Linux 3.12.39
[linux/fpc-iii.git] / fs / btrfs / backref.c
blob53039de1495ded653f57f9e81a7be363e7527420
1 /*
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 "ctree.h"
21 #include "disk-io.h"
22 #include "backref.h"
23 #include "ulist.h"
24 #include "transaction.h"
25 #include "delayed-ref.h"
26 #include "locking.h"
28 struct extent_inode_elem {
29 u64 inum;
30 u64 offset;
31 struct extent_inode_elem *next;
34 static int check_extent_in_eb(struct btrfs_key *key, struct extent_buffer *eb,
35 struct btrfs_file_extent_item *fi,
36 u64 extent_item_pos,
37 struct extent_inode_elem **eie)
39 u64 offset = 0;
40 struct extent_inode_elem *e;
42 if (!btrfs_file_extent_compression(eb, fi) &&
43 !btrfs_file_extent_encryption(eb, fi) &&
44 !btrfs_file_extent_other_encoding(eb, fi)) {
45 u64 data_offset;
46 u64 data_len;
48 data_offset = btrfs_file_extent_offset(eb, fi);
49 data_len = btrfs_file_extent_num_bytes(eb, fi);
51 if (extent_item_pos < data_offset ||
52 extent_item_pos >= data_offset + data_len)
53 return 1;
54 offset = extent_item_pos - data_offset;
57 e = kmalloc(sizeof(*e), GFP_NOFS);
58 if (!e)
59 return -ENOMEM;
61 e->next = *eie;
62 e->inum = key->objectid;
63 e->offset = key->offset + offset;
64 *eie = e;
66 return 0;
69 static int find_extent_in_eb(struct extent_buffer *eb, u64 wanted_disk_byte,
70 u64 extent_item_pos,
71 struct extent_inode_elem **eie)
73 u64 disk_byte;
74 struct btrfs_key key;
75 struct btrfs_file_extent_item *fi;
76 int slot;
77 int nritems;
78 int extent_type;
79 int ret;
82 * from the shared data ref, we only have the leaf but we need
83 * the key. thus, we must look into all items and see that we
84 * find one (some) with a reference to our extent item.
86 nritems = btrfs_header_nritems(eb);
87 for (slot = 0; slot < nritems; ++slot) {
88 btrfs_item_key_to_cpu(eb, &key, slot);
89 if (key.type != BTRFS_EXTENT_DATA_KEY)
90 continue;
91 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
92 extent_type = btrfs_file_extent_type(eb, fi);
93 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
94 continue;
95 /* don't skip BTRFS_FILE_EXTENT_PREALLOC, we can handle that */
96 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
97 if (disk_byte != wanted_disk_byte)
98 continue;
100 ret = check_extent_in_eb(&key, eb, fi, extent_item_pos, eie);
101 if (ret < 0)
102 return ret;
105 return 0;
109 * this structure records all encountered refs on the way up to the root
111 struct __prelim_ref {
112 struct list_head list;
113 u64 root_id;
114 struct btrfs_key key_for_search;
115 int level;
116 int count;
117 struct extent_inode_elem *inode_list;
118 u64 parent;
119 u64 wanted_disk_byte;
122 static struct kmem_cache *btrfs_prelim_ref_cache;
124 int __init btrfs_prelim_ref_init(void)
126 btrfs_prelim_ref_cache = kmem_cache_create("btrfs_prelim_ref",
127 sizeof(struct __prelim_ref),
129 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
130 NULL);
131 if (!btrfs_prelim_ref_cache)
132 return -ENOMEM;
133 return 0;
136 void btrfs_prelim_ref_exit(void)
138 if (btrfs_prelim_ref_cache)
139 kmem_cache_destroy(btrfs_prelim_ref_cache);
143 * the rules for all callers of this function are:
144 * - obtaining the parent is the goal
145 * - if you add a key, you must know that it is a correct key
146 * - if you cannot add the parent or a correct key, then we will look into the
147 * block later to set a correct key
149 * delayed refs
150 * ============
151 * backref type | shared | indirect | shared | indirect
152 * information | tree | tree | data | data
153 * --------------------+--------+----------+--------+----------
154 * parent logical | y | - | - | -
155 * key to resolve | - | y | y | y
156 * tree block logical | - | - | - | -
157 * root for resolving | y | y | y | y
159 * - column 1: we've the parent -> done
160 * - column 2, 3, 4: we use the key to find the parent
162 * on disk refs (inline or keyed)
163 * ==============================
164 * backref type | shared | indirect | shared | indirect
165 * information | tree | tree | data | data
166 * --------------------+--------+----------+--------+----------
167 * parent logical | y | - | y | -
168 * key to resolve | - | - | - | y
169 * tree block logical | y | y | y | y
170 * root for resolving | - | y | y | y
172 * - column 1, 3: we've the parent -> done
173 * - column 2: we take the first key from the block to find the parent
174 * (see __add_missing_keys)
175 * - column 4: we use the key to find the parent
177 * additional information that's available but not required to find the parent
178 * block might help in merging entries to gain some speed.
181 static int __add_prelim_ref(struct list_head *head, u64 root_id,
182 struct btrfs_key *key, int level,
183 u64 parent, u64 wanted_disk_byte, int count,
184 gfp_t gfp_mask)
186 struct __prelim_ref *ref;
188 if (root_id == BTRFS_DATA_RELOC_TREE_OBJECTID)
189 return 0;
191 ref = kmem_cache_alloc(btrfs_prelim_ref_cache, gfp_mask);
192 if (!ref)
193 return -ENOMEM;
195 ref->root_id = root_id;
196 if (key)
197 ref->key_for_search = *key;
198 else
199 memset(&ref->key_for_search, 0, sizeof(ref->key_for_search));
201 ref->inode_list = NULL;
202 ref->level = level;
203 ref->count = count;
204 ref->parent = parent;
205 ref->wanted_disk_byte = wanted_disk_byte;
206 list_add_tail(&ref->list, head);
208 return 0;
211 static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path,
212 struct ulist *parents, int level,
213 struct btrfs_key *key_for_search, u64 time_seq,
214 u64 wanted_disk_byte,
215 const u64 *extent_item_pos)
217 int ret = 0;
218 int slot;
219 struct extent_buffer *eb;
220 struct btrfs_key key;
221 struct btrfs_file_extent_item *fi;
222 struct extent_inode_elem *eie = NULL, *old = NULL;
223 u64 disk_byte;
225 if (level != 0) {
226 eb = path->nodes[level];
227 ret = ulist_add(parents, eb->start, 0, GFP_NOFS);
228 if (ret < 0)
229 return ret;
230 return 0;
234 * We normally enter this function with the path already pointing to
235 * the first item to check. But sometimes, we may enter it with
236 * slot==nritems. In that case, go to the next leaf before we continue.
238 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0]))
239 ret = btrfs_next_old_leaf(root, path, time_seq);
241 while (!ret) {
242 eb = path->nodes[0];
243 slot = path->slots[0];
245 btrfs_item_key_to_cpu(eb, &key, slot);
247 if (key.objectid != key_for_search->objectid ||
248 key.type != BTRFS_EXTENT_DATA_KEY)
249 break;
251 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
252 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
254 if (disk_byte == wanted_disk_byte) {
255 eie = NULL;
256 old = NULL;
257 if (extent_item_pos) {
258 ret = check_extent_in_eb(&key, eb, fi,
259 *extent_item_pos,
260 &eie);
261 if (ret < 0)
262 break;
264 if (ret > 0)
265 goto next;
266 ret = ulist_add_merge_ptr(parents, eb->start,
267 eie, (void **)&old, GFP_NOFS);
268 if (ret < 0)
269 break;
270 if (!ret && extent_item_pos) {
271 while (old->next)
272 old = old->next;
273 old->next = eie;
276 next:
277 ret = btrfs_next_old_item(root, path, time_seq);
280 if (ret > 0)
281 ret = 0;
282 return ret;
286 * resolve an indirect backref in the form (root_id, key, level)
287 * to a logical address
289 static int __resolve_indirect_ref(struct btrfs_fs_info *fs_info,
290 struct btrfs_path *path, u64 time_seq,
291 struct __prelim_ref *ref,
292 struct ulist *parents,
293 const u64 *extent_item_pos)
295 struct btrfs_root *root;
296 struct btrfs_key root_key;
297 struct extent_buffer *eb;
298 int ret = 0;
299 int root_level;
300 int level = ref->level;
302 root_key.objectid = ref->root_id;
303 root_key.type = BTRFS_ROOT_ITEM_KEY;
304 root_key.offset = (u64)-1;
305 root = btrfs_read_fs_root_no_name(fs_info, &root_key);
306 if (IS_ERR(root)) {
307 ret = PTR_ERR(root);
308 goto out;
311 root_level = btrfs_old_root_level(root, time_seq);
313 if (root_level + 1 == level)
314 goto out;
316 path->lowest_level = level;
317 ret = btrfs_search_old_slot(root, &ref->key_for_search, path, time_seq);
318 pr_debug("search slot in root %llu (level %d, ref count %d) returned "
319 "%d for key (%llu %u %llu)\n",
320 ref->root_id, level, ref->count, ret,
321 ref->key_for_search.objectid, ref->key_for_search.type,
322 ref->key_for_search.offset);
323 if (ret < 0)
324 goto out;
326 eb = path->nodes[level];
327 while (!eb) {
328 if (!level) {
329 WARN_ON(1);
330 ret = 1;
331 goto out;
333 level--;
334 eb = path->nodes[level];
337 ret = add_all_parents(root, path, parents, level, &ref->key_for_search,
338 time_seq, ref->wanted_disk_byte,
339 extent_item_pos);
340 out:
341 path->lowest_level = 0;
342 btrfs_release_path(path);
343 return ret;
347 * resolve all indirect backrefs from the list
349 static int __resolve_indirect_refs(struct btrfs_fs_info *fs_info,
350 struct btrfs_path *path, u64 time_seq,
351 struct list_head *head,
352 const u64 *extent_item_pos)
354 int err;
355 int ret = 0;
356 struct __prelim_ref *ref;
357 struct __prelim_ref *ref_safe;
358 struct __prelim_ref *new_ref;
359 struct ulist *parents;
360 struct ulist_node *node;
361 struct ulist_iterator uiter;
363 parents = ulist_alloc(GFP_NOFS);
364 if (!parents)
365 return -ENOMEM;
368 * _safe allows us to insert directly after the current item without
369 * iterating over the newly inserted items.
370 * we're also allowed to re-assign ref during iteration.
372 list_for_each_entry_safe(ref, ref_safe, head, list) {
373 if (ref->parent) /* already direct */
374 continue;
375 if (ref->count == 0)
376 continue;
377 err = __resolve_indirect_ref(fs_info, path, time_seq, ref,
378 parents, extent_item_pos);
379 if (err == -ENOMEM)
380 goto out;
381 if (err)
382 continue;
384 /* we put the first parent into the ref at hand */
385 ULIST_ITER_INIT(&uiter);
386 node = ulist_next(parents, &uiter);
387 ref->parent = node ? node->val : 0;
388 ref->inode_list = node ?
389 (struct extent_inode_elem *)(uintptr_t)node->aux : NULL;
391 /* additional parents require new refs being added here */
392 while ((node = ulist_next(parents, &uiter))) {
393 new_ref = kmem_cache_alloc(btrfs_prelim_ref_cache,
394 GFP_NOFS);
395 if (!new_ref) {
396 ret = -ENOMEM;
397 goto out;
399 memcpy(new_ref, ref, sizeof(*ref));
400 new_ref->parent = node->val;
401 new_ref->inode_list = (struct extent_inode_elem *)
402 (uintptr_t)node->aux;
403 list_add(&new_ref->list, &ref->list);
405 ulist_reinit(parents);
407 out:
408 ulist_free(parents);
409 return ret;
412 static inline int ref_for_same_block(struct __prelim_ref *ref1,
413 struct __prelim_ref *ref2)
415 if (ref1->level != ref2->level)
416 return 0;
417 if (ref1->root_id != ref2->root_id)
418 return 0;
419 if (ref1->key_for_search.type != ref2->key_for_search.type)
420 return 0;
421 if (ref1->key_for_search.objectid != ref2->key_for_search.objectid)
422 return 0;
423 if (ref1->key_for_search.offset != ref2->key_for_search.offset)
424 return 0;
425 if (ref1->parent != ref2->parent)
426 return 0;
428 return 1;
432 * read tree blocks and add keys where required.
434 static int __add_missing_keys(struct btrfs_fs_info *fs_info,
435 struct list_head *head)
437 struct list_head *pos;
438 struct extent_buffer *eb;
440 list_for_each(pos, head) {
441 struct __prelim_ref *ref;
442 ref = list_entry(pos, struct __prelim_ref, list);
444 if (ref->parent)
445 continue;
446 if (ref->key_for_search.type)
447 continue;
448 BUG_ON(!ref->wanted_disk_byte);
449 eb = read_tree_block(fs_info->tree_root, ref->wanted_disk_byte,
450 fs_info->tree_root->leafsize, 0);
451 if (!eb || !extent_buffer_uptodate(eb)) {
452 free_extent_buffer(eb);
453 return -EIO;
455 btrfs_tree_read_lock(eb);
456 if (btrfs_header_level(eb) == 0)
457 btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
458 else
459 btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
460 btrfs_tree_read_unlock(eb);
461 free_extent_buffer(eb);
463 return 0;
467 * merge two lists of backrefs and adjust counts accordingly
469 * mode = 1: merge identical keys, if key is set
470 * FIXME: if we add more keys in __add_prelim_ref, we can merge more here.
471 * additionally, we could even add a key range for the blocks we
472 * looked into to merge even more (-> replace unresolved refs by those
473 * having a parent).
474 * mode = 2: merge identical parents
476 static void __merge_refs(struct list_head *head, int mode)
478 struct list_head *pos1;
480 list_for_each(pos1, head) {
481 struct list_head *n2;
482 struct list_head *pos2;
483 struct __prelim_ref *ref1;
485 ref1 = list_entry(pos1, struct __prelim_ref, list);
487 for (pos2 = pos1->next, n2 = pos2->next; pos2 != head;
488 pos2 = n2, n2 = pos2->next) {
489 struct __prelim_ref *ref2;
490 struct __prelim_ref *xchg;
491 struct extent_inode_elem *eie;
493 ref2 = list_entry(pos2, struct __prelim_ref, list);
495 if (mode == 1) {
496 if (!ref_for_same_block(ref1, ref2))
497 continue;
498 if (!ref1->parent && ref2->parent) {
499 xchg = ref1;
500 ref1 = ref2;
501 ref2 = xchg;
503 } else {
504 if (ref1->parent != ref2->parent)
505 continue;
508 eie = ref1->inode_list;
509 while (eie && eie->next)
510 eie = eie->next;
511 if (eie)
512 eie->next = ref2->inode_list;
513 else
514 ref1->inode_list = ref2->inode_list;
515 ref1->count += ref2->count;
517 list_del(&ref2->list);
518 kmem_cache_free(btrfs_prelim_ref_cache, ref2);
525 * add all currently queued delayed refs from this head whose seq nr is
526 * smaller or equal that seq to the list
528 static int __add_delayed_refs(struct btrfs_delayed_ref_head *head, u64 seq,
529 struct list_head *prefs)
531 struct btrfs_delayed_extent_op *extent_op = head->extent_op;
532 struct rb_node *n = &head->node.rb_node;
533 struct btrfs_key key;
534 struct btrfs_key op_key = {0};
535 int sgn;
536 int ret = 0;
538 if (extent_op && extent_op->update_key)
539 btrfs_disk_key_to_cpu(&op_key, &extent_op->key);
541 while ((n = rb_prev(n))) {
542 struct btrfs_delayed_ref_node *node;
543 node = rb_entry(n, struct btrfs_delayed_ref_node,
544 rb_node);
545 if (node->bytenr != head->node.bytenr)
546 break;
547 WARN_ON(node->is_head);
549 if (node->seq > seq)
550 continue;
552 switch (node->action) {
553 case BTRFS_ADD_DELAYED_EXTENT:
554 case BTRFS_UPDATE_DELAYED_HEAD:
555 WARN_ON(1);
556 continue;
557 case BTRFS_ADD_DELAYED_REF:
558 sgn = 1;
559 break;
560 case BTRFS_DROP_DELAYED_REF:
561 sgn = -1;
562 break;
563 default:
564 BUG_ON(1);
566 switch (node->type) {
567 case BTRFS_TREE_BLOCK_REF_KEY: {
568 struct btrfs_delayed_tree_ref *ref;
570 ref = btrfs_delayed_node_to_tree_ref(node);
571 ret = __add_prelim_ref(prefs, ref->root, &op_key,
572 ref->level + 1, 0, node->bytenr,
573 node->ref_mod * sgn, GFP_ATOMIC);
574 break;
576 case BTRFS_SHARED_BLOCK_REF_KEY: {
577 struct btrfs_delayed_tree_ref *ref;
579 ref = btrfs_delayed_node_to_tree_ref(node);
580 ret = __add_prelim_ref(prefs, ref->root, NULL,
581 ref->level + 1, ref->parent,
582 node->bytenr,
583 node->ref_mod * sgn, GFP_ATOMIC);
584 break;
586 case BTRFS_EXTENT_DATA_REF_KEY: {
587 struct btrfs_delayed_data_ref *ref;
588 ref = btrfs_delayed_node_to_data_ref(node);
590 key.objectid = ref->objectid;
591 key.type = BTRFS_EXTENT_DATA_KEY;
592 key.offset = ref->offset;
593 ret = __add_prelim_ref(prefs, ref->root, &key, 0, 0,
594 node->bytenr,
595 node->ref_mod * sgn, GFP_ATOMIC);
596 break;
598 case BTRFS_SHARED_DATA_REF_KEY: {
599 struct btrfs_delayed_data_ref *ref;
601 ref = btrfs_delayed_node_to_data_ref(node);
603 key.objectid = ref->objectid;
604 key.type = BTRFS_EXTENT_DATA_KEY;
605 key.offset = ref->offset;
606 ret = __add_prelim_ref(prefs, ref->root, &key, 0,
607 ref->parent, node->bytenr,
608 node->ref_mod * sgn, GFP_ATOMIC);
609 break;
611 default:
612 WARN_ON(1);
614 if (ret)
615 return ret;
618 return 0;
622 * add all inline backrefs for bytenr to the list
624 static int __add_inline_refs(struct btrfs_fs_info *fs_info,
625 struct btrfs_path *path, u64 bytenr,
626 int *info_level, struct list_head *prefs)
628 int ret = 0;
629 int slot;
630 struct extent_buffer *leaf;
631 struct btrfs_key key;
632 struct btrfs_key found_key;
633 unsigned long ptr;
634 unsigned long end;
635 struct btrfs_extent_item *ei;
636 u64 flags;
637 u64 item_size;
640 * enumerate all inline refs
642 leaf = path->nodes[0];
643 slot = path->slots[0];
645 item_size = btrfs_item_size_nr(leaf, slot);
646 BUG_ON(item_size < sizeof(*ei));
648 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
649 flags = btrfs_extent_flags(leaf, ei);
650 btrfs_item_key_to_cpu(leaf, &found_key, slot);
652 ptr = (unsigned long)(ei + 1);
653 end = (unsigned long)ei + item_size;
655 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
656 flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
657 struct btrfs_tree_block_info *info;
659 info = (struct btrfs_tree_block_info *)ptr;
660 *info_level = btrfs_tree_block_level(leaf, info);
661 ptr += sizeof(struct btrfs_tree_block_info);
662 BUG_ON(ptr > end);
663 } else if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
664 *info_level = found_key.offset;
665 } else {
666 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
669 while (ptr < end) {
670 struct btrfs_extent_inline_ref *iref;
671 u64 offset;
672 int type;
674 iref = (struct btrfs_extent_inline_ref *)ptr;
675 type = btrfs_extent_inline_ref_type(leaf, iref);
676 offset = btrfs_extent_inline_ref_offset(leaf, iref);
678 switch (type) {
679 case BTRFS_SHARED_BLOCK_REF_KEY:
680 ret = __add_prelim_ref(prefs, 0, NULL,
681 *info_level + 1, offset,
682 bytenr, 1, GFP_NOFS);
683 break;
684 case BTRFS_SHARED_DATA_REF_KEY: {
685 struct btrfs_shared_data_ref *sdref;
686 int count;
688 sdref = (struct btrfs_shared_data_ref *)(iref + 1);
689 count = btrfs_shared_data_ref_count(leaf, sdref);
690 ret = __add_prelim_ref(prefs, 0, NULL, 0, offset,
691 bytenr, count, GFP_NOFS);
692 break;
694 case BTRFS_TREE_BLOCK_REF_KEY:
695 ret = __add_prelim_ref(prefs, offset, NULL,
696 *info_level + 1, 0,
697 bytenr, 1, GFP_NOFS);
698 break;
699 case BTRFS_EXTENT_DATA_REF_KEY: {
700 struct btrfs_extent_data_ref *dref;
701 int count;
702 u64 root;
704 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
705 count = btrfs_extent_data_ref_count(leaf, dref);
706 key.objectid = btrfs_extent_data_ref_objectid(leaf,
707 dref);
708 key.type = BTRFS_EXTENT_DATA_KEY;
709 key.offset = btrfs_extent_data_ref_offset(leaf, dref);
710 root = btrfs_extent_data_ref_root(leaf, dref);
711 ret = __add_prelim_ref(prefs, root, &key, 0, 0,
712 bytenr, count, GFP_NOFS);
713 break;
715 default:
716 WARN_ON(1);
718 if (ret)
719 return ret;
720 ptr += btrfs_extent_inline_ref_size(type);
723 return 0;
727 * add all non-inline backrefs for bytenr to the list
729 static int __add_keyed_refs(struct btrfs_fs_info *fs_info,
730 struct btrfs_path *path, u64 bytenr,
731 int info_level, struct list_head *prefs)
733 struct btrfs_root *extent_root = fs_info->extent_root;
734 int ret;
735 int slot;
736 struct extent_buffer *leaf;
737 struct btrfs_key key;
739 while (1) {
740 ret = btrfs_next_item(extent_root, path);
741 if (ret < 0)
742 break;
743 if (ret) {
744 ret = 0;
745 break;
748 slot = path->slots[0];
749 leaf = path->nodes[0];
750 btrfs_item_key_to_cpu(leaf, &key, slot);
752 if (key.objectid != bytenr)
753 break;
754 if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
755 continue;
756 if (key.type > BTRFS_SHARED_DATA_REF_KEY)
757 break;
759 switch (key.type) {
760 case BTRFS_SHARED_BLOCK_REF_KEY:
761 ret = __add_prelim_ref(prefs, 0, NULL,
762 info_level + 1, key.offset,
763 bytenr, 1, GFP_NOFS);
764 break;
765 case BTRFS_SHARED_DATA_REF_KEY: {
766 struct btrfs_shared_data_ref *sdref;
767 int count;
769 sdref = btrfs_item_ptr(leaf, slot,
770 struct btrfs_shared_data_ref);
771 count = btrfs_shared_data_ref_count(leaf, sdref);
772 ret = __add_prelim_ref(prefs, 0, NULL, 0, key.offset,
773 bytenr, count, GFP_NOFS);
774 break;
776 case BTRFS_TREE_BLOCK_REF_KEY:
777 ret = __add_prelim_ref(prefs, key.offset, NULL,
778 info_level + 1, 0,
779 bytenr, 1, GFP_NOFS);
780 break;
781 case BTRFS_EXTENT_DATA_REF_KEY: {
782 struct btrfs_extent_data_ref *dref;
783 int count;
784 u64 root;
786 dref = btrfs_item_ptr(leaf, slot,
787 struct btrfs_extent_data_ref);
788 count = btrfs_extent_data_ref_count(leaf, dref);
789 key.objectid = btrfs_extent_data_ref_objectid(leaf,
790 dref);
791 key.type = BTRFS_EXTENT_DATA_KEY;
792 key.offset = btrfs_extent_data_ref_offset(leaf, dref);
793 root = btrfs_extent_data_ref_root(leaf, dref);
794 ret = __add_prelim_ref(prefs, root, &key, 0, 0,
795 bytenr, count, GFP_NOFS);
796 break;
798 default:
799 WARN_ON(1);
801 if (ret)
802 return ret;
806 return ret;
810 * this adds all existing backrefs (inline backrefs, backrefs and delayed
811 * refs) for the given bytenr to the refs list, merges duplicates and resolves
812 * indirect refs to their parent bytenr.
813 * When roots are found, they're added to the roots list
815 * FIXME some caching might speed things up
817 static int find_parent_nodes(struct btrfs_trans_handle *trans,
818 struct btrfs_fs_info *fs_info, u64 bytenr,
819 u64 time_seq, struct ulist *refs,
820 struct ulist *roots, const u64 *extent_item_pos)
822 struct btrfs_key key;
823 struct btrfs_path *path;
824 struct btrfs_delayed_ref_root *delayed_refs = NULL;
825 struct btrfs_delayed_ref_head *head;
826 int info_level = 0;
827 int ret;
828 struct list_head prefs_delayed;
829 struct list_head prefs;
830 struct __prelim_ref *ref;
832 INIT_LIST_HEAD(&prefs);
833 INIT_LIST_HEAD(&prefs_delayed);
835 key.objectid = bytenr;
836 key.offset = (u64)-1;
837 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
838 key.type = BTRFS_METADATA_ITEM_KEY;
839 else
840 key.type = BTRFS_EXTENT_ITEM_KEY;
842 path = btrfs_alloc_path();
843 if (!path)
844 return -ENOMEM;
845 if (!trans)
846 path->search_commit_root = 1;
849 * grab both a lock on the path and a lock on the delayed ref head.
850 * We need both to get a consistent picture of how the refs look
851 * at a specified point in time
853 again:
854 head = NULL;
856 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
857 if (ret < 0)
858 goto out;
859 BUG_ON(ret == 0);
861 if (trans) {
863 * look if there are updates for this ref queued and lock the
864 * head
866 delayed_refs = &trans->transaction->delayed_refs;
867 spin_lock(&delayed_refs->lock);
868 head = btrfs_find_delayed_ref_head(trans, bytenr);
869 if (head) {
870 if (!mutex_trylock(&head->mutex)) {
871 atomic_inc(&head->node.refs);
872 spin_unlock(&delayed_refs->lock);
874 btrfs_release_path(path);
877 * Mutex was contended, block until it's
878 * released and try again
880 mutex_lock(&head->mutex);
881 mutex_unlock(&head->mutex);
882 btrfs_put_delayed_ref(&head->node);
883 goto again;
885 ret = __add_delayed_refs(head, time_seq,
886 &prefs_delayed);
887 mutex_unlock(&head->mutex);
888 if (ret) {
889 spin_unlock(&delayed_refs->lock);
890 goto out;
893 spin_unlock(&delayed_refs->lock);
896 if (path->slots[0]) {
897 struct extent_buffer *leaf;
898 int slot;
900 path->slots[0]--;
901 leaf = path->nodes[0];
902 slot = path->slots[0];
903 btrfs_item_key_to_cpu(leaf, &key, slot);
904 if (key.objectid == bytenr &&
905 (key.type == BTRFS_EXTENT_ITEM_KEY ||
906 key.type == BTRFS_METADATA_ITEM_KEY)) {
907 ret = __add_inline_refs(fs_info, path, bytenr,
908 &info_level, &prefs);
909 if (ret)
910 goto out;
911 ret = __add_keyed_refs(fs_info, path, bytenr,
912 info_level, &prefs);
913 if (ret)
914 goto out;
917 btrfs_release_path(path);
919 list_splice_init(&prefs_delayed, &prefs);
921 ret = __add_missing_keys(fs_info, &prefs);
922 if (ret)
923 goto out;
925 __merge_refs(&prefs, 1);
927 ret = __resolve_indirect_refs(fs_info, path, time_seq, &prefs,
928 extent_item_pos);
929 if (ret)
930 goto out;
932 __merge_refs(&prefs, 2);
934 while (!list_empty(&prefs)) {
935 ref = list_first_entry(&prefs, struct __prelim_ref, list);
936 WARN_ON(ref->count < 0);
937 if (ref->count && ref->root_id && ref->parent == 0) {
938 /* no parent == root of tree */
939 ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS);
940 if (ret < 0)
941 goto out;
943 if (ref->count && ref->parent) {
944 struct extent_inode_elem *eie = NULL;
945 if (extent_item_pos && !ref->inode_list) {
946 u32 bsz;
947 struct extent_buffer *eb;
948 bsz = btrfs_level_size(fs_info->extent_root,
949 info_level);
950 eb = read_tree_block(fs_info->extent_root,
951 ref->parent, bsz, 0);
952 if (!eb || !extent_buffer_uptodate(eb)) {
953 free_extent_buffer(eb);
954 ret = -EIO;
955 goto out;
957 btrfs_tree_read_lock(eb);
958 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
959 ret = find_extent_in_eb(eb, bytenr,
960 *extent_item_pos, &eie);
961 btrfs_tree_read_unlock_blocking(eb);
962 free_extent_buffer(eb);
963 if (ret < 0)
964 goto out;
965 ref->inode_list = eie;
967 ret = ulist_add_merge_ptr(refs, ref->parent,
968 ref->inode_list,
969 (void **)&eie, GFP_NOFS);
970 if (ret < 0)
971 goto out;
972 if (!ret && extent_item_pos) {
974 * we've recorded that parent, so we must extend
975 * its inode list here
977 BUG_ON(!eie);
978 while (eie->next)
979 eie = eie->next;
980 eie->next = ref->inode_list;
983 list_del(&ref->list);
984 kmem_cache_free(btrfs_prelim_ref_cache, ref);
987 out:
988 btrfs_free_path(path);
989 while (!list_empty(&prefs)) {
990 ref = list_first_entry(&prefs, struct __prelim_ref, list);
991 list_del(&ref->list);
992 kmem_cache_free(btrfs_prelim_ref_cache, ref);
994 while (!list_empty(&prefs_delayed)) {
995 ref = list_first_entry(&prefs_delayed, struct __prelim_ref,
996 list);
997 list_del(&ref->list);
998 kmem_cache_free(btrfs_prelim_ref_cache, ref);
1001 return ret;
1004 static void free_leaf_list(struct ulist *blocks)
1006 struct ulist_node *node = NULL;
1007 struct extent_inode_elem *eie;
1008 struct extent_inode_elem *eie_next;
1009 struct ulist_iterator uiter;
1011 ULIST_ITER_INIT(&uiter);
1012 while ((node = ulist_next(blocks, &uiter))) {
1013 if (!node->aux)
1014 continue;
1015 eie = (struct extent_inode_elem *)(uintptr_t)node->aux;
1016 for (; eie; eie = eie_next) {
1017 eie_next = eie->next;
1018 kfree(eie);
1020 node->aux = 0;
1023 ulist_free(blocks);
1027 * Finds all leafs with a reference to the specified combination of bytenr and
1028 * offset. key_list_head will point to a list of corresponding keys (caller must
1029 * free each list element). The leafs will be stored in the leafs ulist, which
1030 * must be freed with ulist_free.
1032 * returns 0 on success, <0 on error
1034 static int btrfs_find_all_leafs(struct btrfs_trans_handle *trans,
1035 struct btrfs_fs_info *fs_info, u64 bytenr,
1036 u64 time_seq, struct ulist **leafs,
1037 const u64 *extent_item_pos)
1039 struct ulist *tmp;
1040 int ret;
1042 tmp = ulist_alloc(GFP_NOFS);
1043 if (!tmp)
1044 return -ENOMEM;
1045 *leafs = ulist_alloc(GFP_NOFS);
1046 if (!*leafs) {
1047 ulist_free(tmp);
1048 return -ENOMEM;
1051 ret = find_parent_nodes(trans, fs_info, bytenr,
1052 time_seq, *leafs, tmp, extent_item_pos);
1053 ulist_free(tmp);
1055 if (ret < 0 && ret != -ENOENT) {
1056 free_leaf_list(*leafs);
1057 return ret;
1060 return 0;
1064 * walk all backrefs for a given extent to find all roots that reference this
1065 * extent. Walking a backref means finding all extents that reference this
1066 * extent and in turn walk the backrefs of those, too. Naturally this is a
1067 * recursive process, but here it is implemented in an iterative fashion: We
1068 * find all referencing extents for the extent in question and put them on a
1069 * list. In turn, we find all referencing extents for those, further appending
1070 * to the list. The way we iterate the list allows adding more elements after
1071 * the current while iterating. The process stops when we reach the end of the
1072 * list. Found roots are added to the roots list.
1074 * returns 0 on success, < 0 on error.
1076 int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1077 struct btrfs_fs_info *fs_info, u64 bytenr,
1078 u64 time_seq, struct ulist **roots)
1080 struct ulist *tmp;
1081 struct ulist_node *node = NULL;
1082 struct ulist_iterator uiter;
1083 int ret;
1085 tmp = ulist_alloc(GFP_NOFS);
1086 if (!tmp)
1087 return -ENOMEM;
1088 *roots = ulist_alloc(GFP_NOFS);
1089 if (!*roots) {
1090 ulist_free(tmp);
1091 return -ENOMEM;
1094 ULIST_ITER_INIT(&uiter);
1095 while (1) {
1096 ret = find_parent_nodes(trans, fs_info, bytenr,
1097 time_seq, tmp, *roots, NULL);
1098 if (ret < 0 && ret != -ENOENT) {
1099 ulist_free(tmp);
1100 ulist_free(*roots);
1101 return ret;
1103 node = ulist_next(tmp, &uiter);
1104 if (!node)
1105 break;
1106 bytenr = node->val;
1109 ulist_free(tmp);
1110 return 0;
1114 static int __inode_info(u64 inum, u64 ioff, u8 key_type,
1115 struct btrfs_root *fs_root, struct btrfs_path *path,
1116 struct btrfs_key *found_key)
1118 int ret;
1119 struct btrfs_key key;
1120 struct extent_buffer *eb;
1122 key.type = key_type;
1123 key.objectid = inum;
1124 key.offset = ioff;
1126 ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
1127 if (ret < 0)
1128 return ret;
1130 eb = path->nodes[0];
1131 if (ret && path->slots[0] >= btrfs_header_nritems(eb)) {
1132 ret = btrfs_next_leaf(fs_root, path);
1133 if (ret)
1134 return ret;
1135 eb = path->nodes[0];
1138 btrfs_item_key_to_cpu(eb, found_key, path->slots[0]);
1139 if (found_key->type != key.type || found_key->objectid != key.objectid)
1140 return 1;
1142 return 0;
1146 * this makes the path point to (inum INODE_ITEM ioff)
1148 int inode_item_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
1149 struct btrfs_path *path)
1151 struct btrfs_key key;
1152 return __inode_info(inum, ioff, BTRFS_INODE_ITEM_KEY, fs_root, path,
1153 &key);
1156 static int inode_ref_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
1157 struct btrfs_path *path,
1158 struct btrfs_key *found_key)
1160 return __inode_info(inum, ioff, BTRFS_INODE_REF_KEY, fs_root, path,
1161 found_key);
1164 int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,
1165 u64 start_off, struct btrfs_path *path,
1166 struct btrfs_inode_extref **ret_extref,
1167 u64 *found_off)
1169 int ret, slot;
1170 struct btrfs_key key;
1171 struct btrfs_key found_key;
1172 struct btrfs_inode_extref *extref;
1173 struct extent_buffer *leaf;
1174 unsigned long ptr;
1176 key.objectid = inode_objectid;
1177 btrfs_set_key_type(&key, BTRFS_INODE_EXTREF_KEY);
1178 key.offset = start_off;
1180 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1181 if (ret < 0)
1182 return ret;
1184 while (1) {
1185 leaf = path->nodes[0];
1186 slot = path->slots[0];
1187 if (slot >= btrfs_header_nritems(leaf)) {
1189 * If the item at offset is not found,
1190 * btrfs_search_slot will point us to the slot
1191 * where it should be inserted. In our case
1192 * that will be the slot directly before the
1193 * next INODE_REF_KEY_V2 item. In the case
1194 * that we're pointing to the last slot in a
1195 * leaf, we must move one leaf over.
1197 ret = btrfs_next_leaf(root, path);
1198 if (ret) {
1199 if (ret >= 1)
1200 ret = -ENOENT;
1201 break;
1203 continue;
1206 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1209 * Check that we're still looking at an extended ref key for
1210 * this particular objectid. If we have different
1211 * objectid or type then there are no more to be found
1212 * in the tree and we can exit.
1214 ret = -ENOENT;
1215 if (found_key.objectid != inode_objectid)
1216 break;
1217 if (btrfs_key_type(&found_key) != BTRFS_INODE_EXTREF_KEY)
1218 break;
1220 ret = 0;
1221 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1222 extref = (struct btrfs_inode_extref *)ptr;
1223 *ret_extref = extref;
1224 if (found_off)
1225 *found_off = found_key.offset;
1226 break;
1229 return ret;
1233 * this iterates to turn a name (from iref/extref) into a full filesystem path.
1234 * Elements of the path are separated by '/' and the path is guaranteed to be
1235 * 0-terminated. the path is only given within the current file system.
1236 * Therefore, it never starts with a '/'. the caller is responsible to provide
1237 * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
1238 * the start point of the resulting string is returned. this pointer is within
1239 * dest, normally.
1240 * in case the path buffer would overflow, the pointer is decremented further
1241 * as if output was written to the buffer, though no more output is actually
1242 * generated. that way, the caller can determine how much space would be
1243 * required for the path to fit into the buffer. in that case, the returned
1244 * value will be smaller than dest. callers must check this!
1246 char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
1247 u32 name_len, unsigned long name_off,
1248 struct extent_buffer *eb_in, u64 parent,
1249 char *dest, u32 size)
1251 int slot;
1252 u64 next_inum;
1253 int ret;
1254 s64 bytes_left = ((s64)size) - 1;
1255 struct extent_buffer *eb = eb_in;
1256 struct btrfs_key found_key;
1257 int leave_spinning = path->leave_spinning;
1258 struct btrfs_inode_ref *iref;
1260 if (bytes_left >= 0)
1261 dest[bytes_left] = '\0';
1263 path->leave_spinning = 1;
1264 while (1) {
1265 bytes_left -= name_len;
1266 if (bytes_left >= 0)
1267 read_extent_buffer(eb, dest + bytes_left,
1268 name_off, name_len);
1269 if (eb != eb_in) {
1270 btrfs_tree_read_unlock_blocking(eb);
1271 free_extent_buffer(eb);
1273 ret = inode_ref_info(parent, 0, fs_root, path, &found_key);
1274 if (ret > 0)
1275 ret = -ENOENT;
1276 if (ret)
1277 break;
1279 next_inum = found_key.offset;
1281 /* regular exit ahead */
1282 if (parent == next_inum)
1283 break;
1285 slot = path->slots[0];
1286 eb = path->nodes[0];
1287 /* make sure we can use eb after releasing the path */
1288 if (eb != eb_in) {
1289 atomic_inc(&eb->refs);
1290 btrfs_tree_read_lock(eb);
1291 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1293 btrfs_release_path(path);
1294 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1296 name_len = btrfs_inode_ref_name_len(eb, iref);
1297 name_off = (unsigned long)(iref + 1);
1299 parent = next_inum;
1300 --bytes_left;
1301 if (bytes_left >= 0)
1302 dest[bytes_left] = '/';
1305 btrfs_release_path(path);
1306 path->leave_spinning = leave_spinning;
1308 if (ret)
1309 return ERR_PTR(ret);
1311 return dest + bytes_left;
1315 * this makes the path point to (logical EXTENT_ITEM *)
1316 * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
1317 * tree blocks and <0 on error.
1319 int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
1320 struct btrfs_path *path, struct btrfs_key *found_key,
1321 u64 *flags_ret)
1323 int ret;
1324 u64 flags;
1325 u64 size = 0;
1326 u32 item_size;
1327 struct extent_buffer *eb;
1328 struct btrfs_extent_item *ei;
1329 struct btrfs_key key;
1331 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1332 key.type = BTRFS_METADATA_ITEM_KEY;
1333 else
1334 key.type = BTRFS_EXTENT_ITEM_KEY;
1335 key.objectid = logical;
1336 key.offset = (u64)-1;
1338 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
1339 if (ret < 0)
1340 return ret;
1341 ret = btrfs_previous_item(fs_info->extent_root, path,
1342 0, BTRFS_EXTENT_ITEM_KEY);
1343 if (ret < 0)
1344 return ret;
1346 btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]);
1347 if (found_key->type == BTRFS_METADATA_ITEM_KEY)
1348 size = fs_info->extent_root->leafsize;
1349 else if (found_key->type == BTRFS_EXTENT_ITEM_KEY)
1350 size = found_key->offset;
1352 if ((found_key->type != BTRFS_EXTENT_ITEM_KEY &&
1353 found_key->type != BTRFS_METADATA_ITEM_KEY) ||
1354 found_key->objectid > logical ||
1355 found_key->objectid + size <= logical) {
1356 pr_debug("logical %llu is not within any extent\n", logical);
1357 return -ENOENT;
1360 eb = path->nodes[0];
1361 item_size = btrfs_item_size_nr(eb, path->slots[0]);
1362 BUG_ON(item_size < sizeof(*ei));
1364 ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
1365 flags = btrfs_extent_flags(eb, ei);
1367 pr_debug("logical %llu is at position %llu within the extent (%llu "
1368 "EXTENT_ITEM %llu) flags %#llx size %u\n",
1369 logical, logical - found_key->objectid, found_key->objectid,
1370 found_key->offset, flags, item_size);
1372 WARN_ON(!flags_ret);
1373 if (flags_ret) {
1374 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1375 *flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK;
1376 else if (flags & BTRFS_EXTENT_FLAG_DATA)
1377 *flags_ret = BTRFS_EXTENT_FLAG_DATA;
1378 else
1379 BUG_ON(1);
1380 return 0;
1383 return -EIO;
1387 * helper function to iterate extent inline refs. ptr must point to a 0 value
1388 * for the first call and may be modified. it is used to track state.
1389 * if more refs exist, 0 is returned and the next call to
1390 * __get_extent_inline_ref must pass the modified ptr parameter to get the
1391 * next ref. after the last ref was processed, 1 is returned.
1392 * returns <0 on error
1394 static int __get_extent_inline_ref(unsigned long *ptr, struct extent_buffer *eb,
1395 struct btrfs_key *key,
1396 struct btrfs_extent_item *ei, u32 item_size,
1397 struct btrfs_extent_inline_ref **out_eiref,
1398 int *out_type)
1400 unsigned long end;
1401 u64 flags;
1402 struct btrfs_tree_block_info *info;
1404 if (!*ptr) {
1405 /* first call */
1406 flags = btrfs_extent_flags(eb, ei);
1407 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1408 if (key->type == BTRFS_METADATA_ITEM_KEY) {
1409 /* a skinny metadata extent */
1410 *out_eiref =
1411 (struct btrfs_extent_inline_ref *)(ei + 1);
1412 } else {
1413 WARN_ON(key->type != BTRFS_EXTENT_ITEM_KEY);
1414 info = (struct btrfs_tree_block_info *)(ei + 1);
1415 *out_eiref =
1416 (struct btrfs_extent_inline_ref *)(info + 1);
1418 } else {
1419 *out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1);
1421 *ptr = (unsigned long)*out_eiref;
1422 if ((unsigned long)(*ptr) >= (unsigned long)ei + item_size)
1423 return -ENOENT;
1426 end = (unsigned long)ei + item_size;
1427 *out_eiref = (struct btrfs_extent_inline_ref *)(*ptr);
1428 *out_type = btrfs_extent_inline_ref_type(eb, *out_eiref);
1430 *ptr += btrfs_extent_inline_ref_size(*out_type);
1431 WARN_ON(*ptr > end);
1432 if (*ptr == end)
1433 return 1; /* last */
1435 return 0;
1439 * reads the tree block backref for an extent. tree level and root are returned
1440 * through out_level and out_root. ptr must point to a 0 value for the first
1441 * call and may be modified (see __get_extent_inline_ref comment).
1442 * returns 0 if data was provided, 1 if there was no more data to provide or
1443 * <0 on error.
1445 int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
1446 struct btrfs_key *key, struct btrfs_extent_item *ei,
1447 u32 item_size, u64 *out_root, u8 *out_level)
1449 int ret;
1450 int type;
1451 struct btrfs_tree_block_info *info;
1452 struct btrfs_extent_inline_ref *eiref;
1454 if (*ptr == (unsigned long)-1)
1455 return 1;
1457 while (1) {
1458 ret = __get_extent_inline_ref(ptr, eb, key, ei, item_size,
1459 &eiref, &type);
1460 if (ret < 0)
1461 return ret;
1463 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1464 type == BTRFS_SHARED_BLOCK_REF_KEY)
1465 break;
1467 if (ret == 1)
1468 return 1;
1471 /* we can treat both ref types equally here */
1472 info = (struct btrfs_tree_block_info *)(ei + 1);
1473 *out_root = btrfs_extent_inline_ref_offset(eb, eiref);
1474 *out_level = btrfs_tree_block_level(eb, info);
1476 if (ret == 1)
1477 *ptr = (unsigned long)-1;
1479 return 0;
1482 static int iterate_leaf_refs(struct extent_inode_elem *inode_list,
1483 u64 root, u64 extent_item_objectid,
1484 iterate_extent_inodes_t *iterate, void *ctx)
1486 struct extent_inode_elem *eie;
1487 int ret = 0;
1489 for (eie = inode_list; eie; eie = eie->next) {
1490 pr_debug("ref for %llu resolved, key (%llu EXTEND_DATA %llu), "
1491 "root %llu\n", extent_item_objectid,
1492 eie->inum, eie->offset, root);
1493 ret = iterate(eie->inum, eie->offset, root, ctx);
1494 if (ret) {
1495 pr_debug("stopping iteration for %llu due to ret=%d\n",
1496 extent_item_objectid, ret);
1497 break;
1501 return ret;
1505 * calls iterate() for every inode that references the extent identified by
1506 * the given parameters.
1507 * when the iterator function returns a non-zero value, iteration stops.
1509 int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
1510 u64 extent_item_objectid, u64 extent_item_pos,
1511 int search_commit_root,
1512 iterate_extent_inodes_t *iterate, void *ctx)
1514 int ret;
1515 struct btrfs_trans_handle *trans = NULL;
1516 struct ulist *refs = NULL;
1517 struct ulist *roots = NULL;
1518 struct ulist_node *ref_node = NULL;
1519 struct ulist_node *root_node = NULL;
1520 struct seq_list tree_mod_seq_elem = {};
1521 struct ulist_iterator ref_uiter;
1522 struct ulist_iterator root_uiter;
1524 pr_debug("resolving all inodes for extent %llu\n",
1525 extent_item_objectid);
1527 if (!search_commit_root) {
1528 trans = btrfs_join_transaction(fs_info->extent_root);
1529 if (IS_ERR(trans))
1530 return PTR_ERR(trans);
1531 btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1534 ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid,
1535 tree_mod_seq_elem.seq, &refs,
1536 &extent_item_pos);
1537 if (ret)
1538 goto out;
1540 ULIST_ITER_INIT(&ref_uiter);
1541 while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) {
1542 ret = btrfs_find_all_roots(trans, fs_info, ref_node->val,
1543 tree_mod_seq_elem.seq, &roots);
1544 if (ret)
1545 break;
1546 ULIST_ITER_INIT(&root_uiter);
1547 while (!ret && (root_node = ulist_next(roots, &root_uiter))) {
1548 pr_debug("root %llu references leaf %llu, data list "
1549 "%#llx\n", root_node->val, ref_node->val,
1550 ref_node->aux);
1551 ret = iterate_leaf_refs((struct extent_inode_elem *)
1552 (uintptr_t)ref_node->aux,
1553 root_node->val,
1554 extent_item_objectid,
1555 iterate, ctx);
1557 ulist_free(roots);
1560 free_leaf_list(refs);
1561 out:
1562 if (!search_commit_root) {
1563 btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1564 btrfs_end_transaction(trans, fs_info->extent_root);
1567 return ret;
1570 int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
1571 struct btrfs_path *path,
1572 iterate_extent_inodes_t *iterate, void *ctx)
1574 int ret;
1575 u64 extent_item_pos;
1576 u64 flags = 0;
1577 struct btrfs_key found_key;
1578 int search_commit_root = path->search_commit_root;
1580 ret = extent_from_logical(fs_info, logical, path, &found_key, &flags);
1581 btrfs_release_path(path);
1582 if (ret < 0)
1583 return ret;
1584 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1585 return -EINVAL;
1587 extent_item_pos = logical - found_key.objectid;
1588 ret = iterate_extent_inodes(fs_info, found_key.objectid,
1589 extent_item_pos, search_commit_root,
1590 iterate, ctx);
1592 return ret;
1595 typedef int (iterate_irefs_t)(u64 parent, u32 name_len, unsigned long name_off,
1596 struct extent_buffer *eb, void *ctx);
1598 static int iterate_inode_refs(u64 inum, struct btrfs_root *fs_root,
1599 struct btrfs_path *path,
1600 iterate_irefs_t *iterate, void *ctx)
1602 int ret = 0;
1603 int slot;
1604 u32 cur;
1605 u32 len;
1606 u32 name_len;
1607 u64 parent = 0;
1608 int found = 0;
1609 struct extent_buffer *eb;
1610 struct btrfs_item *item;
1611 struct btrfs_inode_ref *iref;
1612 struct btrfs_key found_key;
1614 while (!ret) {
1615 path->leave_spinning = 1;
1616 ret = inode_ref_info(inum, parent ? parent+1 : 0, fs_root, path,
1617 &found_key);
1618 if (ret < 0)
1619 break;
1620 if (ret) {
1621 ret = found ? 0 : -ENOENT;
1622 break;
1624 ++found;
1626 parent = found_key.offset;
1627 slot = path->slots[0];
1628 eb = path->nodes[0];
1629 /* make sure we can use eb after releasing the path */
1630 atomic_inc(&eb->refs);
1631 btrfs_tree_read_lock(eb);
1632 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1633 btrfs_release_path(path);
1635 item = btrfs_item_nr(eb, slot);
1636 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1638 for (cur = 0; cur < btrfs_item_size(eb, item); cur += len) {
1639 name_len = btrfs_inode_ref_name_len(eb, iref);
1640 /* path must be released before calling iterate()! */
1641 pr_debug("following ref at offset %u for inode %llu in "
1642 "tree %llu\n", cur, found_key.objectid,
1643 fs_root->objectid);
1644 ret = iterate(parent, name_len,
1645 (unsigned long)(iref + 1), eb, ctx);
1646 if (ret)
1647 break;
1648 len = sizeof(*iref) + name_len;
1649 iref = (struct btrfs_inode_ref *)((char *)iref + len);
1651 btrfs_tree_read_unlock_blocking(eb);
1652 free_extent_buffer(eb);
1655 btrfs_release_path(path);
1657 return ret;
1660 static int iterate_inode_extrefs(u64 inum, struct btrfs_root *fs_root,
1661 struct btrfs_path *path,
1662 iterate_irefs_t *iterate, void *ctx)
1664 int ret;
1665 int slot;
1666 u64 offset = 0;
1667 u64 parent;
1668 int found = 0;
1669 struct extent_buffer *eb;
1670 struct btrfs_inode_extref *extref;
1671 struct extent_buffer *leaf;
1672 u32 item_size;
1673 u32 cur_offset;
1674 unsigned long ptr;
1676 while (1) {
1677 ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref,
1678 &offset);
1679 if (ret < 0)
1680 break;
1681 if (ret) {
1682 ret = found ? 0 : -ENOENT;
1683 break;
1685 ++found;
1687 slot = path->slots[0];
1688 eb = path->nodes[0];
1689 /* make sure we can use eb after releasing the path */
1690 atomic_inc(&eb->refs);
1692 btrfs_tree_read_lock(eb);
1693 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1694 btrfs_release_path(path);
1696 leaf = path->nodes[0];
1697 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1698 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1699 cur_offset = 0;
1701 while (cur_offset < item_size) {
1702 u32 name_len;
1704 extref = (struct btrfs_inode_extref *)(ptr + cur_offset);
1705 parent = btrfs_inode_extref_parent(eb, extref);
1706 name_len = btrfs_inode_extref_name_len(eb, extref);
1707 ret = iterate(parent, name_len,
1708 (unsigned long)&extref->name, eb, ctx);
1709 if (ret)
1710 break;
1712 cur_offset += btrfs_inode_extref_name_len(leaf, extref);
1713 cur_offset += sizeof(*extref);
1715 btrfs_tree_read_unlock_blocking(eb);
1716 free_extent_buffer(eb);
1718 offset++;
1721 btrfs_release_path(path);
1723 return ret;
1726 static int iterate_irefs(u64 inum, struct btrfs_root *fs_root,
1727 struct btrfs_path *path, iterate_irefs_t *iterate,
1728 void *ctx)
1730 int ret;
1731 int found_refs = 0;
1733 ret = iterate_inode_refs(inum, fs_root, path, iterate, ctx);
1734 if (!ret)
1735 ++found_refs;
1736 else if (ret != -ENOENT)
1737 return ret;
1739 ret = iterate_inode_extrefs(inum, fs_root, path, iterate, ctx);
1740 if (ret == -ENOENT && found_refs)
1741 return 0;
1743 return ret;
1747 * returns 0 if the path could be dumped (probably truncated)
1748 * returns <0 in case of an error
1750 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
1751 struct extent_buffer *eb, void *ctx)
1753 struct inode_fs_paths *ipath = ctx;
1754 char *fspath;
1755 char *fspath_min;
1756 int i = ipath->fspath->elem_cnt;
1757 const int s_ptr = sizeof(char *);
1758 u32 bytes_left;
1760 bytes_left = ipath->fspath->bytes_left > s_ptr ?
1761 ipath->fspath->bytes_left - s_ptr : 0;
1763 fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
1764 fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len,
1765 name_off, eb, inum, fspath_min, bytes_left);
1766 if (IS_ERR(fspath))
1767 return PTR_ERR(fspath);
1769 if (fspath > fspath_min) {
1770 ipath->fspath->val[i] = (u64)(unsigned long)fspath;
1771 ++ipath->fspath->elem_cnt;
1772 ipath->fspath->bytes_left = fspath - fspath_min;
1773 } else {
1774 ++ipath->fspath->elem_missed;
1775 ipath->fspath->bytes_missing += fspath_min - fspath;
1776 ipath->fspath->bytes_left = 0;
1779 return 0;
1783 * this dumps all file system paths to the inode into the ipath struct, provided
1784 * is has been created large enough. each path is zero-terminated and accessed
1785 * from ipath->fspath->val[i].
1786 * when it returns, there are ipath->fspath->elem_cnt number of paths available
1787 * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the
1788 * number of missed paths in recored in ipath->fspath->elem_missed, otherwise,
1789 * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
1790 * have been needed to return all paths.
1792 int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
1794 return iterate_irefs(inum, ipath->fs_root, ipath->btrfs_path,
1795 inode_to_path, ipath);
1798 struct btrfs_data_container *init_data_container(u32 total_bytes)
1800 struct btrfs_data_container *data;
1801 size_t alloc_bytes;
1803 alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
1804 data = vmalloc(alloc_bytes);
1805 if (!data)
1806 return ERR_PTR(-ENOMEM);
1808 if (total_bytes >= sizeof(*data)) {
1809 data->bytes_left = total_bytes - sizeof(*data);
1810 data->bytes_missing = 0;
1811 } else {
1812 data->bytes_missing = sizeof(*data) - total_bytes;
1813 data->bytes_left = 0;
1816 data->elem_cnt = 0;
1817 data->elem_missed = 0;
1819 return data;
1823 * allocates space to return multiple file system paths for an inode.
1824 * total_bytes to allocate are passed, note that space usable for actual path
1825 * information will be total_bytes - sizeof(struct inode_fs_paths).
1826 * the returned pointer must be freed with free_ipath() in the end.
1828 struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
1829 struct btrfs_path *path)
1831 struct inode_fs_paths *ifp;
1832 struct btrfs_data_container *fspath;
1834 fspath = init_data_container(total_bytes);
1835 if (IS_ERR(fspath))
1836 return (void *)fspath;
1838 ifp = kmalloc(sizeof(*ifp), GFP_NOFS);
1839 if (!ifp) {
1840 kfree(fspath);
1841 return ERR_PTR(-ENOMEM);
1844 ifp->btrfs_path = path;
1845 ifp->fspath = fspath;
1846 ifp->fs_root = fs_root;
1848 return ifp;
1851 void free_ipath(struct inode_fs_paths *ipath)
1853 if (!ipath)
1854 return;
1855 vfree(ipath->fspath);
1856 kfree(ipath);