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                 /*
 212                  * We can often find data backrefs with an offset that is too
 213                  * large (>= LLONG_MAX, maximum allowed file offset) due to
 214                  * underflows when subtracting a file's offset with the data
 215                  * offset of its corresponding extent data item. This can
 216                  * happen for example in the clone ioctl.
 217                  * So if we detect such case we set the search key's offset to
 218                  * zero to make sure we will find the matching file extent item
 219                  * at add_all_parents(), otherwise we will miss it because the
 220                  * offset taken form the backref is much larger then the offset
 221                  * of the file extent item. This can make us scan a very large
 222                  * number of file extent items, but at least it will not make
 223                  * us miss any.
 224                  * This is an ugly workaround for a behaviour that should have
 225                  * never existed, but it does and a fix for the clone ioctl
 226                  * would touch a lot of places, cause backwards incompatibility
 227                  * and would not fix the problem for extents cloned with older
 228                  * kernels.
 229                  */
 230                 if (ref->key_for_search.type == BTRFS_EXTENT_DATA_KEY &&
 231                     ref->key_for_search.offset >= LLONG_MAX)
 232                         ref->key_for_search.offset = 0;
 233         } else {
 234                 memset(&ref->key_for_search, 0, sizeof(ref->key_for_search));
 235         }
 236
 237         ref->inode_list = NULL;
 238         ref->level = level;
 239         ref->count = count;
 240         ref->parent = parent;
 241         ref->wanted_disk_byte = wanted_disk_byte;
 242         list_add_tail(&ref->list, head);
 243
 244         return 0;
 245 }
 246
 247 static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path,
 248                            struct ulist *parents, struct __prelim_ref *ref,
 249                            int level, u64 time_seq, const u64 *extent_item_pos,
 250                            u64 total_refs)
 251 {
 252         int ret = 0;
 253         int slot;
 254         struct extent_buffer *eb;
 255         struct btrfs_key key;
 256         struct btrfs_key *key_for_search = &ref->key_for_search;
 257         struct btrfs_file_extent_item *fi;
 258         struct extent_inode_elem *eie = NULL, *old = NULL;
 259         u64 disk_byte;
 260         u64 wanted_disk_byte = ref->wanted_disk_byte;
 261         u64 count = 0;
 262
 263         if (level != 0) {
 264                 eb = path->nodes[level];
 265                 ret = ulist_add(parents, eb->start, 0, GFP_NOFS);
 266                 if (ret < 0)
 267                         return ret;
 268                 return 0;
 269         }
 270
 271         /*
 272          * We normally enter this function with the path already pointing to
 273          * the first item to check. But sometimes, we may enter it with
 274          * slot==nritems. In that case, go to the next leaf before we continue.
 275          */
 276         if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
 277                 if (time_seq == (u64)-1)
 278                         ret = btrfs_next_leaf(root, path);
 279                 else
 280                         ret = btrfs_next_old_leaf(root, path, time_seq);
 281         }
 282
 283         while (!ret && count < total_refs) {
 284                 eb = path->nodes[0];
 285                 slot = path->slots[0];
 286
 287                 btrfs_item_key_to_cpu(eb, &key, slot);
 288
 289                 if (key.objectid != key_for_search->objectid ||
 290                     key.type != BTRFS_EXTENT_DATA_KEY)
 291                         break;
 292
 293                 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
 294                 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
 295
 296                 if (disk_byte == wanted_disk_byte) {
 297                         eie = NULL;
 298                         old = NULL;
 299                         count++;
 300                         if (extent_item_pos) {
 301                                 ret = check_extent_in_eb(&key, eb, fi,
 302                                                 *extent_item_pos,
 303                                                 &eie);
 304                                 if (ret < 0)
 305                                         break;
 306                         }
 307                         if (ret > 0)
 308                                 goto next;
 309                         ret = ulist_add_merge_ptr(parents, eb->start,
 310                                                   eie, (void **)&old, GFP_NOFS);
 311                         if (ret < 0)
 312                                 break;
 313                         if (!ret && extent_item_pos) {
 314                                 while (old->next)
 315                                         old = old->next;
 316                                 old->next = eie;
 317                         }
 318                         eie = NULL;
 319                 }
 320 next:
 321                 if (time_seq == (u64)-1)
 322                         ret = btrfs_next_item(root, path);
 323                 else
 324                         ret = btrfs_next_old_item(root, path, time_seq);
 325         }
 326
 327         if (ret > 0)
 328                 ret = 0;
 329         else if (ret < 0)
 330                 free_inode_elem_list(eie);
 331         return ret;
 332 }
 333
 334 /*
 335  * resolve an indirect backref in the form (root_id, key, level)
 336  * to a logical address
 337  */
 338 static int __resolve_indirect_ref(struct btrfs_fs_info *fs_info,
 339                                   struct btrfs_path *path, u64 time_seq,
 340                                   struct __prelim_ref *ref,
 341                                   struct ulist *parents,
 342                                   const u64 *extent_item_pos, u64 total_refs)
 343 {
 344         struct btrfs_root *root;
 345         struct btrfs_key root_key;
 346         struct extent_buffer *eb;
 347         int ret = 0;
 348         int root_level;
 349         int level = ref->level;
 350         int index;
 351
 352         root_key.objectid = ref->root_id;
 353         root_key.type = BTRFS_ROOT_ITEM_KEY;
 354         root_key.offset = (u64)-1;
 355
 356         index = srcu_read_lock(&fs_info->subvol_srcu);
 357
 358         root = btrfs_get_fs_root(fs_info, &root_key, false);
 359         if (IS_ERR(root)) {
 360                 srcu_read_unlock(&fs_info->subvol_srcu, index);
 361                 ret = PTR_ERR(root);
 362                 goto out;
 363         }
 364
 365         if (btrfs_test_is_dummy_root(root)) {
 366                 srcu_read_unlock(&fs_info->subvol_srcu, index);
 367                 ret = -ENOENT;
 368                 goto out;
 369         }
 370
 371         if (path->search_commit_root)
 372                 root_level = btrfs_header_level(root->commit_root);
 373         else if (time_seq == (u64)-1)
 374                 root_level = btrfs_header_level(root->node);
 375         else
 376                 root_level = btrfs_old_root_level(root, time_seq);
 377
 378         if (root_level + 1 == level) {
 379                 srcu_read_unlock(&fs_info->subvol_srcu, index);
 380                 goto out;
 381         }
 382
 383         path->lowest_level = level;
 384         if (time_seq == (u64)-1)
 385                 ret = btrfs_search_slot(NULL, root, &ref->key_for_search, path,
 386                                         0, 0);
 387         else
 388                 ret = btrfs_search_old_slot(root, &ref->key_for_search, path,
 389                                             time_seq);
 390
 391         /* root node has been locked, we can release @subvol_srcu safely here */
 392         srcu_read_unlock(&fs_info->subvol_srcu, index);
 393
 394         pr_debug("search slot in root %llu (level %d, ref count %d) returned "
 395                  "%d for key (%llu %u %llu)\n",
 396                  ref->root_id, level, ref->count, ret,
 397                  ref->key_for_search.objectid, ref->key_for_search.type,
 398                  ref->key_for_search.offset);
 399         if (ret < 0)
 400                 goto out;
 401
 402         eb = path->nodes[level];
 403         while (!eb) {
 404                 if (WARN_ON(!level)) {
 405                         ret = 1;
 406                         goto out;
 407                 }
 408                 level--;
 409                 eb = path->nodes[level];
 410         }
 411
 412         ret = add_all_parents(root, path, parents, ref, level, time_seq,
 413                               extent_item_pos, total_refs);
 414 out:
 415         path->lowest_level = 0;
 416         btrfs_release_path(path);
 417         return ret;
 418 }
 419
 420 /*
 421  * resolve all indirect backrefs from the list
 422  */
 423 static int __resolve_indirect_refs(struct btrfs_fs_info *fs_info,
 424                                    struct btrfs_path *path, u64 time_seq,
 425                                    struct list_head *head,
 426                                    const u64 *extent_item_pos, u64 total_refs,
 427                                    u64 root_objectid)
 428 {
 429         int err;
 430         int ret = 0;
 431         struct __prelim_ref *ref;
 432         struct __prelim_ref *ref_safe;
 433         struct __prelim_ref *new_ref;
 434         struct ulist *parents;
 435         struct ulist_node *node;
 436         struct ulist_iterator uiter;
 437
 438         parents = ulist_alloc(GFP_NOFS);
 439         if (!parents)
 440                 return -ENOMEM;
 441
 442         /*
 443          * _safe allows us to insert directly after the current item without
 444          * iterating over the newly inserted items.
 445          * we're also allowed to re-assign ref during iteration.
 446          */
 447         list_for_each_entry_safe(ref, ref_safe, head, list) {
 448                 if (ref->parent)        /* already direct */
 449                         continue;
 450                 if (ref->count == 0)
 451                         continue;
 452                 if (root_objectid && ref->root_id != root_objectid) {
 453                         ret = BACKREF_FOUND_SHARED;
 454                         goto out;
 455                 }
 456                 err = __resolve_indirect_ref(fs_info, path, time_seq, ref,
 457                                              parents, extent_item_pos,
 458                                              total_refs);
 459                 /*
 460                  * we can only tolerate ENOENT,otherwise,we should catch error
 461                  * and return directly.
 462                  */
 463                 if (err == -ENOENT) {
 464                         continue;
 465                 } else if (err) {
 466                         ret = err;
 467                         goto out;
 468                 }
 469
 470                 /* we put the first parent into the ref at hand */
 471                 ULIST_ITER_INIT(&uiter);
 472                 node = ulist_next(parents, &uiter);
 473                 ref->parent = node ? node->val : 0;
 474                 ref->inode_list = node ?
 475                         (struct extent_inode_elem *)(uintptr_t)node->aux : NULL;
 476
 477                 /* additional parents require new refs being added here */
 478                 while ((node = ulist_next(parents, &uiter))) {
 479                         new_ref = kmem_cache_alloc(btrfs_prelim_ref_cache,
 480                                                    GFP_NOFS);
 481                         if (!new_ref) {
 482                                 ret = -ENOMEM;
 483                                 goto out;
 484                         }
 485                         memcpy(new_ref, ref, sizeof(*ref));
 486                         new_ref->parent = node->val;
 487                         new_ref->inode_list = (struct extent_inode_elem *)
 488                                                         (uintptr_t)node->aux;
 489                         list_add(&new_ref->list, &ref->list);
 490                 }
 491                 ulist_reinit(parents);
 492         }
 493 out:
 494         ulist_free(parents);
 495         return ret;
 496 }
 497
 498 static inline int ref_for_same_block(struct __prelim_ref *ref1,
 499                                      struct __prelim_ref *ref2)
 500 {
 501         if (ref1->level != ref2->level)
 502                 return 0;
 503         if (ref1->root_id != ref2->root_id)
 504                 return 0;
 505         if (ref1->key_for_search.type != ref2->key_for_search.type)
 506                 return 0;
 507         if (ref1->key_for_search.objectid != ref2->key_for_search.objectid)
 508                 return 0;
 509         if (ref1->key_for_search.offset != ref2->key_for_search.offset)
 510                 return 0;
 511         if (ref1->parent != ref2->parent)
 512                 return 0;
 513
 514         return 1;
 515 }
 516
 517 /*
 518  * read tree blocks and add keys where required.
 519  */
 520 static int __add_missing_keys(struct btrfs_fs_info *fs_info,
 521                               struct list_head *head)
 522 {
 523         struct list_head *pos;
 524         struct extent_buffer *eb;
 525
 526         list_for_each(pos, head) {
 527                 struct __prelim_ref *ref;
 528                 ref = list_entry(pos, struct __prelim_ref, list);
 529
 530                 if (ref->parent)
 531                         continue;
 532                 if (ref->key_for_search.type)
 533                         continue;
 534                 BUG_ON(!ref->wanted_disk_byte);
 535                 eb = read_tree_block(fs_info->tree_root, ref->wanted_disk_byte,
 536                                      0);
 537                 if (IS_ERR(eb)) {
 538                         return PTR_ERR(eb);
 539                 } else if (!extent_buffer_uptodate(eb)) {
 540                         free_extent_buffer(eb);
 541                         return -EIO;
 542                 }
 543                 btrfs_tree_read_lock(eb);
 544                 if (btrfs_header_level(eb) == 0)
 545                         btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
 546                 else
 547                         btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
 548                 btrfs_tree_read_unlock(eb);
 549                 free_extent_buffer(eb);
 550         }
 551         return 0;
 552 }
 553
 554 /*
 555  * merge backrefs and adjust counts accordingly
 556  *
 557  * mode = 1: merge identical keys, if key is set
 558  *    FIXME: if we add more keys in __add_prelim_ref, we can merge more here.
 559  *           additionally, we could even add a key range for the blocks we
 560  *           looked into to merge even more (-> replace unresolved refs by those
 561  *           having a parent).
 562  * mode = 2: merge identical parents
 563  */
 564 static void __merge_refs(struct list_head *head, int mode)
 565 {
 566         struct list_head *pos1;
 567
 568         list_for_each(pos1, head) {
 569                 struct list_head *n2;
 570                 struct list_head *pos2;
 571                 struct __prelim_ref *ref1;
 572
 573                 ref1 = list_entry(pos1, struct __prelim_ref, list);
 574
 575                 for (pos2 = pos1->next, n2 = pos2->next; pos2 != head;
 576                      pos2 = n2, n2 = pos2->next) {
 577                         struct __prelim_ref *ref2;
 578                         struct __prelim_ref *xchg;
 579                         struct extent_inode_elem *eie;
 580
 581                         ref2 = list_entry(pos2, struct __prelim_ref, list);
 582
 583                         if (!ref_for_same_block(ref1, ref2))
 584                                 continue;
 585                         if (mode == 1) {
 586                                 if (!ref1->parent && ref2->parent) {
 587                                         xchg = ref1;
 588                                         ref1 = ref2;
 589                                         ref2 = xchg;
 590                                 }
 591                         } else {
 592                                 if (ref1->parent != ref2->parent)
 593                                         continue;
 594                         }
 595
 596                         eie = ref1->inode_list;
 597                         while (eie && eie->next)
 598                                 eie = eie->next;
 599                         if (eie)
 600                                 eie->next = ref2->inode_list;
 601                         else
 602                                 ref1->inode_list = ref2->inode_list;
 603                         ref1->count += ref2->count;
 604
 605                         list_del(&ref2->list);
 606                         kmem_cache_free(btrfs_prelim_ref_cache, ref2);
 607                 }
 608
 609         }
 610 }
 611
 612 /*
 613  * add all currently queued delayed refs from this head whose seq nr is
 614  * smaller or equal that seq to the list
 615  */
 616 static int __add_delayed_refs(struct btrfs_delayed_ref_head *head, u64 seq,
 617                               struct list_head *prefs, u64 *total_refs,
 618                               u64 inum)
 619 {
 620         struct btrfs_delayed_ref_node *node;
 621         struct btrfs_delayed_extent_op *extent_op = head->extent_op;
 622         struct btrfs_key key;
 623         struct btrfs_key op_key = {0};
 624         int sgn;
 625         int ret = 0;
 626
 627         if (extent_op && extent_op->update_key)
 628                 btrfs_disk_key_to_cpu(&op_key, &extent_op->key);
 629
 630         spin_lock(&head->lock);
 631         list_for_each_entry(node, &head->ref_list, list) {
 632                 if (node->seq > seq)
 633                         continue;
 634
 635                 switch (node->action) {
 636                 case BTRFS_ADD_DELAYED_EXTENT:
 637                 case BTRFS_UPDATE_DELAYED_HEAD:
 638                         WARN_ON(1);
 639                         continue;
 640                 case BTRFS_ADD_DELAYED_REF:
 641                         sgn = 1;
 642                         break;
 643                 case BTRFS_DROP_DELAYED_REF:
 644                         sgn = -1;
 645                         break;
 646                 default:
 647                         BUG_ON(1);
 648                 }
 649                 *total_refs += (node->ref_mod * sgn);
 650                 switch (node->type) {
 651                 case BTRFS_TREE_BLOCK_REF_KEY: {
 652                         struct btrfs_delayed_tree_ref *ref;
 653
 654                         ref = btrfs_delayed_node_to_tree_ref(node);
 655                         ret = __add_prelim_ref(prefs, ref->root, &op_key,
 656                                                ref->level + 1, 0, node->bytenr,
 657                                                node->ref_mod * sgn, GFP_ATOMIC);
 658                         break;
 659                 }
 660                 case BTRFS_SHARED_BLOCK_REF_KEY: {
 661                         struct btrfs_delayed_tree_ref *ref;
 662
 663                         ref = btrfs_delayed_node_to_tree_ref(node);
 664                         ret = __add_prelim_ref(prefs, 0, NULL,
 665                                                ref->level + 1, ref->parent,
 666                                                node->bytenr,
 667                                                node->ref_mod * sgn, GFP_ATOMIC);
 668                         break;
 669                 }
 670                 case BTRFS_EXTENT_DATA_REF_KEY: {
 671                         struct btrfs_delayed_data_ref *ref;
 672                         ref = btrfs_delayed_node_to_data_ref(node);
 673
 674                         key.objectid = ref->objectid;
 675                         key.type = BTRFS_EXTENT_DATA_KEY;
 676                         key.offset = ref->offset;
 677
 678                         /*
 679                          * Found a inum that doesn't match our known inum, we
 680                          * know it's shared.
 681                          */
 682                         if (inum && ref->objectid != inum) {
 683                                 ret = BACKREF_FOUND_SHARED;
 684                                 break;
 685                         }
 686
 687                         ret = __add_prelim_ref(prefs, ref->root, &key, 0, 0,
 688                                                node->bytenr,
 689                                                node->ref_mod * sgn, GFP_ATOMIC);
 690                         break;
 691                 }
 692                 case BTRFS_SHARED_DATA_REF_KEY: {
 693                         struct btrfs_delayed_data_ref *ref;
 694
 695                         ref = btrfs_delayed_node_to_data_ref(node);
 696                         ret = __add_prelim_ref(prefs, 0, NULL, 0,
 697                                                ref->parent, node->bytenr,
 698                                                node->ref_mod * sgn, GFP_ATOMIC);
 699                         break;
 700                 }
 701                 default:
 702                         WARN_ON(1);
 703                 }
 704                 if (ret)
 705                         break;
 706         }
 707         spin_unlock(&head->lock);
 708         return ret;
 709 }
 710
 711 /*
 712  * add all inline backrefs for bytenr to the list
 713  */
 714 static int __add_inline_refs(struct btrfs_fs_info *fs_info,
 715                              struct btrfs_path *path, u64 bytenr,
 716                              int *info_level, struct list_head *prefs,
 717                              u64 *total_refs, u64 inum)
 718 {
 719         int ret = 0;
 720         int slot;
 721         struct extent_buffer *leaf;
 722         struct btrfs_key key;
 723         struct btrfs_key found_key;
 724         unsigned long ptr;
 725         unsigned long end;
 726         struct btrfs_extent_item *ei;
 727         u64 flags;
 728         u64 item_size;
 729
 730         /*
 731          * enumerate all inline refs
 732          */
 733         leaf = path->nodes[0];
 734         slot = path->slots[0];
 735
 736         item_size = btrfs_item_size_nr(leaf, slot);
 737         BUG_ON(item_size < sizeof(*ei));
 738
 739         ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
 740         flags = btrfs_extent_flags(leaf, ei);
 741         *total_refs += btrfs_extent_refs(leaf, ei);
 742         btrfs_item_key_to_cpu(leaf, &found_key, slot);
 743
 744         ptr = (unsigned long)(ei + 1);
 745         end = (unsigned long)ei + item_size;
 746
 747         if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
 748             flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
 749                 struct btrfs_tree_block_info *info;
 750
 751                 info = (struct btrfs_tree_block_info *)ptr;
 752                 *info_level = btrfs_tree_block_level(leaf, info);
 753                 ptr += sizeof(struct btrfs_tree_block_info);
 754                 BUG_ON(ptr > end);
 755         } else if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
 756                 *info_level = found_key.offset;
 757         } else {
 758                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
 759         }
 760
 761         while (ptr < end) {
 762                 struct btrfs_extent_inline_ref *iref;
 763                 u64 offset;
 764                 int type;
 765
 766                 iref = (struct btrfs_extent_inline_ref *)ptr;
 767                 type = btrfs_extent_inline_ref_type(leaf, iref);
 768                 offset = btrfs_extent_inline_ref_offset(leaf, iref);
 769
 770                 switch (type) {
 771                 case BTRFS_SHARED_BLOCK_REF_KEY:
 772                         ret = __add_prelim_ref(prefs, 0, NULL,
 773                                                 *info_level + 1, offset,
 774                                                 bytenr, 1, GFP_NOFS);
 775                         break;
 776                 case BTRFS_SHARED_DATA_REF_KEY: {
 777                         struct btrfs_shared_data_ref *sdref;
 778                         int count;
 779
 780                         sdref = (struct btrfs_shared_data_ref *)(iref + 1);
 781                         count = btrfs_shared_data_ref_count(leaf, sdref);
 782                         ret = __add_prelim_ref(prefs, 0, NULL, 0, offset,
 783                                                bytenr, count, GFP_NOFS);
 784                         break;
 785                 }
 786                 case BTRFS_TREE_BLOCK_REF_KEY:
 787                         ret = __add_prelim_ref(prefs, offset, NULL,
 788                                                *info_level + 1, 0,
 789                                                bytenr, 1, GFP_NOFS);
 790                         break;
 791                 case BTRFS_EXTENT_DATA_REF_KEY: {
 792                         struct btrfs_extent_data_ref *dref;
 793                         int count;
 794                         u64 root;
 795
 796                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
 797                         count = btrfs_extent_data_ref_count(leaf, dref);
 798                         key.objectid = btrfs_extent_data_ref_objectid(leaf,
 799                                                                       dref);
 800                         key.type = BTRFS_EXTENT_DATA_KEY;
 801                         key.offset = btrfs_extent_data_ref_offset(leaf, dref);
 802
 803                         if (inum && key.objectid != inum) {
 804                                 ret = BACKREF_FOUND_SHARED;
 805                                 break;
 806                         }
 807
 808                         root = btrfs_extent_data_ref_root(leaf, dref);
 809                         ret = __add_prelim_ref(prefs, root, &key, 0, 0,
 810                                                bytenr, count, GFP_NOFS);
 811                         break;
 812                 }
 813                 default:
 814                         WARN_ON(1);
 815                 }
 816                 if (ret)
 817                         return ret;
 818                 ptr += btrfs_extent_inline_ref_size(type);
 819         }
 820
 821         return 0;
 822 }
 823
 824 /*
 825  * add all non-inline backrefs for bytenr to the list
 826  */
 827 static int __add_keyed_refs(struct btrfs_fs_info *fs_info,
 828                             struct btrfs_path *path, u64 bytenr,
 829                             int info_level, struct list_head *prefs, u64 inum)
 830 {
 831         struct btrfs_root *extent_root = fs_info->extent_root;
 832         int ret;
 833         int slot;
 834         struct extent_buffer *leaf;
 835         struct btrfs_key key;
 836
 837         while (1) {
 838                 ret = btrfs_next_item(extent_root, path);
 839                 if (ret < 0)
 840                         break;
 841                 if (ret) {
 842                         ret = 0;
 843                         break;
 844                 }
 845
 846                 slot = path->slots[0];
 847                 leaf = path->nodes[0];
 848                 btrfs_item_key_to_cpu(leaf, &key, slot);
 849
 850                 if (key.objectid != bytenr)
 851                         break;
 852                 if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
 853                         continue;
 854                 if (key.type > BTRFS_SHARED_DATA_REF_KEY)
 855                         break;
 856
 857                 switch (key.type) {
 858                 case BTRFS_SHARED_BLOCK_REF_KEY:
 859                         ret = __add_prelim_ref(prefs, 0, NULL,
 860                                                 info_level + 1, key.offset,
 861                                                 bytenr, 1, GFP_NOFS);
 862                         break;
 863                 case BTRFS_SHARED_DATA_REF_KEY: {
 864                         struct btrfs_shared_data_ref *sdref;
 865                         int count;
 866
 867                         sdref = btrfs_item_ptr(leaf, slot,
 868                                               struct btrfs_shared_data_ref);
 869                         count = btrfs_shared_data_ref_count(leaf, sdref);
 870                         ret = __add_prelim_ref(prefs, 0, NULL, 0, key.offset,
 871                                                 bytenr, count, GFP_NOFS);
 872                         break;
 873                 }
 874                 case BTRFS_TREE_BLOCK_REF_KEY:
 875                         ret = __add_prelim_ref(prefs, key.offset, NULL,
 876                                                info_level + 1, 0,
 877                                                bytenr, 1, GFP_NOFS);
 878                         break;
 879                 case BTRFS_EXTENT_DATA_REF_KEY: {
 880                         struct btrfs_extent_data_ref *dref;
 881                         int count;
 882                         u64 root;
 883
 884                         dref = btrfs_item_ptr(leaf, slot,
 885                                               struct btrfs_extent_data_ref);
 886                         count = btrfs_extent_data_ref_count(leaf, dref);
 887                         key.objectid = btrfs_extent_data_ref_objectid(leaf,
 888                                                                       dref);
 889                         key.type = BTRFS_EXTENT_DATA_KEY;
 890                         key.offset = btrfs_extent_data_ref_offset(leaf, dref);
 891
 892                         if (inum && key.objectid != inum) {
 893                                 ret = BACKREF_FOUND_SHARED;
 894                                 break;
 895                         }
 896
 897                         root = btrfs_extent_data_ref_root(leaf, dref);
 898                         ret = __add_prelim_ref(prefs, root, &key, 0, 0,
 899                                                bytenr, count, GFP_NOFS);
 900                         break;
 901                 }
 902                 default:
 903                         WARN_ON(1);
 904                 }
 905                 if (ret)
 906                         return ret;
 907
 908         }
 909
 910         return ret;
 911 }
 912
 913 /*
 914  * this adds all existing backrefs (inline backrefs, backrefs and delayed
 915  * refs) for the given bytenr to the refs list, merges duplicates and resolves
 916  * indirect refs to their parent bytenr.
 917  * When roots are found, they're added to the roots list
 918  *
 919  * NOTE: This can return values > 0
 920  *
 921  * If time_seq is set to (u64)-1, it will not search delayed_refs, and behave
 922  * much like trans == NULL case, the difference only lies in it will not
 923  * commit root.
 924  * The special case is for qgroup to search roots in commit_transaction().
 925  *
 926  * FIXME some caching might speed things up
 927  */
 928 static int find_parent_nodes(struct btrfs_trans_handle *trans,
 929                              struct btrfs_fs_info *fs_info, u64 bytenr,
 930                              u64 time_seq, struct ulist *refs,
 931                              struct ulist *roots, const u64 *extent_item_pos,
 932                              u64 root_objectid, u64 inum)
 933 {
 934         struct btrfs_key key;
 935         struct btrfs_path *path;
 936         struct btrfs_delayed_ref_root *delayed_refs = NULL;
 937         struct btrfs_delayed_ref_head *head;
 938         int info_level = 0;
 939         int ret;
 940         struct list_head prefs_delayed;
 941         struct list_head prefs;
 942         struct __prelim_ref *ref;
 943         struct extent_inode_elem *eie = NULL;
 944         u64 total_refs = 0;
 945
 946         INIT_LIST_HEAD(&prefs);
 947         INIT_LIST_HEAD(&prefs_delayed);
 948
 949         key.objectid = bytenr;
 950         key.offset = (u64)-1;
 951         if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
 952                 key.type = BTRFS_METADATA_ITEM_KEY;
 953         else
 954                 key.type = BTRFS_EXTENT_ITEM_KEY;
 955
 956         path = btrfs_alloc_path();
 957         if (!path)
 958                 return -ENOMEM;
 959         if (!trans) {
 960                 path->search_commit_root = 1;
 961                 path->skip_locking = 1;
 962         }
 963
 964         if (time_seq == (u64)-1)
 965                 path->skip_locking = 1;
 966
 967         /*
 968          * grab both a lock on the path and a lock on the delayed ref head.
 969          * We need both to get a consistent picture of how the refs look
 970          * at a specified point in time
 971          */
 972 again:
 973         head = NULL;
 974
 975         ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
 976         if (ret < 0)
 977                 goto out;
 978         BUG_ON(ret == 0);
 979
 980 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
 981         if (trans && likely(trans->type != __TRANS_DUMMY) &&
 982             time_seq != (u64)-1) {
 983 #else
 984         if (trans && time_seq != (u64)-1) {
 985 #endif
 986                 /*
 987                  * look if there are updates for this ref queued and lock the
 988                  * head
 989                  */
 990                 delayed_refs = &trans->transaction->delayed_refs;
 991                 spin_lock(&delayed_refs->lock);
 992                 head = btrfs_find_delayed_ref_head(trans, bytenr);
 993                 if (head) {
 994                         if (!mutex_trylock(&head->mutex)) {
 995                                 atomic_inc(&head->node.refs);
 996                                 spin_unlock(&delayed_refs->lock);
 997
 998                                 btrfs_release_path(path);
 999
1000                                 /*
1001                                  * Mutex was contended, block until it's
1002                                  * released and try again
1003                                  */
1004                                 mutex_lock(&head->mutex);
1005                                 mutex_unlock(&head->mutex);
1006                                 btrfs_put_delayed_ref(&head->node);
1007                                 goto again;
1008                         }
1009                         spin_unlock(&delayed_refs->lock);
1010                         ret = __add_delayed_refs(head, time_seq,
1011                                                  &prefs_delayed, &total_refs,
1012                                                  inum);
1013                         mutex_unlock(&head->mutex);
1014                         if (ret)
1015                                 goto out;
1016                 } else {
1017                         spin_unlock(&delayed_refs->lock);
1018                 }
1019         }
1020
1021         if (path->slots[0]) {
1022                 struct extent_buffer *leaf;
1023                 int slot;
1024
1025                 path->slots[0]--;
1026                 leaf = path->nodes[0];
1027                 slot = path->slots[0];
1028                 btrfs_item_key_to_cpu(leaf, &key, slot);
1029                 if (key.objectid == bytenr &&
1030                     (key.type == BTRFS_EXTENT_ITEM_KEY ||
1031                      key.type == BTRFS_METADATA_ITEM_KEY)) {
1032                         ret = __add_inline_refs(fs_info, path, bytenr,
1033                                                 &info_level, &prefs,
1034                                                 &total_refs, inum);
1035                         if (ret)
1036                                 goto out;
1037                         ret = __add_keyed_refs(fs_info, path, bytenr,
1038                                                info_level, &prefs, inum);
1039                         if (ret)
1040                                 goto out;
1041                 }
1042         }
1043         btrfs_release_path(path);
1044
1045         list_splice_init(&prefs_delayed, &prefs);
1046
1047         ret = __add_missing_keys(fs_info, &prefs);
1048         if (ret)
1049                 goto out;
1050
1051         __merge_refs(&prefs, 1);
1052
1053         ret = __resolve_indirect_refs(fs_info, path, time_seq, &prefs,
1054                                       extent_item_pos, total_refs,
1055                                       root_objectid);
1056         if (ret)
1057                 goto out;
1058
1059         __merge_refs(&prefs, 2);
1060
1061         while (!list_empty(&prefs)) {
1062                 ref = list_first_entry(&prefs, struct __prelim_ref, list);
1063                 WARN_ON(ref->count < 0);
1064                 if (roots && ref->count && ref->root_id && ref->parent == 0) {
1065                         if (root_objectid && ref->root_id != root_objectid) {
1066                                 ret = BACKREF_FOUND_SHARED;
1067                                 goto out;
1068                         }
1069
1070                         /* no parent == root of tree */
1071                         ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS);
1072                         if (ret < 0)
1073                                 goto out;
1074                 }
1075                 if (ref->count && ref->parent) {
1076                         if (extent_item_pos && !ref->inode_list &&
1077                             ref->level == 0) {
1078                                 struct extent_buffer *eb;
1079
1080                                 eb = read_tree_block(fs_info->extent_root,
1081                                                            ref->parent, 0);
1082                                 if (IS_ERR(eb)) {
1083                                         ret = PTR_ERR(eb);
1084                                         goto out;
1085                                 } else if (!extent_buffer_uptodate(eb)) {
1086                                         free_extent_buffer(eb);
1087                                         ret = -EIO;
1088                                         goto out;
1089                                 }
1090                                 btrfs_tree_read_lock(eb);
1091                                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1092                                 ret = find_extent_in_eb(eb, bytenr,
1093                                                         *extent_item_pos, &eie);
1094                                 btrfs_tree_read_unlock_blocking(eb);
1095                                 free_extent_buffer(eb);
1096                                 if (ret < 0)
1097                                         goto out;
1098                                 ref->inode_list = eie;
1099                         }
1100                         ret = ulist_add_merge_ptr(refs, ref->parent,
1101                                                   ref->inode_list,
1102                                                   (void **)&eie, GFP_NOFS);
1103                         if (ret < 0)
1104                                 goto out;
1105                         if (!ret && extent_item_pos) {
1106                                 /*
1107                                  * we've recorded that parent, so we must extend
1108                                  * its inode list here
1109                                  */
1110                                 BUG_ON(!eie);
1111                                 while (eie->next)
1112                                         eie = eie->next;
1113                                 eie->next = ref->inode_list;
1114                         }
1115                         eie = NULL;
1116                 }
1117                 list_del(&ref->list);
1118                 kmem_cache_free(btrfs_prelim_ref_cache, ref);
1119         }
1120
1121 out:
1122         btrfs_free_path(path);
1123         while (!list_empty(&prefs)) {
1124                 ref = list_first_entry(&prefs, struct __prelim_ref, list);
1125                 list_del(&ref->list);
1126                 kmem_cache_free(btrfs_prelim_ref_cache, ref);
1127         }
1128         while (!list_empty(&prefs_delayed)) {
1129                 ref = list_first_entry(&prefs_delayed, struct __prelim_ref,
1130                                        list);
1131                 list_del(&ref->list);
1132                 kmem_cache_free(btrfs_prelim_ref_cache, ref);
1133         }
1134         if (ret < 0)
1135                 free_inode_elem_list(eie);
1136         return ret;
1137 }
1138
1139 static void free_leaf_list(struct ulist *blocks)
1140 {
1141         struct ulist_node *node = NULL;
1142         struct extent_inode_elem *eie;
1143         struct ulist_iterator uiter;
1144
1145         ULIST_ITER_INIT(&uiter);
1146         while ((node = ulist_next(blocks, &uiter))) {
1147                 if (!node->aux)
1148                         continue;
1149                 eie = (struct extent_inode_elem *)(uintptr_t)node->aux;
1150                 free_inode_elem_list(eie);
1151                 node->aux = 0;
1152         }
1153
1154         ulist_free(blocks);
1155 }
1156
1157 /*
1158  * Finds all leafs with a reference to the specified combination of bytenr and
1159  * offset. key_list_head will point to a list of corresponding keys (caller must
1160  * free each list element). The leafs will be stored in the leafs ulist, which
1161  * must be freed with ulist_free.
1162  *
1163  * returns 0 on success, <0 on error
1164  */
1165 static int btrfs_find_all_leafs(struct btrfs_trans_handle *trans,
1166                                 struct btrfs_fs_info *fs_info, u64 bytenr,
1167                                 u64 time_seq, struct ulist **leafs,
1168                                 const u64 *extent_item_pos)
1169 {
1170         int ret;
1171
1172         *leafs = ulist_alloc(GFP_NOFS);
1173         if (!*leafs)
1174                 return -ENOMEM;
1175
1176         ret = find_parent_nodes(trans, fs_info, bytenr,
1177                                 time_seq, *leafs, NULL, extent_item_pos, 0, 0);
1178         if (ret < 0 && ret != -ENOENT) {
1179                 free_leaf_list(*leafs);
1180                 return ret;
1181         }
1182
1183         return 0;
1184 }
1185
1186 /*
1187  * walk all backrefs for a given extent to find all roots that reference this
1188  * extent. Walking a backref means finding all extents that reference this
1189  * extent and in turn walk the backrefs of those, too. Naturally this is a
1190  * recursive process, but here it is implemented in an iterative fashion: We
1191  * find all referencing extents for the extent in question and put them on a
1192  * list. In turn, we find all referencing extents for those, further appending
1193  * to the list. The way we iterate the list allows adding more elements after
1194  * the current while iterating. The process stops when we reach the end of the
1195  * list. Found roots are added to the roots list.
1196  *
1197  * returns 0 on success, < 0 on error.
1198  */
1199 static int __btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1200                                   struct btrfs_fs_info *fs_info, u64 bytenr,
1201                                   u64 time_seq, struct ulist **roots)
1202 {
1203         struct ulist *tmp;
1204         struct ulist_node *node = NULL;
1205         struct ulist_iterator uiter;
1206         int ret;
1207
1208         tmp = ulist_alloc(GFP_NOFS);
1209         if (!tmp)
1210                 return -ENOMEM;
1211         *roots = ulist_alloc(GFP_NOFS);
1212         if (!*roots) {
1213                 ulist_free(tmp);
1214                 return -ENOMEM;
1215         }
1216
1217         ULIST_ITER_INIT(&uiter);
1218         while (1) {
1219                 ret = find_parent_nodes(trans, fs_info, bytenr,
1220                                         time_seq, tmp, *roots, NULL, 0, 0);
1221                 if (ret < 0 && ret != -ENOENT) {
1222                         ulist_free(tmp);
1223                         ulist_free(*roots);
1224                         return ret;
1225                 }
1226                 node = ulist_next(tmp, &uiter);
1227                 if (!node)
1228                         break;
1229                 bytenr = node->val;
1230                 cond_resched();
1231         }
1232
1233         ulist_free(tmp);
1234         return 0;
1235 }
1236
1237 int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1238                          struct btrfs_fs_info *fs_info, u64 bytenr,
1239                          u64 time_seq, struct ulist **roots)
1240 {
1241         int ret;
1242
1243         if (!trans)
1244                 down_read(&fs_info->commit_root_sem);
1245         ret = __btrfs_find_all_roots(trans, fs_info, bytenr, time_seq, roots);
1246         if (!trans)
1247                 up_read(&fs_info->commit_root_sem);
1248         return ret;
1249 }
1250
1251 /**
1252  * btrfs_check_shared - tell us whether an extent is shared
1253  *
1254  * @trans: optional trans handle
1255  *
1256  * btrfs_check_shared uses the backref walking code but will short
1257  * circuit as soon as it finds a root or inode that doesn't match the
1258  * one passed in. This provides a significant performance benefit for
1259  * callers (such as fiemap) which want to know whether the extent is
1260  * shared but do not need a ref count.
1261  *
1262  * Return: 0 if extent is not shared, 1 if it is shared, < 0 on error.
1263  */
1264 int btrfs_check_shared(struct btrfs_trans_handle *trans,
1265                        struct btrfs_fs_info *fs_info, u64 root_objectid,
1266                        u64 inum, u64 bytenr)
1267 {
1268         struct ulist *tmp = NULL;
1269         struct ulist *roots = NULL;
1270         struct ulist_iterator uiter;
1271         struct ulist_node *node;
1272         struct seq_list elem = SEQ_LIST_INIT(elem);
1273         int ret = 0;
1274
1275         tmp = ulist_alloc(GFP_NOFS);
1276         roots = ulist_alloc(GFP_NOFS);
1277         if (!tmp || !roots) {
1278                 ulist_free(tmp);
1279                 ulist_free(roots);
1280                 return -ENOMEM;
1281         }
1282
1283         if (trans)
1284                 btrfs_get_tree_mod_seq(fs_info, &elem);
1285         else
1286                 down_read(&fs_info->commit_root_sem);
1287         ULIST_ITER_INIT(&uiter);
1288         while (1) {
1289                 ret = find_parent_nodes(trans, fs_info, bytenr, elem.seq, tmp,
1290                                         roots, NULL, root_objectid, inum);
1291                 if (ret == BACKREF_FOUND_SHARED) {
1292                         /* this is the only condition under which we return 1 */
1293                         ret = 1;
1294                         break;
1295                 }
1296                 if (ret < 0 && ret != -ENOENT)
1297                         break;
1298                 ret = 0;
1299                 node = ulist_next(tmp, &uiter);
1300                 if (!node)
1301                         break;
1302                 bytenr = node->val;
1303                 cond_resched();
1304         }
1305         if (trans)
1306                 btrfs_put_tree_mod_seq(fs_info, &elem);
1307         else
1308                 up_read(&fs_info->commit_root_sem);
1309         ulist_free(tmp);
1310         ulist_free(roots);
1311         return ret;
1312 }
1313
1314 int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,
1315                           u64 start_off, struct btrfs_path *path,
1316                           struct btrfs_inode_extref **ret_extref,
1317                           u64 *found_off)
1318 {
1319         int ret, slot;
1320         struct btrfs_key key;
1321         struct btrfs_key found_key;
1322         struct btrfs_inode_extref *extref;
1323         struct extent_buffer *leaf;
1324         unsigned long ptr;
1325
1326         key.objectid = inode_objectid;
1327         key.type = BTRFS_INODE_EXTREF_KEY;
1328         key.offset = start_off;
1329
1330         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1331         if (ret < 0)
1332                 return ret;
1333
1334         while (1) {
1335                 leaf = path->nodes[0];
1336                 slot = path->slots[0];
1337                 if (slot >= btrfs_header_nritems(leaf)) {
1338                         /*
1339                          * If the item at offset is not found,
1340                          * btrfs_search_slot will point us to the slot
1341                          * where it should be inserted. In our case
1342                          * that will be the slot directly before the
1343                          * next INODE_REF_KEY_V2 item. In the case
1344                          * that we're pointing to the last slot in a
1345                          * leaf, we must move one leaf over.
1346                          */
1347                         ret = btrfs_next_leaf(root, path);
1348                         if (ret) {
1349                                 if (ret >= 1)
1350                                         ret = -ENOENT;
1351                                 break;
1352                         }
1353                         continue;
1354                 }
1355
1356                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1357
1358                 /*
1359                  * Check that we're still looking at an extended ref key for
1360                  * this particular objectid. If we have different
1361                  * objectid or type then there are no more to be found
1362                  * in the tree and we can exit.
1363                  */
1364                 ret = -ENOENT;
1365                 if (found_key.objectid != inode_objectid)
1366                         break;
1367                 if (found_key.type != BTRFS_INODE_EXTREF_KEY)
1368                         break;
1369
1370                 ret = 0;
1371                 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1372                 extref = (struct btrfs_inode_extref *)ptr;
1373                 *ret_extref = extref;
1374                 if (found_off)
1375                         *found_off = found_key.offset;
1376                 break;
1377         }
1378
1379         return ret;
1380 }
1381
1382 /*
1383  * this iterates to turn a name (from iref/extref) into a full filesystem path.
1384  * Elements of the path are separated by '/' and the path is guaranteed to be
1385  * 0-terminated. the path is only given within the current file system.
1386  * Therefore, it never starts with a '/'. the caller is responsible to provide
1387  * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
1388  * the start point of the resulting string is returned. this pointer is within
1389  * dest, normally.
1390  * in case the path buffer would overflow, the pointer is decremented further
1391  * as if output was written to the buffer, though no more output is actually
1392  * generated. that way, the caller can determine how much space would be
1393  * required for the path to fit into the buffer. in that case, the returned
1394  * value will be smaller than dest. callers must check this!
1395  */
1396 char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
1397                         u32 name_len, unsigned long name_off,
1398                         struct extent_buffer *eb_in, u64 parent,
1399                         char *dest, u32 size)
1400 {
1401         int slot;
1402         u64 next_inum;
1403         int ret;
1404         s64 bytes_left = ((s64)size) - 1;
1405         struct extent_buffer *eb = eb_in;
1406         struct btrfs_key found_key;
1407         int leave_spinning = path->leave_spinning;
1408         struct btrfs_inode_ref *iref;
1409
1410         if (bytes_left >= 0)
1411                 dest[bytes_left] = '\0';
1412
1413         path->leave_spinning = 1;
1414         while (1) {
1415                 bytes_left -= name_len;
1416                 if (bytes_left >= 0)
1417                         read_extent_buffer(eb, dest + bytes_left,
1418                                            name_off, name_len);
1419                 if (eb != eb_in) {
1420                         btrfs_tree_read_unlock_blocking(eb);
1421                         free_extent_buffer(eb);
1422                 }
1423                 ret = btrfs_find_item(fs_root, path, parent, 0,
1424                                 BTRFS_INODE_REF_KEY, &found_key);
1425                 if (ret > 0)
1426                         ret = -ENOENT;
1427                 if (ret)
1428                         break;
1429
1430                 next_inum = found_key.offset;
1431
1432                 /* regular exit ahead */
1433                 if (parent == next_inum)
1434                         break;
1435
1436                 slot = path->slots[0];
1437                 eb = path->nodes[0];
1438                 /* make sure we can use eb after releasing the path */
1439                 if (eb != eb_in) {
1440                         atomic_inc(&eb->refs);
1441                         btrfs_tree_read_lock(eb);
1442                         btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1443                 }
1444                 btrfs_release_path(path);
1445                 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1446
1447                 name_len = btrfs_inode_ref_name_len(eb, iref);
1448                 name_off = (unsigned long)(iref + 1);
1449
1450                 parent = next_inum;
1451                 --bytes_left;
1452                 if (bytes_left >= 0)
1453                         dest[bytes_left] = '/';
1454         }
1455
1456         btrfs_release_path(path);
1457         path->leave_spinning = leave_spinning;
1458
1459         if (ret)
1460                 return ERR_PTR(ret);
1461
1462         return dest + bytes_left;
1463 }
1464
1465 /*
1466  * this makes the path point to (logical EXTENT_ITEM *)
1467  * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
1468  * tree blocks and <0 on error.
1469  */
1470 int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
1471                         struct btrfs_path *path, struct btrfs_key *found_key,
1472                         u64 *flags_ret)
1473 {
1474         int ret;
1475         u64 flags;
1476         u64 size = 0;
1477         u32 item_size;
1478         struct extent_buffer *eb;
1479         struct btrfs_extent_item *ei;
1480         struct btrfs_key key;
1481
1482         if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1483                 key.type = BTRFS_METADATA_ITEM_KEY;
1484         else
1485                 key.type = BTRFS_EXTENT_ITEM_KEY;
1486         key.objectid = logical;
1487         key.offset = (u64)-1;
1488
1489         ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
1490         if (ret < 0)
1491                 return ret;
1492
1493         ret = btrfs_previous_extent_item(fs_info->extent_root, path, 0);
1494         if (ret) {
1495                 if (ret > 0)
1496                         ret = -ENOENT;
1497                 return ret;
1498         }
1499         btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]);
1500         if (found_key->type == BTRFS_METADATA_ITEM_KEY)
1501                 size = fs_info->extent_root->nodesize;
1502         else if (found_key->type == BTRFS_EXTENT_ITEM_KEY)
1503                 size = found_key->offset;
1504
1505         if (found_key->objectid > logical ||
1506             found_key->objectid + size <= logical) {
1507                 pr_debug("logical %llu is not within any extent\n", logical);
1508                 return -ENOENT;
1509         }
1510
1511         eb = path->nodes[0];
1512         item_size = btrfs_item_size_nr(eb, path->slots[0]);
1513         BUG_ON(item_size < sizeof(*ei));
1514
1515         ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
1516         flags = btrfs_extent_flags(eb, ei);
1517
1518         pr_debug("logical %llu is at position %llu within the extent (%llu "
1519                  "EXTENT_ITEM %llu) flags %#llx size %u\n",
1520                  logical, logical - found_key->objectid, found_key->objectid,
1521                  found_key->offset, flags, item_size);
1522
1523         WARN_ON(!flags_ret);
1524         if (flags_ret) {
1525                 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1526                         *flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK;
1527                 else if (flags & BTRFS_EXTENT_FLAG_DATA)
1528                         *flags_ret = BTRFS_EXTENT_FLAG_DATA;
1529                 else
1530                         BUG_ON(1);
1531                 return 0;
1532         }
1533
1534         return -EIO;
1535 }
1536
1537 /*
1538  * helper function to iterate extent inline refs. ptr must point to a 0 value
1539  * for the first call and may be modified. it is used to track state.
1540  * if more refs exist, 0 is returned and the next call to
1541  * __get_extent_inline_ref must pass the modified ptr parameter to get the
1542  * next ref. after the last ref was processed, 1 is returned.
1543  * returns <0 on error
1544  */
1545 static int __get_extent_inline_ref(unsigned long *ptr, struct extent_buffer *eb,
1546                                    struct btrfs_key *key,
1547                                    struct btrfs_extent_item *ei, u32 item_size,
1548                                    struct btrfs_extent_inline_ref **out_eiref,
1549                                    int *out_type)
1550 {
1551         unsigned long end;
1552         u64 flags;
1553         struct btrfs_tree_block_info *info;
1554
1555         if (!*ptr) {
1556                 /* first call */
1557                 flags = btrfs_extent_flags(eb, ei);
1558                 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1559                         if (key->type == BTRFS_METADATA_ITEM_KEY) {
1560                                 /* a skinny metadata extent */
1561                                 *out_eiref =
1562                                      (struct btrfs_extent_inline_ref *)(ei + 1);
1563                         } else {
1564                                 WARN_ON(key->type != BTRFS_EXTENT_ITEM_KEY);
1565                                 info = (struct btrfs_tree_block_info *)(ei + 1);
1566                                 *out_eiref =
1567                                    (struct btrfs_extent_inline_ref *)(info + 1);
1568                         }
1569                 } else {
1570                         *out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1);
1571                 }
1572                 *ptr = (unsigned long)*out_eiref;
1573                 if ((unsigned long)(*ptr) >= (unsigned long)ei + item_size)
1574                         return -ENOENT;
1575         }
1576
1577         end = (unsigned long)ei + item_size;
1578         *out_eiref = (struct btrfs_extent_inline_ref *)(*ptr);
1579         *out_type = btrfs_extent_inline_ref_type(eb, *out_eiref);
1580
1581         *ptr += btrfs_extent_inline_ref_size(*out_type);
1582         WARN_ON(*ptr > end);
1583         if (*ptr == end)
1584                 return 1; /* last */
1585
1586         return 0;
1587 }
1588
1589 /*
1590  * reads the tree block backref for an extent. tree level and root are returned
1591  * through out_level and out_root. ptr must point to a 0 value for the first
1592  * call and may be modified (see __get_extent_inline_ref comment).
1593  * returns 0 if data was provided, 1 if there was no more data to provide or
1594  * <0 on error.
1595  */
1596 int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
1597                             struct btrfs_key *key, struct btrfs_extent_item *ei,
1598                             u32 item_size, u64 *out_root, u8 *out_level)
1599 {
1600         int ret;
1601         int type;
1602         struct btrfs_extent_inline_ref *eiref;
1603
1604         if (*ptr == (unsigned long)-1)
1605                 return 1;
1606
1607         while (1) {
1608                 ret = __get_extent_inline_ref(ptr, eb, key, ei, item_size,
1609                                               &eiref, &type);
1610                 if (ret < 0)
1611                         return ret;
1612
1613                 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1614                     type == BTRFS_SHARED_BLOCK_REF_KEY)
1615                         break;
1616
1617                 if (ret == 1)
1618                         return 1;
1619         }
1620
1621         /* we can treat both ref types equally here */
1622         *out_root = btrfs_extent_inline_ref_offset(eb, eiref);
1623
1624         if (key->type == BTRFS_EXTENT_ITEM_KEY) {
1625                 struct btrfs_tree_block_info *info;
1626
1627                 info = (struct btrfs_tree_block_info *)(ei + 1);
1628                 *out_level = btrfs_tree_block_level(eb, info);
1629         } else {
1630                 ASSERT(key->type == BTRFS_METADATA_ITEM_KEY);
1631                 *out_level = (u8)key->offset;
1632         }
1633
1634         if (ret == 1)
1635                 *ptr = (unsigned long)-1;
1636
1637         return 0;
1638 }
1639
1640 static int iterate_leaf_refs(struct extent_inode_elem *inode_list,
1641                                 u64 root, u64 extent_item_objectid,
1642                                 iterate_extent_inodes_t *iterate, void *ctx)
1643 {
1644         struct extent_inode_elem *eie;
1645         int ret = 0;
1646
1647         for (eie = inode_list; eie; eie = eie->next) {
1648                 pr_debug("ref for %llu resolved, key (%llu EXTEND_DATA %llu), "
1649                          "root %llu\n", extent_item_objectid,
1650                          eie->inum, eie->offset, root);
1651                 ret = iterate(eie->inum, eie->offset, root, ctx);
1652                 if (ret) {
1653                         pr_debug("stopping iteration for %llu due to ret=%d\n",
1654                                  extent_item_objectid, ret);
1655                         break;
1656                 }
1657         }
1658
1659         return ret;
1660 }
1661
1662 /*
1663  * calls iterate() for every inode that references the extent identified by
1664  * the given parameters.
1665  * when the iterator function returns a non-zero value, iteration stops.
1666  */
1667 int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
1668                                 u64 extent_item_objectid, u64 extent_item_pos,
1669                                 int search_commit_root,
1670                                 iterate_extent_inodes_t *iterate, void *ctx)
1671 {
1672         int ret;
1673         struct btrfs_trans_handle *trans = NULL;
1674         struct ulist *refs = NULL;
1675         struct ulist *roots = NULL;
1676         struct ulist_node *ref_node = NULL;
1677         struct ulist_node *root_node = NULL;
1678         struct seq_list tree_mod_seq_elem = SEQ_LIST_INIT(tree_mod_seq_elem);
1679         struct ulist_iterator ref_uiter;
1680         struct ulist_iterator root_uiter;
1681
1682         pr_debug("resolving all inodes for extent %llu\n",
1683                         extent_item_objectid);
1684
1685         if (!search_commit_root) {
1686                 trans = btrfs_join_transaction(fs_info->extent_root);
1687                 if (IS_ERR(trans))
1688                         return PTR_ERR(trans);
1689                 btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1690         } else {
1691                 down_read(&fs_info->commit_root_sem);
1692         }
1693
1694         ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid,
1695                                    tree_mod_seq_elem.seq, &refs,
1696                                    &extent_item_pos);
1697         if (ret)
1698                 goto out;
1699
1700         ULIST_ITER_INIT(&ref_uiter);
1701         while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) {
1702                 ret = __btrfs_find_all_roots(trans, fs_info, ref_node->val,
1703                                              tree_mod_seq_elem.seq, &roots);
1704                 if (ret)
1705                         break;
1706                 ULIST_ITER_INIT(&root_uiter);
1707                 while (!ret && (root_node = ulist_next(roots, &root_uiter))) {
1708                         pr_debug("root %llu references leaf %llu, data list "
1709                                  "%#llx\n", root_node->val, ref_node->val,
1710                                  ref_node->aux);
1711                         ret = iterate_leaf_refs((struct extent_inode_elem *)
1712                                                 (uintptr_t)ref_node->aux,
1713                                                 root_node->val,
1714                                                 extent_item_objectid,
1715                                                 iterate, ctx);
1716                 }
1717                 ulist_free(roots);
1718         }
1719
1720         free_leaf_list(refs);
1721 out:
1722         if (!search_commit_root) {
1723                 btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1724                 btrfs_end_transaction(trans, fs_info->extent_root);
1725         } else {
1726                 up_read(&fs_info->commit_root_sem);
1727         }
1728
1729         return ret;
1730 }
1731
1732 int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
1733                                 struct btrfs_path *path,
1734                                 iterate_extent_inodes_t *iterate, void *ctx)
1735 {
1736         int ret;
1737         u64 extent_item_pos;
1738         u64 flags = 0;
1739         struct btrfs_key found_key;
1740         int search_commit_root = path->search_commit_root;
1741
1742         ret = extent_from_logical(fs_info, logical, path, &found_key, &flags);
1743         btrfs_release_path(path);
1744         if (ret < 0)
1745                 return ret;
1746         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1747                 return -EINVAL;
1748
1749         extent_item_pos = logical - found_key.objectid;
1750         ret = iterate_extent_inodes(fs_info, found_key.objectid,
1751                                         extent_item_pos, search_commit_root,
1752                                         iterate, ctx);
1753
1754         return ret;
1755 }
1756
1757 typedef int (iterate_irefs_t)(u64 parent, u32 name_len, unsigned long name_off,
1758                               struct extent_buffer *eb, void *ctx);
1759
1760 static int iterate_inode_refs(u64 inum, struct btrfs_root *fs_root,
1761                               struct btrfs_path *path,
1762                               iterate_irefs_t *iterate, void *ctx)
1763 {
1764         int ret = 0;
1765         int slot;
1766         u32 cur;
1767         u32 len;
1768         u32 name_len;
1769         u64 parent = 0;
1770         int found = 0;
1771         struct extent_buffer *eb;
1772         struct btrfs_item *item;
1773         struct btrfs_inode_ref *iref;
1774         struct btrfs_key found_key;
1775
1776         while (!ret) {
1777                 ret = btrfs_find_item(fs_root, path, inum,
1778                                 parent ? parent + 1 : 0, BTRFS_INODE_REF_KEY,
1779                                 &found_key);
1780
1781                 if (ret < 0)
1782                         break;
1783                 if (ret) {
1784                         ret = found ? 0 : -ENOENT;
1785                         break;
1786                 }
1787                 ++found;
1788
1789                 parent = found_key.offset;
1790                 slot = path->slots[0];
1791                 eb = btrfs_clone_extent_buffer(path->nodes[0]);
1792                 if (!eb) {
1793                         ret = -ENOMEM;
1794                         break;
1795                 }
1796                 extent_buffer_get(eb);
1797                 btrfs_tree_read_lock(eb);
1798                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1799                 btrfs_release_path(path);
1800
1801                 item = btrfs_item_nr(slot);
1802                 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1803
1804                 for (cur = 0; cur < btrfs_item_size(eb, item); cur += len) {
1805                         name_len = btrfs_inode_ref_name_len(eb, iref);
1806                         /* path must be released before calling iterate()! */
1807                         pr_debug("following ref at offset %u for inode %llu in "
1808                                  "tree %llu\n", cur, found_key.objectid,
1809                                  fs_root->objectid);
1810                         ret = iterate(parent, name_len,
1811                                       (unsigned long)(iref + 1), eb, ctx);
1812                         if (ret)
1813                                 break;
1814                         len = sizeof(*iref) + name_len;
1815                         iref = (struct btrfs_inode_ref *)((char *)iref + len);
1816                 }
1817                 btrfs_tree_read_unlock_blocking(eb);
1818                 free_extent_buffer(eb);
1819         }
1820
1821         btrfs_release_path(path);
1822
1823         return ret;
1824 }
1825
1826 static int iterate_inode_extrefs(u64 inum, struct btrfs_root *fs_root,
1827                                  struct btrfs_path *path,
1828                                  iterate_irefs_t *iterate, void *ctx)
1829 {
1830         int ret;
1831         int slot;
1832         u64 offset = 0;
1833         u64 parent;
1834         int found = 0;
1835         struct extent_buffer *eb;
1836         struct btrfs_inode_extref *extref;
1837         u32 item_size;
1838         u32 cur_offset;
1839         unsigned long ptr;
1840
1841         while (1) {
1842                 ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref,
1843                                             &offset);
1844                 if (ret < 0)
1845                         break;
1846                 if (ret) {
1847                         ret = found ? 0 : -ENOENT;
1848                         break;
1849                 }
1850                 ++found;
1851
1852                 slot = path->slots[0];
1853                 eb = btrfs_clone_extent_buffer(path->nodes[0]);
1854                 if (!eb) {
1855                         ret = -ENOMEM;
1856                         break;
1857                 }
1858                 extent_buffer_get(eb);
1859
1860                 btrfs_tree_read_lock(eb);
1861                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1862                 btrfs_release_path(path);
1863
1864                 item_size = btrfs_item_size_nr(eb, slot);
1865                 ptr = btrfs_item_ptr_offset(eb, slot);
1866                 cur_offset = 0;
1867
1868                 while (cur_offset < item_size) {
1869                         u32 name_len;
1870
1871                         extref = (struct btrfs_inode_extref *)(ptr + cur_offset);
1872                         parent = btrfs_inode_extref_parent(eb, extref);
1873                         name_len = btrfs_inode_extref_name_len(eb, extref);
1874                         ret = iterate(parent, name_len,
1875                                       (unsigned long)&extref->name, eb, ctx);
1876                         if (ret)
1877                                 break;
1878
1879                         cur_offset += btrfs_inode_extref_name_len(eb, extref);
1880                         cur_offset += sizeof(*extref);
1881                 }
1882                 btrfs_tree_read_unlock_blocking(eb);
1883                 free_extent_buffer(eb);
1884
1885                 offset++;
1886         }
1887
1888         btrfs_release_path(path);
1889
1890         return ret;
1891 }
1892
1893 static int iterate_irefs(u64 inum, struct btrfs_root *fs_root,
1894                          struct btrfs_path *path, iterate_irefs_t *iterate,
1895                          void *ctx)
1896 {
1897         int ret;
1898         int found_refs = 0;
1899
1900         ret = iterate_inode_refs(inum, fs_root, path, iterate, ctx);
1901         if (!ret)
1902                 ++found_refs;
1903         else if (ret != -ENOENT)
1904                 return ret;
1905
1906         ret = iterate_inode_extrefs(inum, fs_root, path, iterate, ctx);
1907         if (ret == -ENOENT && found_refs)
1908                 return 0;
1909
1910         return ret;
1911 }
1912
1913 /*
1914  * returns 0 if the path could be dumped (probably truncated)
1915  * returns <0 in case of an error
1916  */
1917 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
1918                          struct extent_buffer *eb, void *ctx)
1919 {
1920         struct inode_fs_paths *ipath = ctx;
1921         char *fspath;
1922         char *fspath_min;
1923         int i = ipath->fspath->elem_cnt;
1924         const int s_ptr = sizeof(char *);
1925         u32 bytes_left;
1926
1927         bytes_left = ipath->fspath->bytes_left > s_ptr ?
1928                                         ipath->fspath->bytes_left - s_ptr : 0;
1929
1930         fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
1931         fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len,
1932                                    name_off, eb, inum, fspath_min, bytes_left);
1933         if (IS_ERR(fspath))
1934                 return PTR_ERR(fspath);
1935
1936         if (fspath > fspath_min) {
1937                 ipath->fspath->val[i] = (u64)(unsigned long)fspath;
1938                 ++ipath->fspath->elem_cnt;
1939                 ipath->fspath->bytes_left = fspath - fspath_min;
1940         } else {
1941                 ++ipath->fspath->elem_missed;
1942                 ipath->fspath->bytes_missing += fspath_min - fspath;
1943                 ipath->fspath->bytes_left = 0;
1944         }
1945
1946         return 0;
1947 }
1948
1949 /*
1950  * this dumps all file system paths to the inode into the ipath struct, provided
1951  * is has been created large enough. each path is zero-terminated and accessed
1952  * from ipath->fspath->val[i].
1953  * when it returns, there are ipath->fspath->elem_cnt number of paths available
1954  * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the
1955  * number of missed paths in recored in ipath->fspath->elem_missed, otherwise,
1956  * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
1957  * have been needed to return all paths.
1958  */
1959 int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
1960 {
1961         return iterate_irefs(inum, ipath->fs_root, ipath->btrfs_path,
1962                              inode_to_path, ipath);
1963 }
1964
1965 struct btrfs_data_container *init_data_container(u32 total_bytes)
1966 {
1967         struct btrfs_data_container *data;
1968         size_t alloc_bytes;
1969
1970         alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
1971         data = vmalloc(alloc_bytes);
1972         if (!data)
1973                 return ERR_PTR(-ENOMEM);
1974
1975         if (total_bytes >= sizeof(*data)) {
1976                 data->bytes_left = total_bytes - sizeof(*data);
1977                 data->bytes_missing = 0;
1978         } else {
1979                 data->bytes_missing = sizeof(*data) - total_bytes;
1980                 data->bytes_left = 0;
1981         }
1982
1983         data->elem_cnt = 0;
1984         data->elem_missed = 0;
1985
1986         return data;
1987 }
1988
1989 /*
1990  * allocates space to return multiple file system paths for an inode.
1991  * total_bytes to allocate are passed, note that space usable for actual path
1992  * information will be total_bytes - sizeof(struct inode_fs_paths).
1993  * the returned pointer must be freed with free_ipath() in the end.
1994  */
1995 struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
1996                                         struct btrfs_path *path)
1997 {
1998         struct inode_fs_paths *ifp;
1999         struct btrfs_data_container *fspath;
2000
2001         fspath = init_data_container(total_bytes);
2002         if (IS_ERR(fspath))
2003                 return (void *)fspath;
2004
2005         ifp = kmalloc(sizeof(*ifp), GFP_NOFS);
2006         if (!ifp) {
2007                 kfree(fspath);
2008                 return ERR_PTR(-ENOMEM);
2009         }
2010
2011         ifp->btrfs_path = path;
2012         ifp->fspath = fspath;
2013         ifp->fs_root = fs_root;
2014
2015         return ifp;
2016 }
2017
2018 void free_ipath(struct inode_fs_paths *ipath)
2019 {
2020         if (!ipath)
2021                 return;
2022         vfree(ipath->fspath);
2023         kfree(ipath);
2024 }