fs/btrfs/backref.c

   1 /*
   2  * Copyright (C) 2011 STRATO.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/vmalloc.h>
  20 #include "ctree.h"
  21 #include "disk-io.h"
  22 #include "backref.h"
  23 #include "ulist.h"
  24 #include "transaction.h"
  25 #include "delayed-ref.h"
  26 #include "locking.h"
  27
  28 /* Just an arbitrary number so we can be sure this happened */
  29 #define BACKREF_FOUND_SHARED 6
  30
  31 struct extent_inode_elem {
  32         u64 inum;
  33         u64 offset;
  34         struct extent_inode_elem *next;
  35 };
  36
  37 static int check_extent_in_eb(struct btrfs_key *key, struct extent_buffer *eb,
  38                                 struct btrfs_file_extent_item *fi,
  39                                 u64 extent_item_pos,
  40                                 struct extent_inode_elem **eie)
  41 {
  42         u64 offset = 0;
  43         struct extent_inode_elem *e;
  44
  45         if (!btrfs_file_extent_compression(eb, fi) &&
  46             !btrfs_file_extent_encryption(eb, fi) &&
  47             !btrfs_file_extent_other_encoding(eb, fi)) {
  48                 u64 data_offset;
  49                 u64 data_len;
  50
  51                 data_offset = btrfs_file_extent_offset(eb, fi);
  52                 data_len = btrfs_file_extent_num_bytes(eb, fi);
  53
  54                 if (extent_item_pos < data_offset ||
  55                     extent_item_pos >= data_offset + data_len)
  56                         return 1;
  57                 offset = extent_item_pos - data_offset;
  58         }
  59
  60         e = kmalloc(sizeof(*e), GFP_NOFS);
  61         if (!e)
  62                 return -ENOMEM;
  63
  64         e->next = *eie;
  65         e->inum = key->objectid;
  66         e->offset = key->offset + offset;
  67         *eie = e;
  68
  69         return 0;
  70 }
  71
  72 static void free_inode_elem_list(struct extent_inode_elem *eie)
  73 {
  74         struct extent_inode_elem *eie_next;
  75
  76         for (; eie; eie = eie_next) {
  77                 eie_next = eie->next;
  78                 kfree(eie);
  79         }
  80 }
  81
  82 static int find_extent_in_eb(struct extent_buffer *eb, u64 wanted_disk_byte,
  83                                 u64 extent_item_pos,
  84                                 struct extent_inode_elem **eie)
  85 {
  86         u64 disk_byte;
  87         struct btrfs_key key;
  88         struct btrfs_file_extent_item *fi;
  89         int slot;
  90         int nritems;
  91         int extent_type;
  92         int ret;
  93
  94         /*
  95          * from the shared data ref, we only have the leaf but we need
  96          * the key. thus, we must look into all items and see that we
  97          * find one (some) with a reference to our extent item.
  98          */
  99         nritems = btrfs_header_nritems(eb);
 100         for (slot = 0; slot < nritems; ++slot) {
 101                 btrfs_item_key_to_cpu(eb, &key, slot);
 102                 if (key.type != BTRFS_EXTENT_DATA_KEY)
 103                         continue;
 104                 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
 105                 extent_type = btrfs_file_extent_type(eb, fi);
 106                 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
 107                         continue;
 108                 /* don't skip BTRFS_FILE_EXTENT_PREALLOC, we can handle that */
 109                 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
 110                 if (disk_byte != wanted_disk_byte)
 111                         continue;
 112
 113                 ret = check_extent_in_eb(&key, eb, fi, extent_item_pos, eie);
 114                 if (ret < 0)
 115                         return ret;
 116         }
 117
 118         return 0;
 119 }
 120
 121 /*
 122  * this structure records all encountered refs on the way up to the root
 123  */
 124 struct __prelim_ref {
 125         struct list_head list;
 126         u64 root_id;
 127         struct btrfs_key key_for_search;
 128         int level;
 129         int count;
 130         struct extent_inode_elem *inode_list;
 131         u64 parent;
 132         u64 wanted_disk_byte;
 133 };
 134
 135 static struct kmem_cache *btrfs_prelim_ref_cache;
 136
 137 int __init btrfs_prelim_ref_init(void)
 138 {
 139         btrfs_prelim_ref_cache = kmem_cache_create("btrfs_prelim_ref",
 140                                         sizeof(struct __prelim_ref),
 141                                         0,
 142                                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
 143                                         NULL);
 144         if (!btrfs_prelim_ref_cache)
 145                 return -ENOMEM;
 146         return 0;
 147 }
 148
 149 void btrfs_prelim_ref_exit(void)
 150 {
 151         if (btrfs_prelim_ref_cache)
 152                 kmem_cache_destroy(btrfs_prelim_ref_cache);
 153 }
 154
 155 /*
 156  * the rules for all callers of this function are:
 157  * - obtaining the parent is the goal
 158  * - if you add a key, you must know that it is a correct key
 159  * - if you cannot add the parent or a correct key, then we will look into the
 160  *   block later to set a correct key
 161  *
 162  * delayed refs
 163  * ============
 164  *        backref type | shared | indirect | shared | indirect
 165  * information         |   tree |     tree |   data |     data
 166  * --------------------+--------+----------+--------+----------
 167  *      parent logical |    y   |     -    |    -   |     -
 168  *      key to resolve |    -   |     y    |    y   |     y
 169  *  tree block logical |    -   |     -    |    -   |     -
 170  *  root for resolving |    y   |     y    |    y   |     y
 171  *
 172  * - column 1:       we've the parent -> done
 173  * - column 2, 3, 4: we use the key to find the parent
 174  *
 175  * on disk refs (inline or keyed)
 176  * ==============================
 177  *        backref type | shared | indirect | shared | indirect
 178  * information         |   tree |     tree |   data |     data
 179  * --------------------+--------+----------+--------+----------
 180  *      parent logical |    y   |     -    |    y   |     -
 181  *      key to resolve |    -   |     -    |    -   |     y
 182  *  tree block logical |    y   |     y    |    y   |     y
 183  *  root for resolving |    -   |     y    |    y   |     y
 184  *
 185  * - column 1, 3: we've the parent -> done
 186  * - column 2:    we take the first key from the block to find the parent
 187  *                (see __add_missing_keys)
 188  * - column 4:    we use the key to find the parent
 189  *
 190  * additional information that's available but not required to find the parent
 191  * block might help in merging entries to gain some speed.
 192  */
 193
 194 static int __add_prelim_ref(struct list_head *head, u64 root_id,
 195                             struct btrfs_key *key, int level,
 196                             u64 parent, u64 wanted_disk_byte, int count,
 197                             gfp_t gfp_mask)
 198 {
 199         struct __prelim_ref *ref;
 200
 201         if (root_id == BTRFS_DATA_RELOC_TREE_OBJECTID)
 202                 return 0;
 203
 204         ref = kmem_cache_alloc(btrfs_prelim_ref_cache, gfp_mask);
 205         if (!ref)
 206                 return -ENOMEM;
 207
 208         ref->root_id = root_id;
 209         if (key) {
 210                 ref->key_for_search = *key;
 211                 /*
 212                  * We can often find data backrefs with an offset that is too
 213                  * large (>= LLONG_MAX, maximum allowed file offset) due to
 214                  * underflows when subtracting a file's offset with the data
 215                  * offset of its corresponding extent data item. This can
 216                  * happen for example in the clone ioctl.
 217                  * So if we detect such case we set the search key's offset to
 218                  * zero to make sure we will find the matching file extent item
 219                  * at add_all_parents(), otherwise we will miss it because the
 220                  * offset taken form the backref is much larger then the offset
 221                  * of the file extent item. This can make us scan a very large
 222                  * number of file extent items, but at least it will not make
 223                  * us miss any.
 224                  * This is an ugly workaround for a behaviour that should have
 225                  * never existed, but it does and a fix for the clone ioctl
 226                  * would touch a lot of places, cause backwards incompatibility
 227                  * and would not fix the problem for extents cloned with older
 228                  * kernels.
 229                  */
 230                 if (ref->key_for_search.type == BTRFS_EXTENT_DATA_KEY &&
 231                     ref->key_for_search.offset >= LLONG_MAX)
 232                         ref->key_for_search.offset = 0;
 233         } else {
 234                 memset(&ref->key_for_search, 0, sizeof(ref->key_for_search));
 235         }
 236
 237         ref->inode_list = NULL;
 238         ref->level = level;
 239         ref->count = count;
 240         ref->parent = parent;
 241         ref->wanted_disk_byte = wanted_disk_byte;
 242         list_add_tail(&ref->list, head);
 243
 244         return 0;
 245 }
 246
 247 static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path,
 248                            struct ulist *parents, struct __prelim_ref *ref,
 249                            int level, u64 time_seq, const u64 *extent_item_pos,
 250                            u64 total_refs)
 251 {
 252         int ret = 0;
 253         int slot;
 254         struct extent_buffer *eb;
 255         struct btrfs_key key;
 256         struct btrfs_key *key_for_search = &ref->key_for_search;
 257         struct btrfs_file_extent_item *fi;
 258         struct extent_inode_elem *eie = NULL, *old = NULL;
 259         u64 disk_byte;
 260         u64 wanted_disk_byte = ref->wanted_disk_byte;
 261         u64 count = 0;
 262
 263         if (level != 0) {
 264                 eb = path->nodes[level];
 265                 ret = ulist_add(parents, eb->start, 0, GFP_NOFS);
 266                 if (ret < 0)
 267                         return ret;
 268                 return 0;
 269         }
 270
 271         /*
 272          * We normally enter this function with the path already pointing to
 273          * the first item to check. But sometimes, we may enter it with
 274          * slot==nritems. In that case, go to the next leaf before we continue.
 275          */
 276         if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
 277                 if (time_seq == (u64)-1)
 278                         ret = btrfs_next_leaf(root, path);
 279                 else
 280                         ret = btrfs_next_old_leaf(root, path, time_seq);
 281         }
 282
 283         while (!ret && count < total_refs) {
 284                 eb = path->nodes[0];
 285                 slot = path->slots[0];
 286
 287                 btrfs_item_key_to_cpu(eb, &key, slot);
 288
 289                 if (key.objectid != key_for_search->objectid ||
 290                     key.type != BTRFS_EXTENT_DATA_KEY)
 291                         break;
 292
 293                 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
 294                 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
 295
 296                 if (disk_byte == wanted_disk_byte) {
 297                         eie = NULL;
 298                         old = NULL;
 299                         count++;
 300                         if (extent_item_pos) {
 301                                 ret = check_extent_in_eb(&key, eb, fi,
 302                                                 *extent_item_pos,
 303                                                 &eie);
 304                                 if (ret < 0)
 305                                         break;
 306                         }
 307                         if (ret > 0)
 308                                 goto next;
 309                         ret = ulist_add_merge_ptr(parents, eb->start,
 310                                                   eie, (void **)&old, GFP_NOFS);
 311                         if (ret < 0)
 312                                 break;
 313                         if (!ret && extent_item_pos) {
 314                                 while (old->next)
 315                                         old = old->next;
 316                                 old->next = eie;
 317                         }
 318                         eie = NULL;
 319                 }
 320 next:
 321                 if (time_seq == (u64)-1)
 322                         ret = btrfs_next_item(root, path);
 323                 else
 324                         ret = btrfs_next_old_item(root, path, time_seq);
 325         }
 326
 327         if (ret > 0)
 328                 ret = 0;
 329         else if (ret < 0)
 330                 free_inode_elem_list(eie);
 331         return ret;
 332 }
 333
 334 /*
 335  * resolve an indirect backref in the form (root_id, key, level)
 336  * to a logical address
 337  */
 338 static int __resolve_indirect_ref(struct btrfs_fs_info *fs_info,
 339                                   struct btrfs_path *path, u64 time_seq,
 340                                   struct __prelim_ref *ref,
 341                                   struct ulist *parents,
 342                                   const u64 *extent_item_pos, u64 total_refs)
 343 {
 344         struct btrfs_root *root;
 345         struct btrfs_key root_key;
 346         struct extent_buffer *eb;
 347         int ret = 0;
 348         int root_level;
 349         int level = ref->level;
 350         int index;
 351
 352         root_key.objectid = ref->root_id;
 353         root_key.type = BTRFS_ROOT_ITEM_KEY;
 354         root_key.offset = (u64)-1;
 355
 356         index = srcu_read_lock(&fs_info->subvol_srcu);
 357
 358         root = btrfs_get_fs_root(fs_info, &root_key, false);
 359         if (IS_ERR(root)) {
 360                 srcu_read_unlock(&fs_info->subvol_srcu, index);
 361                 ret = PTR_ERR(root);
 362                 goto out;
 363         }
 364
 365         if (btrfs_test_is_dummy_root(root)) {
 366                 srcu_read_unlock(&fs_info->subvol_srcu, index);
 367                 ret = -ENOENT;
 368                 goto out;
 369         }
 370
 371         if (path->search_commit_root)
 372                 root_level = btrfs_header_level(root->commit_root);
 373         else if (time_seq == (u64)-1)
 374                 root_level = btrfs_header_level(root->node);
 375         else
 376                 root_level = btrfs_old_root_level(root, time_seq);
 377
 378         if (root_level + 1 == level) {
 379                 srcu_read_unlock(&fs_info->subvol_srcu, index);
 380                 goto out;
 381         }
 382
 383         path->lowest_level = level;
 384         if (time_seq == (u64)-1)
 385                 ret = btrfs_search_slot(NULL, root, &ref->key_for_search, path,
 386                                         0, 0);
 387         else
 388                 ret = btrfs_search_old_slot(root, &ref->key_for_search, path,
 389                                             time_seq);
 390
 391         /* root node has been locked, we can release @subvol_srcu safely here */
 392         srcu_read_unlock(&fs_info->subvol_srcu, index);
 393
 394         pr_debug("search slot in root %llu (level %d, ref count %d) returned "
 395                  "%d for key (%llu %u %llu)\n",
 396                  ref->root_id, level, ref->count, ret,
 397                  ref->key_for_search.objectid, ref->key_for_search.type,
 398                  ref->key_for_search.offset);
 399         if (ret < 0)
 400                 goto out;
 401
 402         eb = path->nodes[level];
 403         while (!eb) {
 404                 if (WARN_ON(!level)) {
 405                         ret = 1;
 406                         goto out;
 407                 }
 408                 level--;
 409                 eb = path->nodes[level];
 410         }
 411
 412         ret = add_all_parents(root, path, parents, ref, level, time_seq,
 413                               extent_item_pos, total_refs);
 414 out:
 415         path->lowest_level = 0;
 416         btrfs_release_path(path);
 417         return ret;
 418 }
 419
 420 /*
 421  * resolve all indirect backrefs from the list
 422  */
 423 static int __resolve_indirect_refs(struct btrfs_fs_info *fs_info,
 424                                    struct btrfs_path *path, u64 time_seq,
 425                                    struct list_head *head,
 426                                    const u64 *extent_item_pos, u64 total_refs,
 427                                    u64 root_objectid)
 428 {
 429         int err;
 430         int ret = 0;
 431         struct __prelim_ref *ref;
 432         struct __prelim_ref *ref_safe;
 433         struct __prelim_ref *new_ref;
 434         struct ulist *parents;
 435         struct ulist_node *node;
 436         struct ulist_iterator uiter;
 437
 438         parents = ulist_alloc(GFP_NOFS);
 439         if (!parents)
 440                 return -ENOMEM;
 441
 442         /*
 443          * _safe allows us to insert directly after the current item without
 444          * iterating over the newly inserted items.
 445          * we're also allowed to re-assign ref during iteration.
 446          */
 447         list_for_each_entry_safe(ref, ref_safe, head, list) {
 448                 if (ref->parent)        /* already direct */
 449                         continue;
 450                 if (ref->count == 0)
 451                         continue;
 452                 if (root_objectid && ref->root_id != root_objectid) {
 453                         ret = BACKREF_FOUND_SHARED;
 454                         goto out;
 455                 }
 456                 err = __resolve_indirect_ref(fs_info, path, time_seq, ref,
 457                                              parents, extent_item_pos,
 458                                              total_refs);
 459                 /*
 460                  * we can only tolerate ENOENT,otherwise,we should catch error
 461                  * and return directly.
 462                  */
 463                 if (err == -ENOENT) {
 464                         continue;
 465                 } else if (err) {
 466                         ret = err;
 467                         goto out;
 468                 }
 469
 470                 /* we put the first parent into the ref at hand */
 471                 ULIST_ITER_INIT(&uiter);
 472                 node = ulist_next(parents, &uiter);
 473                 ref->parent = node ? node->val : 0;
 474                 ref->inode_list = node ?
 475                         (struct extent_inode_elem *)(uintptr_t)node->aux : NULL;
 476
 477                 /* additional parents require new refs being added here */
 478                 while ((node = ulist_next(parents, &uiter))) {
 479                         new_ref = kmem_cache_alloc(btrfs_prelim_ref_cache,
 480                                                    GFP_NOFS);
 481                         if (!new_ref) {
 482                                 ret = -ENOMEM;
 483                                 goto out;
 484                         }
 485                         memcpy(new_ref, ref, sizeof(*ref));
 486                         new_ref->parent = node->val;
 487                         new_ref->inode_list = (struct extent_inode_elem *)
 488                                                         (uintptr_t)node->aux;
 489                         list_add(&new_ref->list, &ref->list);
 490                 }
 491                 ulist_reinit(parents);
 492         }
 493 out:
 494         ulist_free(parents);
 495         return ret;
 496 }
 497
 498 static inline int ref_for_same_block(struct __prelim_ref *ref1,
 499                                      struct __prelim_ref *ref2)
 500 {
 501         if (ref1->level != ref2->level)
 502                 return 0;
 503         if (ref1->root_id != ref2->root_id)
 504                 return 0;
 505         if (ref1->key_for_search.type != ref2->key_for_search.type)
 506                 return 0;
 507         if (ref1->key_for_search.objectid != ref2->key_for_search.objectid)
 508                 return 0;
 509         if (ref1->key_for_search.offset != ref2->key_for_search.offset)
 510                 return 0;
 511         if (ref1->parent != ref2->parent)
 512                 return 0;
 513
 514         return 1;
 515 }
 516
 517 /*
 518  * read tree blocks and add keys where required.
 519  */
 520 static int __add_missing_keys(struct btrfs_fs_info *fs_info,
 521                               struct list_head *head)
 522 {
 523         struct __prelim_ref *ref;
 524         struct extent_buffer *eb;
 525
 526         list_for_each_entry(ref, head, list) {
 527                 if (ref->parent)
 528                         continue;
 529                 if (ref->key_for_search.type)
 530                         continue;
 531                 BUG_ON(!ref->wanted_disk_byte);
 532                 eb = read_tree_block(fs_info->tree_root, ref->wanted_disk_byte,
 533                                      0);
 534                 if (IS_ERR(eb)) {
 535                         return PTR_ERR(eb);
 536                 } else if (!extent_buffer_uptodate(eb)) {
 537                         free_extent_buffer(eb);
 538                         return -EIO;
 539                 }
 540                 btrfs_tree_read_lock(eb);
 541                 if (btrfs_header_level(eb) == 0)
 542                         btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
 543                 else
 544                         btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
 545                 btrfs_tree_read_unlock(eb);
 546                 free_extent_buffer(eb);
 547         }
 548         return 0;
 549 }
 550
 551 /*
 552  * merge backrefs and adjust counts accordingly
 553  *
 554  * mode = 1: merge identical keys, if key is set
 555  *    FIXME: if we add more keys in __add_prelim_ref, we can merge more here.
 556  *           additionally, we could even add a key range for the blocks we
 557  *           looked into to merge even more (-> replace unresolved refs by those
 558  *           having a parent).
 559  * mode = 2: merge identical parents
 560  */
 561 static void __merge_refs(struct list_head *head, int mode)
 562 {
 563         struct __prelim_ref *pos1;
 564
 565         list_for_each_entry(pos1, head, list) {
 566                 struct __prelim_ref *pos2 = pos1, *tmp;
 567
 568                 list_for_each_entry_safe_continue(pos2, tmp, head, list) {
 569                         struct __prelim_ref *xchg, *ref1 = pos1, *ref2 = pos2;
 570                         struct extent_inode_elem *eie;
 571
 572                         if (!ref_for_same_block(ref1, ref2))
 573                                 continue;
 574                         if (mode == 1) {
 575                                 if (!ref1->parent && ref2->parent) {
 576                                         xchg = ref1;
 577                                         ref1 = ref2;
 578                                         ref2 = xchg;
 579                                 }
 580                         } else {
 581                                 if (ref1->parent != ref2->parent)
 582                                         continue;
 583                         }
 584
 585                         eie = ref1->inode_list;
 586                         while (eie && eie->next)
 587                                 eie = eie->next;
 588                         if (eie)
 589                                 eie->next = ref2->inode_list;
 590                         else
 591                                 ref1->inode_list = ref2->inode_list;
 592                         ref1->count += ref2->count;
 593
 594                         list_del(&ref2->list);
 595                         kmem_cache_free(btrfs_prelim_ref_cache, ref2);
 596                 }
 597
 598         }
 599 }
 600
 601 /*
 602  * add all currently queued delayed refs from this head whose seq nr is
 603  * smaller or equal that seq to the list
 604  */
 605 static int __add_delayed_refs(struct btrfs_delayed_ref_head *head, u64 seq,
 606                               struct list_head *prefs, u64 *total_refs,
 607                               u64 inum)
 608 {
 609         struct btrfs_delayed_ref_node *node;
 610         struct btrfs_delayed_extent_op *extent_op = head->extent_op;
 611         struct btrfs_key key;
 612         struct btrfs_key op_key = {0};
 613         int sgn;
 614         int ret = 0;
 615
 616         if (extent_op && extent_op->update_key)
 617                 btrfs_disk_key_to_cpu(&op_key, &extent_op->key);
 618
 619         spin_lock(&head->lock);
 620         list_for_each_entry(node, &head->ref_list, list) {
 621                 if (node->seq > seq)
 622                         continue;
 623
 624                 switch (node->action) {
 625                 case BTRFS_ADD_DELAYED_EXTENT:
 626                 case BTRFS_UPDATE_DELAYED_HEAD:
 627                         WARN_ON(1);
 628                         continue;
 629                 case BTRFS_ADD_DELAYED_REF:
 630                         sgn = 1;
 631                         break;
 632                 case BTRFS_DROP_DELAYED_REF:
 633                         sgn = -1;
 634                         break;
 635                 default:
 636                         BUG_ON(1);
 637                 }
 638                 *total_refs += (node->ref_mod * sgn);
 639                 switch (node->type) {
 640                 case BTRFS_TREE_BLOCK_REF_KEY: {
 641                         struct btrfs_delayed_tree_ref *ref;
 642
 643                         ref = btrfs_delayed_node_to_tree_ref(node);
 644                         ret = __add_prelim_ref(prefs, ref->root, &op_key,
 645                                                ref->level + 1, 0, node->bytenr,
 646                                                node->ref_mod * sgn, GFP_ATOMIC);
 647                         break;
 648                 }
 649                 case BTRFS_SHARED_BLOCK_REF_KEY: {
 650                         struct btrfs_delayed_tree_ref *ref;
 651
 652                         ref = btrfs_delayed_node_to_tree_ref(node);
 653                         ret = __add_prelim_ref(prefs, 0, NULL,
 654                                                ref->level + 1, ref->parent,
 655                                                node->bytenr,
 656                                                node->ref_mod * sgn, GFP_ATOMIC);
 657                         break;
 658                 }
 659                 case BTRFS_EXTENT_DATA_REF_KEY: {
 660                         struct btrfs_delayed_data_ref *ref;
 661                         ref = btrfs_delayed_node_to_data_ref(node);
 662
 663                         key.objectid = ref->objectid;
 664                         key.type = BTRFS_EXTENT_DATA_KEY;
 665                         key.offset = ref->offset;
 666
 667                         /*
 668                          * Found a inum that doesn't match our known inum, we
 669                          * know it's shared.
 670                          */
 671                         if (inum && ref->objectid != inum) {
 672                                 ret = BACKREF_FOUND_SHARED;
 673                                 break;
 674                         }
 675
 676                         ret = __add_prelim_ref(prefs, ref->root, &key, 0, 0,
 677                                                node->bytenr,
 678                                                node->ref_mod * sgn, GFP_ATOMIC);
 679                         break;
 680                 }
 681                 case BTRFS_SHARED_DATA_REF_KEY: {
 682                         struct btrfs_delayed_data_ref *ref;
 683
 684                         ref = btrfs_delayed_node_to_data_ref(node);
 685                         ret = __add_prelim_ref(prefs, 0, NULL, 0,
 686                                                ref->parent, node->bytenr,
 687                                                node->ref_mod * sgn, GFP_ATOMIC);
 688                         break;
 689                 }
 690                 default:
 691                         WARN_ON(1);
 692                 }
 693                 if (ret)
 694                         break;
 695         }
 696         spin_unlock(&head->lock);
 697         return ret;
 698 }
 699
 700 /*
 701  * add all inline backrefs for bytenr to the list
 702  */
 703 static int __add_inline_refs(struct btrfs_fs_info *fs_info,
 704                              struct btrfs_path *path, u64 bytenr,
 705                              int *info_level, struct list_head *prefs,
 706                              u64 *total_refs, u64 inum)
 707 {
 708         int ret = 0;
 709         int slot;
 710         struct extent_buffer *leaf;
 711         struct btrfs_key key;
 712         struct btrfs_key found_key;
 713         unsigned long ptr;
 714         unsigned long end;
 715         struct btrfs_extent_item *ei;
 716         u64 flags;
 717         u64 item_size;
 718
 719         /*
 720          * enumerate all inline refs
 721          */
 722         leaf = path->nodes[0];
 723         slot = path->slots[0];
 724
 725         item_size = btrfs_item_size_nr(leaf, slot);
 726         BUG_ON(item_size < sizeof(*ei));
 727
 728         ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
 729         flags = btrfs_extent_flags(leaf, ei);
 730         *total_refs += btrfs_extent_refs(leaf, ei);
 731         btrfs_item_key_to_cpu(leaf, &found_key, slot);
 732
 733         ptr = (unsigned long)(ei + 1);
 734         end = (unsigned long)ei + item_size;
 735
 736         if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
 737             flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
 738                 struct btrfs_tree_block_info *info;
 739
 740                 info = (struct btrfs_tree_block_info *)ptr;
 741                 *info_level = btrfs_tree_block_level(leaf, info);
 742                 ptr += sizeof(struct btrfs_tree_block_info);
 743                 BUG_ON(ptr > end);
 744         } else if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
 745                 *info_level = found_key.offset;
 746         } else {
 747                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
 748         }
 749
 750         while (ptr < end) {
 751                 struct btrfs_extent_inline_ref *iref;
 752                 u64 offset;
 753                 int type;
 754
 755                 iref = (struct btrfs_extent_inline_ref *)ptr;
 756                 type = btrfs_extent_inline_ref_type(leaf, iref);
 757                 offset = btrfs_extent_inline_ref_offset(leaf, iref);
 758
 759                 switch (type) {
 760                 case BTRFS_SHARED_BLOCK_REF_KEY:
 761                         ret = __add_prelim_ref(prefs, 0, NULL,
 762                                                 *info_level + 1, offset,
 763                                                 bytenr, 1, GFP_NOFS);
 764                         break;
 765                 case BTRFS_SHARED_DATA_REF_KEY: {
 766                         struct btrfs_shared_data_ref *sdref;
 767                         int count;
 768
 769                         sdref = (struct btrfs_shared_data_ref *)(iref + 1);
 770                         count = btrfs_shared_data_ref_count(leaf, sdref);
 771                         ret = __add_prelim_ref(prefs, 0, NULL, 0, offset,
 772                                                bytenr, count, GFP_NOFS);
 773                         break;
 774                 }
 775                 case BTRFS_TREE_BLOCK_REF_KEY:
 776                         ret = __add_prelim_ref(prefs, offset, NULL,
 777                                                *info_level + 1, 0,
 778                                                bytenr, 1, GFP_NOFS);
 779                         break;
 780                 case BTRFS_EXTENT_DATA_REF_KEY: {
 781                         struct btrfs_extent_data_ref *dref;
 782                         int count;
 783                         u64 root;
 784
 785                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
 786                         count = btrfs_extent_data_ref_count(leaf, dref);
 787                         key.objectid = btrfs_extent_data_ref_objectid(leaf,
 788                                                                       dref);
 789                         key.type = BTRFS_EXTENT_DATA_KEY;
 790                         key.offset = btrfs_extent_data_ref_offset(leaf, dref);
 791
 792                         if (inum && key.objectid != inum) {
 793                                 ret = BACKREF_FOUND_SHARED;
 794                                 break;
 795                         }
 796
 797                         root = btrfs_extent_data_ref_root(leaf, dref);
 798                         ret = __add_prelim_ref(prefs, root, &key, 0, 0,
 799                                                bytenr, count, GFP_NOFS);
 800                         break;
 801                 }
 802                 default:
 803                         WARN_ON(1);
 804                 }
 805                 if (ret)
 806                         return ret;
 807                 ptr += btrfs_extent_inline_ref_size(type);
 808         }
 809
 810         return 0;
 811 }
 812
 813 /*
 814  * add all non-inline backrefs for bytenr to the list
 815  */
 816 static int __add_keyed_refs(struct btrfs_fs_info *fs_info,
 817                             struct btrfs_path *path, u64 bytenr,
 818                             int info_level, struct list_head *prefs, u64 inum)
 819 {
 820         struct btrfs_root *extent_root = fs_info->extent_root;
 821         int ret;
 822         int slot;
 823         struct extent_buffer *leaf;
 824         struct btrfs_key key;
 825
 826         while (1) {
 827                 ret = btrfs_next_item(extent_root, path);
 828                 if (ret < 0)
 829                         break;
 830                 if (ret) {
 831                         ret = 0;
 832                         break;
 833                 }
 834
 835                 slot = path->slots[0];
 836                 leaf = path->nodes[0];
 837                 btrfs_item_key_to_cpu(leaf, &key, slot);
 838
 839                 if (key.objectid != bytenr)
 840                         break;
 841                 if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
 842                         continue;
 843                 if (key.type > BTRFS_SHARED_DATA_REF_KEY)
 844                         break;
 845
 846                 switch (key.type) {
 847                 case BTRFS_SHARED_BLOCK_REF_KEY:
 848                         ret = __add_prelim_ref(prefs, 0, NULL,
 849                                                 info_level + 1, key.offset,
 850                                                 bytenr, 1, GFP_NOFS);
 851                         break;
 852                 case BTRFS_SHARED_DATA_REF_KEY: {
 853                         struct btrfs_shared_data_ref *sdref;
 854                         int count;
 855
 856                         sdref = btrfs_item_ptr(leaf, slot,
 857                                               struct btrfs_shared_data_ref);
 858                         count = btrfs_shared_data_ref_count(leaf, sdref);
 859                         ret = __add_prelim_ref(prefs, 0, NULL, 0, key.offset,
 860                                                 bytenr, count, GFP_NOFS);
 861                         break;
 862                 }
 863                 case BTRFS_TREE_BLOCK_REF_KEY:
 864                         ret = __add_prelim_ref(prefs, key.offset, NULL,
 865                                                info_level + 1, 0,
 866                                                bytenr, 1, GFP_NOFS);
 867                         break;
 868                 case BTRFS_EXTENT_DATA_REF_KEY: {
 869                         struct btrfs_extent_data_ref *dref;
 870                         int count;
 871                         u64 root;
 872
 873                         dref = btrfs_item_ptr(leaf, slot,
 874                                               struct btrfs_extent_data_ref);
 875                         count = btrfs_extent_data_ref_count(leaf, dref);
 876                         key.objectid = btrfs_extent_data_ref_objectid(leaf,
 877                                                                       dref);
 878                         key.type = BTRFS_EXTENT_DATA_KEY;
 879                         key.offset = btrfs_extent_data_ref_offset(leaf, dref);
 880
 881                         if (inum && key.objectid != inum) {
 882                                 ret = BACKREF_FOUND_SHARED;
 883                                 break;
 884                         }
 885
 886                         root = btrfs_extent_data_ref_root(leaf, dref);
 887                         ret = __add_prelim_ref(prefs, root, &key, 0, 0,
 888                                                bytenr, count, GFP_NOFS);
 889                         break;
 890                 }
 891                 default:
 892                         WARN_ON(1);
 893                 }
 894                 if (ret)
 895                         return ret;
 896
 897         }
 898
 899         return ret;
 900 }
 901
 902 /*
 903  * this adds all existing backrefs (inline backrefs, backrefs and delayed
 904  * refs) for the given bytenr to the refs list, merges duplicates and resolves
 905  * indirect refs to their parent bytenr.
 906  * When roots are found, they're added to the roots list
 907  *
 908  * NOTE: This can return values > 0
 909  *
 910  * If time_seq is set to (u64)-1, it will not search delayed_refs, and behave
 911  * much like trans == NULL case, the difference only lies in it will not
 912  * commit root.
 913  * The special case is for qgroup to search roots in commit_transaction().
 914  *
 915  * FIXME some caching might speed things up
 916  */
 917 static int find_parent_nodes(struct btrfs_trans_handle *trans,
 918                              struct btrfs_fs_info *fs_info, u64 bytenr,
 919                              u64 time_seq, struct ulist *refs,
 920                              struct ulist *roots, const u64 *extent_item_pos,
 921                              u64 root_objectid, u64 inum)
 922 {
 923         struct btrfs_key key;
 924         struct btrfs_path *path;
 925         struct btrfs_delayed_ref_root *delayed_refs = NULL;
 926         struct btrfs_delayed_ref_head *head;
 927         int info_level = 0;
 928         int ret;
 929         struct list_head prefs_delayed;
 930         struct list_head prefs;
 931         struct __prelim_ref *ref;
 932         struct extent_inode_elem *eie = NULL;
 933         u64 total_refs = 0;
 934
 935         INIT_LIST_HEAD(&prefs);
 936         INIT_LIST_HEAD(&prefs_delayed);
 937
 938         key.objectid = bytenr;
 939         key.offset = (u64)-1;
 940         if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
 941                 key.type = BTRFS_METADATA_ITEM_KEY;
 942         else
 943                 key.type = BTRFS_EXTENT_ITEM_KEY;
 944
 945         path = btrfs_alloc_path();
 946         if (!path)
 947                 return -ENOMEM;
 948         if (!trans) {
 949                 path->search_commit_root = 1;
 950                 path->skip_locking = 1;
 951         }
 952
 953         if (time_seq == (u64)-1)
 954                 path->skip_locking = 1;
 955
 956         /*
 957          * grab both a lock on the path and a lock on the delayed ref head.
 958          * We need both to get a consistent picture of how the refs look
 959          * at a specified point in time
 960          */
 961 again:
 962         head = NULL;
 963
 964         ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
 965         if (ret < 0)
 966                 goto out;
 967         BUG_ON(ret == 0);
 968
 969 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
 970         if (trans && likely(trans->type != __TRANS_DUMMY) &&
 971             time_seq != (u64)-1) {
 972 #else
 973         if (trans && time_seq != (u64)-1) {
 974 #endif
 975                 /*
 976                  * look if there are updates for this ref queued and lock the
 977                  * head
 978                  */
 979                 delayed_refs = &trans->transaction->delayed_refs;
 980                 spin_lock(&delayed_refs->lock);
 981                 head = btrfs_find_delayed_ref_head(trans, bytenr);
 982                 if (head) {
 983                         if (!mutex_trylock(&head->mutex)) {
 984                                 atomic_inc(&head->node.refs);
 985                                 spin_unlock(&delayed_refs->lock);
 986
 987                                 btrfs_release_path(path);
 988
 989                                 /*
 990                                  * Mutex was contended, block until it's
 991                                  * released and try again
 992                                  */
 993                                 mutex_lock(&head->mutex);
 994                                 mutex_unlock(&head->mutex);
 995                                 btrfs_put_delayed_ref(&head->node);
 996                                 goto again;
 997                         }
 998                         spin_unlock(&delayed_refs->lock);
 999                         ret = __add_delayed_refs(head, time_seq,
1000                                                  &prefs_delayed, &total_refs,
1001                                                  inum);
1002                         mutex_unlock(&head->mutex);
1003                         if (ret)
1004                                 goto out;
1005                 } else {
1006                         spin_unlock(&delayed_refs->lock);
1007                 }
1008         }
1009
1010         if (path->slots[0]) {
1011                 struct extent_buffer *leaf;
1012                 int slot;
1013
1014                 path->slots[0]--;
1015                 leaf = path->nodes[0];
1016                 slot = path->slots[0];
1017                 btrfs_item_key_to_cpu(leaf, &key, slot);
1018                 if (key.objectid == bytenr &&
1019                     (key.type == BTRFS_EXTENT_ITEM_KEY ||
1020                      key.type == BTRFS_METADATA_ITEM_KEY)) {
1021                         ret = __add_inline_refs(fs_info, path, bytenr,
1022                                                 &info_level, &prefs,
1023                                                 &total_refs, inum);
1024                         if (ret)
1025                                 goto out;
1026                         ret = __add_keyed_refs(fs_info, path, bytenr,
1027                                                info_level, &prefs, inum);
1028                         if (ret)
1029                                 goto out;
1030                 }
1031         }
1032         btrfs_release_path(path);
1033
1034         list_splice_init(&prefs_delayed, &prefs);
1035
1036         ret = __add_missing_keys(fs_info, &prefs);
1037         if (ret)
1038                 goto out;
1039
1040         __merge_refs(&prefs, 1);
1041
1042         ret = __resolve_indirect_refs(fs_info, path, time_seq, &prefs,
1043                                       extent_item_pos, total_refs,
1044                                       root_objectid);
1045         if (ret)
1046                 goto out;
1047
1048         __merge_refs(&prefs, 2);
1049
1050         while (!list_empty(&prefs)) {
1051                 ref = list_first_entry(&prefs, struct __prelim_ref, list);
1052                 WARN_ON(ref->count < 0);
1053                 if (roots && ref->count && ref->root_id && ref->parent == 0) {
1054                         if (root_objectid && ref->root_id != root_objectid) {
1055                                 ret = BACKREF_FOUND_SHARED;
1056                                 goto out;
1057                         }
1058
1059                         /* no parent == root of tree */
1060                         ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS);
1061                         if (ret < 0)
1062                                 goto out;
1063                 }
1064                 if (ref->count && ref->parent) {
1065                         if (extent_item_pos && !ref->inode_list &&
1066                             ref->level == 0) {
1067                                 struct extent_buffer *eb;
1068
1069                                 eb = read_tree_block(fs_info->extent_root,
1070                                                            ref->parent, 0);
1071                                 if (IS_ERR(eb)) {
1072                                         ret = PTR_ERR(eb);
1073                                         goto out;
1074                                 } else if (!extent_buffer_uptodate(eb)) {
1075                                         free_extent_buffer(eb);
1076                                         ret = -EIO;
1077                                         goto out;
1078                                 }
1079                                 btrfs_tree_read_lock(eb);
1080                                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1081                                 ret = find_extent_in_eb(eb, bytenr,
1082                                                         *extent_item_pos, &eie);
1083                                 btrfs_tree_read_unlock_blocking(eb);
1084                                 free_extent_buffer(eb);
1085                                 if (ret < 0)
1086                                         goto out;
1087                                 ref->inode_list = eie;
1088                         }
1089                         ret = ulist_add_merge_ptr(refs, ref->parent,
1090                                                   ref->inode_list,
1091                                                   (void **)&eie, GFP_NOFS);
1092                         if (ret < 0)
1093                                 goto out;
1094                         if (!ret && extent_item_pos) {
1095                                 /*
1096                                  * we've recorded that parent, so we must extend
1097                                  * its inode list here
1098                                  */
1099                                 BUG_ON(!eie);
1100                                 while (eie->next)
1101                                         eie = eie->next;
1102                                 eie->next = ref->inode_list;
1103                         }
1104                         eie = NULL;
1105                 }
1106                 list_del(&ref->list);
1107                 kmem_cache_free(btrfs_prelim_ref_cache, ref);
1108         }
1109
1110 out:
1111         btrfs_free_path(path);
1112         while (!list_empty(&prefs)) {
1113                 ref = list_first_entry(&prefs, struct __prelim_ref, list);
1114                 list_del(&ref->list);
1115                 kmem_cache_free(btrfs_prelim_ref_cache, ref);
1116         }
1117         while (!list_empty(&prefs_delayed)) {
1118                 ref = list_first_entry(&prefs_delayed, struct __prelim_ref,
1119                                        list);
1120                 list_del(&ref->list);
1121                 kmem_cache_free(btrfs_prelim_ref_cache, ref);
1122         }
1123         if (ret < 0)
1124                 free_inode_elem_list(eie);
1125         return ret;
1126 }
1127
1128 static void free_leaf_list(struct ulist *blocks)
1129 {
1130         struct ulist_node *node = NULL;
1131         struct extent_inode_elem *eie;
1132         struct ulist_iterator uiter;
1133
1134         ULIST_ITER_INIT(&uiter);
1135         while ((node = ulist_next(blocks, &uiter))) {
1136                 if (!node->aux)
1137                         continue;
1138                 eie = (struct extent_inode_elem *)(uintptr_t)node->aux;
1139                 free_inode_elem_list(eie);
1140                 node->aux = 0;
1141         }
1142
1143         ulist_free(blocks);
1144 }
1145
1146 /*
1147  * Finds all leafs with a reference to the specified combination of bytenr and
1148  * offset. key_list_head will point to a list of corresponding keys (caller must
1149  * free each list element). The leafs will be stored in the leafs ulist, which
1150  * must be freed with ulist_free.
1151  *
1152  * returns 0 on success, <0 on error
1153  */
1154 static int btrfs_find_all_leafs(struct btrfs_trans_handle *trans,
1155                                 struct btrfs_fs_info *fs_info, u64 bytenr,
1156                                 u64 time_seq, struct ulist **leafs,
1157                                 const u64 *extent_item_pos)
1158 {
1159         int ret;
1160
1161         *leafs = ulist_alloc(GFP_NOFS);
1162         if (!*leafs)
1163                 return -ENOMEM;
1164
1165         ret = find_parent_nodes(trans, fs_info, bytenr,
1166                                 time_seq, *leafs, NULL, extent_item_pos, 0, 0);
1167         if (ret < 0 && ret != -ENOENT) {
1168                 free_leaf_list(*leafs);
1169                 return ret;
1170         }
1171
1172         return 0;
1173 }
1174
1175 /*
1176  * walk all backrefs for a given extent to find all roots that reference this
1177  * extent. Walking a backref means finding all extents that reference this
1178  * extent and in turn walk the backrefs of those, too. Naturally this is a
1179  * recursive process, but here it is implemented in an iterative fashion: We
1180  * find all referencing extents for the extent in question and put them on a
1181  * list. In turn, we find all referencing extents for those, further appending
1182  * to the list. The way we iterate the list allows adding more elements after
1183  * the current while iterating. The process stops when we reach the end of the
1184  * list. Found roots are added to the roots list.
1185  *
1186  * returns 0 on success, < 0 on error.
1187  */
1188 static int __btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1189                                   struct btrfs_fs_info *fs_info, u64 bytenr,
1190                                   u64 time_seq, struct ulist **roots)
1191 {
1192         struct ulist *tmp;
1193         struct ulist_node *node = NULL;
1194         struct ulist_iterator uiter;
1195         int ret;
1196
1197         tmp = ulist_alloc(GFP_NOFS);
1198         if (!tmp)
1199                 return -ENOMEM;
1200         *roots = ulist_alloc(GFP_NOFS);
1201         if (!*roots) {
1202                 ulist_free(tmp);
1203                 return -ENOMEM;
1204         }
1205
1206         ULIST_ITER_INIT(&uiter);
1207         while (1) {
1208                 ret = find_parent_nodes(trans, fs_info, bytenr,
1209                                         time_seq, tmp, *roots, NULL, 0, 0);
1210                 if (ret < 0 && ret != -ENOENT) {
1211                         ulist_free(tmp);
1212                         ulist_free(*roots);
1213                         return ret;
1214                 }
1215                 node = ulist_next(tmp, &uiter);
1216                 if (!node)
1217                         break;
1218                 bytenr = node->val;
1219                 cond_resched();
1220         }
1221
1222         ulist_free(tmp);
1223         return 0;
1224 }
1225
1226 int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1227                          struct btrfs_fs_info *fs_info, u64 bytenr,
1228                          u64 time_seq, struct ulist **roots)
1229 {
1230         int ret;
1231
1232         if (!trans)
1233                 down_read(&fs_info->commit_root_sem);
1234         ret = __btrfs_find_all_roots(trans, fs_info, bytenr, time_seq, roots);
1235         if (!trans)
1236                 up_read(&fs_info->commit_root_sem);
1237         return ret;
1238 }
1239
1240 /**
1241  * btrfs_check_shared - tell us whether an extent is shared
1242  *
1243  * @trans: optional trans handle
1244  *
1245  * btrfs_check_shared uses the backref walking code but will short
1246  * circuit as soon as it finds a root or inode that doesn't match the
1247  * one passed in. This provides a significant performance benefit for
1248  * callers (such as fiemap) which want to know whether the extent is
1249  * shared but do not need a ref count.
1250  *
1251  * Return: 0 if extent is not shared, 1 if it is shared, < 0 on error.
1252  */
1253 int btrfs_check_shared(struct btrfs_trans_handle *trans,
1254                        struct btrfs_fs_info *fs_info, u64 root_objectid,
1255                        u64 inum, u64 bytenr)
1256 {
1257         struct ulist *tmp = NULL;
1258         struct ulist *roots = NULL;
1259         struct ulist_iterator uiter;
1260         struct ulist_node *node;
1261         struct seq_list elem = SEQ_LIST_INIT(elem);
1262         int ret = 0;
1263
1264         tmp = ulist_alloc(GFP_NOFS);
1265         roots = ulist_alloc(GFP_NOFS);
1266         if (!tmp || !roots) {
1267                 ulist_free(tmp);
1268                 ulist_free(roots);
1269                 return -ENOMEM;
1270         }
1271
1272         if (trans)
1273                 btrfs_get_tree_mod_seq(fs_info, &elem);
1274         else
1275                 down_read(&fs_info->commit_root_sem);
1276         ULIST_ITER_INIT(&uiter);
1277         while (1) {
1278                 ret = find_parent_nodes(trans, fs_info, bytenr, elem.seq, tmp,
1279                                         roots, NULL, root_objectid, inum);
1280                 if (ret == BACKREF_FOUND_SHARED) {
1281                         /* this is the only condition under which we return 1 */
1282                         ret = 1;
1283                         break;
1284                 }
1285                 if (ret < 0 && ret != -ENOENT)
1286                         break;
1287                 ret = 0;
1288                 node = ulist_next(tmp, &uiter);
1289                 if (!node)
1290                         break;
1291                 bytenr = node->val;
1292                 cond_resched();
1293         }
1294         if (trans)
1295                 btrfs_put_tree_mod_seq(fs_info, &elem);
1296         else
1297                 up_read(&fs_info->commit_root_sem);
1298         ulist_free(tmp);
1299         ulist_free(roots);
1300         return ret;
1301 }
1302
1303 int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,
1304                           u64 start_off, struct btrfs_path *path,
1305                           struct btrfs_inode_extref **ret_extref,
1306                           u64 *found_off)
1307 {
1308         int ret, slot;
1309         struct btrfs_key key;
1310         struct btrfs_key found_key;
1311         struct btrfs_inode_extref *extref;
1312         struct extent_buffer *leaf;
1313         unsigned long ptr;
1314
1315         key.objectid = inode_objectid;
1316         key.type = BTRFS_INODE_EXTREF_KEY;
1317         key.offset = start_off;
1318
1319         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1320         if (ret < 0)
1321                 return ret;
1322
1323         while (1) {
1324                 leaf = path->nodes[0];
1325                 slot = path->slots[0];
1326                 if (slot >= btrfs_header_nritems(leaf)) {
1327                         /*
1328                          * If the item at offset is not found,
1329                          * btrfs_search_slot will point us to the slot
1330                          * where it should be inserted. In our case
1331                          * that will be the slot directly before the
1332                          * next INODE_REF_KEY_V2 item. In the case
1333                          * that we're pointing to the last slot in a
1334                          * leaf, we must move one leaf over.
1335                          */
1336                         ret = btrfs_next_leaf(root, path);
1337                         if (ret) {
1338                                 if (ret >= 1)
1339                                         ret = -ENOENT;
1340                                 break;
1341                         }
1342                         continue;
1343                 }
1344
1345                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1346
1347                 /*
1348                  * Check that we're still looking at an extended ref key for
1349                  * this particular objectid. If we have different
1350                  * objectid or type then there are no more to be found
1351                  * in the tree and we can exit.
1352                  */
1353                 ret = -ENOENT;
1354                 if (found_key.objectid != inode_objectid)
1355                         break;
1356                 if (found_key.type != BTRFS_INODE_EXTREF_KEY)
1357                         break;
1358
1359                 ret = 0;
1360                 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1361                 extref = (struct btrfs_inode_extref *)ptr;
1362                 *ret_extref = extref;
1363                 if (found_off)
1364                         *found_off = found_key.offset;
1365                 break;
1366         }
1367
1368         return ret;
1369 }
1370
1371 /*
1372  * this iterates to turn a name (from iref/extref) into a full filesystem path.
1373  * Elements of the path are separated by '/' and the path is guaranteed to be
1374  * 0-terminated. the path is only given within the current file system.
1375  * Therefore, it never starts with a '/'. the caller is responsible to provide
1376  * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
1377  * the start point of the resulting string is returned. this pointer is within
1378  * dest, normally.
1379  * in case the path buffer would overflow, the pointer is decremented further
1380  * as if output was written to the buffer, though no more output is actually
1381  * generated. that way, the caller can determine how much space would be
1382  * required for the path to fit into the buffer. in that case, the returned
1383  * value will be smaller than dest. callers must check this!
1384  */
1385 char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
1386                         u32 name_len, unsigned long name_off,
1387                         struct extent_buffer *eb_in, u64 parent,
1388                         char *dest, u32 size)
1389 {
1390         int slot;
1391         u64 next_inum;
1392         int ret;
1393         s64 bytes_left = ((s64)size) - 1;
1394         struct extent_buffer *eb = eb_in;
1395         struct btrfs_key found_key;
1396         int leave_spinning = path->leave_spinning;
1397         struct btrfs_inode_ref *iref;
1398
1399         if (bytes_left >= 0)
1400                 dest[bytes_left] = '\0';
1401
1402         path->leave_spinning = 1;
1403         while (1) {
1404                 bytes_left -= name_len;
1405                 if (bytes_left >= 0)
1406                         read_extent_buffer(eb, dest + bytes_left,
1407                                            name_off, name_len);
1408                 if (eb != eb_in) {
1409                         if (!path->skip_locking)
1410                                 btrfs_tree_read_unlock_blocking(eb);
1411                         free_extent_buffer(eb);
1412                 }
1413                 ret = btrfs_find_item(fs_root, path, parent, 0,
1414                                 BTRFS_INODE_REF_KEY, &found_key);
1415                 if (ret > 0)
1416                         ret = -ENOENT;
1417                 if (ret)
1418                         break;
1419
1420                 next_inum = found_key.offset;
1421
1422                 /* regular exit ahead */
1423                 if (parent == next_inum)
1424                         break;
1425
1426                 slot = path->slots[0];
1427                 eb = path->nodes[0];
1428                 /* make sure we can use eb after releasing the path */
1429                 if (eb != eb_in) {
1430                         if (!path->skip_locking)
1431                                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1432                         path->nodes[0] = NULL;
1433                         path->locks[0] = 0;
1434                 }
1435                 btrfs_release_path(path);
1436                 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1437
1438                 name_len = btrfs_inode_ref_name_len(eb, iref);
1439                 name_off = (unsigned long)(iref + 1);
1440
1441                 parent = next_inum;
1442                 --bytes_left;
1443                 if (bytes_left >= 0)
1444                         dest[bytes_left] = '/';
1445         }
1446
1447         btrfs_release_path(path);
1448         path->leave_spinning = leave_spinning;
1449
1450         if (ret)
1451                 return ERR_PTR(ret);
1452
1453         return dest + bytes_left;
1454 }
1455
1456 /*
1457  * this makes the path point to (logical EXTENT_ITEM *)
1458  * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
1459  * tree blocks and <0 on error.
1460  */
1461 int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
1462                         struct btrfs_path *path, struct btrfs_key *found_key,
1463                         u64 *flags_ret)
1464 {
1465         int ret;
1466         u64 flags;
1467         u64 size = 0;
1468         u32 item_size;
1469         struct extent_buffer *eb;
1470         struct btrfs_extent_item *ei;
1471         struct btrfs_key key;
1472
1473         if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1474                 key.type = BTRFS_METADATA_ITEM_KEY;
1475         else
1476                 key.type = BTRFS_EXTENT_ITEM_KEY;
1477         key.objectid = logical;
1478         key.offset = (u64)-1;
1479
1480         ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
1481         if (ret < 0)
1482                 return ret;
1483
1484         ret = btrfs_previous_extent_item(fs_info->extent_root, path, 0);
1485         if (ret) {
1486                 if (ret > 0)
1487                         ret = -ENOENT;
1488                 return ret;
1489         }
1490         btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]);
1491         if (found_key->type == BTRFS_METADATA_ITEM_KEY)
1492                 size = fs_info->extent_root->nodesize;
1493         else if (found_key->type == BTRFS_EXTENT_ITEM_KEY)
1494                 size = found_key->offset;
1495
1496         if (found_key->objectid > logical ||
1497             found_key->objectid + size <= logical) {
1498                 pr_debug("logical %llu is not within any extent\n", logical);
1499                 return -ENOENT;
1500         }
1501
1502         eb = path->nodes[0];
1503         item_size = btrfs_item_size_nr(eb, path->slots[0]);
1504         BUG_ON(item_size < sizeof(*ei));
1505
1506         ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
1507         flags = btrfs_extent_flags(eb, ei);
1508
1509         pr_debug("logical %llu is at position %llu within the extent (%llu "
1510                  "EXTENT_ITEM %llu) flags %#llx size %u\n",
1511                  logical, logical - found_key->objectid, found_key->objectid,
1512                  found_key->offset, flags, item_size);
1513
1514         WARN_ON(!flags_ret);
1515         if (flags_ret) {
1516                 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1517                         *flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK;
1518                 else if (flags & BTRFS_EXTENT_FLAG_DATA)
1519                         *flags_ret = BTRFS_EXTENT_FLAG_DATA;
1520                 else
1521                         BUG_ON(1);
1522                 return 0;
1523         }
1524
1525         return -EIO;
1526 }
1527
1528 /*
1529  * helper function to iterate extent inline refs. ptr must point to a 0 value
1530  * for the first call and may be modified. it is used to track state.
1531  * if more refs exist, 0 is returned and the next call to
1532  * __get_extent_inline_ref must pass the modified ptr parameter to get the
1533  * next ref. after the last ref was processed, 1 is returned.
1534  * returns <0 on error
1535  */
1536 static int __get_extent_inline_ref(unsigned long *ptr, struct extent_buffer *eb,
1537                                    struct btrfs_key *key,
1538                                    struct btrfs_extent_item *ei, u32 item_size,
1539                                    struct btrfs_extent_inline_ref **out_eiref,
1540                                    int *out_type)
1541 {
1542         unsigned long end;
1543         u64 flags;
1544         struct btrfs_tree_block_info *info;
1545
1546         if (!*ptr) {
1547                 /* first call */
1548                 flags = btrfs_extent_flags(eb, ei);
1549                 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1550                         if (key->type == BTRFS_METADATA_ITEM_KEY) {
1551                                 /* a skinny metadata extent */
1552                                 *out_eiref =
1553                                      (struct btrfs_extent_inline_ref *)(ei + 1);
1554                         } else {
1555                                 WARN_ON(key->type != BTRFS_EXTENT_ITEM_KEY);
1556                                 info = (struct btrfs_tree_block_info *)(ei + 1);
1557                                 *out_eiref =
1558                                    (struct btrfs_extent_inline_ref *)(info + 1);
1559                         }
1560                 } else {
1561                         *out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1);
1562                 }
1563                 *ptr = (unsigned long)*out_eiref;
1564                 if ((unsigned long)(*ptr) >= (unsigned long)ei + item_size)
1565                         return -ENOENT;
1566         }
1567
1568         end = (unsigned long)ei + item_size;
1569         *out_eiref = (struct btrfs_extent_inline_ref *)(*ptr);
1570         *out_type = btrfs_extent_inline_ref_type(eb, *out_eiref);
1571
1572         *ptr += btrfs_extent_inline_ref_size(*out_type);
1573         WARN_ON(*ptr > end);
1574         if (*ptr == end)
1575                 return 1; /* last */
1576
1577         return 0;
1578 }
1579
1580 /*
1581  * reads the tree block backref for an extent. tree level and root are returned
1582  * through out_level and out_root. ptr must point to a 0 value for the first
1583  * call and may be modified (see __get_extent_inline_ref comment).
1584  * returns 0 if data was provided, 1 if there was no more data to provide or
1585  * <0 on error.
1586  */
1587 int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
1588                             struct btrfs_key *key, struct btrfs_extent_item *ei,
1589                             u32 item_size, u64 *out_root, u8 *out_level)
1590 {
1591         int ret;
1592         int type;
1593         struct btrfs_extent_inline_ref *eiref;
1594
1595         if (*ptr == (unsigned long)-1)
1596                 return 1;
1597
1598         while (1) {
1599                 ret = __get_extent_inline_ref(ptr, eb, key, ei, item_size,
1600                                               &eiref, &type);
1601                 if (ret < 0)
1602                         return ret;
1603
1604                 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1605                     type == BTRFS_SHARED_BLOCK_REF_KEY)
1606                         break;
1607
1608                 if (ret == 1)
1609                         return 1;
1610         }
1611
1612         /* we can treat both ref types equally here */
1613         *out_root = btrfs_extent_inline_ref_offset(eb, eiref);
1614
1615         if (key->type == BTRFS_EXTENT_ITEM_KEY) {
1616                 struct btrfs_tree_block_info *info;
1617
1618                 info = (struct btrfs_tree_block_info *)(ei + 1);
1619                 *out_level = btrfs_tree_block_level(eb, info);
1620         } else {
1621                 ASSERT(key->type == BTRFS_METADATA_ITEM_KEY);
1622                 *out_level = (u8)key->offset;
1623         }
1624
1625         if (ret == 1)
1626                 *ptr = (unsigned long)-1;
1627
1628         return 0;
1629 }
1630
1631 static int iterate_leaf_refs(struct extent_inode_elem *inode_list,
1632                                 u64 root, u64 extent_item_objectid,
1633                                 iterate_extent_inodes_t *iterate, void *ctx)
1634 {
1635         struct extent_inode_elem *eie;
1636         int ret = 0;
1637
1638         for (eie = inode_list; eie; eie = eie->next) {
1639                 pr_debug("ref for %llu resolved, key (%llu EXTEND_DATA %llu), "
1640                          "root %llu\n", extent_item_objectid,
1641                          eie->inum, eie->offset, root);
1642                 ret = iterate(eie->inum, eie->offset, root, ctx);
1643                 if (ret) {
1644                         pr_debug("stopping iteration for %llu due to ret=%d\n",
1645                                  extent_item_objectid, ret);
1646                         break;
1647                 }
1648         }
1649
1650         return ret;
1651 }
1652
1653 /*
1654  * calls iterate() for every inode that references the extent identified by
1655  * the given parameters.
1656  * when the iterator function returns a non-zero value, iteration stops.
1657  */
1658 int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
1659                                 u64 extent_item_objectid, u64 extent_item_pos,
1660                                 int search_commit_root,
1661                                 iterate_extent_inodes_t *iterate, void *ctx)
1662 {
1663         int ret;
1664         struct btrfs_trans_handle *trans = NULL;
1665         struct ulist *refs = NULL;
1666         struct ulist *roots = NULL;
1667         struct ulist_node *ref_node = NULL;
1668         struct ulist_node *root_node = NULL;
1669         struct seq_list tree_mod_seq_elem = SEQ_LIST_INIT(tree_mod_seq_elem);
1670         struct ulist_iterator ref_uiter;
1671         struct ulist_iterator root_uiter;
1672
1673         pr_debug("resolving all inodes for extent %llu\n",
1674                         extent_item_objectid);
1675
1676         if (!search_commit_root) {
1677                 trans = btrfs_join_transaction(fs_info->extent_root);
1678                 if (IS_ERR(trans))
1679                         return PTR_ERR(trans);
1680                 btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1681         } else {
1682                 down_read(&fs_info->commit_root_sem);
1683         }
1684
1685         ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid,
1686                                    tree_mod_seq_elem.seq, &refs,
1687                                    &extent_item_pos);
1688         if (ret)
1689                 goto out;
1690
1691         ULIST_ITER_INIT(&ref_uiter);
1692         while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) {
1693                 ret = __btrfs_find_all_roots(trans, fs_info, ref_node->val,
1694                                              tree_mod_seq_elem.seq, &roots);
1695                 if (ret)
1696                         break;
1697                 ULIST_ITER_INIT(&root_uiter);
1698                 while (!ret && (root_node = ulist_next(roots, &root_uiter))) {
1699                         pr_debug("root %llu references leaf %llu, data list "
1700                                  "%#llx\n", root_node->val, ref_node->val,
1701                                  ref_node->aux);
1702                         ret = iterate_leaf_refs((struct extent_inode_elem *)
1703                                                 (uintptr_t)ref_node->aux,
1704                                                 root_node->val,
1705                                                 extent_item_objectid,
1706                                                 iterate, ctx);
1707                 }
1708                 ulist_free(roots);
1709         }
1710
1711         free_leaf_list(refs);
1712 out:
1713         if (!search_commit_root) {
1714                 btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1715                 btrfs_end_transaction(trans, fs_info->extent_root);
1716         } else {
1717                 up_read(&fs_info->commit_root_sem);
1718         }
1719
1720         return ret;
1721 }
1722
1723 int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
1724                                 struct btrfs_path *path,
1725                                 iterate_extent_inodes_t *iterate, void *ctx)
1726 {
1727         int ret;
1728         u64 extent_item_pos;
1729         u64 flags = 0;
1730         struct btrfs_key found_key;
1731         int search_commit_root = path->search_commit_root;
1732
1733         ret = extent_from_logical(fs_info, logical, path, &found_key, &flags);
1734         btrfs_release_path(path);
1735         if (ret < 0)
1736                 return ret;
1737         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1738                 return -EINVAL;
1739
1740         extent_item_pos = logical - found_key.objectid;
1741         ret = iterate_extent_inodes(fs_info, found_key.objectid,
1742                                         extent_item_pos, search_commit_root,
1743                                         iterate, ctx);
1744
1745         return ret;
1746 }
1747
1748 typedef int (iterate_irefs_t)(u64 parent, u32 name_len, unsigned long name_off,
1749                               struct extent_buffer *eb, void *ctx);
1750
1751 static int iterate_inode_refs(u64 inum, struct btrfs_root *fs_root,
1752                               struct btrfs_path *path,
1753                               iterate_irefs_t *iterate, void *ctx)
1754 {
1755         int ret = 0;
1756         int slot;
1757         u32 cur;
1758         u32 len;
1759         u32 name_len;
1760         u64 parent = 0;
1761         int found = 0;
1762         struct extent_buffer *eb;
1763         struct btrfs_item *item;
1764         struct btrfs_inode_ref *iref;
1765         struct btrfs_key found_key;
1766
1767         while (!ret) {
1768                 ret = btrfs_find_item(fs_root, path, inum,
1769                                 parent ? parent + 1 : 0, BTRFS_INODE_REF_KEY,
1770                                 &found_key);
1771
1772                 if (ret < 0)
1773                         break;
1774                 if (ret) {
1775                         ret = found ? 0 : -ENOENT;
1776                         break;
1777                 }
1778                 ++found;
1779
1780                 parent = found_key.offset;
1781                 slot = path->slots[0];
1782                 eb = btrfs_clone_extent_buffer(path->nodes[0]);
1783                 if (!eb) {
1784                         ret = -ENOMEM;
1785                         break;
1786                 }
1787                 extent_buffer_get(eb);
1788                 btrfs_tree_read_lock(eb);
1789                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1790                 btrfs_release_path(path);
1791
1792                 item = btrfs_item_nr(slot);
1793                 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1794
1795                 for (cur = 0; cur < btrfs_item_size(eb, item); cur += len) {
1796                         name_len = btrfs_inode_ref_name_len(eb, iref);
1797                         /* path must be released before calling iterate()! */
1798                         pr_debug("following ref at offset %u for inode %llu in "
1799                                  "tree %llu\n", cur, found_key.objectid,
1800                                  fs_root->objectid);
1801                         ret = iterate(parent, name_len,
1802                                       (unsigned long)(iref + 1), eb, ctx);
1803                         if (ret)
1804                                 break;
1805                         len = sizeof(*iref) + name_len;
1806                         iref = (struct btrfs_inode_ref *)((char *)iref + len);
1807                 }
1808                 btrfs_tree_read_unlock_blocking(eb);
1809                 free_extent_buffer(eb);
1810         }
1811
1812         btrfs_release_path(path);
1813
1814         return ret;
1815 }
1816
1817 static int iterate_inode_extrefs(u64 inum, struct btrfs_root *fs_root,
1818                                  struct btrfs_path *path,
1819                                  iterate_irefs_t *iterate, void *ctx)
1820 {
1821         int ret;
1822         int slot;
1823         u64 offset = 0;
1824         u64 parent;
1825         int found = 0;
1826         struct extent_buffer *eb;
1827         struct btrfs_inode_extref *extref;
1828         u32 item_size;
1829         u32 cur_offset;
1830         unsigned long ptr;
1831
1832         while (1) {
1833                 ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref,
1834                                             &offset);
1835                 if (ret < 0)
1836                         break;
1837                 if (ret) {
1838                         ret = found ? 0 : -ENOENT;
1839                         break;
1840                 }
1841                 ++found;
1842
1843                 slot = path->slots[0];
1844                 eb = btrfs_clone_extent_buffer(path->nodes[0]);
1845                 if (!eb) {
1846                         ret = -ENOMEM;
1847                         break;
1848                 }
1849                 extent_buffer_get(eb);
1850
1851                 btrfs_tree_read_lock(eb);
1852                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1853                 btrfs_release_path(path);
1854
1855                 item_size = btrfs_item_size_nr(eb, slot);
1856                 ptr = btrfs_item_ptr_offset(eb, slot);
1857                 cur_offset = 0;
1858
1859                 while (cur_offset < item_size) {
1860                         u32 name_len;
1861
1862                         extref = (struct btrfs_inode_extref *)(ptr + cur_offset);
1863                         parent = btrfs_inode_extref_parent(eb, extref);
1864                         name_len = btrfs_inode_extref_name_len(eb, extref);
1865                         ret = iterate(parent, name_len,
1866                                       (unsigned long)&extref->name, eb, ctx);
1867                         if (ret)
1868                                 break;
1869
1870                         cur_offset += btrfs_inode_extref_name_len(eb, extref);
1871                         cur_offset += sizeof(*extref);
1872                 }
1873                 btrfs_tree_read_unlock_blocking(eb);
1874                 free_extent_buffer(eb);
1875
1876                 offset++;
1877         }
1878
1879         btrfs_release_path(path);
1880
1881         return ret;
1882 }
1883
1884 static int iterate_irefs(u64 inum, struct btrfs_root *fs_root,
1885                          struct btrfs_path *path, iterate_irefs_t *iterate,
1886                          void *ctx)
1887 {
1888         int ret;
1889         int found_refs = 0;
1890
1891         ret = iterate_inode_refs(inum, fs_root, path, iterate, ctx);
1892         if (!ret)
1893                 ++found_refs;
1894         else if (ret != -ENOENT)
1895                 return ret;
1896
1897         ret = iterate_inode_extrefs(inum, fs_root, path, iterate, ctx);
1898         if (ret == -ENOENT && found_refs)
1899                 return 0;
1900
1901         return ret;
1902 }
1903
1904 /*
1905  * returns 0 if the path could be dumped (probably truncated)
1906  * returns <0 in case of an error
1907  */
1908 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
1909                          struct extent_buffer *eb, void *ctx)
1910 {
1911         struct inode_fs_paths *ipath = ctx;
1912         char *fspath;
1913         char *fspath_min;
1914         int i = ipath->fspath->elem_cnt;
1915         const int s_ptr = sizeof(char *);
1916         u32 bytes_left;
1917
1918         bytes_left = ipath->fspath->bytes_left > s_ptr ?
1919                                         ipath->fspath->bytes_left - s_ptr : 0;
1920
1921         fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
1922         fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len,
1923                                    name_off, eb, inum, fspath_min, bytes_left);
1924         if (IS_ERR(fspath))
1925                 return PTR_ERR(fspath);
1926
1927         if (fspath > fspath_min) {
1928                 ipath->fspath->val[i] = (u64)(unsigned long)fspath;
1929                 ++ipath->fspath->elem_cnt;
1930                 ipath->fspath->bytes_left = fspath - fspath_min;
1931         } else {
1932                 ++ipath->fspath->elem_missed;
1933                 ipath->fspath->bytes_missing += fspath_min - fspath;
1934                 ipath->fspath->bytes_left = 0;
1935         }
1936
1937         return 0;
1938 }
1939
1940 /*
1941  * this dumps all file system paths to the inode into the ipath struct, provided
1942  * is has been created large enough. each path is zero-terminated and accessed
1943  * from ipath->fspath->val[i].
1944  * when it returns, there are ipath->fspath->elem_cnt number of paths available
1945  * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the
1946  * number of missed paths in recored in ipath->fspath->elem_missed, otherwise,
1947  * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
1948  * have been needed to return all paths.
1949  */
1950 int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
1951 {
1952         return iterate_irefs(inum, ipath->fs_root, ipath->btrfs_path,
1953                              inode_to_path, ipath);
1954 }
1955
1956 struct btrfs_data_container *init_data_container(u32 total_bytes)
1957 {
1958         struct btrfs_data_container *data;
1959         size_t alloc_bytes;
1960
1961         alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
1962         data = vmalloc(alloc_bytes);
1963         if (!data)
1964                 return ERR_PTR(-ENOMEM);
1965
1966         if (total_bytes >= sizeof(*data)) {
1967                 data->bytes_left = total_bytes - sizeof(*data);
1968                 data->bytes_missing = 0;
1969         } else {
1970                 data->bytes_missing = sizeof(*data) - total_bytes;
1971                 data->bytes_left = 0;
1972         }
1973
1974         data->elem_cnt = 0;
1975         data->elem_missed = 0;
1976
1977         return data;
1978 }
1979
1980 /*
1981  * allocates space to return multiple file system paths for an inode.
1982  * total_bytes to allocate are passed, note that space usable for actual path
1983  * information will be total_bytes - sizeof(struct inode_fs_paths).
1984  * the returned pointer must be freed with free_ipath() in the end.
1985  */
1986 struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
1987                                         struct btrfs_path *path)
1988 {
1989         struct inode_fs_paths *ifp;
1990         struct btrfs_data_container *fspath;
1991
1992         fspath = init_data_container(total_bytes);
1993         if (IS_ERR(fspath))
1994                 return (void *)fspath;
1995
1996         ifp = kmalloc(sizeof(*ifp), GFP_NOFS);
1997         if (!ifp) {
1998                 kfree(fspath);
1999                 return ERR_PTR(-ENOMEM);
2000         }
2001
2002         ifp->btrfs_path = path;
2003         ifp->fspath = fspath;
2004         ifp->fs_root = fs_root;
2005
2006         return ifp;
2007 }
2008
2009 void free_ipath(struct inode_fs_paths *ipath)
2010 {
2011         if (!ipath)
2012                 return;
2013         vfree(ipath->fspath);
2014         kfree(ipath);
2015 }