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 struct extent_inode_elem {
  29         u64 inum;
  30         u64 offset;
  31         struct extent_inode_elem *next;
  32 };
  33
  34 static int check_extent_in_eb(struct btrfs_key *key, struct extent_buffer *eb,
  35                                 struct btrfs_file_extent_item *fi,
  36                                 u64 extent_item_pos,
  37                                 struct extent_inode_elem **eie)
  38 {
  39         u64 offset = 0;
  40         struct extent_inode_elem *e;
  41
  42         if (!btrfs_file_extent_compression(eb, fi) &&
  43             !btrfs_file_extent_encryption(eb, fi) &&
  44             !btrfs_file_extent_other_encoding(eb, fi)) {
  45                 u64 data_offset;
  46                 u64 data_len;
  47
  48                 data_offset = btrfs_file_extent_offset(eb, fi);
  49                 data_len = btrfs_file_extent_num_bytes(eb, fi);
  50
  51                 if (extent_item_pos < data_offset ||
  52                     extent_item_pos >= data_offset + data_len)
  53                         return 1;
  54                 offset = extent_item_pos - data_offset;
  55         }
  56
  57         e = kmalloc(sizeof(*e), GFP_NOFS);
  58         if (!e)
  59                 return -ENOMEM;
  60
  61         e->next = *eie;
  62         e->inum = key->objectid;
  63         e->offset = key->offset + offset;
  64         *eie = e;
  65
  66         return 0;
  67 }
  68
  69 static int find_extent_in_eb(struct extent_buffer *eb, u64 wanted_disk_byte,
  70                                 u64 extent_item_pos,
  71                                 struct extent_inode_elem **eie)
  72 {
  73         u64 disk_byte;
  74         struct btrfs_key key;
  75         struct btrfs_file_extent_item *fi;
  76         int slot;
  77         int nritems;
  78         int extent_type;
  79         int ret;
  80
  81         /*
  82          * from the shared data ref, we only have the leaf but we need
  83          * the key. thus, we must look into all items and see that we
  84          * find one (some) with a reference to our extent item.
  85          */
  86         nritems = btrfs_header_nritems(eb);
  87         for (slot = 0; slot < nritems; ++slot) {
  88                 btrfs_item_key_to_cpu(eb, &key, slot);
  89                 if (key.type != BTRFS_EXTENT_DATA_KEY)
  90                         continue;
  91                 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
  92                 extent_type = btrfs_file_extent_type(eb, fi);
  93                 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
  94                         continue;
  95                 /* don't skip BTRFS_FILE_EXTENT_PREALLOC, we can handle that */
  96                 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
  97                 if (disk_byte != wanted_disk_byte)
  98                         continue;
  99
 100                 ret = check_extent_in_eb(&key, eb, fi, extent_item_pos, eie);
 101                 if (ret < 0)
 102                         return ret;
 103         }
 104
 105         return 0;
 106 }
 107
 108 /*
 109  * this structure records all encountered refs on the way up to the root
 110  */
 111 struct __prelim_ref {
 112         struct list_head list;
 113         u64 root_id;
 114         struct btrfs_key key_for_search;
 115         int level;
 116         int count;
 117         struct extent_inode_elem *inode_list;
 118         u64 parent;
 119         u64 wanted_disk_byte;
 120 };
 121
 122 /*
 123  * the rules for all callers of this function are:
 124  * - obtaining the parent is the goal
 125  * - if you add a key, you must know that it is a correct key
 126  * - if you cannot add the parent or a correct key, then we will look into the
 127  *   block later to set a correct key
 128  *
 129  * delayed refs
 130  * ============
 131  *        backref type | shared | indirect | shared | indirect
 132  * information         |   tree |     tree |   data |     data
 133  * --------------------+--------+----------+--------+----------
 134  *      parent logical |    y   |     -    |    -   |     -
 135  *      key to resolve |    -   |     y    |    y   |     y
 136  *  tree block logical |    -   |     -    |    -   |     -
 137  *  root for resolving |    y   |     y    |    y   |     y
 138  *
 139  * - column 1:       we've the parent -> done
 140  * - column 2, 3, 4: we use the key to find the parent
 141  *
 142  * on disk refs (inline or keyed)
 143  * ==============================
 144  *        backref type | shared | indirect | shared | indirect
 145  * information         |   tree |     tree |   data |     data
 146  * --------------------+--------+----------+--------+----------
 147  *      parent logical |    y   |     -    |    y   |     -
 148  *      key to resolve |    -   |     -    |    -   |     y
 149  *  tree block logical |    y   |     y    |    y   |     y
 150  *  root for resolving |    -   |     y    |    y   |     y
 151  *
 152  * - column 1, 3: we've the parent -> done
 153  * - column 2:    we take the first key from the block to find the parent
 154  *                (see __add_missing_keys)
 155  * - column 4:    we use the key to find the parent
 156  *
 157  * additional information that's available but not required to find the parent
 158  * block might help in merging entries to gain some speed.
 159  */
 160
 161 static int __add_prelim_ref(struct list_head *head, u64 root_id,
 162                             struct btrfs_key *key, int level,
 163                             u64 parent, u64 wanted_disk_byte, int count)
 164 {
 165         struct __prelim_ref *ref;
 166
 167         /* in case we're adding delayed refs, we're holding the refs spinlock */
 168         ref = kmalloc(sizeof(*ref), GFP_ATOMIC);
 169         if (!ref)
 170                 return -ENOMEM;
 171
 172         ref->root_id = root_id;
 173         if (key)
 174                 ref->key_for_search = *key;
 175         else
 176                 memset(&ref->key_for_search, 0, sizeof(ref->key_for_search));
 177
 178         ref->inode_list = NULL;
 179         ref->level = level;
 180         ref->count = count;
 181         ref->parent = parent;
 182         ref->wanted_disk_byte = wanted_disk_byte;
 183         list_add_tail(&ref->list, head);
 184
 185         return 0;
 186 }
 187
 188 static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path,
 189                                 struct ulist *parents, int level,
 190                                 struct btrfs_key *key_for_search, u64 time_seq,
 191                                 u64 wanted_disk_byte,
 192                                 const u64 *extent_item_pos)
 193 {
 194         int ret = 0;
 195         int slot;
 196         struct extent_buffer *eb;
 197         struct btrfs_key key;
 198         struct btrfs_file_extent_item *fi;
 199         struct extent_inode_elem *eie = NULL;
 200         u64 disk_byte;
 201
 202         if (level != 0) {
 203                 eb = path->nodes[level];
 204                 ret = ulist_add(parents, eb->start, 0, GFP_NOFS);
 205                 if (ret < 0)
 206                         return ret;
 207                 return 0;
 208         }
 209
 210         /*
 211          * We normally enter this function with the path already pointing to
 212          * the first item to check. But sometimes, we may enter it with
 213          * slot==nritems. In that case, go to the next leaf before we continue.
 214          */
 215         if (path->slots[0] >= btrfs_header_nritems(path->nodes[0]))
 216                 ret = btrfs_next_old_leaf(root, path, time_seq);
 217
 218         while (!ret) {
 219                 eb = path->nodes[0];
 220                 slot = path->slots[0];
 221
 222                 btrfs_item_key_to_cpu(eb, &key, slot);
 223
 224                 if (key.objectid != key_for_search->objectid ||
 225                     key.type != BTRFS_EXTENT_DATA_KEY)
 226                         break;
 227
 228                 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
 229                 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
 230
 231                 if (disk_byte == wanted_disk_byte) {
 232                         eie = NULL;
 233                         if (extent_item_pos) {
 234                                 ret = check_extent_in_eb(&key, eb, fi,
 235                                                 *extent_item_pos,
 236                                                 &eie);
 237                                 if (ret < 0)
 238                                         break;
 239                         }
 240                         if (!ret) {
 241                                 ret = ulist_add(parents, eb->start,
 242                                                 (uintptr_t)eie, GFP_NOFS);
 243                                 if (ret < 0)
 244                                         break;
 245                                 if (!extent_item_pos) {
 246                                         ret = btrfs_next_old_leaf(root, path,
 247                                                         time_seq);
 248                                         continue;
 249                                 }
 250                         }
 251                 }
 252                 ret = btrfs_next_old_item(root, path, time_seq);
 253         }
 254
 255         if (ret > 0)
 256                 ret = 0;
 257         return ret;
 258 }
 259
 260 /*
 261  * resolve an indirect backref in the form (root_id, key, level)
 262  * to a logical address
 263  */
 264 static int __resolve_indirect_ref(struct btrfs_fs_info *fs_info,
 265                                   struct btrfs_path *path, u64 time_seq,
 266                                   struct __prelim_ref *ref,
 267                                   struct ulist *parents,
 268                                   const u64 *extent_item_pos)
 269 {
 270         struct btrfs_root *root;
 271         struct btrfs_key root_key;
 272         struct extent_buffer *eb;
 273         int ret = 0;
 274         int root_level;
 275         int level = ref->level;
 276
 277         root_key.objectid = ref->root_id;
 278         root_key.type = BTRFS_ROOT_ITEM_KEY;
 279         root_key.offset = (u64)-1;
 280         root = btrfs_read_fs_root_no_name(fs_info, &root_key);
 281         if (IS_ERR(root)) {
 282                 ret = PTR_ERR(root);
 283                 goto out;
 284         }
 285
 286         root_level = btrfs_old_root_level(root, time_seq);
 287
 288         if (root_level + 1 == level)
 289                 goto out;
 290
 291         path->lowest_level = level;
 292         ret = btrfs_search_old_slot(root, &ref->key_for_search, path, time_seq);
 293         pr_debug("search slot in root %llu (level %d, ref count %d) returned "
 294                  "%d for key (%llu %u %llu)\n",
 295                  (unsigned long long)ref->root_id, level, ref->count, ret,
 296                  (unsigned long long)ref->key_for_search.objectid,
 297                  ref->key_for_search.type,
 298                  (unsigned long long)ref->key_for_search.offset);
 299         if (ret < 0)
 300                 goto out;
 301
 302         eb = path->nodes[level];
 303         while (!eb) {
 304                 if (!level) {
 305                         WARN_ON(1);
 306                         ret = 1;
 307                         goto out;
 308                 }
 309                 level--;
 310                 eb = path->nodes[level];
 311         }
 312
 313         ret = add_all_parents(root, path, parents, level, &ref->key_for_search,
 314                                 time_seq, ref->wanted_disk_byte,
 315                                 extent_item_pos);
 316 out:
 317         path->lowest_level = 0;
 318         btrfs_release_path(path);
 319         return ret;
 320 }
 321
 322 /*
 323  * resolve all indirect backrefs from the list
 324  */
 325 static int __resolve_indirect_refs(struct btrfs_fs_info *fs_info,
 326                                    struct btrfs_path *path, u64 time_seq,
 327                                    struct list_head *head,
 328                                    const u64 *extent_item_pos)
 329 {
 330         int err;
 331         int ret = 0;
 332         struct __prelim_ref *ref;
 333         struct __prelim_ref *ref_safe;
 334         struct __prelim_ref *new_ref;
 335         struct ulist *parents;
 336         struct ulist_node *node;
 337         struct ulist_iterator uiter;
 338
 339         parents = ulist_alloc(GFP_NOFS);
 340         if (!parents)
 341                 return -ENOMEM;
 342
 343         /*
 344          * _safe allows us to insert directly after the current item without
 345          * iterating over the newly inserted items.
 346          * we're also allowed to re-assign ref during iteration.
 347          */
 348         list_for_each_entry_safe(ref, ref_safe, head, list) {
 349                 if (ref->parent)        /* already direct */
 350                         continue;
 351                 if (ref->count == 0)
 352                         continue;
 353                 err = __resolve_indirect_ref(fs_info, path, time_seq, ref,
 354                                              parents, extent_item_pos);
 355                 if (err == -ENOMEM)
 356                         goto out;
 357                 if (err)
 358                         continue;
 359
 360                 /* we put the first parent into the ref at hand */
 361                 ULIST_ITER_INIT(&uiter);
 362                 node = ulist_next(parents, &uiter);
 363                 ref->parent = node ? node->val : 0;
 364                 ref->inode_list = node ?
 365                         (struct extent_inode_elem *)(uintptr_t)node->aux : 0;
 366
 367                 /* additional parents require new refs being added here */
 368                 while ((node = ulist_next(parents, &uiter))) {
 369                         new_ref = kmalloc(sizeof(*new_ref), GFP_NOFS);
 370                         if (!new_ref) {
 371                                 ret = -ENOMEM;
 372                                 goto out;
 373                         }
 374                         memcpy(new_ref, ref, sizeof(*ref));
 375                         new_ref->parent = node->val;
 376                         new_ref->inode_list = (struct extent_inode_elem *)
 377                                                         (uintptr_t)node->aux;
 378                         list_add(&new_ref->list, &ref->list);
 379                 }
 380                 ulist_reinit(parents);
 381         }
 382 out:
 383         ulist_free(parents);
 384         return ret;
 385 }
 386
 387 static inline int ref_for_same_block(struct __prelim_ref *ref1,
 388                                      struct __prelim_ref *ref2)
 389 {
 390         if (ref1->level != ref2->level)
 391                 return 0;
 392         if (ref1->root_id != ref2->root_id)
 393                 return 0;
 394         if (ref1->key_for_search.type != ref2->key_for_search.type)
 395                 return 0;
 396         if (ref1->key_for_search.objectid != ref2->key_for_search.objectid)
 397                 return 0;
 398         if (ref1->key_for_search.offset != ref2->key_for_search.offset)
 399                 return 0;
 400         if (ref1->parent != ref2->parent)
 401                 return 0;
 402
 403         return 1;
 404 }
 405
 406 /*
 407  * read tree blocks and add keys where required.
 408  */
 409 static int __add_missing_keys(struct btrfs_fs_info *fs_info,
 410                               struct list_head *head)
 411 {
 412         struct list_head *pos;
 413         struct extent_buffer *eb;
 414
 415         list_for_each(pos, head) {
 416                 struct __prelim_ref *ref;
 417                 ref = list_entry(pos, struct __prelim_ref, list);
 418
 419                 if (ref->parent)
 420                         continue;
 421                 if (ref->key_for_search.type)
 422                         continue;
 423                 BUG_ON(!ref->wanted_disk_byte);
 424                 eb = read_tree_block(fs_info->tree_root, ref->wanted_disk_byte,
 425                                      fs_info->tree_root->leafsize, 0);
 426                 if (!eb || !extent_buffer_uptodate(eb)) {
 427                         free_extent_buffer(eb);
 428                         return -EIO;
 429                 }
 430                 btrfs_tree_read_lock(eb);
 431                 if (btrfs_header_level(eb) == 0)
 432                         btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
 433                 else
 434                         btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
 435                 btrfs_tree_read_unlock(eb);
 436                 free_extent_buffer(eb);
 437         }
 438         return 0;
 439 }
 440
 441 /*
 442  * merge two lists of backrefs and adjust counts accordingly
 443  *
 444  * mode = 1: merge identical keys, if key is set
 445  *    FIXME: if we add more keys in __add_prelim_ref, we can merge more here.
 446  *           additionally, we could even add a key range for the blocks we
 447  *           looked into to merge even more (-> replace unresolved refs by those
 448  *           having a parent).
 449  * mode = 2: merge identical parents
 450  */
 451 static void __merge_refs(struct list_head *head, int mode)
 452 {
 453         struct list_head *pos1;
 454
 455         list_for_each(pos1, head) {
 456                 struct list_head *n2;
 457                 struct list_head *pos2;
 458                 struct __prelim_ref *ref1;
 459
 460                 ref1 = list_entry(pos1, struct __prelim_ref, list);
 461
 462                 for (pos2 = pos1->next, n2 = pos2->next; pos2 != head;
 463                      pos2 = n2, n2 = pos2->next) {
 464                         struct __prelim_ref *ref2;
 465                         struct __prelim_ref *xchg;
 466                         struct extent_inode_elem *eie;
 467
 468                         ref2 = list_entry(pos2, struct __prelim_ref, list);
 469
 470                         if (mode == 1) {
 471                                 if (!ref_for_same_block(ref1, ref2))
 472                                         continue;
 473                                 if (!ref1->parent && ref2->parent) {
 474                                         xchg = ref1;
 475                                         ref1 = ref2;
 476                                         ref2 = xchg;
 477                                 }
 478                         } else {
 479                                 if (ref1->parent != ref2->parent)
 480                                         continue;
 481                         }
 482
 483                         eie = ref1->inode_list;
 484                         while (eie && eie->next)
 485                                 eie = eie->next;
 486                         if (eie)
 487                                 eie->next = ref2->inode_list;
 488                         else
 489                                 ref1->inode_list = ref2->inode_list;
 490                         ref1->count += ref2->count;
 491
 492                         list_del(&ref2->list);
 493                         kfree(ref2);
 494                 }
 495
 496         }
 497 }
 498
 499 /*
 500  * add all currently queued delayed refs from this head whose seq nr is
 501  * smaller or equal that seq to the list
 502  */
 503 static int __add_delayed_refs(struct btrfs_delayed_ref_head *head, u64 seq,
 504                               struct list_head *prefs)
 505 {
 506         struct btrfs_delayed_extent_op *extent_op = head->extent_op;
 507         struct rb_node *n = &head->node.rb_node;
 508         struct btrfs_key key;
 509         struct btrfs_key op_key = {0};
 510         int sgn;
 511         int ret = 0;
 512
 513         if (extent_op && extent_op->update_key)
 514                 btrfs_disk_key_to_cpu(&op_key, &extent_op->key);
 515
 516         while ((n = rb_prev(n))) {
 517                 struct btrfs_delayed_ref_node *node;
 518                 node = rb_entry(n, struct btrfs_delayed_ref_node,
 519                                 rb_node);
 520                 if (node->bytenr != head->node.bytenr)
 521                         break;
 522                 WARN_ON(node->is_head);
 523
 524                 if (node->seq > seq)
 525                         continue;
 526
 527                 switch (node->action) {
 528                 case BTRFS_ADD_DELAYED_EXTENT:
 529                 case BTRFS_UPDATE_DELAYED_HEAD:
 530                         WARN_ON(1);
 531                         continue;
 532                 case BTRFS_ADD_DELAYED_REF:
 533                         sgn = 1;
 534                         break;
 535                 case BTRFS_DROP_DELAYED_REF:
 536                         sgn = -1;
 537                         break;
 538                 default:
 539                         BUG_ON(1);
 540                 }
 541                 switch (node->type) {
 542                 case BTRFS_TREE_BLOCK_REF_KEY: {
 543                         struct btrfs_delayed_tree_ref *ref;
 544
 545                         ref = btrfs_delayed_node_to_tree_ref(node);
 546                         ret = __add_prelim_ref(prefs, ref->root, &op_key,
 547                                                ref->level + 1, 0, node->bytenr,
 548                                                node->ref_mod * sgn);
 549                         break;
 550                 }
 551                 case BTRFS_SHARED_BLOCK_REF_KEY: {
 552                         struct btrfs_delayed_tree_ref *ref;
 553
 554                         ref = btrfs_delayed_node_to_tree_ref(node);
 555                         ret = __add_prelim_ref(prefs, ref->root, NULL,
 556                                                ref->level + 1, ref->parent,
 557                                                node->bytenr,
 558                                                node->ref_mod * sgn);
 559                         break;
 560                 }
 561                 case BTRFS_EXTENT_DATA_REF_KEY: {
 562                         struct btrfs_delayed_data_ref *ref;
 563                         ref = btrfs_delayed_node_to_data_ref(node);
 564
 565                         key.objectid = ref->objectid;
 566                         key.type = BTRFS_EXTENT_DATA_KEY;
 567                         key.offset = ref->offset;
 568                         ret = __add_prelim_ref(prefs, ref->root, &key, 0, 0,
 569                                                node->bytenr,
 570                                                node->ref_mod * sgn);
 571                         break;
 572                 }
 573                 case BTRFS_SHARED_DATA_REF_KEY: {
 574                         struct btrfs_delayed_data_ref *ref;
 575
 576                         ref = btrfs_delayed_node_to_data_ref(node);
 577
 578                         key.objectid = ref->objectid;
 579                         key.type = BTRFS_EXTENT_DATA_KEY;
 580                         key.offset = ref->offset;
 581                         ret = __add_prelim_ref(prefs, ref->root, &key, 0,
 582                                                ref->parent, node->bytenr,
 583                                                node->ref_mod * sgn);
 584                         break;
 585                 }
 586                 default:
 587                         WARN_ON(1);
 588                 }
 589                 if (ret)
 590                         return ret;
 591         }
 592
 593         return 0;
 594 }
 595
 596 /*
 597  * add all inline backrefs for bytenr to the list
 598  */
 599 static int __add_inline_refs(struct btrfs_fs_info *fs_info,
 600                              struct btrfs_path *path, u64 bytenr,
 601                              int *info_level, struct list_head *prefs)
 602 {
 603         int ret = 0;
 604         int slot;
 605         struct extent_buffer *leaf;
 606         struct btrfs_key key;
 607         struct btrfs_key found_key;
 608         unsigned long ptr;
 609         unsigned long end;
 610         struct btrfs_extent_item *ei;
 611         u64 flags;
 612         u64 item_size;
 613
 614         /*
 615          * enumerate all inline refs
 616          */
 617         leaf = path->nodes[0];
 618         slot = path->slots[0];
 619
 620         item_size = btrfs_item_size_nr(leaf, slot);
 621         BUG_ON(item_size < sizeof(*ei));
 622
 623         ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
 624         flags = btrfs_extent_flags(leaf, ei);
 625         btrfs_item_key_to_cpu(leaf, &found_key, slot);
 626
 627         ptr = (unsigned long)(ei + 1);
 628         end = (unsigned long)ei + item_size;
 629
 630         if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
 631             flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
 632                 struct btrfs_tree_block_info *info;
 633
 634                 info = (struct btrfs_tree_block_info *)ptr;
 635                 *info_level = btrfs_tree_block_level(leaf, info);
 636                 ptr += sizeof(struct btrfs_tree_block_info);
 637                 BUG_ON(ptr > end);
 638         } else if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
 639                 *info_level = found_key.offset;
 640         } else {
 641                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
 642         }
 643
 644         while (ptr < end) {
 645                 struct btrfs_extent_inline_ref *iref;
 646                 u64 offset;
 647                 int type;
 648
 649                 iref = (struct btrfs_extent_inline_ref *)ptr;
 650                 type = btrfs_extent_inline_ref_type(leaf, iref);
 651                 offset = btrfs_extent_inline_ref_offset(leaf, iref);
 652
 653                 switch (type) {
 654                 case BTRFS_SHARED_BLOCK_REF_KEY:
 655                         ret = __add_prelim_ref(prefs, 0, NULL,
 656                                                 *info_level + 1, offset,
 657                                                 bytenr, 1);
 658                         break;
 659                 case BTRFS_SHARED_DATA_REF_KEY: {
 660                         struct btrfs_shared_data_ref *sdref;
 661                         int count;
 662
 663                         sdref = (struct btrfs_shared_data_ref *)(iref + 1);
 664                         count = btrfs_shared_data_ref_count(leaf, sdref);
 665                         ret = __add_prelim_ref(prefs, 0, NULL, 0, offset,
 666                                                bytenr, count);
 667                         break;
 668                 }
 669                 case BTRFS_TREE_BLOCK_REF_KEY:
 670                         ret = __add_prelim_ref(prefs, offset, NULL,
 671                                                *info_level + 1, 0,
 672                                                bytenr, 1);
 673                         break;
 674                 case BTRFS_EXTENT_DATA_REF_KEY: {
 675                         struct btrfs_extent_data_ref *dref;
 676                         int count;
 677                         u64 root;
 678
 679                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
 680                         count = btrfs_extent_data_ref_count(leaf, dref);
 681                         key.objectid = btrfs_extent_data_ref_objectid(leaf,
 682                                                                       dref);
 683                         key.type = BTRFS_EXTENT_DATA_KEY;
 684                         key.offset = btrfs_extent_data_ref_offset(leaf, dref);
 685                         root = btrfs_extent_data_ref_root(leaf, dref);
 686                         ret = __add_prelim_ref(prefs, root, &key, 0, 0,
 687                                                bytenr, count);
 688                         break;
 689                 }
 690                 default:
 691                         WARN_ON(1);
 692                 }
 693                 if (ret)
 694                         return ret;
 695                 ptr += btrfs_extent_inline_ref_size(type);
 696         }
 697
 698         return 0;
 699 }
 700
 701 /*
 702  * add all non-inline backrefs for bytenr to the list
 703  */
 704 static int __add_keyed_refs(struct btrfs_fs_info *fs_info,
 705                             struct btrfs_path *path, u64 bytenr,
 706                             int info_level, struct list_head *prefs)
 707 {
 708         struct btrfs_root *extent_root = fs_info->extent_root;
 709         int ret;
 710         int slot;
 711         struct extent_buffer *leaf;
 712         struct btrfs_key key;
 713
 714         while (1) {
 715                 ret = btrfs_next_item(extent_root, path);
 716                 if (ret < 0)
 717                         break;
 718                 if (ret) {
 719                         ret = 0;
 720                         break;
 721                 }
 722
 723                 slot = path->slots[0];
 724                 leaf = path->nodes[0];
 725                 btrfs_item_key_to_cpu(leaf, &key, slot);
 726
 727                 if (key.objectid != bytenr)
 728                         break;
 729                 if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
 730                         continue;
 731                 if (key.type > BTRFS_SHARED_DATA_REF_KEY)
 732                         break;
 733
 734                 switch (key.type) {
 735                 case BTRFS_SHARED_BLOCK_REF_KEY:
 736                         ret = __add_prelim_ref(prefs, 0, NULL,
 737                                                 info_level + 1, key.offset,
 738                                                 bytenr, 1);
 739                         break;
 740                 case BTRFS_SHARED_DATA_REF_KEY: {
 741                         struct btrfs_shared_data_ref *sdref;
 742                         int count;
 743
 744                         sdref = btrfs_item_ptr(leaf, slot,
 745                                               struct btrfs_shared_data_ref);
 746                         count = btrfs_shared_data_ref_count(leaf, sdref);
 747                         ret = __add_prelim_ref(prefs, 0, NULL, 0, key.offset,
 748                                                 bytenr, count);
 749                         break;
 750                 }
 751                 case BTRFS_TREE_BLOCK_REF_KEY:
 752                         ret = __add_prelim_ref(prefs, key.offset, NULL,
 753                                                info_level + 1, 0,
 754                                                bytenr, 1);
 755                         break;
 756                 case BTRFS_EXTENT_DATA_REF_KEY: {
 757                         struct btrfs_extent_data_ref *dref;
 758                         int count;
 759                         u64 root;
 760
 761                         dref = btrfs_item_ptr(leaf, slot,
 762                                               struct btrfs_extent_data_ref);
 763                         count = btrfs_extent_data_ref_count(leaf, dref);
 764                         key.objectid = btrfs_extent_data_ref_objectid(leaf,
 765                                                                       dref);
 766                         key.type = BTRFS_EXTENT_DATA_KEY;
 767                         key.offset = btrfs_extent_data_ref_offset(leaf, dref);
 768                         root = btrfs_extent_data_ref_root(leaf, dref);
 769                         ret = __add_prelim_ref(prefs, root, &key, 0, 0,
 770                                                bytenr, count);
 771                         break;
 772                 }
 773                 default:
 774                         WARN_ON(1);
 775                 }
 776                 if (ret)
 777                         return ret;
 778
 779         }
 780
 781         return ret;
 782 }
 783
 784 /*
 785  * this adds all existing backrefs (inline backrefs, backrefs and delayed
 786  * refs) for the given bytenr to the refs list, merges duplicates and resolves
 787  * indirect refs to their parent bytenr.
 788  * When roots are found, they're added to the roots list
 789  *
 790  * FIXME some caching might speed things up
 791  */
 792 static int find_parent_nodes(struct btrfs_trans_handle *trans,
 793                              struct btrfs_fs_info *fs_info, u64 bytenr,
 794                              u64 time_seq, struct ulist *refs,
 795                              struct ulist *roots, const u64 *extent_item_pos)
 796 {
 797         struct btrfs_key key;
 798         struct btrfs_path *path;
 799         struct btrfs_delayed_ref_root *delayed_refs = NULL;
 800         struct btrfs_delayed_ref_head *head;
 801         int info_level = 0;
 802         int ret;
 803         struct list_head prefs_delayed;
 804         struct list_head prefs;
 805         struct __prelim_ref *ref;
 806
 807         INIT_LIST_HEAD(&prefs);
 808         INIT_LIST_HEAD(&prefs_delayed);
 809
 810         key.objectid = bytenr;
 811         key.offset = (u64)-1;
 812         if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
 813                 key.type = BTRFS_METADATA_ITEM_KEY;
 814         else
 815                 key.type = BTRFS_EXTENT_ITEM_KEY;
 816
 817         path = btrfs_alloc_path();
 818         if (!path)
 819                 return -ENOMEM;
 820         if (!trans)
 821                 path->search_commit_root = 1;
 822
 823         /*
 824          * grab both a lock on the path and a lock on the delayed ref head.
 825          * We need both to get a consistent picture of how the refs look
 826          * at a specified point in time
 827          */
 828 again:
 829         head = NULL;
 830
 831         ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
 832         if (ret < 0)
 833                 goto out;
 834         BUG_ON(ret == 0);
 835
 836         if (trans) {
 837                 /*
 838                  * look if there are updates for this ref queued and lock the
 839                  * head
 840                  */
 841                 delayed_refs = &trans->transaction->delayed_refs;
 842                 spin_lock(&delayed_refs->lock);
 843                 head = btrfs_find_delayed_ref_head(trans, bytenr);
 844                 if (head) {
 845                         if (!mutex_trylock(&head->mutex)) {
 846                                 atomic_inc(&head->node.refs);
 847                                 spin_unlock(&delayed_refs->lock);
 848
 849                                 btrfs_release_path(path);
 850
 851                                 /*
 852                                  * Mutex was contended, block until it's
 853                                  * released and try again
 854                                  */
 855                                 mutex_lock(&head->mutex);
 856                                 mutex_unlock(&head->mutex);
 857                                 btrfs_put_delayed_ref(&head->node);
 858                                 goto again;
 859                         }
 860                         ret = __add_delayed_refs(head, time_seq,
 861                                                  &prefs_delayed);
 862                         mutex_unlock(&head->mutex);
 863                         if (ret) {
 864                                 spin_unlock(&delayed_refs->lock);
 865                                 goto out;
 866                         }
 867                 }
 868                 spin_unlock(&delayed_refs->lock);
 869         }
 870
 871         if (path->slots[0]) {
 872                 struct extent_buffer *leaf;
 873                 int slot;
 874
 875                 path->slots[0]--;
 876                 leaf = path->nodes[0];
 877                 slot = path->slots[0];
 878                 btrfs_item_key_to_cpu(leaf, &key, slot);
 879                 if (key.objectid == bytenr &&
 880                     (key.type == BTRFS_EXTENT_ITEM_KEY ||
 881                      key.type == BTRFS_METADATA_ITEM_KEY)) {
 882                         ret = __add_inline_refs(fs_info, path, bytenr,
 883                                                 &info_level, &prefs);
 884                         if (ret)
 885                                 goto out;
 886                         ret = __add_keyed_refs(fs_info, path, bytenr,
 887                                                info_level, &prefs);
 888                         if (ret)
 889                                 goto out;
 890                 }
 891         }
 892         btrfs_release_path(path);
 893
 894         list_splice_init(&prefs_delayed, &prefs);
 895
 896         ret = __add_missing_keys(fs_info, &prefs);
 897         if (ret)
 898                 goto out;
 899
 900         __merge_refs(&prefs, 1);
 901
 902         ret = __resolve_indirect_refs(fs_info, path, time_seq, &prefs,
 903                                       extent_item_pos);
 904         if (ret)
 905                 goto out;
 906
 907         __merge_refs(&prefs, 2);
 908
 909         while (!list_empty(&prefs)) {
 910                 ref = list_first_entry(&prefs, struct __prelim_ref, list);
 911                 list_del(&ref->list);
 912                 WARN_ON(ref->count < 0);
 913                 if (ref->count && ref->root_id && ref->parent == 0) {
 914                         /* no parent == root of tree */
 915                         ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS);
 916                         if (ret < 0)
 917                                 goto out;
 918                 }
 919                 if (ref->count && ref->parent) {
 920                         struct extent_inode_elem *eie = NULL;
 921                         if (extent_item_pos && !ref->inode_list) {
 922                                 u32 bsz;
 923                                 struct extent_buffer *eb;
 924                                 bsz = btrfs_level_size(fs_info->extent_root,
 925                                                         info_level);
 926                                 eb = read_tree_block(fs_info->extent_root,
 927                                                            ref->parent, bsz, 0);
 928                                 if (!eb || !extent_buffer_uptodate(eb)) {
 929                                         free_extent_buffer(eb);
 930                                         ret = -EIO;
 931                                         goto out;
 932                                 }
 933                                 ret = find_extent_in_eb(eb, bytenr,
 934                                                         *extent_item_pos, &eie);
 935                                 ref->inode_list = eie;
 936                                 free_extent_buffer(eb);
 937                         }
 938                         ret = ulist_add_merge(refs, ref->parent,
 939                                               (uintptr_t)ref->inode_list,
 940                                               (u64 *)&eie, GFP_NOFS);
 941                         if (ret < 0)
 942                                 goto out;
 943                         if (!ret && extent_item_pos) {
 944                                 /*
 945                                  * we've recorded that parent, so we must extend
 946                                  * its inode list here
 947                                  */
 948                                 BUG_ON(!eie);
 949                                 while (eie->next)
 950                                         eie = eie->next;
 951                                 eie->next = ref->inode_list;
 952                         }
 953                 }
 954                 kfree(ref);
 955         }
 956
 957 out:
 958         btrfs_free_path(path);
 959         while (!list_empty(&prefs)) {
 960                 ref = list_first_entry(&prefs, struct __prelim_ref, list);
 961                 list_del(&ref->list);
 962                 kfree(ref);
 963         }
 964         while (!list_empty(&prefs_delayed)) {
 965                 ref = list_first_entry(&prefs_delayed, struct __prelim_ref,
 966                                        list);
 967                 list_del(&ref->list);
 968                 kfree(ref);
 969         }
 970
 971         return ret;
 972 }
 973
 974 static void free_leaf_list(struct ulist *blocks)
 975 {
 976         struct ulist_node *node = NULL;
 977         struct extent_inode_elem *eie;
 978         struct extent_inode_elem *eie_next;
 979         struct ulist_iterator uiter;
 980
 981         ULIST_ITER_INIT(&uiter);
 982         while ((node = ulist_next(blocks, &uiter))) {
 983                 if (!node->aux)
 984                         continue;
 985                 eie = (struct extent_inode_elem *)(uintptr_t)node->aux;
 986                 for (; eie; eie = eie_next) {
 987                         eie_next = eie->next;
 988                         kfree(eie);
 989                 }
 990                 node->aux = 0;
 991         }
 992
 993         ulist_free(blocks);
 994 }
 995
 996 /*
 997  * Finds all leafs with a reference to the specified combination of bytenr and
 998  * offset. key_list_head will point to a list of corresponding keys (caller must
 999  * free each list element). The leafs will be stored in the leafs ulist, which
1000  * must be freed with ulist_free.
1001  *
1002  * returns 0 on success, <0 on error
1003  */
1004 static int btrfs_find_all_leafs(struct btrfs_trans_handle *trans,
1005                                 struct btrfs_fs_info *fs_info, u64 bytenr,
1006                                 u64 time_seq, struct ulist **leafs,
1007                                 const u64 *extent_item_pos)
1008 {
1009         struct ulist *tmp;
1010         int ret;
1011
1012         tmp = ulist_alloc(GFP_NOFS);
1013         if (!tmp)
1014                 return -ENOMEM;
1015         *leafs = ulist_alloc(GFP_NOFS);
1016         if (!*leafs) {
1017                 ulist_free(tmp);
1018                 return -ENOMEM;
1019         }
1020
1021         ret = find_parent_nodes(trans, fs_info, bytenr,
1022                                 time_seq, *leafs, tmp, extent_item_pos);
1023         ulist_free(tmp);
1024
1025         if (ret < 0 && ret != -ENOENT) {
1026                 free_leaf_list(*leafs);
1027                 return ret;
1028         }
1029
1030         return 0;
1031 }
1032
1033 /*
1034  * walk all backrefs for a given extent to find all roots that reference this
1035  * extent. Walking a backref means finding all extents that reference this
1036  * extent and in turn walk the backrefs of those, too. Naturally this is a
1037  * recursive process, but here it is implemented in an iterative fashion: We
1038  * find all referencing extents for the extent in question and put them on a
1039  * list. In turn, we find all referencing extents for those, further appending
1040  * to the list. The way we iterate the list allows adding more elements after
1041  * the current while iterating. The process stops when we reach the end of the
1042  * list. Found roots are added to the roots list.
1043  *
1044  * returns 0 on success, < 0 on error.
1045  */
1046 int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1047                                 struct btrfs_fs_info *fs_info, u64 bytenr,
1048                                 u64 time_seq, struct ulist **roots)
1049 {
1050         struct ulist *tmp;
1051         struct ulist_node *node = NULL;
1052         struct ulist_iterator uiter;
1053         int ret;
1054
1055         tmp = ulist_alloc(GFP_NOFS);
1056         if (!tmp)
1057                 return -ENOMEM;
1058         *roots = ulist_alloc(GFP_NOFS);
1059         if (!*roots) {
1060                 ulist_free(tmp);
1061                 return -ENOMEM;
1062         }
1063
1064         ULIST_ITER_INIT(&uiter);
1065         while (1) {
1066                 ret = find_parent_nodes(trans, fs_info, bytenr,
1067                                         time_seq, tmp, *roots, NULL);
1068                 if (ret < 0 && ret != -ENOENT) {
1069                         ulist_free(tmp);
1070                         ulist_free(*roots);
1071                         return ret;
1072                 }
1073                 node = ulist_next(tmp, &uiter);
1074                 if (!node)
1075                         break;
1076                 bytenr = node->val;
1077         }
1078
1079         ulist_free(tmp);
1080         return 0;
1081 }
1082
1083
1084 static int __inode_info(u64 inum, u64 ioff, u8 key_type,
1085                         struct btrfs_root *fs_root, struct btrfs_path *path,
1086                         struct btrfs_key *found_key)
1087 {
1088         int ret;
1089         struct btrfs_key key;
1090         struct extent_buffer *eb;
1091
1092         key.type = key_type;
1093         key.objectid = inum;
1094         key.offset = ioff;
1095
1096         ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
1097         if (ret < 0)
1098                 return ret;
1099
1100         eb = path->nodes[0];
1101         if (ret && path->slots[0] >= btrfs_header_nritems(eb)) {
1102                 ret = btrfs_next_leaf(fs_root, path);
1103                 if (ret)
1104                         return ret;
1105                 eb = path->nodes[0];
1106         }
1107
1108         btrfs_item_key_to_cpu(eb, found_key, path->slots[0]);
1109         if (found_key->type != key.type || found_key->objectid != key.objectid)
1110                 return 1;
1111
1112         return 0;
1113 }
1114
1115 /*
1116  * this makes the path point to (inum INODE_ITEM ioff)
1117  */
1118 int inode_item_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
1119                         struct btrfs_path *path)
1120 {
1121         struct btrfs_key key;
1122         return __inode_info(inum, ioff, BTRFS_INODE_ITEM_KEY, fs_root, path,
1123                                 &key);
1124 }
1125
1126 static int inode_ref_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
1127                                 struct btrfs_path *path,
1128                                 struct btrfs_key *found_key)
1129 {
1130         return __inode_info(inum, ioff, BTRFS_INODE_REF_KEY, fs_root, path,
1131                                 found_key);
1132 }
1133
1134 int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,
1135                           u64 start_off, struct btrfs_path *path,
1136                           struct btrfs_inode_extref **ret_extref,
1137                           u64 *found_off)
1138 {
1139         int ret, slot;
1140         struct btrfs_key key;
1141         struct btrfs_key found_key;
1142         struct btrfs_inode_extref *extref;
1143         struct extent_buffer *leaf;
1144         unsigned long ptr;
1145
1146         key.objectid = inode_objectid;
1147         btrfs_set_key_type(&key, BTRFS_INODE_EXTREF_KEY);
1148         key.offset = start_off;
1149
1150         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1151         if (ret < 0)
1152                 return ret;
1153
1154         while (1) {
1155                 leaf = path->nodes[0];
1156                 slot = path->slots[0];
1157                 if (slot >= btrfs_header_nritems(leaf)) {
1158                         /*
1159                          * If the item at offset is not found,
1160                          * btrfs_search_slot will point us to the slot
1161                          * where it should be inserted. In our case
1162                          * that will be the slot directly before the
1163                          * next INODE_REF_KEY_V2 item. In the case
1164                          * that we're pointing to the last slot in a
1165                          * leaf, we must move one leaf over.
1166                          */
1167                         ret = btrfs_next_leaf(root, path);
1168                         if (ret) {
1169                                 if (ret >= 1)
1170                                         ret = -ENOENT;
1171                                 break;
1172                         }
1173                         continue;
1174                 }
1175
1176                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1177
1178                 /*
1179                  * Check that we're still looking at an extended ref key for
1180                  * this particular objectid. If we have different
1181                  * objectid or type then there are no more to be found
1182                  * in the tree and we can exit.
1183                  */
1184                 ret = -ENOENT;
1185                 if (found_key.objectid != inode_objectid)
1186                         break;
1187                 if (btrfs_key_type(&found_key) != BTRFS_INODE_EXTREF_KEY)
1188                         break;
1189
1190                 ret = 0;
1191                 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1192                 extref = (struct btrfs_inode_extref *)ptr;
1193                 *ret_extref = extref;
1194                 if (found_off)
1195                         *found_off = found_key.offset;
1196                 break;
1197         }
1198
1199         return ret;
1200 }
1201
1202 /*
1203  * this iterates to turn a name (from iref/extref) into a full filesystem path.
1204  * Elements of the path are separated by '/' and the path is guaranteed to be
1205  * 0-terminated. the path is only given within the current file system.
1206  * Therefore, it never starts with a '/'. the caller is responsible to provide
1207  * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
1208  * the start point of the resulting string is returned. this pointer is within
1209  * dest, normally.
1210  * in case the path buffer would overflow, the pointer is decremented further
1211  * as if output was written to the buffer, though no more output is actually
1212  * generated. that way, the caller can determine how much space would be
1213  * required for the path to fit into the buffer. in that case, the returned
1214  * value will be smaller than dest. callers must check this!
1215  */
1216 char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
1217                         u32 name_len, unsigned long name_off,
1218                         struct extent_buffer *eb_in, u64 parent,
1219                         char *dest, u32 size)
1220 {
1221         int slot;
1222         u64 next_inum;
1223         int ret;
1224         s64 bytes_left = ((s64)size) - 1;
1225         struct extent_buffer *eb = eb_in;
1226         struct btrfs_key found_key;
1227         int leave_spinning = path->leave_spinning;
1228         struct btrfs_inode_ref *iref;
1229
1230         if (bytes_left >= 0)
1231                 dest[bytes_left] = '\0';
1232
1233         path->leave_spinning = 1;
1234         while (1) {
1235                 bytes_left -= name_len;
1236                 if (bytes_left >= 0)
1237                         read_extent_buffer(eb, dest + bytes_left,
1238                                            name_off, name_len);
1239                 if (eb != eb_in) {
1240                         btrfs_tree_read_unlock_blocking(eb);
1241                         free_extent_buffer(eb);
1242                 }
1243                 ret = inode_ref_info(parent, 0, fs_root, path, &found_key);
1244                 if (ret > 0)
1245                         ret = -ENOENT;
1246                 if (ret)
1247                         break;
1248
1249                 next_inum = found_key.offset;
1250
1251                 /* regular exit ahead */
1252                 if (parent == next_inum)
1253                         break;
1254
1255                 slot = path->slots[0];
1256                 eb = path->nodes[0];
1257                 /* make sure we can use eb after releasing the path */
1258                 if (eb != eb_in) {
1259                         atomic_inc(&eb->refs);
1260                         btrfs_tree_read_lock(eb);
1261                         btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1262                 }
1263                 btrfs_release_path(path);
1264                 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1265
1266                 name_len = btrfs_inode_ref_name_len(eb, iref);
1267                 name_off = (unsigned long)(iref + 1);
1268
1269                 parent = next_inum;
1270                 --bytes_left;
1271                 if (bytes_left >= 0)
1272                         dest[bytes_left] = '/';
1273         }
1274
1275         btrfs_release_path(path);
1276         path->leave_spinning = leave_spinning;
1277
1278         if (ret)
1279                 return ERR_PTR(ret);
1280
1281         return dest + bytes_left;
1282 }
1283
1284 /*
1285  * this makes the path point to (logical EXTENT_ITEM *)
1286  * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
1287  * tree blocks and <0 on error.
1288  */
1289 int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
1290                         struct btrfs_path *path, struct btrfs_key *found_key,
1291                         u64 *flags_ret)
1292 {
1293         int ret;
1294         u64 flags;
1295         u64 size = 0;
1296         u32 item_size;
1297         struct extent_buffer *eb;
1298         struct btrfs_extent_item *ei;
1299         struct btrfs_key key;
1300
1301         if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1302                 key.type = BTRFS_METADATA_ITEM_KEY;
1303         else
1304                 key.type = BTRFS_EXTENT_ITEM_KEY;
1305         key.objectid = logical;
1306         key.offset = (u64)-1;
1307
1308         ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
1309         if (ret < 0)
1310                 return ret;
1311         ret = btrfs_previous_item(fs_info->extent_root, path,
1312                                         0, BTRFS_EXTENT_ITEM_KEY);
1313         if (ret < 0)
1314                 return ret;
1315
1316         btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]);
1317         if (found_key->type == BTRFS_METADATA_ITEM_KEY)
1318                 size = fs_info->extent_root->leafsize;
1319         else if (found_key->type == BTRFS_EXTENT_ITEM_KEY)
1320                 size = found_key->offset;
1321
1322         if ((found_key->type != BTRFS_EXTENT_ITEM_KEY &&
1323              found_key->type != BTRFS_METADATA_ITEM_KEY) ||
1324             found_key->objectid > logical ||
1325             found_key->objectid + size <= logical) {
1326                 pr_debug("logical %llu is not within any extent\n",
1327                          (unsigned long long)logical);
1328                 return -ENOENT;
1329         }
1330
1331         eb = path->nodes[0];
1332         item_size = btrfs_item_size_nr(eb, path->slots[0]);
1333         BUG_ON(item_size < sizeof(*ei));
1334
1335         ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
1336         flags = btrfs_extent_flags(eb, ei);
1337
1338         pr_debug("logical %llu is at position %llu within the extent (%llu "
1339                  "EXTENT_ITEM %llu) flags %#llx size %u\n",
1340                  (unsigned long long)logical,
1341                  (unsigned long long)(logical - found_key->objectid),
1342                  (unsigned long long)found_key->objectid,
1343                  (unsigned long long)found_key->offset,
1344                  (unsigned long long)flags, item_size);
1345
1346         WARN_ON(!flags_ret);
1347         if (flags_ret) {
1348                 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1349                         *flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK;
1350                 else if (flags & BTRFS_EXTENT_FLAG_DATA)
1351                         *flags_ret = BTRFS_EXTENT_FLAG_DATA;
1352                 else
1353                         BUG_ON(1);
1354                 return 0;
1355         }
1356
1357         return -EIO;
1358 }
1359
1360 /*
1361  * helper function to iterate extent inline refs. ptr must point to a 0 value
1362  * for the first call and may be modified. it is used to track state.
1363  * if more refs exist, 0 is returned and the next call to
1364  * __get_extent_inline_ref must pass the modified ptr parameter to get the
1365  * next ref. after the last ref was processed, 1 is returned.
1366  * returns <0 on error
1367  */
1368 static int __get_extent_inline_ref(unsigned long *ptr, struct extent_buffer *eb,
1369                                 struct btrfs_extent_item *ei, u32 item_size,
1370                                 struct btrfs_extent_inline_ref **out_eiref,
1371                                 int *out_type)
1372 {
1373         unsigned long end;
1374         u64 flags;
1375         struct btrfs_tree_block_info *info;
1376
1377         if (!*ptr) {
1378                 /* first call */
1379                 flags = btrfs_extent_flags(eb, ei);
1380                 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1381                         info = (struct btrfs_tree_block_info *)(ei + 1);
1382                         *out_eiref =
1383                                 (struct btrfs_extent_inline_ref *)(info + 1);
1384                 } else {
1385                         *out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1);
1386                 }
1387                 *ptr = (unsigned long)*out_eiref;
1388                 if ((void *)*ptr >= (void *)ei + item_size)
1389                         return -ENOENT;
1390         }
1391
1392         end = (unsigned long)ei + item_size;
1393         *out_eiref = (struct btrfs_extent_inline_ref *)*ptr;
1394         *out_type = btrfs_extent_inline_ref_type(eb, *out_eiref);
1395
1396         *ptr += btrfs_extent_inline_ref_size(*out_type);
1397         WARN_ON(*ptr > end);
1398         if (*ptr == end)
1399                 return 1; /* last */
1400
1401         return 0;
1402 }
1403
1404 /*
1405  * reads the tree block backref for an extent. tree level and root are returned
1406  * through out_level and out_root. ptr must point to a 0 value for the first
1407  * call and may be modified (see __get_extent_inline_ref comment).
1408  * returns 0 if data was provided, 1 if there was no more data to provide or
1409  * <0 on error.
1410  */
1411 int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
1412                                 struct btrfs_extent_item *ei, u32 item_size,
1413                                 u64 *out_root, u8 *out_level)
1414 {
1415         int ret;
1416         int type;
1417         struct btrfs_tree_block_info *info;
1418         struct btrfs_extent_inline_ref *eiref;
1419
1420         if (*ptr == (unsigned long)-1)
1421                 return 1;
1422
1423         while (1) {
1424                 ret = __get_extent_inline_ref(ptr, eb, ei, item_size,
1425                                                 &eiref, &type);
1426                 if (ret < 0)
1427                         return ret;
1428
1429                 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1430                     type == BTRFS_SHARED_BLOCK_REF_KEY)
1431                         break;
1432
1433                 if (ret == 1)
1434                         return 1;
1435         }
1436
1437         /* we can treat both ref types equally here */
1438         info = (struct btrfs_tree_block_info *)(ei + 1);
1439         *out_root = btrfs_extent_inline_ref_offset(eb, eiref);
1440         *out_level = btrfs_tree_block_level(eb, info);
1441
1442         if (ret == 1)
1443                 *ptr = (unsigned long)-1;
1444
1445         return 0;
1446 }
1447
1448 static int iterate_leaf_refs(struct extent_inode_elem *inode_list,
1449                                 u64 root, u64 extent_item_objectid,
1450                                 iterate_extent_inodes_t *iterate, void *ctx)
1451 {
1452         struct extent_inode_elem *eie;
1453         int ret = 0;
1454
1455         for (eie = inode_list; eie; eie = eie->next) {
1456                 pr_debug("ref for %llu resolved, key (%llu EXTEND_DATA %llu), "
1457                          "root %llu\n", extent_item_objectid,
1458                          eie->inum, eie->offset, root);
1459                 ret = iterate(eie->inum, eie->offset, root, ctx);
1460                 if (ret) {
1461                         pr_debug("stopping iteration for %llu due to ret=%d\n",
1462                                  extent_item_objectid, ret);
1463                         break;
1464                 }
1465         }
1466
1467         return ret;
1468 }
1469
1470 /*
1471  * calls iterate() for every inode that references the extent identified by
1472  * the given parameters.
1473  * when the iterator function returns a non-zero value, iteration stops.
1474  */
1475 int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
1476                                 u64 extent_item_objectid, u64 extent_item_pos,
1477                                 int search_commit_root,
1478                                 iterate_extent_inodes_t *iterate, void *ctx)
1479 {
1480         int ret;
1481         struct btrfs_trans_handle *trans = NULL;
1482         struct ulist *refs = NULL;
1483         struct ulist *roots = NULL;
1484         struct ulist_node *ref_node = NULL;
1485         struct ulist_node *root_node = NULL;
1486         struct seq_list tree_mod_seq_elem = {};
1487         struct ulist_iterator ref_uiter;
1488         struct ulist_iterator root_uiter;
1489
1490         pr_debug("resolving all inodes for extent %llu\n",
1491                         extent_item_objectid);
1492
1493         if (!search_commit_root) {
1494                 trans = btrfs_join_transaction(fs_info->extent_root);
1495                 if (IS_ERR(trans))
1496                         return PTR_ERR(trans);
1497                 btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1498         }
1499
1500         ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid,
1501                                    tree_mod_seq_elem.seq, &refs,
1502                                    &extent_item_pos);
1503         if (ret)
1504                 goto out;
1505
1506         ULIST_ITER_INIT(&ref_uiter);
1507         while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) {
1508                 ret = btrfs_find_all_roots(trans, fs_info, ref_node->val,
1509                                            tree_mod_seq_elem.seq, &roots);
1510                 if (ret)
1511                         break;
1512                 ULIST_ITER_INIT(&root_uiter);
1513                 while (!ret && (root_node = ulist_next(roots, &root_uiter))) {
1514                         pr_debug("root %llu references leaf %llu, data list "
1515                                  "%#llx\n", root_node->val, ref_node->val,
1516                                  (long long)ref_node->aux);
1517                         ret = iterate_leaf_refs((struct extent_inode_elem *)
1518                                                 (uintptr_t)ref_node->aux,
1519                                                 root_node->val,
1520                                                 extent_item_objectid,
1521                                                 iterate, ctx);
1522                 }
1523                 ulist_free(roots);
1524         }
1525
1526         free_leaf_list(refs);
1527 out:
1528         if (!search_commit_root) {
1529                 btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1530                 btrfs_end_transaction(trans, fs_info->extent_root);
1531         }
1532
1533         return ret;
1534 }
1535
1536 int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
1537                                 struct btrfs_path *path,
1538                                 iterate_extent_inodes_t *iterate, void *ctx)
1539 {
1540         int ret;
1541         u64 extent_item_pos;
1542         u64 flags = 0;
1543         struct btrfs_key found_key;
1544         int search_commit_root = path->search_commit_root;
1545
1546         ret = extent_from_logical(fs_info, logical, path, &found_key, &flags);
1547         btrfs_release_path(path);
1548         if (ret < 0)
1549                 return ret;
1550         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1551                 return -EINVAL;
1552
1553         extent_item_pos = logical - found_key.objectid;
1554         ret = iterate_extent_inodes(fs_info, found_key.objectid,
1555                                         extent_item_pos, search_commit_root,
1556                                         iterate, ctx);
1557
1558         return ret;
1559 }
1560
1561 typedef int (iterate_irefs_t)(u64 parent, u32 name_len, unsigned long name_off,
1562                               struct extent_buffer *eb, void *ctx);
1563
1564 static int iterate_inode_refs(u64 inum, struct btrfs_root *fs_root,
1565                               struct btrfs_path *path,
1566                               iterate_irefs_t *iterate, void *ctx)
1567 {
1568         int ret = 0;
1569         int slot;
1570         u32 cur;
1571         u32 len;
1572         u32 name_len;
1573         u64 parent = 0;
1574         int found = 0;
1575         struct extent_buffer *eb;
1576         struct btrfs_item *item;
1577         struct btrfs_inode_ref *iref;
1578         struct btrfs_key found_key;
1579
1580         while (!ret) {
1581                 path->leave_spinning = 1;
1582                 ret = inode_ref_info(inum, parent ? parent+1 : 0, fs_root, path,
1583                                      &found_key);
1584                 if (ret < 0)
1585                         break;
1586                 if (ret) {
1587                         ret = found ? 0 : -ENOENT;
1588                         break;
1589                 }
1590                 ++found;
1591
1592                 parent = found_key.offset;
1593                 slot = path->slots[0];
1594                 eb = path->nodes[0];
1595                 /* make sure we can use eb after releasing the path */
1596                 atomic_inc(&eb->refs);
1597                 btrfs_tree_read_lock(eb);
1598                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1599                 btrfs_release_path(path);
1600
1601                 item = btrfs_item_nr(eb, slot);
1602                 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1603
1604                 for (cur = 0; cur < btrfs_item_size(eb, item); cur += len) {
1605                         name_len = btrfs_inode_ref_name_len(eb, iref);
1606                         /* path must be released before calling iterate()! */
1607                         pr_debug("following ref at offset %u for inode %llu in "
1608                                  "tree %llu\n", cur,
1609                                  (unsigned long long)found_key.objectid,
1610                                  (unsigned long long)fs_root->objectid);
1611                         ret = iterate(parent, name_len,
1612                                       (unsigned long)(iref + 1), eb, ctx);
1613                         if (ret)
1614                                 break;
1615                         len = sizeof(*iref) + name_len;
1616                         iref = (struct btrfs_inode_ref *)((char *)iref + len);
1617                 }
1618                 btrfs_tree_read_unlock_blocking(eb);
1619                 free_extent_buffer(eb);
1620         }
1621
1622         btrfs_release_path(path);
1623
1624         return ret;
1625 }
1626
1627 static int iterate_inode_extrefs(u64 inum, struct btrfs_root *fs_root,
1628                                  struct btrfs_path *path,
1629                                  iterate_irefs_t *iterate, void *ctx)
1630 {
1631         int ret;
1632         int slot;
1633         u64 offset = 0;
1634         u64 parent;
1635         int found = 0;
1636         struct extent_buffer *eb;
1637         struct btrfs_inode_extref *extref;
1638         struct extent_buffer *leaf;
1639         u32 item_size;
1640         u32 cur_offset;
1641         unsigned long ptr;
1642
1643         while (1) {
1644                 ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref,
1645                                             &offset);
1646                 if (ret < 0)
1647                         break;
1648                 if (ret) {
1649                         ret = found ? 0 : -ENOENT;
1650                         break;
1651                 }
1652                 ++found;
1653
1654                 slot = path->slots[0];
1655                 eb = path->nodes[0];
1656                 /* make sure we can use eb after releasing the path */
1657                 atomic_inc(&eb->refs);
1658
1659                 btrfs_tree_read_lock(eb);
1660                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1661                 btrfs_release_path(path);
1662
1663                 leaf = path->nodes[0];
1664                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1665                 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1666                 cur_offset = 0;
1667
1668                 while (cur_offset < item_size) {
1669                         u32 name_len;
1670
1671                         extref = (struct btrfs_inode_extref *)(ptr + cur_offset);
1672                         parent = btrfs_inode_extref_parent(eb, extref);
1673                         name_len = btrfs_inode_extref_name_len(eb, extref);
1674                         ret = iterate(parent, name_len,
1675                                       (unsigned long)&extref->name, eb, ctx);
1676                         if (ret)
1677                                 break;
1678
1679                         cur_offset += btrfs_inode_extref_name_len(leaf, extref);
1680                         cur_offset += sizeof(*extref);
1681                 }
1682                 btrfs_tree_read_unlock_blocking(eb);
1683                 free_extent_buffer(eb);
1684
1685                 offset++;
1686         }
1687
1688         btrfs_release_path(path);
1689
1690         return ret;
1691 }
1692
1693 static int iterate_irefs(u64 inum, struct btrfs_root *fs_root,
1694                          struct btrfs_path *path, iterate_irefs_t *iterate,
1695                          void *ctx)
1696 {
1697         int ret;
1698         int found_refs = 0;
1699
1700         ret = iterate_inode_refs(inum, fs_root, path, iterate, ctx);
1701         if (!ret)
1702                 ++found_refs;
1703         else if (ret != -ENOENT)
1704                 return ret;
1705
1706         ret = iterate_inode_extrefs(inum, fs_root, path, iterate, ctx);
1707         if (ret == -ENOENT && found_refs)
1708                 return 0;
1709
1710         return ret;
1711 }
1712
1713 /*
1714  * returns 0 if the path could be dumped (probably truncated)
1715  * returns <0 in case of an error
1716  */
1717 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
1718                          struct extent_buffer *eb, void *ctx)
1719 {
1720         struct inode_fs_paths *ipath = ctx;
1721         char *fspath;
1722         char *fspath_min;
1723         int i = ipath->fspath->elem_cnt;
1724         const int s_ptr = sizeof(char *);
1725         u32 bytes_left;
1726
1727         bytes_left = ipath->fspath->bytes_left > s_ptr ?
1728                                         ipath->fspath->bytes_left - s_ptr : 0;
1729
1730         fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
1731         fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len,
1732                                    name_off, eb, inum, fspath_min, bytes_left);
1733         if (IS_ERR(fspath))
1734                 return PTR_ERR(fspath);
1735
1736         if (fspath > fspath_min) {
1737                 ipath->fspath->val[i] = (u64)(unsigned long)fspath;
1738                 ++ipath->fspath->elem_cnt;
1739                 ipath->fspath->bytes_left = fspath - fspath_min;
1740         } else {
1741                 ++ipath->fspath->elem_missed;
1742                 ipath->fspath->bytes_missing += fspath_min - fspath;
1743                 ipath->fspath->bytes_left = 0;
1744         }
1745
1746         return 0;
1747 }
1748
1749 /*
1750  * this dumps all file system paths to the inode into the ipath struct, provided
1751  * is has been created large enough. each path is zero-terminated and accessed
1752  * from ipath->fspath->val[i].
1753  * when it returns, there are ipath->fspath->elem_cnt number of paths available
1754  * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the
1755  * number of missed paths in recored in ipath->fspath->elem_missed, otherwise,
1756  * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
1757  * have been needed to return all paths.
1758  */
1759 int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
1760 {
1761         return iterate_irefs(inum, ipath->fs_root, ipath->btrfs_path,
1762                              inode_to_path, ipath);
1763 }
1764
1765 struct btrfs_data_container *init_data_container(u32 total_bytes)
1766 {
1767         struct btrfs_data_container *data;
1768         size_t alloc_bytes;
1769
1770         alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
1771         data = vmalloc(alloc_bytes);
1772         if (!data)
1773                 return ERR_PTR(-ENOMEM);
1774
1775         if (total_bytes >= sizeof(*data)) {
1776                 data->bytes_left = total_bytes - sizeof(*data);
1777                 data->bytes_missing = 0;
1778         } else {
1779                 data->bytes_missing = sizeof(*data) - total_bytes;
1780                 data->bytes_left = 0;
1781         }
1782
1783         data->elem_cnt = 0;
1784         data->elem_missed = 0;
1785
1786         return data;
1787 }
1788
1789 /*
1790  * allocates space to return multiple file system paths for an inode.
1791  * total_bytes to allocate are passed, note that space usable for actual path
1792  * information will be total_bytes - sizeof(struct inode_fs_paths).
1793  * the returned pointer must be freed with free_ipath() in the end.
1794  */
1795 struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
1796                                         struct btrfs_path *path)
1797 {
1798         struct inode_fs_paths *ifp;
1799         struct btrfs_data_container *fspath;
1800
1801         fspath = init_data_container(total_bytes);
1802         if (IS_ERR(fspath))
1803                 return (void *)fspath;
1804
1805         ifp = kmalloc(sizeof(*ifp), GFP_NOFS);
1806         if (!ifp) {
1807                 kfree(fspath);
1808                 return ERR_PTR(-ENOMEM);
1809         }
1810
1811         ifp->btrfs_path = path;
1812         ifp->fspath = fspath;
1813         ifp->fs_root = fs_root;
1814
1815         return ifp;
1816 }
1817
1818 void free_ipath(struct inode_fs_paths *ipath)
1819 {
1820         if (!ipath)
1821                 return;
1822         vfree(ipath->fspath);
1823         kfree(ipath);
1824 }