Merge branches 'arm/rockchip', 'arm/exynos', 'arm/smmu', 'x86/vt-d', 'x86/amd', ...
[sfrench/cifs-2.6.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 /* Just an arbitrary number so we can be sure this happened */
29 #define BACKREF_FOUND_SHARED 6
30
31 struct extent_inode_elem {
32         u64 inum;
33         u64 offset;
34         struct extent_inode_elem *next;
35 };
36
37 static int check_extent_in_eb(struct btrfs_key *key, struct extent_buffer *eb,
38                                 struct btrfs_file_extent_item *fi,
39                                 u64 extent_item_pos,
40                                 struct extent_inode_elem **eie)
41 {
42         u64 offset = 0;
43         struct extent_inode_elem *e;
44
45         if (!btrfs_file_extent_compression(eb, fi) &&
46             !btrfs_file_extent_encryption(eb, fi) &&
47             !btrfs_file_extent_other_encoding(eb, fi)) {
48                 u64 data_offset;
49                 u64 data_len;
50
51                 data_offset = btrfs_file_extent_offset(eb, fi);
52                 data_len = btrfs_file_extent_num_bytes(eb, fi);
53
54                 if (extent_item_pos < data_offset ||
55                     extent_item_pos >= data_offset + data_len)
56                         return 1;
57                 offset = extent_item_pos - data_offset;
58         }
59
60         e = kmalloc(sizeof(*e), GFP_NOFS);
61         if (!e)
62                 return -ENOMEM;
63
64         e->next = *eie;
65         e->inum = key->objectid;
66         e->offset = key->offset + offset;
67         *eie = e;
68
69         return 0;
70 }
71
72 static void free_inode_elem_list(struct extent_inode_elem *eie)
73 {
74         struct extent_inode_elem *eie_next;
75
76         for (; eie; eie = eie_next) {
77                 eie_next = eie->next;
78                 kfree(eie);
79         }
80 }
81
82 static int find_extent_in_eb(struct extent_buffer *eb, u64 wanted_disk_byte,
83                                 u64 extent_item_pos,
84                                 struct extent_inode_elem **eie)
85 {
86         u64 disk_byte;
87         struct btrfs_key key;
88         struct btrfs_file_extent_item *fi;
89         int slot;
90         int nritems;
91         int extent_type;
92         int ret;
93
94         /*
95          * from the shared data ref, we only have the leaf but we need
96          * the key. thus, we must look into all items and see that we
97          * find one (some) with a reference to our extent item.
98          */
99         nritems = btrfs_header_nritems(eb);
100         for (slot = 0; slot < nritems; ++slot) {
101                 btrfs_item_key_to_cpu(eb, &key, slot);
102                 if (key.type != BTRFS_EXTENT_DATA_KEY)
103                         continue;
104                 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
105                 extent_type = btrfs_file_extent_type(eb, fi);
106                 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
107                         continue;
108                 /* don't skip BTRFS_FILE_EXTENT_PREALLOC, we can handle that */
109                 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
110                 if (disk_byte != wanted_disk_byte)
111                         continue;
112
113                 ret = check_extent_in_eb(&key, eb, fi, extent_item_pos, eie);
114                 if (ret < 0)
115                         return ret;
116         }
117
118         return 0;
119 }
120
121 /*
122  * this structure records all encountered refs on the way up to the root
123  */
124 struct __prelim_ref {
125         struct list_head list;
126         u64 root_id;
127         struct btrfs_key key_for_search;
128         int level;
129         int count;
130         struct extent_inode_elem *inode_list;
131         u64 parent;
132         u64 wanted_disk_byte;
133 };
134
135 static struct kmem_cache *btrfs_prelim_ref_cache;
136
137 int __init btrfs_prelim_ref_init(void)
138 {
139         btrfs_prelim_ref_cache = kmem_cache_create("btrfs_prelim_ref",
140                                         sizeof(struct __prelim_ref),
141                                         0,
142                                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
143                                         NULL);
144         if (!btrfs_prelim_ref_cache)
145                 return -ENOMEM;
146         return 0;
147 }
148
149 void btrfs_prelim_ref_exit(void)
150 {
151         if (btrfs_prelim_ref_cache)
152                 kmem_cache_destroy(btrfs_prelim_ref_cache);
153 }
154
155 /*
156  * the rules for all callers of this function are:
157  * - obtaining the parent is the goal
158  * - if you add a key, you must know that it is a correct key
159  * - if you cannot add the parent or a correct key, then we will look into the
160  *   block later to set a correct key
161  *
162  * delayed refs
163  * ============
164  *        backref type | shared | indirect | shared | indirect
165  * information         |   tree |     tree |   data |     data
166  * --------------------+--------+----------+--------+----------
167  *      parent logical |    y   |     -    |    -   |     -
168  *      key to resolve |    -   |     y    |    y   |     y
169  *  tree block logical |    -   |     -    |    -   |     -
170  *  root for resolving |    y   |     y    |    y   |     y
171  *
172  * - column 1:       we've the parent -> done
173  * - column 2, 3, 4: we use the key to find the parent
174  *
175  * on disk refs (inline or keyed)
176  * ==============================
177  *        backref type | shared | indirect | shared | indirect
178  * information         |   tree |     tree |   data |     data
179  * --------------------+--------+----------+--------+----------
180  *      parent logical |    y   |     -    |    y   |     -
181  *      key to resolve |    -   |     -    |    -   |     y
182  *  tree block logical |    y   |     y    |    y   |     y
183  *  root for resolving |    -   |     y    |    y   |     y
184  *
185  * - column 1, 3: we've the parent -> done
186  * - column 2:    we take the first key from the block to find the parent
187  *                (see __add_missing_keys)
188  * - column 4:    we use the key to find the parent
189  *
190  * additional information that's available but not required to find the parent
191  * block might help in merging entries to gain some speed.
192  */
193
194 static int __add_prelim_ref(struct list_head *head, u64 root_id,
195                             struct btrfs_key *key, int level,
196                             u64 parent, u64 wanted_disk_byte, int count,
197                             gfp_t gfp_mask)
198 {
199         struct __prelim_ref *ref;
200
201         if (root_id == BTRFS_DATA_RELOC_TREE_OBJECTID)
202                 return 0;
203
204         ref = kmem_cache_alloc(btrfs_prelim_ref_cache, gfp_mask);
205         if (!ref)
206                 return -ENOMEM;
207
208         ref->root_id = root_id;
209         if (key)
210                 ref->key_for_search = *key;
211         else
212                 memset(&ref->key_for_search, 0, sizeof(ref->key_for_search));
213
214         ref->inode_list = NULL;
215         ref->level = level;
216         ref->count = count;
217         ref->parent = parent;
218         ref->wanted_disk_byte = wanted_disk_byte;
219         list_add_tail(&ref->list, head);
220
221         return 0;
222 }
223
224 static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path,
225                            struct ulist *parents, struct __prelim_ref *ref,
226                            int level, u64 time_seq, const u64 *extent_item_pos,
227                            u64 total_refs)
228 {
229         int ret = 0;
230         int slot;
231         struct extent_buffer *eb;
232         struct btrfs_key key;
233         struct btrfs_key *key_for_search = &ref->key_for_search;
234         struct btrfs_file_extent_item *fi;
235         struct extent_inode_elem *eie = NULL, *old = NULL;
236         u64 disk_byte;
237         u64 wanted_disk_byte = ref->wanted_disk_byte;
238         u64 count = 0;
239
240         if (level != 0) {
241                 eb = path->nodes[level];
242                 ret = ulist_add(parents, eb->start, 0, GFP_NOFS);
243                 if (ret < 0)
244                         return ret;
245                 return 0;
246         }
247
248         /*
249          * We normally enter this function with the path already pointing to
250          * the first item to check. But sometimes, we may enter it with
251          * slot==nritems. In that case, go to the next leaf before we continue.
252          */
253         if (path->slots[0] >= btrfs_header_nritems(path->nodes[0]))
254                 ret = btrfs_next_old_leaf(root, path, time_seq);
255
256         while (!ret && count < total_refs) {
257                 eb = path->nodes[0];
258                 slot = path->slots[0];
259
260                 btrfs_item_key_to_cpu(eb, &key, slot);
261
262                 if (key.objectid != key_for_search->objectid ||
263                     key.type != BTRFS_EXTENT_DATA_KEY)
264                         break;
265
266                 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
267                 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
268
269                 if (disk_byte == wanted_disk_byte) {
270                         eie = NULL;
271                         old = NULL;
272                         count++;
273                         if (extent_item_pos) {
274                                 ret = check_extent_in_eb(&key, eb, fi,
275                                                 *extent_item_pos,
276                                                 &eie);
277                                 if (ret < 0)
278                                         break;
279                         }
280                         if (ret > 0)
281                                 goto next;
282                         ret = ulist_add_merge_ptr(parents, eb->start,
283                                                   eie, (void **)&old, GFP_NOFS);
284                         if (ret < 0)
285                                 break;
286                         if (!ret && extent_item_pos) {
287                                 while (old->next)
288                                         old = old->next;
289                                 old->next = eie;
290                         }
291                         eie = NULL;
292                 }
293 next:
294                 ret = btrfs_next_old_item(root, path, time_seq);
295         }
296
297         if (ret > 0)
298                 ret = 0;
299         else if (ret < 0)
300                 free_inode_elem_list(eie);
301         return ret;
302 }
303
304 /*
305  * resolve an indirect backref in the form (root_id, key, level)
306  * to a logical address
307  */
308 static int __resolve_indirect_ref(struct btrfs_fs_info *fs_info,
309                                   struct btrfs_path *path, u64 time_seq,
310                                   struct __prelim_ref *ref,
311                                   struct ulist *parents,
312                                   const u64 *extent_item_pos, u64 total_refs)
313 {
314         struct btrfs_root *root;
315         struct btrfs_key root_key;
316         struct extent_buffer *eb;
317         int ret = 0;
318         int root_level;
319         int level = ref->level;
320         int index;
321
322         root_key.objectid = ref->root_id;
323         root_key.type = BTRFS_ROOT_ITEM_KEY;
324         root_key.offset = (u64)-1;
325
326         index = srcu_read_lock(&fs_info->subvol_srcu);
327
328         root = btrfs_read_fs_root_no_name(fs_info, &root_key);
329         if (IS_ERR(root)) {
330                 srcu_read_unlock(&fs_info->subvol_srcu, index);
331                 ret = PTR_ERR(root);
332                 goto out;
333         }
334
335         if (path->search_commit_root)
336                 root_level = btrfs_header_level(root->commit_root);
337         else
338                 root_level = btrfs_old_root_level(root, time_seq);
339
340         if (root_level + 1 == level) {
341                 srcu_read_unlock(&fs_info->subvol_srcu, index);
342                 goto out;
343         }
344
345         path->lowest_level = level;
346         ret = btrfs_search_old_slot(root, &ref->key_for_search, path, time_seq);
347
348         /* root node has been locked, we can release @subvol_srcu safely here */
349         srcu_read_unlock(&fs_info->subvol_srcu, index);
350
351         pr_debug("search slot in root %llu (level %d, ref count %d) returned "
352                  "%d for key (%llu %u %llu)\n",
353                  ref->root_id, level, ref->count, ret,
354                  ref->key_for_search.objectid, ref->key_for_search.type,
355                  ref->key_for_search.offset);
356         if (ret < 0)
357                 goto out;
358
359         eb = path->nodes[level];
360         while (!eb) {
361                 if (WARN_ON(!level)) {
362                         ret = 1;
363                         goto out;
364                 }
365                 level--;
366                 eb = path->nodes[level];
367         }
368
369         ret = add_all_parents(root, path, parents, ref, level, time_seq,
370                               extent_item_pos, total_refs);
371 out:
372         path->lowest_level = 0;
373         btrfs_release_path(path);
374         return ret;
375 }
376
377 /*
378  * resolve all indirect backrefs from the list
379  */
380 static int __resolve_indirect_refs(struct btrfs_fs_info *fs_info,
381                                    struct btrfs_path *path, u64 time_seq,
382                                    struct list_head *head,
383                                    const u64 *extent_item_pos, u64 total_refs,
384                                    u64 root_objectid)
385 {
386         int err;
387         int ret = 0;
388         struct __prelim_ref *ref;
389         struct __prelim_ref *ref_safe;
390         struct __prelim_ref *new_ref;
391         struct ulist *parents;
392         struct ulist_node *node;
393         struct ulist_iterator uiter;
394
395         parents = ulist_alloc(GFP_NOFS);
396         if (!parents)
397                 return -ENOMEM;
398
399         /*
400          * _safe allows us to insert directly after the current item without
401          * iterating over the newly inserted items.
402          * we're also allowed to re-assign ref during iteration.
403          */
404         list_for_each_entry_safe(ref, ref_safe, head, list) {
405                 if (ref->parent)        /* already direct */
406                         continue;
407                 if (ref->count == 0)
408                         continue;
409                 if (root_objectid && ref->root_id != root_objectid) {
410                         ret = BACKREF_FOUND_SHARED;
411                         goto out;
412                 }
413                 err = __resolve_indirect_ref(fs_info, path, time_seq, ref,
414                                              parents, extent_item_pos,
415                                              total_refs);
416                 /*
417                  * we can only tolerate ENOENT,otherwise,we should catch error
418                  * and return directly.
419                  */
420                 if (err == -ENOENT) {
421                         continue;
422                 } else if (err) {
423                         ret = err;
424                         goto out;
425                 }
426
427                 /* we put the first parent into the ref at hand */
428                 ULIST_ITER_INIT(&uiter);
429                 node = ulist_next(parents, &uiter);
430                 ref->parent = node ? node->val : 0;
431                 ref->inode_list = node ?
432                         (struct extent_inode_elem *)(uintptr_t)node->aux : NULL;
433
434                 /* additional parents require new refs being added here */
435                 while ((node = ulist_next(parents, &uiter))) {
436                         new_ref = kmem_cache_alloc(btrfs_prelim_ref_cache,
437                                                    GFP_NOFS);
438                         if (!new_ref) {
439                                 ret = -ENOMEM;
440                                 goto out;
441                         }
442                         memcpy(new_ref, ref, sizeof(*ref));
443                         new_ref->parent = node->val;
444                         new_ref->inode_list = (struct extent_inode_elem *)
445                                                         (uintptr_t)node->aux;
446                         list_add(&new_ref->list, &ref->list);
447                 }
448                 ulist_reinit(parents);
449         }
450 out:
451         ulist_free(parents);
452         return ret;
453 }
454
455 static inline int ref_for_same_block(struct __prelim_ref *ref1,
456                                      struct __prelim_ref *ref2)
457 {
458         if (ref1->level != ref2->level)
459                 return 0;
460         if (ref1->root_id != ref2->root_id)
461                 return 0;
462         if (ref1->key_for_search.type != ref2->key_for_search.type)
463                 return 0;
464         if (ref1->key_for_search.objectid != ref2->key_for_search.objectid)
465                 return 0;
466         if (ref1->key_for_search.offset != ref2->key_for_search.offset)
467                 return 0;
468         if (ref1->parent != ref2->parent)
469                 return 0;
470
471         return 1;
472 }
473
474 /*
475  * read tree blocks and add keys where required.
476  */
477 static int __add_missing_keys(struct btrfs_fs_info *fs_info,
478                               struct list_head *head)
479 {
480         struct list_head *pos;
481         struct extent_buffer *eb;
482
483         list_for_each(pos, head) {
484                 struct __prelim_ref *ref;
485                 ref = list_entry(pos, struct __prelim_ref, list);
486
487                 if (ref->parent)
488                         continue;
489                 if (ref->key_for_search.type)
490                         continue;
491                 BUG_ON(!ref->wanted_disk_byte);
492                 eb = read_tree_block(fs_info->tree_root, ref->wanted_disk_byte,
493                                      0);
494                 if (!eb || !extent_buffer_uptodate(eb)) {
495                         free_extent_buffer(eb);
496                         return -EIO;
497                 }
498                 btrfs_tree_read_lock(eb);
499                 if (btrfs_header_level(eb) == 0)
500                         btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
501                 else
502                         btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
503                 btrfs_tree_read_unlock(eb);
504                 free_extent_buffer(eb);
505         }
506         return 0;
507 }
508
509 /*
510  * merge two lists of backrefs and adjust counts accordingly
511  *
512  * mode = 1: merge identical keys, if key is set
513  *    FIXME: if we add more keys in __add_prelim_ref, we can merge more here.
514  *           additionally, we could even add a key range for the blocks we
515  *           looked into to merge even more (-> replace unresolved refs by those
516  *           having a parent).
517  * mode = 2: merge identical parents
518  */
519 static void __merge_refs(struct list_head *head, int mode)
520 {
521         struct list_head *pos1;
522
523         list_for_each(pos1, head) {
524                 struct list_head *n2;
525                 struct list_head *pos2;
526                 struct __prelim_ref *ref1;
527
528                 ref1 = list_entry(pos1, struct __prelim_ref, list);
529
530                 for (pos2 = pos1->next, n2 = pos2->next; pos2 != head;
531                      pos2 = n2, n2 = pos2->next) {
532                         struct __prelim_ref *ref2;
533                         struct __prelim_ref *xchg;
534                         struct extent_inode_elem *eie;
535
536                         ref2 = list_entry(pos2, struct __prelim_ref, list);
537
538                         if (mode == 1) {
539                                 if (!ref_for_same_block(ref1, ref2))
540                                         continue;
541                                 if (!ref1->parent && ref2->parent) {
542                                         xchg = ref1;
543                                         ref1 = ref2;
544                                         ref2 = xchg;
545                                 }
546                         } else {
547                                 if (ref1->parent != ref2->parent)
548                                         continue;
549                         }
550
551                         eie = ref1->inode_list;
552                         while (eie && eie->next)
553                                 eie = eie->next;
554                         if (eie)
555                                 eie->next = ref2->inode_list;
556                         else
557                                 ref1->inode_list = ref2->inode_list;
558                         ref1->count += ref2->count;
559
560                         list_del(&ref2->list);
561                         kmem_cache_free(btrfs_prelim_ref_cache, ref2);
562                 }
563
564         }
565 }
566
567 /*
568  * add all currently queued delayed refs from this head whose seq nr is
569  * smaller or equal that seq to the list
570  */
571 static int __add_delayed_refs(struct btrfs_delayed_ref_head *head, u64 seq,
572                               struct list_head *prefs, u64 *total_refs,
573                               u64 inum)
574 {
575         struct btrfs_delayed_extent_op *extent_op = head->extent_op;
576         struct rb_node *n = &head->node.rb_node;
577         struct btrfs_key key;
578         struct btrfs_key op_key = {0};
579         int sgn;
580         int ret = 0;
581
582         if (extent_op && extent_op->update_key)
583                 btrfs_disk_key_to_cpu(&op_key, &extent_op->key);
584
585         spin_lock(&head->lock);
586         n = rb_first(&head->ref_root);
587         while (n) {
588                 struct btrfs_delayed_ref_node *node;
589                 node = rb_entry(n, struct btrfs_delayed_ref_node,
590                                 rb_node);
591                 n = rb_next(n);
592                 if (node->seq > seq)
593                         continue;
594
595                 switch (node->action) {
596                 case BTRFS_ADD_DELAYED_EXTENT:
597                 case BTRFS_UPDATE_DELAYED_HEAD:
598                         WARN_ON(1);
599                         continue;
600                 case BTRFS_ADD_DELAYED_REF:
601                         sgn = 1;
602                         break;
603                 case BTRFS_DROP_DELAYED_REF:
604                         sgn = -1;
605                         break;
606                 default:
607                         BUG_ON(1);
608                 }
609                 *total_refs += (node->ref_mod * sgn);
610                 switch (node->type) {
611                 case BTRFS_TREE_BLOCK_REF_KEY: {
612                         struct btrfs_delayed_tree_ref *ref;
613
614                         ref = btrfs_delayed_node_to_tree_ref(node);
615                         ret = __add_prelim_ref(prefs, ref->root, &op_key,
616                                                ref->level + 1, 0, node->bytenr,
617                                                node->ref_mod * sgn, GFP_ATOMIC);
618                         break;
619                 }
620                 case BTRFS_SHARED_BLOCK_REF_KEY: {
621                         struct btrfs_delayed_tree_ref *ref;
622
623                         ref = btrfs_delayed_node_to_tree_ref(node);
624                         ret = __add_prelim_ref(prefs, ref->root, NULL,
625                                                ref->level + 1, ref->parent,
626                                                node->bytenr,
627                                                node->ref_mod * sgn, GFP_ATOMIC);
628                         break;
629                 }
630                 case BTRFS_EXTENT_DATA_REF_KEY: {
631                         struct btrfs_delayed_data_ref *ref;
632                         ref = btrfs_delayed_node_to_data_ref(node);
633
634                         key.objectid = ref->objectid;
635                         key.type = BTRFS_EXTENT_DATA_KEY;
636                         key.offset = ref->offset;
637
638                         /*
639                          * Found a inum that doesn't match our known inum, we
640                          * know it's shared.
641                          */
642                         if (inum && ref->objectid != inum) {
643                                 ret = BACKREF_FOUND_SHARED;
644                                 break;
645                         }
646
647                         ret = __add_prelim_ref(prefs, ref->root, &key, 0, 0,
648                                                node->bytenr,
649                                                node->ref_mod * sgn, GFP_ATOMIC);
650                         break;
651                 }
652                 case BTRFS_SHARED_DATA_REF_KEY: {
653                         struct btrfs_delayed_data_ref *ref;
654
655                         ref = btrfs_delayed_node_to_data_ref(node);
656
657                         key.objectid = ref->objectid;
658                         key.type = BTRFS_EXTENT_DATA_KEY;
659                         key.offset = ref->offset;
660                         ret = __add_prelim_ref(prefs, ref->root, &key, 0,
661                                                ref->parent, node->bytenr,
662                                                node->ref_mod * sgn, GFP_ATOMIC);
663                         break;
664                 }
665                 default:
666                         WARN_ON(1);
667                 }
668                 if (ret)
669                         break;
670         }
671         spin_unlock(&head->lock);
672         return ret;
673 }
674
675 /*
676  * add all inline backrefs for bytenr to the list
677  */
678 static int __add_inline_refs(struct btrfs_fs_info *fs_info,
679                              struct btrfs_path *path, u64 bytenr,
680                              int *info_level, struct list_head *prefs,
681                              u64 *total_refs, u64 inum)
682 {
683         int ret = 0;
684         int slot;
685         struct extent_buffer *leaf;
686         struct btrfs_key key;
687         struct btrfs_key found_key;
688         unsigned long ptr;
689         unsigned long end;
690         struct btrfs_extent_item *ei;
691         u64 flags;
692         u64 item_size;
693
694         /*
695          * enumerate all inline refs
696          */
697         leaf = path->nodes[0];
698         slot = path->slots[0];
699
700         item_size = btrfs_item_size_nr(leaf, slot);
701         BUG_ON(item_size < sizeof(*ei));
702
703         ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
704         flags = btrfs_extent_flags(leaf, ei);
705         *total_refs += btrfs_extent_refs(leaf, ei);
706         btrfs_item_key_to_cpu(leaf, &found_key, slot);
707
708         ptr = (unsigned long)(ei + 1);
709         end = (unsigned long)ei + item_size;
710
711         if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
712             flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
713                 struct btrfs_tree_block_info *info;
714
715                 info = (struct btrfs_tree_block_info *)ptr;
716                 *info_level = btrfs_tree_block_level(leaf, info);
717                 ptr += sizeof(struct btrfs_tree_block_info);
718                 BUG_ON(ptr > end);
719         } else if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
720                 *info_level = found_key.offset;
721         } else {
722                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
723         }
724
725         while (ptr < end) {
726                 struct btrfs_extent_inline_ref *iref;
727                 u64 offset;
728                 int type;
729
730                 iref = (struct btrfs_extent_inline_ref *)ptr;
731                 type = btrfs_extent_inline_ref_type(leaf, iref);
732                 offset = btrfs_extent_inline_ref_offset(leaf, iref);
733
734                 switch (type) {
735                 case BTRFS_SHARED_BLOCK_REF_KEY:
736                         ret = __add_prelim_ref(prefs, 0, NULL,
737                                                 *info_level + 1, offset,
738                                                 bytenr, 1, GFP_NOFS);
739                         break;
740                 case BTRFS_SHARED_DATA_REF_KEY: {
741                         struct btrfs_shared_data_ref *sdref;
742                         int count;
743
744                         sdref = (struct btrfs_shared_data_ref *)(iref + 1);
745                         count = btrfs_shared_data_ref_count(leaf, sdref);
746                         ret = __add_prelim_ref(prefs, 0, NULL, 0, offset,
747                                                bytenr, count, GFP_NOFS);
748                         break;
749                 }
750                 case BTRFS_TREE_BLOCK_REF_KEY:
751                         ret = __add_prelim_ref(prefs, offset, NULL,
752                                                *info_level + 1, 0,
753                                                bytenr, 1, GFP_NOFS);
754                         break;
755                 case BTRFS_EXTENT_DATA_REF_KEY: {
756                         struct btrfs_extent_data_ref *dref;
757                         int count;
758                         u64 root;
759
760                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
761                         count = btrfs_extent_data_ref_count(leaf, dref);
762                         key.objectid = btrfs_extent_data_ref_objectid(leaf,
763                                                                       dref);
764                         key.type = BTRFS_EXTENT_DATA_KEY;
765                         key.offset = btrfs_extent_data_ref_offset(leaf, dref);
766
767                         if (inum && key.objectid != inum) {
768                                 ret = BACKREF_FOUND_SHARED;
769                                 break;
770                         }
771
772                         root = btrfs_extent_data_ref_root(leaf, dref);
773                         ret = __add_prelim_ref(prefs, root, &key, 0, 0,
774                                                bytenr, count, GFP_NOFS);
775                         break;
776                 }
777                 default:
778                         WARN_ON(1);
779                 }
780                 if (ret)
781                         return ret;
782                 ptr += btrfs_extent_inline_ref_size(type);
783         }
784
785         return 0;
786 }
787
788 /*
789  * add all non-inline backrefs for bytenr to the list
790  */
791 static int __add_keyed_refs(struct btrfs_fs_info *fs_info,
792                             struct btrfs_path *path, u64 bytenr,
793                             int info_level, struct list_head *prefs, u64 inum)
794 {
795         struct btrfs_root *extent_root = fs_info->extent_root;
796         int ret;
797         int slot;
798         struct extent_buffer *leaf;
799         struct btrfs_key key;
800
801         while (1) {
802                 ret = btrfs_next_item(extent_root, path);
803                 if (ret < 0)
804                         break;
805                 if (ret) {
806                         ret = 0;
807                         break;
808                 }
809
810                 slot = path->slots[0];
811                 leaf = path->nodes[0];
812                 btrfs_item_key_to_cpu(leaf, &key, slot);
813
814                 if (key.objectid != bytenr)
815                         break;
816                 if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
817                         continue;
818                 if (key.type > BTRFS_SHARED_DATA_REF_KEY)
819                         break;
820
821                 switch (key.type) {
822                 case BTRFS_SHARED_BLOCK_REF_KEY:
823                         ret = __add_prelim_ref(prefs, 0, NULL,
824                                                 info_level + 1, key.offset,
825                                                 bytenr, 1, GFP_NOFS);
826                         break;
827                 case BTRFS_SHARED_DATA_REF_KEY: {
828                         struct btrfs_shared_data_ref *sdref;
829                         int count;
830
831                         sdref = btrfs_item_ptr(leaf, slot,
832                                               struct btrfs_shared_data_ref);
833                         count = btrfs_shared_data_ref_count(leaf, sdref);
834                         ret = __add_prelim_ref(prefs, 0, NULL, 0, key.offset,
835                                                 bytenr, count, GFP_NOFS);
836                         break;
837                 }
838                 case BTRFS_TREE_BLOCK_REF_KEY:
839                         ret = __add_prelim_ref(prefs, key.offset, NULL,
840                                                info_level + 1, 0,
841                                                bytenr, 1, GFP_NOFS);
842                         break;
843                 case BTRFS_EXTENT_DATA_REF_KEY: {
844                         struct btrfs_extent_data_ref *dref;
845                         int count;
846                         u64 root;
847
848                         dref = btrfs_item_ptr(leaf, slot,
849                                               struct btrfs_extent_data_ref);
850                         count = btrfs_extent_data_ref_count(leaf, dref);
851                         key.objectid = btrfs_extent_data_ref_objectid(leaf,
852                                                                       dref);
853                         key.type = BTRFS_EXTENT_DATA_KEY;
854                         key.offset = btrfs_extent_data_ref_offset(leaf, dref);
855
856                         if (inum && key.objectid != inum) {
857                                 ret = BACKREF_FOUND_SHARED;
858                                 break;
859                         }
860
861                         root = btrfs_extent_data_ref_root(leaf, dref);
862                         ret = __add_prelim_ref(prefs, root, &key, 0, 0,
863                                                bytenr, count, GFP_NOFS);
864                         break;
865                 }
866                 default:
867                         WARN_ON(1);
868                 }
869                 if (ret)
870                         return ret;
871
872         }
873
874         return ret;
875 }
876
877 /*
878  * this adds all existing backrefs (inline backrefs, backrefs and delayed
879  * refs) for the given bytenr to the refs list, merges duplicates and resolves
880  * indirect refs to their parent bytenr.
881  * When roots are found, they're added to the roots list
882  *
883  * NOTE: This can return values > 0
884  *
885  * FIXME some caching might speed things up
886  */
887 static int find_parent_nodes(struct btrfs_trans_handle *trans,
888                              struct btrfs_fs_info *fs_info, u64 bytenr,
889                              u64 time_seq, struct ulist *refs,
890                              struct ulist *roots, const u64 *extent_item_pos,
891                              u64 root_objectid, u64 inum)
892 {
893         struct btrfs_key key;
894         struct btrfs_path *path;
895         struct btrfs_delayed_ref_root *delayed_refs = NULL;
896         struct btrfs_delayed_ref_head *head;
897         int info_level = 0;
898         int ret;
899         struct list_head prefs_delayed;
900         struct list_head prefs;
901         struct __prelim_ref *ref;
902         struct extent_inode_elem *eie = NULL;
903         u64 total_refs = 0;
904
905         INIT_LIST_HEAD(&prefs);
906         INIT_LIST_HEAD(&prefs_delayed);
907
908         key.objectid = bytenr;
909         key.offset = (u64)-1;
910         if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
911                 key.type = BTRFS_METADATA_ITEM_KEY;
912         else
913                 key.type = BTRFS_EXTENT_ITEM_KEY;
914
915         path = btrfs_alloc_path();
916         if (!path)
917                 return -ENOMEM;
918         if (!trans) {
919                 path->search_commit_root = 1;
920                 path->skip_locking = 1;
921         }
922
923         /*
924          * grab both a lock on the path and a lock on the delayed ref head.
925          * We need both to get a consistent picture of how the refs look
926          * at a specified point in time
927          */
928 again:
929         head = NULL;
930
931         ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
932         if (ret < 0)
933                 goto out;
934         BUG_ON(ret == 0);
935
936 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
937         if (trans && likely(trans->type != __TRANS_DUMMY)) {
938 #else
939         if (trans) {
940 #endif
941                 /*
942                  * look if there are updates for this ref queued and lock the
943                  * head
944                  */
945                 delayed_refs = &trans->transaction->delayed_refs;
946                 spin_lock(&delayed_refs->lock);
947                 head = btrfs_find_delayed_ref_head(trans, bytenr);
948                 if (head) {
949                         if (!mutex_trylock(&head->mutex)) {
950                                 atomic_inc(&head->node.refs);
951                                 spin_unlock(&delayed_refs->lock);
952
953                                 btrfs_release_path(path);
954
955                                 /*
956                                  * Mutex was contended, block until it's
957                                  * released and try again
958                                  */
959                                 mutex_lock(&head->mutex);
960                                 mutex_unlock(&head->mutex);
961                                 btrfs_put_delayed_ref(&head->node);
962                                 goto again;
963                         }
964                         spin_unlock(&delayed_refs->lock);
965                         ret = __add_delayed_refs(head, time_seq,
966                                                  &prefs_delayed, &total_refs,
967                                                  inum);
968                         mutex_unlock(&head->mutex);
969                         if (ret)
970                                 goto out;
971                 } else {
972                         spin_unlock(&delayed_refs->lock);
973                 }
974         }
975
976         if (path->slots[0]) {
977                 struct extent_buffer *leaf;
978                 int slot;
979
980                 path->slots[0]--;
981                 leaf = path->nodes[0];
982                 slot = path->slots[0];
983                 btrfs_item_key_to_cpu(leaf, &key, slot);
984                 if (key.objectid == bytenr &&
985                     (key.type == BTRFS_EXTENT_ITEM_KEY ||
986                      key.type == BTRFS_METADATA_ITEM_KEY)) {
987                         ret = __add_inline_refs(fs_info, path, bytenr,
988                                                 &info_level, &prefs,
989                                                 &total_refs, inum);
990                         if (ret)
991                                 goto out;
992                         ret = __add_keyed_refs(fs_info, path, bytenr,
993                                                info_level, &prefs, inum);
994                         if (ret)
995                                 goto out;
996                 }
997         }
998         btrfs_release_path(path);
999
1000         list_splice_init(&prefs_delayed, &prefs);
1001
1002         ret = __add_missing_keys(fs_info, &prefs);
1003         if (ret)
1004                 goto out;
1005
1006         __merge_refs(&prefs, 1);
1007
1008         ret = __resolve_indirect_refs(fs_info, path, time_seq, &prefs,
1009                                       extent_item_pos, total_refs,
1010                                       root_objectid);
1011         if (ret)
1012                 goto out;
1013
1014         __merge_refs(&prefs, 2);
1015
1016         while (!list_empty(&prefs)) {
1017                 ref = list_first_entry(&prefs, struct __prelim_ref, list);
1018                 WARN_ON(ref->count < 0);
1019                 if (roots && ref->count && ref->root_id && ref->parent == 0) {
1020                         if (root_objectid && ref->root_id != root_objectid) {
1021                                 ret = BACKREF_FOUND_SHARED;
1022                                 goto out;
1023                         }
1024
1025                         /* no parent == root of tree */
1026                         ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS);
1027                         if (ret < 0)
1028                                 goto out;
1029                 }
1030                 if (ref->count && ref->parent) {
1031                         if (extent_item_pos && !ref->inode_list &&
1032                             ref->level == 0) {
1033                                 struct extent_buffer *eb;
1034
1035                                 eb = read_tree_block(fs_info->extent_root,
1036                                                            ref->parent, 0);
1037                                 if (!eb || !extent_buffer_uptodate(eb)) {
1038                                         free_extent_buffer(eb);
1039                                         ret = -EIO;
1040                                         goto out;
1041                                 }
1042                                 btrfs_tree_read_lock(eb);
1043                                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1044                                 ret = find_extent_in_eb(eb, bytenr,
1045                                                         *extent_item_pos, &eie);
1046                                 btrfs_tree_read_unlock_blocking(eb);
1047                                 free_extent_buffer(eb);
1048                                 if (ret < 0)
1049                                         goto out;
1050                                 ref->inode_list = eie;
1051                         }
1052                         ret = ulist_add_merge_ptr(refs, ref->parent,
1053                                                   ref->inode_list,
1054                                                   (void **)&eie, GFP_NOFS);
1055                         if (ret < 0)
1056                                 goto out;
1057                         if (!ret && extent_item_pos) {
1058                                 /*
1059                                  * we've recorded that parent, so we must extend
1060                                  * its inode list here
1061                                  */
1062                                 BUG_ON(!eie);
1063                                 while (eie->next)
1064                                         eie = eie->next;
1065                                 eie->next = ref->inode_list;
1066                         }
1067                         eie = NULL;
1068                 }
1069                 list_del(&ref->list);
1070                 kmem_cache_free(btrfs_prelim_ref_cache, ref);
1071         }
1072
1073 out:
1074         btrfs_free_path(path);
1075         while (!list_empty(&prefs)) {
1076                 ref = list_first_entry(&prefs, struct __prelim_ref, list);
1077                 list_del(&ref->list);
1078                 kmem_cache_free(btrfs_prelim_ref_cache, ref);
1079         }
1080         while (!list_empty(&prefs_delayed)) {
1081                 ref = list_first_entry(&prefs_delayed, struct __prelim_ref,
1082                                        list);
1083                 list_del(&ref->list);
1084                 kmem_cache_free(btrfs_prelim_ref_cache, ref);
1085         }
1086         if (ret < 0)
1087                 free_inode_elem_list(eie);
1088         return ret;
1089 }
1090
1091 static void free_leaf_list(struct ulist *blocks)
1092 {
1093         struct ulist_node *node = NULL;
1094         struct extent_inode_elem *eie;
1095         struct ulist_iterator uiter;
1096
1097         ULIST_ITER_INIT(&uiter);
1098         while ((node = ulist_next(blocks, &uiter))) {
1099                 if (!node->aux)
1100                         continue;
1101                 eie = (struct extent_inode_elem *)(uintptr_t)node->aux;
1102                 free_inode_elem_list(eie);
1103                 node->aux = 0;
1104         }
1105
1106         ulist_free(blocks);
1107 }
1108
1109 /*
1110  * Finds all leafs with a reference to the specified combination of bytenr and
1111  * offset. key_list_head will point to a list of corresponding keys (caller must
1112  * free each list element). The leafs will be stored in the leafs ulist, which
1113  * must be freed with ulist_free.
1114  *
1115  * returns 0 on success, <0 on error
1116  */
1117 static int btrfs_find_all_leafs(struct btrfs_trans_handle *trans,
1118                                 struct btrfs_fs_info *fs_info, u64 bytenr,
1119                                 u64 time_seq, struct ulist **leafs,
1120                                 const u64 *extent_item_pos)
1121 {
1122         int ret;
1123
1124         *leafs = ulist_alloc(GFP_NOFS);
1125         if (!*leafs)
1126                 return -ENOMEM;
1127
1128         ret = find_parent_nodes(trans, fs_info, bytenr,
1129                                 time_seq, *leafs, NULL, extent_item_pos, 0, 0);
1130         if (ret < 0 && ret != -ENOENT) {
1131                 free_leaf_list(*leafs);
1132                 return ret;
1133         }
1134
1135         return 0;
1136 }
1137
1138 /*
1139  * walk all backrefs for a given extent to find all roots that reference this
1140  * extent. Walking a backref means finding all extents that reference this
1141  * extent and in turn walk the backrefs of those, too. Naturally this is a
1142  * recursive process, but here it is implemented in an iterative fashion: We
1143  * find all referencing extents for the extent in question and put them on a
1144  * list. In turn, we find all referencing extents for those, further appending
1145  * to the list. The way we iterate the list allows adding more elements after
1146  * the current while iterating. The process stops when we reach the end of the
1147  * list. Found roots are added to the roots list.
1148  *
1149  * returns 0 on success, < 0 on error.
1150  */
1151 static int __btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1152                                   struct btrfs_fs_info *fs_info, u64 bytenr,
1153                                   u64 time_seq, struct ulist **roots)
1154 {
1155         struct ulist *tmp;
1156         struct ulist_node *node = NULL;
1157         struct ulist_iterator uiter;
1158         int ret;
1159
1160         tmp = ulist_alloc(GFP_NOFS);
1161         if (!tmp)
1162                 return -ENOMEM;
1163         *roots = ulist_alloc(GFP_NOFS);
1164         if (!*roots) {
1165                 ulist_free(tmp);
1166                 return -ENOMEM;
1167         }
1168
1169         ULIST_ITER_INIT(&uiter);
1170         while (1) {
1171                 ret = find_parent_nodes(trans, fs_info, bytenr,
1172                                         time_seq, tmp, *roots, NULL, 0, 0);
1173                 if (ret < 0 && ret != -ENOENT) {
1174                         ulist_free(tmp);
1175                         ulist_free(*roots);
1176                         return ret;
1177                 }
1178                 node = ulist_next(tmp, &uiter);
1179                 if (!node)
1180                         break;
1181                 bytenr = node->val;
1182                 cond_resched();
1183         }
1184
1185         ulist_free(tmp);
1186         return 0;
1187 }
1188
1189 int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1190                          struct btrfs_fs_info *fs_info, u64 bytenr,
1191                          u64 time_seq, struct ulist **roots)
1192 {
1193         int ret;
1194
1195         if (!trans)
1196                 down_read(&fs_info->commit_root_sem);
1197         ret = __btrfs_find_all_roots(trans, fs_info, bytenr, time_seq, roots);
1198         if (!trans)
1199                 up_read(&fs_info->commit_root_sem);
1200         return ret;
1201 }
1202
1203 /**
1204  * btrfs_check_shared - tell us whether an extent is shared
1205  *
1206  * @trans: optional trans handle
1207  *
1208  * btrfs_check_shared uses the backref walking code but will short
1209  * circuit as soon as it finds a root or inode that doesn't match the
1210  * one passed in. This provides a significant performance benefit for
1211  * callers (such as fiemap) which want to know whether the extent is
1212  * shared but do not need a ref count.
1213  *
1214  * Return: 0 if extent is not shared, 1 if it is shared, < 0 on error.
1215  */
1216 int btrfs_check_shared(struct btrfs_trans_handle *trans,
1217                        struct btrfs_fs_info *fs_info, u64 root_objectid,
1218                        u64 inum, u64 bytenr)
1219 {
1220         struct ulist *tmp = NULL;
1221         struct ulist *roots = NULL;
1222         struct ulist_iterator uiter;
1223         struct ulist_node *node;
1224         struct seq_list elem = SEQ_LIST_INIT(elem);
1225         int ret = 0;
1226
1227         tmp = ulist_alloc(GFP_NOFS);
1228         roots = ulist_alloc(GFP_NOFS);
1229         if (!tmp || !roots) {
1230                 ulist_free(tmp);
1231                 ulist_free(roots);
1232                 return -ENOMEM;
1233         }
1234
1235         if (trans)
1236                 btrfs_get_tree_mod_seq(fs_info, &elem);
1237         else
1238                 down_read(&fs_info->commit_root_sem);
1239         ULIST_ITER_INIT(&uiter);
1240         while (1) {
1241                 ret = find_parent_nodes(trans, fs_info, bytenr, elem.seq, tmp,
1242                                         roots, NULL, root_objectid, inum);
1243                 if (ret == BACKREF_FOUND_SHARED) {
1244                         /* this is the only condition under which we return 1 */
1245                         ret = 1;
1246                         break;
1247                 }
1248                 if (ret < 0 && ret != -ENOENT)
1249                         break;
1250                 ret = 0;
1251                 node = ulist_next(tmp, &uiter);
1252                 if (!node)
1253                         break;
1254                 bytenr = node->val;
1255                 cond_resched();
1256         }
1257         if (trans)
1258                 btrfs_put_tree_mod_seq(fs_info, &elem);
1259         else
1260                 up_read(&fs_info->commit_root_sem);
1261         ulist_free(tmp);
1262         ulist_free(roots);
1263         return ret;
1264 }
1265
1266 int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,
1267                           u64 start_off, struct btrfs_path *path,
1268                           struct btrfs_inode_extref **ret_extref,
1269                           u64 *found_off)
1270 {
1271         int ret, slot;
1272         struct btrfs_key key;
1273         struct btrfs_key found_key;
1274         struct btrfs_inode_extref *extref;
1275         struct extent_buffer *leaf;
1276         unsigned long ptr;
1277
1278         key.objectid = inode_objectid;
1279         key.type = BTRFS_INODE_EXTREF_KEY;
1280         key.offset = start_off;
1281
1282         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1283         if (ret < 0)
1284                 return ret;
1285
1286         while (1) {
1287                 leaf = path->nodes[0];
1288                 slot = path->slots[0];
1289                 if (slot >= btrfs_header_nritems(leaf)) {
1290                         /*
1291                          * If the item at offset is not found,
1292                          * btrfs_search_slot will point us to the slot
1293                          * where it should be inserted. In our case
1294                          * that will be the slot directly before the
1295                          * next INODE_REF_KEY_V2 item. In the case
1296                          * that we're pointing to the last slot in a
1297                          * leaf, we must move one leaf over.
1298                          */
1299                         ret = btrfs_next_leaf(root, path);
1300                         if (ret) {
1301                                 if (ret >= 1)
1302                                         ret = -ENOENT;
1303                                 break;
1304                         }
1305                         continue;
1306                 }
1307
1308                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1309
1310                 /*
1311                  * Check that we're still looking at an extended ref key for
1312                  * this particular objectid. If we have different
1313                  * objectid or type then there are no more to be found
1314                  * in the tree and we can exit.
1315                  */
1316                 ret = -ENOENT;
1317                 if (found_key.objectid != inode_objectid)
1318                         break;
1319                 if (found_key.type != BTRFS_INODE_EXTREF_KEY)
1320                         break;
1321
1322                 ret = 0;
1323                 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1324                 extref = (struct btrfs_inode_extref *)ptr;
1325                 *ret_extref = extref;
1326                 if (found_off)
1327                         *found_off = found_key.offset;
1328                 break;
1329         }
1330
1331         return ret;
1332 }
1333
1334 /*
1335  * this iterates to turn a name (from iref/extref) into a full filesystem path.
1336  * Elements of the path are separated by '/' and the path is guaranteed to be
1337  * 0-terminated. the path is only given within the current file system.
1338  * Therefore, it never starts with a '/'. the caller is responsible to provide
1339  * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
1340  * the start point of the resulting string is returned. this pointer is within
1341  * dest, normally.
1342  * in case the path buffer would overflow, the pointer is decremented further
1343  * as if output was written to the buffer, though no more output is actually
1344  * generated. that way, the caller can determine how much space would be
1345  * required for the path to fit into the buffer. in that case, the returned
1346  * value will be smaller than dest. callers must check this!
1347  */
1348 char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
1349                         u32 name_len, unsigned long name_off,
1350                         struct extent_buffer *eb_in, u64 parent,
1351                         char *dest, u32 size)
1352 {
1353         int slot;
1354         u64 next_inum;
1355         int ret;
1356         s64 bytes_left = ((s64)size) - 1;
1357         struct extent_buffer *eb = eb_in;
1358         struct btrfs_key found_key;
1359         int leave_spinning = path->leave_spinning;
1360         struct btrfs_inode_ref *iref;
1361
1362         if (bytes_left >= 0)
1363                 dest[bytes_left] = '\0';
1364
1365         path->leave_spinning = 1;
1366         while (1) {
1367                 bytes_left -= name_len;
1368                 if (bytes_left >= 0)
1369                         read_extent_buffer(eb, dest + bytes_left,
1370                                            name_off, name_len);
1371                 if (eb != eb_in) {
1372                         btrfs_tree_read_unlock_blocking(eb);
1373                         free_extent_buffer(eb);
1374                 }
1375                 ret = btrfs_find_item(fs_root, path, parent, 0,
1376                                 BTRFS_INODE_REF_KEY, &found_key);
1377                 if (ret > 0)
1378                         ret = -ENOENT;
1379                 if (ret)
1380                         break;
1381
1382                 next_inum = found_key.offset;
1383
1384                 /* regular exit ahead */
1385                 if (parent == next_inum)
1386                         break;
1387
1388                 slot = path->slots[0];
1389                 eb = path->nodes[0];
1390                 /* make sure we can use eb after releasing the path */
1391                 if (eb != eb_in) {
1392                         atomic_inc(&eb->refs);
1393                         btrfs_tree_read_lock(eb);
1394                         btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1395                 }
1396                 btrfs_release_path(path);
1397                 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1398
1399                 name_len = btrfs_inode_ref_name_len(eb, iref);
1400                 name_off = (unsigned long)(iref + 1);
1401
1402                 parent = next_inum;
1403                 --bytes_left;
1404                 if (bytes_left >= 0)
1405                         dest[bytes_left] = '/';
1406         }
1407
1408         btrfs_release_path(path);
1409         path->leave_spinning = leave_spinning;
1410
1411         if (ret)
1412                 return ERR_PTR(ret);
1413
1414         return dest + bytes_left;
1415 }
1416
1417 /*
1418  * this makes the path point to (logical EXTENT_ITEM *)
1419  * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
1420  * tree blocks and <0 on error.
1421  */
1422 int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
1423                         struct btrfs_path *path, struct btrfs_key *found_key,
1424                         u64 *flags_ret)
1425 {
1426         int ret;
1427         u64 flags;
1428         u64 size = 0;
1429         u32 item_size;
1430         struct extent_buffer *eb;
1431         struct btrfs_extent_item *ei;
1432         struct btrfs_key key;
1433
1434         if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1435                 key.type = BTRFS_METADATA_ITEM_KEY;
1436         else
1437                 key.type = BTRFS_EXTENT_ITEM_KEY;
1438         key.objectid = logical;
1439         key.offset = (u64)-1;
1440
1441         ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
1442         if (ret < 0)
1443                 return ret;
1444
1445         ret = btrfs_previous_extent_item(fs_info->extent_root, path, 0);
1446         if (ret) {
1447                 if (ret > 0)
1448                         ret = -ENOENT;
1449                 return ret;
1450         }
1451         btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]);
1452         if (found_key->type == BTRFS_METADATA_ITEM_KEY)
1453                 size = fs_info->extent_root->nodesize;
1454         else if (found_key->type == BTRFS_EXTENT_ITEM_KEY)
1455                 size = found_key->offset;
1456
1457         if (found_key->objectid > logical ||
1458             found_key->objectid + size <= logical) {
1459                 pr_debug("logical %llu is not within any extent\n", logical);
1460                 return -ENOENT;
1461         }
1462
1463         eb = path->nodes[0];
1464         item_size = btrfs_item_size_nr(eb, path->slots[0]);
1465         BUG_ON(item_size < sizeof(*ei));
1466
1467         ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
1468         flags = btrfs_extent_flags(eb, ei);
1469
1470         pr_debug("logical %llu is at position %llu within the extent (%llu "
1471                  "EXTENT_ITEM %llu) flags %#llx size %u\n",
1472                  logical, logical - found_key->objectid, found_key->objectid,
1473                  found_key->offset, flags, item_size);
1474
1475         WARN_ON(!flags_ret);
1476         if (flags_ret) {
1477                 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1478                         *flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK;
1479                 else if (flags & BTRFS_EXTENT_FLAG_DATA)
1480                         *flags_ret = BTRFS_EXTENT_FLAG_DATA;
1481                 else
1482                         BUG_ON(1);
1483                 return 0;
1484         }
1485
1486         return -EIO;
1487 }
1488
1489 /*
1490  * helper function to iterate extent inline refs. ptr must point to a 0 value
1491  * for the first call and may be modified. it is used to track state.
1492  * if more refs exist, 0 is returned and the next call to
1493  * __get_extent_inline_ref must pass the modified ptr parameter to get the
1494  * next ref. after the last ref was processed, 1 is returned.
1495  * returns <0 on error
1496  */
1497 static int __get_extent_inline_ref(unsigned long *ptr, struct extent_buffer *eb,
1498                                    struct btrfs_key *key,
1499                                    struct btrfs_extent_item *ei, u32 item_size,
1500                                    struct btrfs_extent_inline_ref **out_eiref,
1501                                    int *out_type)
1502 {
1503         unsigned long end;
1504         u64 flags;
1505         struct btrfs_tree_block_info *info;
1506
1507         if (!*ptr) {
1508                 /* first call */
1509                 flags = btrfs_extent_flags(eb, ei);
1510                 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1511                         if (key->type == BTRFS_METADATA_ITEM_KEY) {
1512                                 /* a skinny metadata extent */
1513                                 *out_eiref =
1514                                      (struct btrfs_extent_inline_ref *)(ei + 1);
1515                         } else {
1516                                 WARN_ON(key->type != BTRFS_EXTENT_ITEM_KEY);
1517                                 info = (struct btrfs_tree_block_info *)(ei + 1);
1518                                 *out_eiref =
1519                                    (struct btrfs_extent_inline_ref *)(info + 1);
1520                         }
1521                 } else {
1522                         *out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1);
1523                 }
1524                 *ptr = (unsigned long)*out_eiref;
1525                 if ((unsigned long)(*ptr) >= (unsigned long)ei + item_size)
1526                         return -ENOENT;
1527         }
1528
1529         end = (unsigned long)ei + item_size;
1530         *out_eiref = (struct btrfs_extent_inline_ref *)(*ptr);
1531         *out_type = btrfs_extent_inline_ref_type(eb, *out_eiref);
1532
1533         *ptr += btrfs_extent_inline_ref_size(*out_type);
1534         WARN_ON(*ptr > end);
1535         if (*ptr == end)
1536                 return 1; /* last */
1537
1538         return 0;
1539 }
1540
1541 /*
1542  * reads the tree block backref for an extent. tree level and root are returned
1543  * through out_level and out_root. ptr must point to a 0 value for the first
1544  * call and may be modified (see __get_extent_inline_ref comment).
1545  * returns 0 if data was provided, 1 if there was no more data to provide or
1546  * <0 on error.
1547  */
1548 int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
1549                             struct btrfs_key *key, struct btrfs_extent_item *ei,
1550                             u32 item_size, u64 *out_root, u8 *out_level)
1551 {
1552         int ret;
1553         int type;
1554         struct btrfs_extent_inline_ref *eiref;
1555
1556         if (*ptr == (unsigned long)-1)
1557                 return 1;
1558
1559         while (1) {
1560                 ret = __get_extent_inline_ref(ptr, eb, key, ei, item_size,
1561                                               &eiref, &type);
1562                 if (ret < 0)
1563                         return ret;
1564
1565                 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1566                     type == BTRFS_SHARED_BLOCK_REF_KEY)
1567                         break;
1568
1569                 if (ret == 1)
1570                         return 1;
1571         }
1572
1573         /* we can treat both ref types equally here */
1574         *out_root = btrfs_extent_inline_ref_offset(eb, eiref);
1575
1576         if (key->type == BTRFS_EXTENT_ITEM_KEY) {
1577                 struct btrfs_tree_block_info *info;
1578
1579                 info = (struct btrfs_tree_block_info *)(ei + 1);
1580                 *out_level = btrfs_tree_block_level(eb, info);
1581         } else {
1582                 ASSERT(key->type == BTRFS_METADATA_ITEM_KEY);
1583                 *out_level = (u8)key->offset;
1584         }
1585
1586         if (ret == 1)
1587                 *ptr = (unsigned long)-1;
1588
1589         return 0;
1590 }
1591
1592 static int iterate_leaf_refs(struct extent_inode_elem *inode_list,
1593                                 u64 root, u64 extent_item_objectid,
1594                                 iterate_extent_inodes_t *iterate, void *ctx)
1595 {
1596         struct extent_inode_elem *eie;
1597         int ret = 0;
1598
1599         for (eie = inode_list; eie; eie = eie->next) {
1600                 pr_debug("ref for %llu resolved, key (%llu EXTEND_DATA %llu), "
1601                          "root %llu\n", extent_item_objectid,
1602                          eie->inum, eie->offset, root);
1603                 ret = iterate(eie->inum, eie->offset, root, ctx);
1604                 if (ret) {
1605                         pr_debug("stopping iteration for %llu due to ret=%d\n",
1606                                  extent_item_objectid, ret);
1607                         break;
1608                 }
1609         }
1610
1611         return ret;
1612 }
1613
1614 /*
1615  * calls iterate() for every inode that references the extent identified by
1616  * the given parameters.
1617  * when the iterator function returns a non-zero value, iteration stops.
1618  */
1619 int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
1620                                 u64 extent_item_objectid, u64 extent_item_pos,
1621                                 int search_commit_root,
1622                                 iterate_extent_inodes_t *iterate, void *ctx)
1623 {
1624         int ret;
1625         struct btrfs_trans_handle *trans = NULL;
1626         struct ulist *refs = NULL;
1627         struct ulist *roots = NULL;
1628         struct ulist_node *ref_node = NULL;
1629         struct ulist_node *root_node = NULL;
1630         struct seq_list tree_mod_seq_elem = SEQ_LIST_INIT(tree_mod_seq_elem);
1631         struct ulist_iterator ref_uiter;
1632         struct ulist_iterator root_uiter;
1633
1634         pr_debug("resolving all inodes for extent %llu\n",
1635                         extent_item_objectid);
1636
1637         if (!search_commit_root) {
1638                 trans = btrfs_join_transaction(fs_info->extent_root);
1639                 if (IS_ERR(trans))
1640                         return PTR_ERR(trans);
1641                 btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1642         } else {
1643                 down_read(&fs_info->commit_root_sem);
1644         }
1645
1646         ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid,
1647                                    tree_mod_seq_elem.seq, &refs,
1648                                    &extent_item_pos);
1649         if (ret)
1650                 goto out;
1651
1652         ULIST_ITER_INIT(&ref_uiter);
1653         while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) {
1654                 ret = __btrfs_find_all_roots(trans, fs_info, ref_node->val,
1655                                              tree_mod_seq_elem.seq, &roots);
1656                 if (ret)
1657                         break;
1658                 ULIST_ITER_INIT(&root_uiter);
1659                 while (!ret && (root_node = ulist_next(roots, &root_uiter))) {
1660                         pr_debug("root %llu references leaf %llu, data list "
1661                                  "%#llx\n", root_node->val, ref_node->val,
1662                                  ref_node->aux);
1663                         ret = iterate_leaf_refs((struct extent_inode_elem *)
1664                                                 (uintptr_t)ref_node->aux,
1665                                                 root_node->val,
1666                                                 extent_item_objectid,
1667                                                 iterate, ctx);
1668                 }
1669                 ulist_free(roots);
1670         }
1671
1672         free_leaf_list(refs);
1673 out:
1674         if (!search_commit_root) {
1675                 btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1676                 btrfs_end_transaction(trans, fs_info->extent_root);
1677         } else {
1678                 up_read(&fs_info->commit_root_sem);
1679         }
1680
1681         return ret;
1682 }
1683
1684 int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
1685                                 struct btrfs_path *path,
1686                                 iterate_extent_inodes_t *iterate, void *ctx)
1687 {
1688         int ret;
1689         u64 extent_item_pos;
1690         u64 flags = 0;
1691         struct btrfs_key found_key;
1692         int search_commit_root = path->search_commit_root;
1693
1694         ret = extent_from_logical(fs_info, logical, path, &found_key, &flags);
1695         btrfs_release_path(path);
1696         if (ret < 0)
1697                 return ret;
1698         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1699                 return -EINVAL;
1700
1701         extent_item_pos = logical - found_key.objectid;
1702         ret = iterate_extent_inodes(fs_info, found_key.objectid,
1703                                         extent_item_pos, search_commit_root,
1704                                         iterate, ctx);
1705
1706         return ret;
1707 }
1708
1709 typedef int (iterate_irefs_t)(u64 parent, u32 name_len, unsigned long name_off,
1710                               struct extent_buffer *eb, void *ctx);
1711
1712 static int iterate_inode_refs(u64 inum, struct btrfs_root *fs_root,
1713                               struct btrfs_path *path,
1714                               iterate_irefs_t *iterate, void *ctx)
1715 {
1716         int ret = 0;
1717         int slot;
1718         u32 cur;
1719         u32 len;
1720         u32 name_len;
1721         u64 parent = 0;
1722         int found = 0;
1723         struct extent_buffer *eb;
1724         struct btrfs_item *item;
1725         struct btrfs_inode_ref *iref;
1726         struct btrfs_key found_key;
1727
1728         while (!ret) {
1729                 ret = btrfs_find_item(fs_root, path, inum,
1730                                 parent ? parent + 1 : 0, BTRFS_INODE_REF_KEY,
1731                                 &found_key);
1732
1733                 if (ret < 0)
1734                         break;
1735                 if (ret) {
1736                         ret = found ? 0 : -ENOENT;
1737                         break;
1738                 }
1739                 ++found;
1740
1741                 parent = found_key.offset;
1742                 slot = path->slots[0];
1743                 eb = btrfs_clone_extent_buffer(path->nodes[0]);
1744                 if (!eb) {
1745                         ret = -ENOMEM;
1746                         break;
1747                 }
1748                 extent_buffer_get(eb);
1749                 btrfs_tree_read_lock(eb);
1750                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1751                 btrfs_release_path(path);
1752
1753                 item = btrfs_item_nr(slot);
1754                 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1755
1756                 for (cur = 0; cur < btrfs_item_size(eb, item); cur += len) {
1757                         name_len = btrfs_inode_ref_name_len(eb, iref);
1758                         /* path must be released before calling iterate()! */
1759                         pr_debug("following ref at offset %u for inode %llu in "
1760                                  "tree %llu\n", cur, found_key.objectid,
1761                                  fs_root->objectid);
1762                         ret = iterate(parent, name_len,
1763                                       (unsigned long)(iref + 1), eb, ctx);
1764                         if (ret)
1765                                 break;
1766                         len = sizeof(*iref) + name_len;
1767                         iref = (struct btrfs_inode_ref *)((char *)iref + len);
1768                 }
1769                 btrfs_tree_read_unlock_blocking(eb);
1770                 free_extent_buffer(eb);
1771         }
1772
1773         btrfs_release_path(path);
1774
1775         return ret;
1776 }
1777
1778 static int iterate_inode_extrefs(u64 inum, struct btrfs_root *fs_root,
1779                                  struct btrfs_path *path,
1780                                  iterate_irefs_t *iterate, void *ctx)
1781 {
1782         int ret;
1783         int slot;
1784         u64 offset = 0;
1785         u64 parent;
1786         int found = 0;
1787         struct extent_buffer *eb;
1788         struct btrfs_inode_extref *extref;
1789         struct extent_buffer *leaf;
1790         u32 item_size;
1791         u32 cur_offset;
1792         unsigned long ptr;
1793
1794         while (1) {
1795                 ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref,
1796                                             &offset);
1797                 if (ret < 0)
1798                         break;
1799                 if (ret) {
1800                         ret = found ? 0 : -ENOENT;
1801                         break;
1802                 }
1803                 ++found;
1804
1805                 slot = path->slots[0];
1806                 eb = btrfs_clone_extent_buffer(path->nodes[0]);
1807                 if (!eb) {
1808                         ret = -ENOMEM;
1809                         break;
1810                 }
1811                 extent_buffer_get(eb);
1812
1813                 btrfs_tree_read_lock(eb);
1814                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1815                 btrfs_release_path(path);
1816
1817                 leaf = path->nodes[0];
1818                 item_size = btrfs_item_size_nr(leaf, slot);
1819                 ptr = btrfs_item_ptr_offset(leaf, slot);
1820                 cur_offset = 0;
1821
1822                 while (cur_offset < item_size) {
1823                         u32 name_len;
1824
1825                         extref = (struct btrfs_inode_extref *)(ptr + cur_offset);
1826                         parent = btrfs_inode_extref_parent(eb, extref);
1827                         name_len = btrfs_inode_extref_name_len(eb, extref);
1828                         ret = iterate(parent, name_len,
1829                                       (unsigned long)&extref->name, eb, ctx);
1830                         if (ret)
1831                                 break;
1832
1833                         cur_offset += btrfs_inode_extref_name_len(leaf, extref);
1834                         cur_offset += sizeof(*extref);
1835                 }
1836                 btrfs_tree_read_unlock_blocking(eb);
1837                 free_extent_buffer(eb);
1838
1839                 offset++;
1840         }
1841
1842         btrfs_release_path(path);
1843
1844         return ret;
1845 }
1846
1847 static int iterate_irefs(u64 inum, struct btrfs_root *fs_root,
1848                          struct btrfs_path *path, iterate_irefs_t *iterate,
1849                          void *ctx)
1850 {
1851         int ret;
1852         int found_refs = 0;
1853
1854         ret = iterate_inode_refs(inum, fs_root, path, iterate, ctx);
1855         if (!ret)
1856                 ++found_refs;
1857         else if (ret != -ENOENT)
1858                 return ret;
1859
1860         ret = iterate_inode_extrefs(inum, fs_root, path, iterate, ctx);
1861         if (ret == -ENOENT && found_refs)
1862                 return 0;
1863
1864         return ret;
1865 }
1866
1867 /*
1868  * returns 0 if the path could be dumped (probably truncated)
1869  * returns <0 in case of an error
1870  */
1871 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
1872                          struct extent_buffer *eb, void *ctx)
1873 {
1874         struct inode_fs_paths *ipath = ctx;
1875         char *fspath;
1876         char *fspath_min;
1877         int i = ipath->fspath->elem_cnt;
1878         const int s_ptr = sizeof(char *);
1879         u32 bytes_left;
1880
1881         bytes_left = ipath->fspath->bytes_left > s_ptr ?
1882                                         ipath->fspath->bytes_left - s_ptr : 0;
1883
1884         fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
1885         fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len,
1886                                    name_off, eb, inum, fspath_min, bytes_left);
1887         if (IS_ERR(fspath))
1888                 return PTR_ERR(fspath);
1889
1890         if (fspath > fspath_min) {
1891                 ipath->fspath->val[i] = (u64)(unsigned long)fspath;
1892                 ++ipath->fspath->elem_cnt;
1893                 ipath->fspath->bytes_left = fspath - fspath_min;
1894         } else {
1895                 ++ipath->fspath->elem_missed;
1896                 ipath->fspath->bytes_missing += fspath_min - fspath;
1897                 ipath->fspath->bytes_left = 0;
1898         }
1899
1900         return 0;
1901 }
1902
1903 /*
1904  * this dumps all file system paths to the inode into the ipath struct, provided
1905  * is has been created large enough. each path is zero-terminated and accessed
1906  * from ipath->fspath->val[i].
1907  * when it returns, there are ipath->fspath->elem_cnt number of paths available
1908  * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the
1909  * number of missed paths in recored in ipath->fspath->elem_missed, otherwise,
1910  * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
1911  * have been needed to return all paths.
1912  */
1913 int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
1914 {
1915         return iterate_irefs(inum, ipath->fs_root, ipath->btrfs_path,
1916                              inode_to_path, ipath);
1917 }
1918
1919 struct btrfs_data_container *init_data_container(u32 total_bytes)
1920 {
1921         struct btrfs_data_container *data;
1922         size_t alloc_bytes;
1923
1924         alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
1925         data = vmalloc(alloc_bytes);
1926         if (!data)
1927                 return ERR_PTR(-ENOMEM);
1928
1929         if (total_bytes >= sizeof(*data)) {
1930                 data->bytes_left = total_bytes - sizeof(*data);
1931                 data->bytes_missing = 0;
1932         } else {
1933                 data->bytes_missing = sizeof(*data) - total_bytes;
1934                 data->bytes_left = 0;
1935         }
1936
1937         data->elem_cnt = 0;
1938         data->elem_missed = 0;
1939
1940         return data;
1941 }
1942
1943 /*
1944  * allocates space to return multiple file system paths for an inode.
1945  * total_bytes to allocate are passed, note that space usable for actual path
1946  * information will be total_bytes - sizeof(struct inode_fs_paths).
1947  * the returned pointer must be freed with free_ipath() in the end.
1948  */
1949 struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
1950                                         struct btrfs_path *path)
1951 {
1952         struct inode_fs_paths *ifp;
1953         struct btrfs_data_container *fspath;
1954
1955         fspath = init_data_container(total_bytes);
1956         if (IS_ERR(fspath))
1957                 return (void *)fspath;
1958
1959         ifp = kmalloc(sizeof(*ifp), GFP_NOFS);
1960         if (!ifp) {
1961                 kfree(fspath);
1962                 return ERR_PTR(-ENOMEM);
1963         }
1964
1965         ifp->btrfs_path = path;
1966         ifp->fspath = fspath;
1967         ifp->fs_root = fs_root;
1968
1969         return ifp;
1970 }
1971
1972 void free_ipath(struct inode_fs_paths *ipath)
1973 {
1974         if (!ipath)
1975                 return;
1976         vfree(ipath->fspath);
1977         kfree(ipath);
1978 }