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"
23 #include "kernel-shared/ulist.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;
133 static struct __prelim_ref *list_first_pref(struct list_head *head)
135 return list_first_entry(head, struct __prelim_ref, list);
139 struct list_head pending;
142 static void init_pref_state(struct pref_state *prefstate)
144 INIT_LIST_HEAD(&prefstate->pending);
148 * the rules for all callers of this function are:
149 * - obtaining the parent is the goal
150 * - if you add a key, you must know that it is a correct key
151 * - if you cannot add the parent or a correct key, then we will look into the
152 * block later to set a correct key
156 * backref type | shared | indirect | shared | indirect
157 * information | tree | tree | data | data
158 * --------------------+--------+----------+--------+----------
159 * parent logical | y | - | - | -
160 * key to resolve | - | y | y | y
161 * tree block logical | - | - | - | -
162 * root for resolving | y | y | y | y
164 * - column 1: we've the parent -> done
165 * - column 2, 3, 4: we use the key to find the parent
167 * on disk refs (inline or keyed)
168 * ==============================
169 * backref type | shared | indirect | shared | indirect
170 * information | tree | tree | data | data
171 * --------------------+--------+----------+--------+----------
172 * parent logical | y | - | y | -
173 * key to resolve | - | - | - | y
174 * tree block logical | y | y | y | y
175 * root for resolving | - | y | y | y
177 * - column 1, 3: we've the parent -> done
178 * - column 2: we take the first key from the block to find the parent
179 * (see __add_missing_keys)
180 * - column 4: we use the key to find the parent
182 * additional information that's available but not required to find the parent
183 * block might help in merging entries to gain some speed.
186 static int __add_prelim_ref(struct pref_state *prefstate, u64 root_id,
187 struct btrfs_key *key, int level,
188 u64 parent, u64 wanted_disk_byte, int count,
191 struct list_head *head = &prefstate->pending;
192 struct __prelim_ref *ref;
194 if (root_id == BTRFS_DATA_RELOC_TREE_OBJECTID)
197 ref = kmalloc(sizeof(*ref), gfp_mask);
201 ref->root_id = root_id;
203 ref->key_for_search = *key;
205 memset(&ref->key_for_search, 0, sizeof(ref->key_for_search));
207 ref->inode_list = NULL;
210 ref->parent = parent;
211 ref->wanted_disk_byte = wanted_disk_byte;
212 list_add_tail(&ref->list, head);
217 static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path,
218 struct ulist *parents, struct __prelim_ref *ref,
219 int level, u64 time_seq, const u64 *extent_item_pos,
224 struct extent_buffer *eb;
225 struct btrfs_key key;
226 struct btrfs_key *key_for_search = &ref->key_for_search;
227 struct btrfs_file_extent_item *fi;
228 struct extent_inode_elem *eie = NULL, *old = NULL;
230 u64 wanted_disk_byte = ref->wanted_disk_byte;
234 eb = path->nodes[level];
235 ret = ulist_add(parents, eb->start, 0, GFP_NOFS);
242 * We normally enter this function with the path already pointing to
243 * the first item to check. But sometimes, we may enter it with
244 * slot==nritems. In that case, go to the next leaf before we continue.
246 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0]))
247 ret = btrfs_next_leaf(root, path);
249 while (!ret && count < total_refs) {
251 slot = path->slots[0];
253 btrfs_item_key_to_cpu(eb, &key, slot);
255 if (key.objectid != key_for_search->objectid ||
256 key.type != BTRFS_EXTENT_DATA_KEY)
259 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
260 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
262 if (disk_byte == wanted_disk_byte) {
266 if (extent_item_pos) {
267 ret = check_extent_in_eb(&key, eb, fi,
275 ret = ulist_add_merge_ptr(parents, eb->start,
276 eie, (void **)&old, GFP_NOFS);
279 if (!ret && extent_item_pos) {
287 ret = btrfs_next_item(root, path);
293 free_inode_elem_list(eie);
298 * resolve an indirect backref in the form (root_id, key, level)
299 * to a logical address
301 static int __resolve_indirect_ref(struct btrfs_fs_info *fs_info,
302 struct btrfs_path *path, u64 time_seq,
303 struct __prelim_ref *ref,
304 struct ulist *parents,
305 const u64 *extent_item_pos, u64 total_refs)
307 struct btrfs_root *root;
308 struct btrfs_key root_key;
309 struct extent_buffer *eb;
312 int level = ref->level;
314 root_key.objectid = ref->root_id;
315 root_key.type = BTRFS_ROOT_ITEM_KEY;
316 root_key.offset = (u64)-1;
318 root = btrfs_read_fs_root(fs_info, &root_key);
324 root_level = btrfs_root_level(&root->root_item);
326 if (root_level + 1 == level)
329 path->lowest_level = level;
330 ret = btrfs_search_slot(NULL, root, &ref->key_for_search, path, 0, 0);
332 pr_debug("search slot in root %llu (level %d, ref count %d) returned "
333 "%d for key (%llu %u %llu)\n",
334 ref->root_id, level, ref->count, ret,
335 ref->key_for_search.objectid, ref->key_for_search.type,
336 ref->key_for_search.offset);
340 eb = path->nodes[level];
348 eb = path->nodes[level];
351 ret = add_all_parents(root, path, parents, ref, level, time_seq,
352 extent_item_pos, total_refs);
354 path->lowest_level = 0;
355 btrfs_release_path(path);
360 * resolve all indirect backrefs from the list
362 static int __resolve_indirect_refs(struct btrfs_fs_info *fs_info,
363 struct pref_state *prefstate,
364 struct btrfs_path *path, u64 time_seq,
365 const u64 *extent_item_pos, u64 total_refs)
367 struct list_head *head = &prefstate->pending;
370 struct __prelim_ref *ref;
371 struct __prelim_ref *ref_safe;
372 struct __prelim_ref *new_ref;
373 struct ulist *parents;
374 struct ulist_node *node;
375 struct ulist_iterator uiter;
377 parents = ulist_alloc(GFP_NOFS);
382 * _safe allows us to insert directly after the current item without
383 * iterating over the newly inserted items.
384 * we're also allowed to re-assign ref during iteration.
386 list_for_each_entry_safe(ref, ref_safe, head, list) {
387 if (ref->parent) /* already direct */
391 err = __resolve_indirect_ref(fs_info, path, time_seq, ref,
392 parents, extent_item_pos,
395 * we can only tolerate ENOENT,otherwise,we should catch error
396 * and return directly.
398 if (err == -ENOENT) {
405 /* we put the first parent into the ref at hand */
406 ULIST_ITER_INIT(&uiter);
407 node = ulist_next(parents, &uiter);
408 ref->parent = node ? node->val : 0;
409 ref->inode_list = node ?
410 (struct extent_inode_elem *)(uintptr_t)node->aux : NULL;
412 /* additional parents require new refs being added here */
413 while ((node = ulist_next(parents, &uiter))) {
414 new_ref = kmalloc(sizeof(*new_ref), GFP_NOFS);
419 memcpy(new_ref, ref, sizeof(*ref));
420 new_ref->parent = node->val;
421 new_ref->inode_list = (struct extent_inode_elem *)
422 (uintptr_t)node->aux;
423 list_add(&new_ref->list, &ref->list);
425 ulist_reinit(parents);
432 static inline int ref_for_same_block(struct __prelim_ref *ref1,
433 struct __prelim_ref *ref2)
435 if (ref1->level != ref2->level)
437 if (ref1->root_id != ref2->root_id)
439 if (ref1->key_for_search.type != ref2->key_for_search.type)
441 if (ref1->key_for_search.objectid != ref2->key_for_search.objectid)
443 if (ref1->key_for_search.offset != ref2->key_for_search.offset)
445 if (ref1->parent != ref2->parent)
452 * read tree blocks and add keys where required.
454 static int __add_missing_keys(struct btrfs_fs_info *fs_info,
455 struct pref_state *prefstate)
457 struct list_head *head = &prefstate->pending;
458 struct list_head *pos;
459 struct extent_buffer *eb;
461 list_for_each(pos, head) {
462 struct __prelim_ref *ref;
463 ref = list_entry(pos, struct __prelim_ref, list);
467 if (ref->key_for_search.type)
469 BUG_ON(!ref->wanted_disk_byte);
470 eb = read_tree_block(fs_info, ref->wanted_disk_byte, 0);
471 if (!extent_buffer_uptodate(eb)) {
472 free_extent_buffer(eb);
475 if (btrfs_header_level(eb) == 0)
476 btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
478 btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
479 free_extent_buffer(eb);
485 * merge two lists of backrefs and adjust counts accordingly
487 * mode = 1: merge identical keys, if key is set
488 * FIXME: if we add more keys in __add_prelim_ref, we can merge more here.
489 * additionally, we could even add a key range for the blocks we
490 * looked into to merge even more (-> replace unresolved refs by those
492 * mode = 2: merge identical parents
494 static void __merge_refs(struct pref_state *prefstate, int mode)
496 struct list_head *head = &prefstate->pending;
497 struct list_head *pos1;
499 list_for_each(pos1, head) {
500 struct list_head *n2;
501 struct list_head *pos2;
502 struct __prelim_ref *ref1;
504 ref1 = list_entry(pos1, struct __prelim_ref, list);
506 for (pos2 = pos1->next, n2 = pos2->next; pos2 != head;
507 pos2 = n2, n2 = pos2->next) {
508 struct __prelim_ref *ref2;
509 struct __prelim_ref *xchg;
510 struct extent_inode_elem *eie;
512 ref2 = list_entry(pos2, struct __prelim_ref, list);
515 if (!ref_for_same_block(ref1, ref2))
517 if (!ref1->parent && ref2->parent) {
523 if (ref1->parent != ref2->parent)
527 eie = ref1->inode_list;
528 while (eie && eie->next)
531 eie->next = ref2->inode_list;
533 ref1->inode_list = ref2->inode_list;
534 ref1->count += ref2->count;
536 list_del(&ref2->list);
544 * add all inline backrefs for bytenr to the list
546 static int __add_inline_refs(struct btrfs_fs_info *fs_info,
547 struct pref_state *prefstate,
548 struct btrfs_path *path, u64 bytenr,
549 int *info_level, u64 *total_refs)
553 struct extent_buffer *leaf;
554 struct btrfs_key key;
555 struct btrfs_key found_key;
558 struct btrfs_extent_item *ei;
562 * enumerate all inline refs
564 leaf = path->nodes[0];
565 slot = path->slots[0];
567 item_size = btrfs_item_size_nr(leaf, slot);
568 BUG_ON(item_size < sizeof(*ei));
570 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
571 flags = btrfs_extent_flags(leaf, ei);
572 *total_refs += btrfs_extent_refs(leaf, ei);
573 btrfs_item_key_to_cpu(leaf, &found_key, slot);
575 ptr = (unsigned long)(ei + 1);
576 end = (unsigned long)ei + item_size;
578 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
579 flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
580 struct btrfs_tree_block_info *info;
582 info = (struct btrfs_tree_block_info *)ptr;
583 *info_level = btrfs_tree_block_level(leaf, info);
584 ptr += sizeof(struct btrfs_tree_block_info);
586 } else if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
587 *info_level = found_key.offset;
589 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
593 struct btrfs_extent_inline_ref *iref;
597 iref = (struct btrfs_extent_inline_ref *)ptr;
598 type = btrfs_extent_inline_ref_type(leaf, iref);
599 offset = btrfs_extent_inline_ref_offset(leaf, iref);
602 case BTRFS_SHARED_BLOCK_REF_KEY:
603 ret = __add_prelim_ref(prefstate, 0, NULL,
604 *info_level + 1, offset,
605 bytenr, 1, GFP_NOFS);
607 case BTRFS_SHARED_DATA_REF_KEY: {
608 struct btrfs_shared_data_ref *sdref;
611 sdref = (struct btrfs_shared_data_ref *)(iref + 1);
612 count = btrfs_shared_data_ref_count(leaf, sdref);
613 ret = __add_prelim_ref(prefstate, 0, NULL, 0, offset,
614 bytenr, count, GFP_NOFS);
617 case BTRFS_TREE_BLOCK_REF_KEY:
618 ret = __add_prelim_ref(prefstate, offset, NULL,
620 bytenr, 1, GFP_NOFS);
622 case BTRFS_EXTENT_DATA_REF_KEY: {
623 struct btrfs_extent_data_ref *dref;
627 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
628 count = btrfs_extent_data_ref_count(leaf, dref);
629 key.objectid = btrfs_extent_data_ref_objectid(leaf,
631 key.type = BTRFS_EXTENT_DATA_KEY;
632 key.offset = btrfs_extent_data_ref_offset(leaf, dref);
633 root = btrfs_extent_data_ref_root(leaf, dref);
634 ret = __add_prelim_ref(prefstate, root, &key, 0, 0,
635 bytenr, count, GFP_NOFS);
643 ptr += btrfs_extent_inline_ref_size(type);
650 * add all non-inline backrefs for bytenr to the list
652 static int __add_keyed_refs(struct btrfs_fs_info *fs_info,
653 struct pref_state *prefstate,
654 struct btrfs_path *path, u64 bytenr,
657 struct btrfs_root *extent_root = fs_info->extent_root;
660 struct extent_buffer *leaf;
661 struct btrfs_key key;
664 ret = btrfs_next_item(extent_root, path);
672 slot = path->slots[0];
673 leaf = path->nodes[0];
674 btrfs_item_key_to_cpu(leaf, &key, slot);
676 if (key.objectid != bytenr)
678 if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
680 if (key.type > BTRFS_SHARED_DATA_REF_KEY)
684 case BTRFS_SHARED_BLOCK_REF_KEY:
685 ret = __add_prelim_ref(prefstate, 0, NULL,
686 info_level + 1, key.offset,
687 bytenr, 1, GFP_NOFS);
689 case BTRFS_SHARED_DATA_REF_KEY: {
690 struct btrfs_shared_data_ref *sdref;
693 sdref = btrfs_item_ptr(leaf, slot,
694 struct btrfs_shared_data_ref);
695 count = btrfs_shared_data_ref_count(leaf, sdref);
696 ret = __add_prelim_ref(prefstate, 0, NULL, 0, key.offset,
697 bytenr, count, GFP_NOFS);
700 case BTRFS_TREE_BLOCK_REF_KEY:
701 ret = __add_prelim_ref(prefstate, key.offset, NULL,
703 bytenr, 1, GFP_NOFS);
705 case BTRFS_EXTENT_DATA_REF_KEY: {
706 struct btrfs_extent_data_ref *dref;
710 dref = btrfs_item_ptr(leaf, slot,
711 struct btrfs_extent_data_ref);
712 count = btrfs_extent_data_ref_count(leaf, dref);
713 key.objectid = btrfs_extent_data_ref_objectid(leaf,
715 key.type = BTRFS_EXTENT_DATA_KEY;
716 key.offset = btrfs_extent_data_ref_offset(leaf, dref);
717 root = btrfs_extent_data_ref_root(leaf, dref);
718 ret = __add_prelim_ref(prefstate, root, &key, 0, 0,
719 bytenr, count, GFP_NOFS);
734 * this adds all existing backrefs (inline backrefs, backrefs and delayed
735 * refs) for the given bytenr to the refs list, merges duplicates and resolves
736 * indirect refs to their parent bytenr.
737 * When roots are found, they're added to the roots list
739 * FIXME some caching might speed things up
741 static int find_parent_nodes(struct btrfs_trans_handle *trans,
742 struct btrfs_fs_info *fs_info, u64 bytenr,
743 u64 time_seq, struct ulist *refs,
744 struct ulist *roots, const u64 *extent_item_pos)
746 struct btrfs_key key;
747 struct btrfs_path *path;
750 struct pref_state prefstate;
751 struct __prelim_ref *ref;
752 struct extent_inode_elem *eie = NULL;
755 init_pref_state(&prefstate);
757 key.objectid = bytenr;
758 key.offset = (u64)-1;
759 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
760 key.type = BTRFS_METADATA_ITEM_KEY;
762 key.type = BTRFS_EXTENT_ITEM_KEY;
764 path = btrfs_alloc_path();
768 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
773 if (path->slots[0]) {
774 struct extent_buffer *leaf;
778 leaf = path->nodes[0];
779 slot = path->slots[0];
780 btrfs_item_key_to_cpu(leaf, &key, slot);
781 if (key.objectid == bytenr &&
782 (key.type == BTRFS_EXTENT_ITEM_KEY ||
783 key.type == BTRFS_METADATA_ITEM_KEY)) {
784 ret = __add_inline_refs(fs_info, &prefstate, path,
789 ret = __add_keyed_refs(fs_info, &prefstate, path,
795 btrfs_release_path(path);
797 ret = __add_missing_keys(fs_info, &prefstate);
801 __merge_refs(&prefstate, 1);
803 ret = __resolve_indirect_refs(fs_info, &prefstate, path, time_seq,
804 extent_item_pos, total_refs);
808 __merge_refs(&prefstate, 2);
810 while (!list_empty(&prefstate.pending)) {
811 ref = list_first_pref(&prefstate.pending);
812 WARN_ON(ref->count < 0);
813 if (roots && ref->count && ref->root_id && ref->parent == 0) {
814 /* no parent == root of tree */
815 ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS);
819 if (ref->count && ref->parent) {
820 if (extent_item_pos && !ref->inode_list &&
822 struct extent_buffer *eb;
824 eb = read_tree_block(fs_info, ref->parent, 0);
825 if (!extent_buffer_uptodate(eb)) {
826 free_extent_buffer(eb);
830 ret = find_extent_in_eb(eb, bytenr,
831 *extent_item_pos, &eie);
832 free_extent_buffer(eb);
835 ref->inode_list = eie;
837 ret = ulist_add_merge_ptr(refs, ref->parent,
839 (void **)&eie, GFP_NOFS);
842 if (!ret && extent_item_pos) {
844 * we've recorded that parent, so we must extend
845 * its inode list here
850 eie->next = ref->inode_list;
854 list_del(&ref->list);
859 btrfs_free_path(path);
860 while (!list_empty(&prefstate.pending)) {
861 ref = list_first_pref(&prefstate.pending);
862 list_del(&ref->list);
866 free_inode_elem_list(eie);
870 static void free_leaf_list(struct ulist *blocks)
872 struct ulist_node *node = NULL;
873 struct extent_inode_elem *eie;
874 struct ulist_iterator uiter;
876 ULIST_ITER_INIT(&uiter);
877 while ((node = ulist_next(blocks, &uiter))) {
880 eie = (struct extent_inode_elem *)(uintptr_t)node->aux;
881 free_inode_elem_list(eie);
889 * Finds all leafs with a reference to the specified combination of bytenr and
890 * offset. key_list_head will point to a list of corresponding keys (caller must
891 * free each list element). The leafs will be stored in the leafs ulist, which
892 * must be freed with ulist_free.
894 * returns 0 on success, <0 on error
896 static int btrfs_find_all_leafs(struct btrfs_trans_handle *trans,
897 struct btrfs_fs_info *fs_info, u64 bytenr,
898 u64 time_seq, struct ulist **leafs,
899 const u64 *extent_item_pos)
903 *leafs = ulist_alloc(GFP_NOFS);
907 ret = find_parent_nodes(trans, fs_info, bytenr,
908 time_seq, *leafs, NULL, extent_item_pos);
909 if (ret < 0 && ret != -ENOENT) {
910 free_leaf_list(*leafs);
918 * walk all backrefs for a given extent to find all roots that reference this
919 * extent. Walking a backref means finding all extents that reference this
920 * extent and in turn walk the backrefs of those, too. Naturally this is a
921 * recursive process, but here it is implemented in an iterative fashion: We
922 * find all referencing extents for the extent in question and put them on a
923 * list. In turn, we find all referencing extents for those, further appending
924 * to the list. The way we iterate the list allows adding more elements after
925 * the current while iterating. The process stops when we reach the end of the
926 * list. Found roots are added to the roots list.
928 * returns 0 on success, < 0 on error.
930 static int __btrfs_find_all_roots(struct btrfs_trans_handle *trans,
931 struct btrfs_fs_info *fs_info, u64 bytenr,
932 u64 time_seq, struct ulist **roots)
935 struct ulist_node *node = NULL;
936 struct ulist_iterator uiter;
939 tmp = ulist_alloc(GFP_NOFS);
942 *roots = ulist_alloc(GFP_NOFS);
948 ULIST_ITER_INIT(&uiter);
950 ret = find_parent_nodes(trans, fs_info, bytenr,
951 time_seq, tmp, *roots, NULL);
952 if (ret < 0 && ret != -ENOENT) {
957 node = ulist_next(tmp, &uiter);
968 int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
969 struct btrfs_fs_info *fs_info, u64 bytenr,
970 u64 time_seq, struct ulist **roots)
972 return __btrfs_find_all_roots(trans, fs_info, bytenr, time_seq, roots);
976 * this makes the path point to (inum INODE_ITEM ioff)
978 int inode_item_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
979 struct btrfs_path *path)
981 struct btrfs_key key;
982 return btrfs_find_item(fs_root, path, inum, ioff,
983 BTRFS_INODE_ITEM_KEY, &key);
986 static int inode_ref_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
987 struct btrfs_path *path,
988 struct btrfs_key *found_key)
990 return btrfs_find_item(fs_root, path, inum, ioff,
991 BTRFS_INODE_REF_KEY, found_key);
994 int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,
995 u64 start_off, struct btrfs_path *path,
996 struct btrfs_inode_extref **ret_extref,
1000 struct btrfs_key key;
1001 struct btrfs_key found_key;
1002 struct btrfs_inode_extref *extref;
1003 struct extent_buffer *leaf;
1006 key.objectid = inode_objectid;
1007 key.type = BTRFS_INODE_EXTREF_KEY;
1008 key.offset = start_off;
1010 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1015 leaf = path->nodes[0];
1016 slot = path->slots[0];
1017 if (slot >= btrfs_header_nritems(leaf)) {
1019 * If the item at offset is not found,
1020 * btrfs_search_slot will point us to the slot
1021 * where it should be inserted. In our case
1022 * that will be the slot directly before the
1023 * next INODE_REF_KEY_V2 item. In the case
1024 * that we're pointing to the last slot in a
1025 * leaf, we must move one leaf over.
1027 ret = btrfs_next_leaf(root, path);
1036 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1039 * Check that we're still looking at an extended ref key for
1040 * this particular objectid. If we have different
1041 * objectid or type then there are no more to be found
1042 * in the tree and we can exit.
1045 if (found_key.objectid != inode_objectid)
1047 if (found_key.type != BTRFS_INODE_EXTREF_KEY)
1051 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1052 extref = (struct btrfs_inode_extref *)ptr;
1053 *ret_extref = extref;
1055 *found_off = found_key.offset;
1063 * this iterates to turn a name (from iref/extref) into a full filesystem path.
1064 * Elements of the path are separated by '/' and the path is guaranteed to be
1065 * 0-terminated. the path is only given within the current file system.
1066 * Therefore, it never starts with a '/'. the caller is responsible to provide
1067 * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
1068 * the start point of the resulting string is returned. this pointer is within
1070 * in case the path buffer would overflow, the pointer is decremented further
1071 * as if output was written to the buffer, though no more output is actually
1072 * generated. that way, the caller can determine how much space would be
1073 * required for the path to fit into the buffer. in that case, the returned
1074 * value will be smaller than dest. callers must check this!
1076 char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
1077 u32 name_len, unsigned long name_off,
1078 struct extent_buffer *eb_in, u64 parent,
1079 char *dest, u32 size)
1084 s64 bytes_left = ((s64)size) - 1;
1085 struct extent_buffer *eb = eb_in;
1086 struct btrfs_key found_key;
1087 struct btrfs_inode_ref *iref;
1089 if (bytes_left >= 0)
1090 dest[bytes_left] = '\0';
1093 bytes_left -= name_len;
1094 if (bytes_left >= 0)
1095 read_extent_buffer(eb, dest + bytes_left,
1096 name_off, name_len);
1098 free_extent_buffer(eb);
1099 ret = inode_ref_info(parent, 0, fs_root, path, &found_key);
1105 next_inum = found_key.offset;
1107 /* regular exit ahead */
1108 if (parent == next_inum)
1111 slot = path->slots[0];
1112 eb = path->nodes[0];
1113 /* make sure we can use eb after releasing the path */
1116 btrfs_release_path(path);
1117 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1119 name_len = btrfs_inode_ref_name_len(eb, iref);
1120 name_off = (unsigned long)(iref + 1);
1124 if (bytes_left >= 0)
1125 dest[bytes_left] = '/';
1128 btrfs_release_path(path);
1131 return ERR_PTR(ret);
1133 return dest + bytes_left;
1137 * this makes the path point to (logical EXTENT_ITEM *)
1138 * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
1139 * tree blocks and <0 on error.
1141 int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
1142 struct btrfs_path *path, struct btrfs_key *found_key,
1149 struct extent_buffer *eb;
1150 struct btrfs_extent_item *ei;
1151 struct btrfs_key key;
1153 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1154 key.type = BTRFS_METADATA_ITEM_KEY;
1156 key.type = BTRFS_EXTENT_ITEM_KEY;
1157 key.objectid = logical;
1158 key.offset = (u64)-1;
1160 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
1164 ret = btrfs_previous_extent_item(fs_info->extent_root, path, 0);
1170 btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]);
1171 if (found_key->type == BTRFS_METADATA_ITEM_KEY)
1172 size = fs_info->nodesize;
1173 else if (found_key->type == BTRFS_EXTENT_ITEM_KEY)
1174 size = found_key->offset;
1176 if (found_key->objectid > logical ||
1177 found_key->objectid + size <= logical) {
1178 pr_debug("logical %llu is not within any extent\n", logical);
1182 eb = path->nodes[0];
1183 item_size = btrfs_item_size_nr(eb, path->slots[0]);
1184 BUG_ON(item_size < sizeof(*ei));
1186 ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
1187 flags = btrfs_extent_flags(eb, ei);
1189 pr_debug("logical %llu is at position %llu within the extent (%llu "
1190 "EXTENT_ITEM %llu) flags %#llx size %u\n",
1191 logical, logical - found_key->objectid, found_key->objectid,
1192 found_key->offset, flags, item_size);
1195 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1196 *flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK;
1197 else if (flags & BTRFS_EXTENT_FLAG_DATA)
1198 *flags_ret = BTRFS_EXTENT_FLAG_DATA;
1209 * helper function to iterate extent inline refs. ptr must point to a 0 value
1210 * for the first call and may be modified. it is used to track state.
1211 * if more refs exist, 0 is returned and the next call to
1212 * __get_extent_inline_ref must pass the modified ptr parameter to get the
1213 * next ref. after the last ref was processed, 1 is returned.
1214 * returns <0 on error
1216 static int __get_extent_inline_ref(unsigned long *ptr, struct extent_buffer *eb,
1217 struct btrfs_key *key,
1218 struct btrfs_extent_item *ei, u32 item_size,
1219 struct btrfs_extent_inline_ref **out_eiref,
1224 struct btrfs_tree_block_info *info;
1228 flags = btrfs_extent_flags(eb, ei);
1229 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1230 if (key->type == BTRFS_METADATA_ITEM_KEY) {
1231 /* a skinny metadata extent */
1233 (struct btrfs_extent_inline_ref *)(ei + 1);
1235 WARN_ON(key->type != BTRFS_EXTENT_ITEM_KEY);
1236 info = (struct btrfs_tree_block_info *)(ei + 1);
1238 (struct btrfs_extent_inline_ref *)(info + 1);
1241 *out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1);
1243 *ptr = (unsigned long)*out_eiref;
1244 if ((unsigned long)(*ptr) >= (unsigned long)ei + item_size)
1248 end = (unsigned long)ei + item_size;
1249 *out_eiref = (struct btrfs_extent_inline_ref *)(*ptr);
1250 *out_type = btrfs_extent_inline_ref_type(eb, *out_eiref);
1252 *ptr += btrfs_extent_inline_ref_size(*out_type);
1253 WARN_ON(*ptr > end);
1255 return 1; /* last */
1261 * reads the tree block backref for an extent. tree level and root are returned
1262 * through out_level and out_root. ptr must point to a 0 value for the first
1263 * call and may be modified (see __get_extent_inline_ref comment).
1264 * returns 0 if data was provided, 1 if there was no more data to provide or
1267 int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
1268 struct btrfs_key *key, struct btrfs_extent_item *ei,
1269 u32 item_size, u64 *out_root, u8 *out_level)
1273 struct btrfs_tree_block_info *info;
1274 struct btrfs_extent_inline_ref *eiref;
1276 if (*ptr == (unsigned long)-1)
1280 ret = __get_extent_inline_ref(ptr, eb, key, ei, item_size,
1285 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1286 type == BTRFS_SHARED_BLOCK_REF_KEY)
1293 /* we can treat both ref types equally here */
1294 info = (struct btrfs_tree_block_info *)(ei + 1);
1295 *out_root = btrfs_extent_inline_ref_offset(eb, eiref);
1296 *out_level = btrfs_tree_block_level(eb, info);
1299 *ptr = (unsigned long)-1;
1304 static int iterate_leaf_refs(struct extent_inode_elem *inode_list,
1305 u64 root, u64 extent_item_objectid,
1306 iterate_extent_inodes_t *iterate, void *ctx)
1308 struct extent_inode_elem *eie;
1311 for (eie = inode_list; eie; eie = eie->next) {
1312 pr_debug("ref for %llu resolved, key (%llu EXTEND_DATA %llu), "
1313 "root %llu\n", extent_item_objectid,
1314 eie->inum, eie->offset, root);
1315 ret = iterate(eie->inum, eie->offset, root, ctx);
1317 pr_debug("stopping iteration for %llu due to ret=%d\n",
1318 extent_item_objectid, ret);
1327 * calls iterate() for every inode that references the extent identified by
1328 * the given parameters.
1329 * when the iterator function returns a non-zero value, iteration stops.
1331 int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
1332 u64 extent_item_objectid, u64 extent_item_pos,
1333 int search_commit_root,
1334 iterate_extent_inodes_t *iterate, void *ctx)
1337 struct btrfs_trans_handle *trans = NULL;
1338 struct ulist *refs = NULL;
1339 struct ulist *roots = NULL;
1340 struct ulist_node *ref_node = NULL;
1341 struct ulist_node *root_node = NULL;
1342 struct ulist_iterator ref_uiter;
1343 struct ulist_iterator root_uiter;
1345 pr_debug("resolving all inodes for extent %llu\n",
1346 extent_item_objectid);
1348 ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid,
1349 0, &refs, &extent_item_pos);
1353 ULIST_ITER_INIT(&ref_uiter);
1354 while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) {
1355 ret = __btrfs_find_all_roots(trans, fs_info, ref_node->val,
1359 ULIST_ITER_INIT(&root_uiter);
1360 while (!ret && (root_node = ulist_next(roots, &root_uiter))) {
1361 pr_debug("root %llu references leaf %llu, data list "
1362 "%#llx\n", root_node->val, ref_node->val,
1364 ret = iterate_leaf_refs((struct extent_inode_elem *)
1365 (uintptr_t)ref_node->aux,
1367 extent_item_objectid,
1373 free_leaf_list(refs);
1378 int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
1379 struct btrfs_path *path,
1380 iterate_extent_inodes_t *iterate, void *ctx)
1383 u64 extent_item_pos;
1385 struct btrfs_key found_key;
1386 int search_commit_root = 0;
1388 ret = extent_from_logical(fs_info, logical, path, &found_key, &flags);
1389 btrfs_release_path(path);
1392 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1395 extent_item_pos = logical - found_key.objectid;
1396 ret = iterate_extent_inodes(fs_info, found_key.objectid,
1397 extent_item_pos, search_commit_root,
1403 typedef int (iterate_irefs_t)(u64 parent, u32 name_len, unsigned long name_off,
1404 struct extent_buffer *eb, void *ctx);
1406 static int iterate_inode_refs(u64 inum, struct btrfs_root *fs_root,
1407 struct btrfs_path *path,
1408 iterate_irefs_t *iterate, void *ctx)
1417 struct extent_buffer *eb;
1418 struct btrfs_item *item;
1419 struct btrfs_inode_ref *iref;
1420 struct btrfs_key found_key;
1423 ret = inode_ref_info(inum, parent ? parent+1 : 0, fs_root, path,
1428 ret = found ? 0 : -ENOENT;
1433 parent = found_key.offset;
1434 slot = path->slots[0];
1435 eb = btrfs_clone_extent_buffer(path->nodes[0]);
1440 extent_buffer_get(eb);
1441 btrfs_release_path(path);
1443 item = btrfs_item_nr(slot);
1444 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1446 for (cur = 0; cur < btrfs_item_size(eb, item); cur += len) {
1447 name_len = btrfs_inode_ref_name_len(eb, iref);
1448 /* path must be released before calling iterate()! */
1449 pr_debug("following ref at offset %u for inode %llu in "
1450 "tree %llu\n", cur, found_key.objectid,
1452 ret = iterate(parent, name_len,
1453 (unsigned long)(iref + 1), eb, ctx);
1456 len = sizeof(*iref) + name_len;
1457 iref = (struct btrfs_inode_ref *)((char *)iref + len);
1459 free_extent_buffer(eb);
1462 btrfs_release_path(path);
1467 static int iterate_inode_extrefs(u64 inum, struct btrfs_root *fs_root,
1468 struct btrfs_path *path,
1469 iterate_irefs_t *iterate, void *ctx)
1476 struct extent_buffer *eb;
1477 struct btrfs_inode_extref *extref;
1478 struct extent_buffer *leaf;
1484 ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref,
1489 ret = found ? 0 : -ENOENT;
1494 slot = path->slots[0];
1495 eb = btrfs_clone_extent_buffer(path->nodes[0]);
1500 extent_buffer_get(eb);
1502 btrfs_release_path(path);
1504 leaf = path->nodes[0];
1505 item_size = btrfs_item_size_nr(leaf, slot);
1506 ptr = btrfs_item_ptr_offset(leaf, slot);
1509 while (cur_offset < item_size) {
1512 extref = (struct btrfs_inode_extref *)(ptr + cur_offset);
1513 parent = btrfs_inode_extref_parent(eb, extref);
1514 name_len = btrfs_inode_extref_name_len(eb, extref);
1515 ret = iterate(parent, name_len,
1516 (unsigned long)&extref->name, eb, ctx);
1520 cur_offset += btrfs_inode_extref_name_len(leaf, extref);
1521 cur_offset += sizeof(*extref);
1523 free_extent_buffer(eb);
1528 btrfs_release_path(path);
1533 static int iterate_irefs(u64 inum, struct btrfs_root *fs_root,
1534 struct btrfs_path *path, iterate_irefs_t *iterate,
1540 ret = iterate_inode_refs(inum, fs_root, path, iterate, ctx);
1543 else if (ret != -ENOENT)
1546 ret = iterate_inode_extrefs(inum, fs_root, path, iterate, ctx);
1547 if (ret == -ENOENT && found_refs)
1554 * returns 0 if the path could be dumped (probably truncated)
1555 * returns <0 in case of an error
1557 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
1558 struct extent_buffer *eb, void *ctx)
1560 struct inode_fs_paths *ipath = ctx;
1563 int i = ipath->fspath->elem_cnt;
1564 const int s_ptr = sizeof(char *);
1567 bytes_left = ipath->fspath->bytes_left > s_ptr ?
1568 ipath->fspath->bytes_left - s_ptr : 0;
1570 fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
1571 fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len,
1572 name_off, eb, inum, fspath_min, bytes_left);
1574 return PTR_ERR(fspath);
1576 if (fspath > fspath_min) {
1577 ipath->fspath->val[i] = (u64)(unsigned long)fspath;
1578 ++ipath->fspath->elem_cnt;
1579 ipath->fspath->bytes_left = fspath - fspath_min;
1581 ++ipath->fspath->elem_missed;
1582 ipath->fspath->bytes_missing += fspath_min - fspath;
1583 ipath->fspath->bytes_left = 0;
1590 * this dumps all file system paths to the inode into the ipath struct, provided
1591 * is has been created large enough. each path is zero-terminated and accessed
1592 * from ipath->fspath->val[i].
1593 * when it returns, there are ipath->fspath->elem_cnt number of paths available
1594 * in ipath->fspath->val[]. When the allocated space wasn't sufficient, the
1595 * number of missed paths is recorded in ipath->fspath->elem_missed, otherwise,
1596 * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
1597 * have been needed to return all paths.
1599 int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
1601 return iterate_irefs(inum, ipath->fs_root, ipath->btrfs_path,
1602 inode_to_path, ipath);
1605 struct btrfs_data_container *init_data_container(u32 total_bytes)
1607 struct btrfs_data_container *data;
1610 alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
1611 data = vmalloc(alloc_bytes);
1613 return ERR_PTR(-ENOMEM);
1615 if (total_bytes >= sizeof(*data)) {
1616 data->bytes_left = total_bytes - sizeof(*data);
1617 data->bytes_missing = 0;
1619 data->bytes_missing = sizeof(*data) - total_bytes;
1620 data->bytes_left = 0;
1624 data->elem_missed = 0;
1630 * allocates space to return multiple file system paths for an inode.
1631 * total_bytes to allocate are passed, note that space usable for actual path
1632 * information will be total_bytes - sizeof(struct inode_fs_paths).
1633 * the returned pointer must be freed with free_ipath() in the end.
1635 struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
1636 struct btrfs_path *path)
1638 struct inode_fs_paths *ifp;
1639 struct btrfs_data_container *fspath;
1641 fspath = init_data_container(total_bytes);
1643 return (void *)fspath;
1645 ifp = kmalloc(sizeof(*ifp), GFP_NOFS);
1648 return ERR_PTR(-ENOMEM);
1651 ifp->btrfs_path = path;
1652 ifp->fspath = fspath;
1653 ifp->fs_root = fs_root;
1658 void free_ipath(struct inode_fs_paths *ipath)
1662 vfree(ipath->fspath);