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         ref = kmem_cache_alloc(btrfs_prelim_ref_cache, gfp_mask);
 189         if (!ref)
 190                 return -ENOMEM;
 191
 192         ref->root_id = root_id;
 193         if (key)
 194                 ref->key_for_search = *key;
 195         else
 196                 memset(&ref->key_for_search, 0, sizeof(ref->key_for_search));
 197
 198         ref->inode_list = NULL;
 199         ref->level = level;
 200         ref->count = count;
 201         ref->parent = parent;
 202         ref->wanted_disk_byte = wanted_disk_byte;
 203         list_add_tail(&ref->list, head);
 204
 205         return 0;
 206 }
 207
 208 static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path,
 209                                 struct ulist *parents, int level,
 210                                 struct btrfs_key *key_for_search, u64 time_seq,
 211                                 u64 wanted_disk_byte,
 212                                 const u64 *extent_item_pos)
 213 {
 214         int ret = 0;
 215         int slot;
 216         struct extent_buffer *eb;
 217         struct btrfs_key key;
 218         struct btrfs_file_extent_item *fi;
 219         struct extent_inode_elem *eie = NULL, *old = NULL;
 220         u64 disk_byte;
 221
 222         if (level != 0) {
 223                 eb = path->nodes[level];
 224                 ret = ulist_add(parents, eb->start, 0, GFP_NOFS);
 225                 if (ret < 0)
 226                         return ret;
 227                 return 0;
 228         }
 229
 230         /*
 231          * We normally enter this function with the path already pointing to
 232          * the first item to check. But sometimes, we may enter it with
 233          * slot==nritems. In that case, go to the next leaf before we continue.
 234          */
 235         if (path->slots[0] >= btrfs_header_nritems(path->nodes[0]))
 236                 ret = btrfs_next_old_leaf(root, path, time_seq);
 237
 238         while (!ret) {
 239                 eb = path->nodes[0];
 240                 slot = path->slots[0];
 241
 242                 btrfs_item_key_to_cpu(eb, &key, slot);
 243
 244                 if (key.objectid != key_for_search->objectid ||
 245                     key.type != BTRFS_EXTENT_DATA_KEY)
 246                         break;
 247
 248                 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
 249                 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
 250
 251                 if (disk_byte == wanted_disk_byte) {
 252                         eie = NULL;
 253                         old = NULL;
 254                         if (extent_item_pos) {
 255                                 ret = check_extent_in_eb(&key, eb, fi,
 256                                                 *extent_item_pos,
 257                                                 &eie);
 258                                 if (ret < 0)
 259                                         break;
 260                         }
 261                         if (ret > 0)
 262                                 goto next;
 263                         ret = ulist_add_merge(parents, eb->start,
 264                                               (uintptr_t)eie,
 265                                               (u64 *)&old, GFP_NOFS);
 266                         if (ret < 0)
 267                                 break;
 268                         if (!ret && extent_item_pos) {
 269                                 while (old->next)
 270                                         old = old->next;
 271                                 old->next = eie;
 272                         }
 273                 }
 274 next:
 275                 ret = btrfs_next_old_item(root, path, time_seq);
 276         }
 277
 278         if (ret > 0)
 279                 ret = 0;
 280         return ret;
 281 }
 282
 283 /*
 284  * resolve an indirect backref in the form (root_id, key, level)
 285  * to a logical address
 286  */
 287 static int __resolve_indirect_ref(struct btrfs_fs_info *fs_info,
 288                                   struct btrfs_path *path, u64 time_seq,
 289                                   struct __prelim_ref *ref,
 290                                   struct ulist *parents,
 291                                   const u64 *extent_item_pos)
 292 {
 293         struct btrfs_root *root;
 294         struct btrfs_key root_key;
 295         struct extent_buffer *eb;
 296         int ret = 0;
 297         int root_level;
 298         int level = ref->level;
 299
 300         root_key.objectid = ref->root_id;
 301         root_key.type = BTRFS_ROOT_ITEM_KEY;
 302         root_key.offset = (u64)-1;
 303         root = btrfs_read_fs_root_no_name(fs_info, &root_key);
 304         if (IS_ERR(root)) {
 305                 ret = PTR_ERR(root);
 306                 goto out;
 307         }
 308
 309         root_level = btrfs_old_root_level(root, time_seq);
 310
 311         if (root_level + 1 == level)
 312                 goto out;
 313
 314         path->lowest_level = level;
 315         ret = btrfs_search_old_slot(root, &ref->key_for_search, path, time_seq);
 316         pr_debug("search slot in root %llu (level %d, ref count %d) returned "
 317                  "%d for key (%llu %u %llu)\n",
 318                  ref->root_id, level, ref->count, ret,
 319                  ref->key_for_search.objectid, ref->key_for_search.type,
 320                  ref->key_for_search.offset);
 321         if (ret < 0)
 322                 goto out;
 323
 324         eb = path->nodes[level];
 325         while (!eb) {
 326                 if (!level) {
 327                         WARN_ON(1);
 328                         ret = 1;
 329                         goto out;
 330                 }
 331                 level--;
 332                 eb = path->nodes[level];
 333         }
 334
 335         ret = add_all_parents(root, path, parents, level, &ref->key_for_search,
 336                                 time_seq, ref->wanted_disk_byte,
 337                                 extent_item_pos);
 338 out:
 339         path->lowest_level = 0;
 340         btrfs_release_path(path);
 341         return ret;
 342 }
 343
 344 /*
 345  * resolve all indirect backrefs from the list
 346  */
 347 static int __resolve_indirect_refs(struct btrfs_fs_info *fs_info,
 348                                    struct btrfs_path *path, u64 time_seq,
 349                                    struct list_head *head,
 350                                    const u64 *extent_item_pos)
 351 {
 352         int err;
 353         int ret = 0;
 354         struct __prelim_ref *ref;
 355         struct __prelim_ref *ref_safe;
 356         struct __prelim_ref *new_ref;
 357         struct ulist *parents;
 358         struct ulist_node *node;
 359         struct ulist_iterator uiter;
 360
 361         parents = ulist_alloc(GFP_NOFS);
 362         if (!parents)
 363                 return -ENOMEM;
 364
 365         /*
 366          * _safe allows us to insert directly after the current item without
 367          * iterating over the newly inserted items.
 368          * we're also allowed to re-assign ref during iteration.
 369          */
 370         list_for_each_entry_safe(ref, ref_safe, head, list) {
 371                 if (ref->parent)        /* already direct */
 372                         continue;
 373                 if (ref->count == 0)
 374                         continue;
 375                 err = __resolve_indirect_ref(fs_info, path, time_seq, ref,
 376                                              parents, extent_item_pos);
 377                 if (err == -ENOMEM)
 378                         goto out;
 379                 if (err)
 380                         continue;
 381
 382                 /* we put the first parent into the ref at hand */
 383                 ULIST_ITER_INIT(&uiter);
 384                 node = ulist_next(parents, &uiter);
 385                 ref->parent = node ? node->val : 0;
 386                 ref->inode_list = node ?
 387                         (struct extent_inode_elem *)(uintptr_t)node->aux : NULL;
 388
 389                 /* additional parents require new refs being added here */
 390                 while ((node = ulist_next(parents, &uiter))) {
 391                         new_ref = kmem_cache_alloc(btrfs_prelim_ref_cache,
 392                                                    GFP_NOFS);
 393                         if (!new_ref) {
 394                                 ret = -ENOMEM;
 395                                 goto out;
 396                         }
 397                         memcpy(new_ref, ref, sizeof(*ref));
 398                         new_ref->parent = node->val;
 399                         new_ref->inode_list = (struct extent_inode_elem *)
 400                                                         (uintptr_t)node->aux;
 401                         list_add(&new_ref->list, &ref->list);
 402                 }
 403                 ulist_reinit(parents);
 404         }
 405 out:
 406         ulist_free(parents);
 407         return ret;
 408 }
 409
 410 static inline int ref_for_same_block(struct __prelim_ref *ref1,
 411                                      struct __prelim_ref *ref2)
 412 {
 413         if (ref1->level != ref2->level)
 414                 return 0;
 415         if (ref1->root_id != ref2->root_id)
 416                 return 0;
 417         if (ref1->key_for_search.type != ref2->key_for_search.type)
 418                 return 0;
 419         if (ref1->key_for_search.objectid != ref2->key_for_search.objectid)
 420                 return 0;
 421         if (ref1->key_for_search.offset != ref2->key_for_search.offset)
 422                 return 0;
 423         if (ref1->parent != ref2->parent)
 424                 return 0;
 425
 426         return 1;
 427 }
 428
 429 /*
 430  * read tree blocks and add keys where required.
 431  */
 432 static int __add_missing_keys(struct btrfs_fs_info *fs_info,
 433                               struct list_head *head)
 434 {
 435         struct list_head *pos;
 436         struct extent_buffer *eb;
 437
 438         list_for_each(pos, head) {
 439                 struct __prelim_ref *ref;
 440                 ref = list_entry(pos, struct __prelim_ref, list);
 441
 442                 if (ref->parent)
 443                         continue;
 444                 if (ref->key_for_search.type)
 445                         continue;
 446                 BUG_ON(!ref->wanted_disk_byte);
 447                 eb = read_tree_block(fs_info->tree_root, ref->wanted_disk_byte,
 448                                      fs_info->tree_root->leafsize, 0);
 449                 if (!eb || !extent_buffer_uptodate(eb)) {
 450                         free_extent_buffer(eb);
 451                         return -EIO;
 452                 }
 453                 btrfs_tree_read_lock(eb);
 454                 if (btrfs_header_level(eb) == 0)
 455                         btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
 456                 else
 457                         btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
 458                 btrfs_tree_read_unlock(eb);
 459                 free_extent_buffer(eb);
 460         }
 461         return 0;
 462 }
 463
 464 /*
 465  * merge two lists of backrefs and adjust counts accordingly
 466  *
 467  * mode = 1: merge identical keys, if key is set
 468  *    FIXME: if we add more keys in __add_prelim_ref, we can merge more here.
 469  *           additionally, we could even add a key range for the blocks we
 470  *           looked into to merge even more (-> replace unresolved refs by those
 471  *           having a parent).
 472  * mode = 2: merge identical parents
 473  */
 474 static void __merge_refs(struct list_head *head, int mode)
 475 {
 476         struct list_head *pos1;
 477
 478         list_for_each(pos1, head) {
 479                 struct list_head *n2;
 480                 struct list_head *pos2;
 481                 struct __prelim_ref *ref1;
 482
 483                 ref1 = list_entry(pos1, struct __prelim_ref, list);
 484
 485                 for (pos2 = pos1->next, n2 = pos2->next; pos2 != head;
 486                      pos2 = n2, n2 = pos2->next) {
 487                         struct __prelim_ref *ref2;
 488                         struct __prelim_ref *xchg;
 489                         struct extent_inode_elem *eie;
 490
 491                         ref2 = list_entry(pos2, struct __prelim_ref, list);
 492
 493                         if (mode == 1) {
 494                                 if (!ref_for_same_block(ref1, ref2))
 495                                         continue;
 496                                 if (!ref1->parent && ref2->parent) {
 497                                         xchg = ref1;
 498                                         ref1 = ref2;
 499                                         ref2 = xchg;
 500                                 }
 501                         } else {
 502                                 if (ref1->parent != ref2->parent)
 503                                         continue;
 504                         }
 505
 506                         eie = ref1->inode_list;
 507                         while (eie && eie->next)
 508                                 eie = eie->next;
 509                         if (eie)
 510                                 eie->next = ref2->inode_list;
 511                         else
 512                                 ref1->inode_list = ref2->inode_list;
 513                         ref1->count += ref2->count;
 514
 515                         list_del(&ref2->list);
 516                         kmem_cache_free(btrfs_prelim_ref_cache, ref2);
 517                 }
 518
 519         }
 520 }
 521
 522 /*
 523  * add all currently queued delayed refs from this head whose seq nr is
 524  * smaller or equal that seq to the list
 525  */
 526 static int __add_delayed_refs(struct btrfs_delayed_ref_head *head, u64 seq,
 527                               struct list_head *prefs)
 528 {
 529         struct btrfs_delayed_extent_op *extent_op = head->extent_op;
 530         struct rb_node *n = &head->node.rb_node;
 531         struct btrfs_key key;
 532         struct btrfs_key op_key = {0};
 533         int sgn;
 534         int ret = 0;
 535
 536         if (extent_op && extent_op->update_key)
 537                 btrfs_disk_key_to_cpu(&op_key, &extent_op->key);
 538
 539         while ((n = rb_prev(n))) {
 540                 struct btrfs_delayed_ref_node *node;
 541                 node = rb_entry(n, struct btrfs_delayed_ref_node,
 542                                 rb_node);
 543                 if (node->bytenr != head->node.bytenr)
 544                         break;
 545                 WARN_ON(node->is_head);
 546
 547                 if (node->seq > seq)
 548                         continue;
 549
 550                 switch (node->action) {
 551                 case BTRFS_ADD_DELAYED_EXTENT:
 552                 case BTRFS_UPDATE_DELAYED_HEAD:
 553                         WARN_ON(1);
 554                         continue;
 555                 case BTRFS_ADD_DELAYED_REF:
 556                         sgn = 1;
 557                         break;
 558                 case BTRFS_DROP_DELAYED_REF:
 559                         sgn = -1;
 560                         break;
 561                 default:
 562                         BUG_ON(1);
 563                 }
 564                 switch (node->type) {
 565                 case BTRFS_TREE_BLOCK_REF_KEY: {
 566                         struct btrfs_delayed_tree_ref *ref;
 567
 568                         ref = btrfs_delayed_node_to_tree_ref(node);
 569                         ret = __add_prelim_ref(prefs, ref->root, &op_key,
 570                                                ref->level + 1, 0, node->bytenr,
 571                                                node->ref_mod * sgn, GFP_ATOMIC);
 572                         break;
 573                 }
 574                 case BTRFS_SHARED_BLOCK_REF_KEY: {
 575                         struct btrfs_delayed_tree_ref *ref;
 576
 577                         ref = btrfs_delayed_node_to_tree_ref(node);
 578                         ret = __add_prelim_ref(prefs, ref->root, NULL,
 579                                                ref->level + 1, ref->parent,
 580                                                node->bytenr,
 581                                                node->ref_mod * sgn, GFP_ATOMIC);
 582                         break;
 583                 }
 584                 case BTRFS_EXTENT_DATA_REF_KEY: {
 585                         struct btrfs_delayed_data_ref *ref;
 586                         ref = btrfs_delayed_node_to_data_ref(node);
 587
 588                         key.objectid = ref->objectid;
 589                         key.type = BTRFS_EXTENT_DATA_KEY;
 590                         key.offset = ref->offset;
 591                         ret = __add_prelim_ref(prefs, ref->root, &key, 0, 0,
 592                                                node->bytenr,
 593                                                node->ref_mod * sgn, GFP_ATOMIC);
 594                         break;
 595                 }
 596                 case BTRFS_SHARED_DATA_REF_KEY: {
 597                         struct btrfs_delayed_data_ref *ref;
 598
 599                         ref = btrfs_delayed_node_to_data_ref(node);
 600
 601                         key.objectid = ref->objectid;
 602                         key.type = BTRFS_EXTENT_DATA_KEY;
 603                         key.offset = ref->offset;
 604                         ret = __add_prelim_ref(prefs, ref->root, &key, 0,
 605                                                ref->parent, node->bytenr,
 606                                                node->ref_mod * sgn, GFP_ATOMIC);
 607                         break;
 608                 }
 609                 default:
 610                         WARN_ON(1);
 611                 }
 612                 if (ret)
 613                         return ret;
 614         }
 615
 616         return 0;
 617 }
 618
 619 /*
 620  * add all inline backrefs for bytenr to the list
 621  */
 622 static int __add_inline_refs(struct btrfs_fs_info *fs_info,
 623                              struct btrfs_path *path, u64 bytenr,
 624                              int *info_level, struct list_head *prefs)
 625 {
 626         int ret = 0;
 627         int slot;
 628         struct extent_buffer *leaf;
 629         struct btrfs_key key;
 630         struct btrfs_key found_key;
 631         unsigned long ptr;
 632         unsigned long end;
 633         struct btrfs_extent_item *ei;
 634         u64 flags;
 635         u64 item_size;
 636
 637         /*
 638          * enumerate all inline refs
 639          */
 640         leaf = path->nodes[0];
 641         slot = path->slots[0];
 642
 643         item_size = btrfs_item_size_nr(leaf, slot);
 644         BUG_ON(item_size < sizeof(*ei));
 645
 646         ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
 647         flags = btrfs_extent_flags(leaf, ei);
 648         btrfs_item_key_to_cpu(leaf, &found_key, slot);
 649
 650         ptr = (unsigned long)(ei + 1);
 651         end = (unsigned long)ei + item_size;
 652
 653         if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
 654             flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
 655                 struct btrfs_tree_block_info *info;
 656
 657                 info = (struct btrfs_tree_block_info *)ptr;
 658                 *info_level = btrfs_tree_block_level(leaf, info);
 659                 ptr += sizeof(struct btrfs_tree_block_info);
 660                 BUG_ON(ptr > end);
 661         } else if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
 662                 *info_level = found_key.offset;
 663         } else {
 664                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
 665         }
 666
 667         while (ptr < end) {
 668                 struct btrfs_extent_inline_ref *iref;
 669                 u64 offset;
 670                 int type;
 671
 672                 iref = (struct btrfs_extent_inline_ref *)ptr;
 673                 type = btrfs_extent_inline_ref_type(leaf, iref);
 674                 offset = btrfs_extent_inline_ref_offset(leaf, iref);
 675
 676                 switch (type) {
 677                 case BTRFS_SHARED_BLOCK_REF_KEY:
 678                         ret = __add_prelim_ref(prefs, 0, NULL,
 679                                                 *info_level + 1, offset,
 680                                                 bytenr, 1, GFP_NOFS);
 681                         break;
 682                 case BTRFS_SHARED_DATA_REF_KEY: {
 683                         struct btrfs_shared_data_ref *sdref;
 684                         int count;
 685
 686                         sdref = (struct btrfs_shared_data_ref *)(iref + 1);
 687                         count = btrfs_shared_data_ref_count(leaf, sdref);
 688                         ret = __add_prelim_ref(prefs, 0, NULL, 0, offset,
 689                                                bytenr, count, GFP_NOFS);
 690                         break;
 691                 }
 692                 case BTRFS_TREE_BLOCK_REF_KEY:
 693                         ret = __add_prelim_ref(prefs, offset, NULL,
 694                                                *info_level + 1, 0,
 695                                                bytenr, 1, GFP_NOFS);
 696                         break;
 697                 case BTRFS_EXTENT_DATA_REF_KEY: {
 698                         struct btrfs_extent_data_ref *dref;
 699                         int count;
 700                         u64 root;
 701
 702                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
 703                         count = btrfs_extent_data_ref_count(leaf, dref);
 704                         key.objectid = btrfs_extent_data_ref_objectid(leaf,
 705                                                                       dref);
 706                         key.type = BTRFS_EXTENT_DATA_KEY;
 707                         key.offset = btrfs_extent_data_ref_offset(leaf, dref);
 708                         root = btrfs_extent_data_ref_root(leaf, dref);
 709                         ret = __add_prelim_ref(prefs, root, &key, 0, 0,
 710                                                bytenr, count, GFP_NOFS);
 711                         break;
 712                 }
 713                 default:
 714                         WARN_ON(1);
 715                 }
 716                 if (ret)
 717                         return ret;
 718                 ptr += btrfs_extent_inline_ref_size(type);
 719         }
 720
 721         return 0;
 722 }
 723
 724 /*
 725  * add all non-inline backrefs for bytenr to the list
 726  */
 727 static int __add_keyed_refs(struct btrfs_fs_info *fs_info,
 728                             struct btrfs_path *path, u64 bytenr,
 729                             int info_level, struct list_head *prefs)
 730 {
 731         struct btrfs_root *extent_root = fs_info->extent_root;
 732         int ret;
 733         int slot;
 734         struct extent_buffer *leaf;
 735         struct btrfs_key key;
 736
 737         while (1) {
 738                 ret = btrfs_next_item(extent_root, path);
 739                 if (ret < 0)
 740                         break;
 741                 if (ret) {
 742                         ret = 0;
 743                         break;
 744                 }
 745
 746                 slot = path->slots[0];
 747                 leaf = path->nodes[0];
 748                 btrfs_item_key_to_cpu(leaf, &key, slot);
 749
 750                 if (key.objectid != bytenr)
 751                         break;
 752                 if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
 753                         continue;
 754                 if (key.type > BTRFS_SHARED_DATA_REF_KEY)
 755                         break;
 756
 757                 switch (key.type) {
 758                 case BTRFS_SHARED_BLOCK_REF_KEY:
 759                         ret = __add_prelim_ref(prefs, 0, NULL,
 760                                                 info_level + 1, key.offset,
 761                                                 bytenr, 1, GFP_NOFS);
 762                         break;
 763                 case BTRFS_SHARED_DATA_REF_KEY: {
 764                         struct btrfs_shared_data_ref *sdref;
 765                         int count;
 766
 767                         sdref = btrfs_item_ptr(leaf, slot,
 768                                               struct btrfs_shared_data_ref);
 769                         count = btrfs_shared_data_ref_count(leaf, sdref);
 770                         ret = __add_prelim_ref(prefs, 0, NULL, 0, key.offset,
 771                                                 bytenr, count, GFP_NOFS);
 772                         break;
 773                 }
 774                 case BTRFS_TREE_BLOCK_REF_KEY:
 775                         ret = __add_prelim_ref(prefs, key.offset, NULL,
 776                                                info_level + 1, 0,
 777                                                bytenr, 1, GFP_NOFS);
 778                         break;
 779                 case BTRFS_EXTENT_DATA_REF_KEY: {
 780                         struct btrfs_extent_data_ref *dref;
 781                         int count;
 782                         u64 root;
 783
 784                         dref = btrfs_item_ptr(leaf, slot,
 785                                               struct btrfs_extent_data_ref);
 786                         count = btrfs_extent_data_ref_count(leaf, dref);
 787                         key.objectid = btrfs_extent_data_ref_objectid(leaf,
 788                                                                       dref);
 789                         key.type = BTRFS_EXTENT_DATA_KEY;
 790                         key.offset = btrfs_extent_data_ref_offset(leaf, dref);
 791                         root = btrfs_extent_data_ref_root(leaf, dref);
 792                         ret = __add_prelim_ref(prefs, root, &key, 0, 0,
 793                                                bytenr, count, GFP_NOFS);
 794                         break;
 795                 }
 796                 default:
 797                         WARN_ON(1);
 798                 }
 799                 if (ret)
 800                         return ret;
 801
 802         }
 803
 804         return ret;
 805 }
 806
 807 /*
 808  * this adds all existing backrefs (inline backrefs, backrefs and delayed
 809  * refs) for the given bytenr to the refs list, merges duplicates and resolves
 810  * indirect refs to their parent bytenr.
 811  * When roots are found, they're added to the roots list
 812  *
 813  * FIXME some caching might speed things up
 814  */
 815 static int find_parent_nodes(struct btrfs_trans_handle *trans,
 816                              struct btrfs_fs_info *fs_info, u64 bytenr,
 817                              u64 time_seq, struct ulist *refs,
 818                              struct ulist *roots, const u64 *extent_item_pos)
 819 {
 820         struct btrfs_key key;
 821         struct btrfs_path *path;
 822         struct btrfs_delayed_ref_root *delayed_refs = NULL;
 823         struct btrfs_delayed_ref_head *head;
 824         int info_level = 0;
 825         int ret;
 826         struct list_head prefs_delayed;
 827         struct list_head prefs;
 828         struct __prelim_ref *ref;
 829
 830         INIT_LIST_HEAD(&prefs);
 831         INIT_LIST_HEAD(&prefs_delayed);
 832
 833         key.objectid = bytenr;
 834         key.offset = (u64)-1;
 835         if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
 836                 key.type = BTRFS_METADATA_ITEM_KEY;
 837         else
 838                 key.type = BTRFS_EXTENT_ITEM_KEY;
 839
 840         path = btrfs_alloc_path();
 841         if (!path)
 842                 return -ENOMEM;
 843         if (!trans)
 844                 path->search_commit_root = 1;
 845
 846         /*
 847          * grab both a lock on the path and a lock on the delayed ref head.
 848          * We need both to get a consistent picture of how the refs look
 849          * at a specified point in time
 850          */
 851 again:
 852         head = NULL;
 853
 854         ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
 855         if (ret < 0)
 856                 goto out;
 857         BUG_ON(ret == 0);
 858
 859         if (trans) {
 860                 /*
 861                  * look if there are updates for this ref queued and lock the
 862                  * head
 863                  */
 864                 delayed_refs = &trans->transaction->delayed_refs;
 865                 spin_lock(&delayed_refs->lock);
 866                 head = btrfs_find_delayed_ref_head(trans, bytenr);
 867                 if (head) {
 868                         if (!mutex_trylock(&head->mutex)) {
 869                                 atomic_inc(&head->node.refs);
 870                                 spin_unlock(&delayed_refs->lock);
 871
 872                                 btrfs_release_path(path);
 873
 874                                 /*
 875                                  * Mutex was contended, block until it's
 876                                  * released and try again
 877                                  */
 878                                 mutex_lock(&head->mutex);
 879                                 mutex_unlock(&head->mutex);
 880                                 btrfs_put_delayed_ref(&head->node);
 881                                 goto again;
 882                         }
 883                         ret = __add_delayed_refs(head, time_seq,
 884                                                  &prefs_delayed);
 885                         mutex_unlock(&head->mutex);
 886                         if (ret) {
 887                                 spin_unlock(&delayed_refs->lock);
 888                                 goto out;
 889                         }
 890                 }
 891                 spin_unlock(&delayed_refs->lock);
 892         }
 893
 894         if (path->slots[0]) {
 895                 struct extent_buffer *leaf;
 896                 int slot;
 897
 898                 path->slots[0]--;
 899                 leaf = path->nodes[0];
 900                 slot = path->slots[0];
 901                 btrfs_item_key_to_cpu(leaf, &key, slot);
 902                 if (key.objectid == bytenr &&
 903                     (key.type == BTRFS_EXTENT_ITEM_KEY ||
 904                      key.type == BTRFS_METADATA_ITEM_KEY)) {
 905                         ret = __add_inline_refs(fs_info, path, bytenr,
 906                                                 &info_level, &prefs);
 907                         if (ret)
 908                                 goto out;
 909                         ret = __add_keyed_refs(fs_info, path, bytenr,
 910                                                info_level, &prefs);
 911                         if (ret)
 912                                 goto out;
 913                 }
 914         }
 915         btrfs_release_path(path);
 916
 917         list_splice_init(&prefs_delayed, &prefs);
 918
 919         ret = __add_missing_keys(fs_info, &prefs);
 920         if (ret)
 921                 goto out;
 922
 923         __merge_refs(&prefs, 1);
 924
 925         ret = __resolve_indirect_refs(fs_info, path, time_seq, &prefs,
 926                                       extent_item_pos);
 927         if (ret)
 928                 goto out;
 929
 930         __merge_refs(&prefs, 2);
 931
 932         while (!list_empty(&prefs)) {
 933                 ref = list_first_entry(&prefs, struct __prelim_ref, list);
 934                 WARN_ON(ref->count < 0);
 935                 if (ref->count && ref->root_id && ref->parent == 0) {
 936                         /* no parent == root of tree */
 937                         ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS);
 938                         if (ret < 0)
 939                                 goto out;
 940                 }
 941                 if (ref->count && ref->parent) {
 942                         struct extent_inode_elem *eie = NULL;
 943                         if (extent_item_pos && !ref->inode_list) {
 944                                 u32 bsz;
 945                                 struct extent_buffer *eb;
 946                                 bsz = btrfs_level_size(fs_info->extent_root,
 947                                                         info_level);
 948                                 eb = read_tree_block(fs_info->extent_root,
 949                                                            ref->parent, bsz, 0);
 950                                 if (!eb || !extent_buffer_uptodate(eb)) {
 951                                         free_extent_buffer(eb);
 952                                         ret = -EIO;
 953                                         goto out;
 954                                 }
 955                                 ret = find_extent_in_eb(eb, bytenr,
 956                                                         *extent_item_pos, &eie);
 957                                 free_extent_buffer(eb);
 958                                 if (ret < 0)
 959                                         goto out;
 960                                 ref->inode_list = eie;
 961                         }
 962                         ret = ulist_add_merge(refs, ref->parent,
 963                                               (uintptr_t)ref->inode_list,
 964                                               (u64 *)&eie, GFP_NOFS);
 965                         if (ret < 0)
 966                                 goto out;
 967                         if (!ret && extent_item_pos) {
 968                                 /*
 969                                  * we've recorded that parent, so we must extend
 970                                  * its inode list here
 971                                  */
 972                                 BUG_ON(!eie);
 973                                 while (eie->next)
 974                                         eie = eie->next;
 975                                 eie->next = ref->inode_list;
 976                         }
 977                 }
 978                 list_del(&ref->list);
 979                 kmem_cache_free(btrfs_prelim_ref_cache, ref);
 980         }
 981
 982 out:
 983         btrfs_free_path(path);
 984         while (!list_empty(&prefs)) {
 985                 ref = list_first_entry(&prefs, struct __prelim_ref, list);
 986                 list_del(&ref->list);
 987                 kmem_cache_free(btrfs_prelim_ref_cache, ref);
 988         }
 989         while (!list_empty(&prefs_delayed)) {
 990                 ref = list_first_entry(&prefs_delayed, struct __prelim_ref,
 991                                        list);
 992                 list_del(&ref->list);
 993                 kmem_cache_free(btrfs_prelim_ref_cache, ref);
 994         }
 995
 996         return ret;
 997 }
 998
 999 static void free_leaf_list(struct ulist *blocks)
1000 {
1001         struct ulist_node *node = NULL;
1002         struct extent_inode_elem *eie;
1003         struct extent_inode_elem *eie_next;
1004         struct ulist_iterator uiter;
1005
1006         ULIST_ITER_INIT(&uiter);
1007         while ((node = ulist_next(blocks, &uiter))) {
1008                 if (!node->aux)
1009                         continue;
1010                 eie = (struct extent_inode_elem *)(uintptr_t)node->aux;
1011                 for (; eie; eie = eie_next) {
1012                         eie_next = eie->next;
1013                         kfree(eie);
1014                 }
1015                 node->aux = 0;
1016         }
1017
1018         ulist_free(blocks);
1019 }
1020
1021 /*
1022  * Finds all leafs with a reference to the specified combination of bytenr and
1023  * offset. key_list_head will point to a list of corresponding keys (caller must
1024  * free each list element). The leafs will be stored in the leafs ulist, which
1025  * must be freed with ulist_free.
1026  *
1027  * returns 0 on success, <0 on error
1028  */
1029 static int btrfs_find_all_leafs(struct btrfs_trans_handle *trans,
1030                                 struct btrfs_fs_info *fs_info, u64 bytenr,
1031                                 u64 time_seq, struct ulist **leafs,
1032                                 const u64 *extent_item_pos)
1033 {
1034         struct ulist *tmp;
1035         int ret;
1036
1037         tmp = ulist_alloc(GFP_NOFS);
1038         if (!tmp)
1039                 return -ENOMEM;
1040         *leafs = ulist_alloc(GFP_NOFS);
1041         if (!*leafs) {
1042                 ulist_free(tmp);
1043                 return -ENOMEM;
1044         }
1045
1046         ret = find_parent_nodes(trans, fs_info, bytenr,
1047                                 time_seq, *leafs, tmp, extent_item_pos);
1048         ulist_free(tmp);
1049
1050         if (ret < 0 && ret != -ENOENT) {
1051                 free_leaf_list(*leafs);
1052                 return ret;
1053         }
1054
1055         return 0;
1056 }
1057
1058 /*
1059  * walk all backrefs for a given extent to find all roots that reference this
1060  * extent. Walking a backref means finding all extents that reference this
1061  * extent and in turn walk the backrefs of those, too. Naturally this is a
1062  * recursive process, but here it is implemented in an iterative fashion: We
1063  * find all referencing extents for the extent in question and put them on a
1064  * list. In turn, we find all referencing extents for those, further appending
1065  * to the list. The way we iterate the list allows adding more elements after
1066  * the current while iterating. The process stops when we reach the end of the
1067  * list. Found roots are added to the roots list.
1068  *
1069  * returns 0 on success, < 0 on error.
1070  */
1071 int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1072                                 struct btrfs_fs_info *fs_info, u64 bytenr,
1073                                 u64 time_seq, struct ulist **roots)
1074 {
1075         struct ulist *tmp;
1076         struct ulist_node *node = NULL;
1077         struct ulist_iterator uiter;
1078         int ret;
1079
1080         tmp = ulist_alloc(GFP_NOFS);
1081         if (!tmp)
1082                 return -ENOMEM;
1083         *roots = ulist_alloc(GFP_NOFS);
1084         if (!*roots) {
1085                 ulist_free(tmp);
1086                 return -ENOMEM;
1087         }
1088
1089         ULIST_ITER_INIT(&uiter);
1090         while (1) {
1091                 ret = find_parent_nodes(trans, fs_info, bytenr,
1092                                         time_seq, tmp, *roots, NULL);
1093                 if (ret < 0 && ret != -ENOENT) {
1094                         ulist_free(tmp);
1095                         ulist_free(*roots);
1096                         return ret;
1097                 }
1098                 node = ulist_next(tmp, &uiter);
1099                 if (!node)
1100                         break;
1101                 bytenr = node->val;
1102         }
1103
1104         ulist_free(tmp);
1105         return 0;
1106 }
1107
1108
1109 static int __inode_info(u64 inum, u64 ioff, u8 key_type,
1110                         struct btrfs_root *fs_root, struct btrfs_path *path,
1111                         struct btrfs_key *found_key)
1112 {
1113         int ret;
1114         struct btrfs_key key;
1115         struct extent_buffer *eb;
1116
1117         key.type = key_type;
1118         key.objectid = inum;
1119         key.offset = ioff;
1120
1121         ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
1122         if (ret < 0)
1123                 return ret;
1124
1125         eb = path->nodes[0];
1126         if (ret && path->slots[0] >= btrfs_header_nritems(eb)) {
1127                 ret = btrfs_next_leaf(fs_root, path);
1128                 if (ret)
1129                         return ret;
1130                 eb = path->nodes[0];
1131         }
1132
1133         btrfs_item_key_to_cpu(eb, found_key, path->slots[0]);
1134         if (found_key->type != key.type || found_key->objectid != key.objectid)
1135                 return 1;
1136
1137         return 0;
1138 }
1139
1140 /*
1141  * this makes the path point to (inum INODE_ITEM ioff)
1142  */
1143 int inode_item_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
1144                         struct btrfs_path *path)
1145 {
1146         struct btrfs_key key;
1147         return __inode_info(inum, ioff, BTRFS_INODE_ITEM_KEY, fs_root, path,
1148                                 &key);
1149 }
1150
1151 static int inode_ref_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
1152                                 struct btrfs_path *path,
1153                                 struct btrfs_key *found_key)
1154 {
1155         return __inode_info(inum, ioff, BTRFS_INODE_REF_KEY, fs_root, path,
1156                                 found_key);
1157 }
1158
1159 int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,
1160                           u64 start_off, struct btrfs_path *path,
1161                           struct btrfs_inode_extref **ret_extref,
1162                           u64 *found_off)
1163 {
1164         int ret, slot;
1165         struct btrfs_key key;
1166         struct btrfs_key found_key;
1167         struct btrfs_inode_extref *extref;
1168         struct extent_buffer *leaf;
1169         unsigned long ptr;
1170
1171         key.objectid = inode_objectid;
1172         btrfs_set_key_type(&key, BTRFS_INODE_EXTREF_KEY);
1173         key.offset = start_off;
1174
1175         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1176         if (ret < 0)
1177                 return ret;
1178
1179         while (1) {
1180                 leaf = path->nodes[0];
1181                 slot = path->slots[0];
1182                 if (slot >= btrfs_header_nritems(leaf)) {
1183                         /*
1184                          * If the item at offset is not found,
1185                          * btrfs_search_slot will point us to the slot
1186                          * where it should be inserted. In our case
1187                          * that will be the slot directly before the
1188                          * next INODE_REF_KEY_V2 item. In the case
1189                          * that we're pointing to the last slot in a
1190                          * leaf, we must move one leaf over.
1191                          */
1192                         ret = btrfs_next_leaf(root, path);
1193                         if (ret) {
1194                                 if (ret >= 1)
1195                                         ret = -ENOENT;
1196                                 break;
1197                         }
1198                         continue;
1199                 }
1200
1201                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1202
1203                 /*
1204                  * Check that we're still looking at an extended ref key for
1205                  * this particular objectid. If we have different
1206                  * objectid or type then there are no more to be found
1207                  * in the tree and we can exit.
1208                  */
1209                 ret = -ENOENT;
1210                 if (found_key.objectid != inode_objectid)
1211                         break;
1212                 if (btrfs_key_type(&found_key) != BTRFS_INODE_EXTREF_KEY)
1213                         break;
1214
1215                 ret = 0;
1216                 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1217                 extref = (struct btrfs_inode_extref *)ptr;
1218                 *ret_extref = extref;
1219                 if (found_off)
1220                         *found_off = found_key.offset;
1221                 break;
1222         }
1223
1224         return ret;
1225 }
1226
1227 /*
1228  * this iterates to turn a name (from iref/extref) into a full filesystem path.
1229  * Elements of the path are separated by '/' and the path is guaranteed to be
1230  * 0-terminated. the path is only given within the current file system.
1231  * Therefore, it never starts with a '/'. the caller is responsible to provide
1232  * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
1233  * the start point of the resulting string is returned. this pointer is within
1234  * dest, normally.
1235  * in case the path buffer would overflow, the pointer is decremented further
1236  * as if output was written to the buffer, though no more output is actually
1237  * generated. that way, the caller can determine how much space would be
1238  * required for the path to fit into the buffer. in that case, the returned
1239  * value will be smaller than dest. callers must check this!
1240  */
1241 char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
1242                         u32 name_len, unsigned long name_off,
1243                         struct extent_buffer *eb_in, u64 parent,
1244                         char *dest, u32 size)
1245 {
1246         int slot;
1247         u64 next_inum;
1248         int ret;
1249         s64 bytes_left = ((s64)size) - 1;
1250         struct extent_buffer *eb = eb_in;
1251         struct btrfs_key found_key;
1252         int leave_spinning = path->leave_spinning;
1253         struct btrfs_inode_ref *iref;
1254
1255         if (bytes_left >= 0)
1256                 dest[bytes_left] = '\0';
1257
1258         path->leave_spinning = 1;
1259         while (1) {
1260                 bytes_left -= name_len;
1261                 if (bytes_left >= 0)
1262                         read_extent_buffer(eb, dest + bytes_left,
1263                                            name_off, name_len);
1264                 if (eb != eb_in) {
1265                         btrfs_tree_read_unlock_blocking(eb);
1266                         free_extent_buffer(eb);
1267                 }
1268                 ret = inode_ref_info(parent, 0, fs_root, path, &found_key);
1269                 if (ret > 0)
1270                         ret = -ENOENT;
1271                 if (ret)
1272                         break;
1273
1274                 next_inum = found_key.offset;
1275
1276                 /* regular exit ahead */
1277                 if (parent == next_inum)
1278                         break;
1279
1280                 slot = path->slots[0];
1281                 eb = path->nodes[0];
1282                 /* make sure we can use eb after releasing the path */
1283                 if (eb != eb_in) {
1284                         atomic_inc(&eb->refs);
1285                         btrfs_tree_read_lock(eb);
1286                         btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1287                 }
1288                 btrfs_release_path(path);
1289                 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1290
1291                 name_len = btrfs_inode_ref_name_len(eb, iref);
1292                 name_off = (unsigned long)(iref + 1);
1293
1294                 parent = next_inum;
1295                 --bytes_left;
1296                 if (bytes_left >= 0)
1297                         dest[bytes_left] = '/';
1298         }
1299
1300         btrfs_release_path(path);
1301         path->leave_spinning = leave_spinning;
1302
1303         if (ret)
1304                 return ERR_PTR(ret);
1305
1306         return dest + bytes_left;
1307 }
1308
1309 /*
1310  * this makes the path point to (logical EXTENT_ITEM *)
1311  * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
1312  * tree blocks and <0 on error.
1313  */
1314 int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
1315                         struct btrfs_path *path, struct btrfs_key *found_key,
1316                         u64 *flags_ret)
1317 {
1318         int ret;
1319         u64 flags;
1320         u64 size = 0;
1321         u32 item_size;
1322         struct extent_buffer *eb;
1323         struct btrfs_extent_item *ei;
1324         struct btrfs_key key;
1325
1326         if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1327                 key.type = BTRFS_METADATA_ITEM_KEY;
1328         else
1329                 key.type = BTRFS_EXTENT_ITEM_KEY;
1330         key.objectid = logical;
1331         key.offset = (u64)-1;
1332
1333         ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
1334         if (ret < 0)
1335                 return ret;
1336         ret = btrfs_previous_item(fs_info->extent_root, path,
1337                                         0, BTRFS_EXTENT_ITEM_KEY);
1338         if (ret < 0)
1339                 return ret;
1340
1341         btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]);
1342         if (found_key->type == BTRFS_METADATA_ITEM_KEY)
1343                 size = fs_info->extent_root->leafsize;
1344         else if (found_key->type == BTRFS_EXTENT_ITEM_KEY)
1345                 size = found_key->offset;
1346
1347         if ((found_key->type != BTRFS_EXTENT_ITEM_KEY &&
1348              found_key->type != BTRFS_METADATA_ITEM_KEY) ||
1349             found_key->objectid > logical ||
1350             found_key->objectid + size <= logical) {
1351                 pr_debug("logical %llu is not within any extent\n", logical);
1352                 return -ENOENT;
1353         }
1354
1355         eb = path->nodes[0];
1356         item_size = btrfs_item_size_nr(eb, path->slots[0]);
1357         BUG_ON(item_size < sizeof(*ei));
1358
1359         ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
1360         flags = btrfs_extent_flags(eb, ei);
1361
1362         pr_debug("logical %llu is at position %llu within the extent (%llu "
1363                  "EXTENT_ITEM %llu) flags %#llx size %u\n",
1364                  logical, logical - found_key->objectid, found_key->objectid,
1365                  found_key->offset, flags, item_size);
1366
1367         WARN_ON(!flags_ret);
1368         if (flags_ret) {
1369                 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1370                         *flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK;
1371                 else if (flags & BTRFS_EXTENT_FLAG_DATA)
1372                         *flags_ret = BTRFS_EXTENT_FLAG_DATA;
1373                 else
1374                         BUG_ON(1);
1375                 return 0;
1376         }
1377
1378         return -EIO;
1379 }
1380
1381 /*
1382  * helper function to iterate extent inline refs. ptr must point to a 0 value
1383  * for the first call and may be modified. it is used to track state.
1384  * if more refs exist, 0 is returned and the next call to
1385  * __get_extent_inline_ref must pass the modified ptr parameter to get the
1386  * next ref. after the last ref was processed, 1 is returned.
1387  * returns <0 on error
1388  */
1389 static int __get_extent_inline_ref(unsigned long *ptr, struct extent_buffer *eb,
1390                                 struct btrfs_extent_item *ei, u32 item_size,
1391                                 struct btrfs_extent_inline_ref **out_eiref,
1392                                 int *out_type)
1393 {
1394         unsigned long end;
1395         u64 flags;
1396         struct btrfs_tree_block_info *info;
1397
1398         if (!*ptr) {
1399                 /* first call */
1400                 flags = btrfs_extent_flags(eb, ei);
1401                 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1402                         info = (struct btrfs_tree_block_info *)(ei + 1);
1403                         *out_eiref =
1404                                 (struct btrfs_extent_inline_ref *)(info + 1);
1405                 } else {
1406                         *out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1);
1407                 }
1408                 *ptr = (unsigned long)*out_eiref;
1409                 if ((void *)*ptr >= (void *)ei + item_size)
1410                         return -ENOENT;
1411         }
1412
1413         end = (unsigned long)ei + item_size;
1414         *out_eiref = (struct btrfs_extent_inline_ref *)*ptr;
1415         *out_type = btrfs_extent_inline_ref_type(eb, *out_eiref);
1416
1417         *ptr += btrfs_extent_inline_ref_size(*out_type);
1418         WARN_ON(*ptr > end);
1419         if (*ptr == end)
1420                 return 1; /* last */
1421
1422         return 0;
1423 }
1424
1425 /*
1426  * reads the tree block backref for an extent. tree level and root are returned
1427  * through out_level and out_root. ptr must point to a 0 value for the first
1428  * call and may be modified (see __get_extent_inline_ref comment).
1429  * returns 0 if data was provided, 1 if there was no more data to provide or
1430  * <0 on error.
1431  */
1432 int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
1433                                 struct btrfs_extent_item *ei, u32 item_size,
1434                                 u64 *out_root, u8 *out_level)
1435 {
1436         int ret;
1437         int type;
1438         struct btrfs_tree_block_info *info;
1439         struct btrfs_extent_inline_ref *eiref;
1440
1441         if (*ptr == (unsigned long)-1)
1442                 return 1;
1443
1444         while (1) {
1445                 ret = __get_extent_inline_ref(ptr, eb, ei, item_size,
1446                                                 &eiref, &type);
1447                 if (ret < 0)
1448                         return ret;
1449
1450                 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1451                     type == BTRFS_SHARED_BLOCK_REF_KEY)
1452                         break;
1453
1454                 if (ret == 1)
1455                         return 1;
1456         }
1457
1458         /* we can treat both ref types equally here */
1459         info = (struct btrfs_tree_block_info *)(ei + 1);
1460         *out_root = btrfs_extent_inline_ref_offset(eb, eiref);
1461         *out_level = btrfs_tree_block_level(eb, info);
1462
1463         if (ret == 1)
1464                 *ptr = (unsigned long)-1;
1465
1466         return 0;
1467 }
1468
1469 static int iterate_leaf_refs(struct extent_inode_elem *inode_list,
1470                                 u64 root, u64 extent_item_objectid,
1471                                 iterate_extent_inodes_t *iterate, void *ctx)
1472 {
1473         struct extent_inode_elem *eie;
1474         int ret = 0;
1475
1476         for (eie = inode_list; eie; eie = eie->next) {
1477                 pr_debug("ref for %llu resolved, key (%llu EXTEND_DATA %llu), "
1478                          "root %llu\n", extent_item_objectid,
1479                          eie->inum, eie->offset, root);
1480                 ret = iterate(eie->inum, eie->offset, root, ctx);
1481                 if (ret) {
1482                         pr_debug("stopping iteration for %llu due to ret=%d\n",
1483                                  extent_item_objectid, ret);
1484                         break;
1485                 }
1486         }
1487
1488         return ret;
1489 }
1490
1491 /*
1492  * calls iterate() for every inode that references the extent identified by
1493  * the given parameters.
1494  * when the iterator function returns a non-zero value, iteration stops.
1495  */
1496 int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
1497                                 u64 extent_item_objectid, u64 extent_item_pos,
1498                                 int search_commit_root,
1499                                 iterate_extent_inodes_t *iterate, void *ctx)
1500 {
1501         int ret;
1502         struct btrfs_trans_handle *trans = NULL;
1503         struct ulist *refs = NULL;
1504         struct ulist *roots = NULL;
1505         struct ulist_node *ref_node = NULL;
1506         struct ulist_node *root_node = NULL;
1507         struct seq_list tree_mod_seq_elem = {};
1508         struct ulist_iterator ref_uiter;
1509         struct ulist_iterator root_uiter;
1510
1511         pr_debug("resolving all inodes for extent %llu\n",
1512                         extent_item_objectid);
1513
1514         if (!search_commit_root) {
1515                 trans = btrfs_join_transaction(fs_info->extent_root);
1516                 if (IS_ERR(trans))
1517                         return PTR_ERR(trans);
1518                 btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1519         }
1520
1521         ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid,
1522                                    tree_mod_seq_elem.seq, &refs,
1523                                    &extent_item_pos);
1524         if (ret)
1525                 goto out;
1526
1527         ULIST_ITER_INIT(&ref_uiter);
1528         while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) {
1529                 ret = btrfs_find_all_roots(trans, fs_info, ref_node->val,
1530                                            tree_mod_seq_elem.seq, &roots);
1531                 if (ret)
1532                         break;
1533                 ULIST_ITER_INIT(&root_uiter);
1534                 while (!ret && (root_node = ulist_next(roots, &root_uiter))) {
1535                         pr_debug("root %llu references leaf %llu, data list "
1536                                  "%#llx\n", root_node->val, ref_node->val,
1537                                  ref_node->aux);
1538                         ret = iterate_leaf_refs((struct extent_inode_elem *)
1539                                                 (uintptr_t)ref_node->aux,
1540                                                 root_node->val,
1541                                                 extent_item_objectid,
1542                                                 iterate, ctx);
1543                 }
1544                 ulist_free(roots);
1545         }
1546
1547         free_leaf_list(refs);
1548 out:
1549         if (!search_commit_root) {
1550                 btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1551                 btrfs_end_transaction(trans, fs_info->extent_root);
1552         }
1553
1554         return ret;
1555 }
1556
1557 int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
1558                                 struct btrfs_path *path,
1559                                 iterate_extent_inodes_t *iterate, void *ctx)
1560 {
1561         int ret;
1562         u64 extent_item_pos;
1563         u64 flags = 0;
1564         struct btrfs_key found_key;
1565         int search_commit_root = path->search_commit_root;
1566
1567         ret = extent_from_logical(fs_info, logical, path, &found_key, &flags);
1568         btrfs_release_path(path);
1569         if (ret < 0)
1570                 return ret;
1571         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1572                 return -EINVAL;
1573
1574         extent_item_pos = logical - found_key.objectid;
1575         ret = iterate_extent_inodes(fs_info, found_key.objectid,
1576                                         extent_item_pos, search_commit_root,
1577                                         iterate, ctx);
1578
1579         return ret;
1580 }
1581
1582 typedef int (iterate_irefs_t)(u64 parent, u32 name_len, unsigned long name_off,
1583                               struct extent_buffer *eb, void *ctx);
1584
1585 static int iterate_inode_refs(u64 inum, struct btrfs_root *fs_root,
1586                               struct btrfs_path *path,
1587                               iterate_irefs_t *iterate, void *ctx)
1588 {
1589         int ret = 0;
1590         int slot;
1591         u32 cur;
1592         u32 len;
1593         u32 name_len;
1594         u64 parent = 0;
1595         int found = 0;
1596         struct extent_buffer *eb;
1597         struct btrfs_item *item;
1598         struct btrfs_inode_ref *iref;
1599         struct btrfs_key found_key;
1600
1601         while (!ret) {
1602                 path->leave_spinning = 1;
1603                 ret = inode_ref_info(inum, parent ? parent+1 : 0, fs_root, path,
1604                                      &found_key);
1605                 if (ret < 0)
1606                         break;
1607                 if (ret) {
1608                         ret = found ? 0 : -ENOENT;
1609                         break;
1610                 }
1611                 ++found;
1612
1613                 parent = found_key.offset;
1614                 slot = path->slots[0];
1615                 eb = path->nodes[0];
1616                 /* make sure we can use eb after releasing the path */
1617                 atomic_inc(&eb->refs);
1618                 btrfs_tree_read_lock(eb);
1619                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1620                 btrfs_release_path(path);
1621
1622                 item = btrfs_item_nr(eb, slot);
1623                 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1624
1625                 for (cur = 0; cur < btrfs_item_size(eb, item); cur += len) {
1626                         name_len = btrfs_inode_ref_name_len(eb, iref);
1627                         /* path must be released before calling iterate()! */
1628                         pr_debug("following ref at offset %u for inode %llu in "
1629                                  "tree %llu\n", cur, found_key.objectid,
1630                                  fs_root->objectid);
1631                         ret = iterate(parent, name_len,
1632                                       (unsigned long)(iref + 1), eb, ctx);
1633                         if (ret)
1634                                 break;
1635                         len = sizeof(*iref) + name_len;
1636                         iref = (struct btrfs_inode_ref *)((char *)iref + len);
1637                 }
1638                 btrfs_tree_read_unlock_blocking(eb);
1639                 free_extent_buffer(eb);
1640         }
1641
1642         btrfs_release_path(path);
1643
1644         return ret;
1645 }
1646
1647 static int iterate_inode_extrefs(u64 inum, struct btrfs_root *fs_root,
1648                                  struct btrfs_path *path,
1649                                  iterate_irefs_t *iterate, void *ctx)
1650 {
1651         int ret;
1652         int slot;
1653         u64 offset = 0;
1654         u64 parent;
1655         int found = 0;
1656         struct extent_buffer *eb;
1657         struct btrfs_inode_extref *extref;
1658         struct extent_buffer *leaf;
1659         u32 item_size;
1660         u32 cur_offset;
1661         unsigned long ptr;
1662
1663         while (1) {
1664                 ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref,
1665                                             &offset);
1666                 if (ret < 0)
1667                         break;
1668                 if (ret) {
1669                         ret = found ? 0 : -ENOENT;
1670                         break;
1671                 }
1672                 ++found;
1673
1674                 slot = path->slots[0];
1675                 eb = path->nodes[0];
1676                 /* make sure we can use eb after releasing the path */
1677                 atomic_inc(&eb->refs);
1678
1679                 btrfs_tree_read_lock(eb);
1680                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1681                 btrfs_release_path(path);
1682
1683                 leaf = path->nodes[0];
1684                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1685                 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1686                 cur_offset = 0;
1687
1688                 while (cur_offset < item_size) {
1689                         u32 name_len;
1690
1691                         extref = (struct btrfs_inode_extref *)(ptr + cur_offset);
1692                         parent = btrfs_inode_extref_parent(eb, extref);
1693                         name_len = btrfs_inode_extref_name_len(eb, extref);
1694                         ret = iterate(parent, name_len,
1695                                       (unsigned long)&extref->name, eb, ctx);
1696                         if (ret)
1697                                 break;
1698
1699                         cur_offset += btrfs_inode_extref_name_len(leaf, extref);
1700                         cur_offset += sizeof(*extref);
1701                 }
1702                 btrfs_tree_read_unlock_blocking(eb);
1703                 free_extent_buffer(eb);
1704
1705                 offset++;
1706         }
1707
1708         btrfs_release_path(path);
1709
1710         return ret;
1711 }
1712
1713 static int iterate_irefs(u64 inum, struct btrfs_root *fs_root,
1714                          struct btrfs_path *path, iterate_irefs_t *iterate,
1715                          void *ctx)
1716 {
1717         int ret;
1718         int found_refs = 0;
1719
1720         ret = iterate_inode_refs(inum, fs_root, path, iterate, ctx);
1721         if (!ret)
1722                 ++found_refs;
1723         else if (ret != -ENOENT)
1724                 return ret;
1725
1726         ret = iterate_inode_extrefs(inum, fs_root, path, iterate, ctx);
1727         if (ret == -ENOENT && found_refs)
1728                 return 0;
1729
1730         return ret;
1731 }
1732
1733 /*
1734  * returns 0 if the path could be dumped (probably truncated)
1735  * returns <0 in case of an error
1736  */
1737 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
1738                          struct extent_buffer *eb, void *ctx)
1739 {
1740         struct inode_fs_paths *ipath = ctx;
1741         char *fspath;
1742         char *fspath_min;
1743         int i = ipath->fspath->elem_cnt;
1744         const int s_ptr = sizeof(char *);
1745         u32 bytes_left;
1746
1747         bytes_left = ipath->fspath->bytes_left > s_ptr ?
1748                                         ipath->fspath->bytes_left - s_ptr : 0;
1749
1750         fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
1751         fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len,
1752                                    name_off, eb, inum, fspath_min, bytes_left);
1753         if (IS_ERR(fspath))
1754                 return PTR_ERR(fspath);
1755
1756         if (fspath > fspath_min) {
1757                 ipath->fspath->val[i] = (u64)(unsigned long)fspath;
1758                 ++ipath->fspath->elem_cnt;
1759                 ipath->fspath->bytes_left = fspath - fspath_min;
1760         } else {
1761                 ++ipath->fspath->elem_missed;
1762                 ipath->fspath->bytes_missing += fspath_min - fspath;
1763                 ipath->fspath->bytes_left = 0;
1764         }
1765
1766         return 0;
1767 }
1768
1769 /*
1770  * this dumps all file system paths to the inode into the ipath struct, provided
1771  * is has been created large enough. each path is zero-terminated and accessed
1772  * from ipath->fspath->val[i].
1773  * when it returns, there are ipath->fspath->elem_cnt number of paths available
1774  * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the
1775  * number of missed paths in recored in ipath->fspath->elem_missed, otherwise,
1776  * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
1777  * have been needed to return all paths.
1778  */
1779 int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
1780 {
1781         return iterate_irefs(inum, ipath->fs_root, ipath->btrfs_path,
1782                              inode_to_path, ipath);
1783 }
1784
1785 struct btrfs_data_container *init_data_container(u32 total_bytes)
1786 {
1787         struct btrfs_data_container *data;
1788         size_t alloc_bytes;
1789
1790         alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
1791         data = vmalloc(alloc_bytes);
1792         if (!data)
1793                 return ERR_PTR(-ENOMEM);
1794
1795         if (total_bytes >= sizeof(*data)) {
1796                 data->bytes_left = total_bytes - sizeof(*data);
1797                 data->bytes_missing = 0;
1798         } else {
1799                 data->bytes_missing = sizeof(*data) - total_bytes;
1800                 data->bytes_left = 0;
1801         }
1802
1803         data->elem_cnt = 0;
1804         data->elem_missed = 0;
1805
1806         return data;
1807 }
1808
1809 /*
1810  * allocates space to return multiple file system paths for an inode.
1811  * total_bytes to allocate are passed, note that space usable for actual path
1812  * information will be total_bytes - sizeof(struct inode_fs_paths).
1813  * the returned pointer must be freed with free_ipath() in the end.
1814  */
1815 struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
1816                                         struct btrfs_path *path)
1817 {
1818         struct inode_fs_paths *ifp;
1819         struct btrfs_data_container *fspath;
1820
1821         fspath = init_data_container(total_bytes);
1822         if (IS_ERR(fspath))
1823                 return (void *)fspath;
1824
1825         ifp = kmalloc(sizeof(*ifp), GFP_NOFS);
1826         if (!ifp) {
1827                 kfree(fspath);
1828                 return ERR_PTR(-ENOMEM);
1829         }
1830
1831         ifp->btrfs_path = path;
1832         ifp->fspath = fspath;
1833         ifp->fs_root = fs_root;
1834
1835         return ifp;
1836 }
1837
1838 void free_ipath(struct inode_fs_paths *ipath)
1839 {
1840         if (!ipath)
1841                 return;
1842         vfree(ipath->fspath);
1843         kfree(ipath);
1844 }