?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
btrfs-progs: fsck-tests: add test case with keyed data backref with reloc tree blocks
[btrfs-progs-unstable/devel.git] / backref.c
blob23a394edfd0298f02fa40701e58764e8aa96e09d
1 /*
2 * Copyright (C) 2011 STRATO. All rights reserved.
3 *
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.
7 *
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.
12 *
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.
17 */
18
19 #include "kerncompat.h"
20 #include "ctree.h"
21 #include "disk-io.h"
22 #include "backref.h"
23 #include "kernel-shared/ulist.h"
24 #include "transaction.h"
25 #include "internal.h"
26
27 #define pr_debug(...) do { } while (0)
28
29 struct extent_inode_elem {
30 u64 inum;
31 u64 offset;
32 struct extent_inode_elem *next;
33 };
34
35 static int check_extent_in_eb(struct btrfs_key *key, struct extent_buffer *eb,
36 struct btrfs_file_extent_item *fi,
37 u64 extent_item_pos,
38 struct extent_inode_elem **eie)
39 {
40 u64 offset = 0;
41 struct extent_inode_elem *e;
42
43 if (!btrfs_file_extent_compression(eb, fi) &&
44 !btrfs_file_extent_encryption(eb, fi) &&
45 !btrfs_file_extent_other_encoding(eb, fi)) {
46 u64 data_offset;
47 u64 data_len;
48
49 data_offset = btrfs_file_extent_offset(eb, fi);
50 data_len = btrfs_file_extent_num_bytes(eb, fi);
51
52 if (extent_item_pos < data_offset ||
53 extent_item_pos >= data_offset + data_len)
54 return 1;
55 offset = extent_item_pos - data_offset;
56 }
57
58 e = kmalloc(sizeof(*e), GFP_NOFS);
59 if (!e)
60 return -ENOMEM;
61
62 e->next = *eie;
63 e->inum = key->objectid;
64 e->offset = key->offset + offset;
65 *eie = e;
66
67 return 0;
68 }
69
70 static void free_inode_elem_list(struct extent_inode_elem *eie)
71 {
72 struct extent_inode_elem *eie_next;
73
74 for (; eie; eie = eie_next) {
75 eie_next = eie->next;
76 kfree(eie);
77 }
78 }
79
80 static int find_extent_in_eb(struct extent_buffer *eb, u64 wanted_disk_byte,
81 u64 extent_item_pos,
82 struct extent_inode_elem **eie)
83 {
84 u64 disk_byte;
85 struct btrfs_key key;
86 struct btrfs_file_extent_item *fi;
87 int slot;
88 int nritems;
89 int extent_type;
90 int ret;
91
92 /*
93 * from the shared data ref, we only have the leaf but we need
94 * the key. thus, we must look into all items and see that we
95 * find one (some) with a reference to our extent item.
96 */
97 nritems = btrfs_header_nritems(eb);
98 for (slot = 0; slot < nritems; ++slot) {
99 btrfs_item_key_to_cpu(eb, &key, slot);
100 if (key.type != BTRFS_EXTENT_DATA_KEY)
101 continue;
102 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
103 extent_type = btrfs_file_extent_type(eb, fi);
104 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
105 continue;
106 /* don't skip BTRFS_FILE_EXTENT_PREALLOC, we can handle that */
107 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
108 if (disk_byte != wanted_disk_byte)
109 continue;
110
111 ret = check_extent_in_eb(&key, eb, fi, extent_item_pos, eie);
112 if (ret < 0)
113 return ret;
114 }
115
116 return 0;
117 }
118
119 /*
120 * this structure records all encountered refs on the way up to the root
121 */
122 struct __prelim_ref {
123 struct list_head list;
124 u64 root_id;
125 struct btrfs_key key_for_search;
126 int level;
127 int count;
128 struct extent_inode_elem *inode_list;
129 u64 parent;
130 u64 wanted_disk_byte;
131 };
132
133 static struct __prelim_ref *list_first_pref(struct list_head *head)
134 {
135 return list_first_entry(head, struct __prelim_ref, list);
136 }
137
138 struct pref_state {
139 struct list_head pending;
140 struct list_head pending_missing_keys;
141 struct list_head pending_indirect_refs;
142 };
143
144 static void init_pref_state(struct pref_state *prefstate)
145 {
146 INIT_LIST_HEAD(&prefstate->pending);
147 INIT_LIST_HEAD(&prefstate->pending_missing_keys);
148 INIT_LIST_HEAD(&prefstate->pending_indirect_refs);
149 }
150
151 /*
152 * the rules for all callers of this function are:
153 * - obtaining the parent is the goal
154 * - if you add a key, you must know that it is a correct key
155 * - if you cannot add the parent or a correct key, then we will look into the
156 * block later to set a correct key
157 *
158 * on disk refs (inline or keyed)
159 * ==============================
160 * backref type | shared | indirect | shared | indirect
161 * information | tree | tree | data | data
162 * --------------------+--------+----------+--------+----------
163 * parent logical | y | - | y | -
164 * key to resolve | - | - | - | y
165 * tree block logical | y | y | y | y
166 * root for resolving | - | y | y | y
167 *
168 * - column 1, 3: we've the parent -> done
169 * - column 2: we take the first key from the block to find the parent
170 * (see __add_missing_keys)
171 * - column 4: we use the key to find the parent
172 *
173 * additional information that's available but not required to find the parent
174 * block might help in merging entries to gain some speed.
175 */
176
177 static int __add_prelim_ref(struct pref_state *prefstate, u64 root_id,
178 struct btrfs_key *key, int level,
179 u64 parent, u64 wanted_disk_byte, int count,
180 gfp_t gfp_mask)
181 {
182 struct list_head *head;
183 struct __prelim_ref *ref;
184
185 if (root_id == BTRFS_DATA_RELOC_TREE_OBJECTID)
186 return 0;
187
188 ref = kmalloc(sizeof(*ref), gfp_mask);
189 if (!ref)
190 return -ENOMEM;
191
192 ref->root_id = root_id;
193 if (key) {
194 ref->key_for_search = *key;
195 head = &prefstate->pending;
196 } else if (parent) {
197 memset(&ref->key_for_search, 0, sizeof(ref->key_for_search));
198 head = &prefstate->pending;
199 } else {
200 memset(&ref->key_for_search, 0, sizeof(ref->key_for_search));
201 head = &prefstate->pending_missing_keys;
202 }
203
204 ref->inode_list = NULL;
205 ref->level = level;
206 ref->count = count;
207 ref->parent = parent;
208 ref->wanted_disk_byte = wanted_disk_byte;
209
210 list_add_tail(&ref->list, head);
211
212 return 0;
213 }
214
215 static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path,
216 struct ulist *parents, struct __prelim_ref *ref,
217 int level, u64 time_seq, const u64 *extent_item_pos,
218 u64 total_refs)
219 {
220 int ret = 0;
221 int slot;
222 struct extent_buffer *eb;
223 struct btrfs_key key;
224 struct btrfs_key *key_for_search = &ref->key_for_search;
225 struct btrfs_file_extent_item *fi;
226 struct extent_inode_elem *eie = NULL, *old = NULL;
227 u64 disk_byte;
228 u64 wanted_disk_byte = ref->wanted_disk_byte;
229 u64 count = 0;
230
231 if (level != 0) {
232 eb = path->nodes[level];
233 ret = ulist_add(parents, eb->start, 0, GFP_NOFS);
234 if (ret < 0)
235 return ret;
236 return 0;
237 }
238
239 /*
240 * We normally enter this function with the path already pointing to
241 * the first item to check. But sometimes, we may enter it with
242 * slot==nritems. In that case, go to the next leaf before we continue.
243 */
244 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0]))
245 ret = btrfs_next_leaf(root, path);
246
247 while (!ret && count < total_refs) {
248 eb = path->nodes[0];
249 slot = path->slots[0];
250
251 btrfs_item_key_to_cpu(eb, &key, slot);
252
253 if (key.objectid != key_for_search->objectid ||
254 key.type != BTRFS_EXTENT_DATA_KEY)
255 break;
256
257 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
258 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
259
260 if (disk_byte == wanted_disk_byte) {
261 eie = NULL;
262 old = NULL;
263 count++;
264 if (extent_item_pos) {
265 ret = check_extent_in_eb(&key, eb, fi,
266 *extent_item_pos,
267 &eie);
268 if (ret < 0)
269 break;
270 }
271 if (ret > 0)
272 goto next;
273 ret = ulist_add_merge_ptr(parents, eb->start,
274 eie, (void **)&old, GFP_NOFS);
275 if (ret < 0)
276 break;
277 if (!ret && extent_item_pos) {
278 while (old->next)
279 old = old->next;
280 old->next = eie;
281 }
282 eie = NULL;
283 }
284 next:
285 ret = btrfs_next_item(root, path);
286 }
287
288 if (ret > 0)
289 ret = 0;
290 else if (ret < 0)
291 free_inode_elem_list(eie);
292 return ret;
293 }
294
295 /*
296 * resolve an indirect backref in the form (root_id, key, level)
297 * to a logical address
298 */
299 static int __resolve_indirect_ref(struct btrfs_fs_info *fs_info,
300 struct btrfs_path *path, u64 time_seq,
301 struct __prelim_ref *ref,
302 struct ulist *parents,
303 const u64 *extent_item_pos, u64 total_refs)
304 {
305 struct btrfs_root *root;
306 struct btrfs_key root_key;
307 struct extent_buffer *eb;
308 int ret = 0;
309 int root_level;
310 int level = ref->level;
311
312 root_key.objectid = ref->root_id;
313 root_key.type = BTRFS_ROOT_ITEM_KEY;
314 root_key.offset = (u64)-1;
315
316 root = btrfs_read_fs_root(fs_info, &root_key);
317 if (IS_ERR(root)) {
318 ret = PTR_ERR(root);
319 goto out;
320 }
321
322 root_level = btrfs_root_level(&root->root_item);
323
324 if (root_level + 1 == level)
325 goto out;
326
327 path->lowest_level = level;
328 ret = btrfs_search_slot(NULL, root, &ref->key_for_search, path, 0, 0);
329
330 pr_debug("search slot in root %llu (level %d, ref count %d) returned "
331 "%d for key (%llu %u %llu)\n",
332 ref->root_id, level, ref->count, ret,
333 ref->key_for_search.objectid, ref->key_for_search.type,
334 ref->key_for_search.offset);
335 if (ret < 0)
336 goto out;
337
338 eb = path->nodes[level];
339 while (!eb) {
340 if (!level) {
341 ret = 1;
342 WARN_ON(1);
343 goto out;
344 }
345 level--;
346 eb = path->nodes[level];
347 }
348
349 ret = add_all_parents(root, path, parents, ref, level, time_seq,
350 extent_item_pos, total_refs);
351 out:
352 path->lowest_level = 0;
353 btrfs_release_path(path);
354 return ret;
355 }
356
357 /*
358 * resolve all indirect backrefs from the list
359 */
360 static int __resolve_indirect_refs(struct btrfs_fs_info *fs_info,
361 struct pref_state *prefstate,
362 struct btrfs_path *path, u64 time_seq,
363 const u64 *extent_item_pos, u64 total_refs)
364 {
365 struct list_head *head = &prefstate->pending_indirect_refs;
366 int err;
367 int ret = 0;
368 struct __prelim_ref *ref;
369 struct __prelim_ref *new_ref;
370 struct ulist *parents;
371 struct ulist_node *node;
372 struct ulist_iterator uiter;
373
374 parents = ulist_alloc(GFP_NOFS);
375 if (!parents)
376 return -ENOMEM;
377
378 while (!list_empty(head)) {
379 ref = list_first_pref(head);
380 list_move(&ref->list, &prefstate->pending);
381 ASSERT(!ref->parent); /* already direct */
382 ASSERT(ref->count);
383 err = __resolve_indirect_ref(fs_info, path, time_seq, ref,
384 parents, extent_item_pos,
385 total_refs);
386 /*
387 * we can only tolerate ENOENT,otherwise,we should catch error
388 * and return directly.
389 */
390 if (err == -ENOENT) {
391 continue;
392 } else if (err) {
393 ret = err;
394 goto out;
395 }
396
397 /* we put the first parent into the ref at hand */
398 ULIST_ITER_INIT(&uiter);
399 node = ulist_next(parents, &uiter);
400 ref->parent = node ? node->val : 0;
401 ref->inode_list = node ?
402 (struct extent_inode_elem *)(uintptr_t)node->aux : NULL;
403
404 /* additional parents require new refs being added here */
405 while ((node = ulist_next(parents, &uiter))) {
406 new_ref = kmalloc(sizeof(*new_ref), GFP_NOFS);
407 if (!new_ref) {
408 ret = -ENOMEM;
409 goto out;
410 }
411 memcpy(new_ref, ref, sizeof(*ref));
412 new_ref->parent = node->val;
413 new_ref->inode_list = (struct extent_inode_elem *)
414 (uintptr_t)node->aux;
415 list_add_tail(&new_ref->list, &prefstate->pending);
416 }
417 ulist_reinit(parents);
418 }
419 out:
420 ulist_free(parents);
421 return ret;
422 }
423
424 static inline int ref_for_same_block(struct __prelim_ref *ref1,
425 struct __prelim_ref *ref2)
426 {
427 if (ref1->level != ref2->level)
428 return 0;
429 if (ref1->root_id != ref2->root_id)
430 return 0;
431 if (ref1->key_for_search.type != ref2->key_for_search.type)
432 return 0;
433 if (ref1->key_for_search.objectid != ref2->key_for_search.objectid)
434 return 0;
435 if (ref1->key_for_search.offset != ref2->key_for_search.offset)
436 return 0;
437 if (ref1->parent != ref2->parent)
438 return 0;
439
440 return 1;
441 }
442
443 /*
444 * read tree blocks and add keys where required.
445 */
446 static int __add_missing_keys(struct btrfs_fs_info *fs_info,
447 struct pref_state *prefstate)
448 {
449 struct extent_buffer *eb;
450
451 while (!list_empty(&prefstate->pending_missing_keys)) {
452 struct __prelim_ref *ref;
453
454 ref = list_first_pref(&prefstate->pending_missing_keys);
455
456 ASSERT(ref->root_id);
457 ASSERT(!ref->parent);
458 ASSERT(!ref->key_for_search.type);
459 BUG_ON(!ref->wanted_disk_byte);
460 eb = read_tree_block(fs_info, ref->wanted_disk_byte, 0);
461 if (!extent_buffer_uptodate(eb)) {
462 free_extent_buffer(eb);
463 return -EIO;
464 }
465 if (btrfs_header_level(eb) == 0)
466 btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
467 else
468 btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
469 free_extent_buffer(eb);
470 list_move(&ref->list, &prefstate->pending);
471 }
472 return 0;
473 }
474
475 /*
476 * merge two lists of backrefs and adjust counts accordingly
477 *
478 * mode = 1: merge identical keys, if key is set
479 * FIXME: if we add more keys in __add_prelim_ref, we can merge more here.
480 * additionally, we could even add a key range for the blocks we
481 * looked into to merge even more (-> replace unresolved refs by those
482 * having a parent).
483 * mode = 2: merge identical parents
484 */
485 static void __merge_refs(struct pref_state *prefstate, int mode)
486 {
487 struct list_head *head = &prefstate->pending;
488 struct list_head *pos1;
489
490 list_for_each(pos1, head) {
491 struct list_head *n2;
492 struct list_head *pos2;
493 struct __prelim_ref *ref1;
494
495 ref1 = list_entry(pos1, struct __prelim_ref, list);
496
497 for (pos2 = pos1->next, n2 = pos2->next; pos2 != head;
498 pos2 = n2, n2 = pos2->next) {
499 struct __prelim_ref *ref2;
500 struct extent_inode_elem *eie;
501
502 ref2 = list_entry(pos2, struct __prelim_ref, list);
503
504 if (mode == 1) {
505 if (!ref_for_same_block(ref1, ref2))
506 continue;
507 } else {
508 /*
509 * Parent == 0 means that the ref is tree block
510 * backref or its parent is unresolved.
511 */
512 if (!ref1->parent || !ref2->parent)
513 continue;
514 if (ref1->parent != ref2->parent)
515 continue;
516 }
517
518 eie = ref1->inode_list;
519 while (eie && eie->next)
520 eie = eie->next;
521 if (eie)
522 eie->next = ref2->inode_list;
523 else
524 ref1->inode_list = ref2->inode_list;
525 ref1->count += ref2->count;
526
527 list_del(&ref2->list);
528 kfree(ref2);
529 }
530
531 }
532 }
533
534 /*
535 * add all inline backrefs for bytenr to the list
536 */
537 static int __add_inline_refs(struct btrfs_fs_info *fs_info,
538 struct pref_state *prefstate,
539 struct btrfs_path *path, u64 bytenr,
540 int *info_level, u64 *total_refs)
541 {
542 int ret = 0;
543 int slot;
544 struct extent_buffer *leaf;
545 struct btrfs_key key;
546 struct btrfs_key found_key;
547 unsigned long ptr;
548 unsigned long end;
549 struct btrfs_extent_item *ei;
550 u64 flags;
551 u64 item_size;
552 /*
553 * enumerate all inline refs
554 */
555 leaf = path->nodes[0];
556 slot = path->slots[0];
557
558 item_size = btrfs_item_size_nr(leaf, slot);
559 BUG_ON(item_size < sizeof(*ei));
560
561 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
562 flags = btrfs_extent_flags(leaf, ei);
563 *total_refs += btrfs_extent_refs(leaf, ei);
564 btrfs_item_key_to_cpu(leaf, &found_key, slot);
565
566 ptr = (unsigned long)(ei + 1);
567 end = (unsigned long)ei + item_size;
568
569 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
570 flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
571 struct btrfs_tree_block_info *info;
572
573 info = (struct btrfs_tree_block_info *)ptr;
574 *info_level = btrfs_tree_block_level(leaf, info);
575 ptr += sizeof(struct btrfs_tree_block_info);
576 BUG_ON(ptr > end);
577 } else if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
578 *info_level = found_key.offset;
579 } else {
580 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
581 }
582
583 while (ptr < end) {
584 struct btrfs_extent_inline_ref *iref;
585 u64 offset;
586 int type;
587
588 iref = (struct btrfs_extent_inline_ref *)ptr;
589 type = btrfs_extent_inline_ref_type(leaf, iref);
590 offset = btrfs_extent_inline_ref_offset(leaf, iref);
591
592 switch (type) {
593 case BTRFS_SHARED_BLOCK_REF_KEY:
594 ret = __add_prelim_ref(prefstate, 0, NULL,
595 *info_level + 1, offset,
596 bytenr, 1, GFP_NOFS);
597 break;
598 case BTRFS_SHARED_DATA_REF_KEY: {
599 struct btrfs_shared_data_ref *sdref;
600 int count;
601
602 sdref = (struct btrfs_shared_data_ref *)(iref + 1);
603 count = btrfs_shared_data_ref_count(leaf, sdref);
604 ret = __add_prelim_ref(prefstate, 0, NULL, 0, offset,
605 bytenr, count, GFP_NOFS);
606 break;
607 }
608 case BTRFS_TREE_BLOCK_REF_KEY:
609 ret = __add_prelim_ref(prefstate, offset, NULL,
610 *info_level + 1, 0,
611 bytenr, 1, GFP_NOFS);
612 break;
613 case BTRFS_EXTENT_DATA_REF_KEY: {
614 struct btrfs_extent_data_ref *dref;
615 int count;
616 u64 root;
617
618 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
619 count = btrfs_extent_data_ref_count(leaf, dref);
620 key.objectid = btrfs_extent_data_ref_objectid(leaf,
621 dref);
622 key.type = BTRFS_EXTENT_DATA_KEY;
623 key.offset = btrfs_extent_data_ref_offset(leaf, dref);
624 root = btrfs_extent_data_ref_root(leaf, dref);
625 ret = __add_prelim_ref(prefstate, root, &key, 0, 0,
626 bytenr, count, GFP_NOFS);
627 break;
628 }
629 default:
630 WARN_ON(1);
631 }
632 if (ret)
633 return ret;
634 ptr += btrfs_extent_inline_ref_size(type);
635 }
636
637 return 0;
638 }
639
640 /*
641 * add all non-inline backrefs for bytenr to the list
642 */
643 static int __add_keyed_refs(struct btrfs_fs_info *fs_info,
644 struct pref_state *prefstate,
645 struct btrfs_path *path, u64 bytenr,
646 int info_level)
647 {
648 struct btrfs_root *extent_root = fs_info->extent_root;
649 int ret;
650 int slot;
651 struct extent_buffer *leaf;
652 struct btrfs_key key;
653
654 while (1) {
655 ret = btrfs_next_item(extent_root, path);
656 if (ret < 0)
657 break;
658 if (ret) {
659 ret = 0;
660 break;
661 }
662
663 slot = path->slots[0];
664 leaf = path->nodes[0];
665 btrfs_item_key_to_cpu(leaf, &key, slot);
666
667 if (key.objectid != bytenr)
668 break;
669 if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
670 continue;
671 if (key.type > BTRFS_SHARED_DATA_REF_KEY)
672 break;
673
674 switch (key.type) {
675 case BTRFS_SHARED_BLOCK_REF_KEY:
676 ret = __add_prelim_ref(prefstate, 0, NULL,
677 info_level + 1, key.offset,
678 bytenr, 1, GFP_NOFS);
679 break;
680 case BTRFS_SHARED_DATA_REF_KEY: {
681 struct btrfs_shared_data_ref *sdref;
682 int count;
683
684 sdref = btrfs_item_ptr(leaf, slot,
685 struct btrfs_shared_data_ref);
686 count = btrfs_shared_data_ref_count(leaf, sdref);
687 ret = __add_prelim_ref(prefstate, 0, NULL, 0, key.offset,
688 bytenr, count, GFP_NOFS);
689 break;
690 }
691 case BTRFS_TREE_BLOCK_REF_KEY:
692 ret = __add_prelim_ref(prefstate, key.offset, NULL,
693 info_level + 1, 0,
694 bytenr, 1, GFP_NOFS);
695 break;
696 case BTRFS_EXTENT_DATA_REF_KEY: {
697 struct btrfs_extent_data_ref *dref;
698 int count;
699 u64 root;
700
701 dref = btrfs_item_ptr(leaf, slot,
702 struct btrfs_extent_data_ref);
703 count = btrfs_extent_data_ref_count(leaf, dref);
704 key.objectid = btrfs_extent_data_ref_objectid(leaf,
705 dref);
706 key.type = BTRFS_EXTENT_DATA_KEY;
707 key.offset = btrfs_extent_data_ref_offset(leaf, dref);
708 root = btrfs_extent_data_ref_root(leaf, dref);
709 ret = __add_prelim_ref(prefstate, root, &key, 0, 0,
710 bytenr, count, GFP_NOFS);
711 break;
712 }
713 default:
714 WARN_ON(1);
715 }
716 if (ret)
717 return ret;
718
719 }
720
721 return ret;
722 }
723
724 /*
725 * this adds all existing backrefs (inline backrefs, backrefs for the given
726 * bytenr to the refs list, merges duplicates and resolves indirect refs to
727 * their parent bytenr.
728 * When roots are found, they're added to the roots list
729 *
730 * FIXME some caching might speed things up
731 */
732 static int find_parent_nodes(struct btrfs_trans_handle *trans,
733 struct btrfs_fs_info *fs_info, u64 bytenr,
734 u64 time_seq, struct ulist *refs,
735 struct ulist *roots, const u64 *extent_item_pos)
736 {
737 struct btrfs_key key;
738 struct btrfs_path *path;
739 int info_level = 0;
740 int ret;
741 struct pref_state prefstate;
742 struct __prelim_ref *ref;
743 struct extent_inode_elem *eie = NULL;
744 u64 total_refs = 0;
745
746 init_pref_state(&prefstate);
747
748 key.objectid = bytenr;
749 key.offset = (u64)-1;
750 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
751 key.type = BTRFS_METADATA_ITEM_KEY;
752 else
753 key.type = BTRFS_EXTENT_ITEM_KEY;
754
755 path = btrfs_alloc_path();
756 if (!path)
757 return -ENOMEM;
758
759 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
760 if (ret < 0)
761 goto out;
762 BUG_ON(ret == 0);
763
764 if (path->slots[0]) {
765 struct extent_buffer *leaf;
766 int slot;
767
768 path->slots[0]--;
769 leaf = path->nodes[0];
770 slot = path->slots[0];
771 btrfs_item_key_to_cpu(leaf, &key, slot);
772 if (key.objectid == bytenr &&
773 (key.type == BTRFS_EXTENT_ITEM_KEY ||
774 key.type == BTRFS_METADATA_ITEM_KEY)) {
775 ret = __add_inline_refs(fs_info, &prefstate, path,
776 bytenr, &info_level,
777 &total_refs);
778 if (ret)
779 goto out;
780 ret = __add_keyed_refs(fs_info, &prefstate, path,
781 bytenr, info_level);
782 if (ret)
783 goto out;
784 }
785 }
786 btrfs_release_path(path);
787
788 ret = __add_missing_keys(fs_info, &prefstate);
789 if (ret)
790 goto out;
791
792 __merge_refs(&prefstate, 1);
793
794 ret = __resolve_indirect_refs(fs_info, &prefstate, path, time_seq,
795 extent_item_pos, total_refs);
796 if (ret)
797 goto out;
798
799 __merge_refs(&prefstate, 2);
800
801 BUG_ON(!list_empty(&prefstate.pending_missing_keys));
802 BUG_ON(!list_empty(&prefstate.pending_indirect_refs));
803
804 while (!list_empty(&prefstate.pending)) {
805 ref = list_first_pref(&prefstate.pending);
806 WARN_ON(ref->count < 0);
807 if (roots && ref->count && ref->root_id && ref->parent == 0) {
808 /* no parent == root of tree */
809 ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS);
810 if (ret < 0)
811 goto out;
812 }
813 if (ref->count && ref->parent) {
814 if (extent_item_pos && !ref->inode_list &&
815 ref->level == 0) {
816 struct extent_buffer *eb;
817
818 eb = read_tree_block(fs_info, ref->parent, 0);
819 if (!extent_buffer_uptodate(eb)) {
820 free_extent_buffer(eb);
821 ret = -EIO;
822 goto out;
823 }
824 ret = find_extent_in_eb(eb, bytenr,
825 *extent_item_pos, &eie);
826 free_extent_buffer(eb);
827 if (ret < 0)
828 goto out;
829 ref->inode_list = eie;
830 }
831 ret = ulist_add_merge_ptr(refs, ref->parent,
832 ref->inode_list,
833 (void **)&eie, GFP_NOFS);
834 if (ret < 0)
835 goto out;
836 if (!ret && extent_item_pos) {
837 /*
838 * we've recorded that parent, so we must extend
839 * its inode list here
840 */
841 BUG_ON(!eie);
842 while (eie->next)
843 eie = eie->next;
844 eie->next = ref->inode_list;
845 }
846 eie = NULL;
847 }
848 list_del(&ref->list);
849 kfree(ref);
850 }
851
852 out:
853 btrfs_free_path(path);
854 while (!list_empty(&prefstate.pending)) {
855 ref = list_first_pref(&prefstate.pending);
856 list_del(&ref->list);
857 kfree(ref);
858 }
859 if (ret < 0)
860 free_inode_elem_list(eie);
861 return ret;
862 }
863
864 static void free_leaf_list(struct ulist *blocks)
865 {
866 struct ulist_node *node = NULL;
867 struct extent_inode_elem *eie;
868 struct ulist_iterator uiter;
869
870 ULIST_ITER_INIT(&uiter);
871 while ((node = ulist_next(blocks, &uiter))) {
872 if (!node->aux)
873 continue;
874 eie = (struct extent_inode_elem *)(uintptr_t)node->aux;
875 free_inode_elem_list(eie);
876 node->aux = 0;
877 }
878
879 ulist_free(blocks);
880 }
881
882 /*
883 * Finds all leafs with a reference to the specified combination of bytenr and
884 * offset. key_list_head will point to a list of corresponding keys (caller must
885 * free each list element). The leafs will be stored in the leafs ulist, which
886 * must be freed with ulist_free.
887 *
888 * returns 0 on success, <0 on error
889 */
890 static int btrfs_find_all_leafs(struct btrfs_trans_handle *trans,
891 struct btrfs_fs_info *fs_info, u64 bytenr,
892 u64 time_seq, struct ulist **leafs,
893 const u64 *extent_item_pos)
894 {
895 int ret;
896
897 *leafs = ulist_alloc(GFP_NOFS);
898 if (!*leafs)
899 return -ENOMEM;
900
901 ret = find_parent_nodes(trans, fs_info, bytenr,
902 time_seq, *leafs, NULL, extent_item_pos);
903 if (ret < 0 && ret != -ENOENT) {
904 free_leaf_list(*leafs);
905 return ret;
906 }
907
908 return 0;
909 }
910
911 /*
912 * walk all backrefs for a given extent to find all roots that reference this
913 * extent. Walking a backref means finding all extents that reference this
914 * extent and in turn walk the backrefs of those, too. Naturally this is a
915 * recursive process, but here it is implemented in an iterative fashion: We
916 * find all referencing extents for the extent in question and put them on a
917 * list. In turn, we find all referencing extents for those, further appending
918 * to the list. The way we iterate the list allows adding more elements after
919 * the current while iterating. The process stops when we reach the end of the
920 * list. Found roots are added to the roots list.
921 *
922 * returns 0 on success, < 0 on error.
923 */
924 static int __btrfs_find_all_roots(struct btrfs_trans_handle *trans,
925 struct btrfs_fs_info *fs_info, u64 bytenr,
926 u64 time_seq, struct ulist **roots)
927 {
928 struct ulist *tmp;
929 struct ulist_node *node = NULL;
930 struct ulist_iterator uiter;
931 int ret;
932
933 tmp = ulist_alloc(GFP_NOFS);
934 if (!tmp)
935 return -ENOMEM;
936 *roots = ulist_alloc(GFP_NOFS);
937 if (!*roots) {
938 ulist_free(tmp);
939 return -ENOMEM;
940 }
941
942 ULIST_ITER_INIT(&uiter);
943 while (1) {
944 ret = find_parent_nodes(trans, fs_info, bytenr,
945 time_seq, tmp, *roots, NULL);
946 if (ret < 0 && ret != -ENOENT) {
947 ulist_free(tmp);
948 ulist_free(*roots);
949 return ret;
950 }
951 node = ulist_next(tmp, &uiter);
952 if (!node)
953 break;
954 bytenr = node->val;
955 cond_resched();
956 }
957
958 ulist_free(tmp);
959 return 0;
960 }
961
962 int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
963 struct btrfs_fs_info *fs_info, u64 bytenr,
964 u64 time_seq, struct ulist **roots)
965 {
966 return __btrfs_find_all_roots(trans, fs_info, bytenr, time_seq, roots);
967 }
968
969 /*
970 * this makes the path point to (inum INODE_ITEM ioff)
971 */
972 int inode_item_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
973 struct btrfs_path *path)
974 {
975 struct btrfs_key key;
976 return btrfs_find_item(fs_root, path, inum, ioff,
977 BTRFS_INODE_ITEM_KEY, &key);
978 }
979
980 static int inode_ref_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
981 struct btrfs_path *path,
982 struct btrfs_key *found_key)
983 {
984 return btrfs_find_item(fs_root, path, inum, ioff,
985 BTRFS_INODE_REF_KEY, found_key);
986 }
987
988 int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,
989 u64 start_off, struct btrfs_path *path,
990 struct btrfs_inode_extref **ret_extref,
991 u64 *found_off)
992 {
993 int ret, slot;
994 struct btrfs_key key;
995 struct btrfs_key found_key;
996 struct btrfs_inode_extref *extref;
997 struct extent_buffer *leaf;
998 unsigned long ptr;
999
1000 key.objectid = inode_objectid;
1001 key.type = BTRFS_INODE_EXTREF_KEY;
1002 key.offset = start_off;
1003
1004 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1005 if (ret < 0)
1006 return ret;
1007
1008 while (1) {
1009 leaf = path->nodes[0];
1010 slot = path->slots[0];
1011 if (slot >= btrfs_header_nritems(leaf)) {
1012 /*
1013 * If the item at offset is not found,
1014 * btrfs_search_slot will point us to the slot
1015 * where it should be inserted. In our case
1016 * that will be the slot directly before the
1017 * next INODE_REF_KEY_V2 item. In the case
1018 * that we're pointing to the last slot in a
1019 * leaf, we must move one leaf over.
1020 */
1021 ret = btrfs_next_leaf(root, path);
1022 if (ret) {
1023 if (ret >= 1)
1024 ret = -ENOENT;
1025 break;
1026 }
1027 continue;
1028 }
1029
1030 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1031
1032 /*
1033 * Check that we're still looking at an extended ref key for
1034 * this particular objectid. If we have different
1035 * objectid or type then there are no more to be found
1036 * in the tree and we can exit.
1037 */
1038 ret = -ENOENT;
1039 if (found_key.objectid != inode_objectid)
1040 break;
1041 if (found_key.type != BTRFS_INODE_EXTREF_KEY)
1042 break;
1043
1044 ret = 0;
1045 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1046 extref = (struct btrfs_inode_extref *)ptr;
1047 *ret_extref = extref;
1048 if (found_off)
1049 *found_off = found_key.offset;
1050 break;
1051 }
1052
1053 return ret;
1054 }
1055
1056 /*
1057 * this iterates to turn a name (from iref/extref) into a full filesystem path.
1058 * Elements of the path are separated by '/' and the path is guaranteed to be
1059 * 0-terminated. the path is only given within the current file system.
1060 * Therefore, it never starts with a '/'. the caller is responsible to provide
1061 * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
1062 * the start point of the resulting string is returned. this pointer is within
1063 * dest, normally.
1064 * in case the path buffer would overflow, the pointer is decremented further
1065 * as if output was written to the buffer, though no more output is actually
1066 * generated. that way, the caller can determine how much space would be
1067 * required for the path to fit into the buffer. in that case, the returned
1068 * value will be smaller than dest. callers must check this!
1069 */
1070 char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
1071 u32 name_len, unsigned long name_off,
1072 struct extent_buffer *eb_in, u64 parent,
1073 char *dest, u32 size)
1074 {
1075 int slot;
1076 u64 next_inum;
1077 int ret;
1078 s64 bytes_left = ((s64)size) - 1;
1079 struct extent_buffer *eb = eb_in;
1080 struct btrfs_key found_key;
1081 struct btrfs_inode_ref *iref;
1082
1083 if (bytes_left >= 0)
1084 dest[bytes_left] = '\0';
1085
1086 while (1) {
1087 bytes_left -= name_len;
1088 if (bytes_left >= 0)
1089 read_extent_buffer(eb, dest + bytes_left,
1090 name_off, name_len);
1091 if (eb != eb_in)
1092 free_extent_buffer(eb);
1093 ret = inode_ref_info(parent, 0, fs_root, path, &found_key);
1094 if (ret > 0)
1095 ret = -ENOENT;
1096 if (ret)
1097 break;
1098
1099 next_inum = found_key.offset;
1100
1101 /* regular exit ahead */
1102 if (parent == next_inum)
1103 break;
1104
1105 slot = path->slots[0];
1106 eb = path->nodes[0];
1107 /* make sure we can use eb after releasing the path */
1108 if (eb != eb_in)
1109 eb->refs++;
1110 btrfs_release_path(path);
1111 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1112
1113 name_len = btrfs_inode_ref_name_len(eb, iref);
1114 name_off = (unsigned long)(iref + 1);
1115
1116 parent = next_inum;
1117 --bytes_left;
1118 if (bytes_left >= 0)
1119 dest[bytes_left] = '/';
1120 }
1121
1122 btrfs_release_path(path);
1123
1124 if (ret)
1125 return ERR_PTR(ret);
1126
1127 return dest + bytes_left;
1128 }
1129
1130 /*
1131 * this makes the path point to (logical EXTENT_ITEM *)
1132 * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
1133 * tree blocks and <0 on error.
1134 */
1135 int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
1136 struct btrfs_path *path, struct btrfs_key *found_key,
1137 u64 *flags_ret)
1138 {
1139 int ret;
1140 u64 flags;
1141 u64 size = 0;
1142 u32 item_size;
1143 struct extent_buffer *eb;
1144 struct btrfs_extent_item *ei;
1145 struct btrfs_key key;
1146
1147 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1148 key.type = BTRFS_METADATA_ITEM_KEY;
1149 else
1150 key.type = BTRFS_EXTENT_ITEM_KEY;
1151 key.objectid = logical;
1152 key.offset = (u64)-1;
1153
1154 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
1155 if (ret < 0)
1156 return ret;
1157
1158 ret = btrfs_previous_extent_item(fs_info->extent_root, path, 0);
1159 if (ret) {
1160 if (ret > 0)
1161 ret = -ENOENT;
1162 return ret;
1163 }
1164 btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]);
1165 if (found_key->type == BTRFS_METADATA_ITEM_KEY)
1166 size = fs_info->nodesize;
1167 else if (found_key->type == BTRFS_EXTENT_ITEM_KEY)
1168 size = found_key->offset;
1169
1170 if (found_key->objectid > logical ||
1171 found_key->objectid + size <= logical) {
1172 pr_debug("logical %llu is not within any extent\n", logical);
1173 return -ENOENT;
1174 }
1175
1176 eb = path->nodes[0];
1177 item_size = btrfs_item_size_nr(eb, path->slots[0]);
1178 BUG_ON(item_size < sizeof(*ei));
1179
1180 ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
1181 flags = btrfs_extent_flags(eb, ei);
1182
1183 pr_debug("logical %llu is at position %llu within the extent (%llu "
1184 "EXTENT_ITEM %llu) flags %#llx size %u\n",
1185 logical, logical - found_key->objectid, found_key->objectid,
1186 found_key->offset, flags, item_size);
1187
1188 if (flags_ret) {
1189 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1190 *flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK;
1191 else if (flags & BTRFS_EXTENT_FLAG_DATA)
1192 *flags_ret = BTRFS_EXTENT_FLAG_DATA;
1193 else
1194 BUG_ON(1);
1195 return 0;
1196 } else {
1197 WARN_ON(1);
1198 return -EIO;
1199 }
1200 }
1201
1202 /*
1203 * helper function to iterate extent inline refs. ptr must point to a 0 value
1204 * for the first call and may be modified. it is used to track state.
1205 * if more refs exist, 0 is returned and the next call to
1206 * __get_extent_inline_ref must pass the modified ptr parameter to get the
1207 * next ref. after the last ref was processed, 1 is returned.
1208 * returns <0 on error
1209 */
1210 static int __get_extent_inline_ref(unsigned long *ptr, struct extent_buffer *eb,
1211 struct btrfs_key *key,
1212 struct btrfs_extent_item *ei, u32 item_size,
1213 struct btrfs_extent_inline_ref **out_eiref,
1214 int *out_type)
1215 {
1216 unsigned long end;
1217 u64 flags;
1218 struct btrfs_tree_block_info *info;
1219
1220 if (!*ptr) {
1221 /* first call */
1222 flags = btrfs_extent_flags(eb, ei);
1223 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1224 if (key->type == BTRFS_METADATA_ITEM_KEY) {
1225 /* a skinny metadata extent */
1226 *out_eiref =
1227 (struct btrfs_extent_inline_ref *)(ei + 1);
1228 } else {
1229 WARN_ON(key->type != BTRFS_EXTENT_ITEM_KEY);
1230 info = (struct btrfs_tree_block_info *)(ei + 1);
1231 *out_eiref =
1232 (struct btrfs_extent_inline_ref *)(info + 1);
1233 }
1234 } else {
1235 *out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1);
1236 }
1237 *ptr = (unsigned long)*out_eiref;
1238 if ((unsigned long)(*ptr) >= (unsigned long)ei + item_size)
1239 return -ENOENT;
1240 }
1241
1242 end = (unsigned long)ei + item_size;
1243 *out_eiref = (struct btrfs_extent_inline_ref *)(*ptr);
1244 *out_type = btrfs_extent_inline_ref_type(eb, *out_eiref);
1245
1246 *ptr += btrfs_extent_inline_ref_size(*out_type);
1247 WARN_ON(*ptr > end);
1248 if (*ptr == end)
1249 return 1; /* last */
1250
1251 return 0;
1252 }
1253
1254 /*
1255 * reads the tree block backref for an extent. tree level and root are returned
1256 * through out_level and out_root. ptr must point to a 0 value for the first
1257 * call and may be modified (see __get_extent_inline_ref comment).
1258 * returns 0 if data was provided, 1 if there was no more data to provide or
1259 * <0 on error.
1260 */
1261 int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
1262 struct btrfs_key *key, struct btrfs_extent_item *ei,
1263 u32 item_size, u64 *out_root, u8 *out_level)
1264 {
1265 int ret;
1266 int type;
1267 struct btrfs_tree_block_info *info;
1268 struct btrfs_extent_inline_ref *eiref;
1269
1270 if (*ptr == (unsigned long)-1)
1271 return 1;
1272
1273 while (1) {
1274 ret = __get_extent_inline_ref(ptr, eb, key, ei, item_size,
1275 &eiref, &type);
1276 if (ret < 0)
1277 return ret;
1278
1279 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1280 type == BTRFS_SHARED_BLOCK_REF_KEY)
1281 break;
1282
1283 if (ret == 1)
1284 return 1;
1285 }
1286
1287 /* we can treat both ref types equally here */
1288 info = (struct btrfs_tree_block_info *)(ei + 1);
1289 *out_root = btrfs_extent_inline_ref_offset(eb, eiref);
1290 *out_level = btrfs_tree_block_level(eb, info);
1291
1292 if (ret == 1)
1293 *ptr = (unsigned long)-1;
1294
1295 return 0;
1296 }
1297
1298 static int iterate_leaf_refs(struct extent_inode_elem *inode_list,
1299 u64 root, u64 extent_item_objectid,
1300 iterate_extent_inodes_t *iterate, void *ctx)
1301 {
1302 struct extent_inode_elem *eie;
1303 int ret = 0;
1304
1305 for (eie = inode_list; eie; eie = eie->next) {
1306 pr_debug("ref for %llu resolved, key (%llu EXTEND_DATA %llu), "
1307 "root %llu\n", extent_item_objectid,
1308 eie->inum, eie->offset, root);
1309 ret = iterate(eie->inum, eie->offset, root, ctx);
1310 if (ret) {
1311 pr_debug("stopping iteration for %llu due to ret=%d\n",
1312 extent_item_objectid, ret);
1313 break;
1314 }
1315 }
1316
1317 return ret;
1318 }
1319
1320 /*
1321 * calls iterate() for every inode that references the extent identified by
1322 * the given parameters.
1323 * when the iterator function returns a non-zero value, iteration stops.
1324 */
1325 int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
1326 u64 extent_item_objectid, u64 extent_item_pos,
1327 int search_commit_root,
1328 iterate_extent_inodes_t *iterate, void *ctx)
1329 {
1330 int ret;
1331 struct btrfs_trans_handle *trans = NULL;
1332 struct ulist *refs = NULL;
1333 struct ulist *roots = NULL;
1334 struct ulist_node *ref_node = NULL;
1335 struct ulist_node *root_node = NULL;
1336 struct ulist_iterator ref_uiter;
1337 struct ulist_iterator root_uiter;
1338
1339 pr_debug("resolving all inodes for extent %llu\n",
1340 extent_item_objectid);
1341
1342 ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid,
1343 0, &refs, &extent_item_pos);
1344 if (ret)
1345 goto out;
1346
1347 ULIST_ITER_INIT(&ref_uiter);
1348 while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) {
1349 ret = __btrfs_find_all_roots(trans, fs_info, ref_node->val,
1350 0, &roots);
1351 if (ret)
1352 break;
1353 ULIST_ITER_INIT(&root_uiter);
1354 while (!ret && (root_node = ulist_next(roots, &root_uiter))) {
1355 pr_debug("root %llu references leaf %llu, data list "
1356 "%#llx\n", root_node->val, ref_node->val,
1357 ref_node->aux);
1358 ret = iterate_leaf_refs((struct extent_inode_elem *)
1359 (uintptr_t)ref_node->aux,
1360 root_node->val,
1361 extent_item_objectid,
1362 iterate, ctx);
1363 }
1364 ulist_free(roots);
1365 }
1366
1367 free_leaf_list(refs);
1368 out:
1369 return ret;
1370 }
1371
1372 int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
1373 struct btrfs_path *path,
1374 iterate_extent_inodes_t *iterate, void *ctx)
1375 {
1376 int ret;
1377 u64 extent_item_pos;
1378 u64 flags = 0;
1379 struct btrfs_key found_key;
1380 int search_commit_root = 0;
1381
1382 ret = extent_from_logical(fs_info, logical, path, &found_key, &flags);
1383 btrfs_release_path(path);
1384 if (ret < 0)
1385 return ret;
1386 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1387 return -EINVAL;
1388
1389 extent_item_pos = logical - found_key.objectid;
1390 ret = iterate_extent_inodes(fs_info, found_key.objectid,
1391 extent_item_pos, search_commit_root,
1392 iterate, ctx);
1393
1394 return ret;
1395 }
1396
1397 typedef int (iterate_irefs_t)(u64 parent, u32 name_len, unsigned long name_off,
1398 struct extent_buffer *eb, void *ctx);
1399
1400 static int iterate_inode_refs(u64 inum, struct btrfs_root *fs_root,
1401 struct btrfs_path *path,
1402 iterate_irefs_t *iterate, void *ctx)
1403 {
1404 int ret = 0;
1405 int slot;
1406 u32 cur;
1407 u32 len;
1408 u32 name_len;
1409 u64 parent = 0;
1410 int found = 0;
1411 struct extent_buffer *eb;
1412 struct btrfs_item *item;
1413 struct btrfs_inode_ref *iref;
1414 struct btrfs_key found_key;
1415
1416 while (!ret) {
1417 ret = inode_ref_info(inum, parent ? parent+1 : 0, fs_root, path,
1418 &found_key);
1419 if (ret < 0)
1420 break;
1421 if (ret) {
1422 ret = found ? 0 : -ENOENT;
1423 break;
1424 }
1425 ++found;
1426
1427 parent = found_key.offset;
1428 slot = path->slots[0];
1429 eb = btrfs_clone_extent_buffer(path->nodes[0]);
1430 if (!eb) {
1431 ret = -ENOMEM;
1432 break;
1433 }
1434 extent_buffer_get(eb);
1435 btrfs_release_path(path);
1436
1437 item = btrfs_item_nr(slot);
1438 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1439
1440 for (cur = 0; cur < btrfs_item_size(eb, item); cur += len) {
1441 name_len = btrfs_inode_ref_name_len(eb, iref);
1442 /* path must be released before calling iterate()! */
1443 pr_debug("following ref at offset %u for inode %llu in "
1444 "tree %llu\n", cur, found_key.objectid,
1445 fs_root->objectid);
1446 ret = iterate(parent, name_len,
1447 (unsigned long)(iref + 1), eb, ctx);
1448 if (ret)
1449 break;
1450 len = sizeof(*iref) + name_len;
1451 iref = (struct btrfs_inode_ref *)((char *)iref + len);
1452 }
1453 free_extent_buffer(eb);
1454 }
1455
1456 btrfs_release_path(path);
1457
1458 return ret;
1459 }
1460
1461 static int iterate_inode_extrefs(u64 inum, struct btrfs_root *fs_root,
1462 struct btrfs_path *path,
1463 iterate_irefs_t *iterate, void *ctx)
1464 {
1465 int ret;
1466 int slot;
1467 u64 offset = 0;
1468 u64 parent;
1469 int found = 0;
1470 struct extent_buffer *eb;
1471 struct btrfs_inode_extref *extref;
1472 struct extent_buffer *leaf;
1473 u32 item_size;
1474 u32 cur_offset;
1475 unsigned long ptr;
1476
1477 while (1) {
1478 ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref,
1479 &offset);
1480 if (ret < 0)
1481 break;
1482 if (ret) {
1483 ret = found ? 0 : -ENOENT;
1484 break;
1485 }
1486 ++found;
1487
1488 slot = path->slots[0];
1489 eb = btrfs_clone_extent_buffer(path->nodes[0]);
1490 if (!eb) {
1491 ret = -ENOMEM;
1492 break;
1493 }
1494 extent_buffer_get(eb);
1495
1496 btrfs_release_path(path);
1497
1498 leaf = path->nodes[0];
1499 item_size = btrfs_item_size_nr(leaf, slot);
1500 ptr = btrfs_item_ptr_offset(leaf, slot);
1501 cur_offset = 0;
1502
1503 while (cur_offset < item_size) {
1504 u32 name_len;
1505
1506 extref = (struct btrfs_inode_extref *)(ptr + cur_offset);
1507 parent = btrfs_inode_extref_parent(eb, extref);
1508 name_len = btrfs_inode_extref_name_len(eb, extref);
1509 ret = iterate(parent, name_len,
1510 (unsigned long)&extref->name, eb, ctx);
1511 if (ret)
1512 break;
1513
1514 cur_offset += btrfs_inode_extref_name_len(leaf, extref);
1515 cur_offset += sizeof(*extref);
1516 }
1517 free_extent_buffer(eb);
1518
1519 offset++;
1520 }
1521
1522 btrfs_release_path(path);
1523
1524 return ret;
1525 }
1526
1527 static int iterate_irefs(u64 inum, struct btrfs_root *fs_root,
1528 struct btrfs_path *path, iterate_irefs_t *iterate,
1529 void *ctx)
1530 {
1531 int ret;
1532 int found_refs = 0;
1533
1534 ret = iterate_inode_refs(inum, fs_root, path, iterate, ctx);
1535 if (!ret)
1536 ++found_refs;
1537 else if (ret != -ENOENT)
1538 return ret;
1539
1540 ret = iterate_inode_extrefs(inum, fs_root, path, iterate, ctx);
1541 if (ret == -ENOENT && found_refs)
1542 return 0;
1543
1544 return ret;
1545 }
1546
1547 /*
1548 * returns 0 if the path could be dumped (probably truncated)
1549 * returns <0 in case of an error
1550 */
1551 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
1552 struct extent_buffer *eb, void *ctx)
1553 {
1554 struct inode_fs_paths *ipath = ctx;
1555 char *fspath;
1556 char *fspath_min;
1557 int i = ipath->fspath->elem_cnt;
1558 const int s_ptr = sizeof(char *);
1559 u32 bytes_left;
1560
1561 bytes_left = ipath->fspath->bytes_left > s_ptr ?
1562 ipath->fspath->bytes_left - s_ptr : 0;
1563
1564 fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
1565 fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len,
1566 name_off, eb, inum, fspath_min, bytes_left);
1567 if (IS_ERR(fspath))
1568 return PTR_ERR(fspath);
1569
1570 if (fspath > fspath_min) {
1571 ipath->fspath->val[i] = (u64)(unsigned long)fspath;
1572 ++ipath->fspath->elem_cnt;
1573 ipath->fspath->bytes_left = fspath - fspath_min;
1574 } else {
1575 ++ipath->fspath->elem_missed;
1576 ipath->fspath->bytes_missing += fspath_min - fspath;
1577 ipath->fspath->bytes_left = 0;
1578 }
1579
1580 return 0;
1581 }
1582
1583 /*
1584 * this dumps all file system paths to the inode into the ipath struct, provided
1585 * is has been created large enough. each path is zero-terminated and accessed
1586 * from ipath->fspath->val[i].
1587 * when it returns, there are ipath->fspath->elem_cnt number of paths available
1588 * in ipath->fspath->val[]. When the allocated space wasn't sufficient, the
1589 * number of missed paths is recorded in ipath->fspath->elem_missed, otherwise,
1590 * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
1591 * have been needed to return all paths.
1592 */
1593 int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
1594 {
1595 return iterate_irefs(inum, ipath->fs_root, ipath->btrfs_path,
1596 inode_to_path, ipath);
1597 }
1598
1599 struct btrfs_data_container *init_data_container(u32 total_bytes)
1600 {
1601 struct btrfs_data_container *data;
1602 size_t alloc_bytes;
1603
1604 alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
1605 data = vmalloc(alloc_bytes);
1606 if (!data)
1607 return ERR_PTR(-ENOMEM);
1608
1609 if (total_bytes >= sizeof(*data)) {
1610 data->bytes_left = total_bytes - sizeof(*data);
1611 data->bytes_missing = 0;
1612 } else {
1613 data->bytes_missing = sizeof(*data) - total_bytes;
1614 data->bytes_left = 0;
1615 }
1616
1617 data->elem_cnt = 0;
1618 data->elem_missed = 0;
1619
1620 return data;
1621 }
1622
1623 /*
1624 * allocates space to return multiple file system paths for an inode.
1625 * total_bytes to allocate are passed, note that space usable for actual path
1626 * information will be total_bytes - sizeof(struct inode_fs_paths).
1627 * the returned pointer must be freed with free_ipath() in the end.
1628 */
1629 struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
1630 struct btrfs_path *path)
1631 {
1632 struct inode_fs_paths *ifp;
1633 struct btrfs_data_container *fspath;
1634
1635 fspath = init_data_container(total_bytes);
1636 if (IS_ERR(fspath))
1637 return (void *)fspath;
1638
1639 ifp = kmalloc(sizeof(*ifp), GFP_NOFS);
1640 if (!ifp) {
1641 kfree(fspath);
1642 return ERR_PTR(-ENOMEM);
1643 }
1644
1645 ifp->btrfs_path = path;
1646 ifp->fspath = fspath;
1647 ifp->fs_root = fs_root;
1648
1649 return ifp;
1650 }
1651
1652 void free_ipath(struct inode_fs_paths *ipath)
1653 {
1654 if (!ipath)
1655 return;
1656 vfree(ipath->fspath);
1657 kfree(ipath);
1658 }
(deprecated) Btrfs progs developments and patch integration