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