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 * We can often find data backrefs with an offset that is too
213 * large (>= LLONG_MAX, maximum allowed file offset) due to
214 * underflows when subtracting a file's offset with the data
215 * offset of its corresponding extent data item. This can
216 * happen for example in the clone ioctl.
217 * So if we detect such case we set the search key's offset to
218 * zero to make sure we will find the matching file extent item
219 * at add_all_parents(), otherwise we will miss it because the
220 * offset taken form the backref is much larger then the offset
221 * of the file extent item. This can make us scan a very large
222 * number of file extent items, but at least it will not make
224 * This is an ugly workaround for a behaviour that should have
225 * never existed, but it does and a fix for the clone ioctl
226 * would touch a lot of places, cause backwards incompatibility
227 * and would not fix the problem for extents cloned with older
230 if (ref->key_for_search.type == BTRFS_EXTENT_DATA_KEY &&
231 ref->key_for_search.offset >= LLONG_MAX)
232 ref->key_for_search.offset = 0;
234 memset(&ref->key_for_search, 0, sizeof(ref->key_for_search));
237 ref->inode_list = NULL;
240 ref->parent = parent;
241 ref->wanted_disk_byte = wanted_disk_byte;
242 list_add_tail(&ref->list, head);
247 static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path,
248 struct ulist *parents, struct __prelim_ref *ref,
249 int level, u64 time_seq, const u64 *extent_item_pos,
254 struct extent_buffer *eb;
255 struct btrfs_key key;
256 struct btrfs_key *key_for_search = &ref->key_for_search;
257 struct btrfs_file_extent_item *fi;
258 struct extent_inode_elem *eie = NULL, *old = NULL;
260 u64 wanted_disk_byte = ref->wanted_disk_byte;
264 eb = path->nodes[level];
265 ret = ulist_add(parents, eb->start, 0, GFP_NOFS);
272 * We normally enter this function with the path already pointing to
273 * the first item to check. But sometimes, we may enter it with
274 * slot==nritems. In that case, go to the next leaf before we continue.
276 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
277 if (time_seq == (u64)-1)
278 ret = btrfs_next_leaf(root, path);
280 ret = btrfs_next_old_leaf(root, path, time_seq);
283 while (!ret && count < total_refs) {
285 slot = path->slots[0];
287 btrfs_item_key_to_cpu(eb, &key, slot);
289 if (key.objectid != key_for_search->objectid ||
290 key.type != BTRFS_EXTENT_DATA_KEY)
293 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
294 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
296 if (disk_byte == wanted_disk_byte) {
300 if (extent_item_pos) {
301 ret = check_extent_in_eb(&key, eb, fi,
309 ret = ulist_add_merge_ptr(parents, eb->start,
310 eie, (void **)&old, GFP_NOFS);
313 if (!ret && extent_item_pos) {
321 if (time_seq == (u64)-1)
322 ret = btrfs_next_item(root, path);
324 ret = btrfs_next_old_item(root, path, time_seq);
330 free_inode_elem_list(eie);
335 * resolve an indirect backref in the form (root_id, key, level)
336 * to a logical address
338 static int __resolve_indirect_ref(struct btrfs_fs_info *fs_info,
339 struct btrfs_path *path, u64 time_seq,
340 struct __prelim_ref *ref,
341 struct ulist *parents,
342 const u64 *extent_item_pos, u64 total_refs)
344 struct btrfs_root *root;
345 struct btrfs_key root_key;
346 struct extent_buffer *eb;
349 int level = ref->level;
352 root_key.objectid = ref->root_id;
353 root_key.type = BTRFS_ROOT_ITEM_KEY;
354 root_key.offset = (u64)-1;
356 index = srcu_read_lock(&fs_info->subvol_srcu);
358 root = btrfs_read_fs_root_no_name(fs_info, &root_key);
360 srcu_read_unlock(&fs_info->subvol_srcu, index);
365 if (path->search_commit_root)
366 root_level = btrfs_header_level(root->commit_root);
367 else if (time_seq == (u64)-1)
368 root_level = btrfs_header_level(root->node);
370 root_level = btrfs_old_root_level(root, time_seq);
372 if (root_level + 1 == level) {
373 srcu_read_unlock(&fs_info->subvol_srcu, index);
377 path->lowest_level = level;
378 if (time_seq == (u64)-1)
379 ret = btrfs_search_slot(NULL, root, &ref->key_for_search, path,
382 ret = btrfs_search_old_slot(root, &ref->key_for_search, path,
385 /* root node has been locked, we can release @subvol_srcu safely here */
386 srcu_read_unlock(&fs_info->subvol_srcu, index);
388 pr_debug("search slot in root %llu (level %d, ref count %d) returned "
389 "%d for key (%llu %u %llu)\n",
390 ref->root_id, level, ref->count, ret,
391 ref->key_for_search.objectid, ref->key_for_search.type,
392 ref->key_for_search.offset);
396 eb = path->nodes[level];
398 if (WARN_ON(!level)) {
403 eb = path->nodes[level];
406 ret = add_all_parents(root, path, parents, ref, level, time_seq,
407 extent_item_pos, total_refs);
409 path->lowest_level = 0;
410 btrfs_release_path(path);
415 * resolve all indirect backrefs from the list
417 static int __resolve_indirect_refs(struct btrfs_fs_info *fs_info,
418 struct btrfs_path *path, u64 time_seq,
419 struct list_head *head,
420 const u64 *extent_item_pos, u64 total_refs,
425 struct __prelim_ref *ref;
426 struct __prelim_ref *ref_safe;
427 struct __prelim_ref *new_ref;
428 struct ulist *parents;
429 struct ulist_node *node;
430 struct ulist_iterator uiter;
432 parents = ulist_alloc(GFP_NOFS);
437 * _safe allows us to insert directly after the current item without
438 * iterating over the newly inserted items.
439 * we're also allowed to re-assign ref during iteration.
441 list_for_each_entry_safe(ref, ref_safe, head, list) {
442 if (ref->parent) /* already direct */
446 if (root_objectid && ref->root_id != root_objectid) {
447 ret = BACKREF_FOUND_SHARED;
450 err = __resolve_indirect_ref(fs_info, path, time_seq, ref,
451 parents, extent_item_pos,
454 * we can only tolerate ENOENT,otherwise,we should catch error
455 * and return directly.
457 if (err == -ENOENT) {
464 /* we put the first parent into the ref at hand */
465 ULIST_ITER_INIT(&uiter);
466 node = ulist_next(parents, &uiter);
467 ref->parent = node ? node->val : 0;
468 ref->inode_list = node ?
469 (struct extent_inode_elem *)(uintptr_t)node->aux : NULL;
471 /* additional parents require new refs being added here */
472 while ((node = ulist_next(parents, &uiter))) {
473 new_ref = kmem_cache_alloc(btrfs_prelim_ref_cache,
479 memcpy(new_ref, ref, sizeof(*ref));
480 new_ref->parent = node->val;
481 new_ref->inode_list = (struct extent_inode_elem *)
482 (uintptr_t)node->aux;
483 list_add(&new_ref->list, &ref->list);
485 ulist_reinit(parents);
492 static inline int ref_for_same_block(struct __prelim_ref *ref1,
493 struct __prelim_ref *ref2)
495 if (ref1->level != ref2->level)
497 if (ref1->root_id != ref2->root_id)
499 if (ref1->key_for_search.type != ref2->key_for_search.type)
501 if (ref1->key_for_search.objectid != ref2->key_for_search.objectid)
503 if (ref1->key_for_search.offset != ref2->key_for_search.offset)
505 if (ref1->parent != ref2->parent)
512 * read tree blocks and add keys where required.
514 static int __add_missing_keys(struct btrfs_fs_info *fs_info,
515 struct list_head *head)
517 struct list_head *pos;
518 struct extent_buffer *eb;
520 list_for_each(pos, head) {
521 struct __prelim_ref *ref;
522 ref = list_entry(pos, struct __prelim_ref, list);
526 if (ref->key_for_search.type)
528 BUG_ON(!ref->wanted_disk_byte);
529 eb = read_tree_block(fs_info->tree_root, ref->wanted_disk_byte,
533 } else if (!extent_buffer_uptodate(eb)) {
534 free_extent_buffer(eb);
537 btrfs_tree_read_lock(eb);
538 if (btrfs_header_level(eb) == 0)
539 btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
541 btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
542 btrfs_tree_read_unlock(eb);
543 free_extent_buffer(eb);
549 * merge backrefs and adjust counts accordingly
551 * mode = 1: merge identical keys, if key is set
552 * FIXME: if we add more keys in __add_prelim_ref, we can merge more here.
553 * additionally, we could even add a key range for the blocks we
554 * looked into to merge even more (-> replace unresolved refs by those
556 * mode = 2: merge identical parents
558 static void __merge_refs(struct list_head *head, int mode)
560 struct list_head *pos1;
562 list_for_each(pos1, head) {
563 struct list_head *n2;
564 struct list_head *pos2;
565 struct __prelim_ref *ref1;
567 ref1 = list_entry(pos1, struct __prelim_ref, list);
569 for (pos2 = pos1->next, n2 = pos2->next; pos2 != head;
570 pos2 = n2, n2 = pos2->next) {
571 struct __prelim_ref *ref2;
572 struct __prelim_ref *xchg;
573 struct extent_inode_elem *eie;
575 ref2 = list_entry(pos2, struct __prelim_ref, list);
577 if (!ref_for_same_block(ref1, ref2))
580 if (!ref1->parent && ref2->parent) {
586 if (ref1->parent != ref2->parent)
590 eie = ref1->inode_list;
591 while (eie && eie->next)
594 eie->next = ref2->inode_list;
596 ref1->inode_list = ref2->inode_list;
597 ref1->count += ref2->count;
599 list_del(&ref2->list);
600 kmem_cache_free(btrfs_prelim_ref_cache, ref2);
607 * add all currently queued delayed refs from this head whose seq nr is
608 * smaller or equal that seq to the list
610 static int __add_delayed_refs(struct btrfs_delayed_ref_head *head, u64 seq,
611 struct list_head *prefs, u64 *total_refs,
614 struct btrfs_delayed_ref_node *node;
615 struct btrfs_delayed_extent_op *extent_op = head->extent_op;
616 struct btrfs_key key;
617 struct btrfs_key op_key = {0};
621 if (extent_op && extent_op->update_key)
622 btrfs_disk_key_to_cpu(&op_key, &extent_op->key);
624 spin_lock(&head->lock);
625 list_for_each_entry(node, &head->ref_list, list) {
629 switch (node->action) {
630 case BTRFS_ADD_DELAYED_EXTENT:
631 case BTRFS_UPDATE_DELAYED_HEAD:
634 case BTRFS_ADD_DELAYED_REF:
637 case BTRFS_DROP_DELAYED_REF:
643 *total_refs += (node->ref_mod * sgn);
644 switch (node->type) {
645 case BTRFS_TREE_BLOCK_REF_KEY: {
646 struct btrfs_delayed_tree_ref *ref;
648 ref = btrfs_delayed_node_to_tree_ref(node);
649 ret = __add_prelim_ref(prefs, ref->root, &op_key,
650 ref->level + 1, 0, node->bytenr,
651 node->ref_mod * sgn, GFP_ATOMIC);
654 case BTRFS_SHARED_BLOCK_REF_KEY: {
655 struct btrfs_delayed_tree_ref *ref;
657 ref = btrfs_delayed_node_to_tree_ref(node);
658 ret = __add_prelim_ref(prefs, 0, NULL,
659 ref->level + 1, ref->parent,
661 node->ref_mod * sgn, GFP_ATOMIC);
664 case BTRFS_EXTENT_DATA_REF_KEY: {
665 struct btrfs_delayed_data_ref *ref;
666 ref = btrfs_delayed_node_to_data_ref(node);
668 key.objectid = ref->objectid;
669 key.type = BTRFS_EXTENT_DATA_KEY;
670 key.offset = ref->offset;
673 * Found a inum that doesn't match our known inum, we
676 if (inum && ref->objectid != inum) {
677 ret = BACKREF_FOUND_SHARED;
681 ret = __add_prelim_ref(prefs, ref->root, &key, 0, 0,
683 node->ref_mod * sgn, GFP_ATOMIC);
686 case BTRFS_SHARED_DATA_REF_KEY: {
687 struct btrfs_delayed_data_ref *ref;
689 ref = btrfs_delayed_node_to_data_ref(node);
690 ret = __add_prelim_ref(prefs, 0, NULL, 0,
691 ref->parent, node->bytenr,
692 node->ref_mod * sgn, GFP_ATOMIC);
701 spin_unlock(&head->lock);
706 * add all inline backrefs for bytenr to the list
708 static int __add_inline_refs(struct btrfs_fs_info *fs_info,
709 struct btrfs_path *path, u64 bytenr,
710 int *info_level, struct list_head *prefs,
711 u64 *total_refs, u64 inum)
715 struct extent_buffer *leaf;
716 struct btrfs_key key;
717 struct btrfs_key found_key;
720 struct btrfs_extent_item *ei;
725 * enumerate all inline refs
727 leaf = path->nodes[0];
728 slot = path->slots[0];
730 item_size = btrfs_item_size_nr(leaf, slot);
731 BUG_ON(item_size < sizeof(*ei));
733 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
734 flags = btrfs_extent_flags(leaf, ei);
735 *total_refs += btrfs_extent_refs(leaf, ei);
736 btrfs_item_key_to_cpu(leaf, &found_key, slot);
738 ptr = (unsigned long)(ei + 1);
739 end = (unsigned long)ei + item_size;
741 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
742 flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
743 struct btrfs_tree_block_info *info;
745 info = (struct btrfs_tree_block_info *)ptr;
746 *info_level = btrfs_tree_block_level(leaf, info);
747 ptr += sizeof(struct btrfs_tree_block_info);
749 } else if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
750 *info_level = found_key.offset;
752 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
756 struct btrfs_extent_inline_ref *iref;
760 iref = (struct btrfs_extent_inline_ref *)ptr;
761 type = btrfs_extent_inline_ref_type(leaf, iref);
762 offset = btrfs_extent_inline_ref_offset(leaf, iref);
765 case BTRFS_SHARED_BLOCK_REF_KEY:
766 ret = __add_prelim_ref(prefs, 0, NULL,
767 *info_level + 1, offset,
768 bytenr, 1, GFP_NOFS);
770 case BTRFS_SHARED_DATA_REF_KEY: {
771 struct btrfs_shared_data_ref *sdref;
774 sdref = (struct btrfs_shared_data_ref *)(iref + 1);
775 count = btrfs_shared_data_ref_count(leaf, sdref);
776 ret = __add_prelim_ref(prefs, 0, NULL, 0, offset,
777 bytenr, count, GFP_NOFS);
780 case BTRFS_TREE_BLOCK_REF_KEY:
781 ret = __add_prelim_ref(prefs, offset, NULL,
783 bytenr, 1, GFP_NOFS);
785 case BTRFS_EXTENT_DATA_REF_KEY: {
786 struct btrfs_extent_data_ref *dref;
790 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
791 count = btrfs_extent_data_ref_count(leaf, dref);
792 key.objectid = btrfs_extent_data_ref_objectid(leaf,
794 key.type = BTRFS_EXTENT_DATA_KEY;
795 key.offset = btrfs_extent_data_ref_offset(leaf, dref);
797 if (inum && key.objectid != inum) {
798 ret = BACKREF_FOUND_SHARED;
802 root = btrfs_extent_data_ref_root(leaf, dref);
803 ret = __add_prelim_ref(prefs, root, &key, 0, 0,
804 bytenr, count, GFP_NOFS);
812 ptr += btrfs_extent_inline_ref_size(type);
819 * add all non-inline backrefs for bytenr to the list
821 static int __add_keyed_refs(struct btrfs_fs_info *fs_info,
822 struct btrfs_path *path, u64 bytenr,
823 int info_level, struct list_head *prefs, u64 inum)
825 struct btrfs_root *extent_root = fs_info->extent_root;
828 struct extent_buffer *leaf;
829 struct btrfs_key key;
832 ret = btrfs_next_item(extent_root, path);
840 slot = path->slots[0];
841 leaf = path->nodes[0];
842 btrfs_item_key_to_cpu(leaf, &key, slot);
844 if (key.objectid != bytenr)
846 if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
848 if (key.type > BTRFS_SHARED_DATA_REF_KEY)
852 case BTRFS_SHARED_BLOCK_REF_KEY:
853 ret = __add_prelim_ref(prefs, 0, NULL,
854 info_level + 1, key.offset,
855 bytenr, 1, GFP_NOFS);
857 case BTRFS_SHARED_DATA_REF_KEY: {
858 struct btrfs_shared_data_ref *sdref;
861 sdref = btrfs_item_ptr(leaf, slot,
862 struct btrfs_shared_data_ref);
863 count = btrfs_shared_data_ref_count(leaf, sdref);
864 ret = __add_prelim_ref(prefs, 0, NULL, 0, key.offset,
865 bytenr, count, GFP_NOFS);
868 case BTRFS_TREE_BLOCK_REF_KEY:
869 ret = __add_prelim_ref(prefs, key.offset, NULL,
871 bytenr, 1, GFP_NOFS);
873 case BTRFS_EXTENT_DATA_REF_KEY: {
874 struct btrfs_extent_data_ref *dref;
878 dref = btrfs_item_ptr(leaf, slot,
879 struct btrfs_extent_data_ref);
880 count = btrfs_extent_data_ref_count(leaf, dref);
881 key.objectid = btrfs_extent_data_ref_objectid(leaf,
883 key.type = BTRFS_EXTENT_DATA_KEY;
884 key.offset = btrfs_extent_data_ref_offset(leaf, dref);
886 if (inum && key.objectid != inum) {
887 ret = BACKREF_FOUND_SHARED;
891 root = btrfs_extent_data_ref_root(leaf, dref);
892 ret = __add_prelim_ref(prefs, root, &key, 0, 0,
893 bytenr, count, GFP_NOFS);
908 * this adds all existing backrefs (inline backrefs, backrefs and delayed
909 * refs) for the given bytenr to the refs list, merges duplicates and resolves
910 * indirect refs to their parent bytenr.
911 * When roots are found, they're added to the roots list
913 * NOTE: This can return values > 0
915 * If time_seq is set to (u64)-1, it will not search delayed_refs, and behave
916 * much like trans == NULL case, the difference only lies in it will not
918 * The special case is for qgroup to search roots in commit_transaction().
920 * FIXME some caching might speed things up
922 static int find_parent_nodes(struct btrfs_trans_handle *trans,
923 struct btrfs_fs_info *fs_info, u64 bytenr,
924 u64 time_seq, struct ulist *refs,
925 struct ulist *roots, const u64 *extent_item_pos,
926 u64 root_objectid, u64 inum)
928 struct btrfs_key key;
929 struct btrfs_path *path;
930 struct btrfs_delayed_ref_root *delayed_refs = NULL;
931 struct btrfs_delayed_ref_head *head;
934 struct list_head prefs_delayed;
935 struct list_head prefs;
936 struct __prelim_ref *ref;
937 struct extent_inode_elem *eie = NULL;
940 INIT_LIST_HEAD(&prefs);
941 INIT_LIST_HEAD(&prefs_delayed);
943 key.objectid = bytenr;
944 key.offset = (u64)-1;
945 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
946 key.type = BTRFS_METADATA_ITEM_KEY;
948 key.type = BTRFS_EXTENT_ITEM_KEY;
950 path = btrfs_alloc_path();
954 path->search_commit_root = 1;
955 path->skip_locking = 1;
958 if (time_seq == (u64)-1)
959 path->skip_locking = 1;
962 * grab both a lock on the path and a lock on the delayed ref head.
963 * We need both to get a consistent picture of how the refs look
964 * at a specified point in time
969 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
974 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
975 if (trans && likely(trans->type != __TRANS_DUMMY) &&
976 time_seq != (u64)-1) {
978 if (trans && time_seq != (u64)-1) {
981 * look if there are updates for this ref queued and lock the
984 delayed_refs = &trans->transaction->delayed_refs;
985 spin_lock(&delayed_refs->lock);
986 head = btrfs_find_delayed_ref_head(trans, bytenr);
988 if (!mutex_trylock(&head->mutex)) {
989 atomic_inc(&head->node.refs);
990 spin_unlock(&delayed_refs->lock);
992 btrfs_release_path(path);
995 * Mutex was contended, block until it's
996 * released and try again
998 mutex_lock(&head->mutex);
999 mutex_unlock(&head->mutex);
1000 btrfs_put_delayed_ref(&head->node);
1003 spin_unlock(&delayed_refs->lock);
1004 ret = __add_delayed_refs(head, time_seq,
1005 &prefs_delayed, &total_refs,
1007 mutex_unlock(&head->mutex);
1011 spin_unlock(&delayed_refs->lock);
1015 if (path->slots[0]) {
1016 struct extent_buffer *leaf;
1020 leaf = path->nodes[0];
1021 slot = path->slots[0];
1022 btrfs_item_key_to_cpu(leaf, &key, slot);
1023 if (key.objectid == bytenr &&
1024 (key.type == BTRFS_EXTENT_ITEM_KEY ||
1025 key.type == BTRFS_METADATA_ITEM_KEY)) {
1026 ret = __add_inline_refs(fs_info, path, bytenr,
1027 &info_level, &prefs,
1031 ret = __add_keyed_refs(fs_info, path, bytenr,
1032 info_level, &prefs, inum);
1037 btrfs_release_path(path);
1039 list_splice_init(&prefs_delayed, &prefs);
1041 ret = __add_missing_keys(fs_info, &prefs);
1045 __merge_refs(&prefs, 1);
1047 ret = __resolve_indirect_refs(fs_info, path, time_seq, &prefs,
1048 extent_item_pos, total_refs,
1053 __merge_refs(&prefs, 2);
1055 while (!list_empty(&prefs)) {
1056 ref = list_first_entry(&prefs, struct __prelim_ref, list);
1057 WARN_ON(ref->count < 0);
1058 if (roots && ref->count && ref->root_id && ref->parent == 0) {
1059 if (root_objectid && ref->root_id != root_objectid) {
1060 ret = BACKREF_FOUND_SHARED;
1064 /* no parent == root of tree */
1065 ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS);
1069 if (ref->count && ref->parent) {
1070 if (extent_item_pos && !ref->inode_list &&
1072 struct extent_buffer *eb;
1074 eb = read_tree_block(fs_info->extent_root,
1079 } else if (!extent_buffer_uptodate(eb)) {
1080 free_extent_buffer(eb);
1084 btrfs_tree_read_lock(eb);
1085 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1086 ret = find_extent_in_eb(eb, bytenr,
1087 *extent_item_pos, &eie);
1088 btrfs_tree_read_unlock_blocking(eb);
1089 free_extent_buffer(eb);
1092 ref->inode_list = eie;
1094 ret = ulist_add_merge_ptr(refs, ref->parent,
1096 (void **)&eie, GFP_NOFS);
1099 if (!ret && extent_item_pos) {
1101 * we've recorded that parent, so we must extend
1102 * its inode list here
1107 eie->next = ref->inode_list;
1111 list_del(&ref->list);
1112 kmem_cache_free(btrfs_prelim_ref_cache, ref);
1116 btrfs_free_path(path);
1117 while (!list_empty(&prefs)) {
1118 ref = list_first_entry(&prefs, struct __prelim_ref, list);
1119 list_del(&ref->list);
1120 kmem_cache_free(btrfs_prelim_ref_cache, ref);
1122 while (!list_empty(&prefs_delayed)) {
1123 ref = list_first_entry(&prefs_delayed, struct __prelim_ref,
1125 list_del(&ref->list);
1126 kmem_cache_free(btrfs_prelim_ref_cache, ref);
1129 free_inode_elem_list(eie);
1133 static void free_leaf_list(struct ulist *blocks)
1135 struct ulist_node *node = NULL;
1136 struct extent_inode_elem *eie;
1137 struct ulist_iterator uiter;
1139 ULIST_ITER_INIT(&uiter);
1140 while ((node = ulist_next(blocks, &uiter))) {
1143 eie = (struct extent_inode_elem *)(uintptr_t)node->aux;
1144 free_inode_elem_list(eie);
1152 * Finds all leafs with a reference to the specified combination of bytenr and
1153 * offset. key_list_head will point to a list of corresponding keys (caller must
1154 * free each list element). The leafs will be stored in the leafs ulist, which
1155 * must be freed with ulist_free.
1157 * returns 0 on success, <0 on error
1159 static int btrfs_find_all_leafs(struct btrfs_trans_handle *trans,
1160 struct btrfs_fs_info *fs_info, u64 bytenr,
1161 u64 time_seq, struct ulist **leafs,
1162 const u64 *extent_item_pos)
1166 *leafs = ulist_alloc(GFP_NOFS);
1170 ret = find_parent_nodes(trans, fs_info, bytenr,
1171 time_seq, *leafs, NULL, extent_item_pos, 0, 0);
1172 if (ret < 0 && ret != -ENOENT) {
1173 free_leaf_list(*leafs);
1181 * walk all backrefs for a given extent to find all roots that reference this
1182 * extent. Walking a backref means finding all extents that reference this
1183 * extent and in turn walk the backrefs of those, too. Naturally this is a
1184 * recursive process, but here it is implemented in an iterative fashion: We
1185 * find all referencing extents for the extent in question and put them on a
1186 * list. In turn, we find all referencing extents for those, further appending
1187 * to the list. The way we iterate the list allows adding more elements after
1188 * the current while iterating. The process stops when we reach the end of the
1189 * list. Found roots are added to the roots list.
1191 * returns 0 on success, < 0 on error.
1193 static int __btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1194 struct btrfs_fs_info *fs_info, u64 bytenr,
1195 u64 time_seq, struct ulist **roots)
1198 struct ulist_node *node = NULL;
1199 struct ulist_iterator uiter;
1202 tmp = ulist_alloc(GFP_NOFS);
1205 *roots = ulist_alloc(GFP_NOFS);
1211 ULIST_ITER_INIT(&uiter);
1213 ret = find_parent_nodes(trans, fs_info, bytenr,
1214 time_seq, tmp, *roots, NULL, 0, 0);
1215 if (ret < 0 && ret != -ENOENT) {
1220 node = ulist_next(tmp, &uiter);
1231 int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1232 struct btrfs_fs_info *fs_info, u64 bytenr,
1233 u64 time_seq, struct ulist **roots)
1238 down_read(&fs_info->commit_root_sem);
1239 ret = __btrfs_find_all_roots(trans, fs_info, bytenr, time_seq, roots);
1241 up_read(&fs_info->commit_root_sem);
1246 * btrfs_check_shared - tell us whether an extent is shared
1248 * @trans: optional trans handle
1250 * btrfs_check_shared uses the backref walking code but will short
1251 * circuit as soon as it finds a root or inode that doesn't match the
1252 * one passed in. This provides a significant performance benefit for
1253 * callers (such as fiemap) which want to know whether the extent is
1254 * shared but do not need a ref count.
1256 * Return: 0 if extent is not shared, 1 if it is shared, < 0 on error.
1258 int btrfs_check_shared(struct btrfs_trans_handle *trans,
1259 struct btrfs_fs_info *fs_info, u64 root_objectid,
1260 u64 inum, u64 bytenr)
1262 struct ulist *tmp = NULL;
1263 struct ulist *roots = NULL;
1264 struct ulist_iterator uiter;
1265 struct ulist_node *node;
1266 struct seq_list elem = SEQ_LIST_INIT(elem);
1269 tmp = ulist_alloc(GFP_NOFS);
1270 roots = ulist_alloc(GFP_NOFS);
1271 if (!tmp || !roots) {
1278 btrfs_get_tree_mod_seq(fs_info, &elem);
1280 down_read(&fs_info->commit_root_sem);
1281 ULIST_ITER_INIT(&uiter);
1283 ret = find_parent_nodes(trans, fs_info, bytenr, elem.seq, tmp,
1284 roots, NULL, root_objectid, inum);
1285 if (ret == BACKREF_FOUND_SHARED) {
1286 /* this is the only condition under which we return 1 */
1290 if (ret < 0 && ret != -ENOENT)
1293 node = ulist_next(tmp, &uiter);
1300 btrfs_put_tree_mod_seq(fs_info, &elem);
1302 up_read(&fs_info->commit_root_sem);
1308 int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,
1309 u64 start_off, struct btrfs_path *path,
1310 struct btrfs_inode_extref **ret_extref,
1314 struct btrfs_key key;
1315 struct btrfs_key found_key;
1316 struct btrfs_inode_extref *extref;
1317 struct extent_buffer *leaf;
1320 key.objectid = inode_objectid;
1321 key.type = BTRFS_INODE_EXTREF_KEY;
1322 key.offset = start_off;
1324 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1329 leaf = path->nodes[0];
1330 slot = path->slots[0];
1331 if (slot >= btrfs_header_nritems(leaf)) {
1333 * If the item at offset is not found,
1334 * btrfs_search_slot will point us to the slot
1335 * where it should be inserted. In our case
1336 * that will be the slot directly before the
1337 * next INODE_REF_KEY_V2 item. In the case
1338 * that we're pointing to the last slot in a
1339 * leaf, we must move one leaf over.
1341 ret = btrfs_next_leaf(root, path);
1350 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1353 * Check that we're still looking at an extended ref key for
1354 * this particular objectid. If we have different
1355 * objectid or type then there are no more to be found
1356 * in the tree and we can exit.
1359 if (found_key.objectid != inode_objectid)
1361 if (found_key.type != BTRFS_INODE_EXTREF_KEY)
1365 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1366 extref = (struct btrfs_inode_extref *)ptr;
1367 *ret_extref = extref;
1369 *found_off = found_key.offset;
1377 * this iterates to turn a name (from iref/extref) into a full filesystem path.
1378 * Elements of the path are separated by '/' and the path is guaranteed to be
1379 * 0-terminated. the path is only given within the current file system.
1380 * Therefore, it never starts with a '/'. the caller is responsible to provide
1381 * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
1382 * the start point of the resulting string is returned. this pointer is within
1384 * in case the path buffer would overflow, the pointer is decremented further
1385 * as if output was written to the buffer, though no more output is actually
1386 * generated. that way, the caller can determine how much space would be
1387 * required for the path to fit into the buffer. in that case, the returned
1388 * value will be smaller than dest. callers must check this!
1390 char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
1391 u32 name_len, unsigned long name_off,
1392 struct extent_buffer *eb_in, u64 parent,
1393 char *dest, u32 size)
1398 s64 bytes_left = ((s64)size) - 1;
1399 struct extent_buffer *eb = eb_in;
1400 struct btrfs_key found_key;
1401 int leave_spinning = path->leave_spinning;
1402 struct btrfs_inode_ref *iref;
1404 if (bytes_left >= 0)
1405 dest[bytes_left] = '\0';
1407 path->leave_spinning = 1;
1409 bytes_left -= name_len;
1410 if (bytes_left >= 0)
1411 read_extent_buffer(eb, dest + bytes_left,
1412 name_off, name_len);
1414 btrfs_tree_read_unlock_blocking(eb);
1415 free_extent_buffer(eb);
1417 ret = btrfs_find_item(fs_root, path, parent, 0,
1418 BTRFS_INODE_REF_KEY, &found_key);
1424 next_inum = found_key.offset;
1426 /* regular exit ahead */
1427 if (parent == next_inum)
1430 slot = path->slots[0];
1431 eb = path->nodes[0];
1432 /* make sure we can use eb after releasing the path */
1434 atomic_inc(&eb->refs);
1435 btrfs_tree_read_lock(eb);
1436 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1438 btrfs_release_path(path);
1439 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1441 name_len = btrfs_inode_ref_name_len(eb, iref);
1442 name_off = (unsigned long)(iref + 1);
1446 if (bytes_left >= 0)
1447 dest[bytes_left] = '/';
1450 btrfs_release_path(path);
1451 path->leave_spinning = leave_spinning;
1454 return ERR_PTR(ret);
1456 return dest + bytes_left;
1460 * this makes the path point to (logical EXTENT_ITEM *)
1461 * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
1462 * tree blocks and <0 on error.
1464 int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
1465 struct btrfs_path *path, struct btrfs_key *found_key,
1472 struct extent_buffer *eb;
1473 struct btrfs_extent_item *ei;
1474 struct btrfs_key key;
1476 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1477 key.type = BTRFS_METADATA_ITEM_KEY;
1479 key.type = BTRFS_EXTENT_ITEM_KEY;
1480 key.objectid = logical;
1481 key.offset = (u64)-1;
1483 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
1487 ret = btrfs_previous_extent_item(fs_info->extent_root, path, 0);
1493 btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]);
1494 if (found_key->type == BTRFS_METADATA_ITEM_KEY)
1495 size = fs_info->extent_root->nodesize;
1496 else if (found_key->type == BTRFS_EXTENT_ITEM_KEY)
1497 size = found_key->offset;
1499 if (found_key->objectid > logical ||
1500 found_key->objectid + size <= logical) {
1501 pr_debug("logical %llu is not within any extent\n", logical);
1505 eb = path->nodes[0];
1506 item_size = btrfs_item_size_nr(eb, path->slots[0]);
1507 BUG_ON(item_size < sizeof(*ei));
1509 ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
1510 flags = btrfs_extent_flags(eb, ei);
1512 pr_debug("logical %llu is at position %llu within the extent (%llu "
1513 "EXTENT_ITEM %llu) flags %#llx size %u\n",
1514 logical, logical - found_key->objectid, found_key->objectid,
1515 found_key->offset, flags, item_size);
1517 WARN_ON(!flags_ret);
1519 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1520 *flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK;
1521 else if (flags & BTRFS_EXTENT_FLAG_DATA)
1522 *flags_ret = BTRFS_EXTENT_FLAG_DATA;
1532 * helper function to iterate extent inline refs. ptr must point to a 0 value
1533 * for the first call and may be modified. it is used to track state.
1534 * if more refs exist, 0 is returned and the next call to
1535 * __get_extent_inline_ref must pass the modified ptr parameter to get the
1536 * next ref. after the last ref was processed, 1 is returned.
1537 * returns <0 on error
1539 static int __get_extent_inline_ref(unsigned long *ptr, struct extent_buffer *eb,
1540 struct btrfs_key *key,
1541 struct btrfs_extent_item *ei, u32 item_size,
1542 struct btrfs_extent_inline_ref **out_eiref,
1547 struct btrfs_tree_block_info *info;
1551 flags = btrfs_extent_flags(eb, ei);
1552 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1553 if (key->type == BTRFS_METADATA_ITEM_KEY) {
1554 /* a skinny metadata extent */
1556 (struct btrfs_extent_inline_ref *)(ei + 1);
1558 WARN_ON(key->type != BTRFS_EXTENT_ITEM_KEY);
1559 info = (struct btrfs_tree_block_info *)(ei + 1);
1561 (struct btrfs_extent_inline_ref *)(info + 1);
1564 *out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1);
1566 *ptr = (unsigned long)*out_eiref;
1567 if ((unsigned long)(*ptr) >= (unsigned long)ei + item_size)
1571 end = (unsigned long)ei + item_size;
1572 *out_eiref = (struct btrfs_extent_inline_ref *)(*ptr);
1573 *out_type = btrfs_extent_inline_ref_type(eb, *out_eiref);
1575 *ptr += btrfs_extent_inline_ref_size(*out_type);
1576 WARN_ON(*ptr > end);
1578 return 1; /* last */
1584 * reads the tree block backref for an extent. tree level and root are returned
1585 * through out_level and out_root. ptr must point to a 0 value for the first
1586 * call and may be modified (see __get_extent_inline_ref comment).
1587 * returns 0 if data was provided, 1 if there was no more data to provide or
1590 int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
1591 struct btrfs_key *key, struct btrfs_extent_item *ei,
1592 u32 item_size, u64 *out_root, u8 *out_level)
1596 struct btrfs_extent_inline_ref *eiref;
1598 if (*ptr == (unsigned long)-1)
1602 ret = __get_extent_inline_ref(ptr, eb, key, ei, item_size,
1607 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1608 type == BTRFS_SHARED_BLOCK_REF_KEY)
1615 /* we can treat both ref types equally here */
1616 *out_root = btrfs_extent_inline_ref_offset(eb, eiref);
1618 if (key->type == BTRFS_EXTENT_ITEM_KEY) {
1619 struct btrfs_tree_block_info *info;
1621 info = (struct btrfs_tree_block_info *)(ei + 1);
1622 *out_level = btrfs_tree_block_level(eb, info);
1624 ASSERT(key->type == BTRFS_METADATA_ITEM_KEY);
1625 *out_level = (u8)key->offset;
1629 *ptr = (unsigned long)-1;
1634 static int iterate_leaf_refs(struct extent_inode_elem *inode_list,
1635 u64 root, u64 extent_item_objectid,
1636 iterate_extent_inodes_t *iterate, void *ctx)
1638 struct extent_inode_elem *eie;
1641 for (eie = inode_list; eie; eie = eie->next) {
1642 pr_debug("ref for %llu resolved, key (%llu EXTEND_DATA %llu), "
1643 "root %llu\n", extent_item_objectid,
1644 eie->inum, eie->offset, root);
1645 ret = iterate(eie->inum, eie->offset, root, ctx);
1647 pr_debug("stopping iteration for %llu due to ret=%d\n",
1648 extent_item_objectid, ret);
1657 * calls iterate() for every inode that references the extent identified by
1658 * the given parameters.
1659 * when the iterator function returns a non-zero value, iteration stops.
1661 int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
1662 u64 extent_item_objectid, u64 extent_item_pos,
1663 int search_commit_root,
1664 iterate_extent_inodes_t *iterate, void *ctx)
1667 struct btrfs_trans_handle *trans = NULL;
1668 struct ulist *refs = NULL;
1669 struct ulist *roots = NULL;
1670 struct ulist_node *ref_node = NULL;
1671 struct ulist_node *root_node = NULL;
1672 struct seq_list tree_mod_seq_elem = SEQ_LIST_INIT(tree_mod_seq_elem);
1673 struct ulist_iterator ref_uiter;
1674 struct ulist_iterator root_uiter;
1676 pr_debug("resolving all inodes for extent %llu\n",
1677 extent_item_objectid);
1679 if (!search_commit_root) {
1680 trans = btrfs_join_transaction(fs_info->extent_root);
1682 return PTR_ERR(trans);
1683 btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1685 down_read(&fs_info->commit_root_sem);
1688 ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid,
1689 tree_mod_seq_elem.seq, &refs,
1694 ULIST_ITER_INIT(&ref_uiter);
1695 while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) {
1696 ret = __btrfs_find_all_roots(trans, fs_info, ref_node->val,
1697 tree_mod_seq_elem.seq, &roots);
1700 ULIST_ITER_INIT(&root_uiter);
1701 while (!ret && (root_node = ulist_next(roots, &root_uiter))) {
1702 pr_debug("root %llu references leaf %llu, data list "
1703 "%#llx\n", root_node->val, ref_node->val,
1705 ret = iterate_leaf_refs((struct extent_inode_elem *)
1706 (uintptr_t)ref_node->aux,
1708 extent_item_objectid,
1714 free_leaf_list(refs);
1716 if (!search_commit_root) {
1717 btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1718 btrfs_end_transaction(trans, fs_info->extent_root);
1720 up_read(&fs_info->commit_root_sem);
1726 int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
1727 struct btrfs_path *path,
1728 iterate_extent_inodes_t *iterate, void *ctx)
1731 u64 extent_item_pos;
1733 struct btrfs_key found_key;
1734 int search_commit_root = path->search_commit_root;
1736 ret = extent_from_logical(fs_info, logical, path, &found_key, &flags);
1737 btrfs_release_path(path);
1740 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1743 extent_item_pos = logical - found_key.objectid;
1744 ret = iterate_extent_inodes(fs_info, found_key.objectid,
1745 extent_item_pos, search_commit_root,
1751 typedef int (iterate_irefs_t)(u64 parent, u32 name_len, unsigned long name_off,
1752 struct extent_buffer *eb, void *ctx);
1754 static int iterate_inode_refs(u64 inum, struct btrfs_root *fs_root,
1755 struct btrfs_path *path,
1756 iterate_irefs_t *iterate, void *ctx)
1765 struct extent_buffer *eb;
1766 struct btrfs_item *item;
1767 struct btrfs_inode_ref *iref;
1768 struct btrfs_key found_key;
1771 ret = btrfs_find_item(fs_root, path, inum,
1772 parent ? parent + 1 : 0, BTRFS_INODE_REF_KEY,
1778 ret = found ? 0 : -ENOENT;
1783 parent = found_key.offset;
1784 slot = path->slots[0];
1785 eb = btrfs_clone_extent_buffer(path->nodes[0]);
1790 extent_buffer_get(eb);
1791 btrfs_tree_read_lock(eb);
1792 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1793 btrfs_release_path(path);
1795 item = btrfs_item_nr(slot);
1796 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1798 for (cur = 0; cur < btrfs_item_size(eb, item); cur += len) {
1799 name_len = btrfs_inode_ref_name_len(eb, iref);
1800 /* path must be released before calling iterate()! */
1801 pr_debug("following ref at offset %u for inode %llu in "
1802 "tree %llu\n", cur, found_key.objectid,
1804 ret = iterate(parent, name_len,
1805 (unsigned long)(iref + 1), eb, ctx);
1808 len = sizeof(*iref) + name_len;
1809 iref = (struct btrfs_inode_ref *)((char *)iref + len);
1811 btrfs_tree_read_unlock_blocking(eb);
1812 free_extent_buffer(eb);
1815 btrfs_release_path(path);
1820 static int iterate_inode_extrefs(u64 inum, struct btrfs_root *fs_root,
1821 struct btrfs_path *path,
1822 iterate_irefs_t *iterate, void *ctx)
1829 struct extent_buffer *eb;
1830 struct btrfs_inode_extref *extref;
1836 ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref,
1841 ret = found ? 0 : -ENOENT;
1846 slot = path->slots[0];
1847 eb = btrfs_clone_extent_buffer(path->nodes[0]);
1852 extent_buffer_get(eb);
1854 btrfs_tree_read_lock(eb);
1855 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1856 btrfs_release_path(path);
1858 item_size = btrfs_item_size_nr(eb, slot);
1859 ptr = btrfs_item_ptr_offset(eb, slot);
1862 while (cur_offset < item_size) {
1865 extref = (struct btrfs_inode_extref *)(ptr + cur_offset);
1866 parent = btrfs_inode_extref_parent(eb, extref);
1867 name_len = btrfs_inode_extref_name_len(eb, extref);
1868 ret = iterate(parent, name_len,
1869 (unsigned long)&extref->name, eb, ctx);
1873 cur_offset += btrfs_inode_extref_name_len(eb, extref);
1874 cur_offset += sizeof(*extref);
1876 btrfs_tree_read_unlock_blocking(eb);
1877 free_extent_buffer(eb);
1882 btrfs_release_path(path);
1887 static int iterate_irefs(u64 inum, struct btrfs_root *fs_root,
1888 struct btrfs_path *path, iterate_irefs_t *iterate,
1894 ret = iterate_inode_refs(inum, fs_root, path, iterate, ctx);
1897 else if (ret != -ENOENT)
1900 ret = iterate_inode_extrefs(inum, fs_root, path, iterate, ctx);
1901 if (ret == -ENOENT && found_refs)
1908 * returns 0 if the path could be dumped (probably truncated)
1909 * returns <0 in case of an error
1911 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
1912 struct extent_buffer *eb, void *ctx)
1914 struct inode_fs_paths *ipath = ctx;
1917 int i = ipath->fspath->elem_cnt;
1918 const int s_ptr = sizeof(char *);
1921 bytes_left = ipath->fspath->bytes_left > s_ptr ?
1922 ipath->fspath->bytes_left - s_ptr : 0;
1924 fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
1925 fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len,
1926 name_off, eb, inum, fspath_min, bytes_left);
1928 return PTR_ERR(fspath);
1930 if (fspath > fspath_min) {
1931 ipath->fspath->val[i] = (u64)(unsigned long)fspath;
1932 ++ipath->fspath->elem_cnt;
1933 ipath->fspath->bytes_left = fspath - fspath_min;
1935 ++ipath->fspath->elem_missed;
1936 ipath->fspath->bytes_missing += fspath_min - fspath;
1937 ipath->fspath->bytes_left = 0;
1944 * this dumps all file system paths to the inode into the ipath struct, provided
1945 * is has been created large enough. each path is zero-terminated and accessed
1946 * from ipath->fspath->val[i].
1947 * when it returns, there are ipath->fspath->elem_cnt number of paths available
1948 * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the
1949 * number of missed paths in recored in ipath->fspath->elem_missed, otherwise,
1950 * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
1951 * have been needed to return all paths.
1953 int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
1955 return iterate_irefs(inum, ipath->fs_root, ipath->btrfs_path,
1956 inode_to_path, ipath);
1959 struct btrfs_data_container *init_data_container(u32 total_bytes)
1961 struct btrfs_data_container *data;
1964 alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
1965 data = vmalloc(alloc_bytes);
1967 return ERR_PTR(-ENOMEM);
1969 if (total_bytes >= sizeof(*data)) {
1970 data->bytes_left = total_bytes - sizeof(*data);
1971 data->bytes_missing = 0;
1973 data->bytes_missing = sizeof(*data) - total_bytes;
1974 data->bytes_left = 0;
1978 data->elem_missed = 0;
1984 * allocates space to return multiple file system paths for an inode.
1985 * total_bytes to allocate are passed, note that space usable for actual path
1986 * information will be total_bytes - sizeof(struct inode_fs_paths).
1987 * the returned pointer must be freed with free_ipath() in the end.
1989 struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
1990 struct btrfs_path *path)
1992 struct inode_fs_paths *ifp;
1993 struct btrfs_data_container *fspath;
1995 fspath = init_data_container(total_bytes);
1997 return (void *)fspath;
1999 ifp = kmalloc(sizeof(*ifp), GFP_NOFS);
2002 return ERR_PTR(-ENOMEM);
2005 ifp->btrfs_path = path;
2006 ifp->fspath = fspath;
2007 ifp->fs_root = fs_root;
2012 void free_ipath(struct inode_fs_paths *ipath)
2016 vfree(ipath->fspath);