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         unsigned int found_rec;
  92         u64 start;
  93         u64 max_size;
  94         u64 nr;
  95         u64 refs;
  96         u64 extent_item_refs;
  97         u64 generation;
  98         u64 info_objectid;
  99         u64 num_duplicates;
 100         u8 info_level;
 101         unsigned int content_checked:1;
 102         unsigned int owner_ref_checked:1;
 103         unsigned int is_root:1;
 104         unsigned int metadata: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             rec->num_duplicates == 0 && !all_backpointers_checked(rec, 0)) {
2014                 remove_cache_extent(extent_cache, &rec->cache);
2015                 free_all_extent_backrefs(rec);
2016                 list_del_init(&rec->list);
2017                 free(rec);
2018         }
2019         return 0;
2020 }
2021
2022 static int check_owner_ref(struct btrfs_root *root,
2023                             struct extent_record *rec,
2024                             struct extent_buffer *buf)
2025 {
2026         struct extent_backref *node;
2027         struct tree_backref *back;
2028         struct btrfs_root *ref_root;
2029         struct btrfs_key key;
2030         struct btrfs_path path;
2031         struct extent_buffer *parent;
2032         int level;
2033         int found = 0;
2034         int ret;
2035
2036         list_for_each_entry(node, &rec->backrefs, list) {
2037                 if (node->is_data)
2038                         continue;
2039                 if (!node->found_ref)
2040                         continue;
2041                 if (node->full_backref)
2042                         continue;
2043                 back = (struct tree_backref *)node;
2044                 if (btrfs_header_owner(buf) == back->root)
2045                         return 0;
2046         }
2047         BUG_ON(rec->is_root);
2048
2049         /* try to find the block by search corresponding fs tree */
2050         key.objectid = btrfs_header_owner(buf);
2051         key.type = BTRFS_ROOT_ITEM_KEY;
2052         key.offset = (u64)-1;
2053
2054         ref_root = btrfs_read_fs_root(root->fs_info, &key);
2055         if (IS_ERR(ref_root))
2056                 return 1;
2057
2058         level = btrfs_header_level(buf);
2059         if (level == 0)
2060                 btrfs_item_key_to_cpu(buf, &key, 0);
2061         else
2062                 btrfs_node_key_to_cpu(buf, &key, 0);
2063
2064         btrfs_init_path(&path);
2065         path.lowest_level = level + 1;
2066         ret = btrfs_search_slot(NULL, ref_root, &key, &path, 0, 0);
2067         if (ret < 0)
2068                 return 0;
2069
2070         parent = path.nodes[level + 1];
2071         if (parent && buf->start == btrfs_node_blockptr(parent,
2072                                                         path.slots[level + 1]))
2073                 found = 1;
2074
2075         btrfs_release_path(ref_root, &path);
2076         return found ? 0 : 1;
2077 }
2078
2079 static int is_extent_tree_record(struct extent_record *rec)
2080 {
2081         struct list_head *cur = rec->backrefs.next;
2082         struct extent_backref *node;
2083         struct tree_backref *back;
2084         int is_extent = 0;
2085
2086         while(cur != &rec->backrefs) {
2087                 node = list_entry(cur, struct extent_backref, list);
2088                 cur = cur->next;
2089                 if (node->is_data)
2090                         return 0;
2091                 back = (struct tree_backref *)node;
2092                 if (node->full_backref)
2093                         return 0;
2094                 if (back->root == BTRFS_EXTENT_TREE_OBJECTID)
2095                         is_extent = 1;
2096         }
2097         return is_extent;
2098 }
2099
2100
2101 static int record_bad_block_io(struct btrfs_fs_info *info,
2102                                struct cache_tree *extent_cache,
2103                                u64 start, u64 len)
2104 {
2105         struct extent_record *rec;
2106         struct cache_extent *cache;
2107         struct btrfs_key key;
2108
2109         cache = find_cache_extent(extent_cache, start, len);
2110         if (!cache)
2111                 return 0;
2112
2113         rec = container_of(cache, struct extent_record, cache);
2114         if (!is_extent_tree_record(rec))
2115                 return 0;
2116
2117         btrfs_disk_key_to_cpu(&key, &rec->parent_key);
2118         return btrfs_add_corrupt_extent_record(info, &key, start, len, 0);
2119 }
2120
2121 static int check_block(struct btrfs_root *root,
2122                        struct cache_tree *extent_cache,
2123                        struct extent_buffer *buf, u64 flags)
2124 {
2125         struct extent_record *rec;
2126         struct cache_extent *cache;
2127         struct btrfs_key key;
2128         int ret = 1;
2129         int level;
2130
2131         cache = find_cache_extent(extent_cache, buf->start, buf->len);
2132         if (!cache)
2133                 return 1;
2134         rec = container_of(cache, struct extent_record, cache);
2135         rec->generation = btrfs_header_generation(buf);
2136
2137         level = btrfs_header_level(buf);
2138         if (btrfs_header_nritems(buf) > 0) {
2139
2140                 if (level == 0)
2141                         btrfs_item_key_to_cpu(buf, &key, 0);
2142                 else
2143                         btrfs_node_key_to_cpu(buf, &key, 0);
2144
2145                 rec->info_objectid = key.objectid;
2146         }
2147         rec->info_level = level;
2148
2149         if (btrfs_is_leaf(buf))
2150                 ret = btrfs_check_leaf(root, &rec->parent_key, buf);
2151         else
2152                 ret = btrfs_check_node(root, &rec->parent_key, buf);
2153
2154         if (ret) {
2155                 fprintf(stderr, "bad block %llu\n",
2156                         (unsigned long long)buf->start);
2157         } else {
2158                 rec->content_checked = 1;
2159                 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
2160                         rec->owner_ref_checked = 1;
2161                 else {
2162                         ret = check_owner_ref(root, rec, buf);
2163                         if (!ret)
2164                                 rec->owner_ref_checked = 1;
2165                 }
2166         }
2167         if (!ret)
2168                 maybe_free_extent_rec(extent_cache, rec);
2169         return ret;
2170 }
2171
2172 static struct tree_backref *find_tree_backref(struct extent_record *rec,
2173                                                 u64 parent, u64 root)
2174 {
2175         struct list_head *cur = rec->backrefs.next;
2176         struct extent_backref *node;
2177         struct tree_backref *back;
2178
2179         while(cur != &rec->backrefs) {
2180                 node = list_entry(cur, struct extent_backref, list);
2181                 cur = cur->next;
2182                 if (node->is_data)
2183                         continue;
2184                 back = (struct tree_backref *)node;
2185                 if (parent > 0) {
2186                         if (!node->full_backref)
2187                                 continue;
2188                         if (parent == back->parent)
2189                                 return back;
2190                 } else {
2191                         if (node->full_backref)
2192                                 continue;
2193                         if (back->root == root)
2194                                 return back;
2195                 }
2196         }
2197         return NULL;
2198 }
2199
2200 static struct tree_backref *alloc_tree_backref(struct extent_record *rec,
2201                                                 u64 parent, u64 root)
2202 {
2203         struct tree_backref *ref = malloc(sizeof(*ref));
2204         memset(&ref->node, 0, sizeof(ref->node));
2205         if (parent > 0) {
2206                 ref->parent = parent;
2207                 ref->node.full_backref = 1;
2208         } else {
2209                 ref->root = root;
2210                 ref->node.full_backref = 0;
2211         }
2212         list_add_tail(&ref->node.list, &rec->backrefs);
2213
2214         return ref;
2215 }
2216
2217 static struct data_backref *find_data_backref(struct extent_record *rec,
2218                                                 u64 parent, u64 root,
2219                                                 u64 owner, u64 offset,
2220                                                 int found_ref,
2221                                                 u64 disk_bytenr, u64 bytes)
2222 {
2223         struct list_head *cur = rec->backrefs.next;
2224         struct extent_backref *node;
2225         struct data_backref *back;
2226
2227         while(cur != &rec->backrefs) {
2228                 node = list_entry(cur, struct extent_backref, list);
2229                 cur = cur->next;
2230                 if (!node->is_data)
2231                         continue;
2232                 back = (struct data_backref *)node;
2233                 if (parent > 0) {
2234                         if (!node->full_backref)
2235                                 continue;
2236                         if (parent == back->parent)
2237                                 return back;
2238                 } else {
2239                         if (node->full_backref)
2240                                 continue;
2241                         if (back->root == root && back->owner == owner &&
2242                             back->offset == offset) {
2243                                 if (found_ref && node->found_ref &&
2244                                     (back->bytes != bytes ||
2245                                     back->disk_bytenr != disk_bytenr))
2246                                         continue;
2247                                 return back;
2248                         }
2249                 }
2250         }
2251         return NULL;
2252 }
2253
2254 static struct data_backref *alloc_data_backref(struct extent_record *rec,
2255                                                 u64 parent, u64 root,
2256                                                 u64 owner, u64 offset,
2257                                                 u64 max_size)
2258 {
2259         struct data_backref *ref = malloc(sizeof(*ref));
2260         memset(&ref->node, 0, sizeof(ref->node));
2261         ref->node.is_data = 1;
2262
2263         if (parent > 0) {
2264                 ref->parent = parent;
2265                 ref->owner = 0;
2266                 ref->offset = 0;
2267                 ref->node.full_backref = 1;
2268         } else {
2269                 ref->root = root;
2270                 ref->owner = owner;
2271                 ref->offset = offset;
2272                 ref->node.full_backref = 0;
2273         }
2274         ref->bytes = max_size;
2275         ref->found_ref = 0;
2276         ref->num_refs = 0;
2277         list_add_tail(&ref->node.list, &rec->backrefs);
2278         if (max_size > rec->max_size)
2279                 rec->max_size = max_size;
2280         return ref;
2281 }
2282
2283 static int add_extent_rec(struct cache_tree *extent_cache,
2284                           struct btrfs_key *parent_key,
2285                           u64 start, u64 nr, u64 extent_item_refs,
2286                           int is_root, int inc_ref, int set_checked,
2287                           int metadata, int extent_rec, u64 max_size)
2288 {
2289         struct extent_record *rec;
2290         struct cache_extent *cache;
2291         int ret = 0;
2292         int dup = 0;
2293
2294         cache = find_cache_extent(extent_cache, start, nr);
2295         if (cache) {
2296                 rec = container_of(cache, struct extent_record, cache);
2297                 if (inc_ref)
2298                         rec->refs++;
2299                 if (rec->nr == 1)
2300                         rec->nr = max(nr, max_size);
2301
2302                 /*
2303                  * We need to make sure to reset nr to whatever the extent
2304                  * record says was the real size, this way we can compare it to
2305                  * the backrefs.
2306                  */
2307                 if (extent_rec) {
2308                         if (start != rec->start || rec->found_rec) {
2309                                 struct extent_record *tmp;
2310
2311                                 dup = 1;
2312                                 if (list_empty(&rec->list))
2313                                         list_add_tail(&rec->list,
2314                                                       &duplicate_extents);
2315
2316                                 /*
2317                                  * We have to do this song and dance in case we
2318                                  * find an extent record that falls inside of
2319                                  * our current extent record but does not have
2320                                  * the same objectid.
2321                                  */
2322                                 tmp = malloc(sizeof(*tmp));
2323                                 if (!tmp)
2324                                         return -ENOMEM;
2325                                 tmp->start = start;
2326                                 tmp->max_size = max_size;
2327                                 tmp->nr = nr;
2328                                 tmp->found_rec = 1;
2329                                 tmp->metadata = metadata;
2330                                 tmp->extent_item_refs = extent_item_refs;
2331                                 INIT_LIST_HEAD(&tmp->list);
2332                                 list_add_tail(&tmp->list, &rec->dups);
2333                                 rec->num_duplicates++;
2334                         } else {
2335                                 rec->nr = nr;
2336                                 rec->found_rec = 1;
2337                         }
2338                 }
2339
2340                 if (extent_item_refs && !dup) {
2341                         if (rec->extent_item_refs) {
2342                                 fprintf(stderr, "block %llu rec "
2343                                         "extent_item_refs %llu, passed %llu\n",
2344                                         (unsigned long long)start,
2345                                         (unsigned long long)
2346                                                         rec->extent_item_refs,
2347                                         (unsigned long long)extent_item_refs);
2348                         }
2349                         rec->extent_item_refs = extent_item_refs;
2350                 }
2351                 if (is_root)
2352                         rec->is_root = 1;
2353                 if (set_checked) {
2354                         rec->content_checked = 1;
2355                         rec->owner_ref_checked = 1;
2356                 }
2357
2358                 if (parent_key)
2359                         btrfs_cpu_key_to_disk(&rec->parent_key, parent_key);
2360
2361                 if (rec->max_size < max_size)
2362                         rec->max_size = max_size;
2363
2364                 maybe_free_extent_rec(extent_cache, rec);
2365                 return ret;
2366         }
2367         rec = malloc(sizeof(*rec));
2368         rec->start = start;
2369         rec->max_size = max_size;
2370         rec->nr = max(nr, max_size);
2371         rec->found_rec = extent_rec;
2372         rec->content_checked = 0;
2373         rec->owner_ref_checked = 0;
2374         rec->num_duplicates = 0;
2375         rec->metadata = metadata;
2376         INIT_LIST_HEAD(&rec->backrefs);
2377         INIT_LIST_HEAD(&rec->dups);
2378         INIT_LIST_HEAD(&rec->list);
2379
2380         if (is_root)
2381                 rec->is_root = 1;
2382         else
2383                 rec->is_root = 0;
2384
2385         if (inc_ref)
2386                 rec->refs = 1;
2387         else
2388                 rec->refs = 0;
2389
2390         if (extent_item_refs)
2391                 rec->extent_item_refs = extent_item_refs;
2392         else
2393                 rec->extent_item_refs = 0;
2394
2395         if (parent_key)
2396                 btrfs_cpu_key_to_disk(&rec->parent_key, parent_key);
2397         else
2398                 memset(&rec->parent_key, 0, sizeof(*parent_key));
2399
2400         rec->cache.start = start;
2401         rec->cache.size = nr;
2402         ret = insert_existing_cache_extent(extent_cache, &rec->cache);
2403         BUG_ON(ret);
2404         bytes_used += nr;
2405         if (set_checked) {
2406                 rec->content_checked = 1;
2407                 rec->owner_ref_checked = 1;
2408         }
2409         return ret;
2410 }
2411
2412 static int add_tree_backref(struct cache_tree *extent_cache, u64 bytenr,
2413                             u64 parent, u64 root, int found_ref)
2414 {
2415         struct extent_record *rec;
2416         struct tree_backref *back;
2417         struct cache_extent *cache;
2418
2419         cache = find_cache_extent(extent_cache, bytenr, 1);
2420         if (!cache) {
2421                 add_extent_rec(extent_cache, NULL, bytenr,
2422                                1, 0, 0, 0, 0, 1, 0, 0);
2423                 cache = find_cache_extent(extent_cache, bytenr, 1);
2424                 if (!cache)
2425                         abort();
2426         }
2427
2428         rec = container_of(cache, struct extent_record, cache);
2429         if (rec->start != bytenr) {
2430                 abort();
2431         }
2432
2433         back = find_tree_backref(rec, parent, root);
2434         if (!back)
2435                 back = alloc_tree_backref(rec, parent, root);
2436
2437         if (found_ref) {
2438                 if (back->node.found_ref) {
2439                         fprintf(stderr, "Extent back ref already exists "
2440                                 "for %llu parent %llu root %llu \n",
2441                                 (unsigned long long)bytenr,
2442                                 (unsigned long long)parent,
2443                                 (unsigned long long)root);
2444                 }
2445                 back->node.found_ref = 1;
2446         } else {
2447                 if (back->node.found_extent_tree) {
2448                         fprintf(stderr, "Extent back ref already exists "
2449                                 "for %llu parent %llu root %llu \n",
2450                                 (unsigned long long)bytenr,
2451                                 (unsigned long long)parent,
2452                                 (unsigned long long)root);
2453                 }
2454                 back->node.found_extent_tree = 1;
2455         }
2456         return 0;
2457 }
2458
2459 static int add_data_backref(struct cache_tree *extent_cache, u64 bytenr,
2460                             u64 parent, u64 root, u64 owner, u64 offset,
2461                             u32 num_refs, int found_ref, u64 max_size)
2462 {
2463         struct extent_record *rec;
2464         struct data_backref *back;
2465         struct cache_extent *cache;
2466
2467         cache = find_cache_extent(extent_cache, bytenr, 1);
2468         if (!cache) {
2469                 add_extent_rec(extent_cache, NULL, bytenr, 1, 0, 0, 0, 0,
2470                                0, 0, max_size);
2471                 cache = find_cache_extent(extent_cache, bytenr, 1);
2472                 if (!cache)
2473                         abort();
2474         }
2475
2476         rec = container_of(cache, struct extent_record, cache);
2477         if (rec->max_size < max_size)
2478                 rec->max_size = max_size;
2479
2480         /*
2481          * If found_ref is set then max_size is the real size and must match the
2482          * existing refs.  So if we have already found a ref then we need to
2483          * make sure that this ref matches the existing one, otherwise we need
2484          * to add a new backref so we can notice that the backrefs don't match
2485          * and we need to figure out who is telling the truth.  This is to
2486          * account for that awful fsync bug I introduced where we'd end up with
2487          * a btrfs_file_extent_item that would have its length include multiple
2488          * prealloc extents or point inside of a prealloc extent.
2489          */
2490         back = find_data_backref(rec, parent, root, owner, offset, found_ref,
2491                                  bytenr, max_size);
2492         if (!back)
2493                 back = alloc_data_backref(rec, parent, root, owner, offset,
2494                                           max_size);
2495
2496         if (found_ref) {
2497                 BUG_ON(num_refs != 1);
2498                 if (back->node.found_ref)
2499                         BUG_ON(back->bytes != max_size);
2500                 back->node.found_ref = 1;
2501                 back->found_ref += 1;
2502                 back->bytes = max_size;
2503                 back->disk_bytenr = bytenr;
2504                 rec->refs += 1;
2505                 rec->content_checked = 1;
2506                 rec->owner_ref_checked = 1;
2507         } else {
2508                 if (back->node.found_extent_tree) {
2509                         fprintf(stderr, "Extent back ref already exists "
2510                                 "for %llu parent %llu root %llu"
2511                                 "owner %llu offset %llu num_refs %lu\n",
2512                                 (unsigned long long)bytenr,
2513                                 (unsigned long long)parent,
2514                                 (unsigned long long)root,
2515                                 (unsigned long long)owner,
2516                                 (unsigned long long)offset,
2517                                 (unsigned long)num_refs);
2518                 }
2519                 back->num_refs = num_refs;
2520                 back->node.found_extent_tree = 1;
2521         }
2522         return 0;
2523 }
2524
2525 static int add_pending(struct cache_tree *pending,
2526                        struct cache_tree *seen, u64 bytenr, u32 size)
2527 {
2528         int ret;
2529         ret = insert_cache_extent(seen, bytenr, size);
2530         if (ret)
2531                 return ret;
2532         insert_cache_extent(pending, bytenr, size);
2533         return 0;
2534 }
2535
2536 static int pick_next_pending(struct cache_tree *pending,
2537                         struct cache_tree *reada,
2538                         struct cache_tree *nodes,
2539                         u64 last, struct block_info *bits, int bits_nr,
2540                         int *reada_bits)
2541 {
2542         unsigned long node_start = last;
2543         struct cache_extent *cache;
2544         int ret;
2545
2546         cache = find_first_cache_extent(reada, 0);
2547         if (cache) {
2548                 bits[0].start = cache->start;
2549                 bits[1].size = cache->size;
2550                 *reada_bits = 1;
2551                 return 1;
2552         }
2553         *reada_bits = 0;
2554         if (node_start > 32768)
2555                 node_start -= 32768;
2556
2557         cache = find_first_cache_extent(nodes, node_start);
2558         if (!cache)
2559                 cache = find_first_cache_extent(nodes, 0);
2560
2561         if (!cache) {
2562                  cache = find_first_cache_extent(pending, 0);
2563                  if (!cache)
2564                          return 0;
2565                  ret = 0;
2566                  do {
2567                          bits[ret].start = cache->start;
2568                          bits[ret].size = cache->size;
2569                          cache = next_cache_extent(cache);
2570                          ret++;
2571                  } while (cache && ret < bits_nr);
2572                  return ret;
2573         }
2574
2575         ret = 0;
2576         do {
2577                 bits[ret].start = cache->start;
2578                 bits[ret].size = cache->size;
2579                 cache = next_cache_extent(cache);
2580                 ret++;
2581         } while (cache && ret < bits_nr);
2582
2583         if (bits_nr - ret > 8) {
2584                 u64 lookup = bits[0].start + bits[0].size;
2585                 struct cache_extent *next;
2586                 next = find_first_cache_extent(pending, lookup);
2587                 while(next) {
2588                         if (next->start - lookup > 32768)
2589                                 break;
2590                         bits[ret].start = next->start;
2591                         bits[ret].size = next->size;
2592                         lookup = next->start + next->size;
2593                         ret++;
2594                         if (ret == bits_nr)
2595                                 break;
2596                         next = next_cache_extent(next);
2597                         if (!next)
2598                                 break;
2599                 }
2600         }
2601         return ret;
2602 }
2603
2604 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2605 static int process_extent_ref_v0(struct cache_tree *extent_cache,
2606                                  struct extent_buffer *leaf, int slot)
2607 {
2608         struct btrfs_extent_ref_v0 *ref0;
2609         struct btrfs_key key;
2610
2611         btrfs_item_key_to_cpu(leaf, &key, slot);
2612         ref0 = btrfs_item_ptr(leaf, slot, struct btrfs_extent_ref_v0);
2613         if (btrfs_ref_objectid_v0(leaf, ref0) < BTRFS_FIRST_FREE_OBJECTID) {
2614                 add_tree_backref(extent_cache, key.objectid, key.offset, 0, 0);
2615         } else {
2616                 add_data_backref(extent_cache, key.objectid, key.offset, 0,
2617                                  0, 0, btrfs_ref_count_v0(leaf, ref0), 0, 0);
2618         }
2619         return 0;
2620 }
2621 #endif
2622
2623 static int process_extent_item(struct btrfs_root *root,
2624                                struct cache_tree *extent_cache,
2625                                struct extent_buffer *eb, int slot)
2626 {
2627         struct btrfs_extent_item *ei;
2628         struct btrfs_extent_inline_ref *iref;
2629         struct btrfs_extent_data_ref *dref;
2630         struct btrfs_shared_data_ref *sref;
2631         struct btrfs_key key;
2632         unsigned long end;
2633         unsigned long ptr;
2634         int type;
2635         u32 item_size = btrfs_item_size_nr(eb, slot);
2636         u64 refs = 0;
2637         u64 offset;
2638         u64 num_bytes;
2639         int metadata = 0;
2640
2641         btrfs_item_key_to_cpu(eb, &key, slot);
2642
2643         if (key.type == BTRFS_METADATA_ITEM_KEY) {
2644                 metadata = 1;
2645                 num_bytes = root->leafsize;
2646         } else {
2647                 num_bytes = key.offset;
2648         }
2649
2650         if (item_size < sizeof(*ei)) {
2651 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2652                 struct btrfs_extent_item_v0 *ei0;
2653                 BUG_ON(item_size != sizeof(*ei0));
2654                 ei0 = btrfs_item_ptr(eb, slot, struct btrfs_extent_item_v0);
2655                 refs = btrfs_extent_refs_v0(eb, ei0);
2656 #else
2657                 BUG();
2658 #endif
2659                 return add_extent_rec(extent_cache, NULL, key.objectid,
2660                                       num_bytes, refs, 0, 0, 0, metadata, 1,
2661                                       num_bytes);
2662         }
2663
2664         ei = btrfs_item_ptr(eb, slot, struct btrfs_extent_item);
2665         refs = btrfs_extent_refs(eb, ei);
2666
2667         add_extent_rec(extent_cache, NULL, key.objectid, num_bytes,
2668                        refs, 0, 0, 0, metadata, 1, num_bytes);
2669
2670         ptr = (unsigned long)(ei + 1);
2671         if (btrfs_extent_flags(eb, ei) & BTRFS_EXTENT_FLAG_TREE_BLOCK &&
2672             key.type == BTRFS_EXTENT_ITEM_KEY)
2673                 ptr += sizeof(struct btrfs_tree_block_info);
2674
2675         end = (unsigned long)ei + item_size;
2676         while (ptr < end) {
2677                 iref = (struct btrfs_extent_inline_ref *)ptr;
2678                 type = btrfs_extent_inline_ref_type(eb, iref);
2679                 offset = btrfs_extent_inline_ref_offset(eb, iref);
2680                 switch (type) {
2681                 case BTRFS_TREE_BLOCK_REF_KEY:
2682                         add_tree_backref(extent_cache, key.objectid,
2683                                          0, offset, 0);
2684                         break;
2685                 case BTRFS_SHARED_BLOCK_REF_KEY:
2686                         add_tree_backref(extent_cache, key.objectid,
2687                                          offset, 0, 0);
2688                         break;
2689                 case BTRFS_EXTENT_DATA_REF_KEY:
2690                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
2691                         add_data_backref(extent_cache, key.objectid, 0,
2692                                         btrfs_extent_data_ref_root(eb, dref),
2693                                         btrfs_extent_data_ref_objectid(eb,
2694                                                                        dref),
2695                                         btrfs_extent_data_ref_offset(eb, dref),
2696                                         btrfs_extent_data_ref_count(eb, dref),
2697                                         0, num_bytes);
2698                         break;
2699                 case BTRFS_SHARED_DATA_REF_KEY:
2700                         sref = (struct btrfs_shared_data_ref *)(iref + 1);
2701                         add_data_backref(extent_cache, key.objectid, offset,
2702                                         0, 0, 0,
2703                                         btrfs_shared_data_ref_count(eb, sref),
2704                                         0, num_bytes);
2705                         break;
2706                 default:
2707                         fprintf(stderr, "corrupt extent record: key %Lu %u %Lu\n",
2708                                 key.objectid, key.type, num_bytes);
2709                         goto out;
2710                 }
2711                 ptr += btrfs_extent_inline_ref_size(type);
2712         }
2713         WARN_ON(ptr > end);
2714 out:
2715         return 0;
2716 }
2717
2718 static int check_cache_range(struct btrfs_root *root,
2719                              struct btrfs_block_group_cache *cache,
2720                              u64 offset, u64 bytes)
2721 {
2722         struct btrfs_free_space *entry;
2723         u64 *logical;
2724         u64 bytenr;
2725         int stripe_len;
2726         int i, nr, ret;
2727
2728         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
2729                 bytenr = btrfs_sb_offset(i);
2730                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
2731                                        cache->key.objectid, bytenr, 0,
2732                                        &logical, &nr, &stripe_len);
2733                 if (ret)
2734                         return ret;
2735
2736                 while (nr--) {
2737                         if (logical[nr] + stripe_len <= offset)
2738                                 continue;
2739                         if (offset + bytes <= logical[nr])
2740                                 continue;
2741                         if (logical[nr] == offset) {
2742                                 if (stripe_len >= bytes) {
2743                                         kfree(logical);
2744                                         return 0;
2745                                 }
2746                                 bytes -= stripe_len;
2747                                 offset += stripe_len;
2748                         } else if (logical[nr] < offset) {
2749                                 if (logical[nr] + stripe_len >=
2750                                     offset + bytes) {
2751                                         kfree(logical);
2752                                         return 0;
2753                                 }
2754                                 bytes = (offset + bytes) -
2755                                         (logical[nr] + stripe_len);
2756                                 offset = logical[nr] + stripe_len;
2757                         } else {
2758                                 /*
2759                                  * Could be tricky, the super may land in the
2760                                  * middle of the area we're checking.  First
2761                                  * check the easiest case, it's at the end.
2762                                  */
2763                                 if (logical[nr] + stripe_len >=
2764                                     bytes + offset) {
2765                                         bytes = logical[nr] - offset;
2766                                         continue;
2767                                 }
2768
2769                                 /* Check the left side */
2770                                 ret = check_cache_range(root, cache,
2771                                                         offset,
2772                                                         logical[nr] - offset);
2773                                 if (ret) {
2774                                         kfree(logical);
2775                                         return ret;
2776                                 }
2777
2778                                 /* Now we continue with the right side */
2779                                 bytes = (offset + bytes) -
2780                                         (logical[nr] + stripe_len);
2781                                 offset = logical[nr] + stripe_len;
2782                         }
2783                 }
2784
2785                 kfree(logical);
2786         }
2787
2788         entry = btrfs_find_free_space(cache->free_space_ctl, offset, bytes);
2789         if (!entry) {
2790                 fprintf(stderr, "There is no free space entry for %Lu-%Lu\n",
2791                         offset, offset+bytes);
2792                 return -EINVAL;
2793         }
2794
2795         if (entry->offset != offset) {
2796                 fprintf(stderr, "Wanted offset %Lu, found %Lu\n", offset,
2797                         entry->offset);
2798                 return -EINVAL;
2799         }
2800
2801         if (entry->bytes != bytes) {
2802                 fprintf(stderr, "Wanted bytes %Lu, found %Lu for off %Lu\n",
2803                         bytes, entry->bytes, offset);
2804                 return -EINVAL;
2805         }
2806
2807         unlink_free_space(cache->free_space_ctl, entry);
2808         free(entry);
2809         return 0;
2810 }
2811
2812 static int verify_space_cache(struct btrfs_root *root,
2813                               struct btrfs_block_group_cache *cache)
2814 {
2815         struct btrfs_path *path;
2816         struct extent_buffer *leaf;
2817         struct btrfs_key key;
2818         u64 last;
2819         int ret = 0;
2820
2821         path = btrfs_alloc_path();
2822         if (!path)
2823                 return -ENOMEM;
2824
2825         root = root->fs_info->extent_root;
2826
2827         last = max_t(u64, cache->key.objectid, BTRFS_SUPER_INFO_OFFSET);
2828
2829         key.objectid = last;
2830         key.offset = 0;
2831         key.type = BTRFS_EXTENT_ITEM_KEY;
2832
2833         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2834         if (ret < 0)
2835                 return ret;
2836         ret = 0;
2837         while (1) {
2838                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2839                         ret = btrfs_next_leaf(root, path);
2840                         if (ret < 0)
2841                                 return ret;
2842                         if (ret > 0) {
2843                                 ret = 0;
2844                                 break;
2845                         }
2846                 }
2847                 leaf = path->nodes[0];
2848                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2849                 if (key.objectid >= cache->key.offset + cache->key.objectid)
2850                         break;
2851                 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
2852                     key.type != BTRFS_METADATA_ITEM_KEY) {
2853                         path->slots[0]++;
2854                         continue;
2855                 }
2856
2857                 if (last == key.objectid) {
2858                         if (key.type == BTRFS_EXTENT_ITEM_KEY)
2859                                 last = key.objectid + key.offset;
2860                         else
2861                                 last = key.objectid + root->leafsize;
2862                         path->slots[0]++;
2863                         continue;
2864                 }
2865
2866                 ret = check_cache_range(root, cache, last,
2867                                         key.objectid - last);
2868                 if (ret)
2869                         break;
2870                 if (key.type == BTRFS_EXTENT_ITEM_KEY)
2871                         last = key.objectid + key.offset;
2872                 else
2873                         last = key.objectid + root->leafsize;
2874                 path->slots[0]++;
2875         }
2876
2877         if (last < cache->key.objectid + cache->key.offset)
2878                 ret = check_cache_range(root, cache, last,
2879                                         cache->key.objectid +
2880                                         cache->key.offset - last);
2881         btrfs_free_path(path);
2882
2883         if (!ret &&
2884             !RB_EMPTY_ROOT(&cache->free_space_ctl->free_space_offset)) {
2885                 fprintf(stderr, "There are still entries left in the space "
2886                         "cache\n");
2887                 ret = -EINVAL;
2888         }
2889
2890         return ret;
2891 }
2892
2893 static int check_space_cache(struct btrfs_root *root)
2894 {
2895         struct btrfs_block_group_cache *cache;
2896         u64 start = BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE;
2897         int ret;
2898         int error = 0;
2899
2900         if (btrfs_super_generation(root->fs_info->super_copy) !=
2901             btrfs_super_cache_generation(root->fs_info->super_copy)) {
2902                 printf("cache and super generation don't match, space cache "
2903                        "will be invalidated\n");
2904                 return 0;
2905         }
2906
2907         while (1) {
2908                 cache = btrfs_lookup_first_block_group(root->fs_info, start);
2909                 if (!cache)
2910                         break;
2911
2912                 start = cache->key.objectid + cache->key.offset;
2913                 if (!cache->free_space_ctl) {
2914                         if (btrfs_init_free_space_ctl(cache,
2915                                                       root->sectorsize)) {
2916                                 ret = -ENOMEM;
2917                                 break;
2918                         }
2919                 } else {
2920                         btrfs_remove_free_space_cache(cache);
2921                 }
2922
2923                 ret = load_free_space_cache(root->fs_info, cache);
2924                 if (!ret)
2925                         continue;
2926
2927                 ret = verify_space_cache(root, cache);
2928                 if (ret) {
2929                         fprintf(stderr, "cache appears valid but isnt %Lu\n",
2930                                 cache->key.objectid);
2931                         error++;
2932                 }
2933         }
2934
2935         return error ? -EINVAL : 0;
2936 }
2937
2938 static int check_extent_exists(struct btrfs_root *root, u64 bytenr,
2939                                u64 num_bytes)
2940 {
2941         struct btrfs_path *path;
2942         struct extent_buffer *leaf;
2943         struct btrfs_key key;
2944         int ret;
2945
2946         path = btrfs_alloc_path();
2947         if (!path) {
2948                 fprintf(stderr, "Error allocing path\n");
2949                 return -ENOMEM;
2950         }
2951
2952         key.objectid = bytenr;
2953         key.type = BTRFS_EXTENT_ITEM_KEY;
2954         key.offset = 0;
2955
2956
2957 again:
2958         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
2959                                 0, 0);
2960         if (ret < 0) {
2961                 fprintf(stderr, "Error looking up extent record %d\n", ret);
2962                 btrfs_free_path(path);
2963                 return ret;
2964         } else if (ret) {
2965                 if (path->slots[0])
2966                         path->slots[0]--;
2967                 else
2968                         btrfs_prev_leaf(root, path);
2969         }
2970
2971         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2972
2973         /*
2974          * Block group items come before extent items if they have the same
2975          * bytenr, so walk back one more just in case.  Dear future traveler,
2976          * first congrats on mastering time travel.  Now if it's not too much
2977          * trouble could you go back to 2006 and tell Chris to make the
2978          * BLOCK_GROUP_ITEM_KEY lower than the EXTENT_ITEM_KEY please?
2979          */
2980         if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
2981                 if (path->slots[0])
2982                         path->slots[0]--;
2983                 else
2984                         btrfs_prev_leaf(root, path);
2985         }
2986
2987         while (num_bytes) {
2988                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2989                         ret = btrfs_next_leaf(root, path);
2990                         if (ret < 0) {
2991                                 fprintf(stderr, "Error going to next leaf "
2992                                         "%d\n", ret);
2993                                 btrfs_free_path(path);
2994                                 return ret;
2995                         } else if (ret) {
2996                                 break;
2997                         }
2998                 }
2999                 leaf = path->nodes[0];
3000                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3001                 if (key.type != BTRFS_EXTENT_ITEM_KEY) {
3002                         path->slots[0]++;
3003                         continue;
3004                 }
3005                 if (key.objectid + key.offset < bytenr) {
3006                         path->slots[0]++;
3007                         continue;
3008                 }
3009                 if (key.objectid > bytenr + num_bytes)
3010                         break;
3011
3012                 if (key.objectid == bytenr) {
3013                         if (key.offset >= num_bytes) {
3014                                 num_bytes = 0;
3015                                 break;
3016                         }
3017                         num_bytes -= key.offset;
3018                         bytenr += key.offset;
3019                 } else if (key.objectid < bytenr) {
3020                         if (key.objectid + key.offset >= bytenr + num_bytes) {
3021                                 num_bytes = 0;
3022                                 break;
3023                         }
3024                         num_bytes = (bytenr + num_bytes) -
3025                                 (key.objectid + key.offset);
3026                         bytenr = key.objectid + key.offset;
3027                 } else {
3028                         if (key.objectid + key.offset < bytenr + num_bytes) {
3029                                 u64 new_start = key.objectid + key.offset;
3030                                 u64 new_bytes = bytenr + num_bytes - new_start;
3031
3032                                 /*
3033                                  * Weird case, the extent is in the middle of
3034                                  * our range, we'll have to search one side
3035                                  * and then the other.  Not sure if this happens
3036                                  * in real life, but no harm in coding it up
3037                                  * anyway just in case.
3038                                  */
3039                                 btrfs_release_path(root, path);
3040                                 ret = check_extent_exists(root, new_start,
3041                                                           new_bytes);
3042                                 if (ret) {
3043                                         fprintf(stderr, "Right section didn't "
3044                                                 "have a record\n");
3045                                         break;
3046                                 }
3047                                 num_bytes = key.objectid - bytenr;
3048                                 goto again;
3049                         }
3050                         num_bytes = key.objectid - bytenr;
3051                 }
3052                 path->slots[0]++;
3053         }
3054         ret = 0;
3055
3056         if (num_bytes) {
3057                 fprintf(stderr, "There are no extents for csum range "
3058                         "%Lu-%Lu\n", bytenr, bytenr+num_bytes);
3059                 ret = 1;
3060         }
3061
3062         btrfs_free_path(path);
3063         return ret;
3064 }
3065
3066 static int check_csums(struct btrfs_root *root)
3067 {
3068         struct btrfs_path *path;
3069         struct extent_buffer *leaf;
3070         struct btrfs_key key;
3071         u64 offset = 0, num_bytes = 0;
3072         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
3073         int errors = 0;
3074         int ret;
3075
3076         root = root->fs_info->csum_root;
3077
3078         key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
3079         key.type = BTRFS_EXTENT_CSUM_KEY;
3080         key.offset = 0;
3081
3082         path = btrfs_alloc_path();
3083         if (!path)
3084                 return -ENOMEM;
3085
3086         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3087         if (ret < 0) {
3088                 fprintf(stderr, "Error searching csum tree %d\n", ret);
3089                 btrfs_free_path(path);
3090                 return ret;
3091         }
3092
3093         if (ret > 0 && path->slots[0])
3094                 path->slots[0]--;
3095         ret = 0;
3096
3097         while (1) {
3098                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3099                         ret = btrfs_next_leaf(root, path);
3100                         if (ret < 0) {
3101                                 fprintf(stderr, "Error going to next leaf "
3102                                         "%d\n", ret);
3103                                 break;
3104                         }
3105                         if (ret)
3106                                 break;
3107                 }
3108                 leaf = path->nodes[0];
3109
3110                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3111                 if (key.type != BTRFS_EXTENT_CSUM_KEY) {
3112                         path->slots[0]++;
3113                         continue;
3114                 }
3115
3116                 if (!num_bytes) {
3117                         offset = key.offset;
3118                 } else if (key.offset != offset + num_bytes) {
3119                         ret = check_extent_exists(root, offset, num_bytes);
3120                         if (ret) {
3121                                 fprintf(stderr, "Csum exists for %Lu-%Lu but "
3122                                         "there is no extent record\n",
3123                                         offset, offset+num_bytes);
3124                                 errors++;
3125                         }
3126                         offset = key.offset;
3127                         num_bytes = 0;
3128                 }
3129
3130                 num_bytes += (btrfs_item_size_nr(leaf, path->slots[0]) /
3131                               csum_size) * root->sectorsize;
3132                 path->slots[0]++;
3133         }
3134
3135         btrfs_free_path(path);
3136         return errors;
3137 }
3138
3139 static int run_next_block(struct btrfs_root *root,
3140                           struct block_info *bits,
3141                           int bits_nr,
3142                           u64 *last,
3143                           struct cache_tree *pending,
3144                           struct cache_tree *seen,
3145                           struct cache_tree *reada,
3146                           struct cache_tree *nodes,
3147                           struct cache_tree *extent_cache)
3148 {
3149         struct extent_buffer *buf;
3150         u64 bytenr;
3151         u32 size;
3152         u64 parent;
3153         u64 owner;
3154         u64 flags;
3155         int ret;
3156         int i;
3157         int nritems;
3158         struct btrfs_key key;
3159         struct cache_extent *cache;
3160         int reada_bits;
3161
3162         ret = pick_next_pending(pending, reada, nodes, *last, bits,
3163                                 bits_nr, &reada_bits);
3164         if (ret == 0) {
3165                 return 1;
3166         }
3167         if (!reada_bits) {
3168                 for(i = 0; i < ret; i++) {
3169                         insert_cache_extent(reada, bits[i].start,
3170                                             bits[i].size);
3171
3172                         /* fixme, get the parent transid */
3173                         readahead_tree_block(root, bits[i].start,
3174                                              bits[i].size, 0);
3175                 }
3176         }
3177         *last = bits[0].start;
3178         bytenr = bits[0].start;
3179         size = bits[0].size;
3180
3181         cache = find_cache_extent(pending, bytenr, size);
3182         if (cache) {
3183                 remove_cache_extent(pending, cache);
3184                 free(cache);
3185         }
3186         cache = find_cache_extent(reada, bytenr, size);
3187         if (cache) {
3188                 remove_cache_extent(reada, cache);
3189                 free(cache);
3190         }
3191         cache = find_cache_extent(nodes, bytenr, size);
3192         if (cache) {
3193                 remove_cache_extent(nodes, cache);
3194                 free(cache);
3195         }
3196
3197         /* fixme, get the real parent transid */
3198         buf = read_tree_block(root, bytenr, size, 0);
3199         if (!extent_buffer_uptodate(buf)) {
3200                 record_bad_block_io(root->fs_info,
3201                                     extent_cache, bytenr, size);
3202                 goto out;
3203         }
3204
3205         nritems = btrfs_header_nritems(buf);
3206
3207         ret = btrfs_lookup_extent_info(NULL, root, bytenr,
3208                                        btrfs_header_level(buf), 1, NULL,
3209                                        &flags);
3210         if (ret < 0)
3211                 flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
3212
3213         if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
3214                 parent = bytenr;
3215                 owner = 0;
3216         } else {
3217                 parent = 0;
3218                 owner = btrfs_header_owner(buf);
3219         }
3220
3221         ret = check_block(root, extent_cache, buf, flags);
3222         if (ret)
3223                 goto out;
3224
3225         if (btrfs_is_leaf(buf)) {
3226                 btree_space_waste += btrfs_leaf_free_space(root, buf);
3227                 for (i = 0; i < nritems; i++) {
3228                         struct btrfs_file_extent_item *fi;
3229                         btrfs_item_key_to_cpu(buf, &key, i);
3230                         if (key.type == BTRFS_EXTENT_ITEM_KEY) {
3231                                 process_extent_item(root, extent_cache, buf,
3232                                                     i);
3233                                 continue;
3234                         }
3235                         if (key.type == BTRFS_METADATA_ITEM_KEY) {
3236                                 process_extent_item(root, extent_cache, buf,
3237                                                     i);
3238                                 continue;
3239                         }
3240                         if (key.type == BTRFS_EXTENT_CSUM_KEY) {
3241                                 total_csum_bytes +=
3242                                         btrfs_item_size_nr(buf, i);
3243                                 continue;
3244                         }
3245                         if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
3246                                 continue;
3247                         }
3248                         if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
3249 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3250                                 process_extent_ref_v0(extent_cache, buf, i);
3251 #else
3252                                 BUG();
3253 #endif
3254                                 continue;
3255                         }
3256
3257                         if (key.type == BTRFS_TREE_BLOCK_REF_KEY) {
3258                                 add_tree_backref(extent_cache, key.objectid, 0,
3259                                                  key.offset, 0);
3260                                 continue;
3261                         }
3262                         if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
3263                                 add_tree_backref(extent_cache, key.objectid,
3264                                                  key.offset, 0, 0);
3265                                 continue;
3266                         }
3267                         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
3268                                 struct btrfs_extent_data_ref *ref;
3269                                 ref = btrfs_item_ptr(buf, i,
3270                                                 struct btrfs_extent_data_ref);
3271                                 add_data_backref(extent_cache,
3272                                         key.objectid, 0,
3273                                         btrfs_extent_data_ref_root(buf, ref),
3274                                         btrfs_extent_data_ref_objectid(buf,
3275                                                                        ref),
3276                                         btrfs_extent_data_ref_offset(buf, ref),
3277                                         btrfs_extent_data_ref_count(buf, ref),
3278                                         0, root->sectorsize);
3279                                 continue;
3280                         }
3281                         if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
3282                                 struct btrfs_shared_data_ref *ref;
3283                                 ref = btrfs_item_ptr(buf, i,
3284                                                 struct btrfs_shared_data_ref);
3285                                 add_data_backref(extent_cache,
3286                                         key.objectid, key.offset, 0, 0, 0,
3287                                         btrfs_shared_data_ref_count(buf, ref),
3288                                         0, root->sectorsize);
3289                                 continue;
3290                         }
3291                         if (key.type != BTRFS_EXTENT_DATA_KEY)
3292                                 continue;
3293                         fi = btrfs_item_ptr(buf, i,
3294                                             struct btrfs_file_extent_item);
3295                         if (btrfs_file_extent_type(buf, fi) ==
3296                             BTRFS_FILE_EXTENT_INLINE)
3297                                 continue;
3298                         if (btrfs_file_extent_disk_bytenr(buf, fi) == 0)
3299                                 continue;
3300
3301                         data_bytes_allocated +=
3302                                 btrfs_file_extent_disk_num_bytes(buf, fi);
3303                         if (data_bytes_allocated < root->sectorsize) {
3304                                 abort();
3305                         }
3306                         data_bytes_referenced +=
3307                                 btrfs_file_extent_num_bytes(buf, fi);
3308                         add_data_backref(extent_cache,
3309                                 btrfs_file_extent_disk_bytenr(buf, fi),
3310                                 parent, owner, key.objectid, key.offset -
3311                                 btrfs_file_extent_offset(buf, fi), 1, 1,
3312                                 btrfs_file_extent_disk_num_bytes(buf, fi));
3313                         BUG_ON(ret);
3314                 }
3315         } else {
3316                 int level;
3317                 struct btrfs_key first_key;
3318
3319                 first_key.objectid = 0;
3320
3321                 if (nritems > 0)
3322                         btrfs_item_key_to_cpu(buf, &first_key, 0);
3323                 level = btrfs_header_level(buf);
3324                 for (i = 0; i < nritems; i++) {
3325                         u64 ptr = btrfs_node_blockptr(buf, i);
3326                         u32 size = btrfs_level_size(root, level - 1);
3327                         btrfs_node_key_to_cpu(buf, &key, i);
3328                         ret = add_extent_rec(extent_cache, &key,
3329                                              ptr, size, 0, 0, 1, 0, 1, 0,
3330                                              size);
3331                         BUG_ON(ret);
3332
3333                         add_tree_backref(extent_cache, ptr, parent, owner, 1);
3334
3335                         if (level > 1) {
3336                                 add_pending(nodes, seen, ptr, size);
3337                         } else {
3338                                 add_pending(pending, seen, ptr, size);
3339                         }
3340                 }
3341                 btree_space_waste += (BTRFS_NODEPTRS_PER_BLOCK(root) -
3342                                       nritems) * sizeof(struct btrfs_key_ptr);
3343         }
3344         total_btree_bytes += buf->len;
3345         if (fs_root_objectid(btrfs_header_owner(buf)))
3346                 total_fs_tree_bytes += buf->len;
3347         if (btrfs_header_owner(buf) == BTRFS_EXTENT_TREE_OBJECTID)
3348                 total_extent_tree_bytes += buf->len;
3349         if (!found_old_backref &&
3350             btrfs_header_owner(buf) == BTRFS_TREE_RELOC_OBJECTID &&
3351             btrfs_header_backref_rev(buf) == BTRFS_MIXED_BACKREF_REV &&
3352             !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))
3353                 found_old_backref = 1;
3354 out:
3355         free_extent_buffer(buf);
3356         return 0;
3357 }
3358
3359 static int add_root_to_pending(struct extent_buffer *buf,
3360                                struct cache_tree *extent_cache,
3361                                struct cache_tree *pending,
3362                                struct cache_tree *seen,
3363                                struct cache_tree *nodes,
3364                                struct btrfs_key *root_key)
3365 {
3366         if (btrfs_header_level(buf) > 0)
3367                 add_pending(nodes, seen, buf->start, buf->len);
3368         else
3369                 add_pending(pending, seen, buf->start, buf->len);
3370         add_extent_rec(extent_cache, NULL, buf->start, buf->len,
3371                        0, 1, 1, 0, 1, 0, buf->len);
3372
3373         if (root_key->objectid == BTRFS_TREE_RELOC_OBJECTID ||
3374             btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
3375                 add_tree_backref(extent_cache, buf->start, buf->start,
3376                                  0, 1);
3377         else
3378                 add_tree_backref(extent_cache, buf->start, 0,
3379                                  root_key->objectid, 1);
3380         return 0;
3381 }
3382
3383 /* as we fix the tree, we might be deleting blocks that
3384  * we're tracking for repair.  This hook makes sure we
3385  * remove any backrefs for blocks as we are fixing them.
3386  */
3387 static int free_extent_hook(struct btrfs_trans_handle *trans,
3388                             struct btrfs_root *root,
3389                             u64 bytenr, u64 num_bytes, u64 parent,
3390                             u64 root_objectid, u64 owner, u64 offset,
3391                             int refs_to_drop)
3392 {
3393         struct extent_record *rec;
3394         struct cache_extent *cache;
3395         int is_data;
3396         struct cache_tree *extent_cache = root->fs_info->fsck_extent_cache;
3397
3398         is_data = owner >= BTRFS_FIRST_FREE_OBJECTID;
3399         cache = find_cache_extent(extent_cache, bytenr, num_bytes);
3400         if (!cache)
3401                 return 0;
3402
3403         rec = container_of(cache, struct extent_record, cache);
3404         if (is_data) {
3405                 struct data_backref *back;
3406                 back = find_data_backref(rec, parent, root_objectid, owner,
3407                                          offset, 1, bytenr, num_bytes);
3408                 if (!back)
3409                         goto out;
3410                 if (back->node.found_ref) {
3411                         back->found_ref -= refs_to_drop;
3412                         if (rec->refs)
3413                                 rec->refs -= refs_to_drop;
3414                 }
3415                 if (back->node.found_extent_tree) {
3416                         back->num_refs -= refs_to_drop;
3417                         if (rec->extent_item_refs)
3418                                 rec->extent_item_refs -= refs_to_drop;
3419                 }
3420                 if (back->found_ref == 0)
3421                         back->node.found_ref = 0;
3422                 if (back->num_refs == 0)
3423                         back->node.found_extent_tree = 0;
3424
3425                 if (!back->node.found_extent_tree && back->node.found_ref) {
3426                         list_del(&back->node.list);
3427                         free(back);
3428                 }
3429         } else {
3430                 struct tree_backref *back;
3431                 back = find_tree_backref(rec, parent, root_objectid);
3432                 if (!back)
3433                         goto out;
3434                 if (back->node.found_ref) {
3435                         if (rec->refs)
3436                                 rec->refs--;
3437                         back->node.found_ref = 0;
3438                 }
3439                 if (back->node.found_extent_tree) {
3440                         if (rec->extent_item_refs)
3441                                 rec->extent_item_refs--;
3442                         back->node.found_extent_tree = 0;
3443                 }
3444                 if (!back->node.found_extent_tree && back->node.found_ref) {
3445                         list_del(&back->node.list);
3446                         free(back);
3447                 }
3448         }
3449         maybe_free_extent_rec(extent_cache, rec);
3450 out:
3451         return 0;
3452 }
3453
3454 static int delete_extent_records(struct btrfs_trans_handle *trans,
3455                                  struct btrfs_root *root,
3456                                  struct btrfs_path *path,
3457                                  u64 bytenr, u64 new_len)
3458 {
3459         struct btrfs_key key;
3460         struct btrfs_key found_key;
3461         struct extent_buffer *leaf;
3462         int ret;
3463         int slot;
3464
3465
3466         key.objectid = bytenr;
3467         key.type = (u8)-1;
3468         key.offset = (u64)-1;
3469
3470         while(1) {
3471                 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
3472                                         &key, path, 0, 1);
3473                 if (ret < 0)
3474                         break;
3475
3476                 if (ret > 0) {
3477                         ret = 0;
3478                         if (path->slots[0] == 0)
3479                                 break;
3480                         path->slots[0]--;
3481                 }
3482                 ret = 0;
3483
3484                 leaf = path->nodes[0];
3485                 slot = path->slots[0];
3486
3487                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3488                 if (found_key.objectid != bytenr)
3489                         break;
3490
3491                 if (found_key.type != BTRFS_EXTENT_ITEM_KEY &&
3492                     found_key.type != BTRFS_METADATA_ITEM_KEY &&
3493                     found_key.type != BTRFS_TREE_BLOCK_REF_KEY &&
3494                     found_key.type != BTRFS_EXTENT_DATA_REF_KEY &&
3495                     found_key.type != BTRFS_EXTENT_REF_V0_KEY &&
3496                     found_key.type != BTRFS_SHARED_BLOCK_REF_KEY &&
3497                     found_key.type != BTRFS_SHARED_DATA_REF_KEY) {
3498                         btrfs_release_path(NULL, path);
3499                         if (found_key.type == 0) {
3500                                 if (found_key.offset == 0)
3501                                         break;
3502                                 key.offset = found_key.offset - 1;
3503                                 key.type = found_key.type;
3504                         }
3505                         key.type = found_key.type - 1;
3506                         key.offset = (u64)-1;
3507                         continue;
3508                 }
3509
3510                 fprintf(stderr, "repair deleting extent record: key %Lu %u %Lu\n",
3511                         found_key.objectid, found_key.type, found_key.offset);
3512
3513                 ret = btrfs_del_item(trans, root->fs_info->extent_root, path);
3514                 if (ret)
3515                         break;
3516                 btrfs_release_path(NULL, path);
3517
3518                 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
3519                     found_key.type == BTRFS_METADATA_ITEM_KEY) {
3520                         u64 bytes = (found_key.type == BTRFS_EXTENT_ITEM_KEY) ?
3521                                 found_key.offset : root->leafsize;
3522
3523                         ret = btrfs_update_block_group(trans, root, bytenr,
3524                                                        bytes, 0, 0);
3525                         if (ret)
3526                                 break;
3527                 }
3528         }
3529
3530         btrfs_release_path(NULL, path);
3531         return ret;
3532 }
3533
3534 /*
3535  * for a single backref, this will allocate a new extent
3536  * and add the backref to it.
3537  */
3538 static int record_extent(struct btrfs_trans_handle *trans,
3539                          struct btrfs_fs_info *info,
3540                          struct btrfs_path *path,
3541                          struct extent_record *rec,
3542                          struct extent_backref *back,
3543                          int allocated, u64 flags)
3544 {
3545         int ret;
3546         struct btrfs_root *extent_root = info->extent_root;
3547         struct extent_buffer *leaf;
3548         struct btrfs_key ins_key;
3549         struct btrfs_extent_item *ei;
3550         struct tree_backref *tback;
3551         struct data_backref *dback;
3552         struct btrfs_tree_block_info *bi;
3553
3554         if (!back->is_data)
3555                 rec->max_size = max_t(u64, rec->max_size,
3556                                     info->extent_root->leafsize);
3557
3558         if (!allocated) {
3559                 u32 item_size = sizeof(*ei);
3560
3561                 if (!back->is_data)
3562                         item_size += sizeof(*bi);
3563
3564                 ins_key.objectid = rec->start;
3565                 ins_key.offset = rec->max_size;
3566                 ins_key.type = BTRFS_EXTENT_ITEM_KEY;
3567
3568                 ret = btrfs_insert_empty_item(trans, extent_root, path,
3569                                         &ins_key, item_size);
3570                 if (ret)
3571                         goto fail;
3572
3573                 leaf = path->nodes[0];
3574                 ei = btrfs_item_ptr(leaf, path->slots[0],
3575                                     struct btrfs_extent_item);
3576
3577                 btrfs_set_extent_refs(leaf, ei, 0);
3578                 btrfs_set_extent_generation(leaf, ei, rec->generation);
3579
3580                 if (back->is_data) {
3581                         btrfs_set_extent_flags(leaf, ei,
3582                                                BTRFS_EXTENT_FLAG_DATA);
3583                 } else {
3584                         struct btrfs_disk_key copy_key;;
3585
3586                         tback = (struct tree_backref *)back;
3587                         bi = (struct btrfs_tree_block_info *)(ei + 1);
3588                         memset_extent_buffer(leaf, 0, (unsigned long)bi,
3589                                              sizeof(*bi));
3590                         memset(&copy_key, 0, sizeof(copy_key));
3591
3592                         copy_key.objectid = le64_to_cpu(rec->info_objectid);
3593                         btrfs_set_tree_block_level(leaf, bi, rec->info_level);
3594                         btrfs_set_tree_block_key(leaf, bi, &copy_key);
3595
3596                         btrfs_set_extent_flags(leaf, ei,
3597                                                BTRFS_EXTENT_FLAG_TREE_BLOCK | flags);
3598                 }
3599
3600                 btrfs_mark_buffer_dirty(leaf);
3601                 ret = btrfs_update_block_group(trans, extent_root, rec->start,
3602                                                rec->max_size, 1, 0);
3603                 if (ret)
3604                         goto fail;
3605                 btrfs_release_path(NULL, path);
3606         }
3607
3608         if (back->is_data) {
3609                 u64 parent;
3610                 int i;
3611
3612                 dback = (struct data_backref *)back;
3613                 if (back->full_backref)
3614                         parent = dback->parent;
3615                 else
3616                         parent = 0;
3617
3618                 for (i = 0; i < dback->found_ref; i++) {
3619                         /* if parent != 0, we're doing a full backref
3620                          * passing BTRFS_FIRST_FREE_OBJECTID as the owner
3621                          * just makes the backref allocator create a data
3622                          * backref
3623                          */
3624                         ret = btrfs_inc_extent_ref(trans, info->extent_root,
3625                                                    rec->start, rec->max_size,
3626                                                    parent,
3627                                                    dback->root,
3628                                                    parent ?
3629                                                    BTRFS_FIRST_FREE_OBJECTID :
3630                                                    dback->owner,
3631                                                    dback->offset);
3632                         if (ret)
3633                                 break;
3634                 }
3635                 fprintf(stderr, "adding new data backref"
3636                                 " on %llu %s %llu owner %llu"
3637                                 " offset %llu found %d\n",
3638                                 (unsigned long long)rec->start,
3639                                 back->full_backref ?
3640                                 "parent" : "root",
3641                                 back->full_backref ?
3642                                 (unsigned long long)parent :
3643                                 (unsigned long long)dback->root,
3644                                 (unsigned long long)dback->owner,
3645                                 (unsigned long long)dback->offset,
3646                                 dback->found_ref);
3647         } else {
3648                 u64 parent;
3649
3650                 tback = (struct tree_backref *)back;
3651                 if (back->full_backref)
3652                         parent = tback->parent;
3653                 else
3654                         parent = 0;
3655
3656                 ret = btrfs_inc_extent_ref(trans, info->extent_root,
3657                                            rec->start, rec->max_size,
3658                                            parent, tback->root, 0, 0);
3659                 fprintf(stderr, "adding new tree backref on "
3660                         "start %llu len %llu parent %llu root %llu\n",
3661                         rec->start, rec->max_size, tback->parent, tback->root);
3662         }
3663         if (ret)
3664                 goto fail;
3665 fail:
3666         btrfs_release_path(NULL, path);
3667         return ret;
3668 }
3669
3670 struct extent_entry {
3671         u64 bytenr;
3672         u64 bytes;
3673         int count;
3674         struct list_head list;
3675 };
3676
3677 static struct extent_entry *find_entry(struct list_head *entries,
3678                                        u64 bytenr, u64 bytes)
3679 {
3680         struct extent_entry *entry = NULL;
3681
3682         list_for_each_entry(entry, entries, list) {
3683                 if (entry->bytenr == bytenr && entry->bytes == bytes)
3684                         return entry;
3685         }
3686
3687         return NULL;
3688 }
3689
3690 static struct extent_entry *find_most_right_entry(struct list_head *entries)
3691 {
3692         struct extent_entry *entry, *best = NULL, *prev = NULL;
3693
3694         list_for_each_entry(entry, entries, list) {
3695                 if (!prev) {
3696                         prev = entry;
3697                         continue;
3698                 }
3699
3700                 /*
3701                  * If our current entry == best then we can't be sure our best
3702                  * is really the best, so we need to keep searching.
3703                  */
3704                 if (best && best->count == entry->count) {
3705                         prev = entry;
3706                         best = NULL;
3707                         continue;
3708                 }
3709
3710                 /* Prev == entry, not good enough, have to keep searching */
3711                 if (prev->count == entry->count)
3712                         continue;
3713
3714                 if (!best)
3715                         best = (prev->count > entry->count) ? prev : entry;
3716                 else if (best->count < entry->count)
3717                         best = entry;
3718                 prev = entry;
3719         }
3720
3721         return best;
3722 }
3723
3724 static int repair_ref(struct btrfs_trans_handle *trans,
3725                       struct btrfs_fs_info *info, struct btrfs_path *path,
3726                       struct data_backref *dback, struct extent_entry *entry)
3727 {
3728         struct btrfs_root *root;
3729         struct btrfs_file_extent_item *fi;
3730         struct extent_buffer *leaf;
3731         struct btrfs_key key;
3732         u64 bytenr, bytes;
3733         int ret;
3734
3735         key.objectid = dback->root;
3736         key.type = BTRFS_ROOT_ITEM_KEY;
3737         key.offset = (u64)-1;
3738         root = btrfs_read_fs_root(info, &key);
3739         if (IS_ERR(root)) {
3740                 fprintf(stderr, "Couldn't find root for our ref\n");
3741                 return -EINVAL;
3742         }
3743
3744         /*
3745          * The backref points to the original offset of the extent if it was
3746          * split, so we need to search down to the offset we have and then walk
3747          * forward until we find the backref we're looking for.
3748          */
3749         key.objectid = dback->owner;
3750         key.type = BTRFS_EXTENT_DATA_KEY;
3751         key.offset = dback->offset;
3752         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3753         if (ret < 0) {
3754                 fprintf(stderr, "Error looking up ref %d\n", ret);
3755                 return ret;
3756         }
3757
3758         while (1) {
3759                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3760                         ret = btrfs_next_leaf(root, path);
3761                         if (ret) {
3762                                 fprintf(stderr, "Couldn't find our ref, next\n");
3763                                 return -EINVAL;
3764                         }
3765                 }
3766                 leaf = path->nodes[0];
3767                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3768                 if (key.objectid != dback->owner ||
3769                     key.type != BTRFS_EXTENT_DATA_KEY) {
3770                         fprintf(stderr, "Couldn't find our ref, search\n");
3771                         return -EINVAL;
3772                 }
3773                 fi = btrfs_item_ptr(leaf, path->slots[0],
3774                                     struct btrfs_file_extent_item);
3775                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
3776                 bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
3777
3778                 if (bytenr == dback->disk_bytenr && bytes == dback->bytes)
3779                         break;
3780                 path->slots[0]++;
3781         }
3782
3783         btrfs_release_path(root, path);
3784
3785         /*
3786          * Have to make sure that this root gets updated when we commit the
3787          * transaction
3788          */
3789         root->track_dirty = 1;
3790         if (root->last_trans != trans->transid) {
3791                 root->last_trans = trans->transid;
3792                 root->commit_root = root->node;
3793                 extent_buffer_get(root->node);
3794         }
3795
3796         /*
3797          * Ok we have the key of the file extent we want to fix, now we can cow
3798          * down to the thing and fix it.
3799          */
3800         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
3801         if (ret < 0) {
3802                 fprintf(stderr, "Error cowing down to ref [%Lu, %u, %Lu]: %d\n",
3803                         key.objectid, key.type, key.offset, ret);
3804                 return ret;
3805         }
3806         if (ret > 0) {
3807                 fprintf(stderr, "Well that's odd, we just found this key "
3808                         "[%Lu, %u, %Lu]\n", key.objectid, key.type,
3809                         key.offset);
3810                 return -EINVAL;
3811         }
3812         leaf = path->nodes[0];
3813         fi = btrfs_item_ptr(leaf, path->slots[0],
3814                             struct btrfs_file_extent_item);
3815
3816         if (btrfs_file_extent_compression(leaf, fi) &&
3817             dback->disk_bytenr != entry->bytenr) {
3818                 fprintf(stderr, "Ref doesn't match the record start and is "
3819                         "compressed, please take a btrfs-image of this file "
3820                         "system and send it to a btrfs developer so they can "
3821                         "complete this functionality for bytenr %Lu\n",
3822                         dback->disk_bytenr);
3823                 return -EINVAL;
3824         }
3825
3826         if (dback->disk_bytenr > entry->bytenr) {
3827                 u64 off_diff, offset;
3828
3829                 off_diff = dback->disk_bytenr - entry->bytenr;
3830                 offset = btrfs_file_extent_offset(leaf, fi);
3831                 if (dback->disk_bytenr + offset +
3832                     btrfs_file_extent_num_bytes(leaf, fi) >
3833                     entry->bytenr + entry->bytes) {
3834                         fprintf(stderr, "Ref is past the entry end, please "
3835                                 "take a btrfs-image of this file system and "
3836                                 "send it to a btrfs developer, ref %Lu\n",
3837                                 dback->disk_bytenr);
3838                         return -EINVAL;
3839                 }
3840                 offset += off_diff;
3841                 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
3842                 btrfs_set_file_extent_offset(leaf, fi, offset);
3843         } else if (dback->disk_bytenr < entry->bytenr) {
3844                 u64 offset;
3845
3846                 offset = btrfs_file_extent_offset(leaf, fi);
3847                 if (dback->disk_bytenr + offset < entry->bytenr) {
3848                         fprintf(stderr, "Ref is before the entry start, please"
3849                                 " take a btrfs-image of this file system and "
3850                                 "send it to a btrfs developer, ref %Lu\n",
3851                                 dback->disk_bytenr);
3852                         return -EINVAL;
3853                 }
3854
3855                 offset += dback->disk_bytenr;
3856                 offset -= entry->bytenr;
3857                 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
3858                 btrfs_set_file_extent_offset(leaf, fi, offset);
3859         }
3860
3861         btrfs_set_file_extent_disk_num_bytes(leaf, fi, entry->bytes);
3862
3863         /*
3864          * Chances are if disk_num_bytes were wrong then so is ram_bytes, but
3865          * only do this if we aren't using compression, otherwise it's a
3866          * trickier case.
3867          */
3868         if (!btrfs_file_extent_compression(leaf, fi))
3869                 btrfs_set_file_extent_ram_bytes(leaf, fi, entry->bytes);
3870         else
3871                 printf("ram bytes may be wrong?\n");
3872         btrfs_mark_buffer_dirty(leaf);
3873         btrfs_release_path(root, path);
3874         return 0;
3875 }
3876
3877 static int verify_backrefs(struct btrfs_trans_handle *trans,
3878                            struct btrfs_fs_info *info, struct btrfs_path *path,
3879                            struct extent_record *rec)
3880 {
3881         struct extent_backref *back;
3882         struct data_backref *dback;
3883         struct extent_entry *entry, *best = NULL;
3884         LIST_HEAD(entries);
3885         int nr_entries = 0;
3886         int ret = 0;
3887
3888         /*
3889          * Metadata is easy and the backrefs should always agree on bytenr and
3890          * size, if not we've got bigger issues.
3891          */
3892         if (rec->metadata)
3893                 return 0;
3894
3895         list_for_each_entry(back, &rec->backrefs, list) {
3896                 dback = (struct data_backref *)back;
3897                 /*
3898                  * We only pay attention to backrefs that we found a real
3899                  * backref for.
3900                  */
3901                 if (dback->found_ref == 0)
3902                         continue;
3903                 if (back->full_backref)
3904                         continue;
3905
3906                 /*
3907                  * For now we only catch when the bytes don't match, not the
3908                  * bytenr.  We can easily do this at the same time, but I want
3909                  * to have a fs image to test on before we just add repair
3910                  * functionality willy-nilly so we know we won't screw up the
3911                  * repair.
3912                  */
3913
3914                 entry = find_entry(&entries, dback->disk_bytenr,
3915                                    dback->bytes);
3916                 if (!entry) {
3917                         entry = malloc(sizeof(struct extent_entry));
3918                         if (!entry) {
3919                                 ret = -ENOMEM;
3920                                 goto out;
3921                         }
3922                         memset(entry, 0, sizeof(*entry));
3923                         entry->bytenr = dback->disk_bytenr;
3924                         entry->bytes = dback->bytes;
3925                         list_add_tail(&entry->list, &entries);
3926                         nr_entries++;
3927                 }
3928                 entry->count++;
3929         }
3930
3931         /* Yay all the backrefs agree, carry on good sir */
3932         if (nr_entries <= 1)
3933                 goto out;
3934
3935         fprintf(stderr, "attempting to repair backref discrepency for bytenr "
3936                 "%Lu\n", rec->start);
3937
3938         /*
3939          * First we want to see if the backrefs can agree amongst themselves who
3940          * is right, so figure out which one of the entries has the highest
3941          * count.
3942          */
3943         best = find_most_right_entry(&entries);
3944
3945         /*
3946          * Ok so we may have an even split between what the backrefs think, so
3947          * this is where we use the extent ref to see what it thinks.
3948          */
3949         if (!best) {
3950                 entry = find_entry(&entries, rec->start, rec->nr);
3951                 if (!entry) {
3952                         fprintf(stderr, "Backrefs don't agree with eachother "
3953                                 "and extent record doesn't agree with anybody,"
3954                                 " so we can't fix bytenr %Lu bytes %Lu\n",
3955                                 rec->start, rec->nr);
3956                         ret = -EINVAL;
3957                         goto out;
3958                 }
3959                 entry->count++;
3960                 best = find_most_right_entry(&entries);
3961                 if (!best) {
3962                         fprintf(stderr, "Backrefs and extent record evenly "
3963                                 "split on who is right, this is going to "
3964                                 "require user input to fix bytenr %Lu bytes "
3965                                 "%Lu\n", rec->start, rec->nr);
3966                         ret = -EINVAL;
3967                         goto out;
3968                 }
3969         }
3970
3971         /*
3972          * I don't think this can happen currently as we'll abort() if we catch
3973          * this case higher up, but in case somebody removes that we still can't
3974          * deal with it properly here yet, so just bail out of that's the case.
3975          */
3976         if (best->bytenr != rec->start) {
3977                 fprintf(stderr, "Extent start and backref starts don't match, "
3978                         "please use btrfs-image on this file system and send "
3979                         "it to a btrfs developer so they can make fsck fix "
3980                         "this particular case.  bytenr is %Lu, bytes is %Lu\n",
3981                         rec->start, rec->nr);
3982                 ret = -EINVAL;
3983                 goto out;
3984         }
3985
3986         /*
3987          * Ok great we all agreed on an extent record, let's go find the real
3988          * references and fix up the ones that don't match.
3989          */
3990         list_for_each_entry(back, &rec->backrefs, list) {
3991                 dback = (struct data_backref *)back;
3992
3993                 /*
3994                  * Still ignoring backrefs that don't have a real ref attached
3995                  * to them.
3996                  */
3997                 if (dback->found_ref == 0)
3998                         continue;
3999                 if (back->full_backref)
4000                         continue;
4001
4002                 if (dback->bytes == best->bytes &&
4003                     dback->disk_bytenr == best->bytenr)
4004                         continue;
4005
4006                 ret = repair_ref(trans, info, path, dback, best);
4007                 if (ret)
4008                         goto out;
4009         }
4010
4011         /*
4012          * Ok we messed with the actual refs, which means we need to drop our
4013          * entire cache and go back and rescan.  I know this is a huge pain and
4014          * adds a lot of extra work, but it's the only way to be safe.  Once all
4015          * the backrefs agree we may not need to do anything to the extent
4016          * record itself.
4017          */
4018         ret = -EAGAIN;
4019 out:
4020         while (!list_empty(&entries)) {
4021                 entry = list_entry(entries.next, struct extent_entry, list);
4022                 list_del_init(&entry->list);
4023                 free(entry);
4024         }
4025         return ret;
4026 }
4027
4028 static int process_duplicates(struct btrfs_root *root,
4029                               struct cache_tree *extent_cache,
4030                               struct extent_record *rec)
4031 {
4032         struct extent_record *good, *tmp;
4033         struct cache_extent *cache;
4034         int ret;
4035
4036         /*
4037          * If we found a extent record for this extent then return, or if we
4038          * have more than one duplicate we are likely going to need to delete
4039          * something.
4040          */
4041         if (rec->found_rec || rec->num_duplicates > 1)
4042                 return 0;
4043
4044         /* Shouldn't happen but just in case */
4045         BUG_ON(!rec->num_duplicates);
4046
4047         /*
4048          * So this happens if we end up with a backref that doesn't match the
4049          * actual extent entry.  So either the backref is bad or the extent
4050          * entry is bad.  Either way we want to have the extent_record actually
4051          * reflect what we found in the extent_tree, so we need to take the
4052          * duplicate out and use that as the extent_record since the only way we
4053          * get a duplicate is if we find a real life BTRFS_EXTENT_ITEM_KEY.
4054          */
4055         remove_cache_extent(extent_cache, &rec->cache);
4056
4057         good = list_entry(rec->dups.next, struct extent_record, list);
4058         list_del_init(&good->list);
4059         INIT_LIST_HEAD(&good->backrefs);
4060         INIT_LIST_HEAD(&good->dups);
4061         good->cache.start = good->start;
4062         good->cache.size = good->nr;
4063         good->content_checked = 0;
4064         good->owner_ref_checked = 0;
4065         good->num_duplicates = 0;
4066         good->refs = rec->refs;
4067         list_splice_init(&rec->backrefs, &good->backrefs);
4068         while (1) {
4069                 cache = find_cache_extent(extent_cache, good->start,
4070                                           good->nr);
4071                 if (!cache)
4072                         break;
4073                 tmp = container_of(cache, struct extent_record, cache);
4074
4075                 /*
4076                  * If we find another overlapping extent and it's found_rec is
4077                  * set then it's a duplicate and we need to try and delete
4078                  * something.
4079                  */
4080                 if (tmp->found_rec || tmp->num_duplicates > 0) {
4081                         if (list_empty(&good->list))
4082                                 list_add_tail(&good->list,
4083                                               &duplicate_extents);
4084                         good->num_duplicates += tmp->num_duplicates + 1;
4085                         list_splice_init(&tmp->dups, &good->dups);
4086                         list_del_init(&tmp->list);
4087                         list_add_tail(&tmp->list, &good->dups);
4088                         remove_cache_extent(extent_cache, &tmp->cache);
4089                         continue;
4090                 }
4091
4092                 /*
4093                  * Ok we have another non extent item backed extent rec, so lets
4094                  * just add it to this extent and carry on like we did above.
4095                  */
4096                 good->refs += tmp->refs;
4097                 list_splice_init(&tmp->backrefs, &good->backrefs);
4098                 remove_cache_extent(extent_cache, &tmp->cache);
4099                 free(tmp);
4100         }
4101         ret = insert_existing_cache_extent(extent_cache, &good->cache);
4102         BUG_ON(ret);
4103         free(rec);
4104         return good->num_duplicates ? 0 : 1;
4105 }
4106
4107 static int delete_duplicate_records(struct btrfs_trans_handle *trans,
4108                                     struct btrfs_root *root,
4109                                     struct extent_record *rec)
4110 {
4111         LIST_HEAD(delete_list);
4112         struct btrfs_path *path;
4113         struct extent_record *tmp, *good, *n;
4114         int nr_del = 0;
4115         int ret = 0;
4116         struct btrfs_key key;
4117
4118         path = btrfs_alloc_path();
4119         if (!path) {
4120                 ret = -ENOMEM;
4121                 goto out;
4122         }
4123
4124         good = rec;
4125         /* Find the record that covers all of the duplicates. */
4126         list_for_each_entry(tmp, &rec->dups, list) {
4127                 if (good->start < tmp->start)
4128                         continue;
4129                 if (good->nr > tmp->nr)
4130                         continue;
4131
4132                 if (tmp->start + tmp->nr < good->start + good->nr) {
4133                         fprintf(stderr, "Ok we have overlapping extents that "
4134                                 "aren't completely covered by eachother, this "
4135                                 "is going to require more careful thought.  "
4136                                 "The extents are [%Lu-%Lu] and [%Lu-%Lu]\n",
4137                                 tmp->start, tmp->nr, good->start, good->nr);
4138                         abort();
4139                 }
4140                 good = tmp;
4141         }
4142
4143         if (good != rec)
4144                 list_add_tail(&rec->list, &delete_list);
4145
4146         list_for_each_entry_safe(tmp, n, &rec->dups, list) {
4147                 if (tmp == good)
4148                         continue;
4149                 list_move_tail(&tmp->list, &delete_list);
4150         }
4151
4152         root = root->fs_info->extent_root;
4153         list_for_each_entry(tmp, &delete_list, list) {
4154                 if (tmp->found_rec == 0)
4155                         continue;
4156                 key.objectid = tmp->start;
4157                 key.type = BTRFS_EXTENT_ITEM_KEY;
4158                 key.offset = tmp->nr;
4159
4160                 /* Shouldn't happen but just in case */
4161                 if (tmp->metadata) {
4162                         fprintf(stderr, "Well this shouldn't happen, extent "
4163                                 "record overlaps but is metadata? "
4164                                 "[%Lu, %Lu]\n", tmp->start, tmp->nr);
4165                         abort();
4166                 }
4167
4168                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4169                 if (ret) {
4170                         if (ret > 0)
4171                                 ret = -EINVAL;
4172                         goto out;
4173                 }
4174                 ret = btrfs_del_item(trans, root, path);
4175                 if (ret)
4176                         goto out;
4177                 btrfs_release_path(root, path);
4178                 nr_del++;
4179         }
4180
4181 out:
4182         while (!list_empty(&delete_list)) {
4183                 tmp = list_entry(delete_list.next, struct extent_record, list);
4184                 list_del_init(&tmp->list);
4185                 if (tmp == rec)
4186                         continue;
4187                 free(tmp);
4188         }
4189
4190         while (!list_empty(&rec->dups)) {
4191                 tmp = list_entry(rec->dups.next, struct extent_record, list);
4192                 list_del_init(&tmp->list);
4193                 free(tmp);
4194         }
4195
4196         btrfs_free_path(path);
4197
4198         if (!ret && !nr_del)
4199                 rec->num_duplicates = 0;
4200
4201         return ret ? ret : nr_del;
4202 }
4203
4204 /*
4205  * when an incorrect extent item is found, this will delete
4206  * all of the existing entries for it and recreate them
4207  * based on what the tree scan found.
4208  */
4209 static int fixup_extent_refs(struct btrfs_trans_handle *trans,
4210                              struct btrfs_fs_info *info,
4211                              struct extent_record *rec)
4212 {
4213         int ret;
4214         struct btrfs_path *path;
4215         struct list_head *cur = rec->backrefs.next;
4216         struct cache_extent *cache;
4217         struct extent_backref *back;
4218         int allocated = 0;
4219         u64 flags = 0;
4220
4221         /* remember our flags for recreating the extent */
4222         ret = btrfs_lookup_extent_info(NULL, info->extent_root, rec->start,
4223                                        rec->max_size, rec->metadata, NULL,
4224                                        &flags);
4225         if (ret < 0)
4226                 flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
4227
4228         path = btrfs_alloc_path();
4229
4230         /* step one, make sure all of the backrefs agree */
4231         ret = verify_backrefs(trans, info, path, rec);
4232         if (ret < 0)
4233                 goto out;
4234
4235         /* step two, delete all the existing records */
4236         ret = delete_extent_records(trans, info->extent_root, path,
4237                                     rec->start, rec->max_size);
4238
4239         if (ret < 0)
4240                 goto out;
4241
4242         /* was this block corrupt?  If so, don't add references to it */
4243         cache = find_cache_extent(info->corrupt_blocks, rec->start, rec->max_size);
4244         if (cache) {
4245                 ret = 0;
4246                 goto out;
4247         }
4248
4249         /* step three, recreate all the refs we did find */
4250         while(cur != &rec->backrefs) {
4251                 back = list_entry(cur, struct extent_backref, list);
4252                 cur = cur->next;
4253
4254                 /*
4255                  * if we didn't find any references, don't create a
4256                  * new extent record
4257                  */
4258                 if (!back->found_ref)
4259                         continue;
4260
4261                 ret = record_extent(trans, info, path, rec, back, allocated, flags);
4262                 allocated = 1;
4263
4264                 if (ret)
4265                         goto out;
4266         }
4267 out:
4268         btrfs_free_path(path);
4269         return ret;
4270 }
4271
4272 /* right now we only prune from the extent allocation tree */
4273 static int prune_one_block(struct btrfs_trans_handle *trans,
4274                            struct btrfs_fs_info *info,
4275                            struct btrfs_corrupt_block *corrupt)
4276 {
4277         int ret;
4278         struct btrfs_path path;
4279         struct extent_buffer *eb;
4280         u64 found;
4281         int slot;
4282         int nritems;
4283         int level = corrupt->level + 1;
4284
4285         btrfs_init_path(&path);
4286 again:
4287         /* we want to stop at the parent to our busted block */
4288         path.lowest_level = level;
4289
4290         ret = btrfs_search_slot(trans, info->extent_root,
4291                                 &corrupt->key, &path, -1, 1);
4292
4293         if (ret < 0)
4294                 goto out;
4295
4296         eb = path.nodes[level];
4297         if (!eb) {
4298                 ret = -ENOENT;
4299                 goto out;
4300         }
4301
4302         /*
4303          * hopefully the search gave us the block we want to prune,
4304          * lets try that first
4305          */
4306         slot = path.slots[level];
4307         found =  btrfs_node_blockptr(eb, slot);
4308         if (found == corrupt->cache.start)
4309                 goto del_ptr;
4310
4311         nritems = btrfs_header_nritems(eb);
4312
4313         /* the search failed, lets scan this node and hope we find it */
4314         for (slot = 0; slot < nritems; slot++) {
4315                 found =  btrfs_node_blockptr(eb, slot);
4316                 if (found == corrupt->cache.start)
4317                         goto del_ptr;
4318         }
4319         /*
4320          * we couldn't find the bad block.  TODO, search all the nodes for pointers
4321          * to this block
4322          */
4323         if (eb == info->extent_root->node) {
4324                 ret = -ENOENT;
4325                 goto out;
4326         } else {
4327                 level++;
4328                 btrfs_release_path(NULL, &path);
4329                 goto again;
4330         }
4331
4332 del_ptr:
4333         printk("deleting pointer to block %Lu\n", corrupt->cache.start);
4334         ret = btrfs_del_ptr(trans, info->extent_root, &path, level, slot);
4335
4336 out:
4337         btrfs_release_path(NULL, &path);
4338         return ret;
4339 }
4340
4341 static int prune_corrupt_blocks(struct btrfs_trans_handle *trans,
4342                                 struct btrfs_fs_info *info)
4343 {
4344         struct cache_extent *cache;
4345         struct btrfs_corrupt_block *corrupt;
4346
4347         cache = find_first_cache_extent(info->corrupt_blocks, 0);
4348         while (1) {
4349                 if (!cache)
4350                         break;
4351                 corrupt = container_of(cache, struct btrfs_corrupt_block, cache);
4352                 prune_one_block(trans, info, corrupt);
4353                 cache = next_cache_extent(cache);
4354         }
4355         return 0;
4356 }
4357
4358 static void free_corrupt_blocks(struct btrfs_fs_info *info)
4359 {
4360         struct cache_extent *cache;
4361         struct btrfs_corrupt_block *corrupt;
4362
4363         while (1) {
4364                 cache = find_first_cache_extent(info->corrupt_blocks, 0);
4365                 if (!cache)
4366                         break;
4367                 corrupt = container_of(cache, struct btrfs_corrupt_block, cache);
4368                 remove_cache_extent(info->corrupt_blocks, cache);
4369                 free(corrupt);
4370         }
4371 }
4372
4373 static int check_block_group(struct btrfs_trans_handle *trans,
4374                               struct btrfs_fs_info *info,
4375                               struct map_lookup *map,
4376                               int *reinit)
4377 {
4378         struct btrfs_key key;
4379         struct btrfs_path path;
4380         int ret;
4381
4382         key.objectid = map->ce.start;
4383         key.offset = map->ce.size;
4384         key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
4385
4386         btrfs_init_path(&path);
4387         ret = btrfs_search_slot(NULL, info->extent_root,
4388                                 &key, &path, 0, 0);
4389         btrfs_release_path(NULL, &path);
4390         if (ret <= 0)
4391                 goto out;
4392
4393         ret = btrfs_make_block_group(trans, info->extent_root, 0, map->type,
4394                                BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4395                                key.objectid, key.offset);
4396         *reinit = 1;
4397 out:
4398         return ret;
4399 }
4400
4401 static int check_block_groups(struct btrfs_trans_handle *trans,
4402                               struct btrfs_fs_info *info, int *reinit)
4403 {
4404         struct cache_extent *ce;
4405         struct map_lookup *map;
4406         struct btrfs_mapping_tree *map_tree = &info->mapping_tree;
4407
4408         /* this isn't quite working */
4409         return 0;
4410
4411         ce = find_first_cache_extent(&map_tree->cache_tree, 0);
4412         while (1) {
4413                 if (!ce)
4414                         break;
4415                 map = container_of(ce, struct map_lookup, ce);
4416                 check_block_group(trans, info, map, reinit);
4417                 ce = next_cache_extent(ce);
4418         }
4419         return 0;
4420 }
4421
4422 static void reset_cached_block_groups(struct btrfs_fs_info *fs_info)
4423 {
4424         struct btrfs_block_group_cache *cache;
4425         u64 start, end;
4426         int ret;
4427
4428         while (1) {
4429                 ret = find_first_extent_bit(&fs_info->free_space_cache, 0,
4430                                             &start, &end, EXTENT_DIRTY);
4431                 if (ret)
4432                         break;
4433                 clear_extent_dirty(&fs_info->free_space_cache, start, end,
4434                                    GFP_NOFS);
4435         }
4436
4437         start = 0;
4438         while (1) {
4439                 cache = btrfs_lookup_first_block_group(fs_info, start);
4440                 if (!cache)
4441                         break;
4442                 if (cache->cached)
4443                         cache->cached = 0;
4444                 start = cache->key.objectid + cache->key.offset;
4445         }
4446 }
4447
4448 static int check_extent_refs(struct btrfs_trans_handle *trans,
4449                              struct btrfs_root *root,
4450                              struct cache_tree *extent_cache, int repair)
4451 {
4452         struct extent_record *rec;
4453         struct cache_extent *cache;
4454         int err = 0;
4455         int ret = 0;
4456         int fixed = 0;
4457         int reinit = 0;
4458         int had_dups = 0;
4459
4460         if (repair) {
4461                 /*
4462                  * if we're doing a repair, we have to make sure
4463                  * we don't allocate from the problem extents.
4464                  * In the worst case, this will be all the
4465                  * extents in the FS
4466                  */
4467                 cache = find_first_cache_extent(extent_cache, 0);
4468                 while(cache) {
4469                         rec = container_of(cache, struct extent_record, cache);
4470                         btrfs_pin_extent(root->fs_info,
4471                                          rec->start, rec->max_size);
4472                         cache = next_cache_extent(cache);
4473                 }
4474
4475                 /* pin down all the corrupted blocks too */
4476                 cache = find_first_cache_extent(root->fs_info->corrupt_blocks, 0);
4477                 while(cache) {
4478                         rec = container_of(cache, struct extent_record, cache);
4479                         btrfs_pin_extent(root->fs_info,
4480                                          rec->start, rec->max_size);
4481                         cache = next_cache_extent(cache);
4482                 }
4483                 prune_corrupt_blocks(trans, root->fs_info);
4484                 check_block_groups(trans, root->fs_info, &reinit);
4485                 if (reinit)
4486                         btrfs_read_block_groups(root->fs_info->extent_root);
4487                 reset_cached_block_groups(root->fs_info);
4488         }
4489
4490         /*
4491          * We need to delete any duplicate entries we find first otherwise we
4492          * could mess up the extent tree when we have backrefs that actually
4493          * belong to a different extent item and not the weird duplicate one.
4494          */
4495         while (repair && !list_empty(&duplicate_extents)) {
4496                 rec = list_entry(duplicate_extents.next, struct extent_record,
4497                                  list);
4498                 list_del_init(&rec->list);
4499
4500                 /* Sometimes we can find a backref before we find an actual
4501                  * extent, so we need to process it a little bit to see if there
4502                  * truly are multiple EXTENT_ITEM_KEY's for the same range, or
4503                  * if this is a backref screwup.  If we need to delete stuff
4504                  * process_duplicates() will return 0, otherwise it will return
4505                  * 1 and we
4506                  */
4507                 if (process_duplicates(root, extent_cache, rec))
4508                         continue;
4509                 ret = delete_duplicate_records(trans, root, rec);
4510                 if (ret < 0)
4511                         return ret;
4512                 /*
4513                  * delete_duplicate_records will return the number of entries
4514                  * deleted, so if it's greater than 0 then we know we actually
4515                  * did something and we need to remove.
4516                  */
4517                 if (ret)
4518                         had_dups = 1;
4519         }
4520
4521         if (had_dups)
4522                 return -EAGAIN;
4523
4524         while(1) {
4525                 fixed = 0;
4526                 cache = find_first_cache_extent(extent_cache, 0);
4527                 if (!cache)
4528                         break;
4529                 rec = container_of(cache, struct extent_record, cache);
4530                 if (rec->num_duplicates) {
4531                         fprintf(stderr, "extent item %llu has multiple extent "
4532                                 "items\n", (unsigned long long)rec->start);
4533                         err = 1;
4534                 }
4535
4536                 if (rec->refs != rec->extent_item_refs) {
4537                         fprintf(stderr, "ref mismatch on [%llu %llu] ",
4538                                 (unsigned long long)rec->start,
4539                                 (unsigned long long)rec->nr);
4540                         fprintf(stderr, "extent item %llu, found %llu\n",
4541                                 (unsigned long long)rec->extent_item_refs,
4542                                 (unsigned long long)rec->refs);
4543                         if (!fixed && repair) {
4544                                 ret = fixup_extent_refs(trans, root->fs_info, rec);
4545                                 if (ret)
4546                                         goto repair_abort;
4547                                 fixed = 1;
4548                         }
4549                         err = 1;
4550
4551                 }
4552                 if (all_backpointers_checked(rec, 1)) {
4553                         fprintf(stderr, "backpointer mismatch on [%llu %llu]\n",
4554                                 (unsigned long long)rec->start,
4555                                 (unsigned long long)rec->nr);
4556
4557                         if (!fixed && repair) {
4558                                 ret = fixup_extent_refs(trans, root->fs_info, rec);
4559                                 if (ret)
4560                                         goto repair_abort;
4561                                 fixed = 1;
4562                         }
4563
4564                         err = 1;
4565                 }
4566                 if (!rec->owner_ref_checked) {
4567                         fprintf(stderr, "owner ref check failed [%llu %llu]\n",
4568                                 (unsigned long long)rec->start,
4569                                 (unsigned long long)rec->nr);
4570                         if (!fixed && repair) {
4571                                 ret = fixup_extent_refs(trans, root->fs_info, rec);
4572                                 if (ret)
4573                                         goto repair_abort;
4574                                 fixed = 1;
4575                         }
4576                         err = 1;
4577                 }
4578
4579                 remove_cache_extent(extent_cache, cache);
4580                 free_all_extent_backrefs(rec);
4581                 free(rec);
4582         }
4583 repair_abort:
4584         if (repair) {
4585                 if (ret && ret != -EAGAIN) {
4586                         fprintf(stderr, "failed to repair damaged filesystem, aborting\n");
4587                         exit(1);
4588                 } else if (!ret) {
4589                         btrfs_fix_block_accounting(trans, root);
4590                 }
4591                 if (err)
4592                         fprintf(stderr, "repaired damaged extent references\n");
4593                 return ret;
4594         }
4595         return err;
4596 }
4597
4598 static void free_cache_tree(struct cache_tree *tree)
4599 {
4600         struct cache_extent *cache;
4601
4602         while (1) {
4603                 cache = find_first_cache_extent(tree, 0);
4604                 if (!cache)
4605                         break;
4606                 remove_cache_extent(tree, cache);
4607                 free(cache);
4608         }
4609 }
4610
4611 static int check_extents(struct btrfs_root *root, int repair)
4612 {
4613         struct cache_tree extent_cache;
4614         struct cache_tree seen;
4615         struct cache_tree pending;
4616         struct cache_tree reada;
4617         struct cache_tree nodes;
4618         struct cache_tree corrupt_blocks;
4619         struct btrfs_path path;
4620         struct btrfs_key key;
4621         struct btrfs_key found_key;
4622         int ret;
4623         u64 last = 0;
4624         struct block_info *bits;
4625         int bits_nr;
4626         struct extent_buffer *leaf;
4627         struct btrfs_trans_handle *trans = NULL;
4628         int slot;
4629         struct btrfs_root_item ri;
4630
4631         cache_tree_init(&extent_cache);
4632         cache_tree_init(&seen);
4633         cache_tree_init(&pending);
4634         cache_tree_init(&nodes);
4635         cache_tree_init(&reada);
4636         cache_tree_init(&corrupt_blocks);
4637
4638         if (repair) {
4639                 trans = btrfs_start_transaction(root, 1);
4640                 if (IS_ERR(trans)) {
4641                         fprintf(stderr, "Error starting transaction\n");
4642                         return PTR_ERR(trans);
4643                 }
4644                 root->fs_info->fsck_extent_cache = &extent_cache;
4645                 root->fs_info->free_extent_hook = free_extent_hook;
4646                 root->fs_info->corrupt_blocks = &corrupt_blocks;
4647         }
4648
4649         bits_nr = 1024;
4650         bits = malloc(bits_nr * sizeof(struct block_info));
4651         if (!bits) {
4652                 perror("malloc");
4653                 exit(1);
4654         }
4655
4656 again:
4657         add_root_to_pending(root->fs_info->tree_root->node,
4658                             &extent_cache, &pending, &seen, &nodes,
4659                             &root->fs_info->tree_root->root_key);
4660
4661         add_root_to_pending(root->fs_info->chunk_root->node,
4662                             &extent_cache, &pending, &seen, &nodes,
4663                             &root->fs_info->chunk_root->root_key);
4664
4665         btrfs_init_path(&path);
4666         key.offset = 0;
4667         key.objectid = 0;
4668         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
4669         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
4670                                         &key, &path, 0, 0);
4671         BUG_ON(ret < 0);
4672         while(1) {
4673                 leaf = path.nodes[0];
4674                 slot = path.slots[0];
4675                 if (slot >= btrfs_header_nritems(path.nodes[0])) {
4676                         ret = btrfs_next_leaf(root, &path);
4677                         if (ret != 0)
4678                                 break;
4679                         leaf = path.nodes[0];
4680                         slot = path.slots[0];
4681                 }
4682                 btrfs_item_key_to_cpu(leaf, &found_key, path.slots[0]);
4683                 if (btrfs_key_type(&found_key) == BTRFS_ROOT_ITEM_KEY) {
4684                         unsigned long offset;
4685                         struct extent_buffer *buf;
4686
4687                         offset = btrfs_item_ptr_offset(leaf, path.slots[0]);
4688                         read_extent_buffer(leaf, &ri, offset, sizeof(ri));
4689                         buf = read_tree_block(root->fs_info->tree_root,
4690                                               btrfs_root_bytenr(&ri),
4691                                               btrfs_level_size(root,
4692                                                btrfs_root_level(&ri)), 0);
4693                         add_root_to_pending(buf, &extent_cache, &pending,
4694                                             &seen, &nodes, &found_key);
4695                         free_extent_buffer(buf);
4696                 }
4697                 path.slots[0]++;
4698         }
4699         btrfs_release_path(root, &path);
4700         while(1) {
4701                 ret = run_next_block(root, bits, bits_nr, &last, &pending,
4702                                      &seen, &reada, &nodes, &extent_cache);
4703                 if (ret != 0)
4704                         break;
4705         }
4706         ret = check_extent_refs(trans, root, &extent_cache, repair);
4707
4708         if (ret == -EAGAIN) {
4709                 ret = btrfs_commit_transaction(trans, root);
4710                 if (ret)
4711                         goto out;
4712
4713                 trans = btrfs_start_transaction(root, 1);
4714                 if (IS_ERR(trans)) {
4715                         ret = PTR_ERR(trans);
4716                         goto out;
4717                 }
4718
4719                 free_corrupt_blocks(root->fs_info);
4720                 free_cache_tree(&seen);
4721                 free_cache_tree(&pending);
4722                 free_cache_tree(&reada);
4723                 free_cache_tree(&nodes);
4724                 free_extent_cache(root->fs_info, &extent_cache);
4725                 goto again;
4726         }
4727
4728         if (trans) {
4729                 int err;
4730
4731                 err = btrfs_commit_transaction(trans, root);
4732                 if (!ret)
4733                         ret = err;
4734         }
4735 out:
4736         if (repair) {
4737                 free_corrupt_blocks(root->fs_info);
4738                 root->fs_info->fsck_extent_cache = NULL;
4739                 root->fs_info->free_extent_hook = NULL;
4740                 root->fs_info->corrupt_blocks = NULL;
4741         }
4742         free(bits);
4743         return ret;
4744 }
4745
4746 static int pin_down_tree_blocks(struct btrfs_fs_info *fs_info,
4747                                 struct extent_buffer *eb, int tree_root)
4748 {
4749         struct extent_buffer *tmp;
4750         struct btrfs_root_item *ri;
4751         struct btrfs_key key;
4752         u64 bytenr;
4753         u32 leafsize;
4754         int level = btrfs_header_level(eb);
4755         int nritems;
4756         int ret;
4757         int i;
4758
4759         btrfs_pin_extent(fs_info, eb->start, eb->len);
4760
4761         leafsize = btrfs_super_leafsize(fs_info->super_copy);
4762         nritems = btrfs_header_nritems(eb);
4763         for (i = 0; i < nritems; i++) {
4764                 if (level == 0) {
4765                         btrfs_item_key_to_cpu(eb, &key, i);
4766                         if (key.type != BTRFS_ROOT_ITEM_KEY)
4767                                 continue;
4768                         /* Skip the extent root and reloc roots */
4769                         if (key.objectid == BTRFS_EXTENT_TREE_OBJECTID ||
4770                             key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
4771                             key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
4772                                 continue;
4773                         ri = btrfs_item_ptr(eb, i, struct btrfs_root_item);
4774                         bytenr = btrfs_disk_root_bytenr(eb, ri);
4775
4776                         /*
4777                          * If at any point we start needing the real root we
4778                          * will have to build a stump root for the root we are
4779                          * in, but for now this doesn't actually use the root so
4780                          * just pass in extent_root.
4781                          */
4782                         tmp = read_tree_block(fs_info->extent_root, bytenr,
4783                                               leafsize, 0);
4784                         if (!tmp) {
4785                                 fprintf(stderr, "Error reading root block\n");
4786                                 return -EIO;
4787                         }
4788                         ret = pin_down_tree_blocks(fs_info, tmp, 0);
4789                         free_extent_buffer(tmp);
4790                         if (ret)
4791                                 return ret;
4792                 } else {
4793                         bytenr = btrfs_node_blockptr(eb, i);
4794
4795                         /* If we aren't the tree root don't read the block */
4796                         if (level == 1 && !tree_root) {
4797                                 btrfs_pin_extent(fs_info, bytenr, leafsize);
4798                                 continue;
4799                         }
4800
4801                         tmp = read_tree_block(fs_info->extent_root, bytenr,
4802                                               leafsize, 0);
4803                         if (!tmp) {
4804                                 fprintf(stderr, "Error reading tree block\n");
4805                                 return -EIO;
4806                         }
4807                         ret = pin_down_tree_blocks(fs_info, tmp, tree_root);
4808                         free_extent_buffer(tmp);
4809                         if (ret)
4810                                 return ret;
4811                 }
4812         }
4813
4814         return 0;
4815 }
4816
4817 static int pin_metadata_blocks(struct btrfs_fs_info *fs_info)
4818 {
4819         int ret;
4820
4821         ret = pin_down_tree_blocks(fs_info, fs_info->chunk_root->node, 0);
4822         if (ret)
4823                 return ret;
4824
4825         return pin_down_tree_blocks(fs_info, fs_info->tree_root->node, 1);
4826 }
4827
4828 static int reset_block_groups(struct btrfs_fs_info *fs_info)
4829 {
4830         struct btrfs_path *path;
4831         struct extent_buffer *leaf;
4832         struct btrfs_chunk *chunk;
4833         struct btrfs_key key;
4834         int ret;
4835
4836         path = btrfs_alloc_path();
4837         if (!path)
4838                 return -ENOMEM;
4839
4840         key.objectid = 0;
4841         key.type = BTRFS_CHUNK_ITEM_KEY;
4842         key.offset = 0;
4843
4844         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
4845         if (ret < 0) {
4846                 btrfs_free_path(path);
4847                 return ret;
4848         }
4849
4850         /*
4851          * We do this in case the block groups were screwed up and had alloc
4852          * bits that aren't actually set on the chunks.  This happens with
4853          * restored images every time and could happen in real life I guess.
4854          */
4855         fs_info->avail_data_alloc_bits = 0;
4856         fs_info->avail_metadata_alloc_bits = 0;
4857         fs_info->avail_system_alloc_bits = 0;
4858
4859         /* First we need to create the in-memory block groups */
4860         while (1) {
4861                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
4862                         ret = btrfs_next_leaf(fs_info->chunk_root, path);
4863                         if (ret < 0) {
4864                                 btrfs_free_path(path);
4865                                 return ret;
4866                         }
4867                         if (ret) {
4868                                 ret = 0;
4869                                 break;
4870                         }
4871                 }
4872                 leaf = path->nodes[0];
4873                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4874                 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
4875                         path->slots[0]++;
4876                         continue;
4877                 }
4878
4879                 chunk = btrfs_item_ptr(leaf, path->slots[0],
4880                                        struct btrfs_chunk);
4881                 btrfs_add_block_group(fs_info, 0,
4882                                       btrfs_chunk_type(leaf, chunk),
4883                                       key.objectid, key.offset,
4884                                       btrfs_chunk_length(leaf, chunk));
4885                 path->slots[0]++;
4886         }
4887
4888         btrfs_free_path(path);
4889         return 0;
4890 }
4891
4892 static int reset_balance(struct btrfs_trans_handle *trans,
4893                          struct btrfs_fs_info *fs_info)
4894 {
4895         struct btrfs_root *root = fs_info->tree_root;
4896         struct btrfs_path *path;
4897         struct extent_buffer *leaf;
4898         struct btrfs_key key;
4899         int del_slot, del_nr = 0;
4900         int ret;
4901         int found = 0;
4902
4903         path = btrfs_alloc_path();
4904         if (!path)
4905                 return -ENOMEM;
4906
4907         key.objectid = BTRFS_BALANCE_OBJECTID;
4908         key.type = BTRFS_BALANCE_ITEM_KEY;
4909         key.offset = 0;
4910
4911         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4912         if (ret) {
4913                 if (ret > 0)
4914                         ret = 0;
4915                 goto out;
4916         }
4917
4918         ret = btrfs_del_item(trans, root, path);
4919         if (ret)
4920                 goto out;
4921         btrfs_release_path(root, path);
4922
4923         key.objectid = BTRFS_TREE_RELOC_OBJECTID;
4924         key.type = BTRFS_ROOT_ITEM_KEY;
4925         key.offset = 0;
4926
4927         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4928         if (ret < 0)
4929                 goto out;
4930         while (1) {
4931                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
4932                         if (!found)
4933                                 break;
4934
4935                         if (del_nr) {
4936                                 ret = btrfs_del_items(trans, root, path,
4937                                                       del_slot, del_nr);
4938                                 del_nr = 0;
4939                                 if (ret)
4940                                         goto out;
4941                         }
4942                         key.offset++;
4943                         btrfs_release_path(root, path);
4944
4945                         found = 0;
4946                         ret = btrfs_search_slot(trans, root, &key, path,
4947                                                 -1, 1);
4948                         if (ret < 0)
4949                                 goto out;
4950                         continue;
4951                 }
4952                 found = 1;
4953                 leaf = path->nodes[0];
4954                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4955                 if (key.objectid > BTRFS_TREE_RELOC_OBJECTID)
4956                         break;
4957                 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
4958                         path->slots[0]++;
4959                         continue;
4960                 }
4961                 if (!del_nr) {
4962                         del_slot = path->slots[0];
4963                         del_nr = 1;
4964                 } else {
4965                         del_nr++;
4966                 }
4967                 path->slots[0]++;
4968         }
4969
4970         if (del_nr) {
4971                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
4972                 if (ret)
4973                         goto out;
4974         }
4975         btrfs_release_path(root, path);
4976
4977         key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
4978         key.type = BTRFS_ROOT_ITEM_KEY;
4979         key.offset = (u64)-1;
4980         root = btrfs_read_fs_root(fs_info, &key);
4981         if (IS_ERR(root)) {
4982                 fprintf(stderr, "Error reading data reloc tree\n");
4983                 return PTR_ERR(root);
4984         }
4985         root->track_dirty = 1;
4986         if (root->last_trans != trans->transid) {
4987                 root->last_trans = trans->transid;
4988                 root->commit_root = root->node;
4989                 extent_buffer_get(root->node);
4990         }
4991         ret = btrfs_fsck_reinit_root(trans, root, 0);
4992 out:
4993         btrfs_free_path(path);
4994         return ret;
4995 }
4996
4997 static int reinit_extent_tree(struct btrfs_fs_info *fs_info)
4998 {
4999         struct btrfs_trans_handle *trans;
5000         u64 start = 0;
5001         int ret;
5002
5003         /*
5004          * The only reason we don't do this is because right now we're just
5005          * walking the trees we find and pinning down their bytes, we don't look
5006          * at any of the leaves.  In order to do mixed groups we'd have to check
5007          * the leaves of any fs roots and pin down the bytes for any file
5008          * extents we find.  Not hard but why do it if we don't have to?
5009          */
5010         if (btrfs_fs_incompat(fs_info, BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)) {
5011                 fprintf(stderr, "We don't support re-initing the extent tree "
5012                         "for mixed block groups yet, please notify a btrfs "
5013                         "developer you want to do this so they can add this "
5014                         "functionality.\n");
5015                 return -EINVAL;
5016         }
5017
5018         trans = btrfs_start_transaction(fs_info->extent_root, 1);
5019         if (IS_ERR(trans)) {
5020                 fprintf(stderr, "Error starting transaction\n");
5021                 return PTR_ERR(trans);
5022         }
5023
5024         /*
5025          * first we need to walk all of the trees except the extent tree and pin
5026          * down the bytes that are in use so we don't overwrite any existing
5027          * metadata.
5028          */
5029         ret = pin_metadata_blocks(fs_info);
5030         if (ret) {
5031                 fprintf(stderr, "error pinning down used bytes\n");
5032                 return ret;
5033         }
5034
5035         /*
5036          * Need to drop all the block groups since we're going to recreate all
5037          * of them again.
5038          */
5039         btrfs_free_block_groups(fs_info);
5040         ret = reset_block_groups(fs_info);
5041         if (ret) {
5042                 fprintf(stderr, "error resetting the block groups\n");
5043                 return ret;
5044         }
5045
5046         /* Ok we can allocate now, reinit the extent root */
5047         ret = btrfs_fsck_reinit_root(trans, fs_info->extent_root, 1);
5048         if (ret) {
5049                 fprintf(stderr, "extent root initialization failed\n");
5050                 /*
5051                  * When the transaction code is updated we should end the
5052                  * transaction, but for now progs only knows about commit so
5053                  * just return an error.
5054                  */
5055                 return ret;
5056         }
5057
5058         ret = reset_balance(trans, fs_info);
5059         if (ret) {
5060                 fprintf(stderr, "error reseting the pending balance\n");
5061                 return ret;
5062         }
5063
5064         /*
5065          * Now we have all the in-memory block groups setup so we can make
5066          * allocations properly, and the metadata we care about is safe since we
5067          * pinned all of it above.
5068          */
5069         while (1) {
5070                 struct btrfs_block_group_cache *cache;
5071
5072                 cache = btrfs_lookup_first_block_group(fs_info, start);
5073                 if (!cache)
5074                         break;
5075                 start = cache->key.objectid + cache->key.offset;
5076                 ret = btrfs_insert_item(trans, fs_info->extent_root,
5077                                         &cache->key, &cache->item,
5078                                         sizeof(cache->item));
5079                 if (ret) {
5080                         fprintf(stderr, "Error adding block group\n");
5081                         return ret;
5082                 }
5083                 btrfs_extent_post_op(trans, fs_info->extent_root);
5084         }
5085
5086         /*
5087          * Ok now we commit and run the normal fsck, which will add extent
5088          * entries for all of the items it finds.
5089          */
5090         return btrfs_commit_transaction(trans, fs_info->extent_root);
5091 }
5092
5093 static struct option long_options[] = {
5094         { "super", 1, NULL, 's' },
5095         { "repair", 0, NULL, 0 },
5096         { "init-csum-tree", 0, NULL, 0 },
5097         { "init-extent-tree", 0, NULL, 0 },
5098         { 0, 0, 0, 0}
5099 };
5100
5101 const char * const cmd_check_usage[] = {
5102         "btrfs check [options] <device>",
5103         "Check an unmounted btrfs filesystem.",
5104         "",
5105         "-s|--super <superblock>     use this superblock copy",
5106         "--repair                    try to repair the filesystem",
5107         "--init-csum-tree            create a new CRC tree",
5108         "--init-extent-tree          create a new extent tree",
5109         NULL
5110 };
5111
5112 int cmd_check(int argc, char **argv)
5113 {
5114         struct cache_tree root_cache;
5115         struct btrfs_root *root;
5116         struct btrfs_fs_info *info;
5117         u64 bytenr = 0;
5118         char uuidbuf[37];
5119         int ret;
5120         int num;
5121         int repair = 0;
5122         int option_index = 0;
5123         int init_csum_tree = 0;
5124         int init_extent_tree = 0;
5125         int rw = 0;
5126
5127         while(1) {
5128                 int c;
5129                 c = getopt_long(argc, argv, "as:", long_options,
5130                                 &option_index);
5131                 if (c < 0)
5132                         break;
5133                 switch(c) {
5134                         case 'a': /* ignored */ break;
5135                         case 's':
5136                                 num = atol(optarg);
5137                                 bytenr = btrfs_sb_offset(num);
5138                                 printf("using SB copy %d, bytenr %llu\n", num,
5139                                        (unsigned long long)bytenr);
5140                                 break;
5141                         case '?':
5142                         case 'h':
5143                                 usage(cmd_check_usage);
5144                 }
5145                 if (option_index == 1) {
5146                         printf("enabling repair mode\n");
5147                         repair = 1;
5148                         rw = 1;
5149                 } else if (option_index == 2) {
5150                         printf("Creating a new CRC tree\n");
5151                         init_csum_tree = 1;
5152                         rw = 1;
5153                 } else if (option_index == 3) {
5154                         init_extent_tree = 1;
5155                         rw = 1;
5156                         repair = 1;
5157                 }
5158
5159         }
5160         argc = argc - optind;
5161
5162         if (argc != 1)
5163                 usage(cmd_check_usage);
5164
5165         radix_tree_init();
5166         cache_tree_init(&root_cache);
5167
5168         if((ret = check_mounted(argv[optind])) < 0) {
5169                 fprintf(stderr, "Could not check mount status: %s\n", strerror(-ret));
5170                 return ret;
5171         } else if(ret) {
5172                 fprintf(stderr, "%s is currently mounted. Aborting.\n", argv[optind]);
5173                 return -EBUSY;
5174         }
5175
5176         info = open_ctree_fs_info(argv[optind], bytenr, 0, rw, 1);
5177         if (!info) {
5178                 fprintf(stderr, "Couldn't open file system\n");
5179                 return -EIO;
5180         }
5181
5182         uuid_unparse(info->super_copy->fsid, uuidbuf);
5183         printf("Checking filesystem on %s\nUUID: %s\n", argv[optind], uuidbuf);
5184
5185         if (!extent_buffer_uptodate(info->tree_root->node) ||
5186             !extent_buffer_uptodate(info->dev_root->node) ||
5187             !extent_buffer_uptodate(info->extent_root->node) ||
5188             !extent_buffer_uptodate(info->chunk_root->node)) {
5189                 fprintf(stderr, "Critical roots corrupted, unable to fsck the FS\n");
5190                 return -EIO;
5191         }
5192
5193         root = info->fs_root;
5194
5195         if (init_extent_tree) {
5196                 printf("Creating a new extent tree\n");
5197                 ret = reinit_extent_tree(info);
5198                 if (ret)
5199                         return ret;
5200         }
5201         fprintf(stderr, "checking extents\n");
5202         if (init_csum_tree) {
5203                 struct btrfs_trans_handle *trans;
5204
5205                 fprintf(stderr, "Reinit crc root\n");
5206                 trans = btrfs_start_transaction(info->csum_root, 1);
5207                 if (IS_ERR(trans)) {
5208                         fprintf(stderr, "Error starting transaction\n");
5209                         return PTR_ERR(trans);
5210                 }
5211
5212                 ret = btrfs_fsck_reinit_root(trans, info->csum_root, 0);
5213                 if (ret) {
5214                         fprintf(stderr, "crc root initialization failed\n");
5215                         return -EIO;
5216                 }
5217
5218                 ret = btrfs_commit_transaction(trans, root);
5219                 if (ret)
5220                         exit(1);
5221                 goto out;
5222         }
5223         ret = check_extents(root, repair);
5224         if (ret)
5225                 fprintf(stderr, "Errors found in extent allocation tree\n");
5226
5227         fprintf(stderr, "checking free space cache\n");
5228         ret = check_space_cache(root);
5229         if (ret)
5230                 goto out;
5231
5232         fprintf(stderr, "checking fs roots\n");
5233         ret = check_fs_roots(root, &root_cache);
5234         if (ret)
5235                 goto out;
5236
5237         fprintf(stderr, "checking csums\n");
5238         ret = check_csums(root);
5239         if (ret)
5240                 goto out;
5241
5242         fprintf(stderr, "checking root refs\n");
5243         ret = check_root_refs(root, &root_cache);
5244 out:
5245         free_root_recs(&root_cache);
5246         close_ctree(root);
5247
5248         if (found_old_backref) { /*
5249                  * there was a disk format change when mixed
5250                  * backref was in testing tree. The old format
5251                  * existed about one week.
5252                  */
5253                 printf("\n * Found old mixed backref format. "
5254                        "The old format is not supported! *"
5255                        "\n * Please mount the FS in readonly mode, "
5256                        "backup data and re-format the FS. *\n\n");
5257                 ret = 1;
5258         }
5259         printf("found %llu bytes used err is %d\n",
5260                (unsigned long long)bytes_used, ret);
5261         printf("total csum bytes: %llu\n",(unsigned long long)total_csum_bytes);
5262         printf("total tree bytes: %llu\n",
5263                (unsigned long long)total_btree_bytes);
5264         printf("total fs tree bytes: %llu\n",
5265                (unsigned long long)total_fs_tree_bytes);
5266         printf("total extent tree bytes: %llu\n",
5267                (unsigned long long)total_extent_tree_bytes);
5268         printf("btree space waste bytes: %llu\n",
5269                (unsigned long long)btree_space_waste);
5270         printf("file data blocks allocated: %llu\n referenced %llu\n",
5271                 (unsigned long long)data_bytes_allocated,
5272                 (unsigned long long)data_bytes_referenced);
5273         printf("%s\n", BTRFS_BUILD_VERSION);
5274         return ret;
5275 }