802fabb30e1575338eec8ae53f157e9f2d7a6b56
[muen/linux.git] / fs / btrfs / backref.c
1 /*
2  * Copyright (C) 2011 STRATO.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18
19 #include <linux/vmalloc.h>
20 #include "ctree.h"
21 #include "disk-io.h"
22 #include "backref.h"
23 #include "ulist.h"
24 #include "transaction.h"
25 #include "delayed-ref.h"
26 #include "locking.h"
27
28 /* 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                 if (time_seq == (u64)-1)
255                         ret = btrfs_next_leaf(root, path);
256                 else
257                         ret = btrfs_next_old_leaf(root, path, time_seq);
258         }
259
260         while (!ret && count < total_refs) {
261                 eb = path->nodes[0];
262                 slot = path->slots[0];
263
264                 btrfs_item_key_to_cpu(eb, &key, slot);
265
266                 if (key.objectid != key_for_search->objectid ||
267                     key.type != BTRFS_EXTENT_DATA_KEY)
268                         break;
269
270                 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
271                 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
272
273                 if (disk_byte == wanted_disk_byte) {
274                         eie = NULL;
275                         old = NULL;
276                         count++;
277                         if (extent_item_pos) {
278                                 ret = check_extent_in_eb(&key, eb, fi,
279                                                 *extent_item_pos,
280                                                 &eie);
281                                 if (ret < 0)
282                                         break;
283                         }
284                         if (ret > 0)
285                                 goto next;
286                         ret = ulist_add_merge_ptr(parents, eb->start,
287                                                   eie, (void **)&old, GFP_NOFS);
288                         if (ret < 0)
289                                 break;
290                         if (!ret && extent_item_pos) {
291                                 while (old->next)
292                                         old = old->next;
293                                 old->next = eie;
294                         }
295                         eie = NULL;
296                 }
297 next:
298                 if (time_seq == (u64)-1)
299                         ret = btrfs_next_item(root, path);
300                 else
301                         ret = btrfs_next_old_item(root, path, time_seq);
302         }
303
304         if (ret > 0)
305                 ret = 0;
306         else if (ret < 0)
307                 free_inode_elem_list(eie);
308         return ret;
309 }
310
311 /*
312  * resolve an indirect backref in the form (root_id, key, level)
313  * to a logical address
314  */
315 static int __resolve_indirect_ref(struct btrfs_fs_info *fs_info,
316                                   struct btrfs_path *path, u64 time_seq,
317                                   struct __prelim_ref *ref,
318                                   struct ulist *parents,
319                                   const u64 *extent_item_pos, u64 total_refs)
320 {
321         struct btrfs_root *root;
322         struct btrfs_key root_key;
323         struct extent_buffer *eb;
324         int ret = 0;
325         int root_level;
326         int level = ref->level;
327         int index;
328
329         root_key.objectid = ref->root_id;
330         root_key.type = BTRFS_ROOT_ITEM_KEY;
331         root_key.offset = (u64)-1;
332
333         index = srcu_read_lock(&fs_info->subvol_srcu);
334
335         root = btrfs_read_fs_root_no_name(fs_info, &root_key);
336         if (IS_ERR(root)) {
337                 srcu_read_unlock(&fs_info->subvol_srcu, index);
338                 ret = PTR_ERR(root);
339                 goto out;
340         }
341
342         if (path->search_commit_root)
343                 root_level = btrfs_header_level(root->commit_root);
344         else if (time_seq == (u64)-1)
345                 root_level = btrfs_header_level(root->node);
346         else
347                 root_level = btrfs_old_root_level(root, time_seq);
348
349         if (root_level + 1 == level) {
350                 srcu_read_unlock(&fs_info->subvol_srcu, index);
351                 goto out;
352         }
353
354         path->lowest_level = level;
355         if (time_seq == (u64)-1)
356                 ret = btrfs_search_slot(NULL, root, &ref->key_for_search, path,
357                                         0, 0);
358         else
359                 ret = btrfs_search_old_slot(root, &ref->key_for_search, path,
360                                             time_seq);
361
362         /* root node has been locked, we can release @subvol_srcu safely here */
363         srcu_read_unlock(&fs_info->subvol_srcu, index);
364
365         pr_debug("search slot in root %llu (level %d, ref count %d) returned "
366                  "%d for key (%llu %u %llu)\n",
367                  ref->root_id, level, ref->count, ret,
368                  ref->key_for_search.objectid, ref->key_for_search.type,
369                  ref->key_for_search.offset);
370         if (ret < 0)
371                 goto out;
372
373         eb = path->nodes[level];
374         while (!eb) {
375                 if (WARN_ON(!level)) {
376                         ret = 1;
377                         goto out;
378                 }
379                 level--;
380                 eb = path->nodes[level];
381         }
382
383         ret = add_all_parents(root, path, parents, ref, level, time_seq,
384                               extent_item_pos, total_refs);
385 out:
386         path->lowest_level = 0;
387         btrfs_release_path(path);
388         return ret;
389 }
390
391 /*
392  * resolve all indirect backrefs from the list
393  */
394 static int __resolve_indirect_refs(struct btrfs_fs_info *fs_info,
395                                    struct btrfs_path *path, u64 time_seq,
396                                    struct list_head *head,
397                                    const u64 *extent_item_pos, u64 total_refs,
398                                    u64 root_objectid)
399 {
400         int err;
401         int ret = 0;
402         struct __prelim_ref *ref;
403         struct __prelim_ref *ref_safe;
404         struct __prelim_ref *new_ref;
405         struct ulist *parents;
406         struct ulist_node *node;
407         struct ulist_iterator uiter;
408
409         parents = ulist_alloc(GFP_NOFS);
410         if (!parents)
411                 return -ENOMEM;
412
413         /*
414          * _safe allows us to insert directly after the current item without
415          * iterating over the newly inserted items.
416          * we're also allowed to re-assign ref during iteration.
417          */
418         list_for_each_entry_safe(ref, ref_safe, head, list) {
419                 if (ref->parent)        /* already direct */
420                         continue;
421                 if (ref->count == 0)
422                         continue;
423                 if (root_objectid && ref->root_id != root_objectid) {
424                         ret = BACKREF_FOUND_SHARED;
425                         goto out;
426                 }
427                 err = __resolve_indirect_ref(fs_info, path, time_seq, ref,
428                                              parents, extent_item_pos,
429                                              total_refs);
430                 /*
431                  * we can only tolerate ENOENT,otherwise,we should catch error
432                  * and return directly.
433                  */
434                 if (err == -ENOENT) {
435                         continue;
436                 } else if (err) {
437                         ret = err;
438                         goto out;
439                 }
440
441                 /* we put the first parent into the ref at hand */
442                 ULIST_ITER_INIT(&uiter);
443                 node = ulist_next(parents, &uiter);
444                 ref->parent = node ? node->val : 0;
445                 ref->inode_list = node ?
446                         (struct extent_inode_elem *)(uintptr_t)node->aux : NULL;
447
448                 /* additional parents require new refs being added here */
449                 while ((node = ulist_next(parents, &uiter))) {
450                         new_ref = kmem_cache_alloc(btrfs_prelim_ref_cache,
451                                                    GFP_NOFS);
452                         if (!new_ref) {
453                                 ret = -ENOMEM;
454                                 goto out;
455                         }
456                         memcpy(new_ref, ref, sizeof(*ref));
457                         new_ref->parent = node->val;
458                         new_ref->inode_list = (struct extent_inode_elem *)
459                                                         (uintptr_t)node->aux;
460                         list_add(&new_ref->list, &ref->list);
461                 }
462                 ulist_reinit(parents);
463         }
464 out:
465         ulist_free(parents);
466         return ret;
467 }
468
469 static inline int ref_for_same_block(struct __prelim_ref *ref1,
470                                      struct __prelim_ref *ref2)
471 {
472         if (ref1->level != ref2->level)
473                 return 0;
474         if (ref1->root_id != ref2->root_id)
475                 return 0;
476         if (ref1->key_for_search.type != ref2->key_for_search.type)
477                 return 0;
478         if (ref1->key_for_search.objectid != ref2->key_for_search.objectid)
479                 return 0;
480         if (ref1->key_for_search.offset != ref2->key_for_search.offset)
481                 return 0;
482         if (ref1->parent != ref2->parent)
483                 return 0;
484
485         return 1;
486 }
487
488 /*
489  * read tree blocks and add keys where required.
490  */
491 static int __add_missing_keys(struct btrfs_fs_info *fs_info,
492                               struct list_head *head)
493 {
494         struct list_head *pos;
495         struct extent_buffer *eb;
496
497         list_for_each(pos, head) {
498                 struct __prelim_ref *ref;
499                 ref = list_entry(pos, struct __prelim_ref, list);
500
501                 if (ref->parent)
502                         continue;
503                 if (ref->key_for_search.type)
504                         continue;
505                 BUG_ON(!ref->wanted_disk_byte);
506                 eb = read_tree_block(fs_info->tree_root, ref->wanted_disk_byte,
507                                      0);
508                 if (IS_ERR(eb)) {
509                         return PTR_ERR(eb);
510                 } else if (!extent_buffer_uptodate(eb)) {
511                         free_extent_buffer(eb);
512                         return -EIO;
513                 }
514                 btrfs_tree_read_lock(eb);
515                 if (btrfs_header_level(eb) == 0)
516                         btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
517                 else
518                         btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
519                 btrfs_tree_read_unlock(eb);
520                 free_extent_buffer(eb);
521         }
522         return 0;
523 }
524
525 /*
526  * merge backrefs and adjust counts accordingly
527  *
528  * mode = 1: merge identical keys, if key is set
529  *    FIXME: if we add more keys in __add_prelim_ref, we can merge more here.
530  *           additionally, we could even add a key range for the blocks we
531  *           looked into to merge even more (-> replace unresolved refs by those
532  *           having a parent).
533  * mode = 2: merge identical parents
534  */
535 static void __merge_refs(struct list_head *head, int mode)
536 {
537         struct list_head *pos1;
538
539         list_for_each(pos1, head) {
540                 struct list_head *n2;
541                 struct list_head *pos2;
542                 struct __prelim_ref *ref1;
543
544                 ref1 = list_entry(pos1, struct __prelim_ref, list);
545
546                 for (pos2 = pos1->next, n2 = pos2->next; pos2 != head;
547                      pos2 = n2, n2 = pos2->next) {
548                         struct __prelim_ref *ref2;
549                         struct __prelim_ref *xchg;
550                         struct extent_inode_elem *eie;
551
552                         ref2 = list_entry(pos2, struct __prelim_ref, list);
553
554                         if (!ref_for_same_block(ref1, ref2))
555                                 continue;
556                         if (mode == 1) {
557                                 if (!ref1->parent && ref2->parent) {
558                                         xchg = ref1;
559                                         ref1 = ref2;
560                                         ref2 = xchg;
561                                 }
562                         } else {
563                                 if (ref1->parent != ref2->parent)
564                                         continue;
565                         }
566
567                         eie = ref1->inode_list;
568                         while (eie && eie->next)
569                                 eie = eie->next;
570                         if (eie)
571                                 eie->next = ref2->inode_list;
572                         else
573                                 ref1->inode_list = ref2->inode_list;
574                         ref1->count += ref2->count;
575
576                         list_del(&ref2->list);
577                         kmem_cache_free(btrfs_prelim_ref_cache, ref2);
578                 }
579
580         }
581 }
582
583 /*
584  * add all currently queued delayed refs from this head whose seq nr is
585  * smaller or equal that seq to the list
586  */
587 static int __add_delayed_refs(struct btrfs_delayed_ref_head *head, u64 seq,
588                               struct list_head *prefs, u64 *total_refs,
589                               u64 inum)
590 {
591         struct btrfs_delayed_ref_node *node;
592         struct btrfs_delayed_extent_op *extent_op = head->extent_op;
593         struct btrfs_key key;
594         struct btrfs_key op_key = {0};
595         int sgn;
596         int ret = 0;
597
598         if (extent_op && extent_op->update_key)
599                 btrfs_disk_key_to_cpu(&op_key, &extent_op->key);
600
601         spin_lock(&head->lock);
602         list_for_each_entry(node, &head->ref_list, list) {
603                 if (node->seq > seq)
604                         continue;
605
606                 switch (node->action) {
607                 case BTRFS_ADD_DELAYED_EXTENT:
608                 case BTRFS_UPDATE_DELAYED_HEAD:
609                         WARN_ON(1);
610                         continue;
611                 case BTRFS_ADD_DELAYED_REF:
612                         sgn = 1;
613                         break;
614                 case BTRFS_DROP_DELAYED_REF:
615                         sgn = -1;
616                         break;
617                 default:
618                         BUG_ON(1);
619                 }
620                 *total_refs += (node->ref_mod * sgn);
621                 switch (node->type) {
622                 case BTRFS_TREE_BLOCK_REF_KEY: {
623                         struct btrfs_delayed_tree_ref *ref;
624
625                         ref = btrfs_delayed_node_to_tree_ref(node);
626                         ret = __add_prelim_ref(prefs, ref->root, &op_key,
627                                                ref->level + 1, 0, node->bytenr,
628                                                node->ref_mod * sgn, GFP_ATOMIC);
629                         break;
630                 }
631                 case BTRFS_SHARED_BLOCK_REF_KEY: {
632                         struct btrfs_delayed_tree_ref *ref;
633
634                         ref = btrfs_delayed_node_to_tree_ref(node);
635                         ret = __add_prelim_ref(prefs, ref->root, NULL,
636                                                ref->level + 1, ref->parent,
637                                                node->bytenr,
638                                                node->ref_mod * sgn, GFP_ATOMIC);
639                         break;
640                 }
641                 case BTRFS_EXTENT_DATA_REF_KEY: {
642                         struct btrfs_delayed_data_ref *ref;
643                         ref = btrfs_delayed_node_to_data_ref(node);
644
645                         key.objectid = ref->objectid;
646                         key.type = BTRFS_EXTENT_DATA_KEY;
647                         key.offset = ref->offset;
648
649                         /*
650                          * Found a inum that doesn't match our known inum, we
651                          * know it's shared.
652                          */
653                         if (inum && ref->objectid != inum) {
654                                 ret = BACKREF_FOUND_SHARED;
655                                 break;
656                         }
657
658                         ret = __add_prelim_ref(prefs, ref->root, &key, 0, 0,
659                                                node->bytenr,
660                                                node->ref_mod * sgn, GFP_ATOMIC);
661                         break;
662                 }
663                 case BTRFS_SHARED_DATA_REF_KEY: {
664                         struct btrfs_delayed_data_ref *ref;
665
666                         ref = btrfs_delayed_node_to_data_ref(node);
667
668                         key.objectid = ref->objectid;
669                         key.type = BTRFS_EXTENT_DATA_KEY;
670                         key.offset = ref->offset;
671                         ret = __add_prelim_ref(prefs, ref->root, &key, 0,
672                                                ref->parent, node->bytenr,
673                                                node->ref_mod * sgn, GFP_ATOMIC);
674                         break;
675                 }
676                 default:
677                         WARN_ON(1);
678                 }
679                 if (ret)
680                         break;
681         }
682         spin_unlock(&head->lock);
683         return ret;
684 }
685
686 /*
687  * add all inline backrefs for bytenr to the list
688  */
689 static int __add_inline_refs(struct btrfs_fs_info *fs_info,
690                              struct btrfs_path *path, u64 bytenr,
691                              int *info_level, struct list_head *prefs,
692                              u64 *total_refs, u64 inum)
693 {
694         int ret = 0;
695         int slot;
696         struct extent_buffer *leaf;
697         struct btrfs_key key;
698         struct btrfs_key found_key;
699         unsigned long ptr;
700         unsigned long end;
701         struct btrfs_extent_item *ei;
702         u64 flags;
703         u64 item_size;
704
705         /*
706          * enumerate all inline refs
707          */
708         leaf = path->nodes[0];
709         slot = path->slots[0];
710
711         item_size = btrfs_item_size_nr(leaf, slot);
712         BUG_ON(item_size < sizeof(*ei));
713
714         ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
715         flags = btrfs_extent_flags(leaf, ei);
716         *total_refs += btrfs_extent_refs(leaf, ei);
717         btrfs_item_key_to_cpu(leaf, &found_key, slot);
718
719         ptr = (unsigned long)(ei + 1);
720         end = (unsigned long)ei + item_size;
721
722         if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
723             flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
724                 struct btrfs_tree_block_info *info;
725
726                 info = (struct btrfs_tree_block_info *)ptr;
727                 *info_level = btrfs_tree_block_level(leaf, info);
728                 ptr += sizeof(struct btrfs_tree_block_info);
729                 BUG_ON(ptr > end);
730         } else if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
731                 *info_level = found_key.offset;
732         } else {
733                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
734         }
735
736         while (ptr < end) {
737                 struct btrfs_extent_inline_ref *iref;
738                 u64 offset;
739                 int type;
740
741                 iref = (struct btrfs_extent_inline_ref *)ptr;
742                 type = btrfs_extent_inline_ref_type(leaf, iref);
743                 offset = btrfs_extent_inline_ref_offset(leaf, iref);
744
745                 switch (type) {
746                 case BTRFS_SHARED_BLOCK_REF_KEY:
747                         ret = __add_prelim_ref(prefs, 0, NULL,
748                                                 *info_level + 1, offset,
749                                                 bytenr, 1, GFP_NOFS);
750                         break;
751                 case BTRFS_SHARED_DATA_REF_KEY: {
752                         struct btrfs_shared_data_ref *sdref;
753                         int count;
754
755                         sdref = (struct btrfs_shared_data_ref *)(iref + 1);
756                         count = btrfs_shared_data_ref_count(leaf, sdref);
757                         ret = __add_prelim_ref(prefs, 0, NULL, 0, offset,
758                                                bytenr, count, GFP_NOFS);
759                         break;
760                 }
761                 case BTRFS_TREE_BLOCK_REF_KEY:
762                         ret = __add_prelim_ref(prefs, offset, NULL,
763                                                *info_level + 1, 0,
764                                                bytenr, 1, GFP_NOFS);
765                         break;
766                 case BTRFS_EXTENT_DATA_REF_KEY: {
767                         struct btrfs_extent_data_ref *dref;
768                         int count;
769                         u64 root;
770
771                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
772                         count = btrfs_extent_data_ref_count(leaf, dref);
773                         key.objectid = btrfs_extent_data_ref_objectid(leaf,
774                                                                       dref);
775                         key.type = BTRFS_EXTENT_DATA_KEY;
776                         key.offset = btrfs_extent_data_ref_offset(leaf, dref);
777
778                         if (inum && key.objectid != inum) {
779                                 ret = BACKREF_FOUND_SHARED;
780                                 break;
781                         }
782
783                         root = btrfs_extent_data_ref_root(leaf, dref);
784                         ret = __add_prelim_ref(prefs, root, &key, 0, 0,
785                                                bytenr, count, GFP_NOFS);
786                         break;
787                 }
788                 default:
789                         WARN_ON(1);
790                 }
791                 if (ret)
792                         return ret;
793                 ptr += btrfs_extent_inline_ref_size(type);
794         }
795
796         return 0;
797 }
798
799 /*
800  * add all non-inline backrefs for bytenr to the list
801  */
802 static int __add_keyed_refs(struct btrfs_fs_info *fs_info,
803                             struct btrfs_path *path, u64 bytenr,
804                             int info_level, struct list_head *prefs, u64 inum)
805 {
806         struct btrfs_root *extent_root = fs_info->extent_root;
807         int ret;
808         int slot;
809         struct extent_buffer *leaf;
810         struct btrfs_key key;
811
812         while (1) {
813                 ret = btrfs_next_item(extent_root, path);
814                 if (ret < 0)
815                         break;
816                 if (ret) {
817                         ret = 0;
818                         break;
819                 }
820
821                 slot = path->slots[0];
822                 leaf = path->nodes[0];
823                 btrfs_item_key_to_cpu(leaf, &key, slot);
824
825                 if (key.objectid != bytenr)
826                         break;
827                 if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
828                         continue;
829                 if (key.type > BTRFS_SHARED_DATA_REF_KEY)
830                         break;
831
832                 switch (key.type) {
833                 case BTRFS_SHARED_BLOCK_REF_KEY:
834                         ret = __add_prelim_ref(prefs, 0, NULL,
835                                                 info_level + 1, key.offset,
836                                                 bytenr, 1, GFP_NOFS);
837                         break;
838                 case BTRFS_SHARED_DATA_REF_KEY: {
839                         struct btrfs_shared_data_ref *sdref;
840                         int count;
841
842                         sdref = btrfs_item_ptr(leaf, slot,
843                                               struct btrfs_shared_data_ref);
844                         count = btrfs_shared_data_ref_count(leaf, sdref);
845                         ret = __add_prelim_ref(prefs, 0, NULL, 0, key.offset,
846                                                 bytenr, count, GFP_NOFS);
847                         break;
848                 }
849                 case BTRFS_TREE_BLOCK_REF_KEY:
850                         ret = __add_prelim_ref(prefs, key.offset, NULL,
851                                                info_level + 1, 0,
852                                                bytenr, 1, GFP_NOFS);
853                         break;
854                 case BTRFS_EXTENT_DATA_REF_KEY: {
855                         struct btrfs_extent_data_ref *dref;
856                         int count;
857                         u64 root;
858
859                         dref = btrfs_item_ptr(leaf, slot,
860                                               struct btrfs_extent_data_ref);
861                         count = btrfs_extent_data_ref_count(leaf, dref);
862                         key.objectid = btrfs_extent_data_ref_objectid(leaf,
863                                                                       dref);
864                         key.type = BTRFS_EXTENT_DATA_KEY;
865                         key.offset = btrfs_extent_data_ref_offset(leaf, dref);
866
867                         if (inum && key.objectid != inum) {
868                                 ret = BACKREF_FOUND_SHARED;
869                                 break;
870                         }
871
872                         root = btrfs_extent_data_ref_root(leaf, dref);
873                         ret = __add_prelim_ref(prefs, root, &key, 0, 0,
874                                                bytenr, count, GFP_NOFS);
875                         break;
876                 }
877                 default:
878                         WARN_ON(1);
879                 }
880                 if (ret)
881                         return ret;
882
883         }
884
885         return ret;
886 }
887
888 /*
889  * this adds all existing backrefs (inline backrefs, backrefs and delayed
890  * refs) for the given bytenr to the refs list, merges duplicates and resolves
891  * indirect refs to their parent bytenr.
892  * When roots are found, they're added to the roots list
893  *
894  * NOTE: This can return values > 0
895  *
896  * If time_seq is set to (u64)-1, it will not search delayed_refs, and behave
897  * much like trans == NULL case, the difference only lies in it will not
898  * commit root.
899  * The special case is for qgroup to search roots in commit_transaction().
900  *
901  * FIXME some caching might speed things up
902  */
903 static int find_parent_nodes(struct btrfs_trans_handle *trans,
904                              struct btrfs_fs_info *fs_info, u64 bytenr,
905                              u64 time_seq, struct ulist *refs,
906                              struct ulist *roots, const u64 *extent_item_pos,
907                              u64 root_objectid, u64 inum)
908 {
909         struct btrfs_key key;
910         struct btrfs_path *path;
911         struct btrfs_delayed_ref_root *delayed_refs = NULL;
912         struct btrfs_delayed_ref_head *head;
913         int info_level = 0;
914         int ret;
915         struct list_head prefs_delayed;
916         struct list_head prefs;
917         struct __prelim_ref *ref;
918         struct extent_inode_elem *eie = NULL;
919         u64 total_refs = 0;
920
921         INIT_LIST_HEAD(&prefs);
922         INIT_LIST_HEAD(&prefs_delayed);
923
924         key.objectid = bytenr;
925         key.offset = (u64)-1;
926         if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
927                 key.type = BTRFS_METADATA_ITEM_KEY;
928         else
929                 key.type = BTRFS_EXTENT_ITEM_KEY;
930
931         path = btrfs_alloc_path();
932         if (!path)
933                 return -ENOMEM;
934         if (!trans) {
935                 path->search_commit_root = 1;
936                 path->skip_locking = 1;
937         }
938
939         if (time_seq == (u64)-1)
940                 path->skip_locking = 1;
941
942         /*
943          * grab both a lock on the path and a lock on the delayed ref head.
944          * We need both to get a consistent picture of how the refs look
945          * at a specified point in time
946          */
947 again:
948         head = NULL;
949
950         ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
951         if (ret < 0)
952                 goto out;
953         BUG_ON(ret == 0);
954
955 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
956         if (trans && likely(trans->type != __TRANS_DUMMY) &&
957             time_seq != (u64)-1) {
958 #else
959         if (trans && time_seq != (u64)-1) {
960 #endif
961                 /*
962                  * look if there are updates for this ref queued and lock the
963                  * head
964                  */
965                 delayed_refs = &trans->transaction->delayed_refs;
966                 spin_lock(&delayed_refs->lock);
967                 head = btrfs_find_delayed_ref_head(trans, bytenr);
968                 if (head) {
969                         if (!mutex_trylock(&head->mutex)) {
970                                 atomic_inc(&head->node.refs);
971                                 spin_unlock(&delayed_refs->lock);
972
973                                 btrfs_release_path(path);
974
975                                 /*
976                                  * Mutex was contended, block until it's
977                                  * released and try again
978                                  */
979                                 mutex_lock(&head->mutex);
980                                 mutex_unlock(&head->mutex);
981                                 btrfs_put_delayed_ref(&head->node);
982                                 goto again;
983                         }
984                         spin_unlock(&delayed_refs->lock);
985                         ret = __add_delayed_refs(head, time_seq,
986                                                  &prefs_delayed, &total_refs,
987                                                  inum);
988                         mutex_unlock(&head->mutex);
989                         if (ret)
990                                 goto out;
991                 } else {
992                         spin_unlock(&delayed_refs->lock);
993                 }
994         }
995
996         if (path->slots[0]) {
997                 struct extent_buffer *leaf;
998                 int slot;
999
1000                 path->slots[0]--;
1001                 leaf = path->nodes[0];
1002                 slot = path->slots[0];
1003                 btrfs_item_key_to_cpu(leaf, &key, slot);
1004                 if (key.objectid == bytenr &&
1005                     (key.type == BTRFS_EXTENT_ITEM_KEY ||
1006                      key.type == BTRFS_METADATA_ITEM_KEY)) {
1007                         ret = __add_inline_refs(fs_info, path, bytenr,
1008                                                 &info_level, &prefs,
1009                                                 &total_refs, inum);
1010                         if (ret)
1011                                 goto out;
1012                         ret = __add_keyed_refs(fs_info, path, bytenr,
1013                                                info_level, &prefs, inum);
1014                         if (ret)
1015                                 goto out;
1016                 }
1017         }
1018         btrfs_release_path(path);
1019
1020         list_splice_init(&prefs_delayed, &prefs);
1021
1022         ret = __add_missing_keys(fs_info, &prefs);
1023         if (ret)
1024                 goto out;
1025
1026         __merge_refs(&prefs, 1);
1027
1028         ret = __resolve_indirect_refs(fs_info, path, time_seq, &prefs,
1029                                       extent_item_pos, total_refs,
1030                                       root_objectid);
1031         if (ret)
1032                 goto out;
1033
1034         __merge_refs(&prefs, 2);
1035
1036         while (!list_empty(&prefs)) {
1037                 ref = list_first_entry(&prefs, struct __prelim_ref, list);
1038                 WARN_ON(ref->count < 0);
1039                 if (roots && ref->count && ref->root_id && ref->parent == 0) {
1040                         if (root_objectid && ref->root_id != root_objectid) {
1041                                 ret = BACKREF_FOUND_SHARED;
1042                                 goto out;
1043                         }
1044
1045                         /* no parent == root of tree */
1046                         ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS);
1047                         if (ret < 0)
1048                                 goto out;
1049                 }
1050                 if (ref->count && ref->parent) {
1051                         if (extent_item_pos && !ref->inode_list &&
1052                             ref->level == 0) {
1053                                 struct extent_buffer *eb;
1054
1055                                 eb = read_tree_block(fs_info->extent_root,
1056                                                            ref->parent, 0);
1057                                 if (IS_ERR(eb)) {
1058                                         ret = PTR_ERR(eb);
1059                                         goto out;
1060                                 } else if (!extent_buffer_uptodate(eb)) {
1061                                         free_extent_buffer(eb);
1062                                         ret = -EIO;
1063                                         goto out;
1064                                 }
1065                                 btrfs_tree_read_lock(eb);
1066                                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1067                                 ret = find_extent_in_eb(eb, bytenr,
1068                                                         *extent_item_pos, &eie);
1069                                 btrfs_tree_read_unlock_blocking(eb);
1070                                 free_extent_buffer(eb);
1071                                 if (ret < 0)
1072                                         goto out;
1073                                 ref->inode_list = eie;
1074                         }
1075                         ret = ulist_add_merge_ptr(refs, ref->parent,
1076                                                   ref->inode_list,
1077                                                   (void **)&eie, GFP_NOFS);
1078                         if (ret < 0)
1079                                 goto out;
1080                         if (!ret && extent_item_pos) {
1081                                 /*
1082                                  * we've recorded that parent, so we must extend
1083                                  * its inode list here
1084                                  */
1085                                 BUG_ON(!eie);
1086                                 while (eie->next)
1087                                         eie = eie->next;
1088                                 eie->next = ref->inode_list;
1089                         }
1090                         eie = NULL;
1091                 }
1092                 list_del(&ref->list);
1093                 kmem_cache_free(btrfs_prelim_ref_cache, ref);
1094         }
1095
1096 out:
1097         btrfs_free_path(path);
1098         while (!list_empty(&prefs)) {
1099                 ref = list_first_entry(&prefs, struct __prelim_ref, list);
1100                 list_del(&ref->list);
1101                 kmem_cache_free(btrfs_prelim_ref_cache, ref);
1102         }
1103         while (!list_empty(&prefs_delayed)) {
1104                 ref = list_first_entry(&prefs_delayed, struct __prelim_ref,
1105                                        list);
1106                 list_del(&ref->list);
1107                 kmem_cache_free(btrfs_prelim_ref_cache, ref);
1108         }
1109         if (ret < 0)
1110                 free_inode_elem_list(eie);
1111         return ret;
1112 }
1113
1114 static void free_leaf_list(struct ulist *blocks)
1115 {
1116         struct ulist_node *node = NULL;
1117         struct extent_inode_elem *eie;
1118         struct ulist_iterator uiter;
1119
1120         ULIST_ITER_INIT(&uiter);
1121         while ((node = ulist_next(blocks, &uiter))) {
1122                 if (!node->aux)
1123                         continue;
1124                 eie = (struct extent_inode_elem *)(uintptr_t)node->aux;
1125                 free_inode_elem_list(eie);
1126                 node->aux = 0;
1127         }
1128
1129         ulist_free(blocks);
1130 }
1131
1132 /*
1133  * Finds all leafs with a reference to the specified combination of bytenr and
1134  * offset. key_list_head will point to a list of corresponding keys (caller must
1135  * free each list element). The leafs will be stored in the leafs ulist, which
1136  * must be freed with ulist_free.
1137  *
1138  * returns 0 on success, <0 on error
1139  */
1140 static int btrfs_find_all_leafs(struct btrfs_trans_handle *trans,
1141                                 struct btrfs_fs_info *fs_info, u64 bytenr,
1142                                 u64 time_seq, struct ulist **leafs,
1143                                 const u64 *extent_item_pos)
1144 {
1145         int ret;
1146
1147         *leafs = ulist_alloc(GFP_NOFS);
1148         if (!*leafs)
1149                 return -ENOMEM;
1150
1151         ret = find_parent_nodes(trans, fs_info, bytenr,
1152                                 time_seq, *leafs, NULL, extent_item_pos, 0, 0);
1153         if (ret < 0 && ret != -ENOENT) {
1154                 free_leaf_list(*leafs);
1155                 return ret;
1156         }
1157
1158         return 0;
1159 }
1160
1161 /*
1162  * walk all backrefs for a given extent to find all roots that reference this
1163  * extent. Walking a backref means finding all extents that reference this
1164  * extent and in turn walk the backrefs of those, too. Naturally this is a
1165  * recursive process, but here it is implemented in an iterative fashion: We
1166  * find all referencing extents for the extent in question and put them on a
1167  * list. In turn, we find all referencing extents for those, further appending
1168  * to the list. The way we iterate the list allows adding more elements after
1169  * the current while iterating. The process stops when we reach the end of the
1170  * list. Found roots are added to the roots list.
1171  *
1172  * returns 0 on success, < 0 on error.
1173  */
1174 static int __btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1175                                   struct btrfs_fs_info *fs_info, u64 bytenr,
1176                                   u64 time_seq, struct ulist **roots)
1177 {
1178         struct ulist *tmp;
1179         struct ulist_node *node = NULL;
1180         struct ulist_iterator uiter;
1181         int ret;
1182
1183         tmp = ulist_alloc(GFP_NOFS);
1184         if (!tmp)
1185                 return -ENOMEM;
1186         *roots = ulist_alloc(GFP_NOFS);
1187         if (!*roots) {
1188                 ulist_free(tmp);
1189                 return -ENOMEM;
1190         }
1191
1192         ULIST_ITER_INIT(&uiter);
1193         while (1) {
1194                 ret = find_parent_nodes(trans, fs_info, bytenr,
1195                                         time_seq, tmp, *roots, NULL, 0, 0);
1196                 if (ret < 0 && ret != -ENOENT) {
1197                         ulist_free(tmp);
1198                         ulist_free(*roots);
1199                         return ret;
1200                 }
1201                 node = ulist_next(tmp, &uiter);
1202                 if (!node)
1203                         break;
1204                 bytenr = node->val;
1205                 cond_resched();
1206         }
1207
1208         ulist_free(tmp);
1209         return 0;
1210 }
1211
1212 int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1213                          struct btrfs_fs_info *fs_info, u64 bytenr,
1214                          u64 time_seq, struct ulist **roots)
1215 {
1216         int ret;
1217
1218         if (!trans)
1219                 down_read(&fs_info->commit_root_sem);
1220         ret = __btrfs_find_all_roots(trans, fs_info, bytenr, time_seq, roots);
1221         if (!trans)
1222                 up_read(&fs_info->commit_root_sem);
1223         return ret;
1224 }
1225
1226 /**
1227  * btrfs_check_shared - tell us whether an extent is shared
1228  *
1229  * @trans: optional trans handle
1230  *
1231  * btrfs_check_shared uses the backref walking code but will short
1232  * circuit as soon as it finds a root or inode that doesn't match the
1233  * one passed in. This provides a significant performance benefit for
1234  * callers (such as fiemap) which want to know whether the extent is
1235  * shared but do not need a ref count.
1236  *
1237  * Return: 0 if extent is not shared, 1 if it is shared, < 0 on error.
1238  */
1239 int btrfs_check_shared(struct btrfs_trans_handle *trans,
1240                        struct btrfs_fs_info *fs_info, u64 root_objectid,
1241                        u64 inum, u64 bytenr)
1242 {
1243         struct ulist *tmp = NULL;
1244         struct ulist *roots = NULL;
1245         struct ulist_iterator uiter;
1246         struct ulist_node *node;
1247         struct seq_list elem = SEQ_LIST_INIT(elem);
1248         int ret = 0;
1249
1250         tmp = ulist_alloc(GFP_NOFS);
1251         roots = ulist_alloc(GFP_NOFS);
1252         if (!tmp || !roots) {
1253                 ulist_free(tmp);
1254                 ulist_free(roots);
1255                 return -ENOMEM;
1256         }
1257
1258         if (trans)
1259                 btrfs_get_tree_mod_seq(fs_info, &elem);
1260         else
1261                 down_read(&fs_info->commit_root_sem);
1262         ULIST_ITER_INIT(&uiter);
1263         while (1) {
1264                 ret = find_parent_nodes(trans, fs_info, bytenr, elem.seq, tmp,
1265                                         roots, NULL, root_objectid, inum);
1266                 if (ret == BACKREF_FOUND_SHARED) {
1267                         /* this is the only condition under which we return 1 */
1268                         ret = 1;
1269                         break;
1270                 }
1271                 if (ret < 0 && ret != -ENOENT)
1272                         break;
1273                 ret = 0;
1274                 node = ulist_next(tmp, &uiter);
1275                 if (!node)
1276                         break;
1277                 bytenr = node->val;
1278                 cond_resched();
1279         }
1280         if (trans)
1281                 btrfs_put_tree_mod_seq(fs_info, &elem);
1282         else
1283                 up_read(&fs_info->commit_root_sem);
1284         ulist_free(tmp);
1285         ulist_free(roots);
1286         return ret;
1287 }
1288
1289 int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,
1290                           u64 start_off, struct btrfs_path *path,
1291                           struct btrfs_inode_extref **ret_extref,
1292                           u64 *found_off)
1293 {
1294         int ret, slot;
1295         struct btrfs_key key;
1296         struct btrfs_key found_key;
1297         struct btrfs_inode_extref *extref;
1298         struct extent_buffer *leaf;
1299         unsigned long ptr;
1300
1301         key.objectid = inode_objectid;
1302         key.type = BTRFS_INODE_EXTREF_KEY;
1303         key.offset = start_off;
1304
1305         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1306         if (ret < 0)
1307                 return ret;
1308
1309         while (1) {
1310                 leaf = path->nodes[0];
1311                 slot = path->slots[0];
1312                 if (slot >= btrfs_header_nritems(leaf)) {
1313                         /*
1314                          * If the item at offset is not found,
1315                          * btrfs_search_slot will point us to the slot
1316                          * where it should be inserted. In our case
1317                          * that will be the slot directly before the
1318                          * next INODE_REF_KEY_V2 item. In the case
1319                          * that we're pointing to the last slot in a
1320                          * leaf, we must move one leaf over.
1321                          */
1322                         ret = btrfs_next_leaf(root, path);
1323                         if (ret) {
1324                                 if (ret >= 1)
1325                                         ret = -ENOENT;
1326                                 break;
1327                         }
1328                         continue;
1329                 }
1330
1331                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1332
1333                 /*
1334                  * Check that we're still looking at an extended ref key for
1335                  * this particular objectid. If we have different
1336                  * objectid or type then there are no more to be found
1337                  * in the tree and we can exit.
1338                  */
1339                 ret = -ENOENT;
1340                 if (found_key.objectid != inode_objectid)
1341                         break;
1342                 if (found_key.type != BTRFS_INODE_EXTREF_KEY)
1343                         break;
1344
1345                 ret = 0;
1346                 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1347                 extref = (struct btrfs_inode_extref *)ptr;
1348                 *ret_extref = extref;
1349                 if (found_off)
1350                         *found_off = found_key.offset;
1351                 break;
1352         }
1353
1354         return ret;
1355 }
1356
1357 /*
1358  * this iterates to turn a name (from iref/extref) into a full filesystem path.
1359  * Elements of the path are separated by '/' and the path is guaranteed to be
1360  * 0-terminated. the path is only given within the current file system.
1361  * Therefore, it never starts with a '/'. the caller is responsible to provide
1362  * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
1363  * the start point of the resulting string is returned. this pointer is within
1364  * dest, normally.
1365  * in case the path buffer would overflow, the pointer is decremented further
1366  * as if output was written to the buffer, though no more output is actually
1367  * generated. that way, the caller can determine how much space would be
1368  * required for the path to fit into the buffer. in that case, the returned
1369  * value will be smaller than dest. callers must check this!
1370  */
1371 char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
1372                         u32 name_len, unsigned long name_off,
1373                         struct extent_buffer *eb_in, u64 parent,
1374                         char *dest, u32 size)
1375 {
1376         int slot;
1377         u64 next_inum;
1378         int ret;
1379         s64 bytes_left = ((s64)size) - 1;
1380         struct extent_buffer *eb = eb_in;
1381         struct btrfs_key found_key;
1382         int leave_spinning = path->leave_spinning;
1383         struct btrfs_inode_ref *iref;
1384
1385         if (bytes_left >= 0)
1386                 dest[bytes_left] = '\0';
1387
1388         path->leave_spinning = 1;
1389         while (1) {
1390                 bytes_left -= name_len;
1391                 if (bytes_left >= 0)
1392                         read_extent_buffer(eb, dest + bytes_left,
1393                                            name_off, name_len);
1394                 if (eb != eb_in) {
1395                         btrfs_tree_read_unlock_blocking(eb);
1396                         free_extent_buffer(eb);
1397                 }
1398                 ret = btrfs_find_item(fs_root, path, parent, 0,
1399                                 BTRFS_INODE_REF_KEY, &found_key);
1400                 if (ret > 0)
1401                         ret = -ENOENT;
1402                 if (ret)
1403                         break;
1404
1405                 next_inum = found_key.offset;
1406
1407                 /* regular exit ahead */
1408                 if (parent == next_inum)
1409                         break;
1410
1411                 slot = path->slots[0];
1412                 eb = path->nodes[0];
1413                 /* make sure we can use eb after releasing the path */
1414                 if (eb != eb_in) {
1415                         atomic_inc(&eb->refs);
1416                         btrfs_tree_read_lock(eb);
1417                         btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1418                 }
1419                 btrfs_release_path(path);
1420                 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1421
1422                 name_len = btrfs_inode_ref_name_len(eb, iref);
1423                 name_off = (unsigned long)(iref + 1);
1424
1425                 parent = next_inum;
1426                 --bytes_left;
1427                 if (bytes_left >= 0)
1428                         dest[bytes_left] = '/';
1429         }
1430
1431         btrfs_release_path(path);
1432         path->leave_spinning = leave_spinning;
1433
1434         if (ret)
1435                 return ERR_PTR(ret);
1436
1437         return dest + bytes_left;
1438 }
1439
1440 /*
1441  * this makes the path point to (logical EXTENT_ITEM *)
1442  * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
1443  * tree blocks and <0 on error.
1444  */
1445 int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
1446                         struct btrfs_path *path, struct btrfs_key *found_key,
1447                         u64 *flags_ret)
1448 {
1449         int ret;
1450         u64 flags;
1451         u64 size = 0;
1452         u32 item_size;
1453         struct extent_buffer *eb;
1454         struct btrfs_extent_item *ei;
1455         struct btrfs_key key;
1456
1457         if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1458                 key.type = BTRFS_METADATA_ITEM_KEY;
1459         else
1460                 key.type = BTRFS_EXTENT_ITEM_KEY;
1461         key.objectid = logical;
1462         key.offset = (u64)-1;
1463
1464         ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
1465         if (ret < 0)
1466                 return ret;
1467
1468         ret = btrfs_previous_extent_item(fs_info->extent_root, path, 0);
1469         if (ret) {
1470                 if (ret > 0)
1471                         ret = -ENOENT;
1472                 return ret;
1473         }
1474         btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]);
1475         if (found_key->type == BTRFS_METADATA_ITEM_KEY)
1476                 size = fs_info->extent_root->nodesize;
1477         else if (found_key->type == BTRFS_EXTENT_ITEM_KEY)
1478                 size = found_key->offset;
1479
1480         if (found_key->objectid > logical ||
1481             found_key->objectid + size <= logical) {
1482                 pr_debug("logical %llu is not within any extent\n", logical);
1483                 return -ENOENT;
1484         }
1485
1486         eb = path->nodes[0];
1487         item_size = btrfs_item_size_nr(eb, path->slots[0]);
1488         BUG_ON(item_size < sizeof(*ei));
1489
1490         ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
1491         flags = btrfs_extent_flags(eb, ei);
1492
1493         pr_debug("logical %llu is at position %llu within the extent (%llu "
1494                  "EXTENT_ITEM %llu) flags %#llx size %u\n",
1495                  logical, logical - found_key->objectid, found_key->objectid,
1496                  found_key->offset, flags, item_size);
1497
1498         WARN_ON(!flags_ret);
1499         if (flags_ret) {
1500                 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1501                         *flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK;
1502                 else if (flags & BTRFS_EXTENT_FLAG_DATA)
1503                         *flags_ret = BTRFS_EXTENT_FLAG_DATA;
1504                 else
1505                         BUG_ON(1);
1506                 return 0;
1507         }
1508
1509         return -EIO;
1510 }
1511
1512 /*
1513  * helper function to iterate extent inline refs. ptr must point to a 0 value
1514  * for the first call and may be modified. it is used to track state.
1515  * if more refs exist, 0 is returned and the next call to
1516  * __get_extent_inline_ref must pass the modified ptr parameter to get the
1517  * next ref. after the last ref was processed, 1 is returned.
1518  * returns <0 on error
1519  */
1520 static int __get_extent_inline_ref(unsigned long *ptr, struct extent_buffer *eb,
1521                                    struct btrfs_key *key,
1522                                    struct btrfs_extent_item *ei, u32 item_size,
1523                                    struct btrfs_extent_inline_ref **out_eiref,
1524                                    int *out_type)
1525 {
1526         unsigned long end;
1527         u64 flags;
1528         struct btrfs_tree_block_info *info;
1529
1530         if (!*ptr) {
1531                 /* first call */
1532                 flags = btrfs_extent_flags(eb, ei);
1533                 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1534                         if (key->type == BTRFS_METADATA_ITEM_KEY) {
1535                                 /* a skinny metadata extent */
1536                                 *out_eiref =
1537                                      (struct btrfs_extent_inline_ref *)(ei + 1);
1538                         } else {
1539                                 WARN_ON(key->type != BTRFS_EXTENT_ITEM_KEY);
1540                                 info = (struct btrfs_tree_block_info *)(ei + 1);
1541                                 *out_eiref =
1542                                    (struct btrfs_extent_inline_ref *)(info + 1);
1543                         }
1544                 } else {
1545                         *out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1);
1546                 }
1547                 *ptr = (unsigned long)*out_eiref;
1548                 if ((unsigned long)(*ptr) >= (unsigned long)ei + item_size)
1549                         return -ENOENT;
1550         }
1551
1552         end = (unsigned long)ei + item_size;
1553         *out_eiref = (struct btrfs_extent_inline_ref *)(*ptr);
1554         *out_type = btrfs_extent_inline_ref_type(eb, *out_eiref);
1555
1556         *ptr += btrfs_extent_inline_ref_size(*out_type);
1557         WARN_ON(*ptr > end);
1558         if (*ptr == end)
1559                 return 1; /* last */
1560
1561         return 0;
1562 }
1563
1564 /*
1565  * reads the tree block backref for an extent. tree level and root are returned
1566  * through out_level and out_root. ptr must point to a 0 value for the first
1567  * call and may be modified (see __get_extent_inline_ref comment).
1568  * returns 0 if data was provided, 1 if there was no more data to provide or
1569  * <0 on error.
1570  */
1571 int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
1572                             struct btrfs_key *key, struct btrfs_extent_item *ei,
1573                             u32 item_size, u64 *out_root, u8 *out_level)
1574 {
1575         int ret;
1576         int type;
1577         struct btrfs_extent_inline_ref *eiref;
1578
1579         if (*ptr == (unsigned long)-1)
1580                 return 1;
1581
1582         while (1) {
1583                 ret = __get_extent_inline_ref(ptr, eb, key, ei, item_size,
1584                                               &eiref, &type);
1585                 if (ret < 0)
1586                         return ret;
1587
1588                 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1589                     type == BTRFS_SHARED_BLOCK_REF_KEY)
1590                         break;
1591
1592                 if (ret == 1)
1593                         return 1;
1594         }
1595
1596         /* we can treat both ref types equally here */
1597         *out_root = btrfs_extent_inline_ref_offset(eb, eiref);
1598
1599         if (key->type == BTRFS_EXTENT_ITEM_KEY) {
1600                 struct btrfs_tree_block_info *info;
1601
1602                 info = (struct btrfs_tree_block_info *)(ei + 1);
1603                 *out_level = btrfs_tree_block_level(eb, info);
1604         } else {
1605                 ASSERT(key->type == BTRFS_METADATA_ITEM_KEY);
1606                 *out_level = (u8)key->offset;
1607         }
1608
1609         if (ret == 1)
1610                 *ptr = (unsigned long)-1;
1611
1612         return 0;
1613 }
1614
1615 static int iterate_leaf_refs(struct extent_inode_elem *inode_list,
1616                                 u64 root, u64 extent_item_objectid,
1617                                 iterate_extent_inodes_t *iterate, void *ctx)
1618 {
1619         struct extent_inode_elem *eie;
1620         int ret = 0;
1621
1622         for (eie = inode_list; eie; eie = eie->next) {
1623                 pr_debug("ref for %llu resolved, key (%llu EXTEND_DATA %llu), "
1624                          "root %llu\n", extent_item_objectid,
1625                          eie->inum, eie->offset, root);
1626                 ret = iterate(eie->inum, eie->offset, root, ctx);
1627                 if (ret) {
1628                         pr_debug("stopping iteration for %llu due to ret=%d\n",
1629                                  extent_item_objectid, ret);
1630                         break;
1631                 }
1632         }
1633
1634         return ret;
1635 }
1636
1637 /*
1638  * calls iterate() for every inode that references the extent identified by
1639  * the given parameters.
1640  * when the iterator function returns a non-zero value, iteration stops.
1641  */
1642 int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
1643                                 u64 extent_item_objectid, u64 extent_item_pos,
1644                                 int search_commit_root,
1645                                 iterate_extent_inodes_t *iterate, void *ctx)
1646 {
1647         int ret;
1648         struct btrfs_trans_handle *trans = NULL;
1649         struct ulist *refs = NULL;
1650         struct ulist *roots = NULL;
1651         struct ulist_node *ref_node = NULL;
1652         struct ulist_node *root_node = NULL;
1653         struct seq_list tree_mod_seq_elem = SEQ_LIST_INIT(tree_mod_seq_elem);
1654         struct ulist_iterator ref_uiter;
1655         struct ulist_iterator root_uiter;
1656
1657         pr_debug("resolving all inodes for extent %llu\n",
1658                         extent_item_objectid);
1659
1660         if (!search_commit_root) {
1661                 trans = btrfs_join_transaction(fs_info->extent_root);
1662                 if (IS_ERR(trans))
1663                         return PTR_ERR(trans);
1664                 btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1665         } else {
1666                 down_read(&fs_info->commit_root_sem);
1667         }
1668
1669         ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid,
1670                                    tree_mod_seq_elem.seq, &refs,
1671                                    &extent_item_pos);
1672         if (ret)
1673                 goto out;
1674
1675         ULIST_ITER_INIT(&ref_uiter);
1676         while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) {
1677                 ret = __btrfs_find_all_roots(trans, fs_info, ref_node->val,
1678                                              tree_mod_seq_elem.seq, &roots);
1679                 if (ret)
1680                         break;
1681                 ULIST_ITER_INIT(&root_uiter);
1682                 while (!ret && (root_node = ulist_next(roots, &root_uiter))) {
1683                         pr_debug("root %llu references leaf %llu, data list "
1684                                  "%#llx\n", root_node->val, ref_node->val,
1685                                  ref_node->aux);
1686                         ret = iterate_leaf_refs((struct extent_inode_elem *)
1687                                                 (uintptr_t)ref_node->aux,
1688                                                 root_node->val,
1689                                                 extent_item_objectid,
1690                                                 iterate, ctx);
1691                 }
1692                 ulist_free(roots);
1693         }
1694
1695         free_leaf_list(refs);
1696 out:
1697         if (!search_commit_root) {
1698                 btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1699                 btrfs_end_transaction(trans, fs_info->extent_root);
1700         } else {
1701                 up_read(&fs_info->commit_root_sem);
1702         }
1703
1704         return ret;
1705 }
1706
1707 int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
1708                                 struct btrfs_path *path,
1709                                 iterate_extent_inodes_t *iterate, void *ctx)
1710 {
1711         int ret;
1712         u64 extent_item_pos;
1713         u64 flags = 0;
1714         struct btrfs_key found_key;
1715         int search_commit_root = path->search_commit_root;
1716
1717         ret = extent_from_logical(fs_info, logical, path, &found_key, &flags);
1718         btrfs_release_path(path);
1719         if (ret < 0)
1720                 return ret;
1721         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1722                 return -EINVAL;
1723
1724         extent_item_pos = logical - found_key.objectid;
1725         ret = iterate_extent_inodes(fs_info, found_key.objectid,
1726                                         extent_item_pos, search_commit_root,
1727                                         iterate, ctx);
1728
1729         return ret;
1730 }
1731
1732 typedef int (iterate_irefs_t)(u64 parent, u32 name_len, unsigned long name_off,
1733                               struct extent_buffer *eb, void *ctx);
1734
1735 static int iterate_inode_refs(u64 inum, struct btrfs_root *fs_root,
1736                               struct btrfs_path *path,
1737                               iterate_irefs_t *iterate, void *ctx)
1738 {
1739         int ret = 0;
1740         int slot;
1741         u32 cur;
1742         u32 len;
1743         u32 name_len;
1744         u64 parent = 0;
1745         int found = 0;
1746         struct extent_buffer *eb;
1747         struct btrfs_item *item;
1748         struct btrfs_inode_ref *iref;
1749         struct btrfs_key found_key;
1750
1751         while (!ret) {
1752                 ret = btrfs_find_item(fs_root, path, inum,
1753                                 parent ? parent + 1 : 0, BTRFS_INODE_REF_KEY,
1754                                 &found_key);
1755
1756                 if (ret < 0)
1757                         break;
1758                 if (ret) {
1759                         ret = found ? 0 : -ENOENT;
1760                         break;
1761                 }
1762                 ++found;
1763
1764                 parent = found_key.offset;
1765                 slot = path->slots[0];
1766                 eb = btrfs_clone_extent_buffer(path->nodes[0]);
1767                 if (!eb) {
1768                         ret = -ENOMEM;
1769                         break;
1770                 }
1771                 extent_buffer_get(eb);
1772                 btrfs_tree_read_lock(eb);
1773                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1774                 btrfs_release_path(path);
1775
1776                 item = btrfs_item_nr(slot);
1777                 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1778
1779                 for (cur = 0; cur < btrfs_item_size(eb, item); cur += len) {
1780                         name_len = btrfs_inode_ref_name_len(eb, iref);
1781                         /* path must be released before calling iterate()! */
1782                         pr_debug("following ref at offset %u for inode %llu in "
1783                                  "tree %llu\n", cur, found_key.objectid,
1784                                  fs_root->objectid);
1785                         ret = iterate(parent, name_len,
1786                                       (unsigned long)(iref + 1), eb, ctx);
1787                         if (ret)
1788                                 break;
1789                         len = sizeof(*iref) + name_len;
1790                         iref = (struct btrfs_inode_ref *)((char *)iref + len);
1791                 }
1792                 btrfs_tree_read_unlock_blocking(eb);
1793                 free_extent_buffer(eb);
1794         }
1795
1796         btrfs_release_path(path);
1797
1798         return ret;
1799 }
1800
1801 static int iterate_inode_extrefs(u64 inum, struct btrfs_root *fs_root,
1802                                  struct btrfs_path *path,
1803                                  iterate_irefs_t *iterate, void *ctx)
1804 {
1805         int ret;
1806         int slot;
1807         u64 offset = 0;
1808         u64 parent;
1809         int found = 0;
1810         struct extent_buffer *eb;
1811         struct btrfs_inode_extref *extref;
1812         struct extent_buffer *leaf;
1813         u32 item_size;
1814         u32 cur_offset;
1815         unsigned long ptr;
1816
1817         while (1) {
1818                 ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref,
1819                                             &offset);
1820                 if (ret < 0)
1821                         break;
1822                 if (ret) {
1823                         ret = found ? 0 : -ENOENT;
1824                         break;
1825                 }
1826                 ++found;
1827
1828                 slot = path->slots[0];
1829                 eb = btrfs_clone_extent_buffer(path->nodes[0]);
1830                 if (!eb) {
1831                         ret = -ENOMEM;
1832                         break;
1833                 }
1834                 extent_buffer_get(eb);
1835
1836                 btrfs_tree_read_lock(eb);
1837                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1838                 btrfs_release_path(path);
1839
1840                 leaf = path->nodes[0];
1841                 item_size = btrfs_item_size_nr(leaf, slot);
1842                 ptr = btrfs_item_ptr_offset(leaf, slot);
1843                 cur_offset = 0;
1844
1845                 while (cur_offset < item_size) {
1846                         u32 name_len;
1847
1848                         extref = (struct btrfs_inode_extref *)(ptr + cur_offset);
1849                         parent = btrfs_inode_extref_parent(eb, extref);
1850                         name_len = btrfs_inode_extref_name_len(eb, extref);
1851                         ret = iterate(parent, name_len,
1852                                       (unsigned long)&extref->name, eb, ctx);
1853                         if (ret)
1854                                 break;
1855
1856                         cur_offset += btrfs_inode_extref_name_len(leaf, extref);
1857                         cur_offset += sizeof(*extref);
1858                 }
1859                 btrfs_tree_read_unlock_blocking(eb);
1860                 free_extent_buffer(eb);
1861
1862                 offset++;
1863         }
1864
1865         btrfs_release_path(path);
1866
1867         return ret;
1868 }
1869
1870 static int iterate_irefs(u64 inum, struct btrfs_root *fs_root,
1871                          struct btrfs_path *path, iterate_irefs_t *iterate,
1872                          void *ctx)
1873 {
1874         int ret;
1875         int found_refs = 0;
1876
1877         ret = iterate_inode_refs(inum, fs_root, path, iterate, ctx);
1878         if (!ret)
1879                 ++found_refs;
1880         else if (ret != -ENOENT)
1881                 return ret;
1882
1883         ret = iterate_inode_extrefs(inum, fs_root, path, iterate, ctx);
1884         if (ret == -ENOENT && found_refs)
1885                 return 0;
1886
1887         return ret;
1888 }
1889
1890 /*
1891  * returns 0 if the path could be dumped (probably truncated)
1892  * returns <0 in case of an error
1893  */
1894 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
1895                          struct extent_buffer *eb, void *ctx)
1896 {
1897         struct inode_fs_paths *ipath = ctx;
1898         char *fspath;
1899         char *fspath_min;
1900         int i = ipath->fspath->elem_cnt;
1901         const int s_ptr = sizeof(char *);
1902         u32 bytes_left;
1903
1904         bytes_left = ipath->fspath->bytes_left > s_ptr ?
1905                                         ipath->fspath->bytes_left - s_ptr : 0;
1906
1907         fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
1908         fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len,
1909                                    name_off, eb, inum, fspath_min, bytes_left);
1910         if (IS_ERR(fspath))
1911                 return PTR_ERR(fspath);
1912
1913         if (fspath > fspath_min) {
1914                 ipath->fspath->val[i] = (u64)(unsigned long)fspath;
1915                 ++ipath->fspath->elem_cnt;
1916                 ipath->fspath->bytes_left = fspath - fspath_min;
1917         } else {
1918                 ++ipath->fspath->elem_missed;
1919                 ipath->fspath->bytes_missing += fspath_min - fspath;
1920                 ipath->fspath->bytes_left = 0;
1921         }
1922
1923         return 0;
1924 }
1925
1926 /*
1927  * this dumps all file system paths to the inode into the ipath struct, provided
1928  * is has been created large enough. each path is zero-terminated and accessed
1929  * from ipath->fspath->val[i].
1930  * when it returns, there are ipath->fspath->elem_cnt number of paths available
1931  * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the
1932  * number of missed paths in recored in ipath->fspath->elem_missed, otherwise,
1933  * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
1934  * have been needed to return all paths.
1935  */
1936 int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
1937 {
1938         return iterate_irefs(inum, ipath->fs_root, ipath->btrfs_path,
1939                              inode_to_path, ipath);
1940 }
1941
1942 struct btrfs_data_container *init_data_container(u32 total_bytes)
1943 {
1944         struct btrfs_data_container *data;
1945         size_t alloc_bytes;
1946
1947         alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
1948         data = vmalloc(alloc_bytes);
1949         if (!data)
1950                 return ERR_PTR(-ENOMEM);
1951
1952         if (total_bytes >= sizeof(*data)) {
1953                 data->bytes_left = total_bytes - sizeof(*data);
1954                 data->bytes_missing = 0;
1955         } else {
1956                 data->bytes_missing = sizeof(*data) - total_bytes;
1957                 data->bytes_left = 0;
1958         }
1959
1960         data->elem_cnt = 0;
1961         data->elem_missed = 0;
1962
1963         return data;
1964 }
1965
1966 /*
1967  * allocates space to return multiple file system paths for an inode.
1968  * total_bytes to allocate are passed, note that space usable for actual path
1969  * information will be total_bytes - sizeof(struct inode_fs_paths).
1970  * the returned pointer must be freed with free_ipath() in the end.
1971  */
1972 struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
1973                                         struct btrfs_path *path)
1974 {
1975         struct inode_fs_paths *ifp;
1976         struct btrfs_data_container *fspath;
1977
1978         fspath = init_data_container(total_bytes);
1979         if (IS_ERR(fspath))
1980                 return (void *)fspath;
1981
1982         ifp = kmalloc(sizeof(*ifp), GFP_NOFS);
1983         if (!ifp) {
1984                 kfree(fspath);
1985                 return ERR_PTR(-ENOMEM);
1986         }
1987
1988         ifp->btrfs_path = path;
1989         ifp->fspath = fspath;
1990         ifp->fs_root = fs_root;
1991
1992         return ifp;
1993 }
1994
1995 void free_ipath(struct inode_fs_paths *ipath)
1996 {
1997         if (!ipath)
1998                 return;
1999         vfree(ipath->fspath);
2000         kfree(ipath);
2001 }