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