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 void free_inode_elem_list(struct extent_inode_elem *eie)
  70 {
  71         struct extent_inode_elem *eie_next;
  72
  73         for (; eie; eie = eie_next) {
  74                 eie_next = eie->next;
  75                 kfree(eie);
  76         }
  77 }
  78
  79 static int find_extent_in_eb(struct extent_buffer *eb, u64 wanted_disk_byte,
  80                                 u64 extent_item_pos,
  81                                 struct extent_inode_elem **eie)
  82 {
  83         u64 disk_byte;
  84         struct btrfs_key key;
  85         struct btrfs_file_extent_item *fi;
  86         int slot;
  87         int nritems;
  88         int extent_type;
  89         int ret;
  90
  91         /*
  92          * from the shared data ref, we only have the leaf but we need
  93          * the key. thus, we must look into all items and see that we
  94          * find one (some) with a reference to our extent item.
  95          */
  96         nritems = btrfs_header_nritems(eb);
  97         for (slot = 0; slot < nritems; ++slot) {
  98                 btrfs_item_key_to_cpu(eb, &key, slot);
  99                 if (key.type != BTRFS_EXTENT_DATA_KEY)
 100                         continue;
 101                 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
 102                 extent_type = btrfs_file_extent_type(eb, fi);
 103                 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
 104                         continue;
 105                 /* don't skip BTRFS_FILE_EXTENT_PREALLOC, we can handle that */
 106                 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
 107                 if (disk_byte != wanted_disk_byte)
 108                         continue;
 109
 110                 ret = check_extent_in_eb(&key, eb, fi, extent_item_pos, eie);
 111                 if (ret < 0)
 112                         return ret;
 113         }
 114
 115         return 0;
 116 }
 117
 118 /*
 119  * this structure records all encountered refs on the way up to the root
 120  */
 121 struct __prelim_ref {
 122         struct list_head list;
 123         u64 root_id;
 124         struct btrfs_key key_for_search;
 125         int level;
 126         int count;
 127         struct extent_inode_elem *inode_list;
 128         u64 parent;
 129         u64 wanted_disk_byte;
 130 };
 131
 132 static struct kmem_cache *btrfs_prelim_ref_cache;
 133
 134 int __init btrfs_prelim_ref_init(void)
 135 {
 136         btrfs_prelim_ref_cache = kmem_cache_create("btrfs_prelim_ref",
 137                                         sizeof(struct __prelim_ref),
 138                                         0,
 139                                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
 140                                         NULL);
 141         if (!btrfs_prelim_ref_cache)
 142                 return -ENOMEM;
 143         return 0;
 144 }
 145
 146 void btrfs_prelim_ref_exit(void)
 147 {
 148         if (btrfs_prelim_ref_cache)
 149                 kmem_cache_destroy(btrfs_prelim_ref_cache);
 150 }
 151
 152 /*
 153  * the rules for all callers of this function are:
 154  * - obtaining the parent is the goal
 155  * - if you add a key, you must know that it is a correct key
 156  * - if you cannot add the parent or a correct key, then we will look into the
 157  *   block later to set a correct key
 158  *
 159  * delayed refs
 160  * ============
 161  *        backref type | shared | indirect | shared | indirect
 162  * information         |   tree |     tree |   data |     data
 163  * --------------------+--------+----------+--------+----------
 164  *      parent logical |    y   |     -    |    -   |     -
 165  *      key to resolve |    -   |     y    |    y   |     y
 166  *  tree block logical |    -   |     -    |    -   |     -
 167  *  root for resolving |    y   |     y    |    y   |     y
 168  *
 169  * - column 1:       we've the parent -> done
 170  * - column 2, 3, 4: we use the key to find the parent
 171  *
 172  * on disk refs (inline or keyed)
 173  * ==============================
 174  *        backref type | shared | indirect | shared | indirect
 175  * information         |   tree |     tree |   data |     data
 176  * --------------------+--------+----------+--------+----------
 177  *      parent logical |    y   |     -    |    y   |     -
 178  *      key to resolve |    -   |     -    |    -   |     y
 179  *  tree block logical |    y   |     y    |    y   |     y
 180  *  root for resolving |    -   |     y    |    y   |     y
 181  *
 182  * - column 1, 3: we've the parent -> done
 183  * - column 2:    we take the first key from the block to find the parent
 184  *                (see __add_missing_keys)
 185  * - column 4:    we use the key to find the parent
 186  *
 187  * additional information that's available but not required to find the parent
 188  * block might help in merging entries to gain some speed.
 189  */
 190
 191 static int __add_prelim_ref(struct list_head *head, u64 root_id,
 192                             struct btrfs_key *key, int level,
 193                             u64 parent, u64 wanted_disk_byte, int count,
 194                             gfp_t gfp_mask)
 195 {
 196         struct __prelim_ref *ref;
 197
 198         if (root_id == BTRFS_DATA_RELOC_TREE_OBJECTID)
 199                 return 0;
 200
 201         ref = kmem_cache_alloc(btrfs_prelim_ref_cache, gfp_mask);
 202         if (!ref)
 203                 return -ENOMEM;
 204
 205         ref->root_id = root_id;
 206         if (key)
 207                 ref->key_for_search = *key;
 208         else
 209                 memset(&ref->key_for_search, 0, sizeof(ref->key_for_search));
 210
 211         ref->inode_list = NULL;
 212         ref->level = level;
 213         ref->count = count;
 214         ref->parent = parent;
 215         ref->wanted_disk_byte = wanted_disk_byte;
 216         list_add_tail(&ref->list, head);
 217
 218         return 0;
 219 }
 220
 221 static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path,
 222                            struct ulist *parents, struct __prelim_ref *ref,
 223                            int level, u64 time_seq, const u64 *extent_item_pos,
 224                            u64 total_refs)
 225 {
 226         int ret = 0;
 227         int slot;
 228         struct extent_buffer *eb;
 229         struct btrfs_key key;
 230         struct btrfs_key *key_for_search = &ref->key_for_search;
 231         struct btrfs_file_extent_item *fi;
 232         struct extent_inode_elem *eie = NULL, *old = NULL;
 233         u64 disk_byte;
 234         u64 wanted_disk_byte = ref->wanted_disk_byte;
 235         u64 count = 0;
 236
 237         if (level != 0) {
 238                 eb = path->nodes[level];
 239                 ret = ulist_add(parents, eb->start, 0, GFP_NOFS);
 240                 if (ret < 0)
 241                         return ret;
 242                 return 0;
 243         }
 244
 245         /*
 246          * We normally enter this function with the path already pointing to
 247          * the first item to check. But sometimes, we may enter it with
 248          * slot==nritems. In that case, go to the next leaf before we continue.
 249          */
 250         if (path->slots[0] >= btrfs_header_nritems(path->nodes[0]))
 251                 ret = btrfs_next_old_leaf(root, path, time_seq);
 252
 253         while (!ret && count < total_refs) {
 254                 eb = path->nodes[0];
 255                 slot = path->slots[0];
 256
 257                 btrfs_item_key_to_cpu(eb, &key, slot);
 258
 259                 if (key.objectid != key_for_search->objectid ||
 260                     key.type != BTRFS_EXTENT_DATA_KEY)
 261                         break;
 262
 263                 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
 264                 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
 265
 266                 if (disk_byte == wanted_disk_byte) {
 267                         eie = NULL;
 268                         old = NULL;
 269                         count++;
 270                         if (extent_item_pos) {
 271                                 ret = check_extent_in_eb(&key, eb, fi,
 272                                                 *extent_item_pos,
 273                                                 &eie);
 274                                 if (ret < 0)
 275                                         break;
 276                         }
 277                         if (ret > 0)
 278                                 goto next;
 279                         ret = ulist_add_merge(parents, eb->start,
 280                                               (uintptr_t)eie,
 281                                               (u64 *)&old, GFP_NOFS);
 282                         if (ret < 0)
 283                                 break;
 284                         if (!ret && extent_item_pos) {
 285                                 while (old->next)
 286                                         old = old->next;
 287                                 old->next = eie;
 288                         }
 289                         eie = NULL;
 290                 }
 291 next:
 292                 ret = btrfs_next_old_item(root, path, time_seq);
 293         }
 294
 295         if (ret > 0)
 296                 ret = 0;
 297         else if (ret < 0)
 298                 free_inode_elem_list(eie);
 299         return ret;
 300 }
 301
 302 /*
 303  * resolve an indirect backref in the form (root_id, key, level)
 304  * to a logical address
 305  */
 306 static int __resolve_indirect_ref(struct btrfs_fs_info *fs_info,
 307                                   struct btrfs_path *path, u64 time_seq,
 308                                   struct __prelim_ref *ref,
 309                                   struct ulist *parents,
 310                                   const u64 *extent_item_pos, u64 total_refs)
 311 {
 312         struct btrfs_root *root;
 313         struct btrfs_key root_key;
 314         struct extent_buffer *eb;
 315         int ret = 0;
 316         int root_level;
 317         int level = ref->level;
 318         int index;
 319
 320         root_key.objectid = ref->root_id;
 321         root_key.type = BTRFS_ROOT_ITEM_KEY;
 322         root_key.offset = (u64)-1;
 323
 324         index = srcu_read_lock(&fs_info->subvol_srcu);
 325
 326         root = btrfs_read_fs_root_no_name(fs_info, &root_key);
 327         if (IS_ERR(root)) {
 328                 srcu_read_unlock(&fs_info->subvol_srcu, index);
 329                 ret = PTR_ERR(root);
 330                 goto out;
 331         }
 332
 333         root_level = btrfs_old_root_level(root, time_seq);
 334
 335         if (root_level + 1 == level) {
 336                 srcu_read_unlock(&fs_info->subvol_srcu, index);
 337                 goto out;
 338         }
 339
 340         path->lowest_level = level;
 341         ret = btrfs_search_old_slot(root, &ref->key_for_search, path, time_seq);
 342
 343         /* root node has been locked, we can release @subvol_srcu safely here */
 344         srcu_read_unlock(&fs_info->subvol_srcu, index);
 345
 346         pr_debug("search slot in root %llu (level %d, ref count %d) returned "
 347                  "%d for key (%llu %u %llu)\n",
 348                  ref->root_id, level, ref->count, ret,
 349                  ref->key_for_search.objectid, ref->key_for_search.type,
 350                  ref->key_for_search.offset);
 351         if (ret < 0)
 352                 goto out;
 353
 354         eb = path->nodes[level];
 355         while (!eb) {
 356                 if (WARN_ON(!level)) {
 357                         ret = 1;
 358                         goto out;
 359                 }
 360                 level--;
 361                 eb = path->nodes[level];
 362         }
 363
 364         ret = add_all_parents(root, path, parents, ref, level, time_seq,
 365                               extent_item_pos, total_refs);
 366 out:
 367         path->lowest_level = 0;
 368         btrfs_release_path(path);
 369         return ret;
 370 }
 371
 372 /*
 373  * resolve all indirect backrefs from the list
 374  */
 375 static int __resolve_indirect_refs(struct btrfs_fs_info *fs_info,
 376                                    struct btrfs_path *path, u64 time_seq,
 377                                    struct list_head *head,
 378                                    const u64 *extent_item_pos, u64 total_refs)
 379 {
 380         int err;
 381         int ret = 0;
 382         struct __prelim_ref *ref;
 383         struct __prelim_ref *ref_safe;
 384         struct __prelim_ref *new_ref;
 385         struct ulist *parents;
 386         struct ulist_node *node;
 387         struct ulist_iterator uiter;
 388
 389         parents = ulist_alloc(GFP_NOFS);
 390         if (!parents)
 391                 return -ENOMEM;
 392
 393         /*
 394          * _safe allows us to insert directly after the current item without
 395          * iterating over the newly inserted items.
 396          * we're also allowed to re-assign ref during iteration.
 397          */
 398         list_for_each_entry_safe(ref, ref_safe, head, list) {
 399                 if (ref->parent)        /* already direct */
 400                         continue;
 401                 if (ref->count == 0)
 402                         continue;
 403                 err = __resolve_indirect_ref(fs_info, path, time_seq, ref,
 404                                              parents, extent_item_pos,
 405                                              total_refs);
 406                 /*
 407                  * we can only tolerate ENOENT,otherwise,we should catch error
 408                  * and return directly.
 409                  */
 410                 if (err == -ENOENT) {
 411                         continue;
 412                 } else if (err) {
 413                         ret = err;
 414                         goto out;
 415                 }
 416
 417                 /* we put the first parent into the ref at hand */
 418                 ULIST_ITER_INIT(&uiter);
 419                 node = ulist_next(parents, &uiter);
 420                 ref->parent = node ? node->val : 0;
 421                 ref->inode_list = node ?
 422                         (struct extent_inode_elem *)(uintptr_t)node->aux : NULL;
 423
 424                 /* additional parents require new refs being added here */
 425                 while ((node = ulist_next(parents, &uiter))) {
 426                         new_ref = kmem_cache_alloc(btrfs_prelim_ref_cache,
 427                                                    GFP_NOFS);
 428                         if (!new_ref) {
 429                                 ret = -ENOMEM;
 430                                 goto out;
 431                         }
 432                         memcpy(new_ref, ref, sizeof(*ref));
 433                         new_ref->parent = node->val;
 434                         new_ref->inode_list = (struct extent_inode_elem *)
 435                                                         (uintptr_t)node->aux;
 436                         list_add(&new_ref->list, &ref->list);
 437                 }
 438                 ulist_reinit(parents);
 439         }
 440 out:
 441         ulist_free(parents);
 442         return ret;
 443 }
 444
 445 static inline int ref_for_same_block(struct __prelim_ref *ref1,
 446                                      struct __prelim_ref *ref2)
 447 {
 448         if (ref1->level != ref2->level)
 449                 return 0;
 450         if (ref1->root_id != ref2->root_id)
 451                 return 0;
 452         if (ref1->key_for_search.type != ref2->key_for_search.type)
 453                 return 0;
 454         if (ref1->key_for_search.objectid != ref2->key_for_search.objectid)
 455                 return 0;
 456         if (ref1->key_for_search.offset != ref2->key_for_search.offset)
 457                 return 0;
 458         if (ref1->parent != ref2->parent)
 459                 return 0;
 460
 461         return 1;
 462 }
 463
 464 /*
 465  * read tree blocks and add keys where required.
 466  */
 467 static int __add_missing_keys(struct btrfs_fs_info *fs_info,
 468                               struct list_head *head)
 469 {
 470         struct list_head *pos;
 471         struct extent_buffer *eb;
 472
 473         list_for_each(pos, head) {
 474                 struct __prelim_ref *ref;
 475                 ref = list_entry(pos, struct __prelim_ref, list);
 476
 477                 if (ref->parent)
 478                         continue;
 479                 if (ref->key_for_search.type)
 480                         continue;
 481                 BUG_ON(!ref->wanted_disk_byte);
 482                 eb = read_tree_block(fs_info->tree_root, ref->wanted_disk_byte,
 483                                      fs_info->tree_root->leafsize, 0);
 484                 if (!eb || !extent_buffer_uptodate(eb)) {
 485                         free_extent_buffer(eb);
 486                         return -EIO;
 487                 }
 488                 btrfs_tree_read_lock(eb);
 489                 if (btrfs_header_level(eb) == 0)
 490                         btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
 491                 else
 492                         btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
 493                 btrfs_tree_read_unlock(eb);
 494                 free_extent_buffer(eb);
 495         }
 496         return 0;
 497 }
 498
 499 /*
 500  * merge two lists of backrefs and adjust counts accordingly
 501  *
 502  * mode = 1: merge identical keys, if key is set
 503  *    FIXME: if we add more keys in __add_prelim_ref, we can merge more here.
 504  *           additionally, we could even add a key range for the blocks we
 505  *           looked into to merge even more (-> replace unresolved refs by those
 506  *           having a parent).
 507  * mode = 2: merge identical parents
 508  */
 509 static void __merge_refs(struct list_head *head, int mode)
 510 {
 511         struct list_head *pos1;
 512
 513         list_for_each(pos1, head) {
 514                 struct list_head *n2;
 515                 struct list_head *pos2;
 516                 struct __prelim_ref *ref1;
 517
 518                 ref1 = list_entry(pos1, struct __prelim_ref, list);
 519
 520                 for (pos2 = pos1->next, n2 = pos2->next; pos2 != head;
 521                      pos2 = n2, n2 = pos2->next) {
 522                         struct __prelim_ref *ref2;
 523                         struct __prelim_ref *xchg;
 524                         struct extent_inode_elem *eie;
 525
 526                         ref2 = list_entry(pos2, struct __prelim_ref, list);
 527
 528                         if (mode == 1) {
 529                                 if (!ref_for_same_block(ref1, ref2))
 530                                         continue;
 531                                 if (!ref1->parent && ref2->parent) {
 532                                         xchg = ref1;
 533                                         ref1 = ref2;
 534                                         ref2 = xchg;
 535                                 }
 536                         } else {
 537                                 if (ref1->parent != ref2->parent)
 538                                         continue;
 539                         }
 540
 541                         eie = ref1->inode_list;
 542                         while (eie && eie->next)
 543                                 eie = eie->next;
 544                         if (eie)
 545                                 eie->next = ref2->inode_list;
 546                         else
 547                                 ref1->inode_list = ref2->inode_list;
 548                         ref1->count += ref2->count;
 549
 550                         list_del(&ref2->list);
 551                         kmem_cache_free(btrfs_prelim_ref_cache, ref2);
 552                 }
 553
 554         }
 555 }
 556
 557 /*
 558  * add all currently queued delayed refs from this head whose seq nr is
 559  * smaller or equal that seq to the list
 560  */
 561 static int __add_delayed_refs(struct btrfs_delayed_ref_head *head, u64 seq,
 562                               struct list_head *prefs, u64 *total_refs)
 563 {
 564         struct btrfs_delayed_extent_op *extent_op = head->extent_op;
 565         struct rb_node *n = &head->node.rb_node;
 566         struct btrfs_key key;
 567         struct btrfs_key op_key = {0};
 568         int sgn;
 569         int ret = 0;
 570
 571         if (extent_op && extent_op->update_key)
 572                 btrfs_disk_key_to_cpu(&op_key, &extent_op->key);
 573
 574         spin_lock(&head->lock);
 575         n = rb_first(&head->ref_root);
 576         while (n) {
 577                 struct btrfs_delayed_ref_node *node;
 578                 node = rb_entry(n, struct btrfs_delayed_ref_node,
 579                                 rb_node);
 580                 n = rb_next(n);
 581                 if (node->seq > seq)
 582                         continue;
 583
 584                 switch (node->action) {
 585                 case BTRFS_ADD_DELAYED_EXTENT:
 586                 case BTRFS_UPDATE_DELAYED_HEAD:
 587                         WARN_ON(1);
 588                         continue;
 589                 case BTRFS_ADD_DELAYED_REF:
 590                         sgn = 1;
 591                         break;
 592                 case BTRFS_DROP_DELAYED_REF:
 593                         sgn = -1;
 594                         break;
 595                 default:
 596                         BUG_ON(1);
 597                 }
 598                 *total_refs += (node->ref_mod * sgn);
 599                 switch (node->type) {
 600                 case BTRFS_TREE_BLOCK_REF_KEY: {
 601                         struct btrfs_delayed_tree_ref *ref;
 602
 603                         ref = btrfs_delayed_node_to_tree_ref(node);
 604                         ret = __add_prelim_ref(prefs, ref->root, &op_key,
 605                                                ref->level + 1, 0, node->bytenr,
 606                                                node->ref_mod * sgn, GFP_ATOMIC);
 607                         break;
 608                 }
 609                 case BTRFS_SHARED_BLOCK_REF_KEY: {
 610                         struct btrfs_delayed_tree_ref *ref;
 611
 612                         ref = btrfs_delayed_node_to_tree_ref(node);
 613                         ret = __add_prelim_ref(prefs, ref->root, NULL,
 614                                                ref->level + 1, ref->parent,
 615                                                node->bytenr,
 616                                                node->ref_mod * sgn, GFP_ATOMIC);
 617                         break;
 618                 }
 619                 case BTRFS_EXTENT_DATA_REF_KEY: {
 620                         struct btrfs_delayed_data_ref *ref;
 621                         ref = btrfs_delayed_node_to_data_ref(node);
 622
 623                         key.objectid = ref->objectid;
 624                         key.type = BTRFS_EXTENT_DATA_KEY;
 625                         key.offset = ref->offset;
 626                         ret = __add_prelim_ref(prefs, ref->root, &key, 0, 0,
 627                                                node->bytenr,
 628                                                node->ref_mod * sgn, GFP_ATOMIC);
 629                         break;
 630                 }
 631                 case BTRFS_SHARED_DATA_REF_KEY: {
 632                         struct btrfs_delayed_data_ref *ref;
 633
 634                         ref = btrfs_delayed_node_to_data_ref(node);
 635
 636                         key.objectid = ref->objectid;
 637                         key.type = BTRFS_EXTENT_DATA_KEY;
 638                         key.offset = ref->offset;
 639                         ret = __add_prelim_ref(prefs, ref->root, &key, 0,
 640                                                ref->parent, node->bytenr,
 641                                                node->ref_mod * sgn, GFP_ATOMIC);
 642                         break;
 643                 }
 644                 default:
 645                         WARN_ON(1);
 646                 }
 647                 if (ret)
 648                         break;
 649         }
 650         spin_unlock(&head->lock);
 651         return ret;
 652 }
 653
 654 /*
 655  * add all inline backrefs for bytenr to the list
 656  */
 657 static int __add_inline_refs(struct btrfs_fs_info *fs_info,
 658                              struct btrfs_path *path, u64 bytenr,
 659                              int *info_level, struct list_head *prefs,
 660                              u64 *total_refs)
 661 {
 662         int ret = 0;
 663         int slot;
 664         struct extent_buffer *leaf;
 665         struct btrfs_key key;
 666         struct btrfs_key found_key;
 667         unsigned long ptr;
 668         unsigned long end;
 669         struct btrfs_extent_item *ei;
 670         u64 flags;
 671         u64 item_size;
 672
 673         /*
 674          * enumerate all inline refs
 675          */
 676         leaf = path->nodes[0];
 677         slot = path->slots[0];
 678
 679         item_size = btrfs_item_size_nr(leaf, slot);
 680         BUG_ON(item_size < sizeof(*ei));
 681
 682         ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
 683         flags = btrfs_extent_flags(leaf, ei);
 684         *total_refs += btrfs_extent_refs(leaf, ei);
 685         btrfs_item_key_to_cpu(leaf, &found_key, slot);
 686
 687         ptr = (unsigned long)(ei + 1);
 688         end = (unsigned long)ei + item_size;
 689
 690         if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
 691             flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
 692                 struct btrfs_tree_block_info *info;
 693
 694                 info = (struct btrfs_tree_block_info *)ptr;
 695                 *info_level = btrfs_tree_block_level(leaf, info);
 696                 ptr += sizeof(struct btrfs_tree_block_info);
 697                 BUG_ON(ptr > end);
 698         } else if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
 699                 *info_level = found_key.offset;
 700         } else {
 701                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
 702         }
 703
 704         while (ptr < end) {
 705                 struct btrfs_extent_inline_ref *iref;
 706                 u64 offset;
 707                 int type;
 708
 709                 iref = (struct btrfs_extent_inline_ref *)ptr;
 710                 type = btrfs_extent_inline_ref_type(leaf, iref);
 711                 offset = btrfs_extent_inline_ref_offset(leaf, iref);
 712
 713                 switch (type) {
 714                 case BTRFS_SHARED_BLOCK_REF_KEY:
 715                         ret = __add_prelim_ref(prefs, 0, NULL,
 716                                                 *info_level + 1, offset,
 717                                                 bytenr, 1, GFP_NOFS);
 718                         break;
 719                 case BTRFS_SHARED_DATA_REF_KEY: {
 720                         struct btrfs_shared_data_ref *sdref;
 721                         int count;
 722
 723                         sdref = (struct btrfs_shared_data_ref *)(iref + 1);
 724                         count = btrfs_shared_data_ref_count(leaf, sdref);
 725                         ret = __add_prelim_ref(prefs, 0, NULL, 0, offset,
 726                                                bytenr, count, GFP_NOFS);
 727                         break;
 728                 }
 729                 case BTRFS_TREE_BLOCK_REF_KEY:
 730                         ret = __add_prelim_ref(prefs, offset, NULL,
 731                                                *info_level + 1, 0,
 732                                                bytenr, 1, GFP_NOFS);
 733                         break;
 734                 case BTRFS_EXTENT_DATA_REF_KEY: {
 735                         struct btrfs_extent_data_ref *dref;
 736                         int count;
 737                         u64 root;
 738
 739                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
 740                         count = btrfs_extent_data_ref_count(leaf, dref);
 741                         key.objectid = btrfs_extent_data_ref_objectid(leaf,
 742                                                                       dref);
 743                         key.type = BTRFS_EXTENT_DATA_KEY;
 744                         key.offset = btrfs_extent_data_ref_offset(leaf, dref);
 745                         root = btrfs_extent_data_ref_root(leaf, dref);
 746                         ret = __add_prelim_ref(prefs, root, &key, 0, 0,
 747                                                bytenr, count, GFP_NOFS);
 748                         break;
 749                 }
 750                 default:
 751                         WARN_ON(1);
 752                 }
 753                 if (ret)
 754                         return ret;
 755                 ptr += btrfs_extent_inline_ref_size(type);
 756         }
 757
 758         return 0;
 759 }
 760
 761 /*
 762  * add all non-inline backrefs for bytenr to the list
 763  */
 764 static int __add_keyed_refs(struct btrfs_fs_info *fs_info,
 765                             struct btrfs_path *path, u64 bytenr,
 766                             int info_level, struct list_head *prefs)
 767 {
 768         struct btrfs_root *extent_root = fs_info->extent_root;
 769         int ret;
 770         int slot;
 771         struct extent_buffer *leaf;
 772         struct btrfs_key key;
 773
 774         while (1) {
 775                 ret = btrfs_next_item(extent_root, path);
 776                 if (ret < 0)
 777                         break;
 778                 if (ret) {
 779                         ret = 0;
 780                         break;
 781                 }
 782
 783                 slot = path->slots[0];
 784                 leaf = path->nodes[0];
 785                 btrfs_item_key_to_cpu(leaf, &key, slot);
 786
 787                 if (key.objectid != bytenr)
 788                         break;
 789                 if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
 790                         continue;
 791                 if (key.type > BTRFS_SHARED_DATA_REF_KEY)
 792                         break;
 793
 794                 switch (key.type) {
 795                 case BTRFS_SHARED_BLOCK_REF_KEY:
 796                         ret = __add_prelim_ref(prefs, 0, NULL,
 797                                                 info_level + 1, key.offset,
 798                                                 bytenr, 1, GFP_NOFS);
 799                         break;
 800                 case BTRFS_SHARED_DATA_REF_KEY: {
 801                         struct btrfs_shared_data_ref *sdref;
 802                         int count;
 803
 804                         sdref = btrfs_item_ptr(leaf, slot,
 805                                               struct btrfs_shared_data_ref);
 806                         count = btrfs_shared_data_ref_count(leaf, sdref);
 807                         ret = __add_prelim_ref(prefs, 0, NULL, 0, key.offset,
 808                                                 bytenr, count, GFP_NOFS);
 809                         break;
 810                 }
 811                 case BTRFS_TREE_BLOCK_REF_KEY:
 812                         ret = __add_prelim_ref(prefs, key.offset, NULL,
 813                                                info_level + 1, 0,
 814                                                bytenr, 1, GFP_NOFS);
 815                         break;
 816                 case BTRFS_EXTENT_DATA_REF_KEY: {
 817                         struct btrfs_extent_data_ref *dref;
 818                         int count;
 819                         u64 root;
 820
 821                         dref = btrfs_item_ptr(leaf, slot,
 822                                               struct btrfs_extent_data_ref);
 823                         count = btrfs_extent_data_ref_count(leaf, dref);
 824                         key.objectid = btrfs_extent_data_ref_objectid(leaf,
 825                                                                       dref);
 826                         key.type = BTRFS_EXTENT_DATA_KEY;
 827                         key.offset = btrfs_extent_data_ref_offset(leaf, dref);
 828                         root = btrfs_extent_data_ref_root(leaf, dref);
 829                         ret = __add_prelim_ref(prefs, root, &key, 0, 0,
 830                                                bytenr, count, GFP_NOFS);
 831                         break;
 832                 }
 833                 default:
 834                         WARN_ON(1);
 835                 }
 836                 if (ret)
 837                         return ret;
 838
 839         }
 840
 841         return ret;
 842 }
 843
 844 /*
 845  * this adds all existing backrefs (inline backrefs, backrefs and delayed
 846  * refs) for the given bytenr to the refs list, merges duplicates and resolves
 847  * indirect refs to their parent bytenr.
 848  * When roots are found, they're added to the roots list
 849  *
 850  * FIXME some caching might speed things up
 851  */
 852 static int find_parent_nodes(struct btrfs_trans_handle *trans,
 853                              struct btrfs_fs_info *fs_info, u64 bytenr,
 854                              u64 time_seq, struct ulist *refs,
 855                              struct ulist *roots, const u64 *extent_item_pos)
 856 {
 857         struct btrfs_key key;
 858         struct btrfs_path *path;
 859         struct btrfs_delayed_ref_root *delayed_refs = NULL;
 860         struct btrfs_delayed_ref_head *head;
 861         int info_level = 0;
 862         int ret;
 863         struct list_head prefs_delayed;
 864         struct list_head prefs;
 865         struct __prelim_ref *ref;
 866         struct extent_inode_elem *eie = NULL;
 867         u64 total_refs = 0;
 868
 869         INIT_LIST_HEAD(&prefs);
 870         INIT_LIST_HEAD(&prefs_delayed);
 871
 872         key.objectid = bytenr;
 873         key.offset = (u64)-1;
 874         if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
 875                 key.type = BTRFS_METADATA_ITEM_KEY;
 876         else
 877                 key.type = BTRFS_EXTENT_ITEM_KEY;
 878
 879         path = btrfs_alloc_path();
 880         if (!path)
 881                 return -ENOMEM;
 882         if (!trans) {
 883                 path->search_commit_root = 1;
 884                 path->skip_locking = 1;
 885         }
 886
 887         /*
 888          * grab both a lock on the path and a lock on the delayed ref head.
 889          * We need both to get a consistent picture of how the refs look
 890          * at a specified point in time
 891          */
 892 again:
 893         head = NULL;
 894
 895         ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
 896         if (ret < 0)
 897                 goto out;
 898         BUG_ON(ret == 0);
 899
 900         if (trans) {
 901                 /*
 902                  * look if there are updates for this ref queued and lock the
 903                  * head
 904                  */
 905                 delayed_refs = &trans->transaction->delayed_refs;
 906                 spin_lock(&delayed_refs->lock);
 907                 head = btrfs_find_delayed_ref_head(trans, bytenr);
 908                 if (head) {
 909                         if (!mutex_trylock(&head->mutex)) {
 910                                 atomic_inc(&head->node.refs);
 911                                 spin_unlock(&delayed_refs->lock);
 912
 913                                 btrfs_release_path(path);
 914
 915                                 /*
 916                                  * Mutex was contended, block until it's
 917                                  * released and try again
 918                                  */
 919                                 mutex_lock(&head->mutex);
 920                                 mutex_unlock(&head->mutex);
 921                                 btrfs_put_delayed_ref(&head->node);
 922                                 goto again;
 923                         }
 924                         spin_unlock(&delayed_refs->lock);
 925                         ret = __add_delayed_refs(head, time_seq,
 926                                                  &prefs_delayed, &total_refs);
 927                         mutex_unlock(&head->mutex);
 928                         if (ret)
 929                                 goto out;
 930                 } else {
 931                         spin_unlock(&delayed_refs->lock);
 932                 }
 933         }
 934
 935         if (path->slots[0]) {
 936                 struct extent_buffer *leaf;
 937                 int slot;
 938
 939                 path->slots[0]--;
 940                 leaf = path->nodes[0];
 941                 slot = path->slots[0];
 942                 btrfs_item_key_to_cpu(leaf, &key, slot);
 943                 if (key.objectid == bytenr &&
 944                     (key.type == BTRFS_EXTENT_ITEM_KEY ||
 945                      key.type == BTRFS_METADATA_ITEM_KEY)) {
 946                         ret = __add_inline_refs(fs_info, path, bytenr,
 947                                                 &info_level, &prefs,
 948                                                 &total_refs);
 949                         if (ret)
 950                                 goto out;
 951                         ret = __add_keyed_refs(fs_info, path, bytenr,
 952                                                info_level, &prefs);
 953                         if (ret)
 954                                 goto out;
 955                 }
 956         }
 957         btrfs_release_path(path);
 958
 959         list_splice_init(&prefs_delayed, &prefs);
 960
 961         ret = __add_missing_keys(fs_info, &prefs);
 962         if (ret)
 963                 goto out;
 964
 965         __merge_refs(&prefs, 1);
 966
 967         ret = __resolve_indirect_refs(fs_info, path, time_seq, &prefs,
 968                                       extent_item_pos, total_refs);
 969         if (ret)
 970                 goto out;
 971
 972         __merge_refs(&prefs, 2);
 973
 974         while (!list_empty(&prefs)) {
 975                 ref = list_first_entry(&prefs, struct __prelim_ref, list);
 976                 WARN_ON(ref->count < 0);
 977                 if (roots && ref->count && ref->root_id && ref->parent == 0) {
 978                         /* no parent == root of tree */
 979                         ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS);
 980                         if (ret < 0)
 981                                 goto out;
 982                 }
 983                 if (ref->count && ref->parent) {
 984                         if (extent_item_pos && !ref->inode_list) {
 985                                 u32 bsz;
 986                                 struct extent_buffer *eb;
 987                                 bsz = btrfs_level_size(fs_info->extent_root,
 988                                                         info_level);
 989                                 eb = read_tree_block(fs_info->extent_root,
 990                                                            ref->parent, bsz, 0);
 991                                 if (!eb || !extent_buffer_uptodate(eb)) {
 992                                         free_extent_buffer(eb);
 993                                         ret = -EIO;
 994                                         goto out;
 995                                 }
 996                                 ret = find_extent_in_eb(eb, bytenr,
 997                                                         *extent_item_pos, &eie);
 998                                 free_extent_buffer(eb);
 999                                 if (ret < 0)
1000                                         goto out;
1001                                 ref->inode_list = eie;
1002                         }
1003                         ret = ulist_add_merge(refs, ref->parent,
1004                                               (uintptr_t)ref->inode_list,
1005                                               (u64 *)&eie, GFP_NOFS);
1006                         if (ret < 0)
1007                                 goto out;
1008                         if (!ret && extent_item_pos) {
1009                                 /*
1010                                  * we've recorded that parent, so we must extend
1011                                  * its inode list here
1012                                  */
1013                                 BUG_ON(!eie);
1014                                 while (eie->next)
1015                                         eie = eie->next;
1016                                 eie->next = ref->inode_list;
1017                         }
1018                         eie = NULL;
1019                 }
1020                 list_del(&ref->list);
1021                 kmem_cache_free(btrfs_prelim_ref_cache, ref);
1022         }
1023
1024 out:
1025         btrfs_free_path(path);
1026         while (!list_empty(&prefs)) {
1027                 ref = list_first_entry(&prefs, struct __prelim_ref, list);
1028                 list_del(&ref->list);
1029                 kmem_cache_free(btrfs_prelim_ref_cache, ref);
1030         }
1031         while (!list_empty(&prefs_delayed)) {
1032                 ref = list_first_entry(&prefs_delayed, struct __prelim_ref,
1033                                        list);
1034                 list_del(&ref->list);
1035                 kmem_cache_free(btrfs_prelim_ref_cache, ref);
1036         }
1037         if (ret < 0)
1038                 free_inode_elem_list(eie);
1039         return ret;
1040 }
1041
1042 static void free_leaf_list(struct ulist *blocks)
1043 {
1044         struct ulist_node *node = NULL;
1045         struct extent_inode_elem *eie;
1046         struct ulist_iterator uiter;
1047
1048         ULIST_ITER_INIT(&uiter);
1049         while ((node = ulist_next(blocks, &uiter))) {
1050                 if (!node->aux)
1051                         continue;
1052                 eie = (struct extent_inode_elem *)(uintptr_t)node->aux;
1053                 free_inode_elem_list(eie);
1054                 node->aux = 0;
1055         }
1056
1057         ulist_free(blocks);
1058 }
1059
1060 /*
1061  * Finds all leafs with a reference to the specified combination of bytenr and
1062  * offset. key_list_head will point to a list of corresponding keys (caller must
1063  * free each list element). The leafs will be stored in the leafs ulist, which
1064  * must be freed with ulist_free.
1065  *
1066  * returns 0 on success, <0 on error
1067  */
1068 static int btrfs_find_all_leafs(struct btrfs_trans_handle *trans,
1069                                 struct btrfs_fs_info *fs_info, u64 bytenr,
1070                                 u64 time_seq, struct ulist **leafs,
1071                                 const u64 *extent_item_pos)
1072 {
1073         int ret;
1074
1075         *leafs = ulist_alloc(GFP_NOFS);
1076         if (!*leafs)
1077                 return -ENOMEM;
1078
1079         ret = find_parent_nodes(trans, fs_info, bytenr,
1080                                 time_seq, *leafs, NULL, extent_item_pos);
1081         if (ret < 0 && ret != -ENOENT) {
1082                 free_leaf_list(*leafs);
1083                 return ret;
1084         }
1085
1086         return 0;
1087 }
1088
1089 /*
1090  * walk all backrefs for a given extent to find all roots that reference this
1091  * extent. Walking a backref means finding all extents that reference this
1092  * extent and in turn walk the backrefs of those, too. Naturally this is a
1093  * recursive process, but here it is implemented in an iterative fashion: We
1094  * find all referencing extents for the extent in question and put them on a
1095  * list. In turn, we find all referencing extents for those, further appending
1096  * to the list. The way we iterate the list allows adding more elements after
1097  * the current while iterating. The process stops when we reach the end of the
1098  * list. Found roots are added to the roots list.
1099  *
1100  * returns 0 on success, < 0 on error.
1101  */
1102 int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1103                                 struct btrfs_fs_info *fs_info, u64 bytenr,
1104                                 u64 time_seq, struct ulist **roots)
1105 {
1106         struct ulist *tmp;
1107         struct ulist_node *node = NULL;
1108         struct ulist_iterator uiter;
1109         int ret;
1110
1111         tmp = ulist_alloc(GFP_NOFS);
1112         if (!tmp)
1113                 return -ENOMEM;
1114         *roots = ulist_alloc(GFP_NOFS);
1115         if (!*roots) {
1116                 ulist_free(tmp);
1117                 return -ENOMEM;
1118         }
1119
1120         ULIST_ITER_INIT(&uiter);
1121         while (1) {
1122                 ret = find_parent_nodes(trans, fs_info, bytenr,
1123                                         time_seq, tmp, *roots, NULL);
1124                 if (ret < 0 && ret != -ENOENT) {
1125                         ulist_free(tmp);
1126                         ulist_free(*roots);
1127                         return ret;
1128                 }
1129                 node = ulist_next(tmp, &uiter);
1130                 if (!node)
1131                         break;
1132                 bytenr = node->val;
1133                 cond_resched();
1134         }
1135
1136         ulist_free(tmp);
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 btrfs_find_item(fs_root, path, inum, ioff,
1148                         BTRFS_INODE_ITEM_KEY, &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 btrfs_find_item(fs_root, path, inum, ioff,
1156                         BTRFS_INODE_REF_KEY, 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
1337         ret = btrfs_previous_extent_item(fs_info->extent_root, path, 0);
1338         if (ret) {
1339                 if (ret > 0)
1340                         ret = -ENOENT;
1341                 return ret;
1342         }
1343         btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]);
1344         if (found_key->type == BTRFS_METADATA_ITEM_KEY)
1345                 size = fs_info->extent_root->leafsize;
1346         else if (found_key->type == BTRFS_EXTENT_ITEM_KEY)
1347                 size = found_key->offset;
1348
1349         if (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                 ret = inode_ref_info(inum, parent ? parent+1 : 0, fs_root, path,
1603                                      &found_key);
1604                 if (ret < 0)
1605                         break;
1606                 if (ret) {
1607                         ret = found ? 0 : -ENOENT;
1608                         break;
1609                 }
1610                 ++found;
1611
1612                 parent = found_key.offset;
1613                 slot = path->slots[0];
1614                 eb = btrfs_clone_extent_buffer(path->nodes[0]);
1615                 if (!eb) {
1616                         ret = -ENOMEM;
1617                         break;
1618                 }
1619                 extent_buffer_get(eb);
1620                 btrfs_tree_read_lock(eb);
1621                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1622                 btrfs_release_path(path);
1623
1624                 item = btrfs_item_nr(slot);
1625                 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1626
1627                 for (cur = 0; cur < btrfs_item_size(eb, item); cur += len) {
1628                         name_len = btrfs_inode_ref_name_len(eb, iref);
1629                         /* path must be released before calling iterate()! */
1630                         pr_debug("following ref at offset %u for inode %llu in "
1631                                  "tree %llu\n", cur, found_key.objectid,
1632                                  fs_root->objectid);
1633                         ret = iterate(parent, name_len,
1634                                       (unsigned long)(iref + 1), eb, ctx);
1635                         if (ret)
1636                                 break;
1637                         len = sizeof(*iref) + name_len;
1638                         iref = (struct btrfs_inode_ref *)((char *)iref + len);
1639                 }
1640                 btrfs_tree_read_unlock_blocking(eb);
1641                 free_extent_buffer(eb);
1642         }
1643
1644         btrfs_release_path(path);
1645
1646         return ret;
1647 }
1648
1649 static int iterate_inode_extrefs(u64 inum, struct btrfs_root *fs_root,
1650                                  struct btrfs_path *path,
1651                                  iterate_irefs_t *iterate, void *ctx)
1652 {
1653         int ret;
1654         int slot;
1655         u64 offset = 0;
1656         u64 parent;
1657         int found = 0;
1658         struct extent_buffer *eb;
1659         struct btrfs_inode_extref *extref;
1660         struct extent_buffer *leaf;
1661         u32 item_size;
1662         u32 cur_offset;
1663         unsigned long ptr;
1664
1665         while (1) {
1666                 ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref,
1667                                             &offset);
1668                 if (ret < 0)
1669                         break;
1670                 if (ret) {
1671                         ret = found ? 0 : -ENOENT;
1672                         break;
1673                 }
1674                 ++found;
1675
1676                 slot = path->slots[0];
1677                 eb = btrfs_clone_extent_buffer(path->nodes[0]);
1678                 if (!eb) {
1679                         ret = -ENOMEM;
1680                         break;
1681                 }
1682                 extent_buffer_get(eb);
1683
1684                 btrfs_tree_read_lock(eb);
1685                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1686                 btrfs_release_path(path);
1687
1688                 leaf = path->nodes[0];
1689                 item_size = btrfs_item_size_nr(leaf, slot);
1690                 ptr = btrfs_item_ptr_offset(leaf, slot);
1691                 cur_offset = 0;
1692
1693                 while (cur_offset < item_size) {
1694                         u32 name_len;
1695
1696                         extref = (struct btrfs_inode_extref *)(ptr + cur_offset);
1697                         parent = btrfs_inode_extref_parent(eb, extref);
1698                         name_len = btrfs_inode_extref_name_len(eb, extref);
1699                         ret = iterate(parent, name_len,
1700                                       (unsigned long)&extref->name, eb, ctx);
1701                         if (ret)
1702                                 break;
1703
1704                         cur_offset += btrfs_inode_extref_name_len(leaf, extref);
1705                         cur_offset += sizeof(*extref);
1706                 }
1707                 btrfs_tree_read_unlock_blocking(eb);
1708                 free_extent_buffer(eb);
1709
1710                 offset++;
1711         }
1712
1713         btrfs_release_path(path);
1714
1715         return ret;
1716 }
1717
1718 static int iterate_irefs(u64 inum, struct btrfs_root *fs_root,
1719                          struct btrfs_path *path, iterate_irefs_t *iterate,
1720                          void *ctx)
1721 {
1722         int ret;
1723         int found_refs = 0;
1724
1725         ret = iterate_inode_refs(inum, fs_root, path, iterate, ctx);
1726         if (!ret)
1727                 ++found_refs;
1728         else if (ret != -ENOENT)
1729                 return ret;
1730
1731         ret = iterate_inode_extrefs(inum, fs_root, path, iterate, ctx);
1732         if (ret == -ENOENT && found_refs)
1733                 return 0;
1734
1735         return ret;
1736 }
1737
1738 /*
1739  * returns 0 if the path could be dumped (probably truncated)
1740  * returns <0 in case of an error
1741  */
1742 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
1743                          struct extent_buffer *eb, void *ctx)
1744 {
1745         struct inode_fs_paths *ipath = ctx;
1746         char *fspath;
1747         char *fspath_min;
1748         int i = ipath->fspath->elem_cnt;
1749         const int s_ptr = sizeof(char *);
1750         u32 bytes_left;
1751
1752         bytes_left = ipath->fspath->bytes_left > s_ptr ?
1753                                         ipath->fspath->bytes_left - s_ptr : 0;
1754
1755         fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
1756         fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len,
1757                                    name_off, eb, inum, fspath_min, bytes_left);
1758         if (IS_ERR(fspath))
1759                 return PTR_ERR(fspath);
1760
1761         if (fspath > fspath_min) {
1762                 ipath->fspath->val[i] = (u64)(unsigned long)fspath;
1763                 ++ipath->fspath->elem_cnt;
1764                 ipath->fspath->bytes_left = fspath - fspath_min;
1765         } else {
1766                 ++ipath->fspath->elem_missed;
1767                 ipath->fspath->bytes_missing += fspath_min - fspath;
1768                 ipath->fspath->bytes_left = 0;
1769         }
1770
1771         return 0;
1772 }
1773
1774 /*
1775  * this dumps all file system paths to the inode into the ipath struct, provided
1776  * is has been created large enough. each path is zero-terminated and accessed
1777  * from ipath->fspath->val[i].
1778  * when it returns, there are ipath->fspath->elem_cnt number of paths available
1779  * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the
1780  * number of missed paths in recored in ipath->fspath->elem_missed, otherwise,
1781  * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
1782  * have been needed to return all paths.
1783  */
1784 int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
1785 {
1786         return iterate_irefs(inum, ipath->fs_root, ipath->btrfs_path,
1787                              inode_to_path, ipath);
1788 }
1789
1790 struct btrfs_data_container *init_data_container(u32 total_bytes)
1791 {
1792         struct btrfs_data_container *data;
1793         size_t alloc_bytes;
1794
1795         alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
1796         data = vmalloc(alloc_bytes);
1797         if (!data)
1798                 return ERR_PTR(-ENOMEM);
1799
1800         if (total_bytes >= sizeof(*data)) {
1801                 data->bytes_left = total_bytes - sizeof(*data);
1802                 data->bytes_missing = 0;
1803         } else {
1804                 data->bytes_missing = sizeof(*data) - total_bytes;
1805                 data->bytes_left = 0;
1806         }
1807
1808         data->elem_cnt = 0;
1809         data->elem_missed = 0;
1810
1811         return data;
1812 }
1813
1814 /*
1815  * allocates space to return multiple file system paths for an inode.
1816  * total_bytes to allocate are passed, note that space usable for actual path
1817  * information will be total_bytes - sizeof(struct inode_fs_paths).
1818  * the returned pointer must be freed with free_ipath() in the end.
1819  */
1820 struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
1821                                         struct btrfs_path *path)
1822 {
1823         struct inode_fs_paths *ifp;
1824         struct btrfs_data_container *fspath;
1825
1826         fspath = init_data_container(total_bytes);
1827         if (IS_ERR(fspath))
1828                 return (void *)fspath;
1829
1830         ifp = kmalloc(sizeof(*ifp), GFP_NOFS);
1831         if (!ifp) {
1832                 kfree(fspath);
1833                 return ERR_PTR(-ENOMEM);
1834         }
1835
1836         ifp->btrfs_path = path;
1837         ifp->fspath = fspath;
1838         ifp->fs_root = fs_root;
1839
1840         return ifp;
1841 }
1842
1843 void free_ipath(struct inode_fs_paths *ipath)
1844 {
1845         if (!ipath)
1846                 return;
1847         vfree(ipath->fspath);
1848         kfree(ipath);
1849 }