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 "kerncompat.h"
24 #include "transaction.h"
27 #define pr_debug(...) do { } while (0)
29 struct extent_inode_elem {
32 struct extent_inode_elem *next;
35 static int check_extent_in_eb(struct btrfs_key *key, struct extent_buffer *eb,
36 struct btrfs_file_extent_item *fi,
38 struct extent_inode_elem **eie)
41 struct extent_inode_elem *e;
43 if (!btrfs_file_extent_compression(eb, fi) &&
44 !btrfs_file_extent_encryption(eb, fi) &&
45 !btrfs_file_extent_other_encoding(eb, fi)) {
49 data_offset = btrfs_file_extent_offset(eb, fi);
50 data_len = btrfs_file_extent_num_bytes(eb, fi);
52 if (extent_item_pos < data_offset ||
53 extent_item_pos >= data_offset + data_len)
55 offset = extent_item_pos - data_offset;
58 e = kmalloc(sizeof(*e), GFP_NOFS);
63 e->inum = key->objectid;
64 e->offset = key->offset + offset;
70 static void free_inode_elem_list(struct extent_inode_elem *eie)
72 struct extent_inode_elem *eie_next;
74 for (; eie; eie = eie_next) {
80 static int find_extent_in_eb(struct extent_buffer *eb, u64 wanted_disk_byte,
82 struct extent_inode_elem **eie)
86 struct btrfs_file_extent_item *fi;
93 * from the shared data ref, we only have the leaf but we need
94 * the key. thus, we must look into all items and see that we
95 * find one (some) with a reference to our extent item.
97 nritems = btrfs_header_nritems(eb);
98 for (slot = 0; slot < nritems; ++slot) {
99 btrfs_item_key_to_cpu(eb, &key, slot);
100 if (key.type != BTRFS_EXTENT_DATA_KEY)
102 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
103 extent_type = btrfs_file_extent_type(eb, fi);
104 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
106 /* don't skip BTRFS_FILE_EXTENT_PREALLOC, we can handle that */
107 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
108 if (disk_byte != wanted_disk_byte)
111 ret = check_extent_in_eb(&key, eb, fi, extent_item_pos, eie);
120 * this structure records all encountered refs on the way up to the root
122 struct __prelim_ref {
123 struct list_head list;
125 struct btrfs_key key_for_search;
128 struct extent_inode_elem *inode_list;
130 u64 wanted_disk_byte;
134 * the rules for all callers of this function are:
135 * - obtaining the parent is the goal
136 * - if you add a key, you must know that it is a correct key
137 * - if you cannot add the parent or a correct key, then we will look into the
138 * block later to set a correct key
142 * backref type | shared | indirect | shared | indirect
143 * information | tree | tree | data | data
144 * --------------------+--------+----------+--------+----------
145 * parent logical | y | - | - | -
146 * key to resolve | - | y | y | y
147 * tree block logical | - | - | - | -
148 * root for resolving | y | y | y | y
150 * - column 1: we've the parent -> done
151 * - column 2, 3, 4: we use the key to find the parent
153 * on disk refs (inline or keyed)
154 * ==============================
155 * backref type | shared | indirect | shared | indirect
156 * information | tree | tree | data | data
157 * --------------------+--------+----------+--------+----------
158 * parent logical | y | - | y | -
159 * key to resolve | - | - | - | y
160 * tree block logical | y | y | y | y
161 * root for resolving | - | y | y | y
163 * - column 1, 3: we've the parent -> done
164 * - column 2: we take the first key from the block to find the parent
165 * (see __add_missing_keys)
166 * - column 4: we use the key to find the parent
168 * additional information that's available but not required to find the parent
169 * block might help in merging entries to gain some speed.
172 static int __add_prelim_ref(struct list_head *head, u64 root_id,
173 struct btrfs_key *key, int level,
174 u64 parent, u64 wanted_disk_byte, int count,
177 struct __prelim_ref *ref;
179 if (root_id == BTRFS_DATA_RELOC_TREE_OBJECTID)
182 ref = kmalloc(sizeof(*ref), gfp_mask);
186 ref->root_id = root_id;
188 ref->key_for_search = *key;
190 memset(&ref->key_for_search, 0, sizeof(ref->key_for_search));
192 ref->inode_list = NULL;
195 ref->parent = parent;
196 ref->wanted_disk_byte = wanted_disk_byte;
197 list_add_tail(&ref->list, head);
202 static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path,
203 struct ulist *parents, struct __prelim_ref *ref,
204 int level, u64 time_seq, const u64 *extent_item_pos,
209 struct extent_buffer *eb;
210 struct btrfs_key key;
211 struct btrfs_key *key_for_search = &ref->key_for_search;
212 struct btrfs_file_extent_item *fi;
213 struct extent_inode_elem *eie = NULL, *old = NULL;
215 u64 wanted_disk_byte = ref->wanted_disk_byte;
219 eb = path->nodes[level];
220 ret = ulist_add(parents, eb->start, 0, GFP_NOFS);
227 * We normally enter this function with the path already pointing to
228 * the first item to check. But sometimes, we may enter it with
229 * slot==nritems. In that case, go to the next leaf before we continue.
231 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0]))
232 ret = btrfs_next_leaf(root, path);
234 while (!ret && count < total_refs) {
236 slot = path->slots[0];
238 btrfs_item_key_to_cpu(eb, &key, slot);
240 if (key.objectid != key_for_search->objectid ||
241 key.type != BTRFS_EXTENT_DATA_KEY)
244 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
245 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
247 if (disk_byte == wanted_disk_byte) {
251 if (extent_item_pos) {
252 ret = check_extent_in_eb(&key, eb, fi,
260 ret = ulist_add_merge_ptr(parents, eb->start,
261 eie, (void **)&old, GFP_NOFS);
264 if (!ret && extent_item_pos) {
272 ret = btrfs_next_item(root, path);
278 free_inode_elem_list(eie);
283 * resolve an indirect backref in the form (root_id, key, level)
284 * to a logical address
286 static int __resolve_indirect_ref(struct btrfs_fs_info *fs_info,
287 struct btrfs_path *path, u64 time_seq,
288 struct __prelim_ref *ref,
289 struct ulist *parents,
290 const u64 *extent_item_pos, u64 total_refs)
292 struct btrfs_root *root;
293 struct btrfs_key root_key;
294 struct extent_buffer *eb;
297 int level = ref->level;
299 root_key.objectid = ref->root_id;
300 root_key.type = BTRFS_ROOT_ITEM_KEY;
301 root_key.offset = (u64)-1;
303 root = btrfs_read_fs_root(fs_info, &root_key);
309 root_level = btrfs_root_level(&root->root_item);
311 if (root_level + 1 == level)
314 path->lowest_level = level;
315 ret = btrfs_search_slot(NULL, root, &ref->key_for_search, path, 0, 0);
317 pr_debug("search slot in root %llu (level %d, ref count %d) returned "
318 "%d for key (%llu %u %llu)\n",
319 ref->root_id, level, ref->count, ret,
320 ref->key_for_search.objectid, ref->key_for_search.type,
321 ref->key_for_search.offset);
325 eb = path->nodes[level];
333 eb = path->nodes[level];
336 ret = add_all_parents(root, path, parents, ref, level, time_seq,
337 extent_item_pos, total_refs);
339 path->lowest_level = 0;
340 btrfs_release_path(path);
345 * resolve all indirect backrefs from the list
347 static int __resolve_indirect_refs(struct btrfs_fs_info *fs_info,
348 struct btrfs_path *path, u64 time_seq,
349 struct list_head *head,
350 const u64 *extent_item_pos, u64 total_refs)
354 struct __prelim_ref *ref;
355 struct __prelim_ref *ref_safe;
356 struct __prelim_ref *new_ref;
357 struct ulist *parents;
358 struct ulist_node *node;
359 struct ulist_iterator uiter;
361 parents = ulist_alloc(GFP_NOFS);
366 * _safe allows us to insert directly after the current item without
367 * iterating over the newly inserted items.
368 * we're also allowed to re-assign ref during iteration.
370 list_for_each_entry_safe(ref, ref_safe, head, list) {
371 if (ref->parent) /* already direct */
375 err = __resolve_indirect_ref(fs_info, path, time_seq, ref,
376 parents, extent_item_pos,
379 * we can only tolerate ENOENT,otherwise,we should catch error
380 * and return directly.
382 if (err == -ENOENT) {
389 /* we put the first parent into the ref at hand */
390 ULIST_ITER_INIT(&uiter);
391 node = ulist_next(parents, &uiter);
392 ref->parent = node ? node->val : 0;
393 ref->inode_list = node ?
394 (struct extent_inode_elem *)(uintptr_t)node->aux : NULL;
396 /* additional parents require new refs being added here */
397 while ((node = ulist_next(parents, &uiter))) {
398 new_ref = kmalloc(sizeof(*new_ref), GFP_NOFS);
403 memcpy(new_ref, ref, sizeof(*ref));
404 new_ref->parent = node->val;
405 new_ref->inode_list = (struct extent_inode_elem *)
406 (uintptr_t)node->aux;
407 list_add(&new_ref->list, &ref->list);
409 ulist_reinit(parents);
416 static inline int ref_for_same_block(struct __prelim_ref *ref1,
417 struct __prelim_ref *ref2)
419 if (ref1->level != ref2->level)
421 if (ref1->root_id != ref2->root_id)
423 if (ref1->key_for_search.type != ref2->key_for_search.type)
425 if (ref1->key_for_search.objectid != ref2->key_for_search.objectid)
427 if (ref1->key_for_search.offset != ref2->key_for_search.offset)
429 if (ref1->parent != ref2->parent)
436 * read tree blocks and add keys where required.
438 static int __add_missing_keys(struct btrfs_fs_info *fs_info,
439 struct list_head *head)
441 struct list_head *pos;
442 struct extent_buffer *eb;
444 list_for_each(pos, head) {
445 struct __prelim_ref *ref;
446 ref = list_entry(pos, struct __prelim_ref, list);
450 if (ref->key_for_search.type)
452 BUG_ON(!ref->wanted_disk_byte);
453 eb = read_tree_block(fs_info->tree_root, ref->wanted_disk_byte,
454 fs_info->tree_root->nodesize, 0);
455 if (!extent_buffer_uptodate(eb)) {
456 free_extent_buffer(eb);
459 if (btrfs_header_level(eb) == 0)
460 btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
462 btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
463 free_extent_buffer(eb);
469 * merge two lists of backrefs and adjust counts accordingly
471 * mode = 1: merge identical keys, if key is set
472 * FIXME: if we add more keys in __add_prelim_ref, we can merge more here.
473 * additionally, we could even add a key range for the blocks we
474 * looked into to merge even more (-> replace unresolved refs by those
476 * mode = 2: merge identical parents
478 static void __merge_refs(struct list_head *head, int mode)
480 struct list_head *pos1;
482 list_for_each(pos1, head) {
483 struct list_head *n2;
484 struct list_head *pos2;
485 struct __prelim_ref *ref1;
487 ref1 = list_entry(pos1, struct __prelim_ref, list);
489 for (pos2 = pos1->next, n2 = pos2->next; pos2 != head;
490 pos2 = n2, n2 = pos2->next) {
491 struct __prelim_ref *ref2;
492 struct __prelim_ref *xchg;
493 struct extent_inode_elem *eie;
495 ref2 = list_entry(pos2, struct __prelim_ref, list);
498 if (!ref_for_same_block(ref1, ref2))
500 if (!ref1->parent && ref2->parent) {
506 if (ref1->parent != ref2->parent)
510 eie = ref1->inode_list;
511 while (eie && eie->next)
514 eie->next = ref2->inode_list;
516 ref1->inode_list = ref2->inode_list;
517 ref1->count += ref2->count;
519 list_del(&ref2->list);
527 * add all inline backrefs for bytenr to the list
529 static int __add_inline_refs(struct btrfs_fs_info *fs_info,
530 struct btrfs_path *path, u64 bytenr,
531 int *info_level, struct list_head *prefs,
536 struct extent_buffer *leaf;
537 struct btrfs_key key;
538 struct btrfs_key found_key;
541 struct btrfs_extent_item *ei;
546 * enumerate all inline refs
548 leaf = path->nodes[0];
549 slot = path->slots[0];
551 item_size = btrfs_item_size_nr(leaf, slot);
552 BUG_ON(item_size < sizeof(*ei));
554 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
555 flags = btrfs_extent_flags(leaf, ei);
556 *total_refs += btrfs_extent_refs(leaf, ei);
557 btrfs_item_key_to_cpu(leaf, &found_key, slot);
559 ptr = (unsigned long)(ei + 1);
560 end = (unsigned long)ei + item_size;
562 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
563 flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
564 struct btrfs_tree_block_info *info;
566 info = (struct btrfs_tree_block_info *)ptr;
567 *info_level = btrfs_tree_block_level(leaf, info);
568 ptr += sizeof(struct btrfs_tree_block_info);
570 } else if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
571 *info_level = found_key.offset;
573 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
577 struct btrfs_extent_inline_ref *iref;
581 iref = (struct btrfs_extent_inline_ref *)ptr;
582 type = btrfs_extent_inline_ref_type(leaf, iref);
583 offset = btrfs_extent_inline_ref_offset(leaf, iref);
586 case BTRFS_SHARED_BLOCK_REF_KEY:
587 ret = __add_prelim_ref(prefs, 0, NULL,
588 *info_level + 1, offset,
589 bytenr, 1, GFP_NOFS);
591 case BTRFS_SHARED_DATA_REF_KEY: {
592 struct btrfs_shared_data_ref *sdref;
595 sdref = (struct btrfs_shared_data_ref *)(iref + 1);
596 count = btrfs_shared_data_ref_count(leaf, sdref);
597 ret = __add_prelim_ref(prefs, 0, NULL, 0, offset,
598 bytenr, count, GFP_NOFS);
601 case BTRFS_TREE_BLOCK_REF_KEY:
602 ret = __add_prelim_ref(prefs, offset, NULL,
604 bytenr, 1, GFP_NOFS);
606 case BTRFS_EXTENT_DATA_REF_KEY: {
607 struct btrfs_extent_data_ref *dref;
611 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
612 count = btrfs_extent_data_ref_count(leaf, dref);
613 key.objectid = btrfs_extent_data_ref_objectid(leaf,
615 key.type = BTRFS_EXTENT_DATA_KEY;
616 key.offset = btrfs_extent_data_ref_offset(leaf, dref);
617 root = btrfs_extent_data_ref_root(leaf, dref);
618 ret = __add_prelim_ref(prefs, root, &key, 0, 0,
619 bytenr, count, GFP_NOFS);
627 ptr += btrfs_extent_inline_ref_size(type);
634 * add all non-inline backrefs for bytenr to the list
636 static int __add_keyed_refs(struct btrfs_fs_info *fs_info,
637 struct btrfs_path *path, u64 bytenr,
638 int info_level, struct list_head *prefs)
640 struct btrfs_root *extent_root = fs_info->extent_root;
643 struct extent_buffer *leaf;
644 struct btrfs_key key;
647 ret = btrfs_next_item(extent_root, path);
655 slot = path->slots[0];
656 leaf = path->nodes[0];
657 btrfs_item_key_to_cpu(leaf, &key, slot);
659 if (key.objectid != bytenr)
661 if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
663 if (key.type > BTRFS_SHARED_DATA_REF_KEY)
667 case BTRFS_SHARED_BLOCK_REF_KEY:
668 ret = __add_prelim_ref(prefs, 0, NULL,
669 info_level + 1, key.offset,
670 bytenr, 1, GFP_NOFS);
672 case BTRFS_SHARED_DATA_REF_KEY: {
673 struct btrfs_shared_data_ref *sdref;
676 sdref = btrfs_item_ptr(leaf, slot,
677 struct btrfs_shared_data_ref);
678 count = btrfs_shared_data_ref_count(leaf, sdref);
679 ret = __add_prelim_ref(prefs, 0, NULL, 0, key.offset,
680 bytenr, count, GFP_NOFS);
683 case BTRFS_TREE_BLOCK_REF_KEY:
684 ret = __add_prelim_ref(prefs, key.offset, NULL,
686 bytenr, 1, GFP_NOFS);
688 case BTRFS_EXTENT_DATA_REF_KEY: {
689 struct btrfs_extent_data_ref *dref;
693 dref = btrfs_item_ptr(leaf, slot,
694 struct btrfs_extent_data_ref);
695 count = btrfs_extent_data_ref_count(leaf, dref);
696 key.objectid = btrfs_extent_data_ref_objectid(leaf,
698 key.type = BTRFS_EXTENT_DATA_KEY;
699 key.offset = btrfs_extent_data_ref_offset(leaf, dref);
700 root = btrfs_extent_data_ref_root(leaf, dref);
701 ret = __add_prelim_ref(prefs, root, &key, 0, 0,
702 bytenr, count, GFP_NOFS);
717 * this adds all existing backrefs (inline backrefs, backrefs and delayed
718 * refs) for the given bytenr to the refs list, merges duplicates and resolves
719 * indirect refs to their parent bytenr.
720 * When roots are found, they're added to the roots list
722 * FIXME some caching might speed things up
724 static int find_parent_nodes(struct btrfs_trans_handle *trans,
725 struct btrfs_fs_info *fs_info, u64 bytenr,
726 u64 time_seq, struct ulist *refs,
727 struct ulist *roots, const u64 *extent_item_pos)
729 struct btrfs_key key;
730 struct btrfs_path *path;
733 struct list_head prefs;
734 struct __prelim_ref *ref;
735 struct extent_inode_elem *eie = NULL;
738 INIT_LIST_HEAD(&prefs);
740 key.objectid = bytenr;
741 key.offset = (u64)-1;
742 if (btrfs_fs_incompat(fs_info,
743 BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA))
744 key.type = BTRFS_METADATA_ITEM_KEY;
746 key.type = BTRFS_EXTENT_ITEM_KEY;
748 path = btrfs_alloc_path();
752 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
757 if (path->slots[0]) {
758 struct extent_buffer *leaf;
762 leaf = path->nodes[0];
763 slot = path->slots[0];
764 btrfs_item_key_to_cpu(leaf, &key, slot);
765 if (key.objectid == bytenr &&
766 (key.type == BTRFS_EXTENT_ITEM_KEY ||
767 key.type == BTRFS_METADATA_ITEM_KEY)) {
768 ret = __add_inline_refs(fs_info, path, bytenr,
773 ret = __add_keyed_refs(fs_info, path, bytenr,
779 btrfs_release_path(path);
781 ret = __add_missing_keys(fs_info, &prefs);
785 __merge_refs(&prefs, 1);
787 ret = __resolve_indirect_refs(fs_info, path, time_seq, &prefs,
788 extent_item_pos, total_refs);
792 __merge_refs(&prefs, 2);
794 while (!list_empty(&prefs)) {
795 ref = list_first_entry(&prefs, struct __prelim_ref, list);
796 WARN_ON(ref->count < 0);
797 if (roots && ref->count && ref->root_id && ref->parent == 0) {
798 /* no parent == root of tree */
799 ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS);
803 if (ref->count && ref->parent) {
804 if (extent_item_pos && !ref->inode_list &&
807 struct extent_buffer *eb;
808 bsz = fs_info->extent_root->nodesize;
809 eb = read_tree_block(fs_info->extent_root,
810 ref->parent, bsz, 0);
811 if (!extent_buffer_uptodate(eb)) {
812 free_extent_buffer(eb);
816 ret = find_extent_in_eb(eb, bytenr,
817 *extent_item_pos, &eie);
818 free_extent_buffer(eb);
821 ref->inode_list = eie;
823 ret = ulist_add_merge_ptr(refs, ref->parent,
825 (void **)&eie, GFP_NOFS);
828 if (!ret && extent_item_pos) {
830 * we've recorded that parent, so we must extend
831 * its inode list here
836 eie->next = ref->inode_list;
840 list_del(&ref->list);
845 btrfs_free_path(path);
846 while (!list_empty(&prefs)) {
847 ref = list_first_entry(&prefs, struct __prelim_ref, list);
848 list_del(&ref->list);
852 free_inode_elem_list(eie);
856 static void free_leaf_list(struct ulist *blocks)
858 struct ulist_node *node = NULL;
859 struct extent_inode_elem *eie;
860 struct ulist_iterator uiter;
862 ULIST_ITER_INIT(&uiter);
863 while ((node = ulist_next(blocks, &uiter))) {
866 eie = (struct extent_inode_elem *)(uintptr_t)node->aux;
867 free_inode_elem_list(eie);
875 * Finds all leafs with a reference to the specified combination of bytenr and
876 * offset. key_list_head will point to a list of corresponding keys (caller must
877 * free each list element). The leafs will be stored in the leafs ulist, which
878 * must be freed with ulist_free.
880 * returns 0 on success, <0 on error
882 static int btrfs_find_all_leafs(struct btrfs_trans_handle *trans,
883 struct btrfs_fs_info *fs_info, u64 bytenr,
884 u64 time_seq, struct ulist **leafs,
885 const u64 *extent_item_pos)
889 *leafs = ulist_alloc(GFP_NOFS);
893 ret = find_parent_nodes(trans, fs_info, bytenr,
894 time_seq, *leafs, NULL, extent_item_pos);
895 if (ret < 0 && ret != -ENOENT) {
896 free_leaf_list(*leafs);
904 * walk all backrefs for a given extent to find all roots that reference this
905 * extent. Walking a backref means finding all extents that reference this
906 * extent and in turn walk the backrefs of those, too. Naturally this is a
907 * recursive process, but here it is implemented in an iterative fashion: We
908 * find all referencing extents for the extent in question and put them on a
909 * list. In turn, we find all referencing extents for those, further appending
910 * to the list. The way we iterate the list allows adding more elements after
911 * the current while iterating. The process stops when we reach the end of the
912 * list. Found roots are added to the roots list.
914 * returns 0 on success, < 0 on error.
916 static int __btrfs_find_all_roots(struct btrfs_trans_handle *trans,
917 struct btrfs_fs_info *fs_info, u64 bytenr,
918 u64 time_seq, struct ulist **roots)
921 struct ulist_node *node = NULL;
922 struct ulist_iterator uiter;
925 tmp = ulist_alloc(GFP_NOFS);
928 *roots = ulist_alloc(GFP_NOFS);
934 ULIST_ITER_INIT(&uiter);
936 ret = find_parent_nodes(trans, fs_info, bytenr,
937 time_seq, tmp, *roots, NULL);
938 if (ret < 0 && ret != -ENOENT) {
943 node = ulist_next(tmp, &uiter);
954 int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
955 struct btrfs_fs_info *fs_info, u64 bytenr,
956 u64 time_seq, struct ulist **roots)
958 return __btrfs_find_all_roots(trans, fs_info, bytenr, time_seq, roots);
962 * this makes the path point to (inum INODE_ITEM ioff)
964 int inode_item_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
965 struct btrfs_path *path)
967 struct btrfs_key key;
968 return btrfs_find_item(fs_root, path, inum, ioff,
969 BTRFS_INODE_ITEM_KEY, &key);
972 static int inode_ref_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
973 struct btrfs_path *path,
974 struct btrfs_key *found_key)
976 return btrfs_find_item(fs_root, path, inum, ioff,
977 BTRFS_INODE_REF_KEY, found_key);
980 int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,
981 u64 start_off, struct btrfs_path *path,
982 struct btrfs_inode_extref **ret_extref,
986 struct btrfs_key key;
987 struct btrfs_key found_key;
988 struct btrfs_inode_extref *extref;
989 struct extent_buffer *leaf;
992 key.objectid = inode_objectid;
993 btrfs_set_key_type(&key, BTRFS_INODE_EXTREF_KEY);
994 key.offset = start_off;
996 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1001 leaf = path->nodes[0];
1002 slot = path->slots[0];
1003 if (slot >= btrfs_header_nritems(leaf)) {
1005 * If the item at offset is not found,
1006 * btrfs_search_slot will point us to the slot
1007 * where it should be inserted. In our case
1008 * that will be the slot directly before the
1009 * next INODE_REF_KEY_V2 item. In the case
1010 * that we're pointing to the last slot in a
1011 * leaf, we must move one leaf over.
1013 ret = btrfs_next_leaf(root, path);
1022 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1025 * Check that we're still looking at an extended ref key for
1026 * this particular objectid. If we have different
1027 * objectid or type then there are no more to be found
1028 * in the tree and we can exit.
1031 if (found_key.objectid != inode_objectid)
1033 if (btrfs_key_type(&found_key) != BTRFS_INODE_EXTREF_KEY)
1037 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1038 extref = (struct btrfs_inode_extref *)ptr;
1039 *ret_extref = extref;
1041 *found_off = found_key.offset;
1049 * this iterates to turn a name (from iref/extref) into a full filesystem path.
1050 * Elements of the path are separated by '/' and the path is guaranteed to be
1051 * 0-terminated. the path is only given within the current file system.
1052 * Therefore, it never starts with a '/'. the caller is responsible to provide
1053 * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
1054 * the start point of the resulting string is returned. this pointer is within
1056 * in case the path buffer would overflow, the pointer is decremented further
1057 * as if output was written to the buffer, though no more output is actually
1058 * generated. that way, the caller can determine how much space would be
1059 * required for the path to fit into the buffer. in that case, the returned
1060 * value will be smaller than dest. callers must check this!
1062 char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
1063 u32 name_len, unsigned long name_off,
1064 struct extent_buffer *eb_in, u64 parent,
1065 char *dest, u32 size)
1070 s64 bytes_left = ((s64)size) - 1;
1071 struct extent_buffer *eb = eb_in;
1072 struct btrfs_key found_key;
1073 struct btrfs_inode_ref *iref;
1075 if (bytes_left >= 0)
1076 dest[bytes_left] = '\0';
1079 bytes_left -= name_len;
1080 if (bytes_left >= 0)
1081 read_extent_buffer(eb, dest + bytes_left,
1082 name_off, name_len);
1084 free_extent_buffer(eb);
1085 ret = inode_ref_info(parent, 0, fs_root, path, &found_key);
1091 next_inum = found_key.offset;
1093 /* regular exit ahead */
1094 if (parent == next_inum)
1097 slot = path->slots[0];
1098 eb = path->nodes[0];
1099 /* make sure we can use eb after releasing the path */
1102 btrfs_release_path(path);
1103 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1105 name_len = btrfs_inode_ref_name_len(eb, iref);
1106 name_off = (unsigned long)(iref + 1);
1110 if (bytes_left >= 0)
1111 dest[bytes_left] = '/';
1114 btrfs_release_path(path);
1117 return ERR_PTR(ret);
1119 return dest + bytes_left;
1123 * this makes the path point to (logical EXTENT_ITEM *)
1124 * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
1125 * tree blocks and <0 on error.
1127 int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
1128 struct btrfs_path *path, struct btrfs_key *found_key,
1135 struct extent_buffer *eb;
1136 struct btrfs_extent_item *ei;
1137 struct btrfs_key key;
1139 if (btrfs_fs_incompat(fs_info,
1140 BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA))
1141 key.type = BTRFS_METADATA_ITEM_KEY;
1143 key.type = BTRFS_EXTENT_ITEM_KEY;
1144 key.objectid = logical;
1145 key.offset = (u64)-1;
1147 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
1151 ret = btrfs_previous_extent_item(fs_info->extent_root, path, 0);
1157 btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]);
1158 if (found_key->type == BTRFS_METADATA_ITEM_KEY)
1159 size = fs_info->extent_root->nodesize;
1160 else if (found_key->type == BTRFS_EXTENT_ITEM_KEY)
1161 size = found_key->offset;
1163 if (found_key->objectid > logical ||
1164 found_key->objectid + size <= logical) {
1165 pr_debug("logical %llu is not within any extent\n", logical);
1169 eb = path->nodes[0];
1170 item_size = btrfs_item_size_nr(eb, path->slots[0]);
1171 BUG_ON(item_size < sizeof(*ei));
1173 ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
1174 flags = btrfs_extent_flags(eb, ei);
1176 pr_debug("logical %llu is at position %llu within the extent (%llu "
1177 "EXTENT_ITEM %llu) flags %#llx size %u\n",
1178 logical, logical - found_key->objectid, found_key->objectid,
1179 found_key->offset, flags, item_size);
1182 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1183 *flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK;
1184 else if (flags & BTRFS_EXTENT_FLAG_DATA)
1185 *flags_ret = BTRFS_EXTENT_FLAG_DATA;
1196 * helper function to iterate extent inline refs. ptr must point to a 0 value
1197 * for the first call and may be modified. it is used to track state.
1198 * if more refs exist, 0 is returned and the next call to
1199 * __get_extent_inline_ref must pass the modified ptr parameter to get the
1200 * next ref. after the last ref was processed, 1 is returned.
1201 * returns <0 on error
1203 static int __get_extent_inline_ref(unsigned long *ptr, struct extent_buffer *eb,
1204 struct btrfs_key *key,
1205 struct btrfs_extent_item *ei, u32 item_size,
1206 struct btrfs_extent_inline_ref **out_eiref,
1211 struct btrfs_tree_block_info *info;
1215 flags = btrfs_extent_flags(eb, ei);
1216 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1217 if (key->type == BTRFS_METADATA_ITEM_KEY) {
1218 /* a skinny metadata extent */
1220 (struct btrfs_extent_inline_ref *)(ei + 1);
1222 WARN_ON(key->type != BTRFS_EXTENT_ITEM_KEY);
1223 info = (struct btrfs_tree_block_info *)(ei + 1);
1225 (struct btrfs_extent_inline_ref *)(info + 1);
1228 *out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1);
1230 *ptr = (unsigned long)*out_eiref;
1231 if ((unsigned long)(*ptr) >= (unsigned long)ei + item_size)
1235 end = (unsigned long)ei + item_size;
1236 *out_eiref = (struct btrfs_extent_inline_ref *)(*ptr);
1237 *out_type = btrfs_extent_inline_ref_type(eb, *out_eiref);
1239 *ptr += btrfs_extent_inline_ref_size(*out_type);
1240 WARN_ON(*ptr > end);
1242 return 1; /* last */
1248 * reads the tree block backref for an extent. tree level and root are returned
1249 * through out_level and out_root. ptr must point to a 0 value for the first
1250 * call and may be modified (see __get_extent_inline_ref comment).
1251 * returns 0 if data was provided, 1 if there was no more data to provide or
1254 int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
1255 struct btrfs_key *key, struct btrfs_extent_item *ei,
1256 u32 item_size, u64 *out_root, u8 *out_level)
1260 struct btrfs_tree_block_info *info;
1261 struct btrfs_extent_inline_ref *eiref;
1263 if (*ptr == (unsigned long)-1)
1267 ret = __get_extent_inline_ref(ptr, eb, key, ei, item_size,
1272 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1273 type == BTRFS_SHARED_BLOCK_REF_KEY)
1280 /* we can treat both ref types equally here */
1281 info = (struct btrfs_tree_block_info *)(ei + 1);
1282 *out_root = btrfs_extent_inline_ref_offset(eb, eiref);
1283 *out_level = btrfs_tree_block_level(eb, info);
1286 *ptr = (unsigned long)-1;
1291 static int iterate_leaf_refs(struct extent_inode_elem *inode_list,
1292 u64 root, u64 extent_item_objectid,
1293 iterate_extent_inodes_t *iterate, void *ctx)
1295 struct extent_inode_elem *eie;
1298 for (eie = inode_list; eie; eie = eie->next) {
1299 pr_debug("ref for %llu resolved, key (%llu EXTEND_DATA %llu), "
1300 "root %llu\n", extent_item_objectid,
1301 eie->inum, eie->offset, root);
1302 ret = iterate(eie->inum, eie->offset, root, ctx);
1304 pr_debug("stopping iteration for %llu due to ret=%d\n",
1305 extent_item_objectid, ret);
1314 * calls iterate() for every inode that references the extent identified by
1315 * the given parameters.
1316 * when the iterator function returns a non-zero value, iteration stops.
1318 int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
1319 u64 extent_item_objectid, u64 extent_item_pos,
1320 int search_commit_root,
1321 iterate_extent_inodes_t *iterate, void *ctx)
1324 struct btrfs_trans_handle *trans = NULL;
1325 struct ulist *refs = NULL;
1326 struct ulist *roots = NULL;
1327 struct ulist_node *ref_node = NULL;
1328 struct ulist_node *root_node = NULL;
1329 struct ulist_iterator ref_uiter;
1330 struct ulist_iterator root_uiter;
1332 pr_debug("resolving all inodes for extent %llu\n",
1333 extent_item_objectid);
1335 ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid,
1336 0, &refs, &extent_item_pos);
1340 ULIST_ITER_INIT(&ref_uiter);
1341 while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) {
1342 ret = __btrfs_find_all_roots(trans, fs_info, ref_node->val,
1346 ULIST_ITER_INIT(&root_uiter);
1347 while (!ret && (root_node = ulist_next(roots, &root_uiter))) {
1348 pr_debug("root %llu references leaf %llu, data list "
1349 "%#llx\n", root_node->val, ref_node->val,
1351 ret = iterate_leaf_refs((struct extent_inode_elem *)
1352 (uintptr_t)ref_node->aux,
1354 extent_item_objectid,
1360 free_leaf_list(refs);
1365 int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
1366 struct btrfs_path *path,
1367 iterate_extent_inodes_t *iterate, void *ctx)
1370 u64 extent_item_pos;
1372 struct btrfs_key found_key;
1373 int search_commit_root = 0;
1375 ret = extent_from_logical(fs_info, logical, path, &found_key, &flags);
1376 btrfs_release_path(path);
1379 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1382 extent_item_pos = logical - found_key.objectid;
1383 ret = iterate_extent_inodes(fs_info, found_key.objectid,
1384 extent_item_pos, search_commit_root,
1390 typedef int (iterate_irefs_t)(u64 parent, u32 name_len, unsigned long name_off,
1391 struct extent_buffer *eb, void *ctx);
1393 static int iterate_inode_refs(u64 inum, struct btrfs_root *fs_root,
1394 struct btrfs_path *path,
1395 iterate_irefs_t *iterate, void *ctx)
1404 struct extent_buffer *eb;
1405 struct btrfs_item *item;
1406 struct btrfs_inode_ref *iref;
1407 struct btrfs_key found_key;
1410 ret = inode_ref_info(inum, parent ? parent+1 : 0, fs_root, path,
1415 ret = found ? 0 : -ENOENT;
1420 parent = found_key.offset;
1421 slot = path->slots[0];
1422 eb = btrfs_clone_extent_buffer(path->nodes[0]);
1427 extent_buffer_get(eb);
1428 btrfs_release_path(path);
1430 item = btrfs_item_nr(slot);
1431 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1433 for (cur = 0; cur < btrfs_item_size(eb, item); cur += len) {
1434 name_len = btrfs_inode_ref_name_len(eb, iref);
1435 /* path must be released before calling iterate()! */
1436 pr_debug("following ref at offset %u for inode %llu in "
1437 "tree %llu\n", cur, found_key.objectid,
1439 ret = iterate(parent, name_len,
1440 (unsigned long)(iref + 1), eb, ctx);
1443 len = sizeof(*iref) + name_len;
1444 iref = (struct btrfs_inode_ref *)((char *)iref + len);
1446 free_extent_buffer(eb);
1449 btrfs_release_path(path);
1454 static int iterate_inode_extrefs(u64 inum, struct btrfs_root *fs_root,
1455 struct btrfs_path *path,
1456 iterate_irefs_t *iterate, void *ctx)
1463 struct extent_buffer *eb;
1464 struct btrfs_inode_extref *extref;
1465 struct extent_buffer *leaf;
1471 ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref,
1476 ret = found ? 0 : -ENOENT;
1481 slot = path->slots[0];
1482 eb = btrfs_clone_extent_buffer(path->nodes[0]);
1487 extent_buffer_get(eb);
1489 btrfs_release_path(path);
1491 leaf = path->nodes[0];
1492 item_size = btrfs_item_size_nr(leaf, slot);
1493 ptr = btrfs_item_ptr_offset(leaf, slot);
1496 while (cur_offset < item_size) {
1499 extref = (struct btrfs_inode_extref *)(ptr + cur_offset);
1500 parent = btrfs_inode_extref_parent(eb, extref);
1501 name_len = btrfs_inode_extref_name_len(eb, extref);
1502 ret = iterate(parent, name_len,
1503 (unsigned long)&extref->name, eb, ctx);
1507 cur_offset += btrfs_inode_extref_name_len(leaf, extref);
1508 cur_offset += sizeof(*extref);
1510 free_extent_buffer(eb);
1515 btrfs_release_path(path);
1520 static int iterate_irefs(u64 inum, struct btrfs_root *fs_root,
1521 struct btrfs_path *path, iterate_irefs_t *iterate,
1527 ret = iterate_inode_refs(inum, fs_root, path, iterate, ctx);
1530 else if (ret != -ENOENT)
1533 ret = iterate_inode_extrefs(inum, fs_root, path, iterate, ctx);
1534 if (ret == -ENOENT && found_refs)
1541 * returns 0 if the path could be dumped (probably truncated)
1542 * returns <0 in case of an error
1544 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
1545 struct extent_buffer *eb, void *ctx)
1547 struct inode_fs_paths *ipath = ctx;
1550 int i = ipath->fspath->elem_cnt;
1551 const int s_ptr = sizeof(char *);
1554 bytes_left = ipath->fspath->bytes_left > s_ptr ?
1555 ipath->fspath->bytes_left - s_ptr : 0;
1557 fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
1558 fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len,
1559 name_off, eb, inum, fspath_min, bytes_left);
1561 return PTR_ERR(fspath);
1563 if (fspath > fspath_min) {
1564 ipath->fspath->val[i] = (u64)(unsigned long)fspath;
1565 ++ipath->fspath->elem_cnt;
1566 ipath->fspath->bytes_left = fspath - fspath_min;
1568 ++ipath->fspath->elem_missed;
1569 ipath->fspath->bytes_missing += fspath_min - fspath;
1570 ipath->fspath->bytes_left = 0;
1577 * this dumps all file system paths to the inode into the ipath struct, provided
1578 * is has been created large enough. each path is zero-terminated and accessed
1579 * from ipath->fspath->val[i].
1580 * when it returns, there are ipath->fspath->elem_cnt number of paths available
1581 * in ipath->fspath->val[]. When the allocated space wasn't sufficient, the
1582 * number of missed paths is recorded in ipath->fspath->elem_missed, otherwise,
1583 * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
1584 * have been needed to return all paths.
1586 int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
1588 return iterate_irefs(inum, ipath->fs_root, ipath->btrfs_path,
1589 inode_to_path, ipath);
1592 struct btrfs_data_container *init_data_container(u32 total_bytes)
1594 struct btrfs_data_container *data;
1597 alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
1598 data = vmalloc(alloc_bytes);
1600 return ERR_PTR(-ENOMEM);
1602 if (total_bytes >= sizeof(*data)) {
1603 data->bytes_left = total_bytes - sizeof(*data);
1604 data->bytes_missing = 0;
1606 data->bytes_missing = sizeof(*data) - total_bytes;
1607 data->bytes_left = 0;
1611 data->elem_missed = 0;
1617 * allocates space to return multiple file system paths for an inode.
1618 * total_bytes to allocate are passed, note that space usable for actual path
1619 * information will be total_bytes - sizeof(struct inode_fs_paths).
1620 * the returned pointer must be freed with free_ipath() in the end.
1622 struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
1623 struct btrfs_path *path)
1625 struct inode_fs_paths *ifp;
1626 struct btrfs_data_container *fspath;
1628 fspath = init_data_container(total_bytes);
1630 return (void *)fspath;
1632 ifp = kmalloc(sizeof(*ifp), GFP_NOFS);
1635 return ERR_PTR(-ENOMEM);
1638 ifp->btrfs_path = path;
1639 ifp->fspath = fspath;
1640 ifp->fs_root = fs_root;
1645 void free_ipath(struct inode_fs_paths *ipath)
1649 vfree(ipath->fspath);