2 * Copyright (C) 2011 STRATO. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/vmalloc.h>
24 #include "transaction.h"
25 #include "delayed-ref.h"
28 /* Just an arbitrary number so we can be sure this happened */
29 #define BACKREF_FOUND_SHARED 6
31 struct extent_inode_elem {
34 struct extent_inode_elem *next;
37 static int check_extent_in_eb(struct btrfs_key *key, struct extent_buffer *eb,
38 struct btrfs_file_extent_item *fi,
40 struct extent_inode_elem **eie)
43 struct extent_inode_elem *e;
45 if (!btrfs_file_extent_compression(eb, fi) &&
46 !btrfs_file_extent_encryption(eb, fi) &&
47 !btrfs_file_extent_other_encoding(eb, fi)) {
51 data_offset = btrfs_file_extent_offset(eb, fi);
52 data_len = btrfs_file_extent_num_bytes(eb, fi);
54 if (extent_item_pos < data_offset ||
55 extent_item_pos >= data_offset + data_len)
57 offset = extent_item_pos - data_offset;
60 e = kmalloc(sizeof(*e), GFP_NOFS);
65 e->inum = key->objectid;
66 e->offset = key->offset + offset;
72 static void free_inode_elem_list(struct extent_inode_elem *eie)
74 struct extent_inode_elem *eie_next;
76 for (; eie; eie = eie_next) {
82 static int find_extent_in_eb(struct extent_buffer *eb, u64 wanted_disk_byte,
84 struct extent_inode_elem **eie)
88 struct btrfs_file_extent_item *fi;
95 * from the shared data ref, we only have the leaf but we need
96 * the key. thus, we must look into all items and see that we
97 * find one (some) with a reference to our extent item.
99 nritems = btrfs_header_nritems(eb);
100 for (slot = 0; slot < nritems; ++slot) {
101 btrfs_item_key_to_cpu(eb, &key, slot);
102 if (key.type != BTRFS_EXTENT_DATA_KEY)
104 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
105 extent_type = btrfs_file_extent_type(eb, fi);
106 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
108 /* don't skip BTRFS_FILE_EXTENT_PREALLOC, we can handle that */
109 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
110 if (disk_byte != wanted_disk_byte)
113 ret = check_extent_in_eb(&key, eb, fi, extent_item_pos, eie);
122 * this structure records all encountered refs on the way up to the root
124 struct __prelim_ref {
125 struct list_head list;
127 struct btrfs_key key_for_search;
130 struct extent_inode_elem *inode_list;
132 u64 wanted_disk_byte;
135 static struct kmem_cache *btrfs_prelim_ref_cache;
137 int __init btrfs_prelim_ref_init(void)
139 btrfs_prelim_ref_cache = kmem_cache_create("btrfs_prelim_ref",
140 sizeof(struct __prelim_ref),
142 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
144 if (!btrfs_prelim_ref_cache)
149 void btrfs_prelim_ref_exit(void)
151 if (btrfs_prelim_ref_cache)
152 kmem_cache_destroy(btrfs_prelim_ref_cache);
156 * the rules for all callers of this function are:
157 * - obtaining the parent is the goal
158 * - if you add a key, you must know that it is a correct key
159 * - if you cannot add the parent or a correct key, then we will look into the
160 * block later to set a correct key
164 * backref type | shared | indirect | shared | indirect
165 * information | tree | tree | data | data
166 * --------------------+--------+----------+--------+----------
167 * parent logical | y | - | - | -
168 * key to resolve | - | y | y | y
169 * tree block logical | - | - | - | -
170 * root for resolving | y | y | y | y
172 * - column 1: we've the parent -> done
173 * - column 2, 3, 4: we use the key to find the parent
175 * on disk refs (inline or keyed)
176 * ==============================
177 * backref type | shared | indirect | shared | indirect
178 * information | tree | tree | data | data
179 * --------------------+--------+----------+--------+----------
180 * parent logical | y | - | y | -
181 * key to resolve | - | - | - | y
182 * tree block logical | y | y | y | y
183 * root for resolving | - | y | y | y
185 * - column 1, 3: we've the parent -> done
186 * - column 2: we take the first key from the block to find the parent
187 * (see __add_missing_keys)
188 * - column 4: we use the key to find the parent
190 * additional information that's available but not required to find the parent
191 * block might help in merging entries to gain some speed.
194 static int __add_prelim_ref(struct list_head *head, u64 root_id,
195 struct btrfs_key *key, int level,
196 u64 parent, u64 wanted_disk_byte, int count,
199 struct __prelim_ref *ref;
201 if (root_id == BTRFS_DATA_RELOC_TREE_OBJECTID)
204 ref = kmem_cache_alloc(btrfs_prelim_ref_cache, gfp_mask);
208 ref->root_id = root_id;
210 ref->key_for_search = *key;
212 memset(&ref->key_for_search, 0, sizeof(ref->key_for_search));
214 ref->inode_list = NULL;
217 ref->parent = parent;
218 ref->wanted_disk_byte = wanted_disk_byte;
219 list_add_tail(&ref->list, head);
224 static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path,
225 struct ulist *parents, struct __prelim_ref *ref,
226 int level, u64 time_seq, const u64 *extent_item_pos,
231 struct extent_buffer *eb;
232 struct btrfs_key key;
233 struct btrfs_key *key_for_search = &ref->key_for_search;
234 struct btrfs_file_extent_item *fi;
235 struct extent_inode_elem *eie = NULL, *old = NULL;
237 u64 wanted_disk_byte = ref->wanted_disk_byte;
241 eb = path->nodes[level];
242 ret = ulist_add(parents, eb->start, 0, GFP_NOFS);
249 * We normally enter this function with the path already pointing to
250 * the first item to check. But sometimes, we may enter it with
251 * slot==nritems. In that case, go to the next leaf before we continue.
253 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0]))
254 ret = btrfs_next_old_leaf(root, path, time_seq);
256 while (!ret && count < total_refs) {
258 slot = path->slots[0];
260 btrfs_item_key_to_cpu(eb, &key, slot);
262 if (key.objectid != key_for_search->objectid ||
263 key.type != BTRFS_EXTENT_DATA_KEY)
266 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
267 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
269 if (disk_byte == wanted_disk_byte) {
273 if (extent_item_pos) {
274 ret = check_extent_in_eb(&key, eb, fi,
282 ret = ulist_add_merge_ptr(parents, eb->start,
283 eie, (void **)&old, GFP_NOFS);
286 if (!ret && extent_item_pos) {
294 ret = btrfs_next_old_item(root, path, time_seq);
300 free_inode_elem_list(eie);
305 * resolve an indirect backref in the form (root_id, key, level)
306 * to a logical address
308 static int __resolve_indirect_ref(struct btrfs_fs_info *fs_info,
309 struct btrfs_path *path, u64 time_seq,
310 struct __prelim_ref *ref,
311 struct ulist *parents,
312 const u64 *extent_item_pos, u64 total_refs)
314 struct btrfs_root *root;
315 struct btrfs_key root_key;
316 struct extent_buffer *eb;
319 int level = ref->level;
322 root_key.objectid = ref->root_id;
323 root_key.type = BTRFS_ROOT_ITEM_KEY;
324 root_key.offset = (u64)-1;
326 index = srcu_read_lock(&fs_info->subvol_srcu);
328 root = btrfs_read_fs_root_no_name(fs_info, &root_key);
330 srcu_read_unlock(&fs_info->subvol_srcu, index);
335 if (path->search_commit_root)
336 root_level = btrfs_header_level(root->commit_root);
338 root_level = btrfs_old_root_level(root, time_seq);
340 if (root_level + 1 == level) {
341 srcu_read_unlock(&fs_info->subvol_srcu, index);
345 path->lowest_level = level;
346 ret = btrfs_search_old_slot(root, &ref->key_for_search, path, time_seq);
348 /* root node has been locked, we can release @subvol_srcu safely here */
349 srcu_read_unlock(&fs_info->subvol_srcu, index);
351 pr_debug("search slot in root %llu (level %d, ref count %d) returned "
352 "%d for key (%llu %u %llu)\n",
353 ref->root_id, level, ref->count, ret,
354 ref->key_for_search.objectid, ref->key_for_search.type,
355 ref->key_for_search.offset);
359 eb = path->nodes[level];
361 if (WARN_ON(!level)) {
366 eb = path->nodes[level];
369 ret = add_all_parents(root, path, parents, ref, level, time_seq,
370 extent_item_pos, total_refs);
372 path->lowest_level = 0;
373 btrfs_release_path(path);
378 * resolve all indirect backrefs from the list
380 static int __resolve_indirect_refs(struct btrfs_fs_info *fs_info,
381 struct btrfs_path *path, u64 time_seq,
382 struct list_head *head,
383 const u64 *extent_item_pos, u64 total_refs,
388 struct __prelim_ref *ref;
389 struct __prelim_ref *ref_safe;
390 struct __prelim_ref *new_ref;
391 struct ulist *parents;
392 struct ulist_node *node;
393 struct ulist_iterator uiter;
395 parents = ulist_alloc(GFP_NOFS);
400 * _safe allows us to insert directly after the current item without
401 * iterating over the newly inserted items.
402 * we're also allowed to re-assign ref during iteration.
404 list_for_each_entry_safe(ref, ref_safe, head, list) {
405 if (ref->parent) /* already direct */
409 if (root_objectid && ref->root_id != root_objectid) {
410 ret = BACKREF_FOUND_SHARED;
413 err = __resolve_indirect_ref(fs_info, path, time_seq, ref,
414 parents, extent_item_pos,
417 * we can only tolerate ENOENT,otherwise,we should catch error
418 * and return directly.
420 if (err == -ENOENT) {
427 /* we put the first parent into the ref at hand */
428 ULIST_ITER_INIT(&uiter);
429 node = ulist_next(parents, &uiter);
430 ref->parent = node ? node->val : 0;
431 ref->inode_list = node ?
432 (struct extent_inode_elem *)(uintptr_t)node->aux : NULL;
434 /* additional parents require new refs being added here */
435 while ((node = ulist_next(parents, &uiter))) {
436 new_ref = kmem_cache_alloc(btrfs_prelim_ref_cache,
442 memcpy(new_ref, ref, sizeof(*ref));
443 new_ref->parent = node->val;
444 new_ref->inode_list = (struct extent_inode_elem *)
445 (uintptr_t)node->aux;
446 list_add(&new_ref->list, &ref->list);
448 ulist_reinit(parents);
455 static inline int ref_for_same_block(struct __prelim_ref *ref1,
456 struct __prelim_ref *ref2)
458 if (ref1->level != ref2->level)
460 if (ref1->root_id != ref2->root_id)
462 if (ref1->key_for_search.type != ref2->key_for_search.type)
464 if (ref1->key_for_search.objectid != ref2->key_for_search.objectid)
466 if (ref1->key_for_search.offset != ref2->key_for_search.offset)
468 if (ref1->parent != ref2->parent)
475 * read tree blocks and add keys where required.
477 static int __add_missing_keys(struct btrfs_fs_info *fs_info,
478 struct list_head *head)
480 struct list_head *pos;
481 struct extent_buffer *eb;
483 list_for_each(pos, head) {
484 struct __prelim_ref *ref;
485 ref = list_entry(pos, struct __prelim_ref, list);
489 if (ref->key_for_search.type)
491 BUG_ON(!ref->wanted_disk_byte);
492 eb = read_tree_block(fs_info->tree_root, ref->wanted_disk_byte,
494 if (!eb || !extent_buffer_uptodate(eb)) {
495 free_extent_buffer(eb);
498 btrfs_tree_read_lock(eb);
499 if (btrfs_header_level(eb) == 0)
500 btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
502 btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
503 btrfs_tree_read_unlock(eb);
504 free_extent_buffer(eb);
510 * merge two lists of backrefs and adjust counts accordingly
512 * mode = 1: merge identical keys, if key is set
513 * FIXME: if we add more keys in __add_prelim_ref, we can merge more here.
514 * additionally, we could even add a key range for the blocks we
515 * looked into to merge even more (-> replace unresolved refs by those
517 * mode = 2: merge identical parents
519 static void __merge_refs(struct list_head *head, int mode)
521 struct list_head *pos1;
523 list_for_each(pos1, head) {
524 struct list_head *n2;
525 struct list_head *pos2;
526 struct __prelim_ref *ref1;
528 ref1 = list_entry(pos1, struct __prelim_ref, list);
530 for (pos2 = pos1->next, n2 = pos2->next; pos2 != head;
531 pos2 = n2, n2 = pos2->next) {
532 struct __prelim_ref *ref2;
533 struct __prelim_ref *xchg;
534 struct extent_inode_elem *eie;
536 ref2 = list_entry(pos2, struct __prelim_ref, list);
539 if (!ref_for_same_block(ref1, ref2))
541 if (!ref1->parent && ref2->parent) {
547 if (ref1->parent != ref2->parent)
551 eie = ref1->inode_list;
552 while (eie && eie->next)
555 eie->next = ref2->inode_list;
557 ref1->inode_list = ref2->inode_list;
558 ref1->count += ref2->count;
560 list_del(&ref2->list);
561 kmem_cache_free(btrfs_prelim_ref_cache, ref2);
568 * add all currently queued delayed refs from this head whose seq nr is
569 * smaller or equal that seq to the list
571 static int __add_delayed_refs(struct btrfs_delayed_ref_head *head, u64 seq,
572 struct list_head *prefs, u64 *total_refs,
575 struct btrfs_delayed_extent_op *extent_op = head->extent_op;
576 struct rb_node *n = &head->node.rb_node;
577 struct btrfs_key key;
578 struct btrfs_key op_key = {0};
582 if (extent_op && extent_op->update_key)
583 btrfs_disk_key_to_cpu(&op_key, &extent_op->key);
585 spin_lock(&head->lock);
586 n = rb_first(&head->ref_root);
588 struct btrfs_delayed_ref_node *node;
589 node = rb_entry(n, struct btrfs_delayed_ref_node,
595 switch (node->action) {
596 case BTRFS_ADD_DELAYED_EXTENT:
597 case BTRFS_UPDATE_DELAYED_HEAD:
600 case BTRFS_ADD_DELAYED_REF:
603 case BTRFS_DROP_DELAYED_REF:
609 *total_refs += (node->ref_mod * sgn);
610 switch (node->type) {
611 case BTRFS_TREE_BLOCK_REF_KEY: {
612 struct btrfs_delayed_tree_ref *ref;
614 ref = btrfs_delayed_node_to_tree_ref(node);
615 ret = __add_prelim_ref(prefs, ref->root, &op_key,
616 ref->level + 1, 0, node->bytenr,
617 node->ref_mod * sgn, GFP_ATOMIC);
620 case BTRFS_SHARED_BLOCK_REF_KEY: {
621 struct btrfs_delayed_tree_ref *ref;
623 ref = btrfs_delayed_node_to_tree_ref(node);
624 ret = __add_prelim_ref(prefs, ref->root, NULL,
625 ref->level + 1, ref->parent,
627 node->ref_mod * sgn, GFP_ATOMIC);
630 case BTRFS_EXTENT_DATA_REF_KEY: {
631 struct btrfs_delayed_data_ref *ref;
632 ref = btrfs_delayed_node_to_data_ref(node);
634 key.objectid = ref->objectid;
635 key.type = BTRFS_EXTENT_DATA_KEY;
636 key.offset = ref->offset;
639 * Found a inum that doesn't match our known inum, we
642 if (inum && ref->objectid != inum) {
643 ret = BACKREF_FOUND_SHARED;
647 ret = __add_prelim_ref(prefs, ref->root, &key, 0, 0,
649 node->ref_mod * sgn, GFP_ATOMIC);
652 case BTRFS_SHARED_DATA_REF_KEY: {
653 struct btrfs_delayed_data_ref *ref;
655 ref = btrfs_delayed_node_to_data_ref(node);
657 key.objectid = ref->objectid;
658 key.type = BTRFS_EXTENT_DATA_KEY;
659 key.offset = ref->offset;
660 ret = __add_prelim_ref(prefs, ref->root, &key, 0,
661 ref->parent, node->bytenr,
662 node->ref_mod * sgn, GFP_ATOMIC);
671 spin_unlock(&head->lock);
676 * add all inline backrefs for bytenr to the list
678 static int __add_inline_refs(struct btrfs_fs_info *fs_info,
679 struct btrfs_path *path, u64 bytenr,
680 int *info_level, struct list_head *prefs,
681 u64 *total_refs, u64 inum)
685 struct extent_buffer *leaf;
686 struct btrfs_key key;
687 struct btrfs_key found_key;
690 struct btrfs_extent_item *ei;
695 * enumerate all inline refs
697 leaf = path->nodes[0];
698 slot = path->slots[0];
700 item_size = btrfs_item_size_nr(leaf, slot);
701 BUG_ON(item_size < sizeof(*ei));
703 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
704 flags = btrfs_extent_flags(leaf, ei);
705 *total_refs += btrfs_extent_refs(leaf, ei);
706 btrfs_item_key_to_cpu(leaf, &found_key, slot);
708 ptr = (unsigned long)(ei + 1);
709 end = (unsigned long)ei + item_size;
711 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
712 flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
713 struct btrfs_tree_block_info *info;
715 info = (struct btrfs_tree_block_info *)ptr;
716 *info_level = btrfs_tree_block_level(leaf, info);
717 ptr += sizeof(struct btrfs_tree_block_info);
719 } else if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
720 *info_level = found_key.offset;
722 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
726 struct btrfs_extent_inline_ref *iref;
730 iref = (struct btrfs_extent_inline_ref *)ptr;
731 type = btrfs_extent_inline_ref_type(leaf, iref);
732 offset = btrfs_extent_inline_ref_offset(leaf, iref);
735 case BTRFS_SHARED_BLOCK_REF_KEY:
736 ret = __add_prelim_ref(prefs, 0, NULL,
737 *info_level + 1, offset,
738 bytenr, 1, GFP_NOFS);
740 case BTRFS_SHARED_DATA_REF_KEY: {
741 struct btrfs_shared_data_ref *sdref;
744 sdref = (struct btrfs_shared_data_ref *)(iref + 1);
745 count = btrfs_shared_data_ref_count(leaf, sdref);
746 ret = __add_prelim_ref(prefs, 0, NULL, 0, offset,
747 bytenr, count, GFP_NOFS);
750 case BTRFS_TREE_BLOCK_REF_KEY:
751 ret = __add_prelim_ref(prefs, offset, NULL,
753 bytenr, 1, GFP_NOFS);
755 case BTRFS_EXTENT_DATA_REF_KEY: {
756 struct btrfs_extent_data_ref *dref;
760 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
761 count = btrfs_extent_data_ref_count(leaf, dref);
762 key.objectid = btrfs_extent_data_ref_objectid(leaf,
764 key.type = BTRFS_EXTENT_DATA_KEY;
765 key.offset = btrfs_extent_data_ref_offset(leaf, dref);
767 if (inum && key.objectid != inum) {
768 ret = BACKREF_FOUND_SHARED;
772 root = btrfs_extent_data_ref_root(leaf, dref);
773 ret = __add_prelim_ref(prefs, root, &key, 0, 0,
774 bytenr, count, GFP_NOFS);
782 ptr += btrfs_extent_inline_ref_size(type);
789 * add all non-inline backrefs for bytenr to the list
791 static int __add_keyed_refs(struct btrfs_fs_info *fs_info,
792 struct btrfs_path *path, u64 bytenr,
793 int info_level, struct list_head *prefs, u64 inum)
795 struct btrfs_root *extent_root = fs_info->extent_root;
798 struct extent_buffer *leaf;
799 struct btrfs_key key;
802 ret = btrfs_next_item(extent_root, path);
810 slot = path->slots[0];
811 leaf = path->nodes[0];
812 btrfs_item_key_to_cpu(leaf, &key, slot);
814 if (key.objectid != bytenr)
816 if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
818 if (key.type > BTRFS_SHARED_DATA_REF_KEY)
822 case BTRFS_SHARED_BLOCK_REF_KEY:
823 ret = __add_prelim_ref(prefs, 0, NULL,
824 info_level + 1, key.offset,
825 bytenr, 1, GFP_NOFS);
827 case BTRFS_SHARED_DATA_REF_KEY: {
828 struct btrfs_shared_data_ref *sdref;
831 sdref = btrfs_item_ptr(leaf, slot,
832 struct btrfs_shared_data_ref);
833 count = btrfs_shared_data_ref_count(leaf, sdref);
834 ret = __add_prelim_ref(prefs, 0, NULL, 0, key.offset,
835 bytenr, count, GFP_NOFS);
838 case BTRFS_TREE_BLOCK_REF_KEY:
839 ret = __add_prelim_ref(prefs, key.offset, NULL,
841 bytenr, 1, GFP_NOFS);
843 case BTRFS_EXTENT_DATA_REF_KEY: {
844 struct btrfs_extent_data_ref *dref;
848 dref = btrfs_item_ptr(leaf, slot,
849 struct btrfs_extent_data_ref);
850 count = btrfs_extent_data_ref_count(leaf, dref);
851 key.objectid = btrfs_extent_data_ref_objectid(leaf,
853 key.type = BTRFS_EXTENT_DATA_KEY;
854 key.offset = btrfs_extent_data_ref_offset(leaf, dref);
856 if (inum && key.objectid != inum) {
857 ret = BACKREF_FOUND_SHARED;
861 root = btrfs_extent_data_ref_root(leaf, dref);
862 ret = __add_prelim_ref(prefs, root, &key, 0, 0,
863 bytenr, count, GFP_NOFS);
878 * this adds all existing backrefs (inline backrefs, backrefs and delayed
879 * refs) for the given bytenr to the refs list, merges duplicates and resolves
880 * indirect refs to their parent bytenr.
881 * When roots are found, they're added to the roots list
883 * FIXME some caching might speed things up
885 static int find_parent_nodes(struct btrfs_trans_handle *trans,
886 struct btrfs_fs_info *fs_info, u64 bytenr,
887 u64 time_seq, struct ulist *refs,
888 struct ulist *roots, const u64 *extent_item_pos,
889 u64 root_objectid, u64 inum)
891 struct btrfs_key key;
892 struct btrfs_path *path;
893 struct btrfs_delayed_ref_root *delayed_refs = NULL;
894 struct btrfs_delayed_ref_head *head;
897 struct list_head prefs_delayed;
898 struct list_head prefs;
899 struct __prelim_ref *ref;
900 struct extent_inode_elem *eie = NULL;
903 INIT_LIST_HEAD(&prefs);
904 INIT_LIST_HEAD(&prefs_delayed);
906 key.objectid = bytenr;
907 key.offset = (u64)-1;
908 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
909 key.type = BTRFS_METADATA_ITEM_KEY;
911 key.type = BTRFS_EXTENT_ITEM_KEY;
913 path = btrfs_alloc_path();
917 path->search_commit_root = 1;
918 path->skip_locking = 1;
922 * grab both a lock on the path and a lock on the delayed ref head.
923 * We need both to get a consistent picture of how the refs look
924 * at a specified point in time
929 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
934 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
935 if (trans && likely(trans->type != __TRANS_DUMMY)) {
940 * look if there are updates for this ref queued and lock the
943 delayed_refs = &trans->transaction->delayed_refs;
944 spin_lock(&delayed_refs->lock);
945 head = btrfs_find_delayed_ref_head(trans, bytenr);
947 if (!mutex_trylock(&head->mutex)) {
948 atomic_inc(&head->node.refs);
949 spin_unlock(&delayed_refs->lock);
951 btrfs_release_path(path);
954 * Mutex was contended, block until it's
955 * released and try again
957 mutex_lock(&head->mutex);
958 mutex_unlock(&head->mutex);
959 btrfs_put_delayed_ref(&head->node);
962 spin_unlock(&delayed_refs->lock);
963 ret = __add_delayed_refs(head, time_seq,
964 &prefs_delayed, &total_refs,
966 mutex_unlock(&head->mutex);
970 spin_unlock(&delayed_refs->lock);
974 if (path->slots[0]) {
975 struct extent_buffer *leaf;
979 leaf = path->nodes[0];
980 slot = path->slots[0];
981 btrfs_item_key_to_cpu(leaf, &key, slot);
982 if (key.objectid == bytenr &&
983 (key.type == BTRFS_EXTENT_ITEM_KEY ||
984 key.type == BTRFS_METADATA_ITEM_KEY)) {
985 ret = __add_inline_refs(fs_info, path, bytenr,
990 ret = __add_keyed_refs(fs_info, path, bytenr,
991 info_level, &prefs, inum);
996 btrfs_release_path(path);
998 list_splice_init(&prefs_delayed, &prefs);
1000 ret = __add_missing_keys(fs_info, &prefs);
1004 __merge_refs(&prefs, 1);
1006 ret = __resolve_indirect_refs(fs_info, path, time_seq, &prefs,
1007 extent_item_pos, total_refs,
1012 __merge_refs(&prefs, 2);
1014 while (!list_empty(&prefs)) {
1015 ref = list_first_entry(&prefs, struct __prelim_ref, list);
1016 WARN_ON(ref->count < 0);
1017 if (roots && ref->count && ref->root_id && ref->parent == 0) {
1018 if (root_objectid && ref->root_id != root_objectid) {
1019 ret = BACKREF_FOUND_SHARED;
1023 /* no parent == root of tree */
1024 ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS);
1028 if (ref->count && ref->parent) {
1029 if (extent_item_pos && !ref->inode_list &&
1031 struct extent_buffer *eb;
1033 eb = read_tree_block(fs_info->extent_root,
1035 if (!eb || !extent_buffer_uptodate(eb)) {
1036 free_extent_buffer(eb);
1040 btrfs_tree_read_lock(eb);
1041 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1042 ret = find_extent_in_eb(eb, bytenr,
1043 *extent_item_pos, &eie);
1044 btrfs_tree_read_unlock_blocking(eb);
1045 free_extent_buffer(eb);
1048 ref->inode_list = eie;
1050 ret = ulist_add_merge_ptr(refs, ref->parent,
1052 (void **)&eie, GFP_NOFS);
1055 if (!ret && extent_item_pos) {
1057 * we've recorded that parent, so we must extend
1058 * its inode list here
1063 eie->next = ref->inode_list;
1067 list_del(&ref->list);
1068 kmem_cache_free(btrfs_prelim_ref_cache, ref);
1072 btrfs_free_path(path);
1073 while (!list_empty(&prefs)) {
1074 ref = list_first_entry(&prefs, struct __prelim_ref, list);
1075 list_del(&ref->list);
1076 kmem_cache_free(btrfs_prelim_ref_cache, ref);
1078 while (!list_empty(&prefs_delayed)) {
1079 ref = list_first_entry(&prefs_delayed, struct __prelim_ref,
1081 list_del(&ref->list);
1082 kmem_cache_free(btrfs_prelim_ref_cache, ref);
1085 free_inode_elem_list(eie);
1089 static void free_leaf_list(struct ulist *blocks)
1091 struct ulist_node *node = NULL;
1092 struct extent_inode_elem *eie;
1093 struct ulist_iterator uiter;
1095 ULIST_ITER_INIT(&uiter);
1096 while ((node = ulist_next(blocks, &uiter))) {
1099 eie = (struct extent_inode_elem *)(uintptr_t)node->aux;
1100 free_inode_elem_list(eie);
1108 * Finds all leafs with a reference to the specified combination of bytenr and
1109 * offset. key_list_head will point to a list of corresponding keys (caller must
1110 * free each list element). The leafs will be stored in the leafs ulist, which
1111 * must be freed with ulist_free.
1113 * returns 0 on success, <0 on error
1115 static int btrfs_find_all_leafs(struct btrfs_trans_handle *trans,
1116 struct btrfs_fs_info *fs_info, u64 bytenr,
1117 u64 time_seq, struct ulist **leafs,
1118 const u64 *extent_item_pos)
1122 *leafs = ulist_alloc(GFP_NOFS);
1126 ret = find_parent_nodes(trans, fs_info, bytenr,
1127 time_seq, *leafs, NULL, extent_item_pos, 0, 0);
1128 if (ret < 0 && ret != -ENOENT) {
1129 free_leaf_list(*leafs);
1137 * walk all backrefs for a given extent to find all roots that reference this
1138 * extent. Walking a backref means finding all extents that reference this
1139 * extent and in turn walk the backrefs of those, too. Naturally this is a
1140 * recursive process, but here it is implemented in an iterative fashion: We
1141 * find all referencing extents for the extent in question and put them on a
1142 * list. In turn, we find all referencing extents for those, further appending
1143 * to the list. The way we iterate the list allows adding more elements after
1144 * the current while iterating. The process stops when we reach the end of the
1145 * list. Found roots are added to the roots list.
1147 * returns 0 on success, < 0 on error.
1149 static int __btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1150 struct btrfs_fs_info *fs_info, u64 bytenr,
1151 u64 time_seq, struct ulist **roots)
1154 struct ulist_node *node = NULL;
1155 struct ulist_iterator uiter;
1158 tmp = ulist_alloc(GFP_NOFS);
1161 *roots = ulist_alloc(GFP_NOFS);
1167 ULIST_ITER_INIT(&uiter);
1169 ret = find_parent_nodes(trans, fs_info, bytenr,
1170 time_seq, tmp, *roots, NULL, 0, 0);
1171 if (ret < 0 && ret != -ENOENT) {
1176 node = ulist_next(tmp, &uiter);
1187 int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1188 struct btrfs_fs_info *fs_info, u64 bytenr,
1189 u64 time_seq, struct ulist **roots)
1194 down_read(&fs_info->commit_root_sem);
1195 ret = __btrfs_find_all_roots(trans, fs_info, bytenr, time_seq, roots);
1197 up_read(&fs_info->commit_root_sem);
1201 int btrfs_check_shared(struct btrfs_trans_handle *trans,
1202 struct btrfs_fs_info *fs_info, u64 root_objectid,
1203 u64 inum, u64 bytenr)
1205 struct ulist *tmp = NULL;
1206 struct ulist *roots = NULL;
1207 struct ulist_iterator uiter;
1208 struct ulist_node *node;
1209 struct seq_list elem = {};
1212 tmp = ulist_alloc(GFP_NOFS);
1213 roots = ulist_alloc(GFP_NOFS);
1214 if (!tmp || !roots) {
1221 btrfs_get_tree_mod_seq(fs_info, &elem);
1223 down_read(&fs_info->commit_root_sem);
1224 ULIST_ITER_INIT(&uiter);
1226 ret = find_parent_nodes(trans, fs_info, bytenr, elem.seq, tmp,
1227 roots, NULL, root_objectid, inum);
1228 if (ret == BACKREF_FOUND_SHARED) {
1232 if (ret < 0 && ret != -ENOENT)
1234 node = ulist_next(tmp, &uiter);
1241 btrfs_put_tree_mod_seq(fs_info, &elem);
1243 up_read(&fs_info->commit_root_sem);
1250 * this makes the path point to (inum INODE_ITEM ioff)
1252 int inode_item_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
1253 struct btrfs_path *path)
1255 struct btrfs_key key;
1256 return btrfs_find_item(fs_root, path, inum, ioff,
1257 BTRFS_INODE_ITEM_KEY, &key);
1260 static int inode_ref_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
1261 struct btrfs_path *path,
1262 struct btrfs_key *found_key)
1264 return btrfs_find_item(fs_root, path, inum, ioff,
1265 BTRFS_INODE_REF_KEY, found_key);
1268 int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,
1269 u64 start_off, struct btrfs_path *path,
1270 struct btrfs_inode_extref **ret_extref,
1274 struct btrfs_key key;
1275 struct btrfs_key found_key;
1276 struct btrfs_inode_extref *extref;
1277 struct extent_buffer *leaf;
1280 key.objectid = inode_objectid;
1281 key.type = BTRFS_INODE_EXTREF_KEY;
1282 key.offset = start_off;
1284 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1289 leaf = path->nodes[0];
1290 slot = path->slots[0];
1291 if (slot >= btrfs_header_nritems(leaf)) {
1293 * If the item at offset is not found,
1294 * btrfs_search_slot will point us to the slot
1295 * where it should be inserted. In our case
1296 * that will be the slot directly before the
1297 * next INODE_REF_KEY_V2 item. In the case
1298 * that we're pointing to the last slot in a
1299 * leaf, we must move one leaf over.
1301 ret = btrfs_next_leaf(root, path);
1310 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1313 * Check that we're still looking at an extended ref key for
1314 * this particular objectid. If we have different
1315 * objectid or type then there are no more to be found
1316 * in the tree and we can exit.
1319 if (found_key.objectid != inode_objectid)
1321 if (found_key.type != BTRFS_INODE_EXTREF_KEY)
1325 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1326 extref = (struct btrfs_inode_extref *)ptr;
1327 *ret_extref = extref;
1329 *found_off = found_key.offset;
1337 * this iterates to turn a name (from iref/extref) into a full filesystem path.
1338 * Elements of the path are separated by '/' and the path is guaranteed to be
1339 * 0-terminated. the path is only given within the current file system.
1340 * Therefore, it never starts with a '/'. the caller is responsible to provide
1341 * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
1342 * the start point of the resulting string is returned. this pointer is within
1344 * in case the path buffer would overflow, the pointer is decremented further
1345 * as if output was written to the buffer, though no more output is actually
1346 * generated. that way, the caller can determine how much space would be
1347 * required for the path to fit into the buffer. in that case, the returned
1348 * value will be smaller than dest. callers must check this!
1350 char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
1351 u32 name_len, unsigned long name_off,
1352 struct extent_buffer *eb_in, u64 parent,
1353 char *dest, u32 size)
1358 s64 bytes_left = ((s64)size) - 1;
1359 struct extent_buffer *eb = eb_in;
1360 struct btrfs_key found_key;
1361 int leave_spinning = path->leave_spinning;
1362 struct btrfs_inode_ref *iref;
1364 if (bytes_left >= 0)
1365 dest[bytes_left] = '\0';
1367 path->leave_spinning = 1;
1369 bytes_left -= name_len;
1370 if (bytes_left >= 0)
1371 read_extent_buffer(eb, dest + bytes_left,
1372 name_off, name_len);
1374 btrfs_tree_read_unlock_blocking(eb);
1375 free_extent_buffer(eb);
1377 ret = inode_ref_info(parent, 0, fs_root, path, &found_key);
1383 next_inum = found_key.offset;
1385 /* regular exit ahead */
1386 if (parent == next_inum)
1389 slot = path->slots[0];
1390 eb = path->nodes[0];
1391 /* make sure we can use eb after releasing the path */
1393 atomic_inc(&eb->refs);
1394 btrfs_tree_read_lock(eb);
1395 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1397 btrfs_release_path(path);
1398 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1400 name_len = btrfs_inode_ref_name_len(eb, iref);
1401 name_off = (unsigned long)(iref + 1);
1405 if (bytes_left >= 0)
1406 dest[bytes_left] = '/';
1409 btrfs_release_path(path);
1410 path->leave_spinning = leave_spinning;
1413 return ERR_PTR(ret);
1415 return dest + bytes_left;
1419 * this makes the path point to (logical EXTENT_ITEM *)
1420 * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
1421 * tree blocks and <0 on error.
1423 int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
1424 struct btrfs_path *path, struct btrfs_key *found_key,
1431 struct extent_buffer *eb;
1432 struct btrfs_extent_item *ei;
1433 struct btrfs_key key;
1435 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1436 key.type = BTRFS_METADATA_ITEM_KEY;
1438 key.type = BTRFS_EXTENT_ITEM_KEY;
1439 key.objectid = logical;
1440 key.offset = (u64)-1;
1442 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
1446 ret = btrfs_previous_extent_item(fs_info->extent_root, path, 0);
1452 btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]);
1453 if (found_key->type == BTRFS_METADATA_ITEM_KEY)
1454 size = fs_info->extent_root->nodesize;
1455 else if (found_key->type == BTRFS_EXTENT_ITEM_KEY)
1456 size = found_key->offset;
1458 if (found_key->objectid > logical ||
1459 found_key->objectid + size <= logical) {
1460 pr_debug("logical %llu is not within any extent\n", logical);
1464 eb = path->nodes[0];
1465 item_size = btrfs_item_size_nr(eb, path->slots[0]);
1466 BUG_ON(item_size < sizeof(*ei));
1468 ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
1469 flags = btrfs_extent_flags(eb, ei);
1471 pr_debug("logical %llu is at position %llu within the extent (%llu "
1472 "EXTENT_ITEM %llu) flags %#llx size %u\n",
1473 logical, logical - found_key->objectid, found_key->objectid,
1474 found_key->offset, flags, item_size);
1476 WARN_ON(!flags_ret);
1478 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1479 *flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK;
1480 else if (flags & BTRFS_EXTENT_FLAG_DATA)
1481 *flags_ret = BTRFS_EXTENT_FLAG_DATA;
1491 * helper function to iterate extent inline refs. ptr must point to a 0 value
1492 * for the first call and may be modified. it is used to track state.
1493 * if more refs exist, 0 is returned and the next call to
1494 * __get_extent_inline_ref must pass the modified ptr parameter to get the
1495 * next ref. after the last ref was processed, 1 is returned.
1496 * returns <0 on error
1498 static int __get_extent_inline_ref(unsigned long *ptr, struct extent_buffer *eb,
1499 struct btrfs_key *key,
1500 struct btrfs_extent_item *ei, u32 item_size,
1501 struct btrfs_extent_inline_ref **out_eiref,
1506 struct btrfs_tree_block_info *info;
1510 flags = btrfs_extent_flags(eb, ei);
1511 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1512 if (key->type == BTRFS_METADATA_ITEM_KEY) {
1513 /* a skinny metadata extent */
1515 (struct btrfs_extent_inline_ref *)(ei + 1);
1517 WARN_ON(key->type != BTRFS_EXTENT_ITEM_KEY);
1518 info = (struct btrfs_tree_block_info *)(ei + 1);
1520 (struct btrfs_extent_inline_ref *)(info + 1);
1523 *out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1);
1525 *ptr = (unsigned long)*out_eiref;
1526 if ((unsigned long)(*ptr) >= (unsigned long)ei + item_size)
1530 end = (unsigned long)ei + item_size;
1531 *out_eiref = (struct btrfs_extent_inline_ref *)(*ptr);
1532 *out_type = btrfs_extent_inline_ref_type(eb, *out_eiref);
1534 *ptr += btrfs_extent_inline_ref_size(*out_type);
1535 WARN_ON(*ptr > end);
1537 return 1; /* last */
1543 * reads the tree block backref for an extent. tree level and root are returned
1544 * through out_level and out_root. ptr must point to a 0 value for the first
1545 * call and may be modified (see __get_extent_inline_ref comment).
1546 * returns 0 if data was provided, 1 if there was no more data to provide or
1549 int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
1550 struct btrfs_key *key, struct btrfs_extent_item *ei,
1551 u32 item_size, u64 *out_root, u8 *out_level)
1555 struct btrfs_extent_inline_ref *eiref;
1557 if (*ptr == (unsigned long)-1)
1561 ret = __get_extent_inline_ref(ptr, eb, key, ei, item_size,
1566 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1567 type == BTRFS_SHARED_BLOCK_REF_KEY)
1574 /* we can treat both ref types equally here */
1575 *out_root = btrfs_extent_inline_ref_offset(eb, eiref);
1577 if (key->type == BTRFS_EXTENT_ITEM_KEY) {
1578 struct btrfs_tree_block_info *info;
1580 info = (struct btrfs_tree_block_info *)(ei + 1);
1581 *out_level = btrfs_tree_block_level(eb, info);
1583 ASSERT(key->type == BTRFS_METADATA_ITEM_KEY);
1584 *out_level = (u8)key->offset;
1588 *ptr = (unsigned long)-1;
1593 static int iterate_leaf_refs(struct extent_inode_elem *inode_list,
1594 u64 root, u64 extent_item_objectid,
1595 iterate_extent_inodes_t *iterate, void *ctx)
1597 struct extent_inode_elem *eie;
1600 for (eie = inode_list; eie; eie = eie->next) {
1601 pr_debug("ref for %llu resolved, key (%llu EXTEND_DATA %llu), "
1602 "root %llu\n", extent_item_objectid,
1603 eie->inum, eie->offset, root);
1604 ret = iterate(eie->inum, eie->offset, root, ctx);
1606 pr_debug("stopping iteration for %llu due to ret=%d\n",
1607 extent_item_objectid, ret);
1616 * calls iterate() for every inode that references the extent identified by
1617 * the given parameters.
1618 * when the iterator function returns a non-zero value, iteration stops.
1620 int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
1621 u64 extent_item_objectid, u64 extent_item_pos,
1622 int search_commit_root,
1623 iterate_extent_inodes_t *iterate, void *ctx)
1626 struct btrfs_trans_handle *trans = NULL;
1627 struct ulist *refs = NULL;
1628 struct ulist *roots = NULL;
1629 struct ulist_node *ref_node = NULL;
1630 struct ulist_node *root_node = NULL;
1631 struct seq_list tree_mod_seq_elem = {};
1632 struct ulist_iterator ref_uiter;
1633 struct ulist_iterator root_uiter;
1635 pr_debug("resolving all inodes for extent %llu\n",
1636 extent_item_objectid);
1638 if (!search_commit_root) {
1639 trans = btrfs_join_transaction(fs_info->extent_root);
1641 return PTR_ERR(trans);
1642 btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1644 down_read(&fs_info->commit_root_sem);
1647 ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid,
1648 tree_mod_seq_elem.seq, &refs,
1653 ULIST_ITER_INIT(&ref_uiter);
1654 while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) {
1655 ret = __btrfs_find_all_roots(trans, fs_info, ref_node->val,
1656 tree_mod_seq_elem.seq, &roots);
1659 ULIST_ITER_INIT(&root_uiter);
1660 while (!ret && (root_node = ulist_next(roots, &root_uiter))) {
1661 pr_debug("root %llu references leaf %llu, data list "
1662 "%#llx\n", root_node->val, ref_node->val,
1664 ret = iterate_leaf_refs((struct extent_inode_elem *)
1665 (uintptr_t)ref_node->aux,
1667 extent_item_objectid,
1673 free_leaf_list(refs);
1675 if (!search_commit_root) {
1676 btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1677 btrfs_end_transaction(trans, fs_info->extent_root);
1679 up_read(&fs_info->commit_root_sem);
1685 int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
1686 struct btrfs_path *path,
1687 iterate_extent_inodes_t *iterate, void *ctx)
1690 u64 extent_item_pos;
1692 struct btrfs_key found_key;
1693 int search_commit_root = path->search_commit_root;
1695 ret = extent_from_logical(fs_info, logical, path, &found_key, &flags);
1696 btrfs_release_path(path);
1699 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1702 extent_item_pos = logical - found_key.objectid;
1703 ret = iterate_extent_inodes(fs_info, found_key.objectid,
1704 extent_item_pos, search_commit_root,
1710 typedef int (iterate_irefs_t)(u64 parent, u32 name_len, unsigned long name_off,
1711 struct extent_buffer *eb, void *ctx);
1713 static int iterate_inode_refs(u64 inum, struct btrfs_root *fs_root,
1714 struct btrfs_path *path,
1715 iterate_irefs_t *iterate, void *ctx)
1724 struct extent_buffer *eb;
1725 struct btrfs_item *item;
1726 struct btrfs_inode_ref *iref;
1727 struct btrfs_key found_key;
1730 ret = inode_ref_info(inum, parent ? parent+1 : 0, fs_root, path,
1735 ret = found ? 0 : -ENOENT;
1740 parent = found_key.offset;
1741 slot = path->slots[0];
1742 eb = btrfs_clone_extent_buffer(path->nodes[0]);
1747 extent_buffer_get(eb);
1748 btrfs_tree_read_lock(eb);
1749 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1750 btrfs_release_path(path);
1752 item = btrfs_item_nr(slot);
1753 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1755 for (cur = 0; cur < btrfs_item_size(eb, item); cur += len) {
1756 name_len = btrfs_inode_ref_name_len(eb, iref);
1757 /* path must be released before calling iterate()! */
1758 pr_debug("following ref at offset %u for inode %llu in "
1759 "tree %llu\n", cur, found_key.objectid,
1761 ret = iterate(parent, name_len,
1762 (unsigned long)(iref + 1), eb, ctx);
1765 len = sizeof(*iref) + name_len;
1766 iref = (struct btrfs_inode_ref *)((char *)iref + len);
1768 btrfs_tree_read_unlock_blocking(eb);
1769 free_extent_buffer(eb);
1772 btrfs_release_path(path);
1777 static int iterate_inode_extrefs(u64 inum, struct btrfs_root *fs_root,
1778 struct btrfs_path *path,
1779 iterate_irefs_t *iterate, void *ctx)
1786 struct extent_buffer *eb;
1787 struct btrfs_inode_extref *extref;
1788 struct extent_buffer *leaf;
1794 ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref,
1799 ret = found ? 0 : -ENOENT;
1804 slot = path->slots[0];
1805 eb = btrfs_clone_extent_buffer(path->nodes[0]);
1810 extent_buffer_get(eb);
1812 btrfs_tree_read_lock(eb);
1813 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1814 btrfs_release_path(path);
1816 leaf = path->nodes[0];
1817 item_size = btrfs_item_size_nr(leaf, slot);
1818 ptr = btrfs_item_ptr_offset(leaf, slot);
1821 while (cur_offset < item_size) {
1824 extref = (struct btrfs_inode_extref *)(ptr + cur_offset);
1825 parent = btrfs_inode_extref_parent(eb, extref);
1826 name_len = btrfs_inode_extref_name_len(eb, extref);
1827 ret = iterate(parent, name_len,
1828 (unsigned long)&extref->name, eb, ctx);
1832 cur_offset += btrfs_inode_extref_name_len(leaf, extref);
1833 cur_offset += sizeof(*extref);
1835 btrfs_tree_read_unlock_blocking(eb);
1836 free_extent_buffer(eb);
1841 btrfs_release_path(path);
1846 static int iterate_irefs(u64 inum, struct btrfs_root *fs_root,
1847 struct btrfs_path *path, iterate_irefs_t *iterate,
1853 ret = iterate_inode_refs(inum, fs_root, path, iterate, ctx);
1856 else if (ret != -ENOENT)
1859 ret = iterate_inode_extrefs(inum, fs_root, path, iterate, ctx);
1860 if (ret == -ENOENT && found_refs)
1867 * returns 0 if the path could be dumped (probably truncated)
1868 * returns <0 in case of an error
1870 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
1871 struct extent_buffer *eb, void *ctx)
1873 struct inode_fs_paths *ipath = ctx;
1876 int i = ipath->fspath->elem_cnt;
1877 const int s_ptr = sizeof(char *);
1880 bytes_left = ipath->fspath->bytes_left > s_ptr ?
1881 ipath->fspath->bytes_left - s_ptr : 0;
1883 fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
1884 fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len,
1885 name_off, eb, inum, fspath_min, bytes_left);
1887 return PTR_ERR(fspath);
1889 if (fspath > fspath_min) {
1890 ipath->fspath->val[i] = (u64)(unsigned long)fspath;
1891 ++ipath->fspath->elem_cnt;
1892 ipath->fspath->bytes_left = fspath - fspath_min;
1894 ++ipath->fspath->elem_missed;
1895 ipath->fspath->bytes_missing += fspath_min - fspath;
1896 ipath->fspath->bytes_left = 0;
1903 * this dumps all file system paths to the inode into the ipath struct, provided
1904 * is has been created large enough. each path is zero-terminated and accessed
1905 * from ipath->fspath->val[i].
1906 * when it returns, there are ipath->fspath->elem_cnt number of paths available
1907 * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the
1908 * number of missed paths in recored in ipath->fspath->elem_missed, otherwise,
1909 * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
1910 * have been needed to return all paths.
1912 int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
1914 return iterate_irefs(inum, ipath->fs_root, ipath->btrfs_path,
1915 inode_to_path, ipath);
1918 struct btrfs_data_container *init_data_container(u32 total_bytes)
1920 struct btrfs_data_container *data;
1923 alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
1924 data = vmalloc(alloc_bytes);
1926 return ERR_PTR(-ENOMEM);
1928 if (total_bytes >= sizeof(*data)) {
1929 data->bytes_left = total_bytes - sizeof(*data);
1930 data->bytes_missing = 0;
1932 data->bytes_missing = sizeof(*data) - total_bytes;
1933 data->bytes_left = 0;
1937 data->elem_missed = 0;
1943 * allocates space to return multiple file system paths for an inode.
1944 * total_bytes to allocate are passed, note that space usable for actual path
1945 * information will be total_bytes - sizeof(struct inode_fs_paths).
1946 * the returned pointer must be freed with free_ipath() in the end.
1948 struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
1949 struct btrfs_path *path)
1951 struct inode_fs_paths *ifp;
1952 struct btrfs_data_container *fspath;
1954 fspath = init_data_container(total_bytes);
1956 return (void *)fspath;
1958 ifp = kmalloc(sizeof(*ifp), GFP_NOFS);
1961 return ERR_PTR(-ENOMEM);
1964 ifp->btrfs_path = path;
1965 ifp->fspath = fspath;
1966 ifp->fs_root = fs_root;
1971 void free_ipath(struct inode_fs_paths *ipath)
1975 vfree(ipath->fspath);