fs/btrfs/backref.c

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