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 #include "tree-mod-log.h"
19 /* Just an arbitrary number so we can be sure this happened */
20 #define BACKREF_FOUND_SHARED 6
22 struct extent_inode_elem {
25 struct extent_inode_elem *next;
28 static int check_extent_in_eb(const struct btrfs_key *key,
29 const struct extent_buffer *eb,
30 const struct btrfs_file_extent_item *fi,
32 struct extent_inode_elem **eie,
36 struct extent_inode_elem *e;
39 !btrfs_file_extent_compression(eb, fi) &&
40 !btrfs_file_extent_encryption(eb, fi) &&
41 !btrfs_file_extent_other_encoding(eb, fi)) {
45 data_offset = btrfs_file_extent_offset(eb, fi);
46 data_len = btrfs_file_extent_num_bytes(eb, fi);
48 if (extent_item_pos < data_offset ||
49 extent_item_pos >= data_offset + data_len)
51 offset = extent_item_pos - data_offset;
54 e = kmalloc(sizeof(*e), GFP_NOFS);
59 e->inum = key->objectid;
60 e->offset = key->offset + offset;
66 static void free_inode_elem_list(struct extent_inode_elem *eie)
68 struct extent_inode_elem *eie_next;
70 for (; eie; eie = eie_next) {
76 static int find_extent_in_eb(const struct extent_buffer *eb,
77 u64 wanted_disk_byte, u64 extent_item_pos,
78 struct extent_inode_elem **eie,
83 struct btrfs_file_extent_item *fi;
90 * from the shared data ref, we only have the leaf but we need
91 * the key. thus, we must look into all items and see that we
92 * find one (some) with a reference to our extent item.
94 nritems = btrfs_header_nritems(eb);
95 for (slot = 0; slot < nritems; ++slot) {
96 btrfs_item_key_to_cpu(eb, &key, slot);
97 if (key.type != BTRFS_EXTENT_DATA_KEY)
99 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
100 extent_type = btrfs_file_extent_type(eb, fi);
101 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
103 /* don't skip BTRFS_FILE_EXTENT_PREALLOC, we can handle that */
104 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
105 if (disk_byte != wanted_disk_byte)
108 ret = check_extent_in_eb(&key, eb, fi, extent_item_pos, eie, ignore_offset);
117 struct rb_root_cached root;
121 #define PREFTREE_INIT { .root = RB_ROOT_CACHED, .count = 0 }
124 struct preftree direct; /* BTRFS_SHARED_[DATA|BLOCK]_REF_KEY */
125 struct preftree indirect; /* BTRFS_[TREE_BLOCK|EXTENT_DATA]_REF_KEY */
126 struct preftree indirect_missing_keys;
130 * Checks for a shared extent during backref search.
132 * The share_count tracks prelim_refs (direct and indirect) having a
134 * - incremented when a ref->count transitions to >0
135 * - decremented when a ref->count transitions to <1
141 bool have_delayed_delete_refs;
144 static inline int extent_is_shared(struct share_check *sc)
146 return (sc && sc->share_count > 1) ? BACKREF_FOUND_SHARED : 0;
149 static struct kmem_cache *btrfs_prelim_ref_cache;
151 int __init btrfs_prelim_ref_init(void)
153 btrfs_prelim_ref_cache = kmem_cache_create("btrfs_prelim_ref",
154 sizeof(struct prelim_ref),
158 if (!btrfs_prelim_ref_cache)
163 void __cold btrfs_prelim_ref_exit(void)
165 kmem_cache_destroy(btrfs_prelim_ref_cache);
168 static void free_pref(struct prelim_ref *ref)
170 kmem_cache_free(btrfs_prelim_ref_cache, ref);
174 * Return 0 when both refs are for the same block (and can be merged).
175 * A -1 return indicates ref1 is a 'lower' block than ref2, while 1
176 * indicates a 'higher' block.
178 static int prelim_ref_compare(struct prelim_ref *ref1,
179 struct prelim_ref *ref2)
181 if (ref1->level < ref2->level)
183 if (ref1->level > ref2->level)
185 if (ref1->root_id < ref2->root_id)
187 if (ref1->root_id > ref2->root_id)
189 if (ref1->key_for_search.type < ref2->key_for_search.type)
191 if (ref1->key_for_search.type > ref2->key_for_search.type)
193 if (ref1->key_for_search.objectid < ref2->key_for_search.objectid)
195 if (ref1->key_for_search.objectid > ref2->key_for_search.objectid)
197 if (ref1->key_for_search.offset < ref2->key_for_search.offset)
199 if (ref1->key_for_search.offset > ref2->key_for_search.offset)
201 if (ref1->parent < ref2->parent)
203 if (ref1->parent > ref2->parent)
209 static void update_share_count(struct share_check *sc, int oldcount,
212 if ((!sc) || (oldcount == 0 && newcount < 1))
215 if (oldcount > 0 && newcount < 1)
217 else if (oldcount < 1 && newcount > 0)
222 * Add @newref to the @root rbtree, merging identical refs.
224 * Callers should assume that newref has been freed after calling.
226 static void prelim_ref_insert(const struct btrfs_fs_info *fs_info,
227 struct preftree *preftree,
228 struct prelim_ref *newref,
229 struct share_check *sc)
231 struct rb_root_cached *root;
233 struct rb_node *parent = NULL;
234 struct prelim_ref *ref;
236 bool leftmost = true;
238 root = &preftree->root;
239 p = &root->rb_root.rb_node;
243 ref = rb_entry(parent, struct prelim_ref, rbnode);
244 result = prelim_ref_compare(ref, newref);
247 } else if (result > 0) {
251 /* Identical refs, merge them and free @newref */
252 struct extent_inode_elem *eie = ref->inode_list;
254 while (eie && eie->next)
258 ref->inode_list = newref->inode_list;
260 eie->next = newref->inode_list;
261 trace_btrfs_prelim_ref_merge(fs_info, ref, newref,
264 * A delayed ref can have newref->count < 0.
265 * The ref->count is updated to follow any
266 * BTRFS_[ADD|DROP]_DELAYED_REF actions.
268 update_share_count(sc, ref->count,
269 ref->count + newref->count);
270 ref->count += newref->count;
276 update_share_count(sc, 0, newref->count);
278 trace_btrfs_prelim_ref_insert(fs_info, newref, NULL, preftree->count);
279 rb_link_node(&newref->rbnode, parent, p);
280 rb_insert_color_cached(&newref->rbnode, root, leftmost);
284 * Release the entire tree. We don't care about internal consistency so
285 * just free everything and then reset the tree root.
287 static void prelim_release(struct preftree *preftree)
289 struct prelim_ref *ref, *next_ref;
291 rbtree_postorder_for_each_entry_safe(ref, next_ref,
292 &preftree->root.rb_root, rbnode)
295 preftree->root = RB_ROOT_CACHED;
300 * the rules for all callers of this function are:
301 * - obtaining the parent is the goal
302 * - if you add a key, you must know that it is a correct key
303 * - if you cannot add the parent or a correct key, then we will look into the
304 * block later to set a correct key
308 * backref type | shared | indirect | shared | indirect
309 * information | tree | tree | data | data
310 * --------------------+--------+----------+--------+----------
311 * parent logical | y | - | - | -
312 * key to resolve | - | y | y | y
313 * tree block logical | - | - | - | -
314 * root for resolving | y | y | y | y
316 * - column 1: we've the parent -> done
317 * - column 2, 3, 4: we use the key to find the parent
319 * on disk refs (inline or keyed)
320 * ==============================
321 * backref type | shared | indirect | shared | indirect
322 * information | tree | tree | data | data
323 * --------------------+--------+----------+--------+----------
324 * parent logical | y | - | y | -
325 * key to resolve | - | - | - | y
326 * tree block logical | y | y | y | y
327 * root for resolving | - | y | y | y
329 * - column 1, 3: we've the parent -> done
330 * - column 2: we take the first key from the block to find the parent
331 * (see add_missing_keys)
332 * - column 4: we use the key to find the parent
334 * additional information that's available but not required to find the parent
335 * block might help in merging entries to gain some speed.
337 static int add_prelim_ref(const struct btrfs_fs_info *fs_info,
338 struct preftree *preftree, u64 root_id,
339 const struct btrfs_key *key, int level, u64 parent,
340 u64 wanted_disk_byte, int count,
341 struct share_check *sc, gfp_t gfp_mask)
343 struct prelim_ref *ref;
345 if (root_id == BTRFS_DATA_RELOC_TREE_OBJECTID)
348 ref = kmem_cache_alloc(btrfs_prelim_ref_cache, gfp_mask);
352 ref->root_id = root_id;
354 ref->key_for_search = *key;
356 memset(&ref->key_for_search, 0, sizeof(ref->key_for_search));
358 ref->inode_list = NULL;
361 ref->parent = parent;
362 ref->wanted_disk_byte = wanted_disk_byte;
363 prelim_ref_insert(fs_info, preftree, ref, sc);
364 return extent_is_shared(sc);
367 /* direct refs use root == 0, key == NULL */
368 static int add_direct_ref(const struct btrfs_fs_info *fs_info,
369 struct preftrees *preftrees, int level, u64 parent,
370 u64 wanted_disk_byte, int count,
371 struct share_check *sc, gfp_t gfp_mask)
373 return add_prelim_ref(fs_info, &preftrees->direct, 0, NULL, level,
374 parent, wanted_disk_byte, count, sc, gfp_mask);
377 /* indirect refs use parent == 0 */
378 static int add_indirect_ref(const struct btrfs_fs_info *fs_info,
379 struct preftrees *preftrees, u64 root_id,
380 const struct btrfs_key *key, int level,
381 u64 wanted_disk_byte, int count,
382 struct share_check *sc, gfp_t gfp_mask)
384 struct preftree *tree = &preftrees->indirect;
387 tree = &preftrees->indirect_missing_keys;
388 return add_prelim_ref(fs_info, tree, root_id, key, level, 0,
389 wanted_disk_byte, count, sc, gfp_mask);
392 static int is_shared_data_backref(struct preftrees *preftrees, u64 bytenr)
394 struct rb_node **p = &preftrees->direct.root.rb_root.rb_node;
395 struct rb_node *parent = NULL;
396 struct prelim_ref *ref = NULL;
397 struct prelim_ref target = {};
400 target.parent = bytenr;
404 ref = rb_entry(parent, struct prelim_ref, rbnode);
405 result = prelim_ref_compare(ref, &target);
417 static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path,
418 struct ulist *parents,
419 struct preftrees *preftrees, struct prelim_ref *ref,
420 int level, u64 time_seq, const u64 *extent_item_pos,
425 struct extent_buffer *eb;
426 struct btrfs_key key;
427 struct btrfs_key *key_for_search = &ref->key_for_search;
428 struct btrfs_file_extent_item *fi;
429 struct extent_inode_elem *eie = NULL, *old = NULL;
431 u64 wanted_disk_byte = ref->wanted_disk_byte;
436 eb = path->nodes[level];
437 ret = ulist_add(parents, eb->start, 0, GFP_NOFS);
444 * 1. We normally enter this function with the path already pointing to
445 * the first item to check. But sometimes, we may enter it with
447 * 2. We are searching for normal backref but bytenr of this leaf
448 * matches shared data backref
449 * 3. The leaf owner is not equal to the root we are searching
451 * For these cases, go to the next leaf before we continue.
454 if (path->slots[0] >= btrfs_header_nritems(eb) ||
455 is_shared_data_backref(preftrees, eb->start) ||
456 ref->root_id != btrfs_header_owner(eb)) {
457 if (time_seq == BTRFS_SEQ_LAST)
458 ret = btrfs_next_leaf(root, path);
460 ret = btrfs_next_old_leaf(root, path, time_seq);
463 while (!ret && count < ref->count) {
465 slot = path->slots[0];
467 btrfs_item_key_to_cpu(eb, &key, slot);
469 if (key.objectid != key_for_search->objectid ||
470 key.type != BTRFS_EXTENT_DATA_KEY)
474 * We are searching for normal backref but bytenr of this leaf
475 * matches shared data backref, OR
476 * the leaf owner is not equal to the root we are searching for
479 (is_shared_data_backref(preftrees, eb->start) ||
480 ref->root_id != btrfs_header_owner(eb))) {
481 if (time_seq == BTRFS_SEQ_LAST)
482 ret = btrfs_next_leaf(root, path);
484 ret = btrfs_next_old_leaf(root, path, time_seq);
487 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
488 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
489 data_offset = btrfs_file_extent_offset(eb, fi);
491 if (disk_byte == wanted_disk_byte) {
494 if (ref->key_for_search.offset == key.offset - data_offset)
498 if (extent_item_pos) {
499 ret = check_extent_in_eb(&key, eb, fi,
501 &eie, ignore_offset);
507 ret = ulist_add_merge_ptr(parents, eb->start,
508 eie, (void **)&old, GFP_NOFS);
511 if (!ret && extent_item_pos) {
519 if (time_seq == BTRFS_SEQ_LAST)
520 ret = btrfs_next_item(root, path);
522 ret = btrfs_next_old_item(root, path, time_seq);
528 free_inode_elem_list(eie);
533 * resolve an indirect backref in the form (root_id, key, level)
534 * to a logical address
536 static int resolve_indirect_ref(struct btrfs_fs_info *fs_info,
537 struct btrfs_path *path, u64 time_seq,
538 struct preftrees *preftrees,
539 struct prelim_ref *ref, struct ulist *parents,
540 const u64 *extent_item_pos, bool ignore_offset)
542 struct btrfs_root *root;
543 struct extent_buffer *eb;
546 int level = ref->level;
547 struct btrfs_key search_key = ref->key_for_search;
550 * If we're search_commit_root we could possibly be holding locks on
551 * other tree nodes. This happens when qgroups does backref walks when
552 * adding new delayed refs. To deal with this we need to look in cache
553 * for the root, and if we don't find it then we need to search the
554 * tree_root's commit root, thus the btrfs_get_fs_root_commit_root usage
557 if (path->search_commit_root)
558 root = btrfs_get_fs_root_commit_root(fs_info, path, ref->root_id);
560 root = btrfs_get_fs_root(fs_info, ref->root_id, false);
566 if (!path->search_commit_root &&
567 test_bit(BTRFS_ROOT_DELETING, &root->state)) {
572 if (btrfs_is_testing(fs_info)) {
577 if (path->search_commit_root)
578 root_level = btrfs_header_level(root->commit_root);
579 else if (time_seq == BTRFS_SEQ_LAST)
580 root_level = btrfs_header_level(root->node);
582 root_level = btrfs_old_root_level(root, time_seq);
584 if (root_level + 1 == level)
588 * We can often find data backrefs with an offset that is too large
589 * (>= LLONG_MAX, maximum allowed file offset) due to underflows when
590 * subtracting a file's offset with the data offset of its
591 * corresponding extent data item. This can happen for example in the
594 * So if we detect such case we set the search key's offset to zero to
595 * make sure we will find the matching file extent item at
596 * add_all_parents(), otherwise we will miss it because the offset
597 * taken form the backref is much larger then the offset of the file
598 * extent item. This can make us scan a very large number of file
599 * extent items, but at least it will not make us miss any.
601 * This is an ugly workaround for a behaviour that should have never
602 * existed, but it does and a fix for the clone ioctl would touch a lot
603 * of places, cause backwards incompatibility and would not fix the
604 * problem for extents cloned with older kernels.
606 if (search_key.type == BTRFS_EXTENT_DATA_KEY &&
607 search_key.offset >= LLONG_MAX)
608 search_key.offset = 0;
609 path->lowest_level = level;
610 if (time_seq == BTRFS_SEQ_LAST)
611 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
613 ret = btrfs_search_old_slot(root, &search_key, path, time_seq);
616 "search slot in root %llu (level %d, ref count %d) returned %d for key (%llu %u %llu)",
617 ref->root_id, level, ref->count, ret,
618 ref->key_for_search.objectid, ref->key_for_search.type,
619 ref->key_for_search.offset);
623 eb = path->nodes[level];
625 if (WARN_ON(!level)) {
630 eb = path->nodes[level];
633 ret = add_all_parents(root, path, parents, preftrees, ref, level,
634 time_seq, extent_item_pos, ignore_offset);
636 btrfs_put_root(root);
638 path->lowest_level = 0;
639 btrfs_release_path(path);
643 static struct extent_inode_elem *
644 unode_aux_to_inode_list(struct ulist_node *node)
648 return (struct extent_inode_elem *)(uintptr_t)node->aux;
652 * We maintain three separate rbtrees: one for direct refs, one for
653 * indirect refs which have a key, and one for indirect refs which do not
654 * have a key. Each tree does merge on insertion.
656 * Once all of the references are located, we iterate over the tree of
657 * indirect refs with missing keys. An appropriate key is located and
658 * the ref is moved onto the tree for indirect refs. After all missing
659 * keys are thus located, we iterate over the indirect ref tree, resolve
660 * each reference, and then insert the resolved reference onto the
661 * direct tree (merging there too).
663 * New backrefs (i.e., for parent nodes) are added to the appropriate
664 * rbtree as they are encountered. The new backrefs are subsequently
667 static int resolve_indirect_refs(struct btrfs_fs_info *fs_info,
668 struct btrfs_path *path, u64 time_seq,
669 struct preftrees *preftrees,
670 const u64 *extent_item_pos,
671 struct share_check *sc, bool ignore_offset)
675 struct ulist *parents;
676 struct ulist_node *node;
677 struct ulist_iterator uiter;
678 struct rb_node *rnode;
680 parents = ulist_alloc(GFP_NOFS);
685 * We could trade memory usage for performance here by iterating
686 * the tree, allocating new refs for each insertion, and then
687 * freeing the entire indirect tree when we're done. In some test
688 * cases, the tree can grow quite large (~200k objects).
690 while ((rnode = rb_first_cached(&preftrees->indirect.root))) {
691 struct prelim_ref *ref;
693 ref = rb_entry(rnode, struct prelim_ref, rbnode);
694 if (WARN(ref->parent,
695 "BUG: direct ref found in indirect tree")) {
700 rb_erase_cached(&ref->rbnode, &preftrees->indirect.root);
701 preftrees->indirect.count--;
703 if (ref->count == 0) {
708 if (sc && sc->root_objectid &&
709 ref->root_id != sc->root_objectid) {
711 ret = BACKREF_FOUND_SHARED;
714 err = resolve_indirect_ref(fs_info, path, time_seq, preftrees,
715 ref, parents, extent_item_pos,
718 * we can only tolerate ENOENT,otherwise,we should catch error
719 * and return directly.
721 if (err == -ENOENT) {
722 prelim_ref_insert(fs_info, &preftrees->direct, ref,
731 /* we put the first parent into the ref at hand */
732 ULIST_ITER_INIT(&uiter);
733 node = ulist_next(parents, &uiter);
734 ref->parent = node ? node->val : 0;
735 ref->inode_list = unode_aux_to_inode_list(node);
737 /* Add a prelim_ref(s) for any other parent(s). */
738 while ((node = ulist_next(parents, &uiter))) {
739 struct prelim_ref *new_ref;
741 new_ref = kmem_cache_alloc(btrfs_prelim_ref_cache,
748 memcpy(new_ref, ref, sizeof(*ref));
749 new_ref->parent = node->val;
750 new_ref->inode_list = unode_aux_to_inode_list(node);
751 prelim_ref_insert(fs_info, &preftrees->direct,
756 * Now it's a direct ref, put it in the direct tree. We must
757 * do this last because the ref could be merged/freed here.
759 prelim_ref_insert(fs_info, &preftrees->direct, ref, NULL);
761 ulist_reinit(parents);
770 * read tree blocks and add keys where required.
772 static int add_missing_keys(struct btrfs_fs_info *fs_info,
773 struct preftrees *preftrees, bool lock)
775 struct prelim_ref *ref;
776 struct extent_buffer *eb;
777 struct preftree *tree = &preftrees->indirect_missing_keys;
778 struct rb_node *node;
780 while ((node = rb_first_cached(&tree->root))) {
781 ref = rb_entry(node, struct prelim_ref, rbnode);
782 rb_erase_cached(node, &tree->root);
784 BUG_ON(ref->parent); /* should not be a direct ref */
785 BUG_ON(ref->key_for_search.type);
786 BUG_ON(!ref->wanted_disk_byte);
788 eb = read_tree_block(fs_info, ref->wanted_disk_byte,
789 ref->root_id, 0, ref->level - 1, NULL);
793 } else if (!extent_buffer_uptodate(eb)) {
795 free_extent_buffer(eb);
799 btrfs_tree_read_lock(eb);
800 if (btrfs_header_level(eb) == 0)
801 btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
803 btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
805 btrfs_tree_read_unlock(eb);
806 free_extent_buffer(eb);
807 prelim_ref_insert(fs_info, &preftrees->indirect, ref, NULL);
814 * add all currently queued delayed refs from this head whose seq nr is
815 * smaller or equal that seq to the list
817 static int add_delayed_refs(const struct btrfs_fs_info *fs_info,
818 struct btrfs_delayed_ref_head *head, u64 seq,
819 struct preftrees *preftrees, struct share_check *sc)
821 struct btrfs_delayed_ref_node *node;
822 struct btrfs_delayed_extent_op *extent_op = head->extent_op;
823 struct btrfs_key key;
824 struct btrfs_key tmp_op_key;
829 if (extent_op && extent_op->update_key)
830 btrfs_disk_key_to_cpu(&tmp_op_key, &extent_op->key);
832 spin_lock(&head->lock);
833 for (n = rb_first_cached(&head->ref_tree); n; n = rb_next(n)) {
834 node = rb_entry(n, struct btrfs_delayed_ref_node,
839 switch (node->action) {
840 case BTRFS_ADD_DELAYED_EXTENT:
841 case BTRFS_UPDATE_DELAYED_HEAD:
844 case BTRFS_ADD_DELAYED_REF:
845 count = node->ref_mod;
847 case BTRFS_DROP_DELAYED_REF:
848 count = node->ref_mod * -1;
853 switch (node->type) {
854 case BTRFS_TREE_BLOCK_REF_KEY: {
855 /* NORMAL INDIRECT METADATA backref */
856 struct btrfs_delayed_tree_ref *ref;
858 ref = btrfs_delayed_node_to_tree_ref(node);
859 ret = add_indirect_ref(fs_info, preftrees, ref->root,
860 &tmp_op_key, ref->level + 1,
861 node->bytenr, count, sc,
865 case BTRFS_SHARED_BLOCK_REF_KEY: {
866 /* SHARED DIRECT METADATA backref */
867 struct btrfs_delayed_tree_ref *ref;
869 ref = btrfs_delayed_node_to_tree_ref(node);
871 ret = add_direct_ref(fs_info, preftrees, ref->level + 1,
872 ref->parent, node->bytenr, count,
876 case BTRFS_EXTENT_DATA_REF_KEY: {
877 /* NORMAL INDIRECT DATA backref */
878 struct btrfs_delayed_data_ref *ref;
879 ref = btrfs_delayed_node_to_data_ref(node);
881 key.objectid = ref->objectid;
882 key.type = BTRFS_EXTENT_DATA_KEY;
883 key.offset = ref->offset;
886 * If we have a share check context and a reference for
887 * another inode, we can't exit immediately. This is
888 * because even if this is a BTRFS_ADD_DELAYED_REF
889 * reference we may find next a BTRFS_DROP_DELAYED_REF
890 * which cancels out this ADD reference.
892 * If this is a DROP reference and there was no previous
893 * ADD reference, then we need to signal that when we
894 * process references from the extent tree (through
895 * add_inline_refs() and add_keyed_refs()), we should
896 * not exit early if we find a reference for another
897 * inode, because one of the delayed DROP references
898 * may cancel that reference in the extent tree.
901 sc->have_delayed_delete_refs = true;
903 ret = add_indirect_ref(fs_info, preftrees, ref->root,
904 &key, 0, node->bytenr, count, sc,
908 case BTRFS_SHARED_DATA_REF_KEY: {
909 /* SHARED DIRECT FULL backref */
910 struct btrfs_delayed_data_ref *ref;
912 ref = btrfs_delayed_node_to_data_ref(node);
914 ret = add_direct_ref(fs_info, preftrees, 0, ref->parent,
915 node->bytenr, count, sc,
923 * We must ignore BACKREF_FOUND_SHARED until all delayed
924 * refs have been checked.
926 if (ret && (ret != BACKREF_FOUND_SHARED))
930 ret = extent_is_shared(sc);
932 spin_unlock(&head->lock);
937 * add all inline backrefs for bytenr to the list
939 * Returns 0 on success, <0 on error, or BACKREF_FOUND_SHARED.
941 static int add_inline_refs(const struct btrfs_fs_info *fs_info,
942 struct btrfs_path *path, u64 bytenr,
943 int *info_level, struct preftrees *preftrees,
944 struct share_check *sc)
948 struct extent_buffer *leaf;
949 struct btrfs_key key;
950 struct btrfs_key found_key;
953 struct btrfs_extent_item *ei;
958 * enumerate all inline refs
960 leaf = path->nodes[0];
961 slot = path->slots[0];
963 item_size = btrfs_item_size_nr(leaf, slot);
964 BUG_ON(item_size < sizeof(*ei));
966 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
967 flags = btrfs_extent_flags(leaf, ei);
968 btrfs_item_key_to_cpu(leaf, &found_key, slot);
970 ptr = (unsigned long)(ei + 1);
971 end = (unsigned long)ei + item_size;
973 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
974 flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
975 struct btrfs_tree_block_info *info;
977 info = (struct btrfs_tree_block_info *)ptr;
978 *info_level = btrfs_tree_block_level(leaf, info);
979 ptr += sizeof(struct btrfs_tree_block_info);
981 } else if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
982 *info_level = found_key.offset;
984 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
988 struct btrfs_extent_inline_ref *iref;
992 iref = (struct btrfs_extent_inline_ref *)ptr;
993 type = btrfs_get_extent_inline_ref_type(leaf, iref,
995 if (type == BTRFS_REF_TYPE_INVALID)
998 offset = btrfs_extent_inline_ref_offset(leaf, iref);
1001 case BTRFS_SHARED_BLOCK_REF_KEY:
1002 ret = add_direct_ref(fs_info, preftrees,
1003 *info_level + 1, offset,
1004 bytenr, 1, NULL, GFP_NOFS);
1006 case BTRFS_SHARED_DATA_REF_KEY: {
1007 struct btrfs_shared_data_ref *sdref;
1010 sdref = (struct btrfs_shared_data_ref *)(iref + 1);
1011 count = btrfs_shared_data_ref_count(leaf, sdref);
1013 ret = add_direct_ref(fs_info, preftrees, 0, offset,
1014 bytenr, count, sc, GFP_NOFS);
1017 case BTRFS_TREE_BLOCK_REF_KEY:
1018 ret = add_indirect_ref(fs_info, preftrees, offset,
1019 NULL, *info_level + 1,
1020 bytenr, 1, NULL, GFP_NOFS);
1022 case BTRFS_EXTENT_DATA_REF_KEY: {
1023 struct btrfs_extent_data_ref *dref;
1027 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1028 count = btrfs_extent_data_ref_count(leaf, dref);
1029 key.objectid = btrfs_extent_data_ref_objectid(leaf,
1031 key.type = BTRFS_EXTENT_DATA_KEY;
1032 key.offset = btrfs_extent_data_ref_offset(leaf, dref);
1034 if (sc && sc->inum && key.objectid != sc->inum &&
1035 !sc->have_delayed_delete_refs) {
1036 ret = BACKREF_FOUND_SHARED;
1040 root = btrfs_extent_data_ref_root(leaf, dref);
1042 ret = add_indirect_ref(fs_info, preftrees, root,
1043 &key, 0, bytenr, count,
1053 ptr += btrfs_extent_inline_ref_size(type);
1060 * add all non-inline backrefs for bytenr to the list
1062 * Returns 0 on success, <0 on error, or BACKREF_FOUND_SHARED.
1064 static int add_keyed_refs(struct btrfs_fs_info *fs_info,
1065 struct btrfs_path *path, u64 bytenr,
1066 int info_level, struct preftrees *preftrees,
1067 struct share_check *sc)
1069 struct btrfs_root *extent_root = fs_info->extent_root;
1072 struct extent_buffer *leaf;
1073 struct btrfs_key key;
1076 ret = btrfs_next_item(extent_root, path);
1084 slot = path->slots[0];
1085 leaf = path->nodes[0];
1086 btrfs_item_key_to_cpu(leaf, &key, slot);
1088 if (key.objectid != bytenr)
1090 if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
1092 if (key.type > BTRFS_SHARED_DATA_REF_KEY)
1096 case BTRFS_SHARED_BLOCK_REF_KEY:
1097 /* SHARED DIRECT METADATA backref */
1098 ret = add_direct_ref(fs_info, preftrees,
1099 info_level + 1, key.offset,
1100 bytenr, 1, NULL, GFP_NOFS);
1102 case BTRFS_SHARED_DATA_REF_KEY: {
1103 /* SHARED DIRECT FULL backref */
1104 struct btrfs_shared_data_ref *sdref;
1107 sdref = btrfs_item_ptr(leaf, slot,
1108 struct btrfs_shared_data_ref);
1109 count = btrfs_shared_data_ref_count(leaf, sdref);
1110 ret = add_direct_ref(fs_info, preftrees, 0,
1111 key.offset, bytenr, count,
1115 case BTRFS_TREE_BLOCK_REF_KEY:
1116 /* NORMAL INDIRECT METADATA backref */
1117 ret = add_indirect_ref(fs_info, preftrees, key.offset,
1118 NULL, info_level + 1, bytenr,
1121 case BTRFS_EXTENT_DATA_REF_KEY: {
1122 /* NORMAL INDIRECT DATA backref */
1123 struct btrfs_extent_data_ref *dref;
1127 dref = btrfs_item_ptr(leaf, slot,
1128 struct btrfs_extent_data_ref);
1129 count = btrfs_extent_data_ref_count(leaf, dref);
1130 key.objectid = btrfs_extent_data_ref_objectid(leaf,
1132 key.type = BTRFS_EXTENT_DATA_KEY;
1133 key.offset = btrfs_extent_data_ref_offset(leaf, dref);
1135 if (sc && sc->inum && key.objectid != sc->inum &&
1136 !sc->have_delayed_delete_refs) {
1137 ret = BACKREF_FOUND_SHARED;
1141 root = btrfs_extent_data_ref_root(leaf, dref);
1142 ret = add_indirect_ref(fs_info, preftrees, root,
1143 &key, 0, bytenr, count,
1159 * this adds all existing backrefs (inline backrefs, backrefs and delayed
1160 * refs) for the given bytenr to the refs list, merges duplicates and resolves
1161 * indirect refs to their parent bytenr.
1162 * When roots are found, they're added to the roots list
1164 * If time_seq is set to BTRFS_SEQ_LAST, it will not search delayed_refs, and
1165 * behave much like trans == NULL case, the difference only lies in it will not
1167 * The special case is for qgroup to search roots in commit_transaction().
1169 * @sc - if !NULL, then immediately return BACKREF_FOUND_SHARED when a
1170 * shared extent is detected.
1172 * Otherwise this returns 0 for success and <0 for an error.
1174 * If ignore_offset is set to false, only extent refs whose offsets match
1175 * extent_item_pos are returned. If true, every extent ref is returned
1176 * and extent_item_pos is ignored.
1178 * FIXME some caching might speed things up
1180 static int find_parent_nodes(struct btrfs_trans_handle *trans,
1181 struct btrfs_fs_info *fs_info, u64 bytenr,
1182 u64 time_seq, struct ulist *refs,
1183 struct ulist *roots, const u64 *extent_item_pos,
1184 struct share_check *sc, bool ignore_offset)
1186 struct btrfs_key key;
1187 struct btrfs_path *path;
1188 struct btrfs_delayed_ref_root *delayed_refs = NULL;
1189 struct btrfs_delayed_ref_head *head;
1192 struct prelim_ref *ref;
1193 struct rb_node *node;
1194 struct extent_inode_elem *eie = NULL;
1195 struct preftrees preftrees = {
1196 .direct = PREFTREE_INIT,
1197 .indirect = PREFTREE_INIT,
1198 .indirect_missing_keys = PREFTREE_INIT
1201 key.objectid = bytenr;
1202 key.offset = (u64)-1;
1203 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1204 key.type = BTRFS_METADATA_ITEM_KEY;
1206 key.type = BTRFS_EXTENT_ITEM_KEY;
1208 path = btrfs_alloc_path();
1212 path->search_commit_root = 1;
1213 path->skip_locking = 1;
1216 if (time_seq == BTRFS_SEQ_LAST)
1217 path->skip_locking = 1;
1220 * grab both a lock on the path and a lock on the delayed ref head.
1221 * We need both to get a consistent picture of how the refs look
1222 * at a specified point in time
1227 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
1231 /* This shouldn't happen, indicates a bug or fs corruption. */
1237 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1238 if (trans && likely(trans->type != __TRANS_DUMMY) &&
1239 time_seq != BTRFS_SEQ_LAST) {
1241 if (trans && time_seq != BTRFS_SEQ_LAST) {
1244 * look if there are updates for this ref queued and lock the
1247 delayed_refs = &trans->transaction->delayed_refs;
1248 spin_lock(&delayed_refs->lock);
1249 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
1251 if (!mutex_trylock(&head->mutex)) {
1252 refcount_inc(&head->refs);
1253 spin_unlock(&delayed_refs->lock);
1255 btrfs_release_path(path);
1258 * Mutex was contended, block until it's
1259 * released and try again
1261 mutex_lock(&head->mutex);
1262 mutex_unlock(&head->mutex);
1263 btrfs_put_delayed_ref_head(head);
1266 spin_unlock(&delayed_refs->lock);
1267 ret = add_delayed_refs(fs_info, head, time_seq,
1269 mutex_unlock(&head->mutex);
1273 spin_unlock(&delayed_refs->lock);
1277 if (path->slots[0]) {
1278 struct extent_buffer *leaf;
1282 leaf = path->nodes[0];
1283 slot = path->slots[0];
1284 btrfs_item_key_to_cpu(leaf, &key, slot);
1285 if (key.objectid == bytenr &&
1286 (key.type == BTRFS_EXTENT_ITEM_KEY ||
1287 key.type == BTRFS_METADATA_ITEM_KEY)) {
1288 ret = add_inline_refs(fs_info, path, bytenr,
1289 &info_level, &preftrees, sc);
1292 ret = add_keyed_refs(fs_info, path, bytenr, info_level,
1299 btrfs_release_path(path);
1301 ret = add_missing_keys(fs_info, &preftrees, path->skip_locking == 0);
1305 WARN_ON(!RB_EMPTY_ROOT(&preftrees.indirect_missing_keys.root.rb_root));
1307 ret = resolve_indirect_refs(fs_info, path, time_seq, &preftrees,
1308 extent_item_pos, sc, ignore_offset);
1312 WARN_ON(!RB_EMPTY_ROOT(&preftrees.indirect.root.rb_root));
1315 * This walks the tree of merged and resolved refs. Tree blocks are
1316 * read in as needed. Unique entries are added to the ulist, and
1317 * the list of found roots is updated.
1319 * We release the entire tree in one go before returning.
1321 node = rb_first_cached(&preftrees.direct.root);
1323 ref = rb_entry(node, struct prelim_ref, rbnode);
1324 node = rb_next(&ref->rbnode);
1326 * ref->count < 0 can happen here if there are delayed
1327 * refs with a node->action of BTRFS_DROP_DELAYED_REF.
1328 * prelim_ref_insert() relies on this when merging
1329 * identical refs to keep the overall count correct.
1330 * prelim_ref_insert() will merge only those refs
1331 * which compare identically. Any refs having
1332 * e.g. different offsets would not be merged,
1333 * and would retain their original ref->count < 0.
1335 if (roots && ref->count && ref->root_id && ref->parent == 0) {
1336 if (sc && sc->root_objectid &&
1337 ref->root_id != sc->root_objectid) {
1338 ret = BACKREF_FOUND_SHARED;
1342 /* no parent == root of tree */
1343 ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS);
1347 if (ref->count && ref->parent) {
1348 if (extent_item_pos && !ref->inode_list &&
1350 struct extent_buffer *eb;
1352 eb = read_tree_block(fs_info, ref->parent, 0,
1353 0, ref->level, NULL);
1357 } else if (!extent_buffer_uptodate(eb)) {
1358 free_extent_buffer(eb);
1363 if (!path->skip_locking)
1364 btrfs_tree_read_lock(eb);
1365 ret = find_extent_in_eb(eb, bytenr,
1366 *extent_item_pos, &eie, ignore_offset);
1367 if (!path->skip_locking)
1368 btrfs_tree_read_unlock(eb);
1369 free_extent_buffer(eb);
1372 ref->inode_list = eie;
1374 ret = ulist_add_merge_ptr(refs, ref->parent,
1376 (void **)&eie, GFP_NOFS);
1379 if (!ret && extent_item_pos) {
1381 * We've recorded that parent, so we must extend
1382 * its inode list here.
1384 * However if there was corruption we may not
1385 * have found an eie, return an error in this
1395 eie->next = ref->inode_list;
1403 btrfs_free_path(path);
1405 prelim_release(&preftrees.direct);
1406 prelim_release(&preftrees.indirect);
1407 prelim_release(&preftrees.indirect_missing_keys);
1410 free_inode_elem_list(eie);
1414 static void free_leaf_list(struct ulist *blocks)
1416 struct ulist_node *node = NULL;
1417 struct extent_inode_elem *eie;
1418 struct ulist_iterator uiter;
1420 ULIST_ITER_INIT(&uiter);
1421 while ((node = ulist_next(blocks, &uiter))) {
1424 eie = unode_aux_to_inode_list(node);
1425 free_inode_elem_list(eie);
1433 * Finds all leafs with a reference to the specified combination of bytenr and
1434 * offset. key_list_head will point to a list of corresponding keys (caller must
1435 * free each list element). The leafs will be stored in the leafs ulist, which
1436 * must be freed with ulist_free.
1438 * returns 0 on success, <0 on error
1440 int btrfs_find_all_leafs(struct btrfs_trans_handle *trans,
1441 struct btrfs_fs_info *fs_info, u64 bytenr,
1442 u64 time_seq, struct ulist **leafs,
1443 const u64 *extent_item_pos, bool ignore_offset)
1447 *leafs = ulist_alloc(GFP_NOFS);
1451 ret = find_parent_nodes(trans, fs_info, bytenr, time_seq,
1452 *leafs, NULL, extent_item_pos, NULL, ignore_offset);
1453 if (ret < 0 && ret != -ENOENT) {
1454 free_leaf_list(*leafs);
1462 * walk all backrefs for a given extent to find all roots that reference this
1463 * extent. Walking a backref means finding all extents that reference this
1464 * extent and in turn walk the backrefs of those, too. Naturally this is a
1465 * recursive process, but here it is implemented in an iterative fashion: We
1466 * find all referencing extents for the extent in question and put them on a
1467 * list. In turn, we find all referencing extents for those, further appending
1468 * to the list. The way we iterate the list allows adding more elements after
1469 * the current while iterating. The process stops when we reach the end of the
1470 * list. Found roots are added to the roots list.
1472 * returns 0 on success, < 0 on error.
1474 static int btrfs_find_all_roots_safe(struct btrfs_trans_handle *trans,
1475 struct btrfs_fs_info *fs_info, u64 bytenr,
1476 u64 time_seq, struct ulist **roots,
1480 struct ulist_node *node = NULL;
1481 struct ulist_iterator uiter;
1484 tmp = ulist_alloc(GFP_NOFS);
1487 *roots = ulist_alloc(GFP_NOFS);
1493 ULIST_ITER_INIT(&uiter);
1495 ret = find_parent_nodes(trans, fs_info, bytenr, time_seq,
1496 tmp, *roots, NULL, NULL, ignore_offset);
1497 if (ret < 0 && ret != -ENOENT) {
1503 node = ulist_next(tmp, &uiter);
1514 int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1515 struct btrfs_fs_info *fs_info, u64 bytenr,
1516 u64 time_seq, struct ulist **roots,
1517 bool skip_commit_root_sem)
1521 if (!trans && !skip_commit_root_sem)
1522 down_read(&fs_info->commit_root_sem);
1523 ret = btrfs_find_all_roots_safe(trans, fs_info, bytenr,
1524 time_seq, roots, false);
1525 if (!trans && !skip_commit_root_sem)
1526 up_read(&fs_info->commit_root_sem);
1531 * Check if an extent is shared or not
1533 * @root: root inode belongs to
1534 * @inum: inode number of the inode whose extent we are checking
1535 * @bytenr: logical bytenr of the extent we are checking
1536 * @roots: list of roots this extent is shared among
1537 * @tmp: temporary list used for iteration
1539 * btrfs_check_shared uses the backref walking code but will short
1540 * circuit as soon as it finds a root or inode that doesn't match the
1541 * one passed in. This provides a significant performance benefit for
1542 * callers (such as fiemap) which want to know whether the extent is
1543 * shared but do not need a ref count.
1545 * This attempts to attach to the running transaction in order to account for
1546 * delayed refs, but continues on even when no running transaction exists.
1548 * Return: 0 if extent is not shared, 1 if it is shared, < 0 on error.
1550 int btrfs_check_shared(struct btrfs_root *root, u64 inum, u64 bytenr,
1551 struct ulist *roots, struct ulist *tmp)
1553 struct btrfs_fs_info *fs_info = root->fs_info;
1554 struct btrfs_trans_handle *trans;
1555 struct ulist_iterator uiter;
1556 struct ulist_node *node;
1557 struct btrfs_seq_list elem = BTRFS_SEQ_LIST_INIT(elem);
1559 struct share_check shared = {
1560 .root_objectid = root->root_key.objectid,
1563 .have_delayed_delete_refs = false,
1569 trans = btrfs_join_transaction_nostart(root);
1570 if (IS_ERR(trans)) {
1571 if (PTR_ERR(trans) != -ENOENT && PTR_ERR(trans) != -EROFS) {
1572 ret = PTR_ERR(trans);
1576 down_read(&fs_info->commit_root_sem);
1578 btrfs_get_tree_mod_seq(fs_info, &elem);
1581 ULIST_ITER_INIT(&uiter);
1583 ret = find_parent_nodes(trans, fs_info, bytenr, elem.seq, tmp,
1584 roots, NULL, &shared, false);
1585 if (ret == BACKREF_FOUND_SHARED) {
1586 /* this is the only condition under which we return 1 */
1590 if (ret < 0 && ret != -ENOENT)
1593 node = ulist_next(tmp, &uiter);
1597 shared.share_count = 0;
1598 shared.have_delayed_delete_refs = false;
1603 btrfs_put_tree_mod_seq(fs_info, &elem);
1604 btrfs_end_transaction(trans);
1606 up_read(&fs_info->commit_root_sem);
1609 ulist_release(roots);
1614 int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,
1615 u64 start_off, struct btrfs_path *path,
1616 struct btrfs_inode_extref **ret_extref,
1620 struct btrfs_key key;
1621 struct btrfs_key found_key;
1622 struct btrfs_inode_extref *extref;
1623 const struct extent_buffer *leaf;
1626 key.objectid = inode_objectid;
1627 key.type = BTRFS_INODE_EXTREF_KEY;
1628 key.offset = start_off;
1630 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1635 leaf = path->nodes[0];
1636 slot = path->slots[0];
1637 if (slot >= btrfs_header_nritems(leaf)) {
1639 * If the item at offset is not found,
1640 * btrfs_search_slot will point us to the slot
1641 * where it should be inserted. In our case
1642 * that will be the slot directly before the
1643 * next INODE_REF_KEY_V2 item. In the case
1644 * that we're pointing to the last slot in a
1645 * leaf, we must move one leaf over.
1647 ret = btrfs_next_leaf(root, path);
1656 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1659 * Check that we're still looking at an extended ref key for
1660 * this particular objectid. If we have different
1661 * objectid or type then there are no more to be found
1662 * in the tree and we can exit.
1665 if (found_key.objectid != inode_objectid)
1667 if (found_key.type != BTRFS_INODE_EXTREF_KEY)
1671 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1672 extref = (struct btrfs_inode_extref *)ptr;
1673 *ret_extref = extref;
1675 *found_off = found_key.offset;
1683 * this iterates to turn a name (from iref/extref) into a full filesystem path.
1684 * Elements of the path are separated by '/' and the path is guaranteed to be
1685 * 0-terminated. the path is only given within the current file system.
1686 * Therefore, it never starts with a '/'. the caller is responsible to provide
1687 * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
1688 * the start point of the resulting string is returned. this pointer is within
1690 * in case the path buffer would overflow, the pointer is decremented further
1691 * as if output was written to the buffer, though no more output is actually
1692 * generated. that way, the caller can determine how much space would be
1693 * required for the path to fit into the buffer. in that case, the returned
1694 * value will be smaller than dest. callers must check this!
1696 char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
1697 u32 name_len, unsigned long name_off,
1698 struct extent_buffer *eb_in, u64 parent,
1699 char *dest, u32 size)
1704 s64 bytes_left = ((s64)size) - 1;
1705 struct extent_buffer *eb = eb_in;
1706 struct btrfs_key found_key;
1707 struct btrfs_inode_ref *iref;
1709 if (bytes_left >= 0)
1710 dest[bytes_left] = '\0';
1713 bytes_left -= name_len;
1714 if (bytes_left >= 0)
1715 read_extent_buffer(eb, dest + bytes_left,
1716 name_off, name_len);
1718 if (!path->skip_locking)
1719 btrfs_tree_read_unlock(eb);
1720 free_extent_buffer(eb);
1722 ret = btrfs_find_item(fs_root, path, parent, 0,
1723 BTRFS_INODE_REF_KEY, &found_key);
1729 next_inum = found_key.offset;
1731 /* regular exit ahead */
1732 if (parent == next_inum)
1735 slot = path->slots[0];
1736 eb = path->nodes[0];
1737 /* make sure we can use eb after releasing the path */
1739 path->nodes[0] = NULL;
1742 btrfs_release_path(path);
1743 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1745 name_len = btrfs_inode_ref_name_len(eb, iref);
1746 name_off = (unsigned long)(iref + 1);
1750 if (bytes_left >= 0)
1751 dest[bytes_left] = '/';
1754 btrfs_release_path(path);
1757 return ERR_PTR(ret);
1759 return dest + bytes_left;
1763 * this makes the path point to (logical EXTENT_ITEM *)
1764 * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
1765 * tree blocks and <0 on error.
1767 int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
1768 struct btrfs_path *path, struct btrfs_key *found_key,
1775 const struct extent_buffer *eb;
1776 struct btrfs_extent_item *ei;
1777 struct btrfs_key key;
1779 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1780 key.type = BTRFS_METADATA_ITEM_KEY;
1782 key.type = BTRFS_EXTENT_ITEM_KEY;
1783 key.objectid = logical;
1784 key.offset = (u64)-1;
1786 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
1790 ret = btrfs_previous_extent_item(fs_info->extent_root, path, 0);
1796 btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]);
1797 if (found_key->type == BTRFS_METADATA_ITEM_KEY)
1798 size = fs_info->nodesize;
1799 else if (found_key->type == BTRFS_EXTENT_ITEM_KEY)
1800 size = found_key->offset;
1802 if (found_key->objectid > logical ||
1803 found_key->objectid + size <= logical) {
1804 btrfs_debug(fs_info,
1805 "logical %llu is not within any extent", logical);
1809 eb = path->nodes[0];
1810 item_size = btrfs_item_size_nr(eb, path->slots[0]);
1811 BUG_ON(item_size < sizeof(*ei));
1813 ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
1814 flags = btrfs_extent_flags(eb, ei);
1816 btrfs_debug(fs_info,
1817 "logical %llu is at position %llu within the extent (%llu EXTENT_ITEM %llu) flags %#llx size %u",
1818 logical, logical - found_key->objectid, found_key->objectid,
1819 found_key->offset, flags, item_size);
1821 WARN_ON(!flags_ret);
1823 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1824 *flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK;
1825 else if (flags & BTRFS_EXTENT_FLAG_DATA)
1826 *flags_ret = BTRFS_EXTENT_FLAG_DATA;
1836 * helper function to iterate extent inline refs. ptr must point to a 0 value
1837 * for the first call and may be modified. it is used to track state.
1838 * if more refs exist, 0 is returned and the next call to
1839 * get_extent_inline_ref must pass the modified ptr parameter to get the
1840 * next ref. after the last ref was processed, 1 is returned.
1841 * returns <0 on error
1843 static int get_extent_inline_ref(unsigned long *ptr,
1844 const struct extent_buffer *eb,
1845 const struct btrfs_key *key,
1846 const struct btrfs_extent_item *ei,
1848 struct btrfs_extent_inline_ref **out_eiref,
1853 struct btrfs_tree_block_info *info;
1857 flags = btrfs_extent_flags(eb, ei);
1858 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1859 if (key->type == BTRFS_METADATA_ITEM_KEY) {
1860 /* a skinny metadata extent */
1862 (struct btrfs_extent_inline_ref *)(ei + 1);
1864 WARN_ON(key->type != BTRFS_EXTENT_ITEM_KEY);
1865 info = (struct btrfs_tree_block_info *)(ei + 1);
1867 (struct btrfs_extent_inline_ref *)(info + 1);
1870 *out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1);
1872 *ptr = (unsigned long)*out_eiref;
1873 if ((unsigned long)(*ptr) >= (unsigned long)ei + item_size)
1877 end = (unsigned long)ei + item_size;
1878 *out_eiref = (struct btrfs_extent_inline_ref *)(*ptr);
1879 *out_type = btrfs_get_extent_inline_ref_type(eb, *out_eiref,
1880 BTRFS_REF_TYPE_ANY);
1881 if (*out_type == BTRFS_REF_TYPE_INVALID)
1884 *ptr += btrfs_extent_inline_ref_size(*out_type);
1885 WARN_ON(*ptr > end);
1887 return 1; /* last */
1893 * reads the tree block backref for an extent. tree level and root are returned
1894 * through out_level and out_root. ptr must point to a 0 value for the first
1895 * call and may be modified (see get_extent_inline_ref comment).
1896 * returns 0 if data was provided, 1 if there was no more data to provide or
1899 int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
1900 struct btrfs_key *key, struct btrfs_extent_item *ei,
1901 u32 item_size, u64 *out_root, u8 *out_level)
1905 struct btrfs_extent_inline_ref *eiref;
1907 if (*ptr == (unsigned long)-1)
1911 ret = get_extent_inline_ref(ptr, eb, key, ei, item_size,
1916 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1917 type == BTRFS_SHARED_BLOCK_REF_KEY)
1924 /* we can treat both ref types equally here */
1925 *out_root = btrfs_extent_inline_ref_offset(eb, eiref);
1927 if (key->type == BTRFS_EXTENT_ITEM_KEY) {
1928 struct btrfs_tree_block_info *info;
1930 info = (struct btrfs_tree_block_info *)(ei + 1);
1931 *out_level = btrfs_tree_block_level(eb, info);
1933 ASSERT(key->type == BTRFS_METADATA_ITEM_KEY);
1934 *out_level = (u8)key->offset;
1938 *ptr = (unsigned long)-1;
1943 static int iterate_leaf_refs(struct btrfs_fs_info *fs_info,
1944 struct extent_inode_elem *inode_list,
1945 u64 root, u64 extent_item_objectid,
1946 iterate_extent_inodes_t *iterate, void *ctx)
1948 struct extent_inode_elem *eie;
1951 for (eie = inode_list; eie; eie = eie->next) {
1952 btrfs_debug(fs_info,
1953 "ref for %llu resolved, key (%llu EXTEND_DATA %llu), root %llu",
1954 extent_item_objectid, eie->inum,
1956 ret = iterate(eie->inum, eie->offset, root, ctx);
1958 btrfs_debug(fs_info,
1959 "stopping iteration for %llu due to ret=%d",
1960 extent_item_objectid, ret);
1969 * calls iterate() for every inode that references the extent identified by
1970 * the given parameters.
1971 * when the iterator function returns a non-zero value, iteration stops.
1973 int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
1974 u64 extent_item_objectid, u64 extent_item_pos,
1975 int search_commit_root,
1976 iterate_extent_inodes_t *iterate, void *ctx,
1980 struct btrfs_trans_handle *trans = NULL;
1981 struct ulist *refs = NULL;
1982 struct ulist *roots = NULL;
1983 struct ulist_node *ref_node = NULL;
1984 struct ulist_node *root_node = NULL;
1985 struct btrfs_seq_list seq_elem = BTRFS_SEQ_LIST_INIT(seq_elem);
1986 struct ulist_iterator ref_uiter;
1987 struct ulist_iterator root_uiter;
1989 btrfs_debug(fs_info, "resolving all inodes for extent %llu",
1990 extent_item_objectid);
1992 if (!search_commit_root) {
1993 trans = btrfs_attach_transaction(fs_info->extent_root);
1994 if (IS_ERR(trans)) {
1995 if (PTR_ERR(trans) != -ENOENT &&
1996 PTR_ERR(trans) != -EROFS)
1997 return PTR_ERR(trans);
2003 btrfs_get_tree_mod_seq(fs_info, &seq_elem);
2005 down_read(&fs_info->commit_root_sem);
2007 ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid,
2008 seq_elem.seq, &refs,
2009 &extent_item_pos, ignore_offset);
2013 ULIST_ITER_INIT(&ref_uiter);
2014 while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) {
2015 ret = btrfs_find_all_roots_safe(trans, fs_info, ref_node->val,
2016 seq_elem.seq, &roots,
2020 ULIST_ITER_INIT(&root_uiter);
2021 while (!ret && (root_node = ulist_next(roots, &root_uiter))) {
2022 btrfs_debug(fs_info,
2023 "root %llu references leaf %llu, data list %#llx",
2024 root_node->val, ref_node->val,
2026 ret = iterate_leaf_refs(fs_info,
2027 (struct extent_inode_elem *)
2028 (uintptr_t)ref_node->aux,
2030 extent_item_objectid,
2036 free_leaf_list(refs);
2039 btrfs_put_tree_mod_seq(fs_info, &seq_elem);
2040 btrfs_end_transaction(trans);
2042 up_read(&fs_info->commit_root_sem);
2048 int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
2049 struct btrfs_path *path,
2050 iterate_extent_inodes_t *iterate, void *ctx,
2054 u64 extent_item_pos;
2056 struct btrfs_key found_key;
2057 int search_commit_root = path->search_commit_root;
2059 ret = extent_from_logical(fs_info, logical, path, &found_key, &flags);
2060 btrfs_release_path(path);
2063 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
2066 extent_item_pos = logical - found_key.objectid;
2067 ret = iterate_extent_inodes(fs_info, found_key.objectid,
2068 extent_item_pos, search_commit_root,
2069 iterate, ctx, ignore_offset);
2074 typedef int (iterate_irefs_t)(u64 parent, u32 name_len, unsigned long name_off,
2075 struct extent_buffer *eb, void *ctx);
2077 static int iterate_inode_refs(u64 inum, struct btrfs_root *fs_root,
2078 struct btrfs_path *path,
2079 iterate_irefs_t *iterate, void *ctx)
2088 struct extent_buffer *eb;
2089 struct btrfs_item *item;
2090 struct btrfs_inode_ref *iref;
2091 struct btrfs_key found_key;
2094 ret = btrfs_find_item(fs_root, path, inum,
2095 parent ? parent + 1 : 0, BTRFS_INODE_REF_KEY,
2101 ret = found ? 0 : -ENOENT;
2106 parent = found_key.offset;
2107 slot = path->slots[0];
2108 eb = btrfs_clone_extent_buffer(path->nodes[0]);
2113 btrfs_release_path(path);
2115 item = btrfs_item_nr(slot);
2116 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
2118 for (cur = 0; cur < btrfs_item_size(eb, item); cur += len) {
2119 name_len = btrfs_inode_ref_name_len(eb, iref);
2120 /* path must be released before calling iterate()! */
2121 btrfs_debug(fs_root->fs_info,
2122 "following ref at offset %u for inode %llu in tree %llu",
2123 cur, found_key.objectid,
2124 fs_root->root_key.objectid);
2125 ret = iterate(parent, name_len,
2126 (unsigned long)(iref + 1), eb, ctx);
2129 len = sizeof(*iref) + name_len;
2130 iref = (struct btrfs_inode_ref *)((char *)iref + len);
2132 free_extent_buffer(eb);
2135 btrfs_release_path(path);
2140 static int iterate_inode_extrefs(u64 inum, struct btrfs_root *fs_root,
2141 struct btrfs_path *path,
2142 iterate_irefs_t *iterate, void *ctx)
2149 struct extent_buffer *eb;
2150 struct btrfs_inode_extref *extref;
2156 ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref,
2161 ret = found ? 0 : -ENOENT;
2166 slot = path->slots[0];
2167 eb = btrfs_clone_extent_buffer(path->nodes[0]);
2172 btrfs_release_path(path);
2174 item_size = btrfs_item_size_nr(eb, slot);
2175 ptr = btrfs_item_ptr_offset(eb, slot);
2178 while (cur_offset < item_size) {
2181 extref = (struct btrfs_inode_extref *)(ptr + cur_offset);
2182 parent = btrfs_inode_extref_parent(eb, extref);
2183 name_len = btrfs_inode_extref_name_len(eb, extref);
2184 ret = iterate(parent, name_len,
2185 (unsigned long)&extref->name, eb, ctx);
2189 cur_offset += btrfs_inode_extref_name_len(eb, extref);
2190 cur_offset += sizeof(*extref);
2192 free_extent_buffer(eb);
2197 btrfs_release_path(path);
2202 static int iterate_irefs(u64 inum, struct btrfs_root *fs_root,
2203 struct btrfs_path *path, iterate_irefs_t *iterate,
2209 ret = iterate_inode_refs(inum, fs_root, path, iterate, ctx);
2212 else if (ret != -ENOENT)
2215 ret = iterate_inode_extrefs(inum, fs_root, path, iterate, ctx);
2216 if (ret == -ENOENT && found_refs)
2223 * returns 0 if the path could be dumped (probably truncated)
2224 * returns <0 in case of an error
2226 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
2227 struct extent_buffer *eb, void *ctx)
2229 struct inode_fs_paths *ipath = ctx;
2232 int i = ipath->fspath->elem_cnt;
2233 const int s_ptr = sizeof(char *);
2236 bytes_left = ipath->fspath->bytes_left > s_ptr ?
2237 ipath->fspath->bytes_left - s_ptr : 0;
2239 fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
2240 fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len,
2241 name_off, eb, inum, fspath_min, bytes_left);
2243 return PTR_ERR(fspath);
2245 if (fspath > fspath_min) {
2246 ipath->fspath->val[i] = (u64)(unsigned long)fspath;
2247 ++ipath->fspath->elem_cnt;
2248 ipath->fspath->bytes_left = fspath - fspath_min;
2250 ++ipath->fspath->elem_missed;
2251 ipath->fspath->bytes_missing += fspath_min - fspath;
2252 ipath->fspath->bytes_left = 0;
2259 * this dumps all file system paths to the inode into the ipath struct, provided
2260 * is has been created large enough. each path is zero-terminated and accessed
2261 * from ipath->fspath->val[i].
2262 * when it returns, there are ipath->fspath->elem_cnt number of paths available
2263 * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the
2264 * number of missed paths is recorded in ipath->fspath->elem_missed, otherwise,
2265 * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
2266 * have been needed to return all paths.
2268 int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
2270 return iterate_irefs(inum, ipath->fs_root, ipath->btrfs_path,
2271 inode_to_path, ipath);
2274 struct btrfs_data_container *init_data_container(u32 total_bytes)
2276 struct btrfs_data_container *data;
2279 alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
2280 data = kvmalloc(alloc_bytes, GFP_KERNEL);
2282 return ERR_PTR(-ENOMEM);
2284 if (total_bytes >= sizeof(*data)) {
2285 data->bytes_left = total_bytes - sizeof(*data);
2286 data->bytes_missing = 0;
2288 data->bytes_missing = sizeof(*data) - total_bytes;
2289 data->bytes_left = 0;
2293 data->elem_missed = 0;
2299 * allocates space to return multiple file system paths for an inode.
2300 * total_bytes to allocate are passed, note that space usable for actual path
2301 * information will be total_bytes - sizeof(struct inode_fs_paths).
2302 * the returned pointer must be freed with free_ipath() in the end.
2304 struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
2305 struct btrfs_path *path)
2307 struct inode_fs_paths *ifp;
2308 struct btrfs_data_container *fspath;
2310 fspath = init_data_container(total_bytes);
2312 return ERR_CAST(fspath);
2314 ifp = kmalloc(sizeof(*ifp), GFP_KERNEL);
2317 return ERR_PTR(-ENOMEM);
2320 ifp->btrfs_path = path;
2321 ifp->fspath = fspath;
2322 ifp->fs_root = fs_root;
2327 void free_ipath(struct inode_fs_paths *ipath)
2331 kvfree(ipath->fspath);
2335 struct btrfs_backref_iter *btrfs_backref_iter_alloc(
2336 struct btrfs_fs_info *fs_info, gfp_t gfp_flag)
2338 struct btrfs_backref_iter *ret;
2340 ret = kzalloc(sizeof(*ret), gfp_flag);
2344 ret->path = btrfs_alloc_path();
2350 /* Current backref iterator only supports iteration in commit root */
2351 ret->path->search_commit_root = 1;
2352 ret->path->skip_locking = 1;
2353 ret->fs_info = fs_info;
2358 int btrfs_backref_iter_start(struct btrfs_backref_iter *iter, u64 bytenr)
2360 struct btrfs_fs_info *fs_info = iter->fs_info;
2361 struct btrfs_path *path = iter->path;
2362 struct btrfs_extent_item *ei;
2363 struct btrfs_key key;
2366 key.objectid = bytenr;
2367 key.type = BTRFS_METADATA_ITEM_KEY;
2368 key.offset = (u64)-1;
2369 iter->bytenr = bytenr;
2371 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
2378 if (path->slots[0] == 0) {
2379 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
2385 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2386 if ((key.type != BTRFS_EXTENT_ITEM_KEY &&
2387 key.type != BTRFS_METADATA_ITEM_KEY) || key.objectid != bytenr) {
2391 memcpy(&iter->cur_key, &key, sizeof(key));
2392 iter->item_ptr = (u32)btrfs_item_ptr_offset(path->nodes[0],
2394 iter->end_ptr = (u32)(iter->item_ptr +
2395 btrfs_item_size_nr(path->nodes[0], path->slots[0]));
2396 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
2397 struct btrfs_extent_item);
2400 * Only support iteration on tree backref yet.
2402 * This is an extra precaution for non skinny-metadata, where
2403 * EXTENT_ITEM is also used for tree blocks, that we can only use
2404 * extent flags to determine if it's a tree block.
2406 if (btrfs_extent_flags(path->nodes[0], ei) & BTRFS_EXTENT_FLAG_DATA) {
2410 iter->cur_ptr = (u32)(iter->item_ptr + sizeof(*ei));
2412 /* If there is no inline backref, go search for keyed backref */
2413 if (iter->cur_ptr >= iter->end_ptr) {
2414 ret = btrfs_next_item(fs_info->extent_root, path);
2416 /* No inline nor keyed ref */
2424 btrfs_item_key_to_cpu(path->nodes[0], &iter->cur_key,
2426 if (iter->cur_key.objectid != bytenr ||
2427 (iter->cur_key.type != BTRFS_SHARED_BLOCK_REF_KEY &&
2428 iter->cur_key.type != BTRFS_TREE_BLOCK_REF_KEY)) {
2432 iter->cur_ptr = (u32)btrfs_item_ptr_offset(path->nodes[0],
2434 iter->item_ptr = iter->cur_ptr;
2435 iter->end_ptr = (u32)(iter->item_ptr + btrfs_item_size_nr(
2436 path->nodes[0], path->slots[0]));
2441 btrfs_backref_iter_release(iter);
2446 * Go to the next backref item of current bytenr, can be either inlined or
2449 * Caller needs to check whether it's inline ref or not by iter->cur_key.
2451 * Return 0 if we get next backref without problem.
2452 * Return >0 if there is no extra backref for this bytenr.
2453 * Return <0 if there is something wrong happened.
2455 int btrfs_backref_iter_next(struct btrfs_backref_iter *iter)
2457 struct extent_buffer *eb = btrfs_backref_get_eb(iter);
2458 struct btrfs_path *path = iter->path;
2459 struct btrfs_extent_inline_ref *iref;
2463 if (btrfs_backref_iter_is_inline_ref(iter)) {
2464 /* We're still inside the inline refs */
2465 ASSERT(iter->cur_ptr < iter->end_ptr);
2467 if (btrfs_backref_has_tree_block_info(iter)) {
2468 /* First tree block info */
2469 size = sizeof(struct btrfs_tree_block_info);
2471 /* Use inline ref type to determine the size */
2474 iref = (struct btrfs_extent_inline_ref *)
2475 ((unsigned long)iter->cur_ptr);
2476 type = btrfs_extent_inline_ref_type(eb, iref);
2478 size = btrfs_extent_inline_ref_size(type);
2480 iter->cur_ptr += size;
2481 if (iter->cur_ptr < iter->end_ptr)
2484 /* All inline items iterated, fall through */
2487 /* We're at keyed items, there is no inline item, go to the next one */
2488 ret = btrfs_next_item(iter->fs_info->extent_root, iter->path);
2492 btrfs_item_key_to_cpu(path->nodes[0], &iter->cur_key, path->slots[0]);
2493 if (iter->cur_key.objectid != iter->bytenr ||
2494 (iter->cur_key.type != BTRFS_TREE_BLOCK_REF_KEY &&
2495 iter->cur_key.type != BTRFS_SHARED_BLOCK_REF_KEY))
2497 iter->item_ptr = (u32)btrfs_item_ptr_offset(path->nodes[0],
2499 iter->cur_ptr = iter->item_ptr;
2500 iter->end_ptr = iter->item_ptr + (u32)btrfs_item_size_nr(path->nodes[0],
2505 void btrfs_backref_init_cache(struct btrfs_fs_info *fs_info,
2506 struct btrfs_backref_cache *cache, int is_reloc)
2510 cache->rb_root = RB_ROOT;
2511 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
2512 INIT_LIST_HEAD(&cache->pending[i]);
2513 INIT_LIST_HEAD(&cache->changed);
2514 INIT_LIST_HEAD(&cache->detached);
2515 INIT_LIST_HEAD(&cache->leaves);
2516 INIT_LIST_HEAD(&cache->pending_edge);
2517 INIT_LIST_HEAD(&cache->useless_node);
2518 cache->fs_info = fs_info;
2519 cache->is_reloc = is_reloc;
2522 struct btrfs_backref_node *btrfs_backref_alloc_node(
2523 struct btrfs_backref_cache *cache, u64 bytenr, int level)
2525 struct btrfs_backref_node *node;
2527 ASSERT(level >= 0 && level < BTRFS_MAX_LEVEL);
2528 node = kzalloc(sizeof(*node), GFP_NOFS);
2532 INIT_LIST_HEAD(&node->list);
2533 INIT_LIST_HEAD(&node->upper);
2534 INIT_LIST_HEAD(&node->lower);
2535 RB_CLEAR_NODE(&node->rb_node);
2537 node->level = level;
2538 node->bytenr = bytenr;
2543 struct btrfs_backref_edge *btrfs_backref_alloc_edge(
2544 struct btrfs_backref_cache *cache)
2546 struct btrfs_backref_edge *edge;
2548 edge = kzalloc(sizeof(*edge), GFP_NOFS);
2555 * Drop the backref node from cache, also cleaning up all its
2556 * upper edges and any uncached nodes in the path.
2558 * This cleanup happens bottom up, thus the node should either
2559 * be the lowest node in the cache or a detached node.
2561 void btrfs_backref_cleanup_node(struct btrfs_backref_cache *cache,
2562 struct btrfs_backref_node *node)
2564 struct btrfs_backref_node *upper;
2565 struct btrfs_backref_edge *edge;
2570 BUG_ON(!node->lowest && !node->detached);
2571 while (!list_empty(&node->upper)) {
2572 edge = list_entry(node->upper.next, struct btrfs_backref_edge,
2574 upper = edge->node[UPPER];
2575 list_del(&edge->list[LOWER]);
2576 list_del(&edge->list[UPPER]);
2577 btrfs_backref_free_edge(cache, edge);
2580 * Add the node to leaf node list if no other child block
2583 if (list_empty(&upper->lower)) {
2584 list_add_tail(&upper->lower, &cache->leaves);
2589 btrfs_backref_drop_node(cache, node);
2593 * Release all nodes/edges from current cache
2595 void btrfs_backref_release_cache(struct btrfs_backref_cache *cache)
2597 struct btrfs_backref_node *node;
2600 while (!list_empty(&cache->detached)) {
2601 node = list_entry(cache->detached.next,
2602 struct btrfs_backref_node, list);
2603 btrfs_backref_cleanup_node(cache, node);
2606 while (!list_empty(&cache->leaves)) {
2607 node = list_entry(cache->leaves.next,
2608 struct btrfs_backref_node, lower);
2609 btrfs_backref_cleanup_node(cache, node);
2612 cache->last_trans = 0;
2614 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
2615 ASSERT(list_empty(&cache->pending[i]));
2616 ASSERT(list_empty(&cache->pending_edge));
2617 ASSERT(list_empty(&cache->useless_node));
2618 ASSERT(list_empty(&cache->changed));
2619 ASSERT(list_empty(&cache->detached));
2620 ASSERT(RB_EMPTY_ROOT(&cache->rb_root));
2621 ASSERT(!cache->nr_nodes);
2622 ASSERT(!cache->nr_edges);
2626 * Handle direct tree backref
2628 * Direct tree backref means, the backref item shows its parent bytenr
2629 * directly. This is for SHARED_BLOCK_REF backref (keyed or inlined).
2631 * @ref_key: The converted backref key.
2632 * For keyed backref, it's the item key.
2633 * For inlined backref, objectid is the bytenr,
2634 * type is btrfs_inline_ref_type, offset is
2635 * btrfs_inline_ref_offset.
2637 static int handle_direct_tree_backref(struct btrfs_backref_cache *cache,
2638 struct btrfs_key *ref_key,
2639 struct btrfs_backref_node *cur)
2641 struct btrfs_backref_edge *edge;
2642 struct btrfs_backref_node *upper;
2643 struct rb_node *rb_node;
2645 ASSERT(ref_key->type == BTRFS_SHARED_BLOCK_REF_KEY);
2647 /* Only reloc root uses backref pointing to itself */
2648 if (ref_key->objectid == ref_key->offset) {
2649 struct btrfs_root *root;
2651 cur->is_reloc_root = 1;
2652 /* Only reloc backref cache cares about a specific root */
2653 if (cache->is_reloc) {
2654 root = find_reloc_root(cache->fs_info, cur->bytenr);
2660 * For generic purpose backref cache, reloc root node
2663 list_add(&cur->list, &cache->useless_node);
2668 edge = btrfs_backref_alloc_edge(cache);
2672 rb_node = rb_simple_search(&cache->rb_root, ref_key->offset);
2674 /* Parent node not yet cached */
2675 upper = btrfs_backref_alloc_node(cache, ref_key->offset,
2678 btrfs_backref_free_edge(cache, edge);
2683 * Backrefs for the upper level block isn't cached, add the
2684 * block to pending list
2686 list_add_tail(&edge->list[UPPER], &cache->pending_edge);
2688 /* Parent node already cached */
2689 upper = rb_entry(rb_node, struct btrfs_backref_node, rb_node);
2690 ASSERT(upper->checked);
2691 INIT_LIST_HEAD(&edge->list[UPPER]);
2693 btrfs_backref_link_edge(edge, cur, upper, LINK_LOWER);
2698 * Handle indirect tree backref
2700 * Indirect tree backref means, we only know which tree the node belongs to.
2701 * We still need to do a tree search to find out the parents. This is for
2702 * TREE_BLOCK_REF backref (keyed or inlined).
2704 * @ref_key: The same as @ref_key in handle_direct_tree_backref()
2705 * @tree_key: The first key of this tree block.
2706 * @path: A clean (released) path, to avoid allocating path every time
2707 * the function get called.
2709 static int handle_indirect_tree_backref(struct btrfs_backref_cache *cache,
2710 struct btrfs_path *path,
2711 struct btrfs_key *ref_key,
2712 struct btrfs_key *tree_key,
2713 struct btrfs_backref_node *cur)
2715 struct btrfs_fs_info *fs_info = cache->fs_info;
2716 struct btrfs_backref_node *upper;
2717 struct btrfs_backref_node *lower;
2718 struct btrfs_backref_edge *edge;
2719 struct extent_buffer *eb;
2720 struct btrfs_root *root;
2721 struct rb_node *rb_node;
2723 bool need_check = true;
2726 root = btrfs_get_fs_root(fs_info, ref_key->offset, false);
2728 return PTR_ERR(root);
2729 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
2732 if (btrfs_root_level(&root->root_item) == cur->level) {
2734 ASSERT(btrfs_root_bytenr(&root->root_item) == cur->bytenr);
2736 * For reloc backref cache, we may ignore reloc root. But for
2737 * general purpose backref cache, we can't rely on
2738 * btrfs_should_ignore_reloc_root() as it may conflict with
2739 * current running relocation and lead to missing root.
2741 * For general purpose backref cache, reloc root detection is
2742 * completely relying on direct backref (key->offset is parent
2743 * bytenr), thus only do such check for reloc cache.
2745 if (btrfs_should_ignore_reloc_root(root) && cache->is_reloc) {
2746 btrfs_put_root(root);
2747 list_add(&cur->list, &cache->useless_node);
2754 level = cur->level + 1;
2756 /* Search the tree to find parent blocks referring to the block */
2757 path->search_commit_root = 1;
2758 path->skip_locking = 1;
2759 path->lowest_level = level;
2760 ret = btrfs_search_slot(NULL, root, tree_key, path, 0, 0);
2761 path->lowest_level = 0;
2763 btrfs_put_root(root);
2766 if (ret > 0 && path->slots[level] > 0)
2767 path->slots[level]--;
2769 eb = path->nodes[level];
2770 if (btrfs_node_blockptr(eb, path->slots[level]) != cur->bytenr) {
2772 "couldn't find block (%llu) (level %d) in tree (%llu) with key (%llu %u %llu)",
2773 cur->bytenr, level - 1, root->root_key.objectid,
2774 tree_key->objectid, tree_key->type, tree_key->offset);
2775 btrfs_put_root(root);
2781 /* Add all nodes and edges in the path */
2782 for (; level < BTRFS_MAX_LEVEL; level++) {
2783 if (!path->nodes[level]) {
2784 ASSERT(btrfs_root_bytenr(&root->root_item) ==
2786 /* Same as previous should_ignore_reloc_root() call */
2787 if (btrfs_should_ignore_reloc_root(root) &&
2789 btrfs_put_root(root);
2790 list_add(&lower->list, &cache->useless_node);
2797 edge = btrfs_backref_alloc_edge(cache);
2799 btrfs_put_root(root);
2804 eb = path->nodes[level];
2805 rb_node = rb_simple_search(&cache->rb_root, eb->start);
2807 upper = btrfs_backref_alloc_node(cache, eb->start,
2810 btrfs_put_root(root);
2811 btrfs_backref_free_edge(cache, edge);
2815 upper->owner = btrfs_header_owner(eb);
2816 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
2820 * If we know the block isn't shared we can avoid
2821 * checking its backrefs.
2823 if (btrfs_block_can_be_shared(root, eb))
2829 * Add the block to pending list if we need to check its
2830 * backrefs, we only do this once while walking up a
2831 * tree as we will catch anything else later on.
2833 if (!upper->checked && need_check) {
2835 list_add_tail(&edge->list[UPPER],
2836 &cache->pending_edge);
2840 INIT_LIST_HEAD(&edge->list[UPPER]);
2843 upper = rb_entry(rb_node, struct btrfs_backref_node,
2845 ASSERT(upper->checked);
2846 INIT_LIST_HEAD(&edge->list[UPPER]);
2848 upper->owner = btrfs_header_owner(eb);
2850 btrfs_backref_link_edge(edge, lower, upper, LINK_LOWER);
2853 btrfs_put_root(root);
2860 btrfs_release_path(path);
2865 * Add backref node @cur into @cache.
2867 * NOTE: Even if the function returned 0, @cur is not yet cached as its upper
2868 * links aren't yet bi-directional. Needs to finish such links.
2869 * Use btrfs_backref_finish_upper_links() to finish such linkage.
2871 * @path: Released path for indirect tree backref lookup
2872 * @iter: Released backref iter for extent tree search
2873 * @node_key: The first key of the tree block
2875 int btrfs_backref_add_tree_node(struct btrfs_backref_cache *cache,
2876 struct btrfs_path *path,
2877 struct btrfs_backref_iter *iter,
2878 struct btrfs_key *node_key,
2879 struct btrfs_backref_node *cur)
2881 struct btrfs_fs_info *fs_info = cache->fs_info;
2882 struct btrfs_backref_edge *edge;
2883 struct btrfs_backref_node *exist;
2886 ret = btrfs_backref_iter_start(iter, cur->bytenr);
2890 * We skip the first btrfs_tree_block_info, as we don't use the key
2891 * stored in it, but fetch it from the tree block
2893 if (btrfs_backref_has_tree_block_info(iter)) {
2894 ret = btrfs_backref_iter_next(iter);
2897 /* No extra backref? This means the tree block is corrupted */
2903 WARN_ON(cur->checked);
2904 if (!list_empty(&cur->upper)) {
2906 * The backref was added previously when processing backref of
2907 * type BTRFS_TREE_BLOCK_REF_KEY
2909 ASSERT(list_is_singular(&cur->upper));
2910 edge = list_entry(cur->upper.next, struct btrfs_backref_edge,
2912 ASSERT(list_empty(&edge->list[UPPER]));
2913 exist = edge->node[UPPER];
2915 * Add the upper level block to pending list if we need check
2918 if (!exist->checked)
2919 list_add_tail(&edge->list[UPPER], &cache->pending_edge);
2924 for (; ret == 0; ret = btrfs_backref_iter_next(iter)) {
2925 struct extent_buffer *eb;
2926 struct btrfs_key key;
2930 eb = btrfs_backref_get_eb(iter);
2932 key.objectid = iter->bytenr;
2933 if (btrfs_backref_iter_is_inline_ref(iter)) {
2934 struct btrfs_extent_inline_ref *iref;
2936 /* Update key for inline backref */
2937 iref = (struct btrfs_extent_inline_ref *)
2938 ((unsigned long)iter->cur_ptr);
2939 type = btrfs_get_extent_inline_ref_type(eb, iref,
2940 BTRFS_REF_TYPE_BLOCK);
2941 if (type == BTRFS_REF_TYPE_INVALID) {
2946 key.offset = btrfs_extent_inline_ref_offset(eb, iref);
2948 key.type = iter->cur_key.type;
2949 key.offset = iter->cur_key.offset;
2953 * Parent node found and matches current inline ref, no need to
2954 * rebuild this node for this inline ref
2957 ((key.type == BTRFS_TREE_BLOCK_REF_KEY &&
2958 exist->owner == key.offset) ||
2959 (key.type == BTRFS_SHARED_BLOCK_REF_KEY &&
2960 exist->bytenr == key.offset))) {
2965 /* SHARED_BLOCK_REF means key.offset is the parent bytenr */
2966 if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
2967 ret = handle_direct_tree_backref(cache, &key, cur);
2971 } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) {
2973 btrfs_print_v0_err(fs_info);
2974 btrfs_handle_fs_error(fs_info, ret, NULL);
2976 } else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) {
2981 * key.type == BTRFS_TREE_BLOCK_REF_KEY, inline ref offset
2982 * means the root objectid. We need to search the tree to get
2983 * its parent bytenr.
2985 ret = handle_indirect_tree_backref(cache, path, &key, node_key,
2994 btrfs_backref_iter_release(iter);
2999 * Finish the upwards linkage created by btrfs_backref_add_tree_node()
3001 int btrfs_backref_finish_upper_links(struct btrfs_backref_cache *cache,
3002 struct btrfs_backref_node *start)
3004 struct list_head *useless_node = &cache->useless_node;
3005 struct btrfs_backref_edge *edge;
3006 struct rb_node *rb_node;
3007 LIST_HEAD(pending_edge);
3009 ASSERT(start->checked);
3011 /* Insert this node to cache if it's not COW-only */
3012 if (!start->cowonly) {
3013 rb_node = rb_simple_insert(&cache->rb_root, start->bytenr,
3016 btrfs_backref_panic(cache->fs_info, start->bytenr,
3018 list_add_tail(&start->lower, &cache->leaves);
3022 * Use breadth first search to iterate all related edges.
3024 * The starting points are all the edges of this node
3026 list_for_each_entry(edge, &start->upper, list[LOWER])
3027 list_add_tail(&edge->list[UPPER], &pending_edge);
3029 while (!list_empty(&pending_edge)) {
3030 struct btrfs_backref_node *upper;
3031 struct btrfs_backref_node *lower;
3033 edge = list_first_entry(&pending_edge,
3034 struct btrfs_backref_edge, list[UPPER]);
3035 list_del_init(&edge->list[UPPER]);
3036 upper = edge->node[UPPER];
3037 lower = edge->node[LOWER];
3039 /* Parent is detached, no need to keep any edges */
3040 if (upper->detached) {
3041 list_del(&edge->list[LOWER]);
3042 btrfs_backref_free_edge(cache, edge);
3044 /* Lower node is orphan, queue for cleanup */
3045 if (list_empty(&lower->upper))
3046 list_add(&lower->list, useless_node);
3051 * All new nodes added in current build_backref_tree() haven't
3052 * been linked to the cache rb tree.
3053 * So if we have upper->rb_node populated, this means a cache
3054 * hit. We only need to link the edge, as @upper and all its
3055 * parents have already been linked.
3057 if (!RB_EMPTY_NODE(&upper->rb_node)) {
3058 if (upper->lowest) {
3059 list_del_init(&upper->lower);
3063 list_add_tail(&edge->list[UPPER], &upper->lower);
3067 /* Sanity check, we shouldn't have any unchecked nodes */
3068 if (!upper->checked) {
3073 /* Sanity check, COW-only node has non-COW-only parent */
3074 if (start->cowonly != upper->cowonly) {
3079 /* Only cache non-COW-only (subvolume trees) tree blocks */
3080 if (!upper->cowonly) {
3081 rb_node = rb_simple_insert(&cache->rb_root, upper->bytenr,
3084 btrfs_backref_panic(cache->fs_info,
3085 upper->bytenr, -EEXIST);
3090 list_add_tail(&edge->list[UPPER], &upper->lower);
3093 * Also queue all the parent edges of this uncached node
3094 * to finish the upper linkage
3096 list_for_each_entry(edge, &upper->upper, list[LOWER])
3097 list_add_tail(&edge->list[UPPER], &pending_edge);
3102 void btrfs_backref_error_cleanup(struct btrfs_backref_cache *cache,
3103 struct btrfs_backref_node *node)
3105 struct btrfs_backref_node *lower;
3106 struct btrfs_backref_node *upper;
3107 struct btrfs_backref_edge *edge;
3109 while (!list_empty(&cache->useless_node)) {
3110 lower = list_first_entry(&cache->useless_node,
3111 struct btrfs_backref_node, list);
3112 list_del_init(&lower->list);
3114 while (!list_empty(&cache->pending_edge)) {
3115 edge = list_first_entry(&cache->pending_edge,
3116 struct btrfs_backref_edge, list[UPPER]);
3117 list_del(&edge->list[UPPER]);
3118 list_del(&edge->list[LOWER]);
3119 lower = edge->node[LOWER];
3120 upper = edge->node[UPPER];
3121 btrfs_backref_free_edge(cache, edge);
3124 * Lower is no longer linked to any upper backref nodes and
3125 * isn't in the cache, we can free it ourselves.
3127 if (list_empty(&lower->upper) &&
3128 RB_EMPTY_NODE(&lower->rb_node))
3129 list_add(&lower->list, &cache->useless_node);
3131 if (!RB_EMPTY_NODE(&upper->rb_node))
3134 /* Add this guy's upper edges to the list to process */
3135 list_for_each_entry(edge, &upper->upper, list[LOWER])
3136 list_add_tail(&edge->list[UPPER],
3137 &cache->pending_edge);
3138 if (list_empty(&upper->upper))
3139 list_add(&upper->list, &cache->useless_node);
3142 while (!list_empty(&cache->useless_node)) {
3143 lower = list_first_entry(&cache->useless_node,
3144 struct btrfs_backref_node, list);
3145 list_del_init(&lower->list);
3148 btrfs_backref_drop_node(cache, lower);
3151 btrfs_backref_cleanup_node(cache, node);
3152 ASSERT(list_empty(&cache->useless_node) &&
3153 list_empty(&cache->pending_edge));