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
  42 static u64 bytes_used = 0;
  43 static u64 total_csum_bytes = 0;
  44 static u64 total_btree_bytes = 0;
  45 static u64 total_fs_tree_bytes = 0;
  46 static u64 total_extent_tree_bytes = 0;
  47 static u64 btree_space_waste = 0;
  48 static u64 data_bytes_allocated = 0;
  49 static u64 data_bytes_referenced = 0;
  50 static int found_old_backref = 0;
  51 static LIST_HEAD(duplicate_extents);
  52
  53 struct extent_backref {
  54         struct list_head list;
  55         unsigned int is_data:1;
  56         unsigned int found_extent_tree:1;
  57         unsigned int full_backref:1;
  58         unsigned int found_ref:1;
  59 };
  60
  61 struct data_backref {
  62         struct extent_backref node;
  63         union {
  64                 u64 parent;
  65                 u64 root;
  66         };
  67         u64 owner;
  68         u64 offset;
  69         u64 disk_bytenr;
  70         u64 bytes;
  71         u64 ram_bytes;
  72         u32 num_refs;
  73         u32 found_ref;
  74 };
  75
  76 struct tree_backref {
  77         struct extent_backref node;
  78         union {
  79                 u64 parent;
  80                 u64 root;
  81         };
  82 };
  83
  84 struct extent_record {
  85         struct list_head backrefs;
  86         struct list_head dups;
  87         struct list_head list;
  88         struct cache_extent cache;
  89         struct btrfs_disk_key parent_key;
  90         unsigned int found_rec;
  91         u64 start;
  92         u64 max_size;
  93         u64 nr;
  94         u64 refs;
  95         u64 extent_item_refs;
  96         u64 generation;
  97         u64 info_objectid;
  98         u64 num_duplicates;
  99         u8 info_level;
 100         unsigned int content_checked:1;
 101         unsigned int owner_ref_checked:1;
 102         unsigned int is_root:1;
 103         unsigned int metadata:1;
 104 };
 105
 106 struct inode_backref {
 107         struct list_head list;
 108         unsigned int found_dir_item:1;
 109         unsigned int found_dir_index:1;
 110         unsigned int found_inode_ref:1;
 111         unsigned int filetype:8;
 112         int errors;
 113         unsigned int ref_type;
 114         u64 dir;
 115         u64 index;
 116         u16 namelen;
 117         char name[0];
 118 };
 119
 120 #define REF_ERR_NO_DIR_ITEM             (1 << 0)
 121 #define REF_ERR_NO_DIR_INDEX            (1 << 1)
 122 #define REF_ERR_NO_INODE_REF            (1 << 2)
 123 #define REF_ERR_DUP_DIR_ITEM            (1 << 3)
 124 #define REF_ERR_DUP_DIR_INDEX           (1 << 4)
 125 #define REF_ERR_DUP_INODE_REF           (1 << 5)
 126 #define REF_ERR_INDEX_UNMATCH           (1 << 6)
 127 #define REF_ERR_FILETYPE_UNMATCH        (1 << 7)
 128 #define REF_ERR_NAME_TOO_LONG           (1 << 8) // 100
 129 #define REF_ERR_NO_ROOT_REF             (1 << 9)
 130 #define REF_ERR_NO_ROOT_BACKREF         (1 << 10)
 131 #define REF_ERR_DUP_ROOT_REF            (1 << 11)
 132 #define REF_ERR_DUP_ROOT_BACKREF        (1 << 12)
 133
 134 struct inode_record {
 135         struct list_head backrefs;
 136         unsigned int checked:1;
 137         unsigned int merging:1;
 138         unsigned int found_inode_item:1;
 139         unsigned int found_dir_item:1;
 140         unsigned int found_file_extent:1;
 141         unsigned int found_csum_item:1;
 142         unsigned int some_csum_missing:1;
 143         unsigned int nodatasum:1;
 144         int errors;
 145
 146         u64 ino;
 147         u32 nlink;
 148         u32 imode;
 149         u64 isize;
 150         u64 nbytes;
 151
 152         u32 found_link;
 153         u64 found_size;
 154         u64 extent_start;
 155         u64 extent_end;
 156         u64 first_extent_gap;
 157
 158         u32 refs;
 159 };
 160
 161 #define I_ERR_NO_INODE_ITEM             (1 << 0)
 162 #define I_ERR_NO_ORPHAN_ITEM            (1 << 1)
 163 #define I_ERR_DUP_INODE_ITEM            (1 << 2)
 164 #define I_ERR_DUP_DIR_INDEX             (1 << 3)
 165 #define I_ERR_ODD_DIR_ITEM              (1 << 4)
 166 #define I_ERR_ODD_FILE_EXTENT           (1 << 5)
 167 #define I_ERR_BAD_FILE_EXTENT           (1 << 6)
 168 #define I_ERR_FILE_EXTENT_OVERLAP       (1 << 7)
 169 #define I_ERR_FILE_EXTENT_DISCOUNT      (1 << 8) // 100
 170 #define I_ERR_DIR_ISIZE_WRONG           (1 << 9)
 171 #define I_ERR_FILE_NBYTES_WRONG         (1 << 10) // 400
 172 #define I_ERR_ODD_CSUM_ITEM             (1 << 11)
 173 #define I_ERR_SOME_CSUM_MISSING         (1 << 12)
 174 #define I_ERR_LINK_COUNT_WRONG          (1 << 13)
 175
 176 struct root_backref {
 177         struct list_head list;
 178         unsigned int found_dir_item:1;
 179         unsigned int found_dir_index:1;
 180         unsigned int found_back_ref:1;
 181         unsigned int found_forward_ref:1;
 182         unsigned int reachable:1;
 183         int errors;
 184         u64 ref_root;
 185         u64 dir;
 186         u64 index;
 187         u16 namelen;
 188         char name[0];
 189 };
 190
 191 struct root_record {
 192         struct list_head backrefs;
 193         struct cache_extent cache;
 194         unsigned int found_root_item:1;
 195         u64 objectid;
 196         u32 found_ref;
 197 };
 198
 199 struct ptr_node {
 200         struct cache_extent cache;
 201         void *data;
 202 };
 203
 204 struct shared_node {
 205         struct cache_extent cache;
 206         struct cache_tree root_cache;
 207         struct cache_tree inode_cache;
 208         struct inode_record *current;
 209         u32 refs;
 210 };
 211
 212 struct block_info {
 213         u64 start;
 214         u32 size;
 215 };
 216
 217 struct walk_control {
 218         struct cache_tree shared;
 219         struct shared_node *nodes[BTRFS_MAX_LEVEL];
 220         int active_node;
 221         int root_level;
 222 };
 223
 224 static u8 imode_to_type(u32 imode)
 225 {
 226 #define S_SHIFT 12
 227         static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
 228                 [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
 229                 [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
 230                 [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
 231                 [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
 232                 [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
 233                 [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
 234                 [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
 235         };
 236
 237         return btrfs_type_by_mode[(imode & S_IFMT) >> S_SHIFT];
 238 #undef S_SHIFT
 239 }
 240
 241 static int device_record_compare(struct rb_node *node1, struct rb_node *node2)
 242 {
 243         struct device_record *rec1;
 244         struct device_record *rec2;
 245
 246         rec1 = rb_entry(node1, struct device_record, node);
 247         rec2 = rb_entry(node2, struct device_record, node);
 248         if (rec1->devid > rec2->devid)
 249                 return -1;
 250         else if (rec1->devid < rec2->devid)
 251                 return 1;
 252         else
 253                 return 0;
 254 }
 255
 256 static struct inode_record *clone_inode_rec(struct inode_record *orig_rec)
 257 {
 258         struct inode_record *rec;
 259         struct inode_backref *backref;
 260         struct inode_backref *orig;
 261         size_t size;
 262
 263         rec = malloc(sizeof(*rec));
 264         memcpy(rec, orig_rec, sizeof(*rec));
 265         rec->refs = 1;
 266         INIT_LIST_HEAD(&rec->backrefs);
 267
 268         list_for_each_entry(orig, &orig_rec->backrefs, list) {
 269                 size = sizeof(*orig) + orig->namelen + 1;
 270                 backref = malloc(size);
 271                 memcpy(backref, orig, size);
 272                 list_add_tail(&backref->list, &rec->backrefs);
 273         }
 274         return rec;
 275 }
 276
 277 static struct inode_record *get_inode_rec(struct cache_tree *inode_cache,
 278                                           u64 ino, int mod)
 279 {
 280         struct ptr_node *node;
 281         struct cache_extent *cache;
 282         struct inode_record *rec = NULL;
 283         int ret;
 284
 285         cache = lookup_cache_extent(inode_cache, ino, 1);
 286         if (cache) {
 287                 node = container_of(cache, struct ptr_node, cache);
 288                 rec = node->data;
 289                 if (mod && rec->refs > 1) {
 290                         node->data = clone_inode_rec(rec);
 291                         rec->refs--;
 292                         rec = node->data;
 293                 }
 294         } else if (mod) {
 295                 rec = calloc(1, sizeof(*rec));
 296                 rec->ino = ino;
 297                 rec->extent_start = (u64)-1;
 298                 rec->first_extent_gap = (u64)-1;
 299                 rec->refs = 1;
 300                 INIT_LIST_HEAD(&rec->backrefs);
 301
 302                 node = malloc(sizeof(*node));
 303                 node->cache.start = ino;
 304                 node->cache.size = 1;
 305                 node->data = rec;
 306
 307                 if (ino == BTRFS_FREE_INO_OBJECTID)
 308                         rec->found_link = 1;
 309
 310                 ret = insert_cache_extent(inode_cache, &node->cache);
 311                 BUG_ON(ret);
 312         }
 313         return rec;
 314 }
 315
 316 static void free_inode_rec(struct inode_record *rec)
 317 {
 318         struct inode_backref *backref;
 319
 320         if (--rec->refs > 0)
 321                 return;
 322
 323         while (!list_empty(&rec->backrefs)) {
 324                 backref = list_entry(rec->backrefs.next,
 325                                      struct inode_backref, list);
 326                 list_del(&backref->list);
 327                 free(backref);
 328         }
 329         free(rec);
 330 }
 331
 332 static int can_free_inode_rec(struct inode_record *rec)
 333 {
 334         if (!rec->errors && rec->checked && rec->found_inode_item &&
 335             rec->nlink == rec->found_link && list_empty(&rec->backrefs))
 336                 return 1;
 337         return 0;
 338 }
 339
 340 static void maybe_free_inode_rec(struct cache_tree *inode_cache,
 341                                  struct inode_record *rec)
 342 {
 343         struct cache_extent *cache;
 344         struct inode_backref *tmp, *backref;
 345         struct ptr_node *node;
 346         unsigned char filetype;
 347
 348         if (!rec->found_inode_item)
 349                 return;
 350
 351         filetype = imode_to_type(rec->imode);
 352         list_for_each_entry_safe(backref, tmp, &rec->backrefs, list) {
 353                 if (backref->found_dir_item && backref->found_dir_index) {
 354                         if (backref->filetype != filetype)
 355                                 backref->errors |= REF_ERR_FILETYPE_UNMATCH;
 356                         if (!backref->errors && backref->found_inode_ref) {
 357                                 list_del(&backref->list);
 358                                 free(backref);
 359                         }
 360                 }
 361         }
 362
 363         if (!rec->checked || rec->merging)
 364                 return;
 365
 366         if (S_ISDIR(rec->imode)) {
 367                 if (rec->found_size != rec->isize)
 368                         rec->errors |= I_ERR_DIR_ISIZE_WRONG;
 369                 if (rec->found_file_extent)
 370                         rec->errors |= I_ERR_ODD_FILE_EXTENT;
 371         } else if (S_ISREG(rec->imode) || S_ISLNK(rec->imode)) {
 372                 if (rec->found_dir_item)
 373                         rec->errors |= I_ERR_ODD_DIR_ITEM;
 374                 if (rec->found_size != rec->nbytes)
 375                         rec->errors |= I_ERR_FILE_NBYTES_WRONG;
 376                 if (rec->extent_start == (u64)-1 || rec->extent_start > 0)
 377                         rec->first_extent_gap = 0;
 378                 if (rec->nlink > 0 && (rec->extent_end < rec->isize ||
 379                     rec->first_extent_gap < rec->isize))
 380                         rec->errors |= I_ERR_FILE_EXTENT_DISCOUNT;
 381         }
 382
 383         if (S_ISREG(rec->imode) || S_ISLNK(rec->imode)) {
 384                 if (rec->found_csum_item && rec->nodatasum)
 385                         rec->errors |= I_ERR_ODD_CSUM_ITEM;
 386                 if (rec->some_csum_missing && !rec->nodatasum)
 387                         rec->errors |= I_ERR_SOME_CSUM_MISSING;
 388         }
 389
 390         BUG_ON(rec->refs != 1);
 391         if (can_free_inode_rec(rec)) {
 392                 cache = lookup_cache_extent(inode_cache, rec->ino, 1);
 393                 node = container_of(cache, struct ptr_node, cache);
 394                 BUG_ON(node->data != rec);
 395                 remove_cache_extent(inode_cache, &node->cache);
 396                 free(node);
 397                 free_inode_rec(rec);
 398         }
 399 }
 400
 401 static int check_orphan_item(struct btrfs_root *root, u64 ino)
 402 {
 403         struct btrfs_path path;
 404         struct btrfs_key key;
 405         int ret;
 406
 407         key.objectid = BTRFS_ORPHAN_OBJECTID;
 408         key.type = BTRFS_ORPHAN_ITEM_KEY;
 409         key.offset = ino;
 410
 411         btrfs_init_path(&path);
 412         ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
 413         btrfs_release_path(root, &path);
 414         if (ret > 0)
 415                 ret = -ENOENT;
 416         return ret;
 417 }
 418
 419 static int process_inode_item(struct extent_buffer *eb,
 420                               int slot, struct btrfs_key *key,
 421                               struct shared_node *active_node)
 422 {
 423         struct inode_record *rec;
 424         struct btrfs_inode_item *item;
 425
 426         rec = active_node->current;
 427         BUG_ON(rec->ino != key->objectid || rec->refs > 1);
 428         if (rec->found_inode_item) {
 429                 rec->errors |= I_ERR_DUP_INODE_ITEM;
 430                 return 1;
 431         }
 432         item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
 433         rec->nlink = btrfs_inode_nlink(eb, item);
 434         rec->isize = btrfs_inode_size(eb, item);
 435         rec->nbytes = btrfs_inode_nbytes(eb, item);
 436         rec->imode = btrfs_inode_mode(eb, item);
 437         if (btrfs_inode_flags(eb, item) & BTRFS_INODE_NODATASUM)
 438                 rec->nodatasum = 1;
 439         rec->found_inode_item = 1;
 440         if (rec->nlink == 0)
 441                 rec->errors |= I_ERR_NO_ORPHAN_ITEM;
 442         maybe_free_inode_rec(&active_node->inode_cache, rec);
 443         return 0;
 444 }
 445
 446 static struct inode_backref *get_inode_backref(struct inode_record *rec,
 447                                                 const char *name,
 448                                                 int namelen, u64 dir)
 449 {
 450         struct inode_backref *backref;
 451
 452         list_for_each_entry(backref, &rec->backrefs, list) {
 453                 if (backref->dir != dir || backref->namelen != namelen)
 454                         continue;
 455                 if (memcmp(name, backref->name, namelen))
 456                         continue;
 457                 return backref;
 458         }
 459
 460         backref = malloc(sizeof(*backref) + namelen + 1);
 461         memset(backref, 0, sizeof(*backref));
 462         backref->dir = dir;
 463         backref->namelen = namelen;
 464         memcpy(backref->name, name, namelen);
 465         backref->name[namelen] = '\0';
 466         list_add_tail(&backref->list, &rec->backrefs);
 467         return backref;
 468 }
 469
 470 static int add_inode_backref(struct cache_tree *inode_cache,
 471                              u64 ino, u64 dir, u64 index,
 472                              const char *name, int namelen,
 473                              int filetype, int itemtype, int errors)
 474 {
 475         struct inode_record *rec;
 476         struct inode_backref *backref;
 477
 478         rec = get_inode_rec(inode_cache, ino, 1);
 479         backref = get_inode_backref(rec, name, namelen, dir);
 480         if (errors)
 481                 backref->errors |= errors;
 482         if (itemtype == BTRFS_DIR_INDEX_KEY) {
 483                 if (backref->found_dir_index)
 484                         backref->errors |= REF_ERR_DUP_DIR_INDEX;
 485                 if (backref->found_inode_ref && backref->index != index)
 486                         backref->errors |= REF_ERR_INDEX_UNMATCH;
 487                 if (backref->found_dir_item && backref->filetype != filetype)
 488                         backref->errors |= REF_ERR_FILETYPE_UNMATCH;
 489
 490                 backref->index = index;
 491                 backref->filetype = filetype;
 492                 backref->found_dir_index = 1;
 493         } else if (itemtype == BTRFS_DIR_ITEM_KEY) {
 494                 rec->found_link++;
 495                 if (backref->found_dir_item)
 496                         backref->errors |= REF_ERR_DUP_DIR_ITEM;
 497                 if (backref->found_dir_index && backref->filetype != filetype)
 498                         backref->errors |= REF_ERR_FILETYPE_UNMATCH;
 499
 500                 backref->filetype = filetype;
 501                 backref->found_dir_item = 1;
 502         } else if ((itemtype == BTRFS_INODE_REF_KEY) ||
 503                    (itemtype == BTRFS_INODE_EXTREF_KEY)) {
 504                 if (backref->found_inode_ref)
 505                         backref->errors |= REF_ERR_DUP_INODE_REF;
 506                 if (backref->found_dir_index && backref->index != index)
 507                         backref->errors |= REF_ERR_INDEX_UNMATCH;
 508
 509                 backref->ref_type = itemtype;
 510                 backref->index = index;
 511                 backref->found_inode_ref = 1;
 512         } else {
 513                 BUG_ON(1);
 514         }
 515
 516         maybe_free_inode_rec(inode_cache, rec);
 517         return 0;
 518 }
 519
 520 static int merge_inode_recs(struct inode_record *src, struct inode_record *dst,
 521                             struct cache_tree *dst_cache)
 522 {
 523         struct inode_backref *backref;
 524         u32 dir_count = 0;
 525
 526         dst->merging = 1;
 527         list_for_each_entry(backref, &src->backrefs, list) {
 528                 if (backref->found_dir_index) {
 529                         add_inode_backref(dst_cache, dst->ino, backref->dir,
 530                                         backref->index, backref->name,
 531                                         backref->namelen, backref->filetype,
 532                                         BTRFS_DIR_INDEX_KEY, backref->errors);
 533                 }
 534                 if (backref->found_dir_item) {
 535                         dir_count++;
 536                         add_inode_backref(dst_cache, dst->ino,
 537                                         backref->dir, 0, backref->name,
 538                                         backref->namelen, backref->filetype,
 539                                         BTRFS_DIR_ITEM_KEY, backref->errors);
 540                 }
 541                 if (backref->found_inode_ref) {
 542                         add_inode_backref(dst_cache, dst->ino,
 543                                         backref->dir, backref->index,
 544                                         backref->name, backref->namelen, 0,
 545                                         backref->ref_type, backref->errors);
 546                 }
 547         }
 548
 549         if (src->found_dir_item)
 550                 dst->found_dir_item = 1;
 551         if (src->found_file_extent)
 552                 dst->found_file_extent = 1;
 553         if (src->found_csum_item)
 554                 dst->found_csum_item = 1;
 555         if (src->some_csum_missing)
 556                 dst->some_csum_missing = 1;
 557         if (dst->first_extent_gap > src->first_extent_gap)
 558                 dst->first_extent_gap = src->first_extent_gap;
 559
 560         BUG_ON(src->found_link < dir_count);
 561         dst->found_link += src->found_link - dir_count;
 562         dst->found_size += src->found_size;
 563         if (src->extent_start != (u64)-1) {
 564                 if (dst->extent_start == (u64)-1) {
 565                         dst->extent_start = src->extent_start;
 566                         dst->extent_end = src->extent_end;
 567                 } else {
 568                         if (dst->extent_end > src->extent_start)
 569                                 dst->errors |= I_ERR_FILE_EXTENT_OVERLAP;
 570                         else if (dst->extent_end < src->extent_start &&
 571                                  dst->extent_end < dst->first_extent_gap)
 572                                 dst->first_extent_gap = dst->extent_end;
 573                         if (dst->extent_end < src->extent_end)
 574                                 dst->extent_end = src->extent_end;
 575                 }
 576         }
 577
 578         dst->errors |= src->errors;
 579         if (src->found_inode_item) {
 580                 if (!dst->found_inode_item) {
 581                         dst->nlink = src->nlink;
 582                         dst->isize = src->isize;
 583                         dst->nbytes = src->nbytes;
 584                         dst->imode = src->imode;
 585                         dst->nodatasum = src->nodatasum;
 586                         dst->found_inode_item = 1;
 587                 } else {
 588                         dst->errors |= I_ERR_DUP_INODE_ITEM;
 589                 }
 590         }
 591         dst->merging = 0;
 592
 593         return 0;
 594 }
 595
 596 static int splice_shared_node(struct shared_node *src_node,
 597                               struct shared_node *dst_node)
 598 {
 599         struct cache_extent *cache;
 600         struct ptr_node *node, *ins;
 601         struct cache_tree *src, *dst;
 602         struct inode_record *rec, *conflict;
 603         u64 current_ino = 0;
 604         int splice = 0;
 605         int ret;
 606
 607         if (--src_node->refs == 0)
 608                 splice = 1;
 609         if (src_node->current)
 610                 current_ino = src_node->current->ino;
 611
 612         src = &src_node->root_cache;
 613         dst = &dst_node->root_cache;
 614 again:
 615         cache = search_cache_extent(src, 0);
 616         while (cache) {
 617                 node = container_of(cache, struct ptr_node, cache);
 618                 rec = node->data;
 619                 cache = next_cache_extent(cache);
 620
 621                 if (splice) {
 622                         remove_cache_extent(src, &node->cache);
 623                         ins = node;
 624                 } else {
 625                         ins = malloc(sizeof(*ins));
 626                         ins->cache.start = node->cache.start;
 627                         ins->cache.size = node->cache.size;
 628                         ins->data = rec;
 629                         rec->refs++;
 630                 }
 631                 ret = insert_cache_extent(dst, &ins->cache);
 632                 if (ret == -EEXIST) {
 633                         conflict = get_inode_rec(dst, rec->ino, 1);
 634                         merge_inode_recs(rec, conflict, dst);
 635                         if (rec->checked) {
 636                                 conflict->checked = 1;
 637                                 if (dst_node->current == conflict)
 638                                         dst_node->current = NULL;
 639                         }
 640                         maybe_free_inode_rec(dst, conflict);
 641                         free_inode_rec(rec);
 642                         free(ins);
 643                 } else {
 644                         BUG_ON(ret);
 645                 }
 646         }
 647
 648         if (src == &src_node->root_cache) {
 649                 src = &src_node->inode_cache;
 650                 dst = &dst_node->inode_cache;
 651                 goto again;
 652         }
 653
 654         if (current_ino > 0 && (!dst_node->current ||
 655             current_ino > dst_node->current->ino)) {
 656                 if (dst_node->current) {
 657                         dst_node->current->checked = 1;
 658                         maybe_free_inode_rec(dst, dst_node->current);
 659                 }
 660                 dst_node->current = get_inode_rec(dst, current_ino, 1);
 661         }
 662         return 0;
 663 }
 664
 665 static void free_inode_ptr(struct cache_extent *cache)
 666 {
 667         struct ptr_node *node;
 668         struct inode_record *rec;
 669
 670         node = container_of(cache, struct ptr_node, cache);
 671         rec = node->data;
 672         free_inode_rec(rec);
 673         free(node);
 674 }
 675
 676 FREE_EXTENT_CACHE_BASED_TREE(inode_recs, free_inode_ptr);
 677
 678 static struct shared_node *find_shared_node(struct cache_tree *shared,
 679                                             u64 bytenr)
 680 {
 681         struct cache_extent *cache;
 682         struct shared_node *node;
 683
 684         cache = lookup_cache_extent(shared, bytenr, 1);
 685         if (cache) {
 686                 node = container_of(cache, struct shared_node, cache);
 687                 return node;
 688         }
 689         return NULL;
 690 }
 691
 692 static int add_shared_node(struct cache_tree *shared, u64 bytenr, u32 refs)
 693 {
 694         int ret;
 695         struct shared_node *node;
 696
 697         node = calloc(1, sizeof(*node));
 698         node->cache.start = bytenr;
 699         node->cache.size = 1;
 700         cache_tree_init(&node->root_cache);
 701         cache_tree_init(&node->inode_cache);
 702         node->refs = refs;
 703
 704         ret = insert_cache_extent(shared, &node->cache);
 705         BUG_ON(ret);
 706         return 0;
 707 }
 708
 709 static int enter_shared_node(struct btrfs_root *root, u64 bytenr, u32 refs,
 710                              struct walk_control *wc, int level)
 711 {
 712         struct shared_node *node;
 713         struct shared_node *dest;
 714
 715         if (level == wc->active_node)
 716                 return 0;
 717
 718         BUG_ON(wc->active_node <= level);
 719         node = find_shared_node(&wc->shared, bytenr);
 720         if (!node) {
 721                 add_shared_node(&wc->shared, bytenr, refs);
 722                 node = find_shared_node(&wc->shared, bytenr);
 723                 wc->nodes[level] = node;
 724                 wc->active_node = level;
 725                 return 0;
 726         }
 727
 728         if (wc->root_level == wc->active_node &&
 729             btrfs_root_refs(&root->root_item) == 0) {
 730                 if (--node->refs == 0) {
 731                         free_inode_recs_tree(&node->root_cache);
 732                         free_inode_recs_tree(&node->inode_cache);
 733                         remove_cache_extent(&wc->shared, &node->cache);
 734                         free(node);
 735                 }
 736                 return 1;
 737         }
 738
 739         dest = wc->nodes[wc->active_node];
 740         splice_shared_node(node, dest);
 741         if (node->refs == 0) {
 742                 remove_cache_extent(&wc->shared, &node->cache);
 743                 free(node);
 744         }
 745         return 1;
 746 }
 747
 748 static int leave_shared_node(struct btrfs_root *root,
 749                              struct walk_control *wc, int level)
 750 {
 751         struct shared_node *node;
 752         struct shared_node *dest;
 753         int i;
 754
 755         if (level == wc->root_level)
 756                 return 0;
 757
 758         for (i = level + 1; i < BTRFS_MAX_LEVEL; i++) {
 759                 if (wc->nodes[i])
 760                         break;
 761         }
 762         BUG_ON(i >= BTRFS_MAX_LEVEL);
 763
 764         node = wc->nodes[wc->active_node];
 765         wc->nodes[wc->active_node] = NULL;
 766         wc->active_node = i;
 767
 768         dest = wc->nodes[wc->active_node];
 769         if (wc->active_node < wc->root_level ||
 770             btrfs_root_refs(&root->root_item) > 0) {
 771                 BUG_ON(node->refs <= 1);
 772                 splice_shared_node(node, dest);
 773         } else {
 774                 BUG_ON(node->refs < 2);
 775                 node->refs--;
 776         }
 777         return 0;
 778 }
 779
 780 static int is_child_root(struct btrfs_root *root, u64 parent_root_id,
 781                          u64 child_root_id)
 782 {
 783         struct btrfs_path path;
 784         struct btrfs_key key;
 785         struct extent_buffer *leaf;
 786         int has_parent = 0;
 787         int ret;
 788
 789         btrfs_init_path(&path);
 790
 791         key.objectid = parent_root_id;
 792         key.type = BTRFS_ROOT_REF_KEY;
 793         key.offset = child_root_id;
 794         ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, &path,
 795                                 0, 0);
 796         BUG_ON(ret < 0);
 797         btrfs_release_path(root, &path);
 798         if (!ret)
 799                 return 1;
 800
 801         key.objectid = child_root_id;
 802         key.type = BTRFS_ROOT_BACKREF_KEY;
 803         key.offset = 0;
 804         ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, &path,
 805                                 0, 0);
 806         BUG_ON(ret <= 0);
 807
 808         while (1) {
 809                 leaf = path.nodes[0];
 810                 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
 811                         ret = btrfs_next_leaf(root->fs_info->tree_root, &path);
 812                         BUG_ON(ret < 0);
 813
 814                         if (ret > 0)
 815                                 break;
 816                         leaf = path.nodes[0];
 817                 }
 818
 819                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
 820                 if (key.objectid != child_root_id ||
 821                     key.type != BTRFS_ROOT_BACKREF_KEY)
 822                         break;
 823
 824                 has_parent = 1;
 825
 826                 if (key.offset == parent_root_id) {
 827                         btrfs_release_path(root, &path);
 828                         return 1;
 829                 }
 830
 831                 path.slots[0]++;
 832         }
 833
 834         btrfs_release_path(root, &path);
 835         return has_parent? 0 : -1;
 836 }
 837
 838 static int process_dir_item(struct btrfs_root *root,
 839                             struct extent_buffer *eb,
 840                             int slot, struct btrfs_key *key,
 841                             struct shared_node *active_node)
 842 {
 843         u32 total;
 844         u32 cur = 0;
 845         u32 len;
 846         u32 name_len;
 847         u32 data_len;
 848         int error;
 849         int nritems = 0;
 850         int filetype;
 851         struct btrfs_dir_item *di;
 852         struct inode_record *rec;
 853         struct cache_tree *root_cache;
 854         struct cache_tree *inode_cache;
 855         struct btrfs_key location;
 856         char namebuf[BTRFS_NAME_LEN];
 857
 858         root_cache = &active_node->root_cache;
 859         inode_cache = &active_node->inode_cache;
 860         rec = active_node->current;
 861         rec->found_dir_item = 1;
 862
 863         di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
 864         total = btrfs_item_size_nr(eb, slot);
 865         while (cur < total) {
 866                 nritems++;
 867                 btrfs_dir_item_key_to_cpu(eb, di, &location);
 868                 name_len = btrfs_dir_name_len(eb, di);
 869                 data_len = btrfs_dir_data_len(eb, di);
 870                 filetype = btrfs_dir_type(eb, di);
 871
 872                 rec->found_size += name_len;
 873                 if (name_len <= BTRFS_NAME_LEN) {
 874                         len = name_len;
 875                         error = 0;
 876                 } else {
 877                         len = BTRFS_NAME_LEN;
 878                         error = REF_ERR_NAME_TOO_LONG;
 879                 }
 880                 read_extent_buffer(eb, namebuf, (unsigned long)(di + 1), len);
 881
 882                 if (location.type == BTRFS_INODE_ITEM_KEY) {
 883                         add_inode_backref(inode_cache, location.objectid,
 884                                           key->objectid, key->offset, namebuf,
 885                                           len, filetype, key->type, error);
 886                 } else if (location.type == BTRFS_ROOT_ITEM_KEY) {
 887                         add_inode_backref(root_cache, location.objectid,
 888                                           key->objectid, key->offset,
 889                                           namebuf, len, filetype,
 890                                           key->type, error);
 891                 } else {
 892                         fprintf(stderr, "warning line %d\n", __LINE__);
 893                 }
 894
 895                 len = sizeof(*di) + name_len + data_len;
 896                 di = (struct btrfs_dir_item *)((char *)di + len);
 897                 cur += len;
 898         }
 899         if (key->type == BTRFS_DIR_INDEX_KEY && nritems > 1)
 900                 rec->errors |= I_ERR_DUP_DIR_INDEX;
 901
 902         return 0;
 903 }
 904
 905 static int process_inode_ref(struct extent_buffer *eb,
 906                              int slot, struct btrfs_key *key,
 907                              struct shared_node *active_node)
 908 {
 909         u32 total;
 910         u32 cur = 0;
 911         u32 len;
 912         u32 name_len;
 913         u64 index;
 914         int error;
 915         struct cache_tree *inode_cache;
 916         struct btrfs_inode_ref *ref;
 917         char namebuf[BTRFS_NAME_LEN];
 918
 919         inode_cache = &active_node->inode_cache;
 920
 921         ref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
 922         total = btrfs_item_size_nr(eb, slot);
 923         while (cur < total) {
 924                 name_len = btrfs_inode_ref_name_len(eb, ref);
 925                 index = btrfs_inode_ref_index(eb, ref);
 926                 if (name_len <= BTRFS_NAME_LEN) {
 927                         len = name_len;
 928                         error = 0;
 929                 } else {
 930                         len = BTRFS_NAME_LEN;
 931                         error = REF_ERR_NAME_TOO_LONG;
 932                 }
 933                 read_extent_buffer(eb, namebuf, (unsigned long)(ref + 1), len);
 934                 add_inode_backref(inode_cache, key->objectid, key->offset,
 935                                   index, namebuf, len, 0, key->type, error);
 936
 937                 len = sizeof(*ref) + name_len;
 938                 ref = (struct btrfs_inode_ref *)((char *)ref + len);
 939                 cur += len;
 940         }
 941         return 0;
 942 }
 943
 944 static int process_inode_extref(struct extent_buffer *eb,
 945                                 int slot, struct btrfs_key *key,
 946                                 struct shared_node *active_node)
 947 {
 948         u32 total;
 949         u32 cur = 0;
 950         u32 len;
 951         u32 name_len;
 952         u64 index;
 953         u64 parent;
 954         int error;
 955         struct cache_tree *inode_cache;
 956         struct btrfs_inode_extref *extref;
 957         char namebuf[BTRFS_NAME_LEN];
 958
 959         inode_cache = &active_node->inode_cache;
 960
 961         extref = btrfs_item_ptr(eb, slot, struct btrfs_inode_extref);
 962         total = btrfs_item_size_nr(eb, slot);
 963         while (cur < total) {
 964                 name_len = btrfs_inode_extref_name_len(eb, extref);
 965                 index = btrfs_inode_extref_index(eb, extref);
 966                 parent = btrfs_inode_extref_parent(eb, extref);
 967                 if (name_len <= BTRFS_NAME_LEN) {
 968                         len = name_len;
 969                         error = 0;
 970                 } else {
 971                         len = BTRFS_NAME_LEN;
 972                         error = REF_ERR_NAME_TOO_LONG;
 973                 }
 974                 read_extent_buffer(eb, namebuf,
 975                                    (unsigned long)(extref + 1), len);
 976                 add_inode_backref(inode_cache, key->objectid, parent,
 977                                   index, namebuf, len, 0, key->type, error);
 978
 979                 len = sizeof(*extref) + name_len;
 980                 extref = (struct btrfs_inode_extref *)((char *)extref + len);
 981                 cur += len;
 982         }
 983         return 0;
 984
 985 }
 986
 987 static u64 count_csum_range(struct btrfs_root *root, u64 start, u64 len)
 988 {
 989         struct btrfs_key key;
 990         struct btrfs_path path;
 991         struct extent_buffer *leaf;
 992         int ret ;
 993         size_t size;
 994         u64 found = 0;
 995         u64 csum_end;
 996         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
 997
 998         btrfs_init_path(&path);
 999
1000         key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
1001         key.offset = start;
1002         key.type = BTRFS_EXTENT_CSUM_KEY;
1003
1004         ret = btrfs_search_slot(NULL, root->fs_info->csum_root,
1005                                 &key, &path, 0, 0);
1006         BUG_ON(ret < 0);
1007         if (ret > 0 && path.slots[0] > 0) {
1008                 leaf = path.nodes[0];
1009                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0] - 1);
1010                 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
1011                     key.type == BTRFS_EXTENT_CSUM_KEY)
1012                         path.slots[0]--;
1013         }
1014
1015         while (len > 0) {
1016                 leaf = path.nodes[0];
1017                 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
1018                         ret = btrfs_next_leaf(root->fs_info->csum_root, &path);
1019                         BUG_ON(ret < 0);
1020                         if (ret > 0)
1021                                 break;
1022                         leaf = path.nodes[0];
1023                 }
1024
1025                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
1026                 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1027                     key.type != BTRFS_EXTENT_CSUM_KEY)
1028                         break;
1029
1030                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
1031                 if (key.offset >= start + len)
1032                         break;
1033
1034                 if (key.offset > start)
1035                         start = key.offset;
1036
1037                 size = btrfs_item_size_nr(leaf, path.slots[0]);
1038                 csum_end = key.offset + (size / csum_size) * root->sectorsize;
1039                 if (csum_end > start) {
1040                         size = min(csum_end - start, len);
1041                         len -= size;
1042                         start += size;
1043                         found += size;
1044                 }
1045
1046                 path.slots[0]++;
1047         }
1048         btrfs_release_path(root->fs_info->csum_root, &path);
1049         return found;
1050 }
1051
1052 static int process_file_extent(struct btrfs_root *root,
1053                                 struct extent_buffer *eb,
1054                                 int slot, struct btrfs_key *key,
1055                                 struct shared_node *active_node)
1056 {
1057         struct inode_record *rec;
1058         struct btrfs_file_extent_item *fi;
1059         u64 num_bytes = 0;
1060         u64 disk_bytenr = 0;
1061         u64 extent_offset = 0;
1062         u64 mask = root->sectorsize - 1;
1063         int extent_type;
1064
1065         rec = active_node->current;
1066         BUG_ON(rec->ino != key->objectid || rec->refs > 1);
1067         rec->found_file_extent = 1;
1068
1069         if (rec->extent_start == (u64)-1) {
1070                 rec->extent_start = key->offset;
1071                 rec->extent_end = key->offset;
1072         }
1073
1074         if (rec->extent_end > key->offset)
1075                 rec->errors |= I_ERR_FILE_EXTENT_OVERLAP;
1076         else if (rec->extent_end < key->offset &&
1077                  rec->extent_end < rec->first_extent_gap)
1078                 rec->first_extent_gap = rec->extent_end;
1079
1080         fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
1081         extent_type = btrfs_file_extent_type(eb, fi);
1082
1083         if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1084                 num_bytes = btrfs_file_extent_inline_len(eb, fi);
1085                 if (num_bytes == 0)
1086                         rec->errors |= I_ERR_BAD_FILE_EXTENT;
1087                 rec->found_size += num_bytes;
1088                 num_bytes = (num_bytes + mask) & ~mask;
1089         } else if (extent_type == BTRFS_FILE_EXTENT_REG ||
1090                    extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1091                 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
1092                 disk_bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1093                 extent_offset = btrfs_file_extent_offset(eb, fi);
1094                 if (num_bytes == 0 || (num_bytes & mask))
1095                         rec->errors |= I_ERR_BAD_FILE_EXTENT;
1096                 if (num_bytes + extent_offset >
1097                     btrfs_file_extent_ram_bytes(eb, fi))
1098                         rec->errors |= I_ERR_BAD_FILE_EXTENT;
1099                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC &&
1100                     (btrfs_file_extent_compression(eb, fi) ||
1101                      btrfs_file_extent_encryption(eb, fi) ||
1102                      btrfs_file_extent_other_encoding(eb, fi)))
1103                         rec->errors |= I_ERR_BAD_FILE_EXTENT;
1104                 if (disk_bytenr > 0)
1105                         rec->found_size += num_bytes;
1106         } else {
1107                 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1108         }
1109         rec->extent_end = key->offset + num_bytes;
1110
1111         if (disk_bytenr > 0) {
1112                 u64 found;
1113                 if (btrfs_file_extent_compression(eb, fi))
1114                         num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1115                 else
1116                         disk_bytenr += extent_offset;
1117
1118                 found = count_csum_range(root, disk_bytenr, num_bytes);
1119                 if (extent_type == BTRFS_FILE_EXTENT_REG) {
1120                         if (found > 0)
1121                                 rec->found_csum_item = 1;
1122                         if (found < num_bytes)
1123                                 rec->some_csum_missing = 1;
1124                 } else if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1125                         if (found > 0)
1126                                 rec->errors |= I_ERR_ODD_CSUM_ITEM;
1127                 }
1128         }
1129         return 0;
1130 }
1131
1132 static int process_one_leaf(struct btrfs_root *root, struct extent_buffer *eb,
1133                             struct walk_control *wc)
1134 {
1135         struct btrfs_key key;
1136         u32 nritems;
1137         int i;
1138         int ret = 0;
1139         struct cache_tree *inode_cache;
1140         struct shared_node *active_node;
1141
1142         if (wc->root_level == wc->active_node &&
1143             btrfs_root_refs(&root->root_item) == 0)
1144                 return 0;
1145
1146         active_node = wc->nodes[wc->active_node];
1147         inode_cache = &active_node->inode_cache;
1148         nritems = btrfs_header_nritems(eb);
1149         for (i = 0; i < nritems; i++) {
1150                 btrfs_item_key_to_cpu(eb, &key, i);
1151
1152                 if (key.objectid == BTRFS_FREE_SPACE_OBJECTID)
1153                         continue;
1154
1155                 if (active_node->current == NULL ||
1156                     active_node->current->ino < key.objectid) {
1157                         if (active_node->current) {
1158                                 active_node->current->checked = 1;
1159                                 maybe_free_inode_rec(inode_cache,
1160                                                      active_node->current);
1161                         }
1162                         active_node->current = get_inode_rec(inode_cache,
1163                                                              key.objectid, 1);
1164                 }
1165                 switch (key.type) {
1166                 case BTRFS_DIR_ITEM_KEY:
1167                 case BTRFS_DIR_INDEX_KEY:
1168                         ret = process_dir_item(root, eb, i, &key, active_node);
1169                         break;
1170                 case BTRFS_INODE_REF_KEY:
1171                         ret = process_inode_ref(eb, i, &key, active_node);
1172                         break;
1173                 case BTRFS_INODE_EXTREF_KEY:
1174                         ret = process_inode_extref(eb, i, &key, active_node);
1175                         break;
1176                 case BTRFS_INODE_ITEM_KEY:
1177                         ret = process_inode_item(eb, i, &key, active_node);
1178                         break;
1179                 case BTRFS_EXTENT_DATA_KEY:
1180                         ret = process_file_extent(root, eb, i, &key,
1181                                                   active_node);
1182                         break;
1183                 default:
1184                         break;
1185                 };
1186         }
1187         return ret;
1188 }
1189
1190 static void reada_walk_down(struct btrfs_root *root,
1191                             struct extent_buffer *node, int slot)
1192 {
1193         u64 bytenr;
1194         u64 ptr_gen;
1195         u32 nritems;
1196         u32 blocksize;
1197         int i;
1198         int ret;
1199         int level;
1200
1201         level = btrfs_header_level(node);
1202         if (level != 1)
1203                 return;
1204
1205         nritems = btrfs_header_nritems(node);
1206         blocksize = btrfs_level_size(root, level - 1);
1207         for (i = slot; i < nritems; i++) {
1208                 bytenr = btrfs_node_blockptr(node, i);
1209                 ptr_gen = btrfs_node_ptr_generation(node, i);
1210                 ret = readahead_tree_block(root, bytenr, blocksize, ptr_gen);
1211                 if (ret)
1212                         break;
1213         }
1214 }
1215
1216 static int walk_down_tree(struct btrfs_root *root, struct btrfs_path *path,
1217                           struct walk_control *wc, int *level)
1218 {
1219         u64 bytenr;
1220         u64 ptr_gen;
1221         struct extent_buffer *next;
1222         struct extent_buffer *cur;
1223         u32 blocksize;
1224         int ret;
1225         u64 refs;
1226
1227         WARN_ON(*level < 0);
1228         WARN_ON(*level >= BTRFS_MAX_LEVEL);
1229         ret = btrfs_lookup_extent_info(NULL, root,
1230                                        path->nodes[*level]->start,
1231                                        *level, 1, &refs, NULL);
1232         if (ret < 0)
1233                 goto out;
1234
1235         if (refs > 1) {
1236                 ret = enter_shared_node(root, path->nodes[*level]->start,
1237                                         refs, wc, *level);
1238                 if (ret > 0)
1239                         goto out;
1240         }
1241
1242         while (*level >= 0) {
1243                 WARN_ON(*level < 0);
1244                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1245                 cur = path->nodes[*level];
1246
1247                 if (btrfs_header_level(cur) != *level)
1248                         WARN_ON(1);
1249
1250                 if (path->slots[*level] >= btrfs_header_nritems(cur))
1251                         break;
1252                 if (*level == 0) {
1253                         ret = process_one_leaf(root, cur, wc);
1254                         break;
1255                 }
1256                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1257                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
1258                 blocksize = btrfs_level_size(root, *level - 1);
1259                 ret = btrfs_lookup_extent_info(NULL, root, bytenr, *level - 1,
1260                                                1, &refs, NULL);
1261                 if (ret < 0)
1262                         refs = 0;
1263
1264                 if (refs > 1) {
1265                         ret = enter_shared_node(root, bytenr, refs,
1266                                                 wc, *level - 1);
1267                         if (ret > 0) {
1268                                 path->slots[*level]++;
1269                                 continue;
1270                         }
1271                 }
1272
1273                 next = btrfs_find_tree_block(root, bytenr, blocksize);
1274                 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
1275                         free_extent_buffer(next);
1276                         reada_walk_down(root, cur, path->slots[*level]);
1277                         next = read_tree_block(root, bytenr, blocksize,
1278                                                ptr_gen);
1279                 }
1280
1281                 *level = *level - 1;
1282                 free_extent_buffer(path->nodes[*level]);
1283                 path->nodes[*level] = next;
1284                 path->slots[*level] = 0;
1285         }
1286 out:
1287         path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
1288         return 0;
1289 }
1290
1291 static int walk_up_tree(struct btrfs_root *root, struct btrfs_path *path,
1292                         struct walk_control *wc, int *level)
1293 {
1294         int i;
1295         struct extent_buffer *leaf;
1296
1297         for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
1298                 leaf = path->nodes[i];
1299                 if (path->slots[i] + 1 < btrfs_header_nritems(leaf)) {
1300                         path->slots[i]++;
1301                         *level = i;
1302                         return 0;
1303                 } else {
1304                         free_extent_buffer(path->nodes[*level]);
1305                         path->nodes[*level] = NULL;
1306                         BUG_ON(*level > wc->active_node);
1307                         if (*level == wc->active_node)
1308                                 leave_shared_node(root, wc, *level);
1309                         *level = i + 1;
1310                 }
1311         }
1312         return 1;
1313 }
1314
1315 static int check_root_dir(struct inode_record *rec)
1316 {
1317         struct inode_backref *backref;
1318         int ret = -1;
1319
1320         if (!rec->found_inode_item || rec->errors)
1321                 goto out;
1322         if (rec->nlink != 1 || rec->found_link != 0)
1323                 goto out;
1324         if (list_empty(&rec->backrefs))
1325                 goto out;
1326         backref = list_entry(rec->backrefs.next, struct inode_backref, list);
1327         if (!backref->found_inode_ref)
1328                 goto out;
1329         if (backref->index != 0 || backref->namelen != 2 ||
1330             memcmp(backref->name, "..", 2))
1331                 goto out;
1332         if (backref->found_dir_index || backref->found_dir_item)
1333                 goto out;
1334         ret = 0;
1335 out:
1336         return ret;
1337 }
1338
1339 static int check_inode_recs(struct btrfs_root *root,
1340                             struct cache_tree *inode_cache)
1341 {
1342         struct cache_extent *cache;
1343         struct ptr_node *node;
1344         struct inode_record *rec;
1345         struct inode_backref *backref;
1346         int ret;
1347         u64 error = 0;
1348         u64 root_dirid = btrfs_root_dirid(&root->root_item);
1349
1350         if (btrfs_root_refs(&root->root_item) == 0) {
1351                 if (!cache_tree_empty(inode_cache))
1352                         fprintf(stderr, "warning line %d\n", __LINE__);
1353                 return 0;
1354         }
1355
1356         rec = get_inode_rec(inode_cache, root_dirid, 0);
1357         if (rec) {
1358                 ret = check_root_dir(rec);
1359                 if (ret) {
1360                         fprintf(stderr, "root %llu root dir %llu error\n",
1361                                 (unsigned long long)root->root_key.objectid,
1362                                 (unsigned long long)root_dirid);
1363                         error++;
1364                 }
1365         } else {
1366                 fprintf(stderr, "root %llu root dir %llu not found\n",
1367                         (unsigned long long)root->root_key.objectid,
1368                         (unsigned long long)root_dirid);
1369         }
1370
1371         while (1) {
1372                 cache = search_cache_extent(inode_cache, 0);
1373                 if (!cache)
1374                         break;
1375                 node = container_of(cache, struct ptr_node, cache);
1376                 rec = node->data;
1377                 remove_cache_extent(inode_cache, &node->cache);
1378                 free(node);
1379                 if (rec->ino == root_dirid ||
1380                     rec->ino == BTRFS_ORPHAN_OBJECTID) {
1381                         free_inode_rec(rec);
1382                         continue;
1383                 }
1384
1385                 if (rec->errors & I_ERR_NO_ORPHAN_ITEM) {
1386                         ret = check_orphan_item(root, rec->ino);
1387                         if (ret == 0)
1388                                 rec->errors &= ~I_ERR_NO_ORPHAN_ITEM;
1389                         if (can_free_inode_rec(rec)) {
1390                                 free_inode_rec(rec);
1391                                 continue;
1392                         }
1393                 }
1394
1395                 error++;
1396                 if (!rec->found_inode_item)
1397                         rec->errors |= I_ERR_NO_INODE_ITEM;
1398                 if (rec->found_link != rec->nlink)
1399                         rec->errors |= I_ERR_LINK_COUNT_WRONG;
1400                 fprintf(stderr, "root %llu inode %llu errors %x\n",
1401                         (unsigned long long) root->root_key.objectid,
1402                         (unsigned long long) rec->ino, rec->errors);
1403                 list_for_each_entry(backref, &rec->backrefs, list) {
1404                         if (!backref->found_dir_item)
1405                                 backref->errors |= REF_ERR_NO_DIR_ITEM;
1406                         if (!backref->found_dir_index)
1407                                 backref->errors |= REF_ERR_NO_DIR_INDEX;
1408                         if (!backref->found_inode_ref)
1409                                 backref->errors |= REF_ERR_NO_INODE_REF;
1410                         fprintf(stderr, "\tunresolved ref dir %llu index %llu"
1411                                 " namelen %u name %s filetype %d error %x\n",
1412                                 (unsigned long long)backref->dir,
1413                                 (unsigned long long)backref->index,
1414                                 backref->namelen, backref->name,
1415                                 backref->filetype, backref->errors);
1416                 }
1417                 free_inode_rec(rec);
1418         }
1419         return (error > 0) ? -1 : 0;
1420 }
1421
1422 static struct root_record *get_root_rec(struct cache_tree *root_cache,
1423                                         u64 objectid)
1424 {
1425         struct cache_extent *cache;
1426         struct root_record *rec = NULL;
1427         int ret;
1428
1429         cache = lookup_cache_extent(root_cache, objectid, 1);
1430         if (cache) {
1431                 rec = container_of(cache, struct root_record, cache);
1432         } else {
1433                 rec = calloc(1, sizeof(*rec));
1434                 rec->objectid = objectid;
1435                 INIT_LIST_HEAD(&rec->backrefs);
1436                 rec->cache.start = objectid;
1437                 rec->cache.size = 1;
1438
1439                 ret = insert_cache_extent(root_cache, &rec->cache);
1440                 BUG_ON(ret);
1441         }
1442         return rec;
1443 }
1444
1445 static struct root_backref *get_root_backref(struct root_record *rec,
1446                                              u64 ref_root, u64 dir, u64 index,
1447                                              const char *name, int namelen)
1448 {
1449         struct root_backref *backref;
1450
1451         list_for_each_entry(backref, &rec->backrefs, list) {
1452                 if (backref->ref_root != ref_root || backref->dir != dir ||
1453                     backref->namelen != namelen)
1454                         continue;
1455                 if (memcmp(name, backref->name, namelen))
1456                         continue;
1457                 return backref;
1458         }
1459
1460         backref = malloc(sizeof(*backref) + namelen + 1);
1461         memset(backref, 0, sizeof(*backref));
1462         backref->ref_root = ref_root;
1463         backref->dir = dir;
1464         backref->index = index;
1465         backref->namelen = namelen;
1466         memcpy(backref->name, name, namelen);
1467         backref->name[namelen] = '\0';
1468         list_add_tail(&backref->list, &rec->backrefs);
1469         return backref;
1470 }
1471
1472 static void free_root_record(struct cache_extent *cache)
1473 {
1474         struct root_record *rec;
1475         struct root_backref *backref;
1476
1477         rec = container_of(cache, struct root_record, cache);
1478         while (!list_empty(&rec->backrefs)) {
1479                 backref = list_entry(rec->backrefs.next,
1480                                      struct root_backref, list);
1481                 list_del(&backref->list);
1482                 free(backref);
1483         }
1484
1485         kfree(rec);
1486 }
1487
1488 FREE_EXTENT_CACHE_BASED_TREE(root_recs, free_root_record);
1489
1490 static int add_root_backref(struct cache_tree *root_cache,
1491                             u64 root_id, u64 ref_root, u64 dir, u64 index,
1492                             const char *name, int namelen,
1493                             int item_type, int errors)
1494 {
1495         struct root_record *rec;
1496         struct root_backref *backref;
1497
1498         rec = get_root_rec(root_cache, root_id);
1499         backref = get_root_backref(rec, ref_root, dir, index, name, namelen);
1500
1501         backref->errors |= errors;
1502
1503         if (item_type != BTRFS_DIR_ITEM_KEY) {
1504                 if (backref->found_dir_index || backref->found_back_ref ||
1505                     backref->found_forward_ref) {
1506                         if (backref->index != index)
1507                                 backref->errors |= REF_ERR_INDEX_UNMATCH;
1508                 } else {
1509                         backref->index = index;
1510                 }
1511         }
1512
1513         if (item_type == BTRFS_DIR_ITEM_KEY) {
1514                 if (backref->found_forward_ref)
1515                         rec->found_ref++;
1516                 backref->found_dir_item = 1;
1517         } else if (item_type == BTRFS_DIR_INDEX_KEY) {
1518                 backref->found_dir_index = 1;
1519         } else if (item_type == BTRFS_ROOT_REF_KEY) {
1520                 if (backref->found_forward_ref)
1521                         backref->errors |= REF_ERR_DUP_ROOT_REF;
1522                 else if (backref->found_dir_item)
1523                         rec->found_ref++;
1524                 backref->found_forward_ref = 1;
1525         } else if (item_type == BTRFS_ROOT_BACKREF_KEY) {
1526                 if (backref->found_back_ref)
1527                         backref->errors |= REF_ERR_DUP_ROOT_BACKREF;
1528                 backref->found_back_ref = 1;
1529         } else {
1530                 BUG_ON(1);
1531         }
1532
1533         if (backref->found_forward_ref && backref->found_dir_item)
1534                 backref->reachable = 1;
1535         return 0;
1536 }
1537
1538 static int merge_root_recs(struct btrfs_root *root,
1539                            struct cache_tree *src_cache,
1540                            struct cache_tree *dst_cache)
1541 {
1542         struct cache_extent *cache;
1543         struct ptr_node *node;
1544         struct inode_record *rec;
1545         struct inode_backref *backref;
1546
1547         if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
1548                 free_inode_recs_tree(src_cache);
1549                 return 0;
1550         }
1551
1552         while (1) {
1553                 cache = search_cache_extent(src_cache, 0);
1554                 if (!cache)
1555                         break;
1556                 node = container_of(cache, struct ptr_node, cache);
1557                 rec = node->data;
1558                 remove_cache_extent(src_cache, &node->cache);
1559                 free(node);
1560
1561                 if (!is_child_root(root, root->objectid, rec->ino))
1562                         goto skip;
1563
1564                 list_for_each_entry(backref, &rec->backrefs, list) {
1565                         BUG_ON(backref->found_inode_ref);
1566                         if (backref->found_dir_item)
1567                                 add_root_backref(dst_cache, rec->ino,
1568                                         root->root_key.objectid, backref->dir,
1569                                         backref->index, backref->name,
1570                                         backref->namelen, BTRFS_DIR_ITEM_KEY,
1571                                         backref->errors);
1572                         if (backref->found_dir_index)
1573                                 add_root_backref(dst_cache, rec->ino,
1574                                         root->root_key.objectid, backref->dir,
1575                                         backref->index, backref->name,
1576                                         backref->namelen, BTRFS_DIR_INDEX_KEY,
1577                                         backref->errors);
1578                 }
1579 skip:
1580                 free_inode_rec(rec);
1581         }
1582         return 0;
1583 }
1584
1585 static int check_root_refs(struct btrfs_root *root,
1586                            struct cache_tree *root_cache)
1587 {
1588         struct root_record *rec;
1589         struct root_record *ref_root;
1590         struct root_backref *backref;
1591         struct cache_extent *cache;
1592         int loop = 1;
1593         int ret;
1594         int error;
1595         int errors = 0;
1596
1597         rec = get_root_rec(root_cache, BTRFS_FS_TREE_OBJECTID);
1598         rec->found_ref = 1;
1599
1600         /* fixme: this can not detect circular references */
1601         while (loop) {
1602                 loop = 0;
1603                 cache = search_cache_extent(root_cache, 0);
1604                 while (1) {
1605                         if (!cache)
1606                                 break;
1607                         rec = container_of(cache, struct root_record, cache);
1608                         cache = next_cache_extent(cache);
1609
1610                         if (rec->found_ref == 0)
1611                                 continue;
1612
1613                         list_for_each_entry(backref, &rec->backrefs, list) {
1614                                 if (!backref->reachable)
1615                                         continue;
1616
1617                                 ref_root = get_root_rec(root_cache,
1618                                                         backref->ref_root);
1619                                 if (ref_root->found_ref > 0)
1620                                         continue;
1621
1622                                 backref->reachable = 0;
1623                                 rec->found_ref--;
1624                                 if (rec->found_ref == 0)
1625                                         loop = 1;
1626                         }
1627                 }
1628         }
1629
1630         cache = search_cache_extent(root_cache, 0);
1631         while (1) {
1632                 if (!cache)
1633                         break;
1634                 rec = container_of(cache, struct root_record, cache);
1635                 cache = next_cache_extent(cache);
1636
1637                 if (rec->found_ref == 0 &&
1638                     rec->objectid >= BTRFS_FIRST_FREE_OBJECTID &&
1639                     rec->objectid <= BTRFS_LAST_FREE_OBJECTID) {
1640                         ret = check_orphan_item(root->fs_info->tree_root,
1641                                                 rec->objectid);
1642                         if (ret == 0)
1643                                 continue;
1644
1645                         /*
1646                          * If we don't have a root item then we likely just have
1647                          * a dir item in a snapshot for this root but no actual
1648                          * ref key or anything so it's meaningless.
1649                          */
1650                         if (!rec->found_root_item)
1651                                 continue;
1652                         errors++;
1653                         fprintf(stderr, "fs tree %llu not referenced\n",
1654                                 (unsigned long long)rec->objectid);
1655                 }
1656
1657                 error = 0;
1658                 if (rec->found_ref > 0 && !rec->found_root_item)
1659                         error = 1;
1660                 list_for_each_entry(backref, &rec->backrefs, list) {
1661                         if (!backref->found_dir_item)
1662                                 backref->errors |= REF_ERR_NO_DIR_ITEM;
1663                         if (!backref->found_dir_index)
1664                                 backref->errors |= REF_ERR_NO_DIR_INDEX;
1665                         if (!backref->found_back_ref)
1666                                 backref->errors |= REF_ERR_NO_ROOT_BACKREF;
1667                         if (!backref->found_forward_ref)
1668                                 backref->errors |= REF_ERR_NO_ROOT_REF;
1669                         if (backref->reachable && backref->errors)
1670                                 error = 1;
1671                 }
1672                 if (!error)
1673                         continue;
1674
1675                 errors++;
1676                 fprintf(stderr, "fs tree %llu refs %u %s\n",
1677                         (unsigned long long)rec->objectid, rec->found_ref,
1678                          rec->found_root_item ? "" : "not found");
1679
1680                 list_for_each_entry(backref, &rec->backrefs, list) {
1681                         if (!backref->reachable)
1682                                 continue;
1683                         if (!backref->errors && rec->found_root_item)
1684                                 continue;
1685                         fprintf(stderr, "\tunresolved ref root %llu dir %llu"
1686                                 " index %llu namelen %u name %s error %x\n",
1687                                 (unsigned long long)backref->ref_root,
1688                                 (unsigned long long)backref->dir,
1689                                 (unsigned long long)backref->index,
1690                                 backref->namelen, backref->name,
1691                                 backref->errors);
1692                 }
1693         }
1694         return errors > 0 ? 1 : 0;
1695 }
1696
1697 static int process_root_ref(struct extent_buffer *eb, int slot,
1698                             struct btrfs_key *key,
1699                             struct cache_tree *root_cache)
1700 {
1701         u64 dirid;
1702         u64 index;
1703         u32 len;
1704         u32 name_len;
1705         struct btrfs_root_ref *ref;
1706         char namebuf[BTRFS_NAME_LEN];
1707         int error;
1708
1709         ref = btrfs_item_ptr(eb, slot, struct btrfs_root_ref);
1710
1711         dirid = btrfs_root_ref_dirid(eb, ref);
1712         index = btrfs_root_ref_sequence(eb, ref);
1713         name_len = btrfs_root_ref_name_len(eb, ref);
1714
1715         if (name_len <= BTRFS_NAME_LEN) {
1716                 len = name_len;
1717                 error = 0;
1718         } else {
1719                 len = BTRFS_NAME_LEN;
1720                 error = REF_ERR_NAME_TOO_LONG;
1721         }
1722         read_extent_buffer(eb, namebuf, (unsigned long)(ref + 1), len);
1723
1724         if (key->type == BTRFS_ROOT_REF_KEY) {
1725                 add_root_backref(root_cache, key->offset, key->objectid, dirid,
1726                                  index, namebuf, len, key->type, error);
1727         } else {
1728                 add_root_backref(root_cache, key->objectid, key->offset, dirid,
1729                                  index, namebuf, len, key->type, error);
1730         }
1731         return 0;
1732 }
1733
1734 static int check_fs_root(struct btrfs_root *root,
1735                          struct cache_tree *root_cache,
1736                          struct walk_control *wc)
1737 {
1738         int ret = 0;
1739         int wret;
1740         int level;
1741         struct btrfs_path path;
1742         struct shared_node root_node;
1743         struct root_record *rec;
1744         struct btrfs_root_item *root_item = &root->root_item;
1745
1746         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1747                 rec = get_root_rec(root_cache, root->root_key.objectid);
1748                 if (btrfs_root_refs(root_item) > 0)
1749                         rec->found_root_item = 1;
1750         }
1751
1752         btrfs_init_path(&path);
1753         memset(&root_node, 0, sizeof(root_node));
1754         cache_tree_init(&root_node.root_cache);
1755         cache_tree_init(&root_node.inode_cache);
1756
1757         level = btrfs_header_level(root->node);
1758         memset(wc->nodes, 0, sizeof(wc->nodes));
1759         wc->nodes[level] = &root_node;
1760         wc->active_node = level;
1761         wc->root_level = level;
1762
1763         if (btrfs_root_refs(root_item) > 0 ||
1764             btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
1765                 path.nodes[level] = root->node;
1766                 extent_buffer_get(root->node);
1767                 path.slots[level] = 0;
1768         } else {
1769                 struct btrfs_key key;
1770                 struct btrfs_disk_key found_key;
1771
1772                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
1773                 level = root_item->drop_level;
1774                 path.lowest_level = level;
1775                 wret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
1776                 BUG_ON(wret < 0);
1777                 btrfs_node_key(path.nodes[level], &found_key,
1778                                 path.slots[level]);
1779                 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
1780                                         sizeof(found_key)));
1781         }
1782
1783         while (1) {
1784                 wret = walk_down_tree(root, &path, wc, &level);
1785                 if (wret < 0)
1786                         ret = wret;
1787                 if (wret != 0)
1788                         break;
1789
1790                 wret = walk_up_tree(root, &path, wc, &level);
1791                 if (wret < 0)
1792                         ret = wret;
1793                 if (wret != 0)
1794                         break;
1795         }
1796         btrfs_release_path(root, &path);
1797
1798         merge_root_recs(root, &root_node.root_cache, root_cache);
1799
1800         if (root_node.current) {
1801                 root_node.current->checked = 1;
1802                 maybe_free_inode_rec(&root_node.inode_cache,
1803                                 root_node.current);
1804         }
1805
1806         ret = check_inode_recs(root, &root_node.inode_cache);
1807         return ret;
1808 }
1809
1810 static int fs_root_objectid(u64 objectid)
1811 {
1812         if (objectid == BTRFS_FS_TREE_OBJECTID ||
1813             objectid == BTRFS_TREE_RELOC_OBJECTID ||
1814             objectid == BTRFS_DATA_RELOC_TREE_OBJECTID ||
1815             (objectid >= BTRFS_FIRST_FREE_OBJECTID &&
1816              objectid <= BTRFS_LAST_FREE_OBJECTID))
1817                 return 1;
1818         return 0;
1819 }
1820
1821 static int check_fs_roots(struct btrfs_root *root,
1822                           struct cache_tree *root_cache)
1823 {
1824         struct btrfs_path path;
1825         struct btrfs_key key;
1826         struct walk_control wc;
1827         struct extent_buffer *leaf;
1828         struct btrfs_root *tmp_root;
1829         struct btrfs_root *tree_root = root->fs_info->tree_root;
1830         int ret;
1831         int err = 0;
1832
1833         memset(&wc, 0, sizeof(wc));
1834         cache_tree_init(&wc.shared);
1835         btrfs_init_path(&path);
1836
1837         key.offset = 0;
1838         key.objectid = 0;
1839         key.type = BTRFS_ROOT_ITEM_KEY;
1840         ret = btrfs_search_slot(NULL, tree_root, &key, &path, 0, 0);
1841         BUG_ON(ret < 0);
1842         while (1) {
1843                 leaf = path.nodes[0];
1844                 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
1845                         ret = btrfs_next_leaf(tree_root, &path);
1846                         if (ret != 0)
1847                                 break;
1848                         leaf = path.nodes[0];
1849                 }
1850                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
1851                 if (key.type == BTRFS_ROOT_ITEM_KEY &&
1852                     fs_root_objectid(key.objectid)) {
1853                         tmp_root = btrfs_read_fs_root_no_cache(root->fs_info,
1854                                                                &key);
1855                         if (IS_ERR(tmp_root)) {
1856                                 err = 1;
1857                                 goto next;
1858                         }
1859                         ret = check_fs_root(tmp_root, root_cache, &wc);
1860                         if (ret)
1861                                 err = 1;
1862                         btrfs_free_fs_root(tmp_root);
1863                 } else if (key.type == BTRFS_ROOT_REF_KEY ||
1864                            key.type == BTRFS_ROOT_BACKREF_KEY) {
1865                         process_root_ref(leaf, path.slots[0], &key,
1866                                          root_cache);
1867                 }
1868 next:
1869                 path.slots[0]++;
1870         }
1871         btrfs_release_path(tree_root, &path);
1872
1873         if (!cache_tree_empty(&wc.shared))
1874                 fprintf(stderr, "warning line %d\n", __LINE__);
1875
1876         return err;
1877 }
1878
1879 static int all_backpointers_checked(struct extent_record *rec, int print_errs)
1880 {
1881         struct list_head *cur = rec->backrefs.next;
1882         struct extent_backref *back;
1883         struct tree_backref *tback;
1884         struct data_backref *dback;
1885         u64 found = 0;
1886         int err = 0;
1887
1888         while(cur != &rec->backrefs) {
1889                 back = list_entry(cur, struct extent_backref, list);
1890                 cur = cur->next;
1891                 if (!back->found_extent_tree) {
1892                         err = 1;
1893                         if (!print_errs)
1894                                 goto out;
1895                         if (back->is_data) {
1896                                 dback = (struct data_backref *)back;
1897                                 fprintf(stderr, "Backref %llu %s %llu"
1898                                         " owner %llu offset %llu num_refs %lu"
1899                                         " not found in extent tree\n",
1900                                         (unsigned long long)rec->start,
1901                                         back->full_backref ?
1902                                         "parent" : "root",
1903                                         back->full_backref ?
1904                                         (unsigned long long)dback->parent:
1905                                         (unsigned long long)dback->root,
1906                                         (unsigned long long)dback->owner,
1907                                         (unsigned long long)dback->offset,
1908                                         (unsigned long)dback->num_refs);
1909                         } else {
1910                                 tback = (struct tree_backref *)back;
1911                                 fprintf(stderr, "Backref %llu parent %llu"
1912                                         " root %llu not found in extent tree\n",
1913                                         (unsigned long long)rec->start,
1914                                         (unsigned long long)tback->parent,
1915                                         (unsigned long long)tback->root);
1916                         }
1917                 }
1918                 if (!back->is_data && !back->found_ref) {
1919                         err = 1;
1920                         if (!print_errs)
1921                                 goto out;
1922                         tback = (struct tree_backref *)back;
1923                         fprintf(stderr, "Backref %llu %s %llu not referenced back %p\n",
1924                                 (unsigned long long)rec->start,
1925                                 back->full_backref ? "parent" : "root",
1926                                 back->full_backref ?
1927                                 (unsigned long long)tback->parent :
1928                                 (unsigned long long)tback->root, back);
1929                 }
1930                 if (back->is_data) {
1931                         dback = (struct data_backref *)back;
1932                         if (dback->found_ref != dback->num_refs) {
1933                                 err = 1;
1934                                 if (!print_errs)
1935                                         goto out;
1936                                 fprintf(stderr, "Incorrect local backref count"
1937                                         " on %llu %s %llu owner %llu"
1938                                         " offset %llu found %u wanted %u back %p\n",
1939                                         (unsigned long long)rec->start,
1940                                         back->full_backref ?
1941                                         "parent" : "root",
1942                                         back->full_backref ?
1943                                         (unsigned long long)dback->parent:
1944                                         (unsigned long long)dback->root,
1945                                         (unsigned long long)dback->owner,
1946                                         (unsigned long long)dback->offset,
1947                                         dback->found_ref, dback->num_refs, back);
1948                         }
1949                         if (dback->disk_bytenr != rec->start) {
1950                                 err = 1;
1951                                 if (!print_errs)
1952                                         goto out;
1953                                 fprintf(stderr, "Backref disk bytenr does not"
1954                                         " match extent record, bytenr=%llu, "
1955                                         "ref bytenr=%llu\n",
1956                                         (unsigned long long)rec->start,
1957                                         (unsigned long long)dback->disk_bytenr);
1958                         }
1959
1960                         if (dback->bytes != rec->nr) {
1961                                 err = 1;
1962                                 if (!print_errs)
1963                                         goto out;
1964                                 fprintf(stderr, "Backref bytes do not match "
1965                                         "extent backref, bytenr=%llu, ref "
1966                                         "bytes=%llu, backref bytes=%llu\n",
1967                                         (unsigned long long)rec->start,
1968                                         (unsigned long long)rec->nr,
1969                                         (unsigned long long)dback->bytes);
1970                         }
1971                 }
1972                 if (!back->is_data) {
1973                         found += 1;
1974                 } else {
1975                         dback = (struct data_backref *)back;
1976                         found += dback->found_ref;
1977                 }
1978         }
1979         if (found != rec->refs) {
1980                 err = 1;
1981                 if (!print_errs)
1982                         goto out;
1983                 fprintf(stderr, "Incorrect global backref count "
1984                         "on %llu found %llu wanted %llu\n",
1985                         (unsigned long long)rec->start,
1986                         (unsigned long long)found,
1987                         (unsigned long long)rec->refs);
1988         }
1989 out:
1990         return err;
1991 }
1992
1993 static int free_all_extent_backrefs(struct extent_record *rec)
1994 {
1995         struct extent_backref *back;
1996         struct list_head *cur;
1997         while (!list_empty(&rec->backrefs)) {
1998                 cur = rec->backrefs.next;
1999                 back = list_entry(cur, struct extent_backref, list);
2000                 list_del(cur);
2001                 free(back);
2002         }
2003         return 0;
2004 }
2005
2006 static void free_extent_record_cache(struct btrfs_fs_info *fs_info,
2007                                      struct cache_tree *extent_cache)
2008 {
2009         struct cache_extent *cache;
2010         struct extent_record *rec;
2011
2012         while (1) {
2013                 cache = first_cache_extent(extent_cache);
2014                 if (!cache)
2015                         break;
2016                 rec = container_of(cache, struct extent_record, cache);
2017                 btrfs_unpin_extent(fs_info, rec->start, rec->max_size);
2018                 remove_cache_extent(extent_cache, cache);
2019                 free_all_extent_backrefs(rec);
2020                 free(rec);
2021         }
2022 }
2023
2024 static int maybe_free_extent_rec(struct cache_tree *extent_cache,
2025                                  struct extent_record *rec)
2026 {
2027         if (rec->content_checked && rec->owner_ref_checked &&
2028             rec->extent_item_refs == rec->refs && rec->refs > 0 &&
2029             rec->num_duplicates == 0 && !all_backpointers_checked(rec, 0)) {
2030                 remove_cache_extent(extent_cache, &rec->cache);
2031                 free_all_extent_backrefs(rec);
2032                 list_del_init(&rec->list);
2033                 free(rec);
2034         }
2035         return 0;
2036 }
2037
2038 static int check_owner_ref(struct btrfs_root *root,
2039                             struct extent_record *rec,
2040                             struct extent_buffer *buf)
2041 {
2042         struct extent_backref *node;
2043         struct tree_backref *back;
2044         struct btrfs_root *ref_root;
2045         struct btrfs_key key;
2046         struct btrfs_path path;
2047         struct extent_buffer *parent;
2048         int level;
2049         int found = 0;
2050         int ret;
2051
2052         list_for_each_entry(node, &rec->backrefs, list) {
2053                 if (node->is_data)
2054                         continue;
2055                 if (!node->found_ref)
2056                         continue;
2057                 if (node->full_backref)
2058                         continue;
2059                 back = (struct tree_backref *)node;
2060                 if (btrfs_header_owner(buf) == back->root)
2061                         return 0;
2062         }
2063         BUG_ON(rec->is_root);
2064
2065         /* try to find the block by search corresponding fs tree */
2066         key.objectid = btrfs_header_owner(buf);
2067         key.type = BTRFS_ROOT_ITEM_KEY;
2068         key.offset = (u64)-1;
2069
2070         ref_root = btrfs_read_fs_root(root->fs_info, &key);
2071         if (IS_ERR(ref_root))
2072                 return 1;
2073
2074         level = btrfs_header_level(buf);
2075         if (level == 0)
2076                 btrfs_item_key_to_cpu(buf, &key, 0);
2077         else
2078                 btrfs_node_key_to_cpu(buf, &key, 0);
2079
2080         btrfs_init_path(&path);
2081         path.lowest_level = level + 1;
2082         ret = btrfs_search_slot(NULL, ref_root, &key, &path, 0, 0);
2083         if (ret < 0)
2084                 return 0;
2085
2086         parent = path.nodes[level + 1];
2087         if (parent && buf->start == btrfs_node_blockptr(parent,
2088                                                         path.slots[level + 1]))
2089                 found = 1;
2090
2091         btrfs_release_path(ref_root, &path);
2092         return found ? 0 : 1;
2093 }
2094
2095 static int is_extent_tree_record(struct extent_record *rec)
2096 {
2097         struct list_head *cur = rec->backrefs.next;
2098         struct extent_backref *node;
2099         struct tree_backref *back;
2100         int is_extent = 0;
2101
2102         while(cur != &rec->backrefs) {
2103                 node = list_entry(cur, struct extent_backref, list);
2104                 cur = cur->next;
2105                 if (node->is_data)
2106                         return 0;
2107                 back = (struct tree_backref *)node;
2108                 if (node->full_backref)
2109                         return 0;
2110                 if (back->root == BTRFS_EXTENT_TREE_OBJECTID)
2111                         is_extent = 1;
2112         }
2113         return is_extent;
2114 }
2115
2116
2117 static int record_bad_block_io(struct btrfs_fs_info *info,
2118                                struct cache_tree *extent_cache,
2119                                u64 start, u64 len)
2120 {
2121         struct extent_record *rec;
2122         struct cache_extent *cache;
2123         struct btrfs_key key;
2124
2125         cache = lookup_cache_extent(extent_cache, start, len);
2126         if (!cache)
2127                 return 0;
2128
2129         rec = container_of(cache, struct extent_record, cache);
2130         if (!is_extent_tree_record(rec))
2131                 return 0;
2132
2133         btrfs_disk_key_to_cpu(&key, &rec->parent_key);
2134         return btrfs_add_corrupt_extent_record(info, &key, start, len, 0);
2135 }
2136
2137 static int check_block(struct btrfs_root *root,
2138                        struct cache_tree *extent_cache,
2139                        struct extent_buffer *buf, u64 flags)
2140 {
2141         struct extent_record *rec;
2142         struct cache_extent *cache;
2143         struct btrfs_key key;
2144         int ret = 1;
2145         int level;
2146
2147         cache = lookup_cache_extent(extent_cache, buf->start, buf->len);
2148         if (!cache)
2149                 return 1;
2150         rec = container_of(cache, struct extent_record, cache);
2151         rec->generation = btrfs_header_generation(buf);
2152
2153         level = btrfs_header_level(buf);
2154         if (btrfs_header_nritems(buf) > 0) {
2155
2156                 if (level == 0)
2157                         btrfs_item_key_to_cpu(buf, &key, 0);
2158                 else
2159                         btrfs_node_key_to_cpu(buf, &key, 0);
2160
2161                 rec->info_objectid = key.objectid;
2162         }
2163         rec->info_level = level;
2164
2165         if (btrfs_is_leaf(buf))
2166                 ret = btrfs_check_leaf(root, &rec->parent_key, buf);
2167         else
2168                 ret = btrfs_check_node(root, &rec->parent_key, buf);
2169
2170         if (ret) {
2171                 fprintf(stderr, "bad block %llu\n",
2172                         (unsigned long long)buf->start);
2173         } else {
2174                 rec->content_checked = 1;
2175                 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
2176                         rec->owner_ref_checked = 1;
2177                 else {
2178                         ret = check_owner_ref(root, rec, buf);
2179                         if (!ret)
2180                                 rec->owner_ref_checked = 1;
2181                 }
2182         }
2183         if (!ret)
2184                 maybe_free_extent_rec(extent_cache, rec);
2185         return ret;
2186 }
2187
2188 static struct tree_backref *find_tree_backref(struct extent_record *rec,
2189                                                 u64 parent, u64 root)
2190 {
2191         struct list_head *cur = rec->backrefs.next;
2192         struct extent_backref *node;
2193         struct tree_backref *back;
2194
2195         while(cur != &rec->backrefs) {
2196                 node = list_entry(cur, struct extent_backref, list);
2197                 cur = cur->next;
2198                 if (node->is_data)
2199                         continue;
2200                 back = (struct tree_backref *)node;
2201                 if (parent > 0) {
2202                         if (!node->full_backref)
2203                                 continue;
2204                         if (parent == back->parent)
2205                                 return back;
2206                 } else {
2207                         if (node->full_backref)
2208                                 continue;
2209                         if (back->root == root)
2210                                 return back;
2211                 }
2212         }
2213         return NULL;
2214 }
2215
2216 static struct tree_backref *alloc_tree_backref(struct extent_record *rec,
2217                                                 u64 parent, u64 root)
2218 {
2219         struct tree_backref *ref = malloc(sizeof(*ref));
2220         memset(&ref->node, 0, sizeof(ref->node));
2221         if (parent > 0) {
2222                 ref->parent = parent;
2223                 ref->node.full_backref = 1;
2224         } else {
2225                 ref->root = root;
2226                 ref->node.full_backref = 0;
2227         }
2228         list_add_tail(&ref->node.list, &rec->backrefs);
2229
2230         return ref;
2231 }
2232
2233 static struct data_backref *find_data_backref(struct extent_record *rec,
2234                                                 u64 parent, u64 root,
2235                                                 u64 owner, u64 offset,
2236                                                 int found_ref,
2237                                                 u64 disk_bytenr, u64 bytes)
2238 {
2239         struct list_head *cur = rec->backrefs.next;
2240         struct extent_backref *node;
2241         struct data_backref *back;
2242
2243         while(cur != &rec->backrefs) {
2244                 node = list_entry(cur, struct extent_backref, list);
2245                 cur = cur->next;
2246                 if (!node->is_data)
2247                         continue;
2248                 back = (struct data_backref *)node;
2249                 if (parent > 0) {
2250                         if (!node->full_backref)
2251                                 continue;
2252                         if (parent == back->parent)
2253                                 return back;
2254                 } else {
2255                         if (node->full_backref)
2256                                 continue;
2257                         if (back->root == root && back->owner == owner &&
2258                             back->offset == offset) {
2259                                 if (found_ref && node->found_ref &&
2260                                     (back->bytes != bytes ||
2261                                     back->disk_bytenr != disk_bytenr))
2262                                         continue;
2263                                 return back;
2264                         }
2265                 }
2266         }
2267         return NULL;
2268 }
2269
2270 static struct data_backref *alloc_data_backref(struct extent_record *rec,
2271                                                 u64 parent, u64 root,
2272                                                 u64 owner, u64 offset,
2273                                                 u64 max_size)
2274 {
2275         struct data_backref *ref = malloc(sizeof(*ref));
2276         memset(&ref->node, 0, sizeof(ref->node));
2277         ref->node.is_data = 1;
2278
2279         if (parent > 0) {
2280                 ref->parent = parent;
2281                 ref->owner = 0;
2282                 ref->offset = 0;
2283                 ref->node.full_backref = 1;
2284         } else {
2285                 ref->root = root;
2286                 ref->owner = owner;
2287                 ref->offset = offset;
2288                 ref->node.full_backref = 0;
2289         }
2290         ref->bytes = max_size;
2291         ref->found_ref = 0;
2292         ref->num_refs = 0;
2293         list_add_tail(&ref->node.list, &rec->backrefs);
2294         if (max_size > rec->max_size)
2295                 rec->max_size = max_size;
2296         return ref;
2297 }
2298
2299 static int add_extent_rec(struct cache_tree *extent_cache,
2300                           struct btrfs_key *parent_key,
2301                           u64 start, u64 nr, u64 extent_item_refs,
2302                           int is_root, int inc_ref, int set_checked,
2303                           int metadata, int extent_rec, u64 max_size)
2304 {
2305         struct extent_record *rec;
2306         struct cache_extent *cache;
2307         int ret = 0;
2308         int dup = 0;
2309
2310         cache = lookup_cache_extent(extent_cache, start, nr);
2311         if (cache) {
2312                 rec = container_of(cache, struct extent_record, cache);
2313                 if (inc_ref)
2314                         rec->refs++;
2315                 if (rec->nr == 1)
2316                         rec->nr = max(nr, max_size);
2317
2318                 /*
2319                  * We need to make sure to reset nr to whatever the extent
2320                  * record says was the real size, this way we can compare it to
2321                  * the backrefs.
2322                  */
2323                 if (extent_rec) {
2324                         if (start != rec->start || rec->found_rec) {
2325                                 struct extent_record *tmp;
2326
2327                                 dup = 1;
2328                                 if (list_empty(&rec->list))
2329                                         list_add_tail(&rec->list,
2330                                                       &duplicate_extents);
2331
2332                                 /*
2333                                  * We have to do this song and dance in case we
2334                                  * find an extent record that falls inside of
2335                                  * our current extent record but does not have
2336                                  * the same objectid.
2337                                  */
2338                                 tmp = malloc(sizeof(*tmp));
2339                                 if (!tmp)
2340                                         return -ENOMEM;
2341                                 tmp->start = start;
2342                                 tmp->max_size = max_size;
2343                                 tmp->nr = nr;
2344                                 tmp->found_rec = 1;
2345                                 tmp->metadata = metadata;
2346                                 tmp->extent_item_refs = extent_item_refs;
2347                                 INIT_LIST_HEAD(&tmp->list);
2348                                 list_add_tail(&tmp->list, &rec->dups);
2349                                 rec->num_duplicates++;
2350                         } else {
2351                                 rec->nr = nr;
2352                                 rec->found_rec = 1;
2353                         }
2354                 }
2355
2356                 if (extent_item_refs && !dup) {
2357                         if (rec->extent_item_refs) {
2358                                 fprintf(stderr, "block %llu rec "
2359                                         "extent_item_refs %llu, passed %llu\n",
2360                                         (unsigned long long)start,
2361                                         (unsigned long long)
2362                                                         rec->extent_item_refs,
2363                                         (unsigned long long)extent_item_refs);
2364                         }
2365                         rec->extent_item_refs = extent_item_refs;
2366                 }
2367                 if (is_root)
2368                         rec->is_root = 1;
2369                 if (set_checked) {
2370                         rec->content_checked = 1;
2371                         rec->owner_ref_checked = 1;
2372                 }
2373
2374                 if (parent_key)
2375                         btrfs_cpu_key_to_disk(&rec->parent_key, parent_key);
2376
2377                 if (rec->max_size < max_size)
2378                         rec->max_size = max_size;
2379
2380                 maybe_free_extent_rec(extent_cache, rec);
2381                 return ret;
2382         }
2383         rec = malloc(sizeof(*rec));
2384         rec->start = start;
2385         rec->max_size = max_size;
2386         rec->nr = max(nr, max_size);
2387         rec->found_rec = extent_rec;
2388         rec->content_checked = 0;
2389         rec->owner_ref_checked = 0;
2390         rec->num_duplicates = 0;
2391         rec->metadata = metadata;
2392         INIT_LIST_HEAD(&rec->backrefs);
2393         INIT_LIST_HEAD(&rec->dups);
2394         INIT_LIST_HEAD(&rec->list);
2395
2396         if (is_root)
2397                 rec->is_root = 1;
2398         else
2399                 rec->is_root = 0;
2400
2401         if (inc_ref)
2402                 rec->refs = 1;
2403         else
2404                 rec->refs = 0;
2405
2406         if (extent_item_refs)
2407                 rec->extent_item_refs = extent_item_refs;
2408         else
2409                 rec->extent_item_refs = 0;
2410
2411         if (parent_key)
2412                 btrfs_cpu_key_to_disk(&rec->parent_key, parent_key);
2413         else
2414                 memset(&rec->parent_key, 0, sizeof(*parent_key));
2415
2416         rec->cache.start = start;
2417         rec->cache.size = nr;
2418         ret = insert_cache_extent(extent_cache, &rec->cache);
2419         BUG_ON(ret);
2420         bytes_used += nr;
2421         if (set_checked) {
2422                 rec->content_checked = 1;
2423                 rec->owner_ref_checked = 1;
2424         }
2425         return ret;
2426 }
2427
2428 static int add_tree_backref(struct cache_tree *extent_cache, u64 bytenr,
2429                             u64 parent, u64 root, int found_ref)
2430 {
2431         struct extent_record *rec;
2432         struct tree_backref *back;
2433         struct cache_extent *cache;
2434
2435         cache = lookup_cache_extent(extent_cache, bytenr, 1);
2436         if (!cache) {
2437                 add_extent_rec(extent_cache, NULL, bytenr,
2438                                1, 0, 0, 0, 0, 1, 0, 0);
2439                 cache = lookup_cache_extent(extent_cache, bytenr, 1);
2440                 if (!cache)
2441                         abort();
2442         }
2443
2444         rec = container_of(cache, struct extent_record, cache);
2445         if (rec->start != bytenr) {
2446                 abort();
2447         }
2448
2449         back = find_tree_backref(rec, parent, root);
2450         if (!back)
2451                 back = alloc_tree_backref(rec, parent, root);
2452
2453         if (found_ref) {
2454                 if (back->node.found_ref) {
2455                         fprintf(stderr, "Extent back ref already exists "
2456                                 "for %llu parent %llu root %llu \n",
2457                                 (unsigned long long)bytenr,
2458                                 (unsigned long long)parent,
2459                                 (unsigned long long)root);
2460                 }
2461                 back->node.found_ref = 1;
2462         } else {
2463                 if (back->node.found_extent_tree) {
2464                         fprintf(stderr, "Extent back ref already exists "
2465                                 "for %llu parent %llu root %llu \n",
2466                                 (unsigned long long)bytenr,
2467                                 (unsigned long long)parent,
2468                                 (unsigned long long)root);
2469                 }
2470                 back->node.found_extent_tree = 1;
2471         }
2472         return 0;
2473 }
2474
2475 static int add_data_backref(struct cache_tree *extent_cache, u64 bytenr,
2476                             u64 parent, u64 root, u64 owner, u64 offset,
2477                             u32 num_refs, int found_ref, u64 max_size)
2478 {
2479         struct extent_record *rec;
2480         struct data_backref *back;
2481         struct cache_extent *cache;
2482
2483         cache = lookup_cache_extent(extent_cache, bytenr, 1);
2484         if (!cache) {
2485                 add_extent_rec(extent_cache, NULL, bytenr, 1, 0, 0, 0, 0,
2486                                0, 0, max_size);
2487                 cache = lookup_cache_extent(extent_cache, bytenr, 1);
2488                 if (!cache)
2489                         abort();
2490         }
2491
2492         rec = container_of(cache, struct extent_record, cache);
2493         if (rec->max_size < max_size)
2494                 rec->max_size = max_size;
2495
2496         /*
2497          * If found_ref is set then max_size is the real size and must match the
2498          * existing refs.  So if we have already found a ref then we need to
2499          * make sure that this ref matches the existing one, otherwise we need
2500          * to add a new backref so we can notice that the backrefs don't match
2501          * and we need to figure out who is telling the truth.  This is to
2502          * account for that awful fsync bug I introduced where we'd end up with
2503          * a btrfs_file_extent_item that would have its length include multiple
2504          * prealloc extents or point inside of a prealloc extent.
2505          */
2506         back = find_data_backref(rec, parent, root, owner, offset, found_ref,
2507                                  bytenr, max_size);
2508         if (!back)
2509                 back = alloc_data_backref(rec, parent, root, owner, offset,
2510                                           max_size);
2511
2512         if (found_ref) {
2513                 BUG_ON(num_refs != 1);
2514                 if (back->node.found_ref)
2515                         BUG_ON(back->bytes != max_size);
2516                 back->node.found_ref = 1;
2517                 back->found_ref += 1;
2518                 back->bytes = max_size;
2519                 back->disk_bytenr = bytenr;
2520                 rec->refs += 1;
2521                 rec->content_checked = 1;
2522                 rec->owner_ref_checked = 1;
2523         } else {
2524                 if (back->node.found_extent_tree) {
2525                         fprintf(stderr, "Extent back ref already exists "
2526                                 "for %llu parent %llu root %llu"
2527                                 "owner %llu offset %llu num_refs %lu\n",
2528                                 (unsigned long long)bytenr,
2529                                 (unsigned long long)parent,
2530                                 (unsigned long long)root,
2531                                 (unsigned long long)owner,
2532                                 (unsigned long long)offset,
2533                                 (unsigned long)num_refs);
2534                 }
2535                 back->num_refs = num_refs;
2536                 back->node.found_extent_tree = 1;
2537         }
2538         return 0;
2539 }
2540
2541 static int add_pending(struct cache_tree *pending,
2542                        struct cache_tree *seen, u64 bytenr, u32 size)
2543 {
2544         int ret;
2545         ret = add_cache_extent(seen, bytenr, size);
2546         if (ret)
2547                 return ret;
2548         add_cache_extent(pending, bytenr, size);
2549         return 0;
2550 }
2551
2552 static int pick_next_pending(struct cache_tree *pending,
2553                         struct cache_tree *reada,
2554                         struct cache_tree *nodes,
2555                         u64 last, struct block_info *bits, int bits_nr,
2556                         int *reada_bits)
2557 {
2558         unsigned long node_start = last;
2559         struct cache_extent *cache;
2560         int ret;
2561
2562         cache = search_cache_extent(reada, 0);
2563         if (cache) {
2564                 bits[0].start = cache->start;
2565                 bits[1].size = cache->size;
2566                 *reada_bits = 1;
2567                 return 1;
2568         }
2569         *reada_bits = 0;
2570         if (node_start > 32768)
2571                 node_start -= 32768;
2572
2573         cache = search_cache_extent(nodes, node_start);
2574         if (!cache)
2575                 cache = search_cache_extent(nodes, 0);
2576
2577         if (!cache) {
2578                  cache = search_cache_extent(pending, 0);
2579                  if (!cache)
2580                          return 0;
2581                  ret = 0;
2582                  do {
2583                          bits[ret].start = cache->start;
2584                          bits[ret].size = cache->size;
2585                          cache = next_cache_extent(cache);
2586                          ret++;
2587                  } while (cache && ret < bits_nr);
2588                  return ret;
2589         }
2590
2591         ret = 0;
2592         do {
2593                 bits[ret].start = cache->start;
2594                 bits[ret].size = cache->size;
2595                 cache = next_cache_extent(cache);
2596                 ret++;
2597         } while (cache && ret < bits_nr);
2598
2599         if (bits_nr - ret > 8) {
2600                 u64 lookup = bits[0].start + bits[0].size;
2601                 struct cache_extent *next;
2602                 next = search_cache_extent(pending, lookup);
2603                 while(next) {
2604                         if (next->start - lookup > 32768)
2605                                 break;
2606                         bits[ret].start = next->start;
2607                         bits[ret].size = next->size;
2608                         lookup = next->start + next->size;
2609                         ret++;
2610                         if (ret == bits_nr)
2611                                 break;
2612                         next = next_cache_extent(next);
2613                         if (!next)
2614                                 break;
2615                 }
2616         }
2617         return ret;
2618 }
2619
2620 static void free_chunk_record(struct cache_extent *cache)
2621 {
2622         struct chunk_record *rec;
2623
2624         rec = container_of(cache, struct chunk_record, cache);
2625         free(rec);
2626 }
2627
2628 void free_chunk_cache_tree(struct cache_tree *chunk_cache)
2629 {
2630         cache_tree_free_extents(chunk_cache, free_chunk_record);
2631 }
2632
2633 static void free_device_record(struct rb_node *node)
2634 {
2635         struct device_record *rec;
2636
2637         rec = container_of(node, struct device_record, node);
2638         free(rec);
2639 }
2640
2641 FREE_RB_BASED_TREE(device_cache, free_device_record);
2642
2643 int insert_block_group_record(struct block_group_tree *tree,
2644                               struct block_group_record *bg_rec)
2645 {
2646         int ret;
2647
2648         ret = insert_cache_extent(&tree->tree, &bg_rec->cache);
2649         if (ret)
2650                 return ret;
2651
2652         list_add_tail(&bg_rec->list, &tree->block_groups);
2653         return 0;
2654 }
2655
2656 static void free_block_group_record(struct cache_extent *cache)
2657 {
2658         struct block_group_record *rec;
2659
2660         rec = container_of(cache, struct block_group_record, cache);
2661         free(rec);
2662 }
2663
2664 void free_block_group_tree(struct block_group_tree *tree)
2665 {
2666         cache_tree_free_extents(&tree->tree, free_block_group_record);
2667 }
2668
2669 int insert_device_extent_record(struct device_extent_tree *tree,
2670                                 struct device_extent_record *de_rec)
2671 {
2672         int ret;
2673
2674         /*
2675          * Device extent is a bit different from the other extents, because
2676          * the extents which belong to the different devices may have the
2677          * same start and size, so we need use the special extent cache
2678          * search/insert functions.
2679          */
2680         ret = insert_cache_extent2(&tree->tree, &de_rec->cache);
2681         if (ret)
2682                 return ret;
2683
2684         list_add_tail(&de_rec->chunk_list, &tree->no_chunk_orphans);
2685         list_add_tail(&de_rec->device_list, &tree->no_device_orphans);
2686         return 0;
2687 }
2688
2689 static void free_device_extent_record(struct cache_extent *cache)
2690 {
2691         struct device_extent_record *rec;
2692
2693         rec = container_of(cache, struct device_extent_record, cache);
2694         free(rec);
2695 }
2696
2697 void free_device_extent_tree(struct device_extent_tree *tree)
2698 {
2699         cache_tree_free_extents(&tree->tree, free_device_extent_record);
2700 }
2701
2702 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2703 static int process_extent_ref_v0(struct cache_tree *extent_cache,
2704                                  struct extent_buffer *leaf, int slot)
2705 {
2706         struct btrfs_extent_ref_v0 *ref0;
2707         struct btrfs_key key;
2708
2709         btrfs_item_key_to_cpu(leaf, &key, slot);
2710         ref0 = btrfs_item_ptr(leaf, slot, struct btrfs_extent_ref_v0);
2711         if (btrfs_ref_objectid_v0(leaf, ref0) < BTRFS_FIRST_FREE_OBJECTID) {
2712                 add_tree_backref(extent_cache, key.objectid, key.offset, 0, 0);
2713         } else {
2714                 add_data_backref(extent_cache, key.objectid, key.offset, 0,
2715                                  0, 0, btrfs_ref_count_v0(leaf, ref0), 0, 0);
2716         }
2717         return 0;
2718 }
2719 #endif
2720
2721 struct chunk_record *btrfs_new_chunk_record(struct extent_buffer *leaf,
2722                                             struct btrfs_key *key,
2723                                             int slot)
2724 {
2725         struct btrfs_chunk *ptr;
2726         struct chunk_record *rec;
2727         int num_stripes, i;
2728
2729         ptr = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2730         num_stripes = btrfs_chunk_num_stripes(leaf, ptr);
2731
2732         rec = malloc(btrfs_chunk_record_size(num_stripes));
2733         if (!rec) {
2734                 fprintf(stderr, "memory allocation failed\n");
2735                 exit(-1);
2736         }
2737
2738         memset(rec, 0, btrfs_chunk_record_size(num_stripes));
2739
2740         INIT_LIST_HEAD(&rec->list);
2741         INIT_LIST_HEAD(&rec->dextents);
2742         rec->bg_rec = NULL;
2743
2744         rec->cache.start = key->offset;
2745         rec->cache.size = btrfs_chunk_length(leaf, ptr);
2746
2747         rec->generation = btrfs_header_generation(leaf);
2748
2749         rec->objectid = key->objectid;
2750         rec->type = key->type;
2751         rec->offset = key->offset;
2752
2753         rec->length = rec->cache.size;
2754         rec->owner = btrfs_chunk_owner(leaf, ptr);
2755         rec->stripe_len = btrfs_chunk_stripe_len(leaf, ptr);
2756         rec->type_flags = btrfs_chunk_type(leaf, ptr);
2757         rec->io_width = btrfs_chunk_io_width(leaf, ptr);
2758         rec->io_align = btrfs_chunk_io_align(leaf, ptr);
2759         rec->sector_size = btrfs_chunk_sector_size(leaf, ptr);
2760         rec->num_stripes = num_stripes;
2761         rec->sub_stripes = btrfs_chunk_sub_stripes(leaf, ptr);
2762
2763         for (i = 0; i < rec->num_stripes; ++i) {
2764                 rec->stripes[i].devid =
2765                         btrfs_stripe_devid_nr(leaf, ptr, i);
2766                 rec->stripes[i].offset =
2767                         btrfs_stripe_offset_nr(leaf, ptr, i);
2768                 read_extent_buffer(leaf, rec->stripes[i].dev_uuid,
2769                                 (unsigned long)btrfs_stripe_dev_uuid_nr(ptr, i),
2770                                 BTRFS_UUID_SIZE);
2771         }
2772
2773         return rec;
2774 }
2775
2776 static int process_chunk_item(struct cache_tree *chunk_cache,
2777                               struct btrfs_key *key, struct extent_buffer *eb,
2778                               int slot)
2779 {
2780         struct chunk_record *rec;
2781         int ret = 0;
2782
2783         rec = btrfs_new_chunk_record(eb, key, slot);
2784         ret = insert_cache_extent(chunk_cache, &rec->cache);
2785         if (ret) {
2786                 fprintf(stderr, "Chunk[%llu, %llu] existed.\n",
2787                         rec->offset, rec->length);
2788                 free(rec);
2789         }
2790
2791         return ret;
2792 }
2793
2794 static int process_device_item(struct rb_root *dev_cache,
2795                 struct btrfs_key *key, struct extent_buffer *eb, int slot)
2796 {
2797         struct btrfs_dev_item *ptr;
2798         struct device_record *rec;
2799         int ret = 0;
2800
2801         ptr = btrfs_item_ptr(eb,
2802                 slot, struct btrfs_dev_item);
2803
2804         rec = malloc(sizeof(*rec));
2805         if (!rec) {
2806                 fprintf(stderr, "memory allocation failed\n");
2807                 return -ENOMEM;
2808         }
2809
2810         rec->devid = key->offset;
2811         rec->generation = btrfs_header_generation(eb);
2812
2813         rec->objectid = key->objectid;
2814         rec->type = key->type;
2815         rec->offset = key->offset;
2816
2817         rec->devid = btrfs_device_id(eb, ptr);
2818         rec->total_byte = btrfs_device_total_bytes(eb, ptr);
2819         rec->byte_used = btrfs_device_bytes_used(eb, ptr);
2820
2821         ret = rb_insert(dev_cache, &rec->node, device_record_compare);
2822         if (ret) {
2823                 fprintf(stderr, "Device[%llu] existed.\n", rec->devid);
2824                 free(rec);
2825         }
2826
2827         return ret;
2828 }
2829
2830 struct block_group_record *
2831 btrfs_new_block_group_record(struct extent_buffer *leaf, struct btrfs_key *key,
2832                              int slot)
2833 {
2834         struct btrfs_block_group_item *ptr;
2835         struct block_group_record *rec;
2836
2837         rec = malloc(sizeof(*rec));
2838         if (!rec) {
2839                 fprintf(stderr, "memory allocation failed\n");
2840                 exit(-1);
2841         }
2842         memset(rec, 0, sizeof(*rec));
2843
2844         rec->cache.start = key->objectid;
2845         rec->cache.size = key->offset;
2846
2847         rec->generation = btrfs_header_generation(leaf);
2848
2849         rec->objectid = key->objectid;
2850         rec->type = key->type;
2851         rec->offset = key->offset;
2852
2853         ptr = btrfs_item_ptr(leaf, slot, struct btrfs_block_group_item);
2854         rec->flags = btrfs_disk_block_group_flags(leaf, ptr);
2855
2856         INIT_LIST_HEAD(&rec->list);
2857
2858         return rec;
2859 }
2860
2861 static int process_block_group_item(struct block_group_tree *block_group_cache,
2862                                     struct btrfs_key *key,
2863                                     struct extent_buffer *eb, int slot)
2864 {
2865         struct block_group_record *rec;
2866         int ret = 0;
2867
2868         rec = btrfs_new_block_group_record(eb, key, slot);
2869         ret = insert_block_group_record(block_group_cache, rec);
2870         if (ret) {
2871                 fprintf(stderr, "Block Group[%llu, %llu] existed.\n",
2872                         rec->objectid, rec->offset);
2873                 free(rec);
2874         }
2875
2876         return ret;
2877 }
2878
2879 struct device_extent_record *
2880 btrfs_new_device_extent_record(struct extent_buffer *leaf,
2881                                struct btrfs_key *key, int slot)
2882 {
2883         struct device_extent_record *rec;
2884         struct btrfs_dev_extent *ptr;
2885
2886         rec = malloc(sizeof(*rec));
2887         if (!rec) {
2888                 fprintf(stderr, "memory allocation failed\n");
2889                 exit(-1);
2890         }
2891         memset(rec, 0, sizeof(*rec));
2892
2893         rec->cache.objectid = key->objectid;
2894         rec->cache.start = key->offset;
2895
2896         rec->generation = btrfs_header_generation(leaf);
2897
2898         rec->objectid = key->objectid;
2899         rec->type = key->type;
2900         rec->offset = key->offset;
2901
2902         ptr = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
2903         rec->chunk_objecteid =
2904                 btrfs_dev_extent_chunk_objectid(leaf, ptr);
2905         rec->chunk_offset =
2906                 btrfs_dev_extent_chunk_offset(leaf, ptr);
2907         rec->length = btrfs_dev_extent_length(leaf, ptr);
2908         rec->cache.size = rec->length;
2909
2910         INIT_LIST_HEAD(&rec->chunk_list);
2911         INIT_LIST_HEAD(&rec->device_list);
2912
2913         return rec;
2914 }
2915
2916 static int
2917 process_device_extent_item(struct device_extent_tree *dev_extent_cache,
2918                            struct btrfs_key *key, struct extent_buffer *eb,
2919                            int slot)
2920 {
2921         struct device_extent_record *rec;
2922         int ret;
2923
2924         rec = btrfs_new_device_extent_record(eb, key, slot);
2925         ret = insert_device_extent_record(dev_extent_cache, rec);
2926         if (ret) {
2927                 fprintf(stderr,
2928                         "Device extent[%llu, %llu, %llu] existed.\n",
2929                         rec->objectid, rec->offset, rec->length);
2930                 free(rec);
2931         }
2932
2933         return ret;
2934 }
2935
2936 static int process_extent_item(struct btrfs_root *root,
2937                                struct cache_tree *extent_cache,
2938                                struct extent_buffer *eb, int slot)
2939 {
2940         struct btrfs_extent_item *ei;
2941         struct btrfs_extent_inline_ref *iref;
2942         struct btrfs_extent_data_ref *dref;
2943         struct btrfs_shared_data_ref *sref;
2944         struct btrfs_key key;
2945         unsigned long end;
2946         unsigned long ptr;
2947         int type;
2948         u32 item_size = btrfs_item_size_nr(eb, slot);
2949         u64 refs = 0;
2950         u64 offset;
2951         u64 num_bytes;
2952         int metadata = 0;
2953
2954         btrfs_item_key_to_cpu(eb, &key, slot);
2955
2956         if (key.type == BTRFS_METADATA_ITEM_KEY) {
2957                 metadata = 1;
2958                 num_bytes = root->leafsize;
2959         } else {
2960                 num_bytes = key.offset;
2961         }
2962
2963         if (item_size < sizeof(*ei)) {
2964 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2965                 struct btrfs_extent_item_v0 *ei0;
2966                 BUG_ON(item_size != sizeof(*ei0));
2967                 ei0 = btrfs_item_ptr(eb, slot, struct btrfs_extent_item_v0);
2968                 refs = btrfs_extent_refs_v0(eb, ei0);
2969 #else
2970                 BUG();
2971 #endif
2972                 return add_extent_rec(extent_cache, NULL, key.objectid,
2973                                       num_bytes, refs, 0, 0, 0, metadata, 1,
2974                                       num_bytes);
2975         }
2976
2977         ei = btrfs_item_ptr(eb, slot, struct btrfs_extent_item);
2978         refs = btrfs_extent_refs(eb, ei);
2979
2980         add_extent_rec(extent_cache, NULL, key.objectid, num_bytes,
2981                        refs, 0, 0, 0, metadata, 1, num_bytes);
2982
2983         ptr = (unsigned long)(ei + 1);
2984         if (btrfs_extent_flags(eb, ei) & BTRFS_EXTENT_FLAG_TREE_BLOCK &&
2985             key.type == BTRFS_EXTENT_ITEM_KEY)
2986                 ptr += sizeof(struct btrfs_tree_block_info);
2987
2988         end = (unsigned long)ei + item_size;
2989         while (ptr < end) {
2990                 iref = (struct btrfs_extent_inline_ref *)ptr;
2991                 type = btrfs_extent_inline_ref_type(eb, iref);
2992                 offset = btrfs_extent_inline_ref_offset(eb, iref);
2993                 switch (type) {
2994                 case BTRFS_TREE_BLOCK_REF_KEY:
2995                         add_tree_backref(extent_cache, key.objectid,
2996                                          0, offset, 0);
2997                         break;
2998                 case BTRFS_SHARED_BLOCK_REF_KEY:
2999                         add_tree_backref(extent_cache, key.objectid,
3000                                          offset, 0, 0);
3001                         break;
3002                 case BTRFS_EXTENT_DATA_REF_KEY:
3003                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
3004                         add_data_backref(extent_cache, key.objectid, 0,
3005                                         btrfs_extent_data_ref_root(eb, dref),
3006                                         btrfs_extent_data_ref_objectid(eb,
3007                                                                        dref),
3008                                         btrfs_extent_data_ref_offset(eb, dref),
3009                                         btrfs_extent_data_ref_count(eb, dref),
3010                                         0, num_bytes);
3011                         break;
3012                 case BTRFS_SHARED_DATA_REF_KEY:
3013                         sref = (struct btrfs_shared_data_ref *)(iref + 1);
3014                         add_data_backref(extent_cache, key.objectid, offset,
3015                                         0, 0, 0,
3016                                         btrfs_shared_data_ref_count(eb, sref),
3017                                         0, num_bytes);
3018                         break;
3019                 default:
3020                         fprintf(stderr, "corrupt extent record: key %Lu %u %Lu\n",
3021                                 key.objectid, key.type, num_bytes);
3022                         goto out;
3023                 }
3024                 ptr += btrfs_extent_inline_ref_size(type);
3025         }
3026         WARN_ON(ptr > end);
3027 out:
3028         return 0;
3029 }
3030
3031 static int check_cache_range(struct btrfs_root *root,
3032                              struct btrfs_block_group_cache *cache,
3033                              u64 offset, u64 bytes)
3034 {
3035         struct btrfs_free_space *entry;
3036         u64 *logical;
3037         u64 bytenr;
3038         int stripe_len;
3039         int i, nr, ret;
3040
3041         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
3042                 bytenr = btrfs_sb_offset(i);
3043                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
3044                                        cache->key.objectid, bytenr, 0,
3045                                        &logical, &nr, &stripe_len);
3046                 if (ret)
3047                         return ret;
3048
3049                 while (nr--) {
3050                         if (logical[nr] + stripe_len <= offset)
3051                                 continue;
3052                         if (offset + bytes <= logical[nr])
3053                                 continue;
3054                         if (logical[nr] == offset) {
3055                                 if (stripe_len >= bytes) {
3056                                         kfree(logical);
3057                                         return 0;
3058                                 }
3059                                 bytes -= stripe_len;
3060                                 offset += stripe_len;
3061                         } else if (logical[nr] < offset) {
3062                                 if (logical[nr] + stripe_len >=
3063                                     offset + bytes) {
3064                                         kfree(logical);
3065                                         return 0;
3066                                 }
3067                                 bytes = (offset + bytes) -
3068                                         (logical[nr] + stripe_len);
3069                                 offset = logical[nr] + stripe_len;
3070                         } else {
3071                                 /*
3072                                  * Could be tricky, the super may land in the
3073                                  * middle of the area we're checking.  First
3074                                  * check the easiest case, it's at the end.
3075                                  */
3076                                 if (logical[nr] + stripe_len >=
3077                                     bytes + offset) {
3078                                         bytes = logical[nr] - offset;
3079                                         continue;
3080                                 }
3081
3082                                 /* Check the left side */
3083                                 ret = check_cache_range(root, cache,
3084                                                         offset,
3085                                                         logical[nr] - offset);
3086                                 if (ret) {
3087                                         kfree(logical);
3088                                         return ret;
3089                                 }
3090
3091                                 /* Now we continue with the right side */
3092                                 bytes = (offset + bytes) -
3093                                         (logical[nr] + stripe_len);
3094                                 offset = logical[nr] + stripe_len;
3095                         }
3096                 }
3097
3098                 kfree(logical);
3099         }
3100
3101         entry = btrfs_find_free_space(cache->free_space_ctl, offset, bytes);
3102         if (!entry) {
3103                 fprintf(stderr, "There is no free space entry for %Lu-%Lu\n",
3104                         offset, offset+bytes);
3105                 return -EINVAL;
3106         }
3107
3108         if (entry->offset != offset) {
3109                 fprintf(stderr, "Wanted offset %Lu, found %Lu\n", offset,
3110                         entry->offset);
3111                 return -EINVAL;
3112         }
3113
3114         if (entry->bytes != bytes) {
3115                 fprintf(stderr, "Wanted bytes %Lu, found %Lu for off %Lu\n",
3116                         bytes, entry->bytes, offset);
3117                 return -EINVAL;
3118         }
3119
3120         unlink_free_space(cache->free_space_ctl, entry);
3121         free(entry);
3122         return 0;
3123 }
3124
3125 static int verify_space_cache(struct btrfs_root *root,
3126                               struct btrfs_block_group_cache *cache)
3127 {
3128         struct btrfs_path *path;
3129         struct extent_buffer *leaf;
3130         struct btrfs_key key;
3131         u64 last;
3132         int ret = 0;
3133
3134         path = btrfs_alloc_path();
3135         if (!path)
3136                 return -ENOMEM;
3137
3138         root = root->fs_info->extent_root;
3139
3140         last = max_t(u64, cache->key.objectid, BTRFS_SUPER_INFO_OFFSET);
3141
3142         key.objectid = last;
3143         key.offset = 0;
3144         key.type = BTRFS_EXTENT_ITEM_KEY;
3145
3146         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3147         if (ret < 0)
3148                 return ret;
3149         ret = 0;
3150         while (1) {
3151                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3152                         ret = btrfs_next_leaf(root, path);
3153                         if (ret < 0)
3154                                 return ret;
3155                         if (ret > 0) {
3156                                 ret = 0;
3157                                 break;
3158                         }
3159                 }
3160                 leaf = path->nodes[0];
3161                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3162                 if (key.objectid >= cache->key.offset + cache->key.objectid)
3163                         break;
3164                 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
3165                     key.type != BTRFS_METADATA_ITEM_KEY) {
3166                         path->slots[0]++;
3167                         continue;
3168                 }
3169
3170                 if (last == key.objectid) {
3171                         if (key.type == BTRFS_EXTENT_ITEM_KEY)
3172                                 last = key.objectid + key.offset;
3173                         else
3174                                 last = key.objectid + root->leafsize;
3175                         path->slots[0]++;
3176                         continue;
3177                 }
3178
3179                 ret = check_cache_range(root, cache, last,
3180                                         key.objectid - last);
3181                 if (ret)
3182                         break;
3183                 if (key.type == BTRFS_EXTENT_ITEM_KEY)
3184                         last = key.objectid + key.offset;
3185                 else
3186                         last = key.objectid + root->leafsize;
3187                 path->slots[0]++;
3188         }
3189
3190         if (last < cache->key.objectid + cache->key.offset)
3191                 ret = check_cache_range(root, cache, last,
3192                                         cache->key.objectid +
3193                                         cache->key.offset - last);
3194         btrfs_free_path(path);
3195
3196         if (!ret &&
3197             !RB_EMPTY_ROOT(&cache->free_space_ctl->free_space_offset)) {
3198                 fprintf(stderr, "There are still entries left in the space "
3199                         "cache\n");
3200                 ret = -EINVAL;
3201         }
3202
3203         return ret;
3204 }
3205
3206 static int check_space_cache(struct btrfs_root *root)
3207 {
3208         struct btrfs_block_group_cache *cache;
3209         u64 start = BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE;
3210         int ret;
3211         int error = 0;
3212
3213         if (btrfs_super_generation(root->fs_info->super_copy) !=
3214             btrfs_super_cache_generation(root->fs_info->super_copy)) {
3215                 printf("cache and super generation don't match, space cache "
3216                        "will be invalidated\n");
3217                 return 0;
3218         }
3219
3220         while (1) {
3221                 cache = btrfs_lookup_first_block_group(root->fs_info, start);
3222                 if (!cache)
3223                         break;
3224
3225                 start = cache->key.objectid + cache->key.offset;
3226                 if (!cache->free_space_ctl) {
3227                         if (btrfs_init_free_space_ctl(cache,
3228                                                       root->sectorsize)) {
3229                                 ret = -ENOMEM;
3230                                 break;
3231                         }
3232                 } else {
3233                         btrfs_remove_free_space_cache(cache);
3234                 }
3235
3236                 ret = load_free_space_cache(root->fs_info, cache);
3237                 if (!ret)
3238                         continue;
3239
3240                 ret = verify_space_cache(root, cache);
3241                 if (ret) {
3242                         fprintf(stderr, "cache appears valid but isnt %Lu\n",
3243                                 cache->key.objectid);
3244                         error++;
3245                 }
3246         }
3247
3248         return error ? -EINVAL : 0;
3249 }
3250
3251 static int check_extent_exists(struct btrfs_root *root, u64 bytenr,
3252                                u64 num_bytes)
3253 {
3254         struct btrfs_path *path;
3255         struct extent_buffer *leaf;
3256         struct btrfs_key key;
3257         int ret;
3258
3259         path = btrfs_alloc_path();
3260         if (!path) {
3261                 fprintf(stderr, "Error allocing path\n");
3262                 return -ENOMEM;
3263         }
3264
3265         key.objectid = bytenr;
3266         key.type = BTRFS_EXTENT_ITEM_KEY;
3267         key.offset = 0;
3268
3269
3270 again:
3271         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
3272                                 0, 0);
3273         if (ret < 0) {
3274                 fprintf(stderr, "Error looking up extent record %d\n", ret);
3275                 btrfs_free_path(path);
3276                 return ret;
3277         } else if (ret) {
3278                 if (path->slots[0])
3279                         path->slots[0]--;
3280                 else
3281                         btrfs_prev_leaf(root, path);
3282         }
3283
3284         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3285
3286         /*
3287          * Block group items come before extent items if they have the same
3288          * bytenr, so walk back one more just in case.  Dear future traveler,
3289          * first congrats on mastering time travel.  Now if it's not too much
3290          * trouble could you go back to 2006 and tell Chris to make the
3291          * BLOCK_GROUP_ITEM_KEY lower than the EXTENT_ITEM_KEY please?
3292          */
3293         if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
3294                 if (path->slots[0])
3295                         path->slots[0]--;
3296                 else
3297                         btrfs_prev_leaf(root, path);
3298         }
3299
3300         while (num_bytes) {
3301                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3302                         ret = btrfs_next_leaf(root, path);
3303                         if (ret < 0) {
3304                                 fprintf(stderr, "Error going to next leaf "
3305                                         "%d\n", ret);
3306                                 btrfs_free_path(path);
3307                                 return ret;
3308                         } else if (ret) {
3309                                 break;
3310                         }
3311                 }
3312                 leaf = path->nodes[0];
3313                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3314                 if (key.type != BTRFS_EXTENT_ITEM_KEY) {
3315                         path->slots[0]++;
3316                         continue;
3317                 }
3318                 if (key.objectid + key.offset < bytenr) {
3319                         path->slots[0]++;
3320                         continue;
3321                 }
3322                 if (key.objectid > bytenr + num_bytes)
3323                         break;
3324
3325                 if (key.objectid == bytenr) {
3326                         if (key.offset >= num_bytes) {
3327                                 num_bytes = 0;
3328                                 break;
3329                         }
3330                         num_bytes -= key.offset;
3331                         bytenr += key.offset;
3332                 } else if (key.objectid < bytenr) {
3333                         if (key.objectid + key.offset >= bytenr + num_bytes) {
3334                                 num_bytes = 0;
3335                                 break;
3336                         }
3337                         num_bytes = (bytenr + num_bytes) -
3338                                 (key.objectid + key.offset);
3339                         bytenr = key.objectid + key.offset;
3340                 } else {
3341                         if (key.objectid + key.offset < bytenr + num_bytes) {
3342                                 u64 new_start = key.objectid + key.offset;
3343                                 u64 new_bytes = bytenr + num_bytes - new_start;
3344
3345                                 /*
3346                                  * Weird case, the extent is in the middle of
3347                                  * our range, we'll have to search one side
3348                                  * and then the other.  Not sure if this happens
3349                                  * in real life, but no harm in coding it up
3350                                  * anyway just in case.
3351                                  */
3352                                 btrfs_release_path(root, path);
3353                                 ret = check_extent_exists(root, new_start,
3354                                                           new_bytes);
3355                                 if (ret) {
3356                                         fprintf(stderr, "Right section didn't "
3357                                                 "have a record\n");
3358                                         break;
3359                                 }
3360                                 num_bytes = key.objectid - bytenr;
3361                                 goto again;
3362                         }
3363                         num_bytes = key.objectid - bytenr;
3364                 }
3365                 path->slots[0]++;
3366         }
3367         ret = 0;
3368
3369         if (num_bytes) {
3370                 fprintf(stderr, "There are no extents for csum range "
3371                         "%Lu-%Lu\n", bytenr, bytenr+num_bytes);
3372                 ret = 1;
3373         }
3374
3375         btrfs_free_path(path);
3376         return ret;
3377 }
3378
3379 static int check_csums(struct btrfs_root *root)
3380 {
3381         struct btrfs_path *path;
3382         struct extent_buffer *leaf;
3383         struct btrfs_key key;
3384         u64 offset = 0, num_bytes = 0;
3385         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
3386         int errors = 0;
3387         int ret;
3388
3389         root = root->fs_info->csum_root;
3390
3391         key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
3392         key.type = BTRFS_EXTENT_CSUM_KEY;
3393         key.offset = 0;
3394
3395         path = btrfs_alloc_path();
3396         if (!path)
3397                 return -ENOMEM;
3398
3399         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3400         if (ret < 0) {
3401                 fprintf(stderr, "Error searching csum tree %d\n", ret);
3402                 btrfs_free_path(path);
3403                 return ret;
3404         }
3405
3406         if (ret > 0 && path->slots[0])
3407                 path->slots[0]--;
3408         ret = 0;
3409
3410         while (1) {
3411                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3412                         ret = btrfs_next_leaf(root, path);
3413                         if (ret < 0) {
3414                                 fprintf(stderr, "Error going to next leaf "
3415                                         "%d\n", ret);
3416                                 break;
3417                         }
3418                         if (ret)
3419                                 break;
3420                 }
3421                 leaf = path->nodes[0];
3422
3423                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3424                 if (key.type != BTRFS_EXTENT_CSUM_KEY) {
3425                         path->slots[0]++;
3426                         continue;
3427                 }
3428
3429                 if (!num_bytes) {
3430                         offset = key.offset;
3431                 } else if (key.offset != offset + num_bytes) {
3432                         ret = check_extent_exists(root, offset, num_bytes);
3433                         if (ret) {
3434                                 fprintf(stderr, "Csum exists for %Lu-%Lu but "
3435                                         "there is no extent record\n",
3436                                         offset, offset+num_bytes);
3437                                 errors++;
3438                         }
3439                         offset = key.offset;
3440                         num_bytes = 0;
3441                 }
3442
3443                 num_bytes += (btrfs_item_size_nr(leaf, path->slots[0]) /
3444                               csum_size) * root->sectorsize;
3445                 path->slots[0]++;
3446         }
3447
3448         btrfs_free_path(path);
3449         return errors;
3450 }
3451
3452 static int run_next_block(struct btrfs_root *root,
3453                           struct block_info *bits,
3454                           int bits_nr,
3455                           u64 *last,
3456                           struct cache_tree *pending,
3457                           struct cache_tree *seen,
3458                           struct cache_tree *reada,
3459                           struct cache_tree *nodes,
3460                           struct cache_tree *extent_cache,
3461                           struct cache_tree *chunk_cache,
3462                           struct rb_root *dev_cache,
3463                           struct block_group_tree *block_group_cache,
3464                           struct device_extent_tree *dev_extent_cache)
3465 {
3466         struct extent_buffer *buf;
3467         u64 bytenr;
3468         u32 size;
3469         u64 parent;
3470         u64 owner;
3471         u64 flags;
3472         int ret;
3473         int i;
3474         int nritems;
3475         struct btrfs_key key;
3476         struct cache_extent *cache;
3477         int reada_bits;
3478
3479         nritems = pick_next_pending(pending, reada, nodes, *last, bits,
3480                                     bits_nr, &reada_bits);
3481         if (nritems == 0)
3482                 return 1;
3483
3484         if (!reada_bits) {
3485                 for(i = 0; i < nritems; i++) {
3486                         ret = add_cache_extent(reada, bits[i].start,
3487                                                bits[i].size);
3488                         if (ret == -EEXIST)
3489                                 continue;
3490
3491                         /* fixme, get the parent transid */
3492                         readahead_tree_block(root, bits[i].start,
3493                                              bits[i].size, 0);
3494                 }
3495         }
3496         *last = bits[0].start;
3497         bytenr = bits[0].start;
3498         size = bits[0].size;
3499
3500         cache = lookup_cache_extent(pending, bytenr, size);
3501         if (cache) {
3502                 remove_cache_extent(pending, cache);
3503                 free(cache);
3504         }
3505         cache = lookup_cache_extent(reada, bytenr, size);
3506         if (cache) {
3507                 remove_cache_extent(reada, cache);
3508                 free(cache);
3509         }
3510         cache = lookup_cache_extent(nodes, bytenr, size);
3511         if (cache) {
3512                 remove_cache_extent(nodes, cache);
3513                 free(cache);
3514         }
3515
3516         /* fixme, get the real parent transid */
3517         buf = read_tree_block(root, bytenr, size, 0);
3518         if (!extent_buffer_uptodate(buf)) {
3519                 record_bad_block_io(root->fs_info,
3520                                     extent_cache, bytenr, size);
3521                 goto out;
3522         }
3523
3524         nritems = btrfs_header_nritems(buf);
3525
3526         ret = btrfs_lookup_extent_info(NULL, root, bytenr,
3527                                        btrfs_header_level(buf), 1, NULL,
3528                                        &flags);
3529         if (ret < 0)
3530                 flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
3531
3532         if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
3533                 parent = bytenr;
3534                 owner = 0;
3535         } else {
3536                 parent = 0;
3537                 owner = btrfs_header_owner(buf);
3538         }
3539
3540         ret = check_block(root, extent_cache, buf, flags);
3541         if (ret)
3542                 goto out;
3543
3544         if (btrfs_is_leaf(buf)) {
3545                 btree_space_waste += btrfs_leaf_free_space(root, buf);
3546                 for (i = 0; i < nritems; i++) {
3547                         struct btrfs_file_extent_item *fi;
3548                         btrfs_item_key_to_cpu(buf, &key, i);
3549                         if (key.type == BTRFS_EXTENT_ITEM_KEY) {
3550                                 process_extent_item(root, extent_cache, buf,
3551                                                     i);
3552                                 continue;
3553                         }
3554                         if (key.type == BTRFS_METADATA_ITEM_KEY) {
3555                                 process_extent_item(root, extent_cache, buf,
3556                                                     i);
3557                                 continue;
3558                         }
3559                         if (key.type == BTRFS_EXTENT_CSUM_KEY) {
3560                                 total_csum_bytes +=
3561                                         btrfs_item_size_nr(buf, i);
3562                                 continue;
3563                         }
3564                         if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3565                                 process_chunk_item(chunk_cache, &key, buf, i);
3566                                 continue;
3567                         }
3568                         if (key.type == BTRFS_DEV_ITEM_KEY) {
3569                                 process_device_item(dev_cache, &key, buf, i);
3570                                 continue;
3571                         }
3572                         if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
3573                                 process_block_group_item(block_group_cache,
3574                                         &key, buf, i);
3575                                 continue;
3576                         }
3577                         if (key.type == BTRFS_DEV_EXTENT_KEY) {
3578                                 process_device_extent_item(dev_extent_cache,
3579                                         &key, buf, i);
3580                                 continue;
3581
3582                         }
3583                         if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
3584 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3585                                 process_extent_ref_v0(extent_cache, buf, i);
3586 #else
3587                                 BUG();
3588 #endif
3589                                 continue;
3590                         }
3591
3592                         if (key.type == BTRFS_TREE_BLOCK_REF_KEY) {
3593                                 add_tree_backref(extent_cache, key.objectid, 0,
3594                                                  key.offset, 0);
3595                                 continue;
3596                         }
3597                         if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
3598                                 add_tree_backref(extent_cache, key.objectid,
3599                                                  key.offset, 0, 0);
3600                                 continue;
3601                         }
3602                         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
3603                                 struct btrfs_extent_data_ref *ref;
3604                                 ref = btrfs_item_ptr(buf, i,
3605                                                 struct btrfs_extent_data_ref);
3606                                 add_data_backref(extent_cache,
3607                                         key.objectid, 0,
3608                                         btrfs_extent_data_ref_root(buf, ref),
3609                                         btrfs_extent_data_ref_objectid(buf,
3610                                                                        ref),
3611                                         btrfs_extent_data_ref_offset(buf, ref),
3612                                         btrfs_extent_data_ref_count(buf, ref),
3613                                         0, root->sectorsize);
3614                                 continue;
3615                         }
3616                         if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
3617                                 struct btrfs_shared_data_ref *ref;
3618                                 ref = btrfs_item_ptr(buf, i,
3619                                                 struct btrfs_shared_data_ref);
3620                                 add_data_backref(extent_cache,
3621                                         key.objectid, key.offset, 0, 0, 0,
3622                                         btrfs_shared_data_ref_count(buf, ref),
3623                                         0, root->sectorsize);
3624                                 continue;
3625                         }
3626                         if (key.type != BTRFS_EXTENT_DATA_KEY)
3627                                 continue;
3628                         fi = btrfs_item_ptr(buf, i,
3629                                             struct btrfs_file_extent_item);
3630                         if (btrfs_file_extent_type(buf, fi) ==
3631                             BTRFS_FILE_EXTENT_INLINE)
3632                                 continue;
3633                         if (btrfs_file_extent_disk_bytenr(buf, fi) == 0)
3634                                 continue;
3635
3636                         data_bytes_allocated +=
3637                                 btrfs_file_extent_disk_num_bytes(buf, fi);
3638                         if (data_bytes_allocated < root->sectorsize) {
3639                                 abort();
3640                         }
3641                         data_bytes_referenced +=
3642                                 btrfs_file_extent_num_bytes(buf, fi);
3643                         add_data_backref(extent_cache,
3644                                 btrfs_file_extent_disk_bytenr(buf, fi),
3645                                 parent, owner, key.objectid, key.offset -
3646                                 btrfs_file_extent_offset(buf, fi), 1, 1,
3647                                 btrfs_file_extent_disk_num_bytes(buf, fi));
3648                         BUG_ON(ret);
3649                 }
3650         } else {
3651                 int level;
3652                 struct btrfs_key first_key;
3653
3654                 first_key.objectid = 0;
3655
3656                 if (nritems > 0)
3657                         btrfs_item_key_to_cpu(buf, &first_key, 0);
3658                 level = btrfs_header_level(buf);
3659                 for (i = 0; i < nritems; i++) {
3660                         u64 ptr = btrfs_node_blockptr(buf, i);
3661                         u32 size = btrfs_level_size(root, level - 1);
3662                         btrfs_node_key_to_cpu(buf, &key, i);
3663                         ret = add_extent_rec(extent_cache, &key,
3664                                              ptr, size, 0, 0, 1, 0, 1, 0,
3665                                              size);
3666                         BUG_ON(ret);
3667
3668                         add_tree_backref(extent_cache, ptr, parent, owner, 1);
3669
3670                         if (level > 1) {
3671                                 add_pending(nodes, seen, ptr, size);
3672                         } else {
3673                                 add_pending(pending, seen, ptr, size);
3674                         }
3675                 }
3676                 btree_space_waste += (BTRFS_NODEPTRS_PER_BLOCK(root) -
3677                                       nritems) * sizeof(struct btrfs_key_ptr);
3678         }
3679         total_btree_bytes += buf->len;
3680         if (fs_root_objectid(btrfs_header_owner(buf)))
3681                 total_fs_tree_bytes += buf->len;
3682         if (btrfs_header_owner(buf) == BTRFS_EXTENT_TREE_OBJECTID)
3683                 total_extent_tree_bytes += buf->len;
3684         if (!found_old_backref &&
3685             btrfs_header_owner(buf) == BTRFS_TREE_RELOC_OBJECTID &&
3686             btrfs_header_backref_rev(buf) == BTRFS_MIXED_BACKREF_REV &&
3687             !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))
3688                 found_old_backref = 1;
3689 out:
3690         free_extent_buffer(buf);
3691         return 0;
3692 }
3693
3694 static int add_root_to_pending(struct extent_buffer *buf,
3695                                struct cache_tree *extent_cache,
3696                                struct cache_tree *pending,
3697                                struct cache_tree *seen,
3698                                struct cache_tree *nodes,
3699                                struct btrfs_key *root_key)
3700 {
3701         if (btrfs_header_level(buf) > 0)
3702                 add_pending(nodes, seen, buf->start, buf->len);
3703         else
3704                 add_pending(pending, seen, buf->start, buf->len);
3705         add_extent_rec(extent_cache, NULL, buf->start, buf->len,
3706                        0, 1, 1, 0, 1, 0, buf->len);
3707
3708         if (root_key->objectid == BTRFS_TREE_RELOC_OBJECTID ||
3709             btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
3710                 add_tree_backref(extent_cache, buf->start, buf->start,
3711                                  0, 1);
3712         else
3713                 add_tree_backref(extent_cache, buf->start, 0,
3714                                  root_key->objectid, 1);
3715         return 0;
3716 }
3717
3718 /* as we fix the tree, we might be deleting blocks that
3719  * we're tracking for repair.  This hook makes sure we
3720  * remove any backrefs for blocks as we are fixing them.
3721  */
3722 static int free_extent_hook(struct btrfs_trans_handle *trans,
3723                             struct btrfs_root *root,
3724                             u64 bytenr, u64 num_bytes, u64 parent,
3725                             u64 root_objectid, u64 owner, u64 offset,
3726                             int refs_to_drop)
3727 {
3728         struct extent_record *rec;
3729         struct cache_extent *cache;
3730         int is_data;
3731         struct cache_tree *extent_cache = root->fs_info->fsck_extent_cache;
3732
3733         is_data = owner >= BTRFS_FIRST_FREE_OBJECTID;
3734         cache = lookup_cache_extent(extent_cache, bytenr, num_bytes);
3735         if (!cache)
3736                 return 0;
3737
3738         rec = container_of(cache, struct extent_record, cache);
3739         if (is_data) {
3740                 struct data_backref *back;
3741                 back = find_data_backref(rec, parent, root_objectid, owner,
3742                                          offset, 1, bytenr, num_bytes);
3743                 if (!back)
3744                         goto out;
3745                 if (back->node.found_ref) {
3746                         back->found_ref -= refs_to_drop;
3747                         if (rec->refs)
3748                                 rec->refs -= refs_to_drop;
3749                 }
3750                 if (back->node.found_extent_tree) {
3751                         back->num_refs -= refs_to_drop;
3752                         if (rec->extent_item_refs)
3753                                 rec->extent_item_refs -= refs_to_drop;
3754                 }
3755                 if (back->found_ref == 0)
3756                         back->node.found_ref = 0;
3757                 if (back->num_refs == 0)
3758                         back->node.found_extent_tree = 0;
3759
3760                 if (!back->node.found_extent_tree && back->node.found_ref) {
3761                         list_del(&back->node.list);
3762                         free(back);
3763                 }
3764         } else {
3765                 struct tree_backref *back;
3766                 back = find_tree_backref(rec, parent, root_objectid);
3767                 if (!back)
3768                         goto out;
3769                 if (back->node.found_ref) {
3770                         if (rec->refs)
3771                                 rec->refs--;
3772                         back->node.found_ref = 0;
3773                 }
3774                 if (back->node.found_extent_tree) {
3775                         if (rec->extent_item_refs)
3776                                 rec->extent_item_refs--;
3777                         back->node.found_extent_tree = 0;
3778                 }
3779                 if (!back->node.found_extent_tree && back->node.found_ref) {
3780                         list_del(&back->node.list);
3781                         free(back);
3782                 }
3783         }
3784         maybe_free_extent_rec(extent_cache, rec);
3785 out:
3786         return 0;
3787 }
3788
3789 static int delete_extent_records(struct btrfs_trans_handle *trans,
3790                                  struct btrfs_root *root,
3791                                  struct btrfs_path *path,
3792                                  u64 bytenr, u64 new_len)
3793 {
3794         struct btrfs_key key;
3795         struct btrfs_key found_key;
3796         struct extent_buffer *leaf;
3797         int ret;
3798         int slot;
3799
3800
3801         key.objectid = bytenr;
3802         key.type = (u8)-1;
3803         key.offset = (u64)-1;
3804
3805         while(1) {
3806                 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
3807                                         &key, path, 0, 1);
3808                 if (ret < 0)
3809                         break;
3810
3811                 if (ret > 0) {
3812                         ret = 0;
3813                         if (path->slots[0] == 0)
3814                                 break;
3815                         path->slots[0]--;
3816                 }
3817                 ret = 0;
3818
3819                 leaf = path->nodes[0];
3820                 slot = path->slots[0];
3821
3822                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3823                 if (found_key.objectid != bytenr)
3824                         break;
3825
3826                 if (found_key.type != BTRFS_EXTENT_ITEM_KEY &&
3827                     found_key.type != BTRFS_METADATA_ITEM_KEY &&
3828                     found_key.type != BTRFS_TREE_BLOCK_REF_KEY &&
3829                     found_key.type != BTRFS_EXTENT_DATA_REF_KEY &&
3830                     found_key.type != BTRFS_EXTENT_REF_V0_KEY &&
3831                     found_key.type != BTRFS_SHARED_BLOCK_REF_KEY &&
3832                     found_key.type != BTRFS_SHARED_DATA_REF_KEY) {
3833                         btrfs_release_path(NULL, path);
3834                         if (found_key.type == 0) {
3835                                 if (found_key.offset == 0)
3836                                         break;
3837                                 key.offset = found_key.offset - 1;
3838                                 key.type = found_key.type;
3839                         }
3840                         key.type = found_key.type - 1;
3841                         key.offset = (u64)-1;
3842                         continue;
3843                 }
3844
3845                 fprintf(stderr, "repair deleting extent record: key %Lu %u %Lu\n",
3846                         found_key.objectid, found_key.type, found_key.offset);
3847
3848                 ret = btrfs_del_item(trans, root->fs_info->extent_root, path);
3849                 if (ret)
3850                         break;
3851                 btrfs_release_path(NULL, path);
3852
3853                 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
3854                     found_key.type == BTRFS_METADATA_ITEM_KEY) {
3855                         u64 bytes = (found_key.type == BTRFS_EXTENT_ITEM_KEY) ?
3856                                 found_key.offset : root->leafsize;
3857
3858                         ret = btrfs_update_block_group(trans, root, bytenr,
3859                                                        bytes, 0, 0);
3860                         if (ret)
3861                                 break;
3862                 }
3863         }
3864
3865         btrfs_release_path(NULL, path);
3866         return ret;
3867 }
3868
3869 /*
3870  * for a single backref, this will allocate a new extent
3871  * and add the backref to it.
3872  */
3873 static int record_extent(struct btrfs_trans_handle *trans,
3874                          struct btrfs_fs_info *info,
3875                          struct btrfs_path *path,
3876                          struct extent_record *rec,
3877                          struct extent_backref *back,
3878                          int allocated, u64 flags)
3879 {
3880         int ret;
3881         struct btrfs_root *extent_root = info->extent_root;
3882         struct extent_buffer *leaf;
3883         struct btrfs_key ins_key;
3884         struct btrfs_extent_item *ei;
3885         struct tree_backref *tback;
3886         struct data_backref *dback;
3887         struct btrfs_tree_block_info *bi;
3888
3889         if (!back->is_data)
3890                 rec->max_size = max_t(u64, rec->max_size,
3891                                     info->extent_root->leafsize);
3892
3893         if (!allocated) {
3894                 u32 item_size = sizeof(*ei);
3895
3896                 if (!back->is_data)
3897                         item_size += sizeof(*bi);
3898
3899                 ins_key.objectid = rec->start;
3900                 ins_key.offset = rec->max_size;
3901                 ins_key.type = BTRFS_EXTENT_ITEM_KEY;
3902
3903                 ret = btrfs_insert_empty_item(trans, extent_root, path,
3904                                         &ins_key, item_size);
3905                 if (ret)
3906                         goto fail;
3907
3908                 leaf = path->nodes[0];
3909                 ei = btrfs_item_ptr(leaf, path->slots[0],
3910                                     struct btrfs_extent_item);
3911
3912                 btrfs_set_extent_refs(leaf, ei, 0);
3913                 btrfs_set_extent_generation(leaf, ei, rec->generation);
3914
3915                 if (back->is_data) {
3916                         btrfs_set_extent_flags(leaf, ei,
3917                                                BTRFS_EXTENT_FLAG_DATA);
3918                 } else {
3919                         struct btrfs_disk_key copy_key;;
3920
3921                         tback = (struct tree_backref *)back;
3922                         bi = (struct btrfs_tree_block_info *)(ei + 1);
3923                         memset_extent_buffer(leaf, 0, (unsigned long)bi,
3924                                              sizeof(*bi));
3925                         memset(&copy_key, 0, sizeof(copy_key));
3926
3927                         copy_key.objectid = le64_to_cpu(rec->info_objectid);
3928                         btrfs_set_tree_block_level(leaf, bi, rec->info_level);
3929                         btrfs_set_tree_block_key(leaf, bi, &copy_key);
3930
3931                         btrfs_set_extent_flags(leaf, ei,
3932                                                BTRFS_EXTENT_FLAG_TREE_BLOCK | flags);
3933                 }
3934
3935                 btrfs_mark_buffer_dirty(leaf);
3936                 ret = btrfs_update_block_group(trans, extent_root, rec->start,
3937                                                rec->max_size, 1, 0);
3938                 if (ret)
3939                         goto fail;
3940                 btrfs_release_path(NULL, path);
3941         }
3942
3943         if (back->is_data) {
3944                 u64 parent;
3945                 int i;
3946
3947                 dback = (struct data_backref *)back;
3948                 if (back->full_backref)
3949                         parent = dback->parent;
3950                 else
3951                         parent = 0;
3952
3953                 for (i = 0; i < dback->found_ref; i++) {
3954                         /* if parent != 0, we're doing a full backref
3955                          * passing BTRFS_FIRST_FREE_OBJECTID as the owner
3956                          * just makes the backref allocator create a data
3957                          * backref
3958                          */
3959                         ret = btrfs_inc_extent_ref(trans, info->extent_root,
3960                                                    rec->start, rec->max_size,
3961                                                    parent,
3962                                                    dback->root,
3963                                                    parent ?
3964                                                    BTRFS_FIRST_FREE_OBJECTID :
3965                                                    dback->owner,
3966                                                    dback->offset);
3967                         if (ret)
3968                                 break;
3969                 }
3970                 fprintf(stderr, "adding new data backref"
3971                                 " on %llu %s %llu owner %llu"
3972                                 " offset %llu found %d\n",
3973                                 (unsigned long long)rec->start,
3974                                 back->full_backref ?
3975                                 "parent" : "root",
3976                                 back->full_backref ?
3977                                 (unsigned long long)parent :
3978                                 (unsigned long long)dback->root,
3979                                 (unsigned long long)dback->owner,
3980                                 (unsigned long long)dback->offset,
3981                                 dback->found_ref);
3982         } else {
3983                 u64 parent;
3984
3985                 tback = (struct tree_backref *)back;
3986                 if (back->full_backref)
3987                         parent = tback->parent;
3988                 else
3989                         parent = 0;
3990
3991                 ret = btrfs_inc_extent_ref(trans, info->extent_root,
3992                                            rec->start, rec->max_size,
3993                                            parent, tback->root, 0, 0);
3994                 fprintf(stderr, "adding new tree backref on "
3995                         "start %llu len %llu parent %llu root %llu\n",
3996                         rec->start, rec->max_size, tback->parent, tback->root);
3997         }
3998         if (ret)
3999                 goto fail;
4000 fail:
4001         btrfs_release_path(NULL, path);
4002         return ret;
4003 }
4004
4005 struct extent_entry {
4006         u64 bytenr;
4007         u64 bytes;
4008         int count;
4009         struct list_head list;
4010 };
4011
4012 static struct extent_entry *find_entry(struct list_head *entries,
4013                                        u64 bytenr, u64 bytes)
4014 {
4015         struct extent_entry *entry = NULL;
4016
4017         list_for_each_entry(entry, entries, list) {
4018                 if (entry->bytenr == bytenr && entry->bytes == bytes)
4019                         return entry;
4020         }
4021
4022         return NULL;
4023 }
4024
4025 static struct extent_entry *find_most_right_entry(struct list_head *entries)
4026 {
4027         struct extent_entry *entry, *best = NULL, *prev = NULL;
4028
4029         list_for_each_entry(entry, entries, list) {
4030                 if (!prev) {
4031                         prev = entry;
4032                         continue;
4033                 }
4034
4035                 /*
4036                  * If our current entry == best then we can't be sure our best
4037                  * is really the best, so we need to keep searching.
4038                  */
4039                 if (best && best->count == entry->count) {
4040                         prev = entry;
4041                         best = NULL;
4042                         continue;
4043                 }
4044
4045                 /* Prev == entry, not good enough, have to keep searching */
4046                 if (prev->count == entry->count)
4047                         continue;
4048
4049                 if (!best)
4050                         best = (prev->count > entry->count) ? prev : entry;
4051                 else if (best->count < entry->count)
4052                         best = entry;
4053                 prev = entry;
4054         }
4055
4056         return best;
4057 }
4058
4059 static int repair_ref(struct btrfs_trans_handle *trans,
4060                       struct btrfs_fs_info *info, struct btrfs_path *path,
4061                       struct data_backref *dback, struct extent_entry *entry)
4062 {
4063         struct btrfs_root *root;
4064         struct btrfs_file_extent_item *fi;
4065         struct extent_buffer *leaf;
4066         struct btrfs_key key;
4067         u64 bytenr, bytes;
4068         int ret;
4069
4070         key.objectid = dback->root;
4071         key.type = BTRFS_ROOT_ITEM_KEY;
4072         key.offset = (u64)-1;
4073         root = btrfs_read_fs_root(info, &key);
4074         if (IS_ERR(root)) {
4075                 fprintf(stderr, "Couldn't find root for our ref\n");
4076                 return -EINVAL;
4077         }
4078
4079         /*
4080          * The backref points to the original offset of the extent if it was
4081          * split, so we need to search down to the offset we have and then walk
4082          * forward until we find the backref we're looking for.
4083          */
4084         key.objectid = dback->owner;
4085         key.type = BTRFS_EXTENT_DATA_KEY;
4086         key.offset = dback->offset;
4087         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4088         if (ret < 0) {
4089                 fprintf(stderr, "Error looking up ref %d\n", ret);
4090                 return ret;
4091         }
4092
4093         while (1) {
4094                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
4095                         ret = btrfs_next_leaf(root, path);
4096                         if (ret) {
4097                                 fprintf(stderr, "Couldn't find our ref, next\n");
4098                                 return -EINVAL;
4099                         }
4100                 }
4101                 leaf = path->nodes[0];
4102                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4103                 if (key.objectid != dback->owner ||
4104                     key.type != BTRFS_EXTENT_DATA_KEY) {
4105                         fprintf(stderr, "Couldn't find our ref, search\n");
4106                         return -EINVAL;
4107                 }
4108                 fi = btrfs_item_ptr(leaf, path->slots[0],
4109                                     struct btrfs_file_extent_item);
4110                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4111                 bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4112
4113                 if (bytenr == dback->disk_bytenr && bytes == dback->bytes)
4114                         break;
4115                 path->slots[0]++;
4116         }
4117
4118         btrfs_release_path(root, path);
4119
4120         /*
4121          * Have to make sure that this root gets updated when we commit the
4122          * transaction
4123          */
4124         root->track_dirty = 1;
4125         if (root->last_trans != trans->transid) {
4126                 root->last_trans = trans->transid;
4127                 root->commit_root = root->node;
4128                 extent_buffer_get(root->node);
4129         }
4130
4131         /*
4132          * Ok we have the key of the file extent we want to fix, now we can cow
4133          * down to the thing and fix it.
4134          */
4135         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4136         if (ret < 0) {
4137                 fprintf(stderr, "Error cowing down to ref [%Lu, %u, %Lu]: %d\n",
4138                         key.objectid, key.type, key.offset, ret);
4139                 return ret;
4140         }
4141         if (ret > 0) {
4142                 fprintf(stderr, "Well that's odd, we just found this key "
4143                         "[%Lu, %u, %Lu]\n", key.objectid, key.type,
4144                         key.offset);
4145                 return -EINVAL;
4146         }
4147         leaf = path->nodes[0];
4148         fi = btrfs_item_ptr(leaf, path->slots[0],
4149                             struct btrfs_file_extent_item);
4150
4151         if (btrfs_file_extent_compression(leaf, fi) &&
4152             dback->disk_bytenr != entry->bytenr) {
4153                 fprintf(stderr, "Ref doesn't match the record start and is "
4154                         "compressed, please take a btrfs-image of this file "
4155                         "system and send it to a btrfs developer so they can "
4156                         "complete this functionality for bytenr %Lu\n",
4157                         dback->disk_bytenr);
4158                 return -EINVAL;
4159         }
4160
4161         if (dback->disk_bytenr > entry->bytenr) {
4162                 u64 off_diff, offset;
4163
4164                 off_diff = dback->disk_bytenr - entry->bytenr;
4165                 offset = btrfs_file_extent_offset(leaf, fi);
4166                 if (dback->disk_bytenr + offset +
4167                     btrfs_file_extent_num_bytes(leaf, fi) >
4168                     entry->bytenr + entry->bytes) {
4169                         fprintf(stderr, "Ref is past the entry end, please "
4170                                 "take a btrfs-image of this file system and "
4171                                 "send it to a btrfs developer, ref %Lu\n",
4172                                 dback->disk_bytenr);
4173                         return -EINVAL;
4174                 }
4175                 offset += off_diff;
4176                 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
4177                 btrfs_set_file_extent_offset(leaf, fi, offset);
4178         } else if (dback->disk_bytenr < entry->bytenr) {
4179                 u64 offset;
4180
4181                 offset = btrfs_file_extent_offset(leaf, fi);
4182                 if (dback->disk_bytenr + offset < entry->bytenr) {
4183                         fprintf(stderr, "Ref is before the entry start, please"
4184                                 " take a btrfs-image of this file system and "
4185                                 "send it to a btrfs developer, ref %Lu\n",
4186                                 dback->disk_bytenr);
4187                         return -EINVAL;
4188                 }
4189
4190                 offset += dback->disk_bytenr;
4191                 offset -= entry->bytenr;
4192                 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
4193                 btrfs_set_file_extent_offset(leaf, fi, offset);
4194         }
4195
4196         btrfs_set_file_extent_disk_num_bytes(leaf, fi, entry->bytes);
4197
4198         /*
4199          * Chances are if disk_num_bytes were wrong then so is ram_bytes, but
4200          * only do this if we aren't using compression, otherwise it's a
4201          * trickier case.
4202          */
4203         if (!btrfs_file_extent_compression(leaf, fi))
4204                 btrfs_set_file_extent_ram_bytes(leaf, fi, entry->bytes);
4205         else
4206                 printf("ram bytes may be wrong?\n");
4207         btrfs_mark_buffer_dirty(leaf);
4208         btrfs_release_path(root, path);
4209         return 0;
4210 }
4211
4212 static int verify_backrefs(struct btrfs_trans_handle *trans,
4213                            struct btrfs_fs_info *info, struct btrfs_path *path,
4214                            struct extent_record *rec)
4215 {
4216         struct extent_backref *back;
4217         struct data_backref *dback;
4218         struct extent_entry *entry, *best = NULL;
4219         LIST_HEAD(entries);
4220         int nr_entries = 0;
4221         int ret = 0;
4222
4223         /*
4224          * Metadata is easy and the backrefs should always agree on bytenr and
4225          * size, if not we've got bigger issues.
4226          */
4227         if (rec->metadata)
4228                 return 0;
4229
4230         list_for_each_entry(back, &rec->backrefs, list) {
4231                 dback = (struct data_backref *)back;
4232                 /*
4233                  * We only pay attention to backrefs that we found a real
4234                  * backref for.
4235                  */
4236                 if (dback->found_ref == 0)
4237                         continue;
4238                 if (back->full_backref)
4239                         continue;
4240
4241                 /*
4242                  * For now we only catch when the bytes don't match, not the
4243                  * bytenr.  We can easily do this at the same time, but I want
4244                  * to have a fs image to test on before we just add repair
4245                  * functionality willy-nilly so we know we won't screw up the
4246                  * repair.
4247                  */
4248
4249                 entry = find_entry(&entries, dback->disk_bytenr,
4250                                    dback->bytes);
4251                 if (!entry) {
4252                         entry = malloc(sizeof(struct extent_entry));
4253                         if (!entry) {
4254                                 ret = -ENOMEM;
4255                                 goto out;
4256                         }
4257                         memset(entry, 0, sizeof(*entry));
4258                         entry->bytenr = dback->disk_bytenr;
4259                         entry->bytes = dback->bytes;
4260                         list_add_tail(&entry->list, &entries);
4261                         nr_entries++;
4262                 }
4263                 entry->count++;
4264         }
4265
4266         /* Yay all the backrefs agree, carry on good sir */
4267         if (nr_entries <= 1)
4268                 goto out;
4269
4270         fprintf(stderr, "attempting to repair backref discrepency for bytenr "
4271                 "%Lu\n", rec->start);
4272
4273         /*
4274          * First we want to see if the backrefs can agree amongst themselves who
4275          * is right, so figure out which one of the entries has the highest
4276          * count.
4277          */
4278         best = find_most_right_entry(&entries);
4279
4280         /*
4281          * Ok so we may have an even split between what the backrefs think, so
4282          * this is where we use the extent ref to see what it thinks.
4283          */
4284         if (!best) {
4285                 entry = find_entry(&entries, rec->start, rec->nr);
4286                 if (!entry) {
4287                         fprintf(stderr, "Backrefs don't agree with eachother "
4288                                 "and extent record doesn't agree with anybody,"
4289                                 " so we can't fix bytenr %Lu bytes %Lu\n",
4290                                 rec->start, rec->nr);
4291                         ret = -EINVAL;
4292                         goto out;
4293                 }
4294                 entry->count++;
4295                 best = find_most_right_entry(&entries);
4296                 if (!best) {
4297                         fprintf(stderr, "Backrefs and extent record evenly "
4298                                 "split on who is right, this is going to "
4299                                 "require user input to fix bytenr %Lu bytes "
4300                                 "%Lu\n", rec->start, rec->nr);
4301                         ret = -EINVAL;
4302                         goto out;
4303                 }
4304         }
4305
4306         /*
4307          * I don't think this can happen currently as we'll abort() if we catch
4308          * this case higher up, but in case somebody removes that we still can't
4309          * deal with it properly here yet, so just bail out of that's the case.
4310          */
4311         if (best->bytenr != rec->start) {
4312                 fprintf(stderr, "Extent start and backref starts don't match, "
4313                         "please use btrfs-image on this file system and send "
4314                         "it to a btrfs developer so they can make fsck fix "
4315                         "this particular case.  bytenr is %Lu, bytes is %Lu\n",
4316                         rec->start, rec->nr);
4317                 ret = -EINVAL;
4318                 goto out;
4319         }
4320
4321         /*
4322          * Ok great we all agreed on an extent record, let's go find the real
4323          * references and fix up the ones that don't match.
4324          */
4325         list_for_each_entry(back, &rec->backrefs, list) {
4326                 dback = (struct data_backref *)back;
4327
4328                 /*
4329                  * Still ignoring backrefs that don't have a real ref attached
4330                  * to them.
4331                  */
4332                 if (dback->found_ref == 0)
4333                         continue;
4334                 if (back->full_backref)
4335                         continue;
4336
4337                 if (dback->bytes == best->bytes &&
4338                     dback->disk_bytenr == best->bytenr)
4339                         continue;
4340
4341                 ret = repair_ref(trans, info, path, dback, best);
4342                 if (ret)
4343                         goto out;
4344         }
4345
4346         /*
4347          * Ok we messed with the actual refs, which means we need to drop our
4348          * entire cache and go back and rescan.  I know this is a huge pain and
4349          * adds a lot of extra work, but it's the only way to be safe.  Once all
4350          * the backrefs agree we may not need to do anything to the extent
4351          * record itself.
4352          */
4353         ret = -EAGAIN;
4354 out:
4355         while (!list_empty(&entries)) {
4356                 entry = list_entry(entries.next, struct extent_entry, list);
4357                 list_del_init(&entry->list);
4358                 free(entry);
4359         }
4360         return ret;
4361 }
4362
4363 static int process_duplicates(struct btrfs_root *root,
4364                               struct cache_tree *extent_cache,
4365                               struct extent_record *rec)
4366 {
4367         struct extent_record *good, *tmp;
4368         struct cache_extent *cache;
4369         int ret;
4370
4371         /*
4372          * If we found a extent record for this extent then return, or if we
4373          * have more than one duplicate we are likely going to need to delete
4374          * something.
4375          */
4376         if (rec->found_rec || rec->num_duplicates > 1)
4377                 return 0;
4378
4379         /* Shouldn't happen but just in case */
4380         BUG_ON(!rec->num_duplicates);
4381
4382         /*
4383          * So this happens if we end up with a backref that doesn't match the
4384          * actual extent entry.  So either the backref is bad or the extent
4385          * entry is bad.  Either way we want to have the extent_record actually
4386          * reflect what we found in the extent_tree, so we need to take the
4387          * duplicate out and use that as the extent_record since the only way we
4388          * get a duplicate is if we find a real life BTRFS_EXTENT_ITEM_KEY.
4389          */
4390         remove_cache_extent(extent_cache, &rec->cache);
4391
4392         good = list_entry(rec->dups.next, struct extent_record, list);
4393         list_del_init(&good->list);
4394         INIT_LIST_HEAD(&good->backrefs);
4395         INIT_LIST_HEAD(&good->dups);
4396         good->cache.start = good->start;
4397         good->cache.size = good->nr;
4398         good->content_checked = 0;
4399         good->owner_ref_checked = 0;
4400         good->num_duplicates = 0;
4401         good->refs = rec->refs;
4402         list_splice_init(&rec->backrefs, &good->backrefs);
4403         while (1) {
4404                 cache = lookup_cache_extent(extent_cache, good->start,
4405                                             good->nr);
4406                 if (!cache)
4407                         break;
4408                 tmp = container_of(cache, struct extent_record, cache);
4409
4410                 /*
4411                  * If we find another overlapping extent and it's found_rec is
4412                  * set then it's a duplicate and we need to try and delete
4413                  * something.
4414                  */
4415                 if (tmp->found_rec || tmp->num_duplicates > 0) {
4416                         if (list_empty(&good->list))
4417                                 list_add_tail(&good->list,
4418                                               &duplicate_extents);
4419                         good->num_duplicates += tmp->num_duplicates + 1;
4420                         list_splice_init(&tmp->dups, &good->dups);
4421                         list_del_init(&tmp->list);
4422                         list_add_tail(&tmp->list, &good->dups);
4423                         remove_cache_extent(extent_cache, &tmp->cache);
4424                         continue;
4425                 }
4426
4427                 /*
4428                  * Ok we have another non extent item backed extent rec, so lets
4429                  * just add it to this extent and carry on like we did above.
4430                  */
4431                 good->refs += tmp->refs;
4432                 list_splice_init(&tmp->backrefs, &good->backrefs);
4433                 remove_cache_extent(extent_cache, &tmp->cache);
4434                 free(tmp);
4435         }
4436         ret = insert_cache_extent(extent_cache, &good->cache);
4437         BUG_ON(ret);
4438         free(rec);
4439         return good->num_duplicates ? 0 : 1;
4440 }
4441
4442 static int delete_duplicate_records(struct btrfs_trans_handle *trans,
4443                                     struct btrfs_root *root,
4444                                     struct extent_record *rec)
4445 {
4446         LIST_HEAD(delete_list);
4447         struct btrfs_path *path;
4448         struct extent_record *tmp, *good, *n;
4449         int nr_del = 0;
4450         int ret = 0;
4451         struct btrfs_key key;
4452
4453         path = btrfs_alloc_path();
4454         if (!path) {
4455                 ret = -ENOMEM;
4456                 goto out;
4457         }
4458
4459         good = rec;
4460         /* Find the record that covers all of the duplicates. */
4461         list_for_each_entry(tmp, &rec->dups, list) {
4462                 if (good->start < tmp->start)
4463                         continue;
4464                 if (good->nr > tmp->nr)
4465                         continue;
4466
4467                 if (tmp->start + tmp->nr < good->start + good->nr) {
4468                         fprintf(stderr, "Ok we have overlapping extents that "
4469                                 "aren't completely covered by eachother, this "
4470                                 "is going to require more careful thought.  "
4471                                 "The extents are [%Lu-%Lu] and [%Lu-%Lu]\n",
4472                                 tmp->start, tmp->nr, good->start, good->nr);
4473                         abort();
4474                 }
4475                 good = tmp;
4476         }
4477
4478         if (good != rec)
4479                 list_add_tail(&rec->list, &delete_list);
4480
4481         list_for_each_entry_safe(tmp, n, &rec->dups, list) {
4482                 if (tmp == good)
4483                         continue;
4484                 list_move_tail(&tmp->list, &delete_list);
4485         }
4486
4487         root = root->fs_info->extent_root;
4488         list_for_each_entry(tmp, &delete_list, list) {
4489                 if (tmp->found_rec == 0)
4490                         continue;
4491                 key.objectid = tmp->start;
4492                 key.type = BTRFS_EXTENT_ITEM_KEY;
4493                 key.offset = tmp->nr;
4494
4495                 /* Shouldn't happen but just in case */
4496                 if (tmp->metadata) {
4497                         fprintf(stderr, "Well this shouldn't happen, extent "
4498                                 "record overlaps but is metadata? "
4499                                 "[%Lu, %Lu]\n", tmp->start, tmp->nr);
4500                         abort();
4501                 }
4502
4503                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4504                 if (ret) {
4505                         if (ret > 0)
4506                                 ret = -EINVAL;
4507                         goto out;
4508                 }
4509                 ret = btrfs_del_item(trans, root, path);
4510                 if (ret)
4511                         goto out;
4512                 btrfs_release_path(root, path);
4513                 nr_del++;
4514         }
4515
4516 out:
4517         while (!list_empty(&delete_list)) {
4518                 tmp = list_entry(delete_list.next, struct extent_record, list);
4519                 list_del_init(&tmp->list);
4520                 if (tmp == rec)
4521                         continue;
4522                 free(tmp);
4523         }
4524
4525         while (!list_empty(&rec->dups)) {
4526                 tmp = list_entry(rec->dups.next, struct extent_record, list);
4527                 list_del_init(&tmp->list);
4528                 free(tmp);
4529         }
4530
4531         btrfs_free_path(path);
4532
4533         if (!ret && !nr_del)
4534                 rec->num_duplicates = 0;
4535
4536         return ret ? ret : nr_del;
4537 }
4538
4539 /*
4540  * when an incorrect extent item is found, this will delete
4541  * all of the existing entries for it and recreate them
4542  * based on what the tree scan found.
4543  */
4544 static int fixup_extent_refs(struct btrfs_trans_handle *trans,
4545                              struct btrfs_fs_info *info,
4546                              struct extent_record *rec)
4547 {
4548         int ret;
4549         struct btrfs_path *path;
4550         struct list_head *cur = rec->backrefs.next;
4551         struct cache_extent *cache;
4552         struct extent_backref *back;
4553         int allocated = 0;
4554         u64 flags = 0;
4555
4556         /* remember our flags for recreating the extent */
4557         ret = btrfs_lookup_extent_info(NULL, info->extent_root, rec->start,
4558                                        rec->max_size, rec->metadata, NULL,
4559                                        &flags);
4560         if (ret < 0)
4561                 flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
4562
4563         path = btrfs_alloc_path();
4564
4565         /* step one, make sure all of the backrefs agree */
4566         ret = verify_backrefs(trans, info, path, rec);
4567         if (ret < 0)
4568                 goto out;
4569
4570         /* step two, delete all the existing records */
4571         ret = delete_extent_records(trans, info->extent_root, path,
4572                                     rec->start, rec->max_size);
4573
4574         if (ret < 0)
4575                 goto out;
4576
4577         /* was this block corrupt?  If so, don't add references to it */
4578         cache = lookup_cache_extent(info->corrupt_blocks,
4579                                     rec->start, rec->max_size);
4580         if (cache) {
4581                 ret = 0;
4582                 goto out;
4583         }
4584
4585         /* step three, recreate all the refs we did find */
4586         while(cur != &rec->backrefs) {
4587                 back = list_entry(cur, struct extent_backref, list);
4588                 cur = cur->next;
4589
4590                 /*
4591                  * if we didn't find any references, don't create a
4592                  * new extent record
4593                  */
4594                 if (!back->found_ref)
4595                         continue;
4596
4597                 ret = record_extent(trans, info, path, rec, back, allocated, flags);
4598                 allocated = 1;
4599
4600                 if (ret)
4601                         goto out;
4602         }
4603 out:
4604         btrfs_free_path(path);
4605         return ret;
4606 }
4607
4608 /* right now we only prune from the extent allocation tree */
4609 static int prune_one_block(struct btrfs_trans_handle *trans,
4610                            struct btrfs_fs_info *info,
4611                            struct btrfs_corrupt_block *corrupt)
4612 {
4613         int ret;
4614         struct btrfs_path path;
4615         struct extent_buffer *eb;
4616         u64 found;
4617         int slot;
4618         int nritems;
4619         int level = corrupt->level + 1;
4620
4621         btrfs_init_path(&path);
4622 again:
4623         /* we want to stop at the parent to our busted block */
4624         path.lowest_level = level;
4625
4626         ret = btrfs_search_slot(trans, info->extent_root,
4627                                 &corrupt->key, &path, -1, 1);
4628
4629         if (ret < 0)
4630                 goto out;
4631
4632         eb = path.nodes[level];
4633         if (!eb) {
4634                 ret = -ENOENT;
4635                 goto out;
4636         }
4637
4638         /*
4639          * hopefully the search gave us the block we want to prune,
4640          * lets try that first
4641          */
4642         slot = path.slots[level];
4643         found =  btrfs_node_blockptr(eb, slot);
4644         if (found == corrupt->cache.start)
4645                 goto del_ptr;
4646
4647         nritems = btrfs_header_nritems(eb);
4648
4649         /* the search failed, lets scan this node and hope we find it */
4650         for (slot = 0; slot < nritems; slot++) {
4651                 found =  btrfs_node_blockptr(eb, slot);
4652                 if (found == corrupt->cache.start)
4653                         goto del_ptr;
4654         }
4655         /*
4656          * we couldn't find the bad block.  TODO, search all the nodes for pointers
4657          * to this block
4658          */
4659         if (eb == info->extent_root->node) {
4660                 ret = -ENOENT;
4661                 goto out;
4662         } else {
4663                 level++;
4664                 btrfs_release_path(NULL, &path);
4665                 goto again;
4666         }
4667
4668 del_ptr:
4669         printk("deleting pointer to block %Lu\n", corrupt->cache.start);
4670         ret = btrfs_del_ptr(trans, info->extent_root, &path, level, slot);
4671
4672 out:
4673         btrfs_release_path(NULL, &path);
4674         return ret;
4675 }
4676
4677 static int prune_corrupt_blocks(struct btrfs_trans_handle *trans,
4678                                 struct btrfs_fs_info *info)
4679 {
4680         struct cache_extent *cache;
4681         struct btrfs_corrupt_block *corrupt;
4682
4683         cache = search_cache_extent(info->corrupt_blocks, 0);
4684         while (1) {
4685                 if (!cache)
4686                         break;
4687                 corrupt = container_of(cache, struct btrfs_corrupt_block, cache);
4688                 prune_one_block(trans, info, corrupt);
4689                 cache = next_cache_extent(cache);
4690         }
4691         return 0;
4692 }
4693
4694 static void free_corrupt_block(struct cache_extent *cache)
4695 {
4696         struct btrfs_corrupt_block *corrupt;
4697
4698         corrupt = container_of(cache, struct btrfs_corrupt_block, cache);
4699         free(corrupt);
4700 }
4701
4702 FREE_EXTENT_CACHE_BASED_TREE(corrupt_blocks, free_corrupt_block);
4703
4704 static int check_block_group(struct btrfs_trans_handle *trans,
4705                               struct btrfs_fs_info *info,
4706                               struct map_lookup *map,
4707                               int *reinit)
4708 {
4709         struct btrfs_key key;
4710         struct btrfs_path path;
4711         int ret;
4712
4713         key.objectid = map->ce.start;
4714         key.offset = map->ce.size;
4715         key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
4716
4717         btrfs_init_path(&path);
4718         ret = btrfs_search_slot(NULL, info->extent_root,
4719                                 &key, &path, 0, 0);
4720         btrfs_release_path(NULL, &path);
4721         if (ret <= 0)
4722                 goto out;
4723
4724         ret = btrfs_make_block_group(trans, info->extent_root, 0, map->type,
4725                                BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4726                                key.objectid, key.offset);
4727         *reinit = 1;
4728 out:
4729         return ret;
4730 }
4731
4732 static int check_block_groups(struct btrfs_trans_handle *trans,
4733                               struct btrfs_fs_info *info, int *reinit)
4734 {
4735         struct cache_extent *ce;
4736         struct map_lookup *map;
4737         struct btrfs_mapping_tree *map_tree = &info->mapping_tree;
4738
4739         /* this isn't quite working */
4740         return 0;
4741
4742         ce = search_cache_extent(&map_tree->cache_tree, 0);
4743         while (1) {
4744                 if (!ce)
4745                         break;
4746                 map = container_of(ce, struct map_lookup, ce);
4747                 check_block_group(trans, info, map, reinit);
4748                 ce = next_cache_extent(ce);
4749         }
4750         return 0;
4751 }
4752
4753 static void reset_cached_block_groups(struct btrfs_fs_info *fs_info)
4754 {
4755         struct btrfs_block_group_cache *cache;
4756         u64 start, end;
4757         int ret;
4758
4759         while (1) {
4760                 ret = find_first_extent_bit(&fs_info->free_space_cache, 0,
4761                                             &start, &end, EXTENT_DIRTY);
4762                 if (ret)
4763                         break;
4764                 clear_extent_dirty(&fs_info->free_space_cache, start, end,
4765                                    GFP_NOFS);
4766         }
4767
4768         start = 0;
4769         while (1) {
4770                 cache = btrfs_lookup_first_block_group(fs_info, start);
4771                 if (!cache)
4772                         break;
4773                 if (cache->cached)
4774                         cache->cached = 0;
4775                 start = cache->key.objectid + cache->key.offset;
4776         }
4777 }
4778
4779 static int check_extent_refs(struct btrfs_trans_handle *trans,
4780                              struct btrfs_root *root,
4781                              struct cache_tree *extent_cache, int repair)
4782 {
4783         struct extent_record *rec;
4784         struct cache_extent *cache;
4785         int err = 0;
4786         int ret = 0;
4787         int fixed = 0;
4788         int reinit = 0;
4789         int had_dups = 0;
4790
4791         if (repair) {
4792                 /*
4793                  * if we're doing a repair, we have to make sure
4794                  * we don't allocate from the problem extents.
4795                  * In the worst case, this will be all the
4796                  * extents in the FS
4797                  */
4798                 cache = search_cache_extent(extent_cache, 0);
4799                 while(cache) {
4800                         rec = container_of(cache, struct extent_record, cache);
4801                         btrfs_pin_extent(root->fs_info,
4802                                          rec->start, rec->max_size);
4803                         cache = next_cache_extent(cache);
4804                 }
4805
4806                 /* pin down all the corrupted blocks too */
4807                 cache = search_cache_extent(root->fs_info->corrupt_blocks, 0);
4808                 while(cache) {
4809                         rec = container_of(cache, struct extent_record, cache);
4810                         btrfs_pin_extent(root->fs_info,
4811                                          rec->start, rec->max_size);
4812                         cache = next_cache_extent(cache);
4813                 }
4814                 prune_corrupt_blocks(trans, root->fs_info);
4815                 check_block_groups(trans, root->fs_info, &reinit);
4816                 if (reinit)
4817                         btrfs_read_block_groups(root->fs_info->extent_root);
4818                 reset_cached_block_groups(root->fs_info);
4819         }
4820
4821         /*
4822          * We need to delete any duplicate entries we find first otherwise we
4823          * could mess up the extent tree when we have backrefs that actually
4824          * belong to a different extent item and not the weird duplicate one.
4825          */
4826         while (repair && !list_empty(&duplicate_extents)) {
4827                 rec = list_entry(duplicate_extents.next, struct extent_record,
4828                                  list);
4829                 list_del_init(&rec->list);
4830
4831                 /* Sometimes we can find a backref before we find an actual
4832                  * extent, so we need to process it a little bit to see if there
4833                  * truly are multiple EXTENT_ITEM_KEY's for the same range, or
4834                  * if this is a backref screwup.  If we need to delete stuff
4835                  * process_duplicates() will return 0, otherwise it will return
4836                  * 1 and we
4837                  */
4838                 if (process_duplicates(root, extent_cache, rec))
4839                         continue;
4840                 ret = delete_duplicate_records(trans, root, rec);
4841                 if (ret < 0)
4842                         return ret;
4843                 /*
4844                  * delete_duplicate_records will return the number of entries
4845                  * deleted, so if it's greater than 0 then we know we actually
4846                  * did something and we need to remove.
4847                  */
4848                 if (ret)
4849                         had_dups = 1;
4850         }
4851
4852         if (had_dups)
4853                 return -EAGAIN;
4854
4855         while(1) {
4856                 fixed = 0;
4857                 cache = search_cache_extent(extent_cache, 0);
4858                 if (!cache)
4859                         break;
4860                 rec = container_of(cache, struct extent_record, cache);
4861                 if (rec->num_duplicates) {
4862                         fprintf(stderr, "extent item %llu has multiple extent "
4863                                 "items\n", (unsigned long long)rec->start);
4864                         err = 1;
4865                 }
4866
4867                 if (rec->refs != rec->extent_item_refs) {
4868                         fprintf(stderr, "ref mismatch on [%llu %llu] ",
4869                                 (unsigned long long)rec->start,
4870                                 (unsigned long long)rec->nr);
4871                         fprintf(stderr, "extent item %llu, found %llu\n",
4872                                 (unsigned long long)rec->extent_item_refs,
4873                                 (unsigned long long)rec->refs);
4874                         if (!fixed && repair) {
4875                                 ret = fixup_extent_refs(trans, root->fs_info, rec);
4876                                 if (ret)
4877                                         goto repair_abort;
4878                                 fixed = 1;
4879                         }
4880                         err = 1;
4881
4882                 }
4883                 if (all_backpointers_checked(rec, 1)) {
4884                         fprintf(stderr, "backpointer mismatch on [%llu %llu]\n",
4885                                 (unsigned long long)rec->start,
4886                                 (unsigned long long)rec->nr);
4887
4888                         if (!fixed && repair) {
4889                                 ret = fixup_extent_refs(trans, root->fs_info, rec);
4890                                 if (ret)
4891                                         goto repair_abort;
4892                                 fixed = 1;
4893                         }
4894
4895                         err = 1;
4896                 }
4897                 if (!rec->owner_ref_checked) {
4898                         fprintf(stderr, "owner ref check failed [%llu %llu]\n",
4899                                 (unsigned long long)rec->start,
4900                                 (unsigned long long)rec->nr);
4901                         if (!fixed && repair) {
4902                                 ret = fixup_extent_refs(trans, root->fs_info, rec);
4903                                 if (ret)
4904                                         goto repair_abort;
4905                                 fixed = 1;
4906                         }
4907                         err = 1;
4908                 }
4909
4910                 remove_cache_extent(extent_cache, cache);
4911                 free_all_extent_backrefs(rec);
4912                 free(rec);
4913         }
4914 repair_abort:
4915         if (repair) {
4916                 if (ret && ret != -EAGAIN) {
4917                         fprintf(stderr, "failed to repair damaged filesystem, aborting\n");
4918                         exit(1);
4919                 } else if (!ret) {
4920                         btrfs_fix_block_accounting(trans, root);
4921                 }
4922                 if (err)
4923                         fprintf(stderr, "repaired damaged extent references\n");
4924                 return ret;
4925         }
4926         return err;
4927 }
4928
4929 u64 calc_stripe_length(u64 type, u64 length, int num_stripes)
4930 {
4931         u64 stripe_size;
4932
4933         if (type & BTRFS_BLOCK_GROUP_RAID0) {
4934                 stripe_size = length;
4935                 stripe_size /= num_stripes;
4936         } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
4937                 stripe_size = length * 2;
4938                 stripe_size /= num_stripes;
4939         } else if (type & BTRFS_BLOCK_GROUP_RAID5) {
4940                 stripe_size = length;
4941                 stripe_size /= (num_stripes - 1);
4942         } else if (type & BTRFS_BLOCK_GROUP_RAID6) {
4943                 stripe_size = length;
4944                 stripe_size /= (num_stripes - 2);
4945         } else {
4946                 stripe_size = length;
4947         }
4948         return stripe_size;
4949 }
4950
4951 static int check_chunk_refs(struct chunk_record *chunk_rec,
4952                             struct block_group_tree *block_group_cache,
4953                             struct device_extent_tree *dev_extent_cache,
4954                             int silent)
4955 {
4956         struct cache_extent *block_group_item;
4957         struct block_group_record *block_group_rec;
4958         struct cache_extent *dev_extent_item;
4959         struct device_extent_record *dev_extent_rec;
4960         u64 devid;
4961         u64 offset;
4962         u64 length;
4963         int i;
4964         int ret = 0;
4965
4966         block_group_item = lookup_cache_extent(&block_group_cache->tree,
4967                                                chunk_rec->offset,
4968                                                chunk_rec->length);
4969         if (block_group_item) {
4970                 block_group_rec = container_of(block_group_item,
4971                                                struct block_group_record,
4972                                                cache);
4973                 if (chunk_rec->length != block_group_rec->offset ||
4974                     chunk_rec->offset != block_group_rec->objectid ||
4975                     chunk_rec->type_flags != block_group_rec->flags) {
4976                         if (!silent)
4977                                 fprintf(stderr,
4978                                         "Chunk[%llu, %u, %llu]: length(%llu), offset(%llu), type(%llu) mismatch with block group[%llu, %u, %llu]: offset(%llu), objectid(%llu), flags(%llu)\n",
4979                                         chunk_rec->objectid,
4980                                         chunk_rec->type,
4981                                         chunk_rec->offset,
4982                                         chunk_rec->length,
4983                                         chunk_rec->offset,
4984                                         chunk_rec->type_flags,
4985                                         block_group_rec->objectid,
4986                                         block_group_rec->type,
4987                                         block_group_rec->offset,
4988                                         block_group_rec->offset,
4989                                         block_group_rec->objectid,
4990                                         block_group_rec->flags);
4991                         ret = -1;
4992                 } else {
4993                         list_del_init(&block_group_rec->list);
4994                         chunk_rec->bg_rec = block_group_rec;
4995                 }
4996         } else {
4997                 if (!silent)
4998                         fprintf(stderr,
4999                                 "Chunk[%llu, %u, %llu]: length(%llu), offset(%llu), type(%llu) is not found in block group\n",
5000                                 chunk_rec->objectid,
5001                                 chunk_rec->type,
5002                                 chunk_rec->offset,
5003                                 chunk_rec->length,
5004                                 chunk_rec->offset,
5005                                 chunk_rec->type_flags);
5006                 ret = -1;
5007         }
5008
5009         length = calc_stripe_length(chunk_rec->type_flags, chunk_rec->length,
5010                                     chunk_rec->num_stripes);
5011         for (i = 0; i < chunk_rec->num_stripes; ++i) {
5012                 devid = chunk_rec->stripes[i].devid;
5013                 offset = chunk_rec->stripes[i].offset;
5014                 dev_extent_item = lookup_cache_extent2(&dev_extent_cache->tree,
5015                                                        devid, offset, length);
5016                 if (dev_extent_item) {
5017                         dev_extent_rec = container_of(dev_extent_item,
5018                                                 struct device_extent_record,
5019                                                 cache);
5020                         if (dev_extent_rec->objectid != devid ||
5021                             dev_extent_rec->offset != offset ||
5022                             dev_extent_rec->chunk_offset != chunk_rec->offset ||
5023                             dev_extent_rec->length != length) {
5024                                 if (!silent)
5025                                         fprintf(stderr,
5026                                                 "Chunk[%llu, %u, %llu] stripe[%llu, %llu] dismatch dev extent[%llu, %llu, %llu]\n",
5027                                                 chunk_rec->objectid,
5028                                                 chunk_rec->type,
5029                                                 chunk_rec->offset,
5030                                                 chunk_rec->stripes[i].devid,
5031                                                 chunk_rec->stripes[i].offset,
5032                                                 dev_extent_rec->objectid,
5033                                                 dev_extent_rec->offset,
5034                                                 dev_extent_rec->length);
5035                                 ret = -1;
5036                         } else {
5037                                 list_move(&dev_extent_rec->chunk_list,
5038                                           &chunk_rec->dextents);
5039                         }
5040                 } else {
5041                         if (!silent)
5042                                 fprintf(stderr,
5043                                         "Chunk[%llu, %u, %llu] stripe[%llu, %llu] is not found in dev extent\n",
5044                                         chunk_rec->objectid,
5045                                         chunk_rec->type,
5046                                         chunk_rec->offset,
5047                                         chunk_rec->stripes[i].devid,
5048                                         chunk_rec->stripes[i].offset);
5049                         ret = -1;
5050                 }
5051         }
5052         return ret;
5053 }
5054
5055 /* check btrfs_chunk -> btrfs_dev_extent / btrfs_block_group_item */
5056 int check_chunks(struct cache_tree *chunk_cache,
5057                  struct block_group_tree *block_group_cache,
5058                  struct device_extent_tree *dev_extent_cache,
5059                  struct list_head *good, struct list_head *bad, int silent)
5060 {
5061         struct cache_extent *chunk_item;
5062         struct chunk_record *chunk_rec;
5063         struct block_group_record *bg_rec;
5064         struct device_extent_record *dext_rec;
5065         int err;
5066         int ret = 0;
5067
5068         chunk_item = first_cache_extent(chunk_cache);
5069         while (chunk_item) {
5070                 chunk_rec = container_of(chunk_item, struct chunk_record,
5071                                          cache);
5072                 err = check_chunk_refs(chunk_rec, block_group_cache,
5073                                        dev_extent_cache, silent);
5074                 if (err) {
5075                         ret = err;
5076                         if (bad)
5077                                 list_add_tail(&chunk_rec->list, bad);
5078                 } else {
5079                         if (good)
5080                                 list_add_tail(&chunk_rec->list, good);
5081                 }
5082
5083                 chunk_item = next_cache_extent(chunk_item);
5084         }
5085
5086         list_for_each_entry(bg_rec, &block_group_cache->block_groups, list) {
5087                 if (!silent)
5088                         fprintf(stderr,
5089                                 "Block group[%llu, %llu] (flags = %llu) didn't find the relative chunk.\n",
5090                                 bg_rec->objectid,
5091                                 bg_rec->offset,
5092                                 bg_rec->flags);
5093                 if (!ret)
5094                         ret = 1;
5095         }
5096
5097         list_for_each_entry(dext_rec, &dev_extent_cache->no_chunk_orphans,
5098                             chunk_list) {
5099                 if (!silent)
5100                         fprintf(stderr,
5101                                 "Device extent[%llu, %llu, %llu] didn't find the relative chunk.\n",
5102                                 dext_rec->objectid,
5103                                 dext_rec->offset,
5104                                 dext_rec->length);
5105                 if (!ret)
5106                         ret = 1;
5107         }
5108         return ret;
5109 }
5110
5111
5112 static int check_device_used(struct device_record *dev_rec,
5113                              struct device_extent_tree *dext_cache)
5114 {
5115         struct cache_extent *cache;
5116         struct device_extent_record *dev_extent_rec;
5117         u64 total_byte = 0;
5118
5119         cache = search_cache_extent2(&dext_cache->tree, dev_rec->devid, 0);
5120         while (cache) {
5121                 dev_extent_rec = container_of(cache,
5122                                               struct device_extent_record,
5123                                               cache);
5124                 if (dev_extent_rec->objectid != dev_rec->devid)
5125                         break;
5126
5127                 list_del(&dev_extent_rec->device_list);
5128                 total_byte += dev_extent_rec->length;
5129                 cache = next_cache_extent(cache);
5130         }
5131
5132         if (total_byte != dev_rec->byte_used) {
5133                 fprintf(stderr,
5134                         "Dev extent's total-byte(%llu) is not equal to byte-used(%llu) in dev[%llu, %u, %llu]\n",
5135                         total_byte, dev_rec->byte_used, dev_rec->objectid,
5136                         dev_rec->type, dev_rec->offset);
5137                 return -1;
5138         } else {
5139                 return 0;
5140         }
5141 }
5142
5143 /* check btrfs_dev_item -> btrfs_dev_extent */
5144 static int check_devices(struct rb_root *dev_cache,
5145                          struct device_extent_tree *dev_extent_cache)
5146 {
5147         struct rb_node *dev_node;
5148         struct device_record *dev_rec;
5149         struct device_extent_record *dext_rec;
5150         int err;
5151         int ret = 0;
5152
5153         dev_node = rb_first(dev_cache);
5154         while (dev_node) {
5155                 dev_rec = container_of(dev_node, struct device_record, node);
5156                 err = check_device_used(dev_rec, dev_extent_cache);
5157                 if (err)
5158                         ret = err;
5159
5160                 dev_node = rb_next(dev_node);
5161         }
5162         list_for_each_entry(dext_rec, &dev_extent_cache->no_device_orphans,
5163                             device_list) {
5164                 fprintf(stderr,
5165                         "Device extent[%llu, %llu, %llu] didn't find its device.\n",
5166                         dext_rec->objectid, dext_rec->offset, dext_rec->length);
5167                 if (!ret)
5168                         ret = 1;
5169         }
5170         return ret;
5171 }
5172
5173 static int check_chunks_and_extents(struct btrfs_root *root, int repair)
5174 {
5175         struct rb_root dev_cache;
5176         struct cache_tree chunk_cache;
5177         struct block_group_tree block_group_cache;
5178         struct device_extent_tree dev_extent_cache;
5179         struct cache_tree extent_cache;
5180         struct cache_tree seen;
5181         struct cache_tree pending;
5182         struct cache_tree reada;
5183         struct cache_tree nodes;
5184         struct cache_tree corrupt_blocks;
5185         struct btrfs_path path;
5186         struct btrfs_key key;
5187         struct btrfs_key found_key;
5188         int ret, err = 0;
5189         u64 last = 0;
5190         struct block_info *bits;
5191         int bits_nr;
5192         struct extent_buffer *leaf;
5193         struct btrfs_trans_handle *trans = NULL;
5194         int slot;
5195         struct btrfs_root_item ri;
5196
5197         dev_cache = RB_ROOT;
5198         cache_tree_init(&chunk_cache);
5199         block_group_tree_init(&block_group_cache);
5200         device_extent_tree_init(&dev_extent_cache);
5201
5202         cache_tree_init(&extent_cache);
5203         cache_tree_init(&seen);
5204         cache_tree_init(&pending);
5205         cache_tree_init(&nodes);
5206         cache_tree_init(&reada);
5207         cache_tree_init(&corrupt_blocks);
5208
5209         if (repair) {
5210                 trans = btrfs_start_transaction(root, 1);
5211                 if (IS_ERR(trans)) {
5212                         fprintf(stderr, "Error starting transaction\n");
5213                         return PTR_ERR(trans);
5214                 }
5215                 root->fs_info->fsck_extent_cache = &extent_cache;
5216                 root->fs_info->free_extent_hook = free_extent_hook;
5217                 root->fs_info->corrupt_blocks = &corrupt_blocks;
5218         }
5219
5220         bits_nr = 1024;
5221         bits = malloc(bits_nr * sizeof(struct block_info));
5222         if (!bits) {
5223                 perror("malloc");
5224                 exit(1);
5225         }
5226
5227 again:
5228         add_root_to_pending(root->fs_info->tree_root->node,
5229                             &extent_cache, &pending, &seen, &nodes,
5230                             &root->fs_info->tree_root->root_key);
5231
5232         add_root_to_pending(root->fs_info->chunk_root->node,
5233                             &extent_cache, &pending, &seen, &nodes,
5234                             &root->fs_info->chunk_root->root_key);
5235
5236         btrfs_init_path(&path);
5237         key.offset = 0;
5238         key.objectid = 0;
5239         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
5240         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
5241                                         &key, &path, 0, 0);
5242         BUG_ON(ret < 0);
5243         while(1) {
5244                 leaf = path.nodes[0];
5245                 slot = path.slots[0];
5246                 if (slot >= btrfs_header_nritems(path.nodes[0])) {
5247                         ret = btrfs_next_leaf(root, &path);
5248                         if (ret != 0)
5249                                 break;
5250                         leaf = path.nodes[0];
5251                         slot = path.slots[0];
5252                 }
5253                 btrfs_item_key_to_cpu(leaf, &found_key, path.slots[0]);
5254                 if (btrfs_key_type(&found_key) == BTRFS_ROOT_ITEM_KEY) {
5255                         unsigned long offset;
5256                         struct extent_buffer *buf;
5257
5258                         offset = btrfs_item_ptr_offset(leaf, path.slots[0]);
5259                         read_extent_buffer(leaf, &ri, offset, sizeof(ri));
5260                         buf = read_tree_block(root->fs_info->tree_root,
5261                                               btrfs_root_bytenr(&ri),
5262                                               btrfs_level_size(root,
5263                                                btrfs_root_level(&ri)), 0);
5264                         add_root_to_pending(buf, &extent_cache, &pending,
5265                                             &seen, &nodes, &found_key);
5266                         free_extent_buffer(buf);
5267                 }
5268                 path.slots[0]++;
5269         }
5270         btrfs_release_path(root, &path);
5271         while(1) {
5272                 ret = run_next_block(root, bits, bits_nr, &last, &pending,
5273                                      &seen, &reada, &nodes, &extent_cache,
5274                                      &chunk_cache, &dev_cache,
5275                                      &block_group_cache, &dev_extent_cache);
5276                 if (ret != 0)
5277                         break;
5278         }
5279
5280         ret = check_extent_refs(trans, root, &extent_cache, repair);
5281         if (ret == -EAGAIN) {
5282                 ret = btrfs_commit_transaction(trans, root);
5283                 if (ret)
5284                         goto out;
5285
5286                 trans = btrfs_start_transaction(root, 1);
5287                 if (IS_ERR(trans)) {
5288                         ret = PTR_ERR(trans);
5289                         goto out;
5290                 }
5291
5292                 free_corrupt_blocks_tree(root->fs_info->corrupt_blocks);
5293                 free_extent_cache_tree(&seen);
5294                 free_extent_cache_tree(&pending);
5295                 free_extent_cache_tree(&reada);
5296                 free_extent_cache_tree(&nodes);
5297                 free_extent_record_cache(root->fs_info, &extent_cache);
5298                 goto again;
5299         }
5300
5301         err = check_chunks(&chunk_cache, &block_group_cache,
5302                            &dev_extent_cache, NULL, NULL, 0);
5303         if (err && !ret)
5304                 ret = err;
5305
5306         err = check_devices(&dev_cache, &dev_extent_cache);
5307         if (err && !ret)
5308                 ret = err;
5309
5310         if (trans) {
5311                 int err;
5312
5313                 err = btrfs_commit_transaction(trans, root);
5314                 if (!ret)
5315                         ret = err;
5316         }
5317 out:
5318         if (repair) {
5319                 free_corrupt_blocks_tree(root->fs_info->corrupt_blocks);
5320                 root->fs_info->fsck_extent_cache = NULL;
5321                 root->fs_info->free_extent_hook = NULL;
5322                 root->fs_info->corrupt_blocks = NULL;
5323         }
5324         free(bits);
5325         free_chunk_cache_tree(&chunk_cache);
5326         free_device_cache_tree(&dev_cache);
5327         free_block_group_tree(&block_group_cache);
5328         free_device_extent_tree(&dev_extent_cache);
5329         return ret;
5330 }
5331
5332 static int pin_down_tree_blocks(struct btrfs_fs_info *fs_info,
5333                                 struct extent_buffer *eb, int tree_root)
5334 {
5335         struct extent_buffer *tmp;
5336         struct btrfs_root_item *ri;
5337         struct btrfs_key key;
5338         u64 bytenr;
5339         u32 leafsize;
5340         int level = btrfs_header_level(eb);
5341         int nritems;
5342         int ret;
5343         int i;
5344
5345         btrfs_pin_extent(fs_info, eb->start, eb->len);
5346
5347         leafsize = btrfs_super_leafsize(fs_info->super_copy);
5348         nritems = btrfs_header_nritems(eb);
5349         for (i = 0; i < nritems; i++) {
5350                 if (level == 0) {
5351                         btrfs_item_key_to_cpu(eb, &key, i);
5352                         if (key.type != BTRFS_ROOT_ITEM_KEY)
5353                                 continue;
5354                         /* Skip the extent root and reloc roots */
5355                         if (key.objectid == BTRFS_EXTENT_TREE_OBJECTID ||
5356                             key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
5357                             key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
5358                                 continue;
5359                         ri = btrfs_item_ptr(eb, i, struct btrfs_root_item);
5360                         bytenr = btrfs_disk_root_bytenr(eb, ri);
5361
5362                         /*
5363                          * If at any point we start needing the real root we
5364                          * will have to build a stump root for the root we are
5365                          * in, but for now this doesn't actually use the root so
5366                          * just pass in extent_root.
5367                          */
5368                         tmp = read_tree_block(fs_info->extent_root, bytenr,
5369                                               leafsize, 0);
5370                         if (!tmp) {
5371                                 fprintf(stderr, "Error reading root block\n");
5372                                 return -EIO;
5373                         }
5374                         ret = pin_down_tree_blocks(fs_info, tmp, 0);
5375                         free_extent_buffer(tmp);
5376                         if (ret)
5377                                 return ret;
5378                 } else {
5379                         bytenr = btrfs_node_blockptr(eb, i);
5380
5381                         /* If we aren't the tree root don't read the block */
5382                         if (level == 1 && !tree_root) {
5383                                 btrfs_pin_extent(fs_info, bytenr, leafsize);
5384                                 continue;
5385                         }
5386
5387                         tmp = read_tree_block(fs_info->extent_root, bytenr,
5388                                               leafsize, 0);
5389                         if (!tmp) {
5390                                 fprintf(stderr, "Error reading tree block\n");
5391                                 return -EIO;
5392                         }
5393                         ret = pin_down_tree_blocks(fs_info, tmp, tree_root);
5394                         free_extent_buffer(tmp);
5395                         if (ret)
5396                                 return ret;
5397                 }
5398         }
5399
5400         return 0;
5401 }
5402
5403 static int pin_metadata_blocks(struct btrfs_fs_info *fs_info)
5404 {
5405         int ret;
5406
5407         ret = pin_down_tree_blocks(fs_info, fs_info->chunk_root->node, 0);
5408         if (ret)
5409                 return ret;
5410
5411         return pin_down_tree_blocks(fs_info, fs_info->tree_root->node, 1);
5412 }
5413
5414 static int reset_block_groups(struct btrfs_fs_info *fs_info)
5415 {
5416         struct btrfs_path *path;
5417         struct extent_buffer *leaf;
5418         struct btrfs_chunk *chunk;
5419         struct btrfs_key key;
5420         int ret;
5421
5422         path = btrfs_alloc_path();
5423         if (!path)
5424                 return -ENOMEM;
5425
5426         key.objectid = 0;
5427         key.type = BTRFS_CHUNK_ITEM_KEY;
5428         key.offset = 0;
5429
5430         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
5431         if (ret < 0) {
5432                 btrfs_free_path(path);
5433                 return ret;
5434         }
5435
5436         /*
5437          * We do this in case the block groups were screwed up and had alloc
5438          * bits that aren't actually set on the chunks.  This happens with
5439          * restored images every time and could happen in real life I guess.
5440          */
5441         fs_info->avail_data_alloc_bits = 0;
5442         fs_info->avail_metadata_alloc_bits = 0;
5443         fs_info->avail_system_alloc_bits = 0;
5444
5445         /* First we need to create the in-memory block groups */
5446         while (1) {
5447                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
5448                         ret = btrfs_next_leaf(fs_info->chunk_root, path);
5449                         if (ret < 0) {
5450                                 btrfs_free_path(path);
5451                                 return ret;
5452                         }
5453                         if (ret) {
5454                                 ret = 0;
5455                                 break;
5456                         }
5457                 }
5458                 leaf = path->nodes[0];
5459                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5460                 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
5461                         path->slots[0]++;
5462                         continue;
5463                 }
5464
5465                 chunk = btrfs_item_ptr(leaf, path->slots[0],
5466                                        struct btrfs_chunk);
5467                 btrfs_add_block_group(fs_info, 0,
5468                                       btrfs_chunk_type(leaf, chunk),
5469                                       key.objectid, key.offset,
5470                                       btrfs_chunk_length(leaf, chunk));
5471                 path->slots[0]++;
5472         }
5473
5474         btrfs_free_path(path);
5475         return 0;
5476 }
5477
5478 static int reset_balance(struct btrfs_trans_handle *trans,
5479                          struct btrfs_fs_info *fs_info)
5480 {
5481         struct btrfs_root *root = fs_info->tree_root;
5482         struct btrfs_path *path;
5483         struct extent_buffer *leaf;
5484         struct btrfs_key key;
5485         int del_slot, del_nr = 0;
5486         int ret;
5487         int found = 0;
5488
5489         path = btrfs_alloc_path();
5490         if (!path)
5491                 return -ENOMEM;
5492
5493         key.objectid = BTRFS_BALANCE_OBJECTID;
5494         key.type = BTRFS_BALANCE_ITEM_KEY;
5495         key.offset = 0;
5496
5497         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
5498         if (ret) {
5499                 if (ret > 0)
5500                         ret = 0;
5501                 goto out;
5502         }
5503
5504         ret = btrfs_del_item(trans, root, path);
5505         if (ret)
5506                 goto out;
5507         btrfs_release_path(root, path);
5508
5509         key.objectid = BTRFS_TREE_RELOC_OBJECTID;
5510         key.type = BTRFS_ROOT_ITEM_KEY;
5511         key.offset = 0;
5512
5513         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
5514         if (ret < 0)
5515                 goto out;
5516         while (1) {
5517                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
5518                         if (!found)
5519                                 break;
5520
5521                         if (del_nr) {
5522                                 ret = btrfs_del_items(trans, root, path,
5523                                                       del_slot, del_nr);
5524                                 del_nr = 0;
5525                                 if (ret)
5526                                         goto out;
5527                         }
5528                         key.offset++;
5529                         btrfs_release_path(root, path);
5530
5531                         found = 0;
5532                         ret = btrfs_search_slot(trans, root, &key, path,
5533                                                 -1, 1);
5534                         if (ret < 0)
5535                                 goto out;
5536                         continue;
5537                 }
5538                 found = 1;
5539                 leaf = path->nodes[0];
5540                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5541                 if (key.objectid > BTRFS_TREE_RELOC_OBJECTID)
5542                         break;
5543                 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
5544                         path->slots[0]++;
5545                         continue;
5546                 }
5547                 if (!del_nr) {
5548                         del_slot = path->slots[0];
5549                         del_nr = 1;
5550                 } else {
5551                         del_nr++;
5552                 }
5553                 path->slots[0]++;
5554         }
5555
5556         if (del_nr) {
5557                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
5558                 if (ret)
5559                         goto out;
5560         }
5561         btrfs_release_path(root, path);
5562
5563         key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5564         key.type = BTRFS_ROOT_ITEM_KEY;
5565         key.offset = (u64)-1;
5566         root = btrfs_read_fs_root(fs_info, &key);
5567         if (IS_ERR(root)) {
5568                 fprintf(stderr, "Error reading data reloc tree\n");
5569                 return PTR_ERR(root);
5570         }
5571         root->track_dirty = 1;
5572         if (root->last_trans != trans->transid) {
5573                 root->last_trans = trans->transid;
5574                 root->commit_root = root->node;
5575                 extent_buffer_get(root->node);
5576         }
5577         ret = btrfs_fsck_reinit_root(trans, root, 0);
5578 out:
5579         btrfs_free_path(path);
5580         return ret;
5581 }
5582
5583 static int reinit_extent_tree(struct btrfs_fs_info *fs_info)
5584 {
5585         struct btrfs_trans_handle *trans;
5586         u64 start = 0;
5587         int ret;
5588
5589         /*
5590          * The only reason we don't do this is because right now we're just
5591          * walking the trees we find and pinning down their bytes, we don't look
5592          * at any of the leaves.  In order to do mixed groups we'd have to check
5593          * the leaves of any fs roots and pin down the bytes for any file
5594          * extents we find.  Not hard but why do it if we don't have to?
5595          */
5596         if (btrfs_fs_incompat(fs_info, BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)) {
5597                 fprintf(stderr, "We don't support re-initing the extent tree "
5598                         "for mixed block groups yet, please notify a btrfs "
5599                         "developer you want to do this so they can add this "
5600                         "functionality.\n");
5601                 return -EINVAL;
5602         }
5603
5604         trans = btrfs_start_transaction(fs_info->extent_root, 1);
5605         if (IS_ERR(trans)) {
5606                 fprintf(stderr, "Error starting transaction\n");
5607                 return PTR_ERR(trans);
5608         }
5609
5610         /*
5611          * first we need to walk all of the trees except the extent tree and pin
5612          * down the bytes that are in use so we don't overwrite any existing
5613          * metadata.
5614          */
5615         ret = pin_metadata_blocks(fs_info);
5616         if (ret) {
5617                 fprintf(stderr, "error pinning down used bytes\n");
5618                 return ret;
5619         }
5620
5621         /*
5622          * Need to drop all the block groups since we're going to recreate all
5623          * of them again.
5624          */
5625         btrfs_free_block_groups(fs_info);
5626         ret = reset_block_groups(fs_info);
5627         if (ret) {
5628                 fprintf(stderr, "error resetting the block groups\n");
5629                 return ret;
5630         }
5631
5632         /* Ok we can allocate now, reinit the extent root */
5633         ret = btrfs_fsck_reinit_root(trans, fs_info->extent_root, 1);
5634         if (ret) {
5635                 fprintf(stderr, "extent root initialization failed\n");
5636                 /*
5637                  * When the transaction code is updated we should end the
5638                  * transaction, but for now progs only knows about commit so
5639                  * just return an error.
5640                  */
5641                 return ret;
5642         }
5643
5644         ret = reset_balance(trans, fs_info);
5645         if (ret) {
5646                 fprintf(stderr, "error reseting the pending balance\n");
5647                 return ret;
5648         }
5649
5650         /*
5651          * Now we have all the in-memory block groups setup so we can make
5652          * allocations properly, and the metadata we care about is safe since we
5653          * pinned all of it above.
5654          */
5655         while (1) {
5656                 struct btrfs_block_group_cache *cache;
5657
5658                 cache = btrfs_lookup_first_block_group(fs_info, start);
5659                 if (!cache)
5660                         break;
5661                 start = cache->key.objectid + cache->key.offset;
5662                 ret = btrfs_insert_item(trans, fs_info->extent_root,
5663                                         &cache->key, &cache->item,
5664                                         sizeof(cache->item));
5665                 if (ret) {
5666                         fprintf(stderr, "Error adding block group\n");
5667                         return ret;
5668                 }
5669                 btrfs_extent_post_op(trans, fs_info->extent_root);
5670         }
5671
5672         /*
5673          * Ok now we commit and run the normal fsck, which will add extent
5674          * entries for all of the items it finds.
5675          */
5676         return btrfs_commit_transaction(trans, fs_info->extent_root);
5677 }
5678
5679 static struct option long_options[] = {
5680         { "super", 1, NULL, 's' },
5681         { "repair", 0, NULL, 0 },
5682         { "init-csum-tree", 0, NULL, 0 },
5683         { "init-extent-tree", 0, NULL, 0 },
5684         { 0, 0, 0, 0}
5685 };
5686
5687 const char * const cmd_check_usage[] = {
5688         "btrfs check [options] <device>",
5689         "Check an unmounted btrfs filesystem.",
5690         "",
5691         "-s|--super <superblock>     use this superblock copy",
5692         "--repair                    try to repair the filesystem",
5693         "--init-csum-tree            create a new CRC tree",
5694         "--init-extent-tree          create a new extent tree",
5695         NULL
5696 };
5697
5698 int cmd_check(int argc, char **argv)
5699 {
5700         struct cache_tree root_cache;
5701         struct btrfs_root *root;
5702         struct btrfs_fs_info *info;
5703         u64 bytenr = 0;
5704         char uuidbuf[37];
5705         int ret;
5706         int num;
5707         int repair = 0;
5708         int option_index = 0;
5709         int init_csum_tree = 0;
5710         int init_extent_tree = 0;
5711         int rw = 0;
5712
5713         while(1) {
5714                 int c;
5715                 c = getopt_long(argc, argv, "as:", long_options,
5716                                 &option_index);
5717                 if (c < 0)
5718                         break;
5719                 switch(c) {
5720                         case 'a': /* ignored */ break;
5721                         case 's':
5722                                 num = atol(optarg);
5723                                 bytenr = btrfs_sb_offset(num);
5724                                 printf("using SB copy %d, bytenr %llu\n", num,
5725                                        (unsigned long long)bytenr);
5726                                 break;
5727                         case '?':
5728                         case 'h':
5729                                 usage(cmd_check_usage);
5730                 }
5731                 if (option_index == 1) {
5732                         printf("enabling repair mode\n");
5733                         repair = 1;
5734                         rw = 1;
5735                 } else if (option_index == 2) {
5736                         printf("Creating a new CRC tree\n");
5737                         init_csum_tree = 1;
5738                         rw = 1;
5739                 } else if (option_index == 3) {
5740                         init_extent_tree = 1;
5741                         rw = 1;
5742                         repair = 1;
5743                 }
5744
5745         }
5746         argc = argc - optind;
5747
5748         if (argc != 1)
5749                 usage(cmd_check_usage);
5750
5751         radix_tree_init();
5752         cache_tree_init(&root_cache);
5753
5754         if((ret = check_mounted(argv[optind])) < 0) {
5755                 fprintf(stderr, "Could not check mount status: %s\n", strerror(-ret));
5756                 return ret;
5757         } else if(ret) {
5758                 fprintf(stderr, "%s is currently mounted. Aborting.\n", argv[optind]);
5759                 return -EBUSY;
5760         }
5761
5762         info = open_ctree_fs_info(argv[optind], bytenr, 0, rw, 1);
5763         if (!info) {
5764                 fprintf(stderr, "Couldn't open file system\n");
5765                 return -EIO;
5766         }
5767
5768         uuid_unparse(info->super_copy->fsid, uuidbuf);
5769         printf("Checking filesystem on %s\nUUID: %s\n", argv[optind], uuidbuf);
5770
5771         if (!extent_buffer_uptodate(info->tree_root->node) ||
5772             !extent_buffer_uptodate(info->dev_root->node) ||
5773             !extent_buffer_uptodate(info->extent_root->node) ||
5774             !extent_buffer_uptodate(info->chunk_root->node)) {
5775                 fprintf(stderr, "Critical roots corrupted, unable to fsck the FS\n");
5776                 return -EIO;
5777         }
5778
5779         root = info->fs_root;
5780
5781         if (init_extent_tree) {
5782                 printf("Creating a new extent tree\n");
5783                 ret = reinit_extent_tree(info);
5784                 if (ret)
5785                         return ret;
5786         }
5787         fprintf(stderr, "checking extents\n");
5788         if (init_csum_tree) {
5789                 struct btrfs_trans_handle *trans;
5790
5791                 fprintf(stderr, "Reinit crc root\n");
5792                 trans = btrfs_start_transaction(info->csum_root, 1);
5793                 if (IS_ERR(trans)) {
5794                         fprintf(stderr, "Error starting transaction\n");
5795                         return PTR_ERR(trans);
5796                 }
5797
5798                 ret = btrfs_fsck_reinit_root(trans, info->csum_root, 0);
5799                 if (ret) {
5800                         fprintf(stderr, "crc root initialization failed\n");
5801                         return -EIO;
5802                 }
5803
5804                 ret = btrfs_commit_transaction(trans, root);
5805                 if (ret)
5806                         exit(1);
5807                 goto out;
5808         }
5809         ret = check_chunks_and_extents(root, repair);
5810         if (ret)
5811                 fprintf(stderr, "Errors found in extent allocation tree or chunk allocation\n");
5812
5813         fprintf(stderr, "checking free space cache\n");
5814         ret = check_space_cache(root);
5815         if (ret)
5816                 goto out;
5817
5818         fprintf(stderr, "checking fs roots\n");
5819         ret = check_fs_roots(root, &root_cache);
5820         if (ret)
5821                 goto out;
5822
5823         fprintf(stderr, "checking csums\n");
5824         ret = check_csums(root);
5825         if (ret)
5826                 goto out;
5827
5828         fprintf(stderr, "checking root refs\n");
5829         ret = check_root_refs(root, &root_cache);
5830 out:
5831         free_root_recs_tree(&root_cache);
5832         close_ctree(root);
5833
5834         if (found_old_backref) { /*
5835                  * there was a disk format change when mixed
5836                  * backref was in testing tree. The old format
5837                  * existed about one week.
5838                  */
5839                 printf("\n * Found old mixed backref format. "
5840                        "The old format is not supported! *"
5841                        "\n * Please mount the FS in readonly mode, "
5842                        "backup data and re-format the FS. *\n\n");
5843                 ret = 1;
5844         }
5845         printf("found %llu bytes used err is %d\n",
5846                (unsigned long long)bytes_used, ret);
5847         printf("total csum bytes: %llu\n",(unsigned long long)total_csum_bytes);
5848         printf("total tree bytes: %llu\n",
5849                (unsigned long long)total_btree_bytes);
5850         printf("total fs tree bytes: %llu\n",
5851                (unsigned long long)total_fs_tree_bytes);
5852         printf("total extent tree bytes: %llu\n",
5853                (unsigned long long)total_extent_tree_bytes);
5854         printf("btree space waste bytes: %llu\n",
5855                (unsigned long long)btree_space_waste);
5856         printf("file data blocks allocated: %llu\n referenced %llu\n",
5857                 (unsigned long long)data_bytes_allocated,
5858                 (unsigned long long)data_bytes_referenced);
5859         printf("%s\n", BTRFS_BUILD_VERSION);
5860         return ret;
5861 }