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         if (path->search_commit_root)
 334                 root_level = btrfs_header_level(root->commit_root);
 335         else
 336                 root_level = btrfs_old_root_level(root, time_seq);
 337
 338         if (root_level + 1 == level) {
 339                 srcu_read_unlock(&fs_info->subvol_srcu, index);
 340                 goto out;
 341         }
 342
 343         path->lowest_level = level;
 344         ret = btrfs_search_old_slot(root, &ref->key_for_search, path, time_seq);
 345
 346         /* root node has been locked, we can release @subvol_srcu safely here */
 347         srcu_read_unlock(&fs_info->subvol_srcu, index);
 348
 349         pr_debug("search slot in root %llu (level %d, ref count %d) returned "
 350                  "%d for key (%llu %u %llu)\n",
 351                  ref->root_id, level, ref->count, ret,
 352                  ref->key_for_search.objectid, ref->key_for_search.type,
 353                  ref->key_for_search.offset);
 354         if (ret < 0)
 355                 goto out;
 356
 357         eb = path->nodes[level];
 358         while (!eb) {
 359                 if (WARN_ON(!level)) {
 360                         ret = 1;
 361                         goto out;
 362                 }
 363                 level--;
 364                 eb = path->nodes[level];
 365         }
 366
 367         ret = add_all_parents(root, path, parents, ref, level, time_seq,
 368                               extent_item_pos, total_refs);
 369 out:
 370         path->lowest_level = 0;
 371         btrfs_release_path(path);
 372         return ret;
 373 }
 374
 375 /*
 376  * resolve all indirect backrefs from the list
 377  */
 378 static int __resolve_indirect_refs(struct btrfs_fs_info *fs_info,
 379                                    struct btrfs_path *path, u64 time_seq,
 380                                    struct list_head *head,
 381                                    const u64 *extent_item_pos, u64 total_refs)
 382 {
 383         int err;
 384         int ret = 0;
 385         struct __prelim_ref *ref;
 386         struct __prelim_ref *ref_safe;
 387         struct __prelim_ref *new_ref;
 388         struct ulist *parents;
 389         struct ulist_node *node;
 390         struct ulist_iterator uiter;
 391
 392         parents = ulist_alloc(GFP_NOFS);
 393         if (!parents)
 394                 return -ENOMEM;
 395
 396         /*
 397          * _safe allows us to insert directly after the current item without
 398          * iterating over the newly inserted items.
 399          * we're also allowed to re-assign ref during iteration.
 400          */
 401         list_for_each_entry_safe(ref, ref_safe, head, list) {
 402                 if (ref->parent)        /* already direct */
 403                         continue;
 404                 if (ref->count == 0)
 405                         continue;
 406                 err = __resolve_indirect_ref(fs_info, path, time_seq, ref,
 407                                              parents, extent_item_pos,
 408                                              total_refs);
 409                 /*
 410                  * we can only tolerate ENOENT,otherwise,we should catch error
 411                  * and return directly.
 412                  */
 413                 if (err == -ENOENT) {
 414                         continue;
 415                 } else if (err) {
 416                         ret = err;
 417                         goto out;
 418                 }
 419
 420                 /* we put the first parent into the ref at hand */
 421                 ULIST_ITER_INIT(&uiter);
 422                 node = ulist_next(parents, &uiter);
 423                 ref->parent = node ? node->val : 0;
 424                 ref->inode_list = node ?
 425                         (struct extent_inode_elem *)(uintptr_t)node->aux : NULL;
 426
 427                 /* additional parents require new refs being added here */
 428                 while ((node = ulist_next(parents, &uiter))) {
 429                         new_ref = kmem_cache_alloc(btrfs_prelim_ref_cache,
 430                                                    GFP_NOFS);
 431                         if (!new_ref) {
 432                                 ret = -ENOMEM;
 433                                 goto out;
 434                         }
 435                         memcpy(new_ref, ref, sizeof(*ref));
 436                         new_ref->parent = node->val;
 437                         new_ref->inode_list = (struct extent_inode_elem *)
 438                                                         (uintptr_t)node->aux;
 439                         list_add(&new_ref->list, &ref->list);
 440                 }
 441                 ulist_reinit(parents);
 442         }
 443 out:
 444         ulist_free(parents);
 445         return ret;
 446 }
 447
 448 static inline int ref_for_same_block(struct __prelim_ref *ref1,
 449                                      struct __prelim_ref *ref2)
 450 {
 451         if (ref1->level != ref2->level)
 452                 return 0;
 453         if (ref1->root_id != ref2->root_id)
 454                 return 0;
 455         if (ref1->key_for_search.type != ref2->key_for_search.type)
 456                 return 0;
 457         if (ref1->key_for_search.objectid != ref2->key_for_search.objectid)
 458                 return 0;
 459         if (ref1->key_for_search.offset != ref2->key_for_search.offset)
 460                 return 0;
 461         if (ref1->parent != ref2->parent)
 462                 return 0;
 463
 464         return 1;
 465 }
 466
 467 /*
 468  * read tree blocks and add keys where required.
 469  */
 470 static int __add_missing_keys(struct btrfs_fs_info *fs_info,
 471                               struct list_head *head)
 472 {
 473         struct list_head *pos;
 474         struct extent_buffer *eb;
 475
 476         list_for_each(pos, head) {
 477                 struct __prelim_ref *ref;
 478                 ref = list_entry(pos, struct __prelim_ref, list);
 479
 480                 if (ref->parent)
 481                         continue;
 482                 if (ref->key_for_search.type)
 483                         continue;
 484                 BUG_ON(!ref->wanted_disk_byte);
 485                 eb = read_tree_block(fs_info->tree_root, ref->wanted_disk_byte,
 486                                      fs_info->tree_root->leafsize, 0);
 487                 if (!eb || !extent_buffer_uptodate(eb)) {
 488                         free_extent_buffer(eb);
 489                         return -EIO;
 490                 }
 491                 btrfs_tree_read_lock(eb);
 492                 if (btrfs_header_level(eb) == 0)
 493                         btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
 494                 else
 495                         btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
 496                 btrfs_tree_read_unlock(eb);
 497                 free_extent_buffer(eb);
 498         }
 499         return 0;
 500 }
 501
 502 /*
 503  * merge two lists of backrefs and adjust counts accordingly
 504  *
 505  * mode = 1: merge identical keys, if key is set
 506  *    FIXME: if we add more keys in __add_prelim_ref, we can merge more here.
 507  *           additionally, we could even add a key range for the blocks we
 508  *           looked into to merge even more (-> replace unresolved refs by those
 509  *           having a parent).
 510  * mode = 2: merge identical parents
 511  */
 512 static void __merge_refs(struct list_head *head, int mode)
 513 {
 514         struct list_head *pos1;
 515
 516         list_for_each(pos1, head) {
 517                 struct list_head *n2;
 518                 struct list_head *pos2;
 519                 struct __prelim_ref *ref1;
 520
 521                 ref1 = list_entry(pos1, struct __prelim_ref, list);
 522
 523                 for (pos2 = pos1->next, n2 = pos2->next; pos2 != head;
 524                      pos2 = n2, n2 = pos2->next) {
 525                         struct __prelim_ref *ref2;
 526                         struct __prelim_ref *xchg;
 527                         struct extent_inode_elem *eie;
 528
 529                         ref2 = list_entry(pos2, struct __prelim_ref, list);
 530
 531                         if (mode == 1) {
 532                                 if (!ref_for_same_block(ref1, ref2))
 533                                         continue;
 534                                 if (!ref1->parent && ref2->parent) {
 535                                         xchg = ref1;
 536                                         ref1 = ref2;
 537                                         ref2 = xchg;
 538                                 }
 539                         } else {
 540                                 if (ref1->parent != ref2->parent)
 541                                         continue;
 542                         }
 543
 544                         eie = ref1->inode_list;
 545                         while (eie && eie->next)
 546                                 eie = eie->next;
 547                         if (eie)
 548                                 eie->next = ref2->inode_list;
 549                         else
 550                                 ref1->inode_list = ref2->inode_list;
 551                         ref1->count += ref2->count;
 552
 553                         list_del(&ref2->list);
 554                         kmem_cache_free(btrfs_prelim_ref_cache, ref2);
 555                 }
 556
 557         }
 558 }
 559
 560 /*
 561  * add all currently queued delayed refs from this head whose seq nr is
 562  * smaller or equal that seq to the list
 563  */
 564 static int __add_delayed_refs(struct btrfs_delayed_ref_head *head, u64 seq,
 565                               struct list_head *prefs, u64 *total_refs)
 566 {
 567         struct btrfs_delayed_extent_op *extent_op = head->extent_op;
 568         struct rb_node *n = &head->node.rb_node;
 569         struct btrfs_key key;
 570         struct btrfs_key op_key = {0};
 571         int sgn;
 572         int ret = 0;
 573
 574         if (extent_op && extent_op->update_key)
 575                 btrfs_disk_key_to_cpu(&op_key, &extent_op->key);
 576
 577         spin_lock(&head->lock);
 578         n = rb_first(&head->ref_root);
 579         while (n) {
 580                 struct btrfs_delayed_ref_node *node;
 581                 node = rb_entry(n, struct btrfs_delayed_ref_node,
 582                                 rb_node);
 583                 n = rb_next(n);
 584                 if (node->seq > seq)
 585                         continue;
 586
 587                 switch (node->action) {
 588                 case BTRFS_ADD_DELAYED_EXTENT:
 589                 case BTRFS_UPDATE_DELAYED_HEAD:
 590                         WARN_ON(1);
 591                         continue;
 592                 case BTRFS_ADD_DELAYED_REF:
 593                         sgn = 1;
 594                         break;
 595                 case BTRFS_DROP_DELAYED_REF:
 596                         sgn = -1;
 597                         break;
 598                 default:
 599                         BUG_ON(1);
 600                 }
 601                 *total_refs += (node->ref_mod * sgn);
 602                 switch (node->type) {
 603                 case BTRFS_TREE_BLOCK_REF_KEY: {
 604                         struct btrfs_delayed_tree_ref *ref;
 605
 606                         ref = btrfs_delayed_node_to_tree_ref(node);
 607                         ret = __add_prelim_ref(prefs, ref->root, &op_key,
 608                                                ref->level + 1, 0, node->bytenr,
 609                                                node->ref_mod * sgn, GFP_ATOMIC);
 610                         break;
 611                 }
 612                 case BTRFS_SHARED_BLOCK_REF_KEY: {
 613                         struct btrfs_delayed_tree_ref *ref;
 614
 615                         ref = btrfs_delayed_node_to_tree_ref(node);
 616                         ret = __add_prelim_ref(prefs, ref->root, NULL,
 617                                                ref->level + 1, ref->parent,
 618                                                node->bytenr,
 619                                                node->ref_mod * sgn, GFP_ATOMIC);
 620                         break;
 621                 }
 622                 case BTRFS_EXTENT_DATA_REF_KEY: {
 623                         struct btrfs_delayed_data_ref *ref;
 624                         ref = btrfs_delayed_node_to_data_ref(node);
 625
 626                         key.objectid = ref->objectid;
 627                         key.type = BTRFS_EXTENT_DATA_KEY;
 628                         key.offset = ref->offset;
 629                         ret = __add_prelim_ref(prefs, ref->root, &key, 0, 0,
 630                                                node->bytenr,
 631                                                node->ref_mod * sgn, GFP_ATOMIC);
 632                         break;
 633                 }
 634                 case BTRFS_SHARED_DATA_REF_KEY: {
 635                         struct btrfs_delayed_data_ref *ref;
 636
 637                         ref = btrfs_delayed_node_to_data_ref(node);
 638
 639                         key.objectid = ref->objectid;
 640                         key.type = BTRFS_EXTENT_DATA_KEY;
 641                         key.offset = ref->offset;
 642                         ret = __add_prelim_ref(prefs, ref->root, &key, 0,
 643                                                ref->parent, node->bytenr,
 644                                                node->ref_mod * sgn, GFP_ATOMIC);
 645                         break;
 646                 }
 647                 default:
 648                         WARN_ON(1);
 649                 }
 650                 if (ret)
 651                         break;
 652         }
 653         spin_unlock(&head->lock);
 654         return ret;
 655 }
 656
 657 /*
 658  * add all inline backrefs for bytenr to the list
 659  */
 660 static int __add_inline_refs(struct btrfs_fs_info *fs_info,
 661                              struct btrfs_path *path, u64 bytenr,
 662                              int *info_level, struct list_head *prefs,
 663                              u64 *total_refs)
 664 {
 665         int ret = 0;
 666         int slot;
 667         struct extent_buffer *leaf;
 668         struct btrfs_key key;
 669         struct btrfs_key found_key;
 670         unsigned long ptr;
 671         unsigned long end;
 672         struct btrfs_extent_item *ei;
 673         u64 flags;
 674         u64 item_size;
 675
 676         /*
 677          * enumerate all inline refs
 678          */
 679         leaf = path->nodes[0];
 680         slot = path->slots[0];
 681
 682         item_size = btrfs_item_size_nr(leaf, slot);
 683         BUG_ON(item_size < sizeof(*ei));
 684
 685         ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
 686         flags = btrfs_extent_flags(leaf, ei);
 687         *total_refs += btrfs_extent_refs(leaf, ei);
 688         btrfs_item_key_to_cpu(leaf, &found_key, slot);
 689
 690         ptr = (unsigned long)(ei + 1);
 691         end = (unsigned long)ei + item_size;
 692
 693         if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
 694             flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
 695                 struct btrfs_tree_block_info *info;
 696
 697                 info = (struct btrfs_tree_block_info *)ptr;
 698                 *info_level = btrfs_tree_block_level(leaf, info);
 699                 ptr += sizeof(struct btrfs_tree_block_info);
 700                 BUG_ON(ptr > end);
 701         } else if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
 702                 *info_level = found_key.offset;
 703         } else {
 704                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
 705         }
 706
 707         while (ptr < end) {
 708                 struct btrfs_extent_inline_ref *iref;
 709                 u64 offset;
 710                 int type;
 711
 712                 iref = (struct btrfs_extent_inline_ref *)ptr;
 713                 type = btrfs_extent_inline_ref_type(leaf, iref);
 714                 offset = btrfs_extent_inline_ref_offset(leaf, iref);
 715
 716                 switch (type) {
 717                 case BTRFS_SHARED_BLOCK_REF_KEY:
 718                         ret = __add_prelim_ref(prefs, 0, NULL,
 719                                                 *info_level + 1, offset,
 720                                                 bytenr, 1, GFP_NOFS);
 721                         break;
 722                 case BTRFS_SHARED_DATA_REF_KEY: {
 723                         struct btrfs_shared_data_ref *sdref;
 724                         int count;
 725
 726                         sdref = (struct btrfs_shared_data_ref *)(iref + 1);
 727                         count = btrfs_shared_data_ref_count(leaf, sdref);
 728                         ret = __add_prelim_ref(prefs, 0, NULL, 0, offset,
 729                                                bytenr, count, GFP_NOFS);
 730                         break;
 731                 }
 732                 case BTRFS_TREE_BLOCK_REF_KEY:
 733                         ret = __add_prelim_ref(prefs, offset, NULL,
 734                                                *info_level + 1, 0,
 735                                                bytenr, 1, GFP_NOFS);
 736                         break;
 737                 case BTRFS_EXTENT_DATA_REF_KEY: {
 738                         struct btrfs_extent_data_ref *dref;
 739                         int count;
 740                         u64 root;
 741
 742                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
 743                         count = btrfs_extent_data_ref_count(leaf, dref);
 744                         key.objectid = btrfs_extent_data_ref_objectid(leaf,
 745                                                                       dref);
 746                         key.type = BTRFS_EXTENT_DATA_KEY;
 747                         key.offset = btrfs_extent_data_ref_offset(leaf, dref);
 748                         root = btrfs_extent_data_ref_root(leaf, dref);
 749                         ret = __add_prelim_ref(prefs, root, &key, 0, 0,
 750                                                bytenr, count, GFP_NOFS);
 751                         break;
 752                 }
 753                 default:
 754                         WARN_ON(1);
 755                 }
 756                 if (ret)
 757                         return ret;
 758                 ptr += btrfs_extent_inline_ref_size(type);
 759         }
 760
 761         return 0;
 762 }
 763
 764 /*
 765  * add all non-inline backrefs for bytenr to the list
 766  */
 767 static int __add_keyed_refs(struct btrfs_fs_info *fs_info,
 768                             struct btrfs_path *path, u64 bytenr,
 769                             int info_level, struct list_head *prefs)
 770 {
 771         struct btrfs_root *extent_root = fs_info->extent_root;
 772         int ret;
 773         int slot;
 774         struct extent_buffer *leaf;
 775         struct btrfs_key key;
 776
 777         while (1) {
 778                 ret = btrfs_next_item(extent_root, path);
 779                 if (ret < 0)
 780                         break;
 781                 if (ret) {
 782                         ret = 0;
 783                         break;
 784                 }
 785
 786                 slot = path->slots[0];
 787                 leaf = path->nodes[0];
 788                 btrfs_item_key_to_cpu(leaf, &key, slot);
 789
 790                 if (key.objectid != bytenr)
 791                         break;
 792                 if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
 793                         continue;
 794                 if (key.type > BTRFS_SHARED_DATA_REF_KEY)
 795                         break;
 796
 797                 switch (key.type) {
 798                 case BTRFS_SHARED_BLOCK_REF_KEY:
 799                         ret = __add_prelim_ref(prefs, 0, NULL,
 800                                                 info_level + 1, key.offset,
 801                                                 bytenr, 1, GFP_NOFS);
 802                         break;
 803                 case BTRFS_SHARED_DATA_REF_KEY: {
 804                         struct btrfs_shared_data_ref *sdref;
 805                         int count;
 806
 807                         sdref = btrfs_item_ptr(leaf, slot,
 808                                               struct btrfs_shared_data_ref);
 809                         count = btrfs_shared_data_ref_count(leaf, sdref);
 810                         ret = __add_prelim_ref(prefs, 0, NULL, 0, key.offset,
 811                                                 bytenr, count, GFP_NOFS);
 812                         break;
 813                 }
 814                 case BTRFS_TREE_BLOCK_REF_KEY:
 815                         ret = __add_prelim_ref(prefs, key.offset, NULL,
 816                                                info_level + 1, 0,
 817                                                bytenr, 1, GFP_NOFS);
 818                         break;
 819                 case BTRFS_EXTENT_DATA_REF_KEY: {
 820                         struct btrfs_extent_data_ref *dref;
 821                         int count;
 822                         u64 root;
 823
 824                         dref = btrfs_item_ptr(leaf, slot,
 825                                               struct btrfs_extent_data_ref);
 826                         count = btrfs_extent_data_ref_count(leaf, dref);
 827                         key.objectid = btrfs_extent_data_ref_objectid(leaf,
 828                                                                       dref);
 829                         key.type = BTRFS_EXTENT_DATA_KEY;
 830                         key.offset = btrfs_extent_data_ref_offset(leaf, dref);
 831                         root = btrfs_extent_data_ref_root(leaf, dref);
 832                         ret = __add_prelim_ref(prefs, root, &key, 0, 0,
 833                                                bytenr, count, GFP_NOFS);
 834                         break;
 835                 }
 836                 default:
 837                         WARN_ON(1);
 838                 }
 839                 if (ret)
 840                         return ret;
 841
 842         }
 843
 844         return ret;
 845 }
 846
 847 /*
 848  * this adds all existing backrefs (inline backrefs, backrefs and delayed
 849  * refs) for the given bytenr to the refs list, merges duplicates and resolves
 850  * indirect refs to their parent bytenr.
 851  * When roots are found, they're added to the roots list
 852  *
 853  * FIXME some caching might speed things up
 854  */
 855 static int find_parent_nodes(struct btrfs_trans_handle *trans,
 856                              struct btrfs_fs_info *fs_info, u64 bytenr,
 857                              u64 time_seq, struct ulist *refs,
 858                              struct ulist *roots, const u64 *extent_item_pos)
 859 {
 860         struct btrfs_key key;
 861         struct btrfs_path *path;
 862         struct btrfs_delayed_ref_root *delayed_refs = NULL;
 863         struct btrfs_delayed_ref_head *head;
 864         int info_level = 0;
 865         int ret;
 866         struct list_head prefs_delayed;
 867         struct list_head prefs;
 868         struct __prelim_ref *ref;
 869         struct extent_inode_elem *eie = NULL;
 870         u64 total_refs = 0;
 871
 872         INIT_LIST_HEAD(&prefs);
 873         INIT_LIST_HEAD(&prefs_delayed);
 874
 875         key.objectid = bytenr;
 876         key.offset = (u64)-1;
 877         if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
 878                 key.type = BTRFS_METADATA_ITEM_KEY;
 879         else
 880                 key.type = BTRFS_EXTENT_ITEM_KEY;
 881
 882         path = btrfs_alloc_path();
 883         if (!path)
 884                 return -ENOMEM;
 885         if (!trans) {
 886                 path->search_commit_root = 1;
 887                 path->skip_locking = 1;
 888         }
 889
 890         /*
 891          * grab both a lock on the path and a lock on the delayed ref head.
 892          * We need both to get a consistent picture of how the refs look
 893          * at a specified point in time
 894          */
 895 again:
 896         head = NULL;
 897
 898         ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
 899         if (ret < 0)
 900                 goto out;
 901         BUG_ON(ret == 0);
 902
 903         if (trans) {
 904                 /*
 905                  * look if there are updates for this ref queued and lock the
 906                  * head
 907                  */
 908                 delayed_refs = &trans->transaction->delayed_refs;
 909                 spin_lock(&delayed_refs->lock);
 910                 head = btrfs_find_delayed_ref_head(trans, bytenr);
 911                 if (head) {
 912                         if (!mutex_trylock(&head->mutex)) {
 913                                 atomic_inc(&head->node.refs);
 914                                 spin_unlock(&delayed_refs->lock);
 915
 916                                 btrfs_release_path(path);
 917
 918                                 /*
 919                                  * Mutex was contended, block until it's
 920                                  * released and try again
 921                                  */
 922                                 mutex_lock(&head->mutex);
 923                                 mutex_unlock(&head->mutex);
 924                                 btrfs_put_delayed_ref(&head->node);
 925                                 goto again;
 926                         }
 927                         spin_unlock(&delayed_refs->lock);
 928                         ret = __add_delayed_refs(head, time_seq,
 929                                                  &prefs_delayed, &total_refs);
 930                         mutex_unlock(&head->mutex);
 931                         if (ret)
 932                                 goto out;
 933                 } else {
 934                         spin_unlock(&delayed_refs->lock);
 935                 }
 936         }
 937
 938         if (path->slots[0]) {
 939                 struct extent_buffer *leaf;
 940                 int slot;
 941
 942                 path->slots[0]--;
 943                 leaf = path->nodes[0];
 944                 slot = path->slots[0];
 945                 btrfs_item_key_to_cpu(leaf, &key, slot);
 946                 if (key.objectid == bytenr &&
 947                     (key.type == BTRFS_EXTENT_ITEM_KEY ||
 948                      key.type == BTRFS_METADATA_ITEM_KEY)) {
 949                         ret = __add_inline_refs(fs_info, path, bytenr,
 950                                                 &info_level, &prefs,
 951                                                 &total_refs);
 952                         if (ret)
 953                                 goto out;
 954                         ret = __add_keyed_refs(fs_info, path, bytenr,
 955                                                info_level, &prefs);
 956                         if (ret)
 957                                 goto out;
 958                 }
 959         }
 960         btrfs_release_path(path);
 961
 962         list_splice_init(&prefs_delayed, &prefs);
 963
 964         ret = __add_missing_keys(fs_info, &prefs);
 965         if (ret)
 966                 goto out;
 967
 968         __merge_refs(&prefs, 1);
 969
 970         ret = __resolve_indirect_refs(fs_info, path, time_seq, &prefs,
 971                                       extent_item_pos, total_refs);
 972         if (ret)
 973                 goto out;
 974
 975         __merge_refs(&prefs, 2);
 976
 977         while (!list_empty(&prefs)) {
 978                 ref = list_first_entry(&prefs, struct __prelim_ref, list);
 979                 WARN_ON(ref->count < 0);
 980                 if (roots && ref->count && ref->root_id && ref->parent == 0) {
 981                         /* no parent == root of tree */
 982                         ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS);
 983                         if (ret < 0)
 984                                 goto out;
 985                 }
 986                 if (ref->count && ref->parent) {
 987                         if (extent_item_pos && !ref->inode_list) {
 988                                 u32 bsz;
 989                                 struct extent_buffer *eb;
 990                                 bsz = btrfs_level_size(fs_info->extent_root,
 991                                                         info_level);
 992                                 eb = read_tree_block(fs_info->extent_root,
 993                                                            ref->parent, bsz, 0);
 994                                 if (!eb || !extent_buffer_uptodate(eb)) {
 995                                         free_extent_buffer(eb);
 996                                         ret = -EIO;
 997                                         goto out;
 998                                 }
 999                                 ret = find_extent_in_eb(eb, bytenr,
1000                                                         *extent_item_pos, &eie);
1001                                 free_extent_buffer(eb);
1002                                 if (ret < 0)
1003                                         goto out;
1004                                 ref->inode_list = eie;
1005                         }
1006                         ret = ulist_add_merge(refs, ref->parent,
1007                                               (uintptr_t)ref->inode_list,
1008                                               (u64 *)&eie, GFP_NOFS);
1009                         if (ret < 0)
1010                                 goto out;
1011                         if (!ret && extent_item_pos) {
1012                                 /*
1013                                  * we've recorded that parent, so we must extend
1014                                  * its inode list here
1015                                  */
1016                                 BUG_ON(!eie);
1017                                 while (eie->next)
1018                                         eie = eie->next;
1019                                 eie->next = ref->inode_list;
1020                         }
1021                         eie = NULL;
1022                 }
1023                 list_del(&ref->list);
1024                 kmem_cache_free(btrfs_prelim_ref_cache, ref);
1025         }
1026
1027 out:
1028         btrfs_free_path(path);
1029         while (!list_empty(&prefs)) {
1030                 ref = list_first_entry(&prefs, struct __prelim_ref, list);
1031                 list_del(&ref->list);
1032                 kmem_cache_free(btrfs_prelim_ref_cache, ref);
1033         }
1034         while (!list_empty(&prefs_delayed)) {
1035                 ref = list_first_entry(&prefs_delayed, struct __prelim_ref,
1036                                        list);
1037                 list_del(&ref->list);
1038                 kmem_cache_free(btrfs_prelim_ref_cache, ref);
1039         }
1040         if (ret < 0)
1041                 free_inode_elem_list(eie);
1042         return ret;
1043 }
1044
1045 static void free_leaf_list(struct ulist *blocks)
1046 {
1047         struct ulist_node *node = NULL;
1048         struct extent_inode_elem *eie;
1049         struct ulist_iterator uiter;
1050
1051         ULIST_ITER_INIT(&uiter);
1052         while ((node = ulist_next(blocks, &uiter))) {
1053                 if (!node->aux)
1054                         continue;
1055                 eie = (struct extent_inode_elem *)(uintptr_t)node->aux;
1056                 free_inode_elem_list(eie);
1057                 node->aux = 0;
1058         }
1059
1060         ulist_free(blocks);
1061 }
1062
1063 /*
1064  * Finds all leafs with a reference to the specified combination of bytenr and
1065  * offset. key_list_head will point to a list of corresponding keys (caller must
1066  * free each list element). The leafs will be stored in the leafs ulist, which
1067  * must be freed with ulist_free.
1068  *
1069  * returns 0 on success, <0 on error
1070  */
1071 static int btrfs_find_all_leafs(struct btrfs_trans_handle *trans,
1072                                 struct btrfs_fs_info *fs_info, u64 bytenr,
1073                                 u64 time_seq, struct ulist **leafs,
1074                                 const u64 *extent_item_pos)
1075 {
1076         int ret;
1077
1078         *leafs = ulist_alloc(GFP_NOFS);
1079         if (!*leafs)
1080                 return -ENOMEM;
1081
1082         ret = find_parent_nodes(trans, fs_info, bytenr,
1083                                 time_seq, *leafs, NULL, extent_item_pos);
1084         if (ret < 0 && ret != -ENOENT) {
1085                 free_leaf_list(*leafs);
1086                 return ret;
1087         }
1088
1089         return 0;
1090 }
1091
1092 /*
1093  * walk all backrefs for a given extent to find all roots that reference this
1094  * extent. Walking a backref means finding all extents that reference this
1095  * extent and in turn walk the backrefs of those, too. Naturally this is a
1096  * recursive process, but here it is implemented in an iterative fashion: We
1097  * find all referencing extents for the extent in question and put them on a
1098  * list. In turn, we find all referencing extents for those, further appending
1099  * to the list. The way we iterate the list allows adding more elements after
1100  * the current while iterating. The process stops when we reach the end of the
1101  * list. Found roots are added to the roots list.
1102  *
1103  * returns 0 on success, < 0 on error.
1104  */
1105 static int __btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1106                                   struct btrfs_fs_info *fs_info, u64 bytenr,
1107                                   u64 time_seq, struct ulist **roots)
1108 {
1109         struct ulist *tmp;
1110         struct ulist_node *node = NULL;
1111         struct ulist_iterator uiter;
1112         int ret;
1113
1114         tmp = ulist_alloc(GFP_NOFS);
1115         if (!tmp)
1116                 return -ENOMEM;
1117         *roots = ulist_alloc(GFP_NOFS);
1118         if (!*roots) {
1119                 ulist_free(tmp);
1120                 return -ENOMEM;
1121         }
1122
1123         ULIST_ITER_INIT(&uiter);
1124         while (1) {
1125                 ret = find_parent_nodes(trans, fs_info, bytenr,
1126                                         time_seq, tmp, *roots, NULL);
1127                 if (ret < 0 && ret != -ENOENT) {
1128                         ulist_free(tmp);
1129                         ulist_free(*roots);
1130                         return ret;
1131                 }
1132                 node = ulist_next(tmp, &uiter);
1133                 if (!node)
1134                         break;
1135                 bytenr = node->val;
1136                 cond_resched();
1137         }
1138
1139         ulist_free(tmp);
1140         return 0;
1141 }
1142
1143 int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1144                          struct btrfs_fs_info *fs_info, u64 bytenr,
1145                          u64 time_seq, struct ulist **roots)
1146 {
1147         int ret;
1148
1149         if (!trans)
1150                 down_read(&fs_info->commit_root_sem);
1151         ret = __btrfs_find_all_roots(trans, fs_info, bytenr, time_seq, roots);
1152         if (!trans)
1153                 up_read(&fs_info->commit_root_sem);
1154         return ret;
1155 }
1156
1157 /*
1158  * this makes the path point to (inum INODE_ITEM ioff)
1159  */
1160 int inode_item_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
1161                         struct btrfs_path *path)
1162 {
1163         struct btrfs_key key;
1164         return btrfs_find_item(fs_root, path, inum, ioff,
1165                         BTRFS_INODE_ITEM_KEY, &key);
1166 }
1167
1168 static int inode_ref_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
1169                                 struct btrfs_path *path,
1170                                 struct btrfs_key *found_key)
1171 {
1172         return btrfs_find_item(fs_root, path, inum, ioff,
1173                         BTRFS_INODE_REF_KEY, found_key);
1174 }
1175
1176 int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,
1177                           u64 start_off, struct btrfs_path *path,
1178                           struct btrfs_inode_extref **ret_extref,
1179                           u64 *found_off)
1180 {
1181         int ret, slot;
1182         struct btrfs_key key;
1183         struct btrfs_key found_key;
1184         struct btrfs_inode_extref *extref;
1185         struct extent_buffer *leaf;
1186         unsigned long ptr;
1187
1188         key.objectid = inode_objectid;
1189         btrfs_set_key_type(&key, BTRFS_INODE_EXTREF_KEY);
1190         key.offset = start_off;
1191
1192         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1193         if (ret < 0)
1194                 return ret;
1195
1196         while (1) {
1197                 leaf = path->nodes[0];
1198                 slot = path->slots[0];
1199                 if (slot >= btrfs_header_nritems(leaf)) {
1200                         /*
1201                          * If the item at offset is not found,
1202                          * btrfs_search_slot will point us to the slot
1203                          * where it should be inserted. In our case
1204                          * that will be the slot directly before the
1205                          * next INODE_REF_KEY_V2 item. In the case
1206                          * that we're pointing to the last slot in a
1207                          * leaf, we must move one leaf over.
1208                          */
1209                         ret = btrfs_next_leaf(root, path);
1210                         if (ret) {
1211                                 if (ret >= 1)
1212                                         ret = -ENOENT;
1213                                 break;
1214                         }
1215                         continue;
1216                 }
1217
1218                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1219
1220                 /*
1221                  * Check that we're still looking at an extended ref key for
1222                  * this particular objectid. If we have different
1223                  * objectid or type then there are no more to be found
1224                  * in the tree and we can exit.
1225                  */
1226                 ret = -ENOENT;
1227                 if (found_key.objectid != inode_objectid)
1228                         break;
1229                 if (btrfs_key_type(&found_key) != BTRFS_INODE_EXTREF_KEY)
1230                         break;
1231
1232                 ret = 0;
1233                 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1234                 extref = (struct btrfs_inode_extref *)ptr;
1235                 *ret_extref = extref;
1236                 if (found_off)
1237                         *found_off = found_key.offset;
1238                 break;
1239         }
1240
1241         return ret;
1242 }
1243
1244 /*
1245  * this iterates to turn a name (from iref/extref) into a full filesystem path.
1246  * Elements of the path are separated by '/' and the path is guaranteed to be
1247  * 0-terminated. the path is only given within the current file system.
1248  * Therefore, it never starts with a '/'. the caller is responsible to provide
1249  * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
1250  * the start point of the resulting string is returned. this pointer is within
1251  * dest, normally.
1252  * in case the path buffer would overflow, the pointer is decremented further
1253  * as if output was written to the buffer, though no more output is actually
1254  * generated. that way, the caller can determine how much space would be
1255  * required for the path to fit into the buffer. in that case, the returned
1256  * value will be smaller than dest. callers must check this!
1257  */
1258 char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
1259                         u32 name_len, unsigned long name_off,
1260                         struct extent_buffer *eb_in, u64 parent,
1261                         char *dest, u32 size)
1262 {
1263         int slot;
1264         u64 next_inum;
1265         int ret;
1266         s64 bytes_left = ((s64)size) - 1;
1267         struct extent_buffer *eb = eb_in;
1268         struct btrfs_key found_key;
1269         int leave_spinning = path->leave_spinning;
1270         struct btrfs_inode_ref *iref;
1271
1272         if (bytes_left >= 0)
1273                 dest[bytes_left] = '\0';
1274
1275         path->leave_spinning = 1;
1276         while (1) {
1277                 bytes_left -= name_len;
1278                 if (bytes_left >= 0)
1279                         read_extent_buffer(eb, dest + bytes_left,
1280                                            name_off, name_len);
1281                 if (eb != eb_in) {
1282                         btrfs_tree_read_unlock_blocking(eb);
1283                         free_extent_buffer(eb);
1284                 }
1285                 ret = inode_ref_info(parent, 0, fs_root, path, &found_key);
1286                 if (ret > 0)
1287                         ret = -ENOENT;
1288                 if (ret)
1289                         break;
1290
1291                 next_inum = found_key.offset;
1292
1293                 /* regular exit ahead */
1294                 if (parent == next_inum)
1295                         break;
1296
1297                 slot = path->slots[0];
1298                 eb = path->nodes[0];
1299                 /* make sure we can use eb after releasing the path */
1300                 if (eb != eb_in) {
1301                         atomic_inc(&eb->refs);
1302                         btrfs_tree_read_lock(eb);
1303                         btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1304                 }
1305                 btrfs_release_path(path);
1306                 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1307
1308                 name_len = btrfs_inode_ref_name_len(eb, iref);
1309                 name_off = (unsigned long)(iref + 1);
1310
1311                 parent = next_inum;
1312                 --bytes_left;
1313                 if (bytes_left >= 0)
1314                         dest[bytes_left] = '/';
1315         }
1316
1317         btrfs_release_path(path);
1318         path->leave_spinning = leave_spinning;
1319
1320         if (ret)
1321                 return ERR_PTR(ret);
1322
1323         return dest + bytes_left;
1324 }
1325
1326 /*
1327  * this makes the path point to (logical EXTENT_ITEM *)
1328  * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
1329  * tree blocks and <0 on error.
1330  */
1331 int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
1332                         struct btrfs_path *path, struct btrfs_key *found_key,
1333                         u64 *flags_ret)
1334 {
1335         int ret;
1336         u64 flags;
1337         u64 size = 0;
1338         u32 item_size;
1339         struct extent_buffer *eb;
1340         struct btrfs_extent_item *ei;
1341         struct btrfs_key key;
1342
1343         if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1344                 key.type = BTRFS_METADATA_ITEM_KEY;
1345         else
1346                 key.type = BTRFS_EXTENT_ITEM_KEY;
1347         key.objectid = logical;
1348         key.offset = (u64)-1;
1349
1350         ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
1351         if (ret < 0)
1352                 return ret;
1353
1354         ret = btrfs_previous_extent_item(fs_info->extent_root, path, 0);
1355         if (ret) {
1356                 if (ret > 0)
1357                         ret = -ENOENT;
1358                 return ret;
1359         }
1360         btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]);
1361         if (found_key->type == BTRFS_METADATA_ITEM_KEY)
1362                 size = fs_info->extent_root->leafsize;
1363         else if (found_key->type == BTRFS_EXTENT_ITEM_KEY)
1364                 size = found_key->offset;
1365
1366         if (found_key->objectid > logical ||
1367             found_key->objectid + size <= logical) {
1368                 pr_debug("logical %llu is not within any extent\n", logical);
1369                 return -ENOENT;
1370         }
1371
1372         eb = path->nodes[0];
1373         item_size = btrfs_item_size_nr(eb, path->slots[0]);
1374         BUG_ON(item_size < sizeof(*ei));
1375
1376         ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
1377         flags = btrfs_extent_flags(eb, ei);
1378
1379         pr_debug("logical %llu is at position %llu within the extent (%llu "
1380                  "EXTENT_ITEM %llu) flags %#llx size %u\n",
1381                  logical, logical - found_key->objectid, found_key->objectid,
1382                  found_key->offset, flags, item_size);
1383
1384         WARN_ON(!flags_ret);
1385         if (flags_ret) {
1386                 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1387                         *flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK;
1388                 else if (flags & BTRFS_EXTENT_FLAG_DATA)
1389                         *flags_ret = BTRFS_EXTENT_FLAG_DATA;
1390                 else
1391                         BUG_ON(1);
1392                 return 0;
1393         }
1394
1395         return -EIO;
1396 }
1397
1398 /*
1399  * helper function to iterate extent inline refs. ptr must point to a 0 value
1400  * for the first call and may be modified. it is used to track state.
1401  * if more refs exist, 0 is returned and the next call to
1402  * __get_extent_inline_ref must pass the modified ptr parameter to get the
1403  * next ref. after the last ref was processed, 1 is returned.
1404  * returns <0 on error
1405  */
1406 static int __get_extent_inline_ref(unsigned long *ptr, struct extent_buffer *eb,
1407                                 struct btrfs_extent_item *ei, u32 item_size,
1408                                 struct btrfs_extent_inline_ref **out_eiref,
1409                                 int *out_type)
1410 {
1411         unsigned long end;
1412         u64 flags;
1413         struct btrfs_tree_block_info *info;
1414
1415         if (!*ptr) {
1416                 /* first call */
1417                 flags = btrfs_extent_flags(eb, ei);
1418                 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1419                         info = (struct btrfs_tree_block_info *)(ei + 1);
1420                         *out_eiref =
1421                                 (struct btrfs_extent_inline_ref *)(info + 1);
1422                 } else {
1423                         *out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1);
1424                 }
1425                 *ptr = (unsigned long)*out_eiref;
1426                 if ((void *)*ptr >= (void *)ei + item_size)
1427                         return -ENOENT;
1428         }
1429
1430         end = (unsigned long)ei + item_size;
1431         *out_eiref = (struct btrfs_extent_inline_ref *)*ptr;
1432         *out_type = btrfs_extent_inline_ref_type(eb, *out_eiref);
1433
1434         *ptr += btrfs_extent_inline_ref_size(*out_type);
1435         WARN_ON(*ptr > end);
1436         if (*ptr == end)
1437                 return 1; /* last */
1438
1439         return 0;
1440 }
1441
1442 /*
1443  * reads the tree block backref for an extent. tree level and root are returned
1444  * through out_level and out_root. ptr must point to a 0 value for the first
1445  * call and may be modified (see __get_extent_inline_ref comment).
1446  * returns 0 if data was provided, 1 if there was no more data to provide or
1447  * <0 on error.
1448  */
1449 int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
1450                                 struct btrfs_extent_item *ei, u32 item_size,
1451                                 u64 *out_root, u8 *out_level)
1452 {
1453         int ret;
1454         int type;
1455         struct btrfs_tree_block_info *info;
1456         struct btrfs_extent_inline_ref *eiref;
1457
1458         if (*ptr == (unsigned long)-1)
1459                 return 1;
1460
1461         while (1) {
1462                 ret = __get_extent_inline_ref(ptr, eb, ei, item_size,
1463                                                 &eiref, &type);
1464                 if (ret < 0)
1465                         return ret;
1466
1467                 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1468                     type == BTRFS_SHARED_BLOCK_REF_KEY)
1469                         break;
1470
1471                 if (ret == 1)
1472                         return 1;
1473         }
1474
1475         /* we can treat both ref types equally here */
1476         info = (struct btrfs_tree_block_info *)(ei + 1);
1477         *out_root = btrfs_extent_inline_ref_offset(eb, eiref);
1478         *out_level = btrfs_tree_block_level(eb, info);
1479
1480         if (ret == 1)
1481                 *ptr = (unsigned long)-1;
1482
1483         return 0;
1484 }
1485
1486 static int iterate_leaf_refs(struct extent_inode_elem *inode_list,
1487                                 u64 root, u64 extent_item_objectid,
1488                                 iterate_extent_inodes_t *iterate, void *ctx)
1489 {
1490         struct extent_inode_elem *eie;
1491         int ret = 0;
1492
1493         for (eie = inode_list; eie; eie = eie->next) {
1494                 pr_debug("ref for %llu resolved, key (%llu EXTEND_DATA %llu), "
1495                          "root %llu\n", extent_item_objectid,
1496                          eie->inum, eie->offset, root);
1497                 ret = iterate(eie->inum, eie->offset, root, ctx);
1498                 if (ret) {
1499                         pr_debug("stopping iteration for %llu due to ret=%d\n",
1500                                  extent_item_objectid, ret);
1501                         break;
1502                 }
1503         }
1504
1505         return ret;
1506 }
1507
1508 /*
1509  * calls iterate() for every inode that references the extent identified by
1510  * the given parameters.
1511  * when the iterator function returns a non-zero value, iteration stops.
1512  */
1513 int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
1514                                 u64 extent_item_objectid, u64 extent_item_pos,
1515                                 int search_commit_root,
1516                                 iterate_extent_inodes_t *iterate, void *ctx)
1517 {
1518         int ret;
1519         struct btrfs_trans_handle *trans = NULL;
1520         struct ulist *refs = NULL;
1521         struct ulist *roots = NULL;
1522         struct ulist_node *ref_node = NULL;
1523         struct ulist_node *root_node = NULL;
1524         struct seq_list tree_mod_seq_elem = {};
1525         struct ulist_iterator ref_uiter;
1526         struct ulist_iterator root_uiter;
1527
1528         pr_debug("resolving all inodes for extent %llu\n",
1529                         extent_item_objectid);
1530
1531         if (!search_commit_root) {
1532                 trans = btrfs_join_transaction(fs_info->extent_root);
1533                 if (IS_ERR(trans))
1534                         return PTR_ERR(trans);
1535                 btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1536         } else {
1537                 down_read(&fs_info->commit_root_sem);
1538         }
1539
1540         ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid,
1541                                    tree_mod_seq_elem.seq, &refs,
1542                                    &extent_item_pos);
1543         if (ret)
1544                 goto out;
1545
1546         ULIST_ITER_INIT(&ref_uiter);
1547         while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) {
1548                 ret = __btrfs_find_all_roots(trans, fs_info, ref_node->val,
1549                                              tree_mod_seq_elem.seq, &roots);
1550                 if (ret)
1551                         break;
1552                 ULIST_ITER_INIT(&root_uiter);
1553                 while (!ret && (root_node = ulist_next(roots, &root_uiter))) {
1554                         pr_debug("root %llu references leaf %llu, data list "
1555                                  "%#llx\n", root_node->val, ref_node->val,
1556                                  ref_node->aux);
1557                         ret = iterate_leaf_refs((struct extent_inode_elem *)
1558                                                 (uintptr_t)ref_node->aux,
1559                                                 root_node->val,
1560                                                 extent_item_objectid,
1561                                                 iterate, ctx);
1562                 }
1563                 ulist_free(roots);
1564         }
1565
1566         free_leaf_list(refs);
1567 out:
1568         if (!search_commit_root) {
1569                 btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1570                 btrfs_end_transaction(trans, fs_info->extent_root);
1571         } else {
1572                 up_read(&fs_info->commit_root_sem);
1573         }
1574
1575         return ret;
1576 }
1577
1578 int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
1579                                 struct btrfs_path *path,
1580                                 iterate_extent_inodes_t *iterate, void *ctx)
1581 {
1582         int ret;
1583         u64 extent_item_pos;
1584         u64 flags = 0;
1585         struct btrfs_key found_key;
1586         int search_commit_root = path->search_commit_root;
1587
1588         ret = extent_from_logical(fs_info, logical, path, &found_key, &flags);
1589         btrfs_release_path(path);
1590         if (ret < 0)
1591                 return ret;
1592         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1593                 return -EINVAL;
1594
1595         extent_item_pos = logical - found_key.objectid;
1596         ret = iterate_extent_inodes(fs_info, found_key.objectid,
1597                                         extent_item_pos, search_commit_root,
1598                                         iterate, ctx);
1599
1600         return ret;
1601 }
1602
1603 typedef int (iterate_irefs_t)(u64 parent, u32 name_len, unsigned long name_off,
1604                               struct extent_buffer *eb, void *ctx);
1605
1606 static int iterate_inode_refs(u64 inum, struct btrfs_root *fs_root,
1607                               struct btrfs_path *path,
1608                               iterate_irefs_t *iterate, void *ctx)
1609 {
1610         int ret = 0;
1611         int slot;
1612         u32 cur;
1613         u32 len;
1614         u32 name_len;
1615         u64 parent = 0;
1616         int found = 0;
1617         struct extent_buffer *eb;
1618         struct btrfs_item *item;
1619         struct btrfs_inode_ref *iref;
1620         struct btrfs_key found_key;
1621
1622         while (!ret) {
1623                 ret = inode_ref_info(inum, parent ? parent+1 : 0, fs_root, path,
1624                                      &found_key);
1625                 if (ret < 0)
1626                         break;
1627                 if (ret) {
1628                         ret = found ? 0 : -ENOENT;
1629                         break;
1630                 }
1631                 ++found;
1632
1633                 parent = found_key.offset;
1634                 slot = path->slots[0];
1635                 eb = btrfs_clone_extent_buffer(path->nodes[0]);
1636                 if (!eb) {
1637                         ret = -ENOMEM;
1638                         break;
1639                 }
1640                 extent_buffer_get(eb);
1641                 btrfs_tree_read_lock(eb);
1642                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1643                 btrfs_release_path(path);
1644
1645                 item = btrfs_item_nr(slot);
1646                 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1647
1648                 for (cur = 0; cur < btrfs_item_size(eb, item); cur += len) {
1649                         name_len = btrfs_inode_ref_name_len(eb, iref);
1650                         /* path must be released before calling iterate()! */
1651                         pr_debug("following ref at offset %u for inode %llu in "
1652                                  "tree %llu\n", cur, found_key.objectid,
1653                                  fs_root->objectid);
1654                         ret = iterate(parent, name_len,
1655                                       (unsigned long)(iref + 1), eb, ctx);
1656                         if (ret)
1657                                 break;
1658                         len = sizeof(*iref) + name_len;
1659                         iref = (struct btrfs_inode_ref *)((char *)iref + len);
1660                 }
1661                 btrfs_tree_read_unlock_blocking(eb);
1662                 free_extent_buffer(eb);
1663         }
1664
1665         btrfs_release_path(path);
1666
1667         return ret;
1668 }
1669
1670 static int iterate_inode_extrefs(u64 inum, struct btrfs_root *fs_root,
1671                                  struct btrfs_path *path,
1672                                  iterate_irefs_t *iterate, void *ctx)
1673 {
1674         int ret;
1675         int slot;
1676         u64 offset = 0;
1677         u64 parent;
1678         int found = 0;
1679         struct extent_buffer *eb;
1680         struct btrfs_inode_extref *extref;
1681         struct extent_buffer *leaf;
1682         u32 item_size;
1683         u32 cur_offset;
1684         unsigned long ptr;
1685
1686         while (1) {
1687                 ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref,
1688                                             &offset);
1689                 if (ret < 0)
1690                         break;
1691                 if (ret) {
1692                         ret = found ? 0 : -ENOENT;
1693                         break;
1694                 }
1695                 ++found;
1696
1697                 slot = path->slots[0];
1698                 eb = btrfs_clone_extent_buffer(path->nodes[0]);
1699                 if (!eb) {
1700                         ret = -ENOMEM;
1701                         break;
1702                 }
1703                 extent_buffer_get(eb);
1704
1705                 btrfs_tree_read_lock(eb);
1706                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1707                 btrfs_release_path(path);
1708
1709                 leaf = path->nodes[0];
1710                 item_size = btrfs_item_size_nr(leaf, slot);
1711                 ptr = btrfs_item_ptr_offset(leaf, slot);
1712                 cur_offset = 0;
1713
1714                 while (cur_offset < item_size) {
1715                         u32 name_len;
1716
1717                         extref = (struct btrfs_inode_extref *)(ptr + cur_offset);
1718                         parent = btrfs_inode_extref_parent(eb, extref);
1719                         name_len = btrfs_inode_extref_name_len(eb, extref);
1720                         ret = iterate(parent, name_len,
1721                                       (unsigned long)&extref->name, eb, ctx);
1722                         if (ret)
1723                                 break;
1724
1725                         cur_offset += btrfs_inode_extref_name_len(leaf, extref);
1726                         cur_offset += sizeof(*extref);
1727                 }
1728                 btrfs_tree_read_unlock_blocking(eb);
1729                 free_extent_buffer(eb);
1730
1731                 offset++;
1732         }
1733
1734         btrfs_release_path(path);
1735
1736         return ret;
1737 }
1738
1739 static int iterate_irefs(u64 inum, struct btrfs_root *fs_root,
1740                          struct btrfs_path *path, iterate_irefs_t *iterate,
1741                          void *ctx)
1742 {
1743         int ret;
1744         int found_refs = 0;
1745
1746         ret = iterate_inode_refs(inum, fs_root, path, iterate, ctx);
1747         if (!ret)
1748                 ++found_refs;
1749         else if (ret != -ENOENT)
1750                 return ret;
1751
1752         ret = iterate_inode_extrefs(inum, fs_root, path, iterate, ctx);
1753         if (ret == -ENOENT && found_refs)
1754                 return 0;
1755
1756         return ret;
1757 }
1758
1759 /*
1760  * returns 0 if the path could be dumped (probably truncated)
1761  * returns <0 in case of an error
1762  */
1763 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
1764                          struct extent_buffer *eb, void *ctx)
1765 {
1766         struct inode_fs_paths *ipath = ctx;
1767         char *fspath;
1768         char *fspath_min;
1769         int i = ipath->fspath->elem_cnt;
1770         const int s_ptr = sizeof(char *);
1771         u32 bytes_left;
1772
1773         bytes_left = ipath->fspath->bytes_left > s_ptr ?
1774                                         ipath->fspath->bytes_left - s_ptr : 0;
1775
1776         fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
1777         fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len,
1778                                    name_off, eb, inum, fspath_min, bytes_left);
1779         if (IS_ERR(fspath))
1780                 return PTR_ERR(fspath);
1781
1782         if (fspath > fspath_min) {
1783                 ipath->fspath->val[i] = (u64)(unsigned long)fspath;
1784                 ++ipath->fspath->elem_cnt;
1785                 ipath->fspath->bytes_left = fspath - fspath_min;
1786         } else {
1787                 ++ipath->fspath->elem_missed;
1788                 ipath->fspath->bytes_missing += fspath_min - fspath;
1789                 ipath->fspath->bytes_left = 0;
1790         }
1791
1792         return 0;
1793 }
1794
1795 /*
1796  * this dumps all file system paths to the inode into the ipath struct, provided
1797  * is has been created large enough. each path is zero-terminated and accessed
1798  * from ipath->fspath->val[i].
1799  * when it returns, there are ipath->fspath->elem_cnt number of paths available
1800  * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the
1801  * number of missed paths in recored in ipath->fspath->elem_missed, otherwise,
1802  * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
1803  * have been needed to return all paths.
1804  */
1805 int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
1806 {
1807         return iterate_irefs(inum, ipath->fs_root, ipath->btrfs_path,
1808                              inode_to_path, ipath);
1809 }
1810
1811 struct btrfs_data_container *init_data_container(u32 total_bytes)
1812 {
1813         struct btrfs_data_container *data;
1814         size_t alloc_bytes;
1815
1816         alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
1817         data = vmalloc(alloc_bytes);
1818         if (!data)
1819                 return ERR_PTR(-ENOMEM);
1820
1821         if (total_bytes >= sizeof(*data)) {
1822                 data->bytes_left = total_bytes - sizeof(*data);
1823                 data->bytes_missing = 0;
1824         } else {
1825                 data->bytes_missing = sizeof(*data) - total_bytes;
1826                 data->bytes_left = 0;
1827         }
1828
1829         data->elem_cnt = 0;
1830         data->elem_missed = 0;
1831
1832         return data;
1833 }
1834
1835 /*
1836  * allocates space to return multiple file system paths for an inode.
1837  * total_bytes to allocate are passed, note that space usable for actual path
1838  * information will be total_bytes - sizeof(struct inode_fs_paths).
1839  * the returned pointer must be freed with free_ipath() in the end.
1840  */
1841 struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
1842                                         struct btrfs_path *path)
1843 {
1844         struct inode_fs_paths *ifp;
1845         struct btrfs_data_container *fspath;
1846
1847         fspath = init_data_container(total_bytes);
1848         if (IS_ERR(fspath))
1849                 return (void *)fspath;
1850
1851         ifp = kmalloc(sizeof(*ifp), GFP_NOFS);
1852         if (!ifp) {
1853                 kfree(fspath);
1854                 return ERR_PTR(-ENOMEM);
1855         }
1856
1857         ifp->btrfs_path = path;
1858         ifp->fspath = fspath;
1859         ifp->fs_root = fs_root;
1860
1861         return ifp;
1862 }
1863
1864 void free_ipath(struct inode_fs_paths *ipath)
1865 {
1866         if (!ipath)
1867                 return;
1868         vfree(ipath->fspath);
1869         kfree(ipath);
1870 }