fs/btrfs/backref.c

   1 /*
   2  * Copyright (C) 2011 STRATO.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/vmalloc.h>
  20 #include "ctree.h"
  21 #include "disk-io.h"
  22 #include "backref.h"
  23 #include "ulist.h"
  24 #include "transaction.h"
  25 #include "delayed-ref.h"
  26 #include "locking.h"
  27
  28 struct extent_inode_elem {
  29         u64 inum;
  30         u64 offset;
  31         struct extent_inode_elem *next;
  32 };
  33
  34 static int check_extent_in_eb(struct btrfs_key *key, struct extent_buffer *eb,
  35                                 struct btrfs_file_extent_item *fi,
  36                                 u64 extent_item_pos,
  37                                 struct extent_inode_elem **eie)
  38 {
  39         u64 offset = 0;
  40         struct extent_inode_elem *e;
  41
  42         if (!btrfs_file_extent_compression(eb, fi) &&
  43             !btrfs_file_extent_encryption(eb, fi) &&
  44             !btrfs_file_extent_other_encoding(eb, fi)) {
  45                 u64 data_offset;
  46                 u64 data_len;
  47
  48                 data_offset = btrfs_file_extent_offset(eb, fi);
  49                 data_len = btrfs_file_extent_num_bytes(eb, fi);
  50
  51                 if (extent_item_pos < data_offset ||
  52                     extent_item_pos >= data_offset + data_len)
  53                         return 1;
  54                 offset = extent_item_pos - data_offset;
  55         }
  56
  57         e = kmalloc(sizeof(*e), GFP_NOFS);
  58         if (!e)
  59                 return -ENOMEM;
  60
  61         e->next = *eie;
  62         e->inum = key->objectid;
  63         e->offset = key->offset + offset;
  64         *eie = e;
  65
  66         return 0;
  67 }
  68
  69 static int find_extent_in_eb(struct extent_buffer *eb, u64 wanted_disk_byte,
  70                                 u64 extent_item_pos,
  71                                 struct extent_inode_elem **eie)
  72 {
  73         u64 disk_byte;
  74         struct btrfs_key key;
  75         struct btrfs_file_extent_item *fi;
  76         int slot;
  77         int nritems;
  78         int extent_type;
  79         int ret;
  80
  81         /*
  82          * from the shared data ref, we only have the leaf but we need
  83          * the key. thus, we must look into all items and see that we
  84          * find one (some) with a reference to our extent item.
  85          */
  86         nritems = btrfs_header_nritems(eb);
  87         for (slot = 0; slot < nritems; ++slot) {
  88                 btrfs_item_key_to_cpu(eb, &key, slot);
  89                 if (key.type != BTRFS_EXTENT_DATA_KEY)
  90                         continue;
  91                 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
  92                 extent_type = btrfs_file_extent_type(eb, fi);
  93                 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
  94                         continue;
  95                 /* don't skip BTRFS_FILE_EXTENT_PREALLOC, we can handle that */
  96                 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
  97                 if (disk_byte != wanted_disk_byte)
  98                         continue;
  99
 100                 ret = check_extent_in_eb(&key, eb, fi, extent_item_pos, eie);
 101                 if (ret < 0)
 102                         return ret;
 103         }
 104
 105         return 0;
 106 }
 107
 108 /*
 109  * this structure records all encountered refs on the way up to the root
 110  */
 111 struct __prelim_ref {
 112         struct list_head list;
 113         u64 root_id;
 114         struct btrfs_key key_for_search;
 115         int level;
 116         int count;
 117         struct extent_inode_elem *inode_list;
 118         u64 parent;
 119         u64 wanted_disk_byte;
 120 };
 121
 122 static struct kmem_cache *btrfs_prelim_ref_cache;
 123
 124 int __init btrfs_prelim_ref_init(void)
 125 {
 126         btrfs_prelim_ref_cache = kmem_cache_create("btrfs_prelim_ref",
 127                                         sizeof(struct __prelim_ref),
 128                                         0,
 129                                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
 130                                         NULL);
 131         if (!btrfs_prelim_ref_cache)
 132                 return -ENOMEM;
 133         return 0;
 134 }
 135
 136 void btrfs_prelim_ref_exit(void)
 137 {
 138         if (btrfs_prelim_ref_cache)
 139                 kmem_cache_destroy(btrfs_prelim_ref_cache);
 140 }
 141
 142 /*
 143  * the rules for all callers of this function are:
 144  * - obtaining the parent is the goal
 145  * - if you add a key, you must know that it is a correct key
 146  * - if you cannot add the parent or a correct key, then we will look into the
 147  *   block later to set a correct key
 148  *
 149  * delayed refs
 150  * ============
 151  *        backref type | shared | indirect | shared | indirect
 152  * information         |   tree |     tree |   data |     data
 153  * --------------------+--------+----------+--------+----------
 154  *      parent logical |    y   |     -    |    -   |     -
 155  *      key to resolve |    -   |     y    |    y   |     y
 156  *  tree block logical |    -   |     -    |    -   |     -
 157  *  root for resolving |    y   |     y    |    y   |     y
 158  *
 159  * - column 1:       we've the parent -> done
 160  * - column 2, 3, 4: we use the key to find the parent
 161  *
 162  * on disk refs (inline or keyed)
 163  * ==============================
 164  *        backref type | shared | indirect | shared | indirect
 165  * information         |   tree |     tree |   data |     data
 166  * --------------------+--------+----------+--------+----------
 167  *      parent logical |    y   |     -    |    y   |     -
 168  *      key to resolve |    -   |     -    |    -   |     y
 169  *  tree block logical |    y   |     y    |    y   |     y
 170  *  root for resolving |    -   |     y    |    y   |     y
 171  *
 172  * - column 1, 3: we've the parent -> done
 173  * - column 2:    we take the first key from the block to find the parent
 174  *                (see __add_missing_keys)
 175  * - column 4:    we use the key to find the parent
 176  *
 177  * additional information that's available but not required to find the parent
 178  * block might help in merging entries to gain some speed.
 179  */
 180
 181 static int __add_prelim_ref(struct list_head *head, u64 root_id,
 182                             struct btrfs_key *key, int level,
 183                             u64 parent, u64 wanted_disk_byte, int count,
 184                             gfp_t gfp_mask)
 185 {
 186         struct __prelim_ref *ref;
 187
 188         if (root_id == BTRFS_DATA_RELOC_TREE_OBJECTID)
 189                 return 0;
 190
 191         ref = kmem_cache_alloc(btrfs_prelim_ref_cache, gfp_mask);
 192         if (!ref)
 193                 return -ENOMEM;
 194
 195         ref->root_id = root_id;
 196         if (key)
 197                 ref->key_for_search = *key;
 198         else
 199                 memset(&ref->key_for_search, 0, sizeof(ref->key_for_search));
 200
 201         ref->inode_list = NULL;
 202         ref->level = level;
 203         ref->count = count;
 204         ref->parent = parent;
 205         ref->wanted_disk_byte = wanted_disk_byte;
 206         list_add_tail(&ref->list, head);
 207
 208         return 0;
 209 }
 210
 211 static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path,
 212                            struct ulist *parents, struct __prelim_ref *ref,
 213                            int level, u64 time_seq, const u64 *extent_item_pos)
 214 {
 215         int ret = 0;
 216         int slot;
 217         struct extent_buffer *eb;
 218         struct btrfs_key key;
 219         struct btrfs_key *key_for_search = &ref->key_for_search;
 220         struct btrfs_file_extent_item *fi;
 221         struct extent_inode_elem *eie = NULL, *old = NULL;
 222         u64 disk_byte;
 223         u64 wanted_disk_byte = ref->wanted_disk_byte;
 224         u64 count = 0;
 225
 226         if (level != 0) {
 227                 eb = path->nodes[level];
 228                 ret = ulist_add(parents, eb->start, 0, GFP_NOFS);
 229                 if (ret < 0)
 230                         return ret;
 231                 return 0;
 232         }
 233
 234         /*
 235          * We normally enter this function with the path already pointing to
 236          * the first item to check. But sometimes, we may enter it with
 237          * slot==nritems. In that case, go to the next leaf before we continue.
 238          */
 239         if (path->slots[0] >= btrfs_header_nritems(path->nodes[0]))
 240                 ret = btrfs_next_old_leaf(root, path, time_seq);
 241
 242         while (!ret && count < ref->count) {
 243                 eb = path->nodes[0];
 244                 slot = path->slots[0];
 245
 246                 btrfs_item_key_to_cpu(eb, &key, slot);
 247
 248                 if (key.objectid != key_for_search->objectid ||
 249                     key.type != BTRFS_EXTENT_DATA_KEY)
 250                         break;
 251
 252                 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
 253                 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
 254
 255                 if (disk_byte == wanted_disk_byte) {
 256                         eie = NULL;
 257                         old = NULL;
 258                         count++;
 259                         if (extent_item_pos) {
 260                                 ret = check_extent_in_eb(&key, eb, fi,
 261                                                 *extent_item_pos,
 262                                                 &eie);
 263                                 if (ret < 0)
 264                                         break;
 265                         }
 266                         if (ret > 0)
 267                                 goto next;
 268                         ret = ulist_add_merge(parents, eb->start,
 269                                               (uintptr_t)eie,
 270                                               (u64 *)&old, GFP_NOFS);
 271                         if (ret < 0)
 272                                 break;
 273                         if (!ret && extent_item_pos) {
 274                                 while (old->next)
 275                                         old = old->next;
 276                                 old->next = eie;
 277                         }
 278                 }
 279 next:
 280                 ret = btrfs_next_old_item(root, path, time_seq);
 281         }
 282
 283         if (ret > 0)
 284                 ret = 0;
 285         return ret;
 286 }
 287
 288 /*
 289  * resolve an indirect backref in the form (root_id, key, level)
 290  * to a logical address
 291  */
 292 static int __resolve_indirect_ref(struct btrfs_fs_info *fs_info,
 293                                   struct btrfs_path *path, u64 time_seq,
 294                                   struct __prelim_ref *ref,
 295                                   struct ulist *parents,
 296                                   const u64 *extent_item_pos)
 297 {
 298         struct btrfs_root *root;
 299         struct btrfs_key root_key;
 300         struct extent_buffer *eb;
 301         int ret = 0;
 302         int root_level;
 303         int level = ref->level;
 304         int index;
 305
 306         root_key.objectid = ref->root_id;
 307         root_key.type = BTRFS_ROOT_ITEM_KEY;
 308         root_key.offset = (u64)-1;
 309
 310         index = srcu_read_lock(&fs_info->subvol_srcu);
 311
 312         root = btrfs_read_fs_root_no_name(fs_info, &root_key);
 313         if (IS_ERR(root)) {
 314                 srcu_read_unlock(&fs_info->subvol_srcu, index);
 315                 ret = PTR_ERR(root);
 316                 goto out;
 317         }
 318
 319         root_level = btrfs_old_root_level(root, time_seq);
 320
 321         if (root_level + 1 == level) {
 322                 srcu_read_unlock(&fs_info->subvol_srcu, index);
 323                 goto out;
 324         }
 325
 326         path->lowest_level = level;
 327         ret = btrfs_search_old_slot(root, &ref->key_for_search, path, time_seq);
 328
 329         /* root node has been locked, we can release @subvol_srcu safely here */
 330         srcu_read_unlock(&fs_info->subvol_srcu, index);
 331
 332         pr_debug("search slot in root %llu (level %d, ref count %d) returned "
 333                  "%d for key (%llu %u %llu)\n",
 334                  ref->root_id, level, ref->count, ret,
 335                  ref->key_for_search.objectid, ref->key_for_search.type,
 336                  ref->key_for_search.offset);
 337         if (ret < 0)
 338                 goto out;
 339
 340         eb = path->nodes[level];
 341         while (!eb) {
 342                 if (WARN_ON(!level)) {
 343                         ret = 1;
 344                         goto out;
 345                 }
 346                 level--;
 347                 eb = path->nodes[level];
 348         }
 349
 350         ret = add_all_parents(root, path, parents, ref, level, time_seq,
 351                               extent_item_pos);
 352 out:
 353         path->lowest_level = 0;
 354         btrfs_release_path(path);
 355         return ret;
 356 }
 357
 358 /*
 359  * resolve all indirect backrefs from the list
 360  */
 361 static int __resolve_indirect_refs(struct btrfs_fs_info *fs_info,
 362                                    struct btrfs_path *path, u64 time_seq,
 363                                    struct list_head *head,
 364                                    const u64 *extent_item_pos)
 365 {
 366         int err;
 367         int ret = 0;
 368         struct __prelim_ref *ref;
 369         struct __prelim_ref *ref_safe;
 370         struct __prelim_ref *new_ref;
 371         struct ulist *parents;
 372         struct ulist_node *node;
 373         struct ulist_iterator uiter;
 374
 375         parents = ulist_alloc(GFP_NOFS);
 376         if (!parents)
 377                 return -ENOMEM;
 378
 379         /*
 380          * _safe allows us to insert directly after the current item without
 381          * iterating over the newly inserted items.
 382          * we're also allowed to re-assign ref during iteration.
 383          */
 384         list_for_each_entry_safe(ref, ref_safe, head, list) {
 385                 if (ref->parent)        /* already direct */
 386                         continue;
 387                 if (ref->count == 0)
 388                         continue;
 389                 err = __resolve_indirect_ref(fs_info, path, time_seq, ref,
 390                                              parents, extent_item_pos);
 391                 /*
 392                  * we can only tolerate ENOENT,otherwise,we should catch error
 393                  * and return directly.
 394                  */
 395                 if (err == -ENOENT) {
 396                         continue;
 397                 } else if (err) {
 398                         ret = err;
 399                         goto out;
 400                 }
 401
 402                 /* we put the first parent into the ref at hand */
 403                 ULIST_ITER_INIT(&uiter);
 404                 node = ulist_next(parents, &uiter);
 405                 ref->parent = node ? node->val : 0;
 406                 ref->inode_list = node ?
 407                         (struct extent_inode_elem *)(uintptr_t)node->aux : NULL;
 408
 409                 /* additional parents require new refs being added here */
 410                 while ((node = ulist_next(parents, &uiter))) {
 411                         new_ref = kmem_cache_alloc(btrfs_prelim_ref_cache,
 412                                                    GFP_NOFS);
 413                         if (!new_ref) {
 414                                 ret = -ENOMEM;
 415                                 goto out;
 416                         }
 417                         memcpy(new_ref, ref, sizeof(*ref));
 418                         new_ref->parent = node->val;
 419                         new_ref->inode_list = (struct extent_inode_elem *)
 420                                                         (uintptr_t)node->aux;
 421                         list_add(&new_ref->list, &ref->list);
 422                 }
 423                 ulist_reinit(parents);
 424         }
 425 out:
 426         ulist_free(parents);
 427         return ret;
 428 }
 429
 430 static inline int ref_for_same_block(struct __prelim_ref *ref1,
 431                                      struct __prelim_ref *ref2)
 432 {
 433         if (ref1->level != ref2->level)
 434                 return 0;
 435         if (ref1->root_id != ref2->root_id)
 436                 return 0;
 437         if (ref1->key_for_search.type != ref2->key_for_search.type)
 438                 return 0;
 439         if (ref1->key_for_search.objectid != ref2->key_for_search.objectid)
 440                 return 0;
 441         if (ref1->key_for_search.offset != ref2->key_for_search.offset)
 442                 return 0;
 443         if (ref1->parent != ref2->parent)
 444                 return 0;
 445
 446         return 1;
 447 }
 448
 449 /*
 450  * read tree blocks and add keys where required.
 451  */
 452 static int __add_missing_keys(struct btrfs_fs_info *fs_info,
 453                               struct list_head *head)
 454 {
 455         struct list_head *pos;
 456         struct extent_buffer *eb;
 457
 458         list_for_each(pos, head) {
 459                 struct __prelim_ref *ref;
 460                 ref = list_entry(pos, struct __prelim_ref, list);
 461
 462                 if (ref->parent)
 463                         continue;
 464                 if (ref->key_for_search.type)
 465                         continue;
 466                 BUG_ON(!ref->wanted_disk_byte);
 467                 eb = read_tree_block(fs_info->tree_root, ref->wanted_disk_byte,
 468                                      fs_info->tree_root->leafsize, 0);
 469                 if (!eb || !extent_buffer_uptodate(eb)) {
 470                         free_extent_buffer(eb);
 471                         return -EIO;
 472                 }
 473                 btrfs_tree_read_lock(eb);
 474                 if (btrfs_header_level(eb) == 0)
 475                         btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
 476                 else
 477                         btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
 478                 btrfs_tree_read_unlock(eb);
 479                 free_extent_buffer(eb);
 480         }
 481         return 0;
 482 }
 483
 484 /*
 485  * merge two lists of backrefs and adjust counts accordingly
 486  *
 487  * mode = 1: merge identical keys, if key is set
 488  *    FIXME: if we add more keys in __add_prelim_ref, we can merge more here.
 489  *           additionally, we could even add a key range for the blocks we
 490  *           looked into to merge even more (-> replace unresolved refs by those
 491  *           having a parent).
 492  * mode = 2: merge identical parents
 493  */
 494 static void __merge_refs(struct list_head *head, int mode)
 495 {
 496         struct list_head *pos1;
 497
 498         list_for_each(pos1, head) {
 499                 struct list_head *n2;
 500                 struct list_head *pos2;
 501                 struct __prelim_ref *ref1;
 502
 503                 ref1 = list_entry(pos1, struct __prelim_ref, list);
 504
 505                 for (pos2 = pos1->next, n2 = pos2->next; pos2 != head;
 506                      pos2 = n2, n2 = pos2->next) {
 507                         struct __prelim_ref *ref2;
 508                         struct __prelim_ref *xchg;
 509                         struct extent_inode_elem *eie;
 510
 511                         ref2 = list_entry(pos2, struct __prelim_ref, list);
 512
 513                         if (mode == 1) {
 514                                 if (!ref_for_same_block(ref1, ref2))
 515                                         continue;
 516                                 if (!ref1->parent && ref2->parent) {
 517                                         xchg = ref1;
 518                                         ref1 = ref2;
 519                                         ref2 = xchg;
 520                                 }
 521                         } else {
 522                                 if (ref1->parent != ref2->parent)
 523                                         continue;
 524                         }
 525
 526                         eie = ref1->inode_list;
 527                         while (eie && eie->next)
 528                                 eie = eie->next;
 529                         if (eie)
 530                                 eie->next = ref2->inode_list;
 531                         else
 532                                 ref1->inode_list = ref2->inode_list;
 533                         ref1->count += ref2->count;
 534
 535                         list_del(&ref2->list);
 536                         kmem_cache_free(btrfs_prelim_ref_cache, ref2);
 537                 }
 538
 539         }
 540 }
 541
 542 /*
 543  * add all currently queued delayed refs from this head whose seq nr is
 544  * smaller or equal that seq to the list
 545  */
 546 static int __add_delayed_refs(struct btrfs_delayed_ref_head *head, u64 seq,
 547                               struct list_head *prefs)
 548 {
 549         struct btrfs_delayed_extent_op *extent_op = head->extent_op;
 550         struct rb_node *n = &head->node.rb_node;
 551         struct btrfs_key key;
 552         struct btrfs_key op_key = {0};
 553         int sgn;
 554         int ret = 0;
 555
 556         if (extent_op && extent_op->update_key)
 557                 btrfs_disk_key_to_cpu(&op_key, &extent_op->key);
 558
 559         spin_lock(&head->lock);
 560         n = rb_first(&head->ref_root);
 561         while (n) {
 562                 struct btrfs_delayed_ref_node *node;
 563                 node = rb_entry(n, struct btrfs_delayed_ref_node,
 564                                 rb_node);
 565                 n = rb_next(n);
 566                 if (node->seq > seq)
 567                         continue;
 568
 569                 switch (node->action) {
 570                 case BTRFS_ADD_DELAYED_EXTENT:
 571                 case BTRFS_UPDATE_DELAYED_HEAD:
 572                         WARN_ON(1);
 573                         continue;
 574                 case BTRFS_ADD_DELAYED_REF:
 575                         sgn = 1;
 576                         break;
 577                 case BTRFS_DROP_DELAYED_REF:
 578                         sgn = -1;
 579                         break;
 580                 default:
 581                         BUG_ON(1);
 582                 }
 583                 switch (node->type) {
 584                 case BTRFS_TREE_BLOCK_REF_KEY: {
 585                         struct btrfs_delayed_tree_ref *ref;
 586
 587                         ref = btrfs_delayed_node_to_tree_ref(node);
 588                         ret = __add_prelim_ref(prefs, ref->root, &op_key,
 589                                                ref->level + 1, 0, node->bytenr,
 590                                                node->ref_mod * sgn, GFP_ATOMIC);
 591                         break;
 592                 }
 593                 case BTRFS_SHARED_BLOCK_REF_KEY: {
 594                         struct btrfs_delayed_tree_ref *ref;
 595
 596                         ref = btrfs_delayed_node_to_tree_ref(node);
 597                         ret = __add_prelim_ref(prefs, ref->root, NULL,
 598                                                ref->level + 1, ref->parent,
 599                                                node->bytenr,
 600                                                node->ref_mod * sgn, GFP_ATOMIC);
 601                         break;
 602                 }
 603                 case BTRFS_EXTENT_DATA_REF_KEY: {
 604                         struct btrfs_delayed_data_ref *ref;
 605                         ref = btrfs_delayed_node_to_data_ref(node);
 606
 607                         key.objectid = ref->objectid;
 608                         key.type = BTRFS_EXTENT_DATA_KEY;
 609                         key.offset = ref->offset;
 610                         ret = __add_prelim_ref(prefs, ref->root, &key, 0, 0,
 611                                                node->bytenr,
 612                                                node->ref_mod * sgn, GFP_ATOMIC);
 613                         break;
 614                 }
 615                 case BTRFS_SHARED_DATA_REF_KEY: {
 616                         struct btrfs_delayed_data_ref *ref;
 617
 618                         ref = btrfs_delayed_node_to_data_ref(node);
 619
 620                         key.objectid = ref->objectid;
 621                         key.type = BTRFS_EXTENT_DATA_KEY;
 622                         key.offset = ref->offset;
 623                         ret = __add_prelim_ref(prefs, ref->root, &key, 0,
 624                                                ref->parent, node->bytenr,
 625                                                node->ref_mod * sgn, GFP_ATOMIC);
 626                         break;
 627                 }
 628                 default:
 629                         WARN_ON(1);
 630                 }
 631                 if (ret)
 632                         break;
 633         }
 634         spin_unlock(&head->lock);
 635         return ret;
 636 }
 637
 638 /*
 639  * add all inline backrefs for bytenr to the list
 640  */
 641 static int __add_inline_refs(struct btrfs_fs_info *fs_info,
 642                              struct btrfs_path *path, u64 bytenr,
 643                              int *info_level, struct list_head *prefs)
 644 {
 645         int ret = 0;
 646         int slot;
 647         struct extent_buffer *leaf;
 648         struct btrfs_key key;
 649         struct btrfs_key found_key;
 650         unsigned long ptr;
 651         unsigned long end;
 652         struct btrfs_extent_item *ei;
 653         u64 flags;
 654         u64 item_size;
 655
 656         /*
 657          * enumerate all inline refs
 658          */
 659         leaf = path->nodes[0];
 660         slot = path->slots[0];
 661
 662         item_size = btrfs_item_size_nr(leaf, slot);
 663         BUG_ON(item_size < sizeof(*ei));
 664
 665         ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
 666         flags = btrfs_extent_flags(leaf, ei);
 667         btrfs_item_key_to_cpu(leaf, &found_key, slot);
 668
 669         ptr = (unsigned long)(ei + 1);
 670         end = (unsigned long)ei + item_size;
 671
 672         if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
 673             flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
 674                 struct btrfs_tree_block_info *info;
 675
 676                 info = (struct btrfs_tree_block_info *)ptr;
 677                 *info_level = btrfs_tree_block_level(leaf, info);
 678                 ptr += sizeof(struct btrfs_tree_block_info);
 679                 BUG_ON(ptr > end);
 680         } else if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
 681                 *info_level = found_key.offset;
 682         } else {
 683                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
 684         }
 685
 686         while (ptr < end) {
 687                 struct btrfs_extent_inline_ref *iref;
 688                 u64 offset;
 689                 int type;
 690
 691                 iref = (struct btrfs_extent_inline_ref *)ptr;
 692                 type = btrfs_extent_inline_ref_type(leaf, iref);
 693                 offset = btrfs_extent_inline_ref_offset(leaf, iref);
 694
 695                 switch (type) {
 696                 case BTRFS_SHARED_BLOCK_REF_KEY:
 697                         ret = __add_prelim_ref(prefs, 0, NULL,
 698                                                 *info_level + 1, offset,
 699                                                 bytenr, 1, GFP_NOFS);
 700                         break;
 701                 case BTRFS_SHARED_DATA_REF_KEY: {
 702                         struct btrfs_shared_data_ref *sdref;
 703                         int count;
 704
 705                         sdref = (struct btrfs_shared_data_ref *)(iref + 1);
 706                         count = btrfs_shared_data_ref_count(leaf, sdref);
 707                         ret = __add_prelim_ref(prefs, 0, NULL, 0, offset,
 708                                                bytenr, count, GFP_NOFS);
 709                         break;
 710                 }
 711                 case BTRFS_TREE_BLOCK_REF_KEY:
 712                         ret = __add_prelim_ref(prefs, offset, NULL,
 713                                                *info_level + 1, 0,
 714                                                bytenr, 1, GFP_NOFS);
 715                         break;
 716                 case BTRFS_EXTENT_DATA_REF_KEY: {
 717                         struct btrfs_extent_data_ref *dref;
 718                         int count;
 719                         u64 root;
 720
 721                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
 722                         count = btrfs_extent_data_ref_count(leaf, dref);
 723                         key.objectid = btrfs_extent_data_ref_objectid(leaf,
 724                                                                       dref);
 725                         key.type = BTRFS_EXTENT_DATA_KEY;
 726                         key.offset = btrfs_extent_data_ref_offset(leaf, dref);
 727                         root = btrfs_extent_data_ref_root(leaf, dref);
 728                         ret = __add_prelim_ref(prefs, root, &key, 0, 0,
 729                                                bytenr, count, GFP_NOFS);
 730                         break;
 731                 }
 732                 default:
 733                         WARN_ON(1);
 734                 }
 735                 if (ret)
 736                         return ret;
 737                 ptr += btrfs_extent_inline_ref_size(type);
 738         }
 739
 740         return 0;
 741 }
 742
 743 /*
 744  * add all non-inline backrefs for bytenr to the list
 745  */
 746 static int __add_keyed_refs(struct btrfs_fs_info *fs_info,
 747                             struct btrfs_path *path, u64 bytenr,
 748                             int info_level, struct list_head *prefs)
 749 {
 750         struct btrfs_root *extent_root = fs_info->extent_root;
 751         int ret;
 752         int slot;
 753         struct extent_buffer *leaf;
 754         struct btrfs_key key;
 755
 756         while (1) {
 757                 ret = btrfs_next_item(extent_root, path);
 758                 if (ret < 0)
 759                         break;
 760                 if (ret) {
 761                         ret = 0;
 762                         break;
 763                 }
 764
 765                 slot = path->slots[0];
 766                 leaf = path->nodes[0];
 767                 btrfs_item_key_to_cpu(leaf, &key, slot);
 768
 769                 if (key.objectid != bytenr)
 770                         break;
 771                 if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
 772                         continue;
 773                 if (key.type > BTRFS_SHARED_DATA_REF_KEY)
 774                         break;
 775
 776                 switch (key.type) {
 777                 case BTRFS_SHARED_BLOCK_REF_KEY:
 778                         ret = __add_prelim_ref(prefs, 0, NULL,
 779                                                 info_level + 1, key.offset,
 780                                                 bytenr, 1, GFP_NOFS);
 781                         break;
 782                 case BTRFS_SHARED_DATA_REF_KEY: {
 783                         struct btrfs_shared_data_ref *sdref;
 784                         int count;
 785
 786                         sdref = btrfs_item_ptr(leaf, slot,
 787                                               struct btrfs_shared_data_ref);
 788                         count = btrfs_shared_data_ref_count(leaf, sdref);
 789                         ret = __add_prelim_ref(prefs, 0, NULL, 0, key.offset,
 790                                                 bytenr, count, GFP_NOFS);
 791                         break;
 792                 }
 793                 case BTRFS_TREE_BLOCK_REF_KEY:
 794                         ret = __add_prelim_ref(prefs, key.offset, NULL,
 795                                                info_level + 1, 0,
 796                                                bytenr, 1, GFP_NOFS);
 797                         break;
 798                 case BTRFS_EXTENT_DATA_REF_KEY: {
 799                         struct btrfs_extent_data_ref *dref;
 800                         int count;
 801                         u64 root;
 802
 803                         dref = btrfs_item_ptr(leaf, slot,
 804                                               struct btrfs_extent_data_ref);
 805                         count = btrfs_extent_data_ref_count(leaf, dref);
 806                         key.objectid = btrfs_extent_data_ref_objectid(leaf,
 807                                                                       dref);
 808                         key.type = BTRFS_EXTENT_DATA_KEY;
 809                         key.offset = btrfs_extent_data_ref_offset(leaf, dref);
 810                         root = btrfs_extent_data_ref_root(leaf, dref);
 811                         ret = __add_prelim_ref(prefs, root, &key, 0, 0,
 812                                                bytenr, count, GFP_NOFS);
 813                         break;
 814                 }
 815                 default:
 816                         WARN_ON(1);
 817                 }
 818                 if (ret)
 819                         return ret;
 820
 821         }
 822
 823         return ret;
 824 }
 825
 826 /*
 827  * this adds all existing backrefs (inline backrefs, backrefs and delayed
 828  * refs) for the given bytenr to the refs list, merges duplicates and resolves
 829  * indirect refs to their parent bytenr.
 830  * When roots are found, they're added to the roots list
 831  *
 832  * FIXME some caching might speed things up
 833  */
 834 static int find_parent_nodes(struct btrfs_trans_handle *trans,
 835                              struct btrfs_fs_info *fs_info, u64 bytenr,
 836                              u64 time_seq, struct ulist *refs,
 837                              struct ulist *roots, const u64 *extent_item_pos)
 838 {
 839         struct btrfs_key key;
 840         struct btrfs_path *path;
 841         struct btrfs_delayed_ref_root *delayed_refs = NULL;
 842         struct btrfs_delayed_ref_head *head;
 843         int info_level = 0;
 844         int ret;
 845         struct list_head prefs_delayed;
 846         struct list_head prefs;
 847         struct __prelim_ref *ref;
 848
 849         INIT_LIST_HEAD(&prefs);
 850         INIT_LIST_HEAD(&prefs_delayed);
 851
 852         key.objectid = bytenr;
 853         key.offset = (u64)-1;
 854         if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
 855                 key.type = BTRFS_METADATA_ITEM_KEY;
 856         else
 857                 key.type = BTRFS_EXTENT_ITEM_KEY;
 858
 859         path = btrfs_alloc_path();
 860         if (!path)
 861                 return -ENOMEM;
 862         if (!trans)
 863                 path->search_commit_root = 1;
 864
 865         /*
 866          * grab both a lock on the path and a lock on the delayed ref head.
 867          * We need both to get a consistent picture of how the refs look
 868          * at a specified point in time
 869          */
 870 again:
 871         head = NULL;
 872
 873         ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
 874         if (ret < 0)
 875                 goto out;
 876         BUG_ON(ret == 0);
 877
 878         if (trans) {
 879                 /*
 880                  * look if there are updates for this ref queued and lock the
 881                  * head
 882                  */
 883                 delayed_refs = &trans->transaction->delayed_refs;
 884                 spin_lock(&delayed_refs->lock);
 885                 head = btrfs_find_delayed_ref_head(trans, bytenr);
 886                 if (head) {
 887                         if (!mutex_trylock(&head->mutex)) {
 888                                 atomic_inc(&head->node.refs);
 889                                 spin_unlock(&delayed_refs->lock);
 890
 891                                 btrfs_release_path(path);
 892
 893                                 /*
 894                                  * Mutex was contended, block until it's
 895                                  * released and try again
 896                                  */
 897                                 mutex_lock(&head->mutex);
 898                                 mutex_unlock(&head->mutex);
 899                                 btrfs_put_delayed_ref(&head->node);
 900                                 goto again;
 901                         }
 902                         spin_unlock(&delayed_refs->lock);
 903                         ret = __add_delayed_refs(head, time_seq,
 904                                                  &prefs_delayed);
 905                         mutex_unlock(&head->mutex);
 906                         if (ret)
 907                                 goto out;
 908                 } else {
 909                         spin_unlock(&delayed_refs->lock);
 910                 }
 911         }
 912
 913         if (path->slots[0]) {
 914                 struct extent_buffer *leaf;
 915                 int slot;
 916
 917                 path->slots[0]--;
 918                 leaf = path->nodes[0];
 919                 slot = path->slots[0];
 920                 btrfs_item_key_to_cpu(leaf, &key, slot);
 921                 if (key.objectid == bytenr &&
 922                     (key.type == BTRFS_EXTENT_ITEM_KEY ||
 923                      key.type == BTRFS_METADATA_ITEM_KEY)) {
 924                         ret = __add_inline_refs(fs_info, path, bytenr,
 925                                                 &info_level, &prefs);
 926                         if (ret)
 927                                 goto out;
 928                         ret = __add_keyed_refs(fs_info, path, bytenr,
 929                                                info_level, &prefs);
 930                         if (ret)
 931                                 goto out;
 932                 }
 933         }
 934         btrfs_release_path(path);
 935
 936         list_splice_init(&prefs_delayed, &prefs);
 937
 938         ret = __add_missing_keys(fs_info, &prefs);
 939         if (ret)
 940                 goto out;
 941
 942         __merge_refs(&prefs, 1);
 943
 944         ret = __resolve_indirect_refs(fs_info, path, time_seq, &prefs,
 945                                       extent_item_pos);
 946         if (ret)
 947                 goto out;
 948
 949         __merge_refs(&prefs, 2);
 950
 951         while (!list_empty(&prefs)) {
 952                 ref = list_first_entry(&prefs, struct __prelim_ref, list);
 953                 WARN_ON(ref->count < 0);
 954                 if (ref->count && ref->root_id && ref->parent == 0) {
 955                         /* no parent == root of tree */
 956                         ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS);
 957                         if (ret < 0)
 958                                 goto out;
 959                 }
 960                 if (ref->count && ref->parent) {
 961                         struct extent_inode_elem *eie = NULL;
 962                         if (extent_item_pos && !ref->inode_list) {
 963                                 u32 bsz;
 964                                 struct extent_buffer *eb;
 965                                 bsz = btrfs_level_size(fs_info->extent_root,
 966                                                         info_level);
 967                                 eb = read_tree_block(fs_info->extent_root,
 968                                                            ref->parent, bsz, 0);
 969                                 if (!eb || !extent_buffer_uptodate(eb)) {
 970                                         free_extent_buffer(eb);
 971                                         ret = -EIO;
 972                                         goto out;
 973                                 }
 974                                 ret = find_extent_in_eb(eb, bytenr,
 975                                                         *extent_item_pos, &eie);
 976                                 free_extent_buffer(eb);
 977                                 if (ret < 0)
 978                                         goto out;
 979                                 ref->inode_list = eie;
 980                         }
 981                         ret = ulist_add_merge(refs, ref->parent,
 982                                               (uintptr_t)ref->inode_list,
 983                                               (u64 *)&eie, GFP_NOFS);
 984                         if (ret < 0)
 985                                 goto out;
 986                         if (!ret && extent_item_pos) {
 987                                 /*
 988                                  * we've recorded that parent, so we must extend
 989                                  * its inode list here
 990                                  */
 991                                 BUG_ON(!eie);
 992                                 while (eie->next)
 993                                         eie = eie->next;
 994                                 eie->next = ref->inode_list;
 995                         }
 996                 }
 997                 list_del(&ref->list);
 998                 kmem_cache_free(btrfs_prelim_ref_cache, ref);
 999         }
1000
1001 out:
1002         btrfs_free_path(path);
1003         while (!list_empty(&prefs)) {
1004                 ref = list_first_entry(&prefs, struct __prelim_ref, list);
1005                 list_del(&ref->list);
1006                 kmem_cache_free(btrfs_prelim_ref_cache, ref);
1007         }
1008         while (!list_empty(&prefs_delayed)) {
1009                 ref = list_first_entry(&prefs_delayed, struct __prelim_ref,
1010                                        list);
1011                 list_del(&ref->list);
1012                 kmem_cache_free(btrfs_prelim_ref_cache, ref);
1013         }
1014
1015         return ret;
1016 }
1017
1018 static void free_leaf_list(struct ulist *blocks)
1019 {
1020         struct ulist_node *node = NULL;
1021         struct extent_inode_elem *eie;
1022         struct extent_inode_elem *eie_next;
1023         struct ulist_iterator uiter;
1024
1025         ULIST_ITER_INIT(&uiter);
1026         while ((node = ulist_next(blocks, &uiter))) {
1027                 if (!node->aux)
1028                         continue;
1029                 eie = (struct extent_inode_elem *)(uintptr_t)node->aux;
1030                 for (; eie; eie = eie_next) {
1031                         eie_next = eie->next;
1032                         kfree(eie);
1033                 }
1034                 node->aux = 0;
1035         }
1036
1037         ulist_free(blocks);
1038 }
1039
1040 /*
1041  * Finds all leafs with a reference to the specified combination of bytenr and
1042  * offset. key_list_head will point to a list of corresponding keys (caller must
1043  * free each list element). The leafs will be stored in the leafs ulist, which
1044  * must be freed with ulist_free.
1045  *
1046  * returns 0 on success, <0 on error
1047  */
1048 static int btrfs_find_all_leafs(struct btrfs_trans_handle *trans,
1049                                 struct btrfs_fs_info *fs_info, u64 bytenr,
1050                                 u64 time_seq, struct ulist **leafs,
1051                                 const u64 *extent_item_pos)
1052 {
1053         struct ulist *tmp;
1054         int ret;
1055
1056         tmp = ulist_alloc(GFP_NOFS);
1057         if (!tmp)
1058                 return -ENOMEM;
1059         *leafs = ulist_alloc(GFP_NOFS);
1060         if (!*leafs) {
1061                 ulist_free(tmp);
1062                 return -ENOMEM;
1063         }
1064
1065         ret = find_parent_nodes(trans, fs_info, bytenr,
1066                                 time_seq, *leafs, tmp, extent_item_pos);
1067         ulist_free(tmp);
1068
1069         if (ret < 0 && ret != -ENOENT) {
1070                 free_leaf_list(*leafs);
1071                 return ret;
1072         }
1073
1074         return 0;
1075 }
1076
1077 /*
1078  * walk all backrefs for a given extent to find all roots that reference this
1079  * extent. Walking a backref means finding all extents that reference this
1080  * extent and in turn walk the backrefs of those, too. Naturally this is a
1081  * recursive process, but here it is implemented in an iterative fashion: We
1082  * find all referencing extents for the extent in question and put them on a
1083  * list. In turn, we find all referencing extents for those, further appending
1084  * to the list. The way we iterate the list allows adding more elements after
1085  * the current while iterating. The process stops when we reach the end of the
1086  * list. Found roots are added to the roots list.
1087  *
1088  * returns 0 on success, < 0 on error.
1089  */
1090 int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1091                                 struct btrfs_fs_info *fs_info, u64 bytenr,
1092                                 u64 time_seq, struct ulist **roots)
1093 {
1094         struct ulist *tmp;
1095         struct ulist_node *node = NULL;
1096         struct ulist_iterator uiter;
1097         int ret;
1098
1099         tmp = ulist_alloc(GFP_NOFS);
1100         if (!tmp)
1101                 return -ENOMEM;
1102         *roots = ulist_alloc(GFP_NOFS);
1103         if (!*roots) {
1104                 ulist_free(tmp);
1105                 return -ENOMEM;
1106         }
1107
1108         ULIST_ITER_INIT(&uiter);
1109         while (1) {
1110                 ret = find_parent_nodes(trans, fs_info, bytenr,
1111                                         time_seq, tmp, *roots, NULL);
1112                 if (ret < 0 && ret != -ENOENT) {
1113                         ulist_free(tmp);
1114                         ulist_free(*roots);
1115                         return ret;
1116                 }
1117                 node = ulist_next(tmp, &uiter);
1118                 if (!node)
1119                         break;
1120                 bytenr = node->val;
1121                 cond_resched();
1122         }
1123
1124         ulist_free(tmp);
1125         return 0;
1126 }
1127
1128 /*
1129  * this makes the path point to (inum INODE_ITEM ioff)
1130  */
1131 int inode_item_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
1132                         struct btrfs_path *path)
1133 {
1134         struct btrfs_key key;
1135         return btrfs_find_item(fs_root, path, inum, ioff,
1136                         BTRFS_INODE_ITEM_KEY, &key);
1137 }
1138
1139 static int inode_ref_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
1140                                 struct btrfs_path *path,
1141                                 struct btrfs_key *found_key)
1142 {
1143         return btrfs_find_item(fs_root, path, inum, ioff,
1144                         BTRFS_INODE_REF_KEY, found_key);
1145 }
1146
1147 int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,
1148                           u64 start_off, struct btrfs_path *path,
1149                           struct btrfs_inode_extref **ret_extref,
1150                           u64 *found_off)
1151 {
1152         int ret, slot;
1153         struct btrfs_key key;
1154         struct btrfs_key found_key;
1155         struct btrfs_inode_extref *extref;
1156         struct extent_buffer *leaf;
1157         unsigned long ptr;
1158
1159         key.objectid = inode_objectid;
1160         btrfs_set_key_type(&key, BTRFS_INODE_EXTREF_KEY);
1161         key.offset = start_off;
1162
1163         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1164         if (ret < 0)
1165                 return ret;
1166
1167         while (1) {
1168                 leaf = path->nodes[0];
1169                 slot = path->slots[0];
1170                 if (slot >= btrfs_header_nritems(leaf)) {
1171                         /*
1172                          * If the item at offset is not found,
1173                          * btrfs_search_slot will point us to the slot
1174                          * where it should be inserted. In our case
1175                          * that will be the slot directly before the
1176                          * next INODE_REF_KEY_V2 item. In the case
1177                          * that we're pointing to the last slot in a
1178                          * leaf, we must move one leaf over.
1179                          */
1180                         ret = btrfs_next_leaf(root, path);
1181                         if (ret) {
1182                                 if (ret >= 1)
1183                                         ret = -ENOENT;
1184                                 break;
1185                         }
1186                         continue;
1187                 }
1188
1189                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1190
1191                 /*
1192                  * Check that we're still looking at an extended ref key for
1193                  * this particular objectid. If we have different
1194                  * objectid or type then there are no more to be found
1195                  * in the tree and we can exit.
1196                  */
1197                 ret = -ENOENT;
1198                 if (found_key.objectid != inode_objectid)
1199                         break;
1200                 if (btrfs_key_type(&found_key) != BTRFS_INODE_EXTREF_KEY)
1201                         break;
1202
1203                 ret = 0;
1204                 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1205                 extref = (struct btrfs_inode_extref *)ptr;
1206                 *ret_extref = extref;
1207                 if (found_off)
1208                         *found_off = found_key.offset;
1209                 break;
1210         }
1211
1212         return ret;
1213 }
1214
1215 /*
1216  * this iterates to turn a name (from iref/extref) into a full filesystem path.
1217  * Elements of the path are separated by '/' and the path is guaranteed to be
1218  * 0-terminated. the path is only given within the current file system.
1219  * Therefore, it never starts with a '/'. the caller is responsible to provide
1220  * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
1221  * the start point of the resulting string is returned. this pointer is within
1222  * dest, normally.
1223  * in case the path buffer would overflow, the pointer is decremented further
1224  * as if output was written to the buffer, though no more output is actually
1225  * generated. that way, the caller can determine how much space would be
1226  * required for the path to fit into the buffer. in that case, the returned
1227  * value will be smaller than dest. callers must check this!
1228  */
1229 char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
1230                         u32 name_len, unsigned long name_off,
1231                         struct extent_buffer *eb_in, u64 parent,
1232                         char *dest, u32 size)
1233 {
1234         int slot;
1235         u64 next_inum;
1236         int ret;
1237         s64 bytes_left = ((s64)size) - 1;
1238         struct extent_buffer *eb = eb_in;
1239         struct btrfs_key found_key;
1240         int leave_spinning = path->leave_spinning;
1241         struct btrfs_inode_ref *iref;
1242
1243         if (bytes_left >= 0)
1244                 dest[bytes_left] = '\0';
1245
1246         path->leave_spinning = 1;
1247         while (1) {
1248                 bytes_left -= name_len;
1249                 if (bytes_left >= 0)
1250                         read_extent_buffer(eb, dest + bytes_left,
1251                                            name_off, name_len);
1252                 if (eb != eb_in) {
1253                         btrfs_tree_read_unlock_blocking(eb);
1254                         free_extent_buffer(eb);
1255                 }
1256                 ret = inode_ref_info(parent, 0, fs_root, path, &found_key);
1257                 if (ret > 0)
1258                         ret = -ENOENT;
1259                 if (ret)
1260                         break;
1261
1262                 next_inum = found_key.offset;
1263
1264                 /* regular exit ahead */
1265                 if (parent == next_inum)
1266                         break;
1267
1268                 slot = path->slots[0];
1269                 eb = path->nodes[0];
1270                 /* make sure we can use eb after releasing the path */
1271                 if (eb != eb_in) {
1272                         atomic_inc(&eb->refs);
1273                         btrfs_tree_read_lock(eb);
1274                         btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1275                 }
1276                 btrfs_release_path(path);
1277                 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1278
1279                 name_len = btrfs_inode_ref_name_len(eb, iref);
1280                 name_off = (unsigned long)(iref + 1);
1281
1282                 parent = next_inum;
1283                 --bytes_left;
1284                 if (bytes_left >= 0)
1285                         dest[bytes_left] = '/';
1286         }
1287
1288         btrfs_release_path(path);
1289         path->leave_spinning = leave_spinning;
1290
1291         if (ret)
1292                 return ERR_PTR(ret);
1293
1294         return dest + bytes_left;
1295 }
1296
1297 /*
1298  * this makes the path point to (logical EXTENT_ITEM *)
1299  * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
1300  * tree blocks and <0 on error.
1301  */
1302 int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
1303                         struct btrfs_path *path, struct btrfs_key *found_key,
1304                         u64 *flags_ret)
1305 {
1306         int ret;
1307         u64 flags;
1308         u64 size = 0;
1309         u32 item_size;
1310         struct extent_buffer *eb;
1311         struct btrfs_extent_item *ei;
1312         struct btrfs_key key;
1313
1314         if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1315                 key.type = BTRFS_METADATA_ITEM_KEY;
1316         else
1317                 key.type = BTRFS_EXTENT_ITEM_KEY;
1318         key.objectid = logical;
1319         key.offset = (u64)-1;
1320
1321         ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
1322         if (ret < 0)
1323                 return ret;
1324
1325         while (1) {
1326                 u32 nritems;
1327                 if (path->slots[0] == 0) {
1328                         btrfs_set_path_blocking(path);
1329                         ret = btrfs_prev_leaf(fs_info->extent_root, path);
1330                         if (ret != 0) {
1331                                 if (ret > 0) {
1332                                         pr_debug("logical %llu is not within "
1333                                                  "any extent\n", logical);
1334                                         ret = -ENOENT;
1335                                 }
1336                                 return ret;
1337                         }
1338                 } else {
1339                         path->slots[0]--;
1340                 }
1341                 nritems = btrfs_header_nritems(path->nodes[0]);
1342                 if (nritems == 0) {
1343                         pr_debug("logical %llu is not within any extent\n",
1344                                  logical);
1345                         return -ENOENT;
1346                 }
1347                 if (path->slots[0] == nritems)
1348                         path->slots[0]--;
1349
1350                 btrfs_item_key_to_cpu(path->nodes[0], found_key,
1351                                       path->slots[0]);
1352                 if (found_key->type == BTRFS_EXTENT_ITEM_KEY ||
1353                     found_key->type == BTRFS_METADATA_ITEM_KEY)
1354                         break;
1355         }
1356
1357         if (found_key->type == BTRFS_METADATA_ITEM_KEY)
1358                 size = fs_info->extent_root->leafsize;
1359         else if (found_key->type == BTRFS_EXTENT_ITEM_KEY)
1360                 size = found_key->offset;
1361
1362         if (found_key->objectid > logical ||
1363             found_key->objectid + size <= logical) {
1364                 pr_debug("logical %llu is not within any extent\n", logical);
1365                 return -ENOENT;
1366         }
1367
1368         eb = path->nodes[0];
1369         item_size = btrfs_item_size_nr(eb, path->slots[0]);
1370         BUG_ON(item_size < sizeof(*ei));
1371
1372         ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
1373         flags = btrfs_extent_flags(eb, ei);
1374
1375         pr_debug("logical %llu is at position %llu within the extent (%llu "
1376                  "EXTENT_ITEM %llu) flags %#llx size %u\n",
1377                  logical, logical - found_key->objectid, found_key->objectid,
1378                  found_key->offset, flags, item_size);
1379
1380         WARN_ON(!flags_ret);
1381         if (flags_ret) {
1382                 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1383                         *flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK;
1384                 else if (flags & BTRFS_EXTENT_FLAG_DATA)
1385                         *flags_ret = BTRFS_EXTENT_FLAG_DATA;
1386                 else
1387                         BUG_ON(1);
1388                 return 0;
1389         }
1390
1391         return -EIO;
1392 }
1393
1394 /*
1395  * helper function to iterate extent inline refs. ptr must point to a 0 value
1396  * for the first call and may be modified. it is used to track state.
1397  * if more refs exist, 0 is returned and the next call to
1398  * __get_extent_inline_ref must pass the modified ptr parameter to get the
1399  * next ref. after the last ref was processed, 1 is returned.
1400  * returns <0 on error
1401  */
1402 static int __get_extent_inline_ref(unsigned long *ptr, struct extent_buffer *eb,
1403                                 struct btrfs_extent_item *ei, u32 item_size,
1404                                 struct btrfs_extent_inline_ref **out_eiref,
1405                                 int *out_type)
1406 {
1407         unsigned long end;
1408         u64 flags;
1409         struct btrfs_tree_block_info *info;
1410
1411         if (!*ptr) {
1412                 /* first call */
1413                 flags = btrfs_extent_flags(eb, ei);
1414                 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1415                         info = (struct btrfs_tree_block_info *)(ei + 1);
1416                         *out_eiref =
1417                                 (struct btrfs_extent_inline_ref *)(info + 1);
1418                 } else {
1419                         *out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1);
1420                 }
1421                 *ptr = (unsigned long)*out_eiref;
1422                 if ((void *)*ptr >= (void *)ei + item_size)
1423                         return -ENOENT;
1424         }
1425
1426         end = (unsigned long)ei + item_size;
1427         *out_eiref = (struct btrfs_extent_inline_ref *)*ptr;
1428         *out_type = btrfs_extent_inline_ref_type(eb, *out_eiref);
1429
1430         *ptr += btrfs_extent_inline_ref_size(*out_type);
1431         WARN_ON(*ptr > end);
1432         if (*ptr == end)
1433                 return 1; /* last */
1434
1435         return 0;
1436 }
1437
1438 /*
1439  * reads the tree block backref for an extent. tree level and root are returned
1440  * through out_level and out_root. ptr must point to a 0 value for the first
1441  * call and may be modified (see __get_extent_inline_ref comment).
1442  * returns 0 if data was provided, 1 if there was no more data to provide or
1443  * <0 on error.
1444  */
1445 int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
1446                                 struct btrfs_extent_item *ei, u32 item_size,
1447                                 u64 *out_root, u8 *out_level)
1448 {
1449         int ret;
1450         int type;
1451         struct btrfs_tree_block_info *info;
1452         struct btrfs_extent_inline_ref *eiref;
1453
1454         if (*ptr == (unsigned long)-1)
1455                 return 1;
1456
1457         while (1) {
1458                 ret = __get_extent_inline_ref(ptr, eb, ei, item_size,
1459                                                 &eiref, &type);
1460                 if (ret < 0)
1461                         return ret;
1462
1463                 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1464                     type == BTRFS_SHARED_BLOCK_REF_KEY)
1465                         break;
1466
1467                 if (ret == 1)
1468                         return 1;
1469         }
1470
1471         /* we can treat both ref types equally here */
1472         info = (struct btrfs_tree_block_info *)(ei + 1);
1473         *out_root = btrfs_extent_inline_ref_offset(eb, eiref);
1474         *out_level = btrfs_tree_block_level(eb, info);
1475
1476         if (ret == 1)
1477                 *ptr = (unsigned long)-1;
1478
1479         return 0;
1480 }
1481
1482 static int iterate_leaf_refs(struct extent_inode_elem *inode_list,
1483                                 u64 root, u64 extent_item_objectid,
1484                                 iterate_extent_inodes_t *iterate, void *ctx)
1485 {
1486         struct extent_inode_elem *eie;
1487         int ret = 0;
1488
1489         for (eie = inode_list; eie; eie = eie->next) {
1490                 pr_debug("ref for %llu resolved, key (%llu EXTEND_DATA %llu), "
1491                          "root %llu\n", extent_item_objectid,
1492                          eie->inum, eie->offset, root);
1493                 ret = iterate(eie->inum, eie->offset, root, ctx);
1494                 if (ret) {
1495                         pr_debug("stopping iteration for %llu due to ret=%d\n",
1496                                  extent_item_objectid, ret);
1497                         break;
1498                 }
1499         }
1500
1501         return ret;
1502 }
1503
1504 /*
1505  * calls iterate() for every inode that references the extent identified by
1506  * the given parameters.
1507  * when the iterator function returns a non-zero value, iteration stops.
1508  */
1509 int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
1510                                 u64 extent_item_objectid, u64 extent_item_pos,
1511                                 int search_commit_root,
1512                                 iterate_extent_inodes_t *iterate, void *ctx)
1513 {
1514         int ret;
1515         struct btrfs_trans_handle *trans = NULL;
1516         struct ulist *refs = NULL;
1517         struct ulist *roots = NULL;
1518         struct ulist_node *ref_node = NULL;
1519         struct ulist_node *root_node = NULL;
1520         struct seq_list tree_mod_seq_elem = {};
1521         struct ulist_iterator ref_uiter;
1522         struct ulist_iterator root_uiter;
1523
1524         pr_debug("resolving all inodes for extent %llu\n",
1525                         extent_item_objectid);
1526
1527         if (!search_commit_root) {
1528                 trans = btrfs_join_transaction(fs_info->extent_root);
1529                 if (IS_ERR(trans))
1530                         return PTR_ERR(trans);
1531                 btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1532         }
1533
1534         ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid,
1535                                    tree_mod_seq_elem.seq, &refs,
1536                                    &extent_item_pos);
1537         if (ret)
1538                 goto out;
1539
1540         ULIST_ITER_INIT(&ref_uiter);
1541         while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) {
1542                 ret = btrfs_find_all_roots(trans, fs_info, ref_node->val,
1543                                            tree_mod_seq_elem.seq, &roots);
1544                 if (ret)
1545                         break;
1546                 ULIST_ITER_INIT(&root_uiter);
1547                 while (!ret && (root_node = ulist_next(roots, &root_uiter))) {
1548                         pr_debug("root %llu references leaf %llu, data list "
1549                                  "%#llx\n", root_node->val, ref_node->val,
1550                                  ref_node->aux);
1551                         ret = iterate_leaf_refs((struct extent_inode_elem *)
1552                                                 (uintptr_t)ref_node->aux,
1553                                                 root_node->val,
1554                                                 extent_item_objectid,
1555                                                 iterate, ctx);
1556                 }
1557                 ulist_free(roots);
1558         }
1559
1560         free_leaf_list(refs);
1561 out:
1562         if (!search_commit_root) {
1563                 btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1564                 btrfs_end_transaction(trans, fs_info->extent_root);
1565         }
1566
1567         return ret;
1568 }
1569
1570 int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
1571                                 struct btrfs_path *path,
1572                                 iterate_extent_inodes_t *iterate, void *ctx)
1573 {
1574         int ret;
1575         u64 extent_item_pos;
1576         u64 flags = 0;
1577         struct btrfs_key found_key;
1578         int search_commit_root = path->search_commit_root;
1579
1580         ret = extent_from_logical(fs_info, logical, path, &found_key, &flags);
1581         btrfs_release_path(path);
1582         if (ret < 0)
1583                 return ret;
1584         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1585                 return -EINVAL;
1586
1587         extent_item_pos = logical - found_key.objectid;
1588         ret = iterate_extent_inodes(fs_info, found_key.objectid,
1589                                         extent_item_pos, search_commit_root,
1590                                         iterate, ctx);
1591
1592         return ret;
1593 }
1594
1595 typedef int (iterate_irefs_t)(u64 parent, u32 name_len, unsigned long name_off,
1596                               struct extent_buffer *eb, void *ctx);
1597
1598 static int iterate_inode_refs(u64 inum, struct btrfs_root *fs_root,
1599                               struct btrfs_path *path,
1600                               iterate_irefs_t *iterate, void *ctx)
1601 {
1602         int ret = 0;
1603         int slot;
1604         u32 cur;
1605         u32 len;
1606         u32 name_len;
1607         u64 parent = 0;
1608         int found = 0;
1609         struct extent_buffer *eb;
1610         struct btrfs_item *item;
1611         struct btrfs_inode_ref *iref;
1612         struct btrfs_key found_key;
1613
1614         while (!ret) {
1615                 ret = inode_ref_info(inum, parent ? parent+1 : 0, fs_root, path,
1616                                      &found_key);
1617                 if (ret < 0)
1618                         break;
1619                 if (ret) {
1620                         ret = found ? 0 : -ENOENT;
1621                         break;
1622                 }
1623                 ++found;
1624
1625                 parent = found_key.offset;
1626                 slot = path->slots[0];
1627                 eb = btrfs_clone_extent_buffer(path->nodes[0]);
1628                 if (!eb) {
1629                         ret = -ENOMEM;
1630                         break;
1631                 }
1632                 extent_buffer_get(eb);
1633                 btrfs_tree_read_lock(eb);
1634                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1635                 btrfs_release_path(path);
1636
1637                 item = btrfs_item_nr(slot);
1638                 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1639
1640                 for (cur = 0; cur < btrfs_item_size(eb, item); cur += len) {
1641                         name_len = btrfs_inode_ref_name_len(eb, iref);
1642                         /* path must be released before calling iterate()! */
1643                         pr_debug("following ref at offset %u for inode %llu in "
1644                                  "tree %llu\n", cur, found_key.objectid,
1645                                  fs_root->objectid);
1646                         ret = iterate(parent, name_len,
1647                                       (unsigned long)(iref + 1), eb, ctx);
1648                         if (ret)
1649                                 break;
1650                         len = sizeof(*iref) + name_len;
1651                         iref = (struct btrfs_inode_ref *)((char *)iref + len);
1652                 }
1653                 btrfs_tree_read_unlock_blocking(eb);
1654                 free_extent_buffer(eb);
1655         }
1656
1657         btrfs_release_path(path);
1658
1659         return ret;
1660 }
1661
1662 static int iterate_inode_extrefs(u64 inum, struct btrfs_root *fs_root,
1663                                  struct btrfs_path *path,
1664                                  iterate_irefs_t *iterate, void *ctx)
1665 {
1666         int ret;
1667         int slot;
1668         u64 offset = 0;
1669         u64 parent;
1670         int found = 0;
1671         struct extent_buffer *eb;
1672         struct btrfs_inode_extref *extref;
1673         struct extent_buffer *leaf;
1674         u32 item_size;
1675         u32 cur_offset;
1676         unsigned long ptr;
1677
1678         while (1) {
1679                 ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref,
1680                                             &offset);
1681                 if (ret < 0)
1682                         break;
1683                 if (ret) {
1684                         ret = found ? 0 : -ENOENT;
1685                         break;
1686                 }
1687                 ++found;
1688
1689                 slot = path->slots[0];
1690                 eb = btrfs_clone_extent_buffer(path->nodes[0]);
1691                 if (!eb) {
1692                         ret = -ENOMEM;
1693                         break;
1694                 }
1695                 extent_buffer_get(eb);
1696
1697                 btrfs_tree_read_lock(eb);
1698                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1699                 btrfs_release_path(path);
1700
1701                 leaf = path->nodes[0];
1702                 item_size = btrfs_item_size_nr(leaf, slot);
1703                 ptr = btrfs_item_ptr_offset(leaf, slot);
1704                 cur_offset = 0;
1705
1706                 while (cur_offset < item_size) {
1707                         u32 name_len;
1708
1709                         extref = (struct btrfs_inode_extref *)(ptr + cur_offset);
1710                         parent = btrfs_inode_extref_parent(eb, extref);
1711                         name_len = btrfs_inode_extref_name_len(eb, extref);
1712                         ret = iterate(parent, name_len,
1713                                       (unsigned long)&extref->name, eb, ctx);
1714                         if (ret)
1715                                 break;
1716
1717                         cur_offset += btrfs_inode_extref_name_len(leaf, extref);
1718                         cur_offset += sizeof(*extref);
1719                 }
1720                 btrfs_tree_read_unlock_blocking(eb);
1721                 free_extent_buffer(eb);
1722
1723                 offset++;
1724         }
1725
1726         btrfs_release_path(path);
1727
1728         return ret;
1729 }
1730
1731 static int iterate_irefs(u64 inum, struct btrfs_root *fs_root,
1732                          struct btrfs_path *path, iterate_irefs_t *iterate,
1733                          void *ctx)
1734 {
1735         int ret;
1736         int found_refs = 0;
1737
1738         ret = iterate_inode_refs(inum, fs_root, path, iterate, ctx);
1739         if (!ret)
1740                 ++found_refs;
1741         else if (ret != -ENOENT)
1742                 return ret;
1743
1744         ret = iterate_inode_extrefs(inum, fs_root, path, iterate, ctx);
1745         if (ret == -ENOENT && found_refs)
1746                 return 0;
1747
1748         return ret;
1749 }
1750
1751 /*
1752  * returns 0 if the path could be dumped (probably truncated)
1753  * returns <0 in case of an error
1754  */
1755 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
1756                          struct extent_buffer *eb, void *ctx)
1757 {
1758         struct inode_fs_paths *ipath = ctx;
1759         char *fspath;
1760         char *fspath_min;
1761         int i = ipath->fspath->elem_cnt;
1762         const int s_ptr = sizeof(char *);
1763         u32 bytes_left;
1764
1765         bytes_left = ipath->fspath->bytes_left > s_ptr ?
1766                                         ipath->fspath->bytes_left - s_ptr : 0;
1767
1768         fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
1769         fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len,
1770                                    name_off, eb, inum, fspath_min, bytes_left);
1771         if (IS_ERR(fspath))
1772                 return PTR_ERR(fspath);
1773
1774         if (fspath > fspath_min) {
1775                 ipath->fspath->val[i] = (u64)(unsigned long)fspath;
1776                 ++ipath->fspath->elem_cnt;
1777                 ipath->fspath->bytes_left = fspath - fspath_min;
1778         } else {
1779                 ++ipath->fspath->elem_missed;
1780                 ipath->fspath->bytes_missing += fspath_min - fspath;
1781                 ipath->fspath->bytes_left = 0;
1782         }
1783
1784         return 0;
1785 }
1786
1787 /*
1788  * this dumps all file system paths to the inode into the ipath struct, provided
1789  * is has been created large enough. each path is zero-terminated and accessed
1790  * from ipath->fspath->val[i].
1791  * when it returns, there are ipath->fspath->elem_cnt number of paths available
1792  * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the
1793  * number of missed paths in recored in ipath->fspath->elem_missed, otherwise,
1794  * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
1795  * have been needed to return all paths.
1796  */
1797 int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
1798 {
1799         return iterate_irefs(inum, ipath->fs_root, ipath->btrfs_path,
1800                              inode_to_path, ipath);
1801 }
1802
1803 struct btrfs_data_container *init_data_container(u32 total_bytes)
1804 {
1805         struct btrfs_data_container *data;
1806         size_t alloc_bytes;
1807
1808         alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
1809         data = vmalloc(alloc_bytes);
1810         if (!data)
1811                 return ERR_PTR(-ENOMEM);
1812
1813         if (total_bytes >= sizeof(*data)) {
1814                 data->bytes_left = total_bytes - sizeof(*data);
1815                 data->bytes_missing = 0;
1816         } else {
1817                 data->bytes_missing = sizeof(*data) - total_bytes;
1818                 data->bytes_left = 0;
1819         }
1820
1821         data->elem_cnt = 0;
1822         data->elem_missed = 0;
1823
1824         return data;
1825 }
1826
1827 /*
1828  * allocates space to return multiple file system paths for an inode.
1829  * total_bytes to allocate are passed, note that space usable for actual path
1830  * information will be total_bytes - sizeof(struct inode_fs_paths).
1831  * the returned pointer must be freed with free_ipath() in the end.
1832  */
1833 struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
1834                                         struct btrfs_path *path)
1835 {
1836         struct inode_fs_paths *ifp;
1837         struct btrfs_data_container *fspath;
1838
1839         fspath = init_data_container(total_bytes);
1840         if (IS_ERR(fspath))
1841                 return (void *)fspath;
1842
1843         ifp = kmalloc(sizeof(*ifp), GFP_NOFS);
1844         if (!ifp) {
1845                 kfree(fspath);
1846                 return ERR_PTR(-ENOMEM);
1847         }
1848
1849         ifp->btrfs_path = path;
1850         ifp->fspath = fspath;
1851         ifp->fs_root = fs_root;
1852
1853         return ifp;
1854 }
1855
1856 void free_ipath(struct inode_fs_paths *ipath)
1857 {
1858         if (!ipath)
1859                 return;
1860         vfree(ipath->fspath);
1861         kfree(ipath);
1862 }