */
1704 if (parent == next_inum)
1705 break;
1706
1707 slot = path->slots[0];
1708 eb = path->nodes[0];
1709 /* make sure we can use eb after releasing the path */
1710 if (eb != eb_in) {
1711 path->nodes[0] = NULL;
1712 path->locks[0] = 0;
1713 }
1714 btrfs_release_path(path);
1715 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1716
1717 name_len = btrfs_inode_ref_name_len(eb, iref);
1718 name_off = (unsigned long)(iref + 1);
1719
1720 parent = next_inum;
1721 --bytes_left;
1722 if (bytes_left >= 0)
1723 dest[bytes_left] = '/';
1724 }
1725
1726 btrfs_release_path(path);
1727
1728 if (ret)
1729 return ERR_PTR(ret);
1730
1731 return dest + bytes_left;
1732}
1733
1734/*
1735 * this makes the path point to (logical EXTENT_ITEM *)
1736 * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
1737 * tree blocks and <0 on error.
1738 */
1739int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
1740 struct btrfs_path *path, struct btrfs_key *found_key,
1741 u64 *flags_ret)
1742{
1743 int ret;
1744 u64 flags;
1745 u64 size = 0;
1746 u32 item_size;
1747 const struct extent_buffer *eb;
1748 struct btrfs_extent_item *ei;
1749 struct btrfs_key key;
1750
1751 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1752 key.type = BTRFS_METADATA_ITEM_KEY;
1753 else
1754 key.type = BTRFS_EXTENT_ITEM_KEY;
1755 key.objectid = logical;
1756 key.offset = (u64)-1;
1757
1758 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
1759 if (ret < 0)
1760 return ret;
1761
1762 ret = btrfs_previous_extent_item(fs_info->extent_root, path, 0);
1763 if (ret) {
1764 if (ret > 0)
1765 ret = -ENOENT;
1766 return ret;
1767 }
1768 btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]);
1769 if (found_key->type == BTRFS_METADATA_ITEM_KEY)
1770 size = fs_info->nodesize;
1771 else if (found_key->type == BTRFS_EXTENT_ITEM_KEY)
1772 size = found_key->offset;
1773
1774 if (found_key->objectid > logical ||
1775 found_key->objectid + size <= logical) {
1776 btrfs_debug(fs_info,
1777 "logical %llu is not within any extent", logical);
1778 return -ENOENT;
1779 }
1780
1781 eb = path->nodes[0];
1782 item_size = btrfs_item_size_nr(eb, path->slots[0]);
1783 BUG_ON(item_size < sizeof(*ei));
1784
1785 ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
1786 flags = btrfs_extent_flags(eb, ei);
1787
1788 btrfs_debug(fs_info,
1789 "logical %llu is at position %llu within the extent (%llu EXTENT_ITEM %llu) flags %#llx size %u",
1790 logical, logical - found_key->objectid, found_key->objectid,
1791 found_key->offset, flags, item_size);
1792
1793 WARN_ON(!flags_ret);
1794 if (flags_ret) {
1795 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1796 *flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK;
1797 else if (flags & BTRFS_EXTENT_FLAG_DATA)
1798 *flags_ret = BTRFS_EXTENT_FLAG_DATA;
1799 else
1800 BUG();
1801 return 0;
1802 }
1803
1804 return -EIO;
1805}
1806
1807/*
1808 * helper function to iterate extent inline refs. ptr must point to a 0 value
1809 * for the first call and may be modified. it is used to track state.
1810 * if more refs exist, 0 is returned and the next call to
1811 * get_extent_inline_ref must pass the modified ptr parameter to get the
1812 * next ref. after the last ref was processed, 1 is returned.
1813 * returns <0 on error
1814 */
1815static int get_extent_inline_ref(unsigned long *ptr,
1816 const struct extent_buffer *eb,
1817 const struct btrfs_key *key,
1818 const struct btrfs_extent_item *ei,
1819 u32 item_size,
1820 struct btrfs_extent_inline_ref **out_eiref,
1821 int *out_type)
1822{
1823 unsigned long end;
1824 u64 flags;
1825 struct btrfs_tree_block_info *info;
1826
1827 if (!*ptr) {
1828 /* first call */
1829 flags = btrfs_extent_flags(eb, ei);
1830 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1831 if (key->type == BTRFS_METADATA_ITEM_KEY) {
1832 /* a skinny metadata extent */
1833 *out_eiref =
1834 (struct btrfs_extent_inline_ref *)(ei + 1);
1835 } else {
1836 WARN_ON(key->type != BTRFS_EXTENT_ITEM_KEY);
1837 info = (struct btrfs_tree_block_info *)(ei + 1);
1838 *out_eiref =
1839 (struct btrfs_extent_inline_ref *)(info + 1);
1840 }
1841 } else {
1842 *out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1);
1843 }
1844 *ptr = (unsigned long)*out_eiref;
1845 if ((unsigned long)(*ptr) >= (unsigned long)ei + item_size)
1846 return -ENOENT;
1847 }
1848
1849 end = (unsigned long)ei + item_size;
1850 *out_eiref = (struct btrfs_extent_inline_ref *)(*ptr);
1851 *out_type = btrfs_get_extent_inline_ref_type(eb, *out_eiref,
1852 BTRFS_REF_TYPE_ANY);
1853 if (*out_type == BTRFS_REF_TYPE_INVALID)
1854 return -EUCLEAN;
1855
1856 *ptr += btrfs_extent_inline_ref_size(*out_type);
1857 WARN_ON(*ptr > end);
1858 if (*ptr == end)
1859 return 1; /* last */
1860
1861 return 0;
1862}
1863
1864/*
1865 * reads the tree block backref for an extent. tree level and root are returned
1866 * through out_level and out_root. ptr must point to a 0 value for the first
1867 * call and may be modified (see get_extent_inline_ref comment).
1868 * returns 0 if data was provided, 1 if there was no more data to provide or
1869 * <0 on error.
1870 */
1871int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
1872 struct btrfs_key *key, struct btrfs_extent_item *ei,
1873 u32 item_size, u64 *out_root, u8 *out_level)
1874{
1875 int ret;
1876 int type;
1877 struct btrfs_extent_inline_ref *eiref;
1878
1879 if (*ptr == (unsigned long)-1)
1880 return 1;
1881
1882 while (1) {
1883 ret = get_extent_inline_ref(ptr, eb, key, ei, item_size,
1884 &eiref, &type);
1885 if (ret < 0)
1886 return ret;
1887
1888 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1889 type == BTRFS_SHARED_BLOCK_REF_KEY)
1890 break;
1891
1892 if (ret == 1)
1893 return 1;
1894 }
1895
1896 /* we can treat both ref types equally here */
1897 *out_root = btrfs_extent_inline_ref_offset(eb, eiref);
1898
1899 if (key->type == BTRFS_EXTENT_ITEM_KEY) {
1900 struct btrfs_tree_block_info *info;
1901
1902 info = (struct btrfs_tree_block_info *)(ei + 1);
1903 *out_level = btrfs_tree_block_level(eb, info);
1904 } else {
1905 ASSERT(key->type == BTRFS_METADATA_ITEM_KEY);
1906 *out_level = (u8)key->offset;
1907 }
1908
1909 if (ret == 1)
1910 *ptr = (unsigned long)-1;
1911
1912 return 0;
1913}
1914
1915static int iterate_leaf_refs(struct btrfs_fs_info *fs_info,
1916 struct extent_inode_elem *inode_list,
1917 u64 root, u64 extent_item_objectid,
1918 iterate_extent_inodes_t *iterate, void *ctx)
1919{
1920 struct extent_inode_elem *eie;
1921 int ret = 0;
1922
1923 for (eie = inode_list; eie; eie = eie->next) {
1924 btrfs_debug(fs_info,
1925 "ref for %llu resolved, key (%llu EXTEND_DATA %llu), root %llu",
1926 extent_item_objectid, eie->inum,
1927 eie->offset, root);
1928 ret = iterate(eie->inum, eie->offset, root, ctx);
1929 if (ret) {
1930 btrfs_debug(fs_info,
1931 "stopping iteration for %llu due to ret=%d",
1932 extent_item_objectid, ret);
1933 break;
1934 }
1935 }
1936
1937 return ret;
1938}
1939
1940/*
1941 * calls iterate() for every inode that references the extent identified by
1942 * the given parameters.
1943 * when the iterator function returns a non-zero value, iteration stops.
1944 */
1945int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
1946 u64 extent_item_objectid, u64 extent_item_pos,
1947 int search_commit_root,
1948 iterate_extent_inodes_t *iterate, void *ctx,
1949 bool ignore_offset)
1950{
1951 int ret;
1952 struct btrfs_trans_handle *trans = NULL;
1953 struct ulist *refs = NULL;
1954 struct ulist *roots = NULL;
1955 struct ulist_node *ref_node = NULL;
1956 struct ulist_node *root_node = NULL;
1957 struct btrfs_seq_list seq_elem = BTRFS_SEQ_LIST_INIT(seq_elem);
1958 struct ulist_iterator ref_uiter;
1959 struct ulist_iterator root_uiter;
1960
1961 btrfs_debug(fs_info, "resolving all inodes for extent %llu",
1962 extent_item_objectid);
1963
1964 if (!search_commit_root) {
1965 trans = btrfs_attach_transaction(fs_info->extent_root);
1966 if (IS_ERR(trans)) {
1967 if (PTR_ERR(trans) != -ENOENT &&
1968 PTR_ERR(trans) != -EROFS)
1969 return PTR_ERR(trans);
1970 trans = NULL;
1971 }
1972 }
1973
1974 if (trans)
1975 btrfs_get_tree_mod_seq(fs_info, &seq_elem);
1976 else
1977 down_read(&fs_info->commit_root_sem);
1978
1979 ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid,
1980 seq_elem.seq, &refs,
1981 &extent_item_pos, ignore_offset);
1982 if (ret)
1983 goto out;
1984
1985 ULIST_ITER_INIT(&ref_uiter);
1986 while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) {
1987 ret = btrfs_find_all_roots_safe(trans, fs_info, ref_node->val,
1988 seq_elem.seq, &roots,
1989 ignore_offset);
1990 if (ret)
1991 break;
1992 ULIST_ITER_INIT(&root_uiter);
1993 while (!ret && (root_node = ulist_next(roots, &root_uiter))) {
1994 btrfs_debug(fs_info,
1995 "root %llu references leaf %llu, data list %#llx",
1996 root_node->val, ref_node->val,
1997 ref_node->aux);
1998 ret = iterate_leaf_refs(fs_info,
1999 (struct extent_inode_elem *)
2000 (uintptr_t)ref_node->aux,
2001 root_node->val,
2002 extent_item_objectid,
2003 iterate, ctx);
2004 }
2005 ulist_free(roots);
2006 }
2007
2008 free_leaf_list(refs);
2009out:
2010 if (trans) {
2011 btrfs_put_tree_mod_seq(fs_info, &seq_elem);
2012 btrfs_end_transaction(trans);
2013 } else {
2014 up_read(&fs_info->commit_root_sem);
2015 }
2016
2017 return ret;
2018}
2019
2020int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
2021 struct btrfs_path *path,
2022 iterate_extent_inodes_t *iterate, void *ctx,
2023 bool ignore_offset)
2024{
2025 int ret;
2026 u64 extent_item_pos;
2027 u64 flags = 0;
2028 struct btrfs_key found_key;
2029 int search_commit_root = path->search_commit_root;
2030
2031 ret = extent_from_logical(fs_info, logical, path, &found_key, &flags);
2032 btrfs_release_path(path);
2033 if (ret < 0)
2034 return ret;
2035 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
2036 return -EINVAL;
2037
2038 extent_item_pos = logical - found_key.objectid;
2039 ret = iterate_extent_inodes(fs_info, found_key.objectid,
2040 extent_item_pos, search_commit_root,
2041 iterate, ctx, ignore_offset);
2042
2043 return ret;
2044}
2045
2046typedef int (iterate_irefs_t)(u64 parent, u32 name_len, unsigned long name_off,
2047 struct extent_buffer *eb, void *ctx);
2048
2049static int iterate_inode_refs(u64 inum, struct btrfs_root *fs_root,
2050 struct btrfs_path *path,
2051 iterate_irefs_t *iterate, void *ctx)
2052{
2053 int ret = 0;
2054 int slot;
2055 u32 cur;
2056 u32 len;
2057 u32 name_len;
2058 u64 parent = 0;
2059 int found = 0;
2060 struct extent_buffer *eb;
2061 struct btrfs_item *item;
2062 struct btrfs_inode_ref *iref;
2063 struct btrfs_key found_key;
2064
2065 while (!ret) {
2066 ret = btrfs_find_item(fs_root, path, inum,
2067 parent ? parent + 1 : 0, BTRFS_INODE_REF_KEY,
2068 &found_key);
2069
2070 if (ret < 0)
2071 break;
2072 if (ret) {
2073 ret = found ? 0 : -ENOENT;
2074 break;
2075 }
2076 ++found;
2077
2078 parent = found_key.offset;
2079 slot = path->slots[0];
2080 eb = btrfs_clone_extent_buffer(path->nodes[0]);
2081 if (!eb) {
2082 ret = -ENOMEM;
2083 break;
2084 }
2085 btrfs_release_path(path);
2086
2087 item = btrfs_item_nr(slot);
2088 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
2089
2090 for (cur = 0; cur < btrfs_item_size(eb, item); cur += len) {
2091 name_len = btrfs_inode_ref_name_len(eb, iref);
2092 /* path must be released before calling iterate()! */
2093 btrfs_debug(fs_root->fs_info,
2094 "following ref at offset %u for inode %llu in tree %llu",
2095 cur, found_key.objectid,
2096 fs_root->root_key.objectid);
2097 ret = iterate(parent, name_len,
2098 (unsigned long)(iref + 1), eb, ctx);
2099 if (ret)
2100 break;
2101 len = sizeof(*iref) + name_len;
2102 iref = (struct btrfs_inode_ref *)((char *)iref + len);
2103 }
2104 free_extent_buffer(eb);
2105 }
2106
2107 btrfs_release_path(path);
2108
2109 return ret;
2110}
2111
2112static int iterate_inode_extrefs(u64 inum, struct btrfs_root *fs_root,
2113 struct btrfs_path *path,
2114 iterate_irefs_t *iterate, void *ctx)
2115{
2116 int ret;
2117 int slot;
2118 u64 offset = 0;
2119 u64 parent;
2120 int found = 0;
2121 struct extent_buffer *eb;
2122 struct btrfs_inode_extref *extref;
2123 u32 item_size;
2124 u32 cur_offset;
2125 unsigned long ptr;
2126
2127 while (1) {
2128 ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref,
2129 &offset);
2130 if (ret < 0)
2131 break;
2132 if (ret) {
2133 ret = found ? 0 : -ENOENT;
2134 break;
2135 }
2136 ++found;
2137
2138 slot = path->slots[0];
2139 eb = btrfs_clone_extent_buffer(path->nodes[0]);
2140 if (!eb) {
2141 ret = -ENOMEM;
2142 break;
2143 }
2144 btrfs_release_path(path);
2145
2146 item_size = btrfs_item_size_nr(eb, slot);
2147 ptr = btrfs_item_ptr_offset(eb, slot);
2148 cur_offset = 0;
2149
2150 while (cur_offset < item_size) {
2151 u32 name_len;
2152
2153 extref = (struct btrfs_inode_extref *)(ptr + cur_offset);
2154 parent = btrfs_inode_extref_parent(eb, extref);
2155 name_len = btrfs_inode_extref_name_len(eb, extref);
2156 ret = iterate(parent, name_len,
2157 (unsigned long)&extref->name, eb, ctx);
2158 if (ret)
2159 break;
2160
2161 cur_offset += btrfs_inode_extref_name_len(eb, extref);
2162 cur_offset += sizeof(*extref);
2163 }
2164 free_extent_buffer(eb);
2165
2166 offset++;
2167 }
2168
2169 btrfs_release_path(path);
2170
2171 return ret;
2172}
2173
2174static int iterate_irefs(u64 inum, struct btrfs_root *fs_root,
2175 struct btrfs_path *path, iterate_irefs_t *iterate,
2176 void *ctx)
2177{
2178 int ret;
2179 int found_refs = 0;
2180
2181 ret = iterate_inode_refs(inum, fs_root, path, iterate, ctx);
2182 if (!ret)
2183 ++found_refs;
2184 else if (ret != -ENOENT)
2185 return ret;
2186
2187 ret = iterate_inode_extrefs(inum, fs_root, path, iterate, ctx);
2188 if (ret == -ENOENT && found_refs)
2189 return 0;
2190
2191 return ret;
2192}
2193
2194/*
2195 * returns 0 if the path could be dumped (probably truncated)
2196 * returns <0 in case of an error
2197 */
2198static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
2199 struct extent_buffer *eb, void *ctx)
2200{
2201 struct inode_fs_paths *ipath = ctx;
2202 char *fspath;
2203 char *fspath_min;
2204 int i = ipath->fspath->elem_cnt;
2205 const int s_ptr = sizeof(char *);
2206 u32 bytes_left;
2207
2208 bytes_left = ipath->fspath->bytes_left > s_ptr ?
2209 ipath->fspath->bytes_left - s_ptr : 0;
2210
2211 fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
2212 fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len,
2213 name_off, eb, inum, fspath_min, bytes_left);
2214 if (IS_ERR(fspath))
2215 return PTR_ERR(fspath);
2216
2217 if (fspath > fspath_min) {
2218 ipath->fspath->val[i] = (u64)(unsigned long)fspath;
2219 ++ipath->fspath->elem_cnt;
2220 ipath->fspath->bytes_left = fspath - fspath_min;
2221 } else {
2222 ++ipath->fspath->elem_missed;
2223 ipath->fspath->bytes_missing += fspath_min - fspath;
2224 ipath->fspath->bytes_left = 0;
2225 }
2226
2227 return 0;
2228}
2229
2230/*
2231 * this dumps all file system paths to the inode into the ipath struct, provided
2232 * is has been created large enough. each path is zero-terminated and accessed
2233 * from ipath->fspath->val[i].
2234 * when it returns, there are ipath->fspath->elem_cnt number of paths available
2235 * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the
2236 * number of missed paths is recorded in ipath->fspath->elem_missed, otherwise,
2237 * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
2238 * have been needed to return all paths.
2239 */
2240int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
2241{
2242 return iterate_irefs(inum, ipath->fs_root, ipath->btrfs_path,
2243 inode_to_path, ipath);
2244}
2245
2246struct btrfs_data_container *init_data_container(u32 total_bytes)
2247{
2248 struct btrfs_data_container *data;
2249 size_t alloc_bytes;
2250
2251 alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
2252 data = kvmalloc(alloc_bytes, GFP_KERNEL);
2253 if (!data)
2254 return ERR_PTR(-ENOMEM);
2255
2256 if (total_bytes >= sizeof(*data)) {
2257 data->bytes_left = total_bytes - sizeof(*data);
2258 data->bytes_missing = 0;
2259 } else {
2260 data->bytes_missing = sizeof(*data) - total_bytes;
2261 data->bytes_left = 0;
2262 }
2263
2264 data->elem_cnt = 0;
2265 data->elem_missed = 0;
2266
2267 return data;
2268}
2269
2270/*
2271 * allocates space to return multiple file system paths for an inode.
2272 * total_bytes to allocate are passed, note that space usable for actual path
2273 * information will be total_bytes - sizeof(struct inode_fs_paths).
2274 * the returned pointer must be freed with free_ipath() in the end.
2275 */
2276struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
2277 struct btrfs_path *path)
2278{
2279 struct inode_fs_paths *ifp;
2280 struct btrfs_data_container *fspath;
2281
2282 fspath = init_data_container(total_bytes);
2283 if (IS_ERR(fspath))
2284 return ERR_CAST(fspath);
2285
2286 ifp = kmalloc(sizeof(*ifp), GFP_KERNEL);
2287 if (!ifp) {
2288 kvfree(fspath);
2289 return ERR_PTR(-ENOMEM);
2290 }
2291
2292 ifp->btrfs_path = path;
2293 ifp->fspath = fspath;
2294 ifp->fs_root = fs_root;
2295
2296 return ifp;
2297}
2298
2299void free_ipath(struct inode_fs_paths *ipath)
2300{
2301 if (!ipath)
2302 return;
2303 kvfree(ipath->fspath);
2304 kfree(ipath);
2305}
2306
2307struct btrfs_backref_iter *btrfs_backref_iter_alloc(
2308 struct btrfs_fs_info *fs_info, gfp_t gfp_flag)
2309{
2310 struct btrfs_backref_iter *ret;
2311
2312 ret = kzalloc(sizeof(*ret), gfp_flag);
2313 if (!ret)
2314 return NULL;
2315
2316 ret->path = btrfs_alloc_path();
2317 if (!ret->path) {
2318 kfree(ret);
2319 return NULL;
2320 }
2321
2322 /* Current backref iterator only supports iteration in commit root */
2323 ret->path->search_commit_root = 1;
2324 ret->path->skip_locking = 1;
2325 ret->fs_info = fs_info;
2326
2327 return ret;
2328}
2329
2330int btrfs_backref_iter_start(struct btrfs_backref_iter *iter, u64 bytenr)
2331{
2332 struct btrfs_fs_info *fs_info = iter->fs_info;
2333 struct btrfs_path *path = iter->path;
2334 struct btrfs_extent_item *ei;
2335 struct btrfs_key key;
2336 int ret;
2337
2338 key.objectid = bytenr;
2339 key.type = BTRFS_METADATA_ITEM_KEY;
2340 key.offset = (u64)-1;
2341 iter->bytenr = bytenr;
2342
2343 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
2344 if (ret < 0)
2345 return ret;
2346 if (ret == 0) {
2347 ret = -EUCLEAN;
2348 goto release;
2349 }
2350 if (path->slots[0] == 0) {
2351 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
2352 ret = -EUCLEAN;
2353 goto release;
2354 }
2355 path->slots[0]--;
2356
2357 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2358 if ((key.type != BTRFS_EXTENT_ITEM_KEY &&
2359 key.type != BTRFS_METADATA_ITEM_KEY) || key.objectid != bytenr) {
2360 ret = -ENOENT;
2361 goto release;
2362 }
2363 memcpy(&iter->cur_key, &key, sizeof(key));
2364 iter->item_ptr = (u32)btrfs_item_ptr_offset(path->nodes[0],
2365 path->slots[0]);
2366 iter->end_ptr = (u32)(iter->item_ptr +
2367 btrfs_item_size_nr(path->nodes[0], path->slots[0]));
2368 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
2369 struct btrfs_extent_item);
2370
2371 /*
2372 * Only support iteration on tree backref yet.
2373 *
2374 * This is an extra precaution for non skinny-metadata, where
2375 * EXTENT_ITEM is also used for tree blocks, that we can only use
2376 * extent flags to determine if it's a tree block.
2377 */
2378 if (btrfs_extent_flags(path->nodes[0], ei) & BTRFS_EXTENT_FLAG_DATA) {
2379 ret = -ENOTSUPP;
2380 goto release;
2381 }
2382 iter->cur_ptr = (u32)(iter->item_ptr + sizeof(*ei));
2383
2384 /* If there is no inline backref, go search for keyed backref */
2385 if (iter->cur_ptr >= iter->end_ptr) {
2386 ret = btrfs_next_item(fs_info->extent_root, path);
2387
2388 /* No inline nor keyed ref */
2389 if (ret > 0) {
2390 ret = -ENOENT;
2391 goto release;
2392 }
2393 if (ret < 0)
2394 goto release;
2395
2396 btrfs_item_key_to_cpu(path->nodes[0], &iter->cur_key,
2397 path->slots[0]);
2398 if (iter->cur_key.objectid != bytenr ||
2399 (iter->cur_key.type != BTRFS_SHARED_BLOCK_REF_KEY &&
2400 iter->cur_key.type != BTRFS_TREE_BLOCK_REF_KEY)) {
2401 ret = -ENOENT;
2402 goto release;
2403 }
2404 iter->cur_ptr = (u32)btrfs_item_ptr_offset(path->nodes[0],
2405 path->slots[0]);
2406 iter->item_ptr = iter->cur_ptr;
2407 iter->end_ptr = (u32)(iter->item_ptr + btrfs_item_size_nr(
2408 path->nodes[0], path->slots[0]));
2409 }
2410
2411 return 0;
2412release:
2413 btrfs_backref_iter_release(iter);
2414 return ret;
2415}
2416
2417/*
2418 * Go to the next backref item of current bytenr, can be either inlined or
2419 * keyed.
2420 *
2421 * Caller needs to check whether it's inline ref or not by iter->cur_key.
2422 *
2423 * Return 0 if we get next backref without problem.
2424 * Return >0 if there is no extra backref for this bytenr.
2425 * Return <0 if there is something wrong happened.
2426 */
2427int btrfs_backref_iter_next(struct btrfs_backref_iter *iter)
2428{
2429 struct extent_buffer *eb = btrfs_backref_get_eb(iter);
2430 struct btrfs_path *path = iter->path;
2431 struct btrfs_extent_inline_ref *iref;
2432 int ret;
2433 u32 size;
2434
2435 if (btrfs_backref_iter_is_inline_ref(iter)) {
2436 /* We're still inside the inline refs */
2437 ASSERT(iter->cur_ptr < iter->end_ptr);
2438
2439 if (btrfs_backref_has_tree_block_info(iter)) {
2440 /* First tree block info */
2441 size = sizeof(struct btrfs_tree_block_info);
2442 } else {
2443 /* Use inline ref type to determine the size */
2444 int type;
2445
2446 iref = (struct btrfs_extent_inline_ref *)
2447 ((unsigned long)iter->cur_ptr);
2448 type = btrfs_extent_inline_ref_type(eb, iref);
2449
2450 size = btrfs_extent_inline_ref_size(type);
2451 }
2452 iter->cur_ptr += size;
2453 if (iter->cur_ptr < iter->end_ptr)
2454 return 0;
2455
2456 /* All inline items iterated, fall through */
2457 }
2458
2459 /* We're at keyed items, there is no inline item, go to the next one */
2460 ret = btrfs_next_item(iter->fs_info->extent_root, iter->path);
2461 if (ret)
2462 return ret;
2463
2464 btrfs_item_key_to_cpu(path->nodes[0], &iter->cur_key, path->slots[0]);
2465 if (iter->cur_key.objectid != iter->bytenr ||
2466 (iter->cur_key.type != BTRFS_TREE_BLOCK_REF_KEY &&
2467 iter->cur_key.type != BTRFS_SHARED_BLOCK_REF_KEY))
2468 return 1;
2469 iter->item_ptr = (u32)btrfs_item_ptr_offset(path->nodes[0],
2470 path->slots[0]);
2471 iter->cur_ptr = iter->item_ptr;
2472 iter->end_ptr = iter->item_ptr + (u32)btrfs_item_size_nr(path->nodes[0],
2473 path->slots[0]);
2474 return 0;
2475}
2476
2477void btrfs_backref_init_cache(struct btrfs_fs_info *fs_info,
2478 struct btrfs_backref_cache *cache, int is_reloc)
2479{
2480 int i;
2481
2482 cache->rb_root = RB_ROOT;
2483 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
2484 INIT_LIST_HEAD(&cache->pending[i]);
2485 INIT_LIST_HEAD(&cache->changed);
2486 INIT_LIST_HEAD(&cache->detached);
2487 INIT_LIST_HEAD(&cache->leaves);
2488 INIT_LIST_HEAD(&cache->pending_edge);
2489 INIT_LIST_HEAD(&cache->useless_node);
2490 cache->fs_info = fs_info;
2491 cache->is_reloc = is_reloc;
2492}
2493
2494struct btrfs_backref_node *btrfs_backref_alloc_node(
2495 struct btrfs_backref_cache *cache, u64 bytenr, int level)
2496{
2497 struct btrfs_backref_node *node;
2498
2499 ASSERT(level >= 0 && level < BTRFS_MAX_LEVEL);
2500 node = kzalloc(sizeof(*node), GFP_NOFS);
2501 if (!node)
2502 return node;
2503
2504 INIT_LIST_HEAD(&node->list);
2505 INIT_LIST_HEAD(&node->upper);
2506 INIT_LIST_HEAD(&node->lower);
2507 RB_CLEAR_NODE(&node->rb_node);
2508 cache->nr_nodes++;
2509 node->level = level;
2510 node->bytenr = bytenr;
2511
2512 return node;
2513}
2514
2515struct btrfs_backref_edge *btrfs_backref_alloc_edge(
2516 struct btrfs_backref_cache *cache)
2517{
2518 struct btrfs_backref_edge *edge;
2519
2520 edge = kzalloc(sizeof(*edge), GFP_NOFS);
2521 if (edge)
2522 cache->nr_edges++;
2523 return edge;
2524}
2525
2526/*
2527 * Drop the backref node from cache, also cleaning up all its
2528 * upper edges and any uncached nodes in the path.
2529 *
2530 * This cleanup happens bottom up, thus the node should either
2531 * be the lowest node in the cache or a detached node.
2532 */
2533void btrfs_backref_cleanup_node(struct btrfs_backref_cache *cache,
2534 struct btrfs_backref_node *node)
2535{
2536 struct btrfs_backref_node *upper;
2537 struct btrfs_backref_edge *edge;
2538
2539 if (!node)
2540 return;
2541
2542 BUG_ON(!node->lowest && !node->detached);
2543 while (!list_empty(&node->upper)) {
2544 edge = list_entry(node->upper.next, struct btrfs_backref_edge,
2545 list[LOWER]);
2546 upper = edge->node[UPPER];
2547 list_del(&edge->list[LOWER]);
2548 list_del(&edge->list[UPPER]);
2549 btrfs_backref_free_edge(cache, edge);
2550
2551 /*
2552 * Add the node to leaf node list if no other child block
2553 * cached.
2554 */
2555 if (list_empty(&upper->lower)) {
2556 list_add_tail(&upper->lower, &cache->leaves);
2557 upper->lowest = 1;
2558 }
2559 }
2560
2561 btrfs_backref_drop_node(cache, node);
2562}
2563
2564/*
2565 * Release all nodes/edges from current cache
2566 */
2567void btrfs_backref_release_cache(struct btrfs_backref_cache *cache)
2568{
2569 struct btrfs_backref_node *node;
2570 int i;
2571
2572 while (!list_empty(&cache->detached)) {
2573 node = list_entry(cache->detached.next,
2574 struct btrfs_backref_node, list);
2575 btrfs_backref_cleanup_node(cache, node);
2576 }
2577
2578 while (!list_empty(&cache->leaves)) {
2579 node = list_entry(cache->leaves.next,
2580 struct btrfs_backref_node, lower);
2581 btrfs_backref_cleanup_node(cache, node);
2582 }
2583
2584 cache->last_trans = 0;
2585
2586 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
2587 ASSERT(list_empty(&cache->pending[i]));
2588 ASSERT(list_empty(&cache->pending_edge));
2589 ASSERT(list_empty(&cache->useless_node));
2590 ASSERT(list_empty(&cache->changed));
2591 ASSERT(list_empty(&cache->detached));
2592 ASSERT(RB_EMPTY_ROOT(&cache->rb_root));
2593 ASSERT(!cache->nr_nodes);
2594 ASSERT(!cache->nr_edges);
2595}
2596
2597/*
2598 * Handle direct tree backref
2599 *
2600 * Direct tree backref means, the backref item shows its parent bytenr
2601 * directly. This is for SHARED_BLOCK_REF backref (keyed or inlined).
2602 *
2603 * @ref_key: The converted backref key.
2604 * For keyed backref, it's the item key.
2605 * For inlined backref, objectid is the bytenr,
2606 * type is btrfs_inline_ref_type, offset is
2607 * btrfs_inline_ref_offset.
2608 */
2609static int handle_direct_tree_backref(struct btrfs_backref_cache *cache,
2610 struct btrfs_key *ref_key,
2611 struct btrfs_backref_node *cur)
2612{
2613 struct btrfs_backref_edge *edge;
2614 struct btrfs_backref_node *upper;
2615 struct rb_node *rb_node;
2616
2617 ASSERT(ref_key->type == BTRFS_SHARED_BLOCK_REF_KEY);
2618
2619 /* Only reloc root uses backref pointing to itself */
2620 if (ref_key->objectid == ref_key->offset) {
2621 struct btrfs_root *root;
2622
2623 cur->is_reloc_root = 1;
2624 /* Only reloc backref cache cares about a specific root */
2625 if (cache->is_reloc) {
2626 root = find_reloc_root(cache->fs_info, cur->bytenr);
2627 if (!root)
2628 return -ENOENT;
2629 cur->root = root;
2630 } else {
2631 /*
2632 * For generic purpose backref cache, reloc root node
2633 * is useless.
2634 */
2635 list_add(&cur->list, &cache->useless_node);
2636 }
2637 return 0;
2638 }
2639
2640 edge = btrfs_backref_alloc_edge(cache);
2641 if (!edge)
2642 return -ENOMEM;
2643
2644 rb_node = rb_simple_search(&cache->rb_root, ref_key->offset);
2645 if (!rb_node) {
2646 /* Parent node not yet cached */
2647 upper = btrfs_backref_alloc_node(cache, ref_key->offset,
2648 cur->level + 1);
2649 if (!upper) {
2650 btrfs_backref_free_edge(cache, edge);
2651 return -ENOMEM;
2652 }
2653
2654 /*
2655 * Backrefs for the upper level block isn't cached, add the
2656 * block to pending list
2657 */
2658 list_add_tail(&edge->list[UPPER], &cache->pending_edge);
2659 } else {
2660 /* Parent node already cached */
2661 upper = rb_entry(rb_node, struct btrfs_backref_node, rb_node);
2662 ASSERT(upper->checked);
2663 INIT_LIST_HEAD(&edge->list[UPPER]);
2664 }
2665 btrfs_backref_link_edge(edge, cur, upper, LINK_LOWER);
2666 return 0;
2667}
2668
2669/*
2670 * Handle indirect tree backref
2671 *
2672 * Indirect tree backref means, we only know which tree the node belongs to.
2673 * We still need to do a tree search to find out the parents. This is for
2674 * TREE_BLOCK_REF backref (keyed or inlined).
2675 *
2676 * @ref_key: The same as @ref_key in handle_direct_tree_backref()
2677 * @tree_key: The first key of this tree block.
2678 * @path: A clean (released) path, to avoid allocating path every time
2679 * the function get called.
2680 */
2681static int handle_indirect_tree_backref(struct btrfs_backref_cache *cache,
2682 struct btrfs_path *path,
2683 struct btrfs_key *ref_key,
2684 struct btrfs_key *tree_key,
2685 struct btrfs_backref_node *cur)
2686{
2687 struct btrfs_fs_info *fs_info = cache->fs_info;
2688 struct btrfs_backref_node *upper;
2689 struct btrfs_backref_node *lower;
2690 struct btrfs_backref_edge *edge;
2691 struct extent_buffer *eb;
2692 struct btrfs_root *root;
2693 struct rb_node *rb_node;
2694 int level;
2695 bool need_check = true;
2696 int ret;
2697
2698 root = btrfs_get_fs_root(fs_info, ref_key->offset, false);
2699 if (IS_ERR(root))
2700 return PTR_ERR(root);
2701 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
2702 cur->cowonly = 1;
2703
2704 if (btrfs_root_level(&root->root_item) == cur->level) {
2705 /* Tree root */
2706 ASSERT(btrfs_root_bytenr(&root->root_item) == cur->bytenr);
2707 /*
2708 * For reloc backref cache, we may ignore reloc root. But for
2709 * general purpose backref cache, we can't rely on
2710 * btrfs_should_ignore_reloc_root() as it may conflict with
2711 * current running relocation and lead to missing root.
2712 *
2713 * For general purpose backref cache, reloc root detection is
2714 * completely relying on direct backref (key->offset is parent
2715 * bytenr), thus only do such check for reloc cache.
2716 */
2717 if (btrfs_should_ignore_reloc_root(root) && cache->is_reloc) {
2718 btrfs_put_root(root);
2719 list_add(&cur->list, &cache->useless_node);
2720 } else {
2721 cur->root = root;
2722 }
2723 return 0;
2724 }
2725
2726 level = cur->level + 1;
2727
2728 /* Search the tree to find parent blocks referring to the block */
2729 path->search_commit_root = 1;
2730 path->skip_locking = 1;
2731 path->lowest_level = level;
2732 ret = btrfs_search_slot(NULL, root, tree_key, path, 0, 0);
2733 path->lowest_level = 0;
2734 if (ret < 0) {
2735 btrfs_put_root(root);
2736 return ret;
2737 }
2738 if (ret > 0 && path->slots[level] > 0)
2739 path->slots[level]--;
2740
2741 eb = path->nodes[level];
2742 if (btrfs_node_blockptr(eb, path->slots[level]) != cur->bytenr) {
2743 btrfs_err(fs_info,
2744"couldn't find block (%llu) (level %d) in tree (%llu) with key (%llu %u %llu)",
2745 cur->bytenr, level - 1, root->root_key.objectid,
2746 tree_key->objectid, tree_key->type, tree_key->offset);
2747 btrfs_put_root(root);
2748 ret = -ENOENT;
2749 goto out;
2750 }
2751 lower = cur;
2752
2753 /* Add all nodes and edges in the path */
2754 for (; level < BTRFS_MAX_LEVEL; level++) {
2755 if (!path->nodes[level]) {
2756 ASSERT(btrfs_root_bytenr(&root->root_item) ==
2757 lower->bytenr);
2758 /* Same as previous should_ignore_reloc_root() call */
2759 if (btrfs_should_ignore_reloc_root(root) &&
2760 cache->is_reloc) {
2761 btrfs_put_root(root);
2762 list_add(&lower->list, &cache->useless_node);
2763 } else {
2764 lower->root = root;
2765 }
2766 break;
2767 }
2768
2769 edge = btrfs_backref_alloc_edge(cache);
2770 if (!edge) {
2771 btrfs_put_root(root);
2772 ret = -ENOMEM;
2773 goto out;
2774 }
2775
2776 eb = path->nodes[level];
2777 rb_node = rb_simple_search(&cache->rb_root, eb->start);
2778 if (!rb_node) {
2779 upper = btrfs_backref_alloc_node(cache, eb->start,
2780 lower->level + 1);
2781 if (!upper) {
2782 btrfs_put_root(root);
2783 btrfs_backref_free_edge(cache, edge);
2784 ret = -ENOMEM;
2785 goto out;
2786 }
2787 upper->owner = btrfs_header_owner(eb);
2788 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
2789 upper->cowonly = 1;
2790
2791 /*
2792 * If we know the block isn't shared we can avoid
2793 * checking its backrefs.
2794 */
2795 if (btrfs_block_can_be_shared(root, eb))
2796 upper->checked = 0;
2797 else
2798 upper->checked = 1;
2799
2800 /*
2801 * Add the block to pending list if we need to check its
2802 * backrefs, we only do this once while walking up a
2803 * tree as we will catch anything else later on.
2804 */
2805 if (!upper->checked && need_check) {
2806 need_check = false;
2807 list_add_tail(&edge->list[UPPER],
2808 &cache->pending_edge);
2809 } else {
2810 if (upper->checked)
2811 need_check = true;
2812 INIT_LIST_HEAD(&edge->list[UPPER]);
2813 }
2814 } else {
2815 upper = rb_entry(rb_node, struct btrfs_backref_node,
2816 rb_node);
2817 ASSERT(upper->checked);
2818 INIT_LIST_HEAD(&edge->list[UPPER]);
2819 if (!upper->owner)
2820 upper->owner = btrfs_header_owner(eb);
2821 }
2822 btrfs_backref_link_edge(edge, lower, upper, LINK_LOWER);
2823
2824 if (rb_node) {
2825 btrfs_put_root(root);
2826 break;
2827 }
2828 lower = upper;
2829 upper = NULL;
2830 }
2831out:
2832 btrfs_release_path(path);
2833 return ret;
2834}
2835
2836/*
2837 * Add backref node @cur into @cache.
2838 *
2839 * NOTE: Even if the function returned 0, @cur is not yet cached as its upper
2840 * links aren't yet bi-directional. Needs to finish such links.
2841 * Use btrfs_backref_finish_upper_links() to finish such linkage.
2842 *
2843 * @path: Released path for indirect tree backref lookup
2844 * @iter: Released backref iter for extent tree search
2845 * @node_key: The first key of the tree block
2846 */
2847int btrfs_backref_add_tree_node(struct btrfs_backref_cache *cache,
2848 struct btrfs_path *path,
2849 struct btrfs_backref_iter *iter,
2850 struct btrfs_key *node_key,
2851 struct btrfs_backref_node *cur)
2852{
2853 struct btrfs_fs_info *fs_info = cache->fs_info;
2854 struct btrfs_backref_edge *edge;
2855 struct btrfs_backref_node *exist;
2856 int ret;
2857
2858 ret = btrfs_backref_iter_start(iter, cur->bytenr);
2859 if (ret < 0)
2860 return ret;
2861 /*
2862 * We skip the first btrfs_tree_block_info, as we don't use the key
2863 * stored in it, but fetch it from the tree block
2864 */
2865 if (btrfs_backref_has_tree_block_info(iter)) {
2866 ret = btrfs_backref_iter_next(iter);
2867 if (ret < 0)
2868 goto out;
2869 /* No extra backref? This means the tree block is corrupted */
2870 if (ret > 0) {
2871 ret = -EUCLEAN;
2872 goto out;
2873 }
2874 }
2875 WARN_ON(cur->checked);
2876 if (!list_empty(&cur->upper)) {
2877 /*
2878 * The backref was added previously when processing backref of
2879 * type BTRFS_TREE_BLOCK_REF_KEY
2880 */
2881 ASSERT(list_is_singular(&cur->upper));
2882 edge = list_entry(cur->upper.next, struct btrfs_backref_edge,
2883 list[LOWER]);
2884 ASSERT(list_empty(&edge->list[UPPER]));
2885 exist = edge->node[UPPER];
2886 /*
2887 * Add the upper level block to pending list if we need check
2888 * its backrefs
2889 */
2890 if (!exist->checked)
2891 list_add_tail(&edge->list[UPPER], &cache->pending_edge);
2892 } else {
2893 exist = NULL;
2894 }
2895
2896 for (; ret == 0; ret = btrfs_backref_iter_next(iter)) {
2897 struct extent_buffer *eb;
2898 struct btrfs_key key;
2899 int type;
2900
2901 cond_resched();
2902 eb = btrfs_backref_get_eb(iter);
2903
2904 key.objectid = iter->bytenr;
2905 if (btrfs_backref_iter_is_inline_ref(iter)) {
2906 struct btrfs_extent_inline_ref *iref;
2907
2908 /* Update key for inline backref */
2909 iref = (struct btrfs_extent_inline_ref *)
2910 ((unsigned long)iter->cur_ptr);
2911 type = btrfs_get_extent_inline_ref_type(eb, iref,
2912 BTRFS_REF_TYPE_BLOCK);
2913 if (type == BTRFS_REF_TYPE_INVALID) {
2914 ret = -EUCLEAN;
2915 goto out;
2916 }
2917 key.type = type;
2918 key.offset = btrfs_extent_inline_ref_offset(eb, iref);
2919 } else {
2920 key.type = iter->cur_key.type;
2921 key.offset = iter->cur_key.offset;
2922 }
2923
2924 /*
2925 * Parent node found and matches current inline ref, no need to
2926 * rebuild this node for this inline ref
2927 */
2928 if (exist &&
2929 ((key.type == BTRFS_TREE_BLOCK_REF_KEY &&
2930 exist->owner == key.offset) ||
2931 (key.type == BTRFS_SHARED_BLOCK_REF_KEY &&
2932 exist->bytenr == key.offset))) {
2933 exist = NULL;
2934 continue;
2935 }
2936
2937 /* SHARED_BLOCK_REF means key.offset is the parent bytenr */
2938 if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
2939 ret = handle_direct_tree_backref(cache, &key, cur);
2940 if (ret < 0)
2941 goto out;
2942 continue;
2943 } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) {
2944 ret = -EINVAL;
2945 btrfs_print_v0_err(fs_info);
2946 btrfs_handle_fs_error(fs_info, ret, NULL);
2947 goto out;
2948 } else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) {
2949 continue;
2950 }
2951
2952 /*
2953 * key.type == BTRFS_TREE_BLOCK_REF_KEY, inline ref offset
2954 * means the root objectid. We need to search the tree to get
2955 * its parent bytenr.
2956 */
2957 ret = handle_indirect_tree_backref(cache, path, &key, node_key,
2958 cur);
2959 if (ret < 0)
2960 goto out;
2961 }
2962 ret = 0;
2963 cur->checked = 1;
2964 WARN_ON(exist);
2965out:
2966 btrfs_backref_iter_release(iter);
2967 return ret;
2968}
2969
2970/*
2971 * Finish the upwards linkage created by btrfs_backref_add_tree_node()
2972 */
2973int btrfs_backref_finish_upper_links(struct btrfs_backref_cache *cache,
2974 struct btrfs_backref_node *start)
2975{
2976 struct list_head *useless_node = &cache->useless_node;
2977 struct btrfs_backref_edge *edge;
2978 struct rb_node *rb_node;
2979 LIST_HEAD(pending_edge);
2980
2981 ASSERT(start->checked);
2982
2983 /* Insert this node to cache if it's not COW-only */
2984 if (!start->cowonly) {
2985 rb_node = rb_simple_insert(&cache->rb_root, start->bytenr,
2986 &start->rb_node);
2987 if (rb_node)
2988 btrfs_backref_panic(cache->fs_info, start->bytenr,
2989 -EEXIST);
2990 list_add_tail(&start->lower, &cache->leaves);
2991 }
2992
2993 /*
2994 * Use breadth first search to iterate all related edges.
2995 *
2996 * The starting points are all the edges of this node
2997 */
2998 list_for_each_entry(edge, &start->upper, list[LOWER])
2999 list_add_tail(&edge->list[UPPER], &pending_edge);
3000
3001 while (!list_empty(&pending_edge)) {
3002 struct btrfs_backref_node *upper;
3003 struct btrfs_backref_node *lower;
3004
3005 edge = list_first_entry(&pending_edge,
3006 struct btrfs_backref_edge, list[UPPER]);
3007 list_del_init(&edge->list[UPPER]);
3008 upper = edge->node[UPPER];
3009 lower = edge->node[LOWER];
3010
3011 /* Parent is detached, no need to keep any edges */
3012 if (upper->detached) {
3013 list_del(&edge->list[LOWER]);
3014 btrfs_backref_free_edge(cache, edge);
3015
3016 /* Lower node is orphan, queue for cleanup */
3017 if (list_empty(&lower->upper))
3018 list_add(&lower->list, useless_node);
3019 continue;
3020 }
3021
3022 /*
3023 * All new nodes added in current build_backref_tree() haven't
3024 * been linked to the cache rb tree.
3025 * So if we have upper->rb_node populated, this means a cache
3026 * hit. We only need to link the edge, as @upper and all its
3027 * parents have already been linked.
3028 */
3029 if (!RB_EMPTY_NODE(&upper->rb_node)) {
3030 if (upper->lowest) {
3031 list_del_init(&upper->lower);
3032 upper->lowest = 0;
3033 }
3034
3035 list_add_tail(&edge->list[UPPER], &upper->lower);
3036 continue;
3037 }
3038
3039 /* Sanity check, we shouldn't have any unchecked nodes */
3040 if (!upper->checked) {
3041 ASSERT(0);
3042 return -EUCLEAN;
3043 }
3044
3045 /* Sanity check, COW-only node has non-COW-only parent */
3046 if (start->cowonly != upper->cowonly) {
3047 ASSERT(0);
3048 return -EUCLEAN;
3049 }
3050
3051 /* Only cache non-COW-only (subvolume trees) tree blocks */
3052 if (!upper->cowonly) {
3053 rb_node = rb_simple_insert(&cache->rb_root, upper->bytenr,
3054 &upper->rb_node);
3055 if (rb_node) {
3056 btrfs_backref_panic(cache->fs_info,
3057 upper->bytenr, -EEXIST);
3058 return -EUCLEAN;
3059 }
3060 }
3061
3062 list_add_tail(&edge->list[UPPER], &upper->lower);
3063
3064 /*
3065 * Also queue all the parent edges of this uncached node
3066 * to finish the upper linkage
3067 */
3068 list_for_each_entry(edge, &upper->upper, list[LOWER])
3069 list_add_tail(&edge->list[UPPER], &pending_edge);
3070 }
3071 return 0;
3072}
3073
3074void btrfs_backref_error_cleanup(struct btrfs_backref_cache *cache,
3075 struct btrfs_backref_node *node)
3076{
3077 struct btrfs_backref_node *lower;
3078 struct btrfs_backref_node *upper;
3079 struct btrfs_backref_edge *edge;
3080
3081 while (!list_empty(&cache->useless_node)) {
3082 lower = list_first_entry(&cache->useless_node,
3083 struct btrfs_backref_node, list);
3084 list_del_init(&lower->list);
3085 }
3086 while (!list_empty(&cache->pending_edge)) {
3087 edge = list_first_entry(&cache->pending_edge,
3088 struct btrfs_backref_edge, list[UPPER]);
3089 list_del(&edge->list[UPPER]);
3090 list_del(&edge->list[LOWER]);
3091 lower = edge->node[LOWER];
3092 upper = edge->node[UPPER];
3093 btrfs_backref_free_edge(cache, edge);
3094
3095 /*
3096 * Lower is no longer linked to any upper backref nodes and
3097 * isn't in the cache, we can free it ourselves.
3098 */
3099 if (list_empty(&lower->upper) &&
3100 RB_EMPTY_NODE(&lower->rb_node))
3101 list_add(&lower->list, &cache->useless_node);
3102
3103 if (!RB_EMPTY_NODE(&upper->rb_node))
3104 continue;
3105
3106 /* Add this guy's upper edges to the list to process */
3107 list_for_each_entry(edge, &upper->upper, list[LOWER])
3108 list_add_tail(&edge->list[UPPER],
3109 &cache->pending_edge);
3110 if (list_empty(&upper->upper))
3111 list_add(&upper->list, &cache->useless_node);
3112 }
3113
3114 while (!list_empty(&cache->useless_node)) {
3115 lower = list_first_entry(&cache->useless_node,
3116 struct btrfs_backref_node, list);
3117 list_del_init(&lower->list);
3118 if (lower == node)
3119 node = NULL;
3120 btrfs_backref_drop_node(cache, lower);
3121 }
3122
3123 btrfs_backref_cleanup_node(cache, node);
3124 ASSERT(list_empty(&cache->useless_node) &&
3125 list_empty(&cache->pending_edge));
3126}