fs/btrfs/backref.c

   1 /*
   2  * Copyright (C) 2011 STRATO.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/vmalloc.h>
  20 #include "ctree.h"
  21 #include "disk-io.h"
  22 #include "backref.h"
  23 #include "ulist.h"
  24 #include "transaction.h"
  25 #include "delayed-ref.h"
  26 #include "locking.h"
  27
  28 struct extent_inode_elem {
  29         u64 inum;
  30         u64 offset;
  31         struct extent_inode_elem *next;
  32 };
  33
  34 static int check_extent_in_eb(struct btrfs_key *key, struct extent_buffer *eb,
  35                                 struct btrfs_file_extent_item *fi,
  36                                 u64 extent_item_pos,
  37                                 struct extent_inode_elem **eie)
  38 {
  39         u64 offset = 0;
  40         struct extent_inode_elem *e;
  41
  42         if (!btrfs_file_extent_compression(eb, fi) &&
  43             !btrfs_file_extent_encryption(eb, fi) &&
  44             !btrfs_file_extent_other_encoding(eb, fi)) {
  45                 u64 data_offset;
  46                 u64 data_len;
  47
  48                 data_offset = btrfs_file_extent_offset(eb, fi);
  49                 data_len = btrfs_file_extent_num_bytes(eb, fi);
  50
  51                 if (extent_item_pos < data_offset ||
  52                     extent_item_pos >= data_offset + data_len)
  53                         return 1;
  54                 offset = extent_item_pos - data_offset;
  55         }
  56
  57         e = kmalloc(sizeof(*e), GFP_NOFS);
  58         if (!e)
  59                 return -ENOMEM;
  60
  61         e->next = *eie;
  62         e->inum = key->objectid;
  63         e->offset = key->offset + offset;
  64         *eie = e;
  65
  66         return 0;
  67 }
  68
  69 static int find_extent_in_eb(struct extent_buffer *eb, u64 wanted_disk_byte,
  70                                 u64 extent_item_pos,
  71                                 struct extent_inode_elem **eie)
  72 {
  73         u64 disk_byte;
  74         struct btrfs_key key;
  75         struct btrfs_file_extent_item *fi;
  76         int slot;
  77         int nritems;
  78         int extent_type;
  79         int ret;
  80
  81         /*
  82          * from the shared data ref, we only have the leaf but we need
  83          * the key. thus, we must look into all items and see that we
  84          * find one (some) with a reference to our extent item.
  85          */
  86         nritems = btrfs_header_nritems(eb);
  87         for (slot = 0; slot < nritems; ++slot) {
  88                 btrfs_item_key_to_cpu(eb, &key, slot);
  89                 if (key.type != BTRFS_EXTENT_DATA_KEY)
  90                         continue;
  91                 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
  92                 extent_type = btrfs_file_extent_type(eb, fi);
  93                 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
  94                         continue;
  95                 /* don't skip BTRFS_FILE_EXTENT_PREALLOC, we can handle that */
  96                 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
  97                 if (disk_byte != wanted_disk_byte)
  98                         continue;
  99
 100                 ret = check_extent_in_eb(&key, eb, fi, extent_item_pos, eie);
 101                 if (ret < 0)
 102                         return ret;
 103         }
 104
 105         return 0;
 106 }
 107
 108 /*
 109  * this structure records all encountered refs on the way up to the root
 110  */
 111 struct __prelim_ref {
 112         struct list_head list;
 113         u64 root_id;
 114         struct btrfs_key key_for_search;
 115         int level;
 116         int count;
 117         struct extent_inode_elem *inode_list;
 118         u64 parent;
 119         u64 wanted_disk_byte;
 120 };
 121
 122 /*
 123  * the rules for all callers of this function are:
 124  * - obtaining the parent is the goal
 125  * - if you add a key, you must know that it is a correct key
 126  * - if you cannot add the parent or a correct key, then we will look into the
 127  *   block later to set a correct key
 128  *
 129  * delayed refs
 130  * ============
 131  *        backref type | shared | indirect | shared | indirect
 132  * information         |   tree |     tree |   data |     data
 133  * --------------------+--------+----------+--------+----------
 134  *      parent logical |    y   |     -    |    -   |     -
 135  *      key to resolve |    -   |     y    |    y   |     y
 136  *  tree block logical |    -   |     -    |    -   |     -
 137  *  root for resolving |    y   |     y    |    y   |     y
 138  *
 139  * - column 1:       we've the parent -> done
 140  * - column 2, 3, 4: we use the key to find the parent
 141  *
 142  * on disk refs (inline or keyed)
 143  * ==============================
 144  *        backref type | shared | indirect | shared | indirect
 145  * information         |   tree |     tree |   data |     data
 146  * --------------------+--------+----------+--------+----------
 147  *      parent logical |    y   |     -    |    y   |     -
 148  *      key to resolve |    -   |     -    |    -   |     y
 149  *  tree block logical |    y   |     y    |    y   |     y
 150  *  root for resolving |    -   |     y    |    y   |     y
 151  *
 152  * - column 1, 3: we've the parent -> done
 153  * - column 2:    we take the first key from the block to find the parent
 154  *                (see __add_missing_keys)
 155  * - column 4:    we use the key to find the parent
 156  *
 157  * additional information that's available but not required to find the parent
 158  * block might help in merging entries to gain some speed.
 159  */
 160
 161 static int __add_prelim_ref(struct list_head *head, u64 root_id,
 162                             struct btrfs_key *key, int level,
 163                             u64 parent, u64 wanted_disk_byte, int count)
 164 {
 165         struct __prelim_ref *ref;
 166
 167         /* in case we're adding delayed refs, we're holding the refs spinlock */
 168         ref = kmalloc(sizeof(*ref), GFP_ATOMIC);
 169         if (!ref)
 170                 return -ENOMEM;
 171
 172         ref->root_id = root_id;
 173         if (key)
 174                 ref->key_for_search = *key;
 175         else
 176                 memset(&ref->key_for_search, 0, sizeof(ref->key_for_search));
 177
 178         ref->inode_list = NULL;
 179         ref->level = level;
 180         ref->count = count;
 181         ref->parent = parent;
 182         ref->wanted_disk_byte = wanted_disk_byte;
 183         list_add_tail(&ref->list, head);
 184
 185         return 0;
 186 }
 187
 188 static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path,
 189                                 struct ulist *parents, int level,
 190                                 struct btrfs_key *key_for_search, u64 time_seq,
 191                                 u64 wanted_disk_byte,
 192                                 const u64 *extent_item_pos)
 193 {
 194         int ret = 0;
 195         int slot;
 196         struct extent_buffer *eb;
 197         struct btrfs_key key;
 198         struct btrfs_file_extent_item *fi;
 199         struct extent_inode_elem *eie = NULL, *old = NULL;
 200         u64 disk_byte;
 201
 202         if (level != 0) {
 203                 eb = path->nodes[level];
 204                 ret = ulist_add(parents, eb->start, 0, GFP_NOFS);
 205                 if (ret < 0)
 206                         return ret;
 207                 return 0;
 208         }
 209
 210         /*
 211          * We normally enter this function with the path already pointing to
 212          * the first item to check. But sometimes, we may enter it with
 213          * slot==nritems. In that case, go to the next leaf before we continue.
 214          */
 215         if (path->slots[0] >= btrfs_header_nritems(path->nodes[0]))
 216                 ret = btrfs_next_old_leaf(root, path, time_seq);
 217
 218         while (!ret) {
 219                 eb = path->nodes[0];
 220                 slot = path->slots[0];
 221
 222                 btrfs_item_key_to_cpu(eb, &key, slot);
 223
 224                 if (key.objectid != key_for_search->objectid ||
 225                     key.type != BTRFS_EXTENT_DATA_KEY)
 226                         break;
 227
 228                 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
 229                 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
 230
 231                 if (disk_byte == wanted_disk_byte) {
 232                         eie = NULL;
 233                         old = NULL;
 234                         if (extent_item_pos) {
 235                                 ret = check_extent_in_eb(&key, eb, fi,
 236                                                 *extent_item_pos,
 237                                                 &eie);
 238                                 if (ret < 0)
 239                                         break;
 240                         }
 241                         if (ret > 0)
 242                                 goto next;
 243                         ret = ulist_add_merge(parents, eb->start,
 244                                               (uintptr_t)eie,
 245                                               (u64 *)&old, GFP_NOFS);
 246                         if (ret < 0)
 247                                 break;
 248                         if (!ret && extent_item_pos) {
 249                                 while (old->next)
 250                                         old = old->next;
 251                                 old->next = eie;
 252                         }
 253                 }
 254 next:
 255                 ret = btrfs_next_old_item(root, path, time_seq);
 256         }
 257
 258         if (ret > 0)
 259                 ret = 0;
 260         return ret;
 261 }
 262
 263 /*
 264  * resolve an indirect backref in the form (root_id, key, level)
 265  * to a logical address
 266  */
 267 static int __resolve_indirect_ref(struct btrfs_fs_info *fs_info,
 268                                   struct btrfs_path *path, u64 time_seq,
 269                                   struct __prelim_ref *ref,
 270                                   struct ulist *parents,
 271                                   const u64 *extent_item_pos)
 272 {
 273         struct btrfs_root *root;
 274         struct btrfs_key root_key;
 275         struct extent_buffer *eb;
 276         int ret = 0;
 277         int root_level;
 278         int level = ref->level;
 279
 280         root_key.objectid = ref->root_id;
 281         root_key.type = BTRFS_ROOT_ITEM_KEY;
 282         root_key.offset = (u64)-1;
 283         root = btrfs_read_fs_root_no_name(fs_info, &root_key);
 284         if (IS_ERR(root)) {
 285                 ret = PTR_ERR(root);
 286                 goto out;
 287         }
 288
 289         root_level = btrfs_old_root_level(root, time_seq);
 290
 291         if (root_level + 1 == level)
 292                 goto out;
 293
 294         path->lowest_level = level;
 295         ret = btrfs_search_old_slot(root, &ref->key_for_search, path, time_seq);
 296         pr_debug("search slot in root %llu (level %d, ref count %d) returned "
 297                  "%d for key (%llu %u %llu)\n",
 298                  ref->root_id, level, ref->count, ret,
 299                  ref->key_for_search.objectid, ref->key_for_search.type,
 300                  ref->key_for_search.offset);
 301         if (ret < 0)
 302                 goto out;
 303
 304         eb = path->nodes[level];
 305         while (!eb) {
 306                 if (!level) {
 307                         WARN_ON(1);
 308                         ret = 1;
 309                         goto out;
 310                 }
 311                 level--;
 312                 eb = path->nodes[level];
 313         }
 314
 315         ret = add_all_parents(root, path, parents, level, &ref->key_for_search,
 316                                 time_seq, ref->wanted_disk_byte,
 317                                 extent_item_pos);
 318 out:
 319         path->lowest_level = 0;
 320         btrfs_release_path(path);
 321         return ret;
 322 }
 323
 324 /*
 325  * resolve all indirect backrefs from the list
 326  */
 327 static int __resolve_indirect_refs(struct btrfs_fs_info *fs_info,
 328                                    struct btrfs_path *path, u64 time_seq,
 329                                    struct list_head *head,
 330                                    const u64 *extent_item_pos)
 331 {
 332         int err;
 333         int ret = 0;
 334         struct __prelim_ref *ref;
 335         struct __prelim_ref *ref_safe;
 336         struct __prelim_ref *new_ref;
 337         struct ulist *parents;
 338         struct ulist_node *node;
 339         struct ulist_iterator uiter;
 340
 341         parents = ulist_alloc(GFP_NOFS);
 342         if (!parents)
 343                 return -ENOMEM;
 344
 345         /*
 346          * _safe allows us to insert directly after the current item without
 347          * iterating over the newly inserted items.
 348          * we're also allowed to re-assign ref during iteration.
 349          */
 350         list_for_each_entry_safe(ref, ref_safe, head, list) {
 351                 if (ref->parent)        /* already direct */
 352                         continue;
 353                 if (ref->count == 0)
 354                         continue;
 355                 err = __resolve_indirect_ref(fs_info, path, time_seq, ref,
 356                                              parents, extent_item_pos);
 357                 if (err == -ENOMEM)
 358                         goto out;
 359                 if (err)
 360                         continue;
 361
 362                 /* we put the first parent into the ref at hand */
 363                 ULIST_ITER_INIT(&uiter);
 364                 node = ulist_next(parents, &uiter);
 365                 ref->parent = node ? node->val : 0;
 366                 ref->inode_list = node ?
 367                         (struct extent_inode_elem *)(uintptr_t)node->aux : NULL;
 368
 369                 /* additional parents require new refs being added here */
 370                 while ((node = ulist_next(parents, &uiter))) {
 371                         new_ref = kmalloc(sizeof(*new_ref), GFP_NOFS);
 372                         if (!new_ref) {
 373                                 ret = -ENOMEM;
 374                                 goto out;
 375                         }
 376                         memcpy(new_ref, ref, sizeof(*ref));
 377                         new_ref->parent = node->val;
 378                         new_ref->inode_list = (struct extent_inode_elem *)
 379                                                         (uintptr_t)node->aux;
 380                         list_add(&new_ref->list, &ref->list);
 381                 }
 382                 ulist_reinit(parents);
 383         }
 384 out:
 385         ulist_free(parents);
 386         return ret;
 387 }
 388
 389 static inline int ref_for_same_block(struct __prelim_ref *ref1,
 390                                      struct __prelim_ref *ref2)
 391 {
 392         if (ref1->level != ref2->level)
 393                 return 0;
 394         if (ref1->root_id != ref2->root_id)
 395                 return 0;
 396         if (ref1->key_for_search.type != ref2->key_for_search.type)
 397                 return 0;
 398         if (ref1->key_for_search.objectid != ref2->key_for_search.objectid)
 399                 return 0;
 400         if (ref1->key_for_search.offset != ref2->key_for_search.offset)
 401                 return 0;
 402         if (ref1->parent != ref2->parent)
 403                 return 0;
 404
 405         return 1;
 406 }
 407
 408 /*
 409  * read tree blocks and add keys where required.
 410  */
 411 static int __add_missing_keys(struct btrfs_fs_info *fs_info,
 412                               struct list_head *head)
 413 {
 414         struct list_head *pos;
 415         struct extent_buffer *eb;
 416
 417         list_for_each(pos, head) {
 418                 struct __prelim_ref *ref;
 419                 ref = list_entry(pos, struct __prelim_ref, list);
 420
 421                 if (ref->parent)
 422                         continue;
 423                 if (ref->key_for_search.type)
 424                         continue;
 425                 BUG_ON(!ref->wanted_disk_byte);
 426                 eb = read_tree_block(fs_info->tree_root, ref->wanted_disk_byte,
 427                                      fs_info->tree_root->leafsize, 0);
 428                 if (!eb || !extent_buffer_uptodate(eb)) {
 429                         free_extent_buffer(eb);
 430                         return -EIO;
 431                 }
 432                 btrfs_tree_read_lock(eb);
 433                 if (btrfs_header_level(eb) == 0)
 434                         btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
 435                 else
 436                         btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
 437                 btrfs_tree_read_unlock(eb);
 438                 free_extent_buffer(eb);
 439         }
 440         return 0;
 441 }
 442
 443 /*
 444  * merge two lists of backrefs and adjust counts accordingly
 445  *
 446  * mode = 1: merge identical keys, if key is set
 447  *    FIXME: if we add more keys in __add_prelim_ref, we can merge more here.
 448  *           additionally, we could even add a key range for the blocks we
 449  *           looked into to merge even more (-> replace unresolved refs by those
 450  *           having a parent).
 451  * mode = 2: merge identical parents
 452  */
 453 static void __merge_refs(struct list_head *head, int mode)
 454 {
 455         struct list_head *pos1;
 456
 457         list_for_each(pos1, head) {
 458                 struct list_head *n2;
 459                 struct list_head *pos2;
 460                 struct __prelim_ref *ref1;
 461
 462                 ref1 = list_entry(pos1, struct __prelim_ref, list);
 463
 464                 for (pos2 = pos1->next, n2 = pos2->next; pos2 != head;
 465                      pos2 = n2, n2 = pos2->next) {
 466                         struct __prelim_ref *ref2;
 467                         struct __prelim_ref *xchg;
 468                         struct extent_inode_elem *eie;
 469
 470                         ref2 = list_entry(pos2, struct __prelim_ref, list);
 471
 472                         if (mode == 1) {
 473                                 if (!ref_for_same_block(ref1, ref2))
 474                                         continue;
 475                                 if (!ref1->parent && ref2->parent) {
 476                                         xchg = ref1;
 477                                         ref1 = ref2;
 478                                         ref2 = xchg;
 479                                 }
 480                         } else {
 481                                 if (ref1->parent != ref2->parent)
 482                                         continue;
 483                         }
 484
 485                         eie = ref1->inode_list;
 486                         while (eie && eie->next)
 487                                 eie = eie->next;
 488                         if (eie)
 489                                 eie->next = ref2->inode_list;
 490                         else
 491                                 ref1->inode_list = ref2->inode_list;
 492                         ref1->count += ref2->count;
 493
 494                         list_del(&ref2->list);
 495                         kfree(ref2);
 496                 }
 497
 498         }
 499 }
 500
 501 /*
 502  * add all currently queued delayed refs from this head whose seq nr is
 503  * smaller or equal that seq to the list
 504  */
 505 static int __add_delayed_refs(struct btrfs_delayed_ref_head *head, u64 seq,
 506                               struct list_head *prefs)
 507 {
 508         struct btrfs_delayed_extent_op *extent_op = head->extent_op;
 509         struct rb_node *n = &head->node.rb_node;
 510         struct btrfs_key key;
 511         struct btrfs_key op_key = {0};
 512         int sgn;
 513         int ret = 0;
 514
 515         if (extent_op && extent_op->update_key)
 516                 btrfs_disk_key_to_cpu(&op_key, &extent_op->key);
 517
 518         while ((n = rb_prev(n))) {
 519                 struct btrfs_delayed_ref_node *node;
 520                 node = rb_entry(n, struct btrfs_delayed_ref_node,
 521                                 rb_node);
 522                 if (node->bytenr != head->node.bytenr)
 523                         break;
 524                 WARN_ON(node->is_head);
 525
 526                 if (node->seq > seq)
 527                         continue;
 528
 529                 switch (node->action) {
 530                 case BTRFS_ADD_DELAYED_EXTENT:
 531                 case BTRFS_UPDATE_DELAYED_HEAD:
 532                         WARN_ON(1);
 533                         continue;
 534                 case BTRFS_ADD_DELAYED_REF:
 535                         sgn = 1;
 536                         break;
 537                 case BTRFS_DROP_DELAYED_REF:
 538                         sgn = -1;
 539                         break;
 540                 default:
 541                         BUG_ON(1);
 542                 }
 543                 switch (node->type) {
 544                 case BTRFS_TREE_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, &op_key,
 549                                                ref->level + 1, 0, node->bytenr,
 550                                                node->ref_mod * sgn);
 551                         break;
 552                 }
 553                 case BTRFS_SHARED_BLOCK_REF_KEY: {
 554                         struct btrfs_delayed_tree_ref *ref;
 555
 556                         ref = btrfs_delayed_node_to_tree_ref(node);
 557                         ret = __add_prelim_ref(prefs, ref->root, NULL,
 558                                                ref->level + 1, ref->parent,
 559                                                node->bytenr,
 560                                                node->ref_mod * sgn);
 561                         break;
 562                 }
 563                 case BTRFS_EXTENT_DATA_REF_KEY: {
 564                         struct btrfs_delayed_data_ref *ref;
 565                         ref = btrfs_delayed_node_to_data_ref(node);
 566
 567                         key.objectid = ref->objectid;
 568                         key.type = BTRFS_EXTENT_DATA_KEY;
 569                         key.offset = ref->offset;
 570                         ret = __add_prelim_ref(prefs, ref->root, &key, 0, 0,
 571                                                node->bytenr,
 572                                                node->ref_mod * sgn);
 573                         break;
 574                 }
 575                 case BTRFS_SHARED_DATA_REF_KEY: {
 576                         struct btrfs_delayed_data_ref *ref;
 577
 578                         ref = btrfs_delayed_node_to_data_ref(node);
 579
 580                         key.objectid = ref->objectid;
 581                         key.type = BTRFS_EXTENT_DATA_KEY;
 582                         key.offset = ref->offset;
 583                         ret = __add_prelim_ref(prefs, ref->root, &key, 0,
 584                                                ref->parent, node->bytenr,
 585                                                node->ref_mod * sgn);
 586                         break;
 587                 }
 588                 default:
 589                         WARN_ON(1);
 590                 }
 591                 if (ret)
 592                         return ret;
 593         }
 594
 595         return 0;
 596 }
 597
 598 /*
 599  * add all inline backrefs for bytenr to the list
 600  */
 601 static int __add_inline_refs(struct btrfs_fs_info *fs_info,
 602                              struct btrfs_path *path, u64 bytenr,
 603                              int *info_level, struct list_head *prefs)
 604 {
 605         int ret = 0;
 606         int slot;
 607         struct extent_buffer *leaf;
 608         struct btrfs_key key;
 609         struct btrfs_key found_key;
 610         unsigned long ptr;
 611         unsigned long end;
 612         struct btrfs_extent_item *ei;
 613         u64 flags;
 614         u64 item_size;
 615
 616         /*
 617          * enumerate all inline refs
 618          */
 619         leaf = path->nodes[0];
 620         slot = path->slots[0];
 621
 622         item_size = btrfs_item_size_nr(leaf, slot);
 623         BUG_ON(item_size < sizeof(*ei));
 624
 625         ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
 626         flags = btrfs_extent_flags(leaf, ei);
 627         btrfs_item_key_to_cpu(leaf, &found_key, slot);
 628
 629         ptr = (unsigned long)(ei + 1);
 630         end = (unsigned long)ei + item_size;
 631
 632         if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
 633             flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
 634                 struct btrfs_tree_block_info *info;
 635
 636                 info = (struct btrfs_tree_block_info *)ptr;
 637                 *info_level = btrfs_tree_block_level(leaf, info);
 638                 ptr += sizeof(struct btrfs_tree_block_info);
 639                 BUG_ON(ptr > end);
 640         } else if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
 641                 *info_level = found_key.offset;
 642         } else {
 643                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
 644         }
 645
 646         while (ptr < end) {
 647                 struct btrfs_extent_inline_ref *iref;
 648                 u64 offset;
 649                 int type;
 650
 651                 iref = (struct btrfs_extent_inline_ref *)ptr;
 652                 type = btrfs_extent_inline_ref_type(leaf, iref);
 653                 offset = btrfs_extent_inline_ref_offset(leaf, iref);
 654
 655                 switch (type) {
 656                 case BTRFS_SHARED_BLOCK_REF_KEY:
 657                         ret = __add_prelim_ref(prefs, 0, NULL,
 658                                                 *info_level + 1, offset,
 659                                                 bytenr, 1);
 660                         break;
 661                 case BTRFS_SHARED_DATA_REF_KEY: {
 662                         struct btrfs_shared_data_ref *sdref;
 663                         int count;
 664
 665                         sdref = (struct btrfs_shared_data_ref *)(iref + 1);
 666                         count = btrfs_shared_data_ref_count(leaf, sdref);
 667                         ret = __add_prelim_ref(prefs, 0, NULL, 0, offset,
 668                                                bytenr, count);
 669                         break;
 670                 }
 671                 case BTRFS_TREE_BLOCK_REF_KEY:
 672                         ret = __add_prelim_ref(prefs, offset, NULL,
 673                                                *info_level + 1, 0,
 674                                                bytenr, 1);
 675                         break;
 676                 case BTRFS_EXTENT_DATA_REF_KEY: {
 677                         struct btrfs_extent_data_ref *dref;
 678                         int count;
 679                         u64 root;
 680
 681                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
 682                         count = btrfs_extent_data_ref_count(leaf, dref);
 683                         key.objectid = btrfs_extent_data_ref_objectid(leaf,
 684                                                                       dref);
 685                         key.type = BTRFS_EXTENT_DATA_KEY;
 686                         key.offset = btrfs_extent_data_ref_offset(leaf, dref);
 687                         root = btrfs_extent_data_ref_root(leaf, dref);
 688                         ret = __add_prelim_ref(prefs, root, &key, 0, 0,
 689                                                bytenr, count);
 690                         break;
 691                 }
 692                 default:
 693                         WARN_ON(1);
 694                 }
 695                 if (ret)
 696                         return ret;
 697                 ptr += btrfs_extent_inline_ref_size(type);
 698         }
 699
 700         return 0;
 701 }
 702
 703 /*
 704  * add all non-inline backrefs for bytenr to the list
 705  */
 706 static int __add_keyed_refs(struct btrfs_fs_info *fs_info,
 707                             struct btrfs_path *path, u64 bytenr,
 708                             int info_level, struct list_head *prefs)
 709 {
 710         struct btrfs_root *extent_root = fs_info->extent_root;
 711         int ret;
 712         int slot;
 713         struct extent_buffer *leaf;
 714         struct btrfs_key key;
 715
 716         while (1) {
 717                 ret = btrfs_next_item(extent_root, path);
 718                 if (ret < 0)
 719                         break;
 720                 if (ret) {
 721                         ret = 0;
 722                         break;
 723                 }
 724
 725                 slot = path->slots[0];
 726                 leaf = path->nodes[0];
 727                 btrfs_item_key_to_cpu(leaf, &key, slot);
 728
 729                 if (key.objectid != bytenr)
 730                         break;
 731                 if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
 732                         continue;
 733                 if (key.type > BTRFS_SHARED_DATA_REF_KEY)
 734                         break;
 735
 736                 switch (key.type) {
 737                 case BTRFS_SHARED_BLOCK_REF_KEY:
 738                         ret = __add_prelim_ref(prefs, 0, NULL,
 739                                                 info_level + 1, key.offset,
 740                                                 bytenr, 1);
 741                         break;
 742                 case BTRFS_SHARED_DATA_REF_KEY: {
 743                         struct btrfs_shared_data_ref *sdref;
 744                         int count;
 745
 746                         sdref = btrfs_item_ptr(leaf, slot,
 747                                               struct btrfs_shared_data_ref);
 748                         count = btrfs_shared_data_ref_count(leaf, sdref);
 749                         ret = __add_prelim_ref(prefs, 0, NULL, 0, key.offset,
 750                                                 bytenr, count);
 751                         break;
 752                 }
 753                 case BTRFS_TREE_BLOCK_REF_KEY:
 754                         ret = __add_prelim_ref(prefs, key.offset, NULL,
 755                                                info_level + 1, 0,
 756                                                bytenr, 1);
 757                         break;
 758                 case BTRFS_EXTENT_DATA_REF_KEY: {
 759                         struct btrfs_extent_data_ref *dref;
 760                         int count;
 761                         u64 root;
 762
 763                         dref = btrfs_item_ptr(leaf, slot,
 764                                               struct btrfs_extent_data_ref);
 765                         count = btrfs_extent_data_ref_count(leaf, dref);
 766                         key.objectid = btrfs_extent_data_ref_objectid(leaf,
 767                                                                       dref);
 768                         key.type = BTRFS_EXTENT_DATA_KEY;
 769                         key.offset = btrfs_extent_data_ref_offset(leaf, dref);
 770                         root = btrfs_extent_data_ref_root(leaf, dref);
 771                         ret = __add_prelim_ref(prefs, root, &key, 0, 0,
 772                                                bytenr, count);
 773                         break;
 774                 }
 775                 default:
 776                         WARN_ON(1);
 777                 }
 778                 if (ret)
 779                         return ret;
 780
 781         }
 782
 783         return ret;
 784 }
 785
 786 /*
 787  * this adds all existing backrefs (inline backrefs, backrefs and delayed
 788  * refs) for the given bytenr to the refs list, merges duplicates and resolves
 789  * indirect refs to their parent bytenr.
 790  * When roots are found, they're added to the roots list
 791  *
 792  * FIXME some caching might speed things up
 793  */
 794 static int find_parent_nodes(struct btrfs_trans_handle *trans,
 795                              struct btrfs_fs_info *fs_info, u64 bytenr,
 796                              u64 time_seq, struct ulist *refs,
 797                              struct ulist *roots, const u64 *extent_item_pos)
 798 {
 799         struct btrfs_key key;
 800         struct btrfs_path *path;
 801         struct btrfs_delayed_ref_root *delayed_refs = NULL;
 802         struct btrfs_delayed_ref_head *head;
 803         int info_level = 0;
 804         int ret;
 805         struct list_head prefs_delayed;
 806         struct list_head prefs;
 807         struct __prelim_ref *ref;
 808
 809         INIT_LIST_HEAD(&prefs);
 810         INIT_LIST_HEAD(&prefs_delayed);
 811
 812         key.objectid = bytenr;
 813         key.offset = (u64)-1;
 814         if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
 815                 key.type = BTRFS_METADATA_ITEM_KEY;
 816         else
 817                 key.type = BTRFS_EXTENT_ITEM_KEY;
 818
 819         path = btrfs_alloc_path();
 820         if (!path)
 821                 return -ENOMEM;
 822         if (!trans)
 823                 path->search_commit_root = 1;
 824
 825         /*
 826          * grab both a lock on the path and a lock on the delayed ref head.
 827          * We need both to get a consistent picture of how the refs look
 828          * at a specified point in time
 829          */
 830 again:
 831         head = NULL;
 832
 833         ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
 834         if (ret < 0)
 835                 goto out;
 836         BUG_ON(ret == 0);
 837
 838         if (trans) {
 839                 /*
 840                  * look if there are updates for this ref queued and lock the
 841                  * head
 842                  */
 843                 delayed_refs = &trans->transaction->delayed_refs;
 844                 spin_lock(&delayed_refs->lock);
 845                 head = btrfs_find_delayed_ref_head(trans, bytenr);
 846                 if (head) {
 847                         if (!mutex_trylock(&head->mutex)) {
 848                                 atomic_inc(&head->node.refs);
 849                                 spin_unlock(&delayed_refs->lock);
 850
 851                                 btrfs_release_path(path);
 852
 853                                 /*
 854                                  * Mutex was contended, block until it's
 855                                  * released and try again
 856                                  */
 857                                 mutex_lock(&head->mutex);
 858                                 mutex_unlock(&head->mutex);
 859                                 btrfs_put_delayed_ref(&head->node);
 860                                 goto again;
 861                         }
 862                         ret = __add_delayed_refs(head, time_seq,
 863                                                  &prefs_delayed);
 864                         mutex_unlock(&head->mutex);
 865                         if (ret) {
 866                                 spin_unlock(&delayed_refs->lock);
 867                                 goto out;
 868                         }
 869                 }
 870                 spin_unlock(&delayed_refs->lock);
 871         }
 872
 873         if (path->slots[0]) {
 874                 struct extent_buffer *leaf;
 875                 int slot;
 876
 877                 path->slots[0]--;
 878                 leaf = path->nodes[0];
 879                 slot = path->slots[0];
 880                 btrfs_item_key_to_cpu(leaf, &key, slot);
 881                 if (key.objectid == bytenr &&
 882                     (key.type == BTRFS_EXTENT_ITEM_KEY ||
 883                      key.type == BTRFS_METADATA_ITEM_KEY)) {
 884                         ret = __add_inline_refs(fs_info, path, bytenr,
 885                                                 &info_level, &prefs);
 886                         if (ret)
 887                                 goto out;
 888                         ret = __add_keyed_refs(fs_info, path, bytenr,
 889                                                info_level, &prefs);
 890                         if (ret)
 891                                 goto out;
 892                 }
 893         }
 894         btrfs_release_path(path);
 895
 896         list_splice_init(&prefs_delayed, &prefs);
 897
 898         ret = __add_missing_keys(fs_info, &prefs);
 899         if (ret)
 900                 goto out;
 901
 902         __merge_refs(&prefs, 1);
 903
 904         ret = __resolve_indirect_refs(fs_info, path, time_seq, &prefs,
 905                                       extent_item_pos);
 906         if (ret)
 907                 goto out;
 908
 909         __merge_refs(&prefs, 2);
 910
 911         while (!list_empty(&prefs)) {
 912                 ref = list_first_entry(&prefs, struct __prelim_ref, list);
 913                 WARN_ON(ref->count < 0);
 914                 if (ref->count && ref->root_id && ref->parent == 0) {
 915                         /* no parent == root of tree */
 916                         ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS);
 917                         if (ret < 0)
 918                                 goto out;
 919                 }
 920                 if (ref->count && ref->parent) {
 921                         struct extent_inode_elem *eie = NULL;
 922                         if (extent_item_pos && !ref->inode_list) {
 923                                 u32 bsz;
 924                                 struct extent_buffer *eb;
 925                                 bsz = btrfs_level_size(fs_info->extent_root,
 926                                                         info_level);
 927                                 eb = read_tree_block(fs_info->extent_root,
 928                                                            ref->parent, bsz, 0);
 929                                 if (!eb || !extent_buffer_uptodate(eb)) {
 930                                         free_extent_buffer(eb);
 931                                         ret = -EIO;
 932                                         goto out;
 933                                 }
 934                                 ret = find_extent_in_eb(eb, bytenr,
 935                                                         *extent_item_pos, &eie);
 936                                 free_extent_buffer(eb);
 937                                 if (ret < 0)
 938                                         goto out;
 939                                 ref->inode_list = eie;
 940                         }
 941                         ret = ulist_add_merge(refs, ref->parent,
 942                                               (uintptr_t)ref->inode_list,
 943                                               (u64 *)&eie, GFP_NOFS);
 944                         if (ret < 0)
 945                                 goto out;
 946                         if (!ret && extent_item_pos) {
 947                                 /*
 948                                  * we've recorded that parent, so we must extend
 949                                  * its inode list here
 950                                  */
 951                                 BUG_ON(!eie);
 952                                 while (eie->next)
 953                                         eie = eie->next;
 954                                 eie->next = ref->inode_list;
 955                         }
 956                 }
 957                 list_del(&ref->list);
 958                 kfree(ref);
 959         }
 960
 961 out:
 962         btrfs_free_path(path);
 963         while (!list_empty(&prefs)) {
 964                 ref = list_first_entry(&prefs, struct __prelim_ref, list);
 965                 list_del(&ref->list);
 966                 kfree(ref);
 967         }
 968         while (!list_empty(&prefs_delayed)) {
 969                 ref = list_first_entry(&prefs_delayed, struct __prelim_ref,
 970                                        list);
 971                 list_del(&ref->list);
 972                 kfree(ref);
 973         }
 974
 975         return ret;
 976 }
 977
 978 static void free_leaf_list(struct ulist *blocks)
 979 {
 980         struct ulist_node *node = NULL;
 981         struct extent_inode_elem *eie;
 982         struct extent_inode_elem *eie_next;
 983         struct ulist_iterator uiter;
 984
 985         ULIST_ITER_INIT(&uiter);
 986         while ((node = ulist_next(blocks, &uiter))) {
 987                 if (!node->aux)
 988                         continue;
 989                 eie = (struct extent_inode_elem *)(uintptr_t)node->aux;
 990                 for (; eie; eie = eie_next) {
 991                         eie_next = eie->next;
 992                         kfree(eie);
 993                 }
 994                 node->aux = 0;
 995         }
 996
 997         ulist_free(blocks);
 998 }
 999
1000 /*
1001  * Finds all leafs with a reference to the specified combination of bytenr and
1002  * offset. key_list_head will point to a list of corresponding keys (caller must
1003  * free each list element). The leafs will be stored in the leafs ulist, which
1004  * must be freed with ulist_free.
1005  *
1006  * returns 0 on success, <0 on error
1007  */
1008 static int btrfs_find_all_leafs(struct btrfs_trans_handle *trans,
1009                                 struct btrfs_fs_info *fs_info, u64 bytenr,
1010                                 u64 time_seq, struct ulist **leafs,
1011                                 const u64 *extent_item_pos)
1012 {
1013         struct ulist *tmp;
1014         int ret;
1015
1016         tmp = ulist_alloc(GFP_NOFS);
1017         if (!tmp)
1018                 return -ENOMEM;
1019         *leafs = ulist_alloc(GFP_NOFS);
1020         if (!*leafs) {
1021                 ulist_free(tmp);
1022                 return -ENOMEM;
1023         }
1024
1025         ret = find_parent_nodes(trans, fs_info, bytenr,
1026                                 time_seq, *leafs, tmp, extent_item_pos);
1027         ulist_free(tmp);
1028
1029         if (ret < 0 && ret != -ENOENT) {
1030                 free_leaf_list(*leafs);
1031                 return ret;
1032         }
1033
1034         return 0;
1035 }
1036
1037 /*
1038  * walk all backrefs for a given extent to find all roots that reference this
1039  * extent. Walking a backref means finding all extents that reference this
1040  * extent and in turn walk the backrefs of those, too. Naturally this is a
1041  * recursive process, but here it is implemented in an iterative fashion: We
1042  * find all referencing extents for the extent in question and put them on a
1043  * list. In turn, we find all referencing extents for those, further appending
1044  * to the list. The way we iterate the list allows adding more elements after
1045  * the current while iterating. The process stops when we reach the end of the
1046  * list. Found roots are added to the roots list.
1047  *
1048  * returns 0 on success, < 0 on error.
1049  */
1050 int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1051                                 struct btrfs_fs_info *fs_info, u64 bytenr,
1052                                 u64 time_seq, struct ulist **roots)
1053 {
1054         struct ulist *tmp;
1055         struct ulist_node *node = NULL;
1056         struct ulist_iterator uiter;
1057         int ret;
1058
1059         tmp = ulist_alloc(GFP_NOFS);
1060         if (!tmp)
1061                 return -ENOMEM;
1062         *roots = ulist_alloc(GFP_NOFS);
1063         if (!*roots) {
1064                 ulist_free(tmp);
1065                 return -ENOMEM;
1066         }
1067
1068         ULIST_ITER_INIT(&uiter);
1069         while (1) {
1070                 ret = find_parent_nodes(trans, fs_info, bytenr,
1071                                         time_seq, tmp, *roots, NULL);
1072                 if (ret < 0 && ret != -ENOENT) {
1073                         ulist_free(tmp);
1074                         ulist_free(*roots);
1075                         return ret;
1076                 }
1077                 node = ulist_next(tmp, &uiter);
1078                 if (!node)
1079                         break;
1080                 bytenr = node->val;
1081         }
1082
1083         ulist_free(tmp);
1084         return 0;
1085 }
1086
1087
1088 static int __inode_info(u64 inum, u64 ioff, u8 key_type,
1089                         struct btrfs_root *fs_root, struct btrfs_path *path,
1090                         struct btrfs_key *found_key)
1091 {
1092         int ret;
1093         struct btrfs_key key;
1094         struct extent_buffer *eb;
1095
1096         key.type = key_type;
1097         key.objectid = inum;
1098         key.offset = ioff;
1099
1100         ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
1101         if (ret < 0)
1102                 return ret;
1103
1104         eb = path->nodes[0];
1105         if (ret && path->slots[0] >= btrfs_header_nritems(eb)) {
1106                 ret = btrfs_next_leaf(fs_root, path);
1107                 if (ret)
1108                         return ret;
1109                 eb = path->nodes[0];
1110         }
1111
1112         btrfs_item_key_to_cpu(eb, found_key, path->slots[0]);
1113         if (found_key->type != key.type || found_key->objectid != key.objectid)
1114                 return 1;
1115
1116         return 0;
1117 }
1118
1119 /*
1120  * this makes the path point to (inum INODE_ITEM ioff)
1121  */
1122 int inode_item_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
1123                         struct btrfs_path *path)
1124 {
1125         struct btrfs_key key;
1126         return __inode_info(inum, ioff, BTRFS_INODE_ITEM_KEY, fs_root, path,
1127                                 &key);
1128 }
1129
1130 static int inode_ref_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
1131                                 struct btrfs_path *path,
1132                                 struct btrfs_key *found_key)
1133 {
1134         return __inode_info(inum, ioff, BTRFS_INODE_REF_KEY, fs_root, path,
1135                                 found_key);
1136 }
1137
1138 int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,
1139                           u64 start_off, struct btrfs_path *path,
1140                           struct btrfs_inode_extref **ret_extref,
1141                           u64 *found_off)
1142 {
1143         int ret, slot;
1144         struct btrfs_key key;
1145         struct btrfs_key found_key;
1146         struct btrfs_inode_extref *extref;
1147         struct extent_buffer *leaf;
1148         unsigned long ptr;
1149
1150         key.objectid = inode_objectid;
1151         btrfs_set_key_type(&key, BTRFS_INODE_EXTREF_KEY);
1152         key.offset = start_off;
1153
1154         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1155         if (ret < 0)
1156                 return ret;
1157
1158         while (1) {
1159                 leaf = path->nodes[0];
1160                 slot = path->slots[0];
1161                 if (slot >= btrfs_header_nritems(leaf)) {
1162                         /*
1163                          * If the item at offset is not found,
1164                          * btrfs_search_slot will point us to the slot
1165                          * where it should be inserted. In our case
1166                          * that will be the slot directly before the
1167                          * next INODE_REF_KEY_V2 item. In the case
1168                          * that we're pointing to the last slot in a
1169                          * leaf, we must move one leaf over.
1170                          */
1171                         ret = btrfs_next_leaf(root, path);
1172                         if (ret) {
1173                                 if (ret >= 1)
1174                                         ret = -ENOENT;
1175                                 break;
1176                         }
1177                         continue;
1178                 }
1179
1180                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1181
1182                 /*
1183                  * Check that we're still looking at an extended ref key for
1184                  * this particular objectid. If we have different
1185                  * objectid or type then there are no more to be found
1186                  * in the tree and we can exit.
1187                  */
1188                 ret = -ENOENT;
1189                 if (found_key.objectid != inode_objectid)
1190                         break;
1191                 if (btrfs_key_type(&found_key) != BTRFS_INODE_EXTREF_KEY)
1192                         break;
1193
1194                 ret = 0;
1195                 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1196                 extref = (struct btrfs_inode_extref *)ptr;
1197                 *ret_extref = extref;
1198                 if (found_off)
1199                         *found_off = found_key.offset;
1200                 break;
1201         }
1202
1203         return ret;
1204 }
1205
1206 /*
1207  * this iterates to turn a name (from iref/extref) into a full filesystem path.
1208  * Elements of the path are separated by '/' and the path is guaranteed to be
1209  * 0-terminated. the path is only given within the current file system.
1210  * Therefore, it never starts with a '/'. the caller is responsible to provide
1211  * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
1212  * the start point of the resulting string is returned. this pointer is within
1213  * dest, normally.
1214  * in case the path buffer would overflow, the pointer is decremented further
1215  * as if output was written to the buffer, though no more output is actually
1216  * generated. that way, the caller can determine how much space would be
1217  * required for the path to fit into the buffer. in that case, the returned
1218  * value will be smaller than dest. callers must check this!
1219  */
1220 char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
1221                         u32 name_len, unsigned long name_off,
1222                         struct extent_buffer *eb_in, u64 parent,
1223                         char *dest, u32 size)
1224 {
1225         int slot;
1226         u64 next_inum;
1227         int ret;
1228         s64 bytes_left = ((s64)size) - 1;
1229         struct extent_buffer *eb = eb_in;
1230         struct btrfs_key found_key;
1231         int leave_spinning = path->leave_spinning;
1232         struct btrfs_inode_ref *iref;
1233
1234         if (bytes_left >= 0)
1235                 dest[bytes_left] = '\0';
1236
1237         path->leave_spinning = 1;
1238         while (1) {
1239                 bytes_left -= name_len;
1240                 if (bytes_left >= 0)
1241                         read_extent_buffer(eb, dest + bytes_left,
1242                                            name_off, name_len);
1243                 if (eb != eb_in) {
1244                         btrfs_tree_read_unlock_blocking(eb);
1245                         free_extent_buffer(eb);
1246                 }
1247                 ret = inode_ref_info(parent, 0, fs_root, path, &found_key);
1248                 if (ret > 0)
1249                         ret = -ENOENT;
1250                 if (ret)
1251                         break;
1252
1253                 next_inum = found_key.offset;
1254
1255                 /* regular exit ahead */
1256                 if (parent == next_inum)
1257                         break;
1258
1259                 slot = path->slots[0];
1260                 eb = path->nodes[0];
1261                 /* make sure we can use eb after releasing the path */
1262                 if (eb != eb_in) {
1263                         atomic_inc(&eb->refs);
1264                         btrfs_tree_read_lock(eb);
1265                         btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1266                 }
1267                 btrfs_release_path(path);
1268                 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1269
1270                 name_len = btrfs_inode_ref_name_len(eb, iref);
1271                 name_off = (unsigned long)(iref + 1);
1272
1273                 parent = next_inum;
1274                 --bytes_left;
1275                 if (bytes_left >= 0)
1276                         dest[bytes_left] = '/';
1277         }
1278
1279         btrfs_release_path(path);
1280         path->leave_spinning = leave_spinning;
1281
1282         if (ret)
1283                 return ERR_PTR(ret);
1284
1285         return dest + bytes_left;
1286 }
1287
1288 /*
1289  * this makes the path point to (logical EXTENT_ITEM *)
1290  * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
1291  * tree blocks and <0 on error.
1292  */
1293 int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
1294                         struct btrfs_path *path, struct btrfs_key *found_key,
1295                         u64 *flags_ret)
1296 {
1297         int ret;
1298         u64 flags;
1299         u64 size = 0;
1300         u32 item_size;
1301         struct extent_buffer *eb;
1302         struct btrfs_extent_item *ei;
1303         struct btrfs_key key;
1304
1305         if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1306                 key.type = BTRFS_METADATA_ITEM_KEY;
1307         else
1308                 key.type = BTRFS_EXTENT_ITEM_KEY;
1309         key.objectid = logical;
1310         key.offset = (u64)-1;
1311
1312         ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
1313         if (ret < 0)
1314                 return ret;
1315         ret = btrfs_previous_item(fs_info->extent_root, path,
1316                                         0, BTRFS_EXTENT_ITEM_KEY);
1317         if (ret < 0)
1318                 return ret;
1319
1320         btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]);
1321         if (found_key->type == BTRFS_METADATA_ITEM_KEY)
1322                 size = fs_info->extent_root->leafsize;
1323         else if (found_key->type == BTRFS_EXTENT_ITEM_KEY)
1324                 size = found_key->offset;
1325
1326         if ((found_key->type != BTRFS_EXTENT_ITEM_KEY &&
1327              found_key->type != BTRFS_METADATA_ITEM_KEY) ||
1328             found_key->objectid > logical ||
1329             found_key->objectid + size <= logical) {
1330                 pr_debug("logical %llu is not within any extent\n", logical);
1331                 return -ENOENT;
1332         }
1333
1334         eb = path->nodes[0];
1335         item_size = btrfs_item_size_nr(eb, path->slots[0]);
1336         BUG_ON(item_size < sizeof(*ei));
1337
1338         ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
1339         flags = btrfs_extent_flags(eb, ei);
1340
1341         pr_debug("logical %llu is at position %llu within the extent (%llu "
1342                  "EXTENT_ITEM %llu) flags %#llx size %u\n",
1343                  logical, logical - found_key->objectid, found_key->objectid,
1344                  found_key->offset, flags, item_size);
1345
1346         WARN_ON(!flags_ret);
1347         if (flags_ret) {
1348                 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1349                         *flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK;
1350                 else if (flags & BTRFS_EXTENT_FLAG_DATA)
1351                         *flags_ret = BTRFS_EXTENT_FLAG_DATA;
1352                 else
1353                         BUG_ON(1);
1354                 return 0;
1355         }
1356
1357         return -EIO;
1358 }
1359
1360 /*
1361  * helper function to iterate extent inline refs. ptr must point to a 0 value
1362  * for the first call and may be modified. it is used to track state.
1363  * if more refs exist, 0 is returned and the next call to
1364  * __get_extent_inline_ref must pass the modified ptr parameter to get the
1365  * next ref. after the last ref was processed, 1 is returned.
1366  * returns <0 on error
1367  */
1368 static int __get_extent_inline_ref(unsigned long *ptr, struct extent_buffer *eb,
1369                                 struct btrfs_extent_item *ei, u32 item_size,
1370                                 struct btrfs_extent_inline_ref **out_eiref,
1371                                 int *out_type)
1372 {
1373         unsigned long end;
1374         u64 flags;
1375         struct btrfs_tree_block_info *info;
1376
1377         if (!*ptr) {
1378                 /* first call */
1379                 flags = btrfs_extent_flags(eb, ei);
1380                 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1381                         info = (struct btrfs_tree_block_info *)(ei + 1);
1382                         *out_eiref =
1383                                 (struct btrfs_extent_inline_ref *)(info + 1);
1384                 } else {
1385                         *out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1);
1386                 }
1387                 *ptr = (unsigned long)*out_eiref;
1388                 if ((void *)*ptr >= (void *)ei + item_size)
1389                         return -ENOENT;
1390         }
1391
1392         end = (unsigned long)ei + item_size;
1393         *out_eiref = (struct btrfs_extent_inline_ref *)*ptr;
1394         *out_type = btrfs_extent_inline_ref_type(eb, *out_eiref);
1395
1396         *ptr += btrfs_extent_inline_ref_size(*out_type);
1397         WARN_ON(*ptr > end);
1398         if (*ptr == end)
1399                 return 1; /* last */
1400
1401         return 0;
1402 }
1403
1404 /*
1405  * reads the tree block backref for an extent. tree level and root are returned
1406  * through out_level and out_root. ptr must point to a 0 value for the first
1407  * call and may be modified (see __get_extent_inline_ref comment).
1408  * returns 0 if data was provided, 1 if there was no more data to provide or
1409  * <0 on error.
1410  */
1411 int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
1412                                 struct btrfs_extent_item *ei, u32 item_size,
1413                                 u64 *out_root, u8 *out_level)
1414 {
1415         int ret;
1416         int type;
1417         struct btrfs_tree_block_info *info;
1418         struct btrfs_extent_inline_ref *eiref;
1419
1420         if (*ptr == (unsigned long)-1)
1421                 return 1;
1422
1423         while (1) {
1424                 ret = __get_extent_inline_ref(ptr, eb, ei, item_size,
1425                                                 &eiref, &type);
1426                 if (ret < 0)
1427                         return ret;
1428
1429                 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1430                     type == BTRFS_SHARED_BLOCK_REF_KEY)
1431                         break;
1432
1433                 if (ret == 1)
1434                         return 1;
1435         }
1436
1437         /* we can treat both ref types equally here */
1438         info = (struct btrfs_tree_block_info *)(ei + 1);
1439         *out_root = btrfs_extent_inline_ref_offset(eb, eiref);
1440         *out_level = btrfs_tree_block_level(eb, info);
1441
1442         if (ret == 1)
1443                 *ptr = (unsigned long)-1;
1444
1445         return 0;
1446 }
1447
1448 static int iterate_leaf_refs(struct extent_inode_elem *inode_list,
1449                                 u64 root, u64 extent_item_objectid,
1450                                 iterate_extent_inodes_t *iterate, void *ctx)
1451 {
1452         struct extent_inode_elem *eie;
1453         int ret = 0;
1454
1455         for (eie = inode_list; eie; eie = eie->next) {
1456                 pr_debug("ref for %llu resolved, key (%llu EXTEND_DATA %llu), "
1457                          "root %llu\n", extent_item_objectid,
1458                          eie->inum, eie->offset, root);
1459                 ret = iterate(eie->inum, eie->offset, root, ctx);
1460                 if (ret) {
1461                         pr_debug("stopping iteration for %llu due to ret=%d\n",
1462                                  extent_item_objectid, ret);
1463                         break;
1464                 }
1465         }
1466
1467         return ret;
1468 }
1469
1470 /*
1471  * calls iterate() for every inode that references the extent identified by
1472  * the given parameters.
1473  * when the iterator function returns a non-zero value, iteration stops.
1474  */
1475 int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
1476                                 u64 extent_item_objectid, u64 extent_item_pos,
1477                                 int search_commit_root,
1478                                 iterate_extent_inodes_t *iterate, void *ctx)
1479 {
1480         int ret;
1481         struct btrfs_trans_handle *trans = NULL;
1482         struct ulist *refs = NULL;
1483         struct ulist *roots = NULL;
1484         struct ulist_node *ref_node = NULL;
1485         struct ulist_node *root_node = NULL;
1486         struct seq_list tree_mod_seq_elem = {};
1487         struct ulist_iterator ref_uiter;
1488         struct ulist_iterator root_uiter;
1489
1490         pr_debug("resolving all inodes for extent %llu\n",
1491                         extent_item_objectid);
1492
1493         if (!search_commit_root) {
1494                 trans = btrfs_join_transaction(fs_info->extent_root);
1495                 if (IS_ERR(trans))
1496                         return PTR_ERR(trans);
1497                 btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1498         }
1499
1500         ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid,
1501                                    tree_mod_seq_elem.seq, &refs,
1502                                    &extent_item_pos);
1503         if (ret)
1504                 goto out;
1505
1506         ULIST_ITER_INIT(&ref_uiter);
1507         while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) {
1508                 ret = btrfs_find_all_roots(trans, fs_info, ref_node->val,
1509                                            tree_mod_seq_elem.seq, &roots);
1510                 if (ret)
1511                         break;
1512                 ULIST_ITER_INIT(&root_uiter);
1513                 while (!ret && (root_node = ulist_next(roots, &root_uiter))) {
1514                         pr_debug("root %llu references leaf %llu, data list "
1515                                  "%#llx\n", root_node->val, ref_node->val,
1516                                  ref_node->aux);
1517                         ret = iterate_leaf_refs((struct extent_inode_elem *)
1518                                                 (uintptr_t)ref_node->aux,
1519                                                 root_node->val,
1520                                                 extent_item_objectid,
1521                                                 iterate, ctx);
1522                 }
1523                 ulist_free(roots);
1524         }
1525
1526         free_leaf_list(refs);
1527 out:
1528         if (!search_commit_root) {
1529                 btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1530                 btrfs_end_transaction(trans, fs_info->extent_root);
1531         }
1532
1533         return ret;
1534 }
1535
1536 int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
1537                                 struct btrfs_path *path,
1538                                 iterate_extent_inodes_t *iterate, void *ctx)
1539 {
1540         int ret;
1541         u64 extent_item_pos;
1542         u64 flags = 0;
1543         struct btrfs_key found_key;
1544         int search_commit_root = path->search_commit_root;
1545
1546         ret = extent_from_logical(fs_info, logical, path, &found_key, &flags);
1547         btrfs_release_path(path);
1548         if (ret < 0)
1549                 return ret;
1550         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1551                 return -EINVAL;
1552
1553         extent_item_pos = logical - found_key.objectid;
1554         ret = iterate_extent_inodes(fs_info, found_key.objectid,
1555                                         extent_item_pos, search_commit_root,
1556                                         iterate, ctx);
1557
1558         return ret;
1559 }
1560
1561 typedef int (iterate_irefs_t)(u64 parent, u32 name_len, unsigned long name_off,
1562                               struct extent_buffer *eb, void *ctx);
1563
1564 static int iterate_inode_refs(u64 inum, struct btrfs_root *fs_root,
1565                               struct btrfs_path *path,
1566                               iterate_irefs_t *iterate, void *ctx)
1567 {
1568         int ret = 0;
1569         int slot;
1570         u32 cur;
1571         u32 len;
1572         u32 name_len;
1573         u64 parent = 0;
1574         int found = 0;
1575         struct extent_buffer *eb;
1576         struct btrfs_item *item;
1577         struct btrfs_inode_ref *iref;
1578         struct btrfs_key found_key;
1579
1580         while (!ret) {
1581                 path->leave_spinning = 1;
1582                 ret = inode_ref_info(inum, parent ? parent+1 : 0, fs_root, path,
1583                                      &found_key);
1584                 if (ret < 0)
1585                         break;
1586                 if (ret) {
1587                         ret = found ? 0 : -ENOENT;
1588                         break;
1589                 }
1590                 ++found;
1591
1592                 parent = found_key.offset;
1593                 slot = path->slots[0];
1594                 eb = path->nodes[0];
1595                 /* make sure we can use eb after releasing the path */
1596                 atomic_inc(&eb->refs);
1597                 btrfs_tree_read_lock(eb);
1598                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1599                 btrfs_release_path(path);
1600
1601                 item = btrfs_item_nr(eb, slot);
1602                 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1603
1604                 for (cur = 0; cur < btrfs_item_size(eb, item); cur += len) {
1605                         name_len = btrfs_inode_ref_name_len(eb, iref);
1606                         /* path must be released before calling iterate()! */
1607                         pr_debug("following ref at offset %u for inode %llu in "
1608                                  "tree %llu\n", cur, found_key.objectid,
1609                                  fs_root->objectid);
1610                         ret = iterate(parent, name_len,
1611                                       (unsigned long)(iref + 1), eb, ctx);
1612                         if (ret)
1613                                 break;
1614                         len = sizeof(*iref) + name_len;
1615                         iref = (struct btrfs_inode_ref *)((char *)iref + len);
1616                 }
1617                 btrfs_tree_read_unlock_blocking(eb);
1618                 free_extent_buffer(eb);
1619         }
1620
1621         btrfs_release_path(path);
1622
1623         return ret;
1624 }
1625
1626 static int iterate_inode_extrefs(u64 inum, struct btrfs_root *fs_root,
1627                                  struct btrfs_path *path,
1628                                  iterate_irefs_t *iterate, void *ctx)
1629 {
1630         int ret;
1631         int slot;
1632         u64 offset = 0;
1633         u64 parent;
1634         int found = 0;
1635         struct extent_buffer *eb;
1636         struct btrfs_inode_extref *extref;
1637         struct extent_buffer *leaf;
1638         u32 item_size;
1639         u32 cur_offset;
1640         unsigned long ptr;
1641
1642         while (1) {
1643                 ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref,
1644                                             &offset);
1645                 if (ret < 0)
1646                         break;
1647                 if (ret) {
1648                         ret = found ? 0 : -ENOENT;
1649                         break;
1650                 }
1651                 ++found;
1652
1653                 slot = path->slots[0];
1654                 eb = path->nodes[0];
1655                 /* make sure we can use eb after releasing the path */
1656                 atomic_inc(&eb->refs);
1657
1658                 btrfs_tree_read_lock(eb);
1659                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1660                 btrfs_release_path(path);
1661
1662                 leaf = path->nodes[0];
1663                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1664                 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1665                 cur_offset = 0;
1666
1667                 while (cur_offset < item_size) {
1668                         u32 name_len;
1669
1670                         extref = (struct btrfs_inode_extref *)(ptr + cur_offset);
1671                         parent = btrfs_inode_extref_parent(eb, extref);
1672                         name_len = btrfs_inode_extref_name_len(eb, extref);
1673                         ret = iterate(parent, name_len,
1674                                       (unsigned long)&extref->name, eb, ctx);
1675                         if (ret)
1676                                 break;
1677
1678                         cur_offset += btrfs_inode_extref_name_len(leaf, extref);
1679                         cur_offset += sizeof(*extref);
1680                 }
1681                 btrfs_tree_read_unlock_blocking(eb);
1682                 free_extent_buffer(eb);
1683
1684                 offset++;
1685         }
1686
1687         btrfs_release_path(path);
1688
1689         return ret;
1690 }
1691
1692 static int iterate_irefs(u64 inum, struct btrfs_root *fs_root,
1693                          struct btrfs_path *path, iterate_irefs_t *iterate,
1694                          void *ctx)
1695 {
1696         int ret;
1697         int found_refs = 0;
1698
1699         ret = iterate_inode_refs(inum, fs_root, path, iterate, ctx);
1700         if (!ret)
1701                 ++found_refs;
1702         else if (ret != -ENOENT)
1703                 return ret;
1704
1705         ret = iterate_inode_extrefs(inum, fs_root, path, iterate, ctx);
1706         if (ret == -ENOENT && found_refs)
1707                 return 0;
1708
1709         return ret;
1710 }
1711
1712 /*
1713  * returns 0 if the path could be dumped (probably truncated)
1714  * returns <0 in case of an error
1715  */
1716 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
1717                          struct extent_buffer *eb, void *ctx)
1718 {
1719         struct inode_fs_paths *ipath = ctx;
1720         char *fspath;
1721         char *fspath_min;
1722         int i = ipath->fspath->elem_cnt;
1723         const int s_ptr = sizeof(char *);
1724         u32 bytes_left;
1725
1726         bytes_left = ipath->fspath->bytes_left > s_ptr ?
1727                                         ipath->fspath->bytes_left - s_ptr : 0;
1728
1729         fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
1730         fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len,
1731                                    name_off, eb, inum, fspath_min, bytes_left);
1732         if (IS_ERR(fspath))
1733                 return PTR_ERR(fspath);
1734
1735         if (fspath > fspath_min) {
1736                 ipath->fspath->val[i] = (u64)(unsigned long)fspath;
1737                 ++ipath->fspath->elem_cnt;
1738                 ipath->fspath->bytes_left = fspath - fspath_min;
1739         } else {
1740                 ++ipath->fspath->elem_missed;
1741                 ipath->fspath->bytes_missing += fspath_min - fspath;
1742                 ipath->fspath->bytes_left = 0;
1743         }
1744
1745         return 0;
1746 }
1747
1748 /*
1749  * this dumps all file system paths to the inode into the ipath struct, provided
1750  * is has been created large enough. each path is zero-terminated and accessed
1751  * from ipath->fspath->val[i].
1752  * when it returns, there are ipath->fspath->elem_cnt number of paths available
1753  * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the
1754  * number of missed paths in recored in ipath->fspath->elem_missed, otherwise,
1755  * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
1756  * have been needed to return all paths.
1757  */
1758 int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
1759 {
1760         return iterate_irefs(inum, ipath->fs_root, ipath->btrfs_path,
1761                              inode_to_path, ipath);
1762 }
1763
1764 struct btrfs_data_container *init_data_container(u32 total_bytes)
1765 {
1766         struct btrfs_data_container *data;
1767         size_t alloc_bytes;
1768
1769         alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
1770         data = vmalloc(alloc_bytes);
1771         if (!data)
1772                 return ERR_PTR(-ENOMEM);
1773
1774         if (total_bytes >= sizeof(*data)) {
1775                 data->bytes_left = total_bytes - sizeof(*data);
1776                 data->bytes_missing = 0;
1777         } else {
1778                 data->bytes_missing = sizeof(*data) - total_bytes;
1779                 data->bytes_left = 0;
1780         }
1781
1782         data->elem_cnt = 0;
1783         data->elem_missed = 0;
1784
1785         return data;
1786 }
1787
1788 /*
1789  * allocates space to return multiple file system paths for an inode.
1790  * total_bytes to allocate are passed, note that space usable for actual path
1791  * information will be total_bytes - sizeof(struct inode_fs_paths).
1792  * the returned pointer must be freed with free_ipath() in the end.
1793  */
1794 struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
1795                                         struct btrfs_path *path)
1796 {
1797         struct inode_fs_paths *ifp;
1798         struct btrfs_data_container *fspath;
1799
1800         fspath = init_data_container(total_bytes);
1801         if (IS_ERR(fspath))
1802                 return (void *)fspath;
1803
1804         ifp = kmalloc(sizeof(*ifp), GFP_NOFS);
1805         if (!ifp) {
1806                 kfree(fspath);
1807                 return ERR_PTR(-ENOMEM);
1808         }
1809
1810         ifp->btrfs_path = path;
1811         ifp->fspath = fspath;
1812         ifp->fs_root = fs_root;
1813
1814         return ifp;
1815 }
1816
1817 void free_ipath(struct inode_fs_paths *ipath)
1818 {
1819         if (!ipath)
1820                 return;
1821         vfree(ipath->fspath);
1822         kfree(ipath);
1823 }