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