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;
140 struct list_head pending_missing_keys;
141 struct list_head pending_indirect_refs;
144 static void init_pref_state(struct pref_state *prefstate)
146 INIT_LIST_HEAD(&prefstate->pending);
147 INIT_LIST_HEAD(&prefstate->pending_missing_keys);
148 INIT_LIST_HEAD(&prefstate->pending_indirect_refs);
152 * the rules for all callers of this function are:
153 * - obtaining the parent is the goal
154 * - if you add a key, you must know that it is a correct key
155 * - if you cannot add the parent or a correct key, then we will look into the
156 * block later to set a correct key
160 * backref type | shared | indirect | shared | indirect
161 * information | tree | tree | data | data
162 * --------------------+--------+----------+--------+----------
163 * parent logical | y | - | - | -
164 * key to resolve | - | y | y | y
165 * tree block logical | - | - | - | -
166 * root for resolving | y | y | y | y
168 * - column 1: we've the parent -> done
169 * - column 2, 3, 4: we use the key to find the parent
171 * on disk refs (inline or keyed)
172 * ==============================
173 * backref type | shared | indirect | shared | indirect
174 * information | tree | tree | data | data
175 * --------------------+--------+----------+--------+----------
176 * parent logical | y | - | y | -
177 * key to resolve | - | - | - | y
178 * tree block logical | y | y | y | y
179 * root for resolving | - | y | y | y
181 * - column 1, 3: we've the parent -> done
182 * - column 2: we take the first key from the block to find the parent
183 * (see __add_missing_keys)
184 * - column 4: we use the key to find the parent
186 * additional information that's available but not required to find the parent
187 * block might help in merging entries to gain some speed.
190 static int __add_prelim_ref(struct pref_state *prefstate, u64 root_id,
191 struct btrfs_key *key, int level,
192 u64 parent, u64 wanted_disk_byte, int count,
195 struct list_head *head;
196 struct __prelim_ref *ref;
198 if (root_id == BTRFS_DATA_RELOC_TREE_OBJECTID)
201 ref = kmalloc(sizeof(*ref), gfp_mask);
205 ref->root_id = root_id;
207 ref->key_for_search = *key;
208 head = &prefstate->pending;
210 memset(&ref->key_for_search, 0, sizeof(ref->key_for_search));
211 head = &prefstate->pending_missing_keys;
214 ref->inode_list = NULL;
217 ref->parent = parent;
218 ref->wanted_disk_byte = wanted_disk_byte;
220 list_add_tail(&ref->list, head);
225 static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path,
226 struct ulist *parents, struct __prelim_ref *ref,
227 int level, u64 time_seq, const u64 *extent_item_pos,
232 struct extent_buffer *eb;
233 struct btrfs_key key;
234 struct btrfs_key *key_for_search = &ref->key_for_search;
235 struct btrfs_file_extent_item *fi;
236 struct extent_inode_elem *eie = NULL, *old = NULL;
238 u64 wanted_disk_byte = ref->wanted_disk_byte;
242 eb = path->nodes[level];
243 ret = ulist_add(parents, eb->start, 0, GFP_NOFS);
250 * We normally enter this function with the path already pointing to
251 * the first item to check. But sometimes, we may enter it with
252 * slot==nritems. In that case, go to the next leaf before we continue.
254 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0]))
255 ret = btrfs_next_leaf(root, path);
257 while (!ret && count < total_refs) {
259 slot = path->slots[0];
261 btrfs_item_key_to_cpu(eb, &key, slot);
263 if (key.objectid != key_for_search->objectid ||
264 key.type != BTRFS_EXTENT_DATA_KEY)
267 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
268 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
270 if (disk_byte == wanted_disk_byte) {
274 if (extent_item_pos) {
275 ret = check_extent_in_eb(&key, eb, fi,
283 ret = ulist_add_merge_ptr(parents, eb->start,
284 eie, (void **)&old, GFP_NOFS);
287 if (!ret && extent_item_pos) {
295 ret = btrfs_next_item(root, path);
301 free_inode_elem_list(eie);
306 * resolve an indirect backref in the form (root_id, key, level)
307 * to a logical address
309 static int __resolve_indirect_ref(struct btrfs_fs_info *fs_info,
310 struct btrfs_path *path, u64 time_seq,
311 struct __prelim_ref *ref,
312 struct ulist *parents,
313 const u64 *extent_item_pos, u64 total_refs)
315 struct btrfs_root *root;
316 struct btrfs_key root_key;
317 struct extent_buffer *eb;
320 int level = ref->level;
322 root_key.objectid = ref->root_id;
323 root_key.type = BTRFS_ROOT_ITEM_KEY;
324 root_key.offset = (u64)-1;
326 root = btrfs_read_fs_root(fs_info, &root_key);
332 root_level = btrfs_root_level(&root->root_item);
334 if (root_level + 1 == level)
337 path->lowest_level = level;
338 ret = btrfs_search_slot(NULL, root, &ref->key_for_search, path, 0, 0);
340 pr_debug("search slot in root %llu (level %d, ref count %d) returned "
341 "%d for key (%llu %u %llu)\n",
342 ref->root_id, level, ref->count, ret,
343 ref->key_for_search.objectid, ref->key_for_search.type,
344 ref->key_for_search.offset);
348 eb = path->nodes[level];
356 eb = path->nodes[level];
359 ret = add_all_parents(root, path, parents, ref, level, time_seq,
360 extent_item_pos, total_refs);
362 path->lowest_level = 0;
363 btrfs_release_path(path);
368 * resolve all indirect backrefs from the list
370 static int __resolve_indirect_refs(struct btrfs_fs_info *fs_info,
371 struct pref_state *prefstate,
372 struct btrfs_path *path, u64 time_seq,
373 const u64 *extent_item_pos, u64 total_refs)
375 struct list_head *head = &prefstate->pending_indirect_refs;
378 struct __prelim_ref *ref;
379 struct __prelim_ref *new_ref;
380 struct ulist *parents;
381 struct ulist_node *node;
382 struct ulist_iterator uiter;
384 parents = ulist_alloc(GFP_NOFS);
388 while (!list_empty(head)) {
389 ref = list_first_pref(head);
390 list_move(&ref->list, &prefstate->pending);
391 ASSERT(!ref->parent); /* already direct */
393 err = __resolve_indirect_ref(fs_info, path, time_seq, ref,
394 parents, extent_item_pos,
397 * we can only tolerate ENOENT,otherwise,we should catch error
398 * and return directly.
400 if (err == -ENOENT) {
407 /* we put the first parent into the ref at hand */
408 ULIST_ITER_INIT(&uiter);
409 node = ulist_next(parents, &uiter);
410 ref->parent = node ? node->val : 0;
411 ref->inode_list = node ?
412 (struct extent_inode_elem *)(uintptr_t)node->aux : NULL;
414 /* additional parents require new refs being added here */
415 while ((node = ulist_next(parents, &uiter))) {
416 new_ref = kmalloc(sizeof(*new_ref), GFP_NOFS);
421 memcpy(new_ref, ref, sizeof(*ref));
422 new_ref->parent = node->val;
423 new_ref->inode_list = (struct extent_inode_elem *)
424 (uintptr_t)node->aux;
425 list_add_tail(&new_ref->list, &prefstate->pending);
427 ulist_reinit(parents);
434 static inline int ref_for_same_block(struct __prelim_ref *ref1,
435 struct __prelim_ref *ref2)
437 if (ref1->level != ref2->level)
439 if (ref1->root_id != ref2->root_id)
441 if (ref1->key_for_search.type != ref2->key_for_search.type)
443 if (ref1->key_for_search.objectid != ref2->key_for_search.objectid)
445 if (ref1->key_for_search.offset != ref2->key_for_search.offset)
447 if (ref1->parent != ref2->parent)
454 * read tree blocks and add keys where required.
456 static int __add_missing_keys(struct btrfs_fs_info *fs_info,
457 struct pref_state *prefstate)
459 struct extent_buffer *eb;
461 while (!list_empty(&prefstate->pending_missing_keys)) {
462 struct __prelim_ref *ref;
464 ref = list_first_pref(&prefstate->pending_missing_keys);
466 ASSERT(ref->root_id);
467 ASSERT(!ref->parent);
468 ASSERT(!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);
480 list_move(&ref->list, &prefstate->pending);
486 * merge two lists of backrefs and adjust counts accordingly
488 * mode = 1: merge identical keys, if key is set
489 * FIXME: if we add more keys in __add_prelim_ref, we can merge more here.
490 * additionally, we could even add a key range for the blocks we
491 * looked into to merge even more (-> replace unresolved refs by those
493 * mode = 2: merge identical parents
495 static void __merge_refs(struct pref_state *prefstate, int mode)
497 struct list_head *head = &prefstate->pending;
498 struct list_head *pos1;
500 list_for_each(pos1, head) {
501 struct list_head *n2;
502 struct list_head *pos2;
503 struct __prelim_ref *ref1;
505 ref1 = list_entry(pos1, struct __prelim_ref, list);
507 for (pos2 = pos1->next, n2 = pos2->next; pos2 != head;
508 pos2 = n2, n2 = pos2->next) {
509 struct __prelim_ref *ref2;
510 struct __prelim_ref *xchg;
511 struct extent_inode_elem *eie;
513 ref2 = list_entry(pos2, struct __prelim_ref, list);
516 if (!ref_for_same_block(ref1, ref2))
518 if (!ref1->parent && ref2->parent) {
524 if (ref1->parent != ref2->parent)
528 eie = ref1->inode_list;
529 while (eie && eie->next)
532 eie->next = ref2->inode_list;
534 ref1->inode_list = ref2->inode_list;
535 ref1->count += ref2->count;
537 list_del(&ref2->list);
545 * add all inline backrefs for bytenr to the list
547 static int __add_inline_refs(struct btrfs_fs_info *fs_info,
548 struct pref_state *prefstate,
549 struct btrfs_path *path, u64 bytenr,
550 int *info_level, u64 *total_refs)
554 struct extent_buffer *leaf;
555 struct btrfs_key key;
556 struct btrfs_key found_key;
559 struct btrfs_extent_item *ei;
563 * enumerate all inline refs
565 leaf = path->nodes[0];
566 slot = path->slots[0];
568 item_size = btrfs_item_size_nr(leaf, slot);
569 BUG_ON(item_size < sizeof(*ei));
571 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
572 flags = btrfs_extent_flags(leaf, ei);
573 *total_refs += btrfs_extent_refs(leaf, ei);
574 btrfs_item_key_to_cpu(leaf, &found_key, slot);
576 ptr = (unsigned long)(ei + 1);
577 end = (unsigned long)ei + item_size;
579 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
580 flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
581 struct btrfs_tree_block_info *info;
583 info = (struct btrfs_tree_block_info *)ptr;
584 *info_level = btrfs_tree_block_level(leaf, info);
585 ptr += sizeof(struct btrfs_tree_block_info);
587 } else if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
588 *info_level = found_key.offset;
590 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
594 struct btrfs_extent_inline_ref *iref;
598 iref = (struct btrfs_extent_inline_ref *)ptr;
599 type = btrfs_extent_inline_ref_type(leaf, iref);
600 offset = btrfs_extent_inline_ref_offset(leaf, iref);
603 case BTRFS_SHARED_BLOCK_REF_KEY:
604 ret = __add_prelim_ref(prefstate, 0, NULL,
605 *info_level + 1, offset,
606 bytenr, 1, GFP_NOFS);
608 case BTRFS_SHARED_DATA_REF_KEY: {
609 struct btrfs_shared_data_ref *sdref;
612 sdref = (struct btrfs_shared_data_ref *)(iref + 1);
613 count = btrfs_shared_data_ref_count(leaf, sdref);
614 ret = __add_prelim_ref(prefstate, 0, NULL, 0, offset,
615 bytenr, count, GFP_NOFS);
618 case BTRFS_TREE_BLOCK_REF_KEY:
619 ret = __add_prelim_ref(prefstate, offset, NULL,
621 bytenr, 1, GFP_NOFS);
623 case BTRFS_EXTENT_DATA_REF_KEY: {
624 struct btrfs_extent_data_ref *dref;
628 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
629 count = btrfs_extent_data_ref_count(leaf, dref);
630 key.objectid = btrfs_extent_data_ref_objectid(leaf,
632 key.type = BTRFS_EXTENT_DATA_KEY;
633 key.offset = btrfs_extent_data_ref_offset(leaf, dref);
634 root = btrfs_extent_data_ref_root(leaf, dref);
635 ret = __add_prelim_ref(prefstate, root, &key, 0, 0,
636 bytenr, count, GFP_NOFS);
644 ptr += btrfs_extent_inline_ref_size(type);
651 * add all non-inline backrefs for bytenr to the list
653 static int __add_keyed_refs(struct btrfs_fs_info *fs_info,
654 struct pref_state *prefstate,
655 struct btrfs_path *path, u64 bytenr,
658 struct btrfs_root *extent_root = fs_info->extent_root;
661 struct extent_buffer *leaf;
662 struct btrfs_key key;
665 ret = btrfs_next_item(extent_root, path);
673 slot = path->slots[0];
674 leaf = path->nodes[0];
675 btrfs_item_key_to_cpu(leaf, &key, slot);
677 if (key.objectid != bytenr)
679 if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
681 if (key.type > BTRFS_SHARED_DATA_REF_KEY)
685 case BTRFS_SHARED_BLOCK_REF_KEY:
686 ret = __add_prelim_ref(prefstate, 0, NULL,
687 info_level + 1, key.offset,
688 bytenr, 1, GFP_NOFS);
690 case BTRFS_SHARED_DATA_REF_KEY: {
691 struct btrfs_shared_data_ref *sdref;
694 sdref = btrfs_item_ptr(leaf, slot,
695 struct btrfs_shared_data_ref);
696 count = btrfs_shared_data_ref_count(leaf, sdref);
697 ret = __add_prelim_ref(prefstate, 0, NULL, 0, key.offset,
698 bytenr, count, GFP_NOFS);
701 case BTRFS_TREE_BLOCK_REF_KEY:
702 ret = __add_prelim_ref(prefstate, key.offset, NULL,
704 bytenr, 1, GFP_NOFS);
706 case BTRFS_EXTENT_DATA_REF_KEY: {
707 struct btrfs_extent_data_ref *dref;
711 dref = btrfs_item_ptr(leaf, slot,
712 struct btrfs_extent_data_ref);
713 count = btrfs_extent_data_ref_count(leaf, dref);
714 key.objectid = btrfs_extent_data_ref_objectid(leaf,
716 key.type = BTRFS_EXTENT_DATA_KEY;
717 key.offset = btrfs_extent_data_ref_offset(leaf, dref);
718 root = btrfs_extent_data_ref_root(leaf, dref);
719 ret = __add_prelim_ref(prefstate, root, &key, 0, 0,
720 bytenr, count, GFP_NOFS);
735 * this adds all existing backrefs (inline backrefs, backrefs and delayed
736 * refs) for the given bytenr to the refs list, merges duplicates and resolves
737 * indirect refs to their parent bytenr.
738 * When roots are found, they're added to the roots list
740 * FIXME some caching might speed things up
742 static int find_parent_nodes(struct btrfs_trans_handle *trans,
743 struct btrfs_fs_info *fs_info, u64 bytenr,
744 u64 time_seq, struct ulist *refs,
745 struct ulist *roots, const u64 *extent_item_pos)
747 struct btrfs_key key;
748 struct btrfs_path *path;
751 struct pref_state prefstate;
752 struct __prelim_ref *ref;
753 struct extent_inode_elem *eie = NULL;
756 init_pref_state(&prefstate);
758 key.objectid = bytenr;
759 key.offset = (u64)-1;
760 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
761 key.type = BTRFS_METADATA_ITEM_KEY;
763 key.type = BTRFS_EXTENT_ITEM_KEY;
765 path = btrfs_alloc_path();
769 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
774 if (path->slots[0]) {
775 struct extent_buffer *leaf;
779 leaf = path->nodes[0];
780 slot = path->slots[0];
781 btrfs_item_key_to_cpu(leaf, &key, slot);
782 if (key.objectid == bytenr &&
783 (key.type == BTRFS_EXTENT_ITEM_KEY ||
784 key.type == BTRFS_METADATA_ITEM_KEY)) {
785 ret = __add_inline_refs(fs_info, &prefstate, path,
790 ret = __add_keyed_refs(fs_info, &prefstate, path,
796 btrfs_release_path(path);
798 ret = __add_missing_keys(fs_info, &prefstate);
802 __merge_refs(&prefstate, 1);
804 ret = __resolve_indirect_refs(fs_info, &prefstate, path, time_seq,
805 extent_item_pos, total_refs);
809 __merge_refs(&prefstate, 2);
811 BUG_ON(!list_empty(&prefstate.pending_missing_keys));
812 BUG_ON(!list_empty(&prefstate.pending_indirect_refs));
814 while (!list_empty(&prefstate.pending)) {
815 ref = list_first_pref(&prefstate.pending);
816 WARN_ON(ref->count < 0);
817 if (roots && ref->count && ref->root_id && ref->parent == 0) {
818 /* no parent == root of tree */
819 ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS);
823 if (ref->count && ref->parent) {
824 if (extent_item_pos && !ref->inode_list &&
826 struct extent_buffer *eb;
828 eb = read_tree_block(fs_info, ref->parent, 0);
829 if (!extent_buffer_uptodate(eb)) {
830 free_extent_buffer(eb);
834 ret = find_extent_in_eb(eb, bytenr,
835 *extent_item_pos, &eie);
836 free_extent_buffer(eb);
839 ref->inode_list = eie;
841 ret = ulist_add_merge_ptr(refs, ref->parent,
843 (void **)&eie, GFP_NOFS);
846 if (!ret && extent_item_pos) {
848 * we've recorded that parent, so we must extend
849 * its inode list here
854 eie->next = ref->inode_list;
858 list_del(&ref->list);
863 btrfs_free_path(path);
864 while (!list_empty(&prefstate.pending)) {
865 ref = list_first_pref(&prefstate.pending);
866 list_del(&ref->list);
870 free_inode_elem_list(eie);
874 static void free_leaf_list(struct ulist *blocks)
876 struct ulist_node *node = NULL;
877 struct extent_inode_elem *eie;
878 struct ulist_iterator uiter;
880 ULIST_ITER_INIT(&uiter);
881 while ((node = ulist_next(blocks, &uiter))) {
884 eie = (struct extent_inode_elem *)(uintptr_t)node->aux;
885 free_inode_elem_list(eie);
893 * Finds all leafs with a reference to the specified combination of bytenr and
894 * offset. key_list_head will point to a list of corresponding keys (caller must
895 * free each list element). The leafs will be stored in the leafs ulist, which
896 * must be freed with ulist_free.
898 * returns 0 on success, <0 on error
900 static int btrfs_find_all_leafs(struct btrfs_trans_handle *trans,
901 struct btrfs_fs_info *fs_info, u64 bytenr,
902 u64 time_seq, struct ulist **leafs,
903 const u64 *extent_item_pos)
907 *leafs = ulist_alloc(GFP_NOFS);
911 ret = find_parent_nodes(trans, fs_info, bytenr,
912 time_seq, *leafs, NULL, extent_item_pos);
913 if (ret < 0 && ret != -ENOENT) {
914 free_leaf_list(*leafs);
922 * walk all backrefs for a given extent to find all roots that reference this
923 * extent. Walking a backref means finding all extents that reference this
924 * extent and in turn walk the backrefs of those, too. Naturally this is a
925 * recursive process, but here it is implemented in an iterative fashion: We
926 * find all referencing extents for the extent in question and put them on a
927 * list. In turn, we find all referencing extents for those, further appending
928 * to the list. The way we iterate the list allows adding more elements after
929 * the current while iterating. The process stops when we reach the end of the
930 * list. Found roots are added to the roots list.
932 * returns 0 on success, < 0 on error.
934 static int __btrfs_find_all_roots(struct btrfs_trans_handle *trans,
935 struct btrfs_fs_info *fs_info, u64 bytenr,
936 u64 time_seq, struct ulist **roots)
939 struct ulist_node *node = NULL;
940 struct ulist_iterator uiter;
943 tmp = ulist_alloc(GFP_NOFS);
946 *roots = ulist_alloc(GFP_NOFS);
952 ULIST_ITER_INIT(&uiter);
954 ret = find_parent_nodes(trans, fs_info, bytenr,
955 time_seq, tmp, *roots, NULL);
956 if (ret < 0 && ret != -ENOENT) {
961 node = ulist_next(tmp, &uiter);
972 int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
973 struct btrfs_fs_info *fs_info, u64 bytenr,
974 u64 time_seq, struct ulist **roots)
976 return __btrfs_find_all_roots(trans, fs_info, bytenr, time_seq, roots);
980 * this makes the path point to (inum INODE_ITEM ioff)
982 int inode_item_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
983 struct btrfs_path *path)
985 struct btrfs_key key;
986 return btrfs_find_item(fs_root, path, inum, ioff,
987 BTRFS_INODE_ITEM_KEY, &key);
990 static int inode_ref_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
991 struct btrfs_path *path,
992 struct btrfs_key *found_key)
994 return btrfs_find_item(fs_root, path, inum, ioff,
995 BTRFS_INODE_REF_KEY, found_key);
998 int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,
999 u64 start_off, struct btrfs_path *path,
1000 struct btrfs_inode_extref **ret_extref,
1004 struct btrfs_key key;
1005 struct btrfs_key found_key;
1006 struct btrfs_inode_extref *extref;
1007 struct extent_buffer *leaf;
1010 key.objectid = inode_objectid;
1011 key.type = BTRFS_INODE_EXTREF_KEY;
1012 key.offset = start_off;
1014 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1019 leaf = path->nodes[0];
1020 slot = path->slots[0];
1021 if (slot >= btrfs_header_nritems(leaf)) {
1023 * If the item at offset is not found,
1024 * btrfs_search_slot will point us to the slot
1025 * where it should be inserted. In our case
1026 * that will be the slot directly before the
1027 * next INODE_REF_KEY_V2 item. In the case
1028 * that we're pointing to the last slot in a
1029 * leaf, we must move one leaf over.
1031 ret = btrfs_next_leaf(root, path);
1040 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1043 * Check that we're still looking at an extended ref key for
1044 * this particular objectid. If we have different
1045 * objectid or type then there are no more to be found
1046 * in the tree and we can exit.
1049 if (found_key.objectid != inode_objectid)
1051 if (found_key.type != BTRFS_INODE_EXTREF_KEY)
1055 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1056 extref = (struct btrfs_inode_extref *)ptr;
1057 *ret_extref = extref;
1059 *found_off = found_key.offset;
1067 * this iterates to turn a name (from iref/extref) into a full filesystem path.
1068 * Elements of the path are separated by '/' and the path is guaranteed to be
1069 * 0-terminated. the path is only given within the current file system.
1070 * Therefore, it never starts with a '/'. the caller is responsible to provide
1071 * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
1072 * the start point of the resulting string is returned. this pointer is within
1074 * in case the path buffer would overflow, the pointer is decremented further
1075 * as if output was written to the buffer, though no more output is actually
1076 * generated. that way, the caller can determine how much space would be
1077 * required for the path to fit into the buffer. in that case, the returned
1078 * value will be smaller than dest. callers must check this!
1080 char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
1081 u32 name_len, unsigned long name_off,
1082 struct extent_buffer *eb_in, u64 parent,
1083 char *dest, u32 size)
1088 s64 bytes_left = ((s64)size) - 1;
1089 struct extent_buffer *eb = eb_in;
1090 struct btrfs_key found_key;
1091 struct btrfs_inode_ref *iref;
1093 if (bytes_left >= 0)
1094 dest[bytes_left] = '\0';
1097 bytes_left -= name_len;
1098 if (bytes_left >= 0)
1099 read_extent_buffer(eb, dest + bytes_left,
1100 name_off, name_len);
1102 free_extent_buffer(eb);
1103 ret = inode_ref_info(parent, 0, fs_root, path, &found_key);
1109 next_inum = found_key.offset;
1111 /* regular exit ahead */
1112 if (parent == next_inum)
1115 slot = path->slots[0];
1116 eb = path->nodes[0];
1117 /* make sure we can use eb after releasing the path */
1120 btrfs_release_path(path);
1121 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1123 name_len = btrfs_inode_ref_name_len(eb, iref);
1124 name_off = (unsigned long)(iref + 1);
1128 if (bytes_left >= 0)
1129 dest[bytes_left] = '/';
1132 btrfs_release_path(path);
1135 return ERR_PTR(ret);
1137 return dest + bytes_left;
1141 * this makes the path point to (logical EXTENT_ITEM *)
1142 * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
1143 * tree blocks and <0 on error.
1145 int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
1146 struct btrfs_path *path, struct btrfs_key *found_key,
1153 struct extent_buffer *eb;
1154 struct btrfs_extent_item *ei;
1155 struct btrfs_key key;
1157 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1158 key.type = BTRFS_METADATA_ITEM_KEY;
1160 key.type = BTRFS_EXTENT_ITEM_KEY;
1161 key.objectid = logical;
1162 key.offset = (u64)-1;
1164 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
1168 ret = btrfs_previous_extent_item(fs_info->extent_root, path, 0);
1174 btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]);
1175 if (found_key->type == BTRFS_METADATA_ITEM_KEY)
1176 size = fs_info->nodesize;
1177 else if (found_key->type == BTRFS_EXTENT_ITEM_KEY)
1178 size = found_key->offset;
1180 if (found_key->objectid > logical ||
1181 found_key->objectid + size <= logical) {
1182 pr_debug("logical %llu is not within any extent\n", logical);
1186 eb = path->nodes[0];
1187 item_size = btrfs_item_size_nr(eb, path->slots[0]);
1188 BUG_ON(item_size < sizeof(*ei));
1190 ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
1191 flags = btrfs_extent_flags(eb, ei);
1193 pr_debug("logical %llu is at position %llu within the extent (%llu "
1194 "EXTENT_ITEM %llu) flags %#llx size %u\n",
1195 logical, logical - found_key->objectid, found_key->objectid,
1196 found_key->offset, flags, item_size);
1199 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1200 *flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK;
1201 else if (flags & BTRFS_EXTENT_FLAG_DATA)
1202 *flags_ret = BTRFS_EXTENT_FLAG_DATA;
1213 * helper function to iterate extent inline refs. ptr must point to a 0 value
1214 * for the first call and may be modified. it is used to track state.
1215 * if more refs exist, 0 is returned and the next call to
1216 * __get_extent_inline_ref must pass the modified ptr parameter to get the
1217 * next ref. after the last ref was processed, 1 is returned.
1218 * returns <0 on error
1220 static int __get_extent_inline_ref(unsigned long *ptr, struct extent_buffer *eb,
1221 struct btrfs_key *key,
1222 struct btrfs_extent_item *ei, u32 item_size,
1223 struct btrfs_extent_inline_ref **out_eiref,
1228 struct btrfs_tree_block_info *info;
1232 flags = btrfs_extent_flags(eb, ei);
1233 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1234 if (key->type == BTRFS_METADATA_ITEM_KEY) {
1235 /* a skinny metadata extent */
1237 (struct btrfs_extent_inline_ref *)(ei + 1);
1239 WARN_ON(key->type != BTRFS_EXTENT_ITEM_KEY);
1240 info = (struct btrfs_tree_block_info *)(ei + 1);
1242 (struct btrfs_extent_inline_ref *)(info + 1);
1245 *out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1);
1247 *ptr = (unsigned long)*out_eiref;
1248 if ((unsigned long)(*ptr) >= (unsigned long)ei + item_size)
1252 end = (unsigned long)ei + item_size;
1253 *out_eiref = (struct btrfs_extent_inline_ref *)(*ptr);
1254 *out_type = btrfs_extent_inline_ref_type(eb, *out_eiref);
1256 *ptr += btrfs_extent_inline_ref_size(*out_type);
1257 WARN_ON(*ptr > end);
1259 return 1; /* last */
1265 * reads the tree block backref for an extent. tree level and root are returned
1266 * through out_level and out_root. ptr must point to a 0 value for the first
1267 * call and may be modified (see __get_extent_inline_ref comment).
1268 * returns 0 if data was provided, 1 if there was no more data to provide or
1271 int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
1272 struct btrfs_key *key, struct btrfs_extent_item *ei,
1273 u32 item_size, u64 *out_root, u8 *out_level)
1277 struct btrfs_tree_block_info *info;
1278 struct btrfs_extent_inline_ref *eiref;
1280 if (*ptr == (unsigned long)-1)
1284 ret = __get_extent_inline_ref(ptr, eb, key, ei, item_size,
1289 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1290 type == BTRFS_SHARED_BLOCK_REF_KEY)
1297 /* we can treat both ref types equally here */
1298 info = (struct btrfs_tree_block_info *)(ei + 1);
1299 *out_root = btrfs_extent_inline_ref_offset(eb, eiref);
1300 *out_level = btrfs_tree_block_level(eb, info);
1303 *ptr = (unsigned long)-1;
1308 static int iterate_leaf_refs(struct extent_inode_elem *inode_list,
1309 u64 root, u64 extent_item_objectid,
1310 iterate_extent_inodes_t *iterate, void *ctx)
1312 struct extent_inode_elem *eie;
1315 for (eie = inode_list; eie; eie = eie->next) {
1316 pr_debug("ref for %llu resolved, key (%llu EXTEND_DATA %llu), "
1317 "root %llu\n", extent_item_objectid,
1318 eie->inum, eie->offset, root);
1319 ret = iterate(eie->inum, eie->offset, root, ctx);
1321 pr_debug("stopping iteration for %llu due to ret=%d\n",
1322 extent_item_objectid, ret);
1331 * calls iterate() for every inode that references the extent identified by
1332 * the given parameters.
1333 * when the iterator function returns a non-zero value, iteration stops.
1335 int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
1336 u64 extent_item_objectid, u64 extent_item_pos,
1337 int search_commit_root,
1338 iterate_extent_inodes_t *iterate, void *ctx)
1341 struct btrfs_trans_handle *trans = NULL;
1342 struct ulist *refs = NULL;
1343 struct ulist *roots = NULL;
1344 struct ulist_node *ref_node = NULL;
1345 struct ulist_node *root_node = NULL;
1346 struct ulist_iterator ref_uiter;
1347 struct ulist_iterator root_uiter;
1349 pr_debug("resolving all inodes for extent %llu\n",
1350 extent_item_objectid);
1352 ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid,
1353 0, &refs, &extent_item_pos);
1357 ULIST_ITER_INIT(&ref_uiter);
1358 while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) {
1359 ret = __btrfs_find_all_roots(trans, fs_info, ref_node->val,
1363 ULIST_ITER_INIT(&root_uiter);
1364 while (!ret && (root_node = ulist_next(roots, &root_uiter))) {
1365 pr_debug("root %llu references leaf %llu, data list "
1366 "%#llx\n", root_node->val, ref_node->val,
1368 ret = iterate_leaf_refs((struct extent_inode_elem *)
1369 (uintptr_t)ref_node->aux,
1371 extent_item_objectid,
1377 free_leaf_list(refs);
1382 int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
1383 struct btrfs_path *path,
1384 iterate_extent_inodes_t *iterate, void *ctx)
1387 u64 extent_item_pos;
1389 struct btrfs_key found_key;
1390 int search_commit_root = 0;
1392 ret = extent_from_logical(fs_info, logical, path, &found_key, &flags);
1393 btrfs_release_path(path);
1396 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1399 extent_item_pos = logical - found_key.objectid;
1400 ret = iterate_extent_inodes(fs_info, found_key.objectid,
1401 extent_item_pos, search_commit_root,
1407 typedef int (iterate_irefs_t)(u64 parent, u32 name_len, unsigned long name_off,
1408 struct extent_buffer *eb, void *ctx);
1410 static int iterate_inode_refs(u64 inum, struct btrfs_root *fs_root,
1411 struct btrfs_path *path,
1412 iterate_irefs_t *iterate, void *ctx)
1421 struct extent_buffer *eb;
1422 struct btrfs_item *item;
1423 struct btrfs_inode_ref *iref;
1424 struct btrfs_key found_key;
1427 ret = inode_ref_info(inum, parent ? parent+1 : 0, fs_root, path,
1432 ret = found ? 0 : -ENOENT;
1437 parent = found_key.offset;
1438 slot = path->slots[0];
1439 eb = btrfs_clone_extent_buffer(path->nodes[0]);
1444 extent_buffer_get(eb);
1445 btrfs_release_path(path);
1447 item = btrfs_item_nr(slot);
1448 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1450 for (cur = 0; cur < btrfs_item_size(eb, item); cur += len) {
1451 name_len = btrfs_inode_ref_name_len(eb, iref);
1452 /* path must be released before calling iterate()! */
1453 pr_debug("following ref at offset %u for inode %llu in "
1454 "tree %llu\n", cur, found_key.objectid,
1456 ret = iterate(parent, name_len,
1457 (unsigned long)(iref + 1), eb, ctx);
1460 len = sizeof(*iref) + name_len;
1461 iref = (struct btrfs_inode_ref *)((char *)iref + len);
1463 free_extent_buffer(eb);
1466 btrfs_release_path(path);
1471 static int iterate_inode_extrefs(u64 inum, struct btrfs_root *fs_root,
1472 struct btrfs_path *path,
1473 iterate_irefs_t *iterate, void *ctx)
1480 struct extent_buffer *eb;
1481 struct btrfs_inode_extref *extref;
1482 struct extent_buffer *leaf;
1488 ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref,
1493 ret = found ? 0 : -ENOENT;
1498 slot = path->slots[0];
1499 eb = btrfs_clone_extent_buffer(path->nodes[0]);
1504 extent_buffer_get(eb);
1506 btrfs_release_path(path);
1508 leaf = path->nodes[0];
1509 item_size = btrfs_item_size_nr(leaf, slot);
1510 ptr = btrfs_item_ptr_offset(leaf, slot);
1513 while (cur_offset < item_size) {
1516 extref = (struct btrfs_inode_extref *)(ptr + cur_offset);
1517 parent = btrfs_inode_extref_parent(eb, extref);
1518 name_len = btrfs_inode_extref_name_len(eb, extref);
1519 ret = iterate(parent, name_len,
1520 (unsigned long)&extref->name, eb, ctx);
1524 cur_offset += btrfs_inode_extref_name_len(leaf, extref);
1525 cur_offset += sizeof(*extref);
1527 free_extent_buffer(eb);
1532 btrfs_release_path(path);
1537 static int iterate_irefs(u64 inum, struct btrfs_root *fs_root,
1538 struct btrfs_path *path, iterate_irefs_t *iterate,
1544 ret = iterate_inode_refs(inum, fs_root, path, iterate, ctx);
1547 else if (ret != -ENOENT)
1550 ret = iterate_inode_extrefs(inum, fs_root, path, iterate, ctx);
1551 if (ret == -ENOENT && found_refs)
1558 * returns 0 if the path could be dumped (probably truncated)
1559 * returns <0 in case of an error
1561 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
1562 struct extent_buffer *eb, void *ctx)
1564 struct inode_fs_paths *ipath = ctx;
1567 int i = ipath->fspath->elem_cnt;
1568 const int s_ptr = sizeof(char *);
1571 bytes_left = ipath->fspath->bytes_left > s_ptr ?
1572 ipath->fspath->bytes_left - s_ptr : 0;
1574 fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
1575 fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len,
1576 name_off, eb, inum, fspath_min, bytes_left);
1578 return PTR_ERR(fspath);
1580 if (fspath > fspath_min) {
1581 ipath->fspath->val[i] = (u64)(unsigned long)fspath;
1582 ++ipath->fspath->elem_cnt;
1583 ipath->fspath->bytes_left = fspath - fspath_min;
1585 ++ipath->fspath->elem_missed;
1586 ipath->fspath->bytes_missing += fspath_min - fspath;
1587 ipath->fspath->bytes_left = 0;
1594 * this dumps all file system paths to the inode into the ipath struct, provided
1595 * is has been created large enough. each path is zero-terminated and accessed
1596 * from ipath->fspath->val[i].
1597 * when it returns, there are ipath->fspath->elem_cnt number of paths available
1598 * in ipath->fspath->val[]. When the allocated space wasn't sufficient, the
1599 * number of missed paths is recorded in ipath->fspath->elem_missed, otherwise,
1600 * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
1601 * have been needed to return all paths.
1603 int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
1605 return iterate_irefs(inum, ipath->fs_root, ipath->btrfs_path,
1606 inode_to_path, ipath);
1609 struct btrfs_data_container *init_data_container(u32 total_bytes)
1611 struct btrfs_data_container *data;
1614 alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
1615 data = vmalloc(alloc_bytes);
1617 return ERR_PTR(-ENOMEM);
1619 if (total_bytes >= sizeof(*data)) {
1620 data->bytes_left = total_bytes - sizeof(*data);
1621 data->bytes_missing = 0;
1623 data->bytes_missing = sizeof(*data) - total_bytes;
1624 data->bytes_left = 0;
1628 data->elem_missed = 0;
1634 * allocates space to return multiple file system paths for an inode.
1635 * total_bytes to allocate are passed, note that space usable for actual path
1636 * information will be total_bytes - sizeof(struct inode_fs_paths).
1637 * the returned pointer must be freed with free_ipath() in the end.
1639 struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
1640 struct btrfs_path *path)
1642 struct inode_fs_paths *ifp;
1643 struct btrfs_data_container *fspath;
1645 fspath = init_data_container(total_bytes);
1647 return (void *)fspath;
1649 ifp = kmalloc(sizeof(*ifp), GFP_NOFS);
1652 return ERR_PTR(-ENOMEM);
1655 ifp->btrfs_path = path;
1656 ifp->fspath = fspath;
1657 ifp->fs_root = fs_root;
1662 void free_ipath(struct inode_fs_paths *ipath)
1666 vfree(ipath->fspath);