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.
23 #include "transaction.h"
24 #include "delayed-ref.h"
27 struct extent_inode_elem {
30 struct extent_inode_elem *next;
33 static int check_extent_in_eb(struct btrfs_key *key, struct extent_buffer *eb,
34 struct btrfs_file_extent_item *fi,
36 struct extent_inode_elem **eie)
40 struct extent_inode_elem *e;
42 data_offset = btrfs_file_extent_offset(eb, fi);
43 data_len = btrfs_file_extent_num_bytes(eb, fi);
45 if (extent_item_pos < data_offset ||
46 extent_item_pos >= data_offset + data_len)
49 e = kmalloc(sizeof(*e), GFP_NOFS);
54 e->inum = key->objectid;
55 e->offset = key->offset + (extent_item_pos - data_offset);
61 static int find_extent_in_eb(struct extent_buffer *eb, u64 wanted_disk_byte,
63 struct extent_inode_elem **eie)
67 struct btrfs_file_extent_item *fi;
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.
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)
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)
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)
92 ret = check_extent_in_eb(&key, eb, fi, extent_item_pos, eie);
101 * this structure records all encountered refs on the way up to the root
103 struct __prelim_ref {
104 struct list_head list;
106 struct btrfs_key key_for_search;
109 struct extent_inode_elem *inode_list;
111 u64 wanted_disk_byte;
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
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
131 * - column 1: we've the parent -> done
132 * - column 2, 3, 4: we use the key to find the parent
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
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
149 * additional information that's available but not required to find the parent
150 * block might help in merging entries to gain some speed.
153 static 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)
157 struct __prelim_ref *ref;
159 /* in case we're adding delayed refs, we're holding the refs spinlock */
160 ref = kmalloc(sizeof(*ref), GFP_ATOMIC);
164 ref->root_id = root_id;
166 ref->key_for_search = *key;
168 memset(&ref->key_for_search, 0, sizeof(ref->key_for_search));
170 ref->inode_list = NULL;
173 ref->parent = parent;
174 ref->wanted_disk_byte = wanted_disk_byte;
175 list_add_tail(&ref->list, head);
180 static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path,
181 struct ulist *parents, int level,
182 struct btrfs_key *key, u64 time_seq,
183 u64 wanted_disk_byte,
184 const u64 *extent_item_pos)
187 int slot = path->slots[level];
188 struct extent_buffer *eb = path->nodes[level];
189 struct btrfs_file_extent_item *fi;
190 struct extent_inode_elem *eie = NULL;
192 u64 wanted_objectid = key->objectid;
195 if (level == 0 && extent_item_pos) {
196 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
197 ret = check_extent_in_eb(key, eb, fi, *extent_item_pos, &eie);
201 ret = ulist_add(parents, eb->start, (unsigned long)eie, GFP_NOFS);
209 * if the current leaf is full with EXTENT_DATA items, we must
210 * check the next one if that holds a reference as well.
211 * ref->count cannot be used to skip this check.
212 * repeat this until we don't find any additional EXTENT_DATA items.
216 ret = btrfs_next_old_leaf(root, path, time_seq);
223 for (slot = 0; slot < btrfs_header_nritems(eb); ++slot) {
224 btrfs_item_key_to_cpu(eb, key, slot);
225 if (key->objectid != wanted_objectid ||
226 key->type != BTRFS_EXTENT_DATA_KEY)
228 fi = btrfs_item_ptr(eb, slot,
229 struct btrfs_file_extent_item);
230 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
231 if (disk_byte == wanted_disk_byte)
240 * resolve an indirect backref in the form (root_id, key, level)
241 * to a logical address
243 static int __resolve_indirect_ref(struct btrfs_fs_info *fs_info,
244 int search_commit_root,
246 struct __prelim_ref *ref,
247 struct ulist *parents,
248 const u64 *extent_item_pos)
250 struct btrfs_path *path;
251 struct btrfs_root *root;
252 struct btrfs_key root_key;
253 struct btrfs_key key = {0};
254 struct extent_buffer *eb;
257 int level = ref->level;
259 path = btrfs_alloc_path();
262 path->search_commit_root = !!search_commit_root;
264 root_key.objectid = ref->root_id;
265 root_key.type = BTRFS_ROOT_ITEM_KEY;
266 root_key.offset = (u64)-1;
267 root = btrfs_read_fs_root_no_name(fs_info, &root_key);
274 root_level = btrfs_header_level(root->node);
277 if (root_level + 1 == level)
280 path->lowest_level = level;
281 ret = btrfs_search_old_slot(root, &ref->key_for_search, path, time_seq);
282 pr_debug("search slot in root %llu (level %d, ref count %d) returned "
283 "%d for key (%llu %u %llu)\n",
284 (unsigned long long)ref->root_id, level, ref->count, ret,
285 (unsigned long long)ref->key_for_search.objectid,
286 ref->key_for_search.type,
287 (unsigned long long)ref->key_for_search.offset);
291 eb = path->nodes[level];
299 btrfs_item_key_to_cpu(eb, &key, path->slots[0]);
301 ret = add_all_parents(root, path, parents, level, &key, time_seq,
302 ref->wanted_disk_byte, extent_item_pos);
304 btrfs_free_path(path);
309 * resolve all indirect backrefs from the list
311 static int __resolve_indirect_refs(struct btrfs_fs_info *fs_info,
312 int search_commit_root, u64 time_seq,
313 struct list_head *head,
314 const u64 *extent_item_pos)
318 struct __prelim_ref *ref;
319 struct __prelim_ref *ref_safe;
320 struct __prelim_ref *new_ref;
321 struct ulist *parents;
322 struct ulist_node *node;
323 struct ulist_iterator uiter;
325 parents = ulist_alloc(GFP_NOFS);
330 * _safe allows us to insert directly after the current item without
331 * iterating over the newly inserted items.
332 * we're also allowed to re-assign ref during iteration.
334 list_for_each_entry_safe(ref, ref_safe, head, list) {
335 if (ref->parent) /* already direct */
339 err = __resolve_indirect_ref(fs_info, search_commit_root,
340 time_seq, ref, parents,
348 /* we put the first parent into the ref at hand */
349 ULIST_ITER_INIT(&uiter);
350 node = ulist_next(parents, &uiter);
351 ref->parent = node ? node->val : 0;
353 node ? (struct extent_inode_elem *)node->aux : 0;
355 /* additional parents require new refs being added here */
356 while ((node = ulist_next(parents, &uiter))) {
357 new_ref = kmalloc(sizeof(*new_ref), GFP_NOFS);
362 memcpy(new_ref, ref, sizeof(*ref));
363 new_ref->parent = node->val;
364 new_ref->inode_list =
365 (struct extent_inode_elem *)node->aux;
366 list_add(&new_ref->list, &ref->list);
368 ulist_reinit(parents);
375 static inline int ref_for_same_block(struct __prelim_ref *ref1,
376 struct __prelim_ref *ref2)
378 if (ref1->level != ref2->level)
380 if (ref1->root_id != ref2->root_id)
382 if (ref1->key_for_search.type != ref2->key_for_search.type)
384 if (ref1->key_for_search.objectid != ref2->key_for_search.objectid)
386 if (ref1->key_for_search.offset != ref2->key_for_search.offset)
388 if (ref1->parent != ref2->parent)
395 * read tree blocks and add keys where required.
397 static int __add_missing_keys(struct btrfs_fs_info *fs_info,
398 struct list_head *head)
400 struct list_head *pos;
401 struct extent_buffer *eb;
403 list_for_each(pos, head) {
404 struct __prelim_ref *ref;
405 ref = list_entry(pos, struct __prelim_ref, list);
409 if (ref->key_for_search.type)
411 BUG_ON(!ref->wanted_disk_byte);
412 eb = read_tree_block(fs_info->tree_root, ref->wanted_disk_byte,
413 fs_info->tree_root->leafsize, 0);
415 btrfs_tree_read_lock(eb);
416 if (btrfs_header_level(eb) == 0)
417 btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
419 btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
420 btrfs_tree_read_unlock(eb);
421 free_extent_buffer(eb);
427 * merge two lists of backrefs and adjust counts accordingly
429 * mode = 1: merge identical keys, if key is set
430 * FIXME: if we add more keys in __add_prelim_ref, we can merge more here.
431 * additionally, we could even add a key range for the blocks we
432 * looked into to merge even more (-> replace unresolved refs by those
434 * mode = 2: merge identical parents
436 static int __merge_refs(struct list_head *head, int mode)
438 struct list_head *pos1;
440 list_for_each(pos1, head) {
441 struct list_head *n2;
442 struct list_head *pos2;
443 struct __prelim_ref *ref1;
445 ref1 = list_entry(pos1, struct __prelim_ref, list);
447 for (pos2 = pos1->next, n2 = pos2->next; pos2 != head;
448 pos2 = n2, n2 = pos2->next) {
449 struct __prelim_ref *ref2;
450 struct __prelim_ref *xchg;
452 ref2 = list_entry(pos2, struct __prelim_ref, list);
455 if (!ref_for_same_block(ref1, ref2))
457 if (!ref1->parent && ref2->parent) {
462 ref1->count += ref2->count;
464 if (ref1->parent != ref2->parent)
466 ref1->count += ref2->count;
468 list_del(&ref2->list);
477 * add all currently queued delayed refs from this head whose seq nr is
478 * smaller or equal that seq to the list
480 static int __add_delayed_refs(struct btrfs_delayed_ref_head *head, u64 seq,
481 struct list_head *prefs)
483 struct btrfs_delayed_extent_op *extent_op = head->extent_op;
484 struct rb_node *n = &head->node.rb_node;
485 struct btrfs_key key;
486 struct btrfs_key op_key = {0};
490 if (extent_op && extent_op->update_key)
491 btrfs_disk_key_to_cpu(&op_key, &extent_op->key);
493 while ((n = rb_prev(n))) {
494 struct btrfs_delayed_ref_node *node;
495 node = rb_entry(n, struct btrfs_delayed_ref_node,
497 if (node->bytenr != head->node.bytenr)
499 WARN_ON(node->is_head);
504 switch (node->action) {
505 case BTRFS_ADD_DELAYED_EXTENT:
506 case BTRFS_UPDATE_DELAYED_HEAD:
509 case BTRFS_ADD_DELAYED_REF:
512 case BTRFS_DROP_DELAYED_REF:
518 switch (node->type) {
519 case BTRFS_TREE_BLOCK_REF_KEY: {
520 struct btrfs_delayed_tree_ref *ref;
522 ref = btrfs_delayed_node_to_tree_ref(node);
523 ret = __add_prelim_ref(prefs, ref->root, &op_key,
524 ref->level + 1, 0, node->bytenr,
525 node->ref_mod * sgn);
528 case BTRFS_SHARED_BLOCK_REF_KEY: {
529 struct btrfs_delayed_tree_ref *ref;
531 ref = btrfs_delayed_node_to_tree_ref(node);
532 ret = __add_prelim_ref(prefs, ref->root, NULL,
533 ref->level + 1, ref->parent,
535 node->ref_mod * sgn);
538 case BTRFS_EXTENT_DATA_REF_KEY: {
539 struct btrfs_delayed_data_ref *ref;
540 ref = btrfs_delayed_node_to_data_ref(node);
542 key.objectid = ref->objectid;
543 key.type = BTRFS_EXTENT_DATA_KEY;
544 key.offset = ref->offset;
545 ret = __add_prelim_ref(prefs, ref->root, &key, 0, 0,
547 node->ref_mod * sgn);
550 case BTRFS_SHARED_DATA_REF_KEY: {
551 struct btrfs_delayed_data_ref *ref;
553 ref = btrfs_delayed_node_to_data_ref(node);
555 key.objectid = ref->objectid;
556 key.type = BTRFS_EXTENT_DATA_KEY;
557 key.offset = ref->offset;
558 ret = __add_prelim_ref(prefs, ref->root, &key, 0,
559 ref->parent, node->bytenr,
560 node->ref_mod * sgn);
573 * add all inline backrefs for bytenr to the list
575 static int __add_inline_refs(struct btrfs_fs_info *fs_info,
576 struct btrfs_path *path, u64 bytenr,
577 int *info_level, struct list_head *prefs)
581 struct extent_buffer *leaf;
582 struct btrfs_key key;
585 struct btrfs_extent_item *ei;
590 * enumerate all inline refs
592 leaf = path->nodes[0];
593 slot = path->slots[0];
595 item_size = btrfs_item_size_nr(leaf, slot);
596 BUG_ON(item_size < sizeof(*ei));
598 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
599 flags = btrfs_extent_flags(leaf, ei);
601 ptr = (unsigned long)(ei + 1);
602 end = (unsigned long)ei + item_size;
604 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
605 struct btrfs_tree_block_info *info;
607 info = (struct btrfs_tree_block_info *)ptr;
608 *info_level = btrfs_tree_block_level(leaf, info);
609 ptr += sizeof(struct btrfs_tree_block_info);
612 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
616 struct btrfs_extent_inline_ref *iref;
620 iref = (struct btrfs_extent_inline_ref *)ptr;
621 type = btrfs_extent_inline_ref_type(leaf, iref);
622 offset = btrfs_extent_inline_ref_offset(leaf, iref);
625 case BTRFS_SHARED_BLOCK_REF_KEY:
626 ret = __add_prelim_ref(prefs, 0, NULL,
627 *info_level + 1, offset,
630 case BTRFS_SHARED_DATA_REF_KEY: {
631 struct btrfs_shared_data_ref *sdref;
634 sdref = (struct btrfs_shared_data_ref *)(iref + 1);
635 count = btrfs_shared_data_ref_count(leaf, sdref);
636 ret = __add_prelim_ref(prefs, 0, NULL, 0, offset,
640 case BTRFS_TREE_BLOCK_REF_KEY:
641 ret = __add_prelim_ref(prefs, offset, NULL,
645 case BTRFS_EXTENT_DATA_REF_KEY: {
646 struct btrfs_extent_data_ref *dref;
650 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
651 count = btrfs_extent_data_ref_count(leaf, dref);
652 key.objectid = btrfs_extent_data_ref_objectid(leaf,
654 key.type = BTRFS_EXTENT_DATA_KEY;
655 key.offset = btrfs_extent_data_ref_offset(leaf, dref);
656 root = btrfs_extent_data_ref_root(leaf, dref);
657 ret = __add_prelim_ref(prefs, root, &key, 0, 0,
665 ptr += btrfs_extent_inline_ref_size(type);
672 * add all non-inline backrefs for bytenr to the list
674 static int __add_keyed_refs(struct btrfs_fs_info *fs_info,
675 struct btrfs_path *path, u64 bytenr,
676 int info_level, struct list_head *prefs)
678 struct btrfs_root *extent_root = fs_info->extent_root;
681 struct extent_buffer *leaf;
682 struct btrfs_key key;
685 ret = btrfs_next_item(extent_root, path);
693 slot = path->slots[0];
694 leaf = path->nodes[0];
695 btrfs_item_key_to_cpu(leaf, &key, slot);
697 if (key.objectid != bytenr)
699 if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
701 if (key.type > BTRFS_SHARED_DATA_REF_KEY)
705 case BTRFS_SHARED_BLOCK_REF_KEY:
706 ret = __add_prelim_ref(prefs, 0, NULL,
707 info_level + 1, key.offset,
710 case BTRFS_SHARED_DATA_REF_KEY: {
711 struct btrfs_shared_data_ref *sdref;
714 sdref = btrfs_item_ptr(leaf, slot,
715 struct btrfs_shared_data_ref);
716 count = btrfs_shared_data_ref_count(leaf, sdref);
717 ret = __add_prelim_ref(prefs, 0, NULL, 0, key.offset,
721 case BTRFS_TREE_BLOCK_REF_KEY:
722 ret = __add_prelim_ref(prefs, key.offset, NULL,
726 case BTRFS_EXTENT_DATA_REF_KEY: {
727 struct btrfs_extent_data_ref *dref;
731 dref = btrfs_item_ptr(leaf, slot,
732 struct btrfs_extent_data_ref);
733 count = btrfs_extent_data_ref_count(leaf, dref);
734 key.objectid = btrfs_extent_data_ref_objectid(leaf,
736 key.type = BTRFS_EXTENT_DATA_KEY;
737 key.offset = btrfs_extent_data_ref_offset(leaf, dref);
738 root = btrfs_extent_data_ref_root(leaf, dref);
739 ret = __add_prelim_ref(prefs, root, &key, 0, 0,
753 * this adds all existing backrefs (inline backrefs, backrefs and delayed
754 * refs) for the given bytenr to the refs list, merges duplicates and resolves
755 * indirect refs to their parent bytenr.
756 * When roots are found, they're added to the roots list
758 * FIXME some caching might speed things up
760 static int find_parent_nodes(struct btrfs_trans_handle *trans,
761 struct btrfs_fs_info *fs_info, u64 bytenr,
762 u64 delayed_ref_seq, u64 time_seq,
763 struct ulist *refs, struct ulist *roots,
764 const u64 *extent_item_pos)
766 struct btrfs_key key;
767 struct btrfs_path *path;
768 struct btrfs_delayed_ref_root *delayed_refs = NULL;
769 struct btrfs_delayed_ref_head *head;
772 int search_commit_root = (trans == BTRFS_BACKREF_SEARCH_COMMIT_ROOT);
773 struct list_head prefs_delayed;
774 struct list_head prefs;
775 struct __prelim_ref *ref;
777 INIT_LIST_HEAD(&prefs);
778 INIT_LIST_HEAD(&prefs_delayed);
780 key.objectid = bytenr;
781 key.type = BTRFS_EXTENT_ITEM_KEY;
782 key.offset = (u64)-1;
784 path = btrfs_alloc_path();
787 path->search_commit_root = !!search_commit_root;
790 * grab both a lock on the path and a lock on the delayed ref head.
791 * We need both to get a consistent picture of how the refs look
792 * at a specified point in time
797 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
802 if (trans != BTRFS_BACKREF_SEARCH_COMMIT_ROOT) {
804 * look if there are updates for this ref queued and lock the
807 delayed_refs = &trans->transaction->delayed_refs;
808 spin_lock(&delayed_refs->lock);
809 head = btrfs_find_delayed_ref_head(trans, bytenr);
811 if (!mutex_trylock(&head->mutex)) {
812 atomic_inc(&head->node.refs);
813 spin_unlock(&delayed_refs->lock);
815 btrfs_release_path(path);
818 * Mutex was contended, block until it's
819 * released and try again
821 mutex_lock(&head->mutex);
822 mutex_unlock(&head->mutex);
823 btrfs_put_delayed_ref(&head->node);
826 ret = __add_delayed_refs(head, delayed_ref_seq,
829 spin_unlock(&delayed_refs->lock);
833 spin_unlock(&delayed_refs->lock);
836 if (path->slots[0]) {
837 struct extent_buffer *leaf;
841 leaf = path->nodes[0];
842 slot = path->slots[0];
843 btrfs_item_key_to_cpu(leaf, &key, slot);
844 if (key.objectid == bytenr &&
845 key.type == BTRFS_EXTENT_ITEM_KEY) {
846 ret = __add_inline_refs(fs_info, path, bytenr,
847 &info_level, &prefs);
850 ret = __add_keyed_refs(fs_info, path, bytenr,
856 btrfs_release_path(path);
858 list_splice_init(&prefs_delayed, &prefs);
860 ret = __add_missing_keys(fs_info, &prefs);
864 ret = __merge_refs(&prefs, 1);
868 ret = __resolve_indirect_refs(fs_info, search_commit_root, time_seq,
869 &prefs, extent_item_pos);
873 ret = __merge_refs(&prefs, 2);
877 while (!list_empty(&prefs)) {
878 ref = list_first_entry(&prefs, struct __prelim_ref, list);
879 list_del(&ref->list);
882 if (ref->count && ref->root_id && ref->parent == 0) {
883 /* no parent == root of tree */
884 ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS);
887 if (ref->count && ref->parent) {
888 struct extent_inode_elem *eie = NULL;
889 if (extent_item_pos && !ref->inode_list) {
891 struct extent_buffer *eb;
892 bsz = btrfs_level_size(fs_info->extent_root,
894 eb = read_tree_block(fs_info->extent_root,
895 ref->parent, bsz, 0);
897 ret = find_extent_in_eb(eb, bytenr,
898 *extent_item_pos, &eie);
899 ref->inode_list = eie;
900 free_extent_buffer(eb);
902 ret = ulist_add_merge(refs, ref->parent,
903 (unsigned long)ref->inode_list,
904 (unsigned long *)&eie, GFP_NOFS);
905 if (!ret && extent_item_pos) {
907 * we've recorded that parent, so we must extend
908 * its inode list here
913 eie->next = ref->inode_list;
922 mutex_unlock(&head->mutex);
923 btrfs_free_path(path);
924 while (!list_empty(&prefs)) {
925 ref = list_first_entry(&prefs, struct __prelim_ref, list);
926 list_del(&ref->list);
929 while (!list_empty(&prefs_delayed)) {
930 ref = list_first_entry(&prefs_delayed, struct __prelim_ref,
932 list_del(&ref->list);
939 static void free_leaf_list(struct ulist *blocks)
941 struct ulist_node *node = NULL;
942 struct extent_inode_elem *eie;
943 struct extent_inode_elem *eie_next;
944 struct ulist_iterator uiter;
946 ULIST_ITER_INIT(&uiter);
947 while ((node = ulist_next(blocks, &uiter))) {
950 eie = (struct extent_inode_elem *)node->aux;
951 for (; eie; eie = eie_next) {
952 eie_next = eie->next;
962 * Finds all leafs with a reference to the specified combination of bytenr and
963 * offset. key_list_head will point to a list of corresponding keys (caller must
964 * free each list element). The leafs will be stored in the leafs ulist, which
965 * must be freed with ulist_free.
967 * returns 0 on success, <0 on error
969 static int btrfs_find_all_leafs(struct btrfs_trans_handle *trans,
970 struct btrfs_fs_info *fs_info, u64 bytenr,
971 u64 delayed_ref_seq, u64 time_seq,
972 struct ulist **leafs,
973 const u64 *extent_item_pos)
978 tmp = ulist_alloc(GFP_NOFS);
981 *leafs = ulist_alloc(GFP_NOFS);
987 ret = find_parent_nodes(trans, fs_info, bytenr, delayed_ref_seq,
988 time_seq, *leafs, tmp, extent_item_pos);
991 if (ret < 0 && ret != -ENOENT) {
992 free_leaf_list(*leafs);
1000 * walk all backrefs for a given extent to find all roots that reference this
1001 * extent. Walking a backref means finding all extents that reference this
1002 * extent and in turn walk the backrefs of those, too. Naturally this is a
1003 * recursive process, but here it is implemented in an iterative fashion: We
1004 * find all referencing extents for the extent in question and put them on a
1005 * list. In turn, we find all referencing extents for those, further appending
1006 * to the list. The way we iterate the list allows adding more elements after
1007 * the current while iterating. The process stops when we reach the end of the
1008 * list. Found roots are added to the roots list.
1010 * returns 0 on success, < 0 on error.
1012 int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1013 struct btrfs_fs_info *fs_info, u64 bytenr,
1014 u64 delayed_ref_seq, u64 time_seq,
1015 struct ulist **roots)
1018 struct ulist_node *node = NULL;
1019 struct ulist_iterator uiter;
1022 tmp = ulist_alloc(GFP_NOFS);
1025 *roots = ulist_alloc(GFP_NOFS);
1031 ULIST_ITER_INIT(&uiter);
1033 ret = find_parent_nodes(trans, fs_info, bytenr, delayed_ref_seq,
1034 time_seq, tmp, *roots, NULL);
1035 if (ret < 0 && ret != -ENOENT) {
1040 node = ulist_next(tmp, &uiter);
1051 static int __inode_info(u64 inum, u64 ioff, u8 key_type,
1052 struct btrfs_root *fs_root, struct btrfs_path *path,
1053 struct btrfs_key *found_key)
1056 struct btrfs_key key;
1057 struct extent_buffer *eb;
1059 key.type = key_type;
1060 key.objectid = inum;
1063 ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
1067 eb = path->nodes[0];
1068 if (ret && path->slots[0] >= btrfs_header_nritems(eb)) {
1069 ret = btrfs_next_leaf(fs_root, path);
1072 eb = path->nodes[0];
1075 btrfs_item_key_to_cpu(eb, found_key, path->slots[0]);
1076 if (found_key->type != key.type || found_key->objectid != key.objectid)
1083 * this makes the path point to (inum INODE_ITEM ioff)
1085 int inode_item_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
1086 struct btrfs_path *path)
1088 struct btrfs_key key;
1089 return __inode_info(inum, ioff, BTRFS_INODE_ITEM_KEY, fs_root, path,
1093 static int inode_ref_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
1094 struct btrfs_path *path,
1095 struct btrfs_key *found_key)
1097 return __inode_info(inum, ioff, BTRFS_INODE_REF_KEY, fs_root, path,
1102 * this iterates to turn a btrfs_inode_ref into a full filesystem path. elements
1103 * of the path are separated by '/' and the path is guaranteed to be
1104 * 0-terminated. the path is only given within the current file system.
1105 * Therefore, it never starts with a '/'. the caller is responsible to provide
1106 * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
1107 * the start point of the resulting string is returned. this pointer is within
1109 * in case the path buffer would overflow, the pointer is decremented further
1110 * as if output was written to the buffer, though no more output is actually
1111 * generated. that way, the caller can determine how much space would be
1112 * required for the path to fit into the buffer. in that case, the returned
1113 * value will be smaller than dest. callers must check this!
1115 static char *iref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
1116 struct btrfs_inode_ref *iref,
1117 struct extent_buffer *eb_in, u64 parent,
1118 char *dest, u32 size)
1124 s64 bytes_left = size - 1;
1125 struct extent_buffer *eb = eb_in;
1126 struct btrfs_key found_key;
1127 int leave_spinning = path->leave_spinning;
1129 if (bytes_left >= 0)
1130 dest[bytes_left] = '\0';
1132 path->leave_spinning = 1;
1134 len = btrfs_inode_ref_name_len(eb, iref);
1136 if (bytes_left >= 0)
1137 read_extent_buffer(eb, dest + bytes_left,
1138 (unsigned long)(iref + 1), len);
1140 btrfs_tree_read_unlock_blocking(eb);
1141 free_extent_buffer(eb);
1143 ret = inode_ref_info(parent, 0, fs_root, path, &found_key);
1148 next_inum = found_key.offset;
1150 /* regular exit ahead */
1151 if (parent == next_inum)
1154 slot = path->slots[0];
1155 eb = path->nodes[0];
1156 /* make sure we can use eb after releasing the path */
1158 atomic_inc(&eb->refs);
1159 btrfs_tree_read_lock(eb);
1160 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1162 btrfs_release_path(path);
1164 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1167 if (bytes_left >= 0)
1168 dest[bytes_left] = '/';
1171 btrfs_release_path(path);
1172 path->leave_spinning = leave_spinning;
1175 return ERR_PTR(ret);
1177 return dest + bytes_left;
1181 * this makes the path point to (logical EXTENT_ITEM *)
1182 * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
1183 * tree blocks and <0 on error.
1185 int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
1186 struct btrfs_path *path, struct btrfs_key *found_key)
1191 struct extent_buffer *eb;
1192 struct btrfs_extent_item *ei;
1193 struct btrfs_key key;
1195 key.type = BTRFS_EXTENT_ITEM_KEY;
1196 key.objectid = logical;
1197 key.offset = (u64)-1;
1199 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
1202 ret = btrfs_previous_item(fs_info->extent_root, path,
1203 0, BTRFS_EXTENT_ITEM_KEY);
1207 btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]);
1208 if (found_key->type != BTRFS_EXTENT_ITEM_KEY ||
1209 found_key->objectid > logical ||
1210 found_key->objectid + found_key->offset <= logical) {
1211 pr_debug("logical %llu is not within any extent\n",
1212 (unsigned long long)logical);
1216 eb = path->nodes[0];
1217 item_size = btrfs_item_size_nr(eb, path->slots[0]);
1218 BUG_ON(item_size < sizeof(*ei));
1220 ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
1221 flags = btrfs_extent_flags(eb, ei);
1223 pr_debug("logical %llu is at position %llu within the extent (%llu "
1224 "EXTENT_ITEM %llu) flags %#llx size %u\n",
1225 (unsigned long long)logical,
1226 (unsigned long long)(logical - found_key->objectid),
1227 (unsigned long long)found_key->objectid,
1228 (unsigned long long)found_key->offset,
1229 (unsigned long long)flags, item_size);
1230 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1231 return BTRFS_EXTENT_FLAG_TREE_BLOCK;
1232 if (flags & BTRFS_EXTENT_FLAG_DATA)
1233 return BTRFS_EXTENT_FLAG_DATA;
1239 * helper function to iterate extent inline refs. ptr must point to a 0 value
1240 * for the first call and may be modified. it is used to track state.
1241 * if more refs exist, 0 is returned and the next call to
1242 * __get_extent_inline_ref must pass the modified ptr parameter to get the
1243 * next ref. after the last ref was processed, 1 is returned.
1244 * returns <0 on error
1246 static int __get_extent_inline_ref(unsigned long *ptr, struct extent_buffer *eb,
1247 struct btrfs_extent_item *ei, u32 item_size,
1248 struct btrfs_extent_inline_ref **out_eiref,
1253 struct btrfs_tree_block_info *info;
1257 flags = btrfs_extent_flags(eb, ei);
1258 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1259 info = (struct btrfs_tree_block_info *)(ei + 1);
1261 (struct btrfs_extent_inline_ref *)(info + 1);
1263 *out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1);
1265 *ptr = (unsigned long)*out_eiref;
1266 if ((void *)*ptr >= (void *)ei + item_size)
1270 end = (unsigned long)ei + item_size;
1271 *out_eiref = (struct btrfs_extent_inline_ref *)*ptr;
1272 *out_type = btrfs_extent_inline_ref_type(eb, *out_eiref);
1274 *ptr += btrfs_extent_inline_ref_size(*out_type);
1275 WARN_ON(*ptr > end);
1277 return 1; /* last */
1283 * reads the tree block backref for an extent. tree level and root are returned
1284 * through out_level and out_root. ptr must point to a 0 value for the first
1285 * call and may be modified (see __get_extent_inline_ref comment).
1286 * returns 0 if data was provided, 1 if there was no more data to provide or
1289 int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
1290 struct btrfs_extent_item *ei, u32 item_size,
1291 u64 *out_root, u8 *out_level)
1295 struct btrfs_tree_block_info *info;
1296 struct btrfs_extent_inline_ref *eiref;
1298 if (*ptr == (unsigned long)-1)
1302 ret = __get_extent_inline_ref(ptr, eb, ei, item_size,
1307 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1308 type == BTRFS_SHARED_BLOCK_REF_KEY)
1315 /* we can treat both ref types equally here */
1316 info = (struct btrfs_tree_block_info *)(ei + 1);
1317 *out_root = btrfs_extent_inline_ref_offset(eb, eiref);
1318 *out_level = btrfs_tree_block_level(eb, info);
1321 *ptr = (unsigned long)-1;
1326 static int iterate_leaf_refs(struct extent_inode_elem *inode_list,
1327 u64 root, u64 extent_item_objectid,
1328 iterate_extent_inodes_t *iterate, void *ctx)
1330 struct extent_inode_elem *eie;
1333 for (eie = inode_list; eie; eie = eie->next) {
1334 pr_debug("ref for %llu resolved, key (%llu EXTEND_DATA %llu), "
1335 "root %llu\n", extent_item_objectid,
1336 eie->inum, eie->offset, root);
1337 ret = iterate(eie->inum, eie->offset, root, ctx);
1339 pr_debug("stopping iteration for %llu due to ret=%d\n",
1340 extent_item_objectid, ret);
1349 * calls iterate() for every inode that references the extent identified by
1350 * the given parameters.
1351 * when the iterator function returns a non-zero value, iteration stops.
1353 int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
1354 u64 extent_item_objectid, u64 extent_item_pos,
1355 int search_commit_root,
1356 iterate_extent_inodes_t *iterate, void *ctx)
1359 struct list_head data_refs = LIST_HEAD_INIT(data_refs);
1360 struct list_head shared_refs = LIST_HEAD_INIT(shared_refs);
1361 struct btrfs_trans_handle *trans;
1362 struct ulist *refs = NULL;
1363 struct ulist *roots = NULL;
1364 struct ulist_node *ref_node = NULL;
1365 struct ulist_node *root_node = NULL;
1366 struct seq_list seq_elem = {};
1367 struct seq_list tree_mod_seq_elem = {};
1368 struct ulist_iterator ref_uiter;
1369 struct ulist_iterator root_uiter;
1370 struct btrfs_delayed_ref_root *delayed_refs = NULL;
1372 pr_debug("resolving all inodes for extent %llu\n",
1373 extent_item_objectid);
1375 if (search_commit_root) {
1376 trans = BTRFS_BACKREF_SEARCH_COMMIT_ROOT;
1378 trans = btrfs_join_transaction(fs_info->extent_root);
1380 return PTR_ERR(trans);
1382 delayed_refs = &trans->transaction->delayed_refs;
1383 spin_lock(&delayed_refs->lock);
1384 btrfs_get_delayed_seq(delayed_refs, &seq_elem);
1385 spin_unlock(&delayed_refs->lock);
1386 btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1389 ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid,
1390 seq_elem.seq, tree_mod_seq_elem.seq, &refs,
1395 ULIST_ITER_INIT(&ref_uiter);
1396 while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) {
1397 ret = btrfs_find_all_roots(trans, fs_info, ref_node->val,
1399 tree_mod_seq_elem.seq, &roots);
1402 ULIST_ITER_INIT(&root_uiter);
1403 while (!ret && (root_node = ulist_next(roots, &root_uiter))) {
1404 pr_debug("root %llu references leaf %llu, data list "
1405 "%#lx\n", root_node->val, ref_node->val,
1407 ret = iterate_leaf_refs(
1408 (struct extent_inode_elem *)ref_node->aux,
1409 root_node->val, extent_item_objectid,
1416 free_leaf_list(refs);
1419 if (!search_commit_root) {
1420 btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1421 btrfs_put_delayed_seq(delayed_refs, &seq_elem);
1422 btrfs_end_transaction(trans, fs_info->extent_root);
1428 int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
1429 struct btrfs_path *path,
1430 iterate_extent_inodes_t *iterate, void *ctx)
1433 u64 extent_item_pos;
1434 struct btrfs_key found_key;
1435 int search_commit_root = path->search_commit_root;
1437 ret = extent_from_logical(fs_info, logical, path,
1439 btrfs_release_path(path);
1440 if (ret & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1445 extent_item_pos = logical - found_key.objectid;
1446 ret = iterate_extent_inodes(fs_info, found_key.objectid,
1447 extent_item_pos, search_commit_root,
1453 static int iterate_irefs(u64 inum, struct btrfs_root *fs_root,
1454 struct btrfs_path *path,
1455 iterate_irefs_t *iterate, void *ctx)
1464 struct extent_buffer *eb;
1465 struct btrfs_item *item;
1466 struct btrfs_inode_ref *iref;
1467 struct btrfs_key found_key;
1470 path->leave_spinning = 1;
1471 ret = inode_ref_info(inum, parent ? parent+1 : 0, fs_root, path,
1476 ret = found ? 0 : -ENOENT;
1481 parent = found_key.offset;
1482 slot = path->slots[0];
1483 eb = path->nodes[0];
1484 /* make sure we can use eb after releasing the path */
1485 atomic_inc(&eb->refs);
1486 btrfs_tree_read_lock(eb);
1487 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1488 btrfs_release_path(path);
1490 item = btrfs_item_nr(eb, slot);
1491 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1493 for (cur = 0; cur < btrfs_item_size(eb, item); cur += len) {
1494 name_len = btrfs_inode_ref_name_len(eb, iref);
1495 /* path must be released before calling iterate()! */
1496 pr_debug("following ref at offset %u for inode %llu in "
1498 (unsigned long long)found_key.objectid,
1499 (unsigned long long)fs_root->objectid);
1500 ret = iterate(parent, iref, eb, ctx);
1503 len = sizeof(*iref) + name_len;
1504 iref = (struct btrfs_inode_ref *)((char *)iref + len);
1506 btrfs_tree_read_unlock_blocking(eb);
1507 free_extent_buffer(eb);
1510 btrfs_release_path(path);
1516 * returns 0 if the path could be dumped (probably truncated)
1517 * returns <0 in case of an error
1519 static int inode_to_path(u64 inum, struct btrfs_inode_ref *iref,
1520 struct extent_buffer *eb, void *ctx)
1522 struct inode_fs_paths *ipath = ctx;
1525 int i = ipath->fspath->elem_cnt;
1526 const int s_ptr = sizeof(char *);
1529 bytes_left = ipath->fspath->bytes_left > s_ptr ?
1530 ipath->fspath->bytes_left - s_ptr : 0;
1532 fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
1533 fspath = iref_to_path(ipath->fs_root, ipath->btrfs_path, iref, eb,
1534 inum, fspath_min, bytes_left);
1536 return PTR_ERR(fspath);
1538 if (fspath > fspath_min) {
1539 pr_debug("path resolved: %s\n", fspath);
1540 ipath->fspath->val[i] = (u64)(unsigned long)fspath;
1541 ++ipath->fspath->elem_cnt;
1542 ipath->fspath->bytes_left = fspath - fspath_min;
1544 pr_debug("missed path, not enough space. missing bytes: %lu, "
1545 "constructed so far: %s\n",
1546 (unsigned long)(fspath_min - fspath), fspath_min);
1547 ++ipath->fspath->elem_missed;
1548 ipath->fspath->bytes_missing += fspath_min - fspath;
1549 ipath->fspath->bytes_left = 0;
1556 * this dumps all file system paths to the inode into the ipath struct, provided
1557 * is has been created large enough. each path is zero-terminated and accessed
1558 * from ipath->fspath->val[i].
1559 * when it returns, there are ipath->fspath->elem_cnt number of paths available
1560 * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the
1561 * number of missed paths in recored in ipath->fspath->elem_missed, otherwise,
1562 * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
1563 * have been needed to return all paths.
1565 int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
1567 return iterate_irefs(inum, ipath->fs_root, ipath->btrfs_path,
1568 inode_to_path, ipath);
1571 struct btrfs_data_container *init_data_container(u32 total_bytes)
1573 struct btrfs_data_container *data;
1576 alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
1577 data = kmalloc(alloc_bytes, GFP_NOFS);
1579 return ERR_PTR(-ENOMEM);
1581 if (total_bytes >= sizeof(*data)) {
1582 data->bytes_left = total_bytes - sizeof(*data);
1583 data->bytes_missing = 0;
1585 data->bytes_missing = sizeof(*data) - total_bytes;
1586 data->bytes_left = 0;
1590 data->elem_missed = 0;
1596 * allocates space to return multiple file system paths for an inode.
1597 * total_bytes to allocate are passed, note that space usable for actual path
1598 * information will be total_bytes - sizeof(struct inode_fs_paths).
1599 * the returned pointer must be freed with free_ipath() in the end.
1601 struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
1602 struct btrfs_path *path)
1604 struct inode_fs_paths *ifp;
1605 struct btrfs_data_container *fspath;
1607 fspath = init_data_container(total_bytes);
1609 return (void *)fspath;
1611 ifp = kmalloc(sizeof(*ifp), GFP_NOFS);
1614 return ERR_PTR(-ENOMEM);
1617 ifp->btrfs_path = path;
1618 ifp->fspath = fspath;
1619 ifp->fs_root = fs_root;
1624 void free_ipath(struct inode_fs_paths *ipath)
1628 kfree(ipath->fspath);