projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Btrfs: kill some BUG_ONs() in the find_parent_nodes()
[deliverable/linux.git] / fs / btrfs / backref.c
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 <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"
27
28 struct extent_inode_elem {
29 u64 inum;
30 u64 offset;
31 struct extent_inode_elem *next;
32 };
33
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)
38 {
39 u64 data_offset;
40 u64 data_len;
41 struct extent_inode_elem *e;
42
43 data_offset = btrfs_file_extent_offset(eb, fi);
44 data_len = btrfs_file_extent_num_bytes(eb, fi);
45
46 if (extent_item_pos < data_offset ||
47 extent_item_pos >= data_offset + data_len)
48 return 1;
49
50 e = kmalloc(sizeof(*e), GFP_NOFS);
51 if (!e)
52 return -ENOMEM;
53
54 e->next = *eie;
55 e->inum = key->objectid;
56 e->offset = key->offset + (extent_item_pos - data_offset);
57 *eie = e;
58
59 return 0;
60 }
61
62 static int find_extent_in_eb(struct extent_buffer *eb, u64 wanted_disk_byte,
63 u64 extent_item_pos,
64 struct extent_inode_elem **eie)
65 {
66 u64 disk_byte;
67 struct btrfs_key key;
68 struct btrfs_file_extent_item *fi;
69 int slot;
70 int nritems;
71 int extent_type;
72 int ret;
73
74 /*
75 * from the shared data ref, we only have the leaf but we need
76 * the key. thus, we must look into all items and see that we
77 * find one (some) with a reference to our extent item.
78 */
79 nritems = btrfs_header_nritems(eb);
80 for (slot = 0; slot < nritems; ++slot) {
81 btrfs_item_key_to_cpu(eb, &key, slot);
82 if (key.type != BTRFS_EXTENT_DATA_KEY)
83 continue;
84 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
85 extent_type = btrfs_file_extent_type(eb, fi);
86 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
87 continue;
88 /* don't skip BTRFS_FILE_EXTENT_PREALLOC, we can handle that */
89 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
90 if (disk_byte != wanted_disk_byte)
91 continue;
92
93 ret = check_extent_in_eb(&key, eb, fi, extent_item_pos, eie);
94 if (ret < 0)
95 return ret;
96 }
97
98 return 0;
99 }
100
101 /*
102 * this structure records all encountered refs on the way up to the root
103 */
104 struct __prelim_ref {
105 struct list_head list;
106 u64 root_id;
107 struct btrfs_key key_for_search;
108 int level;
109 int count;
110 struct extent_inode_elem *inode_list;
111 u64 parent;
112 u64 wanted_disk_byte;
113 };
114
115 /*
116 * the rules for all callers of this function are:
117 * - obtaining the parent is the goal
118 * - if you add a key, you must know that it is a correct key
119 * - if you cannot add the parent or a correct key, then we will look into the
120 * block later to set a correct key
121 *
122 * delayed refs
123 * ============
124 * backref type | shared | indirect | shared | indirect
125 * information | tree | tree | data | data
126 * --------------------+--------+----------+--------+----------
127 * parent logical | y | - | - | -
128 * key to resolve | - | y | y | y
129 * tree block logical | - | - | - | -
130 * root for resolving | y | y | y | y
131 *
132 * - column 1: we've the parent -> done
133 * - column 2, 3, 4: we use the key to find the parent
134 *
135 * on disk refs (inline or keyed)
136 * ==============================
137 * backref type | shared | indirect | shared | indirect
138 * information | tree | tree | data | data
139 * --------------------+--------+----------+--------+----------
140 * parent logical | y | - | y | -
141 * key to resolve | - | - | - | y
142 * tree block logical | y | y | y | y
143 * root for resolving | - | y | y | y
144 *
145 * - column 1, 3: we've the parent -> done
146 * - column 2: we take the first key from the block to find the parent
147 * (see __add_missing_keys)
148 * - column 4: we use the key to find the parent
149 *
150 * additional information that's available but not required to find the parent
151 * block might help in merging entries to gain some speed.
152 */
153
154 static int __add_prelim_ref(struct list_head *head, u64 root_id,
155 struct btrfs_key *key, int level,
156 u64 parent, u64 wanted_disk_byte, int count)
157 {
158 struct __prelim_ref *ref;
159
160 /* in case we're adding delayed refs, we're holding the refs spinlock */
161 ref = kmalloc(sizeof(*ref), GFP_ATOMIC);
162 if (!ref)
163 return -ENOMEM;
164
165 ref->root_id = root_id;
166 if (key)
167 ref->key_for_search = *key;
168 else
169 memset(&ref->key_for_search, 0, sizeof(ref->key_for_search));
170
171 ref->inode_list = NULL;
172 ref->level = level;
173 ref->count = count;
174 ref->parent = parent;
175 ref->wanted_disk_byte = wanted_disk_byte;
176 list_add_tail(&ref->list, head);
177
178 return 0;
179 }
180
181 static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path,
182 struct ulist *parents, int level,
183 struct btrfs_key *key_for_search, u64 time_seq,
184 u64 wanted_disk_byte,
185 const u64 *extent_item_pos)
186 {
187 int ret = 0;
188 int slot;
189 struct extent_buffer *eb;
190 struct btrfs_key key;
191 struct btrfs_file_extent_item *fi;
192 struct extent_inode_elem *eie = NULL;
193 u64 disk_byte;
194
195 if (level != 0) {
196 eb = path->nodes[level];
197 ret = ulist_add(parents, eb->start, 0, GFP_NOFS);
198 if (ret < 0)
199 return ret;
200 return 0;
201 }
202
203 /*
204 * We normally enter this function with the path already pointing to
205 * the first item to check. But sometimes, we may enter it with
206 * slot==nritems. In that case, go to the next leaf before we continue.
207 */
208 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0]))
209 ret = btrfs_next_old_leaf(root, path, time_seq);
210
211 while (!ret) {
212 eb = path->nodes[0];
213 slot = path->slots[0];
214
215 btrfs_item_key_to_cpu(eb, &key, slot);
216
217 if (key.objectid != key_for_search->objectid ||
218 key.type != BTRFS_EXTENT_DATA_KEY)
219 break;
220
221 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
222 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
223
224 if (disk_byte == wanted_disk_byte) {
225 eie = NULL;
226 if (extent_item_pos) {
227 ret = check_extent_in_eb(&key, eb, fi,
228 *extent_item_pos,
229 &eie);
230 if (ret < 0)
231 break;
232 }
233 if (!ret) {
234 ret = ulist_add(parents, eb->start,
235 (uintptr_t)eie, GFP_NOFS);
236 if (ret < 0)
237 break;
238 if (!extent_item_pos) {
239 ret = btrfs_next_old_leaf(root, path,
240 time_seq);
241 continue;
242 }
243 }
244 }
245 ret = btrfs_next_old_item(root, path, time_seq);
246 }
247
248 if (ret > 0)
249 ret = 0;
250 return ret;
251 }
252
253 /*
254 * resolve an indirect backref in the form (root_id, key, level)
255 * to a logical address
256 */
257 static int __resolve_indirect_ref(struct btrfs_fs_info *fs_info,
258 int search_commit_root,
259 u64 time_seq,
260 struct __prelim_ref *ref,
261 struct ulist *parents,
262 const u64 *extent_item_pos)
263 {
264 struct btrfs_path *path;
265 struct btrfs_root *root;
266 struct btrfs_key root_key;
267 struct extent_buffer *eb;
268 int ret = 0;
269 int root_level;
270 int level = ref->level;
271
272 path = btrfs_alloc_path();
273 if (!path)
274 return -ENOMEM;
275 path->search_commit_root = !!search_commit_root;
276
277 root_key.objectid = ref->root_id;
278 root_key.type = BTRFS_ROOT_ITEM_KEY;
279 root_key.offset = (u64)-1;
280 root = btrfs_read_fs_root_no_name(fs_info, &root_key);
281 if (IS_ERR(root)) {
282 ret = PTR_ERR(root);
283 goto out;
284 }
285
286 root_level = btrfs_old_root_level(root, time_seq);
287
288 if (root_level + 1 == level)
289 goto out;
290
291 path->lowest_level = level;
292 ret = btrfs_search_old_slot(root, &ref->key_for_search, path, time_seq);
293 pr_debug("search slot in root %llu (level %d, ref count %d) returned "
294 "%d for key (%llu %u %llu)\n",
295 (unsigned long long)ref->root_id, level, ref->count, ret,
296 (unsigned long long)ref->key_for_search.objectid,
297 ref->key_for_search.type,
298 (unsigned long long)ref->key_for_search.offset);
299 if (ret < 0)
300 goto out;
301
302 eb = path->nodes[level];
303 while (!eb) {
304 if (!level) {
305 WARN_ON(1);
306 ret = 1;
307 goto out;
308 }
309 level--;
310 eb = path->nodes[level];
311 }
312
313 ret = add_all_parents(root, path, parents, level, &ref->key_for_search,
314 time_seq, ref->wanted_disk_byte,
315 extent_item_pos);
316 out:
317 btrfs_free_path(path);
318 return ret;
319 }
320
321 /*
322 * resolve all indirect backrefs from the list
323 */
324 static int __resolve_indirect_refs(struct btrfs_fs_info *fs_info,
325 int search_commit_root, u64 time_seq,
326 struct list_head *head,
327 const u64 *extent_item_pos)
328 {
329 int err;
330 int ret = 0;
331 struct __prelim_ref *ref;
332 struct __prelim_ref *ref_safe;
333 struct __prelim_ref *new_ref;
334 struct ulist *parents;
335 struct ulist_node *node;
336 struct ulist_iterator uiter;
337
338 parents = ulist_alloc(GFP_NOFS);
339 if (!parents)
340 return -ENOMEM;
341
342 /*
343 * _safe allows us to insert directly after the current item without
344 * iterating over the newly inserted items.
345 * we're also allowed to re-assign ref during iteration.
346 */
347 list_for_each_entry_safe(ref, ref_safe, head, list) {
348 if (ref->parent) /* already direct */
349 continue;
350 if (ref->count == 0)
351 continue;
352 err = __resolve_indirect_ref(fs_info, search_commit_root,
353 time_seq, ref, parents,
354 extent_item_pos);
355 if (err)
356 continue;
357
358 /* we put the first parent into the ref at hand */
359 ULIST_ITER_INIT(&uiter);
360 node = ulist_next(parents, &uiter);
361 ref->parent = node ? node->val : 0;
362 ref->inode_list = node ?
363 (struct extent_inode_elem *)(uintptr_t)node->aux : 0;
364
365 /* additional parents require new refs being added here */
366 while ((node = ulist_next(parents, &uiter))) {
367 new_ref = kmalloc(sizeof(*new_ref), GFP_NOFS);
368 if (!new_ref) {
369 ret = -ENOMEM;
370 break;
371 }
372 memcpy(new_ref, ref, sizeof(*ref));
373 new_ref->parent = node->val;
374 new_ref->inode_list = (struct extent_inode_elem *)
375 (uintptr_t)node->aux;
376 list_add(&new_ref->list, &ref->list);
377 }
378 ulist_reinit(parents);
379 }
380
381 ulist_free(parents);
382 return ret;
383 }
384
385 static inline int ref_for_same_block(struct __prelim_ref *ref1,
386 struct __prelim_ref *ref2)
387 {
388 if (ref1->level != ref2->level)
389 return 0;
390 if (ref1->root_id != ref2->root_id)
391 return 0;
392 if (ref1->key_for_search.type != ref2->key_for_search.type)
393 return 0;
394 if (ref1->key_for_search.objectid != ref2->key_for_search.objectid)
395 return 0;
396 if (ref1->key_for_search.offset != ref2->key_for_search.offset)
397 return 0;
398 if (ref1->parent != ref2->parent)
399 return 0;
400
401 return 1;
402 }
403
404 /*
405 * read tree blocks and add keys where required.
406 */
407 static int __add_missing_keys(struct btrfs_fs_info *fs_info,
408 struct list_head *head)
409 {
410 struct list_head *pos;
411 struct extent_buffer *eb;
412
413 list_for_each(pos, head) {
414 struct __prelim_ref *ref;
415 ref = list_entry(pos, struct __prelim_ref, list);
416
417 if (ref->parent)
418 continue;
419 if (ref->key_for_search.type)
420 continue;
421 BUG_ON(!ref->wanted_disk_byte);
422 eb = read_tree_block(fs_info->tree_root, ref->wanted_disk_byte,
423 fs_info->tree_root->leafsize, 0);
424 BUG_ON(!eb);
425 btrfs_tree_read_lock(eb);
426 if (btrfs_header_level(eb) == 0)
427 btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
428 else
429 btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
430 btrfs_tree_read_unlock(eb);
431 free_extent_buffer(eb);
432 }
433 return 0;
434 }
435
436 /*
437 * merge two lists of backrefs and adjust counts accordingly
438 *
439 * mode = 1: merge identical keys, if key is set
440 * FIXME: if we add more keys in __add_prelim_ref, we can merge more here.
441 * additionally, we could even add a key range for the blocks we
442 * looked into to merge even more (-> replace unresolved refs by those
443 * having a parent).
444 * mode = 2: merge identical parents
445 */
446 static int __merge_refs(struct list_head *head, int mode)
447 {
448 struct list_head *pos1;
449
450 list_for_each(pos1, head) {
451 struct list_head *n2;
452 struct list_head *pos2;
453 struct __prelim_ref *ref1;
454
455 ref1 = list_entry(pos1, struct __prelim_ref, list);
456
457 for (pos2 = pos1->next, n2 = pos2->next; pos2 != head;
458 pos2 = n2, n2 = pos2->next) {
459 struct __prelim_ref *ref2;
460 struct __prelim_ref *xchg;
461 struct extent_inode_elem *eie;
462
463 ref2 = list_entry(pos2, struct __prelim_ref, list);
464
465 if (mode == 1) {
466 if (!ref_for_same_block(ref1, ref2))
467 continue;
468 if (!ref1->parent && ref2->parent) {
469 xchg = ref1;
470 ref1 = ref2;
471 ref2 = xchg;
472 }
473 } else {
474 if (ref1->parent != ref2->parent)
475 continue;
476 }
477
478 eie = ref1->inode_list;
479 while (eie && eie->next)
480 eie = eie->next;
481 if (eie)
482 eie->next = ref2->inode_list;
483 else
484 ref1->inode_list = ref2->inode_list;
485 ref1->count += ref2->count;
486
487 list_del(&ref2->list);
488 kfree(ref2);
489 }
490
491 }
492 return 0;
493 }
494
495 /*
496 * add all currently queued delayed refs from this head whose seq nr is
497 * smaller or equal that seq to the list
498 */
499 static int __add_delayed_refs(struct btrfs_delayed_ref_head *head, u64 seq,
500 struct list_head *prefs)
501 {
502 struct btrfs_delayed_extent_op *extent_op = head->extent_op;
503 struct rb_node *n = &head->node.rb_node;
504 struct btrfs_key key;
505 struct btrfs_key op_key = {0};
506 int sgn;
507 int ret = 0;
508
509 if (extent_op && extent_op->update_key)
510 btrfs_disk_key_to_cpu(&op_key, &extent_op->key);
511
512 while ((n = rb_prev(n))) {
513 struct btrfs_delayed_ref_node *node;
514 node = rb_entry(n, struct btrfs_delayed_ref_node,
515 rb_node);
516 if (node->bytenr != head->node.bytenr)
517 break;
518 WARN_ON(node->is_head);
519
520 if (node->seq > seq)
521 continue;
522
523 switch (node->action) {
524 case BTRFS_ADD_DELAYED_EXTENT:
525 case BTRFS_UPDATE_DELAYED_HEAD:
526 WARN_ON(1);
527 continue;
528 case BTRFS_ADD_DELAYED_REF:
529 sgn = 1;
530 break;
531 case BTRFS_DROP_DELAYED_REF:
532 sgn = -1;
533 break;
534 default:
535 BUG_ON(1);
536 }
537 switch (node->type) {
538 case BTRFS_TREE_BLOCK_REF_KEY: {
539 struct btrfs_delayed_tree_ref *ref;
540
541 ref = btrfs_delayed_node_to_tree_ref(node);
542 ret = __add_prelim_ref(prefs, ref->root, &op_key,
543 ref->level + 1, 0, node->bytenr,
544 node->ref_mod * sgn);
545 break;
546 }
547 case BTRFS_SHARED_BLOCK_REF_KEY: {
548 struct btrfs_delayed_tree_ref *ref;
549
550 ref = btrfs_delayed_node_to_tree_ref(node);
551 ret = __add_prelim_ref(prefs, ref->root, NULL,
552 ref->level + 1, ref->parent,
553 node->bytenr,
554 node->ref_mod * sgn);
555 break;
556 }
557 case BTRFS_EXTENT_DATA_REF_KEY: {
558 struct btrfs_delayed_data_ref *ref;
559 ref = btrfs_delayed_node_to_data_ref(node);
560
561 key.objectid = ref->objectid;
562 key.type = BTRFS_EXTENT_DATA_KEY;
563 key.offset = ref->offset;
564 ret = __add_prelim_ref(prefs, ref->root, &key, 0, 0,
565 node->bytenr,
566 node->ref_mod * sgn);
567 break;
568 }
569 case BTRFS_SHARED_DATA_REF_KEY: {
570 struct btrfs_delayed_data_ref *ref;
571
572 ref = btrfs_delayed_node_to_data_ref(node);
573
574 key.objectid = ref->objectid;
575 key.type = BTRFS_EXTENT_DATA_KEY;
576 key.offset = ref->offset;
577 ret = __add_prelim_ref(prefs, ref->root, &key, 0,
578 ref->parent, node->bytenr,
579 node->ref_mod * sgn);
580 break;
581 }
582 default:
583 WARN_ON(1);
584 }
585 BUG_ON(ret);
586 }
587
588 return 0;
589 }
590
591 /*
592 * add all inline backrefs for bytenr to the list
593 */
594 static int __add_inline_refs(struct btrfs_fs_info *fs_info,
595 struct btrfs_path *path, u64 bytenr,
596 int *info_level, struct list_head *prefs)
597 {
598 int ret = 0;
599 int slot;
600 struct extent_buffer *leaf;
601 struct btrfs_key key;
602 unsigned long ptr;
603 unsigned long end;
604 struct btrfs_extent_item *ei;
605 u64 flags;
606 u64 item_size;
607
608 /*
609 * enumerate all inline refs
610 */
611 leaf = path->nodes[0];
612 slot = path->slots[0];
613
614 item_size = btrfs_item_size_nr(leaf, slot);
615 BUG_ON(item_size < sizeof(*ei));
616
617 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
618 flags = btrfs_extent_flags(leaf, ei);
619
620 ptr = (unsigned long)(ei + 1);
621 end = (unsigned long)ei + item_size;
622
623 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
624 struct btrfs_tree_block_info *info;
625
626 info = (struct btrfs_tree_block_info *)ptr;
627 *info_level = btrfs_tree_block_level(leaf, info);
628 ptr += sizeof(struct btrfs_tree_block_info);
629 BUG_ON(ptr > end);
630 } else {
631 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
632 }
633
634 while (ptr < end) {
635 struct btrfs_extent_inline_ref *iref;
636 u64 offset;
637 int type;
638
639 iref = (struct btrfs_extent_inline_ref *)ptr;
640 type = btrfs_extent_inline_ref_type(leaf, iref);
641 offset = btrfs_extent_inline_ref_offset(leaf, iref);
642
643 switch (type) {
644 case BTRFS_SHARED_BLOCK_REF_KEY:
645 ret = __add_prelim_ref(prefs, 0, NULL,
646 *info_level + 1, offset,
647 bytenr, 1);
648 break;
649 case BTRFS_SHARED_DATA_REF_KEY: {
650 struct btrfs_shared_data_ref *sdref;
651 int count;
652
653 sdref = (struct btrfs_shared_data_ref *)(iref + 1);
654 count = btrfs_shared_data_ref_count(leaf, sdref);
655 ret = __add_prelim_ref(prefs, 0, NULL, 0, offset,
656 bytenr, count);
657 break;
658 }
659 case BTRFS_TREE_BLOCK_REF_KEY:
660 ret = __add_prelim_ref(prefs, offset, NULL,
661 *info_level + 1, 0,
662 bytenr, 1);
663 break;
664 case BTRFS_EXTENT_DATA_REF_KEY: {
665 struct btrfs_extent_data_ref *dref;
666 int count;
667 u64 root;
668
669 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
670 count = btrfs_extent_data_ref_count(leaf, dref);
671 key.objectid = btrfs_extent_data_ref_objectid(leaf,
672 dref);
673 key.type = BTRFS_EXTENT_DATA_KEY;
674 key.offset = btrfs_extent_data_ref_offset(leaf, dref);
675 root = btrfs_extent_data_ref_root(leaf, dref);
676 ret = __add_prelim_ref(prefs, root, &key, 0, 0,
677 bytenr, count);
678 break;
679 }
680 default:
681 WARN_ON(1);
682 }
683 BUG_ON(ret);
684 ptr += btrfs_extent_inline_ref_size(type);
685 }
686
687 return 0;
688 }
689
690 /*
691 * add all non-inline backrefs for bytenr to the list
692 */
693 static int __add_keyed_refs(struct btrfs_fs_info *fs_info,
694 struct btrfs_path *path, u64 bytenr,
695 int info_level, struct list_head *prefs)
696 {
697 struct btrfs_root *extent_root = fs_info->extent_root;
698 int ret;
699 int slot;
700 struct extent_buffer *leaf;
701 struct btrfs_key key;
702
703 while (1) {
704 ret = btrfs_next_item(extent_root, path);
705 if (ret < 0)
706 break;
707 if (ret) {
708 ret = 0;
709 break;
710 }
711
712 slot = path->slots[0];
713 leaf = path->nodes[0];
714 btrfs_item_key_to_cpu(leaf, &key, slot);
715
716 if (key.objectid != bytenr)
717 break;
718 if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
719 continue;
720 if (key.type > BTRFS_SHARED_DATA_REF_KEY)
721 break;
722
723 switch (key.type) {
724 case BTRFS_SHARED_BLOCK_REF_KEY:
725 ret = __add_prelim_ref(prefs, 0, NULL,
726 info_level + 1, key.offset,
727 bytenr, 1);
728 break;
729 case BTRFS_SHARED_DATA_REF_KEY: {
730 struct btrfs_shared_data_ref *sdref;
731 int count;
732
733 sdref = btrfs_item_ptr(leaf, slot,
734 struct btrfs_shared_data_ref);
735 count = btrfs_shared_data_ref_count(leaf, sdref);
736 ret = __add_prelim_ref(prefs, 0, NULL, 0, key.offset,
737 bytenr, count);
738 break;
739 }
740 case BTRFS_TREE_BLOCK_REF_KEY:
741 ret = __add_prelim_ref(prefs, key.offset, NULL,
742 info_level + 1, 0,
743 bytenr, 1);
744 break;
745 case BTRFS_EXTENT_DATA_REF_KEY: {
746 struct btrfs_extent_data_ref *dref;
747 int count;
748 u64 root;
749
750 dref = btrfs_item_ptr(leaf, slot,
751 struct btrfs_extent_data_ref);
752 count = btrfs_extent_data_ref_count(leaf, dref);
753 key.objectid = btrfs_extent_data_ref_objectid(leaf,
754 dref);
755 key.type = BTRFS_EXTENT_DATA_KEY;
756 key.offset = btrfs_extent_data_ref_offset(leaf, dref);
757 root = btrfs_extent_data_ref_root(leaf, dref);
758 ret = __add_prelim_ref(prefs, root, &key, 0, 0,
759 bytenr, count);
760 break;
761 }
762 default:
763 WARN_ON(1);
764 }
765 BUG_ON(ret);
766 }
767
768 return ret;
769 }
770
771 /*
772 * this adds all existing backrefs (inline backrefs, backrefs and delayed
773 * refs) for the given bytenr to the refs list, merges duplicates and resolves
774 * indirect refs to their parent bytenr.
775 * When roots are found, they're added to the roots list
776 *
777 * FIXME some caching might speed things up
778 */
779 static int find_parent_nodes(struct btrfs_trans_handle *trans,
780 struct btrfs_fs_info *fs_info, u64 bytenr,
781 u64 time_seq, struct ulist *refs,
782 struct ulist *roots, const u64 *extent_item_pos)
783 {
784 struct btrfs_key key;
785 struct btrfs_path *path;
786 struct btrfs_delayed_ref_root *delayed_refs = NULL;
787 struct btrfs_delayed_ref_head *head;
788 int info_level = 0;
789 int ret;
790 int search_commit_root = (trans == BTRFS_BACKREF_SEARCH_COMMIT_ROOT);
791 struct list_head prefs_delayed;
792 struct list_head prefs;
793 struct __prelim_ref *ref;
794
795 INIT_LIST_HEAD(&prefs);
796 INIT_LIST_HEAD(&prefs_delayed);
797
798 key.objectid = bytenr;
799 key.type = BTRFS_EXTENT_ITEM_KEY;
800 key.offset = (u64)-1;
801
802 path = btrfs_alloc_path();
803 if (!path)
804 return -ENOMEM;
805 path->search_commit_root = !!search_commit_root;
806
807 /*
808 * grab both a lock on the path and a lock on the delayed ref head.
809 * We need both to get a consistent picture of how the refs look
810 * at a specified point in time
811 */
812 again:
813 head = NULL;
814
815 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
816 if (ret < 0)
817 goto out;
818 BUG_ON(ret == 0);
819
820 if (trans != BTRFS_BACKREF_SEARCH_COMMIT_ROOT) {
821 /*
822 * look if there are updates for this ref queued and lock the
823 * head
824 */
825 delayed_refs = &trans->transaction->delayed_refs;
826 spin_lock(&delayed_refs->lock);
827 head = btrfs_find_delayed_ref_head(trans, bytenr);
828 if (head) {
829 if (!mutex_trylock(&head->mutex)) {
830 atomic_inc(&head->node.refs);
831 spin_unlock(&delayed_refs->lock);
832
833 btrfs_release_path(path);
834
835 /*
836 * Mutex was contended, block until it's
837 * released and try again
838 */
839 mutex_lock(&head->mutex);
840 mutex_unlock(&head->mutex);
841 btrfs_put_delayed_ref(&head->node);
842 goto again;
843 }
844 ret = __add_delayed_refs(head, time_seq,
845 &prefs_delayed);
846 mutex_unlock(&head->mutex);
847 if (ret) {
848 spin_unlock(&delayed_refs->lock);
849 goto out;
850 }
851 }
852 spin_unlock(&delayed_refs->lock);
853 }
854
855 if (path->slots[0]) {
856 struct extent_buffer *leaf;
857 int slot;
858
859 path->slots[0]--;
860 leaf = path->nodes[0];
861 slot = path->slots[0];
862 btrfs_item_key_to_cpu(leaf, &key, slot);
863 if (key.objectid == bytenr &&
864 key.type == BTRFS_EXTENT_ITEM_KEY) {
865 ret = __add_inline_refs(fs_info, path, bytenr,
866 &info_level, &prefs);
867 if (ret)
868 goto out;
869 ret = __add_keyed_refs(fs_info, path, bytenr,
870 info_level, &prefs);
871 if (ret)
872 goto out;
873 }
874 }
875 btrfs_release_path(path);
876
877 list_splice_init(&prefs_delayed, &prefs);
878
879 ret = __add_missing_keys(fs_info, &prefs);
880 if (ret)
881 goto out;
882
883 ret = __merge_refs(&prefs, 1);
884 if (ret)
885 goto out;
886
887 ret = __resolve_indirect_refs(fs_info, search_commit_root, time_seq,
888 &prefs, extent_item_pos);
889 if (ret)
890 goto out;
891
892 ret = __merge_refs(&prefs, 2);
893 if (ret)
894 goto out;
895
896 while (!list_empty(&prefs)) {
897 ref = list_first_entry(&prefs, struct __prelim_ref, list);
898 list_del(&ref->list);
899 WARN_ON(ref->count < 0);
900 if (ref->count && ref->root_id && ref->parent == 0) {
901 /* no parent == root of tree */
902 ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS);
903 if (ret < 0)
904 goto out;
905 }
906 if (ref->count && ref->parent) {
907 struct extent_inode_elem *eie = NULL;
908 if (extent_item_pos && !ref->inode_list) {
909 u32 bsz;
910 struct extent_buffer *eb;
911 bsz = btrfs_level_size(fs_info->extent_root,
912 info_level);
913 eb = read_tree_block(fs_info->extent_root,
914 ref->parent, bsz, 0);
915 BUG_ON(!eb);
916 ret = find_extent_in_eb(eb, bytenr,
917 *extent_item_pos, &eie);
918 ref->inode_list = eie;
919 free_extent_buffer(eb);
920 }
921 ret = ulist_add_merge(refs, ref->parent,
922 (uintptr_t)ref->inode_list,
923 (u64 *)&eie, GFP_NOFS);
924 if (ret < 0)
925 goto out;
926 if (!ret && extent_item_pos) {
927 /*
928 * we've recorded that parent, so we must extend
929 * its inode list here
930 */
931 BUG_ON(!eie);
932 while (eie->next)
933 eie = eie->next;
934 eie->next = ref->inode_list;
935 }
936 }
937 kfree(ref);
938 }
939
940 out:
941 btrfs_free_path(path);
942 while (!list_empty(&prefs)) {
943 ref = list_first_entry(&prefs, struct __prelim_ref, list);
944 list_del(&ref->list);
945 kfree(ref);
946 }
947 while (!list_empty(&prefs_delayed)) {
948 ref = list_first_entry(&prefs_delayed, struct __prelim_ref,
949 list);
950 list_del(&ref->list);
951 kfree(ref);
952 }
953
954 return ret;
955 }
956
957 static void free_leaf_list(struct ulist *blocks)
958 {
959 struct ulist_node *node = NULL;
960 struct extent_inode_elem *eie;
961 struct extent_inode_elem *eie_next;
962 struct ulist_iterator uiter;
963
964 ULIST_ITER_INIT(&uiter);
965 while ((node = ulist_next(blocks, &uiter))) {
966 if (!node->aux)
967 continue;
968 eie = (struct extent_inode_elem *)(uintptr_t)node->aux;
969 for (; eie; eie = eie_next) {
970 eie_next = eie->next;
971 kfree(eie);
972 }
973 node->aux = 0;
974 }
975
976 ulist_free(blocks);
977 }
978
979 /*
980 * Finds all leafs with a reference to the specified combination of bytenr and
981 * offset. key_list_head will point to a list of corresponding keys (caller must
982 * free each list element). The leafs will be stored in the leafs ulist, which
983 * must be freed with ulist_free.
984 *
985 * returns 0 on success, <0 on error
986 */
987 static int btrfs_find_all_leafs(struct btrfs_trans_handle *trans,
988 struct btrfs_fs_info *fs_info, u64 bytenr,
989 u64 time_seq, struct ulist **leafs,
990 const u64 *extent_item_pos)
991 {
992 struct ulist *tmp;
993 int ret;
994
995 tmp = ulist_alloc(GFP_NOFS);
996 if (!tmp)
997 return -ENOMEM;
998 *leafs = ulist_alloc(GFP_NOFS);
999 if (!*leafs) {
1000 ulist_free(tmp);
1001 return -ENOMEM;
1002 }
1003
1004 ret = find_parent_nodes(trans, fs_info, bytenr,
1005 time_seq, *leafs, tmp, extent_item_pos);
1006 ulist_free(tmp);
1007
1008 if (ret < 0 && ret != -ENOENT) {
1009 free_leaf_list(*leafs);
1010 return ret;
1011 }
1012
1013 return 0;
1014 }
1015
1016 /*
1017 * walk all backrefs for a given extent to find all roots that reference this
1018 * extent. Walking a backref means finding all extents that reference this
1019 * extent and in turn walk the backrefs of those, too. Naturally this is a
1020 * recursive process, but here it is implemented in an iterative fashion: We
1021 * find all referencing extents for the extent in question and put them on a
1022 * list. In turn, we find all referencing extents for those, further appending
1023 * to the list. The way we iterate the list allows adding more elements after
1024 * the current while iterating. The process stops when we reach the end of the
1025 * list. Found roots are added to the roots list.
1026 *
1027 * returns 0 on success, < 0 on error.
1028 */
1029 int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1030 struct btrfs_fs_info *fs_info, u64 bytenr,
1031 u64 time_seq, struct ulist **roots)
1032 {
1033 struct ulist *tmp;
1034 struct ulist_node *node = NULL;
1035 struct ulist_iterator uiter;
1036 int ret;
1037
1038 tmp = ulist_alloc(GFP_NOFS);
1039 if (!tmp)
1040 return -ENOMEM;
1041 *roots = ulist_alloc(GFP_NOFS);
1042 if (!*roots) {
1043 ulist_free(tmp);
1044 return -ENOMEM;
1045 }
1046
1047 ULIST_ITER_INIT(&uiter);
1048 while (1) {
1049 ret = find_parent_nodes(trans, fs_info, bytenr,
1050 time_seq, tmp, *roots, NULL);
1051 if (ret < 0 && ret != -ENOENT) {
1052 ulist_free(tmp);
1053 ulist_free(*roots);
1054 return ret;
1055 }
1056 node = ulist_next(tmp, &uiter);
1057 if (!node)
1058 break;
1059 bytenr = node->val;
1060 }
1061
1062 ulist_free(tmp);
1063 return 0;
1064 }
1065
1066
1067 static int __inode_info(u64 inum, u64 ioff, u8 key_type,
1068 struct btrfs_root *fs_root, struct btrfs_path *path,
1069 struct btrfs_key *found_key)
1070 {
1071 int ret;
1072 struct btrfs_key key;
1073 struct extent_buffer *eb;
1074
1075 key.type = key_type;
1076 key.objectid = inum;
1077 key.offset = ioff;
1078
1079 ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
1080 if (ret < 0)
1081 return ret;
1082
1083 eb = path->nodes[0];
1084 if (ret && path->slots[0] >= btrfs_header_nritems(eb)) {
1085 ret = btrfs_next_leaf(fs_root, path);
1086 if (ret)
1087 return ret;
1088 eb = path->nodes[0];
1089 }
1090
1091 btrfs_item_key_to_cpu(eb, found_key, path->slots[0]);
1092 if (found_key->type != key.type || found_key->objectid != key.objectid)
1093 return 1;
1094
1095 return 0;
1096 }
1097
1098 /*
1099 * this makes the path point to (inum INODE_ITEM ioff)
1100 */
1101 int inode_item_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
1102 struct btrfs_path *path)
1103 {
1104 struct btrfs_key key;
1105 return __inode_info(inum, ioff, BTRFS_INODE_ITEM_KEY, fs_root, path,
1106 &key);
1107 }
1108
1109 static int inode_ref_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
1110 struct btrfs_path *path,
1111 struct btrfs_key *found_key)
1112 {
1113 return __inode_info(inum, ioff, BTRFS_INODE_REF_KEY, fs_root, path,
1114 found_key);
1115 }
1116
1117 int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,
1118 u64 start_off, struct btrfs_path *path,
1119 struct btrfs_inode_extref **ret_extref,
1120 u64 *found_off)
1121 {
1122 int ret, slot;
1123 struct btrfs_key key;
1124 struct btrfs_key found_key;
1125 struct btrfs_inode_extref *extref;
1126 struct extent_buffer *leaf;
1127 unsigned long ptr;
1128
1129 key.objectid = inode_objectid;
1130 btrfs_set_key_type(&key, BTRFS_INODE_EXTREF_KEY);
1131 key.offset = start_off;
1132
1133 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1134 if (ret < 0)
1135 return ret;
1136
1137 while (1) {
1138 leaf = path->nodes[0];
1139 slot = path->slots[0];
1140 if (slot >= btrfs_header_nritems(leaf)) {
1141 /*
1142 * If the item at offset is not found,
1143 * btrfs_search_slot will point us to the slot
1144 * where it should be inserted. In our case
1145 * that will be the slot directly before the
1146 * next INODE_REF_KEY_V2 item. In the case
1147 * that we're pointing to the last slot in a
1148 * leaf, we must move one leaf over.
1149 */
1150 ret = btrfs_next_leaf(root, path);
1151 if (ret) {
1152 if (ret >= 1)
1153 ret = -ENOENT;
1154 break;
1155 }
1156 continue;
1157 }
1158
1159 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1160
1161 /*
1162 * Check that we're still looking at an extended ref key for
1163 * this particular objectid. If we have different
1164 * objectid or type then there are no more to be found
1165 * in the tree and we can exit.
1166 */
1167 ret = -ENOENT;
1168 if (found_key.objectid != inode_objectid)
1169 break;
1170 if (btrfs_key_type(&found_key) != BTRFS_INODE_EXTREF_KEY)
1171 break;
1172
1173 ret = 0;
1174 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1175 extref = (struct btrfs_inode_extref *)ptr;
1176 *ret_extref = extref;
1177 if (found_off)
1178 *found_off = found_key.offset;
1179 break;
1180 }
1181
1182 return ret;
1183 }
1184
1185 char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
1186 u32 name_len, unsigned long name_off,
1187 struct extent_buffer *eb_in, u64 parent,
1188 char *dest, u32 size)
1189 {
1190 int slot;
1191 u64 next_inum;
1192 int ret;
1193 s64 bytes_left = ((s64)size) - 1;
1194 struct extent_buffer *eb = eb_in;
1195 struct btrfs_key found_key;
1196 int leave_spinning = path->leave_spinning;
1197 struct btrfs_inode_ref *iref;
1198
1199 if (bytes_left >= 0)
1200 dest[bytes_left] = '\0';
1201
1202 path->leave_spinning = 1;
1203 while (1) {
1204 bytes_left -= name_len;
1205 if (bytes_left >= 0)
1206 read_extent_buffer(eb, dest + bytes_left,
1207 name_off, name_len);
1208 if (eb != eb_in) {
1209 btrfs_tree_read_unlock_blocking(eb);
1210 free_extent_buffer(eb);
1211 }
1212 ret = inode_ref_info(parent, 0, fs_root, path, &found_key);
1213 if (ret > 0)
1214 ret = -ENOENT;
1215 if (ret)
1216 break;
1217
1218 next_inum = found_key.offset;
1219
1220 /* regular exit ahead */
1221 if (parent == next_inum)
1222 break;
1223
1224 slot = path->slots[0];
1225 eb = path->nodes[0];
1226 /* make sure we can use eb after releasing the path */
1227 if (eb != eb_in) {
1228 atomic_inc(&eb->refs);
1229 btrfs_tree_read_lock(eb);
1230 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1231 }
1232 btrfs_release_path(path);
1233 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1234
1235 name_len = btrfs_inode_ref_name_len(eb, iref);
1236 name_off = (unsigned long)(iref + 1);
1237
1238 parent = next_inum;
1239 --bytes_left;
1240 if (bytes_left >= 0)
1241 dest[bytes_left] = '/';
1242 }
1243
1244 btrfs_release_path(path);
1245 path->leave_spinning = leave_spinning;
1246
1247 if (ret)
1248 return ERR_PTR(ret);
1249
1250 return dest + bytes_left;
1251 }
1252
1253 /*
1254 * this iterates to turn a btrfs_inode_ref into a full filesystem path. elements
1255 * of the path are separated by '/' and the path is guaranteed to be
1256 * 0-terminated. the path is only given within the current file system.
1257 * Therefore, it never starts with a '/'. the caller is responsible to provide
1258 * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
1259 * the start point of the resulting string is returned. this pointer is within
1260 * dest, normally.
1261 * in case the path buffer would overflow, the pointer is decremented further
1262 * as if output was written to the buffer, though no more output is actually
1263 * generated. that way, the caller can determine how much space would be
1264 * required for the path to fit into the buffer. in that case, the returned
1265 * value will be smaller than dest. callers must check this!
1266 */
1267 char *btrfs_iref_to_path(struct btrfs_root *fs_root,
1268 struct btrfs_path *path,
1269 struct btrfs_inode_ref *iref,
1270 struct extent_buffer *eb_in, u64 parent,
1271 char *dest, u32 size)
1272 {
1273 return btrfs_ref_to_path(fs_root, path,
1274 btrfs_inode_ref_name_len(eb_in, iref),
1275 (unsigned long)(iref + 1),
1276 eb_in, parent, dest, size);
1277 }
1278
1279 /*
1280 * this makes the path point to (logical EXTENT_ITEM *)
1281 * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
1282 * tree blocks and <0 on error.
1283 */
1284 int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
1285 struct btrfs_path *path, struct btrfs_key *found_key,
1286 u64 *flags_ret)
1287 {
1288 int ret;
1289 u64 flags;
1290 u32 item_size;
1291 struct extent_buffer *eb;
1292 struct btrfs_extent_item *ei;
1293 struct btrfs_key key;
1294
1295 key.type = BTRFS_EXTENT_ITEM_KEY;
1296 key.objectid = logical;
1297 key.offset = (u64)-1;
1298
1299 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
1300 if (ret < 0)
1301 return ret;
1302 ret = btrfs_previous_item(fs_info->extent_root, path,
1303 0, BTRFS_EXTENT_ITEM_KEY);
1304 if (ret < 0)
1305 return ret;
1306
1307 btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]);
1308 if (found_key->type != BTRFS_EXTENT_ITEM_KEY ||
1309 found_key->objectid > logical ||
1310 found_key->objectid + found_key->offset <= logical) {
1311 pr_debug("logical %llu is not within any extent\n",
1312 (unsigned long long)logical);
1313 return -ENOENT;
1314 }
1315
1316 eb = path->nodes[0];
1317 item_size = btrfs_item_size_nr(eb, path->slots[0]);
1318 BUG_ON(item_size < sizeof(*ei));
1319
1320 ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
1321 flags = btrfs_extent_flags(eb, ei);
1322
1323 pr_debug("logical %llu is at position %llu within the extent (%llu "
1324 "EXTENT_ITEM %llu) flags %#llx size %u\n",
1325 (unsigned long long)logical,
1326 (unsigned long long)(logical - found_key->objectid),
1327 (unsigned long long)found_key->objectid,
1328 (unsigned long long)found_key->offset,
1329 (unsigned long long)flags, item_size);
1330
1331 WARN_ON(!flags_ret);
1332 if (flags_ret) {
1333 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1334 *flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK;
1335 else if (flags & BTRFS_EXTENT_FLAG_DATA)
1336 *flags_ret = BTRFS_EXTENT_FLAG_DATA;
1337 else
1338 BUG_ON(1);
1339 return 0;
1340 }
1341
1342 return -EIO;
1343 }
1344
1345 /*
1346 * helper function to iterate extent inline refs. ptr must point to a 0 value
1347 * for the first call and may be modified. it is used to track state.
1348 * if more refs exist, 0 is returned and the next call to
1349 * __get_extent_inline_ref must pass the modified ptr parameter to get the
1350 * next ref. after the last ref was processed, 1 is returned.
1351 * returns <0 on error
1352 */
1353 static int __get_extent_inline_ref(unsigned long *ptr, struct extent_buffer *eb,
1354 struct btrfs_extent_item *ei, u32 item_size,
1355 struct btrfs_extent_inline_ref **out_eiref,
1356 int *out_type)
1357 {
1358 unsigned long end;
1359 u64 flags;
1360 struct btrfs_tree_block_info *info;
1361
1362 if (!*ptr) {
1363 /* first call */
1364 flags = btrfs_extent_flags(eb, ei);
1365 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1366 info = (struct btrfs_tree_block_info *)(ei + 1);
1367 *out_eiref =
1368 (struct btrfs_extent_inline_ref *)(info + 1);
1369 } else {
1370 *out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1);
1371 }
1372 *ptr = (unsigned long)*out_eiref;
1373 if ((void *)*ptr >= (void *)ei + item_size)
1374 return -ENOENT;
1375 }
1376
1377 end = (unsigned long)ei + item_size;
1378 *out_eiref = (struct btrfs_extent_inline_ref *)*ptr;
1379 *out_type = btrfs_extent_inline_ref_type(eb, *out_eiref);
1380
1381 *ptr += btrfs_extent_inline_ref_size(*out_type);
1382 WARN_ON(*ptr > end);
1383 if (*ptr == end)
1384 return 1; /* last */
1385
1386 return 0;
1387 }
1388
1389 /*
1390 * reads the tree block backref for an extent. tree level and root are returned
1391 * through out_level and out_root. ptr must point to a 0 value for the first
1392 * call and may be modified (see __get_extent_inline_ref comment).
1393 * returns 0 if data was provided, 1 if there was no more data to provide or
1394 * <0 on error.
1395 */
1396 int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
1397 struct btrfs_extent_item *ei, u32 item_size,
1398 u64 *out_root, u8 *out_level)
1399 {
1400 int ret;
1401 int type;
1402 struct btrfs_tree_block_info *info;
1403 struct btrfs_extent_inline_ref *eiref;
1404
1405 if (*ptr == (unsigned long)-1)
1406 return 1;
1407
1408 while (1) {
1409 ret = __get_extent_inline_ref(ptr, eb, ei, item_size,
1410 &eiref, &type);
1411 if (ret < 0)
1412 return ret;
1413
1414 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1415 type == BTRFS_SHARED_BLOCK_REF_KEY)
1416 break;
1417
1418 if (ret == 1)
1419 return 1;
1420 }
1421
1422 /* we can treat both ref types equally here */
1423 info = (struct btrfs_tree_block_info *)(ei + 1);
1424 *out_root = btrfs_extent_inline_ref_offset(eb, eiref);
1425 *out_level = btrfs_tree_block_level(eb, info);
1426
1427 if (ret == 1)
1428 *ptr = (unsigned long)-1;
1429
1430 return 0;
1431 }
1432
1433 static int iterate_leaf_refs(struct extent_inode_elem *inode_list,
1434 u64 root, u64 extent_item_objectid,
1435 iterate_extent_inodes_t *iterate, void *ctx)
1436 {
1437 struct extent_inode_elem *eie;
1438 int ret = 0;
1439
1440 for (eie = inode_list; eie; eie = eie->next) {
1441 pr_debug("ref for %llu resolved, key (%llu EXTEND_DATA %llu), "
1442 "root %llu\n", extent_item_objectid,
1443 eie->inum, eie->offset, root);
1444 ret = iterate(eie->inum, eie->offset, root, ctx);
1445 if (ret) {
1446 pr_debug("stopping iteration for %llu due to ret=%d\n",
1447 extent_item_objectid, ret);
1448 break;
1449 }
1450 }
1451
1452 return ret;
1453 }
1454
1455 /*
1456 * calls iterate() for every inode that references the extent identified by
1457 * the given parameters.
1458 * when the iterator function returns a non-zero value, iteration stops.
1459 */
1460 int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
1461 u64 extent_item_objectid, u64 extent_item_pos,
1462 int search_commit_root,
1463 iterate_extent_inodes_t *iterate, void *ctx)
1464 {
1465 int ret;
1466 struct list_head data_refs = LIST_HEAD_INIT(data_refs);
1467 struct list_head shared_refs = LIST_HEAD_INIT(shared_refs);
1468 struct btrfs_trans_handle *trans;
1469 struct ulist *refs = NULL;
1470 struct ulist *roots = NULL;
1471 struct ulist_node *ref_node = NULL;
1472 struct ulist_node *root_node = NULL;
1473 struct seq_list tree_mod_seq_elem = {};
1474 struct ulist_iterator ref_uiter;
1475 struct ulist_iterator root_uiter;
1476
1477 pr_debug("resolving all inodes for extent %llu\n",
1478 extent_item_objectid);
1479
1480 if (search_commit_root) {
1481 trans = BTRFS_BACKREF_SEARCH_COMMIT_ROOT;
1482 } else {
1483 trans = btrfs_join_transaction(fs_info->extent_root);
1484 if (IS_ERR(trans))
1485 return PTR_ERR(trans);
1486 btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1487 }
1488
1489 ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid,
1490 tree_mod_seq_elem.seq, &refs,
1491 &extent_item_pos);
1492 if (ret)
1493 goto out;
1494
1495 ULIST_ITER_INIT(&ref_uiter);
1496 while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) {
1497 ret = btrfs_find_all_roots(trans, fs_info, ref_node->val,
1498 tree_mod_seq_elem.seq, &roots);
1499 if (ret)
1500 break;
1501 ULIST_ITER_INIT(&root_uiter);
1502 while (!ret && (root_node = ulist_next(roots, &root_uiter))) {
1503 pr_debug("root %llu references leaf %llu, data list "
1504 "%#llx\n", root_node->val, ref_node->val,
1505 (long long)ref_node->aux);
1506 ret = iterate_leaf_refs((struct extent_inode_elem *)
1507 (uintptr_t)ref_node->aux,
1508 root_node->val,
1509 extent_item_objectid,
1510 iterate, ctx);
1511 }
1512 ulist_free(roots);
1513 roots = NULL;
1514 }
1515
1516 free_leaf_list(refs);
1517 ulist_free(roots);
1518 out:
1519 if (!search_commit_root) {
1520 btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1521 btrfs_end_transaction(trans, fs_info->extent_root);
1522 }
1523
1524 return ret;
1525 }
1526
1527 int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
1528 struct btrfs_path *path,
1529 iterate_extent_inodes_t *iterate, void *ctx)
1530 {
1531 int ret;
1532 u64 extent_item_pos;
1533 u64 flags = 0;
1534 struct btrfs_key found_key;
1535 int search_commit_root = path->search_commit_root;
1536
1537 ret = extent_from_logical(fs_info, logical, path, &found_key, &flags);
1538 btrfs_release_path(path);
1539 if (ret < 0)
1540 return ret;
1541 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1542 return -EINVAL;
1543
1544 extent_item_pos = logical - found_key.objectid;
1545 ret = iterate_extent_inodes(fs_info, found_key.objectid,
1546 extent_item_pos, search_commit_root,
1547 iterate, ctx);
1548
1549 return ret;
1550 }
1551
1552 typedef int (iterate_irefs_t)(u64 parent, u32 name_len, unsigned long name_off,
1553 struct extent_buffer *eb, void *ctx);
1554
1555 static int iterate_inode_refs(u64 inum, struct btrfs_root *fs_root,
1556 struct btrfs_path *path,
1557 iterate_irefs_t *iterate, void *ctx)
1558 {
1559 int ret = 0;
1560 int slot;
1561 u32 cur;
1562 u32 len;
1563 u32 name_len;
1564 u64 parent = 0;
1565 int found = 0;
1566 struct extent_buffer *eb;
1567 struct btrfs_item *item;
1568 struct btrfs_inode_ref *iref;
1569 struct btrfs_key found_key;
1570
1571 while (!ret) {
1572 path->leave_spinning = 1;
1573 ret = inode_ref_info(inum, parent ? parent+1 : 0, fs_root, path,
1574 &found_key);
1575 if (ret < 0)
1576 break;
1577 if (ret) {
1578 ret = found ? 0 : -ENOENT;
1579 break;
1580 }
1581 ++found;
1582
1583 parent = found_key.offset;
1584 slot = path->slots[0];
1585 eb = path->nodes[0];
1586 /* make sure we can use eb after releasing the path */
1587 atomic_inc(&eb->refs);
1588 btrfs_tree_read_lock(eb);
1589 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1590 btrfs_release_path(path);
1591
1592 item = btrfs_item_nr(eb, slot);
1593 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1594
1595 for (cur = 0; cur < btrfs_item_size(eb, item); cur += len) {
1596 name_len = btrfs_inode_ref_name_len(eb, iref);
1597 /* path must be released before calling iterate()! */
1598 pr_debug("following ref at offset %u for inode %llu in "
1599 "tree %llu\n", cur,
1600 (unsigned long long)found_key.objectid,
1601 (unsigned long long)fs_root->objectid);
1602 ret = iterate(parent, name_len,
1603 (unsigned long)(iref + 1), eb, ctx);
1604 if (ret)
1605 break;
1606 len = sizeof(*iref) + name_len;
1607 iref = (struct btrfs_inode_ref *)((char *)iref + len);
1608 }
1609 btrfs_tree_read_unlock_blocking(eb);
1610 free_extent_buffer(eb);
1611 }
1612
1613 btrfs_release_path(path);
1614
1615 return ret;
1616 }
1617
1618 static int iterate_inode_extrefs(u64 inum, struct btrfs_root *fs_root,
1619 struct btrfs_path *path,
1620 iterate_irefs_t *iterate, void *ctx)
1621 {
1622 int ret;
1623 int slot;
1624 u64 offset = 0;
1625 u64 parent;
1626 int found = 0;
1627 struct extent_buffer *eb;
1628 struct btrfs_inode_extref *extref;
1629 struct extent_buffer *leaf;
1630 u32 item_size;
1631 u32 cur_offset;
1632 unsigned long ptr;
1633
1634 while (1) {
1635 ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref,
1636 &offset);
1637 if (ret < 0)
1638 break;
1639 if (ret) {
1640 ret = found ? 0 : -ENOENT;
1641 break;
1642 }
1643 ++found;
1644
1645 slot = path->slots[0];
1646 eb = path->nodes[0];
1647 /* make sure we can use eb after releasing the path */
1648 atomic_inc(&eb->refs);
1649
1650 btrfs_tree_read_lock(eb);
1651 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1652 btrfs_release_path(path);
1653
1654 leaf = path->nodes[0];
1655 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1656 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1657 cur_offset = 0;
1658
1659 while (cur_offset < item_size) {
1660 u32 name_len;
1661
1662 extref = (struct btrfs_inode_extref *)(ptr + cur_offset);
1663 parent = btrfs_inode_extref_parent(eb, extref);
1664 name_len = btrfs_inode_extref_name_len(eb, extref);
1665 ret = iterate(parent, name_len,
1666 (unsigned long)&extref->name, eb, ctx);
1667 if (ret)
1668 break;
1669
1670 cur_offset += btrfs_inode_extref_name_len(leaf, extref);
1671 cur_offset += sizeof(*extref);
1672 }
1673 btrfs_tree_read_unlock_blocking(eb);
1674 free_extent_buffer(eb);
1675
1676 offset++;
1677 }
1678
1679 btrfs_release_path(path);
1680
1681 return ret;
1682 }
1683
1684 static int iterate_irefs(u64 inum, struct btrfs_root *fs_root,
1685 struct btrfs_path *path, iterate_irefs_t *iterate,
1686 void *ctx)
1687 {
1688 int ret;
1689 int found_refs = 0;
1690
1691 ret = iterate_inode_refs(inum, fs_root, path, iterate, ctx);
1692 if (!ret)
1693 ++found_refs;
1694 else if (ret != -ENOENT)
1695 return ret;
1696
1697 ret = iterate_inode_extrefs(inum, fs_root, path, iterate, ctx);
1698 if (ret == -ENOENT && found_refs)
1699 return 0;
1700
1701 return ret;
1702 }
1703
1704 /*
1705 * returns 0 if the path could be dumped (probably truncated)
1706 * returns <0 in case of an error
1707 */
1708 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
1709 struct extent_buffer *eb, void *ctx)
1710 {
1711 struct inode_fs_paths *ipath = ctx;
1712 char *fspath;
1713 char *fspath_min;
1714 int i = ipath->fspath->elem_cnt;
1715 const int s_ptr = sizeof(char *);
1716 u32 bytes_left;
1717
1718 bytes_left = ipath->fspath->bytes_left > s_ptr ?
1719 ipath->fspath->bytes_left - s_ptr : 0;
1720
1721 fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
1722 fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len,
1723 name_off, eb, inum, fspath_min, bytes_left);
1724 if (IS_ERR(fspath))
1725 return PTR_ERR(fspath);
1726
1727 if (fspath > fspath_min) {
1728 ipath->fspath->val[i] = (u64)(unsigned long)fspath;
1729 ++ipath->fspath->elem_cnt;
1730 ipath->fspath->bytes_left = fspath - fspath_min;
1731 } else {
1732 ++ipath->fspath->elem_missed;
1733 ipath->fspath->bytes_missing += fspath_min - fspath;
1734 ipath->fspath->bytes_left = 0;
1735 }
1736
1737 return 0;
1738 }
1739
1740 /*
1741 * this dumps all file system paths to the inode into the ipath struct, provided
1742 * is has been created large enough. each path is zero-terminated and accessed
1743 * from ipath->fspath->val[i].
1744 * when it returns, there are ipath->fspath->elem_cnt number of paths available
1745 * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the
1746 * number of missed paths in recored in ipath->fspath->elem_missed, otherwise,
1747 * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
1748 * have been needed to return all paths.
1749 */
1750 int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
1751 {
1752 return iterate_irefs(inum, ipath->fs_root, ipath->btrfs_path,
1753 inode_to_path, ipath);
1754 }
1755
1756 struct btrfs_data_container *init_data_container(u32 total_bytes)
1757 {
1758 struct btrfs_data_container *data;
1759 size_t alloc_bytes;
1760
1761 alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
1762 data = vmalloc(alloc_bytes);
1763 if (!data)
1764 return ERR_PTR(-ENOMEM);
1765
1766 if (total_bytes >= sizeof(*data)) {
1767 data->bytes_left = total_bytes - sizeof(*data);
1768 data->bytes_missing = 0;
1769 } else {
1770 data->bytes_missing = sizeof(*data) - total_bytes;
1771 data->bytes_left = 0;
1772 }
1773
1774 data->elem_cnt = 0;
1775 data->elem_missed = 0;
1776
1777 return data;
1778 }
1779
1780 /*
1781 * allocates space to return multiple file system paths for an inode.
1782 * total_bytes to allocate are passed, note that space usable for actual path
1783 * information will be total_bytes - sizeof(struct inode_fs_paths).
1784 * the returned pointer must be freed with free_ipath() in the end.
1785 */
1786 struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
1787 struct btrfs_path *path)
1788 {
1789 struct inode_fs_paths *ifp;
1790 struct btrfs_data_container *fspath;
1791
1792 fspath = init_data_container(total_bytes);
1793 if (IS_ERR(fspath))
1794 return (void *)fspath;
1795
1796 ifp = kmalloc(sizeof(*ifp), GFP_NOFS);
1797 if (!ifp) {
1798 kfree(fspath);
1799 return ERR_PTR(-ENOMEM);
1800 }
1801
1802 ifp->btrfs_path = path;
1803 ifp->fspath = fspath;
1804 ifp->fs_root = fs_root;
1805
1806 return ifp;
1807 }
1808
1809 void free_ipath(struct inode_fs_paths *ipath)
1810 {
1811 if (!ipath)
1812 return;
1813 vfree(ipath->fspath);
1814 kfree(ipath);
1815 }
Linux kernel - Deliverables
This page took 0.083985 seconds and 5 git commands to generate.