2 * Copyright (C) 2011 STRATO. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/vmalloc.h>
20 #include <linux/rbtree.h>
25 #include "transaction.h"
26 #include "delayed-ref.h"
29 /* Just an arbitrary number so we can be sure this happened */
30 #define BACKREF_FOUND_SHARED 6
32 struct extent_inode_elem
{
35 struct extent_inode_elem
*next
;
39 * ref_root is used as the root of the ref tree that hold a collection
40 * of unique references.
43 struct rb_root rb_root
;
46 * The unique_refs represents the number of ref_nodes with a positive
47 * count stored in the tree. Even if a ref_node (the count is greater
48 * than one) is added, the unique_refs will only increase by one.
50 unsigned int unique_refs
;
53 /* ref_node is used to store a unique reference to the ref tree. */
55 struct rb_node rb_node
;
57 /* For NORMAL_REF, otherwise all these fields should be set to 0 */
62 /* For SHARED_REF, otherwise parent field should be set to 0 */
65 /* Ref to the ref_mod of btrfs_delayed_ref_node */
69 /* Dynamically allocate and initialize a ref_root */
70 static struct ref_root
*ref_root_alloc(void)
72 struct ref_root
*ref_tree
;
74 ref_tree
= kmalloc(sizeof(*ref_tree
), GFP_NOFS
);
78 ref_tree
->rb_root
= RB_ROOT
;
79 ref_tree
->unique_refs
= 0;
84 /* Free all nodes in the ref tree, and reinit ref_root */
85 static void ref_root_fini(struct ref_root
*ref_tree
)
87 struct ref_node
*node
;
90 while ((next
= rb_first(&ref_tree
->rb_root
)) != NULL
) {
91 node
= rb_entry(next
, struct ref_node
, rb_node
);
92 rb_erase(next
, &ref_tree
->rb_root
);
96 ref_tree
->rb_root
= RB_ROOT
;
97 ref_tree
->unique_refs
= 0;
100 static void ref_root_free(struct ref_root
*ref_tree
)
105 ref_root_fini(ref_tree
);
110 * Compare ref_node with (root_id, object_id, offset, parent)
112 * The function compares two ref_node a and b. It returns an integer less
113 * than, equal to, or greater than zero , respectively, to be less than, to
114 * equal, or be greater than b.
116 static int ref_node_cmp(struct ref_node
*a
, struct ref_node
*b
)
118 if (a
->root_id
< b
->root_id
)
120 else if (a
->root_id
> b
->root_id
)
123 if (a
->object_id
< b
->object_id
)
125 else if (a
->object_id
> b
->object_id
)
128 if (a
->offset
< b
->offset
)
130 else if (a
->offset
> b
->offset
)
133 if (a
->parent
< b
->parent
)
135 else if (a
->parent
> b
->parent
)
142 * Search ref_node with (root_id, object_id, offset, parent) in the tree
144 * if found, the pointer of the ref_node will be returned;
145 * if not found, NULL will be returned and pos will point to the rb_node for
146 * insert, pos_parent will point to pos'parent for insert;
148 static struct ref_node
*__ref_tree_search(struct ref_root
*ref_tree
,
149 struct rb_node
***pos
,
150 struct rb_node
**pos_parent
,
151 u64 root_id
, u64 object_id
,
152 u64 offset
, u64 parent
)
154 struct ref_node
*cur
= NULL
;
155 struct ref_node entry
;
158 entry
.root_id
= root_id
;
159 entry
.object_id
= object_id
;
160 entry
.offset
= offset
;
161 entry
.parent
= parent
;
163 *pos
= &ref_tree
->rb_root
.rb_node
;
167 cur
= rb_entry(*pos_parent
, struct ref_node
, rb_node
);
169 ret
= ref_node_cmp(cur
, &entry
);
171 *pos
= &(**pos
)->rb_left
;
173 *pos
= &(**pos
)->rb_right
;
182 * Insert a ref_node to the ref tree
183 * @pos used for specifiy the position to insert
184 * @pos_parent for specifiy pos's parent
187 * ref_node already exists, return -EEXIST;
189 static int ref_tree_insert(struct ref_root
*ref_tree
, struct rb_node
**pos
,
190 struct rb_node
*pos_parent
, struct ref_node
*ins
)
192 struct rb_node
**p
= NULL
;
193 struct rb_node
*parent
= NULL
;
194 struct ref_node
*cur
= NULL
;
197 cur
= __ref_tree_search(ref_tree
, &p
, &parent
, ins
->root_id
,
198 ins
->object_id
, ins
->offset
,
207 rb_link_node(&ins
->rb_node
, parent
, p
);
208 rb_insert_color(&ins
->rb_node
, &ref_tree
->rb_root
);
213 /* Erase and free ref_node, caller should update ref_root->unique_refs */
214 static void ref_tree_remove(struct ref_root
*ref_tree
, struct ref_node
*node
)
216 rb_erase(&node
->rb_node
, &ref_tree
->rb_root
);
221 * Update ref_root->unique_refs
223 * Call __ref_tree_search
224 * 1. if ref_node doesn't exist, ref_tree_insert this node, and update
225 * ref_root->unique_refs:
226 * if ref_node->ref_mod > 0, ref_root->unique_refs++;
227 * if ref_node->ref_mod < 0, do noting;
229 * 2. if ref_node is found, then get origin ref_node->ref_mod, and update
231 * if ref_node->ref_mod is equal to 0,then call ref_tree_remove
233 * according to origin_mod and new_mod, update ref_root->items
234 * +----------------+--------------+-------------+
235 * | |new_count <= 0|new_count > 0|
236 * +----------------+--------------+-------------+
237 * |origin_count < 0| 0 | 1 |
238 * +----------------+--------------+-------------+
239 * |origin_count > 0| -1 | 0 |
240 * +----------------+--------------+-------------+
242 * In case of allocation failure, -ENOMEM is returned and the ref_tree stays
246 static int ref_tree_add(struct ref_root
*ref_tree
, u64 root_id
, u64 object_id
,
247 u64 offset
, u64 parent
, int count
)
249 struct ref_node
*node
= NULL
;
250 struct rb_node
**pos
= NULL
;
251 struct rb_node
*pos_parent
= NULL
;
258 node
= __ref_tree_search(ref_tree
, &pos
, &pos_parent
, root_id
,
259 object_id
, offset
, parent
);
261 node
= kmalloc(sizeof(*node
), GFP_NOFS
);
265 node
->root_id
= root_id
;
266 node
->object_id
= object_id
;
267 node
->offset
= offset
;
268 node
->parent
= parent
;
269 node
->ref_mod
= count
;
271 ret
= ref_tree_insert(ref_tree
, pos
, pos_parent
, node
);
278 ref_tree
->unique_refs
+= node
->ref_mod
> 0 ? 1 : 0;
283 origin_count
= node
->ref_mod
;
284 node
->ref_mod
+= count
;
286 if (node
->ref_mod
> 0)
287 ref_tree
->unique_refs
+= origin_count
> 0 ? 0 : 1;
288 else if (node
->ref_mod
<= 0)
289 ref_tree
->unique_refs
+= origin_count
> 0 ? -1 : 0;
292 ref_tree_remove(ref_tree
, node
);
297 static int check_extent_in_eb(struct btrfs_key
*key
, struct extent_buffer
*eb
,
298 struct btrfs_file_extent_item
*fi
,
300 struct extent_inode_elem
**eie
)
303 struct extent_inode_elem
*e
;
305 if (!btrfs_file_extent_compression(eb
, fi
) &&
306 !btrfs_file_extent_encryption(eb
, fi
) &&
307 !btrfs_file_extent_other_encoding(eb
, fi
)) {
311 data_offset
= btrfs_file_extent_offset(eb
, fi
);
312 data_len
= btrfs_file_extent_num_bytes(eb
, fi
);
314 if (extent_item_pos
< data_offset
||
315 extent_item_pos
>= data_offset
+ data_len
)
317 offset
= extent_item_pos
- data_offset
;
320 e
= kmalloc(sizeof(*e
), GFP_NOFS
);
325 e
->inum
= key
->objectid
;
326 e
->offset
= key
->offset
+ offset
;
332 static void free_inode_elem_list(struct extent_inode_elem
*eie
)
334 struct extent_inode_elem
*eie_next
;
336 for (; eie
; eie
= eie_next
) {
337 eie_next
= eie
->next
;
342 static int find_extent_in_eb(struct extent_buffer
*eb
, u64 wanted_disk_byte
,
344 struct extent_inode_elem
**eie
)
347 struct btrfs_key key
;
348 struct btrfs_file_extent_item
*fi
;
355 * from the shared data ref, we only have the leaf but we need
356 * the key. thus, we must look into all items and see that we
357 * find one (some) with a reference to our extent item.
359 nritems
= btrfs_header_nritems(eb
);
360 for (slot
= 0; slot
< nritems
; ++slot
) {
361 btrfs_item_key_to_cpu(eb
, &key
, slot
);
362 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
364 fi
= btrfs_item_ptr(eb
, slot
, struct btrfs_file_extent_item
);
365 extent_type
= btrfs_file_extent_type(eb
, fi
);
366 if (extent_type
== BTRFS_FILE_EXTENT_INLINE
)
368 /* don't skip BTRFS_FILE_EXTENT_PREALLOC, we can handle that */
369 disk_byte
= btrfs_file_extent_disk_bytenr(eb
, fi
);
370 if (disk_byte
!= wanted_disk_byte
)
373 ret
= check_extent_in_eb(&key
, eb
, fi
, extent_item_pos
, eie
);
382 * this structure records all encountered refs on the way up to the root
384 struct __prelim_ref
{
385 struct list_head list
;
387 struct btrfs_key key_for_search
;
390 struct extent_inode_elem
*inode_list
;
392 u64 wanted_disk_byte
;
395 static struct kmem_cache
*btrfs_prelim_ref_cache
;
397 int __init
btrfs_prelim_ref_init(void)
399 btrfs_prelim_ref_cache
= kmem_cache_create("btrfs_prelim_ref",
400 sizeof(struct __prelim_ref
),
404 if (!btrfs_prelim_ref_cache
)
409 void btrfs_prelim_ref_exit(void)
411 kmem_cache_destroy(btrfs_prelim_ref_cache
);
415 * the rules for all callers of this function are:
416 * - obtaining the parent is the goal
417 * - if you add a key, you must know that it is a correct key
418 * - if you cannot add the parent or a correct key, then we will look into the
419 * block later to set a correct key
423 * backref type | shared | indirect | shared | indirect
424 * information | tree | tree | data | data
425 * --------------------+--------+----------+--------+----------
426 * parent logical | y | - | - | -
427 * key to resolve | - | y | y | y
428 * tree block logical | - | - | - | -
429 * root for resolving | y | y | y | y
431 * - column 1: we've the parent -> done
432 * - column 2, 3, 4: we use the key to find the parent
434 * on disk refs (inline or keyed)
435 * ==============================
436 * backref type | shared | indirect | shared | indirect
437 * information | tree | tree | data | data
438 * --------------------+--------+----------+--------+----------
439 * parent logical | y | - | y | -
440 * key to resolve | - | - | - | y
441 * tree block logical | y | y | y | y
442 * root for resolving | - | y | y | y
444 * - column 1, 3: we've the parent -> done
445 * - column 2: we take the first key from the block to find the parent
446 * (see __add_missing_keys)
447 * - column 4: we use the key to find the parent
449 * additional information that's available but not required to find the parent
450 * block might help in merging entries to gain some speed.
453 static int __add_prelim_ref(struct list_head
*head
, u64 root_id
,
454 struct btrfs_key
*key
, int level
,
455 u64 parent
, u64 wanted_disk_byte
, int count
,
458 struct __prelim_ref
*ref
;
460 if (root_id
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
463 ref
= kmem_cache_alloc(btrfs_prelim_ref_cache
, gfp_mask
);
467 ref
->root_id
= root_id
;
469 ref
->key_for_search
= *key
;
471 * We can often find data backrefs with an offset that is too
472 * large (>= LLONG_MAX, maximum allowed file offset) due to
473 * underflows when subtracting a file's offset with the data
474 * offset of its corresponding extent data item. This can
475 * happen for example in the clone ioctl.
476 * So if we detect such case we set the search key's offset to
477 * zero to make sure we will find the matching file extent item
478 * at add_all_parents(), otherwise we will miss it because the
479 * offset taken form the backref is much larger then the offset
480 * of the file extent item. This can make us scan a very large
481 * number of file extent items, but at least it will not make
483 * This is an ugly workaround for a behaviour that should have
484 * never existed, but it does and a fix for the clone ioctl
485 * would touch a lot of places, cause backwards incompatibility
486 * and would not fix the problem for extents cloned with older
489 if (ref
->key_for_search
.type
== BTRFS_EXTENT_DATA_KEY
&&
490 ref
->key_for_search
.offset
>= LLONG_MAX
)
491 ref
->key_for_search
.offset
= 0;
493 memset(&ref
->key_for_search
, 0, sizeof(ref
->key_for_search
));
496 ref
->inode_list
= NULL
;
499 ref
->parent
= parent
;
500 ref
->wanted_disk_byte
= wanted_disk_byte
;
501 list_add_tail(&ref
->list
, head
);
506 static int add_all_parents(struct btrfs_root
*root
, struct btrfs_path
*path
,
507 struct ulist
*parents
, struct __prelim_ref
*ref
,
508 int level
, u64 time_seq
, const u64
*extent_item_pos
,
513 struct extent_buffer
*eb
;
514 struct btrfs_key key
;
515 struct btrfs_key
*key_for_search
= &ref
->key_for_search
;
516 struct btrfs_file_extent_item
*fi
;
517 struct extent_inode_elem
*eie
= NULL
, *old
= NULL
;
519 u64 wanted_disk_byte
= ref
->wanted_disk_byte
;
523 eb
= path
->nodes
[level
];
524 ret
= ulist_add(parents
, eb
->start
, 0, GFP_NOFS
);
531 * We normally enter this function with the path already pointing to
532 * the first item to check. But sometimes, we may enter it with
533 * slot==nritems. In that case, go to the next leaf before we continue.
535 if (path
->slots
[0] >= btrfs_header_nritems(path
->nodes
[0])) {
536 if (time_seq
== (u64
)-1)
537 ret
= btrfs_next_leaf(root
, path
);
539 ret
= btrfs_next_old_leaf(root
, path
, time_seq
);
542 while (!ret
&& count
< total_refs
) {
544 slot
= path
->slots
[0];
546 btrfs_item_key_to_cpu(eb
, &key
, slot
);
548 if (key
.objectid
!= key_for_search
->objectid
||
549 key
.type
!= BTRFS_EXTENT_DATA_KEY
)
552 fi
= btrfs_item_ptr(eb
, slot
, struct btrfs_file_extent_item
);
553 disk_byte
= btrfs_file_extent_disk_bytenr(eb
, fi
);
555 if (disk_byte
== wanted_disk_byte
) {
559 if (extent_item_pos
) {
560 ret
= check_extent_in_eb(&key
, eb
, fi
,
568 ret
= ulist_add_merge_ptr(parents
, eb
->start
,
569 eie
, (void **)&old
, GFP_NOFS
);
572 if (!ret
&& extent_item_pos
) {
580 if (time_seq
== (u64
)-1)
581 ret
= btrfs_next_item(root
, path
);
583 ret
= btrfs_next_old_item(root
, path
, time_seq
);
589 free_inode_elem_list(eie
);
594 * resolve an indirect backref in the form (root_id, key, level)
595 * to a logical address
597 static int __resolve_indirect_ref(struct btrfs_fs_info
*fs_info
,
598 struct btrfs_path
*path
, u64 time_seq
,
599 struct __prelim_ref
*ref
,
600 struct ulist
*parents
,
601 const u64
*extent_item_pos
, u64 total_refs
)
603 struct btrfs_root
*root
;
604 struct btrfs_key root_key
;
605 struct extent_buffer
*eb
;
608 int level
= ref
->level
;
611 root_key
.objectid
= ref
->root_id
;
612 root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
613 root_key
.offset
= (u64
)-1;
615 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
617 root
= btrfs_get_fs_root(fs_info
, &root_key
, false);
619 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
624 if (btrfs_is_testing(fs_info
)) {
625 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
630 if (path
->search_commit_root
)
631 root_level
= btrfs_header_level(root
->commit_root
);
632 else if (time_seq
== (u64
)-1)
633 root_level
= btrfs_header_level(root
->node
);
635 root_level
= btrfs_old_root_level(root
, time_seq
);
637 if (root_level
+ 1 == level
) {
638 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
642 path
->lowest_level
= level
;
643 if (time_seq
== (u64
)-1)
644 ret
= btrfs_search_slot(NULL
, root
, &ref
->key_for_search
, path
,
647 ret
= btrfs_search_old_slot(root
, &ref
->key_for_search
, path
,
650 /* root node has been locked, we can release @subvol_srcu safely here */
651 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
654 "search slot in root %llu (level %d, ref count %d) returned %d for key (%llu %u %llu)",
655 ref
->root_id
, level
, ref
->count
, ret
,
656 ref
->key_for_search
.objectid
, ref
->key_for_search
.type
,
657 ref
->key_for_search
.offset
);
661 eb
= path
->nodes
[level
];
663 if (WARN_ON(!level
)) {
668 eb
= path
->nodes
[level
];
671 ret
= add_all_parents(root
, path
, parents
, ref
, level
, time_seq
,
672 extent_item_pos
, total_refs
);
674 path
->lowest_level
= 0;
675 btrfs_release_path(path
);
680 * resolve all indirect backrefs from the list
682 static int __resolve_indirect_refs(struct btrfs_fs_info
*fs_info
,
683 struct btrfs_path
*path
, u64 time_seq
,
684 struct list_head
*head
,
685 const u64
*extent_item_pos
, u64 total_refs
,
690 struct __prelim_ref
*ref
;
691 struct __prelim_ref
*ref_safe
;
692 struct __prelim_ref
*new_ref
;
693 struct ulist
*parents
;
694 struct ulist_node
*node
;
695 struct ulist_iterator uiter
;
697 parents
= ulist_alloc(GFP_NOFS
);
702 * _safe allows us to insert directly after the current item without
703 * iterating over the newly inserted items.
704 * we're also allowed to re-assign ref during iteration.
706 list_for_each_entry_safe(ref
, ref_safe
, head
, list
) {
707 if (ref
->parent
) /* already direct */
711 if (root_objectid
&& ref
->root_id
!= root_objectid
) {
712 ret
= BACKREF_FOUND_SHARED
;
715 err
= __resolve_indirect_ref(fs_info
, path
, time_seq
, ref
,
716 parents
, extent_item_pos
,
719 * we can only tolerate ENOENT,otherwise,we should catch error
720 * and return directly.
722 if (err
== -ENOENT
) {
729 /* we put the first parent into the ref at hand */
730 ULIST_ITER_INIT(&uiter
);
731 node
= ulist_next(parents
, &uiter
);
732 ref
->parent
= node
? node
->val
: 0;
733 ref
->inode_list
= node
?
734 (struct extent_inode_elem
*)(uintptr_t)node
->aux
: NULL
;
736 /* additional parents require new refs being added here */
737 while ((node
= ulist_next(parents
, &uiter
))) {
738 new_ref
= kmem_cache_alloc(btrfs_prelim_ref_cache
,
744 memcpy(new_ref
, ref
, sizeof(*ref
));
745 new_ref
->parent
= node
->val
;
746 new_ref
->inode_list
= (struct extent_inode_elem
*)
747 (uintptr_t)node
->aux
;
748 list_add(&new_ref
->list
, &ref
->list
);
750 ulist_reinit(parents
);
757 static inline int ref_for_same_block(struct __prelim_ref
*ref1
,
758 struct __prelim_ref
*ref2
)
760 if (ref1
->level
!= ref2
->level
)
762 if (ref1
->root_id
!= ref2
->root_id
)
764 if (ref1
->key_for_search
.type
!= ref2
->key_for_search
.type
)
766 if (ref1
->key_for_search
.objectid
!= ref2
->key_for_search
.objectid
)
768 if (ref1
->key_for_search
.offset
!= ref2
->key_for_search
.offset
)
770 if (ref1
->parent
!= ref2
->parent
)
777 * read tree blocks and add keys where required.
779 static int __add_missing_keys(struct btrfs_fs_info
*fs_info
,
780 struct list_head
*head
)
782 struct __prelim_ref
*ref
;
783 struct extent_buffer
*eb
;
785 list_for_each_entry(ref
, head
, list
) {
788 if (ref
->key_for_search
.type
)
790 BUG_ON(!ref
->wanted_disk_byte
);
791 eb
= read_tree_block(fs_info
->tree_root
, ref
->wanted_disk_byte
,
795 } else if (!extent_buffer_uptodate(eb
)) {
796 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);
804 btrfs_tree_read_unlock(eb
);
805 free_extent_buffer(eb
);
811 * merge backrefs and adjust counts accordingly
813 * mode = 1: merge identical keys, if key is set
814 * FIXME: if we add more keys in __add_prelim_ref, we can merge more here.
815 * additionally, we could even add a key range for the blocks we
816 * looked into to merge even more (-> replace unresolved refs by those
818 * mode = 2: merge identical parents
820 static void __merge_refs(struct list_head
*head
, int mode
)
822 struct __prelim_ref
*pos1
;
824 list_for_each_entry(pos1
, head
, list
) {
825 struct __prelim_ref
*pos2
= pos1
, *tmp
;
827 list_for_each_entry_safe_continue(pos2
, tmp
, head
, list
) {
828 struct __prelim_ref
*ref1
= pos1
, *ref2
= pos2
;
829 struct extent_inode_elem
*eie
;
831 if (!ref_for_same_block(ref1
, ref2
))
834 if (!ref1
->parent
&& ref2
->parent
)
837 if (ref1
->parent
!= ref2
->parent
)
841 eie
= ref1
->inode_list
;
842 while (eie
&& eie
->next
)
845 eie
->next
= ref2
->inode_list
;
847 ref1
->inode_list
= ref2
->inode_list
;
848 ref1
->count
+= ref2
->count
;
850 list_del(&ref2
->list
);
851 kmem_cache_free(btrfs_prelim_ref_cache
, ref2
);
859 * add all currently queued delayed refs from this head whose seq nr is
860 * smaller or equal that seq to the list
862 static int __add_delayed_refs(struct btrfs_delayed_ref_head
*head
, u64 seq
,
863 struct list_head
*prefs
, u64
*total_refs
,
866 struct btrfs_delayed_ref_node
*node
;
867 struct btrfs_delayed_extent_op
*extent_op
= head
->extent_op
;
868 struct btrfs_key key
;
869 struct btrfs_key op_key
= {0};
873 if (extent_op
&& extent_op
->update_key
)
874 btrfs_disk_key_to_cpu(&op_key
, &extent_op
->key
);
876 spin_lock(&head
->lock
);
877 list_for_each_entry(node
, &head
->ref_list
, list
) {
881 switch (node
->action
) {
882 case BTRFS_ADD_DELAYED_EXTENT
:
883 case BTRFS_UPDATE_DELAYED_HEAD
:
886 case BTRFS_ADD_DELAYED_REF
:
889 case BTRFS_DROP_DELAYED_REF
:
895 *total_refs
+= (node
->ref_mod
* sgn
);
896 switch (node
->type
) {
897 case BTRFS_TREE_BLOCK_REF_KEY
: {
898 struct btrfs_delayed_tree_ref
*ref
;
900 ref
= btrfs_delayed_node_to_tree_ref(node
);
901 ret
= __add_prelim_ref(prefs
, ref
->root
, &op_key
,
902 ref
->level
+ 1, 0, node
->bytenr
,
903 node
->ref_mod
* sgn
, GFP_ATOMIC
);
906 case BTRFS_SHARED_BLOCK_REF_KEY
: {
907 struct btrfs_delayed_tree_ref
*ref
;
909 ref
= btrfs_delayed_node_to_tree_ref(node
);
910 ret
= __add_prelim_ref(prefs
, 0, NULL
,
911 ref
->level
+ 1, ref
->parent
,
913 node
->ref_mod
* sgn
, GFP_ATOMIC
);
916 case BTRFS_EXTENT_DATA_REF_KEY
: {
917 struct btrfs_delayed_data_ref
*ref
;
918 ref
= btrfs_delayed_node_to_data_ref(node
);
920 key
.objectid
= ref
->objectid
;
921 key
.type
= BTRFS_EXTENT_DATA_KEY
;
922 key
.offset
= ref
->offset
;
925 * Found a inum that doesn't match our known inum, we
928 if (inum
&& ref
->objectid
!= inum
) {
929 ret
= BACKREF_FOUND_SHARED
;
933 ret
= __add_prelim_ref(prefs
, ref
->root
, &key
, 0, 0,
935 node
->ref_mod
* sgn
, GFP_ATOMIC
);
938 case BTRFS_SHARED_DATA_REF_KEY
: {
939 struct btrfs_delayed_data_ref
*ref
;
941 ref
= btrfs_delayed_node_to_data_ref(node
);
942 ret
= __add_prelim_ref(prefs
, 0, NULL
, 0,
943 ref
->parent
, node
->bytenr
,
944 node
->ref_mod
* sgn
, GFP_ATOMIC
);
953 spin_unlock(&head
->lock
);
958 * add all inline backrefs for bytenr to the list
960 static int __add_inline_refs(struct btrfs_fs_info
*fs_info
,
961 struct btrfs_path
*path
, u64 bytenr
,
962 int *info_level
, struct list_head
*prefs
,
963 struct ref_root
*ref_tree
,
964 u64
*total_refs
, u64 inum
)
968 struct extent_buffer
*leaf
;
969 struct btrfs_key key
;
970 struct btrfs_key found_key
;
973 struct btrfs_extent_item
*ei
;
978 * enumerate all inline refs
980 leaf
= path
->nodes
[0];
981 slot
= path
->slots
[0];
983 item_size
= btrfs_item_size_nr(leaf
, slot
);
984 BUG_ON(item_size
< sizeof(*ei
));
986 ei
= btrfs_item_ptr(leaf
, slot
, struct btrfs_extent_item
);
987 flags
= btrfs_extent_flags(leaf
, ei
);
988 *total_refs
+= btrfs_extent_refs(leaf
, ei
);
989 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
991 ptr
= (unsigned long)(ei
+ 1);
992 end
= (unsigned long)ei
+ item_size
;
994 if (found_key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
995 flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
) {
996 struct btrfs_tree_block_info
*info
;
998 info
= (struct btrfs_tree_block_info
*)ptr
;
999 *info_level
= btrfs_tree_block_level(leaf
, info
);
1000 ptr
+= sizeof(struct btrfs_tree_block_info
);
1002 } else if (found_key
.type
== BTRFS_METADATA_ITEM_KEY
) {
1003 *info_level
= found_key
.offset
;
1005 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_DATA
));
1009 struct btrfs_extent_inline_ref
*iref
;
1013 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1014 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1015 offset
= btrfs_extent_inline_ref_offset(leaf
, iref
);
1018 case BTRFS_SHARED_BLOCK_REF_KEY
:
1019 ret
= __add_prelim_ref(prefs
, 0, NULL
,
1020 *info_level
+ 1, offset
,
1021 bytenr
, 1, GFP_NOFS
);
1023 case BTRFS_SHARED_DATA_REF_KEY
: {
1024 struct btrfs_shared_data_ref
*sdref
;
1027 sdref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1028 count
= btrfs_shared_data_ref_count(leaf
, sdref
);
1029 ret
= __add_prelim_ref(prefs
, 0, NULL
, 0, offset
,
1030 bytenr
, count
, GFP_NOFS
);
1033 ret
= ref_tree_add(ref_tree
, 0, 0, 0,
1035 if (!ret
&& ref_tree
->unique_refs
> 1)
1036 ret
= BACKREF_FOUND_SHARED
;
1040 case BTRFS_TREE_BLOCK_REF_KEY
:
1041 ret
= __add_prelim_ref(prefs
, offset
, NULL
,
1043 bytenr
, 1, GFP_NOFS
);
1045 case BTRFS_EXTENT_DATA_REF_KEY
: {
1046 struct btrfs_extent_data_ref
*dref
;
1050 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1051 count
= btrfs_extent_data_ref_count(leaf
, dref
);
1052 key
.objectid
= btrfs_extent_data_ref_objectid(leaf
,
1054 key
.type
= BTRFS_EXTENT_DATA_KEY
;
1055 key
.offset
= btrfs_extent_data_ref_offset(leaf
, dref
);
1057 if (inum
&& key
.objectid
!= inum
) {
1058 ret
= BACKREF_FOUND_SHARED
;
1062 root
= btrfs_extent_data_ref_root(leaf
, dref
);
1063 ret
= __add_prelim_ref(prefs
, root
, &key
, 0, 0,
1064 bytenr
, count
, GFP_NOFS
);
1067 ret
= ref_tree_add(ref_tree
, root
,
1071 if (!ret
&& ref_tree
->unique_refs
> 1)
1072 ret
= BACKREF_FOUND_SHARED
;
1081 ptr
+= btrfs_extent_inline_ref_size(type
);
1088 * add all non-inline backrefs for bytenr to the list
1090 static int __add_keyed_refs(struct btrfs_fs_info
*fs_info
,
1091 struct btrfs_path
*path
, u64 bytenr
,
1092 int info_level
, struct list_head
*prefs
,
1093 struct ref_root
*ref_tree
, u64 inum
)
1095 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
1098 struct extent_buffer
*leaf
;
1099 struct btrfs_key key
;
1102 ret
= btrfs_next_item(extent_root
, path
);
1110 slot
= path
->slots
[0];
1111 leaf
= path
->nodes
[0];
1112 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
1114 if (key
.objectid
!= bytenr
)
1116 if (key
.type
< BTRFS_TREE_BLOCK_REF_KEY
)
1118 if (key
.type
> BTRFS_SHARED_DATA_REF_KEY
)
1122 case BTRFS_SHARED_BLOCK_REF_KEY
:
1123 ret
= __add_prelim_ref(prefs
, 0, NULL
,
1124 info_level
+ 1, key
.offset
,
1125 bytenr
, 1, GFP_NOFS
);
1127 case BTRFS_SHARED_DATA_REF_KEY
: {
1128 struct btrfs_shared_data_ref
*sdref
;
1131 sdref
= btrfs_item_ptr(leaf
, slot
,
1132 struct btrfs_shared_data_ref
);
1133 count
= btrfs_shared_data_ref_count(leaf
, sdref
);
1134 ret
= __add_prelim_ref(prefs
, 0, NULL
, 0, key
.offset
,
1135 bytenr
, count
, GFP_NOFS
);
1138 ret
= ref_tree_add(ref_tree
, 0, 0, 0,
1140 if (!ret
&& ref_tree
->unique_refs
> 1)
1141 ret
= BACKREF_FOUND_SHARED
;
1145 case BTRFS_TREE_BLOCK_REF_KEY
:
1146 ret
= __add_prelim_ref(prefs
, key
.offset
, NULL
,
1148 bytenr
, 1, GFP_NOFS
);
1150 case BTRFS_EXTENT_DATA_REF_KEY
: {
1151 struct btrfs_extent_data_ref
*dref
;
1155 dref
= btrfs_item_ptr(leaf
, slot
,
1156 struct btrfs_extent_data_ref
);
1157 count
= btrfs_extent_data_ref_count(leaf
, dref
);
1158 key
.objectid
= btrfs_extent_data_ref_objectid(leaf
,
1160 key
.type
= BTRFS_EXTENT_DATA_KEY
;
1161 key
.offset
= btrfs_extent_data_ref_offset(leaf
, dref
);
1163 if (inum
&& key
.objectid
!= inum
) {
1164 ret
= BACKREF_FOUND_SHARED
;
1168 root
= btrfs_extent_data_ref_root(leaf
, dref
);
1169 ret
= __add_prelim_ref(prefs
, root
, &key
, 0, 0,
1170 bytenr
, count
, GFP_NOFS
);
1173 ret
= ref_tree_add(ref_tree
, root
,
1177 if (!ret
&& ref_tree
->unique_refs
> 1)
1178 ret
= BACKREF_FOUND_SHARED
;
1194 * this adds all existing backrefs (inline backrefs, backrefs and delayed
1195 * refs) for the given bytenr to the refs list, merges duplicates and resolves
1196 * indirect refs to their parent bytenr.
1197 * When roots are found, they're added to the roots list
1199 * NOTE: This can return values > 0
1201 * If time_seq is set to (u64)-1, it will not search delayed_refs, and behave
1202 * much like trans == NULL case, the difference only lies in it will not
1204 * The special case is for qgroup to search roots in commit_transaction().
1206 * If check_shared is set to 1, any extent has more than one ref item, will
1207 * be returned BACKREF_FOUND_SHARED immediately.
1209 * FIXME some caching might speed things up
1211 static int find_parent_nodes(struct btrfs_trans_handle
*trans
,
1212 struct btrfs_fs_info
*fs_info
, u64 bytenr
,
1213 u64 time_seq
, struct ulist
*refs
,
1214 struct ulist
*roots
, const u64
*extent_item_pos
,
1215 u64 root_objectid
, u64 inum
, int check_shared
)
1217 struct btrfs_key key
;
1218 struct btrfs_path
*path
;
1219 struct btrfs_delayed_ref_root
*delayed_refs
= NULL
;
1220 struct btrfs_delayed_ref_head
*head
;
1223 struct list_head prefs_delayed
;
1224 struct list_head prefs
;
1225 struct __prelim_ref
*ref
;
1226 struct extent_inode_elem
*eie
= NULL
;
1227 struct ref_root
*ref_tree
= NULL
;
1230 INIT_LIST_HEAD(&prefs
);
1231 INIT_LIST_HEAD(&prefs_delayed
);
1233 key
.objectid
= bytenr
;
1234 key
.offset
= (u64
)-1;
1235 if (btrfs_fs_incompat(fs_info
, SKINNY_METADATA
))
1236 key
.type
= BTRFS_METADATA_ITEM_KEY
;
1238 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1240 path
= btrfs_alloc_path();
1244 path
->search_commit_root
= 1;
1245 path
->skip_locking
= 1;
1248 if (time_seq
== (u64
)-1)
1249 path
->skip_locking
= 1;
1252 * grab both a lock on the path and a lock on the delayed ref head.
1253 * We need both to get a consistent picture of how the refs look
1254 * at a specified point in time
1261 ref_tree
= ref_root_alloc();
1267 ref_root_fini(ref_tree
);
1271 ret
= btrfs_search_slot(trans
, fs_info
->extent_root
, &key
, path
, 0, 0);
1276 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1277 if (trans
&& likely(trans
->type
!= __TRANS_DUMMY
) &&
1278 time_seq
!= (u64
)-1) {
1280 if (trans
&& time_seq
!= (u64
)-1) {
1283 * look if there are updates for this ref queued and lock the
1286 delayed_refs
= &trans
->transaction
->delayed_refs
;
1287 spin_lock(&delayed_refs
->lock
);
1288 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
1290 if (!mutex_trylock(&head
->mutex
)) {
1291 atomic_inc(&head
->node
.refs
);
1292 spin_unlock(&delayed_refs
->lock
);
1294 btrfs_release_path(path
);
1297 * Mutex was contended, block until it's
1298 * released and try again
1300 mutex_lock(&head
->mutex
);
1301 mutex_unlock(&head
->mutex
);
1302 btrfs_put_delayed_ref(&head
->node
);
1305 spin_unlock(&delayed_refs
->lock
);
1306 ret
= __add_delayed_refs(head
, time_seq
,
1307 &prefs_delayed
, &total_refs
,
1309 mutex_unlock(&head
->mutex
);
1313 spin_unlock(&delayed_refs
->lock
);
1316 if (check_shared
&& !list_empty(&prefs_delayed
)) {
1318 * Add all delay_ref to the ref_tree and check if there
1319 * are multiple ref items added.
1321 list_for_each_entry(ref
, &prefs_delayed
, list
) {
1322 if (ref
->key_for_search
.type
) {
1323 ret
= ref_tree_add(ref_tree
,
1325 ref
->key_for_search
.objectid
,
1326 ref
->key_for_search
.offset
,
1331 ret
= ref_tree_add(ref_tree
, 0, 0, 0,
1332 ref
->parent
, ref
->count
);
1339 if (ref_tree
->unique_refs
> 1) {
1340 ret
= BACKREF_FOUND_SHARED
;
1347 if (path
->slots
[0]) {
1348 struct extent_buffer
*leaf
;
1352 leaf
= path
->nodes
[0];
1353 slot
= path
->slots
[0];
1354 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
1355 if (key
.objectid
== bytenr
&&
1356 (key
.type
== BTRFS_EXTENT_ITEM_KEY
||
1357 key
.type
== BTRFS_METADATA_ITEM_KEY
)) {
1358 ret
= __add_inline_refs(fs_info
, path
, bytenr
,
1359 &info_level
, &prefs
,
1360 ref_tree
, &total_refs
,
1364 ret
= __add_keyed_refs(fs_info
, path
, bytenr
,
1371 btrfs_release_path(path
);
1373 list_splice_init(&prefs_delayed
, &prefs
);
1375 ret
= __add_missing_keys(fs_info
, &prefs
);
1379 __merge_refs(&prefs
, 1);
1381 ret
= __resolve_indirect_refs(fs_info
, path
, time_seq
, &prefs
,
1382 extent_item_pos
, total_refs
,
1387 __merge_refs(&prefs
, 2);
1389 while (!list_empty(&prefs
)) {
1390 ref
= list_first_entry(&prefs
, struct __prelim_ref
, list
);
1391 WARN_ON(ref
->count
< 0);
1392 if (roots
&& ref
->count
&& ref
->root_id
&& ref
->parent
== 0) {
1393 if (root_objectid
&& ref
->root_id
!= root_objectid
) {
1394 ret
= BACKREF_FOUND_SHARED
;
1398 /* no parent == root of tree */
1399 ret
= ulist_add(roots
, ref
->root_id
, 0, GFP_NOFS
);
1403 if (ref
->count
&& ref
->parent
) {
1404 if (extent_item_pos
&& !ref
->inode_list
&&
1406 struct extent_buffer
*eb
;
1408 eb
= read_tree_block(fs_info
->extent_root
,
1413 } else if (!extent_buffer_uptodate(eb
)) {
1414 free_extent_buffer(eb
);
1418 btrfs_tree_read_lock(eb
);
1419 btrfs_set_lock_blocking_rw(eb
, BTRFS_READ_LOCK
);
1420 ret
= find_extent_in_eb(eb
, bytenr
,
1421 *extent_item_pos
, &eie
);
1422 btrfs_tree_read_unlock_blocking(eb
);
1423 free_extent_buffer(eb
);
1426 ref
->inode_list
= eie
;
1428 ret
= ulist_add_merge_ptr(refs
, ref
->parent
,
1430 (void **)&eie
, GFP_NOFS
);
1433 if (!ret
&& extent_item_pos
) {
1435 * we've recorded that parent, so we must extend
1436 * its inode list here
1441 eie
->next
= ref
->inode_list
;
1445 list_del(&ref
->list
);
1446 kmem_cache_free(btrfs_prelim_ref_cache
, ref
);
1450 btrfs_free_path(path
);
1451 ref_root_free(ref_tree
);
1452 while (!list_empty(&prefs
)) {
1453 ref
= list_first_entry(&prefs
, struct __prelim_ref
, list
);
1454 list_del(&ref
->list
);
1455 kmem_cache_free(btrfs_prelim_ref_cache
, ref
);
1457 while (!list_empty(&prefs_delayed
)) {
1458 ref
= list_first_entry(&prefs_delayed
, struct __prelim_ref
,
1460 list_del(&ref
->list
);
1461 kmem_cache_free(btrfs_prelim_ref_cache
, ref
);
1464 free_inode_elem_list(eie
);
1468 static void free_leaf_list(struct ulist
*blocks
)
1470 struct ulist_node
*node
= NULL
;
1471 struct extent_inode_elem
*eie
;
1472 struct ulist_iterator uiter
;
1474 ULIST_ITER_INIT(&uiter
);
1475 while ((node
= ulist_next(blocks
, &uiter
))) {
1478 eie
= (struct extent_inode_elem
*)(uintptr_t)node
->aux
;
1479 free_inode_elem_list(eie
);
1487 * Finds all leafs with a reference to the specified combination of bytenr and
1488 * offset. key_list_head will point to a list of corresponding keys (caller must
1489 * free each list element). The leafs will be stored in the leafs ulist, which
1490 * must be freed with ulist_free.
1492 * returns 0 on success, <0 on error
1494 static int btrfs_find_all_leafs(struct btrfs_trans_handle
*trans
,
1495 struct btrfs_fs_info
*fs_info
, u64 bytenr
,
1496 u64 time_seq
, struct ulist
**leafs
,
1497 const u64
*extent_item_pos
)
1501 *leafs
= ulist_alloc(GFP_NOFS
);
1505 ret
= find_parent_nodes(trans
, fs_info
, bytenr
, time_seq
,
1506 *leafs
, NULL
, extent_item_pos
, 0, 0, 0);
1507 if (ret
< 0 && ret
!= -ENOENT
) {
1508 free_leaf_list(*leafs
);
1516 * walk all backrefs for a given extent to find all roots that reference this
1517 * extent. Walking a backref means finding all extents that reference this
1518 * extent and in turn walk the backrefs of those, too. Naturally this is a
1519 * recursive process, but here it is implemented in an iterative fashion: We
1520 * find all referencing extents for the extent in question and put them on a
1521 * list. In turn, we find all referencing extents for those, further appending
1522 * to the list. The way we iterate the list allows adding more elements after
1523 * the current while iterating. The process stops when we reach the end of the
1524 * list. Found roots are added to the roots list.
1526 * returns 0 on success, < 0 on error.
1528 static int __btrfs_find_all_roots(struct btrfs_trans_handle
*trans
,
1529 struct btrfs_fs_info
*fs_info
, u64 bytenr
,
1530 u64 time_seq
, struct ulist
**roots
)
1533 struct ulist_node
*node
= NULL
;
1534 struct ulist_iterator uiter
;
1537 tmp
= ulist_alloc(GFP_NOFS
);
1540 *roots
= ulist_alloc(GFP_NOFS
);
1546 ULIST_ITER_INIT(&uiter
);
1548 ret
= find_parent_nodes(trans
, fs_info
, bytenr
, time_seq
,
1549 tmp
, *roots
, NULL
, 0, 0, 0);
1550 if (ret
< 0 && ret
!= -ENOENT
) {
1555 node
= ulist_next(tmp
, &uiter
);
1566 int btrfs_find_all_roots(struct btrfs_trans_handle
*trans
,
1567 struct btrfs_fs_info
*fs_info
, u64 bytenr
,
1568 u64 time_seq
, struct ulist
**roots
)
1573 down_read(&fs_info
->commit_root_sem
);
1574 ret
= __btrfs_find_all_roots(trans
, fs_info
, bytenr
, time_seq
, roots
);
1576 up_read(&fs_info
->commit_root_sem
);
1581 * btrfs_check_shared - tell us whether an extent is shared
1583 * @trans: optional trans handle
1585 * btrfs_check_shared uses the backref walking code but will short
1586 * circuit as soon as it finds a root or inode that doesn't match the
1587 * one passed in. This provides a significant performance benefit for
1588 * callers (such as fiemap) which want to know whether the extent is
1589 * shared but do not need a ref count.
1591 * Return: 0 if extent is not shared, 1 if it is shared, < 0 on error.
1593 int btrfs_check_shared(struct btrfs_trans_handle
*trans
,
1594 struct btrfs_fs_info
*fs_info
, u64 root_objectid
,
1595 u64 inum
, u64 bytenr
)
1597 struct ulist
*tmp
= NULL
;
1598 struct ulist
*roots
= NULL
;
1599 struct ulist_iterator uiter
;
1600 struct ulist_node
*node
;
1601 struct seq_list elem
= SEQ_LIST_INIT(elem
);
1604 tmp
= ulist_alloc(GFP_NOFS
);
1605 roots
= ulist_alloc(GFP_NOFS
);
1606 if (!tmp
|| !roots
) {
1613 btrfs_get_tree_mod_seq(fs_info
, &elem
);
1615 down_read(&fs_info
->commit_root_sem
);
1616 ULIST_ITER_INIT(&uiter
);
1618 ret
= find_parent_nodes(trans
, fs_info
, bytenr
, elem
.seq
, tmp
,
1619 roots
, NULL
, root_objectid
, inum
, 1);
1620 if (ret
== BACKREF_FOUND_SHARED
) {
1621 /* this is the only condition under which we return 1 */
1625 if (ret
< 0 && ret
!= -ENOENT
)
1628 node
= ulist_next(tmp
, &uiter
);
1635 btrfs_put_tree_mod_seq(fs_info
, &elem
);
1637 up_read(&fs_info
->commit_root_sem
);
1643 int btrfs_find_one_extref(struct btrfs_root
*root
, u64 inode_objectid
,
1644 u64 start_off
, struct btrfs_path
*path
,
1645 struct btrfs_inode_extref
**ret_extref
,
1649 struct btrfs_key key
;
1650 struct btrfs_key found_key
;
1651 struct btrfs_inode_extref
*extref
;
1652 struct extent_buffer
*leaf
;
1655 key
.objectid
= inode_objectid
;
1656 key
.type
= BTRFS_INODE_EXTREF_KEY
;
1657 key
.offset
= start_off
;
1659 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1664 leaf
= path
->nodes
[0];
1665 slot
= path
->slots
[0];
1666 if (slot
>= btrfs_header_nritems(leaf
)) {
1668 * If the item at offset is not found,
1669 * btrfs_search_slot will point us to the slot
1670 * where it should be inserted. In our case
1671 * that will be the slot directly before the
1672 * next INODE_REF_KEY_V2 item. In the case
1673 * that we're pointing to the last slot in a
1674 * leaf, we must move one leaf over.
1676 ret
= btrfs_next_leaf(root
, path
);
1685 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
1688 * Check that we're still looking at an extended ref key for
1689 * this particular objectid. If we have different
1690 * objectid or type then there are no more to be found
1691 * in the tree and we can exit.
1694 if (found_key
.objectid
!= inode_objectid
)
1696 if (found_key
.type
!= BTRFS_INODE_EXTREF_KEY
)
1700 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
1701 extref
= (struct btrfs_inode_extref
*)ptr
;
1702 *ret_extref
= extref
;
1704 *found_off
= found_key
.offset
;
1712 * this iterates to turn a name (from iref/extref) into a full filesystem path.
1713 * Elements of the path are separated by '/' and the path is guaranteed to be
1714 * 0-terminated. the path is only given within the current file system.
1715 * Therefore, it never starts with a '/'. the caller is responsible to provide
1716 * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
1717 * the start point of the resulting string is returned. this pointer is within
1719 * in case the path buffer would overflow, the pointer is decremented further
1720 * as if output was written to the buffer, though no more output is actually
1721 * generated. that way, the caller can determine how much space would be
1722 * required for the path to fit into the buffer. in that case, the returned
1723 * value will be smaller than dest. callers must check this!
1725 char *btrfs_ref_to_path(struct btrfs_root
*fs_root
, struct btrfs_path
*path
,
1726 u32 name_len
, unsigned long name_off
,
1727 struct extent_buffer
*eb_in
, u64 parent
,
1728 char *dest
, u32 size
)
1733 s64 bytes_left
= ((s64
)size
) - 1;
1734 struct extent_buffer
*eb
= eb_in
;
1735 struct btrfs_key found_key
;
1736 int leave_spinning
= path
->leave_spinning
;
1737 struct btrfs_inode_ref
*iref
;
1739 if (bytes_left
>= 0)
1740 dest
[bytes_left
] = '\0';
1742 path
->leave_spinning
= 1;
1744 bytes_left
-= name_len
;
1745 if (bytes_left
>= 0)
1746 read_extent_buffer(eb
, dest
+ bytes_left
,
1747 name_off
, name_len
);
1749 if (!path
->skip_locking
)
1750 btrfs_tree_read_unlock_blocking(eb
);
1751 free_extent_buffer(eb
);
1753 ret
= btrfs_find_item(fs_root
, path
, parent
, 0,
1754 BTRFS_INODE_REF_KEY
, &found_key
);
1760 next_inum
= found_key
.offset
;
1762 /* regular exit ahead */
1763 if (parent
== next_inum
)
1766 slot
= path
->slots
[0];
1767 eb
= path
->nodes
[0];
1768 /* make sure we can use eb after releasing the path */
1770 if (!path
->skip_locking
)
1771 btrfs_set_lock_blocking_rw(eb
, BTRFS_READ_LOCK
);
1772 path
->nodes
[0] = NULL
;
1775 btrfs_release_path(path
);
1776 iref
= btrfs_item_ptr(eb
, slot
, struct btrfs_inode_ref
);
1778 name_len
= btrfs_inode_ref_name_len(eb
, iref
);
1779 name_off
= (unsigned long)(iref
+ 1);
1783 if (bytes_left
>= 0)
1784 dest
[bytes_left
] = '/';
1787 btrfs_release_path(path
);
1788 path
->leave_spinning
= leave_spinning
;
1791 return ERR_PTR(ret
);
1793 return dest
+ bytes_left
;
1797 * this makes the path point to (logical EXTENT_ITEM *)
1798 * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
1799 * tree blocks and <0 on error.
1801 int extent_from_logical(struct btrfs_fs_info
*fs_info
, u64 logical
,
1802 struct btrfs_path
*path
, struct btrfs_key
*found_key
,
1809 struct extent_buffer
*eb
;
1810 struct btrfs_extent_item
*ei
;
1811 struct btrfs_key key
;
1813 if (btrfs_fs_incompat(fs_info
, SKINNY_METADATA
))
1814 key
.type
= BTRFS_METADATA_ITEM_KEY
;
1816 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1817 key
.objectid
= logical
;
1818 key
.offset
= (u64
)-1;
1820 ret
= btrfs_search_slot(NULL
, fs_info
->extent_root
, &key
, path
, 0, 0);
1824 ret
= btrfs_previous_extent_item(fs_info
->extent_root
, path
, 0);
1830 btrfs_item_key_to_cpu(path
->nodes
[0], found_key
, path
->slots
[0]);
1831 if (found_key
->type
== BTRFS_METADATA_ITEM_KEY
)
1832 size
= fs_info
->extent_root
->nodesize
;
1833 else if (found_key
->type
== BTRFS_EXTENT_ITEM_KEY
)
1834 size
= found_key
->offset
;
1836 if (found_key
->objectid
> logical
||
1837 found_key
->objectid
+ size
<= logical
) {
1838 btrfs_debug(fs_info
,
1839 "logical %llu is not within any extent", logical
);
1843 eb
= path
->nodes
[0];
1844 item_size
= btrfs_item_size_nr(eb
, path
->slots
[0]);
1845 BUG_ON(item_size
< sizeof(*ei
));
1847 ei
= btrfs_item_ptr(eb
, path
->slots
[0], struct btrfs_extent_item
);
1848 flags
= btrfs_extent_flags(eb
, ei
);
1850 btrfs_debug(fs_info
,
1851 "logical %llu is at position %llu within the extent (%llu EXTENT_ITEM %llu) flags %#llx size %u",
1852 logical
, logical
- found_key
->objectid
, found_key
->objectid
,
1853 found_key
->offset
, flags
, item_size
);
1855 WARN_ON(!flags_ret
);
1857 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
)
1858 *flags_ret
= BTRFS_EXTENT_FLAG_TREE_BLOCK
;
1859 else if (flags
& BTRFS_EXTENT_FLAG_DATA
)
1860 *flags_ret
= BTRFS_EXTENT_FLAG_DATA
;
1870 * helper function to iterate extent inline refs. ptr must point to a 0 value
1871 * for the first call and may be modified. it is used to track state.
1872 * if more refs exist, 0 is returned and the next call to
1873 * __get_extent_inline_ref must pass the modified ptr parameter to get the
1874 * next ref. after the last ref was processed, 1 is returned.
1875 * returns <0 on error
1877 static int __get_extent_inline_ref(unsigned long *ptr
, struct extent_buffer
*eb
,
1878 struct btrfs_key
*key
,
1879 struct btrfs_extent_item
*ei
, u32 item_size
,
1880 struct btrfs_extent_inline_ref
**out_eiref
,
1885 struct btrfs_tree_block_info
*info
;
1889 flags
= btrfs_extent_flags(eb
, ei
);
1890 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
) {
1891 if (key
->type
== BTRFS_METADATA_ITEM_KEY
) {
1892 /* a skinny metadata extent */
1894 (struct btrfs_extent_inline_ref
*)(ei
+ 1);
1896 WARN_ON(key
->type
!= BTRFS_EXTENT_ITEM_KEY
);
1897 info
= (struct btrfs_tree_block_info
*)(ei
+ 1);
1899 (struct btrfs_extent_inline_ref
*)(info
+ 1);
1902 *out_eiref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
1904 *ptr
= (unsigned long)*out_eiref
;
1905 if ((unsigned long)(*ptr
) >= (unsigned long)ei
+ item_size
)
1909 end
= (unsigned long)ei
+ item_size
;
1910 *out_eiref
= (struct btrfs_extent_inline_ref
*)(*ptr
);
1911 *out_type
= btrfs_extent_inline_ref_type(eb
, *out_eiref
);
1913 *ptr
+= btrfs_extent_inline_ref_size(*out_type
);
1914 WARN_ON(*ptr
> end
);
1916 return 1; /* last */
1922 * reads the tree block backref for an extent. tree level and root are returned
1923 * through out_level and out_root. ptr must point to a 0 value for the first
1924 * call and may be modified (see __get_extent_inline_ref comment).
1925 * returns 0 if data was provided, 1 if there was no more data to provide or
1928 int tree_backref_for_extent(unsigned long *ptr
, struct extent_buffer
*eb
,
1929 struct btrfs_key
*key
, struct btrfs_extent_item
*ei
,
1930 u32 item_size
, u64
*out_root
, u8
*out_level
)
1934 struct btrfs_extent_inline_ref
*eiref
;
1936 if (*ptr
== (unsigned long)-1)
1940 ret
= __get_extent_inline_ref(ptr
, eb
, key
, ei
, item_size
,
1945 if (type
== BTRFS_TREE_BLOCK_REF_KEY
||
1946 type
== BTRFS_SHARED_BLOCK_REF_KEY
)
1953 /* we can treat both ref types equally here */
1954 *out_root
= btrfs_extent_inline_ref_offset(eb
, eiref
);
1956 if (key
->type
== BTRFS_EXTENT_ITEM_KEY
) {
1957 struct btrfs_tree_block_info
*info
;
1959 info
= (struct btrfs_tree_block_info
*)(ei
+ 1);
1960 *out_level
= btrfs_tree_block_level(eb
, info
);
1962 ASSERT(key
->type
== BTRFS_METADATA_ITEM_KEY
);
1963 *out_level
= (u8
)key
->offset
;
1967 *ptr
= (unsigned long)-1;
1972 static int iterate_leaf_refs(struct btrfs_fs_info
*fs_info
,
1973 struct extent_inode_elem
*inode_list
,
1974 u64 root
, u64 extent_item_objectid
,
1975 iterate_extent_inodes_t
*iterate
, void *ctx
)
1977 struct extent_inode_elem
*eie
;
1980 for (eie
= inode_list
; eie
; eie
= eie
->next
) {
1981 btrfs_debug(fs_info
,
1982 "ref for %llu resolved, key (%llu EXTEND_DATA %llu), root %llu",
1983 extent_item_objectid
, eie
->inum
,
1985 ret
= iterate(eie
->inum
, eie
->offset
, root
, ctx
);
1987 btrfs_debug(fs_info
,
1988 "stopping iteration for %llu due to ret=%d",
1989 extent_item_objectid
, ret
);
1998 * calls iterate() for every inode that references the extent identified by
1999 * the given parameters.
2000 * when the iterator function returns a non-zero value, iteration stops.
2002 int iterate_extent_inodes(struct btrfs_fs_info
*fs_info
,
2003 u64 extent_item_objectid
, u64 extent_item_pos
,
2004 int search_commit_root
,
2005 iterate_extent_inodes_t
*iterate
, void *ctx
)
2008 struct btrfs_trans_handle
*trans
= NULL
;
2009 struct ulist
*refs
= NULL
;
2010 struct ulist
*roots
= NULL
;
2011 struct ulist_node
*ref_node
= NULL
;
2012 struct ulist_node
*root_node
= NULL
;
2013 struct seq_list tree_mod_seq_elem
= SEQ_LIST_INIT(tree_mod_seq_elem
);
2014 struct ulist_iterator ref_uiter
;
2015 struct ulist_iterator root_uiter
;
2017 btrfs_debug(fs_info
, "resolving all inodes for extent %llu",
2018 extent_item_objectid
);
2020 if (!search_commit_root
) {
2021 trans
= btrfs_join_transaction(fs_info
->extent_root
);
2023 return PTR_ERR(trans
);
2024 btrfs_get_tree_mod_seq(fs_info
, &tree_mod_seq_elem
);
2026 down_read(&fs_info
->commit_root_sem
);
2029 ret
= btrfs_find_all_leafs(trans
, fs_info
, extent_item_objectid
,
2030 tree_mod_seq_elem
.seq
, &refs
,
2035 ULIST_ITER_INIT(&ref_uiter
);
2036 while (!ret
&& (ref_node
= ulist_next(refs
, &ref_uiter
))) {
2037 ret
= __btrfs_find_all_roots(trans
, fs_info
, ref_node
->val
,
2038 tree_mod_seq_elem
.seq
, &roots
);
2041 ULIST_ITER_INIT(&root_uiter
);
2042 while (!ret
&& (root_node
= ulist_next(roots
, &root_uiter
))) {
2043 btrfs_debug(fs_info
,
2044 "root %llu references leaf %llu, data list %#llx",
2045 root_node
->val
, ref_node
->val
,
2047 ret
= iterate_leaf_refs(fs_info
,
2048 (struct extent_inode_elem
*)
2049 (uintptr_t)ref_node
->aux
,
2051 extent_item_objectid
,
2057 free_leaf_list(refs
);
2059 if (!search_commit_root
) {
2060 btrfs_put_tree_mod_seq(fs_info
, &tree_mod_seq_elem
);
2061 btrfs_end_transaction(trans
, fs_info
->extent_root
);
2063 up_read(&fs_info
->commit_root_sem
);
2069 int iterate_inodes_from_logical(u64 logical
, struct btrfs_fs_info
*fs_info
,
2070 struct btrfs_path
*path
,
2071 iterate_extent_inodes_t
*iterate
, void *ctx
)
2074 u64 extent_item_pos
;
2076 struct btrfs_key found_key
;
2077 int search_commit_root
= path
->search_commit_root
;
2079 ret
= extent_from_logical(fs_info
, logical
, path
, &found_key
, &flags
);
2080 btrfs_release_path(path
);
2083 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
)
2086 extent_item_pos
= logical
- found_key
.objectid
;
2087 ret
= iterate_extent_inodes(fs_info
, found_key
.objectid
,
2088 extent_item_pos
, search_commit_root
,
2094 typedef int (iterate_irefs_t
)(u64 parent
, u32 name_len
, unsigned long name_off
,
2095 struct extent_buffer
*eb
, void *ctx
);
2097 static int iterate_inode_refs(u64 inum
, struct btrfs_root
*fs_root
,
2098 struct btrfs_path
*path
,
2099 iterate_irefs_t
*iterate
, void *ctx
)
2108 struct extent_buffer
*eb
;
2109 struct btrfs_item
*item
;
2110 struct btrfs_inode_ref
*iref
;
2111 struct btrfs_key found_key
;
2114 ret
= btrfs_find_item(fs_root
, path
, inum
,
2115 parent
? parent
+ 1 : 0, BTRFS_INODE_REF_KEY
,
2121 ret
= found
? 0 : -ENOENT
;
2126 parent
= found_key
.offset
;
2127 slot
= path
->slots
[0];
2128 eb
= btrfs_clone_extent_buffer(path
->nodes
[0]);
2133 extent_buffer_get(eb
);
2134 btrfs_tree_read_lock(eb
);
2135 btrfs_set_lock_blocking_rw(eb
, BTRFS_READ_LOCK
);
2136 btrfs_release_path(path
);
2138 item
= btrfs_item_nr(slot
);
2139 iref
= btrfs_item_ptr(eb
, slot
, struct btrfs_inode_ref
);
2141 for (cur
= 0; cur
< btrfs_item_size(eb
, item
); cur
+= len
) {
2142 name_len
= btrfs_inode_ref_name_len(eb
, iref
);
2143 /* path must be released before calling iterate()! */
2144 btrfs_debug(fs_root
->fs_info
,
2145 "following ref at offset %u for inode %llu in tree %llu",
2146 cur
, found_key
.objectid
, fs_root
->objectid
);
2147 ret
= iterate(parent
, name_len
,
2148 (unsigned long)(iref
+ 1), eb
, ctx
);
2151 len
= sizeof(*iref
) + name_len
;
2152 iref
= (struct btrfs_inode_ref
*)((char *)iref
+ len
);
2154 btrfs_tree_read_unlock_blocking(eb
);
2155 free_extent_buffer(eb
);
2158 btrfs_release_path(path
);
2163 static int iterate_inode_extrefs(u64 inum
, struct btrfs_root
*fs_root
,
2164 struct btrfs_path
*path
,
2165 iterate_irefs_t
*iterate
, void *ctx
)
2172 struct extent_buffer
*eb
;
2173 struct btrfs_inode_extref
*extref
;
2179 ret
= btrfs_find_one_extref(fs_root
, inum
, offset
, path
, &extref
,
2184 ret
= found
? 0 : -ENOENT
;
2189 slot
= path
->slots
[0];
2190 eb
= btrfs_clone_extent_buffer(path
->nodes
[0]);
2195 extent_buffer_get(eb
);
2197 btrfs_tree_read_lock(eb
);
2198 btrfs_set_lock_blocking_rw(eb
, BTRFS_READ_LOCK
);
2199 btrfs_release_path(path
);
2201 item_size
= btrfs_item_size_nr(eb
, slot
);
2202 ptr
= btrfs_item_ptr_offset(eb
, slot
);
2205 while (cur_offset
< item_size
) {
2208 extref
= (struct btrfs_inode_extref
*)(ptr
+ cur_offset
);
2209 parent
= btrfs_inode_extref_parent(eb
, extref
);
2210 name_len
= btrfs_inode_extref_name_len(eb
, extref
);
2211 ret
= iterate(parent
, name_len
,
2212 (unsigned long)&extref
->name
, eb
, ctx
);
2216 cur_offset
+= btrfs_inode_extref_name_len(eb
, extref
);
2217 cur_offset
+= sizeof(*extref
);
2219 btrfs_tree_read_unlock_blocking(eb
);
2220 free_extent_buffer(eb
);
2225 btrfs_release_path(path
);
2230 static int iterate_irefs(u64 inum
, struct btrfs_root
*fs_root
,
2231 struct btrfs_path
*path
, iterate_irefs_t
*iterate
,
2237 ret
= iterate_inode_refs(inum
, fs_root
, path
, iterate
, ctx
);
2240 else if (ret
!= -ENOENT
)
2243 ret
= iterate_inode_extrefs(inum
, fs_root
, path
, iterate
, ctx
);
2244 if (ret
== -ENOENT
&& found_refs
)
2251 * returns 0 if the path could be dumped (probably truncated)
2252 * returns <0 in case of an error
2254 static int inode_to_path(u64 inum
, u32 name_len
, unsigned long name_off
,
2255 struct extent_buffer
*eb
, void *ctx
)
2257 struct inode_fs_paths
*ipath
= ctx
;
2260 int i
= ipath
->fspath
->elem_cnt
;
2261 const int s_ptr
= sizeof(char *);
2264 bytes_left
= ipath
->fspath
->bytes_left
> s_ptr
?
2265 ipath
->fspath
->bytes_left
- s_ptr
: 0;
2267 fspath_min
= (char *)ipath
->fspath
->val
+ (i
+ 1) * s_ptr
;
2268 fspath
= btrfs_ref_to_path(ipath
->fs_root
, ipath
->btrfs_path
, name_len
,
2269 name_off
, eb
, inum
, fspath_min
, bytes_left
);
2271 return PTR_ERR(fspath
);
2273 if (fspath
> fspath_min
) {
2274 ipath
->fspath
->val
[i
] = (u64
)(unsigned long)fspath
;
2275 ++ipath
->fspath
->elem_cnt
;
2276 ipath
->fspath
->bytes_left
= fspath
- fspath_min
;
2278 ++ipath
->fspath
->elem_missed
;
2279 ipath
->fspath
->bytes_missing
+= fspath_min
- fspath
;
2280 ipath
->fspath
->bytes_left
= 0;
2287 * this dumps all file system paths to the inode into the ipath struct, provided
2288 * is has been created large enough. each path is zero-terminated and accessed
2289 * from ipath->fspath->val[i].
2290 * when it returns, there are ipath->fspath->elem_cnt number of paths available
2291 * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the
2292 * number of missed paths is recorded in ipath->fspath->elem_missed, otherwise,
2293 * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
2294 * have been needed to return all paths.
2296 int paths_from_inode(u64 inum
, struct inode_fs_paths
*ipath
)
2298 return iterate_irefs(inum
, ipath
->fs_root
, ipath
->btrfs_path
,
2299 inode_to_path
, ipath
);
2302 struct btrfs_data_container
*init_data_container(u32 total_bytes
)
2304 struct btrfs_data_container
*data
;
2307 alloc_bytes
= max_t(size_t, total_bytes
, sizeof(*data
));
2308 data
= vmalloc(alloc_bytes
);
2310 return ERR_PTR(-ENOMEM
);
2312 if (total_bytes
>= sizeof(*data
)) {
2313 data
->bytes_left
= total_bytes
- sizeof(*data
);
2314 data
->bytes_missing
= 0;
2316 data
->bytes_missing
= sizeof(*data
) - total_bytes
;
2317 data
->bytes_left
= 0;
2321 data
->elem_missed
= 0;
2327 * allocates space to return multiple file system paths for an inode.
2328 * total_bytes to allocate are passed, note that space usable for actual path
2329 * information will be total_bytes - sizeof(struct inode_fs_paths).
2330 * the returned pointer must be freed with free_ipath() in the end.
2332 struct inode_fs_paths
*init_ipath(s32 total_bytes
, struct btrfs_root
*fs_root
,
2333 struct btrfs_path
*path
)
2335 struct inode_fs_paths
*ifp
;
2336 struct btrfs_data_container
*fspath
;
2338 fspath
= init_data_container(total_bytes
);
2340 return (void *)fspath
;
2342 ifp
= kmalloc(sizeof(*ifp
), GFP_NOFS
);
2345 return ERR_PTR(-ENOMEM
);
2348 ifp
->btrfs_path
= path
;
2349 ifp
->fspath
= fspath
;
2350 ifp
->fs_root
= fs_root
;
2355 void free_ipath(struct inode_fs_paths
*ipath
)
2359 vfree(ipath
->fspath
);