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 "ctree.h"
  31 #include "volumes.h"
  32 #include "repair.h"
  33 #include "disk-io.h"
  34 #include "print-tree.h"
  35 #include "transaction.h"
  36 #include "version.h"
  37 #include "utils.h"
  38 #include "commands.h"
  39 #include "free-space-cache.h"
  40 #include "btrfsck.h"
  41 #include "qgroup-verify.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 static LIST_HEAD(delete_items);
  54 static int repair = 0;
  55 static int no_holes = 0;
  56 static int init_extent_tree = 0;
  57 static int check_data_csum = 0;
  58
  59 struct extent_backref {
  60         struct list_head list;
  61         unsigned int is_data:1;
  62         unsigned int found_extent_tree:1;
  63         unsigned int full_backref:1;
  64         unsigned int found_ref:1;
  65         unsigned int broken:1;
  66 };
  67
  68 struct data_backref {
  69         struct extent_backref node;
  70         union {
  71                 u64 parent;
  72                 u64 root;
  73         };
  74         u64 owner;
  75         u64 offset;
  76         u64 disk_bytenr;
  77         u64 bytes;
  78         u64 ram_bytes;
  79         u32 num_refs;
  80         u32 found_ref;
  81 };
  82
  83 struct tree_backref {
  84         struct extent_backref node;
  85         union {
  86                 u64 parent;
  87                 u64 root;
  88         };
  89 };
  90
  91 struct extent_record {
  92         struct list_head backrefs;
  93         struct list_head dups;
  94         struct list_head list;
  95         struct cache_extent cache;
  96         struct btrfs_disk_key parent_key;
  97         u64 start;
  98         u64 max_size;
  99         u64 nr;
 100         u64 refs;
 101         u64 extent_item_refs;
 102         u64 generation;
 103         u64 parent_generation;
 104         u64 info_objectid;
 105         u32 num_duplicates;
 106         u8 info_level;
 107         unsigned int found_rec:1;
 108         unsigned int content_checked:1;
 109         unsigned int owner_ref_checked:1;
 110         unsigned int is_root:1;
 111         unsigned int metadata:1;
 112 };
 113
 114 struct inode_backref {
 115         struct list_head list;
 116         unsigned int found_dir_item:1;
 117         unsigned int found_dir_index:1;
 118         unsigned int found_inode_ref:1;
 119         unsigned int filetype:8;
 120         int errors;
 121         unsigned int ref_type;
 122         u64 dir;
 123         u64 index;
 124         u16 namelen;
 125         char name[0];
 126 };
 127
 128 struct dropping_root_item_record {
 129         struct list_head list;
 130         struct btrfs_root_item ri;
 131         struct btrfs_key found_key;
 132 };
 133
 134 #define REF_ERR_NO_DIR_ITEM             (1 << 0)
 135 #define REF_ERR_NO_DIR_INDEX            (1 << 1)
 136 #define REF_ERR_NO_INODE_REF            (1 << 2)
 137 #define REF_ERR_DUP_DIR_ITEM            (1 << 3)
 138 #define REF_ERR_DUP_DIR_INDEX           (1 << 4)
 139 #define REF_ERR_DUP_INODE_REF           (1 << 5)
 140 #define REF_ERR_INDEX_UNMATCH           (1 << 6)
 141 #define REF_ERR_FILETYPE_UNMATCH        (1 << 7)
 142 #define REF_ERR_NAME_TOO_LONG           (1 << 8) // 100
 143 #define REF_ERR_NO_ROOT_REF             (1 << 9)
 144 #define REF_ERR_NO_ROOT_BACKREF         (1 << 10)
 145 #define REF_ERR_DUP_ROOT_REF            (1 << 11)
 146 #define REF_ERR_DUP_ROOT_BACKREF        (1 << 12)
 147
 148 struct inode_record {
 149         struct list_head backrefs;
 150         unsigned int checked:1;
 151         unsigned int merging:1;
 152         unsigned int found_inode_item:1;
 153         unsigned int found_dir_item:1;
 154         unsigned int found_file_extent:1;
 155         unsigned int found_csum_item:1;
 156         unsigned int some_csum_missing:1;
 157         unsigned int nodatasum:1;
 158         int errors;
 159
 160         u64 ino;
 161         u32 nlink;
 162         u32 imode;
 163         u64 isize;
 164         u64 nbytes;
 165
 166         u32 found_link;
 167         u64 found_size;
 168         u64 extent_start;
 169         u64 extent_end;
 170         u64 first_extent_gap;
 171
 172         u32 refs;
 173 };
 174
 175 #define I_ERR_NO_INODE_ITEM             (1 << 0)
 176 #define I_ERR_NO_ORPHAN_ITEM            (1 << 1)
 177 #define I_ERR_DUP_INODE_ITEM            (1 << 2)
 178 #define I_ERR_DUP_DIR_INDEX             (1 << 3)
 179 #define I_ERR_ODD_DIR_ITEM              (1 << 4)
 180 #define I_ERR_ODD_FILE_EXTENT           (1 << 5)
 181 #define I_ERR_BAD_FILE_EXTENT           (1 << 6)
 182 #define I_ERR_FILE_EXTENT_OVERLAP       (1 << 7)
 183 #define I_ERR_FILE_EXTENT_DISCOUNT      (1 << 8) // 100
 184 #define I_ERR_DIR_ISIZE_WRONG           (1 << 9)
 185 #define I_ERR_FILE_NBYTES_WRONG         (1 << 10) // 400
 186 #define I_ERR_ODD_CSUM_ITEM             (1 << 11)
 187 #define I_ERR_SOME_CSUM_MISSING         (1 << 12)
 188 #define I_ERR_LINK_COUNT_WRONG          (1 << 13)
 189
 190 struct root_backref {
 191         struct list_head list;
 192         unsigned int found_dir_item:1;
 193         unsigned int found_dir_index:1;
 194         unsigned int found_back_ref:1;
 195         unsigned int found_forward_ref:1;
 196         unsigned int reachable:1;
 197         int errors;
 198         u64 ref_root;
 199         u64 dir;
 200         u64 index;
 201         u16 namelen;
 202         char name[0];
 203 };
 204
 205 struct root_record {
 206         struct list_head backrefs;
 207         struct cache_extent cache;
 208         unsigned int found_root_item:1;
 209         u64 objectid;
 210         u32 found_ref;
 211 };
 212
 213 struct ptr_node {
 214         struct cache_extent cache;
 215         void *data;
 216 };
 217
 218 struct shared_node {
 219         struct cache_extent cache;
 220         struct cache_tree root_cache;
 221         struct cache_tree inode_cache;
 222         struct inode_record *current;
 223         u32 refs;
 224 };
 225
 226 struct block_info {
 227         u64 start;
 228         u32 size;
 229 };
 230
 231 struct walk_control {
 232         struct cache_tree shared;
 233         struct shared_node *nodes[BTRFS_MAX_LEVEL];
 234         int active_node;
 235         int root_level;
 236 };
 237
 238 struct bad_item {
 239         struct btrfs_key key;
 240         u64 root_id;
 241         struct list_head list;
 242 };
 243
 244 static void reset_cached_block_groups(struct btrfs_fs_info *fs_info);
 245
 246 static u8 imode_to_type(u32 imode)
 247 {
 248 #define S_SHIFT 12
 249         static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
 250                 [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
 251                 [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
 252                 [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
 253                 [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
 254                 [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
 255                 [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
 256                 [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
 257         };
 258
 259         return btrfs_type_by_mode[(imode & S_IFMT) >> S_SHIFT];
 260 #undef S_SHIFT
 261 }
 262
 263 static int device_record_compare(struct rb_node *node1, struct rb_node *node2)
 264 {
 265         struct device_record *rec1;
 266         struct device_record *rec2;
 267
 268         rec1 = rb_entry(node1, struct device_record, node);
 269         rec2 = rb_entry(node2, struct device_record, node);
 270         if (rec1->devid > rec2->devid)
 271                 return -1;
 272         else if (rec1->devid < rec2->devid)
 273                 return 1;
 274         else
 275                 return 0;
 276 }
 277
 278 static struct inode_record *clone_inode_rec(struct inode_record *orig_rec)
 279 {
 280         struct inode_record *rec;
 281         struct inode_backref *backref;
 282         struct inode_backref *orig;
 283         size_t size;
 284
 285         rec = malloc(sizeof(*rec));
 286         memcpy(rec, orig_rec, sizeof(*rec));
 287         rec->refs = 1;
 288         INIT_LIST_HEAD(&rec->backrefs);
 289
 290         list_for_each_entry(orig, &orig_rec->backrefs, list) {
 291                 size = sizeof(*orig) + orig->namelen + 1;
 292                 backref = malloc(size);
 293                 memcpy(backref, orig, size);
 294                 list_add_tail(&backref->list, &rec->backrefs);
 295         }
 296         return rec;
 297 }
 298
 299 static void print_inode_error(int errors)
 300 {
 301         if (errors & I_ERR_NO_INODE_ITEM)
 302                 fprintf(stderr, ", no inode item");
 303         if (errors & I_ERR_NO_ORPHAN_ITEM)
 304                 fprintf(stderr, ", no orphan item");
 305         if (errors & I_ERR_DUP_INODE_ITEM)
 306                 fprintf(stderr, ", dup inode item");
 307         if (errors & I_ERR_DUP_DIR_INDEX)
 308                 fprintf(stderr, ", dup dir index");
 309         if (errors & I_ERR_ODD_DIR_ITEM)
 310                 fprintf(stderr, ", odd dir item");
 311         if (errors & I_ERR_ODD_FILE_EXTENT)
 312                 fprintf(stderr, ", odd file extent");
 313         if (errors & I_ERR_BAD_FILE_EXTENT)
 314                 fprintf(stderr, ", bad file extent");
 315         if (errors & I_ERR_FILE_EXTENT_OVERLAP)
 316                 fprintf(stderr, ", file extent overlap");
 317         if (errors & I_ERR_FILE_EXTENT_DISCOUNT)
 318                 fprintf(stderr, ", file extent discount");
 319         if (errors & I_ERR_DIR_ISIZE_WRONG)
 320                 fprintf(stderr, ", dir isize wrong");
 321         if (errors & I_ERR_FILE_NBYTES_WRONG)
 322                 fprintf(stderr, ", nbytes wrong");
 323         if (errors & I_ERR_ODD_CSUM_ITEM)
 324                 fprintf(stderr, ", odd csum item");
 325         if (errors & I_ERR_SOME_CSUM_MISSING)
 326                 fprintf(stderr, ", some csum missing");
 327         if (errors & I_ERR_LINK_COUNT_WRONG)
 328                 fprintf(stderr, ", link count wrong");
 329         fprintf(stderr, "\n");
 330 }
 331
 332 static void print_ref_error(int errors)
 333 {
 334         if (errors & REF_ERR_NO_DIR_ITEM)
 335                 fprintf(stderr, ", no dir item");
 336         if (errors & REF_ERR_NO_DIR_INDEX)
 337                 fprintf(stderr, ", no dir index");
 338         if (errors & REF_ERR_NO_INODE_REF)
 339                 fprintf(stderr, ", no inode ref");
 340         if (errors & REF_ERR_DUP_DIR_ITEM)
 341                 fprintf(stderr, ", dup dir item");
 342         if (errors & REF_ERR_DUP_DIR_INDEX)
 343                 fprintf(stderr, ", dup dir index");
 344         if (errors & REF_ERR_DUP_INODE_REF)
 345                 fprintf(stderr, ", dup inode ref");
 346         if (errors & REF_ERR_INDEX_UNMATCH)
 347                 fprintf(stderr, ", index unmatch");
 348         if (errors & REF_ERR_FILETYPE_UNMATCH)
 349                 fprintf(stderr, ", filetype unmatch");
 350         if (errors & REF_ERR_NAME_TOO_LONG)
 351                 fprintf(stderr, ", name too long");
 352         if (errors & REF_ERR_NO_ROOT_REF)
 353                 fprintf(stderr, ", no root ref");
 354         if (errors & REF_ERR_NO_ROOT_BACKREF)
 355                 fprintf(stderr, ", no root backref");
 356         if (errors & REF_ERR_DUP_ROOT_REF)
 357                 fprintf(stderr, ", dup root ref");
 358         if (errors & REF_ERR_DUP_ROOT_BACKREF)
 359                 fprintf(stderr, ", dup root backref");
 360         fprintf(stderr, "\n");
 361 }
 362
 363 static struct inode_record *get_inode_rec(struct cache_tree *inode_cache,
 364                                           u64 ino, int mod)
 365 {
 366         struct ptr_node *node;
 367         struct cache_extent *cache;
 368         struct inode_record *rec = NULL;
 369         int ret;
 370
 371         cache = lookup_cache_extent(inode_cache, ino, 1);
 372         if (cache) {
 373                 node = container_of(cache, struct ptr_node, cache);
 374                 rec = node->data;
 375                 if (mod && rec->refs > 1) {
 376                         node->data = clone_inode_rec(rec);
 377                         rec->refs--;
 378                         rec = node->data;
 379                 }
 380         } else if (mod) {
 381                 rec = calloc(1, sizeof(*rec));
 382                 rec->ino = ino;
 383                 rec->extent_start = (u64)-1;
 384                 rec->first_extent_gap = (u64)-1;
 385                 rec->refs = 1;
 386                 INIT_LIST_HEAD(&rec->backrefs);
 387
 388                 node = malloc(sizeof(*node));
 389                 node->cache.start = ino;
 390                 node->cache.size = 1;
 391                 node->data = rec;
 392
 393                 if (ino == BTRFS_FREE_INO_OBJECTID)
 394                         rec->found_link = 1;
 395
 396                 ret = insert_cache_extent(inode_cache, &node->cache);
 397                 BUG_ON(ret);
 398         }
 399         return rec;
 400 }
 401
 402 static void free_inode_rec(struct inode_record *rec)
 403 {
 404         struct inode_backref *backref;
 405
 406         if (--rec->refs > 0)
 407                 return;
 408
 409         while (!list_empty(&rec->backrefs)) {
 410                 backref = list_entry(rec->backrefs.next,
 411                                      struct inode_backref, list);
 412                 list_del(&backref->list);
 413                 free(backref);
 414         }
 415         free(rec);
 416 }
 417
 418 static int can_free_inode_rec(struct inode_record *rec)
 419 {
 420         if (!rec->errors && rec->checked && rec->found_inode_item &&
 421             rec->nlink == rec->found_link && list_empty(&rec->backrefs))
 422                 return 1;
 423         return 0;
 424 }
 425
 426 static void maybe_free_inode_rec(struct cache_tree *inode_cache,
 427                                  struct inode_record *rec)
 428 {
 429         struct cache_extent *cache;
 430         struct inode_backref *tmp, *backref;
 431         struct ptr_node *node;
 432         unsigned char filetype;
 433
 434         if (!rec->found_inode_item)
 435                 return;
 436
 437         filetype = imode_to_type(rec->imode);
 438         list_for_each_entry_safe(backref, tmp, &rec->backrefs, list) {
 439                 if (backref->found_dir_item && backref->found_dir_index) {
 440                         if (backref->filetype != filetype)
 441                                 backref->errors |= REF_ERR_FILETYPE_UNMATCH;
 442                         if (!backref->errors && backref->found_inode_ref) {
 443                                 list_del(&backref->list);
 444                                 free(backref);
 445                         }
 446                 }
 447         }
 448
 449         if (!rec->checked || rec->merging)
 450                 return;
 451
 452         if (S_ISDIR(rec->imode)) {
 453                 if (rec->found_size != rec->isize)
 454                         rec->errors |= I_ERR_DIR_ISIZE_WRONG;
 455                 if (rec->found_file_extent)
 456                         rec->errors |= I_ERR_ODD_FILE_EXTENT;
 457         } else if (S_ISREG(rec->imode) || S_ISLNK(rec->imode)) {
 458                 if (rec->found_dir_item)
 459                         rec->errors |= I_ERR_ODD_DIR_ITEM;
 460                 if (rec->found_size != rec->nbytes)
 461                         rec->errors |= I_ERR_FILE_NBYTES_WRONG;
 462                 if (rec->extent_start == (u64)-1 || rec->extent_start > 0)
 463                         rec->first_extent_gap = 0;
 464                 if (rec->nlink > 0 && !no_holes &&
 465                     (rec->extent_end < rec->isize ||
 466                      rec->first_extent_gap < rec->isize))
 467                         rec->errors |= I_ERR_FILE_EXTENT_DISCOUNT;
 468         }
 469
 470         if (S_ISREG(rec->imode) || S_ISLNK(rec->imode)) {
 471                 if (rec->found_csum_item && rec->nodatasum)
 472                         rec->errors |= I_ERR_ODD_CSUM_ITEM;
 473                 if (rec->some_csum_missing && !rec->nodatasum)
 474                         rec->errors |= I_ERR_SOME_CSUM_MISSING;
 475         }
 476
 477         BUG_ON(rec->refs != 1);
 478         if (can_free_inode_rec(rec)) {
 479                 cache = lookup_cache_extent(inode_cache, rec->ino, 1);
 480                 node = container_of(cache, struct ptr_node, cache);
 481                 BUG_ON(node->data != rec);
 482                 remove_cache_extent(inode_cache, &node->cache);
 483                 free(node);
 484                 free_inode_rec(rec);
 485         }
 486 }
 487
 488 static int check_orphan_item(struct btrfs_root *root, u64 ino)
 489 {
 490         struct btrfs_path path;
 491         struct btrfs_key key;
 492         int ret;
 493
 494         key.objectid = BTRFS_ORPHAN_OBJECTID;
 495         key.type = BTRFS_ORPHAN_ITEM_KEY;
 496         key.offset = ino;
 497
 498         btrfs_init_path(&path);
 499         ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
 500         btrfs_release_path(&path);
 501         if (ret > 0)
 502                 ret = -ENOENT;
 503         return ret;
 504 }
 505
 506 static int process_inode_item(struct extent_buffer *eb,
 507                               int slot, struct btrfs_key *key,
 508                               struct shared_node *active_node)
 509 {
 510         struct inode_record *rec;
 511         struct btrfs_inode_item *item;
 512
 513         rec = active_node->current;
 514         BUG_ON(rec->ino != key->objectid || rec->refs > 1);
 515         if (rec->found_inode_item) {
 516                 rec->errors |= I_ERR_DUP_INODE_ITEM;
 517                 return 1;
 518         }
 519         item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
 520         rec->nlink = btrfs_inode_nlink(eb, item);
 521         rec->isize = btrfs_inode_size(eb, item);
 522         rec->nbytes = btrfs_inode_nbytes(eb, item);
 523         rec->imode = btrfs_inode_mode(eb, item);
 524         if (btrfs_inode_flags(eb, item) & BTRFS_INODE_NODATASUM)
 525                 rec->nodatasum = 1;
 526         rec->found_inode_item = 1;
 527         if (rec->nlink == 0)
 528                 rec->errors |= I_ERR_NO_ORPHAN_ITEM;
 529         maybe_free_inode_rec(&active_node->inode_cache, rec);
 530         return 0;
 531 }
 532
 533 static struct inode_backref *get_inode_backref(struct inode_record *rec,
 534                                                 const char *name,
 535                                                 int namelen, u64 dir)
 536 {
 537         struct inode_backref *backref;
 538
 539         list_for_each_entry(backref, &rec->backrefs, list) {
 540                 if (backref->dir != dir || backref->namelen != namelen)
 541                         continue;
 542                 if (memcmp(name, backref->name, namelen))
 543                         continue;
 544                 return backref;
 545         }
 546
 547         backref = malloc(sizeof(*backref) + namelen + 1);
 548         memset(backref, 0, sizeof(*backref));
 549         backref->dir = dir;
 550         backref->namelen = namelen;
 551         memcpy(backref->name, name, namelen);
 552         backref->name[namelen] = '\0';
 553         list_add_tail(&backref->list, &rec->backrefs);
 554         return backref;
 555 }
 556
 557 static int add_inode_backref(struct cache_tree *inode_cache,
 558                              u64 ino, u64 dir, u64 index,
 559                              const char *name, int namelen,
 560                              int filetype, int itemtype, int errors)
 561 {
 562         struct inode_record *rec;
 563         struct inode_backref *backref;
 564
 565         rec = get_inode_rec(inode_cache, ino, 1);
 566         backref = get_inode_backref(rec, name, namelen, dir);
 567         if (errors)
 568                 backref->errors |= errors;
 569         if (itemtype == BTRFS_DIR_INDEX_KEY) {
 570                 if (backref->found_dir_index)
 571                         backref->errors |= REF_ERR_DUP_DIR_INDEX;
 572                 if (backref->found_inode_ref && backref->index != index)
 573                         backref->errors |= REF_ERR_INDEX_UNMATCH;
 574                 if (backref->found_dir_item && backref->filetype != filetype)
 575                         backref->errors |= REF_ERR_FILETYPE_UNMATCH;
 576
 577                 backref->index = index;
 578                 backref->filetype = filetype;
 579                 backref->found_dir_index = 1;
 580         } else if (itemtype == BTRFS_DIR_ITEM_KEY) {
 581                 rec->found_link++;
 582                 if (backref->found_dir_item)
 583                         backref->errors |= REF_ERR_DUP_DIR_ITEM;
 584                 if (backref->found_dir_index && backref->filetype != filetype)
 585                         backref->errors |= REF_ERR_FILETYPE_UNMATCH;
 586
 587                 backref->filetype = filetype;
 588                 backref->found_dir_item = 1;
 589         } else if ((itemtype == BTRFS_INODE_REF_KEY) ||
 590                    (itemtype == BTRFS_INODE_EXTREF_KEY)) {
 591                 if (backref->found_inode_ref)
 592                         backref->errors |= REF_ERR_DUP_INODE_REF;
 593                 if (backref->found_dir_index && backref->index != index)
 594                         backref->errors |= REF_ERR_INDEX_UNMATCH;
 595
 596                 backref->ref_type = itemtype;
 597                 backref->index = index;
 598                 backref->found_inode_ref = 1;
 599         } else {
 600                 BUG_ON(1);
 601         }
 602
 603         maybe_free_inode_rec(inode_cache, rec);
 604         return 0;
 605 }
 606
 607 static int merge_inode_recs(struct inode_record *src, struct inode_record *dst,
 608                             struct cache_tree *dst_cache)
 609 {
 610         struct inode_backref *backref;
 611         u32 dir_count = 0;
 612
 613         dst->merging = 1;
 614         list_for_each_entry(backref, &src->backrefs, list) {
 615                 if (backref->found_dir_index) {
 616                         add_inode_backref(dst_cache, dst->ino, backref->dir,
 617                                         backref->index, backref->name,
 618                                         backref->namelen, backref->filetype,
 619                                         BTRFS_DIR_INDEX_KEY, backref->errors);
 620                 }
 621                 if (backref->found_dir_item) {
 622                         dir_count++;
 623                         add_inode_backref(dst_cache, dst->ino,
 624                                         backref->dir, 0, backref->name,
 625                                         backref->namelen, backref->filetype,
 626                                         BTRFS_DIR_ITEM_KEY, backref->errors);
 627                 }
 628                 if (backref->found_inode_ref) {
 629                         add_inode_backref(dst_cache, dst->ino,
 630                                         backref->dir, backref->index,
 631                                         backref->name, backref->namelen, 0,
 632                                         backref->ref_type, backref->errors);
 633                 }
 634         }
 635
 636         if (src->found_dir_item)
 637                 dst->found_dir_item = 1;
 638         if (src->found_file_extent)
 639                 dst->found_file_extent = 1;
 640         if (src->found_csum_item)
 641                 dst->found_csum_item = 1;
 642         if (src->some_csum_missing)
 643                 dst->some_csum_missing = 1;
 644         if (dst->first_extent_gap > src->first_extent_gap)
 645                 dst->first_extent_gap = src->first_extent_gap;
 646
 647         BUG_ON(src->found_link < dir_count);
 648         dst->found_link += src->found_link - dir_count;
 649         dst->found_size += src->found_size;
 650         if (src->extent_start != (u64)-1) {
 651                 if (dst->extent_start == (u64)-1) {
 652                         dst->extent_start = src->extent_start;
 653                         dst->extent_end = src->extent_end;
 654                 } else {
 655                         if (dst->extent_end > src->extent_start)
 656                                 dst->errors |= I_ERR_FILE_EXTENT_OVERLAP;
 657                         else if (dst->extent_end < src->extent_start &&
 658                                  dst->extent_end < dst->first_extent_gap)
 659                                 dst->first_extent_gap = dst->extent_end;
 660                         if (dst->extent_end < src->extent_end)
 661                                 dst->extent_end = src->extent_end;
 662                 }
 663         }
 664
 665         dst->errors |= src->errors;
 666         if (src->found_inode_item) {
 667                 if (!dst->found_inode_item) {
 668                         dst->nlink = src->nlink;
 669                         dst->isize = src->isize;
 670                         dst->nbytes = src->nbytes;
 671                         dst->imode = src->imode;
 672                         dst->nodatasum = src->nodatasum;
 673                         dst->found_inode_item = 1;
 674                 } else {
 675                         dst->errors |= I_ERR_DUP_INODE_ITEM;
 676                 }
 677         }
 678         dst->merging = 0;
 679
 680         return 0;
 681 }
 682
 683 static int splice_shared_node(struct shared_node *src_node,
 684                               struct shared_node *dst_node)
 685 {
 686         struct cache_extent *cache;
 687         struct ptr_node *node, *ins;
 688         struct cache_tree *src, *dst;
 689         struct inode_record *rec, *conflict;
 690         u64 current_ino = 0;
 691         int splice = 0;
 692         int ret;
 693
 694         if (--src_node->refs == 0)
 695                 splice = 1;
 696         if (src_node->current)
 697                 current_ino = src_node->current->ino;
 698
 699         src = &src_node->root_cache;
 700         dst = &dst_node->root_cache;
 701 again:
 702         cache = search_cache_extent(src, 0);
 703         while (cache) {
 704                 node = container_of(cache, struct ptr_node, cache);
 705                 rec = node->data;
 706                 cache = next_cache_extent(cache);
 707
 708                 if (splice) {
 709                         remove_cache_extent(src, &node->cache);
 710                         ins = node;
 711                 } else {
 712                         ins = malloc(sizeof(*ins));
 713                         ins->cache.start = node->cache.start;
 714                         ins->cache.size = node->cache.size;
 715                         ins->data = rec;
 716                         rec->refs++;
 717                 }
 718                 ret = insert_cache_extent(dst, &ins->cache);
 719                 if (ret == -EEXIST) {
 720                         conflict = get_inode_rec(dst, rec->ino, 1);
 721                         merge_inode_recs(rec, conflict, dst);
 722                         if (rec->checked) {
 723                                 conflict->checked = 1;
 724                                 if (dst_node->current == conflict)
 725                                         dst_node->current = NULL;
 726                         }
 727                         maybe_free_inode_rec(dst, conflict);
 728                         free_inode_rec(rec);
 729                         free(ins);
 730                 } else {
 731                         BUG_ON(ret);
 732                 }
 733         }
 734
 735         if (src == &src_node->root_cache) {
 736                 src = &src_node->inode_cache;
 737                 dst = &dst_node->inode_cache;
 738                 goto again;
 739         }
 740
 741         if (current_ino > 0 && (!dst_node->current ||
 742             current_ino > dst_node->current->ino)) {
 743                 if (dst_node->current) {
 744                         dst_node->current->checked = 1;
 745                         maybe_free_inode_rec(dst, dst_node->current);
 746                 }
 747                 dst_node->current = get_inode_rec(dst, current_ino, 1);
 748         }
 749         return 0;
 750 }
 751
 752 static void free_inode_ptr(struct cache_extent *cache)
 753 {
 754         struct ptr_node *node;
 755         struct inode_record *rec;
 756
 757         node = container_of(cache, struct ptr_node, cache);
 758         rec = node->data;
 759         free_inode_rec(rec);
 760         free(node);
 761 }
 762
 763 FREE_EXTENT_CACHE_BASED_TREE(inode_recs, free_inode_ptr);
 764
 765 static struct shared_node *find_shared_node(struct cache_tree *shared,
 766                                             u64 bytenr)
 767 {
 768         struct cache_extent *cache;
 769         struct shared_node *node;
 770
 771         cache = lookup_cache_extent(shared, bytenr, 1);
 772         if (cache) {
 773                 node = container_of(cache, struct shared_node, cache);
 774                 return node;
 775         }
 776         return NULL;
 777 }
 778
 779 static int add_shared_node(struct cache_tree *shared, u64 bytenr, u32 refs)
 780 {
 781         int ret;
 782         struct shared_node *node;
 783
 784         node = calloc(1, sizeof(*node));
 785         node->cache.start = bytenr;
 786         node->cache.size = 1;
 787         cache_tree_init(&node->root_cache);
 788         cache_tree_init(&node->inode_cache);
 789         node->refs = refs;
 790
 791         ret = insert_cache_extent(shared, &node->cache);
 792         BUG_ON(ret);
 793         return 0;
 794 }
 795
 796 static int enter_shared_node(struct btrfs_root *root, u64 bytenr, u32 refs,
 797                              struct walk_control *wc, int level)
 798 {
 799         struct shared_node *node;
 800         struct shared_node *dest;
 801
 802         if (level == wc->active_node)
 803                 return 0;
 804
 805         BUG_ON(wc->active_node <= level);
 806         node = find_shared_node(&wc->shared, bytenr);
 807         if (!node) {
 808                 add_shared_node(&wc->shared, bytenr, refs);
 809                 node = find_shared_node(&wc->shared, bytenr);
 810                 wc->nodes[level] = node;
 811                 wc->active_node = level;
 812                 return 0;
 813         }
 814
 815         if (wc->root_level == wc->active_node &&
 816             btrfs_root_refs(&root->root_item) == 0) {
 817                 if (--node->refs == 0) {
 818                         free_inode_recs_tree(&node->root_cache);
 819                         free_inode_recs_tree(&node->inode_cache);
 820                         remove_cache_extent(&wc->shared, &node->cache);
 821                         free(node);
 822                 }
 823                 return 1;
 824         }
 825
 826         dest = wc->nodes[wc->active_node];
 827         splice_shared_node(node, dest);
 828         if (node->refs == 0) {
 829                 remove_cache_extent(&wc->shared, &node->cache);
 830                 free(node);
 831         }
 832         return 1;
 833 }
 834
 835 static int leave_shared_node(struct btrfs_root *root,
 836                              struct walk_control *wc, int level)
 837 {
 838         struct shared_node *node;
 839         struct shared_node *dest;
 840         int i;
 841
 842         if (level == wc->root_level)
 843                 return 0;
 844
 845         for (i = level + 1; i < BTRFS_MAX_LEVEL; i++) {
 846                 if (wc->nodes[i])
 847                         break;
 848         }
 849         BUG_ON(i >= BTRFS_MAX_LEVEL);
 850
 851         node = wc->nodes[wc->active_node];
 852         wc->nodes[wc->active_node] = NULL;
 853         wc->active_node = i;
 854
 855         dest = wc->nodes[wc->active_node];
 856         if (wc->active_node < wc->root_level ||
 857             btrfs_root_refs(&root->root_item) > 0) {
 858                 BUG_ON(node->refs <= 1);
 859                 splice_shared_node(node, dest);
 860         } else {
 861                 BUG_ON(node->refs < 2);
 862                 node->refs--;
 863         }
 864         return 0;
 865 }
 866
 867 static int is_child_root(struct btrfs_root *root, u64 parent_root_id,
 868                          u64 child_root_id)
 869 {
 870         struct btrfs_path path;
 871         struct btrfs_key key;
 872         struct extent_buffer *leaf;
 873         int has_parent = 0;
 874         int ret;
 875
 876         btrfs_init_path(&path);
 877
 878         key.objectid = parent_root_id;
 879         key.type = BTRFS_ROOT_REF_KEY;
 880         key.offset = child_root_id;
 881         ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, &path,
 882                                 0, 0);
 883         BUG_ON(ret < 0);
 884         btrfs_release_path(&path);
 885         if (!ret)
 886                 return 1;
 887
 888         key.objectid = child_root_id;
 889         key.type = BTRFS_ROOT_BACKREF_KEY;
 890         key.offset = 0;
 891         ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, &path,
 892                                 0, 0);
 893         BUG_ON(ret <= 0);
 894
 895         while (1) {
 896                 leaf = path.nodes[0];
 897                 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
 898                         ret = btrfs_next_leaf(root->fs_info->tree_root, &path);
 899                         BUG_ON(ret < 0);
 900
 901                         if (ret > 0)
 902                                 break;
 903                         leaf = path.nodes[0];
 904                 }
 905
 906                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
 907                 if (key.objectid != child_root_id ||
 908                     key.type != BTRFS_ROOT_BACKREF_KEY)
 909                         break;
 910
 911                 has_parent = 1;
 912
 913                 if (key.offset == parent_root_id) {
 914                         btrfs_release_path(&path);
 915                         return 1;
 916                 }
 917
 918                 path.slots[0]++;
 919         }
 920
 921         btrfs_release_path(&path);
 922         return has_parent? 0 : -1;
 923 }
 924
 925 static int process_dir_item(struct btrfs_root *root,
 926                             struct extent_buffer *eb,
 927                             int slot, struct btrfs_key *key,
 928                             struct shared_node *active_node)
 929 {
 930         u32 total;
 931         u32 cur = 0;
 932         u32 len;
 933         u32 name_len;
 934         u32 data_len;
 935         int error;
 936         int nritems = 0;
 937         int filetype;
 938         struct btrfs_dir_item *di;
 939         struct inode_record *rec;
 940         struct cache_tree *root_cache;
 941         struct cache_tree *inode_cache;
 942         struct btrfs_key location;
 943         char namebuf[BTRFS_NAME_LEN];
 944
 945         root_cache = &active_node->root_cache;
 946         inode_cache = &active_node->inode_cache;
 947         rec = active_node->current;
 948         rec->found_dir_item = 1;
 949
 950         di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
 951         total = btrfs_item_size_nr(eb, slot);
 952         while (cur < total) {
 953                 nritems++;
 954                 btrfs_dir_item_key_to_cpu(eb, di, &location);
 955                 name_len = btrfs_dir_name_len(eb, di);
 956                 data_len = btrfs_dir_data_len(eb, di);
 957                 filetype = btrfs_dir_type(eb, di);
 958
 959                 rec->found_size += name_len;
 960                 if (name_len <= BTRFS_NAME_LEN) {
 961                         len = name_len;
 962                         error = 0;
 963                 } else {
 964                         len = BTRFS_NAME_LEN;
 965                         error = REF_ERR_NAME_TOO_LONG;
 966                 }
 967                 read_extent_buffer(eb, namebuf, (unsigned long)(di + 1), len);
 968
 969                 if (location.type == BTRFS_INODE_ITEM_KEY) {
 970                         add_inode_backref(inode_cache, location.objectid,
 971                                           key->objectid, key->offset, namebuf,
 972                                           len, filetype, key->type, error);
 973                 } else if (location.type == BTRFS_ROOT_ITEM_KEY) {
 974                         add_inode_backref(root_cache, location.objectid,
 975                                           key->objectid, key->offset,
 976                                           namebuf, len, filetype,
 977                                           key->type, error);
 978                 } else {
 979                         fprintf(stderr, "warning line %d\n", __LINE__);
 980                 }
 981
 982                 len = sizeof(*di) + name_len + data_len;
 983                 di = (struct btrfs_dir_item *)((char *)di + len);
 984                 cur += len;
 985         }
 986         if (key->type == BTRFS_DIR_INDEX_KEY && nritems > 1)
 987                 rec->errors |= I_ERR_DUP_DIR_INDEX;
 988
 989         return 0;
 990 }
 991
 992 static int process_inode_ref(struct extent_buffer *eb,
 993                              int slot, struct btrfs_key *key,
 994                              struct shared_node *active_node)
 995 {
 996         u32 total;
 997         u32 cur = 0;
 998         u32 len;
 999         u32 name_len;
1000         u64 index;
1001         int error;
1002         struct cache_tree *inode_cache;
1003         struct btrfs_inode_ref *ref;
1004         char namebuf[BTRFS_NAME_LEN];
1005
1006         inode_cache = &active_node->inode_cache;
1007
1008         ref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1009         total = btrfs_item_size_nr(eb, slot);
1010         while (cur < total) {
1011                 name_len = btrfs_inode_ref_name_len(eb, ref);
1012                 index = btrfs_inode_ref_index(eb, ref);
1013                 if (name_len <= BTRFS_NAME_LEN) {
1014                         len = name_len;
1015                         error = 0;
1016                 } else {
1017                         len = BTRFS_NAME_LEN;
1018                         error = REF_ERR_NAME_TOO_LONG;
1019                 }
1020                 read_extent_buffer(eb, namebuf, (unsigned long)(ref + 1), len);
1021                 add_inode_backref(inode_cache, key->objectid, key->offset,
1022                                   index, namebuf, len, 0, key->type, error);
1023
1024                 len = sizeof(*ref) + name_len;
1025                 ref = (struct btrfs_inode_ref *)((char *)ref + len);
1026                 cur += len;
1027         }
1028         return 0;
1029 }
1030
1031 static int process_inode_extref(struct extent_buffer *eb,
1032                                 int slot, struct btrfs_key *key,
1033                                 struct shared_node *active_node)
1034 {
1035         u32 total;
1036         u32 cur = 0;
1037         u32 len;
1038         u32 name_len;
1039         u64 index;
1040         u64 parent;
1041         int error;
1042         struct cache_tree *inode_cache;
1043         struct btrfs_inode_extref *extref;
1044         char namebuf[BTRFS_NAME_LEN];
1045
1046         inode_cache = &active_node->inode_cache;
1047
1048         extref = btrfs_item_ptr(eb, slot, struct btrfs_inode_extref);
1049         total = btrfs_item_size_nr(eb, slot);
1050         while (cur < total) {
1051                 name_len = btrfs_inode_extref_name_len(eb, extref);
1052                 index = btrfs_inode_extref_index(eb, extref);
1053                 parent = btrfs_inode_extref_parent(eb, extref);
1054                 if (name_len <= BTRFS_NAME_LEN) {
1055                         len = name_len;
1056                         error = 0;
1057                 } else {
1058                         len = BTRFS_NAME_LEN;
1059                         error = REF_ERR_NAME_TOO_LONG;
1060                 }
1061                 read_extent_buffer(eb, namebuf,
1062                                    (unsigned long)(extref + 1), len);
1063                 add_inode_backref(inode_cache, key->objectid, parent,
1064                                   index, namebuf, len, 0, key->type, error);
1065
1066                 len = sizeof(*extref) + name_len;
1067                 extref = (struct btrfs_inode_extref *)((char *)extref + len);
1068                 cur += len;
1069         }
1070         return 0;
1071
1072 }
1073
1074 static u64 count_csum_range(struct btrfs_root *root, u64 start, u64 len)
1075 {
1076         struct btrfs_key key;
1077         struct btrfs_path path;
1078         struct extent_buffer *leaf;
1079         int ret ;
1080         size_t size;
1081         u64 found = 0;
1082         u64 csum_end;
1083         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
1084
1085         btrfs_init_path(&path);
1086
1087         key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
1088         key.offset = start;
1089         key.type = BTRFS_EXTENT_CSUM_KEY;
1090
1091         ret = btrfs_search_slot(NULL, root->fs_info->csum_root,
1092                                 &key, &path, 0, 0);
1093         BUG_ON(ret < 0);
1094         if (ret > 0 && path.slots[0] > 0) {
1095                 leaf = path.nodes[0];
1096                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0] - 1);
1097                 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
1098                     key.type == BTRFS_EXTENT_CSUM_KEY)
1099                         path.slots[0]--;
1100         }
1101
1102         while (len > 0) {
1103                 leaf = path.nodes[0];
1104                 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
1105                         ret = btrfs_next_leaf(root->fs_info->csum_root, &path);
1106                         BUG_ON(ret < 0);
1107                         if (ret > 0)
1108                                 break;
1109                         leaf = path.nodes[0];
1110                 }
1111
1112                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
1113                 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1114                     key.type != BTRFS_EXTENT_CSUM_KEY)
1115                         break;
1116
1117                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
1118                 if (key.offset >= start + len)
1119                         break;
1120
1121                 if (key.offset > start)
1122                         start = key.offset;
1123
1124                 size = btrfs_item_size_nr(leaf, path.slots[0]);
1125                 csum_end = key.offset + (size / csum_size) * root->sectorsize;
1126                 if (csum_end > start) {
1127                         size = min(csum_end - start, len);
1128                         len -= size;
1129                         start += size;
1130                         found += size;
1131                 }
1132
1133                 path.slots[0]++;
1134         }
1135         btrfs_release_path(&path);
1136         return found;
1137 }
1138
1139 static int process_file_extent(struct btrfs_root *root,
1140                                 struct extent_buffer *eb,
1141                                 int slot, struct btrfs_key *key,
1142                                 struct shared_node *active_node)
1143 {
1144         struct inode_record *rec;
1145         struct btrfs_file_extent_item *fi;
1146         u64 num_bytes = 0;
1147         u64 disk_bytenr = 0;
1148         u64 extent_offset = 0;
1149         u64 mask = root->sectorsize - 1;
1150         int extent_type;
1151
1152         rec = active_node->current;
1153         BUG_ON(rec->ino != key->objectid || rec->refs > 1);
1154         rec->found_file_extent = 1;
1155
1156         if (rec->extent_start == (u64)-1) {
1157                 rec->extent_start = key->offset;
1158                 rec->extent_end = key->offset;
1159         }
1160
1161         if (rec->extent_end > key->offset)
1162                 rec->errors |= I_ERR_FILE_EXTENT_OVERLAP;
1163         else if (rec->extent_end < key->offset &&
1164                  rec->extent_end < rec->first_extent_gap)
1165                 rec->first_extent_gap = rec->extent_end;
1166
1167         fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
1168         extent_type = btrfs_file_extent_type(eb, fi);
1169
1170         if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1171                 num_bytes = btrfs_file_extent_inline_len(eb, slot, fi);
1172                 if (num_bytes == 0)
1173                         rec->errors |= I_ERR_BAD_FILE_EXTENT;
1174                 rec->found_size += num_bytes;
1175                 num_bytes = (num_bytes + mask) & ~mask;
1176         } else if (extent_type == BTRFS_FILE_EXTENT_REG ||
1177                    extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1178                 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
1179                 disk_bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1180                 extent_offset = btrfs_file_extent_offset(eb, fi);
1181                 if (num_bytes == 0 || (num_bytes & mask))
1182                         rec->errors |= I_ERR_BAD_FILE_EXTENT;
1183                 if (num_bytes + extent_offset >
1184                     btrfs_file_extent_ram_bytes(eb, fi))
1185                         rec->errors |= I_ERR_BAD_FILE_EXTENT;
1186                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC &&
1187                     (btrfs_file_extent_compression(eb, fi) ||
1188                      btrfs_file_extent_encryption(eb, fi) ||
1189                      btrfs_file_extent_other_encoding(eb, fi)))
1190                         rec->errors |= I_ERR_BAD_FILE_EXTENT;
1191                 if (disk_bytenr > 0)
1192                         rec->found_size += num_bytes;
1193         } else {
1194                 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1195         }
1196         rec->extent_end = key->offset + num_bytes;
1197
1198         if (disk_bytenr > 0) {
1199                 u64 found;
1200                 if (btrfs_file_extent_compression(eb, fi))
1201                         num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1202                 else
1203                         disk_bytenr += extent_offset;
1204
1205                 found = count_csum_range(root, disk_bytenr, num_bytes);
1206                 if (extent_type == BTRFS_FILE_EXTENT_REG) {
1207                         if (found > 0)
1208                                 rec->found_csum_item = 1;
1209                         if (found < num_bytes)
1210                                 rec->some_csum_missing = 1;
1211                 } else if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1212                         if (found > 0)
1213                                 rec->errors |= I_ERR_ODD_CSUM_ITEM;
1214                 }
1215         }
1216         return 0;
1217 }
1218
1219 static int process_one_leaf(struct btrfs_root *root, struct extent_buffer *eb,
1220                             struct walk_control *wc)
1221 {
1222         struct btrfs_key key;
1223         u32 nritems;
1224         int i;
1225         int ret = 0;
1226         int error = 0;
1227         struct cache_tree *inode_cache;
1228         struct shared_node *active_node;
1229
1230         if (wc->root_level == wc->active_node &&
1231             btrfs_root_refs(&root->root_item) == 0)
1232                 return 0;
1233
1234         active_node = wc->nodes[wc->active_node];
1235         inode_cache = &active_node->inode_cache;
1236         nritems = btrfs_header_nritems(eb);
1237         for (i = 0; i < nritems; i++) {
1238                 btrfs_item_key_to_cpu(eb, &key, i);
1239
1240                 if (key.objectid == BTRFS_FREE_SPACE_OBJECTID)
1241                         continue;
1242                 if (key.type == BTRFS_ORPHAN_ITEM_KEY)
1243                         continue;
1244
1245                 if (active_node->current == NULL ||
1246                     active_node->current->ino < key.objectid) {
1247                         if (active_node->current) {
1248                                 active_node->current->checked = 1;
1249                                 maybe_free_inode_rec(inode_cache,
1250                                                      active_node->current);
1251                         }
1252                         active_node->current = get_inode_rec(inode_cache,
1253                                                              key.objectid, 1);
1254                 }
1255                 switch (key.type) {
1256                 case BTRFS_DIR_ITEM_KEY:
1257                 case BTRFS_DIR_INDEX_KEY:
1258                         ret = process_dir_item(root, eb, i, &key, active_node);
1259                         break;
1260                 case BTRFS_INODE_REF_KEY:
1261                         ret = process_inode_ref(eb, i, &key, active_node);
1262                         break;
1263                 case BTRFS_INODE_EXTREF_KEY:
1264                         ret = process_inode_extref(eb, i, &key, active_node);
1265                         break;
1266                 case BTRFS_INODE_ITEM_KEY:
1267                         ret = process_inode_item(eb, i, &key, active_node);
1268                         break;
1269                 case BTRFS_EXTENT_DATA_KEY:
1270                         ret = process_file_extent(root, eb, i, &key,
1271                                                   active_node);
1272                         break;
1273                 default:
1274                         break;
1275                 };
1276                 if (ret != 0)
1277                         error = 1;
1278         }
1279         return error;
1280 }
1281
1282 static void reada_walk_down(struct btrfs_root *root,
1283                             struct extent_buffer *node, int slot)
1284 {
1285         u64 bytenr;
1286         u64 ptr_gen;
1287         u32 nritems;
1288         u32 blocksize;
1289         int i;
1290         int ret;
1291         int level;
1292
1293         level = btrfs_header_level(node);
1294         if (level != 1)
1295                 return;
1296
1297         nritems = btrfs_header_nritems(node);
1298         blocksize = btrfs_level_size(root, level - 1);
1299         for (i = slot; i < nritems; i++) {
1300                 bytenr = btrfs_node_blockptr(node, i);
1301                 ptr_gen = btrfs_node_ptr_generation(node, i);
1302                 ret = readahead_tree_block(root, bytenr, blocksize, ptr_gen);
1303                 if (ret)
1304                         break;
1305         }
1306 }
1307
1308 static int walk_down_tree(struct btrfs_root *root, struct btrfs_path *path,
1309                           struct walk_control *wc, int *level)
1310 {
1311         u64 bytenr;
1312         u64 ptr_gen;
1313         struct extent_buffer *next;
1314         struct extent_buffer *cur;
1315         u32 blocksize;
1316         int ret, err = 0;
1317         u64 refs;
1318
1319         WARN_ON(*level < 0);
1320         WARN_ON(*level >= BTRFS_MAX_LEVEL);
1321         ret = btrfs_lookup_extent_info(NULL, root,
1322                                        path->nodes[*level]->start,
1323                                        *level, 1, &refs, NULL);
1324         if (ret < 0) {
1325                 err = ret;
1326                 goto out;
1327         }
1328
1329         if (refs > 1) {
1330                 ret = enter_shared_node(root, path->nodes[*level]->start,
1331                                         refs, wc, *level);
1332                 if (ret > 0) {
1333                         err = ret;
1334                         goto out;
1335                 }
1336         }
1337
1338         while (*level >= 0) {
1339                 WARN_ON(*level < 0);
1340                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1341                 cur = path->nodes[*level];
1342
1343                 if (btrfs_header_level(cur) != *level)
1344                         WARN_ON(1);
1345
1346                 if (path->slots[*level] >= btrfs_header_nritems(cur))
1347                         break;
1348                 if (*level == 0) {
1349                         ret = process_one_leaf(root, cur, wc);
1350                         break;
1351                 }
1352                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1353                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
1354                 blocksize = btrfs_level_size(root, *level - 1);
1355                 ret = btrfs_lookup_extent_info(NULL, root, bytenr, *level - 1,
1356                                                1, &refs, NULL);
1357                 if (ret < 0)
1358                         refs = 0;
1359
1360                 if (refs > 1) {
1361                         ret = enter_shared_node(root, bytenr, refs,
1362                                                 wc, *level - 1);
1363                         if (ret > 0) {
1364                                 path->slots[*level]++;
1365                                 continue;
1366                         }
1367                 }
1368
1369                 next = btrfs_find_tree_block(root, bytenr, blocksize);
1370                 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
1371                         free_extent_buffer(next);
1372                         reada_walk_down(root, cur, path->slots[*level]);
1373                         next = read_tree_block(root, bytenr, blocksize,
1374                                                ptr_gen);
1375                         if (!next) {
1376                                 err = -EIO;
1377                                 goto out;
1378                         }
1379                 }
1380
1381                 *level = *level - 1;
1382                 free_extent_buffer(path->nodes[*level]);
1383                 path->nodes[*level] = next;
1384                 path->slots[*level] = 0;
1385         }
1386 out:
1387         path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
1388         return err;
1389 }
1390
1391 static int walk_up_tree(struct btrfs_root *root, struct btrfs_path *path,
1392                         struct walk_control *wc, int *level)
1393 {
1394         int i;
1395         struct extent_buffer *leaf;
1396
1397         for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
1398                 leaf = path->nodes[i];
1399                 if (path->slots[i] + 1 < btrfs_header_nritems(leaf)) {
1400                         path->slots[i]++;
1401                         *level = i;
1402                         return 0;
1403                 } else {
1404                         free_extent_buffer(path->nodes[*level]);
1405                         path->nodes[*level] = NULL;
1406                         BUG_ON(*level > wc->active_node);
1407                         if (*level == wc->active_node)
1408                                 leave_shared_node(root, wc, *level);
1409                         *level = i + 1;
1410                 }
1411         }
1412         return 1;
1413 }
1414
1415 static int check_root_dir(struct inode_record *rec)
1416 {
1417         struct inode_backref *backref;
1418         int ret = -1;
1419
1420         if (!rec->found_inode_item || rec->errors)
1421                 goto out;
1422         if (rec->nlink != 1 || rec->found_link != 0)
1423                 goto out;
1424         if (list_empty(&rec->backrefs))
1425                 goto out;
1426         backref = list_entry(rec->backrefs.next, struct inode_backref, list);
1427         if (!backref->found_inode_ref)
1428                 goto out;
1429         if (backref->index != 0 || backref->namelen != 2 ||
1430             memcmp(backref->name, "..", 2))
1431                 goto out;
1432         if (backref->found_dir_index || backref->found_dir_item)
1433                 goto out;
1434         ret = 0;
1435 out:
1436         return ret;
1437 }
1438
1439 static int repair_inode_isize(struct btrfs_trans_handle *trans,
1440                               struct btrfs_root *root, struct btrfs_path *path,
1441                               struct inode_record *rec)
1442 {
1443         struct btrfs_inode_item *ei;
1444         struct btrfs_key key;
1445         int ret;
1446
1447         key.objectid = rec->ino;
1448         key.type = BTRFS_INODE_ITEM_KEY;
1449         key.offset = (u64)-1;
1450
1451         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1452         if (ret < 0)
1453                 goto out;
1454         if (ret) {
1455                 if (!path->slots[0]) {
1456                         ret = -ENOENT;
1457                         goto out;
1458                 }
1459                 path->slots[0]--;
1460                 ret = 0;
1461         }
1462         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1463         if (key.objectid != rec->ino) {
1464                 ret = -ENOENT;
1465                 goto out;
1466         }
1467
1468         ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
1469                             struct btrfs_inode_item);
1470         btrfs_set_inode_size(path->nodes[0], ei, rec->found_size);
1471         btrfs_mark_buffer_dirty(path->nodes[0]);
1472         rec->errors &= ~I_ERR_DIR_ISIZE_WRONG;
1473         printf("reset isize for dir %Lu root %Lu\n", rec->ino,
1474                root->root_key.objectid);
1475 out:
1476         btrfs_release_path(path);
1477         return ret;
1478 }
1479
1480 static int repair_inode_orphan_item(struct btrfs_trans_handle *trans,
1481                                     struct btrfs_root *root,
1482                                     struct btrfs_path *path,
1483                                     struct inode_record *rec)
1484 {
1485         struct btrfs_key key;
1486         int ret;
1487
1488         key.objectid = BTRFS_ORPHAN_OBJECTID;
1489         key.type = BTRFS_ORPHAN_ITEM_KEY;
1490         key.offset = rec->ino;
1491
1492         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1493         btrfs_release_path(path);
1494         if (!ret)
1495                 rec->errors &= ~I_ERR_NO_ORPHAN_ITEM;
1496         return ret;
1497 }
1498
1499 static int try_repair_inode(struct btrfs_root *root, struct inode_record *rec)
1500 {
1501         struct btrfs_trans_handle *trans;
1502         struct btrfs_path *path;
1503         int ret = 0;
1504
1505         /* So far we just fix dir isize wrong */
1506         if (!(rec->errors & (I_ERR_DIR_ISIZE_WRONG | I_ERR_NO_ORPHAN_ITEM)))
1507                 return 1;
1508
1509         path = btrfs_alloc_path();
1510         if (!path)
1511                 return -ENOMEM;
1512
1513         trans = btrfs_start_transaction(root, 1);
1514         if (IS_ERR(trans)) {
1515                 btrfs_free_path(path);
1516                 return PTR_ERR(trans);
1517         }
1518
1519         if (rec->errors & I_ERR_DIR_ISIZE_WRONG)
1520                 ret = repair_inode_isize(trans, root, path, rec);
1521         if (!ret && rec->errors & I_ERR_NO_ORPHAN_ITEM)
1522                 ret = repair_inode_orphan_item(trans, root, path, rec);
1523         btrfs_commit_transaction(trans, root);
1524         btrfs_free_path(path);
1525         return ret;
1526 }
1527
1528 static int check_inode_recs(struct btrfs_root *root,
1529                             struct cache_tree *inode_cache)
1530 {
1531         struct cache_extent *cache;
1532         struct ptr_node *node;
1533         struct inode_record *rec;
1534         struct inode_backref *backref;
1535         int ret;
1536         u64 error = 0;
1537         u64 root_dirid = btrfs_root_dirid(&root->root_item);
1538
1539         if (btrfs_root_refs(&root->root_item) == 0) {
1540                 if (!cache_tree_empty(inode_cache))
1541                         fprintf(stderr, "warning line %d\n", __LINE__);
1542                 return 0;
1543         }
1544
1545         rec = get_inode_rec(inode_cache, root_dirid, 0);
1546         if (rec) {
1547                 ret = check_root_dir(rec);
1548                 if (ret) {
1549                         fprintf(stderr, "root %llu root dir %llu error\n",
1550                                 (unsigned long long)root->root_key.objectid,
1551                                 (unsigned long long)root_dirid);
1552                         error++;
1553                 }
1554         } else {
1555                 fprintf(stderr, "root %llu root dir %llu not found\n",
1556                         (unsigned long long)root->root_key.objectid,
1557                         (unsigned long long)root_dirid);
1558         }
1559
1560         while (1) {
1561                 cache = search_cache_extent(inode_cache, 0);
1562                 if (!cache)
1563                         break;
1564                 node = container_of(cache, struct ptr_node, cache);
1565                 rec = node->data;
1566                 remove_cache_extent(inode_cache, &node->cache);
1567                 free(node);
1568                 if (rec->ino == root_dirid ||
1569                     rec->ino == BTRFS_ORPHAN_OBJECTID) {
1570                         free_inode_rec(rec);
1571                         continue;
1572                 }
1573
1574                 if (rec->errors & I_ERR_NO_ORPHAN_ITEM) {
1575                         ret = check_orphan_item(root, rec->ino);
1576                         if (ret == 0)
1577                                 rec->errors &= ~I_ERR_NO_ORPHAN_ITEM;
1578                         if (can_free_inode_rec(rec)) {
1579                                 free_inode_rec(rec);
1580                                 continue;
1581                         }
1582                 }
1583
1584                 if (repair) {
1585                         ret = try_repair_inode(root, rec);
1586                         if (ret == 0 && can_free_inode_rec(rec)) {
1587                                 free_inode_rec(rec);
1588                                 continue;
1589                         }
1590                         ret = 0;
1591                 }
1592
1593                 error++;
1594                 if (!rec->found_inode_item)
1595                         rec->errors |= I_ERR_NO_INODE_ITEM;
1596                 if (rec->found_link != rec->nlink)
1597                         rec->errors |= I_ERR_LINK_COUNT_WRONG;
1598                 fprintf(stderr, "root %llu inode %llu errors %x",
1599                         (unsigned long long) root->root_key.objectid,
1600                         (unsigned long long) rec->ino, rec->errors);
1601                 print_inode_error(rec->errors);
1602                 list_for_each_entry(backref, &rec->backrefs, list) {
1603                         if (!backref->found_dir_item)
1604                                 backref->errors |= REF_ERR_NO_DIR_ITEM;
1605                         if (!backref->found_dir_index)
1606                                 backref->errors |= REF_ERR_NO_DIR_INDEX;
1607                         if (!backref->found_inode_ref)
1608                                 backref->errors |= REF_ERR_NO_INODE_REF;
1609                         fprintf(stderr, "\tunresolved ref dir %llu index %llu"
1610                                 " namelen %u name %s filetype %d errors %x",
1611                                 (unsigned long long)backref->dir,
1612                                 (unsigned long long)backref->index,
1613                                 backref->namelen, backref->name,
1614                                 backref->filetype, backref->errors);
1615                         print_ref_error(backref->errors);
1616                 }
1617                 free_inode_rec(rec);
1618         }
1619         return (error > 0) ? -1 : 0;
1620 }
1621
1622 static struct root_record *get_root_rec(struct cache_tree *root_cache,
1623                                         u64 objectid)
1624 {
1625         struct cache_extent *cache;
1626         struct root_record *rec = NULL;
1627         int ret;
1628
1629         cache = lookup_cache_extent(root_cache, objectid, 1);
1630         if (cache) {
1631                 rec = container_of(cache, struct root_record, cache);
1632         } else {
1633                 rec = calloc(1, sizeof(*rec));
1634                 rec->objectid = objectid;
1635                 INIT_LIST_HEAD(&rec->backrefs);
1636                 rec->cache.start = objectid;
1637                 rec->cache.size = 1;
1638
1639                 ret = insert_cache_extent(root_cache, &rec->cache);
1640                 BUG_ON(ret);
1641         }
1642         return rec;
1643 }
1644
1645 static struct root_backref *get_root_backref(struct root_record *rec,
1646                                              u64 ref_root, u64 dir, u64 index,
1647                                              const char *name, int namelen)
1648 {
1649         struct root_backref *backref;
1650
1651         list_for_each_entry(backref, &rec->backrefs, list) {
1652                 if (backref->ref_root != ref_root || backref->dir != dir ||
1653                     backref->namelen != namelen)
1654                         continue;
1655                 if (memcmp(name, backref->name, namelen))
1656                         continue;
1657                 return backref;
1658         }
1659
1660         backref = malloc(sizeof(*backref) + namelen + 1);
1661         memset(backref, 0, sizeof(*backref));
1662         backref->ref_root = ref_root;
1663         backref->dir = dir;
1664         backref->index = index;
1665         backref->namelen = namelen;
1666         memcpy(backref->name, name, namelen);
1667         backref->name[namelen] = '\0';
1668         list_add_tail(&backref->list, &rec->backrefs);
1669         return backref;
1670 }
1671
1672 static void free_root_record(struct cache_extent *cache)
1673 {
1674         struct root_record *rec;
1675         struct root_backref *backref;
1676
1677         rec = container_of(cache, struct root_record, cache);
1678         while (!list_empty(&rec->backrefs)) {
1679                 backref = list_entry(rec->backrefs.next,
1680                                      struct root_backref, list);
1681                 list_del(&backref->list);
1682                 free(backref);
1683         }
1684
1685         kfree(rec);
1686 }
1687
1688 FREE_EXTENT_CACHE_BASED_TREE(root_recs, free_root_record);
1689
1690 static int add_root_backref(struct cache_tree *root_cache,
1691                             u64 root_id, u64 ref_root, u64 dir, u64 index,
1692                             const char *name, int namelen,
1693                             int item_type, int errors)
1694 {
1695         struct root_record *rec;
1696         struct root_backref *backref;
1697
1698         rec = get_root_rec(root_cache, root_id);
1699         backref = get_root_backref(rec, ref_root, dir, index, name, namelen);
1700
1701         backref->errors |= errors;
1702
1703         if (item_type != BTRFS_DIR_ITEM_KEY) {
1704                 if (backref->found_dir_index || backref->found_back_ref ||
1705                     backref->found_forward_ref) {
1706                         if (backref->index != index)
1707                                 backref->errors |= REF_ERR_INDEX_UNMATCH;
1708                 } else {
1709                         backref->index = index;
1710                 }
1711         }
1712
1713         if (item_type == BTRFS_DIR_ITEM_KEY) {
1714                 if (backref->found_forward_ref)
1715                         rec->found_ref++;
1716                 backref->found_dir_item = 1;
1717         } else if (item_type == BTRFS_DIR_INDEX_KEY) {
1718                 backref->found_dir_index = 1;
1719         } else if (item_type == BTRFS_ROOT_REF_KEY) {
1720                 if (backref->found_forward_ref)
1721                         backref->errors |= REF_ERR_DUP_ROOT_REF;
1722                 else if (backref->found_dir_item)
1723                         rec->found_ref++;
1724                 backref->found_forward_ref = 1;
1725         } else if (item_type == BTRFS_ROOT_BACKREF_KEY) {
1726                 if (backref->found_back_ref)
1727                         backref->errors |= REF_ERR_DUP_ROOT_BACKREF;
1728                 backref->found_back_ref = 1;
1729         } else {
1730                 BUG_ON(1);
1731         }
1732
1733         if (backref->found_forward_ref && backref->found_dir_item)
1734                 backref->reachable = 1;
1735         return 0;
1736 }
1737
1738 static int merge_root_recs(struct btrfs_root *root,
1739                            struct cache_tree *src_cache,
1740                            struct cache_tree *dst_cache)
1741 {
1742         struct cache_extent *cache;
1743         struct ptr_node *node;
1744         struct inode_record *rec;
1745         struct inode_backref *backref;
1746
1747         if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
1748                 free_inode_recs_tree(src_cache);
1749                 return 0;
1750         }
1751
1752         while (1) {
1753                 cache = search_cache_extent(src_cache, 0);
1754                 if (!cache)
1755                         break;
1756                 node = container_of(cache, struct ptr_node, cache);
1757                 rec = node->data;
1758                 remove_cache_extent(src_cache, &node->cache);
1759                 free(node);
1760
1761                 if (!is_child_root(root, root->objectid, rec->ino))
1762                         goto skip;
1763
1764                 list_for_each_entry(backref, &rec->backrefs, list) {
1765                         BUG_ON(backref->found_inode_ref);
1766                         if (backref->found_dir_item)
1767                                 add_root_backref(dst_cache, rec->ino,
1768                                         root->root_key.objectid, backref->dir,
1769                                         backref->index, backref->name,
1770                                         backref->namelen, BTRFS_DIR_ITEM_KEY,
1771                                         backref->errors);
1772                         if (backref->found_dir_index)
1773                                 add_root_backref(dst_cache, rec->ino,
1774                                         root->root_key.objectid, backref->dir,
1775                                         backref->index, backref->name,
1776                                         backref->namelen, BTRFS_DIR_INDEX_KEY,
1777                                         backref->errors);
1778                 }
1779 skip:
1780                 free_inode_rec(rec);
1781         }
1782         return 0;
1783 }
1784
1785 static int check_root_refs(struct btrfs_root *root,
1786                            struct cache_tree *root_cache)
1787 {
1788         struct root_record *rec;
1789         struct root_record *ref_root;
1790         struct root_backref *backref;
1791         struct cache_extent *cache;
1792         int loop = 1;
1793         int ret;
1794         int error;
1795         int errors = 0;
1796
1797         rec = get_root_rec(root_cache, BTRFS_FS_TREE_OBJECTID);
1798         rec->found_ref = 1;
1799
1800         /* fixme: this can not detect circular references */
1801         while (loop) {
1802                 loop = 0;
1803                 cache = search_cache_extent(root_cache, 0);
1804                 while (1) {
1805                         if (!cache)
1806                                 break;
1807                         rec = container_of(cache, struct root_record, cache);
1808                         cache = next_cache_extent(cache);
1809
1810                         if (rec->found_ref == 0)
1811                                 continue;
1812
1813                         list_for_each_entry(backref, &rec->backrefs, list) {
1814                                 if (!backref->reachable)
1815                                         continue;
1816
1817                                 ref_root = get_root_rec(root_cache,
1818                                                         backref->ref_root);
1819                                 if (ref_root->found_ref > 0)
1820                                         continue;
1821
1822                                 backref->reachable = 0;
1823                                 rec->found_ref--;
1824                                 if (rec->found_ref == 0)
1825                                         loop = 1;
1826                         }
1827                 }
1828         }
1829
1830         cache = search_cache_extent(root_cache, 0);
1831         while (1) {
1832                 if (!cache)
1833                         break;
1834                 rec = container_of(cache, struct root_record, cache);
1835                 cache = next_cache_extent(cache);
1836
1837                 if (rec->found_ref == 0 &&
1838                     rec->objectid >= BTRFS_FIRST_FREE_OBJECTID &&
1839                     rec->objectid <= BTRFS_LAST_FREE_OBJECTID) {
1840                         ret = check_orphan_item(root->fs_info->tree_root,
1841                                                 rec->objectid);
1842                         if (ret == 0)
1843                                 continue;
1844
1845                         /*
1846                          * If we don't have a root item then we likely just have
1847                          * a dir item in a snapshot for this root but no actual
1848                          * ref key or anything so it's meaningless.
1849                          */
1850                         if (!rec->found_root_item)
1851                                 continue;
1852                         errors++;
1853                         fprintf(stderr, "fs tree %llu not referenced\n",
1854                                 (unsigned long long)rec->objectid);
1855                 }
1856
1857                 error = 0;
1858                 if (rec->found_ref > 0 && !rec->found_root_item)
1859                         error = 1;
1860                 list_for_each_entry(backref, &rec->backrefs, list) {
1861                         if (!backref->found_dir_item)
1862                                 backref->errors |= REF_ERR_NO_DIR_ITEM;
1863                         if (!backref->found_dir_index)
1864                                 backref->errors |= REF_ERR_NO_DIR_INDEX;
1865                         if (!backref->found_back_ref)
1866                                 backref->errors |= REF_ERR_NO_ROOT_BACKREF;
1867                         if (!backref->found_forward_ref)
1868                                 backref->errors |= REF_ERR_NO_ROOT_REF;
1869                         if (backref->reachable && backref->errors)
1870                                 error = 1;
1871                 }
1872                 if (!error)
1873                         continue;
1874
1875                 errors++;
1876                 fprintf(stderr, "fs tree %llu refs %u %s\n",
1877                         (unsigned long long)rec->objectid, rec->found_ref,
1878                          rec->found_root_item ? "" : "not found");
1879
1880                 list_for_each_entry(backref, &rec->backrefs, list) {
1881                         if (!backref->reachable)
1882                                 continue;
1883                         if (!backref->errors && rec->found_root_item)
1884                                 continue;
1885                         fprintf(stderr, "\tunresolved ref root %llu dir %llu"
1886                                 " index %llu namelen %u name %s errors %x\n",
1887                                 (unsigned long long)backref->ref_root,
1888                                 (unsigned long long)backref->dir,
1889                                 (unsigned long long)backref->index,
1890                                 backref->namelen, backref->name,
1891                                 backref->errors);
1892                         print_ref_error(backref->errors);
1893                 }
1894         }
1895         return errors > 0 ? 1 : 0;
1896 }
1897
1898 static int process_root_ref(struct extent_buffer *eb, int slot,
1899                             struct btrfs_key *key,
1900                             struct cache_tree *root_cache)
1901 {
1902         u64 dirid;
1903         u64 index;
1904         u32 len;
1905         u32 name_len;
1906         struct btrfs_root_ref *ref;
1907         char namebuf[BTRFS_NAME_LEN];
1908         int error;
1909
1910         ref = btrfs_item_ptr(eb, slot, struct btrfs_root_ref);
1911
1912         dirid = btrfs_root_ref_dirid(eb, ref);
1913         index = btrfs_root_ref_sequence(eb, ref);
1914         name_len = btrfs_root_ref_name_len(eb, ref);
1915
1916         if (name_len <= BTRFS_NAME_LEN) {
1917                 len = name_len;
1918                 error = 0;
1919         } else {
1920                 len = BTRFS_NAME_LEN;
1921                 error = REF_ERR_NAME_TOO_LONG;
1922         }
1923         read_extent_buffer(eb, namebuf, (unsigned long)(ref + 1), len);
1924
1925         if (key->type == BTRFS_ROOT_REF_KEY) {
1926                 add_root_backref(root_cache, key->offset, key->objectid, dirid,
1927                                  index, namebuf, len, key->type, error);
1928         } else {
1929                 add_root_backref(root_cache, key->objectid, key->offset, dirid,
1930                                  index, namebuf, len, key->type, error);
1931         }
1932         return 0;
1933 }
1934
1935 static int check_fs_root(struct btrfs_root *root,
1936                          struct cache_tree *root_cache,
1937                          struct walk_control *wc)
1938 {
1939         int ret = 0;
1940         int wret;
1941         int level;
1942         struct btrfs_path path;
1943         struct shared_node root_node;
1944         struct root_record *rec;
1945         struct btrfs_root_item *root_item = &root->root_item;
1946
1947         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1948                 rec = get_root_rec(root_cache, root->root_key.objectid);
1949                 if (btrfs_root_refs(root_item) > 0)
1950                         rec->found_root_item = 1;
1951         }
1952
1953         btrfs_init_path(&path);
1954         memset(&root_node, 0, sizeof(root_node));
1955         cache_tree_init(&root_node.root_cache);
1956         cache_tree_init(&root_node.inode_cache);
1957
1958         level = btrfs_header_level(root->node);
1959         memset(wc->nodes, 0, sizeof(wc->nodes));
1960         wc->nodes[level] = &root_node;
1961         wc->active_node = level;
1962         wc->root_level = level;
1963
1964         if (btrfs_root_refs(root_item) > 0 ||
1965             btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
1966                 path.nodes[level] = root->node;
1967                 extent_buffer_get(root->node);
1968                 path.slots[level] = 0;
1969         } else {
1970                 struct btrfs_key key;
1971                 struct btrfs_disk_key found_key;
1972
1973                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
1974                 level = root_item->drop_level;
1975                 path.lowest_level = level;
1976                 wret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
1977                 BUG_ON(wret < 0);
1978                 btrfs_node_key(path.nodes[level], &found_key,
1979                                 path.slots[level]);
1980                 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
1981                                         sizeof(found_key)));
1982         }
1983
1984         while (1) {
1985                 wret = walk_down_tree(root, &path, wc, &level);
1986                 if (wret < 0)
1987                         ret = wret;
1988                 if (wret != 0)
1989                         break;
1990
1991                 wret = walk_up_tree(root, &path, wc, &level);
1992                 if (wret < 0)
1993                         ret = wret;
1994                 if (wret != 0)
1995                         break;
1996         }
1997         btrfs_release_path(&path);
1998
1999         merge_root_recs(root, &root_node.root_cache, root_cache);
2000
2001         if (root_node.current) {
2002                 root_node.current->checked = 1;
2003                 maybe_free_inode_rec(&root_node.inode_cache,
2004                                 root_node.current);
2005         }
2006
2007         ret = check_inode_recs(root, &root_node.inode_cache);
2008         return ret;
2009 }
2010
2011 static int fs_root_objectid(u64 objectid)
2012 {
2013         if (objectid == BTRFS_TREE_RELOC_OBJECTID ||
2014             objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2015                 return 1;
2016         return is_fstree(objectid);
2017 }
2018
2019 static int check_fs_roots(struct btrfs_root *root,
2020                           struct cache_tree *root_cache)
2021 {
2022         struct btrfs_path path;
2023         struct btrfs_key key;
2024         struct walk_control wc;
2025         struct extent_buffer *leaf;
2026         struct btrfs_root *tmp_root;
2027         struct btrfs_root *tree_root = root->fs_info->tree_root;
2028         int ret;
2029         int err = 0;
2030
2031         /*
2032          * Just in case we made any changes to the extent tree that weren't
2033          * reflected into the free space cache yet.
2034          */
2035         if (repair)
2036                 reset_cached_block_groups(root->fs_info);
2037         memset(&wc, 0, sizeof(wc));
2038         cache_tree_init(&wc.shared);
2039         btrfs_init_path(&path);
2040
2041         key.offset = 0;
2042         key.objectid = 0;
2043         key.type = BTRFS_ROOT_ITEM_KEY;
2044         ret = btrfs_search_slot(NULL, tree_root, &key, &path, 0, 0);
2045         BUG_ON(ret < 0);
2046         while (1) {
2047                 leaf = path.nodes[0];
2048                 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
2049                         ret = btrfs_next_leaf(tree_root, &path);
2050                         if (ret != 0)
2051                                 break;
2052                         leaf = path.nodes[0];
2053                 }
2054                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
2055                 if (key.type == BTRFS_ROOT_ITEM_KEY &&
2056                     fs_root_objectid(key.objectid)) {
2057                         key.offset = (u64)-1;
2058                         tmp_root = btrfs_read_fs_root(root->fs_info, &key);
2059                         if (IS_ERR(tmp_root)) {
2060                                 err = 1;
2061                                 goto next;
2062                         }
2063                         ret = check_fs_root(tmp_root, root_cache, &wc);
2064                         if (ret)
2065                                 err = 1;
2066                 } else if (key.type == BTRFS_ROOT_REF_KEY ||
2067                            key.type == BTRFS_ROOT_BACKREF_KEY) {
2068                         process_root_ref(leaf, path.slots[0], &key,
2069                                          root_cache);
2070                 }
2071 next:
2072                 path.slots[0]++;
2073         }
2074         btrfs_release_path(&path);
2075
2076         if (!cache_tree_empty(&wc.shared))
2077                 fprintf(stderr, "warning line %d\n", __LINE__);
2078
2079         return err;
2080 }
2081
2082 static int all_backpointers_checked(struct extent_record *rec, int print_errs)
2083 {
2084         struct list_head *cur = rec->backrefs.next;
2085         struct extent_backref *back;
2086         struct tree_backref *tback;
2087         struct data_backref *dback;
2088         u64 found = 0;
2089         int err = 0;
2090
2091         while(cur != &rec->backrefs) {
2092                 back = list_entry(cur, struct extent_backref, list);
2093                 cur = cur->next;
2094                 if (!back->found_extent_tree) {
2095                         err = 1;
2096                         if (!print_errs)
2097                                 goto out;
2098                         if (back->is_data) {
2099                                 dback = (struct data_backref *)back;
2100                                 fprintf(stderr, "Backref %llu %s %llu"
2101                                         " owner %llu offset %llu num_refs %lu"
2102                                         " not found in extent tree\n",
2103                                         (unsigned long long)rec->start,
2104                                         back->full_backref ?
2105                                         "parent" : "root",
2106                                         back->full_backref ?
2107                                         (unsigned long long)dback->parent:
2108                                         (unsigned long long)dback->root,
2109                                         (unsigned long long)dback->owner,
2110                                         (unsigned long long)dback->offset,
2111                                         (unsigned long)dback->num_refs);
2112                         } else {
2113                                 tback = (struct tree_backref *)back;
2114                                 fprintf(stderr, "Backref %llu parent %llu"
2115                                         " root %llu not found in extent tree\n",
2116                                         (unsigned long long)rec->start,
2117                                         (unsigned long long)tback->parent,
2118                                         (unsigned long long)tback->root);
2119                         }
2120                 }
2121                 if (!back->is_data && !back->found_ref) {
2122                         err = 1;
2123                         if (!print_errs)
2124                                 goto out;
2125                         tback = (struct tree_backref *)back;
2126                         fprintf(stderr, "Backref %llu %s %llu not referenced back %p\n",
2127                                 (unsigned long long)rec->start,
2128                                 back->full_backref ? "parent" : "root",
2129                                 back->full_backref ?
2130                                 (unsigned long long)tback->parent :
2131                                 (unsigned long long)tback->root, back);
2132                 }
2133                 if (back->is_data) {
2134                         dback = (struct data_backref *)back;
2135                         if (dback->found_ref != dback->num_refs) {
2136                                 err = 1;
2137                                 if (!print_errs)
2138                                         goto out;
2139                                 fprintf(stderr, "Incorrect local backref count"
2140                                         " on %llu %s %llu owner %llu"
2141                                         " offset %llu found %u wanted %u back %p\n",
2142                                         (unsigned long long)rec->start,
2143                                         back->full_backref ?
2144                                         "parent" : "root",
2145                                         back->full_backref ?
2146                                         (unsigned long long)dback->parent:
2147                                         (unsigned long long)dback->root,
2148                                         (unsigned long long)dback->owner,
2149                                         (unsigned long long)dback->offset,
2150                                         dback->found_ref, dback->num_refs, back);
2151                         }
2152                         if (dback->disk_bytenr != rec->start) {
2153                                 err = 1;
2154                                 if (!print_errs)
2155                                         goto out;
2156                                 fprintf(stderr, "Backref disk bytenr does not"
2157                                         " match extent record, bytenr=%llu, "
2158                                         "ref bytenr=%llu\n",
2159                                         (unsigned long long)rec->start,
2160                                         (unsigned long long)dback->disk_bytenr);
2161                         }
2162
2163                         if (dback->bytes != rec->nr) {
2164                                 err = 1;
2165                                 if (!print_errs)
2166                                         goto out;
2167                                 fprintf(stderr, "Backref bytes do not match "
2168                                         "extent backref, bytenr=%llu, ref "
2169                                         "bytes=%llu, backref bytes=%llu\n",
2170                                         (unsigned long long)rec->start,
2171                                         (unsigned long long)rec->nr,
2172                                         (unsigned long long)dback->bytes);
2173                         }
2174                 }
2175                 if (!back->is_data) {
2176                         found += 1;
2177                 } else {
2178                         dback = (struct data_backref *)back;
2179                         found += dback->found_ref;
2180                 }
2181         }
2182         if (found != rec->refs) {
2183                 err = 1;
2184                 if (!print_errs)
2185                         goto out;
2186                 fprintf(stderr, "Incorrect global backref count "
2187                         "on %llu found %llu wanted %llu\n",
2188                         (unsigned long long)rec->start,
2189                         (unsigned long long)found,
2190                         (unsigned long long)rec->refs);
2191         }
2192 out:
2193         return err;
2194 }
2195
2196 static int free_all_extent_backrefs(struct extent_record *rec)
2197 {
2198         struct extent_backref *back;
2199         struct list_head *cur;
2200         while (!list_empty(&rec->backrefs)) {
2201                 cur = rec->backrefs.next;
2202                 back = list_entry(cur, struct extent_backref, list);
2203                 list_del(cur);
2204                 free(back);
2205         }
2206         return 0;
2207 }
2208
2209 static void free_extent_record_cache(struct btrfs_fs_info *fs_info,
2210                                      struct cache_tree *extent_cache)
2211 {
2212         struct cache_extent *cache;
2213         struct extent_record *rec;
2214
2215         while (1) {
2216                 cache = first_cache_extent(extent_cache);
2217                 if (!cache)
2218                         break;
2219                 rec = container_of(cache, struct extent_record, cache);
2220                 btrfs_unpin_extent(fs_info, rec->start, rec->max_size);
2221                 remove_cache_extent(extent_cache, cache);
2222                 free_all_extent_backrefs(rec);
2223                 free(rec);
2224         }
2225 }
2226
2227 static int maybe_free_extent_rec(struct cache_tree *extent_cache,
2228                                  struct extent_record *rec)
2229 {
2230         if (rec->content_checked && rec->owner_ref_checked &&
2231             rec->extent_item_refs == rec->refs && rec->refs > 0 &&
2232             rec->num_duplicates == 0 && !all_backpointers_checked(rec, 0)) {
2233                 remove_cache_extent(extent_cache, &rec->cache);
2234                 free_all_extent_backrefs(rec);
2235                 list_del_init(&rec->list);
2236                 free(rec);
2237         }
2238         return 0;
2239 }
2240
2241 static int check_owner_ref(struct btrfs_root *root,
2242                             struct extent_record *rec,
2243                             struct extent_buffer *buf)
2244 {
2245         struct extent_backref *node;
2246         struct tree_backref *back;
2247         struct btrfs_root *ref_root;
2248         struct btrfs_key key;
2249         struct btrfs_path path;
2250         struct extent_buffer *parent;
2251         int level;
2252         int found = 0;
2253         int ret;
2254
2255         list_for_each_entry(node, &rec->backrefs, list) {
2256                 if (node->is_data)
2257                         continue;
2258                 if (!node->found_ref)
2259                         continue;
2260                 if (node->full_backref)
2261                         continue;
2262                 back = (struct tree_backref *)node;
2263                 if (btrfs_header_owner(buf) == back->root)
2264                         return 0;
2265         }
2266         BUG_ON(rec->is_root);
2267
2268         /* try to find the block by search corresponding fs tree */
2269         key.objectid = btrfs_header_owner(buf);
2270         key.type = BTRFS_ROOT_ITEM_KEY;
2271         key.offset = (u64)-1;
2272
2273         ref_root = btrfs_read_fs_root(root->fs_info, &key);
2274         if (IS_ERR(ref_root))
2275                 return 1;
2276
2277         level = btrfs_header_level(buf);
2278         if (level == 0)
2279                 btrfs_item_key_to_cpu(buf, &key, 0);
2280         else
2281                 btrfs_node_key_to_cpu(buf, &key, 0);
2282
2283         btrfs_init_path(&path);
2284         path.lowest_level = level + 1;
2285         ret = btrfs_search_slot(NULL, ref_root, &key, &path, 0, 0);
2286         if (ret < 0)
2287                 return 0;
2288
2289         parent = path.nodes[level + 1];
2290         if (parent && buf->start == btrfs_node_blockptr(parent,
2291                                                         path.slots[level + 1]))
2292                 found = 1;
2293
2294         btrfs_release_path(&path);
2295         return found ? 0 : 1;
2296 }
2297
2298 static int is_extent_tree_record(struct extent_record *rec)
2299 {
2300         struct list_head *cur = rec->backrefs.next;
2301         struct extent_backref *node;
2302         struct tree_backref *back;
2303         int is_extent = 0;
2304
2305         while(cur != &rec->backrefs) {
2306                 node = list_entry(cur, struct extent_backref, list);
2307                 cur = cur->next;
2308                 if (node->is_data)
2309                         return 0;
2310                 back = (struct tree_backref *)node;
2311                 if (node->full_backref)
2312                         return 0;
2313                 if (back->root == BTRFS_EXTENT_TREE_OBJECTID)
2314                         is_extent = 1;
2315         }
2316         return is_extent;
2317 }
2318
2319
2320 static int record_bad_block_io(struct btrfs_fs_info *info,
2321                                struct cache_tree *extent_cache,
2322                                u64 start, u64 len)
2323 {
2324         struct extent_record *rec;
2325         struct cache_extent *cache;
2326         struct btrfs_key key;
2327
2328         cache = lookup_cache_extent(extent_cache, start, len);
2329         if (!cache)
2330                 return 0;
2331
2332         rec = container_of(cache, struct extent_record, cache);
2333         if (!is_extent_tree_record(rec))
2334                 return 0;
2335
2336         btrfs_disk_key_to_cpu(&key, &rec->parent_key);
2337         return btrfs_add_corrupt_extent_record(info, &key, start, len, 0);
2338 }
2339
2340 static int swap_values(struct btrfs_root *root, struct btrfs_path *path,
2341                        struct extent_buffer *buf, int slot)
2342 {
2343         if (btrfs_header_level(buf)) {
2344                 struct btrfs_key_ptr ptr1, ptr2;
2345
2346                 read_extent_buffer(buf, &ptr1, btrfs_node_key_ptr_offset(slot),
2347                                    sizeof(struct btrfs_key_ptr));
2348                 read_extent_buffer(buf, &ptr2,
2349                                    btrfs_node_key_ptr_offset(slot + 1),
2350                                    sizeof(struct btrfs_key_ptr));
2351                 write_extent_buffer(buf, &ptr1,
2352                                     btrfs_node_key_ptr_offset(slot + 1),
2353                                     sizeof(struct btrfs_key_ptr));
2354                 write_extent_buffer(buf, &ptr2,
2355                                     btrfs_node_key_ptr_offset(slot),
2356                                     sizeof(struct btrfs_key_ptr));
2357                 if (slot == 0) {
2358                         struct btrfs_disk_key key;
2359                         btrfs_node_key(buf, &key, 0);
2360                         btrfs_fixup_low_keys(root, path, &key,
2361                                              btrfs_header_level(buf) + 1);
2362                 }
2363         } else {
2364                 struct btrfs_item *item1, *item2;
2365                 struct btrfs_key k1, k2;
2366                 char *item1_data, *item2_data;
2367                 u32 item1_offset, item2_offset, item1_size, item2_size;
2368
2369                 item1 = btrfs_item_nr(slot);
2370                 item2 = btrfs_item_nr(slot + 1);
2371                 btrfs_item_key_to_cpu(buf, &k1, slot);
2372                 btrfs_item_key_to_cpu(buf, &k2, slot + 1);
2373                 item1_offset = btrfs_item_offset(buf, item1);
2374                 item2_offset = btrfs_item_offset(buf, item2);
2375                 item1_size = btrfs_item_size(buf, item1);
2376                 item2_size = btrfs_item_size(buf, item2);
2377
2378                 item1_data = malloc(item1_size);
2379                 if (!item1_data)
2380                         return -ENOMEM;
2381                 item2_data = malloc(item2_size);
2382                 if (!item2_data) {
2383                         free(item1_data);
2384                         return -ENOMEM;
2385                 }
2386
2387                 read_extent_buffer(buf, item1_data, item1_offset, item1_size);
2388                 read_extent_buffer(buf, item2_data, item2_offset, item2_size);
2389
2390                 write_extent_buffer(buf, item1_data, item2_offset, item2_size);
2391                 write_extent_buffer(buf, item2_data, item1_offset, item1_size);
2392                 free(item1_data);
2393                 free(item2_data);
2394
2395                 btrfs_set_item_offset(buf, item1, item2_offset);
2396                 btrfs_set_item_offset(buf, item2, item1_offset);
2397                 btrfs_set_item_size(buf, item1, item2_size);
2398                 btrfs_set_item_size(buf, item2, item1_size);
2399
2400                 path->slots[0] = slot;
2401                 btrfs_set_item_key_unsafe(root, path, &k2);
2402                 path->slots[0] = slot + 1;
2403                 btrfs_set_item_key_unsafe(root, path, &k1);
2404         }
2405         return 0;
2406 }
2407
2408 /*
2409  * Attempt to fix basic block failures.  Currently we only handle bad key
2410  * orders, we will cycle through the keys and swap them if necessary.
2411  */
2412 static int try_to_fix_bad_block(struct btrfs_trans_handle *trans,
2413                                 struct btrfs_root *root,
2414                                 struct extent_buffer *buf,
2415                                 struct btrfs_disk_key *parent_key,
2416                                 enum btrfs_tree_block_status status)
2417 {
2418         struct btrfs_path *path;
2419         struct btrfs_key k1, k2;
2420         int i;
2421         int level;
2422         int ret;
2423
2424         if (status != BTRFS_TREE_BLOCK_BAD_KEY_ORDER)
2425                 return -EIO;
2426
2427         k1.objectid = btrfs_header_owner(buf);
2428         k1.type = BTRFS_ROOT_ITEM_KEY;
2429         k1.offset = (u64)-1;
2430
2431         root = btrfs_read_fs_root(root->fs_info, &k1);
2432         if (IS_ERR(root))
2433                 return -EIO;
2434
2435         path = btrfs_alloc_path();
2436         if (!path)
2437                 return -EIO;
2438
2439         level = btrfs_header_level(buf);
2440         path->lowest_level = level;
2441         path->skip_check_block = 1;
2442         if (level)
2443                 btrfs_node_key_to_cpu(buf, &k1, 0);
2444         else
2445                 btrfs_item_key_to_cpu(buf, &k1, 0);
2446
2447         ret = btrfs_search_slot(trans, root, &k1, path, 0, 1);
2448         if (ret) {
2449                 btrfs_free_path(path);
2450                 return -EIO;
2451         }
2452
2453         buf = path->nodes[level];
2454         for (i = 0; i < btrfs_header_nritems(buf) - 1; i++) {
2455                 if (level) {
2456                         btrfs_node_key_to_cpu(buf, &k1, i);
2457                         btrfs_node_key_to_cpu(buf, &k2, i + 1);
2458                 } else {
2459                         btrfs_item_key_to_cpu(buf, &k1, i);
2460                         btrfs_item_key_to_cpu(buf, &k2, i + 1);
2461                 }
2462                 if (btrfs_comp_cpu_keys(&k1, &k2) < 0)
2463                         continue;
2464                 ret = swap_values(root, path, buf, i);
2465                 if (ret)
2466                         break;
2467                 btrfs_mark_buffer_dirty(buf);
2468                 i = 0;
2469         }
2470
2471         btrfs_free_path(path);
2472         return ret;
2473 }
2474
2475 static int check_block(struct btrfs_trans_handle *trans,
2476                        struct btrfs_root *root,
2477                        struct cache_tree *extent_cache,
2478                        struct extent_buffer *buf, u64 flags)
2479 {
2480         struct extent_record *rec;
2481         struct cache_extent *cache;
2482         struct btrfs_key key;
2483         enum btrfs_tree_block_status status;
2484         int ret = 0;
2485         int level;
2486
2487         cache = lookup_cache_extent(extent_cache, buf->start, buf->len);
2488         if (!cache)
2489                 return 1;
2490         rec = container_of(cache, struct extent_record, cache);
2491         rec->generation = btrfs_header_generation(buf);
2492
2493         level = btrfs_header_level(buf);
2494         if (btrfs_header_nritems(buf) > 0) {
2495
2496                 if (level == 0)
2497                         btrfs_item_key_to_cpu(buf, &key, 0);
2498                 else
2499                         btrfs_node_key_to_cpu(buf, &key, 0);
2500
2501                 rec->info_objectid = key.objectid;
2502         }
2503         rec->info_level = level;
2504
2505         if (btrfs_is_leaf(buf))
2506                 status = btrfs_check_leaf(root, &rec->parent_key, buf);
2507         else
2508                 status = btrfs_check_node(root, &rec->parent_key, buf);
2509
2510         if (status != BTRFS_TREE_BLOCK_CLEAN) {
2511                 if (repair)
2512                         status = try_to_fix_bad_block(trans, root, buf,
2513                                                       &rec->parent_key,
2514                                                       status);
2515                 if (status != BTRFS_TREE_BLOCK_CLEAN) {
2516                         ret = -EIO;
2517                         fprintf(stderr, "bad block %llu\n",
2518                                 (unsigned long long)buf->start);
2519                 } else {
2520                         /*
2521                          * Signal to callers we need to start the scan over
2522                          * again since we'll have cow'ed blocks.
2523                          */
2524                         ret = -EAGAIN;
2525                 }
2526         } else {
2527                 rec->content_checked = 1;
2528                 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
2529                         rec->owner_ref_checked = 1;
2530                 else {
2531                         ret = check_owner_ref(root, rec, buf);
2532                         if (!ret)
2533                                 rec->owner_ref_checked = 1;
2534                 }
2535         }
2536         if (!ret)
2537                 maybe_free_extent_rec(extent_cache, rec);
2538         return ret;
2539 }
2540
2541 static struct tree_backref *find_tree_backref(struct extent_record *rec,
2542                                                 u64 parent, u64 root)
2543 {
2544         struct list_head *cur = rec->backrefs.next;
2545         struct extent_backref *node;
2546         struct tree_backref *back;
2547
2548         while(cur != &rec->backrefs) {
2549                 node = list_entry(cur, struct extent_backref, list);
2550                 cur = cur->next;
2551                 if (node->is_data)
2552                         continue;
2553                 back = (struct tree_backref *)node;
2554                 if (parent > 0) {
2555                         if (!node->full_backref)
2556                                 continue;
2557                         if (parent == back->parent)
2558                                 return back;
2559                 } else {
2560                         if (node->full_backref)
2561                                 continue;
2562                         if (back->root == root)
2563                                 return back;
2564                 }
2565         }
2566         return NULL;
2567 }
2568
2569 static struct tree_backref *alloc_tree_backref(struct extent_record *rec,
2570                                                 u64 parent, u64 root)
2571 {
2572         struct tree_backref *ref = malloc(sizeof(*ref));
2573         memset(&ref->node, 0, sizeof(ref->node));
2574         if (parent > 0) {
2575                 ref->parent = parent;
2576                 ref->node.full_backref = 1;
2577         } else {
2578                 ref->root = root;
2579                 ref->node.full_backref = 0;
2580         }
2581         list_add_tail(&ref->node.list, &rec->backrefs);
2582
2583         return ref;
2584 }
2585
2586 static struct data_backref *find_data_backref(struct extent_record *rec,
2587                                                 u64 parent, u64 root,
2588                                                 u64 owner, u64 offset,
2589                                                 int found_ref,
2590                                                 u64 disk_bytenr, u64 bytes)
2591 {
2592         struct list_head *cur = rec->backrefs.next;
2593         struct extent_backref *node;
2594         struct data_backref *back;
2595
2596         while(cur != &rec->backrefs) {
2597                 node = list_entry(cur, struct extent_backref, list);
2598                 cur = cur->next;
2599                 if (!node->is_data)
2600                         continue;
2601                 back = (struct data_backref *)node;
2602                 if (parent > 0) {
2603                         if (!node->full_backref)
2604                                 continue;
2605                         if (parent == back->parent)
2606                                 return back;
2607                 } else {
2608                         if (node->full_backref)
2609                                 continue;
2610                         if (back->root == root && back->owner == owner &&
2611                             back->offset == offset) {
2612                                 if (found_ref && node->found_ref &&
2613                                     (back->bytes != bytes ||
2614                                     back->disk_bytenr != disk_bytenr))
2615                                         continue;
2616                                 return back;
2617                         }
2618                 }
2619         }
2620         return NULL;
2621 }
2622
2623 static struct data_backref *alloc_data_backref(struct extent_record *rec,
2624                                                 u64 parent, u64 root,
2625                                                 u64 owner, u64 offset,
2626                                                 u64 max_size)
2627 {
2628         struct data_backref *ref = malloc(sizeof(*ref));
2629         memset(&ref->node, 0, sizeof(ref->node));
2630         ref->node.is_data = 1;
2631
2632         if (parent > 0) {
2633                 ref->parent = parent;
2634                 ref->owner = 0;
2635                 ref->offset = 0;
2636                 ref->node.full_backref = 1;
2637         } else {
2638                 ref->root = root;
2639                 ref->owner = owner;
2640                 ref->offset = offset;
2641                 ref->node.full_backref = 0;
2642         }
2643         ref->bytes = max_size;
2644         ref->found_ref = 0;
2645         ref->num_refs = 0;
2646         list_add_tail(&ref->node.list, &rec->backrefs);
2647         if (max_size > rec->max_size)
2648                 rec->max_size = max_size;
2649         return ref;
2650 }
2651
2652 static int add_extent_rec(struct cache_tree *extent_cache,
2653                           struct btrfs_key *parent_key, u64 parent_gen,
2654                           u64 start, u64 nr, u64 extent_item_refs,
2655                           int is_root, int inc_ref, int set_checked,
2656                           int metadata, int extent_rec, u64 max_size)
2657 {
2658         struct extent_record *rec;
2659         struct cache_extent *cache;
2660         int ret = 0;
2661         int dup = 0;
2662
2663         cache = lookup_cache_extent(extent_cache, start, nr);
2664         if (cache) {
2665                 rec = container_of(cache, struct extent_record, cache);
2666                 if (inc_ref)
2667                         rec->refs++;
2668                 if (rec->nr == 1)
2669                         rec->nr = max(nr, max_size);
2670
2671                 /*
2672                  * We need to make sure to reset nr to whatever the extent
2673                  * record says was the real size, this way we can compare it to
2674                  * the backrefs.
2675                  */
2676                 if (extent_rec) {
2677                         if (start != rec->start || rec->found_rec) {
2678                                 struct extent_record *tmp;
2679
2680                                 dup = 1;
2681                                 if (list_empty(&rec->list))
2682                                         list_add_tail(&rec->list,
2683                                                       &duplicate_extents);
2684
2685                                 /*
2686                                  * We have to do this song and dance in case we
2687                                  * find an extent record that falls inside of
2688                                  * our current extent record but does not have
2689                                  * the same objectid.
2690                                  */
2691                                 tmp = malloc(sizeof(*tmp));
2692                                 if (!tmp)
2693                                         return -ENOMEM;
2694                                 tmp->start = start;
2695                                 tmp->max_size = max_size;
2696                                 tmp->nr = nr;
2697                                 tmp->found_rec = 1;
2698                                 tmp->metadata = metadata;
2699                                 tmp->extent_item_refs = extent_item_refs;
2700                                 INIT_LIST_HEAD(&tmp->list);
2701                                 list_add_tail(&tmp->list, &rec->dups);
2702                                 rec->num_duplicates++;
2703                         } else {
2704                                 rec->nr = nr;
2705                                 rec->found_rec = 1;
2706                         }
2707                 }
2708
2709                 if (extent_item_refs && !dup) {
2710                         if (rec->extent_item_refs) {
2711                                 fprintf(stderr, "block %llu rec "
2712                                         "extent_item_refs %llu, passed %llu\n",
2713                                         (unsigned long long)start,
2714                                         (unsigned long long)
2715                                                         rec->extent_item_refs,
2716                                         (unsigned long long)extent_item_refs);
2717                         }
2718                         rec->extent_item_refs = extent_item_refs;
2719                 }
2720                 if (is_root)
2721                         rec->is_root = 1;
2722                 if (set_checked) {
2723                         rec->content_checked = 1;
2724                         rec->owner_ref_checked = 1;
2725                 }
2726
2727                 if (parent_key)
2728                         btrfs_cpu_key_to_disk(&rec->parent_key, parent_key);
2729                 if (parent_gen)
2730                         rec->parent_generation = parent_gen;
2731
2732                 if (rec->max_size < max_size)
2733                         rec->max_size = max_size;
2734
2735                 maybe_free_extent_rec(extent_cache, rec);
2736                 return ret;
2737         }
2738         rec = malloc(sizeof(*rec));
2739         rec->start = start;
2740         rec->max_size = max_size;
2741         rec->nr = max(nr, max_size);
2742         rec->found_rec = !!extent_rec;
2743         rec->content_checked = 0;
2744         rec->owner_ref_checked = 0;
2745         rec->num_duplicates = 0;
2746         rec->metadata = metadata;
2747         INIT_LIST_HEAD(&rec->backrefs);
2748         INIT_LIST_HEAD(&rec->dups);
2749         INIT_LIST_HEAD(&rec->list);
2750
2751         if (is_root)
2752                 rec->is_root = 1;
2753         else
2754                 rec->is_root = 0;
2755
2756         if (inc_ref)
2757                 rec->refs = 1;
2758         else
2759                 rec->refs = 0;
2760
2761         if (extent_item_refs)
2762                 rec->extent_item_refs = extent_item_refs;
2763         else
2764                 rec->extent_item_refs = 0;
2765
2766         if (parent_key)
2767                 btrfs_cpu_key_to_disk(&rec->parent_key, parent_key);
2768         else
2769                 memset(&rec->parent_key, 0, sizeof(*parent_key));
2770
2771         if (parent_gen)
2772                 rec->parent_generation = parent_gen;
2773         else
2774                 rec->parent_generation = 0;
2775
2776         rec->cache.start = start;
2777         rec->cache.size = nr;
2778         ret = insert_cache_extent(extent_cache, &rec->cache);
2779         BUG_ON(ret);
2780         bytes_used += nr;
2781         if (set_checked) {
2782                 rec->content_checked = 1;
2783                 rec->owner_ref_checked = 1;
2784         }
2785         return ret;
2786 }
2787
2788 static int add_tree_backref(struct cache_tree *extent_cache, u64 bytenr,
2789                             u64 parent, u64 root, int found_ref)
2790 {
2791         struct extent_record *rec;
2792         struct tree_backref *back;
2793         struct cache_extent *cache;
2794
2795         cache = lookup_cache_extent(extent_cache, bytenr, 1);
2796         if (!cache) {
2797                 add_extent_rec(extent_cache, NULL, 0, bytenr,
2798                                1, 0, 0, 0, 0, 1, 0, 0);
2799                 cache = lookup_cache_extent(extent_cache, bytenr, 1);
2800                 if (!cache)
2801                         abort();
2802         }
2803
2804         rec = container_of(cache, struct extent_record, cache);
2805         if (rec->start != bytenr) {
2806                 abort();
2807         }
2808
2809         back = find_tree_backref(rec, parent, root);
2810         if (!back)
2811                 back = alloc_tree_backref(rec, parent, root);
2812
2813         if (found_ref) {
2814                 if (back->node.found_ref) {
2815                         fprintf(stderr, "Extent back ref already exists "
2816                                 "for %llu parent %llu root %llu \n",
2817                                 (unsigned long long)bytenr,
2818                                 (unsigned long long)parent,
2819                                 (unsigned long long)root);
2820                 }
2821                 back->node.found_ref = 1;
2822         } else {
2823                 if (back->node.found_extent_tree) {
2824                         fprintf(stderr, "Extent back ref already exists "
2825                                 "for %llu parent %llu root %llu \n",
2826                                 (unsigned long long)bytenr,
2827                                 (unsigned long long)parent,
2828                                 (unsigned long long)root);
2829                 }
2830                 back->node.found_extent_tree = 1;
2831         }
2832         return 0;
2833 }
2834
2835 static int add_data_backref(struct cache_tree *extent_cache, u64 bytenr,
2836                             u64 parent, u64 root, u64 owner, u64 offset,
2837                             u32 num_refs, int found_ref, u64 max_size)
2838 {
2839         struct extent_record *rec;
2840         struct data_backref *back;
2841         struct cache_extent *cache;
2842
2843         cache = lookup_cache_extent(extent_cache, bytenr, 1);
2844         if (!cache) {
2845                 add_extent_rec(extent_cache, NULL, 0, bytenr, 1, 0, 0, 0, 0,
2846                                0, 0, max_size);
2847                 cache = lookup_cache_extent(extent_cache, bytenr, 1);
2848                 if (!cache)
2849                         abort();
2850         }
2851
2852         rec = container_of(cache, struct extent_record, cache);
2853         if (rec->max_size < max_size)
2854                 rec->max_size = max_size;
2855
2856         /*
2857          * If found_ref is set then max_size is the real size and must match the
2858          * existing refs.  So if we have already found a ref then we need to
2859          * make sure that this ref matches the existing one, otherwise we need
2860          * to add a new backref so we can notice that the backrefs don't match
2861          * and we need to figure out who is telling the truth.  This is to
2862          * account for that awful fsync bug I introduced where we'd end up with
2863          * a btrfs_file_extent_item that would have its length include multiple
2864          * prealloc extents or point inside of a prealloc extent.
2865          */
2866         back = find_data_backref(rec, parent, root, owner, offset, found_ref,
2867                                  bytenr, max_size);
2868         if (!back)
2869                 back = alloc_data_backref(rec, parent, root, owner, offset,
2870                                           max_size);
2871
2872         if (found_ref) {
2873                 BUG_ON(num_refs != 1);
2874                 if (back->node.found_ref)
2875                         BUG_ON(back->bytes != max_size);
2876                 back->node.found_ref = 1;
2877                 back->found_ref += 1;
2878                 back->bytes = max_size;
2879                 back->disk_bytenr = bytenr;
2880                 rec->refs += 1;
2881                 rec->content_checked = 1;
2882                 rec->owner_ref_checked = 1;
2883         } else {
2884                 if (back->node.found_extent_tree) {
2885                         fprintf(stderr, "Extent back ref already exists "
2886                                 "for %llu parent %llu root %llu "
2887                                 "owner %llu offset %llu num_refs %lu\n",
2888                                 (unsigned long long)bytenr,
2889                                 (unsigned long long)parent,
2890                                 (unsigned long long)root,
2891                                 (unsigned long long)owner,
2892                                 (unsigned long long)offset,
2893                                 (unsigned long)num_refs);
2894                 }
2895                 back->num_refs = num_refs;
2896                 back->node.found_extent_tree = 1;
2897         }
2898         return 0;
2899 }
2900
2901 static int add_pending(struct cache_tree *pending,
2902                        struct cache_tree *seen, u64 bytenr, u32 size)
2903 {
2904         int ret;
2905         ret = add_cache_extent(seen, bytenr, size);
2906         if (ret)
2907                 return ret;
2908         add_cache_extent(pending, bytenr, size);
2909         return 0;
2910 }
2911
2912 static int pick_next_pending(struct cache_tree *pending,
2913                         struct cache_tree *reada,
2914                         struct cache_tree *nodes,
2915                         u64 last, struct block_info *bits, int bits_nr,
2916                         int *reada_bits)
2917 {
2918         unsigned long node_start = last;
2919         struct cache_extent *cache;
2920         int ret;
2921
2922         cache = search_cache_extent(reada, 0);
2923         if (cache) {
2924                 bits[0].start = cache->start;
2925                 bits[0].size = cache->size;
2926                 *reada_bits = 1;
2927                 return 1;
2928         }
2929         *reada_bits = 0;
2930         if (node_start > 32768)
2931                 node_start -= 32768;
2932
2933         cache = search_cache_extent(nodes, node_start);
2934         if (!cache)
2935                 cache = search_cache_extent(nodes, 0);
2936
2937         if (!cache) {
2938                  cache = search_cache_extent(pending, 0);
2939                  if (!cache)
2940                          return 0;
2941                  ret = 0;
2942                  do {
2943                          bits[ret].start = cache->start;
2944                          bits[ret].size = cache->size;
2945                          cache = next_cache_extent(cache);
2946                          ret++;
2947                  } while (cache && ret < bits_nr);
2948                  return ret;
2949         }
2950
2951         ret = 0;
2952         do {
2953                 bits[ret].start = cache->start;
2954                 bits[ret].size = cache->size;
2955                 cache = next_cache_extent(cache);
2956                 ret++;
2957         } while (cache && ret < bits_nr);
2958
2959         if (bits_nr - ret > 8) {
2960                 u64 lookup = bits[0].start + bits[0].size;
2961                 struct cache_extent *next;
2962                 next = search_cache_extent(pending, lookup);
2963                 while(next) {
2964                         if (next->start - lookup > 32768)
2965                                 break;
2966                         bits[ret].start = next->start;
2967                         bits[ret].size = next->size;
2968                         lookup = next->start + next->size;
2969                         ret++;
2970                         if (ret == bits_nr)
2971                                 break;
2972                         next = next_cache_extent(next);
2973                         if (!next)
2974                                 break;
2975                 }
2976         }
2977         return ret;
2978 }
2979
2980 static void free_chunk_record(struct cache_extent *cache)
2981 {
2982         struct chunk_record *rec;
2983
2984         rec = container_of(cache, struct chunk_record, cache);
2985         free(rec);
2986 }
2987
2988 void free_chunk_cache_tree(struct cache_tree *chunk_cache)
2989 {
2990         cache_tree_free_extents(chunk_cache, free_chunk_record);
2991 }
2992
2993 static void free_device_record(struct rb_node *node)
2994 {
2995         struct device_record *rec;
2996
2997         rec = container_of(node, struct device_record, node);
2998         free(rec);
2999 }
3000
3001 FREE_RB_BASED_TREE(device_cache, free_device_record);
3002
3003 int insert_block_group_record(struct block_group_tree *tree,
3004                               struct block_group_record *bg_rec)
3005 {
3006         int ret;
3007
3008         ret = insert_cache_extent(&tree->tree, &bg_rec->cache);
3009         if (ret)
3010                 return ret;
3011
3012         list_add_tail(&bg_rec->list, &tree->block_groups);
3013         return 0;
3014 }
3015
3016 static void free_block_group_record(struct cache_extent *cache)
3017 {
3018         struct block_group_record *rec;
3019
3020         rec = container_of(cache, struct block_group_record, cache);
3021         free(rec);
3022 }
3023
3024 void free_block_group_tree(struct block_group_tree *tree)
3025 {
3026         cache_tree_free_extents(&tree->tree, free_block_group_record);
3027 }
3028
3029 int insert_device_extent_record(struct device_extent_tree *tree,
3030                                 struct device_extent_record *de_rec)
3031 {
3032         int ret;
3033
3034         /*
3035          * Device extent is a bit different from the other extents, because
3036          * the extents which belong to the different devices may have the
3037          * same start and size, so we need use the special extent cache
3038          * search/insert functions.
3039          */
3040         ret = insert_cache_extent2(&tree->tree, &de_rec->cache);
3041         if (ret)
3042                 return ret;
3043
3044         list_add_tail(&de_rec->chunk_list, &tree->no_chunk_orphans);
3045         list_add_tail(&de_rec->device_list, &tree->no_device_orphans);
3046         return 0;
3047 }
3048
3049 static void free_device_extent_record(struct cache_extent *cache)
3050 {
3051         struct device_extent_record *rec;
3052
3053         rec = container_of(cache, struct device_extent_record, cache);
3054         free(rec);
3055 }
3056
3057 void free_device_extent_tree(struct device_extent_tree *tree)
3058 {
3059         cache_tree_free_extents(&tree->tree, free_device_extent_record);
3060 }
3061
3062 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3063 static int process_extent_ref_v0(struct cache_tree *extent_cache,
3064                                  struct extent_buffer *leaf, int slot)
3065 {
3066         struct btrfs_extent_ref_v0 *ref0;
3067         struct btrfs_key key;
3068
3069         btrfs_item_key_to_cpu(leaf, &key, slot);
3070         ref0 = btrfs_item_ptr(leaf, slot, struct btrfs_extent_ref_v0);
3071         if (btrfs_ref_objectid_v0(leaf, ref0) < BTRFS_FIRST_FREE_OBJECTID) {
3072                 add_tree_backref(extent_cache, key.objectid, key.offset, 0, 0);
3073         } else {
3074                 add_data_backref(extent_cache, key.objectid, key.offset, 0,
3075                                  0, 0, btrfs_ref_count_v0(leaf, ref0), 0, 0);
3076         }
3077         return 0;
3078 }
3079 #endif
3080
3081 struct chunk_record *btrfs_new_chunk_record(struct extent_buffer *leaf,
3082                                             struct btrfs_key *key,
3083                                             int slot)
3084 {
3085         struct btrfs_chunk *ptr;
3086         struct chunk_record *rec;
3087         int num_stripes, i;
3088
3089         ptr = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3090         num_stripes = btrfs_chunk_num_stripes(leaf, ptr);
3091
3092         rec = malloc(btrfs_chunk_record_size(num_stripes));
3093         if (!rec) {
3094                 fprintf(stderr, "memory allocation failed\n");
3095                 exit(-1);
3096         }
3097
3098         memset(rec, 0, btrfs_chunk_record_size(num_stripes));
3099
3100         INIT_LIST_HEAD(&rec->list);
3101         INIT_LIST_HEAD(&rec->dextents);
3102         rec->bg_rec = NULL;
3103
3104         rec->cache.start = key->offset;
3105         rec->cache.size = btrfs_chunk_length(leaf, ptr);
3106
3107         rec->generation = btrfs_header_generation(leaf);
3108
3109         rec->objectid = key->objectid;
3110         rec->type = key->type;
3111         rec->offset = key->offset;
3112
3113         rec->length = rec->cache.size;
3114         rec->owner = btrfs_chunk_owner(leaf, ptr);
3115         rec->stripe_len = btrfs_chunk_stripe_len(leaf, ptr);
3116         rec->type_flags = btrfs_chunk_type(leaf, ptr);
3117         rec->io_width = btrfs_chunk_io_width(leaf, ptr);
3118         rec->io_align = btrfs_chunk_io_align(leaf, ptr);
3119         rec->sector_size = btrfs_chunk_sector_size(leaf, ptr);
3120         rec->num_stripes = num_stripes;
3121         rec->sub_stripes = btrfs_chunk_sub_stripes(leaf, ptr);
3122
3123         for (i = 0; i < rec->num_stripes; ++i) {
3124                 rec->stripes[i].devid =
3125                         btrfs_stripe_devid_nr(leaf, ptr, i);
3126                 rec->stripes[i].offset =
3127                         btrfs_stripe_offset_nr(leaf, ptr, i);
3128                 read_extent_buffer(leaf, rec->stripes[i].dev_uuid,
3129                                 (unsigned long)btrfs_stripe_dev_uuid_nr(ptr, i),
3130                                 BTRFS_UUID_SIZE);
3131         }
3132
3133         return rec;
3134 }
3135
3136 static int process_chunk_item(struct cache_tree *chunk_cache,
3137                               struct btrfs_key *key, struct extent_buffer *eb,
3138                               int slot)
3139 {
3140         struct chunk_record *rec;
3141         int ret = 0;
3142
3143         rec = btrfs_new_chunk_record(eb, key, slot);
3144         ret = insert_cache_extent(chunk_cache, &rec->cache);
3145         if (ret) {
3146                 fprintf(stderr, "Chunk[%llu, %llu] existed.\n",
3147                         rec->offset, rec->length);
3148                 free(rec);
3149         }
3150
3151         return ret;
3152 }
3153
3154 static int process_device_item(struct rb_root *dev_cache,
3155                 struct btrfs_key *key, struct extent_buffer *eb, int slot)
3156 {
3157         struct btrfs_dev_item *ptr;
3158         struct device_record *rec;
3159         int ret = 0;
3160
3161         ptr = btrfs_item_ptr(eb,
3162                 slot, struct btrfs_dev_item);
3163
3164         rec = malloc(sizeof(*rec));
3165         if (!rec) {
3166                 fprintf(stderr, "memory allocation failed\n");
3167                 return -ENOMEM;
3168         }
3169
3170         rec->devid = key->offset;
3171         rec->generation = btrfs_header_generation(eb);
3172
3173         rec->objectid = key->objectid;
3174         rec->type = key->type;
3175         rec->offset = key->offset;
3176
3177         rec->devid = btrfs_device_id(eb, ptr);
3178         rec->total_byte = btrfs_device_total_bytes(eb, ptr);
3179         rec->byte_used = btrfs_device_bytes_used(eb, ptr);
3180
3181         ret = rb_insert(dev_cache, &rec->node, device_record_compare);
3182         if (ret) {
3183                 fprintf(stderr, "Device[%llu] existed.\n", rec->devid);
3184                 free(rec);
3185         }
3186
3187         return ret;
3188 }
3189
3190 struct block_group_record *
3191 btrfs_new_block_group_record(struct extent_buffer *leaf, struct btrfs_key *key,
3192                              int slot)
3193 {
3194         struct btrfs_block_group_item *ptr;
3195         struct block_group_record *rec;
3196
3197         rec = malloc(sizeof(*rec));
3198         if (!rec) {
3199                 fprintf(stderr, "memory allocation failed\n");
3200                 exit(-1);
3201         }
3202         memset(rec, 0, sizeof(*rec));
3203
3204         rec->cache.start = key->objectid;
3205         rec->cache.size = key->offset;
3206
3207         rec->generation = btrfs_header_generation(leaf);
3208
3209         rec->objectid = key->objectid;
3210         rec->type = key->type;
3211         rec->offset = key->offset;
3212
3213         ptr = btrfs_item_ptr(leaf, slot, struct btrfs_block_group_item);
3214         rec->flags = btrfs_disk_block_group_flags(leaf, ptr);
3215
3216         INIT_LIST_HEAD(&rec->list);
3217
3218         return rec;
3219 }
3220
3221 static int process_block_group_item(struct block_group_tree *block_group_cache,
3222                                     struct btrfs_key *key,
3223                                     struct extent_buffer *eb, int slot)
3224 {
3225         struct block_group_record *rec;
3226         int ret = 0;
3227
3228         rec = btrfs_new_block_group_record(eb, key, slot);
3229         ret = insert_block_group_record(block_group_cache, rec);
3230         if (ret) {
3231                 fprintf(stderr, "Block Group[%llu, %llu] existed.\n",
3232                         rec->objectid, rec->offset);
3233                 free(rec);
3234         }
3235
3236         return ret;
3237 }
3238
3239 struct device_extent_record *
3240 btrfs_new_device_extent_record(struct extent_buffer *leaf,
3241                                struct btrfs_key *key, int slot)
3242 {
3243         struct device_extent_record *rec;
3244         struct btrfs_dev_extent *ptr;
3245
3246         rec = malloc(sizeof(*rec));
3247         if (!rec) {
3248                 fprintf(stderr, "memory allocation failed\n");
3249                 exit(-1);
3250         }
3251         memset(rec, 0, sizeof(*rec));
3252
3253         rec->cache.objectid = key->objectid;
3254         rec->cache.start = key->offset;
3255
3256         rec->generation = btrfs_header_generation(leaf);
3257
3258         rec->objectid = key->objectid;
3259         rec->type = key->type;
3260         rec->offset = key->offset;
3261
3262         ptr = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
3263         rec->chunk_objecteid =
3264                 btrfs_dev_extent_chunk_objectid(leaf, ptr);
3265         rec->chunk_offset =
3266                 btrfs_dev_extent_chunk_offset(leaf, ptr);
3267         rec->length = btrfs_dev_extent_length(leaf, ptr);
3268         rec->cache.size = rec->length;
3269
3270         INIT_LIST_HEAD(&rec->chunk_list);
3271         INIT_LIST_HEAD(&rec->device_list);
3272
3273         return rec;
3274 }
3275
3276 static int
3277 process_device_extent_item(struct device_extent_tree *dev_extent_cache,
3278                            struct btrfs_key *key, struct extent_buffer *eb,
3279                            int slot)
3280 {
3281         struct device_extent_record *rec;
3282         int ret;
3283
3284         rec = btrfs_new_device_extent_record(eb, key, slot);
3285         ret = insert_device_extent_record(dev_extent_cache, rec);
3286         if (ret) {
3287                 fprintf(stderr,
3288                         "Device extent[%llu, %llu, %llu] existed.\n",
3289                         rec->objectid, rec->offset, rec->length);
3290                 free(rec);
3291         }
3292
3293         return ret;
3294 }
3295
3296 static int process_extent_item(struct btrfs_root *root,
3297                                struct cache_tree *extent_cache,
3298                                struct extent_buffer *eb, int slot)
3299 {
3300         struct btrfs_extent_item *ei;
3301         struct btrfs_extent_inline_ref *iref;
3302         struct btrfs_extent_data_ref *dref;
3303         struct btrfs_shared_data_ref *sref;
3304         struct btrfs_key key;
3305         unsigned long end;
3306         unsigned long ptr;
3307         int type;
3308         u32 item_size = btrfs_item_size_nr(eb, slot);
3309         u64 refs = 0;
3310         u64 offset;
3311         u64 num_bytes;
3312         int metadata = 0;
3313
3314         btrfs_item_key_to_cpu(eb, &key, slot);
3315
3316         if (key.type == BTRFS_METADATA_ITEM_KEY) {
3317                 metadata = 1;
3318                 num_bytes = root->leafsize;
3319         } else {
3320                 num_bytes = key.offset;
3321         }
3322
3323         if (item_size < sizeof(*ei)) {
3324 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3325                 struct btrfs_extent_item_v0 *ei0;
3326                 BUG_ON(item_size != sizeof(*ei0));
3327                 ei0 = btrfs_item_ptr(eb, slot, struct btrfs_extent_item_v0);
3328                 refs = btrfs_extent_refs_v0(eb, ei0);
3329 #else
3330                 BUG();
3331 #endif
3332                 return add_extent_rec(extent_cache, NULL, 0, key.objectid,
3333                                       num_bytes, refs, 0, 0, 0, metadata, 1,
3334                                       num_bytes);
3335         }
3336
3337         ei = btrfs_item_ptr(eb, slot, struct btrfs_extent_item);
3338         refs = btrfs_extent_refs(eb, ei);
3339
3340         add_extent_rec(extent_cache, NULL, 0, key.objectid, num_bytes,
3341                        refs, 0, 0, 0, metadata, 1, num_bytes);
3342
3343         ptr = (unsigned long)(ei + 1);
3344         if (btrfs_extent_flags(eb, ei) & BTRFS_EXTENT_FLAG_TREE_BLOCK &&
3345             key.type == BTRFS_EXTENT_ITEM_KEY)
3346                 ptr += sizeof(struct btrfs_tree_block_info);
3347
3348         end = (unsigned long)ei + item_size;
3349         while (ptr < end) {
3350                 iref = (struct btrfs_extent_inline_ref *)ptr;
3351                 type = btrfs_extent_inline_ref_type(eb, iref);
3352                 offset = btrfs_extent_inline_ref_offset(eb, iref);
3353                 switch (type) {
3354                 case BTRFS_TREE_BLOCK_REF_KEY:
3355                         add_tree_backref(extent_cache, key.objectid,
3356                                          0, offset, 0);
3357                         break;
3358                 case BTRFS_SHARED_BLOCK_REF_KEY:
3359                         add_tree_backref(extent_cache, key.objectid,
3360                                          offset, 0, 0);
3361                         break;
3362                 case BTRFS_EXTENT_DATA_REF_KEY:
3363                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
3364                         add_data_backref(extent_cache, key.objectid, 0,
3365                                         btrfs_extent_data_ref_root(eb, dref),
3366                                         btrfs_extent_data_ref_objectid(eb,
3367                                                                        dref),
3368                                         btrfs_extent_data_ref_offset(eb, dref),
3369                                         btrfs_extent_data_ref_count(eb, dref),
3370                                         0, num_bytes);
3371                         break;
3372                 case BTRFS_SHARED_DATA_REF_KEY:
3373                         sref = (struct btrfs_shared_data_ref *)(iref + 1);
3374                         add_data_backref(extent_cache, key.objectid, offset,
3375                                         0, 0, 0,
3376                                         btrfs_shared_data_ref_count(eb, sref),
3377                                         0, num_bytes);
3378                         break;
3379                 default:
3380                         fprintf(stderr, "corrupt extent record: key %Lu %u %Lu\n",
3381                                 key.objectid, key.type, num_bytes);
3382                         goto out;
3383                 }
3384                 ptr += btrfs_extent_inline_ref_size(type);
3385         }
3386         WARN_ON(ptr > end);
3387 out:
3388         return 0;
3389 }
3390
3391 static int check_cache_range(struct btrfs_root *root,
3392                              struct btrfs_block_group_cache *cache,
3393                              u64 offset, u64 bytes)
3394 {
3395         struct btrfs_free_space *entry;
3396         u64 *logical;
3397         u64 bytenr;
3398         int stripe_len;
3399         int i, nr, ret;
3400
3401         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
3402                 bytenr = btrfs_sb_offset(i);
3403                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
3404                                        cache->key.objectid, bytenr, 0,
3405                                        &logical, &nr, &stripe_len);
3406                 if (ret)
3407                         return ret;
3408
3409                 while (nr--) {
3410                         if (logical[nr] + stripe_len <= offset)
3411                                 continue;
3412                         if (offset + bytes <= logical[nr])
3413                                 continue;
3414                         if (logical[nr] == offset) {
3415                                 if (stripe_len >= bytes) {
3416                                         kfree(logical);
3417                                         return 0;
3418                                 }
3419                                 bytes -= stripe_len;
3420                                 offset += stripe_len;
3421                         } else if (logical[nr] < offset) {
3422                                 if (logical[nr] + stripe_len >=
3423                                     offset + bytes) {
3424                                         kfree(logical);
3425                                         return 0;
3426                                 }
3427                                 bytes = (offset + bytes) -
3428                                         (logical[nr] + stripe_len);
3429                                 offset = logical[nr] + stripe_len;
3430                         } else {
3431                                 /*
3432                                  * Could be tricky, the super may land in the
3433                                  * middle of the area we're checking.  First
3434                                  * check the easiest case, it's at the end.
3435                                  */
3436                                 if (logical[nr] + stripe_len >=
3437                                     bytes + offset) {
3438                                         bytes = logical[nr] - offset;
3439                                         continue;
3440                                 }
3441
3442                                 /* Check the left side */
3443                                 ret = check_cache_range(root, cache,
3444                                                         offset,
3445                                                         logical[nr] - offset);
3446                                 if (ret) {
3447                                         kfree(logical);
3448                                         return ret;
3449                                 }
3450
3451                                 /* Now we continue with the right side */
3452                                 bytes = (offset + bytes) -
3453                                         (logical[nr] + stripe_len);
3454                                 offset = logical[nr] + stripe_len;
3455                         }
3456                 }
3457
3458                 kfree(logical);
3459         }
3460
3461         entry = btrfs_find_free_space(cache->free_space_ctl, offset, bytes);
3462         if (!entry) {
3463                 fprintf(stderr, "There is no free space entry for %Lu-%Lu\n",
3464                         offset, offset+bytes);
3465                 return -EINVAL;
3466         }
3467
3468         if (entry->offset != offset) {
3469                 fprintf(stderr, "Wanted offset %Lu, found %Lu\n", offset,
3470                         entry->offset);
3471                 return -EINVAL;
3472         }
3473
3474         if (entry->bytes != bytes) {
3475                 fprintf(stderr, "Wanted bytes %Lu, found %Lu for off %Lu\n",
3476                         bytes, entry->bytes, offset);
3477                 return -EINVAL;
3478         }
3479
3480         unlink_free_space(cache->free_space_ctl, entry);
3481         free(entry);
3482         return 0;
3483 }
3484
3485 static int verify_space_cache(struct btrfs_root *root,
3486                               struct btrfs_block_group_cache *cache)
3487 {
3488         struct btrfs_path *path;
3489         struct extent_buffer *leaf;
3490         struct btrfs_key key;
3491         u64 last;
3492         int ret = 0;
3493
3494         path = btrfs_alloc_path();
3495         if (!path)
3496                 return -ENOMEM;
3497
3498         root = root->fs_info->extent_root;
3499
3500         last = max_t(u64, cache->key.objectid, BTRFS_SUPER_INFO_OFFSET);
3501
3502         key.objectid = last;
3503         key.offset = 0;
3504         key.type = BTRFS_EXTENT_ITEM_KEY;
3505
3506         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3507         if (ret < 0)
3508                 goto out;
3509         ret = 0;
3510         while (1) {
3511                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3512                         ret = btrfs_next_leaf(root, path);
3513                         if (ret < 0)
3514                                 goto out;
3515                         if (ret > 0) {
3516                                 ret = 0;
3517                                 break;
3518                         }
3519                 }
3520                 leaf = path->nodes[0];
3521                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3522                 if (key.objectid >= cache->key.offset + cache->key.objectid)
3523                         break;
3524                 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
3525                     key.type != BTRFS_METADATA_ITEM_KEY) {
3526                         path->slots[0]++;
3527                         continue;
3528                 }
3529
3530                 if (last == key.objectid) {
3531                         if (key.type == BTRFS_EXTENT_ITEM_KEY)
3532                                 last = key.objectid + key.offset;
3533                         else
3534                                 last = key.objectid + root->leafsize;
3535                         path->slots[0]++;
3536                         continue;
3537                 }
3538
3539                 ret = check_cache_range(root, cache, last,
3540                                         key.objectid - last);
3541                 if (ret)
3542                         break;
3543                 if (key.type == BTRFS_EXTENT_ITEM_KEY)
3544                         last = key.objectid + key.offset;
3545                 else
3546                         last = key.objectid + root->leafsize;
3547                 path->slots[0]++;
3548         }
3549
3550         if (last < cache->key.objectid + cache->key.offset)
3551                 ret = check_cache_range(root, cache, last,
3552                                         cache->key.objectid +
3553                                         cache->key.offset - last);
3554
3555 out:
3556         btrfs_free_path(path);
3557
3558         if (!ret &&
3559             !RB_EMPTY_ROOT(&cache->free_space_ctl->free_space_offset)) {
3560                 fprintf(stderr, "There are still entries left in the space "
3561                         "cache\n");
3562                 ret = -EINVAL;
3563         }
3564
3565         return ret;
3566 }
3567
3568 static int check_space_cache(struct btrfs_root *root)
3569 {
3570         struct btrfs_block_group_cache *cache;
3571         u64 start = BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE;
3572         int ret;
3573         int error = 0;
3574
3575         if (btrfs_super_cache_generation(root->fs_info->super_copy) != -1ULL &&
3576             btrfs_super_generation(root->fs_info->super_copy) !=
3577             btrfs_super_cache_generation(root->fs_info->super_copy)) {
3578                 printf("cache and super generation don't match, space cache "
3579                        "will be invalidated\n");
3580                 return 0;
3581         }
3582
3583         while (1) {
3584                 cache = btrfs_lookup_first_block_group(root->fs_info, start);
3585                 if (!cache)
3586                         break;
3587
3588                 start = cache->key.objectid + cache->key.offset;
3589                 if (!cache->free_space_ctl) {
3590                         if (btrfs_init_free_space_ctl(cache,
3591                                                       root->sectorsize)) {
3592                                 ret = -ENOMEM;
3593                                 break;
3594                         }
3595                 } else {
3596                         btrfs_remove_free_space_cache(cache);
3597                 }
3598
3599                 ret = load_free_space_cache(root->fs_info, cache);
3600                 if (!ret)
3601                         continue;
3602
3603                 ret = verify_space_cache(root, cache);
3604                 if (ret) {
3605                         fprintf(stderr, "cache appears valid but isnt %Lu\n",
3606                                 cache->key.objectid);
3607                         error++;
3608                 }
3609         }
3610
3611         return error ? -EINVAL : 0;
3612 }
3613
3614 static int read_extent_data(struct btrfs_root *root, char *data,
3615                         u64 logical, u64 *len, int mirror)
3616 {
3617         u64 offset = 0;
3618         struct btrfs_multi_bio *multi = NULL;
3619         struct btrfs_fs_info *info = root->fs_info;
3620         struct btrfs_device *device;
3621         int ret = 0;
3622         u64 max_len = *len;
3623
3624         ret = btrfs_map_block(&info->mapping_tree, READ, logical, len,
3625                               &multi, mirror, NULL);
3626         if (ret) {
3627                 fprintf(stderr, "Couldn't map the block %llu\n",
3628                                 logical + offset);
3629                 goto err;
3630         }
3631         device = multi->stripes[0].dev;
3632
3633         if (device->fd == 0)
3634                 goto err;
3635         if (*len > max_len)
3636                 *len = max_len;
3637
3638         ret = pread64(device->fd, data, *len, multi->stripes[0].physical);
3639         if (ret != *len)
3640                 ret = -EIO;
3641         else
3642                 ret = 0;
3643 err:
3644         kfree(multi);
3645         return ret;
3646 }
3647
3648 static int check_extent_csums(struct btrfs_root *root, u64 bytenr,
3649                         u64 num_bytes, unsigned long leaf_offset,
3650                         struct extent_buffer *eb) {
3651
3652         u64 offset = 0;
3653         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
3654         char *data;
3655         unsigned long csum_offset;
3656         u32 csum;
3657         u32 csum_expected;
3658         u64 read_len;
3659         u64 data_checked = 0;
3660         u64 tmp;
3661         int ret = 0;
3662         int mirror;
3663         int num_copies;
3664
3665         if (num_bytes % root->sectorsize)
3666                 return -EINVAL;
3667
3668         data = malloc(num_bytes);
3669         if (!data)
3670                 return -ENOMEM;
3671
3672         while (offset < num_bytes) {
3673                 mirror = 0;
3674 again:
3675                 read_len = num_bytes - offset;
3676                 /* read as much space once a time */
3677                 ret = read_extent_data(root, data + offset,
3678                                 bytenr + offset, &read_len, mirror);
3679                 if (ret)
3680                         goto out;
3681                 data_checked = 0;
3682                 /* verify every 4k data's checksum */
3683                 while (data_checked < read_len) {
3684                         csum = ~(u32)0;
3685                         tmp = offset + data_checked;
3686
3687                         csum = btrfs_csum_data(NULL, (char *)data + tmp,
3688                                                csum, root->sectorsize);
3689                         btrfs_csum_final(csum, (char *)&csum);
3690
3691                         csum_offset = leaf_offset +
3692                                  tmp / root->sectorsize * csum_size;
3693                         read_extent_buffer(eb, (char *)&csum_expected,
3694                                            csum_offset, csum_size);
3695                         /* try another mirror */
3696                         if (csum != csum_expected) {
3697                                 fprintf(stderr, "mirror %d bytenr %llu csum %u expected csum %u\n",
3698                                                 mirror, bytenr + tmp,
3699                                                 csum, csum_expected);
3700                                 num_copies = btrfs_num_copies(
3701                                                 &root->fs_info->mapping_tree,
3702                                                 bytenr, num_bytes);
3703                                 if (mirror < num_copies - 1) {
3704                                         mirror += 1;
3705                                         goto again;
3706                                 }
3707                         }
3708                         data_checked += root->sectorsize;
3709                 }
3710                 offset += read_len;
3711         }
3712 out:
3713         free(data);
3714         return ret;
3715 }
3716
3717 static int check_extent_exists(struct btrfs_root *root, u64 bytenr,
3718                                u64 num_bytes)
3719 {
3720         struct btrfs_path *path;
3721         struct extent_buffer *leaf;
3722         struct btrfs_key key;
3723         int ret;
3724
3725         path = btrfs_alloc_path();
3726         if (!path) {
3727                 fprintf(stderr, "Error allocing path\n");
3728                 return -ENOMEM;
3729         }
3730
3731         key.objectid = bytenr;
3732         key.type = BTRFS_EXTENT_ITEM_KEY;
3733         key.offset = 0;
3734
3735
3736 again:
3737         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
3738                                 0, 0);
3739         if (ret < 0) {
3740                 fprintf(stderr, "Error looking up extent record %d\n", ret);
3741                 btrfs_free_path(path);
3742                 return ret;
3743         } else if (ret) {
3744                 if (path->slots[0])
3745                         path->slots[0]--;
3746                 else
3747                         btrfs_prev_leaf(root, path);
3748         }
3749
3750         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3751
3752         /*
3753          * Block group items come before extent items if they have the same
3754          * bytenr, so walk back one more just in case.  Dear future traveler,
3755          * first congrats on mastering time travel.  Now if it's not too much
3756          * trouble could you go back to 2006 and tell Chris to make the
3757          * BLOCK_GROUP_ITEM_KEY lower than the EXTENT_ITEM_KEY please?
3758          */
3759         if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
3760                 if (path->slots[0])
3761                         path->slots[0]--;
3762                 else
3763                         btrfs_prev_leaf(root, path);
3764         }
3765
3766         while (num_bytes) {
3767                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3768                         ret = btrfs_next_leaf(root, path);
3769                         if (ret < 0) {
3770                                 fprintf(stderr, "Error going to next leaf "
3771                                         "%d\n", ret);
3772                                 btrfs_free_path(path);
3773                                 return ret;
3774                         } else if (ret) {
3775                                 break;
3776                         }
3777                 }
3778                 leaf = path->nodes[0];
3779                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3780                 if (key.type != BTRFS_EXTENT_ITEM_KEY) {
3781                         path->slots[0]++;
3782                         continue;
3783                 }
3784                 if (key.objectid + key.offset < bytenr) {
3785                         path->slots[0]++;
3786                         continue;
3787                 }
3788                 if (key.objectid > bytenr + num_bytes)
3789                         break;
3790
3791                 if (key.objectid == bytenr) {
3792                         if (key.offset >= num_bytes) {
3793                                 num_bytes = 0;
3794                                 break;
3795                         }
3796                         num_bytes -= key.offset;
3797                         bytenr += key.offset;
3798                 } else if (key.objectid < bytenr) {
3799                         if (key.objectid + key.offset >= bytenr + num_bytes) {
3800                                 num_bytes = 0;
3801                                 break;
3802                         }
3803                         num_bytes = (bytenr + num_bytes) -
3804                                 (key.objectid + key.offset);
3805                         bytenr = key.objectid + key.offset;
3806                 } else {
3807                         if (key.objectid + key.offset < bytenr + num_bytes) {
3808                                 u64 new_start = key.objectid + key.offset;
3809                                 u64 new_bytes = bytenr + num_bytes - new_start;
3810
3811                                 /*
3812                                  * Weird case, the extent is in the middle of
3813                                  * our range, we'll have to search one side
3814                                  * and then the other.  Not sure if this happens
3815                                  * in real life, but no harm in coding it up
3816                                  * anyway just in case.
3817                                  */
3818                                 btrfs_release_path(path);
3819                                 ret = check_extent_exists(root, new_start,
3820                                                           new_bytes);
3821                                 if (ret) {
3822                                         fprintf(stderr, "Right section didn't "
3823                                                 "have a record\n");
3824                                         break;
3825                                 }
3826                                 num_bytes = key.objectid - bytenr;
3827                                 goto again;
3828                         }
3829                         num_bytes = key.objectid - bytenr;
3830                 }
3831                 path->slots[0]++;
3832         }
3833         ret = 0;
3834
3835         if (num_bytes) {
3836                 fprintf(stderr, "There are no extents for csum range "
3837                         "%Lu-%Lu\n", bytenr, bytenr+num_bytes);
3838                 ret = 1;
3839         }
3840
3841         btrfs_free_path(path);
3842         return ret;
3843 }
3844
3845 static int check_csums(struct btrfs_root *root)
3846 {
3847         struct btrfs_path *path;
3848         struct extent_buffer *leaf;
3849         struct btrfs_key key;
3850         u64 offset = 0, num_bytes = 0;
3851         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
3852         int errors = 0;
3853         int ret;
3854         u64 data_len;
3855         unsigned long leaf_offset;
3856
3857         root = root->fs_info->csum_root;
3858
3859         key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
3860         key.type = BTRFS_EXTENT_CSUM_KEY;
3861         key.offset = 0;
3862
3863         path = btrfs_alloc_path();
3864         if (!path)
3865                 return -ENOMEM;
3866
3867         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3868         if (ret < 0) {
3869                 fprintf(stderr, "Error searching csum tree %d\n", ret);
3870                 btrfs_free_path(path);
3871                 return ret;
3872         }
3873
3874         if (ret > 0 && path->slots[0])
3875                 path->slots[0]--;
3876         ret = 0;
3877
3878         while (1) {
3879                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3880                         ret = btrfs_next_leaf(root, path);
3881                         if (ret < 0) {
3882                                 fprintf(stderr, "Error going to next leaf "
3883                                         "%d\n", ret);
3884                                 break;
3885                         }
3886                         if (ret)
3887                                 break;
3888                 }
3889                 leaf = path->nodes[0];
3890
3891                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3892                 if (key.type != BTRFS_EXTENT_CSUM_KEY) {
3893                         path->slots[0]++;
3894                         continue;
3895                 }
3896
3897                 data_len = (btrfs_item_size_nr(leaf, path->slots[0]) /
3898                               csum_size) * root->sectorsize;
3899                 if (!check_data_csum)
3900                         goto skip_csum_check;
3901                 leaf_offset = btrfs_item_ptr_offset(leaf, path->slots[0]);
3902                 ret = check_extent_csums(root, key.offset, data_len,
3903                                          leaf_offset, leaf);
3904                 if (ret)
3905                         break;
3906 skip_csum_check:
3907                 if (!num_bytes) {
3908                         offset = key.offset;
3909                 } else if (key.offset != offset + num_bytes) {
3910                         ret = check_extent_exists(root, offset, num_bytes);
3911                         if (ret) {
3912                                 fprintf(stderr, "Csum exists for %Lu-%Lu but "
3913                                         "there is no extent record\n",
3914                                         offset, offset+num_bytes);
3915                                 errors++;
3916                         }
3917                         offset = key.offset;
3918                         num_bytes = 0;
3919                 }
3920                 num_bytes += data_len;
3921                 path->slots[0]++;
3922         }
3923
3924         btrfs_free_path(path);
3925         return errors;
3926 }
3927
3928 static int is_dropped_key(struct btrfs_key *key,
3929                           struct btrfs_key *drop_key) {
3930         if (key->objectid < drop_key->objectid)
3931                 return 1;
3932         else if (key->objectid == drop_key->objectid) {
3933                 if (key->type < drop_key->type)
3934                         return 1;
3935                 else if (key->type == drop_key->type) {
3936                         if (key->offset < drop_key->offset)
3937                                 return 1;
3938                 }
3939         }
3940         return 0;
3941 }
3942
3943 static int run_next_block(struct btrfs_trans_handle *trans,
3944                           struct btrfs_root *root,
3945                           struct block_info *bits,
3946                           int bits_nr,
3947                           u64 *last,
3948                           struct cache_tree *pending,
3949                           struct cache_tree *seen,
3950                           struct cache_tree *reada,
3951                           struct cache_tree *nodes,
3952                           struct cache_tree *extent_cache,
3953                           struct cache_tree *chunk_cache,
3954                           struct rb_root *dev_cache,
3955                           struct block_group_tree *block_group_cache,
3956                           struct device_extent_tree *dev_extent_cache,
3957                           struct btrfs_root_item *ri)
3958 {
3959         struct extent_buffer *buf;
3960         u64 bytenr;
3961         u32 size;
3962         u64 parent;
3963         u64 owner;
3964         u64 flags;
3965         u64 ptr;
3966         u64 gen = 0;
3967         int ret = 0;
3968         int i;
3969         int nritems;
3970         struct btrfs_key key;
3971         struct cache_extent *cache;
3972         int reada_bits;
3973
3974         nritems = pick_next_pending(pending, reada, nodes, *last, bits,
3975                                     bits_nr, &reada_bits);
3976         if (nritems == 0)
3977                 return 1;
3978
3979         if (!reada_bits) {
3980                 for(i = 0; i < nritems; i++) {
3981                         ret = add_cache_extent(reada, bits[i].start,
3982                                                bits[i].size);
3983                         if (ret == -EEXIST)
3984                                 continue;
3985
3986                         /* fixme, get the parent transid */
3987                         readahead_tree_block(root, bits[i].start,
3988                                              bits[i].size, 0);
3989                 }
3990         }
3991         *last = bits[0].start;
3992         bytenr = bits[0].start;
3993         size = bits[0].size;
3994
3995         cache = lookup_cache_extent(pending, bytenr, size);
3996         if (cache) {
3997                 remove_cache_extent(pending, cache);
3998                 free(cache);
3999         }
4000         cache = lookup_cache_extent(reada, bytenr, size);
4001         if (cache) {
4002                 remove_cache_extent(reada, cache);
4003                 free(cache);
4004         }
4005         cache = lookup_cache_extent(nodes, bytenr, size);
4006         if (cache) {
4007                 remove_cache_extent(nodes, cache);
4008                 free(cache);
4009         }
4010         cache = lookup_cache_extent(extent_cache, bytenr, size);
4011         if (cache) {
4012                 struct extent_record *rec;
4013
4014                 rec = container_of(cache, struct extent_record, cache);
4015                 gen = rec->parent_generation;
4016         }
4017
4018         /* fixme, get the real parent transid */
4019         buf = read_tree_block(root, bytenr, size, gen);
4020         if (!extent_buffer_uptodate(buf)) {
4021                 record_bad_block_io(root->fs_info,
4022                                     extent_cache, bytenr, size);
4023                 goto out;
4024         }
4025
4026         nritems = btrfs_header_nritems(buf);
4027
4028         /*
4029          * FIXME, this only works only if we don't have any full
4030          * backref mode.
4031          */
4032         if (!init_extent_tree) {
4033                 ret = btrfs_lookup_extent_info(NULL, root, bytenr,
4034                                        btrfs_header_level(buf), 1, NULL,
4035                                        &flags);
4036                 if (ret < 0)
4037                         flags = 0;
4038         } else {
4039                 flags = 0;
4040         }
4041
4042         if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
4043                 parent = bytenr;
4044                 owner = 0;
4045         } else {
4046                 parent = 0;
4047                 owner = btrfs_header_owner(buf);
4048         }
4049
4050         ret = check_block(trans, root, extent_cache, buf, flags);
4051         if (ret)
4052                 goto out;
4053
4054         if (btrfs_is_leaf(buf)) {
4055                 btree_space_waste += btrfs_leaf_free_space(root, buf);
4056                 for (i = 0; i < nritems; i++) {
4057                         struct btrfs_file_extent_item *fi;
4058                         btrfs_item_key_to_cpu(buf, &key, i);
4059                         if (key.type == BTRFS_EXTENT_ITEM_KEY) {
4060                                 process_extent_item(root, extent_cache, buf,
4061                                                     i);
4062                                 continue;
4063                         }
4064                         if (key.type == BTRFS_METADATA_ITEM_KEY) {
4065                                 process_extent_item(root, extent_cache, buf,
4066                                                     i);
4067                                 continue;
4068                         }
4069                         if (key.type == BTRFS_EXTENT_CSUM_KEY) {
4070                                 total_csum_bytes +=
4071                                         btrfs_item_size_nr(buf, i);
4072                                 continue;
4073                         }
4074                         if (key.type == BTRFS_CHUNK_ITEM_KEY) {
4075                                 process_chunk_item(chunk_cache, &key, buf, i);
4076                                 continue;
4077                         }
4078                         if (key.type == BTRFS_DEV_ITEM_KEY) {
4079                                 process_device_item(dev_cache, &key, buf, i);
4080                                 continue;
4081                         }
4082                         if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
4083                                 process_block_group_item(block_group_cache,
4084                                         &key, buf, i);
4085                                 continue;
4086                         }
4087                         if (key.type == BTRFS_DEV_EXTENT_KEY) {
4088                                 process_device_extent_item(dev_extent_cache,
4089                                         &key, buf, i);
4090                                 continue;
4091
4092                         }
4093                         if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
4094 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4095                                 process_extent_ref_v0(extent_cache, buf, i);
4096 #else
4097                                 BUG();
4098 #endif
4099                                 continue;
4100                         }
4101
4102                         if (key.type == BTRFS_TREE_BLOCK_REF_KEY) {
4103                                 add_tree_backref(extent_cache, key.objectid, 0,
4104                                                  key.offset, 0);
4105                                 continue;
4106                         }
4107                         if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
4108                                 add_tree_backref(extent_cache, key.objectid,
4109                                                  key.offset, 0, 0);
4110                                 continue;
4111                         }
4112                         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
4113                                 struct btrfs_extent_data_ref *ref;
4114                                 ref = btrfs_item_ptr(buf, i,
4115                                                 struct btrfs_extent_data_ref);
4116                                 add_data_backref(extent_cache,
4117                                         key.objectid, 0,
4118                                         btrfs_extent_data_ref_root(buf, ref),
4119                                         btrfs_extent_data_ref_objectid(buf,
4120                                                                        ref),
4121                                         btrfs_extent_data_ref_offset(buf, ref),
4122                                         btrfs_extent_data_ref_count(buf, ref),
4123                                         0, root->sectorsize);
4124                                 continue;
4125                         }
4126                         if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
4127                                 struct btrfs_shared_data_ref *ref;
4128                                 ref = btrfs_item_ptr(buf, i,
4129                                                 struct btrfs_shared_data_ref);
4130                                 add_data_backref(extent_cache,
4131                                         key.objectid, key.offset, 0, 0, 0,
4132                                         btrfs_shared_data_ref_count(buf, ref),
4133                                         0, root->sectorsize);
4134                                 continue;
4135                         }
4136                         if (key.type == BTRFS_ORPHAN_ITEM_KEY) {
4137                                 struct bad_item *bad;
4138
4139                                 if (key.objectid == BTRFS_ORPHAN_OBJECTID)
4140                                         continue;
4141                                 if (!owner)
4142                                         continue;
4143                                 bad = malloc(sizeof(struct bad_item));
4144                                 if (!bad)
4145                                         continue;
4146                                 INIT_LIST_HEAD(&bad->list);
4147                                 memcpy(&bad->key, &key,
4148                                        sizeof(struct btrfs_key));
4149                                 bad->root_id = owner;
4150                                 list_add_tail(&bad->list, &delete_items);
4151                                 continue;
4152                         }
4153                         if (key.type != BTRFS_EXTENT_DATA_KEY)
4154                                 continue;
4155                         fi = btrfs_item_ptr(buf, i,
4156                                             struct btrfs_file_extent_item);
4157                         if (btrfs_file_extent_type(buf, fi) ==
4158                             BTRFS_FILE_EXTENT_INLINE)
4159                                 continue;
4160                         if (btrfs_file_extent_disk_bytenr(buf, fi) == 0)
4161                                 continue;
4162
4163                         data_bytes_allocated +=
4164                                 btrfs_file_extent_disk_num_bytes(buf, fi);
4165                         if (data_bytes_allocated < root->sectorsize) {
4166                                 abort();
4167                         }
4168                         data_bytes_referenced +=
4169                                 btrfs_file_extent_num_bytes(buf, fi);
4170                         add_data_backref(extent_cache,
4171                                 btrfs_file_extent_disk_bytenr(buf, fi),
4172                                 parent, owner, key.objectid, key.offset -
4173                                 btrfs_file_extent_offset(buf, fi), 1, 1,
4174                                 btrfs_file_extent_disk_num_bytes(buf, fi));
4175                 }
4176         } else {
4177                 int level;
4178                 struct btrfs_key first_key;
4179
4180                 first_key.objectid = 0;
4181
4182                 if (nritems > 0)
4183                         btrfs_item_key_to_cpu(buf, &first_key, 0);
4184                 level = btrfs_header_level(buf);
4185                 for (i = 0; i < nritems; i++) {
4186                         ptr = btrfs_node_blockptr(buf, i);
4187                         size = btrfs_level_size(root, level - 1);
4188                         btrfs_node_key_to_cpu(buf, &key, i);
4189                         if (ri != NULL) {
4190                                 struct btrfs_key drop_key;
4191                                 btrfs_disk_key_to_cpu(&drop_key,
4192                                                       &ri->drop_progress);
4193                                 if ((level == ri->drop_level)
4194                                     && is_dropped_key(&key, &drop_key)) {
4195                                         continue;
4196                                 }
4197                         }
4198                         ret = add_extent_rec(extent_cache, &key,
4199                                              btrfs_node_ptr_generation(buf, i),
4200                                              ptr, size, 0, 0, 1, 0, 1, 0,
4201                                              size);
4202                         BUG_ON(ret);
4203
4204                         add_tree_backref(extent_cache, ptr, parent, owner, 1);
4205
4206                         if (level > 1) {
4207                                 add_pending(nodes, seen, ptr, size);
4208                         } else {
4209                                 add_pending(pending, seen, ptr, size);
4210                         }
4211                 }
4212                 btree_space_waste += (BTRFS_NODEPTRS_PER_BLOCK(root) -
4213                                       nritems) * sizeof(struct btrfs_key_ptr);
4214         }
4215         total_btree_bytes += buf->len;
4216         if (fs_root_objectid(btrfs_header_owner(buf)))
4217                 total_fs_tree_bytes += buf->len;
4218         if (btrfs_header_owner(buf) == BTRFS_EXTENT_TREE_OBJECTID)
4219                 total_extent_tree_bytes += buf->len;
4220         if (!found_old_backref &&
4221             btrfs_header_owner(buf) == BTRFS_TREE_RELOC_OBJECTID &&
4222             btrfs_header_backref_rev(buf) == BTRFS_MIXED_BACKREF_REV &&
4223             !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))
4224                 found_old_backref = 1;
4225 out:
4226         free_extent_buffer(buf);
4227         return ret;
4228 }
4229
4230 static int add_root_to_pending(struct extent_buffer *buf,
4231                                struct cache_tree *extent_cache,
4232                                struct cache_tree *pending,
4233                                struct cache_tree *seen,
4234                                struct cache_tree *nodes,
4235                                struct btrfs_key *root_key)
4236 {
4237         if (btrfs_header_level(buf) > 0)
4238                 add_pending(nodes, seen, buf->start, buf->len);
4239         else
4240                 add_pending(pending, seen, buf->start, buf->len);
4241         add_extent_rec(extent_cache, NULL, 0, buf->start, buf->len,
4242                        0, 1, 1, 0, 1, 0, buf->len);
4243
4244         if (root_key->objectid == BTRFS_TREE_RELOC_OBJECTID ||
4245             btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
4246                 add_tree_backref(extent_cache, buf->start, buf->start,
4247                                  0, 1);
4248         else
4249                 add_tree_backref(extent_cache, buf->start, 0,
4250                                  root_key->objectid, 1);
4251         return 0;
4252 }
4253
4254 /* as we fix the tree, we might be deleting blocks that
4255  * we're tracking for repair.  This hook makes sure we
4256  * remove any backrefs for blocks as we are fixing them.
4257  */
4258 static int free_extent_hook(struct btrfs_trans_handle *trans,
4259                             struct btrfs_root *root,
4260                             u64 bytenr, u64 num_bytes, u64 parent,
4261                             u64 root_objectid, u64 owner, u64 offset,
4262                             int refs_to_drop)
4263 {
4264         struct extent_record *rec;
4265         struct cache_extent *cache;
4266         int is_data;
4267         struct cache_tree *extent_cache = root->fs_info->fsck_extent_cache;
4268
4269         is_data = owner >= BTRFS_FIRST_FREE_OBJECTID;
4270         cache = lookup_cache_extent(extent_cache, bytenr, num_bytes);
4271         if (!cache)
4272                 return 0;
4273
4274         rec = container_of(cache, struct extent_record, cache);
4275         if (is_data) {
4276                 struct data_backref *back;
4277                 back = find_data_backref(rec, parent, root_objectid, owner,
4278                                          offset, 1, bytenr, num_bytes);
4279                 if (!back)
4280                         goto out;
4281                 if (back->node.found_ref) {
4282                         back->found_ref -= refs_to_drop;
4283                         if (rec->refs)
4284                                 rec->refs -= refs_to_drop;
4285                 }
4286                 if (back->node.found_extent_tree) {
4287                         back->num_refs -= refs_to_drop;
4288                         if (rec->extent_item_refs)
4289                                 rec->extent_item_refs -= refs_to_drop;
4290                 }
4291                 if (back->found_ref == 0)
4292                         back->node.found_ref = 0;
4293                 if (back->num_refs == 0)
4294                         back->node.found_extent_tree = 0;
4295
4296                 if (!back->node.found_extent_tree && back->node.found_ref) {
4297                         list_del(&back->node.list);
4298                         free(back);
4299                 }
4300         } else {
4301                 struct tree_backref *back;
4302                 back = find_tree_backref(rec, parent, root_objectid);
4303                 if (!back)
4304                         goto out;
4305                 if (back->node.found_ref) {
4306                         if (rec->refs)
4307                                 rec->refs--;
4308                         back->node.found_ref = 0;
4309                 }
4310                 if (back->node.found_extent_tree) {
4311                         if (rec->extent_item_refs)
4312                                 rec->extent_item_refs--;
4313                         back->node.found_extent_tree = 0;
4314                 }
4315                 if (!back->node.found_extent_tree && back->node.found_ref) {
4316                         list_del(&back->node.list);
4317                         free(back);
4318                 }
4319         }
4320         maybe_free_extent_rec(extent_cache, rec);
4321 out:
4322         return 0;
4323 }
4324
4325 static int delete_extent_records(struct btrfs_trans_handle *trans,
4326                                  struct btrfs_root *root,
4327                                  struct btrfs_path *path,
4328                                  u64 bytenr, u64 new_len)
4329 {
4330         struct btrfs_key key;
4331         struct btrfs_key found_key;
4332         struct extent_buffer *leaf;
4333         int ret;
4334         int slot;
4335
4336
4337         key.objectid = bytenr;
4338         key.type = (u8)-1;
4339         key.offset = (u64)-1;
4340
4341         while(1) {
4342                 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
4343                                         &key, path, 0, 1);
4344                 if (ret < 0)
4345                         break;
4346
4347                 if (ret > 0) {
4348                         ret = 0;
4349                         if (path->slots[0] == 0)
4350                                 break;
4351                         path->slots[0]--;
4352                 }
4353                 ret = 0;
4354
4355                 leaf = path->nodes[0];
4356                 slot = path->slots[0];
4357
4358                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4359                 if (found_key.objectid != bytenr)
4360                         break;
4361
4362                 if (found_key.type != BTRFS_EXTENT_ITEM_KEY &&
4363                     found_key.type != BTRFS_METADATA_ITEM_KEY &&
4364                     found_key.type != BTRFS_TREE_BLOCK_REF_KEY &&
4365                     found_key.type != BTRFS_EXTENT_DATA_REF_KEY &&
4366                     found_key.type != BTRFS_EXTENT_REF_V0_KEY &&
4367                     found_key.type != BTRFS_SHARED_BLOCK_REF_KEY &&
4368                     found_key.type != BTRFS_SHARED_DATA_REF_KEY) {
4369                         btrfs_release_path(path);
4370                         if (found_key.type == 0) {
4371                                 if (found_key.offset == 0)
4372                                         break;
4373                                 key.offset = found_key.offset - 1;
4374                                 key.type = found_key.type;
4375                         }
4376                         key.type = found_key.type - 1;
4377                         key.offset = (u64)-1;
4378                         continue;
4379                 }
4380
4381                 fprintf(stderr, "repair deleting extent record: key %Lu %u %Lu\n",
4382                         found_key.objectid, found_key.type, found_key.offset);
4383
4384                 ret = btrfs_del_item(trans, root->fs_info->extent_root, path);
4385                 if (ret)
4386                         break;
4387                 btrfs_release_path(path);
4388
4389                 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
4390                     found_key.type == BTRFS_METADATA_ITEM_KEY) {
4391                         u64 bytes = (found_key.type == BTRFS_EXTENT_ITEM_KEY) ?
4392                                 found_key.offset : root->leafsize;
4393
4394                         ret = btrfs_update_block_group(trans, root, bytenr,
4395                                                        bytes, 0, 0);
4396                         if (ret)
4397                                 break;
4398                 }
4399         }
4400
4401         btrfs_release_path(path);
4402         return ret;
4403 }
4404
4405 /*
4406  * for a single backref, this will allocate a new extent
4407  * and add the backref to it.
4408  */
4409 static int record_extent(struct btrfs_trans_handle *trans,
4410                          struct btrfs_fs_info *info,
4411                          struct btrfs_path *path,
4412                          struct extent_record *rec,
4413                          struct extent_backref *back,
4414                          int allocated, u64 flags)
4415 {
4416         int ret;
4417         struct btrfs_root *extent_root = info->extent_root;
4418         struct extent_buffer *leaf;
4419         struct btrfs_key ins_key;
4420         struct btrfs_extent_item *ei;
4421         struct tree_backref *tback;
4422         struct data_backref *dback;
4423         struct btrfs_tree_block_info *bi;
4424
4425         if (!back->is_data)
4426                 rec->max_size = max_t(u64, rec->max_size,
4427                                     info->extent_root->leafsize);
4428
4429         if (!allocated) {
4430                 u32 item_size = sizeof(*ei);
4431
4432                 if (!back->is_data)
4433                         item_size += sizeof(*bi);
4434
4435                 ins_key.objectid = rec->start;
4436                 ins_key.offset = rec->max_size;
4437                 ins_key.type = BTRFS_EXTENT_ITEM_KEY;
4438
4439                 ret = btrfs_insert_empty_item(trans, extent_root, path,
4440                                         &ins_key, item_size);
4441                 if (ret)
4442                         goto fail;
4443
4444                 leaf = path->nodes[0];
4445                 ei = btrfs_item_ptr(leaf, path->slots[0],
4446                                     struct btrfs_extent_item);
4447
4448                 btrfs_set_extent_refs(leaf, ei, 0);
4449                 btrfs_set_extent_generation(leaf, ei, rec->generation);
4450
4451                 if (back->is_data) {
4452                         btrfs_set_extent_flags(leaf, ei,
4453                                                BTRFS_EXTENT_FLAG_DATA);
4454                 } else {
4455                         struct btrfs_disk_key copy_key;;
4456
4457                         tback = (struct tree_backref *)back;
4458                         bi = (struct btrfs_tree_block_info *)(ei + 1);
4459                         memset_extent_buffer(leaf, 0, (unsigned long)bi,
4460                                              sizeof(*bi));
4461
4462                         btrfs_set_disk_key_objectid(&copy_key,
4463                                                     rec->info_objectid);
4464                         btrfs_set_disk_key_type(&copy_key, 0);
4465                         btrfs_set_disk_key_offset(&copy_key, 0);
4466
4467                         btrfs_set_tree_block_level(leaf, bi, rec->info_level);
4468                         btrfs_set_tree_block_key(leaf, bi, &copy_key);
4469
4470                         btrfs_set_extent_flags(leaf, ei,
4471                                                BTRFS_EXTENT_FLAG_TREE_BLOCK | flags);
4472                 }
4473
4474                 btrfs_mark_buffer_dirty(leaf);
4475                 ret = btrfs_update_block_group(trans, extent_root, rec->start,
4476                                                rec->max_size, 1, 0);
4477                 if (ret)
4478                         goto fail;
4479                 btrfs_release_path(path);
4480         }
4481
4482         if (back->is_data) {
4483                 u64 parent;
4484                 int i;
4485
4486                 dback = (struct data_backref *)back;
4487                 if (back->full_backref)
4488                         parent = dback->parent;
4489                 else
4490                         parent = 0;
4491
4492                 for (i = 0; i < dback->found_ref; i++) {
4493                         /* if parent != 0, we're doing a full backref
4494                          * passing BTRFS_FIRST_FREE_OBJECTID as the owner
4495                          * just makes the backref allocator create a data
4496                          * backref
4497                          */
4498                         ret = btrfs_inc_extent_ref(trans, info->extent_root,
4499                                                    rec->start, rec->max_size,
4500                                                    parent,
4501                                                    dback->root,
4502                                                    parent ?
4503                                                    BTRFS_FIRST_FREE_OBJECTID :
4504                                                    dback->owner,
4505                                                    dback->offset);
4506                         if (ret)
4507                                 break;
4508                 }
4509                 fprintf(stderr, "adding new data backref"
4510                                 " on %llu %s %llu owner %llu"
4511                                 " offset %llu found %d\n",
4512                                 (unsigned long long)rec->start,
4513                                 back->full_backref ?
4514                                 "parent" : "root",
4515                                 back->full_backref ?
4516                                 (unsigned long long)parent :
4517                                 (unsigned long long)dback->root,
4518                                 (unsigned long long)dback->owner,
4519                                 (unsigned long long)dback->offset,
4520                                 dback->found_ref);
4521         } else {
4522                 u64 parent;
4523
4524                 tback = (struct tree_backref *)back;
4525                 if (back->full_backref)
4526                         parent = tback->parent;
4527                 else
4528                         parent = 0;
4529
4530                 ret = btrfs_inc_extent_ref(trans, info->extent_root,
4531                                            rec->start, rec->max_size,
4532                                            parent, tback->root, 0, 0);
4533                 fprintf(stderr, "adding new tree backref on "
4534                         "start %llu len %llu parent %llu root %llu\n",
4535                         rec->start, rec->max_size, tback->parent, tback->root);
4536         }
4537         if (ret)
4538                 goto fail;
4539 fail:
4540         btrfs_release_path(path);
4541         return ret;
4542 }
4543
4544 struct extent_entry {
4545         u64 bytenr;
4546         u64 bytes;
4547         int count;
4548         int broken;
4549         struct list_head list;
4550 };
4551
4552 static struct extent_entry *find_entry(struct list_head *entries,
4553                                        u64 bytenr, u64 bytes)
4554 {
4555         struct extent_entry *entry = NULL;
4556
4557         list_for_each_entry(entry, entries, list) {
4558                 if (entry->bytenr == bytenr && entry->bytes == bytes)
4559                         return entry;
4560         }
4561
4562         return NULL;
4563 }
4564
4565 static struct extent_entry *find_most_right_entry(struct list_head *entries)
4566 {
4567         struct extent_entry *entry, *best = NULL, *prev = NULL;
4568
4569         list_for_each_entry(entry, entries, list) {
4570                 if (!prev) {
4571                         prev = entry;
4572                         continue;
4573                 }
4574
4575                 /*
4576                  * If there are as many broken entries as entries then we know
4577                  * not to trust this particular entry.
4578                  */
4579                 if (entry->broken == entry->count)
4580                         continue;
4581
4582                 /*
4583                  * If our current entry == best then we can't be sure our best
4584                  * is really the best, so we need to keep searching.
4585                  */
4586                 if (best && best->count == entry->count) {
4587                         prev = entry;
4588                         best = NULL;
4589                         continue;
4590                 }
4591
4592                 /* Prev == entry, not good enough, have to keep searching */
4593                 if (!prev->broken && prev->count == entry->count)
4594                         continue;
4595
4596                 if (!best)
4597                         best = (prev->count > entry->count) ? prev : entry;
4598                 else if (best->count < entry->count)
4599                         best = entry;
4600                 prev = entry;
4601         }
4602
4603         return best;
4604 }
4605
4606 static int repair_ref(struct btrfs_trans_handle *trans,
4607                       struct btrfs_fs_info *info, struct btrfs_path *path,
4608                       struct data_backref *dback, struct extent_entry *entry)
4609 {
4610         struct btrfs_root *root;
4611         struct btrfs_file_extent_item *fi;
4612         struct extent_buffer *leaf;
4613         struct btrfs_key key;
4614         u64 bytenr, bytes;
4615         int ret;
4616
4617         key.objectid = dback->root;
4618         key.type = BTRFS_ROOT_ITEM_KEY;
4619         key.offset = (u64)-1;
4620         root = btrfs_read_fs_root(info, &key);
4621         if (IS_ERR(root)) {
4622                 fprintf(stderr, "Couldn't find root for our ref\n");
4623                 return -EINVAL;
4624         }
4625
4626         /*
4627          * The backref points to the original offset of the extent if it was
4628          * split, so we need to search down to the offset we have and then walk
4629          * forward until we find the backref we're looking for.
4630          */
4631         key.objectid = dback->owner;
4632         key.type = BTRFS_EXTENT_DATA_KEY;
4633         key.offset = dback->offset;
4634         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4635         if (ret < 0) {
4636                 fprintf(stderr, "Error looking up ref %d\n", ret);
4637                 return ret;
4638         }
4639
4640         while (1) {
4641                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
4642                         ret = btrfs_next_leaf(root, path);
4643                         if (ret) {
4644                                 fprintf(stderr, "Couldn't find our ref, next\n");
4645                                 return -EINVAL;
4646                         }
4647                 }
4648                 leaf = path->nodes[0];
4649                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4650                 if (key.objectid != dback->owner ||
4651                     key.type != BTRFS_EXTENT_DATA_KEY) {
4652                         fprintf(stderr, "Couldn't find our ref, search\n");
4653                         return -EINVAL;
4654                 }
4655                 fi = btrfs_item_ptr(leaf, path->slots[0],
4656                                     struct btrfs_file_extent_item);
4657                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4658                 bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4659
4660                 if (bytenr == dback->disk_bytenr && bytes == dback->bytes)
4661                         break;
4662                 path->slots[0]++;
4663         }
4664
4665         btrfs_release_path(path);
4666
4667         /*
4668          * Have to make sure that this root gets updated when we commit the
4669          * transaction
4670          */
4671         root->track_dirty = 1;
4672         if (root->last_trans != trans->transid) {
4673                 root->last_trans = trans->transid;
4674                 root->commit_root = root->node;
4675                 extent_buffer_get(root->node);
4676         }
4677
4678         /*
4679          * Ok we have the key of the file extent we want to fix, now we can cow
4680          * down to the thing and fix it.
4681          */
4682         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4683         if (ret < 0) {
4684                 fprintf(stderr, "Error cowing down to ref [%Lu, %u, %Lu]: %d\n",
4685                         key.objectid, key.type, key.offset, ret);
4686                 return ret;
4687         }
4688         if (ret > 0) {
4689                 fprintf(stderr, "Well that's odd, we just found this key "
4690                         "[%Lu, %u, %Lu]\n", key.objectid, key.type,
4691                         key.offset);
4692                 return -EINVAL;
4693         }
4694         leaf = path->nodes[0];
4695         fi = btrfs_item_ptr(leaf, path->slots[0],
4696                             struct btrfs_file_extent_item);
4697
4698         if (btrfs_file_extent_compression(leaf, fi) &&
4699             dback->disk_bytenr != entry->bytenr) {
4700                 fprintf(stderr, "Ref doesn't match the record start and is "
4701                         "compressed, please take a btrfs-image of this file "
4702                         "system and send it to a btrfs developer so they can "
4703                         "complete this functionality for bytenr %Lu\n",
4704                         dback->disk_bytenr);
4705                 return -EINVAL;
4706         }
4707
4708         if (dback->node.broken && dback->disk_bytenr != entry->bytenr) {
4709                 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
4710         } else if (dback->disk_bytenr > entry->bytenr) {
4711                 u64 off_diff, offset;
4712
4713                 off_diff = dback->disk_bytenr - entry->bytenr;
4714                 offset = btrfs_file_extent_offset(leaf, fi);
4715                 if (dback->disk_bytenr + offset +
4716                     btrfs_file_extent_num_bytes(leaf, fi) >
4717                     entry->bytenr + entry->bytes) {
4718                         fprintf(stderr, "Ref is past the entry end, please "
4719                                 "take a btrfs-image of this file system and "
4720                                 "send it to a btrfs developer, ref %Lu\n",
4721                                 dback->disk_bytenr);
4722                         return -EINVAL;
4723                 }
4724                 offset += off_diff;
4725                 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
4726                 btrfs_set_file_extent_offset(leaf, fi, offset);
4727         } else if (dback->disk_bytenr < entry->bytenr) {
4728                 u64 offset;
4729
4730                 offset = btrfs_file_extent_offset(leaf, fi);
4731                 if (dback->disk_bytenr + offset < entry->bytenr) {
4732                         fprintf(stderr, "Ref is before the entry start, please"
4733                                 " take a btrfs-image of this file system and "
4734                                 "send it to a btrfs developer, ref %Lu\n",
4735                                 dback->disk_bytenr);
4736                         return -EINVAL;
4737                 }
4738
4739                 offset += dback->disk_bytenr;
4740                 offset -= entry->bytenr;
4741                 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
4742                 btrfs_set_file_extent_offset(leaf, fi, offset);
4743         }
4744
4745         btrfs_set_file_extent_disk_num_bytes(leaf, fi, entry->bytes);
4746
4747         /*
4748          * Chances are if disk_num_bytes were wrong then so is ram_bytes, but
4749          * only do this if we aren't using compression, otherwise it's a
4750          * trickier case.
4751          */
4752         if (!btrfs_file_extent_compression(leaf, fi))
4753                 btrfs_set_file_extent_ram_bytes(leaf, fi, entry->bytes);
4754         else
4755                 printf("ram bytes may be wrong?\n");
4756         btrfs_mark_buffer_dirty(leaf);
4757         btrfs_release_path(path);
4758         return 0;
4759 }
4760
4761 static int verify_backrefs(struct btrfs_trans_handle *trans,
4762                            struct btrfs_fs_info *info, struct btrfs_path *path,
4763                            struct extent_record *rec)
4764 {
4765         struct extent_backref *back;
4766         struct data_backref *dback;
4767         struct extent_entry *entry, *best = NULL;
4768         LIST_HEAD(entries);
4769         int nr_entries = 0;
4770         int broken_entries = 0;
4771         int ret = 0;
4772         short mismatch = 0;
4773
4774         /*
4775          * Metadata is easy and the backrefs should always agree on bytenr and
4776          * size, if not we've got bigger issues.
4777          */
4778         if (rec->metadata)
4779                 return 0;
4780
4781         list_for_each_entry(back, &rec->backrefs, list) {
4782                 dback = (struct data_backref *)back;
4783                 /*
4784                  * We only pay attention to backrefs that we found a real
4785                  * backref for.
4786                  */
4787                 if (dback->found_ref == 0)
4788                         continue;
4789                 if (back->full_backref)
4790                         continue;
4791
4792                 /*
4793                  * For now we only catch when the bytes don't match, not the
4794                  * bytenr.  We can easily do this at the same time, but I want
4795                  * to have a fs image to test on before we just add repair
4796                  * functionality willy-nilly so we know we won't screw up the
4797                  * repair.
4798                  */
4799
4800                 entry = find_entry(&entries, dback->disk_bytenr,
4801                                    dback->bytes);
4802                 if (!entry) {
4803                         entry = malloc(sizeof(struct extent_entry));
4804                         if (!entry) {
4805                                 ret = -ENOMEM;
4806                                 goto out;
4807                         }
4808                         memset(entry, 0, sizeof(*entry));
4809                         entry->bytenr = dback->disk_bytenr;
4810                         entry->bytes = dback->bytes;
4811                         list_add_tail(&entry->list, &entries);
4812                         nr_entries++;
4813                 }
4814
4815                 /*
4816                  * If we only have on entry we may think the entries agree when
4817                  * in reality they don't so we have to do some extra checking.
4818                  */
4819                 if (dback->disk_bytenr != rec->start ||
4820                     dback->bytes != rec->nr || back->broken)
4821                         mismatch = 1;
4822
4823                 if (back->broken) {
4824                         entry->broken++;
4825                         broken_entries++;
4826                 }
4827
4828                 entry->count++;
4829         }
4830
4831         /* Yay all the backrefs agree, carry on good sir */
4832         if (nr_entries <= 1 && !mismatch)
4833                 goto out;
4834
4835         fprintf(stderr, "attempting to repair backref discrepency for bytenr "
4836                 "%Lu\n", rec->start);
4837
4838         /*
4839          * First we want to see if the backrefs can agree amongst themselves who
4840          * is right, so figure out which one of the entries has the highest
4841          * count.
4842          */
4843         best = find_most_right_entry(&entries);
4844
4845         /*
4846          * Ok so we may have an even split between what the backrefs think, so
4847          * this is where we use the extent ref to see what it thinks.
4848          */
4849         if (!best) {
4850                 entry = find_entry(&entries, rec->start, rec->nr);
4851                 if (!entry && (!broken_entries || !rec->found_rec)) {
4852                         fprintf(stderr, "Backrefs don't agree with each other "
4853                                 "and extent record doesn't agree with anybody,"
4854                                 " so we can't fix bytenr %Lu bytes %Lu\n",
4855                                 rec->start, rec->nr);
4856                         ret = -EINVAL;
4857                         goto out;
4858                 } else if (!entry) {
4859                         /*
4860                          * Ok our backrefs were broken, we'll assume this is the
4861                          * correct value and add an entry for this range.
4862                          */
4863                         entry = malloc(sizeof(struct extent_entry));
4864                         if (!entry) {
4865                                 ret = -ENOMEM;
4866                                 goto out;
4867                         }
4868                         memset(entry, 0, sizeof(*entry));
4869                         entry->bytenr = rec->start;
4870                         entry->bytes = rec->nr;
4871                         list_add_tail(&entry->list, &entries);
4872                         nr_entries++;
4873                 }
4874                 entry->count++;
4875                 best = find_most_right_entry(&entries);
4876                 if (!best) {
4877                         fprintf(stderr, "Backrefs and extent record evenly "
4878                                 "split on who is right, this is going to "
4879                                 "require user input to fix bytenr %Lu bytes "
4880                                 "%Lu\n", rec->start, rec->nr);
4881                         ret = -EINVAL;
4882                         goto out;
4883                 }
4884         }
4885
4886         /*
4887          * I don't think this can happen currently as we'll abort() if we catch
4888          * this case higher up, but in case somebody removes that we still can't
4889          * deal with it properly here yet, so just bail out of that's the case.
4890          */
4891         if (best->bytenr != rec->start) {
4892                 fprintf(stderr, "Extent start and backref starts don't match, "
4893                         "please use btrfs-image on this file system and send "
4894                         "it to a btrfs developer so they can make fsck fix "
4895                         "this particular case.  bytenr is %Lu, bytes is %Lu\n",
4896                         rec->start, rec->nr);
4897                 ret = -EINVAL;
4898                 goto out;
4899         }
4900
4901         /*
4902          * Ok great we all agreed on an extent record, let's go find the real
4903          * references and fix up the ones that don't match.
4904          */
4905         list_for_each_entry(back, &rec->backrefs, list) {
4906                 dback = (struct data_backref *)back;
4907
4908                 /*
4909                  * Still ignoring backrefs that don't have a real ref attached
4910                  * to them.
4911                  */
4912                 if (dback->found_ref == 0)
4913                         continue;
4914                 if (back->full_backref)
4915                         continue;
4916
4917                 if (dback->bytes == best->bytes &&
4918                     dback->disk_bytenr == best->bytenr)
4919                         continue;
4920
4921                 ret = repair_ref(trans, info, path, dback, best);
4922                 if (ret)
4923                         goto out;
4924         }
4925
4926         /*
4927          * Ok we messed with the actual refs, which means we need to drop our
4928          * entire cache and go back and rescan.  I know this is a huge pain and
4929          * adds a lot of extra work, but it's the only way to be safe.  Once all
4930          * the backrefs agree we may not need to do anything to the extent
4931          * record itself.
4932          */
4933         ret = -EAGAIN;
4934 out:
4935         while (!list_empty(&entries)) {
4936                 entry = list_entry(entries.next, struct extent_entry, list);
4937                 list_del_init(&entry->list);
4938                 free(entry);
4939         }
4940         return ret;
4941 }
4942
4943 static int process_duplicates(struct btrfs_root *root,
4944                               struct cache_tree *extent_cache,
4945                               struct extent_record *rec)
4946 {
4947         struct extent_record *good, *tmp;
4948         struct cache_extent *cache;
4949         int ret;
4950
4951         /*
4952          * If we found a extent record for this extent then return, or if we
4953          * have more than one duplicate we are likely going to need to delete
4954          * something.
4955          */
4956         if (rec->found_rec || rec->num_duplicates > 1)
4957                 return 0;
4958
4959         /* Shouldn't happen but just in case */
4960         BUG_ON(!rec->num_duplicates);
4961
4962         /*
4963          * So this happens if we end up with a backref that doesn't match the
4964          * actual extent entry.  So either the backref is bad or the extent
4965          * entry is bad.  Either way we want to have the extent_record actually
4966          * reflect what we found in the extent_tree, so we need to take the
4967          * duplicate out and use that as the extent_record since the only way we
4968          * get a duplicate is if we find a real life BTRFS_EXTENT_ITEM_KEY.
4969          */
4970         remove_cache_extent(extent_cache, &rec->cache);
4971
4972         good = list_entry(rec->dups.next, struct extent_record, list);
4973         list_del_init(&good->list);
4974         INIT_LIST_HEAD(&good->backrefs);
4975         INIT_LIST_HEAD(&good->dups);
4976         good->cache.start = good->start;
4977         good->cache.size = good->nr;
4978         good->content_checked = 0;
4979         good->owner_ref_checked = 0;
4980         good->num_duplicates = 0;
4981         good->refs = rec->refs;
4982         list_splice_init(&rec->backrefs, &good->backrefs);
4983         while (1) {
4984                 cache = lookup_cache_extent(extent_cache, good->start,
4985                                             good->nr);
4986                 if (!cache)
4987                         break;
4988                 tmp = container_of(cache, struct extent_record, cache);
4989
4990                 /*
4991                  * If we find another overlapping extent and it's found_rec is
4992                  * set then it's a duplicate and we need to try and delete
4993                  * something.
4994                  */
4995                 if (tmp->found_rec || tmp->num_duplicates > 0) {
4996                         if (list_empty(&good->list))
4997                                 list_add_tail(&good->list,
4998                                               &duplicate_extents);
4999                         good->num_duplicates += tmp->num_duplicates + 1;
5000                         list_splice_init(&tmp->dups, &good->dups);
5001                         list_del_init(&tmp->list);
5002                         list_add_tail(&tmp->list, &good->dups);
5003                         remove_cache_extent(extent_cache, &tmp->cache);
5004                         continue;
5005                 }
5006
5007                 /*
5008                  * Ok we have another non extent item backed extent rec, so lets
5009                  * just add it to this extent and carry on like we did above.
5010                  */
5011                 good->refs += tmp->refs;
5012                 list_splice_init(&tmp->backrefs, &good->backrefs);
5013                 remove_cache_extent(extent_cache, &tmp->cache);
5014                 free(tmp);
5015         }
5016         ret = insert_cache_extent(extent_cache, &good->cache);
5017         BUG_ON(ret);
5018         free(rec);
5019         return good->num_duplicates ? 0 : 1;
5020 }
5021
5022 static int delete_duplicate_records(struct btrfs_trans_handle *trans,
5023                                     struct btrfs_root *root,
5024                                     struct extent_record *rec)
5025 {
5026         LIST_HEAD(delete_list);
5027         struct btrfs_path *path;
5028         struct extent_record *tmp, *good, *n;
5029         int nr_del = 0;
5030         int ret = 0;
5031         struct btrfs_key key;
5032
5033         path = btrfs_alloc_path();
5034         if (!path) {
5035                 ret = -ENOMEM;
5036                 goto out;
5037         }
5038
5039         good = rec;
5040         /* Find the record that covers all of the duplicates. */
5041         list_for_each_entry(tmp, &rec->dups, list) {
5042                 if (good->start < tmp->start)
5043                         continue;
5044                 if (good->nr > tmp->nr)
5045                         continue;
5046
5047                 if (tmp->start + tmp->nr < good->start + good->nr) {
5048                         fprintf(stderr, "Ok we have overlapping extents that "
5049                                 "aren't completely covered by eachother, this "
5050                                 "is going to require more careful thought.  "
5051                                 "The extents are [%Lu-%Lu] and [%Lu-%Lu]\n",
5052                                 tmp->start, tmp->nr, good->start, good->nr);
5053                         abort();
5054                 }
5055                 good = tmp;
5056         }
5057
5058         if (good != rec)
5059                 list_add_tail(&rec->list, &delete_list);
5060
5061         list_for_each_entry_safe(tmp, n, &rec->dups, list) {
5062                 if (tmp == good)
5063                         continue;
5064                 list_move_tail(&tmp->list, &delete_list);
5065         }
5066
5067         root = root->fs_info->extent_root;
5068         list_for_each_entry(tmp, &delete_list, list) {
5069                 if (tmp->found_rec == 0)
5070                         continue;
5071                 key.objectid = tmp->start;
5072                 key.type = BTRFS_EXTENT_ITEM_KEY;
5073                 key.offset = tmp->nr;
5074
5075                 /* Shouldn't happen but just in case */
5076                 if (tmp->metadata) {
5077                         fprintf(stderr, "Well this shouldn't happen, extent "
5078                                 "record overlaps but is metadata? "
5079                                 "[%Lu, %Lu]\n", tmp->start, tmp->nr);
5080                         abort();
5081                 }
5082
5083                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
5084                 if (ret) {
5085                         if (ret > 0)
5086                                 ret = -EINVAL;
5087                         goto out;
5088                 }
5089                 ret = btrfs_del_item(trans, root, path);
5090                 if (ret)
5091                         goto out;
5092                 btrfs_release_path(path);
5093                 nr_del++;
5094         }
5095
5096 out:
5097         while (!list_empty(&delete_list)) {
5098                 tmp = list_entry(delete_list.next, struct extent_record, list);
5099                 list_del_init(&tmp->list);
5100                 if (tmp == rec)
5101                         continue;
5102                 free(tmp);
5103         }
5104
5105         while (!list_empty(&rec->dups)) {
5106                 tmp = list_entry(rec->dups.next, struct extent_record, list);
5107                 list_del_init(&tmp->list);
5108                 free(tmp);
5109         }
5110
5111         btrfs_free_path(path);
5112
5113         if (!ret && !nr_del)
5114                 rec->num_duplicates = 0;
5115
5116         return ret ? ret : nr_del;
5117 }
5118
5119 static int find_possible_backrefs(struct btrfs_trans_handle *trans,
5120                                   struct btrfs_fs_info *info,
5121                                   struct btrfs_path *path,
5122                                   struct cache_tree *extent_cache,
5123                                   struct extent_record *rec)
5124 {
5125         struct btrfs_root *root;
5126         struct extent_backref *back;
5127         struct data_backref *dback;
5128         struct cache_extent *cache;
5129         struct btrfs_file_extent_item *fi;
5130         struct btrfs_key key;
5131         u64 bytenr, bytes;
5132         int ret;
5133
5134         list_for_each_entry(back, &rec->backrefs, list) {
5135                 dback = (struct data_backref *)back;
5136
5137                 /* We found this one, we don't need to do a lookup */
5138                 if (dback->found_ref)
5139                         continue;
5140                 /* Don't care about full backrefs (poor unloved backrefs) */
5141                 if (back->full_backref)
5142                         continue;
5143                 key.objectid = dback->root;
5144                 key.type = BTRFS_ROOT_ITEM_KEY;
5145                 key.offset = (u64)-1;
5146
5147                 root = btrfs_read_fs_root(info, &key);
5148
5149                 /* No root, definitely a bad ref, skip */
5150                 if (IS_ERR(root) && PTR_ERR(root) == -ENOENT)
5151                         continue;
5152                 /* Other err, exit */
5153                 if (IS_ERR(root))
5154                         return PTR_ERR(root);
5155
5156                 key.objectid = dback->owner;
5157                 key.type = BTRFS_EXTENT_DATA_KEY;
5158                 key.offset = dback->offset;
5159                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5160                 if (ret) {
5161                         btrfs_release_path(path);
5162                         if (ret < 0)
5163                                 return ret;
5164                         /* Didn't find it, we can carry on */
5165                         ret = 0;
5166                         continue;
5167                 }
5168
5169                 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
5170                                     struct btrfs_file_extent_item);
5171                 bytenr = btrfs_file_extent_disk_bytenr(path->nodes[0], fi);
5172                 bytes = btrfs_file_extent_disk_num_bytes(path->nodes[0], fi);
5173                 btrfs_release_path(path);
5174                 cache = lookup_cache_extent(extent_cache, bytenr, 1);
5175                 if (cache) {
5176                         struct extent_record *tmp;
5177                         tmp = container_of(cache, struct extent_record, cache);
5178
5179                         /*
5180                          * If we found an extent record for the bytenr for this
5181                          * particular backref then we can't add it to our
5182                          * current extent record.  We only want to add backrefs
5183                          * that don't have a corresponding extent item in the
5184                          * extent tree since they likely belong to this record
5185                          * and we need to fix it if it doesn't match bytenrs.
5186                          */
5187                         if  (tmp->found_rec)
5188                                 continue;
5189                 }
5190
5191                 dback->found_ref += 1;
5192                 dback->disk_bytenr = bytenr;
5193                 dback->bytes = bytes;
5194
5195                 /*
5196                  * Set this so the verify backref code knows not to trust the
5197                  * values in this backref.
5198                  */
5199                 back->broken = 1;
5200         }
5201
5202         return 0;
5203 }
5204
5205 /*
5206  * when an incorrect extent item is found, this will delete
5207  * all of the existing entries for it and recreate them
5208  * based on what the tree scan found.
5209  */
5210 static int fixup_extent_refs(struct btrfs_trans_handle *trans,
5211                              struct btrfs_fs_info *info,
5212                              struct cache_tree *extent_cache,
5213                              struct extent_record *rec)
5214 {
5215         int ret;
5216         struct btrfs_path *path;
5217         struct list_head *cur = rec->backrefs.next;
5218         struct cache_extent *cache;
5219         struct extent_backref *back;
5220         int allocated = 0;
5221         u64 flags = 0;
5222
5223         /*
5224          * remember our flags for recreating the extent.
5225          * FIXME, if we have cleared extent tree, we can not
5226          * lookup extent info in extent tree.
5227          */
5228         if (!init_extent_tree) {
5229                 ret = btrfs_lookup_extent_info(NULL, info->extent_root,
5230                                         rec->start, rec->max_size,
5231                                         rec->metadata, NULL, &flags);
5232                 if (ret < 0)
5233                         flags = 0;
5234         } else {
5235                 flags = 0;
5236         }
5237
5238         path = btrfs_alloc_path();
5239         if (!path)
5240                 return -ENOMEM;
5241
5242         if (rec->refs != rec->extent_item_refs && !rec->metadata) {
5243                 /*
5244                  * Sometimes the backrefs themselves are so broken they don't
5245                  * get attached to any meaningful rec, so first go back and
5246                  * check any of our backrefs that we couldn't find and throw
5247                  * them into the list if we find the backref so that
5248                  * verify_backrefs can figure out what to do.
5249                  */
5250                 ret = find_possible_backrefs(trans, info, path, extent_cache,
5251                                              rec);
5252                 if (ret < 0)
5253                         goto out;
5254         }
5255
5256         /* step one, make sure all of the backrefs agree */
5257         ret = verify_backrefs(trans, info, path, rec);
5258         if (ret < 0)
5259                 goto out;
5260
5261         /* step two, delete all the existing records */
5262         ret = delete_extent_records(trans, info->extent_root, path,
5263                                     rec->start, rec->max_size);
5264
5265         if (ret < 0)
5266                 goto out;
5267
5268         /* was this block corrupt?  If so, don't add references to it */
5269         cache = lookup_cache_extent(info->corrupt_blocks,
5270                                     rec->start, rec->max_size);
5271         if (cache) {
5272                 ret = 0;
5273                 goto out;
5274         }
5275
5276         /* step three, recreate all the refs we did find */
5277         while(cur != &rec->backrefs) {
5278                 back = list_entry(cur, struct extent_backref, list);
5279                 cur = cur->next;
5280
5281                 /*
5282                  * if we didn't find any references, don't create a
5283                  * new extent record
5284                  */
5285                 if (!back->found_ref)
5286                         continue;
5287
5288                 ret = record_extent(trans, info, path, rec, back, allocated, flags);
5289                 allocated = 1;
5290
5291                 if (ret)
5292                         goto out;
5293         }
5294 out:
5295         btrfs_free_path(path);
5296         return ret;
5297 }
5298
5299 /* right now we only prune from the extent allocation tree */
5300 static int prune_one_block(struct btrfs_trans_handle *trans,
5301                            struct btrfs_fs_info *info,
5302                            struct btrfs_corrupt_block *corrupt)
5303 {
5304         int ret;
5305         struct btrfs_path path;
5306         struct extent_buffer *eb;
5307         u64 found;
5308         int slot;
5309         int nritems;
5310         int level = corrupt->level + 1;
5311
5312         btrfs_init_path(&path);
5313 again:
5314         /* we want to stop at the parent to our busted block */
5315         path.lowest_level = level;
5316
5317         ret = btrfs_search_slot(trans, info->extent_root,
5318                                 &corrupt->key, &path, -1, 1);
5319
5320         if (ret < 0)
5321                 goto out;
5322
5323         eb = path.nodes[level];
5324         if (!eb) {
5325                 ret = -ENOENT;
5326                 goto out;
5327         }
5328
5329         /*
5330          * hopefully the search gave us the block we want to prune,
5331          * lets try that first
5332          */
5333         slot = path.slots[level];
5334         found =  btrfs_node_blockptr(eb, slot);
5335         if (found == corrupt->cache.start)
5336                 goto del_ptr;
5337
5338         nritems = btrfs_header_nritems(eb);
5339
5340         /* the search failed, lets scan this node and hope we find it */
5341         for (slot = 0; slot < nritems; slot++) {
5342                 found =  btrfs_node_blockptr(eb, slot);
5343                 if (found == corrupt->cache.start)
5344                         goto del_ptr;
5345         }
5346         /*
5347          * we couldn't find the bad block.  TODO, search all the nodes for pointers
5348          * to this block
5349          */
5350         if (eb == info->extent_root->node) {
5351                 ret = -ENOENT;
5352                 goto out;
5353         } else {
5354                 level++;
5355                 btrfs_release_path(&path);
5356                 goto again;
5357         }
5358
5359 del_ptr:
5360         printk("deleting pointer to block %Lu\n", corrupt->cache.start);
5361         ret = btrfs_del_ptr(trans, info->extent_root, &path, level, slot);
5362
5363 out:
5364         btrfs_release_path(&path);
5365         return ret;
5366 }
5367
5368 static int prune_corrupt_blocks(struct btrfs_trans_handle *trans,
5369                                 struct btrfs_fs_info *info)
5370 {
5371         struct cache_extent *cache;
5372         struct btrfs_corrupt_block *corrupt;
5373
5374         cache = search_cache_extent(info->corrupt_blocks, 0);
5375         while (1) {
5376                 if (!cache)
5377                         break;
5378                 corrupt = container_of(cache, struct btrfs_corrupt_block, cache);
5379                 prune_one_block(trans, info, corrupt);
5380                 cache = next_cache_extent(cache);
5381         }
5382         return 0;
5383 }
5384
5385 static void free_corrupt_block(struct cache_extent *cache)
5386 {
5387         struct btrfs_corrupt_block *corrupt;
5388
5389         corrupt = container_of(cache, struct btrfs_corrupt_block, cache);
5390         free(corrupt);
5391 }
5392
5393 FREE_EXTENT_CACHE_BASED_TREE(corrupt_blocks, free_corrupt_block);
5394
5395 static void reset_cached_block_groups(struct btrfs_fs_info *fs_info)
5396 {
5397         struct btrfs_block_group_cache *cache;
5398         u64 start, end;
5399         int ret;
5400
5401         while (1) {
5402                 ret = find_first_extent_bit(&fs_info->free_space_cache, 0,
5403                                             &start, &end, EXTENT_DIRTY);
5404                 if (ret)
5405                         break;
5406                 clear_extent_dirty(&fs_info->free_space_cache, start, end,
5407                                    GFP_NOFS);
5408         }
5409
5410         start = 0;
5411         while (1) {
5412                 cache = btrfs_lookup_first_block_group(fs_info, start);
5413                 if (!cache)
5414                         break;
5415                 if (cache->cached)
5416                         cache->cached = 0;
5417                 start = cache->key.objectid + cache->key.offset;
5418         }
5419 }
5420
5421 static int check_extent_refs(struct btrfs_trans_handle *trans,
5422                              struct btrfs_root *root,
5423                              struct cache_tree *extent_cache)
5424 {
5425         struct extent_record *rec;
5426         struct cache_extent *cache;
5427         int err = 0;
5428         int ret = 0;
5429         int fixed = 0;
5430         int had_dups = 0;
5431
5432         if (repair) {
5433                 /*
5434                  * if we're doing a repair, we have to make sure
5435                  * we don't allocate from the problem extents.
5436                  * In the worst case, this will be all the
5437                  * extents in the FS
5438                  */
5439                 cache = search_cache_extent(extent_cache, 0);
5440                 while(cache) {
5441                         rec = container_of(cache, struct extent_record, cache);
5442                         btrfs_pin_extent(root->fs_info,
5443                                          rec->start, rec->max_size);
5444                         cache = next_cache_extent(cache);
5445                 }
5446
5447                 /* pin down all the corrupted blocks too */
5448                 cache = search_cache_extent(root->fs_info->corrupt_blocks, 0);
5449                 while(cache) {
5450                         btrfs_pin_extent(root->fs_info,
5451                                          cache->start, cache->size);
5452                         cache = next_cache_extent(cache);
5453                 }
5454                 prune_corrupt_blocks(trans, root->fs_info);
5455                 reset_cached_block_groups(root->fs_info);
5456         }
5457
5458         /*
5459          * We need to delete any duplicate entries we find first otherwise we
5460          * could mess up the extent tree when we have backrefs that actually
5461          * belong to a different extent item and not the weird duplicate one.
5462          */
5463         while (repair && !list_empty(&duplicate_extents)) {
5464                 rec = list_entry(duplicate_extents.next, struct extent_record,
5465                                  list);
5466                 list_del_init(&rec->list);
5467
5468                 /* Sometimes we can find a backref before we find an actual
5469                  * extent, so we need to process it a little bit to see if there
5470                  * truly are multiple EXTENT_ITEM_KEY's for the same range, or
5471                  * if this is a backref screwup.  If we need to delete stuff
5472                  * process_duplicates() will return 0, otherwise it will return
5473                  * 1 and we
5474                  */
5475                 if (process_duplicates(root, extent_cache, rec))
5476                         continue;
5477                 ret = delete_duplicate_records(trans, root, rec);
5478                 if (ret < 0)
5479                         return ret;
5480                 /*
5481                  * delete_duplicate_records will return the number of entries
5482                  * deleted, so if it's greater than 0 then we know we actually
5483                  * did something and we need to remove.
5484                  */
5485                 if (ret)
5486                         had_dups = 1;
5487         }
5488
5489         if (had_dups)
5490                 return -EAGAIN;
5491
5492         while(1) {
5493                 fixed = 0;
5494                 cache = search_cache_extent(extent_cache, 0);
5495                 if (!cache)
5496                         break;
5497                 rec = container_of(cache, struct extent_record, cache);
5498                 if (rec->num_duplicates) {
5499                         fprintf(stderr, "extent item %llu has multiple extent "
5500                                 "items\n", (unsigned long long)rec->start);
5501                         err = 1;
5502                 }
5503
5504                 if (rec->refs != rec->extent_item_refs) {
5505                         fprintf(stderr, "ref mismatch on [%llu %llu] ",
5506                                 (unsigned long long)rec->start,
5507                                 (unsigned long long)rec->nr);
5508                         fprintf(stderr, "extent item %llu, found %llu\n",
5509                                 (unsigned long long)rec->extent_item_refs,
5510                                 (unsigned long long)rec->refs);
5511                         if (!fixed && repair) {
5512                                 ret = fixup_extent_refs(trans, root->fs_info,
5513                                                         extent_cache, rec);
5514                                 if (ret)
5515                                         goto repair_abort;
5516                                 fixed = 1;
5517                         }
5518                         err = 1;
5519
5520                 }
5521                 if (all_backpointers_checked(rec, 1)) {
5522                         fprintf(stderr, "backpointer mismatch on [%llu %llu]\n",
5523                                 (unsigned long long)rec->start,
5524                                 (unsigned long long)rec->nr);
5525
5526                         if (!fixed && repair) {
5527                                 ret = fixup_extent_refs(trans, root->fs_info,
5528                                                         extent_cache, rec);
5529                                 if (ret)
5530                                         goto repair_abort;
5531                                 fixed = 1;
5532                         }
5533
5534                         err = 1;
5535                 }
5536                 if (!rec->owner_ref_checked) {
5537                         fprintf(stderr, "owner ref check failed [%llu %llu]\n",
5538                                 (unsigned long long)rec->start,
5539                                 (unsigned long long)rec->nr);
5540                         if (!fixed && repair) {
5541                                 ret = fixup_extent_refs(trans, root->fs_info,
5542                                                         extent_cache, rec);
5543                                 if (ret)
5544                                         goto repair_abort;
5545                                 fixed = 1;
5546                         }
5547                         err = 1;
5548                 }
5549
5550                 remove_cache_extent(extent_cache, cache);
5551                 free_all_extent_backrefs(rec);
5552                 free(rec);
5553         }
5554 repair_abort:
5555         if (repair) {
5556                 if (ret && ret != -EAGAIN) {
5557                         fprintf(stderr, "failed to repair damaged filesystem, aborting\n");
5558                         exit(1);
5559                 } else if (!ret) {
5560                         btrfs_fix_block_accounting(trans, root);
5561                 }
5562                 if (err)
5563                         fprintf(stderr, "repaired damaged extent references\n");
5564                 return ret;
5565         }
5566         return err;
5567 }
5568
5569 u64 calc_stripe_length(u64 type, u64 length, int num_stripes)
5570 {
5571         u64 stripe_size;
5572
5573         if (type & BTRFS_BLOCK_GROUP_RAID0) {
5574                 stripe_size = length;
5575                 stripe_size /= num_stripes;
5576         } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
5577                 stripe_size = length * 2;
5578                 stripe_size /= num_stripes;
5579         } else if (type & BTRFS_BLOCK_GROUP_RAID5) {
5580                 stripe_size = length;
5581                 stripe_size /= (num_stripes - 1);
5582         } else if (type & BTRFS_BLOCK_GROUP_RAID6) {
5583                 stripe_size = length;
5584                 stripe_size /= (num_stripes - 2);
5585         } else {
5586                 stripe_size = length;
5587         }
5588         return stripe_size;
5589 }
5590
5591 static int check_chunk_refs(struct chunk_record *chunk_rec,
5592                             struct block_group_tree *block_group_cache,
5593                             struct device_extent_tree *dev_extent_cache,
5594                             int silent)
5595 {
5596         struct cache_extent *block_group_item;
5597         struct block_group_record *block_group_rec;
5598         struct cache_extent *dev_extent_item;
5599         struct device_extent_record *dev_extent_rec;
5600         u64 devid;
5601         u64 offset;
5602         u64 length;
5603         int i;
5604         int ret = 0;
5605
5606         block_group_item = lookup_cache_extent(&block_group_cache->tree,
5607                                                chunk_rec->offset,
5608                                                chunk_rec->length);
5609         if (block_group_item) {
5610                 block_group_rec = container_of(block_group_item,
5611                                                struct block_group_record,
5612                                                cache);
5613                 if (chunk_rec->length != block_group_rec->offset ||
5614                     chunk_rec->offset != block_group_rec->objectid ||
5615                     chunk_rec->type_flags != block_group_rec->flags) {
5616                         if (!silent)
5617                                 fprintf(stderr,
5618                                         "Chunk[%llu, %u, %llu]: length(%llu), offset(%llu), type(%llu) mismatch with block group[%llu, %u, %llu]: offset(%llu), objectid(%llu), flags(%llu)\n",
5619                                         chunk_rec->objectid,
5620                                         chunk_rec->type,
5621                                         chunk_rec->offset,
5622                                         chunk_rec->length,
5623                                         chunk_rec->offset,
5624                                         chunk_rec->type_flags,
5625                                         block_group_rec->objectid,
5626                                         block_group_rec->type,
5627                                         block_group_rec->offset,
5628                                         block_group_rec->offset,
5629                                         block_group_rec->objectid,
5630                                         block_group_rec->flags);
5631                         ret = -1;
5632                 } else {
5633                         list_del_init(&block_group_rec->list);
5634                         chunk_rec->bg_rec = block_group_rec;
5635                 }
5636         } else {
5637                 if (!silent)
5638                         fprintf(stderr,
5639                                 "Chunk[%llu, %u, %llu]: length(%llu), offset(%llu), type(%llu) is not found in block group\n",
5640                                 chunk_rec->objectid,
5641                                 chunk_rec->type,
5642                                 chunk_rec->offset,
5643                                 chunk_rec->length,
5644                                 chunk_rec->offset,
5645                                 chunk_rec->type_flags);
5646                 ret = -1;
5647         }
5648
5649         length = calc_stripe_length(chunk_rec->type_flags, chunk_rec->length,
5650                                     chunk_rec->num_stripes);
5651         for (i = 0; i < chunk_rec->num_stripes; ++i) {
5652                 devid = chunk_rec->stripes[i].devid;
5653                 offset = chunk_rec->stripes[i].offset;
5654                 dev_extent_item = lookup_cache_extent2(&dev_extent_cache->tree,
5655                                                        devid, offset, length);
5656                 if (dev_extent_item) {
5657                         dev_extent_rec = container_of(dev_extent_item,
5658                                                 struct device_extent_record,
5659                                                 cache);
5660                         if (dev_extent_rec->objectid != devid ||
5661                             dev_extent_rec->offset != offset ||
5662                             dev_extent_rec->chunk_offset != chunk_rec->offset ||
5663                             dev_extent_rec->length != length) {
5664                                 if (!silent)
5665                                         fprintf(stderr,
5666                                                 "Chunk[%llu, %u, %llu] stripe[%llu, %llu] dismatch dev extent[%llu, %llu, %llu]\n",
5667                                                 chunk_rec->objectid,
5668                                                 chunk_rec->type,
5669                                                 chunk_rec->offset,
5670                                                 chunk_rec->stripes[i].devid,
5671                                                 chunk_rec->stripes[i].offset,
5672                                                 dev_extent_rec->objectid,
5673                                                 dev_extent_rec->offset,
5674                                                 dev_extent_rec->length);
5675                                 ret = -1;
5676                         } else {
5677                                 list_move(&dev_extent_rec->chunk_list,
5678                                           &chunk_rec->dextents);
5679                         }
5680                 } else {
5681                         if (!silent)
5682                                 fprintf(stderr,
5683                                         "Chunk[%llu, %u, %llu] stripe[%llu, %llu] is not found in dev extent\n",
5684                                         chunk_rec->objectid,
5685                                         chunk_rec->type,
5686                                         chunk_rec->offset,
5687                                         chunk_rec->stripes[i].devid,
5688                                         chunk_rec->stripes[i].offset);
5689                         ret = -1;
5690                 }
5691         }
5692         return ret;
5693 }
5694
5695 /* check btrfs_chunk -> btrfs_dev_extent / btrfs_block_group_item */
5696 int check_chunks(struct cache_tree *chunk_cache,
5697                  struct block_group_tree *block_group_cache,
5698                  struct device_extent_tree *dev_extent_cache,
5699                  struct list_head *good, struct list_head *bad, int silent)
5700 {
5701         struct cache_extent *chunk_item;
5702         struct chunk_record *chunk_rec;
5703         struct block_group_record *bg_rec;
5704         struct device_extent_record *dext_rec;
5705         int err;
5706         int ret = 0;
5707
5708         chunk_item = first_cache_extent(chunk_cache);
5709         while (chunk_item) {
5710                 chunk_rec = container_of(chunk_item, struct chunk_record,
5711                                          cache);
5712                 err = check_chunk_refs(chunk_rec, block_group_cache,
5713                                        dev_extent_cache, silent);
5714                 if (err) {
5715                         ret = err;
5716                         if (bad)
5717                                 list_add_tail(&chunk_rec->list, bad);
5718                 } else {
5719                         if (good)
5720                                 list_add_tail(&chunk_rec->list, good);
5721                 }
5722
5723                 chunk_item = next_cache_extent(chunk_item);
5724         }
5725
5726         list_for_each_entry(bg_rec, &block_group_cache->block_groups, list) {
5727                 if (!silent)
5728                         fprintf(stderr,
5729                                 "Block group[%llu, %llu] (flags = %llu) didn't find the relative chunk.\n",
5730                                 bg_rec->objectid,
5731                                 bg_rec->offset,
5732                                 bg_rec->flags);
5733                 if (!ret)
5734                         ret = 1;
5735         }
5736
5737         list_for_each_entry(dext_rec, &dev_extent_cache->no_chunk_orphans,
5738                             chunk_list) {
5739                 if (!silent)
5740                         fprintf(stderr,
5741                                 "Device extent[%llu, %llu, %llu] didn't find the relative chunk.\n",
5742                                 dext_rec->objectid,
5743                                 dext_rec->offset,
5744                                 dext_rec->length);
5745                 if (!ret)
5746                         ret = 1;
5747         }
5748         return ret;
5749 }
5750
5751
5752 static int check_device_used(struct device_record *dev_rec,
5753                              struct device_extent_tree *dext_cache)
5754 {
5755         struct cache_extent *cache;
5756         struct device_extent_record *dev_extent_rec;
5757         u64 total_byte = 0;
5758
5759         cache = search_cache_extent2(&dext_cache->tree, dev_rec->devid, 0);
5760         while (cache) {
5761                 dev_extent_rec = container_of(cache,
5762                                               struct device_extent_record,
5763                                               cache);
5764                 if (dev_extent_rec->objectid != dev_rec->devid)
5765                         break;
5766
5767                 list_del(&dev_extent_rec->device_list);
5768                 total_byte += dev_extent_rec->length;
5769                 cache = next_cache_extent(cache);
5770         }
5771
5772         if (total_byte != dev_rec->byte_used) {
5773                 fprintf(stderr,
5774                         "Dev extent's total-byte(%llu) is not equal to byte-used(%llu) in dev[%llu, %u, %llu]\n",
5775                         total_byte, dev_rec->byte_used, dev_rec->objectid,
5776                         dev_rec->type, dev_rec->offset);
5777                 return -1;
5778         } else {
5779                 return 0;
5780         }
5781 }
5782
5783 /* check btrfs_dev_item -> btrfs_dev_extent */
5784 static int check_devices(struct rb_root *dev_cache,
5785                          struct device_extent_tree *dev_extent_cache)
5786 {
5787         struct rb_node *dev_node;
5788         struct device_record *dev_rec;
5789         struct device_extent_record *dext_rec;
5790         int err;
5791         int ret = 0;
5792
5793         dev_node = rb_first(dev_cache);
5794         while (dev_node) {
5795                 dev_rec = container_of(dev_node, struct device_record, node);
5796                 err = check_device_used(dev_rec, dev_extent_cache);
5797                 if (err)
5798                         ret = err;
5799
5800                 dev_node = rb_next(dev_node);
5801         }
5802         list_for_each_entry(dext_rec, &dev_extent_cache->no_device_orphans,
5803                             device_list) {
5804                 fprintf(stderr,
5805                         "Device extent[%llu, %llu, %llu] didn't find its device.\n",
5806                         dext_rec->objectid, dext_rec->offset, dext_rec->length);
5807                 if (!ret)
5808                         ret = 1;
5809         }
5810         return ret;
5811 }
5812
5813 static int check_chunks_and_extents(struct btrfs_root *root)
5814 {
5815         struct rb_root dev_cache;
5816         struct cache_tree chunk_cache;
5817         struct block_group_tree block_group_cache;
5818         struct device_extent_tree dev_extent_cache;
5819         struct cache_tree extent_cache;
5820         struct cache_tree seen;
5821         struct cache_tree pending;
5822         struct cache_tree reada;
5823         struct cache_tree nodes;
5824         struct cache_tree corrupt_blocks;
5825         struct btrfs_path path;
5826         struct btrfs_key key;
5827         struct btrfs_key found_key;
5828         int ret, err = 0;
5829         u64 last = 0;
5830         struct block_info *bits;
5831         int bits_nr;
5832         struct extent_buffer *leaf;
5833         struct btrfs_trans_handle *trans = NULL;
5834         int slot;
5835         struct btrfs_root_item ri;
5836         struct list_head dropping_trees;
5837
5838         dev_cache = RB_ROOT;
5839         cache_tree_init(&chunk_cache);
5840         block_group_tree_init(&block_group_cache);
5841         device_extent_tree_init(&dev_extent_cache);
5842
5843         cache_tree_init(&extent_cache);
5844         cache_tree_init(&seen);
5845         cache_tree_init(&pending);
5846         cache_tree_init(&nodes);
5847         cache_tree_init(&reada);
5848         cache_tree_init(&corrupt_blocks);
5849         INIT_LIST_HEAD(&dropping_trees);
5850
5851         if (repair) {
5852                 trans = btrfs_start_transaction(root, 1);
5853                 if (IS_ERR(trans)) {
5854                         fprintf(stderr, "Error starting transaction\n");
5855                         return PTR_ERR(trans);
5856                 }
5857                 root->fs_info->fsck_extent_cache = &extent_cache;
5858                 root->fs_info->free_extent_hook = free_extent_hook;
5859                 root->fs_info->corrupt_blocks = &corrupt_blocks;
5860         }
5861
5862         bits_nr = 1024;
5863         bits = malloc(bits_nr * sizeof(struct block_info));
5864         if (!bits) {
5865                 perror("malloc");
5866                 exit(1);
5867         }
5868
5869 again:
5870         add_root_to_pending(root->fs_info->tree_root->node,
5871                             &extent_cache, &pending, &seen, &nodes,
5872                             &root->fs_info->tree_root->root_key);
5873
5874         add_root_to_pending(root->fs_info->chunk_root->node,
5875                             &extent_cache, &pending, &seen, &nodes,
5876                             &root->fs_info->chunk_root->root_key);
5877
5878         btrfs_init_path(&path);
5879         key.offset = 0;
5880         key.objectid = 0;
5881         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
5882         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
5883                                         &key, &path, 0, 0);
5884         BUG_ON(ret < 0);
5885         while(1) {
5886                 leaf = path.nodes[0];
5887                 slot = path.slots[0];
5888                 if (slot >= btrfs_header_nritems(path.nodes[0])) {
5889                         ret = btrfs_next_leaf(root, &path);
5890                         if (ret != 0)
5891                                 break;
5892                         leaf = path.nodes[0];
5893                         slot = path.slots[0];
5894                 }
5895                 btrfs_item_key_to_cpu(leaf, &found_key, path.slots[0]);
5896                 if (btrfs_key_type(&found_key) == BTRFS_ROOT_ITEM_KEY) {
5897                         unsigned long offset;
5898                         struct extent_buffer *buf;
5899
5900                         offset = btrfs_item_ptr_offset(leaf, path.slots[0]);
5901                         read_extent_buffer(leaf, &ri, offset, sizeof(ri));
5902                         if (btrfs_disk_key_objectid(&ri.drop_progress) == 0) {
5903                                 buf = read_tree_block(root->fs_info->tree_root,
5904                                                       btrfs_root_bytenr(&ri),
5905                                                       btrfs_level_size(root,
5906                                                       btrfs_root_level(&ri)),
5907                                                       0);
5908                                 if (!buf) {
5909                                         ret = -EIO;
5910                                         goto out;
5911                                 }
5912                                 add_root_to_pending(buf, &extent_cache,
5913                                                     &pending, &seen, &nodes,
5914                                                     &found_key);
5915                                 free_extent_buffer(buf);
5916                         } else {
5917                                 struct dropping_root_item_record *dri_rec;
5918                                 dri_rec = malloc(sizeof(*dri_rec));
5919                                 if (!dri_rec) {
5920                                         perror("malloc");
5921                                         exit(1);
5922                                 }
5923                                 memcpy(&dri_rec->ri, &ri, sizeof(ri));
5924                                 memcpy(&dri_rec->found_key, &found_key,
5925                                        sizeof(found_key));
5926                                 list_add_tail(&dri_rec->list, &dropping_trees);
5927                         }
5928                 }
5929                 path.slots[0]++;
5930         }
5931         btrfs_release_path(&path);
5932         while (1) {
5933                 ret = run_next_block(trans, root, bits, bits_nr, &last,
5934                                      &pending, &seen, &reada, &nodes,
5935                                      &extent_cache, &chunk_cache, &dev_cache,
5936                                      &block_group_cache, &dev_extent_cache,
5937                                      NULL);
5938                 if (ret != 0)
5939                         break;
5940         }
5941
5942         while (!list_empty(&dropping_trees)) {
5943                 struct dropping_root_item_record *rec;
5944                 struct extent_buffer *buf;
5945                 rec = list_entry(dropping_trees.next,
5946                                  struct dropping_root_item_record, list);
5947                 last = 0;
5948                 if (!bits) {
5949                         perror("realloc");
5950                         exit(1);
5951                 }
5952                 buf = read_tree_block(root->fs_info->tree_root,
5953                                       btrfs_root_bytenr(&rec->ri),
5954                                       btrfs_level_size(root,
5955                                       btrfs_root_level(&rec->ri)), 0);
5956                 if (!buf) {
5957                         ret = -EIO;
5958                         goto out;
5959                 }
5960                 add_root_to_pending(buf, &extent_cache, &pending,
5961                                     &seen, &nodes, &rec->found_key);
5962                 while (1) {
5963                         ret = run_next_block(trans, root, bits, bits_nr, &last,
5964                                              &pending, &seen, &reada,
5965                                              &nodes, &extent_cache,
5966                                              &chunk_cache, &dev_cache,
5967                                              &block_group_cache,
5968                                              &dev_extent_cache,
5969                                              &rec->ri);
5970                         if (ret != 0)
5971                                 break;
5972                 }
5973                 free_extent_buffer(buf);
5974                 list_del(&rec->list);
5975                 free(rec);
5976         }
5977
5978         if (ret >= 0)
5979                 ret = check_extent_refs(trans, root, &extent_cache);
5980         if (ret == -EAGAIN) {
5981                 ret = btrfs_commit_transaction(trans, root);
5982                 if (ret)
5983                         goto out;
5984
5985                 trans = btrfs_start_transaction(root, 1);
5986                 if (IS_ERR(trans)) {
5987                         ret = PTR_ERR(trans);
5988                         goto out;
5989                 }
5990
5991                 free_corrupt_blocks_tree(root->fs_info->corrupt_blocks);
5992                 free_extent_cache_tree(&seen);
5993                 free_extent_cache_tree(&pending);
5994                 free_extent_cache_tree(&reada);
5995                 free_extent_cache_tree(&nodes);
5996                 free_extent_record_cache(root->fs_info, &extent_cache);
5997                 goto again;
5998         }
5999
6000         err = check_chunks(&chunk_cache, &block_group_cache,
6001                            &dev_extent_cache, NULL, NULL, 0);
6002         if (err && !ret)
6003                 ret = err;
6004
6005         err = check_devices(&dev_cache, &dev_extent_cache);
6006         if (err && !ret)
6007                 ret = err;
6008
6009         if (trans) {
6010                 err = btrfs_commit_transaction(trans, root);
6011                 if (!ret)
6012                         ret = err;
6013         }
6014 out:
6015         if (repair) {
6016                 free_corrupt_blocks_tree(root->fs_info->corrupt_blocks);
6017                 root->fs_info->fsck_extent_cache = NULL;
6018                 root->fs_info->free_extent_hook = NULL;
6019                 root->fs_info->corrupt_blocks = NULL;
6020         }
6021         free(bits);
6022         free_chunk_cache_tree(&chunk_cache);
6023         free_device_cache_tree(&dev_cache);
6024         free_block_group_tree(&block_group_cache);
6025         free_device_extent_tree(&dev_extent_cache);
6026         free_extent_cache_tree(&seen);
6027         free_extent_cache_tree(&pending);
6028         free_extent_cache_tree(&reada);
6029         free_extent_cache_tree(&nodes);
6030         return ret;
6031 }
6032
6033 static int btrfs_fsck_reinit_root(struct btrfs_trans_handle *trans,
6034                            struct btrfs_root *root, int overwrite)
6035 {
6036         struct extent_buffer *c;
6037         struct extent_buffer *old = root->node;
6038         int level;
6039         int ret;
6040         struct btrfs_disk_key disk_key = {0,0,0};
6041
6042         level = 0;
6043
6044         if (overwrite) {
6045                 c = old;
6046                 extent_buffer_get(c);
6047                 goto init;
6048         }
6049         c = btrfs_alloc_free_block(trans, root,
6050                                    btrfs_level_size(root, 0),
6051                                    root->root_key.objectid,
6052                                    &disk_key, level, 0, 0);
6053         if (IS_ERR(c)) {
6054                 c = old;
6055                 extent_buffer_get(c);
6056                 overwrite = 1;
6057         }
6058 init:
6059         memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
6060         btrfs_set_header_level(c, level);
6061         btrfs_set_header_bytenr(c, c->start);
6062         btrfs_set_header_generation(c, trans->transid);
6063         btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
6064         btrfs_set_header_owner(c, root->root_key.objectid);
6065
6066         write_extent_buffer(c, root->fs_info->fsid,
6067                             btrfs_header_fsid(), BTRFS_FSID_SIZE);
6068
6069         write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
6070                             btrfs_header_chunk_tree_uuid(c),
6071                             BTRFS_UUID_SIZE);
6072
6073         btrfs_mark_buffer_dirty(c);
6074         /*
6075          * this case can happen in the following case:
6076          *
6077          * 1.overwrite previous root.
6078          *
6079          * 2.reinit reloc data root, this is because we skip pin
6080          * down reloc data tree before which means we can allocate
6081          * same block bytenr here.
6082          */
6083         if (old->start == c->start) {
6084                 btrfs_set_root_generation(&root->root_item,
6085                                           trans->transid);
6086                 root->root_item.level = btrfs_header_level(root->node);
6087                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
6088                                         &root->root_key, &root->root_item);
6089                 if (ret) {
6090                         free_extent_buffer(c);
6091                         return ret;
6092                 }
6093         }
6094         free_extent_buffer(old);
6095         root->node = c;
6096         add_root_to_dirty_list(root);
6097         return 0;
6098 }
6099
6100 static int pin_down_tree_blocks(struct btrfs_fs_info *fs_info,
6101                                 struct extent_buffer *eb, int tree_root)
6102 {
6103         struct extent_buffer *tmp;
6104         struct btrfs_root_item *ri;
6105         struct btrfs_key key;
6106         u64 bytenr;
6107         u32 leafsize;
6108         int level = btrfs_header_level(eb);
6109         int nritems;
6110         int ret;
6111         int i;
6112
6113         btrfs_pin_extent(fs_info, eb->start, eb->len);
6114
6115         leafsize = btrfs_super_leafsize(fs_info->super_copy);
6116         nritems = btrfs_header_nritems(eb);
6117         for (i = 0; i < nritems; i++) {
6118                 if (level == 0) {
6119                         btrfs_item_key_to_cpu(eb, &key, i);
6120                         if (key.type != BTRFS_ROOT_ITEM_KEY)
6121                                 continue;
6122                         /* Skip the extent root and reloc roots */
6123                         if (key.objectid == BTRFS_EXTENT_TREE_OBJECTID ||
6124                             key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
6125                             key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
6126                                 continue;
6127                         ri = btrfs_item_ptr(eb, i, struct btrfs_root_item);
6128                         bytenr = btrfs_disk_root_bytenr(eb, ri);
6129
6130                         /*
6131                          * If at any point we start needing the real root we
6132                          * will have to build a stump root for the root we are
6133                          * in, but for now this doesn't actually use the root so
6134                          * just pass in extent_root.
6135                          */
6136                         tmp = read_tree_block(fs_info->extent_root, bytenr,
6137                                               leafsize, 0);
6138                         if (!tmp) {
6139                                 fprintf(stderr, "Error reading root block\n");
6140                                 return -EIO;
6141                         }
6142                         ret = pin_down_tree_blocks(fs_info, tmp, 0);
6143                         free_extent_buffer(tmp);
6144                         if (ret)
6145                                 return ret;
6146                 } else {
6147                         bytenr = btrfs_node_blockptr(eb, i);
6148
6149                         /* If we aren't the tree root don't read the block */
6150                         if (level == 1 && !tree_root) {
6151                                 btrfs_pin_extent(fs_info, bytenr, leafsize);
6152                                 continue;
6153                         }
6154
6155                         tmp = read_tree_block(fs_info->extent_root, bytenr,
6156                                               leafsize, 0);
6157                         if (!tmp) {
6158                                 fprintf(stderr, "Error reading tree block\n");
6159                                 return -EIO;
6160                         }
6161                         ret = pin_down_tree_blocks(fs_info, tmp, tree_root);
6162                         free_extent_buffer(tmp);
6163                         if (ret)
6164                                 return ret;
6165                 }
6166         }
6167
6168         return 0;
6169 }
6170
6171 static int pin_metadata_blocks(struct btrfs_fs_info *fs_info)
6172 {
6173         int ret;
6174
6175         ret = pin_down_tree_blocks(fs_info, fs_info->chunk_root->node, 0);
6176         if (ret)
6177                 return ret;
6178
6179         return pin_down_tree_blocks(fs_info, fs_info->tree_root->node, 1);
6180 }
6181
6182 static int reset_block_groups(struct btrfs_fs_info *fs_info)
6183 {
6184         struct btrfs_block_group_cache *cache;
6185         struct btrfs_path *path;
6186         struct extent_buffer *leaf;
6187         struct btrfs_chunk *chunk;
6188         struct btrfs_key key;
6189         int ret;
6190         u64 start;
6191
6192         path = btrfs_alloc_path();
6193         if (!path)
6194                 return -ENOMEM;
6195
6196         key.objectid = 0;
6197         key.type = BTRFS_CHUNK_ITEM_KEY;
6198         key.offset = 0;
6199
6200         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
6201         if (ret < 0) {
6202                 btrfs_free_path(path);
6203                 return ret;
6204         }
6205
6206         /*
6207          * We do this in case the block groups were screwed up and had alloc
6208          * bits that aren't actually set on the chunks.  This happens with
6209          * restored images every time and could happen in real life I guess.
6210          */
6211         fs_info->avail_data_alloc_bits = 0;
6212         fs_info->avail_metadata_alloc_bits = 0;
6213         fs_info->avail_system_alloc_bits = 0;
6214
6215         /* First we need to create the in-memory block groups */
6216         while (1) {
6217                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
6218                         ret = btrfs_next_leaf(fs_info->chunk_root, path);
6219                         if (ret < 0) {
6220                                 btrfs_free_path(path);
6221                                 return ret;
6222                         }
6223                         if (ret) {
6224                                 ret = 0;
6225                                 break;
6226                         }
6227                 }
6228                 leaf = path->nodes[0];
6229                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
6230                 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
6231                         path->slots[0]++;
6232                         continue;
6233                 }
6234
6235                 chunk = btrfs_item_ptr(leaf, path->slots[0],
6236                                        struct btrfs_chunk);
6237                 btrfs_add_block_group(fs_info, 0,
6238                                       btrfs_chunk_type(leaf, chunk),
6239                                       key.objectid, key.offset,
6240                                       btrfs_chunk_length(leaf, chunk));
6241                 set_extent_dirty(&fs_info->free_space_cache, key.offset,
6242                                  key.offset + btrfs_chunk_length(leaf, chunk),
6243                                  GFP_NOFS);
6244                 path->slots[0]++;
6245         }
6246         start = 0;
6247         while (1) {
6248                 cache = btrfs_lookup_first_block_group(fs_info, start);
6249                 if (!cache)
6250                         break;
6251                 cache->cached = 1;
6252                 start = cache->key.objectid + cache->key.offset;
6253         }
6254
6255         btrfs_free_path(path);
6256         return 0;
6257 }
6258
6259 static int reset_balance(struct btrfs_trans_handle *trans,
6260                          struct btrfs_fs_info *fs_info)
6261 {
6262         struct btrfs_root *root = fs_info->tree_root;
6263         struct btrfs_path *path;
6264         struct extent_buffer *leaf;
6265         struct btrfs_key key;
6266         int del_slot, del_nr = 0;
6267         int ret;
6268         int found = 0;
6269
6270         path = btrfs_alloc_path();
6271         if (!path)
6272                 return -ENOMEM;
6273
6274         key.objectid = BTRFS_BALANCE_OBJECTID;
6275         key.type = BTRFS_BALANCE_ITEM_KEY;
6276         key.offset = 0;
6277
6278         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
6279         if (ret) {
6280                 if (ret > 0)
6281                         ret = 0;
6282                 if (!ret)
6283                         goto reinit_data_reloc;
6284                 else
6285                         goto out;
6286         }
6287
6288         ret = btrfs_del_item(trans, root, path);
6289         if (ret)
6290                 goto out;
6291         btrfs_release_path(path);
6292
6293         key.objectid = BTRFS_TREE_RELOC_OBJECTID;
6294         key.type = BTRFS_ROOT_ITEM_KEY;
6295         key.offset = 0;
6296
6297         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
6298         if (ret < 0)
6299                 goto out;
6300         while (1) {
6301                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
6302                         if (!found)
6303                                 break;
6304
6305                         if (del_nr) {
6306                                 ret = btrfs_del_items(trans, root, path,
6307                                                       del_slot, del_nr);
6308                                 del_nr = 0;
6309                                 if (ret)
6310                                         goto out;
6311                         }
6312                         key.offset++;
6313                         btrfs_release_path(path);
6314
6315                         found = 0;
6316                         ret = btrfs_search_slot(trans, root, &key, path,
6317                                                 -1, 1);
6318                         if (ret < 0)
6319                                 goto out;
6320                         continue;
6321                 }
6322                 found = 1;
6323                 leaf = path->nodes[0];
6324                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
6325                 if (key.objectid > BTRFS_TREE_RELOC_OBJECTID)
6326                         break;
6327                 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6328                         path->slots[0]++;
6329                         continue;
6330                 }
6331                 if (!del_nr) {
6332                         del_slot = path->slots[0];
6333                         del_nr = 1;
6334                 } else {
6335                         del_nr++;
6336                 }
6337                 path->slots[0]++;
6338         }
6339
6340         if (del_nr) {
6341                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
6342                 if (ret)
6343                         goto out;
6344         }
6345         btrfs_release_path(path);
6346
6347 reinit_data_reloc:
6348         key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
6349         key.type = BTRFS_ROOT_ITEM_KEY;
6350         key.offset = (u64)-1;
6351         root = btrfs_read_fs_root(fs_info, &key);
6352         if (IS_ERR(root)) {
6353                 fprintf(stderr, "Error reading data reloc tree\n");
6354                 return PTR_ERR(root);
6355         }
6356         root->track_dirty = 1;
6357         if (root->last_trans != trans->transid) {
6358                 root->last_trans = trans->transid;
6359                 root->commit_root = root->node;
6360                 extent_buffer_get(root->node);
6361         }
6362         ret = btrfs_fsck_reinit_root(trans, root, 0);
6363         if (ret)
6364                 goto out;
6365         ret = btrfs_make_root_dir(trans, root, BTRFS_FIRST_FREE_OBJECTID);
6366 out:
6367         btrfs_free_path(path);
6368         return ret;
6369 }
6370
6371 static int reinit_extent_tree(struct btrfs_trans_handle *trans,
6372                               struct btrfs_fs_info *fs_info)
6373 {
6374         u64 start = 0;
6375         int ret;
6376
6377         /*
6378          * The only reason we don't do this is because right now we're just
6379          * walking the trees we find and pinning down their bytes, we don't look
6380          * at any of the leaves.  In order to do mixed groups we'd have to check
6381          * the leaves of any fs roots and pin down the bytes for any file
6382          * extents we find.  Not hard but why do it if we don't have to?
6383          */
6384         if (btrfs_fs_incompat(fs_info, BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)) {
6385                 fprintf(stderr, "We don't support re-initing the extent tree "
6386                         "for mixed block groups yet, please notify a btrfs "
6387                         "developer you want to do this so they can add this "
6388                         "functionality.\n");
6389                 return -EINVAL;
6390         }
6391
6392         /*
6393          * first we need to walk all of the trees except the extent tree and pin
6394          * down the bytes that are in use so we don't overwrite any existing
6395          * metadata.
6396          */
6397         ret = pin_metadata_blocks(fs_info);
6398         if (ret) {
6399                 fprintf(stderr, "error pinning down used bytes\n");
6400                 return ret;
6401         }
6402
6403         /*
6404          * Need to drop all the block groups since we're going to recreate all
6405          * of them again.
6406          */
6407         btrfs_free_block_groups(fs_info);
6408         ret = reset_block_groups(fs_info);
6409         if (ret) {
6410                 fprintf(stderr, "error resetting the block groups\n");
6411                 return ret;
6412         }
6413
6414         /* Ok we can allocate now, reinit the extent root */
6415         ret = btrfs_fsck_reinit_root(trans, fs_info->extent_root, 0);
6416         if (ret) {
6417                 fprintf(stderr, "extent root initialization failed\n");
6418                 /*
6419                  * When the transaction code is updated we should end the
6420                  * transaction, but for now progs only knows about commit so
6421                  * just return an error.
6422                  */
6423                 return ret;
6424         }
6425
6426         /*
6427          * Now we have all the in-memory block groups setup so we can make
6428          * allocations properly, and the metadata we care about is safe since we
6429          * pinned all of it above.
6430          */
6431         while (1) {
6432                 struct btrfs_block_group_cache *cache;
6433
6434                 cache = btrfs_lookup_first_block_group(fs_info, start);
6435                 if (!cache)
6436                         break;
6437                 start = cache->key.objectid + cache->key.offset;
6438                 ret = btrfs_insert_item(trans, fs_info->extent_root,
6439                                         &cache->key, &cache->item,
6440                                         sizeof(cache->item));
6441                 if (ret) {
6442                         fprintf(stderr, "Error adding block group\n");
6443                         return ret;
6444                 }
6445                 btrfs_extent_post_op(trans, fs_info->extent_root);
6446         }
6447
6448         ret = reset_balance(trans, fs_info);
6449         if (ret)
6450                 fprintf(stderr, "error reseting the pending balance\n");
6451
6452         return ret;
6453 }
6454
6455 static int recow_extent_buffer(struct btrfs_root *root, struct extent_buffer *eb)
6456 {
6457         struct btrfs_path *path;
6458         struct btrfs_trans_handle *trans;
6459         struct btrfs_key key;
6460         int ret;
6461
6462         printf("Recowing metadata block %llu\n", eb->start);
6463         key.objectid = btrfs_header_owner(eb);
6464         key.type = BTRFS_ROOT_ITEM_KEY;
6465         key.offset = (u64)-1;
6466
6467         root = btrfs_read_fs_root(root->fs_info, &key);
6468         if (IS_ERR(root)) {
6469                 fprintf(stderr, "Couldn't find owner root %llu\n",
6470                         key.objectid);
6471                 return PTR_ERR(root);
6472         }
6473
6474         path = btrfs_alloc_path();
6475         if (!path)
6476                 return -ENOMEM;
6477
6478         trans = btrfs_start_transaction(root, 1);
6479         if (IS_ERR(trans)) {
6480                 btrfs_free_path(path);
6481                 return PTR_ERR(trans);
6482         }
6483
6484         path->lowest_level = btrfs_header_level(eb);
6485         if (path->lowest_level)
6486                 btrfs_node_key_to_cpu(eb, &key, 0);
6487         else
6488                 btrfs_item_key_to_cpu(eb, &key, 0);
6489
6490         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
6491         btrfs_commit_transaction(trans, root);
6492         btrfs_free_path(path);
6493         return ret;
6494 }
6495
6496 static int delete_bad_item(struct btrfs_root *root, struct bad_item *bad)
6497 {
6498         struct btrfs_path *path;
6499         struct btrfs_trans_handle *trans;
6500         struct btrfs_key key;
6501         int ret;
6502
6503         printf("Deleting bad item [%llu,%u,%llu]\n", bad->key.objectid,
6504                bad->key.type, bad->key.offset);
6505         key.objectid = bad->root_id;
6506         key.type = BTRFS_ROOT_ITEM_KEY;
6507         key.offset = (u64)-1;
6508
6509         root = btrfs_read_fs_root(root->fs_info, &key);
6510         if (IS_ERR(root)) {
6511                 fprintf(stderr, "Couldn't find owner root %llu\n",
6512                         key.objectid);
6513                 return PTR_ERR(root);
6514         }
6515
6516         path = btrfs_alloc_path();
6517         if (!path)
6518                 return -ENOMEM;
6519
6520         trans = btrfs_start_transaction(root, 1);
6521         if (IS_ERR(trans)) {
6522                 btrfs_free_path(path);
6523                 return PTR_ERR(trans);
6524         }
6525
6526         ret = btrfs_search_slot(trans, root, &bad->key, path, -1, 1);
6527         if (ret) {
6528                 if (ret > 0)
6529                         ret = 0;
6530                 goto out;
6531         }
6532         ret = btrfs_del_item(trans, root, path);
6533 out:
6534         btrfs_commit_transaction(trans, root);
6535         btrfs_free_path(path);
6536         return ret;
6537 }
6538
6539 static struct option long_options[] = {
6540         { "super", 1, NULL, 's' },
6541         { "repair", 0, NULL, 0 },
6542         { "init-csum-tree", 0, NULL, 0 },
6543         { "init-extent-tree", 0, NULL, 0 },
6544         { "check-data-csum", 0, NULL, 0 },
6545         { "backup", 0, NULL, 0 },
6546         { "qgroup-report", 0, NULL, 'Q' },
6547         { NULL, 0, NULL, 0}
6548 };
6549
6550 const char * const cmd_check_usage[] = {
6551         "btrfs check [options] <device>",
6552         "Check an unmounted btrfs filesystem.",
6553         "",
6554         "-s|--super <superblock>     use this superblock copy",
6555         "-b|--backup                 use the backup root copy",
6556         "--repair                    try to repair the filesystem",
6557         "--init-csum-tree            create a new CRC tree",
6558         "--init-extent-tree          create a new extent tree",
6559         "--check-data-csum           verify checkums of data blocks",
6560         "--qgroup-report             print a report on qgroup consistency",
6561         NULL
6562 };
6563
6564 int cmd_check(int argc, char **argv)
6565 {
6566         struct cache_tree root_cache;
6567         struct btrfs_root *root;
6568         struct btrfs_fs_info *info;
6569         u64 bytenr = 0;
6570         char uuidbuf[BTRFS_UUID_UNPARSED_SIZE];
6571         int ret;
6572         u64 num;
6573         int option_index = 0;
6574         int init_csum_tree = 0;
6575         int qgroup_report = 0;
6576         enum btrfs_open_ctree_flags ctree_flags =
6577                 OPEN_CTREE_PARTIAL | OPEN_CTREE_EXCLUSIVE;
6578
6579         while(1) {
6580                 int c;
6581                 c = getopt_long(argc, argv, "as:b", long_options,
6582                                 &option_index);
6583                 if (c < 0)
6584                         break;
6585                 switch(c) {
6586                         case 'a': /* ignored */ break;
6587                         case 'b':
6588                                 ctree_flags |= OPEN_CTREE_BACKUP_ROOT;
6589                                 break;
6590                         case 's':
6591                                 num = arg_strtou64(optarg);
6592                                 if (num >= BTRFS_SUPER_MIRROR_MAX) {
6593                                         fprintf(stderr,
6594                                                 "ERROR: super mirror should be less than: %d\n",
6595                                                 BTRFS_SUPER_MIRROR_MAX);
6596                                         exit(1);
6597                                 }
6598                                 bytenr = btrfs_sb_offset(((int)num));
6599                                 printf("using SB copy %llu, bytenr %llu\n", num,
6600                                        (unsigned long long)bytenr);
6601                                 break;
6602                         case 'Q':
6603                                 qgroup_report = 1;
6604                                 break;
6605                         case '?':
6606                         case 'h':
6607                                 usage(cmd_check_usage);
6608                 }
6609                 if (option_index == 1) {
6610                         printf("enabling repair mode\n");
6611                         repair = 1;
6612                         ctree_flags |= OPEN_CTREE_WRITES;
6613                 } else if (option_index == 2) {
6614                         printf("Creating a new CRC tree\n");
6615                         init_csum_tree = 1;
6616                         repair = 1;
6617                         ctree_flags |= OPEN_CTREE_WRITES;
6618                 } else if (option_index == 3) {
6619                         init_extent_tree = 1;
6620                         ctree_flags |= (OPEN_CTREE_WRITES |
6621                                         OPEN_CTREE_NO_BLOCK_GROUPS);
6622                         repair = 1;
6623                 } else if (option_index == 4) {
6624                         check_data_csum = 1;
6625                 }
6626         }
6627         argc = argc - optind;
6628
6629         if (argc != 1)
6630                 usage(cmd_check_usage);
6631
6632         radix_tree_init();
6633         cache_tree_init(&root_cache);
6634
6635         if((ret = check_mounted(argv[optind])) < 0) {
6636                 fprintf(stderr, "Could not check mount status: %s\n", strerror(-ret));
6637                 goto err_out;
6638         } else if(ret) {
6639                 fprintf(stderr, "%s is currently mounted. Aborting.\n", argv[optind]);
6640                 ret = -EBUSY;
6641                 goto err_out;
6642         }
6643
6644         info = open_ctree_fs_info(argv[optind], bytenr, 0, ctree_flags);
6645         if (!info) {
6646                 fprintf(stderr, "Couldn't open file system\n");
6647                 ret = -EIO;
6648                 goto err_out;
6649         }
6650
6651         root = info->fs_root;
6652         uuid_unparse(info->super_copy->fsid, uuidbuf);
6653         if (qgroup_report) {
6654                 printf("Print quota groups for %s\nUUID: %s\n", argv[optind],
6655                        uuidbuf);
6656                 ret = qgroup_verify_all(info);
6657                 if (ret == 0)
6658                         print_qgroup_report(1);
6659                 goto close_out;
6660         }
6661         printf("Checking filesystem on %s\nUUID: %s\n", argv[optind], uuidbuf);
6662
6663         if (!extent_buffer_uptodate(info->tree_root->node) ||
6664             !extent_buffer_uptodate(info->dev_root->node) ||
6665             !extent_buffer_uptodate(info->chunk_root->node)) {
6666                 fprintf(stderr, "Critical roots corrupted, unable to fsck the FS\n");
6667                 ret = -EIO;
6668                 goto close_out;
6669         }
6670
6671         if (init_extent_tree || init_csum_tree) {
6672                 struct btrfs_trans_handle *trans;
6673
6674                 trans = btrfs_start_transaction(info->extent_root, 0);
6675                 if (IS_ERR(trans)) {
6676                         fprintf(stderr, "Error starting transaction\n");
6677                         ret = PTR_ERR(trans);
6678                         goto close_out;
6679                 }
6680
6681                 if (init_extent_tree) {
6682                         printf("Creating a new extent tree\n");
6683                         ret = reinit_extent_tree(trans, info);
6684                         if (ret)
6685                                 goto close_out;
6686                 }
6687
6688                 if (init_csum_tree) {
6689                         fprintf(stderr, "Reinit crc root\n");
6690                         ret = btrfs_fsck_reinit_root(trans, info->csum_root, 0);
6691                         if (ret) {
6692                                 fprintf(stderr, "crc root initialization failed\n");
6693                                 ret = -EIO;
6694                                 goto close_out;
6695                         }
6696                 }
6697                 /*
6698                  * Ok now we commit and run the normal fsck, which will add
6699                  * extent entries for all of the items it finds.
6700                  */
6701                 ret = btrfs_commit_transaction(trans, info->extent_root);
6702                 if (ret)
6703                         goto close_out;
6704         }
6705         if (!extent_buffer_uptodate(info->extent_root->node)) {
6706                 fprintf(stderr, "Critical roots corrupted, unable to fsck the FS\n");
6707                 ret = -EIO;
6708                 goto close_out;
6709         }
6710
6711         fprintf(stderr, "checking extents\n");
6712         ret = check_chunks_and_extents(root);
6713         if (ret)
6714                 fprintf(stderr, "Errors found in extent allocation tree or chunk allocation\n");
6715
6716         fprintf(stderr, "checking free space cache\n");
6717         ret = check_space_cache(root);
6718         if (ret)
6719                 goto out;
6720
6721         /*
6722          * We used to have to have these hole extents in between our real
6723          * extents so if we don't have this flag set we need to make sure there
6724          * are no gaps in the file extents for inodes, otherwise we can just
6725          * ignore it when this happens.
6726          */
6727         no_holes = btrfs_fs_incompat(root->fs_info,
6728                                      BTRFS_FEATURE_INCOMPAT_NO_HOLES);
6729         fprintf(stderr, "checking fs roots\n");
6730         ret = check_fs_roots(root, &root_cache);
6731         if (ret)
6732                 goto out;
6733
6734         fprintf(stderr, "checking csums\n");
6735         ret = check_csums(root);
6736         if (ret)
6737                 goto out;
6738
6739         fprintf(stderr, "checking root refs\n");
6740         ret = check_root_refs(root, &root_cache);
6741         if (ret)
6742                 goto out;
6743
6744         while (repair && !list_empty(&root->fs_info->recow_ebs)) {
6745                 struct extent_buffer *eb;
6746
6747                 eb = list_first_entry(&root->fs_info->recow_ebs,
6748                                       struct extent_buffer, recow);
6749                 ret = recow_extent_buffer(root, eb);
6750                 if (ret)
6751                         break;
6752         }
6753
6754         while (!list_empty(&delete_items)) {
6755                 struct bad_item *bad;
6756
6757                 bad = list_first_entry(&delete_items, struct bad_item, list);
6758                 list_del_init(&bad->list);
6759                 if (repair)
6760                         ret = delete_bad_item(root, bad);
6761                 free(bad);
6762         }
6763
6764         if (info->quota_enabled) {
6765                 int err;
6766                 fprintf(stderr, "checking quota groups\n");
6767                 err = qgroup_verify_all(info);
6768                 if (err)
6769                         goto out;
6770         }
6771
6772         if (!list_empty(&root->fs_info->recow_ebs)) {
6773                 fprintf(stderr, "Transid errors in file system\n");
6774                 ret = 1;
6775         }
6776 out:
6777         print_qgroup_report(0);
6778         if (found_old_backref) { /*
6779                  * there was a disk format change when mixed
6780                  * backref was in testing tree. The old format
6781                  * existed about one week.
6782                  */
6783                 printf("\n * Found old mixed backref format. "
6784                        "The old format is not supported! *"
6785                        "\n * Please mount the FS in readonly mode, "
6786                        "backup data and re-format the FS. *\n\n");
6787                 ret = 1;
6788         }
6789         printf("found %llu bytes used err is %d\n",
6790                (unsigned long long)bytes_used, ret);
6791         printf("total csum bytes: %llu\n",(unsigned long long)total_csum_bytes);
6792         printf("total tree bytes: %llu\n",
6793                (unsigned long long)total_btree_bytes);
6794         printf("total fs tree bytes: %llu\n",
6795                (unsigned long long)total_fs_tree_bytes);
6796         printf("total extent tree bytes: %llu\n",
6797                (unsigned long long)total_extent_tree_bytes);
6798         printf("btree space waste bytes: %llu\n",
6799                (unsigned long long)btree_space_waste);
6800         printf("file data blocks allocated: %llu\n referenced %llu\n",
6801                 (unsigned long long)data_bytes_allocated,
6802                 (unsigned long long)data_bytes_referenced);
6803         printf("%s\n", BTRFS_BUILD_VERSION);
6804
6805         free_root_recs_tree(&root_cache);
6806 close_out:
6807         close_ctree(root);
6808 err_out:
6809         return ret;
6810 }