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_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         kmem_cache_destroy(btrfs_prelim_ref_cache);
 152 }
 153
 154 /*
 155  * the rules for all callers of this function are:
 156  * - obtaining the parent is the goal
 157  * - if you add a key, you must know that it is a correct key
 158  * - if you cannot add the parent or a correct key, then we will look into the
 159  *   block later to set a correct key
 160  *
 161  * delayed refs
 162  * ============
 163  *        backref type | shared | indirect | shared | indirect
 164  * information         |   tree |     tree |   data |     data
 165  * --------------------+--------+----------+--------+----------
 166  *      parent logical |    y   |     -    |    -   |     -
 167  *      key to resolve |    -   |     y    |    y   |     y
 168  *  tree block logical |    -   |     -    |    -   |     -
 169  *  root for resolving |    y   |     y    |    y   |     y
 170  *
 171  * - column 1:       we've the parent -> done
 172  * - column 2, 3, 4: we use the key to find the parent
 173  *
 174  * on disk refs (inline or keyed)
 175  * ==============================
 176  *        backref type | shared | indirect | shared | indirect
 177  * information         |   tree |     tree |   data |     data
 178  * --------------------+--------+----------+--------+----------
 179  *      parent logical |    y   |     -    |    y   |     -
 180  *      key to resolve |    -   |     -    |    -   |     y
 181  *  tree block logical |    y   |     y    |    y   |     y
 182  *  root for resolving |    -   |     y    |    y   |     y
 183  *
 184  * - column 1, 3: we've the parent -> done
 185  * - column 2:    we take the first key from the block to find the parent
 186  *                (see __add_missing_keys)
 187  * - column 4:    we use the key to find the parent
 188  *
 189  * additional information that's available but not required to find the parent
 190  * block might help in merging entries to gain some speed.
 191  */
 192
 193 static int __add_prelim_ref(struct list_head *head, u64 root_id,
 194                             struct btrfs_key *key, int level,
 195                             u64 parent, u64 wanted_disk_byte, int count,
 196                             gfp_t gfp_mask)
 197 {
 198         struct __prelim_ref *ref;
 199
 200         if (root_id == BTRFS_DATA_RELOC_TREE_OBJECTID)
 201                 return 0;
 202
 203         ref = kmem_cache_alloc(btrfs_prelim_ref_cache, gfp_mask);
 204         if (!ref)
 205                 return -ENOMEM;
 206
 207         ref->root_id = root_id;
 208         if (key) {
 209                 ref->key_for_search = *key;
 210                 /*
 211                  * We can often find data backrefs with an offset that is too
 212                  * large (>= LLONG_MAX, maximum allowed file offset) due to
 213                  * underflows when subtracting a file's offset with the data
 214                  * offset of its corresponding extent data item. This can
 215                  * happen for example in the clone ioctl.
 216                  * So if we detect such case we set the search key's offset to
 217                  * zero to make sure we will find the matching file extent item
 218                  * at add_all_parents(), otherwise we will miss it because the
 219                  * offset taken form the backref is much larger then the offset
 220                  * of the file extent item. This can make us scan a very large
 221                  * number of file extent items, but at least it will not make
 222                  * us miss any.
 223                  * This is an ugly workaround for a behaviour that should have
 224                  * never existed, but it does and a fix for the clone ioctl
 225                  * would touch a lot of places, cause backwards incompatibility
 226                  * and would not fix the problem for extents cloned with older
 227                  * kernels.
 228                  */
 229                 if (ref->key_for_search.type == BTRFS_EXTENT_DATA_KEY &&
 230                     ref->key_for_search.offset >= LLONG_MAX)
 231                         ref->key_for_search.offset = 0;
 232         } else {
 233                 memset(&ref->key_for_search, 0, sizeof(ref->key_for_search));
 234         }
 235
 236         ref->inode_list = NULL;
 237         ref->level = level;
 238         ref->count = count;
 239         ref->parent = parent;
 240         ref->wanted_disk_byte = wanted_disk_byte;
 241         list_add_tail(&ref->list, head);
 242
 243         return 0;
 244 }
 245
 246 static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path,
 247                            struct ulist *parents, struct __prelim_ref *ref,
 248                            int level, u64 time_seq, const u64 *extent_item_pos,
 249                            u64 total_refs)
 250 {
 251         int ret = 0;
 252         int slot;
 253         struct extent_buffer *eb;
 254         struct btrfs_key key;
 255         struct btrfs_key *key_for_search = &ref->key_for_search;
 256         struct btrfs_file_extent_item *fi;
 257         struct extent_inode_elem *eie = NULL, *old = NULL;
 258         u64 disk_byte;
 259         u64 wanted_disk_byte = ref->wanted_disk_byte;
 260         u64 count = 0;
 261
 262         if (level != 0) {
 263                 eb = path->nodes[level];
 264                 ret = ulist_add(parents, eb->start, 0, GFP_NOFS);
 265                 if (ret < 0)
 266                         return ret;
 267                 return 0;
 268         }
 269
 270         /*
 271          * We normally enter this function with the path already pointing to
 272          * the first item to check. But sometimes, we may enter it with
 273          * slot==nritems. In that case, go to the next leaf before we continue.
 274          */
 275         if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
 276                 if (time_seq == (u64)-1)
 277                         ret = btrfs_next_leaf(root, path);
 278                 else
 279                         ret = btrfs_next_old_leaf(root, path, time_seq);
 280         }
 281
 282         while (!ret && count < total_refs) {
 283                 eb = path->nodes[0];
 284                 slot = path->slots[0];
 285
 286                 btrfs_item_key_to_cpu(eb, &key, slot);
 287
 288                 if (key.objectid != key_for_search->objectid ||
 289                     key.type != BTRFS_EXTENT_DATA_KEY)
 290                         break;
 291
 292                 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
 293                 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
 294
 295                 if (disk_byte == wanted_disk_byte) {
 296                         eie = NULL;
 297                         old = NULL;
 298                         count++;
 299                         if (extent_item_pos) {
 300                                 ret = check_extent_in_eb(&key, eb, fi,
 301                                                 *extent_item_pos,
 302                                                 &eie);
 303                                 if (ret < 0)
 304                                         break;
 305                         }
 306                         if (ret > 0)
 307                                 goto next;
 308                         ret = ulist_add_merge_ptr(parents, eb->start,
 309                                                   eie, (void **)&old, GFP_NOFS);
 310                         if (ret < 0)
 311                                 break;
 312                         if (!ret && extent_item_pos) {
 313                                 while (old->next)
 314                                         old = old->next;
 315                                 old->next = eie;
 316                         }
 317                         eie = NULL;
 318                 }
 319 next:
 320                 if (time_seq == (u64)-1)
 321                         ret = btrfs_next_item(root, path);
 322                 else
 323                         ret = btrfs_next_old_item(root, path, time_seq);
 324         }
 325
 326         if (ret > 0)
 327                 ret = 0;
 328         else if (ret < 0)
 329                 free_inode_elem_list(eie);
 330         return ret;
 331 }
 332
 333 /*
 334  * resolve an indirect backref in the form (root_id, key, level)
 335  * to a logical address
 336  */
 337 static int __resolve_indirect_ref(struct btrfs_fs_info *fs_info,
 338                                   struct btrfs_path *path, u64 time_seq,
 339                                   struct __prelim_ref *ref,
 340                                   struct ulist *parents,
 341                                   const u64 *extent_item_pos, u64 total_refs)
 342 {
 343         struct btrfs_root *root;
 344         struct btrfs_key root_key;
 345         struct extent_buffer *eb;
 346         int ret = 0;
 347         int root_level;
 348         int level = ref->level;
 349         int index;
 350
 351         root_key.objectid = ref->root_id;
 352         root_key.type = BTRFS_ROOT_ITEM_KEY;
 353         root_key.offset = (u64)-1;
 354
 355         index = srcu_read_lock(&fs_info->subvol_srcu);
 356
 357         root = btrfs_get_fs_root(fs_info, &root_key, false);
 358         if (IS_ERR(root)) {
 359                 srcu_read_unlock(&fs_info->subvol_srcu, index);
 360                 ret = PTR_ERR(root);
 361                 goto out;
 362         }
 363
 364         if (btrfs_is_testing(fs_info)) {
 365                 srcu_read_unlock(&fs_info->subvol_srcu, index);
 366                 ret = -ENOENT;
 367                 goto out;
 368         }
 369
 370         if (path->search_commit_root)
 371                 root_level = btrfs_header_level(root->commit_root);
 372         else if (time_seq == (u64)-1)
 373                 root_level = btrfs_header_level(root->node);
 374         else
 375                 root_level = btrfs_old_root_level(root, time_seq);
 376
 377         if (root_level + 1 == level) {
 378                 srcu_read_unlock(&fs_info->subvol_srcu, index);
 379                 goto out;
 380         }
 381
 382         path->lowest_level = level;
 383         if (time_seq == (u64)-1)
 384                 ret = btrfs_search_slot(NULL, root, &ref->key_for_search, path,
 385                                         0, 0);
 386         else
 387                 ret = btrfs_search_old_slot(root, &ref->key_for_search, path,
 388                                             time_seq);
 389
 390         /* root node has been locked, we can release @subvol_srcu safely here */
 391         srcu_read_unlock(&fs_info->subvol_srcu, index);
 392
 393         pr_debug("search slot in root %llu (level %d, ref count %d) returned "
 394                  "%d for key (%llu %u %llu)\n",
 395                  ref->root_id, level, ref->count, ret,
 396                  ref->key_for_search.objectid, ref->key_for_search.type,
 397                  ref->key_for_search.offset);
 398         if (ret < 0)
 399                 goto out;
 400
 401         eb = path->nodes[level];
 402         while (!eb) {
 403                 if (WARN_ON(!level)) {
 404                         ret = 1;
 405                         goto out;
 406                 }
 407                 level--;
 408                 eb = path->nodes[level];
 409         }
 410
 411         ret = add_all_parents(root, path, parents, ref, level, time_seq,
 412                               extent_item_pos, total_refs);
 413 out:
 414         path->lowest_level = 0;
 415         btrfs_release_path(path);
 416         return ret;
 417 }
 418
 419 /*
 420  * resolve all indirect backrefs from the list
 421  */
 422 static int __resolve_indirect_refs(struct btrfs_fs_info *fs_info,
 423                                    struct btrfs_path *path, u64 time_seq,
 424                                    struct list_head *head,
 425                                    const u64 *extent_item_pos, u64 total_refs,
 426                                    u64 root_objectid)
 427 {
 428         int err;
 429         int ret = 0;
 430         struct __prelim_ref *ref;
 431         struct __prelim_ref *ref_safe;
 432         struct __prelim_ref *new_ref;
 433         struct ulist *parents;
 434         struct ulist_node *node;
 435         struct ulist_iterator uiter;
 436
 437         parents = ulist_alloc(GFP_NOFS);
 438         if (!parents)
 439                 return -ENOMEM;
 440
 441         /*
 442          * _safe allows us to insert directly after the current item without
 443          * iterating over the newly inserted items.
 444          * we're also allowed to re-assign ref during iteration.
 445          */
 446         list_for_each_entry_safe(ref, ref_safe, head, list) {
 447                 if (ref->parent)        /* already direct */
 448                         continue;
 449                 if (ref->count == 0)
 450                         continue;
 451                 if (root_objectid && ref->root_id != root_objectid) {
 452                         ret = BACKREF_FOUND_SHARED;
 453                         goto out;
 454                 }
 455                 err = __resolve_indirect_ref(fs_info, path, time_seq, ref,
 456                                              parents, extent_item_pos,
 457                                              total_refs);
 458                 /*
 459                  * we can only tolerate ENOENT,otherwise,we should catch error
 460                  * and return directly.
 461                  */
 462                 if (err == -ENOENT) {
 463                         continue;
 464                 } else if (err) {
 465                         ret = err;
 466                         goto out;
 467                 }
 468
 469                 /* we put the first parent into the ref at hand */
 470                 ULIST_ITER_INIT(&uiter);
 471                 node = ulist_next(parents, &uiter);
 472                 ref->parent = node ? node->val : 0;
 473                 ref->inode_list = node ?
 474                         (struct extent_inode_elem *)(uintptr_t)node->aux : NULL;
 475
 476                 /* additional parents require new refs being added here */
 477                 while ((node = ulist_next(parents, &uiter))) {
 478                         new_ref = kmem_cache_alloc(btrfs_prelim_ref_cache,
 479                                                    GFP_NOFS);
 480                         if (!new_ref) {
 481                                 ret = -ENOMEM;
 482                                 goto out;
 483                         }
 484                         memcpy(new_ref, ref, sizeof(*ref));
 485                         new_ref->parent = node->val;
 486                         new_ref->inode_list = (struct extent_inode_elem *)
 487                                                         (uintptr_t)node->aux;
 488                         list_add(&new_ref->list, &ref->list);
 489                 }
 490                 ulist_reinit(parents);
 491         }
 492 out:
 493         ulist_free(parents);
 494         return ret;
 495 }
 496
 497 static inline int ref_for_same_block(struct __prelim_ref *ref1,
 498                                      struct __prelim_ref *ref2)
 499 {
 500         if (ref1->level != ref2->level)
 501                 return 0;
 502         if (ref1->root_id != ref2->root_id)
 503                 return 0;
 504         if (ref1->key_for_search.type != ref2->key_for_search.type)
 505                 return 0;
 506         if (ref1->key_for_search.objectid != ref2->key_for_search.objectid)
 507                 return 0;
 508         if (ref1->key_for_search.offset != ref2->key_for_search.offset)
 509                 return 0;
 510         if (ref1->parent != ref2->parent)
 511                 return 0;
 512
 513         return 1;
 514 }
 515
 516 /*
 517  * read tree blocks and add keys where required.
 518  */
 519 static int __add_missing_keys(struct btrfs_fs_info *fs_info,
 520                               struct list_head *head)
 521 {
 522         struct __prelim_ref *ref;
 523         struct extent_buffer *eb;
 524
 525         list_for_each_entry(ref, head, list) {
 526                 if (ref->parent)
 527                         continue;
 528                 if (ref->key_for_search.type)
 529                         continue;
 530                 BUG_ON(!ref->wanted_disk_byte);
 531                 eb = read_tree_block(fs_info->tree_root, ref->wanted_disk_byte,
 532                                      0);
 533                 if (IS_ERR(eb)) {
 534                         return PTR_ERR(eb);
 535                 } else if (!extent_buffer_uptodate(eb)) {
 536                         free_extent_buffer(eb);
 537                         return -EIO;
 538                 }
 539                 btrfs_tree_read_lock(eb);
 540                 if (btrfs_header_level(eb) == 0)
 541                         btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
 542                 else
 543                         btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
 544                 btrfs_tree_read_unlock(eb);
 545                 free_extent_buffer(eb);
 546         }
 547         return 0;
 548 }
 549
 550 /*
 551  * merge backrefs and adjust counts accordingly
 552  *
 553  * mode = 1: merge identical keys, if key is set
 554  *    FIXME: if we add more keys in __add_prelim_ref, we can merge more here.
 555  *           additionally, we could even add a key range for the blocks we
 556  *           looked into to merge even more (-> replace unresolved refs by those
 557  *           having a parent).
 558  * mode = 2: merge identical parents
 559  */
 560 static void __merge_refs(struct list_head *head, int mode)
 561 {
 562         struct __prelim_ref *pos1;
 563
 564         list_for_each_entry(pos1, head, list) {
 565                 struct __prelim_ref *pos2 = pos1, *tmp;
 566
 567                 list_for_each_entry_safe_continue(pos2, tmp, head, list) {
 568                         struct __prelim_ref *ref1 = pos1, *ref2 = pos2;
 569                         struct extent_inode_elem *eie;
 570
 571                         if (!ref_for_same_block(ref1, ref2))
 572                                 continue;
 573                         if (mode == 1) {
 574                                 if (!ref1->parent && ref2->parent)
 575                                         swap(ref1, ref2);
 576                         } else {
 577                                 if (ref1->parent != ref2->parent)
 578                                         continue;
 579                         }
 580
 581                         eie = ref1->inode_list;
 582                         while (eie && eie->next)
 583                                 eie = eie->next;
 584                         if (eie)
 585                                 eie->next = ref2->inode_list;
 586                         else
 587                                 ref1->inode_list = ref2->inode_list;
 588                         ref1->count += ref2->count;
 589
 590                         list_del(&ref2->list);
 591                         kmem_cache_free(btrfs_prelim_ref_cache, ref2);
 592                 }
 593
 594         }
 595 }
 596
 597 /*
 598  * add all currently queued delayed refs from this head whose seq nr is
 599  * smaller or equal that seq to the list
 600  */
 601 static int __add_delayed_refs(struct btrfs_delayed_ref_head *head, u64 seq,
 602                               struct list_head *prefs, u64 *total_refs,
 603                               u64 inum)
 604 {
 605         struct btrfs_delayed_ref_node *node;
 606         struct btrfs_delayed_extent_op *extent_op = head->extent_op;
 607         struct btrfs_key key;
 608         struct btrfs_key op_key = {0};
 609         int sgn;
 610         int ret = 0;
 611
 612         if (extent_op && extent_op->update_key)
 613                 btrfs_disk_key_to_cpu(&op_key, &extent_op->key);
 614
 615         spin_lock(&head->lock);
 616         list_for_each_entry(node, &head->ref_list, list) {
 617                 if (node->seq > seq)
 618                         continue;
 619
 620                 switch (node->action) {
 621                 case BTRFS_ADD_DELAYED_EXTENT:
 622                 case BTRFS_UPDATE_DELAYED_HEAD:
 623                         WARN_ON(1);
 624                         continue;
 625                 case BTRFS_ADD_DELAYED_REF:
 626                         sgn = 1;
 627                         break;
 628                 case BTRFS_DROP_DELAYED_REF:
 629                         sgn = -1;
 630                         break;
 631                 default:
 632                         BUG_ON(1);
 633                 }
 634                 *total_refs += (node->ref_mod * sgn);
 635                 switch (node->type) {
 636                 case BTRFS_TREE_BLOCK_REF_KEY: {
 637                         struct btrfs_delayed_tree_ref *ref;
 638
 639                         ref = btrfs_delayed_node_to_tree_ref(node);
 640                         ret = __add_prelim_ref(prefs, ref->root, &op_key,
 641                                                ref->level + 1, 0, node->bytenr,
 642                                                node->ref_mod * sgn, GFP_ATOMIC);
 643                         break;
 644                 }
 645                 case BTRFS_SHARED_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, 0, NULL,
 650                                                ref->level + 1, ref->parent,
 651                                                node->bytenr,
 652                                                node->ref_mod * sgn, GFP_ATOMIC);
 653                         break;
 654                 }
 655                 case BTRFS_EXTENT_DATA_REF_KEY: {
 656                         struct btrfs_delayed_data_ref *ref;
 657                         ref = btrfs_delayed_node_to_data_ref(node);
 658
 659                         key.objectid = ref->objectid;
 660                         key.type = BTRFS_EXTENT_DATA_KEY;
 661                         key.offset = ref->offset;
 662
 663                         /*
 664                          * Found a inum that doesn't match our known inum, we
 665                          * know it's shared.
 666                          */
 667                         if (inum && ref->objectid != inum) {
 668                                 ret = BACKREF_FOUND_SHARED;
 669                                 break;
 670                         }
 671
 672                         ret = __add_prelim_ref(prefs, ref->root, &key, 0, 0,
 673                                                node->bytenr,
 674                                                node->ref_mod * sgn, GFP_ATOMIC);
 675                         break;
 676                 }
 677                 case BTRFS_SHARED_DATA_REF_KEY: {
 678                         struct btrfs_delayed_data_ref *ref;
 679
 680                         ref = btrfs_delayed_node_to_data_ref(node);
 681                         ret = __add_prelim_ref(prefs, 0, NULL, 0,
 682                                                ref->parent, node->bytenr,
 683                                                node->ref_mod * sgn, GFP_ATOMIC);
 684                         break;
 685                 }
 686                 default:
 687                         WARN_ON(1);
 688                 }
 689                 if (ret)
 690                         break;
 691         }
 692         spin_unlock(&head->lock);
 693         return ret;
 694 }
 695
 696 /*
 697  * add all inline backrefs for bytenr to the list
 698  */
 699 static int __add_inline_refs(struct btrfs_fs_info *fs_info,
 700                              struct btrfs_path *path, u64 bytenr,
 701                              int *info_level, struct list_head *prefs,
 702                              u64 *total_refs, u64 inum)
 703 {
 704         int ret = 0;
 705         int slot;
 706         struct extent_buffer *leaf;
 707         struct btrfs_key key;
 708         struct btrfs_key found_key;
 709         unsigned long ptr;
 710         unsigned long end;
 711         struct btrfs_extent_item *ei;
 712         u64 flags;
 713         u64 item_size;
 714
 715         /*
 716          * enumerate all inline refs
 717          */
 718         leaf = path->nodes[0];
 719         slot = path->slots[0];
 720
 721         item_size = btrfs_item_size_nr(leaf, slot);
 722         BUG_ON(item_size < sizeof(*ei));
 723
 724         ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
 725         flags = btrfs_extent_flags(leaf, ei);
 726         *total_refs += btrfs_extent_refs(leaf, ei);
 727         btrfs_item_key_to_cpu(leaf, &found_key, slot);
 728
 729         ptr = (unsigned long)(ei + 1);
 730         end = (unsigned long)ei + item_size;
 731
 732         if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
 733             flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
 734                 struct btrfs_tree_block_info *info;
 735
 736                 info = (struct btrfs_tree_block_info *)ptr;
 737                 *info_level = btrfs_tree_block_level(leaf, info);
 738                 ptr += sizeof(struct btrfs_tree_block_info);
 739                 BUG_ON(ptr > end);
 740         } else if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
 741                 *info_level = found_key.offset;
 742         } else {
 743                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
 744         }
 745
 746         while (ptr < end) {
 747                 struct btrfs_extent_inline_ref *iref;
 748                 u64 offset;
 749                 int type;
 750
 751                 iref = (struct btrfs_extent_inline_ref *)ptr;
 752                 type = btrfs_extent_inline_ref_type(leaf, iref);
 753                 offset = btrfs_extent_inline_ref_offset(leaf, iref);
 754
 755                 switch (type) {
 756                 case BTRFS_SHARED_BLOCK_REF_KEY:
 757                         ret = __add_prelim_ref(prefs, 0, NULL,
 758                                                 *info_level + 1, offset,
 759                                                 bytenr, 1, GFP_NOFS);
 760                         break;
 761                 case BTRFS_SHARED_DATA_REF_KEY: {
 762                         struct btrfs_shared_data_ref *sdref;
 763                         int count;
 764
 765                         sdref = (struct btrfs_shared_data_ref *)(iref + 1);
 766                         count = btrfs_shared_data_ref_count(leaf, sdref);
 767                         ret = __add_prelim_ref(prefs, 0, NULL, 0, offset,
 768                                                bytenr, count, GFP_NOFS);
 769                         break;
 770                 }
 771                 case BTRFS_TREE_BLOCK_REF_KEY:
 772                         ret = __add_prelim_ref(prefs, offset, NULL,
 773                                                *info_level + 1, 0,
 774                                                bytenr, 1, GFP_NOFS);
 775                         break;
 776                 case BTRFS_EXTENT_DATA_REF_KEY: {
 777                         struct btrfs_extent_data_ref *dref;
 778                         int count;
 779                         u64 root;
 780
 781                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
 782                         count = btrfs_extent_data_ref_count(leaf, dref);
 783                         key.objectid = btrfs_extent_data_ref_objectid(leaf,
 784                                                                       dref);
 785                         key.type = BTRFS_EXTENT_DATA_KEY;
 786                         key.offset = btrfs_extent_data_ref_offset(leaf, dref);
 787
 788                         if (inum && key.objectid != inum) {
 789                                 ret = BACKREF_FOUND_SHARED;
 790                                 break;
 791                         }
 792
 793                         root = btrfs_extent_data_ref_root(leaf, dref);
 794                         ret = __add_prelim_ref(prefs, root, &key, 0, 0,
 795                                                bytenr, count, GFP_NOFS);
 796                         break;
 797                 }
 798                 default:
 799                         WARN_ON(1);
 800                 }
 801                 if (ret)
 802                         return ret;
 803                 ptr += btrfs_extent_inline_ref_size(type);
 804         }
 805
 806         return 0;
 807 }
 808
 809 /*
 810  * add all non-inline backrefs for bytenr to the list
 811  */
 812 static int __add_keyed_refs(struct btrfs_fs_info *fs_info,
 813                             struct btrfs_path *path, u64 bytenr,
 814                             int info_level, struct list_head *prefs, u64 inum)
 815 {
 816         struct btrfs_root *extent_root = fs_info->extent_root;
 817         int ret;
 818         int slot;
 819         struct extent_buffer *leaf;
 820         struct btrfs_key key;
 821
 822         while (1) {
 823                 ret = btrfs_next_item(extent_root, path);
 824                 if (ret < 0)
 825                         break;
 826                 if (ret) {
 827                         ret = 0;
 828                         break;
 829                 }
 830
 831                 slot = path->slots[0];
 832                 leaf = path->nodes[0];
 833                 btrfs_item_key_to_cpu(leaf, &key, slot);
 834
 835                 if (key.objectid != bytenr)
 836                         break;
 837                 if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
 838                         continue;
 839                 if (key.type > BTRFS_SHARED_DATA_REF_KEY)
 840                         break;
 841
 842                 switch (key.type) {
 843                 case BTRFS_SHARED_BLOCK_REF_KEY:
 844                         ret = __add_prelim_ref(prefs, 0, NULL,
 845                                                 info_level + 1, key.offset,
 846                                                 bytenr, 1, GFP_NOFS);
 847                         break;
 848                 case BTRFS_SHARED_DATA_REF_KEY: {
 849                         struct btrfs_shared_data_ref *sdref;
 850                         int count;
 851
 852                         sdref = btrfs_item_ptr(leaf, slot,
 853                                               struct btrfs_shared_data_ref);
 854                         count = btrfs_shared_data_ref_count(leaf, sdref);
 855                         ret = __add_prelim_ref(prefs, 0, NULL, 0, key.offset,
 856                                                 bytenr, count, GFP_NOFS);
 857                         break;
 858                 }
 859                 case BTRFS_TREE_BLOCK_REF_KEY:
 860                         ret = __add_prelim_ref(prefs, key.offset, NULL,
 861                                                info_level + 1, 0,
 862                                                bytenr, 1, GFP_NOFS);
 863                         break;
 864                 case BTRFS_EXTENT_DATA_REF_KEY: {
 865                         struct btrfs_extent_data_ref *dref;
 866                         int count;
 867                         u64 root;
 868
 869                         dref = btrfs_item_ptr(leaf, slot,
 870                                               struct btrfs_extent_data_ref);
 871                         count = btrfs_extent_data_ref_count(leaf, dref);
 872                         key.objectid = btrfs_extent_data_ref_objectid(leaf,
 873                                                                       dref);
 874                         key.type = BTRFS_EXTENT_DATA_KEY;
 875                         key.offset = btrfs_extent_data_ref_offset(leaf, dref);
 876
 877                         if (inum && key.objectid != inum) {
 878                                 ret = BACKREF_FOUND_SHARED;
 879                                 break;
 880                         }
 881
 882                         root = btrfs_extent_data_ref_root(leaf, dref);
 883                         ret = __add_prelim_ref(prefs, root, &key, 0, 0,
 884                                                bytenr, count, GFP_NOFS);
 885                         break;
 886                 }
 887                 default:
 888                         WARN_ON(1);
 889                 }
 890                 if (ret)
 891                         return ret;
 892
 893         }
 894
 895         return ret;
 896 }
 897
 898 /*
 899  * this adds all existing backrefs (inline backrefs, backrefs and delayed
 900  * refs) for the given bytenr to the refs list, merges duplicates and resolves
 901  * indirect refs to their parent bytenr.
 902  * When roots are found, they're added to the roots list
 903  *
 904  * NOTE: This can return values > 0
 905  *
 906  * If time_seq is set to (u64)-1, it will not search delayed_refs, and behave
 907  * much like trans == NULL case, the difference only lies in it will not
 908  * commit root.
 909  * The special case is for qgroup to search roots in commit_transaction().
 910  *
 911  * FIXME some caching might speed things up
 912  */
 913 static int find_parent_nodes(struct btrfs_trans_handle *trans,
 914                              struct btrfs_fs_info *fs_info, u64 bytenr,
 915                              u64 time_seq, struct ulist *refs,
 916                              struct ulist *roots, const u64 *extent_item_pos,
 917                              u64 root_objectid, u64 inum)
 918 {
 919         struct btrfs_key key;
 920         struct btrfs_path *path;
 921         struct btrfs_delayed_ref_root *delayed_refs = NULL;
 922         struct btrfs_delayed_ref_head *head;
 923         int info_level = 0;
 924         int ret;
 925         struct list_head prefs_delayed;
 926         struct list_head prefs;
 927         struct __prelim_ref *ref;
 928         struct extent_inode_elem *eie = NULL;
 929         u64 total_refs = 0;
 930
 931         INIT_LIST_HEAD(&prefs);
 932         INIT_LIST_HEAD(&prefs_delayed);
 933
 934         key.objectid = bytenr;
 935         key.offset = (u64)-1;
 936         if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
 937                 key.type = BTRFS_METADATA_ITEM_KEY;
 938         else
 939                 key.type = BTRFS_EXTENT_ITEM_KEY;
 940
 941         path = btrfs_alloc_path();
 942         if (!path)
 943                 return -ENOMEM;
 944         if (!trans) {
 945                 path->search_commit_root = 1;
 946                 path->skip_locking = 1;
 947         }
 948
 949         if (time_seq == (u64)-1)
 950                 path->skip_locking = 1;
 951
 952         /*
 953          * grab both a lock on the path and a lock on the delayed ref head.
 954          * We need both to get a consistent picture of how the refs look
 955          * at a specified point in time
 956          */
 957 again:
 958         head = NULL;
 959
 960         ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
 961         if (ret < 0)
 962                 goto out;
 963         BUG_ON(ret == 0);
 964
 965 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
 966         if (trans && likely(trans->type != __TRANS_DUMMY) &&
 967             time_seq != (u64)-1) {
 968 #else
 969         if (trans && time_seq != (u64)-1) {
 970 #endif
 971                 /*
 972                  * look if there are updates for this ref queued and lock the
 973                  * head
 974                  */
 975                 delayed_refs = &trans->transaction->delayed_refs;
 976                 spin_lock(&delayed_refs->lock);
 977                 head = btrfs_find_delayed_ref_head(trans, bytenr);
 978                 if (head) {
 979                         if (!mutex_trylock(&head->mutex)) {
 980                                 atomic_inc(&head->node.refs);
 981                                 spin_unlock(&delayed_refs->lock);
 982
 983                                 btrfs_release_path(path);
 984
 985                                 /*
 986                                  * Mutex was contended, block until it's
 987                                  * released and try again
 988                                  */
 989                                 mutex_lock(&head->mutex);
 990                                 mutex_unlock(&head->mutex);
 991                                 btrfs_put_delayed_ref(&head->node);
 992                                 goto again;
 993                         }
 994                         spin_unlock(&delayed_refs->lock);
 995                         ret = __add_delayed_refs(head, time_seq,
 996                                                  &prefs_delayed, &total_refs,
 997                                                  inum);
 998                         mutex_unlock(&head->mutex);
 999                         if (ret)
1000                                 goto out;
1001                 } else {
1002                         spin_unlock(&delayed_refs->lock);
1003                 }
1004         }
1005
1006         if (path->slots[0]) {
1007                 struct extent_buffer *leaf;
1008                 int slot;
1009
1010                 path->slots[0]--;
1011                 leaf = path->nodes[0];
1012                 slot = path->slots[0];
1013                 btrfs_item_key_to_cpu(leaf, &key, slot);
1014                 if (key.objectid == bytenr &&
1015                     (key.type == BTRFS_EXTENT_ITEM_KEY ||
1016                      key.type == BTRFS_METADATA_ITEM_KEY)) {
1017                         ret = __add_inline_refs(fs_info, path, bytenr,
1018                                                 &info_level, &prefs,
1019                                                 &total_refs, inum);
1020                         if (ret)
1021                                 goto out;
1022                         ret = __add_keyed_refs(fs_info, path, bytenr,
1023                                                info_level, &prefs, inum);
1024                         if (ret)
1025                                 goto out;
1026                 }
1027         }
1028         btrfs_release_path(path);
1029
1030         list_splice_init(&prefs_delayed, &prefs);
1031
1032         ret = __add_missing_keys(fs_info, &prefs);
1033         if (ret)
1034                 goto out;
1035
1036         __merge_refs(&prefs, 1);
1037
1038         ret = __resolve_indirect_refs(fs_info, path, time_seq, &prefs,
1039                                       extent_item_pos, total_refs,
1040                                       root_objectid);
1041         if (ret)
1042                 goto out;
1043
1044         __merge_refs(&prefs, 2);
1045
1046         while (!list_empty(&prefs)) {
1047                 ref = list_first_entry(&prefs, struct __prelim_ref, list);
1048                 WARN_ON(ref->count < 0);
1049                 if (roots && ref->count && ref->root_id && ref->parent == 0) {
1050                         if (root_objectid && ref->root_id != root_objectid) {
1051                                 ret = BACKREF_FOUND_SHARED;
1052                                 goto out;
1053                         }
1054
1055                         /* no parent == root of tree */
1056                         ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS);
1057                         if (ret < 0)
1058                                 goto out;
1059                 }
1060                 if (ref->count && ref->parent) {
1061                         if (extent_item_pos && !ref->inode_list &&
1062                             ref->level == 0) {
1063                                 struct extent_buffer *eb;
1064
1065                                 eb = read_tree_block(fs_info->extent_root,
1066                                                            ref->parent, 0);
1067                                 if (IS_ERR(eb)) {
1068                                         ret = PTR_ERR(eb);
1069                                         goto out;
1070                                 } else if (!extent_buffer_uptodate(eb)) {
1071                                         free_extent_buffer(eb);
1072                                         ret = -EIO;
1073                                         goto out;
1074                                 }
1075                                 btrfs_tree_read_lock(eb);
1076                                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1077                                 ret = find_extent_in_eb(eb, bytenr,
1078                                                         *extent_item_pos, &eie);
1079                                 btrfs_tree_read_unlock_blocking(eb);
1080                                 free_extent_buffer(eb);
1081                                 if (ret < 0)
1082                                         goto out;
1083                                 ref->inode_list = eie;
1084                         }
1085                         ret = ulist_add_merge_ptr(refs, ref->parent,
1086                                                   ref->inode_list,
1087                                                   (void **)&eie, GFP_NOFS);
1088                         if (ret < 0)
1089                                 goto out;
1090                         if (!ret && extent_item_pos) {
1091                                 /*
1092                                  * we've recorded that parent, so we must extend
1093                                  * its inode list here
1094                                  */
1095                                 BUG_ON(!eie);
1096                                 while (eie->next)
1097                                         eie = eie->next;
1098                                 eie->next = ref->inode_list;
1099                         }
1100                         eie = NULL;
1101                 }
1102                 list_del(&ref->list);
1103                 kmem_cache_free(btrfs_prelim_ref_cache, ref);
1104         }
1105
1106 out:
1107         btrfs_free_path(path);
1108         while (!list_empty(&prefs)) {
1109                 ref = list_first_entry(&prefs, struct __prelim_ref, list);
1110                 list_del(&ref->list);
1111                 kmem_cache_free(btrfs_prelim_ref_cache, ref);
1112         }
1113         while (!list_empty(&prefs_delayed)) {
1114                 ref = list_first_entry(&prefs_delayed, struct __prelim_ref,
1115                                        list);
1116                 list_del(&ref->list);
1117                 kmem_cache_free(btrfs_prelim_ref_cache, ref);
1118         }
1119         if (ret < 0)
1120                 free_inode_elem_list(eie);
1121         return ret;
1122 }
1123
1124 static void free_leaf_list(struct ulist *blocks)
1125 {
1126         struct ulist_node *node = NULL;
1127         struct extent_inode_elem *eie;
1128         struct ulist_iterator uiter;
1129
1130         ULIST_ITER_INIT(&uiter);
1131         while ((node = ulist_next(blocks, &uiter))) {
1132                 if (!node->aux)
1133                         continue;
1134                 eie = (struct extent_inode_elem *)(uintptr_t)node->aux;
1135                 free_inode_elem_list(eie);
1136                 node->aux = 0;
1137         }
1138
1139         ulist_free(blocks);
1140 }
1141
1142 /*
1143  * Finds all leafs with a reference to the specified combination of bytenr and
1144  * offset. key_list_head will point to a list of corresponding keys (caller must
1145  * free each list element). The leafs will be stored in the leafs ulist, which
1146  * must be freed with ulist_free.
1147  *
1148  * returns 0 on success, <0 on error
1149  */
1150 static int btrfs_find_all_leafs(struct btrfs_trans_handle *trans,
1151                                 struct btrfs_fs_info *fs_info, u64 bytenr,
1152                                 u64 time_seq, struct ulist **leafs,
1153                                 const u64 *extent_item_pos)
1154 {
1155         int ret;
1156
1157         *leafs = ulist_alloc(GFP_NOFS);
1158         if (!*leafs)
1159                 return -ENOMEM;
1160
1161         ret = find_parent_nodes(trans, fs_info, bytenr,
1162                                 time_seq, *leafs, NULL, extent_item_pos, 0, 0);
1163         if (ret < 0 && ret != -ENOENT) {
1164                 free_leaf_list(*leafs);
1165                 return ret;
1166         }
1167
1168         return 0;
1169 }
1170
1171 /*
1172  * walk all backrefs for a given extent to find all roots that reference this
1173  * extent. Walking a backref means finding all extents that reference this
1174  * extent and in turn walk the backrefs of those, too. Naturally this is a
1175  * recursive process, but here it is implemented in an iterative fashion: We
1176  * find all referencing extents for the extent in question and put them on a
1177  * list. In turn, we find all referencing extents for those, further appending
1178  * to the list. The way we iterate the list allows adding more elements after
1179  * the current while iterating. The process stops when we reach the end of the
1180  * list. Found roots are added to the roots list.
1181  *
1182  * returns 0 on success, < 0 on error.
1183  */
1184 static int __btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1185                                   struct btrfs_fs_info *fs_info, u64 bytenr,
1186                                   u64 time_seq, struct ulist **roots)
1187 {
1188         struct ulist *tmp;
1189         struct ulist_node *node = NULL;
1190         struct ulist_iterator uiter;
1191         int ret;
1192
1193         tmp = ulist_alloc(GFP_NOFS);
1194         if (!tmp)
1195                 return -ENOMEM;
1196         *roots = ulist_alloc(GFP_NOFS);
1197         if (!*roots) {
1198                 ulist_free(tmp);
1199                 return -ENOMEM;
1200         }
1201
1202         ULIST_ITER_INIT(&uiter);
1203         while (1) {
1204                 ret = find_parent_nodes(trans, fs_info, bytenr,
1205                                         time_seq, tmp, *roots, NULL, 0, 0);
1206                 if (ret < 0 && ret != -ENOENT) {
1207                         ulist_free(tmp);
1208                         ulist_free(*roots);
1209                         return ret;
1210                 }
1211                 node = ulist_next(tmp, &uiter);
1212                 if (!node)
1213                         break;
1214                 bytenr = node->val;
1215                 cond_resched();
1216         }
1217
1218         ulist_free(tmp);
1219         return 0;
1220 }
1221
1222 int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1223                          struct btrfs_fs_info *fs_info, u64 bytenr,
1224                          u64 time_seq, struct ulist **roots)
1225 {
1226         int ret;
1227
1228         if (!trans)
1229                 down_read(&fs_info->commit_root_sem);
1230         ret = __btrfs_find_all_roots(trans, fs_info, bytenr, time_seq, roots);
1231         if (!trans)
1232                 up_read(&fs_info->commit_root_sem);
1233         return ret;
1234 }
1235
1236 /**
1237  * btrfs_check_shared - tell us whether an extent is shared
1238  *
1239  * @trans: optional trans handle
1240  *
1241  * btrfs_check_shared uses the backref walking code but will short
1242  * circuit as soon as it finds a root or inode that doesn't match the
1243  * one passed in. This provides a significant performance benefit for
1244  * callers (such as fiemap) which want to know whether the extent is
1245  * shared but do not need a ref count.
1246  *
1247  * Return: 0 if extent is not shared, 1 if it is shared, < 0 on error.
1248  */
1249 int btrfs_check_shared(struct btrfs_trans_handle *trans,
1250                        struct btrfs_fs_info *fs_info, u64 root_objectid,
1251                        u64 inum, u64 bytenr)
1252 {
1253         struct ulist *tmp = NULL;
1254         struct ulist *roots = NULL;
1255         struct ulist_iterator uiter;
1256         struct ulist_node *node;
1257         struct seq_list elem = SEQ_LIST_INIT(elem);
1258         int ret = 0;
1259
1260         tmp = ulist_alloc(GFP_NOFS);
1261         roots = ulist_alloc(GFP_NOFS);
1262         if (!tmp || !roots) {
1263                 ulist_free(tmp);
1264                 ulist_free(roots);
1265                 return -ENOMEM;
1266         }
1267
1268         if (trans)
1269                 btrfs_get_tree_mod_seq(fs_info, &elem);
1270         else
1271                 down_read(&fs_info->commit_root_sem);
1272         ULIST_ITER_INIT(&uiter);
1273         while (1) {
1274                 ret = find_parent_nodes(trans, fs_info, bytenr, elem.seq, tmp,
1275                                         roots, NULL, root_objectid, inum);
1276                 if (ret == BACKREF_FOUND_SHARED) {
1277                         /* this is the only condition under which we return 1 */
1278                         ret = 1;
1279                         break;
1280                 }
1281                 if (ret < 0 && ret != -ENOENT)
1282                         break;
1283                 ret = 0;
1284                 node = ulist_next(tmp, &uiter);
1285                 if (!node)
1286                         break;
1287                 bytenr = node->val;
1288                 cond_resched();
1289         }
1290         if (trans)
1291                 btrfs_put_tree_mod_seq(fs_info, &elem);
1292         else
1293                 up_read(&fs_info->commit_root_sem);
1294         ulist_free(tmp);
1295         ulist_free(roots);
1296         return ret;
1297 }
1298
1299 int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,
1300                           u64 start_off, struct btrfs_path *path,
1301                           struct btrfs_inode_extref **ret_extref,
1302                           u64 *found_off)
1303 {
1304         int ret, slot;
1305         struct btrfs_key key;
1306         struct btrfs_key found_key;
1307         struct btrfs_inode_extref *extref;
1308         struct extent_buffer *leaf;
1309         unsigned long ptr;
1310
1311         key.objectid = inode_objectid;
1312         key.type = BTRFS_INODE_EXTREF_KEY;
1313         key.offset = start_off;
1314
1315         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1316         if (ret < 0)
1317                 return ret;
1318
1319         while (1) {
1320                 leaf = path->nodes[0];
1321                 slot = path->slots[0];
1322                 if (slot >= btrfs_header_nritems(leaf)) {
1323                         /*
1324                          * If the item at offset is not found,
1325                          * btrfs_search_slot will point us to the slot
1326                          * where it should be inserted. In our case
1327                          * that will be the slot directly before the
1328                          * next INODE_REF_KEY_V2 item. In the case
1329                          * that we're pointing to the last slot in a
1330                          * leaf, we must move one leaf over.
1331                          */
1332                         ret = btrfs_next_leaf(root, path);
1333                         if (ret) {
1334                                 if (ret >= 1)
1335                                         ret = -ENOENT;
1336                                 break;
1337                         }
1338                         continue;
1339                 }
1340
1341                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1342
1343                 /*
1344                  * Check that we're still looking at an extended ref key for
1345                  * this particular objectid. If we have different
1346                  * objectid or type then there are no more to be found
1347                  * in the tree and we can exit.
1348                  */
1349                 ret = -ENOENT;
1350                 if (found_key.objectid != inode_objectid)
1351                         break;
1352                 if (found_key.type != BTRFS_INODE_EXTREF_KEY)
1353                         break;
1354
1355                 ret = 0;
1356                 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1357                 extref = (struct btrfs_inode_extref *)ptr;
1358                 *ret_extref = extref;
1359                 if (found_off)
1360                         *found_off = found_key.offset;
1361                 break;
1362         }
1363
1364         return ret;
1365 }
1366
1367 /*
1368  * this iterates to turn a name (from iref/extref) into a full filesystem path.
1369  * Elements of the path are separated by '/' and the path is guaranteed to be
1370  * 0-terminated. the path is only given within the current file system.
1371  * Therefore, it never starts with a '/'. the caller is responsible to provide
1372  * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
1373  * the start point of the resulting string is returned. this pointer is within
1374  * dest, normally.
1375  * in case the path buffer would overflow, the pointer is decremented further
1376  * as if output was written to the buffer, though no more output is actually
1377  * generated. that way, the caller can determine how much space would be
1378  * required for the path to fit into the buffer. in that case, the returned
1379  * value will be smaller than dest. callers must check this!
1380  */
1381 char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
1382                         u32 name_len, unsigned long name_off,
1383                         struct extent_buffer *eb_in, u64 parent,
1384                         char *dest, u32 size)
1385 {
1386         int slot;
1387         u64 next_inum;
1388         int ret;
1389         s64 bytes_left = ((s64)size) - 1;
1390         struct extent_buffer *eb = eb_in;
1391         struct btrfs_key found_key;
1392         int leave_spinning = path->leave_spinning;
1393         struct btrfs_inode_ref *iref;
1394
1395         if (bytes_left >= 0)
1396                 dest[bytes_left] = '\0';
1397
1398         path->leave_spinning = 1;
1399         while (1) {
1400                 bytes_left -= name_len;
1401                 if (bytes_left >= 0)
1402                         read_extent_buffer(eb, dest + bytes_left,
1403                                            name_off, name_len);
1404                 if (eb != eb_in) {
1405                         if (!path->skip_locking)
1406                                 btrfs_tree_read_unlock_blocking(eb);
1407                         free_extent_buffer(eb);
1408                 }
1409                 ret = btrfs_find_item(fs_root, path, parent, 0,
1410                                 BTRFS_INODE_REF_KEY, &found_key);
1411                 if (ret > 0)
1412                         ret = -ENOENT;
1413                 if (ret)
1414                         break;
1415
1416                 next_inum = found_key.offset;
1417
1418                 /* regular exit ahead */
1419                 if (parent == next_inum)
1420                         break;
1421
1422                 slot = path->slots[0];
1423                 eb = path->nodes[0];
1424                 /* make sure we can use eb after releasing the path */
1425                 if (eb != eb_in) {
1426                         if (!path->skip_locking)
1427                                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1428                         path->nodes[0] = NULL;
1429                         path->locks[0] = 0;
1430                 }
1431                 btrfs_release_path(path);
1432                 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1433
1434                 name_len = btrfs_inode_ref_name_len(eb, iref);
1435                 name_off = (unsigned long)(iref + 1);
1436
1437                 parent = next_inum;
1438                 --bytes_left;
1439                 if (bytes_left >= 0)
1440                         dest[bytes_left] = '/';
1441         }
1442
1443         btrfs_release_path(path);
1444         path->leave_spinning = leave_spinning;
1445
1446         if (ret)
1447                 return ERR_PTR(ret);
1448
1449         return dest + bytes_left;
1450 }
1451
1452 /*
1453  * this makes the path point to (logical EXTENT_ITEM *)
1454  * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
1455  * tree blocks and <0 on error.
1456  */
1457 int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
1458                         struct btrfs_path *path, struct btrfs_key *found_key,
1459                         u64 *flags_ret)
1460 {
1461         int ret;
1462         u64 flags;
1463         u64 size = 0;
1464         u32 item_size;
1465         struct extent_buffer *eb;
1466         struct btrfs_extent_item *ei;
1467         struct btrfs_key key;
1468
1469         if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1470                 key.type = BTRFS_METADATA_ITEM_KEY;
1471         else
1472                 key.type = BTRFS_EXTENT_ITEM_KEY;
1473         key.objectid = logical;
1474         key.offset = (u64)-1;
1475
1476         ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
1477         if (ret < 0)
1478                 return ret;
1479
1480         ret = btrfs_previous_extent_item(fs_info->extent_root, path, 0);
1481         if (ret) {
1482                 if (ret > 0)
1483                         ret = -ENOENT;
1484                 return ret;
1485         }
1486         btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]);
1487         if (found_key->type == BTRFS_METADATA_ITEM_KEY)
1488                 size = fs_info->extent_root->nodesize;
1489         else if (found_key->type == BTRFS_EXTENT_ITEM_KEY)
1490                 size = found_key->offset;
1491
1492         if (found_key->objectid > logical ||
1493             found_key->objectid + size <= logical) {
1494                 pr_debug("logical %llu is not within any extent\n", logical);
1495                 return -ENOENT;
1496         }
1497
1498         eb = path->nodes[0];
1499         item_size = btrfs_item_size_nr(eb, path->slots[0]);
1500         BUG_ON(item_size < sizeof(*ei));
1501
1502         ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
1503         flags = btrfs_extent_flags(eb, ei);
1504
1505         pr_debug("logical %llu is at position %llu within the extent (%llu "
1506                  "EXTENT_ITEM %llu) flags %#llx size %u\n",
1507                  logical, logical - found_key->objectid, found_key->objectid,
1508                  found_key->offset, flags, item_size);
1509
1510         WARN_ON(!flags_ret);
1511         if (flags_ret) {
1512                 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1513                         *flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK;
1514                 else if (flags & BTRFS_EXTENT_FLAG_DATA)
1515                         *flags_ret = BTRFS_EXTENT_FLAG_DATA;
1516                 else
1517                         BUG_ON(1);
1518                 return 0;
1519         }
1520
1521         return -EIO;
1522 }
1523
1524 /*
1525  * helper function to iterate extent inline refs. ptr must point to a 0 value
1526  * for the first call and may be modified. it is used to track state.
1527  * if more refs exist, 0 is returned and the next call to
1528  * __get_extent_inline_ref must pass the modified ptr parameter to get the
1529  * next ref. after the last ref was processed, 1 is returned.
1530  * returns <0 on error
1531  */
1532 static int __get_extent_inline_ref(unsigned long *ptr, struct extent_buffer *eb,
1533                                    struct btrfs_key *key,
1534                                    struct btrfs_extent_item *ei, u32 item_size,
1535                                    struct btrfs_extent_inline_ref **out_eiref,
1536                                    int *out_type)
1537 {
1538         unsigned long end;
1539         u64 flags;
1540         struct btrfs_tree_block_info *info;
1541
1542         if (!*ptr) {
1543                 /* first call */
1544                 flags = btrfs_extent_flags(eb, ei);
1545                 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1546                         if (key->type == BTRFS_METADATA_ITEM_KEY) {
1547                                 /* a skinny metadata extent */
1548                                 *out_eiref =
1549                                      (struct btrfs_extent_inline_ref *)(ei + 1);
1550                         } else {
1551                                 WARN_ON(key->type != BTRFS_EXTENT_ITEM_KEY);
1552                                 info = (struct btrfs_tree_block_info *)(ei + 1);
1553                                 *out_eiref =
1554                                    (struct btrfs_extent_inline_ref *)(info + 1);
1555                         }
1556                 } else {
1557                         *out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1);
1558                 }
1559                 *ptr = (unsigned long)*out_eiref;
1560                 if ((unsigned long)(*ptr) >= (unsigned long)ei + item_size)
1561                         return -ENOENT;
1562         }
1563
1564         end = (unsigned long)ei + item_size;
1565         *out_eiref = (struct btrfs_extent_inline_ref *)(*ptr);
1566         *out_type = btrfs_extent_inline_ref_type(eb, *out_eiref);
1567
1568         *ptr += btrfs_extent_inline_ref_size(*out_type);
1569         WARN_ON(*ptr > end);
1570         if (*ptr == end)
1571                 return 1; /* last */
1572
1573         return 0;
1574 }
1575
1576 /*
1577  * reads the tree block backref for an extent. tree level and root are returned
1578  * through out_level and out_root. ptr must point to a 0 value for the first
1579  * call and may be modified (see __get_extent_inline_ref comment).
1580  * returns 0 if data was provided, 1 if there was no more data to provide or
1581  * <0 on error.
1582  */
1583 int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
1584                             struct btrfs_key *key, struct btrfs_extent_item *ei,
1585                             u32 item_size, u64 *out_root, u8 *out_level)
1586 {
1587         int ret;
1588         int type;
1589         struct btrfs_extent_inline_ref *eiref;
1590
1591         if (*ptr == (unsigned long)-1)
1592                 return 1;
1593
1594         while (1) {
1595                 ret = __get_extent_inline_ref(ptr, eb, key, ei, item_size,
1596                                               &eiref, &type);
1597                 if (ret < 0)
1598                         return ret;
1599
1600                 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1601                     type == BTRFS_SHARED_BLOCK_REF_KEY)
1602                         break;
1603
1604                 if (ret == 1)
1605                         return 1;
1606         }
1607
1608         /* we can treat both ref types equally here */
1609         *out_root = btrfs_extent_inline_ref_offset(eb, eiref);
1610
1611         if (key->type == BTRFS_EXTENT_ITEM_KEY) {
1612                 struct btrfs_tree_block_info *info;
1613
1614                 info = (struct btrfs_tree_block_info *)(ei + 1);
1615                 *out_level = btrfs_tree_block_level(eb, info);
1616         } else {
1617                 ASSERT(key->type == BTRFS_METADATA_ITEM_KEY);
1618                 *out_level = (u8)key->offset;
1619         }
1620
1621         if (ret == 1)
1622                 *ptr = (unsigned long)-1;
1623
1624         return 0;
1625 }
1626
1627 static int iterate_leaf_refs(struct extent_inode_elem *inode_list,
1628                                 u64 root, u64 extent_item_objectid,
1629                                 iterate_extent_inodes_t *iterate, void *ctx)
1630 {
1631         struct extent_inode_elem *eie;
1632         int ret = 0;
1633
1634         for (eie = inode_list; eie; eie = eie->next) {
1635                 pr_debug("ref for %llu resolved, key (%llu EXTEND_DATA %llu), "
1636                          "root %llu\n", extent_item_objectid,
1637                          eie->inum, eie->offset, root);
1638                 ret = iterate(eie->inum, eie->offset, root, ctx);
1639                 if (ret) {
1640                         pr_debug("stopping iteration for %llu due to ret=%d\n",
1641                                  extent_item_objectid, ret);
1642                         break;
1643                 }
1644         }
1645
1646         return ret;
1647 }
1648
1649 /*
1650  * calls iterate() for every inode that references the extent identified by
1651  * the given parameters.
1652  * when the iterator function returns a non-zero value, iteration stops.
1653  */
1654 int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
1655                                 u64 extent_item_objectid, u64 extent_item_pos,
1656                                 int search_commit_root,
1657                                 iterate_extent_inodes_t *iterate, void *ctx)
1658 {
1659         int ret;
1660         struct btrfs_trans_handle *trans = NULL;
1661         struct ulist *refs = NULL;
1662         struct ulist *roots = NULL;
1663         struct ulist_node *ref_node = NULL;
1664         struct ulist_node *root_node = NULL;
1665         struct seq_list tree_mod_seq_elem = SEQ_LIST_INIT(tree_mod_seq_elem);
1666         struct ulist_iterator ref_uiter;
1667         struct ulist_iterator root_uiter;
1668
1669         pr_debug("resolving all inodes for extent %llu\n",
1670                         extent_item_objectid);
1671
1672         if (!search_commit_root) {
1673                 trans = btrfs_join_transaction(fs_info->extent_root);
1674                 if (IS_ERR(trans))
1675                         return PTR_ERR(trans);
1676                 btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1677         } else {
1678                 down_read(&fs_info->commit_root_sem);
1679         }
1680
1681         ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid,
1682                                    tree_mod_seq_elem.seq, &refs,
1683                                    &extent_item_pos);
1684         if (ret)
1685                 goto out;
1686
1687         ULIST_ITER_INIT(&ref_uiter);
1688         while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) {
1689                 ret = __btrfs_find_all_roots(trans, fs_info, ref_node->val,
1690                                              tree_mod_seq_elem.seq, &roots);
1691                 if (ret)
1692                         break;
1693                 ULIST_ITER_INIT(&root_uiter);
1694                 while (!ret && (root_node = ulist_next(roots, &root_uiter))) {
1695                         pr_debug("root %llu references leaf %llu, data list "
1696                                  "%#llx\n", root_node->val, ref_node->val,
1697                                  ref_node->aux);
1698                         ret = iterate_leaf_refs((struct extent_inode_elem *)
1699                                                 (uintptr_t)ref_node->aux,
1700                                                 root_node->val,
1701                                                 extent_item_objectid,
1702                                                 iterate, ctx);
1703                 }
1704                 ulist_free(roots);
1705         }
1706
1707         free_leaf_list(refs);
1708 out:
1709         if (!search_commit_root) {
1710                 btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1711                 btrfs_end_transaction(trans, fs_info->extent_root);
1712         } else {
1713                 up_read(&fs_info->commit_root_sem);
1714         }
1715
1716         return ret;
1717 }
1718
1719 int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
1720                                 struct btrfs_path *path,
1721                                 iterate_extent_inodes_t *iterate, void *ctx)
1722 {
1723         int ret;
1724         u64 extent_item_pos;
1725         u64 flags = 0;
1726         struct btrfs_key found_key;
1727         int search_commit_root = path->search_commit_root;
1728
1729         ret = extent_from_logical(fs_info, logical, path, &found_key, &flags);
1730         btrfs_release_path(path);
1731         if (ret < 0)
1732                 return ret;
1733         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1734                 return -EINVAL;
1735
1736         extent_item_pos = logical - found_key.objectid;
1737         ret = iterate_extent_inodes(fs_info, found_key.objectid,
1738                                         extent_item_pos, search_commit_root,
1739                                         iterate, ctx);
1740
1741         return ret;
1742 }
1743
1744 typedef int (iterate_irefs_t)(u64 parent, u32 name_len, unsigned long name_off,
1745                               struct extent_buffer *eb, void *ctx);
1746
1747 static int iterate_inode_refs(u64 inum, struct btrfs_root *fs_root,
1748                               struct btrfs_path *path,
1749                               iterate_irefs_t *iterate, void *ctx)
1750 {
1751         int ret = 0;
1752         int slot;
1753         u32 cur;
1754         u32 len;
1755         u32 name_len;
1756         u64 parent = 0;
1757         int found = 0;
1758         struct extent_buffer *eb;
1759         struct btrfs_item *item;
1760         struct btrfs_inode_ref *iref;
1761         struct btrfs_key found_key;
1762
1763         while (!ret) {
1764                 ret = btrfs_find_item(fs_root, path, inum,
1765                                 parent ? parent + 1 : 0, BTRFS_INODE_REF_KEY,
1766                                 &found_key);
1767
1768                 if (ret < 0)
1769                         break;
1770                 if (ret) {
1771                         ret = found ? 0 : -ENOENT;
1772                         break;
1773                 }
1774                 ++found;
1775
1776                 parent = found_key.offset;
1777                 slot = path->slots[0];
1778                 eb = btrfs_clone_extent_buffer(path->nodes[0]);
1779                 if (!eb) {
1780                         ret = -ENOMEM;
1781                         break;
1782                 }
1783                 extent_buffer_get(eb);
1784                 btrfs_tree_read_lock(eb);
1785                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1786                 btrfs_release_path(path);
1787
1788                 item = btrfs_item_nr(slot);
1789                 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1790
1791                 for (cur = 0; cur < btrfs_item_size(eb, item); cur += len) {
1792                         name_len = btrfs_inode_ref_name_len(eb, iref);
1793                         /* path must be released before calling iterate()! */
1794                         pr_debug("following ref at offset %u for inode %llu in "
1795                                  "tree %llu\n", cur, found_key.objectid,
1796                                  fs_root->objectid);
1797                         ret = iterate(parent, name_len,
1798                                       (unsigned long)(iref + 1), eb, ctx);
1799                         if (ret)
1800                                 break;
1801                         len = sizeof(*iref) + name_len;
1802                         iref = (struct btrfs_inode_ref *)((char *)iref + len);
1803                 }
1804                 btrfs_tree_read_unlock_blocking(eb);
1805                 free_extent_buffer(eb);
1806         }
1807
1808         btrfs_release_path(path);
1809
1810         return ret;
1811 }
1812
1813 static int iterate_inode_extrefs(u64 inum, struct btrfs_root *fs_root,
1814                                  struct btrfs_path *path,
1815                                  iterate_irefs_t *iterate, void *ctx)
1816 {
1817         int ret;
1818         int slot;
1819         u64 offset = 0;
1820         u64 parent;
1821         int found = 0;
1822         struct extent_buffer *eb;
1823         struct btrfs_inode_extref *extref;
1824         u32 item_size;
1825         u32 cur_offset;
1826         unsigned long ptr;
1827
1828         while (1) {
1829                 ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref,
1830                                             &offset);
1831                 if (ret < 0)
1832                         break;
1833                 if (ret) {
1834                         ret = found ? 0 : -ENOENT;
1835                         break;
1836                 }
1837                 ++found;
1838
1839                 slot = path->slots[0];
1840                 eb = btrfs_clone_extent_buffer(path->nodes[0]);
1841                 if (!eb) {
1842                         ret = -ENOMEM;
1843                         break;
1844                 }
1845                 extent_buffer_get(eb);
1846
1847                 btrfs_tree_read_lock(eb);
1848                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1849                 btrfs_release_path(path);
1850
1851                 item_size = btrfs_item_size_nr(eb, slot);
1852                 ptr = btrfs_item_ptr_offset(eb, slot);
1853                 cur_offset = 0;
1854
1855                 while (cur_offset < item_size) {
1856                         u32 name_len;
1857
1858                         extref = (struct btrfs_inode_extref *)(ptr + cur_offset);
1859                         parent = btrfs_inode_extref_parent(eb, extref);
1860                         name_len = btrfs_inode_extref_name_len(eb, extref);
1861                         ret = iterate(parent, name_len,
1862                                       (unsigned long)&extref->name, eb, ctx);
1863                         if (ret)
1864                                 break;
1865
1866                         cur_offset += btrfs_inode_extref_name_len(eb, extref);
1867                         cur_offset += sizeof(*extref);
1868                 }
1869                 btrfs_tree_read_unlock_blocking(eb);
1870                 free_extent_buffer(eb);
1871
1872                 offset++;
1873         }
1874
1875         btrfs_release_path(path);
1876
1877         return ret;
1878 }
1879
1880 static int iterate_irefs(u64 inum, struct btrfs_root *fs_root,
1881                          struct btrfs_path *path, iterate_irefs_t *iterate,
1882                          void *ctx)
1883 {
1884         int ret;
1885         int found_refs = 0;
1886
1887         ret = iterate_inode_refs(inum, fs_root, path, iterate, ctx);
1888         if (!ret)
1889                 ++found_refs;
1890         else if (ret != -ENOENT)
1891                 return ret;
1892
1893         ret = iterate_inode_extrefs(inum, fs_root, path, iterate, ctx);
1894         if (ret == -ENOENT && found_refs)
1895                 return 0;
1896
1897         return ret;
1898 }
1899
1900 /*
1901  * returns 0 if the path could be dumped (probably truncated)
1902  * returns <0 in case of an error
1903  */
1904 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
1905                          struct extent_buffer *eb, void *ctx)
1906 {
1907         struct inode_fs_paths *ipath = ctx;
1908         char *fspath;
1909         char *fspath_min;
1910         int i = ipath->fspath->elem_cnt;
1911         const int s_ptr = sizeof(char *);
1912         u32 bytes_left;
1913
1914         bytes_left = ipath->fspath->bytes_left > s_ptr ?
1915                                         ipath->fspath->bytes_left - s_ptr : 0;
1916
1917         fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
1918         fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len,
1919                                    name_off, eb, inum, fspath_min, bytes_left);
1920         if (IS_ERR(fspath))
1921                 return PTR_ERR(fspath);
1922
1923         if (fspath > fspath_min) {
1924                 ipath->fspath->val[i] = (u64)(unsigned long)fspath;
1925                 ++ipath->fspath->elem_cnt;
1926                 ipath->fspath->bytes_left = fspath - fspath_min;
1927         } else {
1928                 ++ipath->fspath->elem_missed;
1929                 ipath->fspath->bytes_missing += fspath_min - fspath;
1930                 ipath->fspath->bytes_left = 0;
1931         }
1932
1933         return 0;
1934 }
1935
1936 /*
1937  * this dumps all file system paths to the inode into the ipath struct, provided
1938  * is has been created large enough. each path is zero-terminated and accessed
1939  * from ipath->fspath->val[i].
1940  * when it returns, there are ipath->fspath->elem_cnt number of paths available
1941  * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the
1942  * number of missed paths is recorded in ipath->fspath->elem_missed, otherwise,
1943  * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
1944  * have been needed to return all paths.
1945  */
1946 int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
1947 {
1948         return iterate_irefs(inum, ipath->fs_root, ipath->btrfs_path,
1949                              inode_to_path, ipath);
1950 }
1951
1952 struct btrfs_data_container *init_data_container(u32 total_bytes)
1953 {
1954         struct btrfs_data_container *data;
1955         size_t alloc_bytes;
1956
1957         alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
1958         data = vmalloc(alloc_bytes);
1959         if (!data)
1960                 return ERR_PTR(-ENOMEM);
1961
1962         if (total_bytes >= sizeof(*data)) {
1963                 data->bytes_left = total_bytes - sizeof(*data);
1964                 data->bytes_missing = 0;
1965         } else {
1966                 data->bytes_missing = sizeof(*data) - total_bytes;
1967                 data->bytes_left = 0;
1968         }
1969
1970         data->elem_cnt = 0;
1971         data->elem_missed = 0;
1972
1973         return data;
1974 }
1975
1976 /*
1977  * allocates space to return multiple file system paths for an inode.
1978  * total_bytes to allocate are passed, note that space usable for actual path
1979  * information will be total_bytes - sizeof(struct inode_fs_paths).
1980  * the returned pointer must be freed with free_ipath() in the end.
1981  */
1982 struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
1983                                         struct btrfs_path *path)
1984 {
1985         struct inode_fs_paths *ifp;
1986         struct btrfs_data_container *fspath;
1987
1988         fspath = init_data_container(total_bytes);
1989         if (IS_ERR(fspath))
1990                 return (void *)fspath;
1991
1992         ifp = kmalloc(sizeof(*ifp), GFP_NOFS);
1993         if (!ifp) {
1994                 vfree(fspath);
1995                 return ERR_PTR(-ENOMEM);
1996         }
1997
1998         ifp->btrfs_path = path;
1999         ifp->fspath = fspath;
2000         ifp->fs_root = fs_root;
2001
2002         return ifp;
2003 }
2004
2005 void free_ipath(struct inode_fs_paths *ipath)
2006 {
2007         if (!ipath)
2008                 return;
2009         vfree(ipath->fspath);
2010         kfree(ipath);
2011 }