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