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 struct extent_inode_elem {
31 struct extent_inode_elem *next;
34 static int check_extent_in_eb(struct btrfs_key *key, struct extent_buffer *eb,
35 struct btrfs_file_extent_item *fi,
37 struct extent_inode_elem **eie)
41 struct extent_inode_elem *e;
43 data_offset = btrfs_file_extent_offset(eb, fi);
44 data_len = btrfs_file_extent_num_bytes(eb, fi);
46 if (extent_item_pos < data_offset ||
47 extent_item_pos >= data_offset + data_len)
50 e = kmalloc(sizeof(*e), GFP_NOFS);
55 e->inum = key->objectid;
56 e->offset = key->offset + (extent_item_pos - data_offset);
62 static int find_extent_in_eb(struct extent_buffer *eb, u64 wanted_disk_byte,
64 struct extent_inode_elem **eie)
68 struct btrfs_file_extent_item *fi;
75 * from the shared data ref, we only have the leaf but we need
76 * the key. thus, we must look into all items and see that we
77 * find one (some) with a reference to our extent item.
79 nritems = btrfs_header_nritems(eb);
80 for (slot = 0; slot < nritems; ++slot) {
81 btrfs_item_key_to_cpu(eb, &key, slot);
82 if (key.type != BTRFS_EXTENT_DATA_KEY)
84 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
85 extent_type = btrfs_file_extent_type(eb, fi);
86 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
88 /* don't skip BTRFS_FILE_EXTENT_PREALLOC, we can handle that */
89 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
90 if (disk_byte != wanted_disk_byte)
93 ret = check_extent_in_eb(&key, eb, fi, extent_item_pos, eie);
102 * this structure records all encountered refs on the way up to the root
104 struct __prelim_ref {
105 struct list_head list;
107 struct btrfs_key key_for_search;
110 struct extent_inode_elem *inode_list;
112 u64 wanted_disk_byte;
116 * the rules for all callers of this function are:
117 * - obtaining the parent is the goal
118 * - if you add a key, you must know that it is a correct key
119 * - if you cannot add the parent or a correct key, then we will look into the
120 * block later to set a correct key
124 * backref type | shared | indirect | shared | indirect
125 * information | tree | tree | data | data
126 * --------------------+--------+----------+--------+----------
127 * parent logical | y | - | - | -
128 * key to resolve | - | y | y | y
129 * tree block logical | - | - | - | -
130 * root for resolving | y | y | y | y
132 * - column 1: we've the parent -> done
133 * - column 2, 3, 4: we use the key to find the parent
135 * on disk refs (inline or keyed)
136 * ==============================
137 * backref type | shared | indirect | shared | indirect
138 * information | tree | tree | data | data
139 * --------------------+--------+----------+--------+----------
140 * parent logical | y | - | y | -
141 * key to resolve | - | - | - | y
142 * tree block logical | y | y | y | y
143 * root for resolving | - | y | y | y
145 * - column 1, 3: we've the parent -> done
146 * - column 2: we take the first key from the block to find the parent
147 * (see __add_missing_keys)
148 * - column 4: we use the key to find the parent
150 * additional information that's available but not required to find the parent
151 * block might help in merging entries to gain some speed.
154 static int __add_prelim_ref(struct list_head *head, u64 root_id,
155 struct btrfs_key *key, int level,
156 u64 parent, u64 wanted_disk_byte, int count)
158 struct __prelim_ref *ref;
160 /* in case we're adding delayed refs, we're holding the refs spinlock */
161 ref = kmalloc(sizeof(*ref), GFP_ATOMIC);
165 ref->root_id = root_id;
167 ref->key_for_search = *key;
169 memset(&ref->key_for_search, 0, sizeof(ref->key_for_search));
171 ref->inode_list = NULL;
174 ref->parent = parent;
175 ref->wanted_disk_byte = wanted_disk_byte;
176 list_add_tail(&ref->list, head);
181 static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path,
182 struct ulist *parents, int level,
183 struct btrfs_key *key_for_search, u64 time_seq,
184 u64 wanted_disk_byte,
185 const u64 *extent_item_pos)
189 struct extent_buffer *eb;
190 struct btrfs_key key;
191 struct btrfs_file_extent_item *fi;
192 struct extent_inode_elem *eie = NULL;
196 eb = path->nodes[level];
197 ret = ulist_add(parents, eb->start, 0, GFP_NOFS);
204 * We normally enter this function with the path already pointing to
205 * the first item to check. But sometimes, we may enter it with
206 * slot==nritems. In that case, go to the next leaf before we continue.
208 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0]))
209 ret = btrfs_next_old_leaf(root, path, time_seq);
213 slot = path->slots[0];
215 btrfs_item_key_to_cpu(eb, &key, slot);
217 if (key.objectid != key_for_search->objectid ||
218 key.type != BTRFS_EXTENT_DATA_KEY)
221 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
222 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
224 if (disk_byte == wanted_disk_byte) {
226 if (extent_item_pos) {
227 ret = check_extent_in_eb(&key, eb, fi,
234 ret = ulist_add(parents, eb->start,
235 (uintptr_t)eie, GFP_NOFS);
238 if (!extent_item_pos) {
239 ret = btrfs_next_old_leaf(root, path,
245 ret = btrfs_next_old_item(root, path, time_seq);
254 * resolve an indirect backref in the form (root_id, key, level)
255 * to a logical address
257 static int __resolve_indirect_ref(struct btrfs_fs_info *fs_info,
258 int search_commit_root,
260 struct __prelim_ref *ref,
261 struct ulist *parents,
262 const u64 *extent_item_pos)
264 struct btrfs_path *path;
265 struct btrfs_root *root;
266 struct btrfs_key root_key;
267 struct extent_buffer *eb;
270 int level = ref->level;
272 path = btrfs_alloc_path();
275 path->search_commit_root = !!search_commit_root;
277 root_key.objectid = ref->root_id;
278 root_key.type = BTRFS_ROOT_ITEM_KEY;
279 root_key.offset = (u64)-1;
280 root = btrfs_read_fs_root_no_name(fs_info, &root_key);
286 root_level = btrfs_old_root_level(root, time_seq);
288 if (root_level + 1 == level)
291 path->lowest_level = level;
292 ret = btrfs_search_old_slot(root, &ref->key_for_search, path, time_seq);
293 pr_debug("search slot in root %llu (level %d, ref count %d) returned "
294 "%d for key (%llu %u %llu)\n",
295 (unsigned long long)ref->root_id, level, ref->count, ret,
296 (unsigned long long)ref->key_for_search.objectid,
297 ref->key_for_search.type,
298 (unsigned long long)ref->key_for_search.offset);
302 eb = path->nodes[level];
310 eb = path->nodes[level];
313 ret = add_all_parents(root, path, parents, level, &ref->key_for_search,
314 time_seq, ref->wanted_disk_byte,
317 btrfs_free_path(path);
322 * resolve all indirect backrefs from the list
324 static int __resolve_indirect_refs(struct btrfs_fs_info *fs_info,
325 int search_commit_root, u64 time_seq,
326 struct list_head *head,
327 const u64 *extent_item_pos)
331 struct __prelim_ref *ref;
332 struct __prelim_ref *ref_safe;
333 struct __prelim_ref *new_ref;
334 struct ulist *parents;
335 struct ulist_node *node;
336 struct ulist_iterator uiter;
338 parents = ulist_alloc(GFP_NOFS);
343 * _safe allows us to insert directly after the current item without
344 * iterating over the newly inserted items.
345 * we're also allowed to re-assign ref during iteration.
347 list_for_each_entry_safe(ref, ref_safe, head, list) {
348 if (ref->parent) /* already direct */
352 err = __resolve_indirect_ref(fs_info, search_commit_root,
353 time_seq, ref, parents,
361 /* we put the first parent into the ref at hand */
362 ULIST_ITER_INIT(&uiter);
363 node = ulist_next(parents, &uiter);
364 ref->parent = node ? node->val : 0;
365 ref->inode_list = node ?
366 (struct extent_inode_elem *)(uintptr_t)node->aux : 0;
368 /* additional parents require new refs being added here */
369 while ((node = ulist_next(parents, &uiter))) {
370 new_ref = kmalloc(sizeof(*new_ref), GFP_NOFS);
375 memcpy(new_ref, ref, sizeof(*ref));
376 new_ref->parent = node->val;
377 new_ref->inode_list = (struct extent_inode_elem *)
378 (uintptr_t)node->aux;
379 list_add(&new_ref->list, &ref->list);
381 ulist_reinit(parents);
388 static inline int ref_for_same_block(struct __prelim_ref *ref1,
389 struct __prelim_ref *ref2)
391 if (ref1->level != ref2->level)
393 if (ref1->root_id != ref2->root_id)
395 if (ref1->key_for_search.type != ref2->key_for_search.type)
397 if (ref1->key_for_search.objectid != ref2->key_for_search.objectid)
399 if (ref1->key_for_search.offset != ref2->key_for_search.offset)
401 if (ref1->parent != ref2->parent)
408 * read tree blocks and add keys where required.
410 static int __add_missing_keys(struct btrfs_fs_info *fs_info,
411 struct list_head *head)
413 struct list_head *pos;
414 struct extent_buffer *eb;
416 list_for_each(pos, head) {
417 struct __prelim_ref *ref;
418 ref = list_entry(pos, struct __prelim_ref, list);
422 if (ref->key_for_search.type)
424 BUG_ON(!ref->wanted_disk_byte);
425 eb = read_tree_block(fs_info->tree_root, ref->wanted_disk_byte,
426 fs_info->tree_root->leafsize, 0);
428 btrfs_tree_read_lock(eb);
429 if (btrfs_header_level(eb) == 0)
430 btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
432 btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
433 btrfs_tree_read_unlock(eb);
434 free_extent_buffer(eb);
440 * merge two lists of backrefs and adjust counts accordingly
442 * mode = 1: merge identical keys, if key is set
443 * FIXME: if we add more keys in __add_prelim_ref, we can merge more here.
444 * additionally, we could even add a key range for the blocks we
445 * looked into to merge even more (-> replace unresolved refs by those
447 * mode = 2: merge identical parents
449 static int __merge_refs(struct list_head *head, int mode)
451 struct list_head *pos1;
453 list_for_each(pos1, head) {
454 struct list_head *n2;
455 struct list_head *pos2;
456 struct __prelim_ref *ref1;
458 ref1 = list_entry(pos1, struct __prelim_ref, list);
460 for (pos2 = pos1->next, n2 = pos2->next; pos2 != head;
461 pos2 = n2, n2 = pos2->next) {
462 struct __prelim_ref *ref2;
463 struct __prelim_ref *xchg;
465 ref2 = list_entry(pos2, struct __prelim_ref, list);
468 if (!ref_for_same_block(ref1, ref2))
470 if (!ref1->parent && ref2->parent) {
475 ref1->count += ref2->count;
477 if (ref1->parent != ref2->parent)
479 ref1->count += ref2->count;
481 list_del(&ref2->list);
490 * add all currently queued delayed refs from this head whose seq nr is
491 * smaller or equal that seq to the list
493 static int __add_delayed_refs(struct btrfs_delayed_ref_head *head, u64 seq,
494 struct list_head *prefs)
496 struct btrfs_delayed_extent_op *extent_op = head->extent_op;
497 struct rb_node *n = &head->node.rb_node;
498 struct btrfs_key key;
499 struct btrfs_key op_key = {0};
503 if (extent_op && extent_op->update_key)
504 btrfs_disk_key_to_cpu(&op_key, &extent_op->key);
506 while ((n = rb_prev(n))) {
507 struct btrfs_delayed_ref_node *node;
508 node = rb_entry(n, struct btrfs_delayed_ref_node,
510 if (node->bytenr != head->node.bytenr)
512 WARN_ON(node->is_head);
517 switch (node->action) {
518 case BTRFS_ADD_DELAYED_EXTENT:
519 case BTRFS_UPDATE_DELAYED_HEAD:
522 case BTRFS_ADD_DELAYED_REF:
525 case BTRFS_DROP_DELAYED_REF:
531 switch (node->type) {
532 case BTRFS_TREE_BLOCK_REF_KEY: {
533 struct btrfs_delayed_tree_ref *ref;
535 ref = btrfs_delayed_node_to_tree_ref(node);
536 ret = __add_prelim_ref(prefs, ref->root, &op_key,
537 ref->level + 1, 0, node->bytenr,
538 node->ref_mod * sgn);
541 case BTRFS_SHARED_BLOCK_REF_KEY: {
542 struct btrfs_delayed_tree_ref *ref;
544 ref = btrfs_delayed_node_to_tree_ref(node);
545 ret = __add_prelim_ref(prefs, ref->root, NULL,
546 ref->level + 1, ref->parent,
548 node->ref_mod * sgn);
551 case BTRFS_EXTENT_DATA_REF_KEY: {
552 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, 0,
560 node->ref_mod * sgn);
563 case BTRFS_SHARED_DATA_REF_KEY: {
564 struct btrfs_delayed_data_ref *ref;
566 ref = btrfs_delayed_node_to_data_ref(node);
568 key.objectid = ref->objectid;
569 key.type = BTRFS_EXTENT_DATA_KEY;
570 key.offset = ref->offset;
571 ret = __add_prelim_ref(prefs, ref->root, &key, 0,
572 ref->parent, node->bytenr,
573 node->ref_mod * sgn);
586 * add all inline backrefs for bytenr to the list
588 static int __add_inline_refs(struct btrfs_fs_info *fs_info,
589 struct btrfs_path *path, u64 bytenr,
590 int *info_level, struct list_head *prefs)
594 struct extent_buffer *leaf;
595 struct btrfs_key key;
598 struct btrfs_extent_item *ei;
603 * enumerate all inline refs
605 leaf = path->nodes[0];
606 slot = path->slots[0];
608 item_size = btrfs_item_size_nr(leaf, slot);
609 BUG_ON(item_size < sizeof(*ei));
611 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
612 flags = btrfs_extent_flags(leaf, ei);
614 ptr = (unsigned long)(ei + 1);
615 end = (unsigned long)ei + item_size;
617 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
618 struct btrfs_tree_block_info *info;
620 info = (struct btrfs_tree_block_info *)ptr;
621 *info_level = btrfs_tree_block_level(leaf, info);
622 ptr += sizeof(struct btrfs_tree_block_info);
625 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
629 struct btrfs_extent_inline_ref *iref;
633 iref = (struct btrfs_extent_inline_ref *)ptr;
634 type = btrfs_extent_inline_ref_type(leaf, iref);
635 offset = btrfs_extent_inline_ref_offset(leaf, iref);
638 case BTRFS_SHARED_BLOCK_REF_KEY:
639 ret = __add_prelim_ref(prefs, 0, NULL,
640 *info_level + 1, offset,
643 case BTRFS_SHARED_DATA_REF_KEY: {
644 struct btrfs_shared_data_ref *sdref;
647 sdref = (struct btrfs_shared_data_ref *)(iref + 1);
648 count = btrfs_shared_data_ref_count(leaf, sdref);
649 ret = __add_prelim_ref(prefs, 0, NULL, 0, offset,
653 case BTRFS_TREE_BLOCK_REF_KEY:
654 ret = __add_prelim_ref(prefs, offset, NULL,
658 case BTRFS_EXTENT_DATA_REF_KEY: {
659 struct btrfs_extent_data_ref *dref;
663 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
664 count = btrfs_extent_data_ref_count(leaf, dref);
665 key.objectid = btrfs_extent_data_ref_objectid(leaf,
667 key.type = BTRFS_EXTENT_DATA_KEY;
668 key.offset = btrfs_extent_data_ref_offset(leaf, dref);
669 root = btrfs_extent_data_ref_root(leaf, dref);
670 ret = __add_prelim_ref(prefs, root, &key, 0, 0,
678 ptr += btrfs_extent_inline_ref_size(type);
685 * add all non-inline backrefs for bytenr to the list
687 static int __add_keyed_refs(struct btrfs_fs_info *fs_info,
688 struct btrfs_path *path, u64 bytenr,
689 int info_level, struct list_head *prefs)
691 struct btrfs_root *extent_root = fs_info->extent_root;
694 struct extent_buffer *leaf;
695 struct btrfs_key key;
698 ret = btrfs_next_item(extent_root, path);
706 slot = path->slots[0];
707 leaf = path->nodes[0];
708 btrfs_item_key_to_cpu(leaf, &key, slot);
710 if (key.objectid != bytenr)
712 if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
714 if (key.type > BTRFS_SHARED_DATA_REF_KEY)
718 case BTRFS_SHARED_BLOCK_REF_KEY:
719 ret = __add_prelim_ref(prefs, 0, NULL,
720 info_level + 1, key.offset,
723 case BTRFS_SHARED_DATA_REF_KEY: {
724 struct btrfs_shared_data_ref *sdref;
727 sdref = btrfs_item_ptr(leaf, slot,
728 struct btrfs_shared_data_ref);
729 count = btrfs_shared_data_ref_count(leaf, sdref);
730 ret = __add_prelim_ref(prefs, 0, NULL, 0, key.offset,
734 case BTRFS_TREE_BLOCK_REF_KEY:
735 ret = __add_prelim_ref(prefs, key.offset, NULL,
739 case BTRFS_EXTENT_DATA_REF_KEY: {
740 struct btrfs_extent_data_ref *dref;
744 dref = btrfs_item_ptr(leaf, slot,
745 struct btrfs_extent_data_ref);
746 count = btrfs_extent_data_ref_count(leaf, dref);
747 key.objectid = btrfs_extent_data_ref_objectid(leaf,
749 key.type = BTRFS_EXTENT_DATA_KEY;
750 key.offset = btrfs_extent_data_ref_offset(leaf, dref);
751 root = btrfs_extent_data_ref_root(leaf, dref);
752 ret = __add_prelim_ref(prefs, root, &key, 0, 0,
766 * this adds all existing backrefs (inline backrefs, backrefs and delayed
767 * refs) for the given bytenr to the refs list, merges duplicates and resolves
768 * indirect refs to their parent bytenr.
769 * When roots are found, they're added to the roots list
771 * FIXME some caching might speed things up
773 static int find_parent_nodes(struct btrfs_trans_handle *trans,
774 struct btrfs_fs_info *fs_info, u64 bytenr,
775 u64 time_seq, struct ulist *refs,
776 struct ulist *roots, const u64 *extent_item_pos)
778 struct btrfs_key key;
779 struct btrfs_path *path;
780 struct btrfs_delayed_ref_root *delayed_refs = NULL;
781 struct btrfs_delayed_ref_head *head;
784 int search_commit_root = (trans == BTRFS_BACKREF_SEARCH_COMMIT_ROOT);
785 struct list_head prefs_delayed;
786 struct list_head prefs;
787 struct __prelim_ref *ref;
789 INIT_LIST_HEAD(&prefs);
790 INIT_LIST_HEAD(&prefs_delayed);
792 key.objectid = bytenr;
793 key.type = BTRFS_EXTENT_ITEM_KEY;
794 key.offset = (u64)-1;
796 path = btrfs_alloc_path();
799 path->search_commit_root = !!search_commit_root;
802 * grab both a lock on the path and a lock on the delayed ref head.
803 * We need both to get a consistent picture of how the refs look
804 * at a specified point in time
809 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
814 if (trans != BTRFS_BACKREF_SEARCH_COMMIT_ROOT) {
816 * look if there are updates for this ref queued and lock the
819 delayed_refs = &trans->transaction->delayed_refs;
820 spin_lock(&delayed_refs->lock);
821 head = btrfs_find_delayed_ref_head(trans, bytenr);
823 if (!mutex_trylock(&head->mutex)) {
824 atomic_inc(&head->node.refs);
825 spin_unlock(&delayed_refs->lock);
827 btrfs_release_path(path);
830 * Mutex was contended, block until it's
831 * released and try again
833 mutex_lock(&head->mutex);
834 mutex_unlock(&head->mutex);
835 btrfs_put_delayed_ref(&head->node);
838 ret = __add_delayed_refs(head, time_seq,
840 mutex_unlock(&head->mutex);
842 spin_unlock(&delayed_refs->lock);
846 spin_unlock(&delayed_refs->lock);
849 if (path->slots[0]) {
850 struct extent_buffer *leaf;
854 leaf = path->nodes[0];
855 slot = path->slots[0];
856 btrfs_item_key_to_cpu(leaf, &key, slot);
857 if (key.objectid == bytenr &&
858 key.type == BTRFS_EXTENT_ITEM_KEY) {
859 ret = __add_inline_refs(fs_info, path, bytenr,
860 &info_level, &prefs);
863 ret = __add_keyed_refs(fs_info, path, bytenr,
869 btrfs_release_path(path);
871 list_splice_init(&prefs_delayed, &prefs);
873 ret = __add_missing_keys(fs_info, &prefs);
877 ret = __merge_refs(&prefs, 1);
881 ret = __resolve_indirect_refs(fs_info, search_commit_root, time_seq,
882 &prefs, extent_item_pos);
886 ret = __merge_refs(&prefs, 2);
890 while (!list_empty(&prefs)) {
891 ref = list_first_entry(&prefs, struct __prelim_ref, list);
892 list_del(&ref->list);
895 if (ref->count && ref->root_id && ref->parent == 0) {
896 /* no parent == root of tree */
897 ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS);
900 if (ref->count && ref->parent) {
901 struct extent_inode_elem *eie = NULL;
902 if (extent_item_pos && !ref->inode_list) {
904 struct extent_buffer *eb;
905 bsz = btrfs_level_size(fs_info->extent_root,
907 eb = read_tree_block(fs_info->extent_root,
908 ref->parent, bsz, 0);
910 ret = find_extent_in_eb(eb, bytenr,
911 *extent_item_pos, &eie);
912 ref->inode_list = eie;
913 free_extent_buffer(eb);
915 ret = ulist_add_merge(refs, ref->parent,
916 (uintptr_t)ref->inode_list,
917 (u64 *)&eie, GFP_NOFS);
918 if (!ret && extent_item_pos) {
920 * we've recorded that parent, so we must extend
921 * its inode list here
926 eie->next = ref->inode_list;
934 btrfs_free_path(path);
935 while (!list_empty(&prefs)) {
936 ref = list_first_entry(&prefs, struct __prelim_ref, list);
937 list_del(&ref->list);
940 while (!list_empty(&prefs_delayed)) {
941 ref = list_first_entry(&prefs_delayed, struct __prelim_ref,
943 list_del(&ref->list);
950 static void free_leaf_list(struct ulist *blocks)
952 struct ulist_node *node = NULL;
953 struct extent_inode_elem *eie;
954 struct extent_inode_elem *eie_next;
955 struct ulist_iterator uiter;
957 ULIST_ITER_INIT(&uiter);
958 while ((node = ulist_next(blocks, &uiter))) {
961 eie = (struct extent_inode_elem *)(uintptr_t)node->aux;
962 for (; eie; eie = eie_next) {
963 eie_next = eie->next;
973 * Finds all leafs with a reference to the specified combination of bytenr and
974 * offset. key_list_head will point to a list of corresponding keys (caller must
975 * free each list element). The leafs will be stored in the leafs ulist, which
976 * must be freed with ulist_free.
978 * returns 0 on success, <0 on error
980 static int btrfs_find_all_leafs(struct btrfs_trans_handle *trans,
981 struct btrfs_fs_info *fs_info, u64 bytenr,
982 u64 time_seq, struct ulist **leafs,
983 const u64 *extent_item_pos)
988 tmp = ulist_alloc(GFP_NOFS);
991 *leafs = ulist_alloc(GFP_NOFS);
997 ret = find_parent_nodes(trans, fs_info, bytenr,
998 time_seq, *leafs, tmp, extent_item_pos);
1001 if (ret < 0 && ret != -ENOENT) {
1002 free_leaf_list(*leafs);
1010 * walk all backrefs for a given extent to find all roots that reference this
1011 * extent. Walking a backref means finding all extents that reference this
1012 * extent and in turn walk the backrefs of those, too. Naturally this is a
1013 * recursive process, but here it is implemented in an iterative fashion: We
1014 * find all referencing extents for the extent in question and put them on a
1015 * list. In turn, we find all referencing extents for those, further appending
1016 * to the list. The way we iterate the list allows adding more elements after
1017 * the current while iterating. The process stops when we reach the end of the
1018 * list. Found roots are added to the roots list.
1020 * returns 0 on success, < 0 on error.
1022 int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1023 struct btrfs_fs_info *fs_info, u64 bytenr,
1024 u64 time_seq, struct ulist **roots)
1027 struct ulist_node *node = NULL;
1028 struct ulist_iterator uiter;
1031 tmp = ulist_alloc(GFP_NOFS);
1034 *roots = ulist_alloc(GFP_NOFS);
1040 ULIST_ITER_INIT(&uiter);
1042 ret = find_parent_nodes(trans, fs_info, bytenr,
1043 time_seq, tmp, *roots, NULL);
1044 if (ret < 0 && ret != -ENOENT) {
1049 node = ulist_next(tmp, &uiter);
1060 static int __inode_info(u64 inum, u64 ioff, u8 key_type,
1061 struct btrfs_root *fs_root, struct btrfs_path *path,
1062 struct btrfs_key *found_key)
1065 struct btrfs_key key;
1066 struct extent_buffer *eb;
1068 key.type = key_type;
1069 key.objectid = inum;
1072 ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
1076 eb = path->nodes[0];
1077 if (ret && path->slots[0] >= btrfs_header_nritems(eb)) {
1078 ret = btrfs_next_leaf(fs_root, path);
1081 eb = path->nodes[0];
1084 btrfs_item_key_to_cpu(eb, found_key, path->slots[0]);
1085 if (found_key->type != key.type || found_key->objectid != key.objectid)
1092 * this makes the path point to (inum INODE_ITEM ioff)
1094 int inode_item_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
1095 struct btrfs_path *path)
1097 struct btrfs_key key;
1098 return __inode_info(inum, ioff, BTRFS_INODE_ITEM_KEY, fs_root, path,
1102 static int inode_ref_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
1103 struct btrfs_path *path,
1104 struct btrfs_key *found_key)
1106 return __inode_info(inum, ioff, BTRFS_INODE_REF_KEY, fs_root, path,
1110 int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,
1111 u64 start_off, struct btrfs_path *path,
1112 struct btrfs_inode_extref **ret_extref,
1116 struct btrfs_key key;
1117 struct btrfs_key found_key;
1118 struct btrfs_inode_extref *extref;
1119 struct extent_buffer *leaf;
1122 key.objectid = inode_objectid;
1123 btrfs_set_key_type(&key, BTRFS_INODE_EXTREF_KEY);
1124 key.offset = start_off;
1126 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1131 leaf = path->nodes[0];
1132 slot = path->slots[0];
1133 if (slot >= btrfs_header_nritems(leaf)) {
1135 * If the item at offset is not found,
1136 * btrfs_search_slot will point us to the slot
1137 * where it should be inserted. In our case
1138 * that will be the slot directly before the
1139 * next INODE_REF_KEY_V2 item. In the case
1140 * that we're pointing to the last slot in a
1141 * leaf, we must move one leaf over.
1143 ret = btrfs_next_leaf(root, path);
1152 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1155 * Check that we're still looking at an extended ref key for
1156 * this particular objectid. If we have different
1157 * objectid or type then there are no more to be found
1158 * in the tree and we can exit.
1161 if (found_key.objectid != inode_objectid)
1163 if (btrfs_key_type(&found_key) != BTRFS_INODE_EXTREF_KEY)
1167 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1168 extref = (struct btrfs_inode_extref *)ptr;
1169 *ret_extref = extref;
1171 *found_off = found_key.offset;
1178 char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
1179 u32 name_len, unsigned long name_off,
1180 struct extent_buffer *eb_in, u64 parent,
1181 char *dest, u32 size)
1186 s64 bytes_left = ((s64)size) - 1;
1187 struct extent_buffer *eb = eb_in;
1188 struct btrfs_key found_key;
1189 int leave_spinning = path->leave_spinning;
1190 struct btrfs_inode_ref *iref;
1192 if (bytes_left >= 0)
1193 dest[bytes_left] = '\0';
1195 path->leave_spinning = 1;
1197 bytes_left -= name_len;
1198 if (bytes_left >= 0)
1199 read_extent_buffer(eb, dest + bytes_left,
1200 name_off, name_len);
1202 btrfs_tree_read_unlock_blocking(eb);
1203 free_extent_buffer(eb);
1205 ret = inode_ref_info(parent, 0, fs_root, path, &found_key);
1211 next_inum = found_key.offset;
1213 /* regular exit ahead */
1214 if (parent == next_inum)
1217 slot = path->slots[0];
1218 eb = path->nodes[0];
1219 /* make sure we can use eb after releasing the path */
1221 atomic_inc(&eb->refs);
1222 btrfs_tree_read_lock(eb);
1223 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1225 btrfs_release_path(path);
1226 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1228 name_len = btrfs_inode_ref_name_len(eb, iref);
1229 name_off = (unsigned long)(iref + 1);
1233 if (bytes_left >= 0)
1234 dest[bytes_left] = '/';
1237 btrfs_release_path(path);
1238 path->leave_spinning = leave_spinning;
1241 return ERR_PTR(ret);
1243 return dest + bytes_left;
1247 * this iterates to turn a btrfs_inode_ref into a full filesystem path. elements
1248 * of the path are separated by '/' and the path is guaranteed to be
1249 * 0-terminated. the path is only given within the current file system.
1250 * Therefore, it never starts with a '/'. the caller is responsible to provide
1251 * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
1252 * the start point of the resulting string is returned. this pointer is within
1254 * in case the path buffer would overflow, the pointer is decremented further
1255 * as if output was written to the buffer, though no more output is actually
1256 * generated. that way, the caller can determine how much space would be
1257 * required for the path to fit into the buffer. in that case, the returned
1258 * value will be smaller than dest. callers must check this!
1260 char *btrfs_iref_to_path(struct btrfs_root *fs_root,
1261 struct btrfs_path *path,
1262 struct btrfs_inode_ref *iref,
1263 struct extent_buffer *eb_in, u64 parent,
1264 char *dest, u32 size)
1266 return btrfs_ref_to_path(fs_root, path,
1267 btrfs_inode_ref_name_len(eb_in, iref),
1268 (unsigned long)(iref + 1),
1269 eb_in, parent, dest, size);
1273 * this makes the path point to (logical EXTENT_ITEM *)
1274 * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
1275 * tree blocks and <0 on error.
1277 int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
1278 struct btrfs_path *path, struct btrfs_key *found_key,
1284 struct extent_buffer *eb;
1285 struct btrfs_extent_item *ei;
1286 struct btrfs_key key;
1288 key.type = BTRFS_EXTENT_ITEM_KEY;
1289 key.objectid = logical;
1290 key.offset = (u64)-1;
1292 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
1295 ret = btrfs_previous_item(fs_info->extent_root, path,
1296 0, BTRFS_EXTENT_ITEM_KEY);
1300 btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]);
1301 if (found_key->type != BTRFS_EXTENT_ITEM_KEY ||
1302 found_key->objectid > logical ||
1303 found_key->objectid + found_key->offset <= logical) {
1304 pr_debug("logical %llu is not within any extent\n",
1305 (unsigned long long)logical);
1309 eb = path->nodes[0];
1310 item_size = btrfs_item_size_nr(eb, path->slots[0]);
1311 BUG_ON(item_size < sizeof(*ei));
1313 ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
1314 flags = btrfs_extent_flags(eb, ei);
1316 pr_debug("logical %llu is at position %llu within the extent (%llu "
1317 "EXTENT_ITEM %llu) flags %#llx size %u\n",
1318 (unsigned long long)logical,
1319 (unsigned long long)(logical - found_key->objectid),
1320 (unsigned long long)found_key->objectid,
1321 (unsigned long long)found_key->offset,
1322 (unsigned long long)flags, item_size);
1324 WARN_ON(!flags_ret);
1326 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1327 *flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK;
1328 else if (flags & BTRFS_EXTENT_FLAG_DATA)
1329 *flags_ret = BTRFS_EXTENT_FLAG_DATA;
1339 * helper function to iterate extent inline refs. ptr must point to a 0 value
1340 * for the first call and may be modified. it is used to track state.
1341 * if more refs exist, 0 is returned and the next call to
1342 * __get_extent_inline_ref must pass the modified ptr parameter to get the
1343 * next ref. after the last ref was processed, 1 is returned.
1344 * returns <0 on error
1346 static int __get_extent_inline_ref(unsigned long *ptr, struct extent_buffer *eb,
1347 struct btrfs_extent_item *ei, u32 item_size,
1348 struct btrfs_extent_inline_ref **out_eiref,
1353 struct btrfs_tree_block_info *info;
1357 flags = btrfs_extent_flags(eb, ei);
1358 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1359 info = (struct btrfs_tree_block_info *)(ei + 1);
1361 (struct btrfs_extent_inline_ref *)(info + 1);
1363 *out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1);
1365 *ptr = (unsigned long)*out_eiref;
1366 if ((void *)*ptr >= (void *)ei + item_size)
1370 end = (unsigned long)ei + item_size;
1371 *out_eiref = (struct btrfs_extent_inline_ref *)*ptr;
1372 *out_type = btrfs_extent_inline_ref_type(eb, *out_eiref);
1374 *ptr += btrfs_extent_inline_ref_size(*out_type);
1375 WARN_ON(*ptr > end);
1377 return 1; /* last */
1383 * reads the tree block backref for an extent. tree level and root are returned
1384 * through out_level and out_root. ptr must point to a 0 value for the first
1385 * call and may be modified (see __get_extent_inline_ref comment).
1386 * returns 0 if data was provided, 1 if there was no more data to provide or
1389 int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
1390 struct btrfs_extent_item *ei, u32 item_size,
1391 u64 *out_root, u8 *out_level)
1395 struct btrfs_tree_block_info *info;
1396 struct btrfs_extent_inline_ref *eiref;
1398 if (*ptr == (unsigned long)-1)
1402 ret = __get_extent_inline_ref(ptr, eb, ei, item_size,
1407 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1408 type == BTRFS_SHARED_BLOCK_REF_KEY)
1415 /* we can treat both ref types equally here */
1416 info = (struct btrfs_tree_block_info *)(ei + 1);
1417 *out_root = btrfs_extent_inline_ref_offset(eb, eiref);
1418 *out_level = btrfs_tree_block_level(eb, info);
1421 *ptr = (unsigned long)-1;
1426 static int iterate_leaf_refs(struct extent_inode_elem *inode_list,
1427 u64 root, u64 extent_item_objectid,
1428 iterate_extent_inodes_t *iterate, void *ctx)
1430 struct extent_inode_elem *eie;
1433 for (eie = inode_list; eie; eie = eie->next) {
1434 pr_debug("ref for %llu resolved, key (%llu EXTEND_DATA %llu), "
1435 "root %llu\n", extent_item_objectid,
1436 eie->inum, eie->offset, root);
1437 ret = iterate(eie->inum, eie->offset, root, ctx);
1439 pr_debug("stopping iteration for %llu due to ret=%d\n",
1440 extent_item_objectid, ret);
1449 * calls iterate() for every inode that references the extent identified by
1450 * the given parameters.
1451 * when the iterator function returns a non-zero value, iteration stops.
1453 int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
1454 u64 extent_item_objectid, u64 extent_item_pos,
1455 int search_commit_root,
1456 iterate_extent_inodes_t *iterate, void *ctx)
1459 struct list_head data_refs = LIST_HEAD_INIT(data_refs);
1460 struct list_head shared_refs = LIST_HEAD_INIT(shared_refs);
1461 struct btrfs_trans_handle *trans;
1462 struct ulist *refs = NULL;
1463 struct ulist *roots = NULL;
1464 struct ulist_node *ref_node = NULL;
1465 struct ulist_node *root_node = NULL;
1466 struct seq_list tree_mod_seq_elem = {};
1467 struct ulist_iterator ref_uiter;
1468 struct ulist_iterator root_uiter;
1470 pr_debug("resolving all inodes for extent %llu\n",
1471 extent_item_objectid);
1473 if (search_commit_root) {
1474 trans = BTRFS_BACKREF_SEARCH_COMMIT_ROOT;
1476 trans = btrfs_join_transaction(fs_info->extent_root);
1478 return PTR_ERR(trans);
1479 btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1482 ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid,
1483 tree_mod_seq_elem.seq, &refs,
1488 ULIST_ITER_INIT(&ref_uiter);
1489 while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) {
1490 ret = btrfs_find_all_roots(trans, fs_info, ref_node->val,
1491 tree_mod_seq_elem.seq, &roots);
1494 ULIST_ITER_INIT(&root_uiter);
1495 while (!ret && (root_node = ulist_next(roots, &root_uiter))) {
1496 pr_debug("root %llu references leaf %llu, data list "
1497 "%#llx\n", root_node->val, ref_node->val,
1498 (long long)ref_node->aux);
1499 ret = iterate_leaf_refs((struct extent_inode_elem *)
1500 (uintptr_t)ref_node->aux,
1502 extent_item_objectid,
1509 free_leaf_list(refs);
1512 if (!search_commit_root) {
1513 btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1514 btrfs_end_transaction(trans, fs_info->extent_root);
1520 int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
1521 struct btrfs_path *path,
1522 iterate_extent_inodes_t *iterate, void *ctx)
1525 u64 extent_item_pos;
1527 struct btrfs_key found_key;
1528 int search_commit_root = path->search_commit_root;
1530 ret = extent_from_logical(fs_info, logical, path, &found_key, &flags);
1531 btrfs_release_path(path);
1534 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1537 extent_item_pos = logical - found_key.objectid;
1538 ret = iterate_extent_inodes(fs_info, found_key.objectid,
1539 extent_item_pos, search_commit_root,
1545 typedef int (iterate_irefs_t)(u64 parent, u32 name_len, unsigned long name_off,
1546 struct extent_buffer *eb, void *ctx);
1548 static int iterate_inode_refs(u64 inum, struct btrfs_root *fs_root,
1549 struct btrfs_path *path,
1550 iterate_irefs_t *iterate, void *ctx)
1559 struct extent_buffer *eb;
1560 struct btrfs_item *item;
1561 struct btrfs_inode_ref *iref;
1562 struct btrfs_key found_key;
1565 path->leave_spinning = 1;
1566 ret = inode_ref_info(inum, parent ? parent+1 : 0, fs_root, path,
1571 ret = found ? 0 : -ENOENT;
1576 parent = found_key.offset;
1577 slot = path->slots[0];
1578 eb = path->nodes[0];
1579 /* make sure we can use eb after releasing the path */
1580 atomic_inc(&eb->refs);
1581 btrfs_tree_read_lock(eb);
1582 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1583 btrfs_release_path(path);
1585 item = btrfs_item_nr(eb, slot);
1586 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1588 for (cur = 0; cur < btrfs_item_size(eb, item); cur += len) {
1589 name_len = btrfs_inode_ref_name_len(eb, iref);
1590 /* path must be released before calling iterate()! */
1591 pr_debug("following ref at offset %u for inode %llu in "
1593 (unsigned long long)found_key.objectid,
1594 (unsigned long long)fs_root->objectid);
1595 ret = iterate(parent, name_len,
1596 (unsigned long)(iref + 1), eb, ctx);
1599 len = sizeof(*iref) + name_len;
1600 iref = (struct btrfs_inode_ref *)((char *)iref + len);
1602 btrfs_tree_read_unlock_blocking(eb);
1603 free_extent_buffer(eb);
1606 btrfs_release_path(path);
1611 static int iterate_inode_extrefs(u64 inum, struct btrfs_root *fs_root,
1612 struct btrfs_path *path,
1613 iterate_irefs_t *iterate, void *ctx)
1620 struct extent_buffer *eb;
1621 struct btrfs_inode_extref *extref;
1622 struct extent_buffer *leaf;
1628 ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref,
1633 ret = found ? 0 : -ENOENT;
1638 slot = path->slots[0];
1639 eb = path->nodes[0];
1640 /* make sure we can use eb after releasing the path */
1641 atomic_inc(&eb->refs);
1643 btrfs_tree_read_lock(eb);
1644 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1645 btrfs_release_path(path);
1647 leaf = path->nodes[0];
1648 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1649 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1652 while (cur_offset < item_size) {
1655 extref = (struct btrfs_inode_extref *)(ptr + cur_offset);
1656 parent = btrfs_inode_extref_parent(eb, extref);
1657 name_len = btrfs_inode_extref_name_len(eb, extref);
1658 ret = iterate(parent, name_len,
1659 (unsigned long)&extref->name, eb, ctx);
1663 cur_offset += btrfs_inode_extref_name_len(leaf, extref);
1664 cur_offset += sizeof(*extref);
1666 btrfs_tree_read_unlock_blocking(eb);
1667 free_extent_buffer(eb);
1672 btrfs_release_path(path);
1677 static int iterate_irefs(u64 inum, struct btrfs_root *fs_root,
1678 struct btrfs_path *path, iterate_irefs_t *iterate,
1684 ret = iterate_inode_refs(inum, fs_root, path, iterate, ctx);
1687 else if (ret != -ENOENT)
1690 ret = iterate_inode_extrefs(inum, fs_root, path, iterate, ctx);
1691 if (ret == -ENOENT && found_refs)
1698 * returns 0 if the path could be dumped (probably truncated)
1699 * returns <0 in case of an error
1701 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
1702 struct extent_buffer *eb, void *ctx)
1704 struct inode_fs_paths *ipath = ctx;
1707 int i = ipath->fspath->elem_cnt;
1708 const int s_ptr = sizeof(char *);
1711 bytes_left = ipath->fspath->bytes_left > s_ptr ?
1712 ipath->fspath->bytes_left - s_ptr : 0;
1714 fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
1715 fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len,
1716 name_off, eb, inum, fspath_min, bytes_left);
1718 return PTR_ERR(fspath);
1720 if (fspath > fspath_min) {
1721 ipath->fspath->val[i] = (u64)(unsigned long)fspath;
1722 ++ipath->fspath->elem_cnt;
1723 ipath->fspath->bytes_left = fspath - fspath_min;
1725 ++ipath->fspath->elem_missed;
1726 ipath->fspath->bytes_missing += fspath_min - fspath;
1727 ipath->fspath->bytes_left = 0;
1734 * this dumps all file system paths to the inode into the ipath struct, provided
1735 * is has been created large enough. each path is zero-terminated and accessed
1736 * from ipath->fspath->val[i].
1737 * when it returns, there are ipath->fspath->elem_cnt number of paths available
1738 * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the
1739 * number of missed paths in recored in ipath->fspath->elem_missed, otherwise,
1740 * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
1741 * have been needed to return all paths.
1743 int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
1745 return iterate_irefs(inum, ipath->fs_root, ipath->btrfs_path,
1746 inode_to_path, ipath);
1749 struct btrfs_data_container *init_data_container(u32 total_bytes)
1751 struct btrfs_data_container *data;
1754 alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
1755 data = vmalloc(alloc_bytes);
1757 return ERR_PTR(-ENOMEM);
1759 if (total_bytes >= sizeof(*data)) {
1760 data->bytes_left = total_bytes - sizeof(*data);
1761 data->bytes_missing = 0;
1763 data->bytes_missing = sizeof(*data) - total_bytes;
1764 data->bytes_left = 0;
1768 data->elem_missed = 0;
1774 * allocates space to return multiple file system paths for an inode.
1775 * total_bytes to allocate are passed, note that space usable for actual path
1776 * information will be total_bytes - sizeof(struct inode_fs_paths).
1777 * the returned pointer must be freed with free_ipath() in the end.
1779 struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
1780 struct btrfs_path *path)
1782 struct inode_fs_paths *ifp;
1783 struct btrfs_data_container *fspath;
1785 fspath = init_data_container(total_bytes);
1787 return (void *)fspath;
1789 ifp = kmalloc(sizeof(*ifp), GFP_NOFS);
1792 return ERR_PTR(-ENOMEM);
1795 ifp->btrfs_path = path;
1796 ifp->fspath = fspath;
1797 ifp->fs_root = fs_root;
1802 void free_ipath(struct inode_fs_paths *ipath)
1806 vfree(ipath->fspath);