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_ptr(parents, eb->start,
 280                                                   eie, (void **)&old, GFP_NOFS);
 281                         if (ret < 0)
 282                                 break;
 283                         if (!ret && extent_item_pos) {
 284                                 while (old->next)
 285                                         old = old->next;
 286                                 old->next = eie;
 287                         }
 288                         eie = NULL;
 289                 }
 290 next:
 291                 ret = btrfs_next_old_item(root, path, time_seq);
 292         }
 293
 294         if (ret > 0)
 295                 ret = 0;
 296         else if (ret < 0)
 297                 free_inode_elem_list(eie);
 298         return ret;
 299 }
 300
 301 /*
 302  * resolve an indirect backref in the form (root_id, key, level)
 303  * to a logical address
 304  */
 305 static int __resolve_indirect_ref(struct btrfs_fs_info *fs_info,
 306                                   struct btrfs_path *path, u64 time_seq,
 307                                   struct __prelim_ref *ref,
 308                                   struct ulist *parents,
 309                                   const u64 *extent_item_pos, u64 total_refs)
 310 {
 311         struct btrfs_root *root;
 312         struct btrfs_key root_key;
 313         struct extent_buffer *eb;
 314         int ret = 0;
 315         int root_level;
 316         int level = ref->level;
 317         int index;
 318
 319         root_key.objectid = ref->root_id;
 320         root_key.type = BTRFS_ROOT_ITEM_KEY;
 321         root_key.offset = (u64)-1;
 322
 323         index = srcu_read_lock(&fs_info->subvol_srcu);
 324
 325         root = btrfs_read_fs_root_no_name(fs_info, &root_key);
 326         if (IS_ERR(root)) {
 327                 srcu_read_unlock(&fs_info->subvol_srcu, index);
 328                 ret = PTR_ERR(root);
 329                 goto out;
 330         }
 331
 332         if (path->search_commit_root)
 333                 root_level = btrfs_header_level(root->commit_root);
 334         else
 335                 root_level = btrfs_old_root_level(root, time_seq);
 336
 337         if (root_level + 1 == level) {
 338                 srcu_read_unlock(&fs_info->subvol_srcu, index);
 339                 goto out;
 340         }
 341
 342         path->lowest_level = level;
 343         ret = btrfs_search_old_slot(root, &ref->key_for_search, path, time_seq);
 344
 345         /* root node has been locked, we can release @subvol_srcu safely here */
 346         srcu_read_unlock(&fs_info->subvol_srcu, index);
 347
 348         pr_debug("search slot in root %llu (level %d, ref count %d) returned "
 349                  "%d for key (%llu %u %llu)\n",
 350                  ref->root_id, level, ref->count, ret,
 351                  ref->key_for_search.objectid, ref->key_for_search.type,
 352                  ref->key_for_search.offset);
 353         if (ret < 0)
 354                 goto out;
 355
 356         eb = path->nodes[level];
 357         while (!eb) {
 358                 if (WARN_ON(!level)) {
 359                         ret = 1;
 360                         goto out;
 361                 }
 362                 level--;
 363                 eb = path->nodes[level];
 364         }
 365
 366         ret = add_all_parents(root, path, parents, ref, level, time_seq,
 367                               extent_item_pos, total_refs);
 368 out:
 369         path->lowest_level = 0;
 370         btrfs_release_path(path);
 371         return ret;
 372 }
 373
 374 /*
 375  * resolve all indirect backrefs from the list
 376  */
 377 static int __resolve_indirect_refs(struct btrfs_fs_info *fs_info,
 378                                    struct btrfs_path *path, u64 time_seq,
 379                                    struct list_head *head,
 380                                    const u64 *extent_item_pos, u64 total_refs)
 381 {
 382         int err;
 383         int ret = 0;
 384         struct __prelim_ref *ref;
 385         struct __prelim_ref *ref_safe;
 386         struct __prelim_ref *new_ref;
 387         struct ulist *parents;
 388         struct ulist_node *node;
 389         struct ulist_iterator uiter;
 390
 391         parents = ulist_alloc(GFP_NOFS);
 392         if (!parents)
 393                 return -ENOMEM;
 394
 395         /*
 396          * _safe allows us to insert directly after the current item without
 397          * iterating over the newly inserted items.
 398          * we're also allowed to re-assign ref during iteration.
 399          */
 400         list_for_each_entry_safe(ref, ref_safe, head, list) {
 401                 if (ref->parent)        /* already direct */
 402                         continue;
 403                 if (ref->count == 0)
 404                         continue;
 405                 err = __resolve_indirect_ref(fs_info, path, time_seq, ref,
 406                                              parents, extent_item_pos,
 407                                              total_refs);
 408                 /*
 409                  * we can only tolerate ENOENT,otherwise,we should catch error
 410                  * and return directly.
 411                  */
 412                 if (err == -ENOENT) {
 413                         continue;
 414                 } else if (err) {
 415                         ret = err;
 416                         goto out;
 417                 }
 418
 419                 /* we put the first parent into the ref at hand */
 420                 ULIST_ITER_INIT(&uiter);
 421                 node = ulist_next(parents, &uiter);
 422                 ref->parent = node ? node->val : 0;
 423                 ref->inode_list = node ?
 424                         (struct extent_inode_elem *)(uintptr_t)node->aux : NULL;
 425
 426                 /* additional parents require new refs being added here */
 427                 while ((node = ulist_next(parents, &uiter))) {
 428                         new_ref = kmem_cache_alloc(btrfs_prelim_ref_cache,
 429                                                    GFP_NOFS);
 430                         if (!new_ref) {
 431                                 ret = -ENOMEM;
 432                                 goto out;
 433                         }
 434                         memcpy(new_ref, ref, sizeof(*ref));
 435                         new_ref->parent = node->val;
 436                         new_ref->inode_list = (struct extent_inode_elem *)
 437                                                         (uintptr_t)node->aux;
 438                         list_add(&new_ref->list, &ref->list);
 439                 }
 440                 ulist_reinit(parents);
 441         }
 442 out:
 443         ulist_free(parents);
 444         return ret;
 445 }
 446
 447 static inline int ref_for_same_block(struct __prelim_ref *ref1,
 448                                      struct __prelim_ref *ref2)
 449 {
 450         if (ref1->level != ref2->level)
 451                 return 0;
 452         if (ref1->root_id != ref2->root_id)
 453                 return 0;
 454         if (ref1->key_for_search.type != ref2->key_for_search.type)
 455                 return 0;
 456         if (ref1->key_for_search.objectid != ref2->key_for_search.objectid)
 457                 return 0;
 458         if (ref1->key_for_search.offset != ref2->key_for_search.offset)
 459                 return 0;
 460         if (ref1->parent != ref2->parent)
 461                 return 0;
 462
 463         return 1;
 464 }
 465
 466 /*
 467  * read tree blocks and add keys where required.
 468  */
 469 static int __add_missing_keys(struct btrfs_fs_info *fs_info,
 470                               struct list_head *head)
 471 {
 472         struct list_head *pos;
 473         struct extent_buffer *eb;
 474
 475         list_for_each(pos, head) {
 476                 struct __prelim_ref *ref;
 477                 ref = list_entry(pos, struct __prelim_ref, list);
 478
 479                 if (ref->parent)
 480                         continue;
 481                 if (ref->key_for_search.type)
 482                         continue;
 483                 BUG_ON(!ref->wanted_disk_byte);
 484                 eb = read_tree_block(fs_info->tree_root, ref->wanted_disk_byte,
 485                                      fs_info->tree_root->leafsize, 0);
 486                 if (!eb || !extent_buffer_uptodate(eb)) {
 487                         free_extent_buffer(eb);
 488                         return -EIO;
 489                 }
 490                 btrfs_tree_read_lock(eb);
 491                 if (btrfs_header_level(eb) == 0)
 492                         btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
 493                 else
 494                         btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
 495                 btrfs_tree_read_unlock(eb);
 496                 free_extent_buffer(eb);
 497         }
 498         return 0;
 499 }
 500
 501 /*
 502  * merge two lists of backrefs and adjust counts accordingly
 503  *
 504  * mode = 1: merge identical keys, if key is set
 505  *    FIXME: if we add more keys in __add_prelim_ref, we can merge more here.
 506  *           additionally, we could even add a key range for the blocks we
 507  *           looked into to merge even more (-> replace unresolved refs by those
 508  *           having a parent).
 509  * mode = 2: merge identical parents
 510  */
 511 static void __merge_refs(struct list_head *head, int mode)
 512 {
 513         struct list_head *pos1;
 514
 515         list_for_each(pos1, head) {
 516                 struct list_head *n2;
 517                 struct list_head *pos2;
 518                 struct __prelim_ref *ref1;
 519
 520                 ref1 = list_entry(pos1, struct __prelim_ref, list);
 521
 522                 for (pos2 = pos1->next, n2 = pos2->next; pos2 != head;
 523                      pos2 = n2, n2 = pos2->next) {
 524                         struct __prelim_ref *ref2;
 525                         struct __prelim_ref *xchg;
 526                         struct extent_inode_elem *eie;
 527
 528                         ref2 = list_entry(pos2, struct __prelim_ref, list);
 529
 530                         if (mode == 1) {
 531                                 if (!ref_for_same_block(ref1, ref2))
 532                                         continue;
 533                                 if (!ref1->parent && ref2->parent) {
 534                                         xchg = ref1;
 535                                         ref1 = ref2;
 536                                         ref2 = xchg;
 537                                 }
 538                         } else {
 539                                 if (ref1->parent != ref2->parent)
 540                                         continue;
 541                         }
 542
 543                         eie = ref1->inode_list;
 544                         while (eie && eie->next)
 545                                 eie = eie->next;
 546                         if (eie)
 547                                 eie->next = ref2->inode_list;
 548                         else
 549                                 ref1->inode_list = ref2->inode_list;
 550                         ref1->count += ref2->count;
 551
 552                         list_del(&ref2->list);
 553                         kmem_cache_free(btrfs_prelim_ref_cache, ref2);
 554                 }
 555
 556         }
 557 }
 558
 559 /*
 560  * add all currently queued delayed refs from this head whose seq nr is
 561  * smaller or equal that seq to the list
 562  */
 563 static int __add_delayed_refs(struct btrfs_delayed_ref_head *head, u64 seq,
 564                               struct list_head *prefs, u64 *total_refs)
 565 {
 566         struct btrfs_delayed_extent_op *extent_op = head->extent_op;
 567         struct rb_node *n = &head->node.rb_node;
 568         struct btrfs_key key;
 569         struct btrfs_key op_key = {0};
 570         int sgn;
 571         int ret = 0;
 572
 573         if (extent_op && extent_op->update_key)
 574                 btrfs_disk_key_to_cpu(&op_key, &extent_op->key);
 575
 576         spin_lock(&head->lock);
 577         n = rb_first(&head->ref_root);
 578         while (n) {
 579                 struct btrfs_delayed_ref_node *node;
 580                 node = rb_entry(n, struct btrfs_delayed_ref_node,
 581                                 rb_node);
 582                 n = rb_next(n);
 583                 if (node->seq > seq)
 584                         continue;
 585
 586                 switch (node->action) {
 587                 case BTRFS_ADD_DELAYED_EXTENT:
 588                 case BTRFS_UPDATE_DELAYED_HEAD:
 589                         WARN_ON(1);
 590                         continue;
 591                 case BTRFS_ADD_DELAYED_REF:
 592                         sgn = 1;
 593                         break;
 594                 case BTRFS_DROP_DELAYED_REF:
 595                         sgn = -1;
 596                         break;
 597                 default:
 598                         BUG_ON(1);
 599                 }
 600                 *total_refs += (node->ref_mod * sgn);
 601                 switch (node->type) {
 602                 case BTRFS_TREE_BLOCK_REF_KEY: {
 603                         struct btrfs_delayed_tree_ref *ref;
 604
 605                         ref = btrfs_delayed_node_to_tree_ref(node);
 606                         ret = __add_prelim_ref(prefs, ref->root, &op_key,
 607                                                ref->level + 1, 0, node->bytenr,
 608                                                node->ref_mod * sgn, GFP_ATOMIC);
 609                         break;
 610                 }
 611                 case BTRFS_SHARED_BLOCK_REF_KEY: {
 612                         struct btrfs_delayed_tree_ref *ref;
 613
 614                         ref = btrfs_delayed_node_to_tree_ref(node);
 615                         ret = __add_prelim_ref(prefs, ref->root, NULL,
 616                                                ref->level + 1, ref->parent,
 617                                                node->bytenr,
 618                                                node->ref_mod * sgn, GFP_ATOMIC);
 619                         break;
 620                 }
 621                 case BTRFS_EXTENT_DATA_REF_KEY: {
 622                         struct btrfs_delayed_data_ref *ref;
 623                         ref = btrfs_delayed_node_to_data_ref(node);
 624
 625                         key.objectid = ref->objectid;
 626                         key.type = BTRFS_EXTENT_DATA_KEY;
 627                         key.offset = ref->offset;
 628                         ret = __add_prelim_ref(prefs, ref->root, &key, 0, 0,
 629                                                node->bytenr,
 630                                                node->ref_mod * sgn, GFP_ATOMIC);
 631                         break;
 632                 }
 633                 case BTRFS_SHARED_DATA_REF_KEY: {
 634                         struct btrfs_delayed_data_ref *ref;
 635
 636                         ref = btrfs_delayed_node_to_data_ref(node);
 637
 638                         key.objectid = ref->objectid;
 639                         key.type = BTRFS_EXTENT_DATA_KEY;
 640                         key.offset = ref->offset;
 641                         ret = __add_prelim_ref(prefs, ref->root, &key, 0,
 642                                                ref->parent, node->bytenr,
 643                                                node->ref_mod * sgn, GFP_ATOMIC);
 644                         break;
 645                 }
 646                 default:
 647                         WARN_ON(1);
 648                 }
 649                 if (ret)
 650                         break;
 651         }
 652         spin_unlock(&head->lock);
 653         return ret;
 654 }
 655
 656 /*
 657  * add all inline backrefs for bytenr to the list
 658  */
 659 static int __add_inline_refs(struct btrfs_fs_info *fs_info,
 660                              struct btrfs_path *path, u64 bytenr,
 661                              int *info_level, struct list_head *prefs,
 662                              u64 *total_refs)
 663 {
 664         int ret = 0;
 665         int slot;
 666         struct extent_buffer *leaf;
 667         struct btrfs_key key;
 668         struct btrfs_key found_key;
 669         unsigned long ptr;
 670         unsigned long end;
 671         struct btrfs_extent_item *ei;
 672         u64 flags;
 673         u64 item_size;
 674
 675         /*
 676          * enumerate all inline refs
 677          */
 678         leaf = path->nodes[0];
 679         slot = path->slots[0];
 680
 681         item_size = btrfs_item_size_nr(leaf, slot);
 682         BUG_ON(item_size < sizeof(*ei));
 683
 684         ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
 685         flags = btrfs_extent_flags(leaf, ei);
 686         *total_refs += btrfs_extent_refs(leaf, ei);
 687         btrfs_item_key_to_cpu(leaf, &found_key, slot);
 688
 689         ptr = (unsigned long)(ei + 1);
 690         end = (unsigned long)ei + item_size;
 691
 692         if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
 693             flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
 694                 struct btrfs_tree_block_info *info;
 695
 696                 info = (struct btrfs_tree_block_info *)ptr;
 697                 *info_level = btrfs_tree_block_level(leaf, info);
 698                 ptr += sizeof(struct btrfs_tree_block_info);
 699                 BUG_ON(ptr > end);
 700         } else if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
 701                 *info_level = found_key.offset;
 702         } else {
 703                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
 704         }
 705
 706         while (ptr < end) {
 707                 struct btrfs_extent_inline_ref *iref;
 708                 u64 offset;
 709                 int type;
 710
 711                 iref = (struct btrfs_extent_inline_ref *)ptr;
 712                 type = btrfs_extent_inline_ref_type(leaf, iref);
 713                 offset = btrfs_extent_inline_ref_offset(leaf, iref);
 714
 715                 switch (type) {
 716                 case BTRFS_SHARED_BLOCK_REF_KEY:
 717                         ret = __add_prelim_ref(prefs, 0, NULL,
 718                                                 *info_level + 1, offset,
 719                                                 bytenr, 1, GFP_NOFS);
 720                         break;
 721                 case BTRFS_SHARED_DATA_REF_KEY: {
 722                         struct btrfs_shared_data_ref *sdref;
 723                         int count;
 724
 725                         sdref = (struct btrfs_shared_data_ref *)(iref + 1);
 726                         count = btrfs_shared_data_ref_count(leaf, sdref);
 727                         ret = __add_prelim_ref(prefs, 0, NULL, 0, offset,
 728                                                bytenr, count, GFP_NOFS);
 729                         break;
 730                 }
 731                 case BTRFS_TREE_BLOCK_REF_KEY:
 732                         ret = __add_prelim_ref(prefs, offset, NULL,
 733                                                *info_level + 1, 0,
 734                                                bytenr, 1, GFP_NOFS);
 735                         break;
 736                 case BTRFS_EXTENT_DATA_REF_KEY: {
 737                         struct btrfs_extent_data_ref *dref;
 738                         int count;
 739                         u64 root;
 740
 741                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
 742                         count = btrfs_extent_data_ref_count(leaf, dref);
 743                         key.objectid = btrfs_extent_data_ref_objectid(leaf,
 744                                                                       dref);
 745                         key.type = BTRFS_EXTENT_DATA_KEY;
 746                         key.offset = btrfs_extent_data_ref_offset(leaf, dref);
 747                         root = btrfs_extent_data_ref_root(leaf, dref);
 748                         ret = __add_prelim_ref(prefs, root, &key, 0, 0,
 749                                                bytenr, count, GFP_NOFS);
 750                         break;
 751                 }
 752                 default:
 753                         WARN_ON(1);
 754                 }
 755                 if (ret)
 756                         return ret;
 757                 ptr += btrfs_extent_inline_ref_size(type);
 758         }
 759
 760         return 0;
 761 }
 762
 763 /*
 764  * add all non-inline backrefs for bytenr to the list
 765  */
 766 static int __add_keyed_refs(struct btrfs_fs_info *fs_info,
 767                             struct btrfs_path *path, u64 bytenr,
 768                             int info_level, struct list_head *prefs)
 769 {
 770         struct btrfs_root *extent_root = fs_info->extent_root;
 771         int ret;
 772         int slot;
 773         struct extent_buffer *leaf;
 774         struct btrfs_key key;
 775
 776         while (1) {
 777                 ret = btrfs_next_item(extent_root, path);
 778                 if (ret < 0)
 779                         break;
 780                 if (ret) {
 781                         ret = 0;
 782                         break;
 783                 }
 784
 785                 slot = path->slots[0];
 786                 leaf = path->nodes[0];
 787                 btrfs_item_key_to_cpu(leaf, &key, slot);
 788
 789                 if (key.objectid != bytenr)
 790                         break;
 791                 if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
 792                         continue;
 793                 if (key.type > BTRFS_SHARED_DATA_REF_KEY)
 794                         break;
 795
 796                 switch (key.type) {
 797                 case BTRFS_SHARED_BLOCK_REF_KEY:
 798                         ret = __add_prelim_ref(prefs, 0, NULL,
 799                                                 info_level + 1, key.offset,
 800                                                 bytenr, 1, GFP_NOFS);
 801                         break;
 802                 case BTRFS_SHARED_DATA_REF_KEY: {
 803                         struct btrfs_shared_data_ref *sdref;
 804                         int count;
 805
 806                         sdref = btrfs_item_ptr(leaf, slot,
 807                                               struct btrfs_shared_data_ref);
 808                         count = btrfs_shared_data_ref_count(leaf, sdref);
 809                         ret = __add_prelim_ref(prefs, 0, NULL, 0, key.offset,
 810                                                 bytenr, count, GFP_NOFS);
 811                         break;
 812                 }
 813                 case BTRFS_TREE_BLOCK_REF_KEY:
 814                         ret = __add_prelim_ref(prefs, key.offset, NULL,
 815                                                info_level + 1, 0,
 816                                                bytenr, 1, GFP_NOFS);
 817                         break;
 818                 case BTRFS_EXTENT_DATA_REF_KEY: {
 819                         struct btrfs_extent_data_ref *dref;
 820                         int count;
 821                         u64 root;
 822
 823                         dref = btrfs_item_ptr(leaf, slot,
 824                                               struct btrfs_extent_data_ref);
 825                         count = btrfs_extent_data_ref_count(leaf, dref);
 826                         key.objectid = btrfs_extent_data_ref_objectid(leaf,
 827                                                                       dref);
 828                         key.type = BTRFS_EXTENT_DATA_KEY;
 829                         key.offset = btrfs_extent_data_ref_offset(leaf, dref);
 830                         root = btrfs_extent_data_ref_root(leaf, dref);
 831                         ret = __add_prelim_ref(prefs, root, &key, 0, 0,
 832                                                bytenr, count, GFP_NOFS);
 833                         break;
 834                 }
 835                 default:
 836                         WARN_ON(1);
 837                 }
 838                 if (ret)
 839                         return ret;
 840
 841         }
 842
 843         return ret;
 844 }
 845
 846 /*
 847  * this adds all existing backrefs (inline backrefs, backrefs and delayed
 848  * refs) for the given bytenr to the refs list, merges duplicates and resolves
 849  * indirect refs to their parent bytenr.
 850  * When roots are found, they're added to the roots list
 851  *
 852  * FIXME some caching might speed things up
 853  */
 854 static int find_parent_nodes(struct btrfs_trans_handle *trans,
 855                              struct btrfs_fs_info *fs_info, u64 bytenr,
 856                              u64 time_seq, struct ulist *refs,
 857                              struct ulist *roots, const u64 *extent_item_pos)
 858 {
 859         struct btrfs_key key;
 860         struct btrfs_path *path;
 861         struct btrfs_delayed_ref_root *delayed_refs = NULL;
 862         struct btrfs_delayed_ref_head *head;
 863         int info_level = 0;
 864         int ret;
 865         struct list_head prefs_delayed;
 866         struct list_head prefs;
 867         struct __prelim_ref *ref;
 868         struct extent_inode_elem *eie = NULL;
 869         u64 total_refs = 0;
 870
 871         INIT_LIST_HEAD(&prefs);
 872         INIT_LIST_HEAD(&prefs_delayed);
 873
 874         key.objectid = bytenr;
 875         key.offset = (u64)-1;
 876         if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
 877                 key.type = BTRFS_METADATA_ITEM_KEY;
 878         else
 879                 key.type = BTRFS_EXTENT_ITEM_KEY;
 880
 881         path = btrfs_alloc_path();
 882         if (!path)
 883                 return -ENOMEM;
 884         if (!trans) {
 885                 path->search_commit_root = 1;
 886                 path->skip_locking = 1;
 887         }
 888
 889         /*
 890          * grab both a lock on the path and a lock on the delayed ref head.
 891          * We need both to get a consistent picture of how the refs look
 892          * at a specified point in time
 893          */
 894 again:
 895         head = NULL;
 896
 897         ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
 898         if (ret < 0)
 899                 goto out;
 900         BUG_ON(ret == 0);
 901
 902 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
 903         if (trans && likely(trans->type != __TRANS_DUMMY)) {
 904 #else
 905         if (trans) {
 906 #endif
 907                 /*
 908                  * look if there are updates for this ref queued and lock the
 909                  * head
 910                  */
 911                 delayed_refs = &trans->transaction->delayed_refs;
 912                 spin_lock(&delayed_refs->lock);
 913                 head = btrfs_find_delayed_ref_head(trans, bytenr);
 914                 if (head) {
 915                         if (!mutex_trylock(&head->mutex)) {
 916                                 atomic_inc(&head->node.refs);
 917                                 spin_unlock(&delayed_refs->lock);
 918
 919                                 btrfs_release_path(path);
 920
 921                                 /*
 922                                  * Mutex was contended, block until it's
 923                                  * released and try again
 924                                  */
 925                                 mutex_lock(&head->mutex);
 926                                 mutex_unlock(&head->mutex);
 927                                 btrfs_put_delayed_ref(&head->node);
 928                                 goto again;
 929                         }
 930                         spin_unlock(&delayed_refs->lock);
 931                         ret = __add_delayed_refs(head, time_seq,
 932                                                  &prefs_delayed, &total_refs);
 933                         mutex_unlock(&head->mutex);
 934                         if (ret)
 935                                 goto out;
 936                 } else {
 937                         spin_unlock(&delayed_refs->lock);
 938                 }
 939         }
 940
 941         if (path->slots[0]) {
 942                 struct extent_buffer *leaf;
 943                 int slot;
 944
 945                 path->slots[0]--;
 946                 leaf = path->nodes[0];
 947                 slot = path->slots[0];
 948                 btrfs_item_key_to_cpu(leaf, &key, slot);
 949                 if (key.objectid == bytenr &&
 950                     (key.type == BTRFS_EXTENT_ITEM_KEY ||
 951                      key.type == BTRFS_METADATA_ITEM_KEY)) {
 952                         ret = __add_inline_refs(fs_info, path, bytenr,
 953                                                 &info_level, &prefs,
 954                                                 &total_refs);
 955                         if (ret)
 956                                 goto out;
 957                         ret = __add_keyed_refs(fs_info, path, bytenr,
 958                                                info_level, &prefs);
 959                         if (ret)
 960                                 goto out;
 961                 }
 962         }
 963         btrfs_release_path(path);
 964
 965         list_splice_init(&prefs_delayed, &prefs);
 966
 967         ret = __add_missing_keys(fs_info, &prefs);
 968         if (ret)
 969                 goto out;
 970
 971         __merge_refs(&prefs, 1);
 972
 973         ret = __resolve_indirect_refs(fs_info, path, time_seq, &prefs,
 974                                       extent_item_pos, total_refs);
 975         if (ret)
 976                 goto out;
 977
 978         __merge_refs(&prefs, 2);
 979
 980         while (!list_empty(&prefs)) {
 981                 ref = list_first_entry(&prefs, struct __prelim_ref, list);
 982                 WARN_ON(ref->count < 0);
 983                 if (roots && ref->count && ref->root_id && ref->parent == 0) {
 984                         /* no parent == root of tree */
 985                         ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS);
 986                         if (ret < 0)
 987                                 goto out;
 988                 }
 989                 if (ref->count && ref->parent) {
 990                         if (extent_item_pos && !ref->inode_list &&
 991                             ref->level == 0) {
 992                                 u32 bsz;
 993                                 struct extent_buffer *eb;
 994                                 bsz = btrfs_level_size(fs_info->extent_root,
 995                                                         ref->level);
 996                                 eb = read_tree_block(fs_info->extent_root,
 997                                                            ref->parent, bsz, 0);
 998                                 if (!eb || !extent_buffer_uptodate(eb)) {
 999                                         free_extent_buffer(eb);
1000                                         ret = -EIO;
1001                                         goto out;
1002                                 }
1003                                 btrfs_tree_read_lock(eb);
1004                                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1005                                 ret = find_extent_in_eb(eb, bytenr,
1006                                                         *extent_item_pos, &eie);
1007                                 btrfs_tree_read_unlock_blocking(eb);
1008                                 free_extent_buffer(eb);
1009                                 if (ret < 0)
1010                                         goto out;
1011                                 ref->inode_list = eie;
1012                         }
1013                         ret = ulist_add_merge_ptr(refs, ref->parent,
1014                                                   ref->inode_list,
1015                                                   (void **)&eie, GFP_NOFS);
1016                         if (ret < 0)
1017                                 goto out;
1018                         if (!ret && extent_item_pos) {
1019                                 /*
1020                                  * we've recorded that parent, so we must extend
1021                                  * its inode list here
1022                                  */
1023                                 BUG_ON(!eie);
1024                                 while (eie->next)
1025                                         eie = eie->next;
1026                                 eie->next = ref->inode_list;
1027                         }
1028                         eie = NULL;
1029                 }
1030                 list_del(&ref->list);
1031                 kmem_cache_free(btrfs_prelim_ref_cache, ref);
1032         }
1033
1034 out:
1035         btrfs_free_path(path);
1036         while (!list_empty(&prefs)) {
1037                 ref = list_first_entry(&prefs, struct __prelim_ref, list);
1038                 list_del(&ref->list);
1039                 kmem_cache_free(btrfs_prelim_ref_cache, ref);
1040         }
1041         while (!list_empty(&prefs_delayed)) {
1042                 ref = list_first_entry(&prefs_delayed, struct __prelim_ref,
1043                                        list);
1044                 list_del(&ref->list);
1045                 kmem_cache_free(btrfs_prelim_ref_cache, ref);
1046         }
1047         if (ret < 0)
1048                 free_inode_elem_list(eie);
1049         return ret;
1050 }
1051
1052 static void free_leaf_list(struct ulist *blocks)
1053 {
1054         struct ulist_node *node = NULL;
1055         struct extent_inode_elem *eie;
1056         struct ulist_iterator uiter;
1057
1058         ULIST_ITER_INIT(&uiter);
1059         while ((node = ulist_next(blocks, &uiter))) {
1060                 if (!node->aux)
1061                         continue;
1062                 eie = (struct extent_inode_elem *)(uintptr_t)node->aux;
1063                 free_inode_elem_list(eie);
1064                 node->aux = 0;
1065         }
1066
1067         ulist_free(blocks);
1068 }
1069
1070 /*
1071  * Finds all leafs with a reference to the specified combination of bytenr and
1072  * offset. key_list_head will point to a list of corresponding keys (caller must
1073  * free each list element). The leafs will be stored in the leafs ulist, which
1074  * must be freed with ulist_free.
1075  *
1076  * returns 0 on success, <0 on error
1077  */
1078 static int btrfs_find_all_leafs(struct btrfs_trans_handle *trans,
1079                                 struct btrfs_fs_info *fs_info, u64 bytenr,
1080                                 u64 time_seq, struct ulist **leafs,
1081                                 const u64 *extent_item_pos)
1082 {
1083         int ret;
1084
1085         *leafs = ulist_alloc(GFP_NOFS);
1086         if (!*leafs)
1087                 return -ENOMEM;
1088
1089         ret = find_parent_nodes(trans, fs_info, bytenr,
1090                                 time_seq, *leafs, NULL, extent_item_pos);
1091         if (ret < 0 && ret != -ENOENT) {
1092                 free_leaf_list(*leafs);
1093                 return ret;
1094         }
1095
1096         return 0;
1097 }
1098
1099 /*
1100  * walk all backrefs for a given extent to find all roots that reference this
1101  * extent. Walking a backref means finding all extents that reference this
1102  * extent and in turn walk the backrefs of those, too. Naturally this is a
1103  * recursive process, but here it is implemented in an iterative fashion: We
1104  * find all referencing extents for the extent in question and put them on a
1105  * list. In turn, we find all referencing extents for those, further appending
1106  * to the list. The way we iterate the list allows adding more elements after
1107  * the current while iterating. The process stops when we reach the end of the
1108  * list. Found roots are added to the roots list.
1109  *
1110  * returns 0 on success, < 0 on error.
1111  */
1112 static int __btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1113                                   struct btrfs_fs_info *fs_info, u64 bytenr,
1114                                   u64 time_seq, struct ulist **roots)
1115 {
1116         struct ulist *tmp;
1117         struct ulist_node *node = NULL;
1118         struct ulist_iterator uiter;
1119         int ret;
1120
1121         tmp = ulist_alloc(GFP_NOFS);
1122         if (!tmp)
1123                 return -ENOMEM;
1124         *roots = ulist_alloc(GFP_NOFS);
1125         if (!*roots) {
1126                 ulist_free(tmp);
1127                 return -ENOMEM;
1128         }
1129
1130         ULIST_ITER_INIT(&uiter);
1131         while (1) {
1132                 ret = find_parent_nodes(trans, fs_info, bytenr,
1133                                         time_seq, tmp, *roots, NULL);
1134                 if (ret < 0 && ret != -ENOENT) {
1135                         ulist_free(tmp);
1136                         ulist_free(*roots);
1137                         return ret;
1138                 }
1139                 node = ulist_next(tmp, &uiter);
1140                 if (!node)
1141                         break;
1142                 bytenr = node->val;
1143                 cond_resched();
1144         }
1145
1146         ulist_free(tmp);
1147         return 0;
1148 }
1149
1150 int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1151                          struct btrfs_fs_info *fs_info, u64 bytenr,
1152                          u64 time_seq, struct ulist **roots)
1153 {
1154         int ret;
1155
1156         if (!trans)
1157                 down_read(&fs_info->commit_root_sem);
1158         ret = __btrfs_find_all_roots(trans, fs_info, bytenr, time_seq, roots);
1159         if (!trans)
1160                 up_read(&fs_info->commit_root_sem);
1161         return ret;
1162 }
1163
1164 /*
1165  * this makes the path point to (inum INODE_ITEM ioff)
1166  */
1167 int inode_item_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
1168                         struct btrfs_path *path)
1169 {
1170         struct btrfs_key key;
1171         return btrfs_find_item(fs_root, path, inum, ioff,
1172                         BTRFS_INODE_ITEM_KEY, &key);
1173 }
1174
1175 static int inode_ref_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
1176                                 struct btrfs_path *path,
1177                                 struct btrfs_key *found_key)
1178 {
1179         return btrfs_find_item(fs_root, path, inum, ioff,
1180                         BTRFS_INODE_REF_KEY, found_key);
1181 }
1182
1183 int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,
1184                           u64 start_off, struct btrfs_path *path,
1185                           struct btrfs_inode_extref **ret_extref,
1186                           u64 *found_off)
1187 {
1188         int ret, slot;
1189         struct btrfs_key key;
1190         struct btrfs_key found_key;
1191         struct btrfs_inode_extref *extref;
1192         struct extent_buffer *leaf;
1193         unsigned long ptr;
1194
1195         key.objectid = inode_objectid;
1196         btrfs_set_key_type(&key, BTRFS_INODE_EXTREF_KEY);
1197         key.offset = start_off;
1198
1199         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1200         if (ret < 0)
1201                 return ret;
1202
1203         while (1) {
1204                 leaf = path->nodes[0];
1205                 slot = path->slots[0];
1206                 if (slot >= btrfs_header_nritems(leaf)) {
1207                         /*
1208                          * If the item at offset is not found,
1209                          * btrfs_search_slot will point us to the slot
1210                          * where it should be inserted. In our case
1211                          * that will be the slot directly before the
1212                          * next INODE_REF_KEY_V2 item. In the case
1213                          * that we're pointing to the last slot in a
1214                          * leaf, we must move one leaf over.
1215                          */
1216                         ret = btrfs_next_leaf(root, path);
1217                         if (ret) {
1218                                 if (ret >= 1)
1219                                         ret = -ENOENT;
1220                                 break;
1221                         }
1222                         continue;
1223                 }
1224
1225                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1226
1227                 /*
1228                  * Check that we're still looking at an extended ref key for
1229                  * this particular objectid. If we have different
1230                  * objectid or type then there are no more to be found
1231                  * in the tree and we can exit.
1232                  */
1233                 ret = -ENOENT;
1234                 if (found_key.objectid != inode_objectid)
1235                         break;
1236                 if (btrfs_key_type(&found_key) != BTRFS_INODE_EXTREF_KEY)
1237                         break;
1238
1239                 ret = 0;
1240                 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1241                 extref = (struct btrfs_inode_extref *)ptr;
1242                 *ret_extref = extref;
1243                 if (found_off)
1244                         *found_off = found_key.offset;
1245                 break;
1246         }
1247
1248         return ret;
1249 }
1250
1251 /*
1252  * this iterates to turn a name (from iref/extref) into a full filesystem path.
1253  * Elements of the path are separated by '/' and the path is guaranteed to be
1254  * 0-terminated. the path is only given within the current file system.
1255  * Therefore, it never starts with a '/'. the caller is responsible to provide
1256  * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
1257  * the start point of the resulting string is returned. this pointer is within
1258  * dest, normally.
1259  * in case the path buffer would overflow, the pointer is decremented further
1260  * as if output was written to the buffer, though no more output is actually
1261  * generated. that way, the caller can determine how much space would be
1262  * required for the path to fit into the buffer. in that case, the returned
1263  * value will be smaller than dest. callers must check this!
1264  */
1265 char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
1266                         u32 name_len, unsigned long name_off,
1267                         struct extent_buffer *eb_in, u64 parent,
1268                         char *dest, u32 size)
1269 {
1270         int slot;
1271         u64 next_inum;
1272         int ret;
1273         s64 bytes_left = ((s64)size) - 1;
1274         struct extent_buffer *eb = eb_in;
1275         struct btrfs_key found_key;
1276         int leave_spinning = path->leave_spinning;
1277         struct btrfs_inode_ref *iref;
1278
1279         if (bytes_left >= 0)
1280                 dest[bytes_left] = '\0';
1281
1282         path->leave_spinning = 1;
1283         while (1) {
1284                 bytes_left -= name_len;
1285                 if (bytes_left >= 0)
1286                         read_extent_buffer(eb, dest + bytes_left,
1287                                            name_off, name_len);
1288                 if (eb != eb_in) {
1289                         btrfs_tree_read_unlock_blocking(eb);
1290                         free_extent_buffer(eb);
1291                 }
1292                 ret = inode_ref_info(parent, 0, fs_root, path, &found_key);
1293                 if (ret > 0)
1294                         ret = -ENOENT;
1295                 if (ret)
1296                         break;
1297
1298                 next_inum = found_key.offset;
1299
1300                 /* regular exit ahead */
1301                 if (parent == next_inum)
1302                         break;
1303
1304                 slot = path->slots[0];
1305                 eb = path->nodes[0];
1306                 /* make sure we can use eb after releasing the path */
1307                 if (eb != eb_in) {
1308                         atomic_inc(&eb->refs);
1309                         btrfs_tree_read_lock(eb);
1310                         btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1311                 }
1312                 btrfs_release_path(path);
1313                 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1314
1315                 name_len = btrfs_inode_ref_name_len(eb, iref);
1316                 name_off = (unsigned long)(iref + 1);
1317
1318                 parent = next_inum;
1319                 --bytes_left;
1320                 if (bytes_left >= 0)
1321                         dest[bytes_left] = '/';
1322         }
1323
1324         btrfs_release_path(path);
1325         path->leave_spinning = leave_spinning;
1326
1327         if (ret)
1328                 return ERR_PTR(ret);
1329
1330         return dest + bytes_left;
1331 }
1332
1333 /*
1334  * this makes the path point to (logical EXTENT_ITEM *)
1335  * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
1336  * tree blocks and <0 on error.
1337  */
1338 int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
1339                         struct btrfs_path *path, struct btrfs_key *found_key,
1340                         u64 *flags_ret)
1341 {
1342         int ret;
1343         u64 flags;
1344         u64 size = 0;
1345         u32 item_size;
1346         struct extent_buffer *eb;
1347         struct btrfs_extent_item *ei;
1348         struct btrfs_key key;
1349
1350         if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1351                 key.type = BTRFS_METADATA_ITEM_KEY;
1352         else
1353                 key.type = BTRFS_EXTENT_ITEM_KEY;
1354         key.objectid = logical;
1355         key.offset = (u64)-1;
1356
1357         ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
1358         if (ret < 0)
1359                 return ret;
1360
1361         ret = btrfs_previous_extent_item(fs_info->extent_root, path, 0);
1362         if (ret) {
1363                 if (ret > 0)
1364                         ret = -ENOENT;
1365                 return ret;
1366         }
1367         btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]);
1368         if (found_key->type == BTRFS_METADATA_ITEM_KEY)
1369                 size = fs_info->extent_root->leafsize;
1370         else if (found_key->type == BTRFS_EXTENT_ITEM_KEY)
1371                 size = found_key->offset;
1372
1373         if (found_key->objectid > logical ||
1374             found_key->objectid + size <= logical) {
1375                 pr_debug("logical %llu is not within any extent\n", logical);
1376                 return -ENOENT;
1377         }
1378
1379         eb = path->nodes[0];
1380         item_size = btrfs_item_size_nr(eb, path->slots[0]);
1381         BUG_ON(item_size < sizeof(*ei));
1382
1383         ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
1384         flags = btrfs_extent_flags(eb, ei);
1385
1386         pr_debug("logical %llu is at position %llu within the extent (%llu "
1387                  "EXTENT_ITEM %llu) flags %#llx size %u\n",
1388                  logical, logical - found_key->objectid, found_key->objectid,
1389                  found_key->offset, flags, item_size);
1390
1391         WARN_ON(!flags_ret);
1392         if (flags_ret) {
1393                 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1394                         *flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK;
1395                 else if (flags & BTRFS_EXTENT_FLAG_DATA)
1396                         *flags_ret = BTRFS_EXTENT_FLAG_DATA;
1397                 else
1398                         BUG_ON(1);
1399                 return 0;
1400         }
1401
1402         return -EIO;
1403 }
1404
1405 /*
1406  * helper function to iterate extent inline refs. ptr must point to a 0 value
1407  * for the first call and may be modified. it is used to track state.
1408  * if more refs exist, 0 is returned and the next call to
1409  * __get_extent_inline_ref must pass the modified ptr parameter to get the
1410  * next ref. after the last ref was processed, 1 is returned.
1411  * returns <0 on error
1412  */
1413 static int __get_extent_inline_ref(unsigned long *ptr, struct extent_buffer *eb,
1414                                    struct btrfs_key *key,
1415                                    struct btrfs_extent_item *ei, u32 item_size,
1416                                    struct btrfs_extent_inline_ref **out_eiref,
1417                                    int *out_type)
1418 {
1419         unsigned long end;
1420         u64 flags;
1421         struct btrfs_tree_block_info *info;
1422
1423         if (!*ptr) {
1424                 /* first call */
1425                 flags = btrfs_extent_flags(eb, ei);
1426                 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1427                         if (key->type == BTRFS_METADATA_ITEM_KEY) {
1428                                 /* a skinny metadata extent */
1429                                 *out_eiref =
1430                                      (struct btrfs_extent_inline_ref *)(ei + 1);
1431                         } else {
1432                                 WARN_ON(key->type != BTRFS_EXTENT_ITEM_KEY);
1433                                 info = (struct btrfs_tree_block_info *)(ei + 1);
1434                                 *out_eiref =
1435                                    (struct btrfs_extent_inline_ref *)(info + 1);
1436                         }
1437                 } else {
1438                         *out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1);
1439                 }
1440                 *ptr = (unsigned long)*out_eiref;
1441                 if ((unsigned long)(*ptr) >= (unsigned long)ei + item_size)
1442                         return -ENOENT;
1443         }
1444
1445         end = (unsigned long)ei + item_size;
1446         *out_eiref = (struct btrfs_extent_inline_ref *)(*ptr);
1447         *out_type = btrfs_extent_inline_ref_type(eb, *out_eiref);
1448
1449         *ptr += btrfs_extent_inline_ref_size(*out_type);
1450         WARN_ON(*ptr > end);
1451         if (*ptr == end)
1452                 return 1; /* last */
1453
1454         return 0;
1455 }
1456
1457 /*
1458  * reads the tree block backref for an extent. tree level and root are returned
1459  * through out_level and out_root. ptr must point to a 0 value for the first
1460  * call and may be modified (see __get_extent_inline_ref comment).
1461  * returns 0 if data was provided, 1 if there was no more data to provide or
1462  * <0 on error.
1463  */
1464 int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
1465                             struct btrfs_key *key, struct btrfs_extent_item *ei,
1466                             u32 item_size, u64 *out_root, u8 *out_level)
1467 {
1468         int ret;
1469         int type;
1470         struct btrfs_tree_block_info *info;
1471         struct btrfs_extent_inline_ref *eiref;
1472
1473         if (*ptr == (unsigned long)-1)
1474                 return 1;
1475
1476         while (1) {
1477                 ret = __get_extent_inline_ref(ptr, eb, key, ei, item_size,
1478                                               &eiref, &type);
1479                 if (ret < 0)
1480                         return ret;
1481
1482                 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1483                     type == BTRFS_SHARED_BLOCK_REF_KEY)
1484                         break;
1485
1486                 if (ret == 1)
1487                         return 1;
1488         }
1489
1490         /* we can treat both ref types equally here */
1491         info = (struct btrfs_tree_block_info *)(ei + 1);
1492         *out_root = btrfs_extent_inline_ref_offset(eb, eiref);
1493         *out_level = btrfs_tree_block_level(eb, info);
1494
1495         if (ret == 1)
1496                 *ptr = (unsigned long)-1;
1497
1498         return 0;
1499 }
1500
1501 static int iterate_leaf_refs(struct extent_inode_elem *inode_list,
1502                                 u64 root, u64 extent_item_objectid,
1503                                 iterate_extent_inodes_t *iterate, void *ctx)
1504 {
1505         struct extent_inode_elem *eie;
1506         int ret = 0;
1507
1508         for (eie = inode_list; eie; eie = eie->next) {
1509                 pr_debug("ref for %llu resolved, key (%llu EXTEND_DATA %llu), "
1510                          "root %llu\n", extent_item_objectid,
1511                          eie->inum, eie->offset, root);
1512                 ret = iterate(eie->inum, eie->offset, root, ctx);
1513                 if (ret) {
1514                         pr_debug("stopping iteration for %llu due to ret=%d\n",
1515                                  extent_item_objectid, ret);
1516                         break;
1517                 }
1518         }
1519
1520         return ret;
1521 }
1522
1523 /*
1524  * calls iterate() for every inode that references the extent identified by
1525  * the given parameters.
1526  * when the iterator function returns a non-zero value, iteration stops.
1527  */
1528 int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
1529                                 u64 extent_item_objectid, u64 extent_item_pos,
1530                                 int search_commit_root,
1531                                 iterate_extent_inodes_t *iterate, void *ctx)
1532 {
1533         int ret;
1534         struct btrfs_trans_handle *trans = NULL;
1535         struct ulist *refs = NULL;
1536         struct ulist *roots = NULL;
1537         struct ulist_node *ref_node = NULL;
1538         struct ulist_node *root_node = NULL;
1539         struct seq_list tree_mod_seq_elem = {};
1540         struct ulist_iterator ref_uiter;
1541         struct ulist_iterator root_uiter;
1542
1543         pr_debug("resolving all inodes for extent %llu\n",
1544                         extent_item_objectid);
1545
1546         if (!search_commit_root) {
1547                 trans = btrfs_join_transaction(fs_info->extent_root);
1548                 if (IS_ERR(trans))
1549                         return PTR_ERR(trans);
1550                 btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1551         } else {
1552                 down_read(&fs_info->commit_root_sem);
1553         }
1554
1555         ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid,
1556                                    tree_mod_seq_elem.seq, &refs,
1557                                    &extent_item_pos);
1558         if (ret)
1559                 goto out;
1560
1561         ULIST_ITER_INIT(&ref_uiter);
1562         while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) {
1563                 ret = __btrfs_find_all_roots(trans, fs_info, ref_node->val,
1564                                              tree_mod_seq_elem.seq, &roots);
1565                 if (ret)
1566                         break;
1567                 ULIST_ITER_INIT(&root_uiter);
1568                 while (!ret && (root_node = ulist_next(roots, &root_uiter))) {
1569                         pr_debug("root %llu references leaf %llu, data list "
1570                                  "%#llx\n", root_node->val, ref_node->val,
1571                                  ref_node->aux);
1572                         ret = iterate_leaf_refs((struct extent_inode_elem *)
1573                                                 (uintptr_t)ref_node->aux,
1574                                                 root_node->val,
1575                                                 extent_item_objectid,
1576                                                 iterate, ctx);
1577                 }
1578                 ulist_free(roots);
1579         }
1580
1581         free_leaf_list(refs);
1582 out:
1583         if (!search_commit_root) {
1584                 btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1585                 btrfs_end_transaction(trans, fs_info->extent_root);
1586         } else {
1587                 up_read(&fs_info->commit_root_sem);
1588         }
1589
1590         return ret;
1591 }
1592
1593 int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
1594                                 struct btrfs_path *path,
1595                                 iterate_extent_inodes_t *iterate, void *ctx)
1596 {
1597         int ret;
1598         u64 extent_item_pos;
1599         u64 flags = 0;
1600         struct btrfs_key found_key;
1601         int search_commit_root = path->search_commit_root;
1602
1603         ret = extent_from_logical(fs_info, logical, path, &found_key, &flags);
1604         btrfs_release_path(path);
1605         if (ret < 0)
1606                 return ret;
1607         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1608                 return -EINVAL;
1609
1610         extent_item_pos = logical - found_key.objectid;
1611         ret = iterate_extent_inodes(fs_info, found_key.objectid,
1612                                         extent_item_pos, search_commit_root,
1613                                         iterate, ctx);
1614
1615         return ret;
1616 }
1617
1618 typedef int (iterate_irefs_t)(u64 parent, u32 name_len, unsigned long name_off,
1619                               struct extent_buffer *eb, void *ctx);
1620
1621 static int iterate_inode_refs(u64 inum, struct btrfs_root *fs_root,
1622                               struct btrfs_path *path,
1623                               iterate_irefs_t *iterate, void *ctx)
1624 {
1625         int ret = 0;
1626         int slot;
1627         u32 cur;
1628         u32 len;
1629         u32 name_len;
1630         u64 parent = 0;
1631         int found = 0;
1632         struct extent_buffer *eb;
1633         struct btrfs_item *item;
1634         struct btrfs_inode_ref *iref;
1635         struct btrfs_key found_key;
1636
1637         while (!ret) {
1638                 ret = inode_ref_info(inum, parent ? parent+1 : 0, fs_root, path,
1639                                      &found_key);
1640                 if (ret < 0)
1641                         break;
1642                 if (ret) {
1643                         ret = found ? 0 : -ENOENT;
1644                         break;
1645                 }
1646                 ++found;
1647
1648                 parent = found_key.offset;
1649                 slot = path->slots[0];
1650                 eb = btrfs_clone_extent_buffer(path->nodes[0]);
1651                 if (!eb) {
1652                         ret = -ENOMEM;
1653                         break;
1654                 }
1655                 extent_buffer_get(eb);
1656                 btrfs_tree_read_lock(eb);
1657                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1658                 btrfs_release_path(path);
1659
1660                 item = btrfs_item_nr(slot);
1661                 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1662
1663                 for (cur = 0; cur < btrfs_item_size(eb, item); cur += len) {
1664                         name_len = btrfs_inode_ref_name_len(eb, iref);
1665                         /* path must be released before calling iterate()! */
1666                         pr_debug("following ref at offset %u for inode %llu in "
1667                                  "tree %llu\n", cur, found_key.objectid,
1668                                  fs_root->objectid);
1669                         ret = iterate(parent, name_len,
1670                                       (unsigned long)(iref + 1), eb, ctx);
1671                         if (ret)
1672                                 break;
1673                         len = sizeof(*iref) + name_len;
1674                         iref = (struct btrfs_inode_ref *)((char *)iref + len);
1675                 }
1676                 btrfs_tree_read_unlock_blocking(eb);
1677                 free_extent_buffer(eb);
1678         }
1679
1680         btrfs_release_path(path);
1681
1682         return ret;
1683 }
1684
1685 static int iterate_inode_extrefs(u64 inum, struct btrfs_root *fs_root,
1686                                  struct btrfs_path *path,
1687                                  iterate_irefs_t *iterate, void *ctx)
1688 {
1689         int ret;
1690         int slot;
1691         u64 offset = 0;
1692         u64 parent;
1693         int found = 0;
1694         struct extent_buffer *eb;
1695         struct btrfs_inode_extref *extref;
1696         struct extent_buffer *leaf;
1697         u32 item_size;
1698         u32 cur_offset;
1699         unsigned long ptr;
1700
1701         while (1) {
1702                 ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref,
1703                                             &offset);
1704                 if (ret < 0)
1705                         break;
1706                 if (ret) {
1707                         ret = found ? 0 : -ENOENT;
1708                         break;
1709                 }
1710                 ++found;
1711
1712                 slot = path->slots[0];
1713                 eb = btrfs_clone_extent_buffer(path->nodes[0]);
1714                 if (!eb) {
1715                         ret = -ENOMEM;
1716                         break;
1717                 }
1718                 extent_buffer_get(eb);
1719
1720                 btrfs_tree_read_lock(eb);
1721                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1722                 btrfs_release_path(path);
1723
1724                 leaf = path->nodes[0];
1725                 item_size = btrfs_item_size_nr(leaf, slot);
1726                 ptr = btrfs_item_ptr_offset(leaf, slot);
1727                 cur_offset = 0;
1728
1729                 while (cur_offset < item_size) {
1730                         u32 name_len;
1731
1732                         extref = (struct btrfs_inode_extref *)(ptr + cur_offset);
1733                         parent = btrfs_inode_extref_parent(eb, extref);
1734                         name_len = btrfs_inode_extref_name_len(eb, extref);
1735                         ret = iterate(parent, name_len,
1736                                       (unsigned long)&extref->name, eb, ctx);
1737                         if (ret)
1738                                 break;
1739
1740                         cur_offset += btrfs_inode_extref_name_len(leaf, extref);
1741                         cur_offset += sizeof(*extref);
1742                 }
1743                 btrfs_tree_read_unlock_blocking(eb);
1744                 free_extent_buffer(eb);
1745
1746                 offset++;
1747         }
1748
1749         btrfs_release_path(path);
1750
1751         return ret;
1752 }
1753
1754 static int iterate_irefs(u64 inum, struct btrfs_root *fs_root,
1755                          struct btrfs_path *path, iterate_irefs_t *iterate,
1756                          void *ctx)
1757 {
1758         int ret;
1759         int found_refs = 0;
1760
1761         ret = iterate_inode_refs(inum, fs_root, path, iterate, ctx);
1762         if (!ret)
1763                 ++found_refs;
1764         else if (ret != -ENOENT)
1765                 return ret;
1766
1767         ret = iterate_inode_extrefs(inum, fs_root, path, iterate, ctx);
1768         if (ret == -ENOENT && found_refs)
1769                 return 0;
1770
1771         return ret;
1772 }
1773
1774 /*
1775  * returns 0 if the path could be dumped (probably truncated)
1776  * returns <0 in case of an error
1777  */
1778 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
1779                          struct extent_buffer *eb, void *ctx)
1780 {
1781         struct inode_fs_paths *ipath = ctx;
1782         char *fspath;
1783         char *fspath_min;
1784         int i = ipath->fspath->elem_cnt;
1785         const int s_ptr = sizeof(char *);
1786         u32 bytes_left;
1787
1788         bytes_left = ipath->fspath->bytes_left > s_ptr ?
1789                                         ipath->fspath->bytes_left - s_ptr : 0;
1790
1791         fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
1792         fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len,
1793                                    name_off, eb, inum, fspath_min, bytes_left);
1794         if (IS_ERR(fspath))
1795                 return PTR_ERR(fspath);
1796
1797         if (fspath > fspath_min) {
1798                 ipath->fspath->val[i] = (u64)(unsigned long)fspath;
1799                 ++ipath->fspath->elem_cnt;
1800                 ipath->fspath->bytes_left = fspath - fspath_min;
1801         } else {
1802                 ++ipath->fspath->elem_missed;
1803                 ipath->fspath->bytes_missing += fspath_min - fspath;
1804                 ipath->fspath->bytes_left = 0;
1805         }
1806
1807         return 0;
1808 }
1809
1810 /*
1811  * this dumps all file system paths to the inode into the ipath struct, provided
1812  * is has been created large enough. each path is zero-terminated and accessed
1813  * from ipath->fspath->val[i].
1814  * when it returns, there are ipath->fspath->elem_cnt number of paths available
1815  * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the
1816  * number of missed paths in recored in ipath->fspath->elem_missed, otherwise,
1817  * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
1818  * have been needed to return all paths.
1819  */
1820 int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
1821 {
1822         return iterate_irefs(inum, ipath->fs_root, ipath->btrfs_path,
1823                              inode_to_path, ipath);
1824 }
1825
1826 struct btrfs_data_container *init_data_container(u32 total_bytes)
1827 {
1828         struct btrfs_data_container *data;
1829         size_t alloc_bytes;
1830
1831         alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
1832         data = vmalloc(alloc_bytes);
1833         if (!data)
1834                 return ERR_PTR(-ENOMEM);
1835
1836         if (total_bytes >= sizeof(*data)) {
1837                 data->bytes_left = total_bytes - sizeof(*data);
1838                 data->bytes_missing = 0;
1839         } else {
1840                 data->bytes_missing = sizeof(*data) - total_bytes;
1841                 data->bytes_left = 0;
1842         }
1843
1844         data->elem_cnt = 0;
1845         data->elem_missed = 0;
1846
1847         return data;
1848 }
1849
1850 /*
1851  * allocates space to return multiple file system paths for an inode.
1852  * total_bytes to allocate are passed, note that space usable for actual path
1853  * information will be total_bytes - sizeof(struct inode_fs_paths).
1854  * the returned pointer must be freed with free_ipath() in the end.
1855  */
1856 struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
1857                                         struct btrfs_path *path)
1858 {
1859         struct inode_fs_paths *ifp;
1860         struct btrfs_data_container *fspath;
1861
1862         fspath = init_data_container(total_bytes);
1863         if (IS_ERR(fspath))
1864                 return (void *)fspath;
1865
1866         ifp = kmalloc(sizeof(*ifp), GFP_NOFS);
1867         if (!ifp) {
1868                 kfree(fspath);
1869                 return ERR_PTR(-ENOMEM);
1870         }
1871
1872         ifp->btrfs_path = path;
1873         ifp->fspath = fspath;
1874         ifp->fs_root = fs_root;
1875
1876         return ifp;
1877 }
1878
1879 void free_ipath(struct inode_fs_paths *ipath)
1880 {
1881         if (!ipath)
1882                 return;
1883         vfree(ipath->fspath);
1884         kfree(ipath);
1885 }