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"
18 /* Just an arbitrary number so we can be sure this happened */
19 #define BACKREF_FOUND_SHARED 6
21 struct extent_inode_elem {
24 struct extent_inode_elem *next;
27 static int check_extent_in_eb(const struct btrfs_key *key,
28 const struct extent_buffer *eb,
29 const struct btrfs_file_extent_item *fi,
31 struct extent_inode_elem **eie,
35 struct extent_inode_elem *e;
38 !btrfs_file_extent_compression(eb, fi) &&
39 !btrfs_file_extent_encryption(eb, fi) &&
40 !btrfs_file_extent_other_encoding(eb, fi)) {
44 data_offset = btrfs_file_extent_offset(eb, fi);
45 data_len = btrfs_file_extent_num_bytes(eb, fi);
47 if (extent_item_pos < data_offset ||
48 extent_item_pos >= data_offset + data_len)
50 offset = extent_item_pos - data_offset;
53 e = kmalloc(sizeof(*e), GFP_NOFS);
58 e->inum = key->objectid;
59 e->offset = key->offset + offset;
65 static void free_inode_elem_list(struct extent_inode_elem *eie)
67 struct extent_inode_elem *eie_next;
69 for (; eie; eie = eie_next) {
75 static int find_extent_in_eb(const struct extent_buffer *eb,
76 u64 wanted_disk_byte, u64 extent_item_pos,
77 struct extent_inode_elem **eie,
82 struct btrfs_file_extent_item *fi;
89 * from the shared data ref, we only have the leaf but we need
90 * the key. thus, we must look into all items and see that we
91 * find one (some) with a reference to our extent item.
93 nritems = btrfs_header_nritems(eb);
94 for (slot = 0; slot < nritems; ++slot) {
95 btrfs_item_key_to_cpu(eb, &key, slot);
96 if (key.type != BTRFS_EXTENT_DATA_KEY)
98 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
99 extent_type = btrfs_file_extent_type(eb, fi);
100 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
102 /* don't skip BTRFS_FILE_EXTENT_PREALLOC, we can handle that */
103 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
104 if (disk_byte != wanted_disk_byte)
107 ret = check_extent_in_eb(&key, eb, fi, extent_item_pos, eie, ignore_offset);
116 struct rb_root_cached root;
120 #define PREFTREE_INIT { .root = RB_ROOT_CACHED, .count = 0 }
123 struct preftree direct; /* BTRFS_SHARED_[DATA|BLOCK]_REF_KEY */
124 struct preftree indirect; /* BTRFS_[TREE_BLOCK|EXTENT_DATA]_REF_KEY */
125 struct preftree indirect_missing_keys;
129 * Checks for a shared extent during backref search.
131 * The share_count tracks prelim_refs (direct and indirect) having a
133 * - incremented when a ref->count transitions to >0
134 * - decremented when a ref->count transitions to <1
142 static inline int extent_is_shared(struct share_check *sc)
144 return (sc && sc->share_count > 1) ? BACKREF_FOUND_SHARED : 0;
147 static struct kmem_cache *btrfs_prelim_ref_cache;
149 int __init btrfs_prelim_ref_init(void)
151 btrfs_prelim_ref_cache = kmem_cache_create("btrfs_prelim_ref",
152 sizeof(struct prelim_ref),
156 if (!btrfs_prelim_ref_cache)
161 void __cold btrfs_prelim_ref_exit(void)
163 kmem_cache_destroy(btrfs_prelim_ref_cache);
166 static void free_pref(struct prelim_ref *ref)
168 kmem_cache_free(btrfs_prelim_ref_cache, ref);
172 * Return 0 when both refs are for the same block (and can be merged).
173 * A -1 return indicates ref1 is a 'lower' block than ref2, while 1
174 * indicates a 'higher' block.
176 static int prelim_ref_compare(struct prelim_ref *ref1,
177 struct prelim_ref *ref2)
179 if (ref1->level < ref2->level)
181 if (ref1->level > ref2->level)
183 if (ref1->root_id < ref2->root_id)
185 if (ref1->root_id > ref2->root_id)
187 if (ref1->key_for_search.type < ref2->key_for_search.type)
189 if (ref1->key_for_search.type > ref2->key_for_search.type)
191 if (ref1->key_for_search.objectid < ref2->key_for_search.objectid)
193 if (ref1->key_for_search.objectid > ref2->key_for_search.objectid)
195 if (ref1->key_for_search.offset < ref2->key_for_search.offset)
197 if (ref1->key_for_search.offset > ref2->key_for_search.offset)
199 if (ref1->parent < ref2->parent)
201 if (ref1->parent > ref2->parent)
207 static void update_share_count(struct share_check *sc, int oldcount,
210 if ((!sc) || (oldcount == 0 && newcount < 1))
213 if (oldcount > 0 && newcount < 1)
215 else if (oldcount < 1 && newcount > 0)
220 * Add @newref to the @root rbtree, merging identical refs.
222 * Callers should assume that newref has been freed after calling.
224 static void prelim_ref_insert(const struct btrfs_fs_info *fs_info,
225 struct preftree *preftree,
226 struct prelim_ref *newref,
227 struct share_check *sc)
229 struct rb_root_cached *root;
231 struct rb_node *parent = NULL;
232 struct prelim_ref *ref;
234 bool leftmost = true;
236 root = &preftree->root;
237 p = &root->rb_root.rb_node;
241 ref = rb_entry(parent, struct prelim_ref, rbnode);
242 result = prelim_ref_compare(ref, newref);
245 } else if (result > 0) {
249 /* Identical refs, merge them and free @newref */
250 struct extent_inode_elem *eie = ref->inode_list;
252 while (eie && eie->next)
256 ref->inode_list = newref->inode_list;
258 eie->next = newref->inode_list;
259 trace_btrfs_prelim_ref_merge(fs_info, ref, newref,
262 * A delayed ref can have newref->count < 0.
263 * The ref->count is updated to follow any
264 * BTRFS_[ADD|DROP]_DELAYED_REF actions.
266 update_share_count(sc, ref->count,
267 ref->count + newref->count);
268 ref->count += newref->count;
274 update_share_count(sc, 0, newref->count);
276 trace_btrfs_prelim_ref_insert(fs_info, newref, NULL, preftree->count);
277 rb_link_node(&newref->rbnode, parent, p);
278 rb_insert_color_cached(&newref->rbnode, root, leftmost);
282 * Release the entire tree. We don't care about internal consistency so
283 * just free everything and then reset the tree root.
285 static void prelim_release(struct preftree *preftree)
287 struct prelim_ref *ref, *next_ref;
289 rbtree_postorder_for_each_entry_safe(ref, next_ref,
290 &preftree->root.rb_root, rbnode)
293 preftree->root = RB_ROOT_CACHED;
298 * the rules for all callers of this function are:
299 * - obtaining the parent is the goal
300 * - if you add a key, you must know that it is a correct key
301 * - if you cannot add the parent or a correct key, then we will look into the
302 * block later to set a correct key
306 * backref type | shared | indirect | shared | indirect
307 * information | tree | tree | data | data
308 * --------------------+--------+----------+--------+----------
309 * parent logical | y | - | - | -
310 * key to resolve | - | y | y | y
311 * tree block logical | - | - | - | -
312 * root for resolving | y | y | y | y
314 * - column 1: we've the parent -> done
315 * - column 2, 3, 4: we use the key to find the parent
317 * on disk refs (inline or keyed)
318 * ==============================
319 * backref type | shared | indirect | shared | indirect
320 * information | tree | tree | data | data
321 * --------------------+--------+----------+--------+----------
322 * parent logical | y | - | y | -
323 * key to resolve | - | - | - | y
324 * tree block logical | y | y | y | y
325 * root for resolving | - | y | y | y
327 * - column 1, 3: we've the parent -> done
328 * - column 2: we take the first key from the block to find the parent
329 * (see add_missing_keys)
330 * - column 4: we use the key to find the parent
332 * additional information that's available but not required to find the parent
333 * block might help in merging entries to gain some speed.
335 static int add_prelim_ref(const struct btrfs_fs_info *fs_info,
336 struct preftree *preftree, u64 root_id,
337 const struct btrfs_key *key, int level, u64 parent,
338 u64 wanted_disk_byte, int count,
339 struct share_check *sc, gfp_t gfp_mask)
341 struct prelim_ref *ref;
343 if (root_id == BTRFS_DATA_RELOC_TREE_OBJECTID)
346 ref = kmem_cache_alloc(btrfs_prelim_ref_cache, gfp_mask);
350 ref->root_id = root_id;
352 ref->key_for_search = *key;
354 memset(&ref->key_for_search, 0, sizeof(ref->key_for_search));
356 ref->inode_list = NULL;
359 ref->parent = parent;
360 ref->wanted_disk_byte = wanted_disk_byte;
361 prelim_ref_insert(fs_info, preftree, ref, sc);
362 return extent_is_shared(sc);
365 /* direct refs use root == 0, key == NULL */
366 static int add_direct_ref(const struct btrfs_fs_info *fs_info,
367 struct preftrees *preftrees, int level, u64 parent,
368 u64 wanted_disk_byte, int count,
369 struct share_check *sc, gfp_t gfp_mask)
371 return add_prelim_ref(fs_info, &preftrees->direct, 0, NULL, level,
372 parent, wanted_disk_byte, count, sc, gfp_mask);
375 /* indirect refs use parent == 0 */
376 static int add_indirect_ref(const struct btrfs_fs_info *fs_info,
377 struct preftrees *preftrees, u64 root_id,
378 const struct btrfs_key *key, int level,
379 u64 wanted_disk_byte, int count,
380 struct share_check *sc, gfp_t gfp_mask)
382 struct preftree *tree = &preftrees->indirect;
385 tree = &preftrees->indirect_missing_keys;
386 return add_prelim_ref(fs_info, tree, root_id, key, level, 0,
387 wanted_disk_byte, count, sc, gfp_mask);
390 static int is_shared_data_backref(struct preftrees *preftrees, u64 bytenr)
392 struct rb_node **p = &preftrees->direct.root.rb_root.rb_node;
393 struct rb_node *parent = NULL;
394 struct prelim_ref *ref = NULL;
395 struct prelim_ref target = {};
398 target.parent = bytenr;
402 ref = rb_entry(parent, struct prelim_ref, rbnode);
403 result = prelim_ref_compare(ref, &target);
415 static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path,
416 struct ulist *parents,
417 struct preftrees *preftrees, struct prelim_ref *ref,
418 int level, u64 time_seq, const u64 *extent_item_pos,
423 struct extent_buffer *eb;
424 struct btrfs_key key;
425 struct btrfs_key *key_for_search = &ref->key_for_search;
426 struct btrfs_file_extent_item *fi;
427 struct extent_inode_elem *eie = NULL, *old = NULL;
429 u64 wanted_disk_byte = ref->wanted_disk_byte;
434 eb = path->nodes[level];
435 ret = ulist_add(parents, eb->start, 0, GFP_NOFS);
442 * 1. We normally enter this function with the path already pointing to
443 * the first item to check. But sometimes, we may enter it with
445 * 2. We are searching for normal backref but bytenr of this leaf
446 * matches shared data backref
447 * 3. The leaf owner is not equal to the root we are searching
449 * For these cases, go to the next leaf before we continue.
452 if (path->slots[0] >= btrfs_header_nritems(eb) ||
453 is_shared_data_backref(preftrees, eb->start) ||
454 ref->root_id != btrfs_header_owner(eb)) {
455 if (time_seq == SEQ_LAST)
456 ret = btrfs_next_leaf(root, path);
458 ret = btrfs_next_old_leaf(root, path, time_seq);
461 while (!ret && count < ref->count) {
463 slot = path->slots[0];
465 btrfs_item_key_to_cpu(eb, &key, slot);
467 if (key.objectid != key_for_search->objectid ||
468 key.type != BTRFS_EXTENT_DATA_KEY)
472 * We are searching for normal backref but bytenr of this leaf
473 * matches shared data backref, OR
474 * the leaf owner is not equal to the root we are searching for
477 (is_shared_data_backref(preftrees, eb->start) ||
478 ref->root_id != btrfs_header_owner(eb))) {
479 if (time_seq == SEQ_LAST)
480 ret = btrfs_next_leaf(root, path);
482 ret = btrfs_next_old_leaf(root, path, time_seq);
485 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
486 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
487 data_offset = btrfs_file_extent_offset(eb, fi);
489 if (disk_byte == wanted_disk_byte) {
492 if (ref->key_for_search.offset == key.offset - data_offset)
496 if (extent_item_pos) {
497 ret = check_extent_in_eb(&key, eb, fi,
499 &eie, ignore_offset);
505 ret = ulist_add_merge_ptr(parents, eb->start,
506 eie, (void **)&old, GFP_NOFS);
509 if (!ret && extent_item_pos) {
517 if (time_seq == SEQ_LAST)
518 ret = btrfs_next_item(root, path);
520 ret = btrfs_next_old_item(root, path, time_seq);
526 free_inode_elem_list(eie);
531 * resolve an indirect backref in the form (root_id, key, level)
532 * to a logical address
534 static int resolve_indirect_ref(struct btrfs_fs_info *fs_info,
535 struct btrfs_path *path, u64 time_seq,
536 struct preftrees *preftrees,
537 struct prelim_ref *ref, struct ulist *parents,
538 const u64 *extent_item_pos, bool ignore_offset)
540 struct btrfs_root *root;
541 struct extent_buffer *eb;
544 int level = ref->level;
545 struct btrfs_key search_key = ref->key_for_search;
547 root = btrfs_get_fs_root(fs_info, ref->root_id, false);
553 if (!path->search_commit_root &&
554 test_bit(BTRFS_ROOT_DELETING, &root->state)) {
559 if (btrfs_is_testing(fs_info)) {
564 if (path->search_commit_root)
565 root_level = btrfs_header_level(root->commit_root);
566 else if (time_seq == SEQ_LAST)
567 root_level = btrfs_header_level(root->node);
569 root_level = btrfs_old_root_level(root, time_seq);
571 if (root_level + 1 == level)
575 * We can often find data backrefs with an offset that is too large
576 * (>= LLONG_MAX, maximum allowed file offset) due to underflows when
577 * subtracting a file's offset with the data offset of its
578 * corresponding extent data item. This can happen for example in the
581 * So if we detect such case we set the search key's offset to zero to
582 * make sure we will find the matching file extent item at
583 * add_all_parents(), otherwise we will miss it because the offset
584 * taken form the backref is much larger then the offset of the file
585 * extent item. This can make us scan a very large number of file
586 * extent items, but at least it will not make us miss any.
588 * This is an ugly workaround for a behaviour that should have never
589 * existed, but it does and a fix for the clone ioctl would touch a lot
590 * of places, cause backwards incompatibility and would not fix the
591 * problem for extents cloned with older kernels.
593 if (search_key.type == BTRFS_EXTENT_DATA_KEY &&
594 search_key.offset >= LLONG_MAX)
595 search_key.offset = 0;
596 path->lowest_level = level;
597 if (time_seq == SEQ_LAST)
598 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
600 ret = btrfs_search_old_slot(root, &search_key, path, time_seq);
603 "search slot in root %llu (level %d, ref count %d) returned %d for key (%llu %u %llu)",
604 ref->root_id, level, ref->count, ret,
605 ref->key_for_search.objectid, ref->key_for_search.type,
606 ref->key_for_search.offset);
610 eb = path->nodes[level];
612 if (WARN_ON(!level)) {
617 eb = path->nodes[level];
620 ret = add_all_parents(root, path, parents, preftrees, ref, level,
621 time_seq, extent_item_pos, ignore_offset);
623 btrfs_put_root(root);
625 path->lowest_level = 0;
626 btrfs_release_path(path);
630 static struct extent_inode_elem *
631 unode_aux_to_inode_list(struct ulist_node *node)
635 return (struct extent_inode_elem *)(uintptr_t)node->aux;
639 * We maintain three separate rbtrees: one for direct refs, one for
640 * indirect refs which have a key, and one for indirect refs which do not
641 * have a key. Each tree does merge on insertion.
643 * Once all of the references are located, we iterate over the tree of
644 * indirect refs with missing keys. An appropriate key is located and
645 * the ref is moved onto the tree for indirect refs. After all missing
646 * keys are thus located, we iterate over the indirect ref tree, resolve
647 * each reference, and then insert the resolved reference onto the
648 * direct tree (merging there too).
650 * New backrefs (i.e., for parent nodes) are added to the appropriate
651 * rbtree as they are encountered. The new backrefs are subsequently
654 static int resolve_indirect_refs(struct btrfs_fs_info *fs_info,
655 struct btrfs_path *path, u64 time_seq,
656 struct preftrees *preftrees,
657 const u64 *extent_item_pos,
658 struct share_check *sc, bool ignore_offset)
662 struct ulist *parents;
663 struct ulist_node *node;
664 struct ulist_iterator uiter;
665 struct rb_node *rnode;
667 parents = ulist_alloc(GFP_NOFS);
672 * We could trade memory usage for performance here by iterating
673 * the tree, allocating new refs for each insertion, and then
674 * freeing the entire indirect tree when we're done. In some test
675 * cases, the tree can grow quite large (~200k objects).
677 while ((rnode = rb_first_cached(&preftrees->indirect.root))) {
678 struct prelim_ref *ref;
680 ref = rb_entry(rnode, struct prelim_ref, rbnode);
681 if (WARN(ref->parent,
682 "BUG: direct ref found in indirect tree")) {
687 rb_erase_cached(&ref->rbnode, &preftrees->indirect.root);
688 preftrees->indirect.count--;
690 if (ref->count == 0) {
695 if (sc && sc->root_objectid &&
696 ref->root_id != sc->root_objectid) {
698 ret = BACKREF_FOUND_SHARED;
701 err = resolve_indirect_ref(fs_info, path, time_seq, preftrees,
702 ref, parents, extent_item_pos,
705 * we can only tolerate ENOENT,otherwise,we should catch error
706 * and return directly.
708 if (err == -ENOENT) {
709 prelim_ref_insert(fs_info, &preftrees->direct, ref,
718 /* we put the first parent into the ref at hand */
719 ULIST_ITER_INIT(&uiter);
720 node = ulist_next(parents, &uiter);
721 ref->parent = node ? node->val : 0;
722 ref->inode_list = unode_aux_to_inode_list(node);
724 /* Add a prelim_ref(s) for any other parent(s). */
725 while ((node = ulist_next(parents, &uiter))) {
726 struct prelim_ref *new_ref;
728 new_ref = kmem_cache_alloc(btrfs_prelim_ref_cache,
735 memcpy(new_ref, ref, sizeof(*ref));
736 new_ref->parent = node->val;
737 new_ref->inode_list = unode_aux_to_inode_list(node);
738 prelim_ref_insert(fs_info, &preftrees->direct,
743 * Now it's a direct ref, put it in the direct tree. We must
744 * do this last because the ref could be merged/freed here.
746 prelim_ref_insert(fs_info, &preftrees->direct, ref, NULL);
748 ulist_reinit(parents);
757 * read tree blocks and add keys where required.
759 static int add_missing_keys(struct btrfs_fs_info *fs_info,
760 struct preftrees *preftrees, bool lock)
762 struct prelim_ref *ref;
763 struct extent_buffer *eb;
764 struct preftree *tree = &preftrees->indirect_missing_keys;
765 struct rb_node *node;
767 while ((node = rb_first_cached(&tree->root))) {
768 ref = rb_entry(node, struct prelim_ref, rbnode);
769 rb_erase_cached(node, &tree->root);
771 BUG_ON(ref->parent); /* should not be a direct ref */
772 BUG_ON(ref->key_for_search.type);
773 BUG_ON(!ref->wanted_disk_byte);
775 eb = read_tree_block(fs_info, ref->wanted_disk_byte, 0,
776 ref->level - 1, NULL);
780 } else if (!extent_buffer_uptodate(eb)) {
782 free_extent_buffer(eb);
786 btrfs_tree_read_lock(eb);
787 if (btrfs_header_level(eb) == 0)
788 btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
790 btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
792 btrfs_tree_read_unlock(eb);
793 free_extent_buffer(eb);
794 prelim_ref_insert(fs_info, &preftrees->indirect, ref, NULL);
801 * add all currently queued delayed refs from this head whose seq nr is
802 * smaller or equal that seq to the list
804 static int add_delayed_refs(const struct btrfs_fs_info *fs_info,
805 struct btrfs_delayed_ref_head *head, u64 seq,
806 struct preftrees *preftrees, struct share_check *sc)
808 struct btrfs_delayed_ref_node *node;
809 struct btrfs_delayed_extent_op *extent_op = head->extent_op;
810 struct btrfs_key key;
811 struct btrfs_key tmp_op_key;
816 if (extent_op && extent_op->update_key)
817 btrfs_disk_key_to_cpu(&tmp_op_key, &extent_op->key);
819 spin_lock(&head->lock);
820 for (n = rb_first_cached(&head->ref_tree); n; n = rb_next(n)) {
821 node = rb_entry(n, struct btrfs_delayed_ref_node,
826 switch (node->action) {
827 case BTRFS_ADD_DELAYED_EXTENT:
828 case BTRFS_UPDATE_DELAYED_HEAD:
831 case BTRFS_ADD_DELAYED_REF:
832 count = node->ref_mod;
834 case BTRFS_DROP_DELAYED_REF:
835 count = node->ref_mod * -1;
840 switch (node->type) {
841 case BTRFS_TREE_BLOCK_REF_KEY: {
842 /* NORMAL INDIRECT METADATA backref */
843 struct btrfs_delayed_tree_ref *ref;
845 ref = btrfs_delayed_node_to_tree_ref(node);
846 ret = add_indirect_ref(fs_info, preftrees, ref->root,
847 &tmp_op_key, ref->level + 1,
848 node->bytenr, count, sc,
852 case BTRFS_SHARED_BLOCK_REF_KEY: {
853 /* SHARED DIRECT METADATA backref */
854 struct btrfs_delayed_tree_ref *ref;
856 ref = btrfs_delayed_node_to_tree_ref(node);
858 ret = add_direct_ref(fs_info, preftrees, ref->level + 1,
859 ref->parent, node->bytenr, count,
863 case BTRFS_EXTENT_DATA_REF_KEY: {
864 /* NORMAL INDIRECT DATA backref */
865 struct btrfs_delayed_data_ref *ref;
866 ref = btrfs_delayed_node_to_data_ref(node);
868 key.objectid = ref->objectid;
869 key.type = BTRFS_EXTENT_DATA_KEY;
870 key.offset = ref->offset;
873 * Found a inum that doesn't match our known inum, we
876 if (sc && sc->inum && ref->objectid != sc->inum) {
877 ret = BACKREF_FOUND_SHARED;
881 ret = add_indirect_ref(fs_info, preftrees, ref->root,
882 &key, 0, node->bytenr, count, sc,
886 case BTRFS_SHARED_DATA_REF_KEY: {
887 /* SHARED DIRECT FULL backref */
888 struct btrfs_delayed_data_ref *ref;
890 ref = btrfs_delayed_node_to_data_ref(node);
892 ret = add_direct_ref(fs_info, preftrees, 0, ref->parent,
893 node->bytenr, count, sc,
901 * We must ignore BACKREF_FOUND_SHARED until all delayed
902 * refs have been checked.
904 if (ret && (ret != BACKREF_FOUND_SHARED))
908 ret = extent_is_shared(sc);
910 spin_unlock(&head->lock);
915 * add all inline backrefs for bytenr to the list
917 * Returns 0 on success, <0 on error, or BACKREF_FOUND_SHARED.
919 static int add_inline_refs(const struct btrfs_fs_info *fs_info,
920 struct btrfs_path *path, u64 bytenr,
921 int *info_level, struct preftrees *preftrees,
922 struct share_check *sc)
926 struct extent_buffer *leaf;
927 struct btrfs_key key;
928 struct btrfs_key found_key;
931 struct btrfs_extent_item *ei;
936 * enumerate all inline refs
938 leaf = path->nodes[0];
939 slot = path->slots[0];
941 item_size = btrfs_item_size_nr(leaf, slot);
942 BUG_ON(item_size < sizeof(*ei));
944 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
945 flags = btrfs_extent_flags(leaf, ei);
946 btrfs_item_key_to_cpu(leaf, &found_key, slot);
948 ptr = (unsigned long)(ei + 1);
949 end = (unsigned long)ei + item_size;
951 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
952 flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
953 struct btrfs_tree_block_info *info;
955 info = (struct btrfs_tree_block_info *)ptr;
956 *info_level = btrfs_tree_block_level(leaf, info);
957 ptr += sizeof(struct btrfs_tree_block_info);
959 } else if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
960 *info_level = found_key.offset;
962 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
966 struct btrfs_extent_inline_ref *iref;
970 iref = (struct btrfs_extent_inline_ref *)ptr;
971 type = btrfs_get_extent_inline_ref_type(leaf, iref,
973 if (type == BTRFS_REF_TYPE_INVALID)
976 offset = btrfs_extent_inline_ref_offset(leaf, iref);
979 case BTRFS_SHARED_BLOCK_REF_KEY:
980 ret = add_direct_ref(fs_info, preftrees,
981 *info_level + 1, offset,
982 bytenr, 1, NULL, GFP_NOFS);
984 case BTRFS_SHARED_DATA_REF_KEY: {
985 struct btrfs_shared_data_ref *sdref;
988 sdref = (struct btrfs_shared_data_ref *)(iref + 1);
989 count = btrfs_shared_data_ref_count(leaf, sdref);
991 ret = add_direct_ref(fs_info, preftrees, 0, offset,
992 bytenr, count, sc, GFP_NOFS);
995 case BTRFS_TREE_BLOCK_REF_KEY:
996 ret = add_indirect_ref(fs_info, preftrees, offset,
997 NULL, *info_level + 1,
998 bytenr, 1, NULL, GFP_NOFS);
1000 case BTRFS_EXTENT_DATA_REF_KEY: {
1001 struct btrfs_extent_data_ref *dref;
1005 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1006 count = btrfs_extent_data_ref_count(leaf, dref);
1007 key.objectid = btrfs_extent_data_ref_objectid(leaf,
1009 key.type = BTRFS_EXTENT_DATA_KEY;
1010 key.offset = btrfs_extent_data_ref_offset(leaf, dref);
1012 if (sc && sc->inum && key.objectid != sc->inum) {
1013 ret = BACKREF_FOUND_SHARED;
1017 root = btrfs_extent_data_ref_root(leaf, dref);
1019 ret = add_indirect_ref(fs_info, preftrees, root,
1020 &key, 0, bytenr, count,
1029 ptr += btrfs_extent_inline_ref_size(type);
1036 * add all non-inline backrefs for bytenr to the list
1038 * Returns 0 on success, <0 on error, or BACKREF_FOUND_SHARED.
1040 static int add_keyed_refs(struct btrfs_fs_info *fs_info,
1041 struct btrfs_path *path, u64 bytenr,
1042 int info_level, struct preftrees *preftrees,
1043 struct share_check *sc)
1045 struct btrfs_root *extent_root = fs_info->extent_root;
1048 struct extent_buffer *leaf;
1049 struct btrfs_key key;
1052 ret = btrfs_next_item(extent_root, path);
1060 slot = path->slots[0];
1061 leaf = path->nodes[0];
1062 btrfs_item_key_to_cpu(leaf, &key, slot);
1064 if (key.objectid != bytenr)
1066 if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
1068 if (key.type > BTRFS_SHARED_DATA_REF_KEY)
1072 case BTRFS_SHARED_BLOCK_REF_KEY:
1073 /* SHARED DIRECT METADATA backref */
1074 ret = add_direct_ref(fs_info, preftrees,
1075 info_level + 1, key.offset,
1076 bytenr, 1, NULL, GFP_NOFS);
1078 case BTRFS_SHARED_DATA_REF_KEY: {
1079 /* SHARED DIRECT FULL backref */
1080 struct btrfs_shared_data_ref *sdref;
1083 sdref = btrfs_item_ptr(leaf, slot,
1084 struct btrfs_shared_data_ref);
1085 count = btrfs_shared_data_ref_count(leaf, sdref);
1086 ret = add_direct_ref(fs_info, preftrees, 0,
1087 key.offset, bytenr, count,
1091 case BTRFS_TREE_BLOCK_REF_KEY:
1092 /* NORMAL INDIRECT METADATA backref */
1093 ret = add_indirect_ref(fs_info, preftrees, key.offset,
1094 NULL, info_level + 1, bytenr,
1097 case BTRFS_EXTENT_DATA_REF_KEY: {
1098 /* NORMAL INDIRECT DATA backref */
1099 struct btrfs_extent_data_ref *dref;
1103 dref = btrfs_item_ptr(leaf, slot,
1104 struct btrfs_extent_data_ref);
1105 count = btrfs_extent_data_ref_count(leaf, dref);
1106 key.objectid = btrfs_extent_data_ref_objectid(leaf,
1108 key.type = BTRFS_EXTENT_DATA_KEY;
1109 key.offset = btrfs_extent_data_ref_offset(leaf, dref);
1111 if (sc && sc->inum && key.objectid != sc->inum) {
1112 ret = BACKREF_FOUND_SHARED;
1116 root = btrfs_extent_data_ref_root(leaf, dref);
1117 ret = add_indirect_ref(fs_info, preftrees, root,
1118 &key, 0, bytenr, count,
1134 * this adds all existing backrefs (inline backrefs, backrefs and delayed
1135 * refs) for the given bytenr to the refs list, merges duplicates and resolves
1136 * indirect refs to their parent bytenr.
1137 * When roots are found, they're added to the roots list
1139 * If time_seq is set to SEQ_LAST, it will not search delayed_refs, and behave
1140 * much like trans == NULL case, the difference only lies in it will not
1142 * The special case is for qgroup to search roots in commit_transaction().
1144 * @sc - if !NULL, then immediately return BACKREF_FOUND_SHARED when a
1145 * shared extent is detected.
1147 * Otherwise this returns 0 for success and <0 for an error.
1149 * If ignore_offset is set to false, only extent refs whose offsets match
1150 * extent_item_pos are returned. If true, every extent ref is returned
1151 * and extent_item_pos is ignored.
1153 * FIXME some caching might speed things up
1155 static int find_parent_nodes(struct btrfs_trans_handle *trans,
1156 struct btrfs_fs_info *fs_info, u64 bytenr,
1157 u64 time_seq, struct ulist *refs,
1158 struct ulist *roots, const u64 *extent_item_pos,
1159 struct share_check *sc, bool ignore_offset)
1161 struct btrfs_key key;
1162 struct btrfs_path *path;
1163 struct btrfs_delayed_ref_root *delayed_refs = NULL;
1164 struct btrfs_delayed_ref_head *head;
1167 struct prelim_ref *ref;
1168 struct rb_node *node;
1169 struct extent_inode_elem *eie = NULL;
1170 struct preftrees preftrees = {
1171 .direct = PREFTREE_INIT,
1172 .indirect = PREFTREE_INIT,
1173 .indirect_missing_keys = PREFTREE_INIT
1176 key.objectid = bytenr;
1177 key.offset = (u64)-1;
1178 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1179 key.type = BTRFS_METADATA_ITEM_KEY;
1181 key.type = BTRFS_EXTENT_ITEM_KEY;
1183 path = btrfs_alloc_path();
1187 path->search_commit_root = 1;
1188 path->skip_locking = 1;
1191 if (time_seq == SEQ_LAST)
1192 path->skip_locking = 1;
1195 * grab both a lock on the path and a lock on the delayed ref head.
1196 * We need both to get a consistent picture of how the refs look
1197 * at a specified point in time
1202 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
1207 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1208 if (trans && likely(trans->type != __TRANS_DUMMY) &&
1209 time_seq != SEQ_LAST) {
1211 if (trans && time_seq != SEQ_LAST) {
1214 * look if there are updates for this ref queued and lock the
1217 delayed_refs = &trans->transaction->delayed_refs;
1218 spin_lock(&delayed_refs->lock);
1219 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
1221 if (!mutex_trylock(&head->mutex)) {
1222 refcount_inc(&head->refs);
1223 spin_unlock(&delayed_refs->lock);
1225 btrfs_release_path(path);
1228 * Mutex was contended, block until it's
1229 * released and try again
1231 mutex_lock(&head->mutex);
1232 mutex_unlock(&head->mutex);
1233 btrfs_put_delayed_ref_head(head);
1236 spin_unlock(&delayed_refs->lock);
1237 ret = add_delayed_refs(fs_info, head, time_seq,
1239 mutex_unlock(&head->mutex);
1243 spin_unlock(&delayed_refs->lock);
1247 if (path->slots[0]) {
1248 struct extent_buffer *leaf;
1252 leaf = path->nodes[0];
1253 slot = path->slots[0];
1254 btrfs_item_key_to_cpu(leaf, &key, slot);
1255 if (key.objectid == bytenr &&
1256 (key.type == BTRFS_EXTENT_ITEM_KEY ||
1257 key.type == BTRFS_METADATA_ITEM_KEY)) {
1258 ret = add_inline_refs(fs_info, path, bytenr,
1259 &info_level, &preftrees, sc);
1262 ret = add_keyed_refs(fs_info, path, bytenr, info_level,
1269 btrfs_release_path(path);
1271 ret = add_missing_keys(fs_info, &preftrees, path->skip_locking == 0);
1275 WARN_ON(!RB_EMPTY_ROOT(&preftrees.indirect_missing_keys.root.rb_root));
1277 ret = resolve_indirect_refs(fs_info, path, time_seq, &preftrees,
1278 extent_item_pos, sc, ignore_offset);
1282 WARN_ON(!RB_EMPTY_ROOT(&preftrees.indirect.root.rb_root));
1285 * This walks the tree of merged and resolved refs. Tree blocks are
1286 * read in as needed. Unique entries are added to the ulist, and
1287 * the list of found roots is updated.
1289 * We release the entire tree in one go before returning.
1291 node = rb_first_cached(&preftrees.direct.root);
1293 ref = rb_entry(node, struct prelim_ref, rbnode);
1294 node = rb_next(&ref->rbnode);
1296 * ref->count < 0 can happen here if there are delayed
1297 * refs with a node->action of BTRFS_DROP_DELAYED_REF.
1298 * prelim_ref_insert() relies on this when merging
1299 * identical refs to keep the overall count correct.
1300 * prelim_ref_insert() will merge only those refs
1301 * which compare identically. Any refs having
1302 * e.g. different offsets would not be merged,
1303 * and would retain their original ref->count < 0.
1305 if (roots && ref->count && ref->root_id && ref->parent == 0) {
1306 if (sc && sc->root_objectid &&
1307 ref->root_id != sc->root_objectid) {
1308 ret = BACKREF_FOUND_SHARED;
1312 /* no parent == root of tree */
1313 ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS);
1317 if (ref->count && ref->parent) {
1318 if (extent_item_pos && !ref->inode_list &&
1320 struct extent_buffer *eb;
1322 eb = read_tree_block(fs_info, ref->parent, 0,
1327 } else if (!extent_buffer_uptodate(eb)) {
1328 free_extent_buffer(eb);
1333 if (!path->skip_locking) {
1334 btrfs_tree_read_lock(eb);
1335 btrfs_set_lock_blocking_read(eb);
1337 ret = find_extent_in_eb(eb, bytenr,
1338 *extent_item_pos, &eie, ignore_offset);
1339 if (!path->skip_locking)
1340 btrfs_tree_read_unlock_blocking(eb);
1341 free_extent_buffer(eb);
1344 ref->inode_list = eie;
1346 ret = ulist_add_merge_ptr(refs, ref->parent,
1348 (void **)&eie, GFP_NOFS);
1351 if (!ret && extent_item_pos) {
1353 * we've recorded that parent, so we must extend
1354 * its inode list here
1359 eie->next = ref->inode_list;
1367 btrfs_free_path(path);
1369 prelim_release(&preftrees.direct);
1370 prelim_release(&preftrees.indirect);
1371 prelim_release(&preftrees.indirect_missing_keys);
1374 free_inode_elem_list(eie);
1378 static void free_leaf_list(struct ulist *blocks)
1380 struct ulist_node *node = NULL;
1381 struct extent_inode_elem *eie;
1382 struct ulist_iterator uiter;
1384 ULIST_ITER_INIT(&uiter);
1385 while ((node = ulist_next(blocks, &uiter))) {
1388 eie = unode_aux_to_inode_list(node);
1389 free_inode_elem_list(eie);
1397 * Finds all leafs with a reference to the specified combination of bytenr and
1398 * offset. key_list_head will point to a list of corresponding keys (caller must
1399 * free each list element). The leafs will be stored in the leafs ulist, which
1400 * must be freed with ulist_free.
1402 * returns 0 on success, <0 on error
1404 int btrfs_find_all_leafs(struct btrfs_trans_handle *trans,
1405 struct btrfs_fs_info *fs_info, u64 bytenr,
1406 u64 time_seq, struct ulist **leafs,
1407 const u64 *extent_item_pos, bool ignore_offset)
1411 *leafs = ulist_alloc(GFP_NOFS);
1415 ret = find_parent_nodes(trans, fs_info, bytenr, time_seq,
1416 *leafs, NULL, extent_item_pos, NULL, ignore_offset);
1417 if (ret < 0 && ret != -ENOENT) {
1418 free_leaf_list(*leafs);
1426 * walk all backrefs for a given extent to find all roots that reference this
1427 * extent. Walking a backref means finding all extents that reference this
1428 * extent and in turn walk the backrefs of those, too. Naturally this is a
1429 * recursive process, but here it is implemented in an iterative fashion: We
1430 * find all referencing extents for the extent in question and put them on a
1431 * list. In turn, we find all referencing extents for those, further appending
1432 * to the list. The way we iterate the list allows adding more elements after
1433 * the current while iterating. The process stops when we reach the end of the
1434 * list. Found roots are added to the roots list.
1436 * returns 0 on success, < 0 on error.
1438 static int btrfs_find_all_roots_safe(struct btrfs_trans_handle *trans,
1439 struct btrfs_fs_info *fs_info, u64 bytenr,
1440 u64 time_seq, struct ulist **roots,
1444 struct ulist_node *node = NULL;
1445 struct ulist_iterator uiter;
1448 tmp = ulist_alloc(GFP_NOFS);
1451 *roots = ulist_alloc(GFP_NOFS);
1457 ULIST_ITER_INIT(&uiter);
1459 ret = find_parent_nodes(trans, fs_info, bytenr, time_seq,
1460 tmp, *roots, NULL, NULL, ignore_offset);
1461 if (ret < 0 && ret != -ENOENT) {
1466 node = ulist_next(tmp, &uiter);
1477 int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1478 struct btrfs_fs_info *fs_info, u64 bytenr,
1479 u64 time_seq, struct ulist **roots,
1485 down_read(&fs_info->commit_root_sem);
1486 ret = btrfs_find_all_roots_safe(trans, fs_info, bytenr,
1487 time_seq, roots, ignore_offset);
1489 up_read(&fs_info->commit_root_sem);
1494 * btrfs_check_shared - tell us whether an extent is shared
1496 * btrfs_check_shared uses the backref walking code but will short
1497 * circuit as soon as it finds a root or inode that doesn't match the
1498 * one passed in. This provides a significant performance benefit for
1499 * callers (such as fiemap) which want to know whether the extent is
1500 * shared but do not need a ref count.
1502 * This attempts to attach to the running transaction in order to account for
1503 * delayed refs, but continues on even when no running transaction exists.
1505 * Return: 0 if extent is not shared, 1 if it is shared, < 0 on error.
1507 int btrfs_check_shared(struct btrfs_root *root, u64 inum, u64 bytenr,
1508 struct ulist *roots, struct ulist *tmp)
1510 struct btrfs_fs_info *fs_info = root->fs_info;
1511 struct btrfs_trans_handle *trans;
1512 struct ulist_iterator uiter;
1513 struct ulist_node *node;
1514 struct seq_list elem = SEQ_LIST_INIT(elem);
1516 struct share_check shared = {
1517 .root_objectid = root->root_key.objectid,
1525 trans = btrfs_join_transaction_nostart(root);
1526 if (IS_ERR(trans)) {
1527 if (PTR_ERR(trans) != -ENOENT && PTR_ERR(trans) != -EROFS) {
1528 ret = PTR_ERR(trans);
1532 down_read(&fs_info->commit_root_sem);
1534 btrfs_get_tree_mod_seq(fs_info, &elem);
1537 ULIST_ITER_INIT(&uiter);
1539 ret = find_parent_nodes(trans, fs_info, bytenr, elem.seq, tmp,
1540 roots, NULL, &shared, false);
1541 if (ret == BACKREF_FOUND_SHARED) {
1542 /* this is the only condition under which we return 1 */
1546 if (ret < 0 && ret != -ENOENT)
1549 node = ulist_next(tmp, &uiter);
1553 shared.share_count = 0;
1558 btrfs_put_tree_mod_seq(fs_info, &elem);
1559 btrfs_end_transaction(trans);
1561 up_read(&fs_info->commit_root_sem);
1564 ulist_release(roots);
1569 int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,
1570 u64 start_off, struct btrfs_path *path,
1571 struct btrfs_inode_extref **ret_extref,
1575 struct btrfs_key key;
1576 struct btrfs_key found_key;
1577 struct btrfs_inode_extref *extref;
1578 const struct extent_buffer *leaf;
1581 key.objectid = inode_objectid;
1582 key.type = BTRFS_INODE_EXTREF_KEY;
1583 key.offset = start_off;
1585 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1590 leaf = path->nodes[0];
1591 slot = path->slots[0];
1592 if (slot >= btrfs_header_nritems(leaf)) {
1594 * If the item at offset is not found,
1595 * btrfs_search_slot will point us to the slot
1596 * where it should be inserted. In our case
1597 * that will be the slot directly before the
1598 * next INODE_REF_KEY_V2 item. In the case
1599 * that we're pointing to the last slot in a
1600 * leaf, we must move one leaf over.
1602 ret = btrfs_next_leaf(root, path);
1611 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1614 * Check that we're still looking at an extended ref key for
1615 * this particular objectid. If we have different
1616 * objectid or type then there are no more to be found
1617 * in the tree and we can exit.
1620 if (found_key.objectid != inode_objectid)
1622 if (found_key.type != BTRFS_INODE_EXTREF_KEY)
1626 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1627 extref = (struct btrfs_inode_extref *)ptr;
1628 *ret_extref = extref;
1630 *found_off = found_key.offset;
1638 * this iterates to turn a name (from iref/extref) into a full filesystem path.
1639 * Elements of the path are separated by '/' and the path is guaranteed to be
1640 * 0-terminated. the path is only given within the current file system.
1641 * Therefore, it never starts with a '/'. the caller is responsible to provide
1642 * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
1643 * the start point of the resulting string is returned. this pointer is within
1645 * in case the path buffer would overflow, the pointer is decremented further
1646 * as if output was written to the buffer, though no more output is actually
1647 * generated. that way, the caller can determine how much space would be
1648 * required for the path to fit into the buffer. in that case, the returned
1649 * value will be smaller than dest. callers must check this!
1651 char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
1652 u32 name_len, unsigned long name_off,
1653 struct extent_buffer *eb_in, u64 parent,
1654 char *dest, u32 size)
1659 s64 bytes_left = ((s64)size) - 1;
1660 struct extent_buffer *eb = eb_in;
1661 struct btrfs_key found_key;
1662 int leave_spinning = path->leave_spinning;
1663 struct btrfs_inode_ref *iref;
1665 if (bytes_left >= 0)
1666 dest[bytes_left] = '\0';
1668 path->leave_spinning = 1;
1670 bytes_left -= name_len;
1671 if (bytes_left >= 0)
1672 read_extent_buffer(eb, dest + bytes_left,
1673 name_off, name_len);
1675 if (!path->skip_locking)
1676 btrfs_tree_read_unlock_blocking(eb);
1677 free_extent_buffer(eb);
1679 ret = btrfs_find_item(fs_root, path, parent, 0,
1680 BTRFS_INODE_REF_KEY, &found_key);
1686 next_inum = found_key.offset;
1688 /* regular exit ahead */
1689 if (parent == next_inum)
1692 slot = path->slots[0];
1693 eb = path->nodes[0];
1694 /* make sure we can use eb after releasing the path */
1696 if (!path->skip_locking)
1697 btrfs_set_lock_blocking_read(eb);
1698 path->nodes[0] = NULL;
1701 btrfs_release_path(path);
1702 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1704 name_len = btrfs_inode_ref_name_len(eb, iref);
1705 name_off = (unsigned long)(iref + 1);
1709 if (bytes_left >= 0)
1710 dest[bytes_left] = '/';
1713 btrfs_release_path(path);
1714 path->leave_spinning = leave_spinning;
1717 return ERR_PTR(ret);
1719 return dest + bytes_left;
1723 * this makes the path point to (logical EXTENT_ITEM *)
1724 * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
1725 * tree blocks and <0 on error.
1727 int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
1728 struct btrfs_path *path, struct btrfs_key *found_key,
1735 const struct extent_buffer *eb;
1736 struct btrfs_extent_item *ei;
1737 struct btrfs_key key;
1739 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1740 key.type = BTRFS_METADATA_ITEM_KEY;
1742 key.type = BTRFS_EXTENT_ITEM_KEY;
1743 key.objectid = logical;
1744 key.offset = (u64)-1;
1746 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
1750 ret = btrfs_previous_extent_item(fs_info->extent_root, path, 0);
1756 btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]);
1757 if (found_key->type == BTRFS_METADATA_ITEM_KEY)
1758 size = fs_info->nodesize;
1759 else if (found_key->type == BTRFS_EXTENT_ITEM_KEY)
1760 size = found_key->offset;
1762 if (found_key->objectid > logical ||
1763 found_key->objectid + size <= logical) {
1764 btrfs_debug(fs_info,
1765 "logical %llu is not within any extent", logical);
1769 eb = path->nodes[0];
1770 item_size = btrfs_item_size_nr(eb, path->slots[0]);
1771 BUG_ON(item_size < sizeof(*ei));
1773 ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
1774 flags = btrfs_extent_flags(eb, ei);
1776 btrfs_debug(fs_info,
1777 "logical %llu is at position %llu within the extent (%llu EXTENT_ITEM %llu) flags %#llx size %u",
1778 logical, logical - found_key->objectid, found_key->objectid,
1779 found_key->offset, flags, item_size);
1781 WARN_ON(!flags_ret);
1783 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1784 *flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK;
1785 else if (flags & BTRFS_EXTENT_FLAG_DATA)
1786 *flags_ret = BTRFS_EXTENT_FLAG_DATA;
1796 * helper function to iterate extent inline refs. ptr must point to a 0 value
1797 * for the first call and may be modified. it is used to track state.
1798 * if more refs exist, 0 is returned and the next call to
1799 * get_extent_inline_ref must pass the modified ptr parameter to get the
1800 * next ref. after the last ref was processed, 1 is returned.
1801 * returns <0 on error
1803 static int get_extent_inline_ref(unsigned long *ptr,
1804 const struct extent_buffer *eb,
1805 const struct btrfs_key *key,
1806 const struct btrfs_extent_item *ei,
1808 struct btrfs_extent_inline_ref **out_eiref,
1813 struct btrfs_tree_block_info *info;
1817 flags = btrfs_extent_flags(eb, ei);
1818 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1819 if (key->type == BTRFS_METADATA_ITEM_KEY) {
1820 /* a skinny metadata extent */
1822 (struct btrfs_extent_inline_ref *)(ei + 1);
1824 WARN_ON(key->type != BTRFS_EXTENT_ITEM_KEY);
1825 info = (struct btrfs_tree_block_info *)(ei + 1);
1827 (struct btrfs_extent_inline_ref *)(info + 1);
1830 *out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1);
1832 *ptr = (unsigned long)*out_eiref;
1833 if ((unsigned long)(*ptr) >= (unsigned long)ei + item_size)
1837 end = (unsigned long)ei + item_size;
1838 *out_eiref = (struct btrfs_extent_inline_ref *)(*ptr);
1839 *out_type = btrfs_get_extent_inline_ref_type(eb, *out_eiref,
1840 BTRFS_REF_TYPE_ANY);
1841 if (*out_type == BTRFS_REF_TYPE_INVALID)
1844 *ptr += btrfs_extent_inline_ref_size(*out_type);
1845 WARN_ON(*ptr > end);
1847 return 1; /* last */
1853 * reads the tree block backref for an extent. tree level and root are returned
1854 * through out_level and out_root. ptr must point to a 0 value for the first
1855 * call and may be modified (see get_extent_inline_ref comment).
1856 * returns 0 if data was provided, 1 if there was no more data to provide or
1859 int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
1860 struct btrfs_key *key, struct btrfs_extent_item *ei,
1861 u32 item_size, u64 *out_root, u8 *out_level)
1865 struct btrfs_extent_inline_ref *eiref;
1867 if (*ptr == (unsigned long)-1)
1871 ret = get_extent_inline_ref(ptr, eb, key, ei, item_size,
1876 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1877 type == BTRFS_SHARED_BLOCK_REF_KEY)
1884 /* we can treat both ref types equally here */
1885 *out_root = btrfs_extent_inline_ref_offset(eb, eiref);
1887 if (key->type == BTRFS_EXTENT_ITEM_KEY) {
1888 struct btrfs_tree_block_info *info;
1890 info = (struct btrfs_tree_block_info *)(ei + 1);
1891 *out_level = btrfs_tree_block_level(eb, info);
1893 ASSERT(key->type == BTRFS_METADATA_ITEM_KEY);
1894 *out_level = (u8)key->offset;
1898 *ptr = (unsigned long)-1;
1903 static int iterate_leaf_refs(struct btrfs_fs_info *fs_info,
1904 struct extent_inode_elem *inode_list,
1905 u64 root, u64 extent_item_objectid,
1906 iterate_extent_inodes_t *iterate, void *ctx)
1908 struct extent_inode_elem *eie;
1911 for (eie = inode_list; eie; eie = eie->next) {
1912 btrfs_debug(fs_info,
1913 "ref for %llu resolved, key (%llu EXTEND_DATA %llu), root %llu",
1914 extent_item_objectid, eie->inum,
1916 ret = iterate(eie->inum, eie->offset, root, ctx);
1918 btrfs_debug(fs_info,
1919 "stopping iteration for %llu due to ret=%d",
1920 extent_item_objectid, ret);
1929 * calls iterate() for every inode that references the extent identified by
1930 * the given parameters.
1931 * when the iterator function returns a non-zero value, iteration stops.
1933 int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
1934 u64 extent_item_objectid, u64 extent_item_pos,
1935 int search_commit_root,
1936 iterate_extent_inodes_t *iterate, void *ctx,
1940 struct btrfs_trans_handle *trans = NULL;
1941 struct ulist *refs = NULL;
1942 struct ulist *roots = NULL;
1943 struct ulist_node *ref_node = NULL;
1944 struct ulist_node *root_node = NULL;
1945 struct seq_list tree_mod_seq_elem = SEQ_LIST_INIT(tree_mod_seq_elem);
1946 struct ulist_iterator ref_uiter;
1947 struct ulist_iterator root_uiter;
1949 btrfs_debug(fs_info, "resolving all inodes for extent %llu",
1950 extent_item_objectid);
1952 if (!search_commit_root) {
1953 trans = btrfs_attach_transaction(fs_info->extent_root);
1954 if (IS_ERR(trans)) {
1955 if (PTR_ERR(trans) != -ENOENT &&
1956 PTR_ERR(trans) != -EROFS)
1957 return PTR_ERR(trans);
1963 btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1965 down_read(&fs_info->commit_root_sem);
1967 ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid,
1968 tree_mod_seq_elem.seq, &refs,
1969 &extent_item_pos, ignore_offset);
1973 ULIST_ITER_INIT(&ref_uiter);
1974 while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) {
1975 ret = btrfs_find_all_roots_safe(trans, fs_info, ref_node->val,
1976 tree_mod_seq_elem.seq, &roots,
1980 ULIST_ITER_INIT(&root_uiter);
1981 while (!ret && (root_node = ulist_next(roots, &root_uiter))) {
1982 btrfs_debug(fs_info,
1983 "root %llu references leaf %llu, data list %#llx",
1984 root_node->val, ref_node->val,
1986 ret = iterate_leaf_refs(fs_info,
1987 (struct extent_inode_elem *)
1988 (uintptr_t)ref_node->aux,
1990 extent_item_objectid,
1996 free_leaf_list(refs);
1999 btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
2000 btrfs_end_transaction(trans);
2002 up_read(&fs_info->commit_root_sem);
2008 int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
2009 struct btrfs_path *path,
2010 iterate_extent_inodes_t *iterate, void *ctx,
2014 u64 extent_item_pos;
2016 struct btrfs_key found_key;
2017 int search_commit_root = path->search_commit_root;
2019 ret = extent_from_logical(fs_info, logical, path, &found_key, &flags);
2020 btrfs_release_path(path);
2023 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
2026 extent_item_pos = logical - found_key.objectid;
2027 ret = iterate_extent_inodes(fs_info, found_key.objectid,
2028 extent_item_pos, search_commit_root,
2029 iterate, ctx, ignore_offset);
2034 typedef int (iterate_irefs_t)(u64 parent, u32 name_len, unsigned long name_off,
2035 struct extent_buffer *eb, void *ctx);
2037 static int iterate_inode_refs(u64 inum, struct btrfs_root *fs_root,
2038 struct btrfs_path *path,
2039 iterate_irefs_t *iterate, void *ctx)
2048 struct extent_buffer *eb;
2049 struct btrfs_item *item;
2050 struct btrfs_inode_ref *iref;
2051 struct btrfs_key found_key;
2054 ret = btrfs_find_item(fs_root, path, inum,
2055 parent ? parent + 1 : 0, BTRFS_INODE_REF_KEY,
2061 ret = found ? 0 : -ENOENT;
2066 parent = found_key.offset;
2067 slot = path->slots[0];
2068 eb = btrfs_clone_extent_buffer(path->nodes[0]);
2073 btrfs_release_path(path);
2075 item = btrfs_item_nr(slot);
2076 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
2078 for (cur = 0; cur < btrfs_item_size(eb, item); cur += len) {
2079 name_len = btrfs_inode_ref_name_len(eb, iref);
2080 /* path must be released before calling iterate()! */
2081 btrfs_debug(fs_root->fs_info,
2082 "following ref at offset %u for inode %llu in tree %llu",
2083 cur, found_key.objectid,
2084 fs_root->root_key.objectid);
2085 ret = iterate(parent, name_len,
2086 (unsigned long)(iref + 1), eb, ctx);
2089 len = sizeof(*iref) + name_len;
2090 iref = (struct btrfs_inode_ref *)((char *)iref + len);
2092 free_extent_buffer(eb);
2095 btrfs_release_path(path);
2100 static int iterate_inode_extrefs(u64 inum, struct btrfs_root *fs_root,
2101 struct btrfs_path *path,
2102 iterate_irefs_t *iterate, void *ctx)
2109 struct extent_buffer *eb;
2110 struct btrfs_inode_extref *extref;
2116 ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref,
2121 ret = found ? 0 : -ENOENT;
2126 slot = path->slots[0];
2127 eb = btrfs_clone_extent_buffer(path->nodes[0]);
2132 btrfs_release_path(path);
2134 item_size = btrfs_item_size_nr(eb, slot);
2135 ptr = btrfs_item_ptr_offset(eb, slot);
2138 while (cur_offset < item_size) {
2141 extref = (struct btrfs_inode_extref *)(ptr + cur_offset);
2142 parent = btrfs_inode_extref_parent(eb, extref);
2143 name_len = btrfs_inode_extref_name_len(eb, extref);
2144 ret = iterate(parent, name_len,
2145 (unsigned long)&extref->name, eb, ctx);
2149 cur_offset += btrfs_inode_extref_name_len(eb, extref);
2150 cur_offset += sizeof(*extref);
2152 free_extent_buffer(eb);
2157 btrfs_release_path(path);
2162 static int iterate_irefs(u64 inum, struct btrfs_root *fs_root,
2163 struct btrfs_path *path, iterate_irefs_t *iterate,
2169 ret = iterate_inode_refs(inum, fs_root, path, iterate, ctx);
2172 else if (ret != -ENOENT)
2175 ret = iterate_inode_extrefs(inum, fs_root, path, iterate, ctx);
2176 if (ret == -ENOENT && found_refs)
2183 * returns 0 if the path could be dumped (probably truncated)
2184 * returns <0 in case of an error
2186 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
2187 struct extent_buffer *eb, void *ctx)
2189 struct inode_fs_paths *ipath = ctx;
2192 int i = ipath->fspath->elem_cnt;
2193 const int s_ptr = sizeof(char *);
2196 bytes_left = ipath->fspath->bytes_left > s_ptr ?
2197 ipath->fspath->bytes_left - s_ptr : 0;
2199 fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
2200 fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len,
2201 name_off, eb, inum, fspath_min, bytes_left);
2203 return PTR_ERR(fspath);
2205 if (fspath > fspath_min) {
2206 ipath->fspath->val[i] = (u64)(unsigned long)fspath;
2207 ++ipath->fspath->elem_cnt;
2208 ipath->fspath->bytes_left = fspath - fspath_min;
2210 ++ipath->fspath->elem_missed;
2211 ipath->fspath->bytes_missing += fspath_min - fspath;
2212 ipath->fspath->bytes_left = 0;
2219 * this dumps all file system paths to the inode into the ipath struct, provided
2220 * is has been created large enough. each path is zero-terminated and accessed
2221 * from ipath->fspath->val[i].
2222 * when it returns, there are ipath->fspath->elem_cnt number of paths available
2223 * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the
2224 * number of missed paths is recorded in ipath->fspath->elem_missed, otherwise,
2225 * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
2226 * have been needed to return all paths.
2228 int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
2230 return iterate_irefs(inum, ipath->fs_root, ipath->btrfs_path,
2231 inode_to_path, ipath);
2234 struct btrfs_data_container *init_data_container(u32 total_bytes)
2236 struct btrfs_data_container *data;
2239 alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
2240 data = kvmalloc(alloc_bytes, GFP_KERNEL);
2242 return ERR_PTR(-ENOMEM);
2244 if (total_bytes >= sizeof(*data)) {
2245 data->bytes_left = total_bytes - sizeof(*data);
2246 data->bytes_missing = 0;
2248 data->bytes_missing = sizeof(*data) - total_bytes;
2249 data->bytes_left = 0;
2253 data->elem_missed = 0;
2259 * allocates space to return multiple file system paths for an inode.
2260 * total_bytes to allocate are passed, note that space usable for actual path
2261 * information will be total_bytes - sizeof(struct inode_fs_paths).
2262 * the returned pointer must be freed with free_ipath() in the end.
2264 struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
2265 struct btrfs_path *path)
2267 struct inode_fs_paths *ifp;
2268 struct btrfs_data_container *fspath;
2270 fspath = init_data_container(total_bytes);
2272 return ERR_CAST(fspath);
2274 ifp = kmalloc(sizeof(*ifp), GFP_KERNEL);
2277 return ERR_PTR(-ENOMEM);
2280 ifp->btrfs_path = path;
2281 ifp->fspath = fspath;
2282 ifp->fs_root = fs_root;
2287 void free_ipath(struct inode_fs_paths *ipath)
2291 kvfree(ipath->fspath);
2295 struct btrfs_backref_iter *btrfs_backref_iter_alloc(
2296 struct btrfs_fs_info *fs_info, gfp_t gfp_flag)
2298 struct btrfs_backref_iter *ret;
2300 ret = kzalloc(sizeof(*ret), gfp_flag);
2304 ret->path = btrfs_alloc_path();
2310 /* Current backref iterator only supports iteration in commit root */
2311 ret->path->search_commit_root = 1;
2312 ret->path->skip_locking = 1;
2313 ret->fs_info = fs_info;
2318 int btrfs_backref_iter_start(struct btrfs_backref_iter *iter, u64 bytenr)
2320 struct btrfs_fs_info *fs_info = iter->fs_info;
2321 struct btrfs_path *path = iter->path;
2322 struct btrfs_extent_item *ei;
2323 struct btrfs_key key;
2326 key.objectid = bytenr;
2327 key.type = BTRFS_METADATA_ITEM_KEY;
2328 key.offset = (u64)-1;
2329 iter->bytenr = bytenr;
2331 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
2338 if (path->slots[0] == 0) {
2339 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
2345 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2346 if ((key.type != BTRFS_EXTENT_ITEM_KEY &&
2347 key.type != BTRFS_METADATA_ITEM_KEY) || key.objectid != bytenr) {
2351 memcpy(&iter->cur_key, &key, sizeof(key));
2352 iter->item_ptr = (u32)btrfs_item_ptr_offset(path->nodes[0],
2354 iter->end_ptr = (u32)(iter->item_ptr +
2355 btrfs_item_size_nr(path->nodes[0], path->slots[0]));
2356 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
2357 struct btrfs_extent_item);
2360 * Only support iteration on tree backref yet.
2362 * This is an extra precaution for non skinny-metadata, where
2363 * EXTENT_ITEM is also used for tree blocks, that we can only use
2364 * extent flags to determine if it's a tree block.
2366 if (btrfs_extent_flags(path->nodes[0], ei) & BTRFS_EXTENT_FLAG_DATA) {
2370 iter->cur_ptr = (u32)(iter->item_ptr + sizeof(*ei));
2372 /* If there is no inline backref, go search for keyed backref */
2373 if (iter->cur_ptr >= iter->end_ptr) {
2374 ret = btrfs_next_item(fs_info->extent_root, path);
2376 /* No inline nor keyed ref */
2384 btrfs_item_key_to_cpu(path->nodes[0], &iter->cur_key,
2386 if (iter->cur_key.objectid != bytenr ||
2387 (iter->cur_key.type != BTRFS_SHARED_BLOCK_REF_KEY &&
2388 iter->cur_key.type != BTRFS_TREE_BLOCK_REF_KEY)) {
2392 iter->cur_ptr = (u32)btrfs_item_ptr_offset(path->nodes[0],
2394 iter->item_ptr = iter->cur_ptr;
2395 iter->end_ptr = (u32)(iter->item_ptr + btrfs_item_size_nr(
2396 path->nodes[0], path->slots[0]));
2401 btrfs_backref_iter_release(iter);
2406 * Go to the next backref item of current bytenr, can be either inlined or
2409 * Caller needs to check whether it's inline ref or not by iter->cur_key.
2411 * Return 0 if we get next backref without problem.
2412 * Return >0 if there is no extra backref for this bytenr.
2413 * Return <0 if there is something wrong happened.
2415 int btrfs_backref_iter_next(struct btrfs_backref_iter *iter)
2417 struct extent_buffer *eb = btrfs_backref_get_eb(iter);
2418 struct btrfs_path *path = iter->path;
2419 struct btrfs_extent_inline_ref *iref;
2423 if (btrfs_backref_iter_is_inline_ref(iter)) {
2424 /* We're still inside the inline refs */
2425 ASSERT(iter->cur_ptr < iter->end_ptr);
2427 if (btrfs_backref_has_tree_block_info(iter)) {
2428 /* First tree block info */
2429 size = sizeof(struct btrfs_tree_block_info);
2431 /* Use inline ref type to determine the size */
2434 iref = (struct btrfs_extent_inline_ref *)
2435 ((unsigned long)iter->cur_ptr);
2436 type = btrfs_extent_inline_ref_type(eb, iref);
2438 size = btrfs_extent_inline_ref_size(type);
2440 iter->cur_ptr += size;
2441 if (iter->cur_ptr < iter->end_ptr)
2444 /* All inline items iterated, fall through */
2447 /* We're at keyed items, there is no inline item, go to the next one */
2448 ret = btrfs_next_item(iter->fs_info->extent_root, iter->path);
2452 btrfs_item_key_to_cpu(path->nodes[0], &iter->cur_key, path->slots[0]);
2453 if (iter->cur_key.objectid != iter->bytenr ||
2454 (iter->cur_key.type != BTRFS_TREE_BLOCK_REF_KEY &&
2455 iter->cur_key.type != BTRFS_SHARED_BLOCK_REF_KEY))
2457 iter->item_ptr = (u32)btrfs_item_ptr_offset(path->nodes[0],
2459 iter->cur_ptr = iter->item_ptr;
2460 iter->end_ptr = iter->item_ptr + (u32)btrfs_item_size_nr(path->nodes[0],
2465 void btrfs_backref_init_cache(struct btrfs_fs_info *fs_info,
2466 struct btrfs_backref_cache *cache, int is_reloc)
2470 cache->rb_root = RB_ROOT;
2471 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
2472 INIT_LIST_HEAD(&cache->pending[i]);
2473 INIT_LIST_HEAD(&cache->changed);
2474 INIT_LIST_HEAD(&cache->detached);
2475 INIT_LIST_HEAD(&cache->leaves);
2476 INIT_LIST_HEAD(&cache->pending_edge);
2477 INIT_LIST_HEAD(&cache->useless_node);
2478 cache->fs_info = fs_info;
2479 cache->is_reloc = is_reloc;
2482 struct btrfs_backref_node *btrfs_backref_alloc_node(
2483 struct btrfs_backref_cache *cache, u64 bytenr, int level)
2485 struct btrfs_backref_node *node;
2487 ASSERT(level >= 0 && level < BTRFS_MAX_LEVEL);
2488 node = kzalloc(sizeof(*node), GFP_NOFS);
2492 INIT_LIST_HEAD(&node->list);
2493 INIT_LIST_HEAD(&node->upper);
2494 INIT_LIST_HEAD(&node->lower);
2495 RB_CLEAR_NODE(&node->rb_node);
2497 node->level = level;
2498 node->bytenr = bytenr;
2503 struct btrfs_backref_edge *btrfs_backref_alloc_edge(
2504 struct btrfs_backref_cache *cache)
2506 struct btrfs_backref_edge *edge;
2508 edge = kzalloc(sizeof(*edge), GFP_NOFS);
2515 * Drop the backref node from cache, also cleaning up all its
2516 * upper edges and any uncached nodes in the path.
2518 * This cleanup happens bottom up, thus the node should either
2519 * be the lowest node in the cache or a detached node.
2521 void btrfs_backref_cleanup_node(struct btrfs_backref_cache *cache,
2522 struct btrfs_backref_node *node)
2524 struct btrfs_backref_node *upper;
2525 struct btrfs_backref_edge *edge;
2530 BUG_ON(!node->lowest && !node->detached);
2531 while (!list_empty(&node->upper)) {
2532 edge = list_entry(node->upper.next, struct btrfs_backref_edge,
2534 upper = edge->node[UPPER];
2535 list_del(&edge->list[LOWER]);
2536 list_del(&edge->list[UPPER]);
2537 btrfs_backref_free_edge(cache, edge);
2539 if (RB_EMPTY_NODE(&upper->rb_node)) {
2540 BUG_ON(!list_empty(&node->upper));
2541 btrfs_backref_drop_node(cache, node);
2547 * Add the node to leaf node list if no other child block
2550 if (list_empty(&upper->lower)) {
2551 list_add_tail(&upper->lower, &cache->leaves);
2556 btrfs_backref_drop_node(cache, node);
2560 * Release all nodes/edges from current cache
2562 void btrfs_backref_release_cache(struct btrfs_backref_cache *cache)
2564 struct btrfs_backref_node *node;
2567 while (!list_empty(&cache->detached)) {
2568 node = list_entry(cache->detached.next,
2569 struct btrfs_backref_node, list);
2570 btrfs_backref_cleanup_node(cache, node);
2573 while (!list_empty(&cache->leaves)) {
2574 node = list_entry(cache->leaves.next,
2575 struct btrfs_backref_node, lower);
2576 btrfs_backref_cleanup_node(cache, node);
2579 cache->last_trans = 0;
2581 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
2582 ASSERT(list_empty(&cache->pending[i]));
2583 ASSERT(list_empty(&cache->pending_edge));
2584 ASSERT(list_empty(&cache->useless_node));
2585 ASSERT(list_empty(&cache->changed));
2586 ASSERT(list_empty(&cache->detached));
2587 ASSERT(RB_EMPTY_ROOT(&cache->rb_root));
2588 ASSERT(!cache->nr_nodes);
2589 ASSERT(!cache->nr_edges);
2593 * Handle direct tree backref
2595 * Direct tree backref means, the backref item shows its parent bytenr
2596 * directly. This is for SHARED_BLOCK_REF backref (keyed or inlined).
2598 * @ref_key: The converted backref key.
2599 * For keyed backref, it's the item key.
2600 * For inlined backref, objectid is the bytenr,
2601 * type is btrfs_inline_ref_type, offset is
2602 * btrfs_inline_ref_offset.
2604 static int handle_direct_tree_backref(struct btrfs_backref_cache *cache,
2605 struct btrfs_key *ref_key,
2606 struct btrfs_backref_node *cur)
2608 struct btrfs_backref_edge *edge;
2609 struct btrfs_backref_node *upper;
2610 struct rb_node *rb_node;
2612 ASSERT(ref_key->type == BTRFS_SHARED_BLOCK_REF_KEY);
2614 /* Only reloc root uses backref pointing to itself */
2615 if (ref_key->objectid == ref_key->offset) {
2616 struct btrfs_root *root;
2618 cur->is_reloc_root = 1;
2619 /* Only reloc backref cache cares about a specific root */
2620 if (cache->is_reloc) {
2621 root = find_reloc_root(cache->fs_info, cur->bytenr);
2627 * For generic purpose backref cache, reloc root node
2630 list_add(&cur->list, &cache->useless_node);
2635 edge = btrfs_backref_alloc_edge(cache);
2639 rb_node = rb_simple_search(&cache->rb_root, ref_key->offset);
2641 /* Parent node not yet cached */
2642 upper = btrfs_backref_alloc_node(cache, ref_key->offset,
2645 btrfs_backref_free_edge(cache, edge);
2650 * Backrefs for the upper level block isn't cached, add the
2651 * block to pending list
2653 list_add_tail(&edge->list[UPPER], &cache->pending_edge);
2655 /* Parent node already cached */
2656 upper = rb_entry(rb_node, struct btrfs_backref_node, rb_node);
2657 ASSERT(upper->checked);
2658 INIT_LIST_HEAD(&edge->list[UPPER]);
2660 btrfs_backref_link_edge(edge, cur, upper, LINK_LOWER);
2665 * Handle indirect tree backref
2667 * Indirect tree backref means, we only know which tree the node belongs to.
2668 * We still need to do a tree search to find out the parents. This is for
2669 * TREE_BLOCK_REF backref (keyed or inlined).
2671 * @ref_key: The same as @ref_key in handle_direct_tree_backref()
2672 * @tree_key: The first key of this tree block.
2673 * @path: A clean (released) path, to avoid allocating path everytime
2674 * the function get called.
2676 static int handle_indirect_tree_backref(struct btrfs_backref_cache *cache,
2677 struct btrfs_path *path,
2678 struct btrfs_key *ref_key,
2679 struct btrfs_key *tree_key,
2680 struct btrfs_backref_node *cur)
2682 struct btrfs_fs_info *fs_info = cache->fs_info;
2683 struct btrfs_backref_node *upper;
2684 struct btrfs_backref_node *lower;
2685 struct btrfs_backref_edge *edge;
2686 struct extent_buffer *eb;
2687 struct btrfs_root *root;
2688 struct rb_node *rb_node;
2690 bool need_check = true;
2693 root = btrfs_get_fs_root(fs_info, ref_key->offset, false);
2695 return PTR_ERR(root);
2696 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
2699 if (btrfs_root_level(&root->root_item) == cur->level) {
2701 ASSERT(btrfs_root_bytenr(&root->root_item) == cur->bytenr);
2703 * For reloc backref cache, we may ignore reloc root. But for
2704 * general purpose backref cache, we can't rely on
2705 * btrfs_should_ignore_reloc_root() as it may conflict with
2706 * current running relocation and lead to missing root.
2708 * For general purpose backref cache, reloc root detection is
2709 * completely relying on direct backref (key->offset is parent
2710 * bytenr), thus only do such check for reloc cache.
2712 if (btrfs_should_ignore_reloc_root(root) && cache->is_reloc) {
2713 btrfs_put_root(root);
2714 list_add(&cur->list, &cache->useless_node);
2721 level = cur->level + 1;
2723 /* Search the tree to find parent blocks referring to the block */
2724 path->search_commit_root = 1;
2725 path->skip_locking = 1;
2726 path->lowest_level = level;
2727 ret = btrfs_search_slot(NULL, root, tree_key, path, 0, 0);
2728 path->lowest_level = 0;
2730 btrfs_put_root(root);
2733 if (ret > 0 && path->slots[level] > 0)
2734 path->slots[level]--;
2736 eb = path->nodes[level];
2737 if (btrfs_node_blockptr(eb, path->slots[level]) != cur->bytenr) {
2739 "couldn't find block (%llu) (level %d) in tree (%llu) with key (%llu %u %llu)",
2740 cur->bytenr, level - 1, root->root_key.objectid,
2741 tree_key->objectid, tree_key->type, tree_key->offset);
2742 btrfs_put_root(root);
2748 /* Add all nodes and edges in the path */
2749 for (; level < BTRFS_MAX_LEVEL; level++) {
2750 if (!path->nodes[level]) {
2751 ASSERT(btrfs_root_bytenr(&root->root_item) ==
2753 /* Same as previous should_ignore_reloc_root() call */
2754 if (btrfs_should_ignore_reloc_root(root) &&
2756 btrfs_put_root(root);
2757 list_add(&lower->list, &cache->useless_node);
2764 edge = btrfs_backref_alloc_edge(cache);
2766 btrfs_put_root(root);
2771 eb = path->nodes[level];
2772 rb_node = rb_simple_search(&cache->rb_root, eb->start);
2774 upper = btrfs_backref_alloc_node(cache, eb->start,
2777 btrfs_put_root(root);
2778 btrfs_backref_free_edge(cache, edge);
2782 upper->owner = btrfs_header_owner(eb);
2783 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
2787 * If we know the block isn't shared we can avoid
2788 * checking its backrefs.
2790 if (btrfs_block_can_be_shared(root, eb))
2796 * Add the block to pending list if we need to check its
2797 * backrefs, we only do this once while walking up a
2798 * tree as we will catch anything else later on.
2800 if (!upper->checked && need_check) {
2802 list_add_tail(&edge->list[UPPER],
2803 &cache->pending_edge);
2807 INIT_LIST_HEAD(&edge->list[UPPER]);
2810 upper = rb_entry(rb_node, struct btrfs_backref_node,
2812 ASSERT(upper->checked);
2813 INIT_LIST_HEAD(&edge->list[UPPER]);
2815 upper->owner = btrfs_header_owner(eb);
2817 btrfs_backref_link_edge(edge, lower, upper, LINK_LOWER);
2820 btrfs_put_root(root);
2827 btrfs_release_path(path);
2832 * Add backref node @cur into @cache.
2834 * NOTE: Even if the function returned 0, @cur is not yet cached as its upper
2835 * links aren't yet bi-directional. Needs to finish such links.
2836 * Use btrfs_backref_finish_upper_links() to finish such linkage.
2838 * @path: Released path for indirect tree backref lookup
2839 * @iter: Released backref iter for extent tree search
2840 * @node_key: The first key of the tree block
2842 int btrfs_backref_add_tree_node(struct btrfs_backref_cache *cache,
2843 struct btrfs_path *path,
2844 struct btrfs_backref_iter *iter,
2845 struct btrfs_key *node_key,
2846 struct btrfs_backref_node *cur)
2848 struct btrfs_fs_info *fs_info = cache->fs_info;
2849 struct btrfs_backref_edge *edge;
2850 struct btrfs_backref_node *exist;
2853 ret = btrfs_backref_iter_start(iter, cur->bytenr);
2857 * We skip the first btrfs_tree_block_info, as we don't use the key
2858 * stored in it, but fetch it from the tree block
2860 if (btrfs_backref_has_tree_block_info(iter)) {
2861 ret = btrfs_backref_iter_next(iter);
2864 /* No extra backref? This means the tree block is corrupted */
2870 WARN_ON(cur->checked);
2871 if (!list_empty(&cur->upper)) {
2873 * The backref was added previously when processing backref of
2874 * type BTRFS_TREE_BLOCK_REF_KEY
2876 ASSERT(list_is_singular(&cur->upper));
2877 edge = list_entry(cur->upper.next, struct btrfs_backref_edge,
2879 ASSERT(list_empty(&edge->list[UPPER]));
2880 exist = edge->node[UPPER];
2882 * Add the upper level block to pending list if we need check
2885 if (!exist->checked)
2886 list_add_tail(&edge->list[UPPER], &cache->pending_edge);
2891 for (; ret == 0; ret = btrfs_backref_iter_next(iter)) {
2892 struct extent_buffer *eb;
2893 struct btrfs_key key;
2897 eb = btrfs_backref_get_eb(iter);
2899 key.objectid = iter->bytenr;
2900 if (btrfs_backref_iter_is_inline_ref(iter)) {
2901 struct btrfs_extent_inline_ref *iref;
2903 /* Update key for inline backref */
2904 iref = (struct btrfs_extent_inline_ref *)
2905 ((unsigned long)iter->cur_ptr);
2906 type = btrfs_get_extent_inline_ref_type(eb, iref,
2907 BTRFS_REF_TYPE_BLOCK);
2908 if (type == BTRFS_REF_TYPE_INVALID) {
2913 key.offset = btrfs_extent_inline_ref_offset(eb, iref);
2915 key.type = iter->cur_key.type;
2916 key.offset = iter->cur_key.offset;
2920 * Parent node found and matches current inline ref, no need to
2921 * rebuild this node for this inline ref
2924 ((key.type == BTRFS_TREE_BLOCK_REF_KEY &&
2925 exist->owner == key.offset) ||
2926 (key.type == BTRFS_SHARED_BLOCK_REF_KEY &&
2927 exist->bytenr == key.offset))) {
2932 /* SHARED_BLOCK_REF means key.offset is the parent bytenr */
2933 if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
2934 ret = handle_direct_tree_backref(cache, &key, cur);
2938 } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) {
2940 btrfs_print_v0_err(fs_info);
2941 btrfs_handle_fs_error(fs_info, ret, NULL);
2943 } else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) {
2948 * key.type == BTRFS_TREE_BLOCK_REF_KEY, inline ref offset
2949 * means the root objectid. We need to search the tree to get
2950 * its parent bytenr.
2952 ret = handle_indirect_tree_backref(cache, path, &key, node_key,
2961 btrfs_backref_iter_release(iter);
2966 * Finish the upwards linkage created by btrfs_backref_add_tree_node()
2968 int btrfs_backref_finish_upper_links(struct btrfs_backref_cache *cache,
2969 struct btrfs_backref_node *start)
2971 struct list_head *useless_node = &cache->useless_node;
2972 struct btrfs_backref_edge *edge;
2973 struct rb_node *rb_node;
2974 LIST_HEAD(pending_edge);
2976 ASSERT(start->checked);
2978 /* Insert this node to cache if it's not COW-only */
2979 if (!start->cowonly) {
2980 rb_node = rb_simple_insert(&cache->rb_root, start->bytenr,
2983 btrfs_backref_panic(cache->fs_info, start->bytenr,
2985 list_add_tail(&start->lower, &cache->leaves);
2989 * Use breadth first search to iterate all related edges.
2991 * The starting points are all the edges of this node
2993 list_for_each_entry(edge, &start->upper, list[LOWER])
2994 list_add_tail(&edge->list[UPPER], &pending_edge);
2996 while (!list_empty(&pending_edge)) {
2997 struct btrfs_backref_node *upper;
2998 struct btrfs_backref_node *lower;
2999 struct rb_node *rb_node;
3001 edge = list_first_entry(&pending_edge,
3002 struct btrfs_backref_edge, list[UPPER]);
3003 list_del_init(&edge->list[UPPER]);
3004 upper = edge->node[UPPER];
3005 lower = edge->node[LOWER];
3007 /* Parent is detached, no need to keep any edges */
3008 if (upper->detached) {
3009 list_del(&edge->list[LOWER]);
3010 btrfs_backref_free_edge(cache, edge);
3012 /* Lower node is orphan, queue for cleanup */
3013 if (list_empty(&lower->upper))
3014 list_add(&lower->list, useless_node);
3019 * All new nodes added in current build_backref_tree() haven't
3020 * been linked to the cache rb tree.
3021 * So if we have upper->rb_node populated, this means a cache
3022 * hit. We only need to link the edge, as @upper and all its
3023 * parents have already been linked.
3025 if (!RB_EMPTY_NODE(&upper->rb_node)) {
3026 if (upper->lowest) {
3027 list_del_init(&upper->lower);
3031 list_add_tail(&edge->list[UPPER], &upper->lower);
3035 /* Sanity check, we shouldn't have any unchecked nodes */
3036 if (!upper->checked) {
3041 /* Sanity check, COW-only node has non-COW-only parent */
3042 if (start->cowonly != upper->cowonly) {
3047 /* Only cache non-COW-only (subvolume trees) tree blocks */
3048 if (!upper->cowonly) {
3049 rb_node = rb_simple_insert(&cache->rb_root, upper->bytenr,
3052 btrfs_backref_panic(cache->fs_info,
3053 upper->bytenr, -EEXIST);
3058 list_add_tail(&edge->list[UPPER], &upper->lower);
3061 * Also queue all the parent edges of this uncached node
3062 * to finish the upper linkage
3064 list_for_each_entry(edge, &upper->upper, list[LOWER])
3065 list_add_tail(&edge->list[UPPER], &pending_edge);
3070 void btrfs_backref_error_cleanup(struct btrfs_backref_cache *cache,
3071 struct btrfs_backref_node *node)
3073 struct btrfs_backref_node *lower;
3074 struct btrfs_backref_node *upper;
3075 struct btrfs_backref_edge *edge;
3077 while (!list_empty(&cache->useless_node)) {
3078 lower = list_first_entry(&cache->useless_node,
3079 struct btrfs_backref_node, list);
3080 list_del_init(&lower->list);
3082 while (!list_empty(&cache->pending_edge)) {
3083 edge = list_first_entry(&cache->pending_edge,
3084 struct btrfs_backref_edge, list[UPPER]);
3085 list_del(&edge->list[UPPER]);
3086 list_del(&edge->list[LOWER]);
3087 lower = edge->node[LOWER];
3088 upper = edge->node[UPPER];
3089 btrfs_backref_free_edge(cache, edge);
3092 * Lower is no longer linked to any upper backref nodes and
3093 * isn't in the cache, we can free it ourselves.
3095 if (list_empty(&lower->upper) &&
3096 RB_EMPTY_NODE(&lower->rb_node))
3097 list_add(&lower->list, &cache->useless_node);
3099 if (!RB_EMPTY_NODE(&upper->rb_node))
3102 /* Add this guy's upper edges to the list to process */
3103 list_for_each_entry(edge, &upper->upper, list[LOWER])
3104 list_add_tail(&edge->list[UPPER],
3105 &cache->pending_edge);
3106 if (list_empty(&upper->upper))
3107 list_add(&upper->list, &cache->useless_node);
3110 while (!list_empty(&cache->useless_node)) {
3111 lower = list_first_entry(&cache->useless_node,
3112 struct btrfs_backref_node, list);
3113 list_del_init(&lower->list);
3116 btrfs_backref_free_node(cache, lower);
3119 btrfs_backref_cleanup_node(cache, node);
3120 ASSERT(list_empty(&cache->useless_node) &&
3121 list_empty(&cache->pending_edge));