?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 3.12.28
[linux/fpc-iii.git] / fs / btrfs / backref.c
blob1f4ce7ac144d516d6946c323b5c1439dbb4b5e7f
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 offset = 0;
40 struct extent_inode_elem *e;
41
42 if (!btrfs_file_extent_compression(eb, fi) &&
43 !btrfs_file_extent_encryption(eb, fi) &&
44 !btrfs_file_extent_other_encoding(eb, fi)) {
45 u64 data_offset;
46 u64 data_len;
47
48 data_offset = btrfs_file_extent_offset(eb, fi);
49 data_len = btrfs_file_extent_num_bytes(eb, fi);
50
51 if (extent_item_pos < data_offset ||
52 extent_item_pos >= data_offset + data_len)
53 return 1;
54 offset = extent_item_pos - data_offset;
55 }
56
57 e = kmalloc(sizeof(*e), GFP_NOFS);
58 if (!e)
59 return -ENOMEM;
60
61 e->next = *eie;
62 e->inum = key->objectid;
63 e->offset = key->offset + offset;
64 *eie = e;
65
66 return 0;
67 }
68
69 static int find_extent_in_eb(struct extent_buffer *eb, u64 wanted_disk_byte,
70 u64 extent_item_pos,
71 struct extent_inode_elem **eie)
72 {
73 u64 disk_byte;
74 struct btrfs_key key;
75 struct btrfs_file_extent_item *fi;
76 int slot;
77 int nritems;
78 int extent_type;
79 int ret;
80
81 /*
82 * from the shared data ref, we only have the leaf but we need
83 * the key. thus, we must look into all items and see that we
84 * find one (some) with a reference to our extent item.
85 */
86 nritems = btrfs_header_nritems(eb);
87 for (slot = 0; slot < nritems; ++slot) {
88 btrfs_item_key_to_cpu(eb, &key, slot);
89 if (key.type != BTRFS_EXTENT_DATA_KEY)
90 continue;
91 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
92 extent_type = btrfs_file_extent_type(eb, fi);
93 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
94 continue;
95 /* don't skip BTRFS_FILE_EXTENT_PREALLOC, we can handle that */
96 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
97 if (disk_byte != wanted_disk_byte)
98 continue;
99
100 ret = check_extent_in_eb(&key, eb, fi, extent_item_pos, eie);
101 if (ret < 0)
102 return ret;
103 }
104
105 return 0;
106 }
107
108 /*
109 * this structure records all encountered refs on the way up to the root
110 */
111 struct __prelim_ref {
112 struct list_head list;
113 u64 root_id;
114 struct btrfs_key key_for_search;
115 int level;
116 int count;
117 struct extent_inode_elem *inode_list;
118 u64 parent;
119 u64 wanted_disk_byte;
120 };
121
122 static struct kmem_cache *btrfs_prelim_ref_cache;
123
124 int __init btrfs_prelim_ref_init(void)
125 {
126 btrfs_prelim_ref_cache = kmem_cache_create("btrfs_prelim_ref",
127 sizeof(struct __prelim_ref),
128 0,
129 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
130 NULL);
131 if (!btrfs_prelim_ref_cache)
132 return -ENOMEM;
133 return 0;
134 }
135
136 void btrfs_prelim_ref_exit(void)
137 {
138 if (btrfs_prelim_ref_cache)
139 kmem_cache_destroy(btrfs_prelim_ref_cache);
140 }
141
142 /*
143 * the rules for all callers of this function are:
144 * - obtaining the parent is the goal
145 * - if you add a key, you must know that it is a correct key
146 * - if you cannot add the parent or a correct key, then we will look into the
147 * block later to set a correct key
148 *
149 * delayed refs
150 * ============
151 * backref type | shared | indirect | shared | indirect
152 * information | tree | tree | data | data
153 * --------------------+--------+----------+--------+----------
154 * parent logical | y | - | - | -
155 * key to resolve | - | y | y | y
156 * tree block logical | - | - | - | -
157 * root for resolving | y | y | y | y
158 *
159 * - column 1: we've the parent -> done
160 * - column 2, 3, 4: we use the key to find the parent
161 *
162 * on disk refs (inline or keyed)
163 * ==============================
164 * backref type | shared | indirect | shared | indirect
165 * information | tree | tree | data | data
166 * --------------------+--------+----------+--------+----------
167 * parent logical | y | - | y | -
168 * key to resolve | - | - | - | y
169 * tree block logical | y | y | y | y
170 * root for resolving | - | y | y | y
171 *
172 * - column 1, 3: we've the parent -> done
173 * - column 2: we take the first key from the block to find the parent
174 * (see __add_missing_keys)
175 * - column 4: we use the key to find the parent
176 *
177 * additional information that's available but not required to find the parent
178 * block might help in merging entries to gain some speed.
179 */
180
181 static int __add_prelim_ref(struct list_head *head, u64 root_id,
182 struct btrfs_key *key, int level,
183 u64 parent, u64 wanted_disk_byte, int count,
184 gfp_t gfp_mask)
185 {
186 struct __prelim_ref *ref;
187
188 if (root_id == BTRFS_DATA_RELOC_TREE_OBJECTID)
189 return 0;
190
191 ref = kmem_cache_alloc(btrfs_prelim_ref_cache, gfp_mask);
192 if (!ref)
193 return -ENOMEM;
194
195 ref->root_id = root_id;
196 if (key)
197 ref->key_for_search = *key;
198 else
199 memset(&ref->key_for_search, 0, sizeof(ref->key_for_search));
200
201 ref->inode_list = NULL;
202 ref->level = level;
203 ref->count = count;
204 ref->parent = parent;
205 ref->wanted_disk_byte = wanted_disk_byte;
206 list_add_tail(&ref->list, head);
207
208 return 0;
209 }
210
211 static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path,
212 struct ulist *parents, int level,
213 struct btrfs_key *key_for_search, u64 time_seq,
214 u64 wanted_disk_byte,
215 const u64 *extent_item_pos)
216 {
217 int ret = 0;
218 int slot;
219 struct extent_buffer *eb;
220 struct btrfs_key key;
221 struct btrfs_file_extent_item *fi;
222 struct extent_inode_elem *eie = NULL, *old = NULL;
223 u64 disk_byte;
224
225 if (level != 0) {
226 eb = path->nodes[level];
227 ret = ulist_add(parents, eb->start, 0, GFP_NOFS);
228 if (ret < 0)
229 return ret;
230 return 0;
231 }
232
233 /*
234 * We normally enter this function with the path already pointing to
235 * the first item to check. But sometimes, we may enter it with
236 * slot==nritems. In that case, go to the next leaf before we continue.
237 */
238 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0]))
239 ret = btrfs_next_old_leaf(root, path, time_seq);
240
241 while (!ret) {
242 eb = path->nodes[0];
243 slot = path->slots[0];
244
245 btrfs_item_key_to_cpu(eb, &key, slot);
246
247 if (key.objectid != key_for_search->objectid ||
248 key.type != BTRFS_EXTENT_DATA_KEY)
249 break;
250
251 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
252 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
253
254 if (disk_byte == wanted_disk_byte) {
255 eie = NULL;
256 old = NULL;
257 if (extent_item_pos) {
258 ret = check_extent_in_eb(&key, eb, fi,
259 *extent_item_pos,
260 &eie);
261 if (ret < 0)
262 break;
263 }
264 if (ret > 0)
265 goto next;
266 ret = ulist_add_merge(parents, eb->start,
267 (uintptr_t)eie,
268 (u64 *)&old, GFP_NOFS);
269 if (ret < 0)
270 break;
271 if (!ret && extent_item_pos) {
272 while (old->next)
273 old = old->next;
274 old->next = eie;
275 }
276 }
277 next:
278 ret = btrfs_next_old_item(root, path, time_seq);
279 }
280
281 if (ret > 0)
282 ret = 0;
283 return ret;
284 }
285
286 /*
287 * resolve an indirect backref in the form (root_id, key, level)
288 * to a logical address
289 */
290 static int __resolve_indirect_ref(struct btrfs_fs_info *fs_info,
291 struct btrfs_path *path, u64 time_seq,
292 struct __prelim_ref *ref,
293 struct ulist *parents,
294 const u64 *extent_item_pos)
295 {
296 struct btrfs_root *root;
297 struct btrfs_key root_key;
298 struct extent_buffer *eb;
299 int ret = 0;
300 int root_level;
301 int level = ref->level;
302
303 root_key.objectid = ref->root_id;
304 root_key.type = BTRFS_ROOT_ITEM_KEY;
305 root_key.offset = (u64)-1;
306 root = btrfs_read_fs_root_no_name(fs_info, &root_key);
307 if (IS_ERR(root)) {
308 ret = PTR_ERR(root);
309 goto out;
310 }
311
312 root_level = btrfs_old_root_level(root, time_seq);
313
314 if (root_level + 1 == level)
315 goto out;
316
317 path->lowest_level = level;
318 ret = btrfs_search_old_slot(root, &ref->key_for_search, path, time_seq);
319 pr_debug("search slot in root %llu (level %d, ref count %d) returned "
320 "%d for key (%llu %u %llu)\n",
321 ref->root_id, level, ref->count, ret,
322 ref->key_for_search.objectid, ref->key_for_search.type,
323 ref->key_for_search.offset);
324 if (ret < 0)
325 goto out;
326
327 eb = path->nodes[level];
328 while (!eb) {
329 if (!level) {
330 WARN_ON(1);
331 ret = 1;
332 goto out;
333 }
334 level--;
335 eb = path->nodes[level];
336 }
337
338 ret = add_all_parents(root, path, parents, level, &ref->key_for_search,
339 time_seq, ref->wanted_disk_byte,
340 extent_item_pos);
341 out:
342 path->lowest_level = 0;
343 btrfs_release_path(path);
344 return ret;
345 }
346
347 /*
348 * resolve all indirect backrefs from the list
349 */
350 static int __resolve_indirect_refs(struct btrfs_fs_info *fs_info,
351 struct btrfs_path *path, u64 time_seq,
352 struct list_head *head,
353 const u64 *extent_item_pos)
354 {
355 int err;
356 int ret = 0;
357 struct __prelim_ref *ref;
358 struct __prelim_ref *ref_safe;
359 struct __prelim_ref *new_ref;
360 struct ulist *parents;
361 struct ulist_node *node;
362 struct ulist_iterator uiter;
363
364 parents = ulist_alloc(GFP_NOFS);
365 if (!parents)
366 return -ENOMEM;
367
368 /*
369 * _safe allows us to insert directly after the current item without
370 * iterating over the newly inserted items.
371 * we're also allowed to re-assign ref during iteration.
372 */
373 list_for_each_entry_safe(ref, ref_safe, head, list) {
374 if (ref->parent) /* already direct */
375 continue;
376 if (ref->count == 0)
377 continue;
378 err = __resolve_indirect_ref(fs_info, path, time_seq, ref,
379 parents, extent_item_pos);
380 if (err == -ENOMEM)
381 goto out;
382 if (err)
383 continue;
384
385 /* we put the first parent into the ref at hand */
386 ULIST_ITER_INIT(&uiter);
387 node = ulist_next(parents, &uiter);
388 ref->parent = node ? node->val : 0;
389 ref->inode_list = node ?
390 (struct extent_inode_elem *)(uintptr_t)node->aux : NULL;
391
392 /* additional parents require new refs being added here */
393 while ((node = ulist_next(parents, &uiter))) {
394 new_ref = kmem_cache_alloc(btrfs_prelim_ref_cache,
395 GFP_NOFS);
396 if (!new_ref) {
397 ret = -ENOMEM;
398 goto out;
399 }
400 memcpy(new_ref, ref, sizeof(*ref));
401 new_ref->parent = node->val;
402 new_ref->inode_list = (struct extent_inode_elem *)
403 (uintptr_t)node->aux;
404 list_add(&new_ref->list, &ref->list);
405 }
406 ulist_reinit(parents);
407 }
408 out:
409 ulist_free(parents);
410 return ret;
411 }
412
413 static inline int ref_for_same_block(struct __prelim_ref *ref1,
414 struct __prelim_ref *ref2)
415 {
416 if (ref1->level != ref2->level)
417 return 0;
418 if (ref1->root_id != ref2->root_id)
419 return 0;
420 if (ref1->key_for_search.type != ref2->key_for_search.type)
421 return 0;
422 if (ref1->key_for_search.objectid != ref2->key_for_search.objectid)
423 return 0;
424 if (ref1->key_for_search.offset != ref2->key_for_search.offset)
425 return 0;
426 if (ref1->parent != ref2->parent)
427 return 0;
428
429 return 1;
430 }
431
432 /*
433 * read tree blocks and add keys where required.
434 */
435 static int __add_missing_keys(struct btrfs_fs_info *fs_info,
436 struct list_head *head)
437 {
438 struct list_head *pos;
439 struct extent_buffer *eb;
440
441 list_for_each(pos, head) {
442 struct __prelim_ref *ref;
443 ref = list_entry(pos, struct __prelim_ref, list);
444
445 if (ref->parent)
446 continue;
447 if (ref->key_for_search.type)
448 continue;
449 BUG_ON(!ref->wanted_disk_byte);
450 eb = read_tree_block(fs_info->tree_root, ref->wanted_disk_byte,
451 fs_info->tree_root->leafsize, 0);
452 if (!eb || !extent_buffer_uptodate(eb)) {
453 free_extent_buffer(eb);
454 return -EIO;
455 }
456 btrfs_tree_read_lock(eb);
457 if (btrfs_header_level(eb) == 0)
458 btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
459 else
460 btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
461 btrfs_tree_read_unlock(eb);
462 free_extent_buffer(eb);
463 }
464 return 0;
465 }
466
467 /*
468 * merge two lists of backrefs and adjust counts accordingly
469 *
470 * mode = 1: merge identical keys, if key is set
471 * FIXME: if we add more keys in __add_prelim_ref, we can merge more here.
472 * additionally, we could even add a key range for the blocks we
473 * looked into to merge even more (-> replace unresolved refs by those
474 * having a parent).
475 * mode = 2: merge identical parents
476 */
477 static void __merge_refs(struct list_head *head, int mode)
478 {
479 struct list_head *pos1;
480
481 list_for_each(pos1, head) {
482 struct list_head *n2;
483 struct list_head *pos2;
484 struct __prelim_ref *ref1;
485
486 ref1 = list_entry(pos1, struct __prelim_ref, list);
487
488 for (pos2 = pos1->next, n2 = pos2->next; pos2 != head;
489 pos2 = n2, n2 = pos2->next) {
490 struct __prelim_ref *ref2;
491 struct __prelim_ref *xchg;
492 struct extent_inode_elem *eie;
493
494 ref2 = list_entry(pos2, struct __prelim_ref, list);
495
496 if (mode == 1) {
497 if (!ref_for_same_block(ref1, ref2))
498 continue;
499 if (!ref1->parent && ref2->parent) {
500 xchg = ref1;
501 ref1 = ref2;
502 ref2 = xchg;
503 }
504 } else {
505 if (ref1->parent != ref2->parent)
506 continue;
507 }
508
509 eie = ref1->inode_list;
510 while (eie && eie->next)
511 eie = eie->next;
512 if (eie)
513 eie->next = ref2->inode_list;
514 else
515 ref1->inode_list = ref2->inode_list;
516 ref1->count += ref2->count;
517
518 list_del(&ref2->list);
519 kmem_cache_free(btrfs_prelim_ref_cache, ref2);
520 }
521
522 }
523 }
524
525 /*
526 * add all currently queued delayed refs from this head whose seq nr is
527 * smaller or equal that seq to the list
528 */
529 static int __add_delayed_refs(struct btrfs_delayed_ref_head *head, u64 seq,
530 struct list_head *prefs)
531 {
532 struct btrfs_delayed_extent_op *extent_op = head->extent_op;
533 struct rb_node *n = &head->node.rb_node;
534 struct btrfs_key key;
535 struct btrfs_key op_key = {0};
536 int sgn;
537 int ret = 0;
538
539 if (extent_op && extent_op->update_key)
540 btrfs_disk_key_to_cpu(&op_key, &extent_op->key);
541
542 while ((n = rb_prev(n))) {
543 struct btrfs_delayed_ref_node *node;
544 node = rb_entry(n, struct btrfs_delayed_ref_node,
545 rb_node);
546 if (node->bytenr != head->node.bytenr)
547 break;
548 WARN_ON(node->is_head);
549
550 if (node->seq > seq)
551 continue;
552
553 switch (node->action) {
554 case BTRFS_ADD_DELAYED_EXTENT:
555 case BTRFS_UPDATE_DELAYED_HEAD:
556 WARN_ON(1);
557 continue;
558 case BTRFS_ADD_DELAYED_REF:
559 sgn = 1;
560 break;
561 case BTRFS_DROP_DELAYED_REF:
562 sgn = -1;
563 break;
564 default:
565 BUG_ON(1);
566 }
567 switch (node->type) {
568 case BTRFS_TREE_BLOCK_REF_KEY: {
569 struct btrfs_delayed_tree_ref *ref;
570
571 ref = btrfs_delayed_node_to_tree_ref(node);
572 ret = __add_prelim_ref(prefs, ref->root, &op_key,
573 ref->level + 1, 0, node->bytenr,
574 node->ref_mod * sgn, GFP_ATOMIC);
575 break;
576 }
577 case BTRFS_SHARED_BLOCK_REF_KEY: {
578 struct btrfs_delayed_tree_ref *ref;
579
580 ref = btrfs_delayed_node_to_tree_ref(node);
581 ret = __add_prelim_ref(prefs, ref->root, NULL,
582 ref->level + 1, ref->parent,
583 node->bytenr,
584 node->ref_mod * sgn, GFP_ATOMIC);
585 break;
586 }
587 case BTRFS_EXTENT_DATA_REF_KEY: {
588 struct btrfs_delayed_data_ref *ref;
589 ref = btrfs_delayed_node_to_data_ref(node);
590
591 key.objectid = ref->objectid;
592 key.type = BTRFS_EXTENT_DATA_KEY;
593 key.offset = ref->offset;
594 ret = __add_prelim_ref(prefs, ref->root, &key, 0, 0,
595 node->bytenr,
596 node->ref_mod * sgn, GFP_ATOMIC);
597 break;
598 }
599 case BTRFS_SHARED_DATA_REF_KEY: {
600 struct btrfs_delayed_data_ref *ref;
601
602 ref = btrfs_delayed_node_to_data_ref(node);
603
604 key.objectid = ref->objectid;
605 key.type = BTRFS_EXTENT_DATA_KEY;
606 key.offset = ref->offset;
607 ret = __add_prelim_ref(prefs, ref->root, &key, 0,
608 ref->parent, node->bytenr,
609 node->ref_mod * sgn, GFP_ATOMIC);
610 break;
611 }
612 default:
613 WARN_ON(1);
614 }
615 if (ret)
616 return ret;
617 }
618
619 return 0;
620 }
621
622 /*
623 * add all inline backrefs for bytenr to the list
624 */
625 static int __add_inline_refs(struct btrfs_fs_info *fs_info,
626 struct btrfs_path *path, u64 bytenr,
627 int *info_level, struct list_head *prefs)
628 {
629 int ret = 0;
630 int slot;
631 struct extent_buffer *leaf;
632 struct btrfs_key key;
633 struct btrfs_key found_key;
634 unsigned long ptr;
635 unsigned long end;
636 struct btrfs_extent_item *ei;
637 u64 flags;
638 u64 item_size;
639
640 /*
641 * enumerate all inline refs
642 */
643 leaf = path->nodes[0];
644 slot = path->slots[0];
645
646 item_size = btrfs_item_size_nr(leaf, slot);
647 BUG_ON(item_size < sizeof(*ei));
648
649 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
650 flags = btrfs_extent_flags(leaf, ei);
651 btrfs_item_key_to_cpu(leaf, &found_key, slot);
652
653 ptr = (unsigned long)(ei + 1);
654 end = (unsigned long)ei + item_size;
655
656 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
657 flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
658 struct btrfs_tree_block_info *info;
659
660 info = (struct btrfs_tree_block_info *)ptr;
661 *info_level = btrfs_tree_block_level(leaf, info);
662 ptr += sizeof(struct btrfs_tree_block_info);
663 BUG_ON(ptr > end);
664 } else if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
665 *info_level = found_key.offset;
666 } else {
667 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
668 }
669
670 while (ptr < end) {
671 struct btrfs_extent_inline_ref *iref;
672 u64 offset;
673 int type;
674
675 iref = (struct btrfs_extent_inline_ref *)ptr;
676 type = btrfs_extent_inline_ref_type(leaf, iref);
677 offset = btrfs_extent_inline_ref_offset(leaf, iref);
678
679 switch (type) {
680 case BTRFS_SHARED_BLOCK_REF_KEY:
681 ret = __add_prelim_ref(prefs, 0, NULL,
682 *info_level + 1, offset,
683 bytenr, 1, GFP_NOFS);
684 break;
685 case BTRFS_SHARED_DATA_REF_KEY: {
686 struct btrfs_shared_data_ref *sdref;
687 int count;
688
689 sdref = (struct btrfs_shared_data_ref *)(iref + 1);
690 count = btrfs_shared_data_ref_count(leaf, sdref);
691 ret = __add_prelim_ref(prefs, 0, NULL, 0, offset,
692 bytenr, count, GFP_NOFS);
693 break;
694 }
695 case BTRFS_TREE_BLOCK_REF_KEY:
696 ret = __add_prelim_ref(prefs, offset, NULL,
697 *info_level + 1, 0,
698 bytenr, 1, GFP_NOFS);
699 break;
700 case BTRFS_EXTENT_DATA_REF_KEY: {
701 struct btrfs_extent_data_ref *dref;
702 int count;
703 u64 root;
704
705 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
706 count = btrfs_extent_data_ref_count(leaf, dref);
707 key.objectid = btrfs_extent_data_ref_objectid(leaf,
708 dref);
709 key.type = BTRFS_EXTENT_DATA_KEY;
710 key.offset = btrfs_extent_data_ref_offset(leaf, dref);
711 root = btrfs_extent_data_ref_root(leaf, dref);
712 ret = __add_prelim_ref(prefs, root, &key, 0, 0,
713 bytenr, count, GFP_NOFS);
714 break;
715 }
716 default:
717 WARN_ON(1);
718 }
719 if (ret)
720 return ret;
721 ptr += btrfs_extent_inline_ref_size(type);
722 }
723
724 return 0;
725 }
726
727 /*
728 * add all non-inline backrefs for bytenr to the list
729 */
730 static int __add_keyed_refs(struct btrfs_fs_info *fs_info,
731 struct btrfs_path *path, u64 bytenr,
732 int info_level, struct list_head *prefs)
733 {
734 struct btrfs_root *extent_root = fs_info->extent_root;
735 int ret;
736 int slot;
737 struct extent_buffer *leaf;
738 struct btrfs_key key;
739
740 while (1) {
741 ret = btrfs_next_item(extent_root, path);
742 if (ret < 0)
743 break;
744 if (ret) {
745 ret = 0;
746 break;
747 }
748
749 slot = path->slots[0];
750 leaf = path->nodes[0];
751 btrfs_item_key_to_cpu(leaf, &key, slot);
752
753 if (key.objectid != bytenr)
754 break;
755 if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
756 continue;
757 if (key.type > BTRFS_SHARED_DATA_REF_KEY)
758 break;
759
760 switch (key.type) {
761 case BTRFS_SHARED_BLOCK_REF_KEY:
762 ret = __add_prelim_ref(prefs, 0, NULL,
763 info_level + 1, key.offset,
764 bytenr, 1, GFP_NOFS);
765 break;
766 case BTRFS_SHARED_DATA_REF_KEY: {
767 struct btrfs_shared_data_ref *sdref;
768 int count;
769
770 sdref = btrfs_item_ptr(leaf, slot,
771 struct btrfs_shared_data_ref);
772 count = btrfs_shared_data_ref_count(leaf, sdref);
773 ret = __add_prelim_ref(prefs, 0, NULL, 0, key.offset,
774 bytenr, count, GFP_NOFS);
775 break;
776 }
777 case BTRFS_TREE_BLOCK_REF_KEY:
778 ret = __add_prelim_ref(prefs, key.offset, NULL,
779 info_level + 1, 0,
780 bytenr, 1, GFP_NOFS);
781 break;
782 case BTRFS_EXTENT_DATA_REF_KEY: {
783 struct btrfs_extent_data_ref *dref;
784 int count;
785 u64 root;
786
787 dref = btrfs_item_ptr(leaf, slot,
788 struct btrfs_extent_data_ref);
789 count = btrfs_extent_data_ref_count(leaf, dref);
790 key.objectid = btrfs_extent_data_ref_objectid(leaf,
791 dref);
792 key.type = BTRFS_EXTENT_DATA_KEY;
793 key.offset = btrfs_extent_data_ref_offset(leaf, dref);
794 root = btrfs_extent_data_ref_root(leaf, dref);
795 ret = __add_prelim_ref(prefs, root, &key, 0, 0,
796 bytenr, count, GFP_NOFS);
797 break;
798 }
799 default:
800 WARN_ON(1);
801 }
802 if (ret)
803 return ret;
804
805 }
806
807 return ret;
808 }
809
810 /*
811 * this adds all existing backrefs (inline backrefs, backrefs and delayed
812 * refs) for the given bytenr to the refs list, merges duplicates and resolves
813 * indirect refs to their parent bytenr.
814 * When roots are found, they're added to the roots list
815 *
816 * FIXME some caching might speed things up
817 */
818 static int find_parent_nodes(struct btrfs_trans_handle *trans,
819 struct btrfs_fs_info *fs_info, u64 bytenr,
820 u64 time_seq, struct ulist *refs,
821 struct ulist *roots, const u64 *extent_item_pos)
822 {
823 struct btrfs_key key;
824 struct btrfs_path *path;
825 struct btrfs_delayed_ref_root *delayed_refs = NULL;
826 struct btrfs_delayed_ref_head *head;
827 int info_level = 0;
828 int ret;
829 struct list_head prefs_delayed;
830 struct list_head prefs;
831 struct __prelim_ref *ref;
832
833 INIT_LIST_HEAD(&prefs);
834 INIT_LIST_HEAD(&prefs_delayed);
835
836 key.objectid = bytenr;
837 key.offset = (u64)-1;
838 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
839 key.type = BTRFS_METADATA_ITEM_KEY;
840 else
841 key.type = BTRFS_EXTENT_ITEM_KEY;
842
843 path = btrfs_alloc_path();
844 if (!path)
845 return -ENOMEM;
846 if (!trans)
847 path->search_commit_root = 1;
848
849 /*
850 * grab both a lock on the path and a lock on the delayed ref head.
851 * We need both to get a consistent picture of how the refs look
852 * at a specified point in time
853 */
854 again:
855 head = NULL;
856
857 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
858 if (ret < 0)
859 goto out;
860 BUG_ON(ret == 0);
861
862 if (trans) {
863 /*
864 * look if there are updates for this ref queued and lock the
865 * head
866 */
867 delayed_refs = &trans->transaction->delayed_refs;
868 spin_lock(&delayed_refs->lock);
869 head = btrfs_find_delayed_ref_head(trans, bytenr);
870 if (head) {
871 if (!mutex_trylock(&head->mutex)) {
872 atomic_inc(&head->node.refs);
873 spin_unlock(&delayed_refs->lock);
874
875 btrfs_release_path(path);
876
877 /*
878 * Mutex was contended, block until it's
879 * released and try again
880 */
881 mutex_lock(&head->mutex);
882 mutex_unlock(&head->mutex);
883 btrfs_put_delayed_ref(&head->node);
884 goto again;
885 }
886 ret = __add_delayed_refs(head, time_seq,
887 &prefs_delayed);
888 mutex_unlock(&head->mutex);
889 if (ret) {
890 spin_unlock(&delayed_refs->lock);
891 goto out;
892 }
893 }
894 spin_unlock(&delayed_refs->lock);
895 }
896
897 if (path->slots[0]) {
898 struct extent_buffer *leaf;
899 int slot;
900
901 path->slots[0]--;
902 leaf = path->nodes[0];
903 slot = path->slots[0];
904 btrfs_item_key_to_cpu(leaf, &key, slot);
905 if (key.objectid == bytenr &&
906 (key.type == BTRFS_EXTENT_ITEM_KEY ||
907 key.type == BTRFS_METADATA_ITEM_KEY)) {
908 ret = __add_inline_refs(fs_info, path, bytenr,
909 &info_level, &prefs);
910 if (ret)
911 goto out;
912 ret = __add_keyed_refs(fs_info, path, bytenr,
913 info_level, &prefs);
914 if (ret)
915 goto out;
916 }
917 }
918 btrfs_release_path(path);
919
920 list_splice_init(&prefs_delayed, &prefs);
921
922 ret = __add_missing_keys(fs_info, &prefs);
923 if (ret)
924 goto out;
925
926 __merge_refs(&prefs, 1);
927
928 ret = __resolve_indirect_refs(fs_info, path, time_seq, &prefs,
929 extent_item_pos);
930 if (ret)
931 goto out;
932
933 __merge_refs(&prefs, 2);
934
935 while (!list_empty(&prefs)) {
936 ref = list_first_entry(&prefs, struct __prelim_ref, list);
937 WARN_ON(ref->count < 0);
938 if (ref->count && ref->root_id && ref->parent == 0) {
939 /* no parent == root of tree */
940 ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS);
941 if (ret < 0)
942 goto out;
943 }
944 if (ref->count && ref->parent) {
945 struct extent_inode_elem *eie = NULL;
946 if (extent_item_pos && !ref->inode_list) {
947 u32 bsz;
948 struct extent_buffer *eb;
949 bsz = btrfs_level_size(fs_info->extent_root,
950 info_level);
951 eb = read_tree_block(fs_info->extent_root,
952 ref->parent, bsz, 0);
953 if (!eb || !extent_buffer_uptodate(eb)) {
954 free_extent_buffer(eb);
955 ret = -EIO;
956 goto out;
957 }
958 ret = find_extent_in_eb(eb, bytenr,
959 *extent_item_pos, &eie);
960 free_extent_buffer(eb);
961 if (ret < 0)
962 goto out;
963 ref->inode_list = eie;
964 }
965 ret = ulist_add_merge(refs, ref->parent,
966 (uintptr_t)ref->inode_list,
967 (u64 *)&eie, GFP_NOFS);
968 if (ret < 0)
969 goto out;
970 if (!ret && extent_item_pos) {
971 /*
972 * we've recorded that parent, so we must extend
973 * its inode list here
974 */
975 BUG_ON(!eie);
976 while (eie->next)
977 eie = eie->next;
978 eie->next = ref->inode_list;
979 }
980 }
981 list_del(&ref->list);
982 kmem_cache_free(btrfs_prelim_ref_cache, ref);
983 }
984
985 out:
986 btrfs_free_path(path);
987 while (!list_empty(&prefs)) {
988 ref = list_first_entry(&prefs, struct __prelim_ref, list);
989 list_del(&ref->list);
990 kmem_cache_free(btrfs_prelim_ref_cache, ref);
991 }
992 while (!list_empty(&prefs_delayed)) {
993 ref = list_first_entry(&prefs_delayed, struct __prelim_ref,
994 list);
995 list_del(&ref->list);
996 kmem_cache_free(btrfs_prelim_ref_cache, ref);
997 }
998
999 return ret;
1000 }
1001
1002 static void free_leaf_list(struct ulist *blocks)
1003 {
1004 struct ulist_node *node = NULL;
1005 struct extent_inode_elem *eie;
1006 struct extent_inode_elem *eie_next;
1007 struct ulist_iterator uiter;
1008
1009 ULIST_ITER_INIT(&uiter);
1010 while ((node = ulist_next(blocks, &uiter))) {
1011 if (!node->aux)
1012 continue;
1013 eie = (struct extent_inode_elem *)(uintptr_t)node->aux;
1014 for (; eie; eie = eie_next) {
1015 eie_next = eie->next;
1016 kfree(eie);
1017 }
1018 node->aux = 0;
1019 }
1020
1021 ulist_free(blocks);
1022 }
1023
1024 /*
1025 * Finds all leafs with a reference to the specified combination of bytenr and
1026 * offset. key_list_head will point to a list of corresponding keys (caller must
1027 * free each list element). The leafs will be stored in the leafs ulist, which
1028 * must be freed with ulist_free.
1029 *
1030 * returns 0 on success, <0 on error
1031 */
1032 static int btrfs_find_all_leafs(struct btrfs_trans_handle *trans,
1033 struct btrfs_fs_info *fs_info, u64 bytenr,
1034 u64 time_seq, struct ulist **leafs,
1035 const u64 *extent_item_pos)
1036 {
1037 struct ulist *tmp;
1038 int ret;
1039
1040 tmp = ulist_alloc(GFP_NOFS);
1041 if (!tmp)
1042 return -ENOMEM;
1043 *leafs = ulist_alloc(GFP_NOFS);
1044 if (!*leafs) {
1045 ulist_free(tmp);
1046 return -ENOMEM;
1047 }
1048
1049 ret = find_parent_nodes(trans, fs_info, bytenr,
1050 time_seq, *leafs, tmp, extent_item_pos);
1051 ulist_free(tmp);
1052
1053 if (ret < 0 && ret != -ENOENT) {
1054 free_leaf_list(*leafs);
1055 return ret;
1056 }
1057
1058 return 0;
1059 }
1060
1061 /*
1062 * walk all backrefs for a given extent to find all roots that reference this
1063 * extent. Walking a backref means finding all extents that reference this
1064 * extent and in turn walk the backrefs of those, too. Naturally this is a
1065 * recursive process, but here it is implemented in an iterative fashion: We
1066 * find all referencing extents for the extent in question and put them on a
1067 * list. In turn, we find all referencing extents for those, further appending
1068 * to the list. The way we iterate the list allows adding more elements after
1069 * the current while iterating. The process stops when we reach the end of the
1070 * list. Found roots are added to the roots list.
1071 *
1072 * returns 0 on success, < 0 on error.
1073 */
1074 int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1075 struct btrfs_fs_info *fs_info, u64 bytenr,
1076 u64 time_seq, struct ulist **roots)
1077 {
1078 struct ulist *tmp;
1079 struct ulist_node *node = NULL;
1080 struct ulist_iterator uiter;
1081 int ret;
1082
1083 tmp = ulist_alloc(GFP_NOFS);
1084 if (!tmp)
1085 return -ENOMEM;
1086 *roots = ulist_alloc(GFP_NOFS);
1087 if (!*roots) {
1088 ulist_free(tmp);
1089 return -ENOMEM;
1090 }
1091
1092 ULIST_ITER_INIT(&uiter);
1093 while (1) {
1094 ret = find_parent_nodes(trans, fs_info, bytenr,
1095 time_seq, tmp, *roots, NULL);
1096 if (ret < 0 && ret != -ENOENT) {
1097 ulist_free(tmp);
1098 ulist_free(*roots);
1099 return ret;
1100 }
1101 node = ulist_next(tmp, &uiter);
1102 if (!node)
1103 break;
1104 bytenr = node->val;
1105 }
1106
1107 ulist_free(tmp);
1108 return 0;
1109 }
1110
1111
1112 static int __inode_info(u64 inum, u64 ioff, u8 key_type,
1113 struct btrfs_root *fs_root, struct btrfs_path *path,
1114 struct btrfs_key *found_key)
1115 {
1116 int ret;
1117 struct btrfs_key key;
1118 struct extent_buffer *eb;
1119
1120 key.type = key_type;
1121 key.objectid = inum;
1122 key.offset = ioff;
1123
1124 ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
1125 if (ret < 0)
1126 return ret;
1127
1128 eb = path->nodes[0];
1129 if (ret && path->slots[0] >= btrfs_header_nritems(eb)) {
1130 ret = btrfs_next_leaf(fs_root, path);
1131 if (ret)
1132 return ret;
1133 eb = path->nodes[0];
1134 }
1135
1136 btrfs_item_key_to_cpu(eb, found_key, path->slots[0]);
1137 if (found_key->type != key.type || found_key->objectid != key.objectid)
1138 return 1;
1139
1140 return 0;
1141 }
1142
1143 /*
1144 * this makes the path point to (inum INODE_ITEM ioff)
1145 */
1146 int inode_item_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
1147 struct btrfs_path *path)
1148 {
1149 struct btrfs_key key;
1150 return __inode_info(inum, ioff, BTRFS_INODE_ITEM_KEY, fs_root, path,
1151 &key);
1152 }
1153
1154 static int inode_ref_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
1155 struct btrfs_path *path,
1156 struct btrfs_key *found_key)
1157 {
1158 return __inode_info(inum, ioff, BTRFS_INODE_REF_KEY, fs_root, path,
1159 found_key);
1160 }
1161
1162 int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,
1163 u64 start_off, struct btrfs_path *path,
1164 struct btrfs_inode_extref **ret_extref,
1165 u64 *found_off)
1166 {
1167 int ret, slot;
1168 struct btrfs_key key;
1169 struct btrfs_key found_key;
1170 struct btrfs_inode_extref *extref;
1171 struct extent_buffer *leaf;
1172 unsigned long ptr;
1173
1174 key.objectid = inode_objectid;
1175 btrfs_set_key_type(&key, BTRFS_INODE_EXTREF_KEY);
1176 key.offset = start_off;
1177
1178 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1179 if (ret < 0)
1180 return ret;
1181
1182 while (1) {
1183 leaf = path->nodes[0];
1184 slot = path->slots[0];
1185 if (slot >= btrfs_header_nritems(leaf)) {
1186 /*
1187 * If the item at offset is not found,
1188 * btrfs_search_slot will point us to the slot
1189 * where it should be inserted. In our case
1190 * that will be the slot directly before the
1191 * next INODE_REF_KEY_V2 item. In the case
1192 * that we're pointing to the last slot in a
1193 * leaf, we must move one leaf over.
1194 */
1195 ret = btrfs_next_leaf(root, path);
1196 if (ret) {
1197 if (ret >= 1)
1198 ret = -ENOENT;
1199 break;
1200 }
1201 continue;
1202 }
1203
1204 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1205
1206 /*
1207 * Check that we're still looking at an extended ref key for
1208 * this particular objectid. If we have different
1209 * objectid or type then there are no more to be found
1210 * in the tree and we can exit.
1211 */
1212 ret = -ENOENT;
1213 if (found_key.objectid != inode_objectid)
1214 break;
1215 if (btrfs_key_type(&found_key) != BTRFS_INODE_EXTREF_KEY)
1216 break;
1217
1218 ret = 0;
1219 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1220 extref = (struct btrfs_inode_extref *)ptr;
1221 *ret_extref = extref;
1222 if (found_off)
1223 *found_off = found_key.offset;
1224 break;
1225 }
1226
1227 return ret;
1228 }
1229
1230 /*
1231 * this iterates to turn a name (from iref/extref) into a full filesystem path.
1232 * Elements of the path are separated by '/' and the path is guaranteed to be
1233 * 0-terminated. the path is only given within the current file system.
1234 * Therefore, it never starts with a '/'. the caller is responsible to provide
1235 * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
1236 * the start point of the resulting string is returned. this pointer is within
1237 * dest, normally.
1238 * in case the path buffer would overflow, the pointer is decremented further
1239 * as if output was written to the buffer, though no more output is actually
1240 * generated. that way, the caller can determine how much space would be
1241 * required for the path to fit into the buffer. in that case, the returned
1242 * value will be smaller than dest. callers must check this!
1243 */
1244 char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
1245 u32 name_len, unsigned long name_off,
1246 struct extent_buffer *eb_in, u64 parent,
1247 char *dest, u32 size)
1248 {
1249 int slot;
1250 u64 next_inum;
1251 int ret;
1252 s64 bytes_left = ((s64)size) - 1;
1253 struct extent_buffer *eb = eb_in;
1254 struct btrfs_key found_key;
1255 int leave_spinning = path->leave_spinning;
1256 struct btrfs_inode_ref *iref;
1257
1258 if (bytes_left >= 0)
1259 dest[bytes_left] = '\0';
1260
1261 path->leave_spinning = 1;
1262 while (1) {
1263 bytes_left -= name_len;
1264 if (bytes_left >= 0)
1265 read_extent_buffer(eb, dest + bytes_left,
1266 name_off, name_len);
1267 if (eb != eb_in) {
1268 btrfs_tree_read_unlock_blocking(eb);
1269 free_extent_buffer(eb);
1270 }
1271 ret = inode_ref_info(parent, 0, fs_root, path, &found_key);
1272 if (ret > 0)
1273 ret = -ENOENT;
1274 if (ret)
1275 break;
1276
1277 next_inum = found_key.offset;
1278
1279 /* regular exit ahead */
1280 if (parent == next_inum)
1281 break;
1282
1283 slot = path->slots[0];
1284 eb = path->nodes[0];
1285 /* make sure we can use eb after releasing the path */
1286 if (eb != eb_in) {
1287 atomic_inc(&eb->refs);
1288 btrfs_tree_read_lock(eb);
1289 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1290 }
1291 btrfs_release_path(path);
1292 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1293
1294 name_len = btrfs_inode_ref_name_len(eb, iref);
1295 name_off = (unsigned long)(iref + 1);
1296
1297 parent = next_inum;
1298 --bytes_left;
1299 if (bytes_left >= 0)
1300 dest[bytes_left] = '/';
1301 }
1302
1303 btrfs_release_path(path);
1304 path->leave_spinning = leave_spinning;
1305
1306 if (ret)
1307 return ERR_PTR(ret);
1308
1309 return dest + bytes_left;
1310 }
1311
1312 /*
1313 * this makes the path point to (logical EXTENT_ITEM *)
1314 * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
1315 * tree blocks and <0 on error.
1316 */
1317 int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
1318 struct btrfs_path *path, struct btrfs_key *found_key,
1319 u64 *flags_ret)
1320 {
1321 int ret;
1322 u64 flags;
1323 u64 size = 0;
1324 u32 item_size;
1325 struct extent_buffer *eb;
1326 struct btrfs_extent_item *ei;
1327 struct btrfs_key key;
1328
1329 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1330 key.type = BTRFS_METADATA_ITEM_KEY;
1331 else
1332 key.type = BTRFS_EXTENT_ITEM_KEY;
1333 key.objectid = logical;
1334 key.offset = (u64)-1;
1335
1336 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
1337 if (ret < 0)
1338 return ret;
1339 ret = btrfs_previous_item(fs_info->extent_root, path,
1340 0, BTRFS_EXTENT_ITEM_KEY);
1341 if (ret < 0)
1342 return ret;
1343
1344 btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]);
1345 if (found_key->type == BTRFS_METADATA_ITEM_KEY)
1346 size = fs_info->extent_root->leafsize;
1347 else if (found_key->type == BTRFS_EXTENT_ITEM_KEY)
1348 size = found_key->offset;
1349
1350 if ((found_key->type != BTRFS_EXTENT_ITEM_KEY &&
1351 found_key->type != BTRFS_METADATA_ITEM_KEY) ||
1352 found_key->objectid > logical ||
1353 found_key->objectid + size <= logical) {
1354 pr_debug("logical %llu is not within any extent\n", logical);
1355 return -ENOENT;
1356 }
1357
1358 eb = path->nodes[0];
1359 item_size = btrfs_item_size_nr(eb, path->slots[0]);
1360 BUG_ON(item_size < sizeof(*ei));
1361
1362 ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
1363 flags = btrfs_extent_flags(eb, ei);
1364
1365 pr_debug("logical %llu is at position %llu within the extent (%llu "
1366 "EXTENT_ITEM %llu) flags %#llx size %u\n",
1367 logical, logical - found_key->objectid, found_key->objectid,
1368 found_key->offset, flags, item_size);
1369
1370 WARN_ON(!flags_ret);
1371 if (flags_ret) {
1372 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1373 *flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK;
1374 else if (flags & BTRFS_EXTENT_FLAG_DATA)
1375 *flags_ret = BTRFS_EXTENT_FLAG_DATA;
1376 else
1377 BUG_ON(1);
1378 return 0;
1379 }
1380
1381 return -EIO;
1382 }
1383
1384 /*
1385 * helper function to iterate extent inline refs. ptr must point to a 0 value
1386 * for the first call and may be modified. it is used to track state.
1387 * if more refs exist, 0 is returned and the next call to
1388 * __get_extent_inline_ref must pass the modified ptr parameter to get the
1389 * next ref. after the last ref was processed, 1 is returned.
1390 * returns <0 on error
1391 */
1392 static int __get_extent_inline_ref(unsigned long *ptr, struct extent_buffer *eb,
1393 struct btrfs_key *key,
1394 struct btrfs_extent_item *ei, u32 item_size,
1395 struct btrfs_extent_inline_ref **out_eiref,
1396 int *out_type)
1397 {
1398 unsigned long end;
1399 u64 flags;
1400 struct btrfs_tree_block_info *info;
1401
1402 if (!*ptr) {
1403 /* first call */
1404 flags = btrfs_extent_flags(eb, ei);
1405 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1406 if (key->type == BTRFS_METADATA_ITEM_KEY) {
1407 /* a skinny metadata extent */
1408 *out_eiref =
1409 (struct btrfs_extent_inline_ref *)(ei + 1);
1410 } else {
1411 WARN_ON(key->type != BTRFS_EXTENT_ITEM_KEY);
1412 info = (struct btrfs_tree_block_info *)(ei + 1);
1413 *out_eiref =
1414 (struct btrfs_extent_inline_ref *)(info + 1);
1415 }
1416 } else {
1417 *out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1);
1418 }
1419 *ptr = (unsigned long)*out_eiref;
1420 if ((unsigned long)(*ptr) >= (unsigned long)ei + item_size)
1421 return -ENOENT;
1422 }
1423
1424 end = (unsigned long)ei + item_size;
1425 *out_eiref = (struct btrfs_extent_inline_ref *)(*ptr);
1426 *out_type = btrfs_extent_inline_ref_type(eb, *out_eiref);
1427
1428 *ptr += btrfs_extent_inline_ref_size(*out_type);
1429 WARN_ON(*ptr > end);
1430 if (*ptr == end)
1431 return 1; /* last */
1432
1433 return 0;
1434 }
1435
1436 /*
1437 * reads the tree block backref for an extent. tree level and root are returned
1438 * through out_level and out_root. ptr must point to a 0 value for the first
1439 * call and may be modified (see __get_extent_inline_ref comment).
1440 * returns 0 if data was provided, 1 if there was no more data to provide or
1441 * <0 on error.
1442 */
1443 int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
1444 struct btrfs_key *key, struct btrfs_extent_item *ei,
1445 u32 item_size, u64 *out_root, u8 *out_level)
1446 {
1447 int ret;
1448 int type;
1449 struct btrfs_tree_block_info *info;
1450 struct btrfs_extent_inline_ref *eiref;
1451
1452 if (*ptr == (unsigned long)-1)
1453 return 1;
1454
1455 while (1) {
1456 ret = __get_extent_inline_ref(ptr, eb, key, ei, item_size,
1457 &eiref, &type);
1458 if (ret < 0)
1459 return ret;
1460
1461 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1462 type == BTRFS_SHARED_BLOCK_REF_KEY)
1463 break;
1464
1465 if (ret == 1)
1466 return 1;
1467 }
1468
1469 /* we can treat both ref types equally here */
1470 info = (struct btrfs_tree_block_info *)(ei + 1);
1471 *out_root = btrfs_extent_inline_ref_offset(eb, eiref);
1472 *out_level = btrfs_tree_block_level(eb, info);
1473
1474 if (ret == 1)
1475 *ptr = (unsigned long)-1;
1476
1477 return 0;
1478 }
1479
1480 static int iterate_leaf_refs(struct extent_inode_elem *inode_list,
1481 u64 root, u64 extent_item_objectid,
1482 iterate_extent_inodes_t *iterate, void *ctx)
1483 {
1484 struct extent_inode_elem *eie;
1485 int ret = 0;
1486
1487 for (eie = inode_list; eie; eie = eie->next) {
1488 pr_debug("ref for %llu resolved, key (%llu EXTEND_DATA %llu), "
1489 "root %llu\n", extent_item_objectid,
1490 eie->inum, eie->offset, root);
1491 ret = iterate(eie->inum, eie->offset, root, ctx);
1492 if (ret) {
1493 pr_debug("stopping iteration for %llu due to ret=%d\n",
1494 extent_item_objectid, ret);
1495 break;
1496 }
1497 }
1498
1499 return ret;
1500 }
1501
1502 /*
1503 * calls iterate() for every inode that references the extent identified by
1504 * the given parameters.
1505 * when the iterator function returns a non-zero value, iteration stops.
1506 */
1507 int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
1508 u64 extent_item_objectid, u64 extent_item_pos,
1509 int search_commit_root,
1510 iterate_extent_inodes_t *iterate, void *ctx)
1511 {
1512 int ret;
1513 struct btrfs_trans_handle *trans = NULL;
1514 struct ulist *refs = NULL;
1515 struct ulist *roots = NULL;
1516 struct ulist_node *ref_node = NULL;
1517 struct ulist_node *root_node = NULL;
1518 struct seq_list tree_mod_seq_elem = {};
1519 struct ulist_iterator ref_uiter;
1520 struct ulist_iterator root_uiter;
1521
1522 pr_debug("resolving all inodes for extent %llu\n",
1523 extent_item_objectid);
1524
1525 if (!search_commit_root) {
1526 trans = btrfs_join_transaction(fs_info->extent_root);
1527 if (IS_ERR(trans))
1528 return PTR_ERR(trans);
1529 btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1530 }
1531
1532 ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid,
1533 tree_mod_seq_elem.seq, &refs,
1534 &extent_item_pos);
1535 if (ret)
1536 goto out;
1537
1538 ULIST_ITER_INIT(&ref_uiter);
1539 while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) {
1540 ret = btrfs_find_all_roots(trans, fs_info, ref_node->val,
1541 tree_mod_seq_elem.seq, &roots);
1542 if (ret)
1543 break;
1544 ULIST_ITER_INIT(&root_uiter);
1545 while (!ret && (root_node = ulist_next(roots, &root_uiter))) {
1546 pr_debug("root %llu references leaf %llu, data list "
1547 "%#llx\n", root_node->val, ref_node->val,
1548 ref_node->aux);
1549 ret = iterate_leaf_refs((struct extent_inode_elem *)
1550 (uintptr_t)ref_node->aux,
1551 root_node->val,
1552 extent_item_objectid,
1553 iterate, ctx);
1554 }
1555 ulist_free(roots);
1556 }
1557
1558 free_leaf_list(refs);
1559 out:
1560 if (!search_commit_root) {
1561 btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1562 btrfs_end_transaction(trans, fs_info->extent_root);
1563 }
1564
1565 return ret;
1566 }
1567
1568 int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
1569 struct btrfs_path *path,
1570 iterate_extent_inodes_t *iterate, void *ctx)
1571 {
1572 int ret;
1573 u64 extent_item_pos;
1574 u64 flags = 0;
1575 struct btrfs_key found_key;
1576 int search_commit_root = path->search_commit_root;
1577
1578 ret = extent_from_logical(fs_info, logical, path, &found_key, &flags);
1579 btrfs_release_path(path);
1580 if (ret < 0)
1581 return ret;
1582 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1583 return -EINVAL;
1584
1585 extent_item_pos = logical - found_key.objectid;
1586 ret = iterate_extent_inodes(fs_info, found_key.objectid,
1587 extent_item_pos, search_commit_root,
1588 iterate, ctx);
1589
1590 return ret;
1591 }
1592
1593 typedef int (iterate_irefs_t)(u64 parent, u32 name_len, unsigned long name_off,
1594 struct extent_buffer *eb, void *ctx);
1595
1596 static int iterate_inode_refs(u64 inum, struct btrfs_root *fs_root,
1597 struct btrfs_path *path,
1598 iterate_irefs_t *iterate, void *ctx)
1599 {
1600 int ret = 0;
1601 int slot;
1602 u32 cur;
1603 u32 len;
1604 u32 name_len;
1605 u64 parent = 0;
1606 int found = 0;
1607 struct extent_buffer *eb;
1608 struct btrfs_item *item;
1609 struct btrfs_inode_ref *iref;
1610 struct btrfs_key found_key;
1611
1612 while (!ret) {
1613 path->leave_spinning = 1;
1614 ret = inode_ref_info(inum, parent ? parent+1 : 0, fs_root, path,
1615 &found_key);
1616 if (ret < 0)
1617 break;
1618 if (ret) {
1619 ret = found ? 0 : -ENOENT;
1620 break;
1621 }
1622 ++found;
1623
1624 parent = found_key.offset;
1625 slot = path->slots[0];
1626 eb = path->nodes[0];
1627 /* make sure we can use eb after releasing the path */
1628 atomic_inc(&eb->refs);
1629 btrfs_tree_read_lock(eb);
1630 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1631 btrfs_release_path(path);
1632
1633 item = btrfs_item_nr(eb, slot);
1634 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1635
1636 for (cur = 0; cur < btrfs_item_size(eb, item); cur += len) {
1637 name_len = btrfs_inode_ref_name_len(eb, iref);
1638 /* path must be released before calling iterate()! */
1639 pr_debug("following ref at offset %u for inode %llu in "
1640 "tree %llu\n", cur, found_key.objectid,
1641 fs_root->objectid);
1642 ret = iterate(parent, name_len,
1643 (unsigned long)(iref + 1), eb, ctx);
1644 if (ret)
1645 break;
1646 len = sizeof(*iref) + name_len;
1647 iref = (struct btrfs_inode_ref *)((char *)iref + len);
1648 }
1649 btrfs_tree_read_unlock_blocking(eb);
1650 free_extent_buffer(eb);
1651 }
1652
1653 btrfs_release_path(path);
1654
1655 return ret;
1656 }
1657
1658 static int iterate_inode_extrefs(u64 inum, struct btrfs_root *fs_root,
1659 struct btrfs_path *path,
1660 iterate_irefs_t *iterate, void *ctx)
1661 {
1662 int ret;
1663 int slot;
1664 u64 offset = 0;
1665 u64 parent;
1666 int found = 0;
1667 struct extent_buffer *eb;
1668 struct btrfs_inode_extref *extref;
1669 struct extent_buffer *leaf;
1670 u32 item_size;
1671 u32 cur_offset;
1672 unsigned long ptr;
1673
1674 while (1) {
1675 ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref,
1676 &offset);
1677 if (ret < 0)
1678 break;
1679 if (ret) {
1680 ret = found ? 0 : -ENOENT;
1681 break;
1682 }
1683 ++found;
1684
1685 slot = path->slots[0];
1686 eb = path->nodes[0];
1687 /* make sure we can use eb after releasing the path */
1688 atomic_inc(&eb->refs);
1689
1690 btrfs_tree_read_lock(eb);
1691 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1692 btrfs_release_path(path);
1693
1694 leaf = path->nodes[0];
1695 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1696 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1697 cur_offset = 0;
1698
1699 while (cur_offset < item_size) {
1700 u32 name_len;
1701
1702 extref = (struct btrfs_inode_extref *)(ptr + cur_offset);
1703 parent = btrfs_inode_extref_parent(eb, extref);
1704 name_len = btrfs_inode_extref_name_len(eb, extref);
1705 ret = iterate(parent, name_len,
1706 (unsigned long)&extref->name, eb, ctx);
1707 if (ret)
1708 break;
1709
1710 cur_offset += btrfs_inode_extref_name_len(leaf, extref);
1711 cur_offset += sizeof(*extref);
1712 }
1713 btrfs_tree_read_unlock_blocking(eb);
1714 free_extent_buffer(eb);
1715
1716 offset++;
1717 }
1718
1719 btrfs_release_path(path);
1720
1721 return ret;
1722 }
1723
1724 static int iterate_irefs(u64 inum, struct btrfs_root *fs_root,
1725 struct btrfs_path *path, iterate_irefs_t *iterate,
1726 void *ctx)
1727 {
1728 int ret;
1729 int found_refs = 0;
1730
1731 ret = iterate_inode_refs(inum, fs_root, path, iterate, ctx);
1732 if (!ret)
1733 ++found_refs;
1734 else if (ret != -ENOENT)
1735 return ret;
1736
1737 ret = iterate_inode_extrefs(inum, fs_root, path, iterate, ctx);
1738 if (ret == -ENOENT && found_refs)
1739 return 0;
1740
1741 return ret;
1742 }
1743
1744 /*
1745 * returns 0 if the path could be dumped (probably truncated)
1746 * returns <0 in case of an error
1747 */
1748 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
1749 struct extent_buffer *eb, void *ctx)
1750 {
1751 struct inode_fs_paths *ipath = ctx;
1752 char *fspath;
1753 char *fspath_min;
1754 int i = ipath->fspath->elem_cnt;
1755 const int s_ptr = sizeof(char *);
1756 u32 bytes_left;
1757
1758 bytes_left = ipath->fspath->bytes_left > s_ptr ?
1759 ipath->fspath->bytes_left - s_ptr : 0;
1760
1761 fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
1762 fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len,
1763 name_off, eb, inum, fspath_min, bytes_left);
1764 if (IS_ERR(fspath))
1765 return PTR_ERR(fspath);
1766
1767 if (fspath > fspath_min) {
1768 ipath->fspath->val[i] = (u64)(unsigned long)fspath;
1769 ++ipath->fspath->elem_cnt;
1770 ipath->fspath->bytes_left = fspath - fspath_min;
1771 } else {
1772 ++ipath->fspath->elem_missed;
1773 ipath->fspath->bytes_missing += fspath_min - fspath;
1774 ipath->fspath->bytes_left = 0;
1775 }
1776
1777 return 0;
1778 }
1779
1780 /*
1781 * this dumps all file system paths to the inode into the ipath struct, provided
1782 * is has been created large enough. each path is zero-terminated and accessed
1783 * from ipath->fspath->val[i].
1784 * when it returns, there are ipath->fspath->elem_cnt number of paths available
1785 * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the
1786 * number of missed paths in recored in ipath->fspath->elem_missed, otherwise,
1787 * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
1788 * have been needed to return all paths.
1789 */
1790 int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
1791 {
1792 return iterate_irefs(inum, ipath->fs_root, ipath->btrfs_path,
1793 inode_to_path, ipath);
1794 }
1795
1796 struct btrfs_data_container *init_data_container(u32 total_bytes)
1797 {
1798 struct btrfs_data_container *data;
1799 size_t alloc_bytes;
1800
1801 alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
1802 data = vmalloc(alloc_bytes);
1803 if (!data)
1804 return ERR_PTR(-ENOMEM);
1805
1806 if (total_bytes >= sizeof(*data)) {
1807 data->bytes_left = total_bytes - sizeof(*data);
1808 data->bytes_missing = 0;
1809 } else {
1810 data->bytes_missing = sizeof(*data) - total_bytes;
1811 data->bytes_left = 0;
1812 }
1813
1814 data->elem_cnt = 0;
1815 data->elem_missed = 0;
1816
1817 return data;
1818 }
1819
1820 /*
1821 * allocates space to return multiple file system paths for an inode.
1822 * total_bytes to allocate are passed, note that space usable for actual path
1823 * information will be total_bytes - sizeof(struct inode_fs_paths).
1824 * the returned pointer must be freed with free_ipath() in the end.
1825 */
1826 struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
1827 struct btrfs_path *path)
1828 {
1829 struct inode_fs_paths *ifp;
1830 struct btrfs_data_container *fspath;
1831
1832 fspath = init_data_container(total_bytes);
1833 if (IS_ERR(fspath))
1834 return (void *)fspath;
1835
1836 ifp = kmalloc(sizeof(*ifp), GFP_NOFS);
1837 if (!ifp) {
1838 kfree(fspath);
1839 return ERR_PTR(-ENOMEM);
1840 }
1841
1842 ifp->btrfs_path = path;
1843 ifp->fspath = fspath;
1844 ifp->fs_root = fs_root;
1845
1846 return ifp;
1847 }
1848
1849 void free_ipath(struct inode_fs_paths *ipath)
1850 {
1851 if (!ipath)
1852 return;
1853 vfree(ipath->fspath);
1854 kfree(ipath);
1855 }
Linux with added FPC-III support