cmds-check.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  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 #define _XOPEN_SOURCE 500
  20 #define _GNU_SOURCE 1
  21 #include <stdio.h>
  22 #include <stdlib.h>
  23 #include <unistd.h>
  24 #include <fcntl.h>
  25 #include <sys/types.h>
  26 #include <sys/stat.h>
  27 #include <unistd.h>
  28 #include <getopt.h>
  29 #include <uuid/uuid.h>
  30 #include "kerncompat.h"
  31 #include "ctree.h"
  32 #include "volumes.h"
  33 #include "repair.h"
  34 #include "disk-io.h"
  35 #include "print-tree.h"
  36 #include "transaction.h"
  37 #include "list.h"
  38 #include "version.h"
  39 #include "utils.h"
  40 #include "commands.h"
  41 #include "free-space-cache.h"
  42
  43 static u64 bytes_used = 0;
  44 static u64 total_csum_bytes = 0;
  45 static u64 total_btree_bytes = 0;
  46 static u64 total_fs_tree_bytes = 0;
  47 static u64 total_extent_tree_bytes = 0;
  48 static u64 btree_space_waste = 0;
  49 static u64 data_bytes_allocated = 0;
  50 static u64 data_bytes_referenced = 0;
  51 static int found_old_backref = 0;
  52 static LIST_HEAD(duplicate_extents);
  53
  54 struct extent_backref {
  55         struct list_head list;
  56         unsigned int is_data:1;
  57         unsigned int found_extent_tree:1;
  58         unsigned int full_backref:1;
  59         unsigned int found_ref:1;
  60 };
  61
  62 struct data_backref {
  63         struct extent_backref node;
  64         union {
  65                 u64 parent;
  66                 u64 root;
  67         };
  68         u64 owner;
  69         u64 offset;
  70         u64 disk_bytenr;
  71         u64 bytes;
  72         u64 ram_bytes;
  73         u32 num_refs;
  74         u32 found_ref;
  75 };
  76
  77 struct tree_backref {
  78         struct extent_backref node;
  79         union {
  80                 u64 parent;
  81                 u64 root;
  82         };
  83 };
  84
  85 struct extent_record {
  86         struct list_head backrefs;
  87         struct list_head dups;
  88         struct list_head list;
  89         struct cache_extent cache;
  90         struct btrfs_disk_key parent_key;
  91         u64 start;
  92         u64 max_size;
  93         u64 nr;
  94         u64 refs;
  95         u64 extent_item_refs;
  96         u64 generation;
  97         u64 info_objectid;
  98         u8 info_level;
  99         unsigned int content_checked:1;
 100         unsigned int owner_ref_checked:1;
 101         unsigned int is_root:1;
 102         unsigned int metadata:1;
 103         unsigned int found_rec:1;
 104         unsigned int has_duplicate:1;
 105 };
 106
 107 struct inode_backref {
 108         struct list_head list;
 109         unsigned int found_dir_item:1;
 110         unsigned int found_dir_index:1;
 111         unsigned int found_inode_ref:1;
 112         unsigned int filetype:8;
 113         int errors;
 114         unsigned int ref_type;
 115         u64 dir;
 116         u64 index;
 117         u16 namelen;
 118         char name[0];
 119 };
 120
 121 #define REF_ERR_NO_DIR_ITEM             (1 << 0)
 122 #define REF_ERR_NO_DIR_INDEX            (1 << 1)
 123 #define REF_ERR_NO_INODE_REF            (1 << 2)
 124 #define REF_ERR_DUP_DIR_ITEM            (1 << 3)
 125 #define REF_ERR_DUP_DIR_INDEX           (1 << 4)
 126 #define REF_ERR_DUP_INODE_REF           (1 << 5)
 127 #define REF_ERR_INDEX_UNMATCH           (1 << 6)
 128 #define REF_ERR_FILETYPE_UNMATCH        (1 << 7)
 129 #define REF_ERR_NAME_TOO_LONG           (1 << 8) // 100
 130 #define REF_ERR_NO_ROOT_REF             (1 << 9)
 131 #define REF_ERR_NO_ROOT_BACKREF         (1 << 10)
 132 #define REF_ERR_DUP_ROOT_REF            (1 << 11)
 133 #define REF_ERR_DUP_ROOT_BACKREF        (1 << 12)
 134
 135 struct inode_record {
 136         struct list_head backrefs;
 137         unsigned int checked:1;
 138         unsigned int merging:1;
 139         unsigned int found_inode_item:1;
 140         unsigned int found_dir_item:1;
 141         unsigned int found_file_extent:1;
 142         unsigned int found_csum_item:1;
 143         unsigned int some_csum_missing:1;
 144         unsigned int nodatasum:1;
 145         int errors;
 146
 147         u64 ino;
 148         u32 nlink;
 149         u32 imode;
 150         u64 isize;
 151         u64 nbytes;
 152
 153         u32 found_link;
 154         u64 found_size;
 155         u64 extent_start;
 156         u64 extent_end;
 157         u64 first_extent_gap;
 158
 159         u32 refs;
 160 };
 161
 162 #define I_ERR_NO_INODE_ITEM             (1 << 0)
 163 #define I_ERR_NO_ORPHAN_ITEM            (1 << 1)
 164 #define I_ERR_DUP_INODE_ITEM            (1 << 2)
 165 #define I_ERR_DUP_DIR_INDEX             (1 << 3)
 166 #define I_ERR_ODD_DIR_ITEM              (1 << 4)
 167 #define I_ERR_ODD_FILE_EXTENT           (1 << 5)
 168 #define I_ERR_BAD_FILE_EXTENT           (1 << 6)
 169 #define I_ERR_FILE_EXTENT_OVERLAP       (1 << 7)
 170 #define I_ERR_FILE_EXTENT_DISCOUNT      (1 << 8) // 100
 171 #define I_ERR_DIR_ISIZE_WRONG           (1 << 9)
 172 #define I_ERR_FILE_NBYTES_WRONG         (1 << 10) // 400
 173 #define I_ERR_ODD_CSUM_ITEM             (1 << 11)
 174 #define I_ERR_SOME_CSUM_MISSING         (1 << 12)
 175 #define I_ERR_LINK_COUNT_WRONG          (1 << 13)
 176
 177 struct root_backref {
 178         struct list_head list;
 179         unsigned int found_dir_item:1;
 180         unsigned int found_dir_index:1;
 181         unsigned int found_back_ref:1;
 182         unsigned int found_forward_ref:1;
 183         unsigned int reachable:1;
 184         int errors;
 185         u64 ref_root;
 186         u64 dir;
 187         u64 index;
 188         u16 namelen;
 189         char name[0];
 190 };
 191
 192 struct root_record {
 193         struct list_head backrefs;
 194         struct cache_extent cache;
 195         unsigned int found_root_item:1;
 196         u64 objectid;
 197         u32 found_ref;
 198 };
 199
 200 struct ptr_node {
 201         struct cache_extent cache;
 202         void *data;
 203 };
 204
 205 struct shared_node {
 206         struct cache_extent cache;
 207         struct cache_tree root_cache;
 208         struct cache_tree inode_cache;
 209         struct inode_record *current;
 210         u32 refs;
 211 };
 212
 213 struct block_info {
 214         u64 start;
 215         u32 size;
 216 };
 217
 218 struct walk_control {
 219         struct cache_tree shared;
 220         struct shared_node *nodes[BTRFS_MAX_LEVEL];
 221         int active_node;
 222         int root_level;
 223 };
 224
 225 static u8 imode_to_type(u32 imode)
 226 {
 227 #define S_SHIFT 12
 228         static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
 229                 [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
 230                 [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
 231                 [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
 232                 [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
 233                 [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
 234                 [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
 235                 [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
 236         };
 237
 238         return btrfs_type_by_mode[(imode & S_IFMT) >> S_SHIFT];
 239 #undef S_SHIFT
 240 }
 241
 242 static struct inode_record *clone_inode_rec(struct inode_record *orig_rec)
 243 {
 244         struct inode_record *rec;
 245         struct inode_backref *backref;
 246         struct inode_backref *orig;
 247         size_t size;
 248
 249         rec = malloc(sizeof(*rec));
 250         memcpy(rec, orig_rec, sizeof(*rec));
 251         rec->refs = 1;
 252         INIT_LIST_HEAD(&rec->backrefs);
 253
 254         list_for_each_entry(orig, &orig_rec->backrefs, list) {
 255                 size = sizeof(*orig) + orig->namelen + 1;
 256                 backref = malloc(size);
 257                 memcpy(backref, orig, size);
 258                 list_add_tail(&backref->list, &rec->backrefs);
 259         }
 260         return rec;
 261 }
 262
 263 static struct inode_record *get_inode_rec(struct cache_tree *inode_cache,
 264                                           u64 ino, int mod)
 265 {
 266         struct ptr_node *node;
 267         struct cache_extent *cache;
 268         struct inode_record *rec = NULL;
 269         int ret;
 270
 271         cache = find_cache_extent(inode_cache, ino, 1);
 272         if (cache) {
 273                 node = container_of(cache, struct ptr_node, cache);
 274                 rec = node->data;
 275                 if (mod && rec->refs > 1) {
 276                         node->data = clone_inode_rec(rec);
 277                         rec->refs--;
 278                         rec = node->data;
 279                 }
 280         } else if (mod) {
 281                 rec = calloc(1, sizeof(*rec));
 282                 rec->ino = ino;
 283                 rec->extent_start = (u64)-1;
 284                 rec->first_extent_gap = (u64)-1;
 285                 rec->refs = 1;
 286                 INIT_LIST_HEAD(&rec->backrefs);
 287
 288                 node = malloc(sizeof(*node));
 289                 node->cache.start = ino;
 290                 node->cache.size = 1;
 291                 node->data = rec;
 292
 293                 if (ino == BTRFS_FREE_INO_OBJECTID)
 294                         rec->found_link = 1;
 295
 296                 ret = insert_existing_cache_extent(inode_cache, &node->cache);
 297                 BUG_ON(ret);
 298         }
 299         return rec;
 300 }
 301
 302 static void free_inode_rec(struct inode_record *rec)
 303 {
 304         struct inode_backref *backref;
 305
 306         if (--rec->refs > 0)
 307                 return;
 308
 309         while (!list_empty(&rec->backrefs)) {
 310                 backref = list_entry(rec->backrefs.next,
 311                                      struct inode_backref, list);
 312                 list_del(&backref->list);
 313                 free(backref);
 314         }
 315         free(rec);
 316 }
 317
 318 static int can_free_inode_rec(struct inode_record *rec)
 319 {
 320         if (!rec->errors && rec->checked && rec->found_inode_item &&
 321             rec->nlink == rec->found_link && list_empty(&rec->backrefs))
 322                 return 1;
 323         return 0;
 324 }
 325
 326 static void maybe_free_inode_rec(struct cache_tree *inode_cache,
 327                                  struct inode_record *rec)
 328 {
 329         struct cache_extent *cache;
 330         struct inode_backref *tmp, *backref;
 331         struct ptr_node *node;
 332         unsigned char filetype;
 333
 334         if (!rec->found_inode_item)
 335                 return;
 336
 337         filetype = imode_to_type(rec->imode);
 338         list_for_each_entry_safe(backref, tmp, &rec->backrefs, list) {
 339                 if (backref->found_dir_item && backref->found_dir_index) {
 340                         if (backref->filetype != filetype)
 341                                 backref->errors |= REF_ERR_FILETYPE_UNMATCH;
 342                         if (!backref->errors && backref->found_inode_ref) {
 343                                 list_del(&backref->list);
 344                                 free(backref);
 345                         }
 346                 }
 347         }
 348
 349         if (!rec->checked || rec->merging)
 350                 return;
 351
 352         if (S_ISDIR(rec->imode)) {
 353                 if (rec->found_size != rec->isize)
 354                         rec->errors |= I_ERR_DIR_ISIZE_WRONG;
 355                 if (rec->found_file_extent)
 356                         rec->errors |= I_ERR_ODD_FILE_EXTENT;
 357         } else if (S_ISREG(rec->imode) || S_ISLNK(rec->imode)) {
 358                 if (rec->found_dir_item)
 359                         rec->errors |= I_ERR_ODD_DIR_ITEM;
 360                 if (rec->found_size != rec->nbytes)
 361                         rec->errors |= I_ERR_FILE_NBYTES_WRONG;
 362                 if (rec->extent_start == (u64)-1 || rec->extent_start > 0)
 363                         rec->first_extent_gap = 0;
 364                 if (rec->nlink > 0 && (rec->extent_end < rec->isize ||
 365                     rec->first_extent_gap < rec->isize))
 366                         rec->errors |= I_ERR_FILE_EXTENT_DISCOUNT;
 367         }
 368
 369         if (S_ISREG(rec->imode) || S_ISLNK(rec->imode)) {
 370                 if (rec->found_csum_item && rec->nodatasum)
 371                         rec->errors |= I_ERR_ODD_CSUM_ITEM;
 372                 if (rec->some_csum_missing && !rec->nodatasum)
 373                         rec->errors |= I_ERR_SOME_CSUM_MISSING;
 374         }
 375
 376         BUG_ON(rec->refs != 1);
 377         if (can_free_inode_rec(rec)) {
 378                 cache = find_cache_extent(inode_cache, rec->ino, 1);
 379                 node = container_of(cache, struct ptr_node, cache);
 380                 BUG_ON(node->data != rec);
 381                 remove_cache_extent(inode_cache, &node->cache);
 382                 free(node);
 383                 free_inode_rec(rec);
 384         }
 385 }
 386
 387 static int check_orphan_item(struct btrfs_root *root, u64 ino)
 388 {
 389         struct btrfs_path path;
 390         struct btrfs_key key;
 391         int ret;
 392
 393         key.objectid = BTRFS_ORPHAN_OBJECTID;
 394         key.type = BTRFS_ORPHAN_ITEM_KEY;
 395         key.offset = ino;
 396
 397         btrfs_init_path(&path);
 398         ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
 399         btrfs_release_path(root, &path);
 400         if (ret > 0)
 401                 ret = -ENOENT;
 402         return ret;
 403 }
 404
 405 static int process_inode_item(struct extent_buffer *eb,
 406                               int slot, struct btrfs_key *key,
 407                               struct shared_node *active_node)
 408 {
 409         struct inode_record *rec;
 410         struct btrfs_inode_item *item;
 411
 412         rec = active_node->current;
 413         BUG_ON(rec->ino != key->objectid || rec->refs > 1);
 414         if (rec->found_inode_item) {
 415                 rec->errors |= I_ERR_DUP_INODE_ITEM;
 416                 return 1;
 417         }
 418         item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
 419         rec->nlink = btrfs_inode_nlink(eb, item);
 420         rec->isize = btrfs_inode_size(eb, item);
 421         rec->nbytes = btrfs_inode_nbytes(eb, item);
 422         rec->imode = btrfs_inode_mode(eb, item);
 423         if (btrfs_inode_flags(eb, item) & BTRFS_INODE_NODATASUM)
 424                 rec->nodatasum = 1;
 425         rec->found_inode_item = 1;
 426         if (rec->nlink == 0)
 427                 rec->errors |= I_ERR_NO_ORPHAN_ITEM;
 428         maybe_free_inode_rec(&active_node->inode_cache, rec);
 429         return 0;
 430 }
 431
 432 static struct inode_backref *get_inode_backref(struct inode_record *rec,
 433                                                 const char *name,
 434                                                 int namelen, u64 dir)
 435 {
 436         struct inode_backref *backref;
 437
 438         list_for_each_entry(backref, &rec->backrefs, list) {
 439                 if (backref->dir != dir || backref->namelen != namelen)
 440                         continue;
 441                 if (memcmp(name, backref->name, namelen))
 442                         continue;
 443                 return backref;
 444         }
 445
 446         backref = malloc(sizeof(*backref) + namelen + 1);
 447         memset(backref, 0, sizeof(*backref));
 448         backref->dir = dir;
 449         backref->namelen = namelen;
 450         memcpy(backref->name, name, namelen);
 451         backref->name[namelen] = '\0';
 452         list_add_tail(&backref->list, &rec->backrefs);
 453         return backref;
 454 }
 455
 456 static int add_inode_backref(struct cache_tree *inode_cache,
 457                              u64 ino, u64 dir, u64 index,
 458                              const char *name, int namelen,
 459                              int filetype, int itemtype, int errors)
 460 {
 461         struct inode_record *rec;
 462         struct inode_backref *backref;
 463
 464         rec = get_inode_rec(inode_cache, ino, 1);
 465         backref = get_inode_backref(rec, name, namelen, dir);
 466         if (errors)
 467                 backref->errors |= errors;
 468         if (itemtype == BTRFS_DIR_INDEX_KEY) {
 469                 if (backref->found_dir_index)
 470                         backref->errors |= REF_ERR_DUP_DIR_INDEX;
 471                 if (backref->found_inode_ref && backref->index != index)
 472                         backref->errors |= REF_ERR_INDEX_UNMATCH;
 473                 if (backref->found_dir_item && backref->filetype != filetype)
 474                         backref->errors |= REF_ERR_FILETYPE_UNMATCH;
 475
 476                 backref->index = index;
 477                 backref->filetype = filetype;
 478                 backref->found_dir_index = 1;
 479         } else if (itemtype == BTRFS_DIR_ITEM_KEY) {
 480                 rec->found_link++;
 481                 if (backref->found_dir_item)
 482                         backref->errors |= REF_ERR_DUP_DIR_ITEM;
 483                 if (backref->found_dir_index && backref->filetype != filetype)
 484                         backref->errors |= REF_ERR_FILETYPE_UNMATCH;
 485
 486                 backref->filetype = filetype;
 487                 backref->found_dir_item = 1;
 488         } else if ((itemtype == BTRFS_INODE_REF_KEY) ||
 489                    (itemtype == BTRFS_INODE_EXTREF_KEY)) {
 490                 if (backref->found_inode_ref)
 491                         backref->errors |= REF_ERR_DUP_INODE_REF;
 492                 if (backref->found_dir_index && backref->index != index)
 493                         backref->errors |= REF_ERR_INDEX_UNMATCH;
 494
 495                 backref->ref_type = itemtype;
 496                 backref->index = index;
 497                 backref->found_inode_ref = 1;
 498         } else {
 499                 BUG_ON(1);
 500         }
 501
 502         maybe_free_inode_rec(inode_cache, rec);
 503         return 0;
 504 }
 505
 506 static int merge_inode_recs(struct inode_record *src, struct inode_record *dst,
 507                             struct cache_tree *dst_cache)
 508 {
 509         struct inode_backref *backref;
 510         u32 dir_count = 0;
 511
 512         dst->merging = 1;
 513         list_for_each_entry(backref, &src->backrefs, list) {
 514                 if (backref->found_dir_index) {
 515                         add_inode_backref(dst_cache, dst->ino, backref->dir,
 516                                         backref->index, backref->name,
 517                                         backref->namelen, backref->filetype,
 518                                         BTRFS_DIR_INDEX_KEY, backref->errors);
 519                 }
 520                 if (backref->found_dir_item) {
 521                         dir_count++;
 522                         add_inode_backref(dst_cache, dst->ino,
 523                                         backref->dir, 0, backref->name,
 524                                         backref->namelen, backref->filetype,
 525                                         BTRFS_DIR_ITEM_KEY, backref->errors);
 526                 }
 527                 if (backref->found_inode_ref) {
 528                         add_inode_backref(dst_cache, dst->ino,
 529                                         backref->dir, backref->index,
 530                                         backref->name, backref->namelen, 0,
 531                                         backref->ref_type, backref->errors);
 532                 }
 533         }
 534
 535         if (src->found_dir_item)
 536                 dst->found_dir_item = 1;
 537         if (src->found_file_extent)
 538                 dst->found_file_extent = 1;
 539         if (src->found_csum_item)
 540                 dst->found_csum_item = 1;
 541         if (src->some_csum_missing)
 542                 dst->some_csum_missing = 1;
 543         if (dst->first_extent_gap > src->first_extent_gap)
 544                 dst->first_extent_gap = src->first_extent_gap;
 545
 546         BUG_ON(src->found_link < dir_count);
 547         dst->found_link += src->found_link - dir_count;
 548         dst->found_size += src->found_size;
 549         if (src->extent_start != (u64)-1) {
 550                 if (dst->extent_start == (u64)-1) {
 551                         dst->extent_start = src->extent_start;
 552                         dst->extent_end = src->extent_end;
 553                 } else {
 554                         if (dst->extent_end > src->extent_start)
 555                                 dst->errors |= I_ERR_FILE_EXTENT_OVERLAP;
 556                         else if (dst->extent_end < src->extent_start &&
 557                                  dst->extent_end < dst->first_extent_gap)
 558                                 dst->first_extent_gap = dst->extent_end;
 559                         if (dst->extent_end < src->extent_end)
 560                                 dst->extent_end = src->extent_end;
 561                 }
 562         }
 563
 564         dst->errors |= src->errors;
 565         if (src->found_inode_item) {
 566                 if (!dst->found_inode_item) {
 567                         dst->nlink = src->nlink;
 568                         dst->isize = src->isize;
 569                         dst->nbytes = src->nbytes;
 570                         dst->imode = src->imode;
 571                         dst->nodatasum = src->nodatasum;
 572                         dst->found_inode_item = 1;
 573                 } else {
 574                         dst->errors |= I_ERR_DUP_INODE_ITEM;
 575                 }
 576         }
 577         dst->merging = 0;
 578
 579         return 0;
 580 }
 581
 582 static int splice_shared_node(struct shared_node *src_node,
 583                               struct shared_node *dst_node)
 584 {
 585         struct cache_extent *cache;
 586         struct ptr_node *node, *ins;
 587         struct cache_tree *src, *dst;
 588         struct inode_record *rec, *conflict;
 589         u64 current_ino = 0;
 590         int splice = 0;
 591         int ret;
 592
 593         if (--src_node->refs == 0)
 594                 splice = 1;
 595         if (src_node->current)
 596                 current_ino = src_node->current->ino;
 597
 598         src = &src_node->root_cache;
 599         dst = &dst_node->root_cache;
 600 again:
 601         cache = find_first_cache_extent(src, 0);
 602         while (cache) {
 603                 node = container_of(cache, struct ptr_node, cache);
 604                 rec = node->data;
 605                 cache = next_cache_extent(cache);
 606
 607                 if (splice) {
 608                         remove_cache_extent(src, &node->cache);
 609                         ins = node;
 610                 } else {
 611                         ins = malloc(sizeof(*ins));
 612                         ins->cache.start = node->cache.start;
 613                         ins->cache.size = node->cache.size;
 614                         ins->data = rec;
 615                         rec->refs++;
 616                 }
 617                 ret = insert_existing_cache_extent(dst, &ins->cache);
 618                 if (ret == -EEXIST) {
 619                         conflict = get_inode_rec(dst, rec->ino, 1);
 620                         merge_inode_recs(rec, conflict, dst);
 621                         if (rec->checked) {
 622                                 conflict->checked = 1;
 623                                 if (dst_node->current == conflict)
 624                                         dst_node->current = NULL;
 625                         }
 626                         maybe_free_inode_rec(dst, conflict);
 627                         free_inode_rec(rec);
 628                         free(ins);
 629                 } else {
 630                         BUG_ON(ret);
 631                 }
 632         }
 633
 634         if (src == &src_node->root_cache) {
 635                 src = &src_node->inode_cache;
 636                 dst = &dst_node->inode_cache;
 637                 goto again;
 638         }
 639
 640         if (current_ino > 0 && (!dst_node->current ||
 641             current_ino > dst_node->current->ino)) {
 642                 if (dst_node->current) {
 643                         dst_node->current->checked = 1;
 644                         maybe_free_inode_rec(dst, dst_node->current);
 645                 }
 646                 dst_node->current = get_inode_rec(dst, current_ino, 1);
 647         }
 648         return 0;
 649 }
 650
 651 static void free_inode_recs(struct cache_tree *inode_cache)
 652 {
 653         struct cache_extent *cache;
 654         struct ptr_node *node;
 655         struct inode_record *rec;
 656
 657         while (1) {
 658                 cache = find_first_cache_extent(inode_cache, 0);
 659                 if (!cache)
 660                         break;
 661                 node = container_of(cache, struct ptr_node, cache);
 662                 rec = node->data;
 663                 remove_cache_extent(inode_cache, &node->cache);
 664                 free(node);
 665                 free_inode_rec(rec);
 666         }
 667 }
 668
 669 static struct shared_node *find_shared_node(struct cache_tree *shared,
 670                                             u64 bytenr)
 671 {
 672         struct cache_extent *cache;
 673         struct shared_node *node;
 674
 675         cache = find_cache_extent(shared, bytenr, 1);
 676         if (cache) {
 677                 node = container_of(cache, struct shared_node, cache);
 678                 return node;
 679         }
 680         return NULL;
 681 }
 682
 683 static int add_shared_node(struct cache_tree *shared, u64 bytenr, u32 refs)
 684 {
 685         int ret;
 686         struct shared_node *node;
 687
 688         node = calloc(1, sizeof(*node));
 689         node->cache.start = bytenr;
 690         node->cache.size = 1;
 691         cache_tree_init(&node->root_cache);
 692         cache_tree_init(&node->inode_cache);
 693         node->refs = refs;
 694
 695         ret = insert_existing_cache_extent(shared, &node->cache);
 696         BUG_ON(ret);
 697         return 0;
 698 }
 699
 700 static int enter_shared_node(struct btrfs_root *root, u64 bytenr, u32 refs,
 701                              struct walk_control *wc, int level)
 702 {
 703         struct shared_node *node;
 704         struct shared_node *dest;
 705
 706         if (level == wc->active_node)
 707                 return 0;
 708
 709         BUG_ON(wc->active_node <= level);
 710         node = find_shared_node(&wc->shared, bytenr);
 711         if (!node) {
 712                 add_shared_node(&wc->shared, bytenr, refs);
 713                 node = find_shared_node(&wc->shared, bytenr);
 714                 wc->nodes[level] = node;
 715                 wc->active_node = level;
 716                 return 0;
 717         }
 718
 719         if (wc->root_level == wc->active_node &&
 720             btrfs_root_refs(&root->root_item) == 0) {
 721                 if (--node->refs == 0) {
 722                         free_inode_recs(&node->root_cache);
 723                         free_inode_recs(&node->inode_cache);
 724                         remove_cache_extent(&wc->shared, &node->cache);
 725                         free(node);
 726                 }
 727                 return 1;
 728         }
 729
 730         dest = wc->nodes[wc->active_node];
 731         splice_shared_node(node, dest);
 732         if (node->refs == 0) {
 733                 remove_cache_extent(&wc->shared, &node->cache);
 734                 free(node);
 735         }
 736         return 1;
 737 }
 738
 739 static int leave_shared_node(struct btrfs_root *root,
 740                              struct walk_control *wc, int level)
 741 {
 742         struct shared_node *node;
 743         struct shared_node *dest;
 744         int i;
 745
 746         if (level == wc->root_level)
 747                 return 0;
 748
 749         for (i = level + 1; i < BTRFS_MAX_LEVEL; i++) {
 750                 if (wc->nodes[i])
 751                         break;
 752         }
 753         BUG_ON(i >= BTRFS_MAX_LEVEL);
 754
 755         node = wc->nodes[wc->active_node];
 756         wc->nodes[wc->active_node] = NULL;
 757         wc->active_node = i;
 758
 759         dest = wc->nodes[wc->active_node];
 760         if (wc->active_node < wc->root_level ||
 761             btrfs_root_refs(&root->root_item) > 0) {
 762                 BUG_ON(node->refs <= 1);
 763                 splice_shared_node(node, dest);
 764         } else {
 765                 BUG_ON(node->refs < 2);
 766                 node->refs--;
 767         }
 768         return 0;
 769 }
 770
 771 static int is_child_root(struct btrfs_root *root, u64 parent_root_id,
 772                          u64 child_root_id)
 773 {
 774         struct btrfs_path path;
 775         struct btrfs_key key;
 776         struct extent_buffer *leaf;
 777         int has_parent = 0;
 778         int ret;
 779
 780         btrfs_init_path(&path);
 781
 782         key.objectid = parent_root_id;
 783         key.type = BTRFS_ROOT_REF_KEY;
 784         key.offset = child_root_id;
 785         ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, &path,
 786                                 0, 0);
 787         BUG_ON(ret < 0);
 788         btrfs_release_path(root, &path);
 789         if (!ret)
 790                 return 1;
 791
 792         key.objectid = child_root_id;
 793         key.type = BTRFS_ROOT_BACKREF_KEY;
 794         key.offset = 0;
 795         ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, &path,
 796                                 0, 0);
 797         BUG_ON(ret <= 0);
 798
 799         while (1) {
 800                 leaf = path.nodes[0];
 801                 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
 802                         ret = btrfs_next_leaf(root->fs_info->tree_root, &path);
 803                         BUG_ON(ret < 0);
 804
 805                         if (ret > 0)
 806                                 break;
 807                         leaf = path.nodes[0];
 808                 }
 809
 810                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
 811                 if (key.objectid != child_root_id ||
 812                     key.type != BTRFS_ROOT_BACKREF_KEY)
 813                         break;
 814
 815                 has_parent = 1;
 816
 817                 if (key.offset == parent_root_id) {
 818                         btrfs_release_path(root, &path);
 819                         return 1;
 820                 }
 821
 822                 path.slots[0]++;
 823         }
 824
 825         btrfs_release_path(root, &path);
 826         return has_parent? 0 : -1;
 827 }
 828
 829 static int process_dir_item(struct btrfs_root *root,
 830                             struct extent_buffer *eb,
 831                             int slot, struct btrfs_key *key,
 832                             struct shared_node *active_node)
 833 {
 834         u32 total;
 835         u32 cur = 0;
 836         u32 len;
 837         u32 name_len;
 838         u32 data_len;
 839         int error;
 840         int nritems = 0;
 841         int filetype;
 842         struct btrfs_dir_item *di;
 843         struct inode_record *rec;
 844         struct cache_tree *root_cache;
 845         struct cache_tree *inode_cache;
 846         struct btrfs_key location;
 847         char namebuf[BTRFS_NAME_LEN];
 848
 849         root_cache = &active_node->root_cache;
 850         inode_cache = &active_node->inode_cache;
 851         rec = active_node->current;
 852         rec->found_dir_item = 1;
 853
 854         di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
 855         total = btrfs_item_size_nr(eb, slot);
 856         while (cur < total) {
 857                 nritems++;
 858                 btrfs_dir_item_key_to_cpu(eb, di, &location);
 859                 name_len = btrfs_dir_name_len(eb, di);
 860                 data_len = btrfs_dir_data_len(eb, di);
 861                 filetype = btrfs_dir_type(eb, di);
 862
 863                 rec->found_size += name_len;
 864                 if (name_len <= BTRFS_NAME_LEN) {
 865                         len = name_len;
 866                         error = 0;
 867                 } else {
 868                         len = BTRFS_NAME_LEN;
 869                         error = REF_ERR_NAME_TOO_LONG;
 870                 }
 871                 read_extent_buffer(eb, namebuf, (unsigned long)(di + 1), len);
 872
 873                 if (location.type == BTRFS_INODE_ITEM_KEY) {
 874                         add_inode_backref(inode_cache, location.objectid,
 875                                           key->objectid, key->offset, namebuf,
 876                                           len, filetype, key->type, error);
 877                 } else if (location.type == BTRFS_ROOT_ITEM_KEY) {
 878                         add_inode_backref(root_cache, location.objectid,
 879                                           key->objectid, key->offset,
 880                                           namebuf, len, filetype,
 881                                           key->type, error);
 882                 } else {
 883                         fprintf(stderr, "warning line %d\n", __LINE__);
 884                 }
 885
 886                 len = sizeof(*di) + name_len + data_len;
 887                 di = (struct btrfs_dir_item *)((char *)di + len);
 888                 cur += len;
 889         }
 890         if (key->type == BTRFS_DIR_INDEX_KEY && nritems > 1)
 891                 rec->errors |= I_ERR_DUP_DIR_INDEX;
 892
 893         return 0;
 894 }
 895
 896 static int process_inode_ref(struct extent_buffer *eb,
 897                              int slot, struct btrfs_key *key,
 898                              struct shared_node *active_node)
 899 {
 900         u32 total;
 901         u32 cur = 0;
 902         u32 len;
 903         u32 name_len;
 904         u64 index;
 905         int error;
 906         struct cache_tree *inode_cache;
 907         struct btrfs_inode_ref *ref;
 908         char namebuf[BTRFS_NAME_LEN];
 909
 910         inode_cache = &active_node->inode_cache;
 911
 912         ref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
 913         total = btrfs_item_size_nr(eb, slot);
 914         while (cur < total) {
 915                 name_len = btrfs_inode_ref_name_len(eb, ref);
 916                 index = btrfs_inode_ref_index(eb, ref);
 917                 if (name_len <= BTRFS_NAME_LEN) {
 918                         len = name_len;
 919                         error = 0;
 920                 } else {
 921                         len = BTRFS_NAME_LEN;
 922                         error = REF_ERR_NAME_TOO_LONG;
 923                 }
 924                 read_extent_buffer(eb, namebuf, (unsigned long)(ref + 1), len);
 925                 add_inode_backref(inode_cache, key->objectid, key->offset,
 926                                   index, namebuf, len, 0, key->type, error);
 927
 928                 len = sizeof(*ref) + name_len;
 929                 ref = (struct btrfs_inode_ref *)((char *)ref + len);
 930                 cur += len;
 931         }
 932         return 0;
 933 }
 934
 935 static int process_inode_extref(struct extent_buffer *eb,
 936                                 int slot, struct btrfs_key *key,
 937                                 struct shared_node *active_node)
 938 {
 939         u32 total;
 940         u32 cur = 0;
 941         u32 len;
 942         u32 name_len;
 943         u64 index;
 944         u64 parent;
 945         int error;
 946         struct cache_tree *inode_cache;
 947         struct btrfs_inode_extref *extref;
 948         char namebuf[BTRFS_NAME_LEN];
 949
 950         inode_cache = &active_node->inode_cache;
 951
 952         extref = btrfs_item_ptr(eb, slot, struct btrfs_inode_extref);
 953         total = btrfs_item_size_nr(eb, slot);
 954         while (cur < total) {
 955                 name_len = btrfs_inode_extref_name_len(eb, extref);
 956                 index = btrfs_inode_extref_index(eb, extref);
 957                 parent = btrfs_inode_extref_parent(eb, extref);
 958                 if (name_len <= BTRFS_NAME_LEN) {
 959                         len = name_len;
 960                         error = 0;
 961                 } else {
 962                         len = BTRFS_NAME_LEN;
 963                         error = REF_ERR_NAME_TOO_LONG;
 964                 }
 965                 read_extent_buffer(eb, namebuf,
 966                                    (unsigned long)(extref + 1), len);
 967                 add_inode_backref(inode_cache, key->objectid, parent,
 968                                   index, namebuf, len, 0, key->type, error);
 969
 970                 len = sizeof(*extref) + name_len;
 971                 extref = (struct btrfs_inode_extref *)((char *)extref + len);
 972                 cur += len;
 973         }
 974         return 0;
 975
 976 }
 977
 978 static u64 count_csum_range(struct btrfs_root *root, u64 start, u64 len)
 979 {
 980         struct btrfs_key key;
 981         struct btrfs_path path;
 982         struct extent_buffer *leaf;
 983         int ret ;
 984         size_t size;
 985         u64 found = 0;
 986         u64 csum_end;
 987         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
 988
 989         btrfs_init_path(&path);
 990
 991         key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
 992         key.offset = start;
 993         key.type = BTRFS_EXTENT_CSUM_KEY;
 994
 995         ret = btrfs_search_slot(NULL, root->fs_info->csum_root,
 996                                 &key, &path, 0, 0);
 997         BUG_ON(ret < 0);
 998         if (ret > 0 && path.slots[0] > 0) {
 999                 leaf = path.nodes[0];
1000                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0] - 1);
1001                 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
1002                     key.type == BTRFS_EXTENT_CSUM_KEY)
1003                         path.slots[0]--;
1004         }
1005
1006         while (len > 0) {
1007                 leaf = path.nodes[0];
1008                 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
1009                         ret = btrfs_next_leaf(root->fs_info->csum_root, &path);
1010                         BUG_ON(ret < 0);
1011                         if (ret > 0)
1012                                 break;
1013                         leaf = path.nodes[0];
1014                 }
1015
1016                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
1017                 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1018                     key.type != BTRFS_EXTENT_CSUM_KEY)
1019                         break;
1020
1021                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
1022                 if (key.offset >= start + len)
1023                         break;
1024
1025                 if (key.offset > start)
1026                         start = key.offset;
1027
1028                 size = btrfs_item_size_nr(leaf, path.slots[0]);
1029                 csum_end = key.offset + (size / csum_size) * root->sectorsize;
1030                 if (csum_end > start) {
1031                         size = min(csum_end - start, len);
1032                         len -= size;
1033                         start += size;
1034                         found += size;
1035                 }
1036
1037                 path.slots[0]++;
1038         }
1039         btrfs_release_path(root->fs_info->csum_root, &path);
1040         return found;
1041 }
1042
1043 static int process_file_extent(struct btrfs_root *root,
1044                                 struct extent_buffer *eb,
1045                                 int slot, struct btrfs_key *key,
1046                                 struct shared_node *active_node)
1047 {
1048         struct inode_record *rec;
1049         struct btrfs_file_extent_item *fi;
1050         u64 num_bytes = 0;
1051         u64 disk_bytenr = 0;
1052         u64 extent_offset = 0;
1053         u64 mask = root->sectorsize - 1;
1054         int extent_type;
1055
1056         rec = active_node->current;
1057         BUG_ON(rec->ino != key->objectid || rec->refs > 1);
1058         rec->found_file_extent = 1;
1059
1060         if (rec->extent_start == (u64)-1) {
1061                 rec->extent_start = key->offset;
1062                 rec->extent_end = key->offset;
1063         }
1064
1065         if (rec->extent_end > key->offset)
1066                 rec->errors |= I_ERR_FILE_EXTENT_OVERLAP;
1067         else if (rec->extent_end < key->offset &&
1068                  rec->extent_end < rec->first_extent_gap)
1069                 rec->first_extent_gap = rec->extent_end;
1070
1071         fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
1072         extent_type = btrfs_file_extent_type(eb, fi);
1073
1074         if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1075                 num_bytes = btrfs_file_extent_inline_len(eb, fi);
1076                 if (num_bytes == 0)
1077                         rec->errors |= I_ERR_BAD_FILE_EXTENT;
1078                 rec->found_size += num_bytes;
1079                 num_bytes = (num_bytes + mask) & ~mask;
1080         } else if (extent_type == BTRFS_FILE_EXTENT_REG ||
1081                    extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1082                 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
1083                 disk_bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1084                 extent_offset = btrfs_file_extent_offset(eb, fi);
1085                 if (num_bytes == 0 || (num_bytes & mask))
1086                         rec->errors |= I_ERR_BAD_FILE_EXTENT;
1087                 if (num_bytes + extent_offset >
1088                     btrfs_file_extent_ram_bytes(eb, fi))
1089                         rec->errors |= I_ERR_BAD_FILE_EXTENT;
1090                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC &&
1091                     (btrfs_file_extent_compression(eb, fi) ||
1092                      btrfs_file_extent_encryption(eb, fi) ||
1093                      btrfs_file_extent_other_encoding(eb, fi)))
1094                         rec->errors |= I_ERR_BAD_FILE_EXTENT;
1095                 if (disk_bytenr > 0)
1096                         rec->found_size += num_bytes;
1097         } else {
1098                 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1099         }
1100         rec->extent_end = key->offset + num_bytes;
1101
1102         if (disk_bytenr > 0) {
1103                 u64 found;
1104                 if (btrfs_file_extent_compression(eb, fi))
1105                         num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1106                 else
1107                         disk_bytenr += extent_offset;
1108
1109                 found = count_csum_range(root, disk_bytenr, num_bytes);
1110                 if (extent_type == BTRFS_FILE_EXTENT_REG) {
1111                         if (found > 0)
1112                                 rec->found_csum_item = 1;
1113                         if (found < num_bytes)
1114                                 rec->some_csum_missing = 1;
1115                 } else if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1116                         if (found > 0)
1117                                 rec->errors |= I_ERR_ODD_CSUM_ITEM;
1118                 }
1119         }
1120         return 0;
1121 }
1122
1123 static int process_one_leaf(struct btrfs_root *root, struct extent_buffer *eb,
1124                             struct walk_control *wc)
1125 {
1126         struct btrfs_key key;
1127         u32 nritems;
1128         int i;
1129         int ret = 0;
1130         struct cache_tree *inode_cache;
1131         struct shared_node *active_node;
1132
1133         if (wc->root_level == wc->active_node &&
1134             btrfs_root_refs(&root->root_item) == 0)
1135                 return 0;
1136
1137         active_node = wc->nodes[wc->active_node];
1138         inode_cache = &active_node->inode_cache;
1139         nritems = btrfs_header_nritems(eb);
1140         for (i = 0; i < nritems; i++) {
1141                 btrfs_item_key_to_cpu(eb, &key, i);
1142
1143                 if (key.objectid == BTRFS_FREE_SPACE_OBJECTID)
1144                         continue;
1145
1146                 if (active_node->current == NULL ||
1147                     active_node->current->ino < key.objectid) {
1148                         if (active_node->current) {
1149                                 active_node->current->checked = 1;
1150                                 maybe_free_inode_rec(inode_cache,
1151                                                      active_node->current);
1152                         }
1153                         active_node->current = get_inode_rec(inode_cache,
1154                                                              key.objectid, 1);
1155                 }
1156                 switch (key.type) {
1157                 case BTRFS_DIR_ITEM_KEY:
1158                 case BTRFS_DIR_INDEX_KEY:
1159                         ret = process_dir_item(root, eb, i, &key, active_node);
1160                         break;
1161                 case BTRFS_INODE_REF_KEY:
1162                         ret = process_inode_ref(eb, i, &key, active_node);
1163                         break;
1164                 case BTRFS_INODE_EXTREF_KEY:
1165                         ret = process_inode_extref(eb, i, &key, active_node);
1166                         break;
1167                 case BTRFS_INODE_ITEM_KEY:
1168                         ret = process_inode_item(eb, i, &key, active_node);
1169                         break;
1170                 case BTRFS_EXTENT_DATA_KEY:
1171                         ret = process_file_extent(root, eb, i, &key,
1172                                                   active_node);
1173                         break;
1174                 default:
1175                         break;
1176                 };
1177         }
1178         return ret;
1179 }
1180
1181 static void reada_walk_down(struct btrfs_root *root,
1182                             struct extent_buffer *node, int slot)
1183 {
1184         u64 bytenr;
1185         u64 ptr_gen;
1186         u32 nritems;
1187         u32 blocksize;
1188         int i;
1189         int ret;
1190         int level;
1191
1192         level = btrfs_header_level(node);
1193         if (level != 1)
1194                 return;
1195
1196         nritems = btrfs_header_nritems(node);
1197         blocksize = btrfs_level_size(root, level - 1);
1198         for (i = slot; i < nritems; i++) {
1199                 bytenr = btrfs_node_blockptr(node, i);
1200                 ptr_gen = btrfs_node_ptr_generation(node, i);
1201                 ret = readahead_tree_block(root, bytenr, blocksize, ptr_gen);
1202                 if (ret)
1203                         break;
1204         }
1205 }
1206
1207 static int walk_down_tree(struct btrfs_root *root, struct btrfs_path *path,
1208                           struct walk_control *wc, int *level)
1209 {
1210         u64 bytenr;
1211         u64 ptr_gen;
1212         struct extent_buffer *next;
1213         struct extent_buffer *cur;
1214         u32 blocksize;
1215         int ret;
1216         u64 refs;
1217
1218         WARN_ON(*level < 0);
1219         WARN_ON(*level >= BTRFS_MAX_LEVEL);
1220         ret = btrfs_lookup_extent_info(NULL, root,
1221                                        path->nodes[*level]->start,
1222                                        *level, 1, &refs, NULL);
1223         if (ret < 0)
1224                 goto out;
1225
1226         if (refs > 1) {
1227                 ret = enter_shared_node(root, path->nodes[*level]->start,
1228                                         refs, wc, *level);
1229                 if (ret > 0)
1230                         goto out;
1231         }
1232
1233         while (*level >= 0) {
1234                 WARN_ON(*level < 0);
1235                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1236                 cur = path->nodes[*level];
1237
1238                 if (btrfs_header_level(cur) != *level)
1239                         WARN_ON(1);
1240
1241                 if (path->slots[*level] >= btrfs_header_nritems(cur))
1242                         break;
1243                 if (*level == 0) {
1244                         ret = process_one_leaf(root, cur, wc);
1245                         break;
1246                 }
1247                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1248                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
1249                 blocksize = btrfs_level_size(root, *level - 1);
1250                 ret = btrfs_lookup_extent_info(NULL, root, bytenr, *level - 1,
1251                                                1, &refs, NULL);
1252                 if (ret < 0)
1253                         refs = 0;
1254
1255                 if (refs > 1) {
1256                         ret = enter_shared_node(root, bytenr, refs,
1257                                                 wc, *level - 1);
1258                         if (ret > 0) {
1259                                 path->slots[*level]++;
1260                                 continue;
1261                         }
1262                 }
1263
1264                 next = btrfs_find_tree_block(root, bytenr, blocksize);
1265                 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
1266                         free_extent_buffer(next);
1267                         reada_walk_down(root, cur, path->slots[*level]);
1268                         next = read_tree_block(root, bytenr, blocksize,
1269                                                ptr_gen);
1270                 }
1271
1272                 *level = *level - 1;
1273                 free_extent_buffer(path->nodes[*level]);
1274                 path->nodes[*level] = next;
1275                 path->slots[*level] = 0;
1276         }
1277 out:
1278         path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
1279         return 0;
1280 }
1281
1282 static int walk_up_tree(struct btrfs_root *root, struct btrfs_path *path,
1283                         struct walk_control *wc, int *level)
1284 {
1285         int i;
1286         struct extent_buffer *leaf;
1287
1288         for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
1289                 leaf = path->nodes[i];
1290                 if (path->slots[i] + 1 < btrfs_header_nritems(leaf)) {
1291                         path->slots[i]++;
1292                         *level = i;
1293                         return 0;
1294                 } else {
1295                         free_extent_buffer(path->nodes[*level]);
1296                         path->nodes[*level] = NULL;
1297                         BUG_ON(*level > wc->active_node);
1298                         if (*level == wc->active_node)
1299                                 leave_shared_node(root, wc, *level);
1300                         *level = i + 1;
1301                 }
1302         }
1303         return 1;
1304 }
1305
1306 static int check_root_dir(struct inode_record *rec)
1307 {
1308         struct inode_backref *backref;
1309         int ret = -1;
1310
1311         if (!rec->found_inode_item || rec->errors)
1312                 goto out;
1313         if (rec->nlink != 1 || rec->found_link != 0)
1314                 goto out;
1315         if (list_empty(&rec->backrefs))
1316                 goto out;
1317         backref = list_entry(rec->backrefs.next, struct inode_backref, list);
1318         if (!backref->found_inode_ref)
1319                 goto out;
1320         if (backref->index != 0 || backref->namelen != 2 ||
1321             memcmp(backref->name, "..", 2))
1322                 goto out;
1323         if (backref->found_dir_index || backref->found_dir_item)
1324                 goto out;
1325         ret = 0;
1326 out:
1327         return ret;
1328 }
1329
1330 static int check_inode_recs(struct btrfs_root *root,
1331                             struct cache_tree *inode_cache)
1332 {
1333         struct cache_extent *cache;
1334         struct ptr_node *node;
1335         struct inode_record *rec;
1336         struct inode_backref *backref;
1337         int ret;
1338         u64 error = 0;
1339         u64 root_dirid = btrfs_root_dirid(&root->root_item);
1340
1341         if (btrfs_root_refs(&root->root_item) == 0) {
1342                 if (!cache_tree_empty(inode_cache))
1343                         fprintf(stderr, "warning line %d\n", __LINE__);
1344                 return 0;
1345         }
1346
1347         rec = get_inode_rec(inode_cache, root_dirid, 0);
1348         if (rec) {
1349                 ret = check_root_dir(rec);
1350                 if (ret) {
1351                         fprintf(stderr, "root %llu root dir %llu error\n",
1352                                 (unsigned long long)root->root_key.objectid,
1353                                 (unsigned long long)root_dirid);
1354                         error++;
1355                 }
1356         } else {
1357                 fprintf(stderr, "root %llu root dir %llu not found\n",
1358                         (unsigned long long)root->root_key.objectid,
1359                         (unsigned long long)root_dirid);
1360         }
1361
1362         while (1) {
1363                 cache = find_first_cache_extent(inode_cache, 0);
1364                 if (!cache)
1365                         break;
1366                 node = container_of(cache, struct ptr_node, cache);
1367                 rec = node->data;
1368                 remove_cache_extent(inode_cache, &node->cache);
1369                 free(node);
1370                 if (rec->ino == root_dirid ||
1371                     rec->ino == BTRFS_ORPHAN_OBJECTID) {
1372                         free_inode_rec(rec);
1373                         continue;
1374                 }
1375
1376                 if (rec->errors & I_ERR_NO_ORPHAN_ITEM) {
1377                         ret = check_orphan_item(root, rec->ino);
1378                         if (ret == 0)
1379                                 rec->errors &= ~I_ERR_NO_ORPHAN_ITEM;
1380                         if (can_free_inode_rec(rec)) {
1381                                 free_inode_rec(rec);
1382                                 continue;
1383                         }
1384                 }
1385
1386                 error++;
1387                 if (!rec->found_inode_item)
1388                         rec->errors |= I_ERR_NO_INODE_ITEM;
1389                 if (rec->found_link != rec->nlink)
1390                         rec->errors |= I_ERR_LINK_COUNT_WRONG;
1391                 fprintf(stderr, "root %llu inode %llu errors %x\n",
1392                         (unsigned long long) root->root_key.objectid,
1393                         (unsigned long long) rec->ino, rec->errors);
1394                 list_for_each_entry(backref, &rec->backrefs, list) {
1395                         if (!backref->found_dir_item)
1396                                 backref->errors |= REF_ERR_NO_DIR_ITEM;
1397                         if (!backref->found_dir_index)
1398                                 backref->errors |= REF_ERR_NO_DIR_INDEX;
1399                         if (!backref->found_inode_ref)
1400                                 backref->errors |= REF_ERR_NO_INODE_REF;
1401                         fprintf(stderr, "\tunresolved ref dir %llu index %llu"
1402                                 " namelen %u name %s filetype %d error %x\n",
1403                                 (unsigned long long)backref->dir,
1404                                 (unsigned long long)backref->index,
1405                                 backref->namelen, backref->name,
1406                                 backref->filetype, backref->errors);
1407                 }
1408                 free_inode_rec(rec);
1409         }
1410         return (error > 0) ? -1 : 0;
1411 }
1412
1413 static struct root_record *get_root_rec(struct cache_tree *root_cache,
1414                                         u64 objectid)
1415 {
1416         struct cache_extent *cache;
1417         struct root_record *rec = NULL;
1418         int ret;
1419
1420         cache = find_cache_extent(root_cache, objectid, 1);
1421         if (cache) {
1422                 rec = container_of(cache, struct root_record, cache);
1423         } else {
1424                 rec = calloc(1, sizeof(*rec));
1425                 rec->objectid = objectid;
1426                 INIT_LIST_HEAD(&rec->backrefs);
1427                 rec->cache.start = objectid;
1428                 rec->cache.size = 1;
1429
1430                 ret = insert_existing_cache_extent(root_cache, &rec->cache);
1431                 BUG_ON(ret);
1432         }
1433         return rec;
1434 }
1435
1436 static struct root_backref *get_root_backref(struct root_record *rec,
1437                                              u64 ref_root, u64 dir, u64 index,
1438                                              const char *name, int namelen)
1439 {
1440         struct root_backref *backref;
1441
1442         list_for_each_entry(backref, &rec->backrefs, list) {
1443                 if (backref->ref_root != ref_root || backref->dir != dir ||
1444                     backref->namelen != namelen)
1445                         continue;
1446                 if (memcmp(name, backref->name, namelen))
1447                         continue;
1448                 return backref;
1449         }
1450
1451         backref = malloc(sizeof(*backref) + namelen + 1);
1452         memset(backref, 0, sizeof(*backref));
1453         backref->ref_root = ref_root;
1454         backref->dir = dir;
1455         backref->index = index;
1456         backref->namelen = namelen;
1457         memcpy(backref->name, name, namelen);
1458         backref->name[namelen] = '\0';
1459         list_add_tail(&backref->list, &rec->backrefs);
1460         return backref;
1461 }
1462
1463 static void free_root_recs(struct cache_tree *root_cache)
1464 {
1465         struct cache_extent *cache;
1466         struct root_record *rec;
1467         struct root_backref *backref;
1468
1469         while (1) {
1470                 cache = find_first_cache_extent(root_cache, 0);
1471                 if (!cache)
1472                         break;
1473                 rec = container_of(cache, struct root_record, cache);
1474                 remove_cache_extent(root_cache, &rec->cache);
1475
1476                 while (!list_empty(&rec->backrefs)) {
1477                         backref = list_entry(rec->backrefs.next,
1478                                              struct root_backref, list);
1479                         list_del(&backref->list);
1480                         free(backref);
1481                 }
1482                 kfree(rec);
1483         }
1484 }
1485
1486 static int add_root_backref(struct cache_tree *root_cache,
1487                             u64 root_id, u64 ref_root, u64 dir, u64 index,
1488                             const char *name, int namelen,
1489                             int item_type, int errors)
1490 {
1491         struct root_record *rec;
1492         struct root_backref *backref;
1493
1494         rec = get_root_rec(root_cache, root_id);
1495         backref = get_root_backref(rec, ref_root, dir, index, name, namelen);
1496
1497         backref->errors |= errors;
1498
1499         if (item_type != BTRFS_DIR_ITEM_KEY) {
1500                 if (backref->found_dir_index || backref->found_back_ref ||
1501                     backref->found_forward_ref) {
1502                         if (backref->index != index)
1503                                 backref->errors |= REF_ERR_INDEX_UNMATCH;
1504                 } else {
1505                         backref->index = index;
1506                 }
1507         }
1508
1509         if (item_type == BTRFS_DIR_ITEM_KEY) {
1510                 backref->found_dir_item = 1;
1511                 backref->reachable = 1;
1512                 rec->found_ref++;
1513         } else if (item_type == BTRFS_DIR_INDEX_KEY) {
1514                 backref->found_dir_index = 1;
1515         } else if (item_type == BTRFS_ROOT_REF_KEY) {
1516                 if (backref->found_forward_ref)
1517                         backref->errors |= REF_ERR_DUP_ROOT_REF;
1518                 backref->found_forward_ref = 1;
1519         } else if (item_type == BTRFS_ROOT_BACKREF_KEY) {
1520                 if (backref->found_back_ref)
1521                         backref->errors |= REF_ERR_DUP_ROOT_BACKREF;
1522                 backref->found_back_ref = 1;
1523         } else {
1524                 BUG_ON(1);
1525         }
1526
1527         return 0;
1528 }
1529
1530 static int merge_root_recs(struct btrfs_root *root,
1531                            struct cache_tree *src_cache,
1532                            struct cache_tree *dst_cache)
1533 {
1534         struct cache_extent *cache;
1535         struct ptr_node *node;
1536         struct inode_record *rec;
1537         struct inode_backref *backref;
1538
1539         if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
1540                 free_inode_recs(src_cache);
1541                 return 0;
1542         }
1543
1544         while (1) {
1545                 cache = find_first_cache_extent(src_cache, 0);
1546                 if (!cache)
1547                         break;
1548                 node = container_of(cache, struct ptr_node, cache);
1549                 rec = node->data;
1550                 remove_cache_extent(src_cache, &node->cache);
1551                 free(node);
1552
1553                 if (!is_child_root(root, root->objectid, rec->ino))
1554                         goto skip;
1555
1556                 list_for_each_entry(backref, &rec->backrefs, list) {
1557                         BUG_ON(backref->found_inode_ref);
1558                         if (backref->found_dir_item)
1559                                 add_root_backref(dst_cache, rec->ino,
1560                                         root->root_key.objectid, backref->dir,
1561                                         backref->index, backref->name,
1562                                         backref->namelen, BTRFS_DIR_ITEM_KEY,
1563                                         backref->errors);
1564                         if (backref->found_dir_index)
1565                                 add_root_backref(dst_cache, rec->ino,
1566                                         root->root_key.objectid, backref->dir,
1567                                         backref->index, backref->name,
1568                                         backref->namelen, BTRFS_DIR_INDEX_KEY,
1569                                         backref->errors);
1570                 }
1571 skip:
1572                 free_inode_rec(rec);
1573         }
1574         return 0;
1575 }
1576
1577 static int check_root_refs(struct btrfs_root *root,
1578                            struct cache_tree *root_cache)
1579 {
1580         struct root_record *rec;
1581         struct root_record *ref_root;
1582         struct root_backref *backref;
1583         struct cache_extent *cache;
1584         int loop = 1;
1585         int ret;
1586         int error;
1587         int errors = 0;
1588
1589         rec = get_root_rec(root_cache, BTRFS_FS_TREE_OBJECTID);
1590         rec->found_ref = 1;
1591
1592         /* fixme: this can not detect circular references */
1593         while (loop) {
1594                 loop = 0;
1595                 cache = find_first_cache_extent(root_cache, 0);
1596                 while (1) {
1597                         if (!cache)
1598                                 break;
1599                         rec = container_of(cache, struct root_record, cache);
1600                         cache = next_cache_extent(cache);
1601
1602                         if (rec->found_ref == 0)
1603                                 continue;
1604
1605                         list_for_each_entry(backref, &rec->backrefs, list) {
1606                                 if (!backref->reachable)
1607                                         continue;
1608
1609                                 ref_root = get_root_rec(root_cache,
1610                                                         backref->ref_root);
1611                                 if (ref_root->found_ref > 0)
1612                                         continue;
1613
1614                                 backref->reachable = 0;
1615                                 rec->found_ref--;
1616                                 if (rec->found_ref == 0)
1617                                         loop = 1;
1618                         }
1619                 }
1620         }
1621
1622         cache = find_first_cache_extent(root_cache, 0);
1623         while (1) {
1624                 if (!cache)
1625                         break;
1626                 rec = container_of(cache, struct root_record, cache);
1627                 cache = next_cache_extent(cache);
1628
1629                 if (rec->found_ref == 0 &&
1630                     rec->objectid >= BTRFS_FIRST_FREE_OBJECTID &&
1631                     rec->objectid <= BTRFS_LAST_FREE_OBJECTID) {
1632                         ret = check_orphan_item(root->fs_info->tree_root,
1633                                                 rec->objectid);
1634                         if (ret == 0)
1635                                 continue;
1636                         errors++;
1637                         fprintf(stderr, "fs tree %llu not referenced\n",
1638                                 (unsigned long long)rec->objectid);
1639                 }
1640
1641                 error = 0;
1642                 if (rec->found_ref > 0 && !rec->found_root_item)
1643                         error = 1;
1644                 list_for_each_entry(backref, &rec->backrefs, list) {
1645                         if (!backref->found_dir_item)
1646                                 backref->errors |= REF_ERR_NO_DIR_ITEM;
1647                         if (!backref->found_dir_index)
1648                                 backref->errors |= REF_ERR_NO_DIR_INDEX;
1649                         if (!backref->found_back_ref)
1650                                 backref->errors |= REF_ERR_NO_ROOT_BACKREF;
1651                         if (!backref->found_forward_ref)
1652                                 backref->errors |= REF_ERR_NO_ROOT_REF;
1653                         if (backref->reachable && backref->errors)
1654                                 error = 1;
1655                 }
1656                 if (!error)
1657                         continue;
1658
1659                 errors++;
1660                 fprintf(stderr, "fs tree %llu refs %u %s\n",
1661                         (unsigned long long)rec->objectid, rec->found_ref,
1662                          rec->found_root_item ? "" : "not found");
1663
1664                 list_for_each_entry(backref, &rec->backrefs, list) {
1665                         if (!backref->reachable)
1666                                 continue;
1667                         if (!backref->errors && rec->found_root_item)
1668                                 continue;
1669                         fprintf(stderr, "\tunresolved ref root %llu dir %llu"
1670                                 " index %llu namelen %u name %s error %x\n",
1671                                 (unsigned long long)backref->ref_root,
1672                                 (unsigned long long)backref->dir,
1673                                 (unsigned long long)backref->index,
1674                                 backref->namelen, backref->name,
1675                                 backref->errors);
1676                 }
1677         }
1678         return errors > 0 ? 1 : 0;
1679 }
1680
1681 static int process_root_ref(struct extent_buffer *eb, int slot,
1682                             struct btrfs_key *key,
1683                             struct cache_tree *root_cache)
1684 {
1685         u64 dirid;
1686         u64 index;
1687         u32 len;
1688         u32 name_len;
1689         struct btrfs_root_ref *ref;
1690         char namebuf[BTRFS_NAME_LEN];
1691         int error;
1692
1693         ref = btrfs_item_ptr(eb, slot, struct btrfs_root_ref);
1694
1695         dirid = btrfs_root_ref_dirid(eb, ref);
1696         index = btrfs_root_ref_sequence(eb, ref);
1697         name_len = btrfs_root_ref_name_len(eb, ref);
1698
1699         if (name_len <= BTRFS_NAME_LEN) {
1700                 len = name_len;
1701                 error = 0;
1702         } else {
1703                 len = BTRFS_NAME_LEN;
1704                 error = REF_ERR_NAME_TOO_LONG;
1705         }
1706         read_extent_buffer(eb, namebuf, (unsigned long)(ref + 1), len);
1707
1708         if (key->type == BTRFS_ROOT_REF_KEY) {
1709                 add_root_backref(root_cache, key->offset, key->objectid, dirid,
1710                                  index, namebuf, len, key->type, error);
1711         } else {
1712                 add_root_backref(root_cache, key->objectid, key->offset, dirid,
1713                                  index, namebuf, len, key->type, error);
1714         }
1715         return 0;
1716 }
1717
1718 static int check_fs_root(struct btrfs_root *root,
1719                          struct cache_tree *root_cache,
1720                          struct walk_control *wc)
1721 {
1722         int ret = 0;
1723         int wret;
1724         int level;
1725         struct btrfs_path path;
1726         struct shared_node root_node;
1727         struct root_record *rec;
1728         struct btrfs_root_item *root_item = &root->root_item;
1729
1730         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1731                 rec = get_root_rec(root_cache, root->root_key.objectid);
1732                 if (btrfs_root_refs(root_item) > 0)
1733                         rec->found_root_item = 1;
1734         }
1735
1736         btrfs_init_path(&path);
1737         memset(&root_node, 0, sizeof(root_node));
1738         cache_tree_init(&root_node.root_cache);
1739         cache_tree_init(&root_node.inode_cache);
1740
1741         level = btrfs_header_level(root->node);
1742         memset(wc->nodes, 0, sizeof(wc->nodes));
1743         wc->nodes[level] = &root_node;
1744         wc->active_node = level;
1745         wc->root_level = level;
1746
1747         if (btrfs_root_refs(root_item) > 0 ||
1748             btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
1749                 path.nodes[level] = root->node;
1750                 extent_buffer_get(root->node);
1751                 path.slots[level] = 0;
1752         } else {
1753                 struct btrfs_key key;
1754                 struct btrfs_disk_key found_key;
1755
1756                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
1757                 level = root_item->drop_level;
1758                 path.lowest_level = level;
1759                 wret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
1760                 BUG_ON(wret < 0);
1761                 btrfs_node_key(path.nodes[level], &found_key,
1762                                 path.slots[level]);
1763                 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
1764                                         sizeof(found_key)));
1765         }
1766
1767         while (1) {
1768                 wret = walk_down_tree(root, &path, wc, &level);
1769                 if (wret < 0)
1770                         ret = wret;
1771                 if (wret != 0)
1772                         break;
1773
1774                 wret = walk_up_tree(root, &path, wc, &level);
1775                 if (wret < 0)
1776                         ret = wret;
1777                 if (wret != 0)
1778                         break;
1779         }
1780         btrfs_release_path(root, &path);
1781
1782         merge_root_recs(root, &root_node.root_cache, root_cache);
1783
1784         if (root_node.current) {
1785                 root_node.current->checked = 1;
1786                 maybe_free_inode_rec(&root_node.inode_cache,
1787                                 root_node.current);
1788         }
1789
1790         ret = check_inode_recs(root, &root_node.inode_cache);
1791         return ret;
1792 }
1793
1794 static int fs_root_objectid(u64 objectid)
1795 {
1796         if (objectid == BTRFS_FS_TREE_OBJECTID ||
1797             objectid == BTRFS_TREE_RELOC_OBJECTID ||
1798             objectid == BTRFS_DATA_RELOC_TREE_OBJECTID ||
1799             (objectid >= BTRFS_FIRST_FREE_OBJECTID &&
1800              objectid <= BTRFS_LAST_FREE_OBJECTID))
1801                 return 1;
1802         return 0;
1803 }
1804
1805 static int check_fs_roots(struct btrfs_root *root,
1806                           struct cache_tree *root_cache)
1807 {
1808         struct btrfs_path path;
1809         struct btrfs_key key;
1810         struct walk_control wc;
1811         struct extent_buffer *leaf;
1812         struct btrfs_root *tmp_root;
1813         struct btrfs_root *tree_root = root->fs_info->tree_root;
1814         int ret;
1815         int err = 0;
1816
1817         memset(&wc, 0, sizeof(wc));
1818         cache_tree_init(&wc.shared);
1819         btrfs_init_path(&path);
1820
1821         key.offset = 0;
1822         key.objectid = 0;
1823         key.type = BTRFS_ROOT_ITEM_KEY;
1824         ret = btrfs_search_slot(NULL, tree_root, &key, &path, 0, 0);
1825         BUG_ON(ret < 0);
1826         while (1) {
1827                 leaf = path.nodes[0];
1828                 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
1829                         ret = btrfs_next_leaf(tree_root, &path);
1830                         if (ret != 0)
1831                                 break;
1832                         leaf = path.nodes[0];
1833                 }
1834                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
1835                 if (key.type == BTRFS_ROOT_ITEM_KEY &&
1836                     fs_root_objectid(key.objectid)) {
1837                         tmp_root = btrfs_read_fs_root_no_cache(root->fs_info,
1838                                                                &key);
1839                         if (IS_ERR(tmp_root)) {
1840                                 err = 1;
1841                                 goto next;
1842                         }
1843                         ret = check_fs_root(tmp_root, root_cache, &wc);
1844                         if (ret)
1845                                 err = 1;
1846                         btrfs_free_fs_root(root->fs_info, tmp_root);
1847                 } else if (key.type == BTRFS_ROOT_REF_KEY ||
1848                            key.type == BTRFS_ROOT_BACKREF_KEY) {
1849                         process_root_ref(leaf, path.slots[0], &key,
1850                                          root_cache);
1851                 }
1852 next:
1853                 path.slots[0]++;
1854         }
1855         btrfs_release_path(tree_root, &path);
1856
1857         if (!cache_tree_empty(&wc.shared))
1858                 fprintf(stderr, "warning line %d\n", __LINE__);
1859
1860         return err;
1861 }
1862
1863 static int all_backpointers_checked(struct extent_record *rec, int print_errs)
1864 {
1865         struct list_head *cur = rec->backrefs.next;
1866         struct extent_backref *back;
1867         struct tree_backref *tback;
1868         struct data_backref *dback;
1869         u64 found = 0;
1870         int err = 0;
1871
1872         while(cur != &rec->backrefs) {
1873                 back = list_entry(cur, struct extent_backref, list);
1874                 cur = cur->next;
1875                 if (!back->found_extent_tree) {
1876                         err = 1;
1877                         if (!print_errs)
1878                                 goto out;
1879                         if (back->is_data) {
1880                                 dback = (struct data_backref *)back;
1881                                 fprintf(stderr, "Backref %llu %s %llu"
1882                                         " owner %llu offset %llu num_refs %lu"
1883                                         " not found in extent tree\n",
1884                                         (unsigned long long)rec->start,
1885                                         back->full_backref ?
1886                                         "parent" : "root",
1887                                         back->full_backref ?
1888                                         (unsigned long long)dback->parent:
1889                                         (unsigned long long)dback->root,
1890                                         (unsigned long long)dback->owner,
1891                                         (unsigned long long)dback->offset,
1892                                         (unsigned long)dback->num_refs);
1893                         } else {
1894                                 tback = (struct tree_backref *)back;
1895                                 fprintf(stderr, "Backref %llu parent %llu"
1896                                         " root %llu not found in extent tree\n",
1897                                         (unsigned long long)rec->start,
1898                                         (unsigned long long)tback->parent,
1899                                         (unsigned long long)tback->root);
1900                         }
1901                 }
1902                 if (!back->is_data && !back->found_ref) {
1903                         err = 1;
1904                         if (!print_errs)
1905                                 goto out;
1906                         tback = (struct tree_backref *)back;
1907                         fprintf(stderr, "Backref %llu %s %llu not referenced back %p\n",
1908                                 (unsigned long long)rec->start,
1909                                 back->full_backref ? "parent" : "root",
1910                                 back->full_backref ?
1911                                 (unsigned long long)tback->parent :
1912                                 (unsigned long long)tback->root, back);
1913                 }
1914                 if (back->is_data) {
1915                         dback = (struct data_backref *)back;
1916                         if (dback->found_ref != dback->num_refs) {
1917                                 err = 1;
1918                                 if (!print_errs)
1919                                         goto out;
1920                                 fprintf(stderr, "Incorrect local backref count"
1921                                         " on %llu %s %llu owner %llu"
1922                                         " offset %llu found %u wanted %u back %p\n",
1923                                         (unsigned long long)rec->start,
1924                                         back->full_backref ?
1925                                         "parent" : "root",
1926                                         back->full_backref ?
1927                                         (unsigned long long)dback->parent:
1928                                         (unsigned long long)dback->root,
1929                                         (unsigned long long)dback->owner,
1930                                         (unsigned long long)dback->offset,
1931                                         dback->found_ref, dback->num_refs, back);
1932                         }
1933                         if (dback->disk_bytenr != rec->start) {
1934                                 err = 1;
1935                                 if (!print_errs)
1936                                         goto out;
1937                                 fprintf(stderr, "Backref disk bytenr does not"
1938                                         " match extent record, bytenr=%llu, "
1939                                         "ref bytenr=%llu\n",
1940                                         (unsigned long long)rec->start,
1941                                         (unsigned long long)dback->disk_bytenr);
1942                         }
1943
1944                         if (dback->bytes != rec->nr) {
1945                                 err = 1;
1946                                 if (!print_errs)
1947                                         goto out;
1948                                 fprintf(stderr, "Backref bytes do not match "
1949                                         "extent backref, bytenr=%llu, ref "
1950                                         "bytes=%llu, backref bytes=%llu\n",
1951                                         (unsigned long long)rec->start,
1952                                         (unsigned long long)rec->nr,
1953                                         (unsigned long long)dback->bytes);
1954                         }
1955                 }
1956                 if (!back->is_data) {
1957                         found += 1;
1958                 } else {
1959                         dback = (struct data_backref *)back;
1960                         found += dback->found_ref;
1961                 }
1962         }
1963         if (found != rec->refs) {
1964                 err = 1;
1965                 if (!print_errs)
1966                         goto out;
1967                 fprintf(stderr, "Incorrect global backref count "
1968                         "on %llu found %llu wanted %llu\n",
1969                         (unsigned long long)rec->start,
1970                         (unsigned long long)found,
1971                         (unsigned long long)rec->refs);
1972         }
1973 out:
1974         return err;
1975 }
1976
1977 static int free_all_extent_backrefs(struct extent_record *rec)
1978 {
1979         struct extent_backref *back;
1980         struct list_head *cur;
1981         while (!list_empty(&rec->backrefs)) {
1982                 cur = rec->backrefs.next;
1983                 back = list_entry(cur, struct extent_backref, list);
1984                 list_del(cur);
1985                 free(back);
1986         }
1987         return 0;
1988 }
1989
1990 static void free_extent_cache(struct btrfs_fs_info *fs_info,
1991                               struct cache_tree *extent_cache)
1992 {
1993         struct cache_extent *cache;
1994         struct extent_record *rec;
1995
1996         while (1) {
1997                 cache = find_first_cache_extent(extent_cache, 0);
1998                 if (!cache)
1999                         break;
2000                 rec = container_of(cache, struct extent_record, cache);
2001                 btrfs_unpin_extent(fs_info, rec->start, rec->max_size);
2002                 remove_cache_extent(extent_cache, cache);
2003                 free_all_extent_backrefs(rec);
2004                 free(rec);
2005         }
2006 }
2007
2008 static int maybe_free_extent_rec(struct cache_tree *extent_cache,
2009                                  struct extent_record *rec)
2010 {
2011         if (rec->content_checked && rec->owner_ref_checked &&
2012             rec->extent_item_refs == rec->refs && rec->refs > 0 &&
2013             !all_backpointers_checked(rec, 0)) {
2014                 remove_cache_extent(extent_cache, &rec->cache);
2015                 free_all_extent_backrefs(rec);
2016                 free(rec);
2017         }
2018         return 0;
2019 }
2020
2021 static int check_owner_ref(struct btrfs_root *root,
2022                             struct extent_record *rec,
2023                             struct extent_buffer *buf)
2024 {
2025         struct extent_backref *node;
2026         struct tree_backref *back;
2027         struct btrfs_root *ref_root;
2028         struct btrfs_key key;
2029         struct btrfs_path path;
2030         struct extent_buffer *parent;
2031         int level;
2032         int found = 0;
2033         int ret;
2034
2035         list_for_each_entry(node, &rec->backrefs, list) {
2036                 if (node->is_data)
2037                         continue;
2038                 if (!node->found_ref)
2039                         continue;
2040                 if (node->full_backref)
2041                         continue;
2042                 back = (struct tree_backref *)node;
2043                 if (btrfs_header_owner(buf) == back->root)
2044                         return 0;
2045         }
2046         BUG_ON(rec->is_root);
2047
2048         /* try to find the block by search corresponding fs tree */
2049         key.objectid = btrfs_header_owner(buf);
2050         key.type = BTRFS_ROOT_ITEM_KEY;
2051         key.offset = (u64)-1;
2052
2053         ref_root = btrfs_read_fs_root(root->fs_info, &key);
2054         if (IS_ERR(ref_root))
2055                 return 1;
2056
2057         level = btrfs_header_level(buf);
2058         if (level == 0)
2059                 btrfs_item_key_to_cpu(buf, &key, 0);
2060         else
2061                 btrfs_node_key_to_cpu(buf, &key, 0);
2062
2063         btrfs_init_path(&path);
2064         path.lowest_level = level + 1;
2065         ret = btrfs_search_slot(NULL, ref_root, &key, &path, 0, 0);
2066         if (ret < 0)
2067                 return 0;
2068
2069         parent = path.nodes[level + 1];
2070         if (parent && buf->start == btrfs_node_blockptr(parent,
2071                                                         path.slots[level + 1]))
2072                 found = 1;
2073
2074         btrfs_release_path(ref_root, &path);
2075         return found ? 0 : 1;
2076 }
2077
2078 static int is_extent_tree_record(struct extent_record *rec)
2079 {
2080         struct list_head *cur = rec->backrefs.next;
2081         struct extent_backref *node;
2082         struct tree_backref *back;
2083         int is_extent = 0;
2084
2085         while(cur != &rec->backrefs) {
2086                 node = list_entry(cur, struct extent_backref, list);
2087                 cur = cur->next;
2088                 if (node->is_data)
2089                         return 0;
2090                 back = (struct tree_backref *)node;
2091                 if (node->full_backref)
2092                         return 0;
2093                 if (back->root == BTRFS_EXTENT_TREE_OBJECTID)
2094                         is_extent = 1;
2095         }
2096         return is_extent;
2097 }
2098
2099
2100 static int record_bad_block_io(struct btrfs_fs_info *info,
2101                                struct cache_tree *extent_cache,
2102                                u64 start, u64 len)
2103 {
2104         struct extent_record *rec;
2105         struct cache_extent *cache;
2106         struct btrfs_key key;
2107
2108         cache = find_cache_extent(extent_cache, start, len);
2109         if (!cache)
2110                 return 0;
2111
2112         rec = container_of(cache, struct extent_record, cache);
2113         if (!is_extent_tree_record(rec))
2114                 return 0;
2115
2116         btrfs_disk_key_to_cpu(&key, &rec->parent_key);
2117         return btrfs_add_corrupt_extent_record(info, &key, start, len, 0);
2118 }
2119
2120 static int check_block(struct btrfs_root *root,
2121                        struct cache_tree *extent_cache,
2122                        struct extent_buffer *buf, u64 flags)
2123 {
2124         struct extent_record *rec;
2125         struct cache_extent *cache;
2126         struct btrfs_key key;
2127         int ret = 1;
2128         int level;
2129
2130         cache = find_cache_extent(extent_cache, buf->start, buf->len);
2131         if (!cache)
2132                 return 1;
2133         rec = container_of(cache, struct extent_record, cache);
2134         rec->generation = btrfs_header_generation(buf);
2135
2136         level = btrfs_header_level(buf);
2137         if (btrfs_header_nritems(buf) > 0) {
2138
2139                 if (level == 0)
2140                         btrfs_item_key_to_cpu(buf, &key, 0);
2141                 else
2142                         btrfs_node_key_to_cpu(buf, &key, 0);
2143
2144                 rec->info_objectid = key.objectid;
2145         }
2146         rec->info_level = level;
2147
2148         if (btrfs_is_leaf(buf))
2149                 ret = btrfs_check_leaf(root, &rec->parent_key, buf);
2150         else
2151                 ret = btrfs_check_node(root, &rec->parent_key, buf);
2152
2153         if (ret) {
2154                 fprintf(stderr, "bad block %llu\n",
2155                         (unsigned long long)buf->start);
2156         } else {
2157                 rec->content_checked = 1;
2158                 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
2159                         rec->owner_ref_checked = 1;
2160                 else {
2161                         ret = check_owner_ref(root, rec, buf);
2162                         if (!ret)
2163                                 rec->owner_ref_checked = 1;
2164                 }
2165         }
2166         if (!ret)
2167                 maybe_free_extent_rec(extent_cache, rec);
2168         return ret;
2169 }
2170
2171 static struct tree_backref *find_tree_backref(struct extent_record *rec,
2172                                                 u64 parent, u64 root)
2173 {
2174         struct list_head *cur = rec->backrefs.next;
2175         struct extent_backref *node;
2176         struct tree_backref *back;
2177
2178         while(cur != &rec->backrefs) {
2179                 node = list_entry(cur, struct extent_backref, list);
2180                 cur = cur->next;
2181                 if (node->is_data)
2182                         continue;
2183                 back = (struct tree_backref *)node;
2184                 if (parent > 0) {
2185                         if (!node->full_backref)
2186                                 continue;
2187                         if (parent == back->parent)
2188                                 return back;
2189                 } else {
2190                         if (node->full_backref)
2191                                 continue;
2192                         if (back->root == root)
2193                                 return back;
2194                 }
2195         }
2196         return NULL;
2197 }
2198
2199 static struct tree_backref *alloc_tree_backref(struct extent_record *rec,
2200                                                 u64 parent, u64 root)
2201 {
2202         struct tree_backref *ref = malloc(sizeof(*ref));
2203         memset(&ref->node, 0, sizeof(ref->node));
2204         if (parent > 0) {
2205                 ref->parent = parent;
2206                 ref->node.full_backref = 1;
2207         } else {
2208                 ref->root = root;
2209                 ref->node.full_backref = 0;
2210         }
2211         list_add_tail(&ref->node.list, &rec->backrefs);
2212
2213         return ref;
2214 }
2215
2216 static struct data_backref *find_data_backref(struct extent_record *rec,
2217                                                 u64 parent, u64 root,
2218                                                 u64 owner, u64 offset,
2219                                                 int found_ref,
2220                                                 u64 disk_bytenr, u64 bytes)
2221 {
2222         struct list_head *cur = rec->backrefs.next;
2223         struct extent_backref *node;
2224         struct data_backref *back;
2225
2226         while(cur != &rec->backrefs) {
2227                 node = list_entry(cur, struct extent_backref, list);
2228                 cur = cur->next;
2229                 if (!node->is_data)
2230                         continue;
2231                 back = (struct data_backref *)node;
2232                 if (parent > 0) {
2233                         if (!node->full_backref)
2234                                 continue;
2235                         if (parent == back->parent)
2236                                 return back;
2237                 } else {
2238                         if (node->full_backref)
2239                                 continue;
2240                         if (back->root == root && back->owner == owner &&
2241                             back->offset == offset) {
2242                                 if (found_ref && node->found_ref &&
2243                                     (back->bytes != bytes ||
2244                                     back->disk_bytenr != disk_bytenr))
2245                                         continue;
2246                                 return back;
2247                         }
2248                 }
2249         }
2250         return NULL;
2251 }
2252
2253 static struct data_backref *alloc_data_backref(struct extent_record *rec,
2254                                                 u64 parent, u64 root,
2255                                                 u64 owner, u64 offset,
2256                                                 u64 max_size)
2257 {
2258         struct data_backref *ref = malloc(sizeof(*ref));
2259         memset(&ref->node, 0, sizeof(ref->node));
2260         ref->node.is_data = 1;
2261
2262         if (parent > 0) {
2263                 ref->parent = parent;
2264                 ref->owner = 0;
2265                 ref->offset = 0;
2266                 ref->node.full_backref = 1;
2267         } else {
2268                 ref->root = root;
2269                 ref->owner = owner;
2270                 ref->offset = offset;
2271                 ref->node.full_backref = 0;
2272         }
2273         ref->bytes = max_size;
2274         ref->found_ref = 0;
2275         ref->num_refs = 0;
2276         list_add_tail(&ref->node.list, &rec->backrefs);
2277         if (max_size > rec->max_size)
2278                 rec->max_size = max_size;
2279         return ref;
2280 }
2281
2282 static int add_extent_rec(struct cache_tree *extent_cache,
2283                           struct btrfs_key *parent_key,
2284                           u64 start, u64 nr, u64 extent_item_refs,
2285                           int is_root, int inc_ref, int set_checked,
2286                           int metadata, int extent_rec, u64 max_size)
2287 {
2288         struct extent_record *rec;
2289         struct cache_extent *cache;
2290         int ret = 0;
2291         int dup = 0;
2292
2293         cache = find_cache_extent(extent_cache, start, nr);
2294         if (cache) {
2295                 rec = container_of(cache, struct extent_record, cache);
2296                 if (inc_ref)
2297                         rec->refs++;
2298                 if (rec->nr == 1)
2299                         rec->nr = max(nr, max_size);
2300
2301                 /*
2302                  * We need to make sure to reset nr to whatever the extent
2303                  * record says was the real size, this way we can compare it to
2304                  * the backrefs.
2305                  */
2306                 if (extent_rec) {
2307                         if (rec->found_rec) {
2308                                 struct extent_record *tmp;
2309
2310                                 dup = 1;
2311                                 if (list_empty(&rec->list))
2312                                         list_add_tail(&rec->list,
2313                                                       &duplicate_extents);
2314
2315                                 /*
2316                                  * We have to do this song and dance in case we
2317                                  * find an extent record that falls inside of
2318                                  * our current extent record but does not have
2319                                  * the same objectid.
2320                                  */
2321                                 tmp = malloc(sizeof(*tmp));
2322                                 if (!tmp)
2323                                         return -ENOMEM;
2324                                 tmp->start = start;
2325                                 tmp->max_size = max_size;
2326                                 tmp->nr = nr;
2327                                 tmp->found_rec = 1;
2328                                 tmp->metadata = metadata;
2329                                 INIT_LIST_HEAD(&tmp->list);
2330                                 list_add_tail(&tmp->list, &rec->dups);
2331                                 rec->has_duplicate = 1;
2332                         } else {
2333                                 rec->nr = nr;
2334                         }
2335                         rec->found_rec = 1;
2336                 }
2337
2338                 if (extent_item_refs && !dup) {
2339                         if (rec->extent_item_refs) {
2340                                 fprintf(stderr, "block %llu rec "
2341                                         "extent_item_refs %llu, passed %llu\n",
2342                                         (unsigned long long)start,
2343                                         (unsigned long long)
2344                                                         rec->extent_item_refs,
2345                                         (unsigned long long)extent_item_refs);
2346                         }
2347                         rec->extent_item_refs = extent_item_refs;
2348                 }
2349                 if (is_root)
2350                         rec->is_root = 1;
2351                 if (set_checked) {
2352                         rec->content_checked = 1;
2353                         rec->owner_ref_checked = 1;
2354                 }
2355
2356                 if (parent_key)
2357                         btrfs_cpu_key_to_disk(&rec->parent_key, parent_key);
2358
2359                 if (rec->max_size < max_size)
2360                         rec->max_size = max_size;
2361
2362                 maybe_free_extent_rec(extent_cache, rec);
2363                 return ret;
2364         }
2365         rec = malloc(sizeof(*rec));
2366         rec->start = start;
2367         rec->max_size = max_size;
2368         rec->nr = max(nr, max_size);
2369         rec->found_rec = extent_rec;
2370         rec->content_checked = 0;
2371         rec->owner_ref_checked = 0;
2372         rec->has_duplicate = 0;
2373         rec->metadata = metadata;
2374         INIT_LIST_HEAD(&rec->backrefs);
2375         INIT_LIST_HEAD(&rec->dups);
2376         INIT_LIST_HEAD(&rec->list);
2377
2378         if (is_root)
2379                 rec->is_root = 1;
2380         else
2381                 rec->is_root = 0;
2382
2383         if (inc_ref)
2384                 rec->refs = 1;
2385         else
2386                 rec->refs = 0;
2387
2388         if (extent_item_refs)
2389                 rec->extent_item_refs = extent_item_refs;
2390         else
2391                 rec->extent_item_refs = 0;
2392
2393         if (parent_key)
2394                 btrfs_cpu_key_to_disk(&rec->parent_key, parent_key);
2395         else
2396                 memset(&rec->parent_key, 0, sizeof(*parent_key));
2397
2398         rec->cache.start = start;
2399         rec->cache.size = nr;
2400         ret = insert_existing_cache_extent(extent_cache, &rec->cache);
2401         BUG_ON(ret);
2402         bytes_used += nr;
2403         if (set_checked) {
2404                 rec->content_checked = 1;
2405                 rec->owner_ref_checked = 1;
2406         }
2407         return ret;
2408 }
2409
2410 static int add_tree_backref(struct cache_tree *extent_cache, u64 bytenr,
2411                             u64 parent, u64 root, int found_ref)
2412 {
2413         struct extent_record *rec;
2414         struct tree_backref *back;
2415         struct cache_extent *cache;
2416
2417         cache = find_cache_extent(extent_cache, bytenr, 1);
2418         if (!cache) {
2419                 add_extent_rec(extent_cache, NULL, bytenr,
2420                                1, 0, 0, 0, 0, 1, 0, 0);
2421                 cache = find_cache_extent(extent_cache, bytenr, 1);
2422                 if (!cache)
2423                         abort();
2424         }
2425
2426         rec = container_of(cache, struct extent_record, cache);
2427         if (rec->start != bytenr) {
2428                 abort();
2429         }
2430
2431         back = find_tree_backref(rec, parent, root);
2432         if (!back)
2433                 back = alloc_tree_backref(rec, parent, root);
2434
2435         if (found_ref) {
2436                 if (back->node.found_ref) {
2437                         fprintf(stderr, "Extent back ref already exists "
2438                                 "for %llu parent %llu root %llu \n",
2439                                 (unsigned long long)bytenr,
2440                                 (unsigned long long)parent,
2441                                 (unsigned long long)root);
2442                 }
2443                 back->node.found_ref = 1;
2444         } else {
2445                 if (back->node.found_extent_tree) {
2446                         fprintf(stderr, "Extent back ref already exists "
2447                                 "for %llu parent %llu root %llu \n",
2448                                 (unsigned long long)bytenr,
2449                                 (unsigned long long)parent,
2450                                 (unsigned long long)root);
2451                 }
2452                 back->node.found_extent_tree = 1;
2453         }
2454         return 0;
2455 }
2456
2457 static int add_data_backref(struct cache_tree *extent_cache, u64 bytenr,
2458                             u64 parent, u64 root, u64 owner, u64 offset,
2459                             u32 num_refs, int found_ref, u64 max_size)
2460 {
2461         struct extent_record *rec;
2462         struct data_backref *back;
2463         struct cache_extent *cache;
2464
2465         cache = find_cache_extent(extent_cache, bytenr, 1);
2466         if (!cache) {
2467                 add_extent_rec(extent_cache, NULL, bytenr, 1, 0, 0, 0, 0,
2468                                0, 0, max_size);
2469                 cache = find_cache_extent(extent_cache, bytenr, 1);
2470                 if (!cache)
2471                         abort();
2472         }
2473
2474         rec = container_of(cache, struct extent_record, cache);
2475         if (rec->max_size < max_size)
2476                 rec->max_size = max_size;
2477
2478         /*
2479          * If found_ref is set then max_size is the real size and must match the
2480          * existing refs.  So if we have already found a ref then we need to
2481          * make sure that this ref matches the existing one, otherwise we need
2482          * to add a new backref so we can notice that the backrefs don't match
2483          * and we need to figure out who is telling the truth.  This is to
2484          * account for that awful fsync bug I introduced where we'd end up with
2485          * a btrfs_file_extent_item that would have its length include multiple
2486          * prealloc extents or point inside of a prealloc extent.
2487          */
2488         back = find_data_backref(rec, parent, root, owner, offset, found_ref,
2489                                  bytenr, max_size);
2490         if (!back)
2491                 back = alloc_data_backref(rec, parent, root, owner, offset,
2492                                           max_size);
2493
2494         if (found_ref) {
2495                 BUG_ON(num_refs != 1);
2496                 if (back->node.found_ref)
2497                         BUG_ON(back->bytes != max_size);
2498                 back->node.found_ref = 1;
2499                 back->found_ref += 1;
2500                 back->bytes = max_size;
2501                 back->disk_bytenr = bytenr;
2502                 rec->refs += 1;
2503                 rec->content_checked = 1;
2504                 rec->owner_ref_checked = 1;
2505         } else {
2506                 if (back->node.found_extent_tree) {
2507                         fprintf(stderr, "Extent back ref already exists "
2508                                 "for %llu parent %llu root %llu"
2509                                 "owner %llu offset %llu num_refs %lu\n",
2510                                 (unsigned long long)bytenr,
2511                                 (unsigned long long)parent,
2512                                 (unsigned long long)root,
2513                                 (unsigned long long)owner,
2514                                 (unsigned long long)offset,
2515                                 (unsigned long)num_refs);
2516                 }
2517                 back->num_refs = num_refs;
2518                 back->node.found_extent_tree = 1;
2519         }
2520         return 0;
2521 }
2522
2523 static int add_pending(struct cache_tree *pending,
2524                        struct cache_tree *seen, u64 bytenr, u32 size)
2525 {
2526         int ret;
2527         ret = insert_cache_extent(seen, bytenr, size);
2528         if (ret)
2529                 return ret;
2530         insert_cache_extent(pending, bytenr, size);
2531         return 0;
2532 }
2533
2534 static int pick_next_pending(struct cache_tree *pending,
2535                         struct cache_tree *reada,
2536                         struct cache_tree *nodes,
2537                         u64 last, struct block_info *bits, int bits_nr,
2538                         int *reada_bits)
2539 {
2540         unsigned long node_start = last;
2541         struct cache_extent *cache;
2542         int ret;
2543
2544         cache = find_first_cache_extent(reada, 0);
2545         if (cache) {
2546                 bits[0].start = cache->start;
2547                 bits[1].size = cache->size;
2548                 *reada_bits = 1;
2549                 return 1;
2550         }
2551         *reada_bits = 0;
2552         if (node_start > 32768)
2553                 node_start -= 32768;
2554
2555         cache = find_first_cache_extent(nodes, node_start);
2556         if (!cache)
2557                 cache = find_first_cache_extent(nodes, 0);
2558
2559         if (!cache) {
2560                  cache = find_first_cache_extent(pending, 0);
2561                  if (!cache)
2562                          return 0;
2563                  ret = 0;
2564                  do {
2565                          bits[ret].start = cache->start;
2566                          bits[ret].size = cache->size;
2567                          cache = next_cache_extent(cache);
2568                          ret++;
2569                  } while (cache && ret < bits_nr);
2570                  return ret;
2571         }
2572
2573         ret = 0;
2574         do {
2575                 bits[ret].start = cache->start;
2576                 bits[ret].size = cache->size;
2577                 cache = next_cache_extent(cache);
2578                 ret++;
2579         } while (cache && ret < bits_nr);
2580
2581         if (bits_nr - ret > 8) {
2582                 u64 lookup = bits[0].start + bits[0].size;
2583                 struct cache_extent *next;
2584                 next = find_first_cache_extent(pending, lookup);
2585                 while(next) {
2586                         if (next->start - lookup > 32768)
2587                                 break;
2588                         bits[ret].start = next->start;
2589                         bits[ret].size = next->size;
2590                         lookup = next->start + next->size;
2591                         ret++;
2592                         if (ret == bits_nr)
2593                                 break;
2594                         next = next_cache_extent(next);
2595                         if (!next)
2596                                 break;
2597                 }
2598         }
2599         return ret;
2600 }
2601
2602 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2603 static int process_extent_ref_v0(struct cache_tree *extent_cache,
2604                                  struct extent_buffer *leaf, int slot)
2605 {
2606         struct btrfs_extent_ref_v0 *ref0;
2607         struct btrfs_key key;
2608
2609         btrfs_item_key_to_cpu(leaf, &key, slot);
2610         ref0 = btrfs_item_ptr(leaf, slot, struct btrfs_extent_ref_v0);
2611         if (btrfs_ref_objectid_v0(leaf, ref0) < BTRFS_FIRST_FREE_OBJECTID) {
2612                 add_tree_backref(extent_cache, key.objectid, key.offset, 0, 0);
2613         } else {
2614                 add_data_backref(extent_cache, key.objectid, key.offset, 0,
2615                                  0, 0, btrfs_ref_count_v0(leaf, ref0), 0, 0);
2616         }
2617         return 0;
2618 }
2619 #endif
2620
2621 static int process_extent_item(struct btrfs_root *root,
2622                                struct cache_tree *extent_cache,
2623                                struct extent_buffer *eb, int slot)
2624 {
2625         struct btrfs_extent_item *ei;
2626         struct btrfs_extent_inline_ref *iref;
2627         struct btrfs_extent_data_ref *dref;
2628         struct btrfs_shared_data_ref *sref;
2629         struct btrfs_key key;
2630         unsigned long end;
2631         unsigned long ptr;
2632         int type;
2633         u32 item_size = btrfs_item_size_nr(eb, slot);
2634         u64 refs = 0;
2635         u64 offset;
2636         u64 num_bytes;
2637         int metadata = 0;
2638
2639         btrfs_item_key_to_cpu(eb, &key, slot);
2640
2641         if (key.type == BTRFS_METADATA_ITEM_KEY) {
2642                 metadata = 1;
2643                 num_bytes = root->leafsize;
2644         } else {
2645                 num_bytes = key.offset;
2646         }
2647
2648         if (item_size < sizeof(*ei)) {
2649 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2650                 struct btrfs_extent_item_v0 *ei0;
2651                 BUG_ON(item_size != sizeof(*ei0));
2652                 ei0 = btrfs_item_ptr(eb, slot, struct btrfs_extent_item_v0);
2653                 refs = btrfs_extent_refs_v0(eb, ei0);
2654 #else
2655                 BUG();
2656 #endif
2657                 return add_extent_rec(extent_cache, NULL, key.objectid,
2658                                       num_bytes, refs, 0, 0, 0, metadata, 1,
2659                                       num_bytes);
2660         }
2661
2662         ei = btrfs_item_ptr(eb, slot, struct btrfs_extent_item);
2663         refs = btrfs_extent_refs(eb, ei);
2664
2665         add_extent_rec(extent_cache, NULL, key.objectid, num_bytes,
2666                        refs, 0, 0, 0, metadata, 1, num_bytes);
2667
2668         ptr = (unsigned long)(ei + 1);
2669         if (btrfs_extent_flags(eb, ei) & BTRFS_EXTENT_FLAG_TREE_BLOCK &&
2670             key.type == BTRFS_EXTENT_ITEM_KEY)
2671                 ptr += sizeof(struct btrfs_tree_block_info);
2672
2673         end = (unsigned long)ei + item_size;
2674         while (ptr < end) {
2675                 iref = (struct btrfs_extent_inline_ref *)ptr;
2676                 type = btrfs_extent_inline_ref_type(eb, iref);
2677                 offset = btrfs_extent_inline_ref_offset(eb, iref);
2678                 switch (type) {
2679                 case BTRFS_TREE_BLOCK_REF_KEY:
2680                         add_tree_backref(extent_cache, key.objectid,
2681                                          0, offset, 0);
2682                         break;
2683                 case BTRFS_SHARED_BLOCK_REF_KEY:
2684                         add_tree_backref(extent_cache, key.objectid,
2685                                          offset, 0, 0);
2686                         break;
2687                 case BTRFS_EXTENT_DATA_REF_KEY:
2688                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
2689                         add_data_backref(extent_cache, key.objectid, 0,
2690                                         btrfs_extent_data_ref_root(eb, dref),
2691                                         btrfs_extent_data_ref_objectid(eb,
2692                                                                        dref),
2693                                         btrfs_extent_data_ref_offset(eb, dref),
2694                                         btrfs_extent_data_ref_count(eb, dref),
2695                                         0, num_bytes);
2696                         break;
2697                 case BTRFS_SHARED_DATA_REF_KEY:
2698                         sref = (struct btrfs_shared_data_ref *)(iref + 1);
2699                         add_data_backref(extent_cache, key.objectid, offset,
2700                                         0, 0, 0,
2701                                         btrfs_shared_data_ref_count(eb, sref),
2702                                         0, num_bytes);
2703                         break;
2704                 default:
2705                         fprintf(stderr, "corrupt extent record: key %Lu %u %Lu\n",
2706                                 key.objectid, key.type, num_bytes);
2707                         goto out;
2708                 }
2709                 ptr += btrfs_extent_inline_ref_size(type);
2710         }
2711         WARN_ON(ptr > end);
2712 out:
2713         return 0;
2714 }
2715
2716 static int check_cache_range(struct btrfs_root *root,
2717                              struct btrfs_block_group_cache *cache,
2718                              u64 offset, u64 bytes)
2719 {
2720         struct btrfs_free_space *entry;
2721         u64 *logical;
2722         u64 bytenr;
2723         int stripe_len;
2724         int i, nr, ret;
2725
2726         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
2727                 bytenr = btrfs_sb_offset(i);
2728                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
2729                                        cache->key.objectid, bytenr, 0,
2730                                        &logical, &nr, &stripe_len);
2731                 if (ret)
2732                         return ret;
2733
2734                 while (nr--) {
2735                         if (logical[nr] + stripe_len <= offset)
2736                                 continue;
2737                         if (offset + bytes <= logical[nr])
2738                                 continue;
2739                         if (logical[nr] == offset) {
2740                                 if (stripe_len >= bytes) {
2741                                         kfree(logical);
2742                                         return 0;
2743                                 }
2744                                 bytes -= stripe_len;
2745                                 offset += stripe_len;
2746                         } else if (logical[nr] < offset) {
2747                                 if (logical[nr] + stripe_len >=
2748                                     offset + bytes) {
2749                                         kfree(logical);
2750                                         return 0;
2751                                 }
2752                                 bytes = (offset + bytes) -
2753                                         (logical[nr] + stripe_len);
2754                                 offset = logical[nr] + stripe_len;
2755                         } else {
2756                                 /*
2757                                  * Could be tricky, the super may land in the
2758                                  * middle of the area we're checking.  First
2759                                  * check the easiest case, it's at the end.
2760                                  */
2761                                 if (logical[nr] + stripe_len >=
2762                                     bytes + offset) {
2763                                         bytes = logical[nr] - offset;
2764                                         continue;
2765                                 }
2766
2767                                 /* Check the left side */
2768                                 ret = check_cache_range(root, cache,
2769                                                         offset,
2770                                                         logical[nr] - offset);
2771                                 if (ret) {
2772                                         kfree(logical);
2773                                         return ret;
2774                                 }
2775
2776                                 /* Now we continue with the right side */
2777                                 bytes = (offset + bytes) -
2778                                         (logical[nr] + stripe_len);
2779                                 offset = logical[nr] + stripe_len;
2780                         }
2781                 }
2782
2783                 kfree(logical);
2784         }
2785
2786         entry = btrfs_find_free_space(cache->free_space_ctl, offset, bytes);
2787         if (!entry) {
2788                 fprintf(stderr, "There is no free space entry for %Lu-%Lu\n",
2789                         offset, offset+bytes);
2790                 return -EINVAL;
2791         }
2792
2793         if (entry->offset != offset) {
2794                 fprintf(stderr, "Wanted offset %Lu, found %Lu\n", offset,
2795                         entry->offset);
2796                 return -EINVAL;
2797         }
2798
2799         if (entry->bytes != bytes) {
2800                 fprintf(stderr, "Wanted bytes %Lu, found %Lu for off %Lu\n",
2801                         bytes, entry->bytes, offset);
2802                 return -EINVAL;
2803         }
2804
2805         unlink_free_space(cache->free_space_ctl, entry);
2806         free(entry);
2807         return 0;
2808 }
2809
2810 static int verify_space_cache(struct btrfs_root *root,
2811                               struct btrfs_block_group_cache *cache)
2812 {
2813         struct btrfs_path *path;
2814         struct extent_buffer *leaf;
2815         struct btrfs_key key;
2816         u64 last;
2817         int ret = 0;
2818
2819         path = btrfs_alloc_path();
2820         if (!path)
2821                 return -ENOMEM;
2822
2823         root = root->fs_info->extent_root;
2824
2825         last = max_t(u64, cache->key.objectid, BTRFS_SUPER_INFO_OFFSET);
2826
2827         key.objectid = last;
2828         key.offset = 0;
2829         key.type = BTRFS_EXTENT_ITEM_KEY;
2830
2831         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2832         if (ret < 0)
2833                 return ret;
2834         ret = 0;
2835         while (1) {
2836                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2837                         ret = btrfs_next_leaf(root, path);
2838                         if (ret < 0)
2839                                 return ret;
2840                         if (ret > 0) {
2841                                 ret = 0;
2842                                 break;
2843                         }
2844                 }
2845                 leaf = path->nodes[0];
2846                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2847                 if (key.objectid >= cache->key.offset + cache->key.objectid)
2848                         break;
2849                 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
2850                     key.type != BTRFS_METADATA_ITEM_KEY) {
2851                         path->slots[0]++;
2852                         continue;
2853                 }
2854
2855                 if (last == key.objectid) {
2856                         if (key.type == BTRFS_EXTENT_ITEM_KEY)
2857                                 last = key.objectid + key.offset;
2858                         else
2859                                 last = key.objectid + root->leafsize;
2860                         path->slots[0]++;
2861                         continue;
2862                 }
2863
2864                 ret = check_cache_range(root, cache, last,
2865                                         key.objectid - last);
2866                 if (ret)
2867                         break;
2868                 if (key.type == BTRFS_EXTENT_ITEM_KEY)
2869                         last = key.objectid + key.offset;
2870                 else
2871                         last = key.objectid + root->leafsize;
2872                 path->slots[0]++;
2873         }
2874
2875         if (last < cache->key.objectid + cache->key.offset)
2876                 ret = check_cache_range(root, cache, last,
2877                                         cache->key.objectid +
2878                                         cache->key.offset - last);
2879         btrfs_free_path(path);
2880
2881         if (!ret &&
2882             !RB_EMPTY_ROOT(&cache->free_space_ctl->free_space_offset)) {
2883                 fprintf(stderr, "There are still entries left in the space "
2884                         "cache\n");
2885                 ret = -EINVAL;
2886         }
2887
2888         return ret;
2889 }
2890
2891 static int check_space_cache(struct btrfs_root *root)
2892 {
2893         struct btrfs_block_group_cache *cache;
2894         u64 start = BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE;
2895         int ret;
2896         int error = 0;
2897
2898         if (btrfs_super_generation(root->fs_info->super_copy) !=
2899             btrfs_super_cache_generation(root->fs_info->super_copy)) {
2900                 printf("cache and super generation don't match, space cache "
2901                        "will be invalidated\n");
2902                 return 0;
2903         }
2904
2905         while (1) {
2906                 cache = btrfs_lookup_first_block_group(root->fs_info, start);
2907                 if (!cache)
2908                         break;
2909
2910                 start = cache->key.objectid + cache->key.offset;
2911                 if (!cache->free_space_ctl) {
2912                         if (btrfs_init_free_space_ctl(cache,
2913                                                       root->sectorsize)) {
2914                                 ret = -ENOMEM;
2915                                 break;
2916                         }
2917                 } else {
2918                         btrfs_remove_free_space_cache(cache);
2919                 }
2920
2921                 ret = load_free_space_cache(root->fs_info, cache);
2922                 if (!ret)
2923                         continue;
2924
2925                 ret = verify_space_cache(root, cache);
2926                 if (ret) {
2927                         fprintf(stderr, "cache appears valid but isnt %Lu\n",
2928                                 cache->key.objectid);
2929                         error++;
2930                 }
2931         }
2932
2933         return error ? -EINVAL : 0;
2934 }
2935
2936 static int check_extent_exists(struct btrfs_root *root, u64 bytenr,
2937                                u64 num_bytes)
2938 {
2939         struct btrfs_path *path;
2940         struct extent_buffer *leaf;
2941         struct btrfs_key key;
2942         int ret;
2943
2944         path = btrfs_alloc_path();
2945         if (!path) {
2946                 fprintf(stderr, "Error allocing path\n");
2947                 return -ENOMEM;
2948         }
2949
2950         key.objectid = bytenr;
2951         key.type = BTRFS_EXTENT_ITEM_KEY;
2952         key.offset = 0;
2953
2954
2955 again:
2956         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
2957                                 0, 0);
2958         if (ret < 0) {
2959                 fprintf(stderr, "Error looking up extent record %d\n", ret);
2960                 btrfs_free_path(path);
2961                 return ret;
2962         } else if (ret) {
2963                 if (path->slots[0])
2964                         path->slots[0]--;
2965                 else
2966                         btrfs_prev_leaf(root, path);
2967         }
2968
2969         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2970
2971         /*
2972          * Block group items come before extent items if they have the same
2973          * bytenr, so walk back one more just in case.  Dear future traveler,
2974          * first congrats on mastering time travel.  Now if it's not too much
2975          * trouble could you go back to 2006 and tell Chris to make the
2976          * BLOCK_GROUP_ITEM_KEY lower than the EXTENT_ITEM_KEY please?
2977          */
2978         if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
2979                 if (path->slots[0])
2980                         path->slots[0]--;
2981                 else
2982                         btrfs_prev_leaf(root, path);
2983         }
2984
2985         while (num_bytes) {
2986                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2987                         ret = btrfs_next_leaf(root, path);
2988                         if (ret < 0) {
2989                                 fprintf(stderr, "Error going to next leaf "
2990                                         "%d\n", ret);
2991                                 btrfs_free_path(path);
2992                                 return ret;
2993                         } else if (ret) {
2994                                 break;
2995                         }
2996                 }
2997                 leaf = path->nodes[0];
2998                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2999                 if (key.type != BTRFS_EXTENT_ITEM_KEY) {
3000                         path->slots[0]++;
3001                         continue;
3002                 }
3003                 if (key.objectid + key.offset < bytenr) {
3004                         path->slots[0]++;
3005                         continue;
3006                 }
3007                 if (key.objectid > bytenr + num_bytes)
3008                         break;
3009
3010                 if (key.objectid == bytenr) {
3011                         if (key.offset >= num_bytes) {
3012                                 num_bytes = 0;
3013                                 break;
3014                         }
3015                         num_bytes -= key.offset;
3016                         bytenr += key.offset;
3017                 } else if (key.objectid < bytenr) {
3018                         if (key.objectid + key.offset >= bytenr + num_bytes) {
3019                                 num_bytes = 0;
3020                                 break;
3021                         }
3022                         num_bytes = (bytenr + num_bytes) -
3023                                 (key.objectid + key.offset);
3024                         bytenr = key.objectid + key.offset;
3025                 } else {
3026                         if (key.objectid + key.offset < bytenr + num_bytes) {
3027                                 u64 new_start = key.objectid + key.offset;
3028                                 u64 new_bytes = bytenr + num_bytes - new_start;
3029
3030                                 /*
3031                                  * Weird case, the extent is in the middle of
3032                                  * our range, we'll have to search one side
3033                                  * and then the other.  Not sure if this happens
3034                                  * in real life, but no harm in coding it up
3035                                  * anyway just in case.
3036                                  */
3037                                 btrfs_release_path(root, path);
3038                                 ret = check_extent_exists(root, new_start,
3039                                                           new_bytes);
3040                                 if (ret) {
3041                                         fprintf(stderr, "Right section didn't "
3042                                                 "have a record\n");
3043                                         break;
3044                                 }
3045                                 num_bytes = key.objectid - bytenr;
3046                                 goto again;
3047                         }
3048                         num_bytes = key.objectid - bytenr;
3049                 }
3050                 path->slots[0]++;
3051         }
3052         ret = 0;
3053
3054         if (num_bytes) {
3055                 fprintf(stderr, "There are no extents for csum range "
3056                         "%Lu-%Lu\n", bytenr, bytenr+num_bytes);
3057                 ret = 1;
3058         }
3059
3060         btrfs_free_path(path);
3061         return ret;
3062 }
3063
3064 static int check_csums(struct btrfs_root *root)
3065 {
3066         struct btrfs_path *path;
3067         struct extent_buffer *leaf;
3068         struct btrfs_key key;
3069         u64 offset = 0, num_bytes = 0;
3070         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
3071         int errors = 0;
3072         int ret;
3073
3074         root = root->fs_info->csum_root;
3075
3076         key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
3077         key.type = BTRFS_EXTENT_CSUM_KEY;
3078         key.offset = 0;
3079
3080         path = btrfs_alloc_path();
3081         if (!path)
3082                 return -ENOMEM;
3083
3084         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3085         if (ret < 0) {
3086                 fprintf(stderr, "Error searching csum tree %d\n", ret);
3087                 btrfs_free_path(path);
3088                 return ret;
3089         }
3090
3091         if (ret > 0 && path->slots[0])
3092                 path->slots[0]--;
3093         ret = 0;
3094
3095         while (1) {
3096                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3097                         ret = btrfs_next_leaf(root, path);
3098                         if (ret < 0) {
3099                                 fprintf(stderr, "Error going to next leaf "
3100                                         "%d\n", ret);
3101                                 break;
3102                         }
3103                         if (ret)
3104                                 break;
3105                 }
3106                 leaf = path->nodes[0];
3107
3108                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3109                 if (key.type != BTRFS_EXTENT_CSUM_KEY) {
3110                         path->slots[0]++;
3111                         continue;
3112                 }
3113
3114                 if (!num_bytes) {
3115                         offset = key.offset;
3116                 } else if (key.offset != offset + num_bytes) {
3117                         ret = check_extent_exists(root, offset, num_bytes);
3118                         if (ret) {
3119                                 fprintf(stderr, "Csum exists for %Lu-%Lu but "
3120                                         "there is no extent record\n",
3121                                         offset, offset+num_bytes);
3122                                 errors++;
3123                         }
3124                         offset = key.offset;
3125                         num_bytes = 0;
3126                 }
3127
3128                 num_bytes += (btrfs_item_size_nr(leaf, path->slots[0]) /
3129                               csum_size) * root->sectorsize;
3130                 path->slots[0]++;
3131         }
3132
3133         btrfs_free_path(path);
3134         return errors;
3135 }
3136
3137 static int run_next_block(struct btrfs_root *root,
3138                           struct block_info *bits,
3139                           int bits_nr,
3140                           u64 *last,
3141                           struct cache_tree *pending,
3142                           struct cache_tree *seen,
3143                           struct cache_tree *reada,
3144                           struct cache_tree *nodes,
3145                           struct cache_tree *extent_cache)
3146 {
3147         struct extent_buffer *buf;
3148         u64 bytenr;
3149         u32 size;
3150         u64 parent;
3151         u64 owner;
3152         u64 flags;
3153         int ret;
3154         int i;
3155         int nritems;
3156         struct btrfs_key key;
3157         struct cache_extent *cache;
3158         int reada_bits;
3159
3160         ret = pick_next_pending(pending, reada, nodes, *last, bits,
3161                                 bits_nr, &reada_bits);
3162         if (ret == 0) {
3163                 return 1;
3164         }
3165         if (!reada_bits) {
3166                 for(i = 0; i < ret; i++) {
3167                         insert_cache_extent(reada, bits[i].start,
3168                                             bits[i].size);
3169
3170                         /* fixme, get the parent transid */
3171                         readahead_tree_block(root, bits[i].start,
3172                                              bits[i].size, 0);
3173                 }
3174         }
3175         *last = bits[0].start;
3176         bytenr = bits[0].start;
3177         size = bits[0].size;
3178
3179         cache = find_cache_extent(pending, bytenr, size);
3180         if (cache) {
3181                 remove_cache_extent(pending, cache);
3182                 free(cache);
3183         }
3184         cache = find_cache_extent(reada, bytenr, size);
3185         if (cache) {
3186                 remove_cache_extent(reada, cache);
3187                 free(cache);
3188         }
3189         cache = find_cache_extent(nodes, bytenr, size);
3190         if (cache) {
3191                 remove_cache_extent(nodes, cache);
3192                 free(cache);
3193         }
3194
3195         /* fixme, get the real parent transid */
3196         buf = read_tree_block(root, bytenr, size, 0);
3197         if (!extent_buffer_uptodate(buf)) {
3198                 record_bad_block_io(root->fs_info,
3199                                     extent_cache, bytenr, size);
3200                 goto out;
3201         }
3202
3203         nritems = btrfs_header_nritems(buf);
3204
3205         ret = btrfs_lookup_extent_info(NULL, root, bytenr,
3206                                        btrfs_header_level(buf), 1, NULL,
3207                                        &flags);
3208         if (ret < 0)
3209                 flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
3210
3211         if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
3212                 parent = bytenr;
3213                 owner = 0;
3214         } else {
3215                 parent = 0;
3216                 owner = btrfs_header_owner(buf);
3217         }
3218
3219         ret = check_block(root, extent_cache, buf, flags);
3220         if (ret)
3221                 goto out;
3222
3223         if (btrfs_is_leaf(buf)) {
3224                 btree_space_waste += btrfs_leaf_free_space(root, buf);
3225                 for (i = 0; i < nritems; i++) {
3226                         struct btrfs_file_extent_item *fi;
3227                         btrfs_item_key_to_cpu(buf, &key, i);
3228                         if (key.type == BTRFS_EXTENT_ITEM_KEY) {
3229                                 process_extent_item(root, extent_cache, buf,
3230                                                     i);
3231                                 continue;
3232                         }
3233                         if (key.type == BTRFS_METADATA_ITEM_KEY) {
3234                                 process_extent_item(root, extent_cache, buf,
3235                                                     i);
3236                                 continue;
3237                         }
3238                         if (key.type == BTRFS_EXTENT_CSUM_KEY) {
3239                                 total_csum_bytes +=
3240                                         btrfs_item_size_nr(buf, i);
3241                                 continue;
3242                         }
3243                         if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
3244                                 continue;
3245                         }
3246                         if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
3247 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3248                                 process_extent_ref_v0(extent_cache, buf, i);
3249 #else
3250                                 BUG();
3251 #endif
3252                                 continue;
3253                         }
3254
3255                         if (key.type == BTRFS_TREE_BLOCK_REF_KEY) {
3256                                 add_tree_backref(extent_cache, key.objectid, 0,
3257                                                  key.offset, 0);
3258                                 continue;
3259                         }
3260                         if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
3261                                 add_tree_backref(extent_cache, key.objectid,
3262                                                  key.offset, 0, 0);
3263                                 continue;
3264                         }
3265                         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
3266                                 struct btrfs_extent_data_ref *ref;
3267                                 ref = btrfs_item_ptr(buf, i,
3268                                                 struct btrfs_extent_data_ref);
3269                                 add_data_backref(extent_cache,
3270                                         key.objectid, 0,
3271                                         btrfs_extent_data_ref_root(buf, ref),
3272                                         btrfs_extent_data_ref_objectid(buf,
3273                                                                        ref),
3274                                         btrfs_extent_data_ref_offset(buf, ref),
3275                                         btrfs_extent_data_ref_count(buf, ref),
3276                                         0, root->sectorsize);
3277                                 continue;
3278                         }
3279                         if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
3280                                 struct btrfs_shared_data_ref *ref;
3281                                 ref = btrfs_item_ptr(buf, i,
3282                                                 struct btrfs_shared_data_ref);
3283                                 add_data_backref(extent_cache,
3284                                         key.objectid, key.offset, 0, 0, 0,
3285                                         btrfs_shared_data_ref_count(buf, ref),
3286                                         0, root->sectorsize);
3287                                 continue;
3288                         }
3289                         if (key.type != BTRFS_EXTENT_DATA_KEY)
3290                                 continue;
3291                         fi = btrfs_item_ptr(buf, i,
3292                                             struct btrfs_file_extent_item);
3293                         if (btrfs_file_extent_type(buf, fi) ==
3294                             BTRFS_FILE_EXTENT_INLINE)
3295                                 continue;
3296                         if (btrfs_file_extent_disk_bytenr(buf, fi) == 0)
3297                                 continue;
3298
3299                         data_bytes_allocated +=
3300                                 btrfs_file_extent_disk_num_bytes(buf, fi);
3301                         if (data_bytes_allocated < root->sectorsize) {
3302                                 abort();
3303                         }
3304                         data_bytes_referenced +=
3305                                 btrfs_file_extent_num_bytes(buf, fi);
3306                         add_data_backref(extent_cache,
3307                                 btrfs_file_extent_disk_bytenr(buf, fi),
3308                                 parent, owner, key.objectid, key.offset -
3309                                 btrfs_file_extent_offset(buf, fi), 1, 1,
3310                                 btrfs_file_extent_disk_num_bytes(buf, fi));
3311                         BUG_ON(ret);
3312                 }
3313         } else {
3314                 int level;
3315                 struct btrfs_key first_key;
3316
3317                 first_key.objectid = 0;
3318
3319                 if (nritems > 0)
3320                         btrfs_item_key_to_cpu(buf, &first_key, 0);
3321                 level = btrfs_header_level(buf);
3322                 for (i = 0; i < nritems; i++) {
3323                         u64 ptr = btrfs_node_blockptr(buf, i);
3324                         u32 size = btrfs_level_size(root, level - 1);
3325                         btrfs_node_key_to_cpu(buf, &key, i);
3326                         ret = add_extent_rec(extent_cache, &key,
3327                                              ptr, size, 0, 0, 1, 0, 1, 0,
3328                                              size);
3329                         BUG_ON(ret);
3330
3331                         add_tree_backref(extent_cache, ptr, parent, owner, 1);
3332
3333                         if (level > 1) {
3334                                 add_pending(nodes, seen, ptr, size);
3335                         } else {
3336                                 add_pending(pending, seen, ptr, size);
3337                         }
3338                 }
3339                 btree_space_waste += (BTRFS_NODEPTRS_PER_BLOCK(root) -
3340                                       nritems) * sizeof(struct btrfs_key_ptr);
3341         }
3342         total_btree_bytes += buf->len;
3343         if (fs_root_objectid(btrfs_header_owner(buf)))
3344                 total_fs_tree_bytes += buf->len;
3345         if (btrfs_header_owner(buf) == BTRFS_EXTENT_TREE_OBJECTID)
3346                 total_extent_tree_bytes += buf->len;
3347         if (!found_old_backref &&
3348             btrfs_header_owner(buf) == BTRFS_TREE_RELOC_OBJECTID &&
3349             btrfs_header_backref_rev(buf) == BTRFS_MIXED_BACKREF_REV &&
3350             !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))
3351                 found_old_backref = 1;
3352 out:
3353         free_extent_buffer(buf);
3354         return 0;
3355 }
3356
3357 static int add_root_to_pending(struct extent_buffer *buf,
3358                                struct cache_tree *extent_cache,
3359                                struct cache_tree *pending,
3360                                struct cache_tree *seen,
3361                                struct cache_tree *nodes,
3362                                struct btrfs_key *root_key)
3363 {
3364         if (btrfs_header_level(buf) > 0)
3365                 add_pending(nodes, seen, buf->start, buf->len);
3366         else
3367                 add_pending(pending, seen, buf->start, buf->len);
3368         add_extent_rec(extent_cache, NULL, buf->start, buf->len,
3369                        0, 1, 1, 0, 1, 0, buf->len);
3370
3371         if (root_key->objectid == BTRFS_TREE_RELOC_OBJECTID ||
3372             btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
3373                 add_tree_backref(extent_cache, buf->start, buf->start,
3374                                  0, 1);
3375         else
3376                 add_tree_backref(extent_cache, buf->start, 0,
3377                                  root_key->objectid, 1);
3378         return 0;
3379 }
3380
3381 /* as we fix the tree, we might be deleting blocks that
3382  * we're tracking for repair.  This hook makes sure we
3383  * remove any backrefs for blocks as we are fixing them.
3384  */
3385 static int free_extent_hook(struct btrfs_trans_handle *trans,
3386                             struct btrfs_root *root,
3387                             u64 bytenr, u64 num_bytes, u64 parent,
3388                             u64 root_objectid, u64 owner, u64 offset,
3389                             int refs_to_drop)
3390 {
3391         struct extent_record *rec;
3392         struct cache_extent *cache;
3393         int is_data;
3394         struct cache_tree *extent_cache = root->fs_info->fsck_extent_cache;
3395
3396         is_data = owner >= BTRFS_FIRST_FREE_OBJECTID;
3397         cache = find_cache_extent(extent_cache, bytenr, num_bytes);
3398         if (!cache)
3399                 return 0;
3400
3401         rec = container_of(cache, struct extent_record, cache);
3402         if (is_data) {
3403                 struct data_backref *back;
3404                 back = find_data_backref(rec, parent, root_objectid, owner,
3405                                          offset, 1, bytenr, num_bytes);
3406                 if (!back)
3407                         goto out;
3408                 if (back->node.found_ref) {
3409                         back->found_ref -= refs_to_drop;
3410                         if (rec->refs)
3411                                 rec->refs -= refs_to_drop;
3412                 }
3413                 if (back->node.found_extent_tree) {
3414                         back->num_refs -= refs_to_drop;
3415                         if (rec->extent_item_refs)
3416                                 rec->extent_item_refs -= refs_to_drop;
3417                 }
3418                 if (back->found_ref == 0)
3419                         back->node.found_ref = 0;
3420                 if (back->num_refs == 0)
3421                         back->node.found_extent_tree = 0;
3422
3423                 if (!back->node.found_extent_tree && back->node.found_ref) {
3424                         list_del(&back->node.list);
3425                         free(back);
3426                 }
3427         } else {
3428                 struct tree_backref *back;
3429                 back = find_tree_backref(rec, parent, root_objectid);
3430                 if (!back)
3431                         goto out;
3432                 if (back->node.found_ref) {
3433                         if (rec->refs)
3434                                 rec->refs--;
3435                         back->node.found_ref = 0;
3436                 }
3437                 if (back->node.found_extent_tree) {
3438                         if (rec->extent_item_refs)
3439                                 rec->extent_item_refs--;
3440                         back->node.found_extent_tree = 0;
3441                 }
3442                 if (!back->node.found_extent_tree && back->node.found_ref) {
3443                         list_del(&back->node.list);
3444                         free(back);
3445                 }
3446         }
3447         maybe_free_extent_rec(extent_cache, rec);
3448 out:
3449         return 0;
3450 }
3451
3452 static int delete_extent_records(struct btrfs_trans_handle *trans,
3453                                  struct btrfs_root *root,
3454                                  struct btrfs_path *path,
3455                                  u64 bytenr, u64 new_len)
3456 {
3457         struct btrfs_key key;
3458         struct btrfs_key found_key;
3459         struct extent_buffer *leaf;
3460         int ret;
3461         int slot;
3462
3463
3464         key.objectid = bytenr;
3465         key.type = (u8)-1;
3466         key.offset = (u64)-1;
3467
3468         while(1) {
3469                 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
3470                                         &key, path, 0, 1);
3471                 if (ret < 0)
3472                         break;
3473
3474                 if (ret > 0) {
3475                         ret = 0;
3476                         if (path->slots[0] == 0)
3477                                 break;
3478                         path->slots[0]--;
3479                 }
3480                 ret = 0;
3481
3482                 leaf = path->nodes[0];
3483                 slot = path->slots[0];
3484
3485                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3486                 if (found_key.objectid != bytenr)
3487                         break;
3488
3489                 if (found_key.type != BTRFS_EXTENT_ITEM_KEY &&
3490                     found_key.type != BTRFS_METADATA_ITEM_KEY &&
3491                     found_key.type != BTRFS_TREE_BLOCK_REF_KEY &&
3492                     found_key.type != BTRFS_EXTENT_DATA_REF_KEY &&
3493                     found_key.type != BTRFS_EXTENT_REF_V0_KEY &&
3494                     found_key.type != BTRFS_SHARED_BLOCK_REF_KEY &&
3495                     found_key.type != BTRFS_SHARED_DATA_REF_KEY) {
3496                         btrfs_release_path(NULL, path);
3497                         if (found_key.type == 0) {
3498                                 if (found_key.offset == 0)
3499                                         break;
3500                                 key.offset = found_key.offset - 1;
3501                                 key.type = found_key.type;
3502                         }
3503                         key.type = found_key.type - 1;
3504                         key.offset = (u64)-1;
3505                         continue;
3506                 }
3507
3508                 fprintf(stderr, "repair deleting extent record: key %Lu %u %Lu\n",
3509                         found_key.objectid, found_key.type, found_key.offset);
3510
3511                 ret = btrfs_del_item(trans, root->fs_info->extent_root, path);
3512                 if (ret)
3513                         break;
3514                 btrfs_release_path(NULL, path);
3515
3516                 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
3517                     found_key.type == BTRFS_METADATA_ITEM_KEY) {
3518                         u64 bytes = (found_key.type == BTRFS_EXTENT_ITEM_KEY) ?
3519                                 found_key.offset : root->leafsize;
3520
3521                         ret = btrfs_update_block_group(trans, root, bytenr,
3522                                                        bytes, 0, 0);
3523                         if (ret)
3524                                 break;
3525                 }
3526         }
3527
3528         btrfs_release_path(NULL, path);
3529         return ret;
3530 }
3531
3532 /*
3533  * for a single backref, this will allocate a new extent
3534  * and add the backref to it.
3535  */
3536 static int record_extent(struct btrfs_trans_handle *trans,
3537                          struct btrfs_fs_info *info,
3538                          struct btrfs_path *path,
3539                          struct extent_record *rec,
3540                          struct extent_backref *back,
3541                          int allocated, u64 flags)
3542 {
3543         int ret;
3544         struct btrfs_root *extent_root = info->extent_root;
3545         struct extent_buffer *leaf;
3546         struct btrfs_key ins_key;
3547         struct btrfs_extent_item *ei;
3548         struct tree_backref *tback;
3549         struct data_backref *dback;
3550         struct btrfs_tree_block_info *bi;
3551
3552         if (!back->is_data)
3553                 rec->max_size = max_t(u64, rec->max_size,
3554                                     info->extent_root->leafsize);
3555
3556         if (!allocated) {
3557                 u32 item_size = sizeof(*ei);
3558
3559                 if (!back->is_data)
3560                         item_size += sizeof(*bi);
3561
3562                 ins_key.objectid = rec->start;
3563                 ins_key.offset = rec->max_size;
3564                 ins_key.type = BTRFS_EXTENT_ITEM_KEY;
3565
3566                 ret = btrfs_insert_empty_item(trans, extent_root, path,
3567                                         &ins_key, item_size);
3568                 if (ret)
3569                         goto fail;
3570
3571                 leaf = path->nodes[0];
3572                 ei = btrfs_item_ptr(leaf, path->slots[0],
3573                                     struct btrfs_extent_item);
3574
3575                 btrfs_set_extent_refs(leaf, ei, 0);
3576                 btrfs_set_extent_generation(leaf, ei, rec->generation);
3577
3578                 if (back->is_data) {
3579                         btrfs_set_extent_flags(leaf, ei,
3580                                                BTRFS_EXTENT_FLAG_DATA);
3581                 } else {
3582                         struct btrfs_disk_key copy_key;;
3583
3584                         tback = (struct tree_backref *)back;
3585                         bi = (struct btrfs_tree_block_info *)(ei + 1);
3586                         memset_extent_buffer(leaf, 0, (unsigned long)bi,
3587                                              sizeof(*bi));
3588                         memset(&copy_key, 0, sizeof(copy_key));
3589
3590                         copy_key.objectid = le64_to_cpu(rec->info_objectid);
3591                         btrfs_set_tree_block_level(leaf, bi, rec->info_level);
3592                         btrfs_set_tree_block_key(leaf, bi, &copy_key);
3593
3594                         btrfs_set_extent_flags(leaf, ei,
3595                                                BTRFS_EXTENT_FLAG_TREE_BLOCK | flags);
3596                 }
3597
3598                 btrfs_mark_buffer_dirty(leaf);
3599                 ret = btrfs_update_block_group(trans, extent_root, rec->start,
3600                                                rec->max_size, 1, 0);
3601                 if (ret)
3602                         goto fail;
3603                 btrfs_release_path(NULL, path);
3604         }
3605
3606         if (back->is_data) {
3607                 u64 parent;
3608                 int i;
3609
3610                 dback = (struct data_backref *)back;
3611                 if (back->full_backref)
3612                         parent = dback->parent;
3613                 else
3614                         parent = 0;
3615
3616                 for (i = 0; i < dback->found_ref; i++) {
3617                         /* if parent != 0, we're doing a full backref
3618                          * passing BTRFS_FIRST_FREE_OBJECTID as the owner
3619                          * just makes the backref allocator create a data
3620                          * backref
3621                          */
3622                         ret = btrfs_inc_extent_ref(trans, info->extent_root,
3623                                                    rec->start, rec->max_size,
3624                                                    parent,
3625                                                    dback->root,
3626                                                    parent ?
3627                                                    BTRFS_FIRST_FREE_OBJECTID :
3628                                                    dback->owner,
3629                                                    dback->offset);
3630                         if (ret)
3631                                 break;
3632                 }
3633                 fprintf(stderr, "adding new data backref"
3634                                 " on %llu %s %llu owner %llu"
3635                                 " offset %llu found %d\n",
3636                                 (unsigned long long)rec->start,
3637                                 back->full_backref ?
3638                                 "parent" : "root",
3639                                 back->full_backref ?
3640                                 (unsigned long long)parent :
3641                                 (unsigned long long)dback->root,
3642                                 (unsigned long long)dback->owner,
3643                                 (unsigned long long)dback->offset,
3644                                 dback->found_ref);
3645         } else {
3646                 u64 parent;
3647
3648                 tback = (struct tree_backref *)back;
3649                 if (back->full_backref)
3650                         parent = tback->parent;
3651                 else
3652                         parent = 0;
3653
3654                 ret = btrfs_inc_extent_ref(trans, info->extent_root,
3655                                            rec->start, rec->max_size,
3656                                            parent, tback->root, 0, 0);
3657                 fprintf(stderr, "adding new tree backref on "
3658                         "start %llu len %llu parent %llu root %llu\n",
3659                         rec->start, rec->max_size, tback->parent, tback->root);
3660         }
3661         if (ret)
3662                 goto fail;
3663 fail:
3664         btrfs_release_path(NULL, path);
3665         return ret;
3666 }
3667
3668 struct extent_entry {
3669         u64 bytenr;
3670         u64 bytes;
3671         int count;
3672         struct list_head list;
3673 };
3674
3675 static struct extent_entry *find_entry(struct list_head *entries,
3676                                        u64 bytenr, u64 bytes)
3677 {
3678         struct extent_entry *entry = NULL;
3679
3680         list_for_each_entry(entry, entries, list) {
3681                 if (entry->bytenr == bytenr && entry->bytes == bytes)
3682                         return entry;
3683         }
3684
3685         return NULL;
3686 }
3687
3688 static struct extent_entry *find_most_right_entry(struct list_head *entries)
3689 {
3690         struct extent_entry *entry, *best = NULL, *prev = NULL;
3691
3692         list_for_each_entry(entry, entries, list) {
3693                 if (!prev) {
3694                         prev = entry;
3695                         continue;
3696                 }
3697
3698                 /*
3699                  * If our current entry == best then we can't be sure our best
3700                  * is really the best, so we need to keep searching.
3701                  */
3702                 if (best && best->count == entry->count) {
3703                         prev = entry;
3704                         best = NULL;
3705                         continue;
3706                 }
3707
3708                 /* Prev == entry, not good enough, have to keep searching */
3709                 if (prev->count == entry->count)
3710                         continue;
3711
3712                 if (!best)
3713                         best = (prev->count > entry->count) ? prev : entry;
3714                 else if (best->count < entry->count)
3715                         best = entry;
3716                 prev = entry;
3717         }
3718
3719         return best;
3720 }
3721
3722 static int repair_ref(struct btrfs_trans_handle *trans,
3723                       struct btrfs_fs_info *info, struct btrfs_path *path,
3724                       struct data_backref *dback, struct extent_entry *entry)
3725 {
3726         struct btrfs_root *root;
3727         struct btrfs_file_extent_item *fi;
3728         struct extent_buffer *leaf;
3729         struct btrfs_key key;
3730         u64 bytenr, bytes;
3731         int ret;
3732
3733         key.objectid = dback->root;
3734         key.type = BTRFS_ROOT_ITEM_KEY;
3735         key.offset = (u64)-1;
3736         root = btrfs_read_fs_root(info, &key);
3737         if (IS_ERR(root)) {
3738                 fprintf(stderr, "Couldn't find root for our ref\n");
3739                 return -EINVAL;
3740         }
3741
3742         /*
3743          * The backref points to the original offset of the extent if it was
3744          * split, so we need to search down to the offset we have and then walk
3745          * forward until we find the backref we're looking for.
3746          */
3747         key.objectid = dback->owner;
3748         key.type = BTRFS_EXTENT_DATA_KEY;
3749         key.offset = dback->offset;
3750         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3751         if (ret < 0) {
3752                 fprintf(stderr, "Error looking up ref %d\n", ret);
3753                 return ret;
3754         }
3755
3756         while (1) {
3757                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3758                         ret = btrfs_next_leaf(root, path);
3759                         if (ret) {
3760                                 fprintf(stderr, "Couldn't find our ref, next\n");
3761                                 return -EINVAL;
3762                         }
3763                 }
3764                 leaf = path->nodes[0];
3765                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3766                 if (key.objectid != dback->owner ||
3767                     key.type != BTRFS_EXTENT_DATA_KEY) {
3768                         fprintf(stderr, "Couldn't find our ref, search\n");
3769                         return -EINVAL;
3770                 }
3771                 fi = btrfs_item_ptr(leaf, path->slots[0],
3772                                     struct btrfs_file_extent_item);
3773                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
3774                 bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
3775
3776                 if (bytenr == dback->disk_bytenr && bytes == dback->bytes)
3777                         break;
3778                 path->slots[0]++;
3779         }
3780
3781         btrfs_release_path(root, path);
3782
3783         /*
3784          * Have to make sure that this root gets updated when we commit the
3785          * transaction
3786          */
3787         root->track_dirty = 1;
3788         if (root->last_trans != trans->transid) {
3789                 root->last_trans = trans->transid;
3790                 root->commit_root = root->node;
3791                 extent_buffer_get(root->node);
3792         }
3793
3794         /*
3795          * Ok we have the key of the file extent we want to fix, now we can cow
3796          * down to the thing and fix it.
3797          */
3798         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
3799         if (ret < 0) {
3800                 fprintf(stderr, "Error cowing down to ref [%Lu, %u, %Lu]: %d\n",
3801                         key.objectid, key.type, key.offset, ret);
3802                 return ret;
3803         }
3804         if (ret > 0) {
3805                 fprintf(stderr, "Well that's odd, we just found this key "
3806                         "[%Lu, %u, %Lu]\n", key.objectid, key.type,
3807                         key.offset);
3808                 return -EINVAL;
3809         }
3810         leaf = path->nodes[0];
3811         fi = btrfs_item_ptr(leaf, path->slots[0],
3812                             struct btrfs_file_extent_item);
3813
3814         if (btrfs_file_extent_compression(leaf, fi) &&
3815             dback->disk_bytenr != entry->bytenr) {
3816                 fprintf(stderr, "Ref doesn't match the record start and is "
3817                         "compressed, please take a btrfs-image of this file "
3818                         "system and send it to a btrfs developer so they can "
3819                         "complete this functionality for bytenr %Lu\n",
3820                         dback->disk_bytenr);
3821                 return -EINVAL;
3822         }
3823
3824         if (dback->disk_bytenr > entry->bytenr) {
3825                 u64 off_diff, offset;
3826
3827                 off_diff = dback->disk_bytenr - entry->bytenr;
3828                 offset = btrfs_file_extent_offset(leaf, fi);
3829                 if (dback->disk_bytenr + offset +
3830                     btrfs_file_extent_num_bytes(leaf, fi) >
3831                     entry->bytenr + entry->bytes) {
3832                         fprintf(stderr, "Ref is past the entry end, please "
3833                                 "take a btrfs-image of this file system and "
3834                                 "send it to a btrfs developer, ref %Lu\n",
3835                                 dback->disk_bytenr);
3836                         return -EINVAL;
3837                 }
3838                 offset += off_diff;
3839                 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
3840                 btrfs_set_file_extent_offset(leaf, fi, offset);
3841         } else if (dback->disk_bytenr < entry->bytenr) {
3842                 u64 offset;
3843
3844                 offset = btrfs_file_extent_offset(leaf, fi);
3845                 if (dback->disk_bytenr + offset < entry->bytenr) {
3846                         fprintf(stderr, "Ref is before the entry start, please"
3847                                 " take a btrfs-image of this file system and "
3848                                 "send it to a btrfs developer, ref %Lu\n",
3849                                 dback->disk_bytenr);
3850                         return -EINVAL;
3851                 }
3852
3853                 offset += dback->disk_bytenr;
3854                 offset -= entry->bytenr;
3855                 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
3856                 btrfs_set_file_extent_offset(leaf, fi, offset);
3857         }
3858
3859         btrfs_set_file_extent_disk_num_bytes(leaf, fi, entry->bytes);
3860
3861         /*
3862          * Chances are if disk_num_bytes were wrong then so is ram_bytes, but
3863          * only do this if we aren't using compression, otherwise it's a
3864          * trickier case.
3865          */
3866         if (!btrfs_file_extent_compression(leaf, fi))
3867                 btrfs_set_file_extent_ram_bytes(leaf, fi, entry->bytes);
3868         else
3869                 printf("ram bytes may be wrong?\n");
3870         btrfs_mark_buffer_dirty(leaf);
3871         btrfs_release_path(root, path);
3872         return 0;
3873 }
3874
3875 static int verify_backrefs(struct btrfs_trans_handle *trans,
3876                            struct btrfs_fs_info *info, struct btrfs_path *path,
3877                            struct extent_record *rec)
3878 {
3879         struct extent_backref *back;
3880         struct data_backref *dback;
3881         struct extent_entry *entry, *best = NULL;
3882         LIST_HEAD(entries);
3883         int nr_entries = 0;
3884         int ret = 0;
3885
3886         /*
3887          * Metadata is easy and the backrefs should always agree on bytenr and
3888          * size, if not we've got bigger issues.
3889          */
3890         if (rec->metadata)
3891                 return 0;
3892
3893         list_for_each_entry(back, &rec->backrefs, list) {
3894                 dback = (struct data_backref *)back;
3895                 /*
3896                  * We only pay attention to backrefs that we found a real
3897                  * backref for.
3898                  */
3899                 if (dback->found_ref == 0)
3900                         continue;
3901                 if (back->full_backref)
3902                         continue;
3903
3904                 /*
3905                  * For now we only catch when the bytes don't match, not the
3906                  * bytenr.  We can easily do this at the same time, but I want
3907                  * to have a fs image to test on before we just add repair
3908                  * functionality willy-nilly so we know we won't screw up the
3909                  * repair.
3910                  */
3911
3912                 entry = find_entry(&entries, dback->disk_bytenr,
3913                                    dback->bytes);
3914                 if (!entry) {
3915                         entry = malloc(sizeof(struct extent_entry));
3916                         if (!entry) {
3917                                 ret = -ENOMEM;
3918                                 goto out;
3919                         }
3920                         memset(entry, 0, sizeof(*entry));
3921                         entry->bytenr = dback->disk_bytenr;
3922                         entry->bytes = dback->bytes;
3923                         list_add_tail(&entry->list, &entries);
3924                         nr_entries++;
3925                 }
3926                 entry->count++;
3927         }
3928
3929         /* Yay all the backrefs agree, carry on good sir */
3930         if (nr_entries <= 1)
3931                 goto out;
3932
3933         fprintf(stderr, "attempting to repair backref discrepency for bytenr "
3934                 "%Lu\n", rec->start);
3935
3936         /*
3937          * First we want to see if the backrefs can agree amongst themselves who
3938          * is right, so figure out which one of the entries has the highest
3939          * count.
3940          */
3941         best = find_most_right_entry(&entries);
3942
3943         /*
3944          * Ok so we may have an even split between what the backrefs think, so
3945          * this is where we use the extent ref to see what it thinks.
3946          */
3947         if (!best) {
3948                 entry = find_entry(&entries, rec->start, rec->nr);
3949                 if (!entry) {
3950                         fprintf(stderr, "Backrefs don't agree with eachother "
3951                                 "and extent record doesn't agree with anybody,"
3952                                 " so we can't fix bytenr %Lu bytes %Lu\n",
3953                                 rec->start, rec->nr);
3954                         ret = -EINVAL;
3955                         goto out;
3956                 }
3957                 entry->count++;
3958                 best = find_most_right_entry(&entries);
3959                 if (!best) {
3960                         fprintf(stderr, "Backrefs and extent record evenly "
3961                                 "split on who is right, this is going to "
3962                                 "require user input to fix bytenr %Lu bytes "
3963                                 "%Lu\n", rec->start, rec->nr);
3964                         ret = -EINVAL;
3965                         goto out;
3966                 }
3967         }
3968
3969         /*
3970          * I don't think this can happen currently as we'll abort() if we catch
3971          * this case higher up, but in case somebody removes that we still can't
3972          * deal with it properly here yet, so just bail out of that's the case.
3973          */
3974         if (best->bytenr != rec->start) {
3975                 fprintf(stderr, "Extent start and backref starts don't match, "
3976                         "please use btrfs-image on this file system and send "
3977                         "it to a btrfs developer so they can make fsck fix "
3978                         "this particular case.  bytenr is %Lu, bytes is %Lu\n",
3979                         rec->start, rec->nr);
3980                 ret = -EINVAL;
3981                 goto out;
3982         }
3983
3984         /*
3985          * Ok great we all agreed on an extent record, let's go find the real
3986          * references and fix up the ones that don't match.
3987          */
3988         list_for_each_entry(back, &rec->backrefs, list) {
3989                 dback = (struct data_backref *)back;
3990
3991                 /*
3992                  * Still ignoring backrefs that don't have a real ref attached
3993                  * to them.
3994                  */
3995                 if (dback->found_ref == 0)
3996                         continue;
3997                 if (back->full_backref)
3998                         continue;
3999
4000                 if (dback->bytes == best->bytes &&
4001                     dback->disk_bytenr == best->bytenr)
4002                         continue;
4003
4004                 ret = repair_ref(trans, info, path, dback, best);
4005                 if (ret)
4006                         goto out;
4007         }
4008
4009         /*
4010          * Ok we messed with the actual refs, which means we need to drop our
4011          * entire cache and go back and rescan.  I know this is a huge pain and
4012          * adds a lot of extra work, but it's the only way to be safe.  Once all
4013          * the backrefs agree we may not need to do anything to the extent
4014          * record itself.
4015          */
4016         ret = -EAGAIN;
4017 out:
4018         while (!list_empty(&entries)) {
4019                 entry = list_entry(entries.next, struct extent_entry, list);
4020                 list_del_init(&entry->list);
4021                 free(entry);
4022         }
4023         return ret;
4024 }
4025
4026 static int delete_duplicate_records(struct btrfs_trans_handle *trans,
4027                                     struct btrfs_root *root,
4028                                     struct extent_record *rec)
4029 {
4030         LIST_HEAD(delete_list);
4031         struct btrfs_path *path;
4032         struct extent_record *tmp, *good, *n;
4033         int ret = 0;
4034         struct btrfs_key key;
4035
4036         path = btrfs_alloc_path();
4037         if (!path) {
4038                 ret = -ENOMEM;
4039                 goto out;
4040         }
4041
4042         good = rec;
4043         /* Find the record that covers all of the duplicates. */
4044         list_for_each_entry(tmp, &rec->dups, list) {
4045                 if (good->start < tmp->start)
4046                         continue;
4047                 if (good->nr > tmp->nr)
4048                         continue;
4049
4050                 if (tmp->start + tmp->nr < good->start + good->nr) {
4051                         fprintf(stderr, "Ok we have overlapping extents that "
4052                                 "aren't completely covered by eachother, this "
4053                                 "is going to require more careful thought.  "
4054                                 "The extents are [%Lu-%Lu] and [%Lu-%Lu]\n",
4055                                 tmp->start, tmp->nr, good->start, good->nr);
4056                         abort();
4057                 }
4058                 good = tmp;
4059         }
4060
4061         if (good != rec)
4062                 list_add_tail(&rec->list, &delete_list);
4063
4064         list_for_each_entry_safe(tmp, n, &rec->dups, list) {
4065                 if (tmp == good)
4066                         continue;
4067                 list_move_tail(&tmp->list, &delete_list);
4068         }
4069
4070         root = root->fs_info->extent_root;
4071         list_for_each_entry(tmp, &delete_list, list) {
4072                 key.objectid = tmp->start;
4073                 key.type = BTRFS_EXTENT_ITEM_KEY;
4074                 key.offset = tmp->nr;
4075
4076                 /* Shouldn't happen but just in case */
4077                 if (tmp->metadata) {
4078                         fprintf(stderr, "Well this shouldn't happen, extent "
4079                                 "record overlaps but is metadata? "
4080                                 "[%Lu, %Lu]\n", tmp->start, tmp->nr);
4081                         abort();
4082                 }
4083
4084                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4085                 if (ret) {
4086                         if (ret > 0)
4087                                 ret = -EINVAL;
4088                         goto out;
4089                 }
4090                 ret = btrfs_del_item(trans, root, path);
4091                 if (ret)
4092                         goto out;
4093                 btrfs_release_path(root, path);
4094         }
4095
4096 out:
4097         while (!list_empty(&delete_list)) {
4098                 tmp = list_entry(delete_list.next, struct extent_record, list);
4099                 list_del_init(&tmp->list);
4100                 if (tmp == rec)
4101                         continue;
4102                 free(tmp);
4103         }
4104
4105         while (!list_empty(&rec->dups)) {
4106                 tmp = list_entry(rec->dups.next, struct extent_record, list);
4107                 list_del_init(&tmp->list);
4108                 free(tmp);
4109         }
4110
4111         btrfs_free_path(path);
4112
4113         return ret;
4114 }
4115
4116 /*
4117  * when an incorrect extent item is found, this will delete
4118  * all of the existing entries for it and recreate them
4119  * based on what the tree scan found.
4120  */
4121 static int fixup_extent_refs(struct btrfs_trans_handle *trans,
4122                              struct btrfs_fs_info *info,
4123                              struct extent_record *rec)
4124 {
4125         int ret;
4126         struct btrfs_path *path;
4127         struct list_head *cur = rec->backrefs.next;
4128         struct cache_extent *cache;
4129         struct extent_backref *back;
4130         int allocated = 0;
4131         u64 flags = 0;
4132
4133         /* remember our flags for recreating the extent */
4134         ret = btrfs_lookup_extent_info(NULL, info->extent_root, rec->start,
4135                                        rec->max_size, rec->metadata, NULL,
4136                                        &flags);
4137         if (ret < 0)
4138                 flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
4139
4140         path = btrfs_alloc_path();
4141
4142         /* step one, make sure all of the backrefs agree */
4143         ret = verify_backrefs(trans, info, path, rec);
4144         if (ret < 0)
4145                 goto out;
4146
4147         /* step two, delete all the existing records */
4148         ret = delete_extent_records(trans, info->extent_root, path,
4149                                     rec->start, rec->max_size);
4150
4151         if (ret < 0)
4152                 goto out;
4153
4154         /* was this block corrupt?  If so, don't add references to it */
4155         cache = find_cache_extent(info->corrupt_blocks, rec->start, rec->max_size);
4156         if (cache) {
4157                 ret = 0;
4158                 goto out;
4159         }
4160
4161         /* step three, recreate all the refs we did find */
4162         while(cur != &rec->backrefs) {
4163                 back = list_entry(cur, struct extent_backref, list);
4164                 cur = cur->next;
4165
4166                 /*
4167                  * if we didn't find any references, don't create a
4168                  * new extent record
4169                  */
4170                 if (!back->found_ref)
4171                         continue;
4172
4173                 ret = record_extent(trans, info, path, rec, back, allocated, flags);
4174                 allocated = 1;
4175
4176                 if (ret)
4177                         goto out;
4178         }
4179 out:
4180         btrfs_free_path(path);
4181         return ret;
4182 }
4183
4184 /* right now we only prune from the extent allocation tree */
4185 static int prune_one_block(struct btrfs_trans_handle *trans,
4186                            struct btrfs_fs_info *info,
4187                            struct btrfs_corrupt_block *corrupt)
4188 {
4189         int ret;
4190         struct btrfs_path path;
4191         struct extent_buffer *eb;
4192         u64 found;
4193         int slot;
4194         int nritems;
4195         int level = corrupt->level + 1;
4196
4197         btrfs_init_path(&path);
4198 again:
4199         /* we want to stop at the parent to our busted block */
4200         path.lowest_level = level;
4201
4202         ret = btrfs_search_slot(trans, info->extent_root,
4203                                 &corrupt->key, &path, -1, 1);
4204
4205         if (ret < 0)
4206                 goto out;
4207
4208         eb = path.nodes[level];
4209         if (!eb) {
4210                 ret = -ENOENT;
4211                 goto out;
4212         }
4213
4214         /*
4215          * hopefully the search gave us the block we want to prune,
4216          * lets try that first
4217          */
4218         slot = path.slots[level];
4219         found =  btrfs_node_blockptr(eb, slot);
4220         if (found == corrupt->cache.start)
4221                 goto del_ptr;
4222
4223         nritems = btrfs_header_nritems(eb);
4224
4225         /* the search failed, lets scan this node and hope we find it */
4226         for (slot = 0; slot < nritems; slot++) {
4227                 found =  btrfs_node_blockptr(eb, slot);
4228                 if (found == corrupt->cache.start)
4229                         goto del_ptr;
4230         }
4231         /*
4232          * we couldn't find the bad block.  TODO, search all the nodes for pointers
4233          * to this block
4234          */
4235         if (eb == info->extent_root->node) {
4236                 ret = -ENOENT;
4237                 goto out;
4238         } else {
4239                 level++;
4240                 btrfs_release_path(NULL, &path);
4241                 goto again;
4242         }
4243
4244 del_ptr:
4245         printk("deleting pointer to block %Lu\n", corrupt->cache.start);
4246         ret = btrfs_del_ptr(trans, info->extent_root, &path, level, slot);
4247
4248 out:
4249         btrfs_release_path(NULL, &path);
4250         return ret;
4251 }
4252
4253 static int prune_corrupt_blocks(struct btrfs_trans_handle *trans,
4254                                 struct btrfs_fs_info *info)
4255 {
4256         struct cache_extent *cache;
4257         struct btrfs_corrupt_block *corrupt;
4258
4259         cache = find_first_cache_extent(info->corrupt_blocks, 0);
4260         while (1) {
4261                 if (!cache)
4262                         break;
4263                 corrupt = container_of(cache, struct btrfs_corrupt_block, cache);
4264                 prune_one_block(trans, info, corrupt);
4265                 cache = next_cache_extent(cache);
4266         }
4267         return 0;
4268 }
4269
4270 static void free_corrupt_blocks(struct btrfs_fs_info *info)
4271 {
4272         struct cache_extent *cache;
4273         struct btrfs_corrupt_block *corrupt;
4274
4275         while (1) {
4276                 cache = find_first_cache_extent(info->corrupt_blocks, 0);
4277                 if (!cache)
4278                         break;
4279                 corrupt = container_of(cache, struct btrfs_corrupt_block, cache);
4280                 remove_cache_extent(info->corrupt_blocks, cache);
4281                 free(corrupt);
4282         }
4283 }
4284
4285 static int check_block_group(struct btrfs_trans_handle *trans,
4286                               struct btrfs_fs_info *info,
4287                               struct map_lookup *map,
4288                               int *reinit)
4289 {
4290         struct btrfs_key key;
4291         struct btrfs_path path;
4292         int ret;
4293
4294         key.objectid = map->ce.start;
4295         key.offset = map->ce.size;
4296         key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
4297
4298         btrfs_init_path(&path);
4299         ret = btrfs_search_slot(NULL, info->extent_root,
4300                                 &key, &path, 0, 0);
4301         btrfs_release_path(NULL, &path);
4302         if (ret <= 0)
4303                 goto out;
4304
4305         ret = btrfs_make_block_group(trans, info->extent_root, 0, map->type,
4306                                BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4307                                key.objectid, key.offset);
4308         *reinit = 1;
4309 out:
4310         return ret;
4311 }
4312
4313 static int check_block_groups(struct btrfs_trans_handle *trans,
4314                               struct btrfs_fs_info *info, int *reinit)
4315 {
4316         struct cache_extent *ce;
4317         struct map_lookup *map;
4318         struct btrfs_mapping_tree *map_tree = &info->mapping_tree;
4319
4320         /* this isn't quite working */
4321         return 0;
4322
4323         ce = find_first_cache_extent(&map_tree->cache_tree, 0);
4324         while (1) {
4325                 if (!ce)
4326                         break;
4327                 map = container_of(ce, struct map_lookup, ce);
4328                 check_block_group(trans, info, map, reinit);
4329                 ce = next_cache_extent(ce);
4330         }
4331         return 0;
4332 }
4333
4334 static void reset_cached_block_groups(struct btrfs_fs_info *fs_info)
4335 {
4336         struct btrfs_block_group_cache *cache;
4337         u64 start, end;
4338         int ret;
4339
4340         while (1) {
4341                 ret = find_first_extent_bit(&fs_info->free_space_cache, 0,
4342                                             &start, &end, EXTENT_DIRTY);
4343                 if (ret)
4344                         break;
4345                 clear_extent_dirty(&fs_info->free_space_cache, start, end,
4346                                    GFP_NOFS);
4347         }
4348
4349         start = 0;
4350         while (1) {
4351                 cache = btrfs_lookup_first_block_group(fs_info, start);
4352                 if (!cache)
4353                         break;
4354                 if (cache->cached)
4355                         cache->cached = 0;
4356                 start = cache->key.objectid + cache->key.offset;
4357         }
4358 }
4359
4360 static int check_extent_refs(struct btrfs_trans_handle *trans,
4361                              struct btrfs_root *root,
4362                              struct cache_tree *extent_cache, int repair)
4363 {
4364         struct extent_record *rec, *n;
4365         struct cache_extent *cache;
4366         int err = 0;
4367         int ret = 0;
4368         int fixed = 0;
4369         int reinit = 0;
4370         int had_dups = 0;
4371
4372         if (repair) {
4373                 /*
4374                  * if we're doing a repair, we have to make sure
4375                  * we don't allocate from the problem extents.
4376                  * In the worst case, this will be all the
4377                  * extents in the FS
4378                  */
4379                 cache = find_first_cache_extent(extent_cache, 0);
4380                 while(cache) {
4381                         rec = container_of(cache, struct extent_record, cache);
4382                         btrfs_pin_extent(root->fs_info,
4383                                          rec->start, rec->max_size);
4384                         cache = next_cache_extent(cache);
4385                 }
4386
4387                 /* pin down all the corrupted blocks too */
4388                 cache = find_first_cache_extent(root->fs_info->corrupt_blocks, 0);
4389                 while(cache) {
4390                         rec = container_of(cache, struct extent_record, cache);
4391                         btrfs_pin_extent(root->fs_info,
4392                                          rec->start, rec->max_size);
4393                         cache = next_cache_extent(cache);
4394                 }
4395                 prune_corrupt_blocks(trans, root->fs_info);
4396                 check_block_groups(trans, root->fs_info, &reinit);
4397                 if (reinit)
4398                         btrfs_read_block_groups(root->fs_info->extent_root);
4399                 reset_cached_block_groups(root->fs_info);
4400         }
4401
4402         /*
4403          * We need to delete any duplicate entries we find first otherwise we
4404          * could mess up the extent tree when we have backrefs that actually
4405          * belong to a different extent item and not the weird duplicate one.
4406          */
4407         list_for_each_entry_safe(rec, n, &duplicate_extents, list) {
4408                 if (!repair)
4409                         break;
4410                 list_del_init(&rec->list);
4411                 had_dups = 1;
4412                 ret = delete_duplicate_records(trans, root, rec);
4413                 if (ret)
4414                         return ret;
4415         }
4416
4417         if (had_dups)
4418                 return -EAGAIN;
4419
4420         while(1) {
4421                 fixed = 0;
4422                 cache = find_first_cache_extent(extent_cache, 0);
4423                 if (!cache)
4424                         break;
4425                 rec = container_of(cache, struct extent_record, cache);
4426                 if (rec->has_duplicate) {
4427                         fprintf(stderr, "extent item %llu has multiple extent "
4428                                 "items\n", (unsigned long long)rec->start);
4429                         err = 1;
4430                 }
4431
4432                 if (rec->refs != rec->extent_item_refs) {
4433                         fprintf(stderr, "ref mismatch on [%llu %llu] ",
4434                                 (unsigned long long)rec->start,
4435                                 (unsigned long long)rec->nr);
4436                         fprintf(stderr, "extent item %llu, found %llu\n",
4437                                 (unsigned long long)rec->extent_item_refs,
4438                                 (unsigned long long)rec->refs);
4439                         if (!fixed && repair) {
4440                                 ret = fixup_extent_refs(trans, root->fs_info, rec);
4441                                 if (ret)
4442                                         goto repair_abort;
4443                                 fixed = 1;
4444                         }
4445                         err = 1;
4446
4447                 }
4448                 if (all_backpointers_checked(rec, 1)) {
4449                         fprintf(stderr, "backpointer mismatch on [%llu %llu]\n",
4450                                 (unsigned long long)rec->start,
4451                                 (unsigned long long)rec->nr);
4452
4453                         if (!fixed && repair) {
4454                                 ret = fixup_extent_refs(trans, root->fs_info, rec);
4455                                 if (ret)
4456                                         goto repair_abort;
4457                                 fixed = 1;
4458                         }
4459
4460                         err = 1;
4461                 }
4462                 if (!rec->owner_ref_checked) {
4463                         fprintf(stderr, "owner ref check failed [%llu %llu]\n",
4464                                 (unsigned long long)rec->start,
4465                                 (unsigned long long)rec->nr);
4466                         if (!fixed && repair) {
4467                                 ret = fixup_extent_refs(trans, root->fs_info, rec);
4468                                 if (ret)
4469                                         goto repair_abort;
4470                                 fixed = 1;
4471                         }
4472                         err = 1;
4473                 }
4474
4475                 remove_cache_extent(extent_cache, cache);
4476                 free_all_extent_backrefs(rec);
4477                 free(rec);
4478         }
4479 repair_abort:
4480         if (repair) {
4481                 if (ret && ret != -EAGAIN) {
4482                         fprintf(stderr, "failed to repair damaged filesystem, aborting\n");
4483                         exit(1);
4484                 } else if (!ret) {
4485                         btrfs_fix_block_accounting(trans, root);
4486                 }
4487                 if (err)
4488                         fprintf(stderr, "repaired damaged extent references\n");
4489                 return ret;
4490         }
4491         return err;
4492 }
4493
4494 static void free_cache_tree(struct cache_tree *tree)
4495 {
4496         struct cache_extent *cache;
4497
4498         while (1) {
4499                 cache = find_first_cache_extent(tree, 0);
4500                 if (!cache)
4501                         break;
4502                 remove_cache_extent(tree, cache);
4503                 free(cache);
4504         }
4505 }
4506
4507 static int check_extents(struct btrfs_root *root, int repair)
4508 {
4509         struct cache_tree extent_cache;
4510         struct cache_tree seen;
4511         struct cache_tree pending;
4512         struct cache_tree reada;
4513         struct cache_tree nodes;
4514         struct cache_tree corrupt_blocks;
4515         struct btrfs_path path;
4516         struct btrfs_key key;
4517         struct btrfs_key found_key;
4518         int ret;
4519         u64 last = 0;
4520         struct block_info *bits;
4521         int bits_nr;
4522         struct extent_buffer *leaf;
4523         struct btrfs_trans_handle *trans = NULL;
4524         int slot;
4525         struct btrfs_root_item ri;
4526
4527         cache_tree_init(&extent_cache);
4528         cache_tree_init(&seen);
4529         cache_tree_init(&pending);
4530         cache_tree_init(&nodes);
4531         cache_tree_init(&reada);
4532         cache_tree_init(&corrupt_blocks);
4533
4534         if (repair) {
4535                 trans = btrfs_start_transaction(root, 1);
4536                 if (IS_ERR(trans)) {
4537                         fprintf(stderr, "Error starting transaction\n");
4538                         return PTR_ERR(trans);
4539                 }
4540                 root->fs_info->fsck_extent_cache = &extent_cache;
4541                 root->fs_info->free_extent_hook = free_extent_hook;
4542                 root->fs_info->corrupt_blocks = &corrupt_blocks;
4543         }
4544
4545         bits_nr = 1024;
4546         bits = malloc(bits_nr * sizeof(struct block_info));
4547         if (!bits) {
4548                 perror("malloc");
4549                 exit(1);
4550         }
4551
4552 again:
4553         add_root_to_pending(root->fs_info->tree_root->node,
4554                             &extent_cache, &pending, &seen, &nodes,
4555                             &root->fs_info->tree_root->root_key);
4556
4557         add_root_to_pending(root->fs_info->chunk_root->node,
4558                             &extent_cache, &pending, &seen, &nodes,
4559                             &root->fs_info->chunk_root->root_key);
4560
4561         btrfs_init_path(&path);
4562         key.offset = 0;
4563         key.objectid = 0;
4564         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
4565         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
4566                                         &key, &path, 0, 0);
4567         BUG_ON(ret < 0);
4568         while(1) {
4569                 leaf = path.nodes[0];
4570                 slot = path.slots[0];
4571                 if (slot >= btrfs_header_nritems(path.nodes[0])) {
4572                         ret = btrfs_next_leaf(root, &path);
4573                         if (ret != 0)
4574                                 break;
4575                         leaf = path.nodes[0];
4576                         slot = path.slots[0];
4577                 }
4578                 btrfs_item_key_to_cpu(leaf, &found_key, path.slots[0]);
4579                 if (btrfs_key_type(&found_key) == BTRFS_ROOT_ITEM_KEY) {
4580                         unsigned long offset;
4581                         struct extent_buffer *buf;
4582
4583                         offset = btrfs_item_ptr_offset(leaf, path.slots[0]);
4584                         read_extent_buffer(leaf, &ri, offset, sizeof(ri));
4585                         buf = read_tree_block(root->fs_info->tree_root,
4586                                               btrfs_root_bytenr(&ri),
4587                                               btrfs_level_size(root,
4588                                                btrfs_root_level(&ri)), 0);
4589                         add_root_to_pending(buf, &extent_cache, &pending,
4590                                             &seen, &nodes, &found_key);
4591                         free_extent_buffer(buf);
4592                 }
4593                 path.slots[0]++;
4594         }
4595         btrfs_release_path(root, &path);
4596         while(1) {
4597                 ret = run_next_block(root, bits, bits_nr, &last, &pending,
4598                                      &seen, &reada, &nodes, &extent_cache);
4599                 if (ret != 0)
4600                         break;
4601         }
4602         ret = check_extent_refs(trans, root, &extent_cache, repair);
4603
4604         if (ret == -EAGAIN) {
4605                 ret = btrfs_commit_transaction(trans, root);
4606                 if (ret)
4607                         goto out;
4608
4609                 trans = btrfs_start_transaction(root, 1);
4610                 if (IS_ERR(trans)) {
4611                         ret = PTR_ERR(trans);
4612                         goto out;
4613                 }
4614
4615                 free_corrupt_blocks(root->fs_info);
4616                 free_cache_tree(&seen);
4617                 free_cache_tree(&pending);
4618                 free_cache_tree(&reada);
4619                 free_cache_tree(&nodes);
4620                 free_extent_cache(root->fs_info, &extent_cache);
4621                 goto again;
4622         }
4623
4624         if (trans) {
4625                 int err;
4626
4627                 err = btrfs_commit_transaction(trans, root);
4628                 if (!ret)
4629                         ret = err;
4630         }
4631 out:
4632         if (repair) {
4633                 free_corrupt_blocks(root->fs_info);
4634                 root->fs_info->fsck_extent_cache = NULL;
4635                 root->fs_info->free_extent_hook = NULL;
4636                 root->fs_info->corrupt_blocks = NULL;
4637         }
4638         free(bits);
4639         return ret;
4640 }
4641
4642 static int pin_down_tree_blocks(struct btrfs_fs_info *fs_info,
4643                                 struct extent_buffer *eb, int tree_root)
4644 {
4645         struct extent_buffer *tmp;
4646         struct btrfs_root_item *ri;
4647         struct btrfs_key key;
4648         u64 bytenr;
4649         u32 leafsize;
4650         int level = btrfs_header_level(eb);
4651         int nritems;
4652         int ret;
4653         int i;
4654
4655         btrfs_pin_extent(fs_info, eb->start, eb->len);
4656
4657         leafsize = btrfs_super_leafsize(fs_info->super_copy);
4658         nritems = btrfs_header_nritems(eb);
4659         for (i = 0; i < nritems; i++) {
4660                 if (level == 0) {
4661                         btrfs_item_key_to_cpu(eb, &key, i);
4662                         if (key.type != BTRFS_ROOT_ITEM_KEY)
4663                                 continue;
4664                         /* Skip the extent root and reloc roots */
4665                         if (key.objectid == BTRFS_EXTENT_TREE_OBJECTID ||
4666                             key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
4667                             key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
4668                                 continue;
4669                         ri = btrfs_item_ptr(eb, i, struct btrfs_root_item);
4670                         bytenr = btrfs_disk_root_bytenr(eb, ri);
4671
4672                         /*
4673                          * If at any point we start needing the real root we
4674                          * will have to build a stump root for the root we are
4675                          * in, but for now this doesn't actually use the root so
4676                          * just pass in extent_root.
4677                          */
4678                         tmp = read_tree_block(fs_info->extent_root, bytenr,
4679                                               leafsize, 0);
4680                         if (!tmp) {
4681                                 fprintf(stderr, "Error reading root block\n");
4682                                 return -EIO;
4683                         }
4684                         ret = pin_down_tree_blocks(fs_info, tmp, 0);
4685                         free_extent_buffer(tmp);
4686                         if (ret)
4687                                 return ret;
4688                 } else {
4689                         bytenr = btrfs_node_blockptr(eb, i);
4690
4691                         /* If we aren't the tree root don't read the block */
4692                         if (level == 1 && !tree_root) {
4693                                 btrfs_pin_extent(fs_info, bytenr, leafsize);
4694                                 continue;
4695                         }
4696
4697                         tmp = read_tree_block(fs_info->extent_root, bytenr,
4698                                               leafsize, 0);
4699                         if (!tmp) {
4700                                 fprintf(stderr, "Error reading tree block\n");
4701                                 return -EIO;
4702                         }
4703                         ret = pin_down_tree_blocks(fs_info, tmp, tree_root);
4704                         free_extent_buffer(tmp);
4705                         if (ret)
4706                                 return ret;
4707                 }
4708         }
4709
4710         return 0;
4711 }
4712
4713 static int pin_metadata_blocks(struct btrfs_fs_info *fs_info)
4714 {
4715         int ret;
4716
4717         ret = pin_down_tree_blocks(fs_info, fs_info->chunk_root->node, 0);
4718         if (ret)
4719                 return ret;
4720
4721         return pin_down_tree_blocks(fs_info, fs_info->tree_root->node, 1);
4722 }
4723
4724 static int reset_block_groups(struct btrfs_fs_info *fs_info)
4725 {
4726         struct btrfs_path *path;
4727         struct extent_buffer *leaf;
4728         struct btrfs_chunk *chunk;
4729         struct btrfs_key key;
4730         int ret;
4731
4732         path = btrfs_alloc_path();
4733         if (!path)
4734                 return -ENOMEM;
4735
4736         key.objectid = 0;
4737         key.type = BTRFS_CHUNK_ITEM_KEY;
4738         key.offset = 0;
4739
4740         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
4741         if (ret < 0) {
4742                 btrfs_free_path(path);
4743                 return ret;
4744         }
4745
4746         /*
4747          * We do this in case the block groups were screwed up and had alloc
4748          * bits that aren't actually set on the chunks.  This happens with
4749          * restored images every time and could happen in real life I guess.
4750          */
4751         fs_info->avail_data_alloc_bits = 0;
4752         fs_info->avail_metadata_alloc_bits = 0;
4753         fs_info->avail_system_alloc_bits = 0;
4754
4755         /* First we need to create the in-memory block groups */
4756         while (1) {
4757                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
4758                         ret = btrfs_next_leaf(fs_info->chunk_root, path);
4759                         if (ret < 0) {
4760                                 btrfs_free_path(path);
4761                                 return ret;
4762                         }
4763                         if (ret) {
4764                                 ret = 0;
4765                                 break;
4766                         }
4767                 }
4768                 leaf = path->nodes[0];
4769                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4770                 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
4771                         path->slots[0]++;
4772                         continue;
4773                 }
4774
4775                 chunk = btrfs_item_ptr(leaf, path->slots[0],
4776                                        struct btrfs_chunk);
4777                 btrfs_add_block_group(fs_info, 0,
4778                                       btrfs_chunk_type(leaf, chunk),
4779                                       key.objectid, key.offset,
4780                                       btrfs_chunk_length(leaf, chunk));
4781                 path->slots[0]++;
4782         }
4783
4784         btrfs_free_path(path);
4785         return 0;
4786 }
4787
4788 static int reset_balance(struct btrfs_trans_handle *trans,
4789                          struct btrfs_fs_info *fs_info)
4790 {
4791         struct btrfs_root *root = fs_info->tree_root;
4792         struct btrfs_path *path;
4793         struct extent_buffer *leaf;
4794         struct btrfs_key key;
4795         int del_slot, del_nr = 0;
4796         int ret;
4797         int found = 0;
4798
4799         path = btrfs_alloc_path();
4800         if (!path)
4801                 return -ENOMEM;
4802
4803         key.objectid = BTRFS_BALANCE_OBJECTID;
4804         key.type = BTRFS_BALANCE_ITEM_KEY;
4805         key.offset = 0;
4806
4807         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4808         if (ret) {
4809                 if (ret > 0)
4810                         ret = 0;
4811                 goto out;
4812         }
4813
4814         ret = btrfs_del_item(trans, root, path);
4815         if (ret)
4816                 goto out;
4817         btrfs_release_path(root, path);
4818
4819         key.objectid = BTRFS_TREE_RELOC_OBJECTID;
4820         key.type = BTRFS_ROOT_ITEM_KEY;
4821         key.offset = 0;
4822
4823         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4824         if (ret < 0)
4825                 goto out;
4826         while (1) {
4827                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
4828                         if (!found)
4829                                 break;
4830
4831                         if (del_nr) {
4832                                 ret = btrfs_del_items(trans, root, path,
4833                                                       del_slot, del_nr);
4834                                 del_nr = 0;
4835                                 if (ret)
4836                                         goto out;
4837                         }
4838                         key.offset++;
4839                         btrfs_release_path(root, path);
4840
4841                         found = 0;
4842                         ret = btrfs_search_slot(trans, root, &key, path,
4843                                                 -1, 1);
4844                         if (ret < 0)
4845                                 goto out;
4846                         continue;
4847                 }
4848                 found = 1;
4849                 leaf = path->nodes[0];
4850                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4851                 if (key.objectid > BTRFS_TREE_RELOC_OBJECTID)
4852                         break;
4853                 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
4854                         path->slots[0]++;
4855                         continue;
4856                 }
4857                 if (!del_nr) {
4858                         del_slot = path->slots[0];
4859                         del_nr = 1;
4860                 } else {
4861                         del_nr++;
4862                 }
4863                 path->slots[0]++;
4864         }
4865
4866         if (del_nr) {
4867                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
4868                 if (ret)
4869                         goto out;
4870         }
4871         btrfs_release_path(root, path);
4872
4873         key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
4874         key.type = BTRFS_ROOT_ITEM_KEY;
4875         key.offset = (u64)-1;
4876         root = btrfs_read_fs_root(fs_info, &key);
4877         if (IS_ERR(root)) {
4878                 fprintf(stderr, "Error reading data reloc tree\n");
4879                 return PTR_ERR(root);
4880         }
4881         root->track_dirty = 1;
4882         if (root->last_trans != trans->transid) {
4883                 root->last_trans = trans->transid;
4884                 root->commit_root = root->node;
4885                 extent_buffer_get(root->node);
4886         }
4887         ret = btrfs_fsck_reinit_root(trans, root, 0);
4888 out:
4889         btrfs_free_path(path);
4890         return ret;
4891 }
4892
4893 static int reinit_extent_tree(struct btrfs_fs_info *fs_info)
4894 {
4895         struct btrfs_trans_handle *trans;
4896         u64 start = 0;
4897         int ret;
4898
4899         /*
4900          * The only reason we don't do this is because right now we're just
4901          * walking the trees we find and pinning down their bytes, we don't look
4902          * at any of the leaves.  In order to do mixed groups we'd have to check
4903          * the leaves of any fs roots and pin down the bytes for any file
4904          * extents we find.  Not hard but why do it if we don't have to?
4905          */
4906         if (btrfs_fs_incompat(fs_info, BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)) {
4907                 fprintf(stderr, "We don't support re-initing the extent tree "
4908                         "for mixed block groups yet, please notify a btrfs "
4909                         "developer you want to do this so they can add this "
4910                         "functionality.\n");
4911                 return -EINVAL;
4912         }
4913
4914         trans = btrfs_start_transaction(fs_info->extent_root, 1);
4915         if (IS_ERR(trans)) {
4916                 fprintf(stderr, "Error starting transaction\n");
4917                 return PTR_ERR(trans);
4918         }
4919
4920         /*
4921          * first we need to walk all of the trees except the extent tree and pin
4922          * down the bytes that are in use so we don't overwrite any existing
4923          * metadata.
4924          */
4925         ret = pin_metadata_blocks(fs_info);
4926         if (ret) {
4927                 fprintf(stderr, "error pinning down used bytes\n");
4928                 return ret;
4929         }
4930
4931         /*
4932          * Need to drop all the block groups since we're going to recreate all
4933          * of them again.
4934          */
4935         btrfs_free_block_groups(fs_info);
4936         ret = reset_block_groups(fs_info);
4937         if (ret) {
4938                 fprintf(stderr, "error resetting the block groups\n");
4939                 return ret;
4940         }
4941
4942         /* Ok we can allocate now, reinit the extent root */
4943         ret = btrfs_fsck_reinit_root(trans, fs_info->extent_root, 1);
4944         if (ret) {
4945                 fprintf(stderr, "extent root initialization failed\n");
4946                 /*
4947                  * When the transaction code is updated we should end the
4948                  * transaction, but for now progs only knows about commit so
4949                  * just return an error.
4950                  */
4951                 return ret;
4952         }
4953
4954         ret = reset_balance(trans, fs_info);
4955         if (ret) {
4956                 fprintf(stderr, "error reseting the pending balance\n");
4957                 return ret;
4958         }
4959
4960         /*
4961          * Now we have all the in-memory block groups setup so we can make
4962          * allocations properly, and the metadata we care about is safe since we
4963          * pinned all of it above.
4964          */
4965         while (1) {
4966                 struct btrfs_block_group_cache *cache;
4967
4968                 cache = btrfs_lookup_first_block_group(fs_info, start);
4969                 if (!cache)
4970                         break;
4971                 start = cache->key.objectid + cache->key.offset;
4972                 ret = btrfs_insert_item(trans, fs_info->extent_root,
4973                                         &cache->key, &cache->item,
4974                                         sizeof(cache->item));
4975                 if (ret) {
4976                         fprintf(stderr, "Error adding block group\n");
4977                         return ret;
4978                 }
4979                 btrfs_extent_post_op(trans, fs_info->extent_root);
4980         }
4981
4982         /*
4983          * Ok now we commit and run the normal fsck, which will add extent
4984          * entries for all of the items it finds.
4985          */
4986         return btrfs_commit_transaction(trans, fs_info->extent_root);
4987 }
4988
4989 static struct option long_options[] = {
4990         { "super", 1, NULL, 's' },
4991         { "repair", 0, NULL, 0 },
4992         { "init-csum-tree", 0, NULL, 0 },
4993         { "init-extent-tree", 0, NULL, 0 },
4994         { 0, 0, 0, 0}
4995 };
4996
4997 const char * const cmd_check_usage[] = {
4998         "btrfs check [options] <device>",
4999         "Check an unmounted btrfs filesystem.",
5000         "",
5001         "-s|--super <superblock>     use this superblock copy",
5002         "--repair                    try to repair the filesystem",
5003         "--init-csum-tree            create a new CRC tree",
5004         "--init-extent-tree          create a new extent tree",
5005         NULL
5006 };
5007
5008 int cmd_check(int argc, char **argv)
5009 {
5010         struct cache_tree root_cache;
5011         struct btrfs_root *root;
5012         struct btrfs_fs_info *info;
5013         u64 bytenr = 0;
5014         char uuidbuf[37];
5015         int ret;
5016         int num;
5017         int repair = 0;
5018         int option_index = 0;
5019         int init_csum_tree = 0;
5020         int init_extent_tree = 0;
5021         int rw = 0;
5022
5023         while(1) {
5024                 int c;
5025                 c = getopt_long(argc, argv, "as:", long_options,
5026                                 &option_index);
5027                 if (c < 0)
5028                         break;
5029                 switch(c) {
5030                         case 'a': /* ignored */ break;
5031                         case 's':
5032                                 num = atol(optarg);
5033                                 bytenr = btrfs_sb_offset(num);
5034                                 printf("using SB copy %d, bytenr %llu\n", num,
5035                                        (unsigned long long)bytenr);
5036                                 break;
5037                         case '?':
5038                         case 'h':
5039                                 usage(cmd_check_usage);
5040                 }
5041                 if (option_index == 1) {
5042                         printf("enabling repair mode\n");
5043                         repair = 1;
5044                         rw = 1;
5045                 } else if (option_index == 2) {
5046                         printf("Creating a new CRC tree\n");
5047                         init_csum_tree = 1;
5048                         rw = 1;
5049                 } else if (option_index == 3) {
5050                         init_extent_tree = 1;
5051                         rw = 1;
5052                         repair = 1;
5053                 }
5054
5055         }
5056         argc = argc - optind;
5057
5058         if (argc != 1)
5059                 usage(cmd_check_usage);
5060
5061         radix_tree_init();
5062         cache_tree_init(&root_cache);
5063
5064         if((ret = check_mounted(argv[optind])) < 0) {
5065                 fprintf(stderr, "Could not check mount status: %s\n", strerror(-ret));
5066                 return ret;
5067         } else if(ret) {
5068                 fprintf(stderr, "%s is currently mounted. Aborting.\n", argv[optind]);
5069                 return -EBUSY;
5070         }
5071
5072         info = open_ctree_fs_info(argv[optind], bytenr, 0, rw, 1);
5073         if (!info) {
5074                 fprintf(stderr, "Couldn't open file system\n");
5075                 return -EIO;
5076         }
5077
5078         uuid_unparse(info->super_copy->fsid, uuidbuf);
5079         printf("Checking filesystem on %s\nUUID: %s\n", argv[optind], uuidbuf);
5080
5081         if (!extent_buffer_uptodate(info->tree_root->node) ||
5082             !extent_buffer_uptodate(info->dev_root->node) ||
5083             !extent_buffer_uptodate(info->extent_root->node) ||
5084             !extent_buffer_uptodate(info->chunk_root->node)) {
5085                 fprintf(stderr, "Critical roots corrupted, unable to fsck the FS\n");
5086                 return -EIO;
5087         }
5088
5089         root = info->fs_root;
5090
5091         if (init_extent_tree) {
5092                 printf("Creating a new extent tree\n");
5093                 ret = reinit_extent_tree(info);
5094                 if (ret)
5095                         return ret;
5096         }
5097         fprintf(stderr, "checking extents\n");
5098         if (init_csum_tree) {
5099                 struct btrfs_trans_handle *trans;
5100
5101                 fprintf(stderr, "Reinit crc root\n");
5102                 trans = btrfs_start_transaction(info->csum_root, 1);
5103                 if (IS_ERR(trans)) {
5104                         fprintf(stderr, "Error starting transaction\n");
5105                         return PTR_ERR(trans);
5106                 }
5107
5108                 ret = btrfs_fsck_reinit_root(trans, info->csum_root, 0);
5109                 if (ret) {
5110                         fprintf(stderr, "crc root initialization failed\n");
5111                         return -EIO;
5112                 }
5113
5114                 ret = btrfs_commit_transaction(trans, root);
5115                 if (ret)
5116                         exit(1);
5117                 goto out;
5118         }
5119         ret = check_extents(root, repair);
5120         if (ret)
5121                 fprintf(stderr, "Errors found in extent allocation tree\n");
5122
5123         fprintf(stderr, "checking free space cache\n");
5124         ret = check_space_cache(root);
5125         if (ret)
5126                 goto out;
5127
5128         fprintf(stderr, "checking fs roots\n");
5129         ret = check_fs_roots(root, &root_cache);
5130         if (ret)
5131                 goto out;
5132
5133         fprintf(stderr, "checking csums\n");
5134         ret = check_csums(root);
5135         if (ret)
5136                 goto out;
5137
5138         fprintf(stderr, "checking root refs\n");
5139         ret = check_root_refs(root, &root_cache);
5140 out:
5141         free_root_recs(&root_cache);
5142         close_ctree(root);
5143
5144         if (found_old_backref) { /*
5145                  * there was a disk format change when mixed
5146                  * backref was in testing tree. The old format
5147                  * existed about one week.
5148                  */
5149                 printf("\n * Found old mixed backref format. "
5150                        "The old format is not supported! *"
5151                        "\n * Please mount the FS in readonly mode, "
5152                        "backup data and re-format the FS. *\n\n");
5153                 ret = 1;
5154         }
5155         printf("found %llu bytes used err is %d\n",
5156                (unsigned long long)bytes_used, ret);
5157         printf("total csum bytes: %llu\n",(unsigned long long)total_csum_bytes);
5158         printf("total tree bytes: %llu\n",
5159                (unsigned long long)total_btree_bytes);
5160         printf("total fs tree bytes: %llu\n",
5161                (unsigned long long)total_fs_tree_bytes);
5162         printf("total extent tree bytes: %llu\n",
5163                (unsigned long long)total_extent_tree_bytes);
5164         printf("btree space waste bytes: %llu\n",
5165                (unsigned long long)btree_space_waste);
5166         printf("file data blocks allocated: %llu\n referenced %llu\n",
5167                 (unsigned long long)data_bytes_allocated,
5168                 (unsigned long long)data_bytes_referenced);
5169         printf("%s\n", BTRFS_BUILD_VERSION);
5170         return ret;
5171 }