1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2011 STRATO. All rights reserved.
7 #include <linux/rbtree.h>
8 #include <trace/events/btrfs.h>
13 #include "transaction.h"
14 #include "delayed-ref.h"
17 /* Just an arbitrary number so we can be sure this happened */
18 #define BACKREF_FOUND_SHARED 6
20 struct extent_inode_elem {
23 struct extent_inode_elem *next;
26 static int check_extent_in_eb(const struct btrfs_key *key,
27 const struct extent_buffer *eb,
28 const struct btrfs_file_extent_item *fi,
30 struct extent_inode_elem **eie,
34 struct extent_inode_elem *e;
37 !btrfs_file_extent_compression(eb, fi) &&
38 !btrfs_file_extent_encryption(eb, fi) &&
39 !btrfs_file_extent_other_encoding(eb, fi)) {
43 data_offset = btrfs_file_extent_offset(eb, fi);
44 data_len = btrfs_file_extent_num_bytes(eb, fi);
46 if (extent_item_pos < data_offset ||
47 extent_item_pos >= data_offset + data_len)
49 offset = extent_item_pos - data_offset;
52 e = kmalloc(sizeof(*e), GFP_NOFS);
57 e->inum = key->objectid;
58 e->offset = key->offset + offset;
64 static void free_inode_elem_list(struct extent_inode_elem *eie)
66 struct extent_inode_elem *eie_next;
68 for (; eie; eie = eie_next) {
74 static int find_extent_in_eb(const struct extent_buffer *eb,
75 u64 wanted_disk_byte, u64 extent_item_pos,
76 struct extent_inode_elem **eie,
81 struct btrfs_file_extent_item *fi;
88 * from the shared data ref, we only have the leaf but we need
89 * the key. thus, we must look into all items and see that we
90 * find one (some) with a reference to our extent item.
92 nritems = btrfs_header_nritems(eb);
93 for (slot = 0; slot < nritems; ++slot) {
94 btrfs_item_key_to_cpu(eb, &key, slot);
95 if (key.type != BTRFS_EXTENT_DATA_KEY)
97 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
98 extent_type = btrfs_file_extent_type(eb, fi);
99 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
101 /* don't skip BTRFS_FILE_EXTENT_PREALLOC, we can handle that */
102 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
103 if (disk_byte != wanted_disk_byte)
106 ret = check_extent_in_eb(&key, eb, fi, extent_item_pos, eie, ignore_offset);
115 struct rb_root_cached root;
119 #define PREFTREE_INIT { .root = RB_ROOT_CACHED, .count = 0 }
122 struct preftree direct; /* BTRFS_SHARED_[DATA|BLOCK]_REF_KEY */
123 struct preftree indirect; /* BTRFS_[TREE_BLOCK|EXTENT_DATA]_REF_KEY */
124 struct preftree indirect_missing_keys;
128 * Checks for a shared extent during backref search.
130 * The share_count tracks prelim_refs (direct and indirect) having a
132 * - incremented when a ref->count transitions to >0
133 * - decremented when a ref->count transitions to <1
141 static inline int extent_is_shared(struct share_check *sc)
143 return (sc && sc->share_count > 1) ? BACKREF_FOUND_SHARED : 0;
146 static struct kmem_cache *btrfs_prelim_ref_cache;
148 int __init btrfs_prelim_ref_init(void)
150 btrfs_prelim_ref_cache = kmem_cache_create("btrfs_prelim_ref",
151 sizeof(struct prelim_ref),
155 if (!btrfs_prelim_ref_cache)
160 void __cold btrfs_prelim_ref_exit(void)
162 kmem_cache_destroy(btrfs_prelim_ref_cache);
165 static void free_pref(struct prelim_ref *ref)
167 kmem_cache_free(btrfs_prelim_ref_cache, ref);
171 * Return 0 when both refs are for the same block (and can be merged).
172 * A -1 return indicates ref1 is a 'lower' block than ref2, while 1
173 * indicates a 'higher' block.
175 static int prelim_ref_compare(struct prelim_ref *ref1,
176 struct prelim_ref *ref2)
178 if (ref1->level < ref2->level)
180 if (ref1->level > ref2->level)
182 if (ref1->root_id < ref2->root_id)
184 if (ref1->root_id > ref2->root_id)
186 if (ref1->key_for_search.type < ref2->key_for_search.type)
188 if (ref1->key_for_search.type > ref2->key_for_search.type)
190 if (ref1->key_for_search.objectid < ref2->key_for_search.objectid)
192 if (ref1->key_for_search.objectid > ref2->key_for_search.objectid)
194 if (ref1->key_for_search.offset < ref2->key_for_search.offset)
196 if (ref1->key_for_search.offset > ref2->key_for_search.offset)
198 if (ref1->parent < ref2->parent)
200 if (ref1->parent > ref2->parent)
206 static void update_share_count(struct share_check *sc, int oldcount,
209 if ((!sc) || (oldcount == 0 && newcount < 1))
212 if (oldcount > 0 && newcount < 1)
214 else if (oldcount < 1 && newcount > 0)
219 * Add @newref to the @root rbtree, merging identical refs.
221 * Callers should assume that newref has been freed after calling.
223 static void prelim_ref_insert(const struct btrfs_fs_info *fs_info,
224 struct preftree *preftree,
225 struct prelim_ref *newref,
226 struct share_check *sc)
228 struct rb_root_cached *root;
230 struct rb_node *parent = NULL;
231 struct prelim_ref *ref;
233 bool leftmost = true;
235 root = &preftree->root;
236 p = &root->rb_root.rb_node;
240 ref = rb_entry(parent, struct prelim_ref, rbnode);
241 result = prelim_ref_compare(ref, newref);
244 } else if (result > 0) {
248 /* Identical refs, merge them and free @newref */
249 struct extent_inode_elem *eie = ref->inode_list;
251 while (eie && eie->next)
255 ref->inode_list = newref->inode_list;
257 eie->next = newref->inode_list;
258 trace_btrfs_prelim_ref_merge(fs_info, ref, newref,
261 * A delayed ref can have newref->count < 0.
262 * The ref->count is updated to follow any
263 * BTRFS_[ADD|DROP]_DELAYED_REF actions.
265 update_share_count(sc, ref->count,
266 ref->count + newref->count);
267 ref->count += newref->count;
273 update_share_count(sc, 0, newref->count);
275 trace_btrfs_prelim_ref_insert(fs_info, newref, NULL, preftree->count);
276 rb_link_node(&newref->rbnode, parent, p);
277 rb_insert_color_cached(&newref->rbnode, root, leftmost);
281 * Release the entire tree. We don't care about internal consistency so
282 * just free everything and then reset the tree root.
284 static void prelim_release(struct preftree *preftree)
286 struct prelim_ref *ref, *next_ref;
288 rbtree_postorder_for_each_entry_safe(ref, next_ref,
289 &preftree->root.rb_root, rbnode)
292 preftree->root = RB_ROOT_CACHED;
297 * the rules for all callers of this function are:
298 * - obtaining the parent is the goal
299 * - if you add a key, you must know that it is a correct key
300 * - if you cannot add the parent or a correct key, then we will look into the
301 * block later to set a correct key
305 * backref type | shared | indirect | shared | indirect
306 * information | tree | tree | data | data
307 * --------------------+--------+----------+--------+----------
308 * parent logical | y | - | - | -
309 * key to resolve | - | y | y | y
310 * tree block logical | - | - | - | -
311 * root for resolving | y | y | y | y
313 * - column 1: we've the parent -> done
314 * - column 2, 3, 4: we use the key to find the parent
316 * on disk refs (inline or keyed)
317 * ==============================
318 * backref type | shared | indirect | shared | indirect
319 * information | tree | tree | data | data
320 * --------------------+--------+----------+--------+----------
321 * parent logical | y | - | y | -
322 * key to resolve | - | - | - | y
323 * tree block logical | y | y | y | y
324 * root for resolving | - | y | y | y
326 * - column 1, 3: we've the parent -> done
327 * - column 2: we take the first key from the block to find the parent
328 * (see add_missing_keys)
329 * - column 4: we use the key to find the parent
331 * additional information that's available but not required to find the parent
332 * block might help in merging entries to gain some speed.
334 static int add_prelim_ref(const struct btrfs_fs_info *fs_info,
335 struct preftree *preftree, u64 root_id,
336 const struct btrfs_key *key, int level, u64 parent,
337 u64 wanted_disk_byte, int count,
338 struct share_check *sc, gfp_t gfp_mask)
340 struct prelim_ref *ref;
342 if (root_id == BTRFS_DATA_RELOC_TREE_OBJECTID)
345 ref = kmem_cache_alloc(btrfs_prelim_ref_cache, gfp_mask);
349 ref->root_id = root_id;
351 ref->key_for_search = *key;
353 memset(&ref->key_for_search, 0, sizeof(ref->key_for_search));
355 ref->inode_list = NULL;
358 ref->parent = parent;
359 ref->wanted_disk_byte = wanted_disk_byte;
360 prelim_ref_insert(fs_info, preftree, ref, sc);
361 return extent_is_shared(sc);
364 /* direct refs use root == 0, key == NULL */
365 static int add_direct_ref(const struct btrfs_fs_info *fs_info,
366 struct preftrees *preftrees, int level, u64 parent,
367 u64 wanted_disk_byte, int count,
368 struct share_check *sc, gfp_t gfp_mask)
370 return add_prelim_ref(fs_info, &preftrees->direct, 0, NULL, level,
371 parent, wanted_disk_byte, count, sc, gfp_mask);
374 /* indirect refs use parent == 0 */
375 static int add_indirect_ref(const struct btrfs_fs_info *fs_info,
376 struct preftrees *preftrees, u64 root_id,
377 const struct btrfs_key *key, int level,
378 u64 wanted_disk_byte, int count,
379 struct share_check *sc, gfp_t gfp_mask)
381 struct preftree *tree = &preftrees->indirect;
384 tree = &preftrees->indirect_missing_keys;
385 return add_prelim_ref(fs_info, tree, root_id, key, level, 0,
386 wanted_disk_byte, count, sc, gfp_mask);
389 static int is_shared_data_backref(struct preftrees *preftrees, u64 bytenr)
391 struct rb_node **p = &preftrees->direct.root.rb_root.rb_node;
392 struct rb_node *parent = NULL;
393 struct prelim_ref *ref = NULL;
394 struct prelim_ref target = {0};
397 target.parent = bytenr;
401 ref = rb_entry(parent, struct prelim_ref, rbnode);
402 result = prelim_ref_compare(ref, &target);
414 static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path,
415 struct ulist *parents,
416 struct preftrees *preftrees, struct prelim_ref *ref,
417 int level, u64 time_seq, const u64 *extent_item_pos,
422 struct extent_buffer *eb;
423 struct btrfs_key key;
424 struct btrfs_key *key_for_search = &ref->key_for_search;
425 struct btrfs_file_extent_item *fi;
426 struct extent_inode_elem *eie = NULL, *old = NULL;
428 u64 wanted_disk_byte = ref->wanted_disk_byte;
433 eb = path->nodes[level];
434 ret = ulist_add(parents, eb->start, 0, GFP_NOFS);
441 * 1. We normally enter this function with the path already pointing to
442 * the first item to check. But sometimes, we may enter it with
444 * 2. We are searching for normal backref but bytenr of this leaf
445 * matches shared data backref
446 * 3. The leaf owner is not equal to the root we are searching
448 * For these cases, go to the next leaf before we continue.
451 if (path->slots[0] >= btrfs_header_nritems(eb) ||
452 is_shared_data_backref(preftrees, eb->start) ||
453 ref->root_id != btrfs_header_owner(eb)) {
454 if (time_seq == SEQ_LAST)
455 ret = btrfs_next_leaf(root, path);
457 ret = btrfs_next_old_leaf(root, path, time_seq);
460 while (!ret && count < ref->count) {
462 slot = path->slots[0];
464 btrfs_item_key_to_cpu(eb, &key, slot);
466 if (key.objectid != key_for_search->objectid ||
467 key.type != BTRFS_EXTENT_DATA_KEY)
471 * We are searching for normal backref but bytenr of this leaf
472 * matches shared data backref, OR
473 * the leaf owner is not equal to the root we are searching for
476 (is_shared_data_backref(preftrees, eb->start) ||
477 ref->root_id != btrfs_header_owner(eb))) {
478 if (time_seq == SEQ_LAST)
479 ret = btrfs_next_leaf(root, path);
481 ret = btrfs_next_old_leaf(root, path, time_seq);
484 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
485 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
486 data_offset = btrfs_file_extent_offset(eb, fi);
488 if (disk_byte == wanted_disk_byte) {
491 if (ref->key_for_search.offset == key.offset - data_offset)
495 if (extent_item_pos) {
496 ret = check_extent_in_eb(&key, eb, fi,
498 &eie, ignore_offset);
504 ret = ulist_add_merge_ptr(parents, eb->start,
505 eie, (void **)&old, GFP_NOFS);
508 if (!ret && extent_item_pos) {
516 if (time_seq == SEQ_LAST)
517 ret = btrfs_next_item(root, path);
519 ret = btrfs_next_old_item(root, path, time_seq);
525 free_inode_elem_list(eie);
530 * resolve an indirect backref in the form (root_id, key, level)
531 * to a logical address
533 static int resolve_indirect_ref(struct btrfs_fs_info *fs_info,
534 struct btrfs_path *path, u64 time_seq,
535 struct preftrees *preftrees,
536 struct prelim_ref *ref, struct ulist *parents,
537 const u64 *extent_item_pos, bool ignore_offset)
539 struct btrfs_root *root;
540 struct btrfs_key root_key;
541 struct extent_buffer *eb;
544 int level = ref->level;
545 struct btrfs_key search_key = ref->key_for_search;
547 root_key.objectid = ref->root_id;
548 root_key.type = BTRFS_ROOT_ITEM_KEY;
549 root_key.offset = (u64)-1;
551 root = btrfs_get_fs_root(fs_info, &root_key, false);
557 if (btrfs_is_testing(fs_info)) {
562 if (path->search_commit_root)
563 root_level = btrfs_header_level(root->commit_root);
564 else if (time_seq == SEQ_LAST)
565 root_level = btrfs_header_level(root->node);
567 root_level = btrfs_old_root_level(root, time_seq);
569 if (root_level + 1 == level)
573 * We can often find data backrefs with an offset that is too large
574 * (>= LLONG_MAX, maximum allowed file offset) due to underflows when
575 * subtracting a file's offset with the data offset of its
576 * corresponding extent data item. This can happen for example in the
579 * So if we detect such case we set the search key's offset to zero to
580 * make sure we will find the matching file extent item at
581 * add_all_parents(), otherwise we will miss it because the offset
582 * taken form the backref is much larger then the offset of the file
583 * extent item. This can make us scan a very large number of file
584 * extent items, but at least it will not make us miss any.
586 * This is an ugly workaround for a behaviour that should have never
587 * existed, but it does and a fix for the clone ioctl would touch a lot
588 * of places, cause backwards incompatibility and would not fix the
589 * problem for extents cloned with older kernels.
591 if (search_key.type == BTRFS_EXTENT_DATA_KEY &&
592 search_key.offset >= LLONG_MAX)
593 search_key.offset = 0;
594 path->lowest_level = level;
595 if (time_seq == SEQ_LAST)
596 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
598 ret = btrfs_search_old_slot(root, &search_key, path, time_seq);
601 "search slot in root %llu (level %d, ref count %d) returned %d for key (%llu %u %llu)",
602 ref->root_id, level, ref->count, ret,
603 ref->key_for_search.objectid, ref->key_for_search.type,
604 ref->key_for_search.offset);
608 eb = path->nodes[level];
610 if (WARN_ON(!level)) {
615 eb = path->nodes[level];
618 ret = add_all_parents(root, path, parents, preftrees, ref, level,
619 time_seq, extent_item_pos, ignore_offset);
621 btrfs_put_root(root);
623 path->lowest_level = 0;
624 btrfs_release_path(path);
628 static struct extent_inode_elem *
629 unode_aux_to_inode_list(struct ulist_node *node)
633 return (struct extent_inode_elem *)(uintptr_t)node->aux;
637 * We maintain three separate rbtrees: one for direct refs, one for
638 * indirect refs which have a key, and one for indirect refs which do not
639 * have a key. Each tree does merge on insertion.
641 * Once all of the references are located, we iterate over the tree of
642 * indirect refs with missing keys. An appropriate key is located and
643 * the ref is moved onto the tree for indirect refs. After all missing
644 * keys are thus located, we iterate over the indirect ref tree, resolve
645 * each reference, and then insert the resolved reference onto the
646 * direct tree (merging there too).
648 * New backrefs (i.e., for parent nodes) are added to the appropriate
649 * rbtree as they are encountered. The new backrefs are subsequently
652 static int resolve_indirect_refs(struct btrfs_fs_info *fs_info,
653 struct btrfs_path *path, u64 time_seq,
654 struct preftrees *preftrees,
655 const u64 *extent_item_pos,
656 struct share_check *sc, bool ignore_offset)
660 struct ulist *parents;
661 struct ulist_node *node;
662 struct ulist_iterator uiter;
663 struct rb_node *rnode;
665 parents = ulist_alloc(GFP_NOFS);
670 * We could trade memory usage for performance here by iterating
671 * the tree, allocating new refs for each insertion, and then
672 * freeing the entire indirect tree when we're done. In some test
673 * cases, the tree can grow quite large (~200k objects).
675 while ((rnode = rb_first_cached(&preftrees->indirect.root))) {
676 struct prelim_ref *ref;
678 ref = rb_entry(rnode, struct prelim_ref, rbnode);
679 if (WARN(ref->parent,
680 "BUG: direct ref found in indirect tree")) {
685 rb_erase_cached(&ref->rbnode, &preftrees->indirect.root);
686 preftrees->indirect.count--;
688 if (ref->count == 0) {
693 if (sc && sc->root_objectid &&
694 ref->root_id != sc->root_objectid) {
696 ret = BACKREF_FOUND_SHARED;
699 err = resolve_indirect_ref(fs_info, path, time_seq, preftrees,
700 ref, parents, extent_item_pos,
703 * we can only tolerate ENOENT,otherwise,we should catch error
704 * and return directly.
706 if (err == -ENOENT) {
707 prelim_ref_insert(fs_info, &preftrees->direct, ref,
716 /* we put the first parent into the ref at hand */
717 ULIST_ITER_INIT(&uiter);
718 node = ulist_next(parents, &uiter);
719 ref->parent = node ? node->val : 0;
720 ref->inode_list = unode_aux_to_inode_list(node);
722 /* Add a prelim_ref(s) for any other parent(s). */
723 while ((node = ulist_next(parents, &uiter))) {
724 struct prelim_ref *new_ref;
726 new_ref = kmem_cache_alloc(btrfs_prelim_ref_cache,
733 memcpy(new_ref, ref, sizeof(*ref));
734 new_ref->parent = node->val;
735 new_ref->inode_list = unode_aux_to_inode_list(node);
736 prelim_ref_insert(fs_info, &preftrees->direct,
741 * Now it's a direct ref, put it in the direct tree. We must
742 * do this last because the ref could be merged/freed here.
744 prelim_ref_insert(fs_info, &preftrees->direct, ref, NULL);
746 ulist_reinit(parents);
755 * read tree blocks and add keys where required.
757 static int add_missing_keys(struct btrfs_fs_info *fs_info,
758 struct preftrees *preftrees, bool lock)
760 struct prelim_ref *ref;
761 struct extent_buffer *eb;
762 struct preftree *tree = &preftrees->indirect_missing_keys;
763 struct rb_node *node;
765 while ((node = rb_first_cached(&tree->root))) {
766 ref = rb_entry(node, struct prelim_ref, rbnode);
767 rb_erase_cached(node, &tree->root);
769 BUG_ON(ref->parent); /* should not be a direct ref */
770 BUG_ON(ref->key_for_search.type);
771 BUG_ON(!ref->wanted_disk_byte);
773 eb = read_tree_block(fs_info, ref->wanted_disk_byte, 0,
774 ref->level - 1, NULL);
778 } else if (!extent_buffer_uptodate(eb)) {
780 free_extent_buffer(eb);
784 btrfs_tree_read_lock(eb);
785 if (btrfs_header_level(eb) == 0)
786 btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
788 btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
790 btrfs_tree_read_unlock(eb);
791 free_extent_buffer(eb);
792 prelim_ref_insert(fs_info, &preftrees->indirect, ref, NULL);
799 * add all currently queued delayed refs from this head whose seq nr is
800 * smaller or equal that seq to the list
802 static int add_delayed_refs(const struct btrfs_fs_info *fs_info,
803 struct btrfs_delayed_ref_head *head, u64 seq,
804 struct preftrees *preftrees, struct share_check *sc)
806 struct btrfs_delayed_ref_node *node;
807 struct btrfs_delayed_extent_op *extent_op = head->extent_op;
808 struct btrfs_key key;
809 struct btrfs_key tmp_op_key;
814 if (extent_op && extent_op->update_key)
815 btrfs_disk_key_to_cpu(&tmp_op_key, &extent_op->key);
817 spin_lock(&head->lock);
818 for (n = rb_first_cached(&head->ref_tree); n; n = rb_next(n)) {
819 node = rb_entry(n, struct btrfs_delayed_ref_node,
824 switch (node->action) {
825 case BTRFS_ADD_DELAYED_EXTENT:
826 case BTRFS_UPDATE_DELAYED_HEAD:
829 case BTRFS_ADD_DELAYED_REF:
830 count = node->ref_mod;
832 case BTRFS_DROP_DELAYED_REF:
833 count = node->ref_mod * -1;
838 switch (node->type) {
839 case BTRFS_TREE_BLOCK_REF_KEY: {
840 /* NORMAL INDIRECT METADATA backref */
841 struct btrfs_delayed_tree_ref *ref;
843 ref = btrfs_delayed_node_to_tree_ref(node);
844 ret = add_indirect_ref(fs_info, preftrees, ref->root,
845 &tmp_op_key, ref->level + 1,
846 node->bytenr, count, sc,
850 case BTRFS_SHARED_BLOCK_REF_KEY: {
851 /* SHARED DIRECT METADATA backref */
852 struct btrfs_delayed_tree_ref *ref;
854 ref = btrfs_delayed_node_to_tree_ref(node);
856 ret = add_direct_ref(fs_info, preftrees, ref->level + 1,
857 ref->parent, node->bytenr, count,
861 case BTRFS_EXTENT_DATA_REF_KEY: {
862 /* NORMAL INDIRECT DATA backref */
863 struct btrfs_delayed_data_ref *ref;
864 ref = btrfs_delayed_node_to_data_ref(node);
866 key.objectid = ref->objectid;
867 key.type = BTRFS_EXTENT_DATA_KEY;
868 key.offset = ref->offset;
871 * Found a inum that doesn't match our known inum, we
874 if (sc && sc->inum && ref->objectid != sc->inum) {
875 ret = BACKREF_FOUND_SHARED;
879 ret = add_indirect_ref(fs_info, preftrees, ref->root,
880 &key, 0, node->bytenr, count, sc,
884 case BTRFS_SHARED_DATA_REF_KEY: {
885 /* SHARED DIRECT FULL backref */
886 struct btrfs_delayed_data_ref *ref;
888 ref = btrfs_delayed_node_to_data_ref(node);
890 ret = add_direct_ref(fs_info, preftrees, 0, ref->parent,
891 node->bytenr, count, sc,
899 * We must ignore BACKREF_FOUND_SHARED until all delayed
900 * refs have been checked.
902 if (ret && (ret != BACKREF_FOUND_SHARED))
906 ret = extent_is_shared(sc);
908 spin_unlock(&head->lock);
913 * add all inline backrefs for bytenr to the list
915 * Returns 0 on success, <0 on error, or BACKREF_FOUND_SHARED.
917 static int add_inline_refs(const struct btrfs_fs_info *fs_info,
918 struct btrfs_path *path, u64 bytenr,
919 int *info_level, struct preftrees *preftrees,
920 struct share_check *sc)
924 struct extent_buffer *leaf;
925 struct btrfs_key key;
926 struct btrfs_key found_key;
929 struct btrfs_extent_item *ei;
934 * enumerate all inline refs
936 leaf = path->nodes[0];
937 slot = path->slots[0];
939 item_size = btrfs_item_size_nr(leaf, slot);
940 BUG_ON(item_size < sizeof(*ei));
942 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
943 flags = btrfs_extent_flags(leaf, ei);
944 btrfs_item_key_to_cpu(leaf, &found_key, slot);
946 ptr = (unsigned long)(ei + 1);
947 end = (unsigned long)ei + item_size;
949 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
950 flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
951 struct btrfs_tree_block_info *info;
953 info = (struct btrfs_tree_block_info *)ptr;
954 *info_level = btrfs_tree_block_level(leaf, info);
955 ptr += sizeof(struct btrfs_tree_block_info);
957 } else if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
958 *info_level = found_key.offset;
960 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
964 struct btrfs_extent_inline_ref *iref;
968 iref = (struct btrfs_extent_inline_ref *)ptr;
969 type = btrfs_get_extent_inline_ref_type(leaf, iref,
971 if (type == BTRFS_REF_TYPE_INVALID)
974 offset = btrfs_extent_inline_ref_offset(leaf, iref);
977 case BTRFS_SHARED_BLOCK_REF_KEY:
978 ret = add_direct_ref(fs_info, preftrees,
979 *info_level + 1, offset,
980 bytenr, 1, NULL, GFP_NOFS);
982 case BTRFS_SHARED_DATA_REF_KEY: {
983 struct btrfs_shared_data_ref *sdref;
986 sdref = (struct btrfs_shared_data_ref *)(iref + 1);
987 count = btrfs_shared_data_ref_count(leaf, sdref);
989 ret = add_direct_ref(fs_info, preftrees, 0, offset,
990 bytenr, count, sc, GFP_NOFS);
993 case BTRFS_TREE_BLOCK_REF_KEY:
994 ret = add_indirect_ref(fs_info, preftrees, offset,
995 NULL, *info_level + 1,
996 bytenr, 1, NULL, GFP_NOFS);
998 case BTRFS_EXTENT_DATA_REF_KEY: {
999 struct btrfs_extent_data_ref *dref;
1003 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1004 count = btrfs_extent_data_ref_count(leaf, dref);
1005 key.objectid = btrfs_extent_data_ref_objectid(leaf,
1007 key.type = BTRFS_EXTENT_DATA_KEY;
1008 key.offset = btrfs_extent_data_ref_offset(leaf, dref);
1010 if (sc && sc->inum && key.objectid != sc->inum) {
1011 ret = BACKREF_FOUND_SHARED;
1015 root = btrfs_extent_data_ref_root(leaf, dref);
1017 ret = add_indirect_ref(fs_info, preftrees, root,
1018 &key, 0, bytenr, count,
1027 ptr += btrfs_extent_inline_ref_size(type);
1034 * add all non-inline backrefs for bytenr to the list
1036 * Returns 0 on success, <0 on error, or BACKREF_FOUND_SHARED.
1038 static int add_keyed_refs(struct btrfs_fs_info *fs_info,
1039 struct btrfs_path *path, u64 bytenr,
1040 int info_level, struct preftrees *preftrees,
1041 struct share_check *sc)
1043 struct btrfs_root *extent_root = fs_info->extent_root;
1046 struct extent_buffer *leaf;
1047 struct btrfs_key key;
1050 ret = btrfs_next_item(extent_root, path);
1058 slot = path->slots[0];
1059 leaf = path->nodes[0];
1060 btrfs_item_key_to_cpu(leaf, &key, slot);
1062 if (key.objectid != bytenr)
1064 if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
1066 if (key.type > BTRFS_SHARED_DATA_REF_KEY)
1070 case BTRFS_SHARED_BLOCK_REF_KEY:
1071 /* SHARED DIRECT METADATA backref */
1072 ret = add_direct_ref(fs_info, preftrees,
1073 info_level + 1, key.offset,
1074 bytenr, 1, NULL, GFP_NOFS);
1076 case BTRFS_SHARED_DATA_REF_KEY: {
1077 /* SHARED DIRECT FULL backref */
1078 struct btrfs_shared_data_ref *sdref;
1081 sdref = btrfs_item_ptr(leaf, slot,
1082 struct btrfs_shared_data_ref);
1083 count = btrfs_shared_data_ref_count(leaf, sdref);
1084 ret = add_direct_ref(fs_info, preftrees, 0,
1085 key.offset, bytenr, count,
1089 case BTRFS_TREE_BLOCK_REF_KEY:
1090 /* NORMAL INDIRECT METADATA backref */
1091 ret = add_indirect_ref(fs_info, preftrees, key.offset,
1092 NULL, info_level + 1, bytenr,
1095 case BTRFS_EXTENT_DATA_REF_KEY: {
1096 /* NORMAL INDIRECT DATA backref */
1097 struct btrfs_extent_data_ref *dref;
1101 dref = btrfs_item_ptr(leaf, slot,
1102 struct btrfs_extent_data_ref);
1103 count = btrfs_extent_data_ref_count(leaf, dref);
1104 key.objectid = btrfs_extent_data_ref_objectid(leaf,
1106 key.type = BTRFS_EXTENT_DATA_KEY;
1107 key.offset = btrfs_extent_data_ref_offset(leaf, dref);
1109 if (sc && sc->inum && key.objectid != sc->inum) {
1110 ret = BACKREF_FOUND_SHARED;
1114 root = btrfs_extent_data_ref_root(leaf, dref);
1115 ret = add_indirect_ref(fs_info, preftrees, root,
1116 &key, 0, bytenr, count,
1132 * this adds all existing backrefs (inline backrefs, backrefs and delayed
1133 * refs) for the given bytenr to the refs list, merges duplicates and resolves
1134 * indirect refs to their parent bytenr.
1135 * When roots are found, they're added to the roots list
1137 * If time_seq is set to SEQ_LAST, it will not search delayed_refs, and behave
1138 * much like trans == NULL case, the difference only lies in it will not
1140 * The special case is for qgroup to search roots in commit_transaction().
1142 * @sc - if !NULL, then immediately return BACKREF_FOUND_SHARED when a
1143 * shared extent is detected.
1145 * Otherwise this returns 0 for success and <0 for an error.
1147 * If ignore_offset is set to false, only extent refs whose offsets match
1148 * extent_item_pos are returned. If true, every extent ref is returned
1149 * and extent_item_pos is ignored.
1151 * FIXME some caching might speed things up
1153 static int find_parent_nodes(struct btrfs_trans_handle *trans,
1154 struct btrfs_fs_info *fs_info, u64 bytenr,
1155 u64 time_seq, struct ulist *refs,
1156 struct ulist *roots, const u64 *extent_item_pos,
1157 struct share_check *sc, bool ignore_offset)
1159 struct btrfs_key key;
1160 struct btrfs_path *path;
1161 struct btrfs_delayed_ref_root *delayed_refs = NULL;
1162 struct btrfs_delayed_ref_head *head;
1165 struct prelim_ref *ref;
1166 struct rb_node *node;
1167 struct extent_inode_elem *eie = NULL;
1168 struct preftrees preftrees = {
1169 .direct = PREFTREE_INIT,
1170 .indirect = PREFTREE_INIT,
1171 .indirect_missing_keys = PREFTREE_INIT
1174 key.objectid = bytenr;
1175 key.offset = (u64)-1;
1176 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1177 key.type = BTRFS_METADATA_ITEM_KEY;
1179 key.type = BTRFS_EXTENT_ITEM_KEY;
1181 path = btrfs_alloc_path();
1185 path->search_commit_root = 1;
1186 path->skip_locking = 1;
1189 if (time_seq == SEQ_LAST)
1190 path->skip_locking = 1;
1193 * grab both a lock on the path and a lock on the delayed ref head.
1194 * We need both to get a consistent picture of how the refs look
1195 * at a specified point in time
1200 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
1205 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1206 if (trans && likely(trans->type != __TRANS_DUMMY) &&
1207 time_seq != SEQ_LAST) {
1209 if (trans && time_seq != SEQ_LAST) {
1212 * look if there are updates for this ref queued and lock the
1215 delayed_refs = &trans->transaction->delayed_refs;
1216 spin_lock(&delayed_refs->lock);
1217 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
1219 if (!mutex_trylock(&head->mutex)) {
1220 refcount_inc(&head->refs);
1221 spin_unlock(&delayed_refs->lock);
1223 btrfs_release_path(path);
1226 * Mutex was contended, block until it's
1227 * released and try again
1229 mutex_lock(&head->mutex);
1230 mutex_unlock(&head->mutex);
1231 btrfs_put_delayed_ref_head(head);
1234 spin_unlock(&delayed_refs->lock);
1235 ret = add_delayed_refs(fs_info, head, time_seq,
1237 mutex_unlock(&head->mutex);
1241 spin_unlock(&delayed_refs->lock);
1245 if (path->slots[0]) {
1246 struct extent_buffer *leaf;
1250 leaf = path->nodes[0];
1251 slot = path->slots[0];
1252 btrfs_item_key_to_cpu(leaf, &key, slot);
1253 if (key.objectid == bytenr &&
1254 (key.type == BTRFS_EXTENT_ITEM_KEY ||
1255 key.type == BTRFS_METADATA_ITEM_KEY)) {
1256 ret = add_inline_refs(fs_info, path, bytenr,
1257 &info_level, &preftrees, sc);
1260 ret = add_keyed_refs(fs_info, path, bytenr, info_level,
1267 btrfs_release_path(path);
1269 ret = add_missing_keys(fs_info, &preftrees, path->skip_locking == 0);
1273 WARN_ON(!RB_EMPTY_ROOT(&preftrees.indirect_missing_keys.root.rb_root));
1275 ret = resolve_indirect_refs(fs_info, path, time_seq, &preftrees,
1276 extent_item_pos, sc, ignore_offset);
1280 WARN_ON(!RB_EMPTY_ROOT(&preftrees.indirect.root.rb_root));
1283 * This walks the tree of merged and resolved refs. Tree blocks are
1284 * read in as needed. Unique entries are added to the ulist, and
1285 * the list of found roots is updated.
1287 * We release the entire tree in one go before returning.
1289 node = rb_first_cached(&preftrees.direct.root);
1291 ref = rb_entry(node, struct prelim_ref, rbnode);
1292 node = rb_next(&ref->rbnode);
1294 * ref->count < 0 can happen here if there are delayed
1295 * refs with a node->action of BTRFS_DROP_DELAYED_REF.
1296 * prelim_ref_insert() relies on this when merging
1297 * identical refs to keep the overall count correct.
1298 * prelim_ref_insert() will merge only those refs
1299 * which compare identically. Any refs having
1300 * e.g. different offsets would not be merged,
1301 * and would retain their original ref->count < 0.
1303 if (roots && ref->count && ref->root_id && ref->parent == 0) {
1304 if (sc && sc->root_objectid &&
1305 ref->root_id != sc->root_objectid) {
1306 ret = BACKREF_FOUND_SHARED;
1310 /* no parent == root of tree */
1311 ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS);
1315 if (ref->count && ref->parent) {
1316 if (extent_item_pos && !ref->inode_list &&
1318 struct extent_buffer *eb;
1320 eb = read_tree_block(fs_info, ref->parent, 0,
1325 } else if (!extent_buffer_uptodate(eb)) {
1326 free_extent_buffer(eb);
1331 if (!path->skip_locking) {
1332 btrfs_tree_read_lock(eb);
1333 btrfs_set_lock_blocking_read(eb);
1335 ret = find_extent_in_eb(eb, bytenr,
1336 *extent_item_pos, &eie, ignore_offset);
1337 if (!path->skip_locking)
1338 btrfs_tree_read_unlock_blocking(eb);
1339 free_extent_buffer(eb);
1342 ref->inode_list = eie;
1344 ret = ulist_add_merge_ptr(refs, ref->parent,
1346 (void **)&eie, GFP_NOFS);
1349 if (!ret && extent_item_pos) {
1351 * we've recorded that parent, so we must extend
1352 * its inode list here
1357 eie->next = ref->inode_list;
1365 btrfs_free_path(path);
1367 prelim_release(&preftrees.direct);
1368 prelim_release(&preftrees.indirect);
1369 prelim_release(&preftrees.indirect_missing_keys);
1372 free_inode_elem_list(eie);
1376 static void free_leaf_list(struct ulist *blocks)
1378 struct ulist_node *node = NULL;
1379 struct extent_inode_elem *eie;
1380 struct ulist_iterator uiter;
1382 ULIST_ITER_INIT(&uiter);
1383 while ((node = ulist_next(blocks, &uiter))) {
1386 eie = unode_aux_to_inode_list(node);
1387 free_inode_elem_list(eie);
1395 * Finds all leafs with a reference to the specified combination of bytenr and
1396 * offset. key_list_head will point to a list of corresponding keys (caller must
1397 * free each list element). The leafs will be stored in the leafs ulist, which
1398 * must be freed with ulist_free.
1400 * returns 0 on success, <0 on error
1402 int btrfs_find_all_leafs(struct btrfs_trans_handle *trans,
1403 struct btrfs_fs_info *fs_info, u64 bytenr,
1404 u64 time_seq, struct ulist **leafs,
1405 const u64 *extent_item_pos, bool ignore_offset)
1409 *leafs = ulist_alloc(GFP_NOFS);
1413 ret = find_parent_nodes(trans, fs_info, bytenr, time_seq,
1414 *leafs, NULL, extent_item_pos, NULL, ignore_offset);
1415 if (ret < 0 && ret != -ENOENT) {
1416 free_leaf_list(*leafs);
1424 * walk all backrefs for a given extent to find all roots that reference this
1425 * extent. Walking a backref means finding all extents that reference this
1426 * extent and in turn walk the backrefs of those, too. Naturally this is a
1427 * recursive process, but here it is implemented in an iterative fashion: We
1428 * find all referencing extents for the extent in question and put them on a
1429 * list. In turn, we find all referencing extents for those, further appending
1430 * to the list. The way we iterate the list allows adding more elements after
1431 * the current while iterating. The process stops when we reach the end of the
1432 * list. Found roots are added to the roots list.
1434 * returns 0 on success, < 0 on error.
1436 static int btrfs_find_all_roots_safe(struct btrfs_trans_handle *trans,
1437 struct btrfs_fs_info *fs_info, u64 bytenr,
1438 u64 time_seq, struct ulist **roots,
1442 struct ulist_node *node = NULL;
1443 struct ulist_iterator uiter;
1446 tmp = ulist_alloc(GFP_NOFS);
1449 *roots = ulist_alloc(GFP_NOFS);
1455 ULIST_ITER_INIT(&uiter);
1457 ret = find_parent_nodes(trans, fs_info, bytenr, time_seq,
1458 tmp, *roots, NULL, NULL, ignore_offset);
1459 if (ret < 0 && ret != -ENOENT) {
1464 node = ulist_next(tmp, &uiter);
1475 int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1476 struct btrfs_fs_info *fs_info, u64 bytenr,
1477 u64 time_seq, struct ulist **roots,
1483 down_read(&fs_info->commit_root_sem);
1484 ret = btrfs_find_all_roots_safe(trans, fs_info, bytenr,
1485 time_seq, roots, ignore_offset);
1487 up_read(&fs_info->commit_root_sem);
1492 * btrfs_check_shared - tell us whether an extent is shared
1494 * btrfs_check_shared uses the backref walking code but will short
1495 * circuit as soon as it finds a root or inode that doesn't match the
1496 * one passed in. This provides a significant performance benefit for
1497 * callers (such as fiemap) which want to know whether the extent is
1498 * shared but do not need a ref count.
1500 * This attempts to attach to the running transaction in order to account for
1501 * delayed refs, but continues on even when no running transaction exists.
1503 * Return: 0 if extent is not shared, 1 if it is shared, < 0 on error.
1505 int btrfs_check_shared(struct btrfs_root *root, u64 inum, u64 bytenr,
1506 struct ulist *roots, struct ulist *tmp)
1508 struct btrfs_fs_info *fs_info = root->fs_info;
1509 struct btrfs_trans_handle *trans;
1510 struct ulist_iterator uiter;
1511 struct ulist_node *node;
1512 struct seq_list elem = SEQ_LIST_INIT(elem);
1514 struct share_check shared = {
1515 .root_objectid = root->root_key.objectid,
1523 trans = btrfs_join_transaction_nostart(root);
1524 if (IS_ERR(trans)) {
1525 if (PTR_ERR(trans) != -ENOENT && PTR_ERR(trans) != -EROFS) {
1526 ret = PTR_ERR(trans);
1530 down_read(&fs_info->commit_root_sem);
1532 btrfs_get_tree_mod_seq(fs_info, &elem);
1535 ULIST_ITER_INIT(&uiter);
1537 ret = find_parent_nodes(trans, fs_info, bytenr, elem.seq, tmp,
1538 roots, NULL, &shared, false);
1539 if (ret == BACKREF_FOUND_SHARED) {
1540 /* this is the only condition under which we return 1 */
1544 if (ret < 0 && ret != -ENOENT)
1547 node = ulist_next(tmp, &uiter);
1551 shared.share_count = 0;
1556 btrfs_put_tree_mod_seq(fs_info, &elem);
1557 btrfs_end_transaction(trans);
1559 up_read(&fs_info->commit_root_sem);
1562 ulist_release(roots);
1567 int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,
1568 u64 start_off, struct btrfs_path *path,
1569 struct btrfs_inode_extref **ret_extref,
1573 struct btrfs_key key;
1574 struct btrfs_key found_key;
1575 struct btrfs_inode_extref *extref;
1576 const struct extent_buffer *leaf;
1579 key.objectid = inode_objectid;
1580 key.type = BTRFS_INODE_EXTREF_KEY;
1581 key.offset = start_off;
1583 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1588 leaf = path->nodes[0];
1589 slot = path->slots[0];
1590 if (slot >= btrfs_header_nritems(leaf)) {
1592 * If the item at offset is not found,
1593 * btrfs_search_slot will point us to the slot
1594 * where it should be inserted. In our case
1595 * that will be the slot directly before the
1596 * next INODE_REF_KEY_V2 item. In the case
1597 * that we're pointing to the last slot in a
1598 * leaf, we must move one leaf over.
1600 ret = btrfs_next_leaf(root, path);
1609 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1612 * Check that we're still looking at an extended ref key for
1613 * this particular objectid. If we have different
1614 * objectid or type then there are no more to be found
1615 * in the tree and we can exit.
1618 if (found_key.objectid != inode_objectid)
1620 if (found_key.type != BTRFS_INODE_EXTREF_KEY)
1624 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1625 extref = (struct btrfs_inode_extref *)ptr;
1626 *ret_extref = extref;
1628 *found_off = found_key.offset;
1636 * this iterates to turn a name (from iref/extref) into a full filesystem path.
1637 * Elements of the path are separated by '/' and the path is guaranteed to be
1638 * 0-terminated. the path is only given within the current file system.
1639 * Therefore, it never starts with a '/'. the caller is responsible to provide
1640 * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
1641 * the start point of the resulting string is returned. this pointer is within
1643 * in case the path buffer would overflow, the pointer is decremented further
1644 * as if output was written to the buffer, though no more output is actually
1645 * generated. that way, the caller can determine how much space would be
1646 * required for the path to fit into the buffer. in that case, the returned
1647 * value will be smaller than dest. callers must check this!
1649 char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
1650 u32 name_len, unsigned long name_off,
1651 struct extent_buffer *eb_in, u64 parent,
1652 char *dest, u32 size)
1657 s64 bytes_left = ((s64)size) - 1;
1658 struct extent_buffer *eb = eb_in;
1659 struct btrfs_key found_key;
1660 int leave_spinning = path->leave_spinning;
1661 struct btrfs_inode_ref *iref;
1663 if (bytes_left >= 0)
1664 dest[bytes_left] = '\0';
1666 path->leave_spinning = 1;
1668 bytes_left -= name_len;
1669 if (bytes_left >= 0)
1670 read_extent_buffer(eb, dest + bytes_left,
1671 name_off, name_len);
1673 if (!path->skip_locking)
1674 btrfs_tree_read_unlock_blocking(eb);
1675 free_extent_buffer(eb);
1677 ret = btrfs_find_item(fs_root, path, parent, 0,
1678 BTRFS_INODE_REF_KEY, &found_key);
1684 next_inum = found_key.offset;
1686 /* regular exit ahead */
1687 if (parent == next_inum)
1690 slot = path->slots[0];
1691 eb = path->nodes[0];
1692 /* make sure we can use eb after releasing the path */
1694 if (!path->skip_locking)
1695 btrfs_set_lock_blocking_read(eb);
1696 path->nodes[0] = NULL;
1699 btrfs_release_path(path);
1700 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1702 name_len = btrfs_inode_ref_name_len(eb, iref);
1703 name_off = (unsigned long)(iref + 1);
1707 if (bytes_left >= 0)
1708 dest[bytes_left] = '/';
1711 btrfs_release_path(path);
1712 path->leave_spinning = leave_spinning;
1715 return ERR_PTR(ret);
1717 return dest + bytes_left;
1721 * this makes the path point to (logical EXTENT_ITEM *)
1722 * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
1723 * tree blocks and <0 on error.
1725 int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
1726 struct btrfs_path *path, struct btrfs_key *found_key,
1733 const struct extent_buffer *eb;
1734 struct btrfs_extent_item *ei;
1735 struct btrfs_key key;
1737 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1738 key.type = BTRFS_METADATA_ITEM_KEY;
1740 key.type = BTRFS_EXTENT_ITEM_KEY;
1741 key.objectid = logical;
1742 key.offset = (u64)-1;
1744 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
1748 ret = btrfs_previous_extent_item(fs_info->extent_root, path, 0);
1754 btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]);
1755 if (found_key->type == BTRFS_METADATA_ITEM_KEY)
1756 size = fs_info->nodesize;
1757 else if (found_key->type == BTRFS_EXTENT_ITEM_KEY)
1758 size = found_key->offset;
1760 if (found_key->objectid > logical ||
1761 found_key->objectid + size <= logical) {
1762 btrfs_debug(fs_info,
1763 "logical %llu is not within any extent", logical);
1767 eb = path->nodes[0];
1768 item_size = btrfs_item_size_nr(eb, path->slots[0]);
1769 BUG_ON(item_size < sizeof(*ei));
1771 ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
1772 flags = btrfs_extent_flags(eb, ei);
1774 btrfs_debug(fs_info,
1775 "logical %llu is at position %llu within the extent (%llu EXTENT_ITEM %llu) flags %#llx size %u",
1776 logical, logical - found_key->objectid, found_key->objectid,
1777 found_key->offset, flags, item_size);
1779 WARN_ON(!flags_ret);
1781 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1782 *flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK;
1783 else if (flags & BTRFS_EXTENT_FLAG_DATA)
1784 *flags_ret = BTRFS_EXTENT_FLAG_DATA;
1794 * helper function to iterate extent inline refs. ptr must point to a 0 value
1795 * for the first call and may be modified. it is used to track state.
1796 * if more refs exist, 0 is returned and the next call to
1797 * get_extent_inline_ref must pass the modified ptr parameter to get the
1798 * next ref. after the last ref was processed, 1 is returned.
1799 * returns <0 on error
1801 static int get_extent_inline_ref(unsigned long *ptr,
1802 const struct extent_buffer *eb,
1803 const struct btrfs_key *key,
1804 const struct btrfs_extent_item *ei,
1806 struct btrfs_extent_inline_ref **out_eiref,
1811 struct btrfs_tree_block_info *info;
1815 flags = btrfs_extent_flags(eb, ei);
1816 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1817 if (key->type == BTRFS_METADATA_ITEM_KEY) {
1818 /* a skinny metadata extent */
1820 (struct btrfs_extent_inline_ref *)(ei + 1);
1822 WARN_ON(key->type != BTRFS_EXTENT_ITEM_KEY);
1823 info = (struct btrfs_tree_block_info *)(ei + 1);
1825 (struct btrfs_extent_inline_ref *)(info + 1);
1828 *out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1);
1830 *ptr = (unsigned long)*out_eiref;
1831 if ((unsigned long)(*ptr) >= (unsigned long)ei + item_size)
1835 end = (unsigned long)ei + item_size;
1836 *out_eiref = (struct btrfs_extent_inline_ref *)(*ptr);
1837 *out_type = btrfs_get_extent_inline_ref_type(eb, *out_eiref,
1838 BTRFS_REF_TYPE_ANY);
1839 if (*out_type == BTRFS_REF_TYPE_INVALID)
1842 *ptr += btrfs_extent_inline_ref_size(*out_type);
1843 WARN_ON(*ptr > end);
1845 return 1; /* last */
1851 * reads the tree block backref for an extent. tree level and root are returned
1852 * through out_level and out_root. ptr must point to a 0 value for the first
1853 * call and may be modified (see get_extent_inline_ref comment).
1854 * returns 0 if data was provided, 1 if there was no more data to provide or
1857 int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
1858 struct btrfs_key *key, struct btrfs_extent_item *ei,
1859 u32 item_size, u64 *out_root, u8 *out_level)
1863 struct btrfs_extent_inline_ref *eiref;
1865 if (*ptr == (unsigned long)-1)
1869 ret = get_extent_inline_ref(ptr, eb, key, ei, item_size,
1874 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1875 type == BTRFS_SHARED_BLOCK_REF_KEY)
1882 /* we can treat both ref types equally here */
1883 *out_root = btrfs_extent_inline_ref_offset(eb, eiref);
1885 if (key->type == BTRFS_EXTENT_ITEM_KEY) {
1886 struct btrfs_tree_block_info *info;
1888 info = (struct btrfs_tree_block_info *)(ei + 1);
1889 *out_level = btrfs_tree_block_level(eb, info);
1891 ASSERT(key->type == BTRFS_METADATA_ITEM_KEY);
1892 *out_level = (u8)key->offset;
1896 *ptr = (unsigned long)-1;
1901 static int iterate_leaf_refs(struct btrfs_fs_info *fs_info,
1902 struct extent_inode_elem *inode_list,
1903 u64 root, u64 extent_item_objectid,
1904 iterate_extent_inodes_t *iterate, void *ctx)
1906 struct extent_inode_elem *eie;
1909 for (eie = inode_list; eie; eie = eie->next) {
1910 btrfs_debug(fs_info,
1911 "ref for %llu resolved, key (%llu EXTEND_DATA %llu), root %llu",
1912 extent_item_objectid, eie->inum,
1914 ret = iterate(eie->inum, eie->offset, root, ctx);
1916 btrfs_debug(fs_info,
1917 "stopping iteration for %llu due to ret=%d",
1918 extent_item_objectid, ret);
1927 * calls iterate() for every inode that references the extent identified by
1928 * the given parameters.
1929 * when the iterator function returns a non-zero value, iteration stops.
1931 int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
1932 u64 extent_item_objectid, u64 extent_item_pos,
1933 int search_commit_root,
1934 iterate_extent_inodes_t *iterate, void *ctx,
1938 struct btrfs_trans_handle *trans = NULL;
1939 struct ulist *refs = NULL;
1940 struct ulist *roots = NULL;
1941 struct ulist_node *ref_node = NULL;
1942 struct ulist_node *root_node = NULL;
1943 struct seq_list tree_mod_seq_elem = SEQ_LIST_INIT(tree_mod_seq_elem);
1944 struct ulist_iterator ref_uiter;
1945 struct ulist_iterator root_uiter;
1947 btrfs_debug(fs_info, "resolving all inodes for extent %llu",
1948 extent_item_objectid);
1950 if (!search_commit_root) {
1951 trans = btrfs_attach_transaction(fs_info->extent_root);
1952 if (IS_ERR(trans)) {
1953 if (PTR_ERR(trans) != -ENOENT &&
1954 PTR_ERR(trans) != -EROFS)
1955 return PTR_ERR(trans);
1961 btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1963 down_read(&fs_info->commit_root_sem);
1965 ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid,
1966 tree_mod_seq_elem.seq, &refs,
1967 &extent_item_pos, ignore_offset);
1971 ULIST_ITER_INIT(&ref_uiter);
1972 while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) {
1973 ret = btrfs_find_all_roots_safe(trans, fs_info, ref_node->val,
1974 tree_mod_seq_elem.seq, &roots,
1978 ULIST_ITER_INIT(&root_uiter);
1979 while (!ret && (root_node = ulist_next(roots, &root_uiter))) {
1980 btrfs_debug(fs_info,
1981 "root %llu references leaf %llu, data list %#llx",
1982 root_node->val, ref_node->val,
1984 ret = iterate_leaf_refs(fs_info,
1985 (struct extent_inode_elem *)
1986 (uintptr_t)ref_node->aux,
1988 extent_item_objectid,
1994 free_leaf_list(refs);
1997 btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1998 btrfs_end_transaction(trans);
2000 up_read(&fs_info->commit_root_sem);
2006 int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
2007 struct btrfs_path *path,
2008 iterate_extent_inodes_t *iterate, void *ctx,
2012 u64 extent_item_pos;
2014 struct btrfs_key found_key;
2015 int search_commit_root = path->search_commit_root;
2017 ret = extent_from_logical(fs_info, logical, path, &found_key, &flags);
2018 btrfs_release_path(path);
2021 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
2024 extent_item_pos = logical - found_key.objectid;
2025 ret = iterate_extent_inodes(fs_info, found_key.objectid,
2026 extent_item_pos, search_commit_root,
2027 iterate, ctx, ignore_offset);
2032 typedef int (iterate_irefs_t)(u64 parent, u32 name_len, unsigned long name_off,
2033 struct extent_buffer *eb, void *ctx);
2035 static int iterate_inode_refs(u64 inum, struct btrfs_root *fs_root,
2036 struct btrfs_path *path,
2037 iterate_irefs_t *iterate, void *ctx)
2046 struct extent_buffer *eb;
2047 struct btrfs_item *item;
2048 struct btrfs_inode_ref *iref;
2049 struct btrfs_key found_key;
2052 ret = btrfs_find_item(fs_root, path, inum,
2053 parent ? parent + 1 : 0, BTRFS_INODE_REF_KEY,
2059 ret = found ? 0 : -ENOENT;
2064 parent = found_key.offset;
2065 slot = path->slots[0];
2066 eb = btrfs_clone_extent_buffer(path->nodes[0]);
2071 btrfs_release_path(path);
2073 item = btrfs_item_nr(slot);
2074 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
2076 for (cur = 0; cur < btrfs_item_size(eb, item); cur += len) {
2077 name_len = btrfs_inode_ref_name_len(eb, iref);
2078 /* path must be released before calling iterate()! */
2079 btrfs_debug(fs_root->fs_info,
2080 "following ref at offset %u for inode %llu in tree %llu",
2081 cur, found_key.objectid,
2082 fs_root->root_key.objectid);
2083 ret = iterate(parent, name_len,
2084 (unsigned long)(iref + 1), eb, ctx);
2087 len = sizeof(*iref) + name_len;
2088 iref = (struct btrfs_inode_ref *)((char *)iref + len);
2090 free_extent_buffer(eb);
2093 btrfs_release_path(path);
2098 static int iterate_inode_extrefs(u64 inum, struct btrfs_root *fs_root,
2099 struct btrfs_path *path,
2100 iterate_irefs_t *iterate, void *ctx)
2107 struct extent_buffer *eb;
2108 struct btrfs_inode_extref *extref;
2114 ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref,
2119 ret = found ? 0 : -ENOENT;
2124 slot = path->slots[0];
2125 eb = btrfs_clone_extent_buffer(path->nodes[0]);
2130 btrfs_release_path(path);
2132 item_size = btrfs_item_size_nr(eb, slot);
2133 ptr = btrfs_item_ptr_offset(eb, slot);
2136 while (cur_offset < item_size) {
2139 extref = (struct btrfs_inode_extref *)(ptr + cur_offset);
2140 parent = btrfs_inode_extref_parent(eb, extref);
2141 name_len = btrfs_inode_extref_name_len(eb, extref);
2142 ret = iterate(parent, name_len,
2143 (unsigned long)&extref->name, eb, ctx);
2147 cur_offset += btrfs_inode_extref_name_len(eb, extref);
2148 cur_offset += sizeof(*extref);
2150 free_extent_buffer(eb);
2155 btrfs_release_path(path);
2160 static int iterate_irefs(u64 inum, struct btrfs_root *fs_root,
2161 struct btrfs_path *path, iterate_irefs_t *iterate,
2167 ret = iterate_inode_refs(inum, fs_root, path, iterate, ctx);
2170 else if (ret != -ENOENT)
2173 ret = iterate_inode_extrefs(inum, fs_root, path, iterate, ctx);
2174 if (ret == -ENOENT && found_refs)
2181 * returns 0 if the path could be dumped (probably truncated)
2182 * returns <0 in case of an error
2184 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
2185 struct extent_buffer *eb, void *ctx)
2187 struct inode_fs_paths *ipath = ctx;
2190 int i = ipath->fspath->elem_cnt;
2191 const int s_ptr = sizeof(char *);
2194 bytes_left = ipath->fspath->bytes_left > s_ptr ?
2195 ipath->fspath->bytes_left - s_ptr : 0;
2197 fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
2198 fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len,
2199 name_off, eb, inum, fspath_min, bytes_left);
2201 return PTR_ERR(fspath);
2203 if (fspath > fspath_min) {
2204 ipath->fspath->val[i] = (u64)(unsigned long)fspath;
2205 ++ipath->fspath->elem_cnt;
2206 ipath->fspath->bytes_left = fspath - fspath_min;
2208 ++ipath->fspath->elem_missed;
2209 ipath->fspath->bytes_missing += fspath_min - fspath;
2210 ipath->fspath->bytes_left = 0;
2217 * this dumps all file system paths to the inode into the ipath struct, provided
2218 * is has been created large enough. each path is zero-terminated and accessed
2219 * from ipath->fspath->val[i].
2220 * when it returns, there are ipath->fspath->elem_cnt number of paths available
2221 * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the
2222 * number of missed paths is recorded in ipath->fspath->elem_missed, otherwise,
2223 * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
2224 * have been needed to return all paths.
2226 int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
2228 return iterate_irefs(inum, ipath->fs_root, ipath->btrfs_path,
2229 inode_to_path, ipath);
2232 struct btrfs_data_container *init_data_container(u32 total_bytes)
2234 struct btrfs_data_container *data;
2237 alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
2238 data = kvmalloc(alloc_bytes, GFP_KERNEL);
2240 return ERR_PTR(-ENOMEM);
2242 if (total_bytes >= sizeof(*data)) {
2243 data->bytes_left = total_bytes - sizeof(*data);
2244 data->bytes_missing = 0;
2246 data->bytes_missing = sizeof(*data) - total_bytes;
2247 data->bytes_left = 0;
2251 data->elem_missed = 0;
2257 * allocates space to return multiple file system paths for an inode.
2258 * total_bytes to allocate are passed, note that space usable for actual path
2259 * information will be total_bytes - sizeof(struct inode_fs_paths).
2260 * the returned pointer must be freed with free_ipath() in the end.
2262 struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
2263 struct btrfs_path *path)
2265 struct inode_fs_paths *ifp;
2266 struct btrfs_data_container *fspath;
2268 fspath = init_data_container(total_bytes);
2270 return ERR_CAST(fspath);
2272 ifp = kmalloc(sizeof(*ifp), GFP_KERNEL);
2275 return ERR_PTR(-ENOMEM);
2278 ifp->btrfs_path = path;
2279 ifp->fspath = fspath;
2280 ifp->fs_root = fs_root;
2285 void free_ipath(struct inode_fs_paths *ipath)
2289 kvfree(ipath->fspath);