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 #include <stdio.h>
  20 #include <stdlib.h>
  21 #include <unistd.h>
  22 #include <fcntl.h>
  23 #include <sys/types.h>
  24 #include <sys/stat.h>
  25 #include <unistd.h>
  26 #include <getopt.h>
  27 #include <uuid/uuid.h>
  28 #include "ctree.h"
  29 #include "volumes.h"
  30 #include "repair.h"
  31 #include "disk-io.h"
  32 #include "print-tree.h"
  33 #include "task-utils.h"
  34 #include "transaction.h"
  35 #include "utils.h"
  36 #include "commands.h"
  37 #include "free-space-cache.h"
  38 #include "free-space-tree.h"
  39 #include "btrfsck.h"
  40 #include "qgroup-verify.h"
  41 #include "rbtree-utils.h"
  42 #include "backref.h"
  43 #include "kernel-shared/ulist.h"
  44 #include "hash.h"
  45 #include "help.h"
  46
  47 enum task_position {
  48         TASK_EXTENTS,
  49         TASK_FREE_SPACE,
  50         TASK_FS_ROOTS,
  51         TASK_NOTHING, /* have to be the last element */
  52 };
  53
  54 struct task_ctx {
  55         int progress_enabled;
  56         enum task_position tp;
  57
  58         struct task_info *info;
  59 };
  60
  61 static u64 bytes_used = 0;
  62 static u64 total_csum_bytes = 0;
  63 static u64 total_btree_bytes = 0;
  64 static u64 total_fs_tree_bytes = 0;
  65 static u64 total_extent_tree_bytes = 0;
  66 static u64 btree_space_waste = 0;
  67 static u64 data_bytes_allocated = 0;
  68 static u64 data_bytes_referenced = 0;
  69 static LIST_HEAD(duplicate_extents);
  70 static LIST_HEAD(delete_items);
  71 static int no_holes = 0;
  72 static int init_extent_tree = 0;
  73 static int check_data_csum = 0;
  74 static struct btrfs_fs_info *global_info;
  75 static struct task_ctx ctx = { 0 };
  76 static struct cache_tree *roots_info_cache = NULL;
  77
  78 enum btrfs_check_mode {
  79         CHECK_MODE_ORIGINAL,
  80         CHECK_MODE_LOWMEM,
  81         CHECK_MODE_UNKNOWN,
  82         CHECK_MODE_DEFAULT = CHECK_MODE_ORIGINAL
  83 };
  84
  85 static enum btrfs_check_mode check_mode = CHECK_MODE_DEFAULT;
  86
  87 struct extent_backref {
  88         struct rb_node node;
  89         unsigned int is_data:1;
  90         unsigned int found_extent_tree:1;
  91         unsigned int full_backref:1;
  92         unsigned int found_ref:1;
  93         unsigned int broken:1;
  94 };
  95
  96 static inline struct extent_backref* rb_node_to_extent_backref(struct rb_node *node)
  97 {
  98         return rb_entry(node, struct extent_backref, node);
  99 }
 100
 101 struct data_backref {
 102         struct extent_backref node;
 103         union {
 104                 u64 parent;
 105                 u64 root;
 106         };
 107         u64 owner;
 108         u64 offset;
 109         u64 disk_bytenr;
 110         u64 bytes;
 111         u64 ram_bytes;
 112         u32 num_refs;
 113         u32 found_ref;
 114 };
 115
 116 #define ROOT_DIR_ERROR          (1<<1)  /* bad ROOT_DIR */
 117 #define DIR_ITEM_MISSING        (1<<2)  /* DIR_ITEM not found */
 118 #define DIR_ITEM_MISMATCH       (1<<3)  /* DIR_ITEM found but not match */
 119 #define INODE_REF_MISSING       (1<<4)  /* INODE_REF/INODE_EXTREF not found */
 120 #define INODE_ITEM_MISSING      (1<<5)  /* INODE_ITEM not found */
 121 #define INODE_ITEM_MISMATCH     (1<<6)  /* INODE_ITEM found but not match */
 122 #define FILE_EXTENT_ERROR       (1<<7)  /* bad FILE_EXTENT */
 123 #define ODD_CSUM_ITEM           (1<<8)  /* CSUM_ITEM error */
 124 #define CSUM_ITEM_MISSING       (1<<9)  /* CSUM_ITEM not found */
 125 #define LINK_COUNT_ERROR        (1<<10) /* INODE_ITEM nlink count error */
 126 #define NBYTES_ERROR            (1<<11) /* INODE_ITEM nbytes count error */
 127 #define ISIZE_ERROR             (1<<12) /* INODE_ITEM size count error */
 128 #define ORPHAN_ITEM             (1<<13) /* INODE_ITEM no reference */
 129 #define NO_INODE_ITEM           (1<<14) /* no inode_item */
 130 #define LAST_ITEM               (1<<15) /* Complete this tree traversal */
 131 #define ROOT_REF_MISSING        (1<<16) /* ROOT_REF not found */
 132 #define ROOT_REF_MISMATCH       (1<<17) /* ROOT_REF found but not match */
 133 #define DIR_INDEX_MISSING       (1<<18) /* INODE_INDEX not found */
 134 #define DIR_INDEX_MISMATCH      (1<<19) /* INODE_INDEX found but not match */
 135 #define DIR_COUNT_AGAIN         (1<<20) /* DIR isize should be recalculated */
 136 #define BG_ACCOUNTING_ERROR     (1<<21) /* Block group accounting error */
 137
 138 static inline struct data_backref* to_data_backref(struct extent_backref *back)
 139 {
 140         return container_of(back, struct data_backref, node);
 141 }
 142
 143 static int compare_data_backref(struct rb_node *node1, struct rb_node *node2)
 144 {
 145         struct extent_backref *ext1 = rb_node_to_extent_backref(node1);
 146         struct extent_backref *ext2 = rb_node_to_extent_backref(node2);
 147         struct data_backref *back1 = to_data_backref(ext1);
 148         struct data_backref *back2 = to_data_backref(ext2);
 149
 150         WARN_ON(!ext1->is_data);
 151         WARN_ON(!ext2->is_data);
 152
 153         /* parent and root are a union, so this covers both */
 154         if (back1->parent > back2->parent)
 155                 return 1;
 156         if (back1->parent < back2->parent)
 157                 return -1;
 158
 159         /* This is a full backref and the parents match. */
 160         if (back1->node.full_backref)
 161                 return 0;
 162
 163         if (back1->owner > back2->owner)
 164                 return 1;
 165         if (back1->owner < back2->owner)
 166                 return -1;
 167
 168         if (back1->offset > back2->offset)
 169                 return 1;
 170         if (back1->offset < back2->offset)
 171                 return -1;
 172
 173         if (back1->found_ref && back2->found_ref) {
 174                 if (back1->disk_bytenr > back2->disk_bytenr)
 175                         return 1;
 176                 if (back1->disk_bytenr < back2->disk_bytenr)
 177                         return -1;
 178
 179                 if (back1->bytes > back2->bytes)
 180                         return 1;
 181                 if (back1->bytes < back2->bytes)
 182                         return -1;
 183         }
 184
 185         return 0;
 186 }
 187
 188 /*
 189  * Much like data_backref, just removed the undetermined members
 190  * and change it to use list_head.
 191  * During extent scan, it is stored in root->orphan_data_extent.
 192  * During fs tree scan, it is then moved to inode_rec->orphan_data_extents.
 193  */
 194 struct orphan_data_extent {
 195         struct list_head list;
 196         u64 root;
 197         u64 objectid;
 198         u64 offset;
 199         u64 disk_bytenr;
 200         u64 disk_len;
 201 };
 202
 203 struct tree_backref {
 204         struct extent_backref node;
 205         union {
 206                 u64 parent;
 207                 u64 root;
 208         };
 209 };
 210
 211 static inline struct tree_backref* to_tree_backref(struct extent_backref *back)
 212 {
 213         return container_of(back, struct tree_backref, node);
 214 }
 215
 216 static int compare_tree_backref(struct rb_node *node1, struct rb_node *node2)
 217 {
 218         struct extent_backref *ext1 = rb_node_to_extent_backref(node1);
 219         struct extent_backref *ext2 = rb_node_to_extent_backref(node2);
 220         struct tree_backref *back1 = to_tree_backref(ext1);
 221         struct tree_backref *back2 = to_tree_backref(ext2);
 222
 223         WARN_ON(ext1->is_data);
 224         WARN_ON(ext2->is_data);
 225
 226         /* parent and root are a union, so this covers both */
 227         if (back1->parent > back2->parent)
 228                 return 1;
 229         if (back1->parent < back2->parent)
 230                 return -1;
 231
 232         return 0;
 233 }
 234
 235 static int compare_extent_backref(struct rb_node *node1, struct rb_node *node2)
 236 {
 237         struct extent_backref *ext1 = rb_node_to_extent_backref(node1);
 238         struct extent_backref *ext2 = rb_node_to_extent_backref(node2);
 239
 240         if (ext1->is_data > ext2->is_data)
 241                 return 1;
 242
 243         if (ext1->is_data < ext2->is_data)
 244                 return -1;
 245
 246         if (ext1->full_backref > ext2->full_backref)
 247                 return 1;
 248         if (ext1->full_backref < ext2->full_backref)
 249                 return -1;
 250
 251         if (ext1->is_data)
 252                 return compare_data_backref(node1, node2);
 253         else
 254                 return compare_tree_backref(node1, node2);
 255 }
 256
 257 /* Explicit initialization for extent_record::flag_block_full_backref */
 258 enum { FLAG_UNSET = 2 };
 259
 260 struct extent_record {
 261         struct list_head backrefs;
 262         struct list_head dups;
 263         struct rb_root backref_tree;
 264         struct list_head list;
 265         struct cache_extent cache;
 266         struct btrfs_disk_key parent_key;
 267         u64 start;
 268         u64 max_size;
 269         u64 nr;
 270         u64 refs;
 271         u64 extent_item_refs;
 272         u64 generation;
 273         u64 parent_generation;
 274         u64 info_objectid;
 275         u32 num_duplicates;
 276         u8 info_level;
 277         unsigned int flag_block_full_backref:2;
 278         unsigned int found_rec:1;
 279         unsigned int content_checked:1;
 280         unsigned int owner_ref_checked:1;
 281         unsigned int is_root:1;
 282         unsigned int metadata:1;
 283         unsigned int bad_full_backref:1;
 284         unsigned int crossing_stripes:1;
 285         unsigned int wrong_chunk_type:1;
 286 };
 287
 288 static inline struct extent_record* to_extent_record(struct list_head *entry)
 289 {
 290         return container_of(entry, struct extent_record, list);
 291 }
 292
 293 struct inode_backref {
 294         struct list_head list;
 295         unsigned int found_dir_item:1;
 296         unsigned int found_dir_index:1;
 297         unsigned int found_inode_ref:1;
 298         u8 filetype;
 299         u8 ref_type;
 300         int errors;
 301         u64 dir;
 302         u64 index;
 303         u16 namelen;
 304         char name[0];
 305 };
 306
 307 static inline struct inode_backref* to_inode_backref(struct list_head *entry)
 308 {
 309         return list_entry(entry, struct inode_backref, list);
 310 }
 311
 312 struct root_item_record {
 313         struct list_head list;
 314         u64 objectid;
 315         u64 bytenr;
 316         u64 last_snapshot;
 317         u8 level;
 318         u8 drop_level;
 319         struct btrfs_key drop_key;
 320 };
 321
 322 #define REF_ERR_NO_DIR_ITEM             (1 << 0)
 323 #define REF_ERR_NO_DIR_INDEX            (1 << 1)
 324 #define REF_ERR_NO_INODE_REF            (1 << 2)
 325 #define REF_ERR_DUP_DIR_ITEM            (1 << 3)
 326 #define REF_ERR_DUP_DIR_INDEX           (1 << 4)
 327 #define REF_ERR_DUP_INODE_REF           (1 << 5)
 328 #define REF_ERR_INDEX_UNMATCH           (1 << 6)
 329 #define REF_ERR_FILETYPE_UNMATCH        (1 << 7)
 330 #define REF_ERR_NAME_TOO_LONG           (1 << 8) // 100
 331 #define REF_ERR_NO_ROOT_REF             (1 << 9)
 332 #define REF_ERR_NO_ROOT_BACKREF         (1 << 10)
 333 #define REF_ERR_DUP_ROOT_REF            (1 << 11)
 334 #define REF_ERR_DUP_ROOT_BACKREF        (1 << 12)
 335
 336 struct file_extent_hole {
 337         struct rb_node node;
 338         u64 start;
 339         u64 len;
 340 };
 341
 342 struct inode_record {
 343         struct list_head backrefs;
 344         unsigned int checked:1;
 345         unsigned int merging:1;
 346         unsigned int found_inode_item:1;
 347         unsigned int found_dir_item:1;
 348         unsigned int found_file_extent:1;
 349         unsigned int found_csum_item:1;
 350         unsigned int some_csum_missing:1;
 351         unsigned int nodatasum:1;
 352         int errors;
 353
 354         u64 ino;
 355         u32 nlink;
 356         u32 imode;
 357         u64 isize;
 358         u64 nbytes;
 359
 360         u32 found_link;
 361         u64 found_size;
 362         u64 extent_start;
 363         u64 extent_end;
 364         struct rb_root holes;
 365         struct list_head orphan_extents;
 366
 367         u32 refs;
 368 };
 369
 370 #define I_ERR_NO_INODE_ITEM             (1 << 0)
 371 #define I_ERR_NO_ORPHAN_ITEM            (1 << 1)
 372 #define I_ERR_DUP_INODE_ITEM            (1 << 2)
 373 #define I_ERR_DUP_DIR_INDEX             (1 << 3)
 374 #define I_ERR_ODD_DIR_ITEM              (1 << 4)
 375 #define I_ERR_ODD_FILE_EXTENT           (1 << 5)
 376 #define I_ERR_BAD_FILE_EXTENT           (1 << 6)
 377 #define I_ERR_FILE_EXTENT_OVERLAP       (1 << 7)
 378 #define I_ERR_FILE_EXTENT_DISCOUNT      (1 << 8) // 100
 379 #define I_ERR_DIR_ISIZE_WRONG           (1 << 9)
 380 #define I_ERR_FILE_NBYTES_WRONG         (1 << 10) // 400
 381 #define I_ERR_ODD_CSUM_ITEM             (1 << 11)
 382 #define I_ERR_SOME_CSUM_MISSING         (1 << 12)
 383 #define I_ERR_LINK_COUNT_WRONG          (1 << 13)
 384 #define I_ERR_FILE_EXTENT_ORPHAN        (1 << 14)
 385
 386 struct root_backref {
 387         struct list_head list;
 388         unsigned int found_dir_item:1;
 389         unsigned int found_dir_index:1;
 390         unsigned int found_back_ref:1;
 391         unsigned int found_forward_ref:1;
 392         unsigned int reachable:1;
 393         int errors;
 394         u64 ref_root;
 395         u64 dir;
 396         u64 index;
 397         u16 namelen;
 398         char name[0];
 399 };
 400
 401 static inline struct root_backref* to_root_backref(struct list_head *entry)
 402 {
 403         return list_entry(entry, struct root_backref, list);
 404 }
 405
 406 struct root_record {
 407         struct list_head backrefs;
 408         struct cache_extent cache;
 409         unsigned int found_root_item:1;
 410         u64 objectid;
 411         u32 found_ref;
 412 };
 413
 414 struct ptr_node {
 415         struct cache_extent cache;
 416         void *data;
 417 };
 418
 419 struct shared_node {
 420         struct cache_extent cache;
 421         struct cache_tree root_cache;
 422         struct cache_tree inode_cache;
 423         struct inode_record *current;
 424         u32 refs;
 425 };
 426
 427 struct block_info {
 428         u64 start;
 429         u32 size;
 430 };
 431
 432 struct walk_control {
 433         struct cache_tree shared;
 434         struct shared_node *nodes[BTRFS_MAX_LEVEL];
 435         int active_node;
 436         int root_level;
 437 };
 438
 439 struct bad_item {
 440         struct btrfs_key key;
 441         u64 root_id;
 442         struct list_head list;
 443 };
 444
 445 struct extent_entry {
 446         u64 bytenr;
 447         u64 bytes;
 448         int count;
 449         int broken;
 450         struct list_head list;
 451 };
 452
 453 struct root_item_info {
 454         /* level of the root */
 455         u8 level;
 456         /* number of nodes at this level, must be 1 for a root */
 457         int node_count;
 458         u64 bytenr;
 459         u64 gen;
 460         struct cache_extent cache_extent;
 461 };
 462
 463 /*
 464  * Error bit for low memory mode check.
 465  *
 466  * Currently no caller cares about it yet.  Just internal use for error
 467  * classification.
 468  */
 469 #define BACKREF_MISSING         (1 << 0) /* Backref missing in extent tree */
 470 #define BACKREF_MISMATCH        (1 << 1) /* Backref exists but does not match */
 471 #define BYTES_UNALIGNED         (1 << 2) /* Some bytes are not aligned */
 472 #define REFERENCER_MISSING      (1 << 3) /* Referencer not found */
 473 #define REFERENCER_MISMATCH     (1 << 4) /* Referenceer found but does not match */
 474 #define CROSSING_STRIPE_BOUNDARY (1 << 4) /* For kernel scrub workaround */
 475 #define ITEM_SIZE_MISMATCH      (1 << 5) /* Bad item size */
 476 #define UNKNOWN_TYPE            (1 << 6) /* Unknown type */
 477 #define ACCOUNTING_MISMATCH     (1 << 7) /* Used space accounting error */
 478 #define CHUNK_TYPE_MISMATCH     (1 << 8)
 479
 480 static void *print_status_check(void *p)
 481 {
 482         struct task_ctx *priv = p;
 483         const char work_indicator[] = { '.', 'o', 'O', 'o' };
 484         uint32_t count = 0;
 485         static char *task_position_string[] = {
 486                 "checking extents",
 487                 "checking free space cache",
 488                 "checking fs roots",
 489         };
 490
 491         task_period_start(priv->info, 1000 /* 1s */);
 492
 493         if (priv->tp == TASK_NOTHING)
 494                 return NULL;
 495
 496         while (1) {
 497                 printf("%s [%c]\r", task_position_string[priv->tp],
 498                                 work_indicator[count % 4]);
 499                 count++;
 500                 fflush(stdout);
 501                 task_period_wait(priv->info);
 502         }
 503         return NULL;
 504 }
 505
 506 static int print_status_return(void *p)
 507 {
 508         printf("\n");
 509         fflush(stdout);
 510
 511         return 0;
 512 }
 513
 514 static enum btrfs_check_mode parse_check_mode(const char *str)
 515 {
 516         if (strcmp(str, "lowmem") == 0)
 517                 return CHECK_MODE_LOWMEM;
 518         if (strcmp(str, "orig") == 0)
 519                 return CHECK_MODE_ORIGINAL;
 520         if (strcmp(str, "original") == 0)
 521                 return CHECK_MODE_ORIGINAL;
 522
 523         return CHECK_MODE_UNKNOWN;
 524 }
 525
 526 /* Compatible function to allow reuse of old codes */
 527 static u64 first_extent_gap(struct rb_root *holes)
 528 {
 529         struct file_extent_hole *hole;
 530
 531         if (RB_EMPTY_ROOT(holes))
 532                 return (u64)-1;
 533
 534         hole = rb_entry(rb_first(holes), struct file_extent_hole, node);
 535         return hole->start;
 536 }
 537
 538 static int compare_hole(struct rb_node *node1, struct rb_node *node2)
 539 {
 540         struct file_extent_hole *hole1;
 541         struct file_extent_hole *hole2;
 542
 543         hole1 = rb_entry(node1, struct file_extent_hole, node);
 544         hole2 = rb_entry(node2, struct file_extent_hole, node);
 545
 546         if (hole1->start > hole2->start)
 547                 return -1;
 548         if (hole1->start < hole2->start)
 549                 return 1;
 550         /* Now hole1->start == hole2->start */
 551         if (hole1->len >= hole2->len)
 552                 /*
 553                  * Hole 1 will be merge center
 554                  * Same hole will be merged later
 555                  */
 556                 return -1;
 557         /* Hole 2 will be merge center */
 558         return 1;
 559 }
 560
 561 /*
 562  * Add a hole to the record
 563  *
 564  * This will do hole merge for copy_file_extent_holes(),
 565  * which will ensure there won't be continuous holes.
 566  */
 567 static int add_file_extent_hole(struct rb_root *holes,
 568                                 u64 start, u64 len)
 569 {
 570         struct file_extent_hole *hole;
 571         struct file_extent_hole *prev = NULL;
 572         struct file_extent_hole *next = NULL;
 573
 574         hole = malloc(sizeof(*hole));
 575         if (!hole)
 576                 return -ENOMEM;
 577         hole->start = start;
 578         hole->len = len;
 579         /* Since compare will not return 0, no -EEXIST will happen */
 580         rb_insert(holes, &hole->node, compare_hole);
 581
 582         /* simple merge with previous hole */
 583         if (rb_prev(&hole->node))
 584                 prev = rb_entry(rb_prev(&hole->node), struct file_extent_hole,
 585                                 node);
 586         if (prev && prev->start + prev->len >= hole->start) {
 587                 hole->len = hole->start + hole->len - prev->start;
 588                 hole->start = prev->start;
 589                 rb_erase(&prev->node, holes);
 590                 free(prev);
 591                 prev = NULL;
 592         }
 593
 594         /* iterate merge with next holes */
 595         while (1) {
 596                 if (!rb_next(&hole->node))
 597                         break;
 598                 next = rb_entry(rb_next(&hole->node), struct file_extent_hole,
 599                                         node);
 600                 if (hole->start + hole->len >= next->start) {
 601                         if (hole->start + hole->len <= next->start + next->len)
 602                                 hole->len = next->start + next->len -
 603                                             hole->start;
 604                         rb_erase(&next->node, holes);
 605                         free(next);
 606                         next = NULL;
 607                 } else
 608                         break;
 609         }
 610         return 0;
 611 }
 612
 613 static int compare_hole_range(struct rb_node *node, void *data)
 614 {
 615         struct file_extent_hole *hole;
 616         u64 start;
 617
 618         hole = (struct file_extent_hole *)data;
 619         start = hole->start;
 620
 621         hole = rb_entry(node, struct file_extent_hole, node);
 622         if (start < hole->start)
 623                 return -1;
 624         if (start >= hole->start && start < hole->start + hole->len)
 625                 return 0;
 626         return 1;
 627 }
 628
 629 /*
 630  * Delete a hole in the record
 631  *
 632  * This will do the hole split and is much restrict than add.
 633  */
 634 static int del_file_extent_hole(struct rb_root *holes,
 635                                 u64 start, u64 len)
 636 {
 637         struct file_extent_hole *hole;
 638         struct file_extent_hole tmp;
 639         u64 prev_start = 0;
 640         u64 prev_len = 0;
 641         u64 next_start = 0;
 642         u64 next_len = 0;
 643         struct rb_node *node;
 644         int have_prev = 0;
 645         int have_next = 0;
 646         int ret = 0;
 647
 648         tmp.start = start;
 649         tmp.len = len;
 650         node = rb_search(holes, &tmp, compare_hole_range, NULL);
 651         if (!node)
 652                 return -EEXIST;
 653         hole = rb_entry(node, struct file_extent_hole, node);
 654         if (start + len > hole->start + hole->len)
 655                 return -EEXIST;
 656
 657         /*
 658          * Now there will be no overlap, delete the hole and re-add the
 659          * split(s) if they exists.
 660          */
 661         if (start > hole->start) {
 662                 prev_start = hole->start;
 663                 prev_len = start - hole->start;
 664                 have_prev = 1;
 665         }
 666         if (hole->start + hole->len > start + len) {
 667                 next_start = start + len;
 668                 next_len = hole->start + hole->len - start - len;
 669                 have_next = 1;
 670         }
 671         rb_erase(node, holes);
 672         free(hole);
 673         if (have_prev) {
 674                 ret = add_file_extent_hole(holes, prev_start, prev_len);
 675                 if (ret < 0)
 676                         return ret;
 677         }
 678         if (have_next) {
 679                 ret = add_file_extent_hole(holes, next_start, next_len);
 680                 if (ret < 0)
 681                         return ret;
 682         }
 683         return 0;
 684 }
 685
 686 static int copy_file_extent_holes(struct rb_root *dst,
 687                                   struct rb_root *src)
 688 {
 689         struct file_extent_hole *hole;
 690         struct rb_node *node;
 691         int ret = 0;
 692
 693         node = rb_first(src);
 694         while (node) {
 695                 hole = rb_entry(node, struct file_extent_hole, node);
 696                 ret = add_file_extent_hole(dst, hole->start, hole->len);
 697                 if (ret)
 698                         break;
 699                 node = rb_next(node);
 700         }
 701         return ret;
 702 }
 703
 704 static void free_file_extent_holes(struct rb_root *holes)
 705 {
 706         struct rb_node *node;
 707         struct file_extent_hole *hole;
 708
 709         node = rb_first(holes);
 710         while (node) {
 711                 hole = rb_entry(node, struct file_extent_hole, node);
 712                 rb_erase(node, holes);
 713                 free(hole);
 714                 node = rb_first(holes);
 715         }
 716 }
 717
 718 static void reset_cached_block_groups(struct btrfs_fs_info *fs_info);
 719
 720 static void record_root_in_trans(struct btrfs_trans_handle *trans,
 721                                  struct btrfs_root *root)
 722 {
 723         if (root->last_trans != trans->transid) {
 724                 root->track_dirty = 1;
 725                 root->last_trans = trans->transid;
 726                 root->commit_root = root->node;
 727                 extent_buffer_get(root->node);
 728         }
 729 }
 730
 731 static u8 imode_to_type(u32 imode)
 732 {
 733 #define S_SHIFT 12
 734         static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
 735                 [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
 736                 [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
 737                 [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
 738                 [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
 739                 [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
 740                 [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
 741                 [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
 742         };
 743
 744         return btrfs_type_by_mode[(imode & S_IFMT) >> S_SHIFT];
 745 #undef S_SHIFT
 746 }
 747
 748 static int device_record_compare(struct rb_node *node1, struct rb_node *node2)
 749 {
 750         struct device_record *rec1;
 751         struct device_record *rec2;
 752
 753         rec1 = rb_entry(node1, struct device_record, node);
 754         rec2 = rb_entry(node2, struct device_record, node);
 755         if (rec1->devid > rec2->devid)
 756                 return -1;
 757         else if (rec1->devid < rec2->devid)
 758                 return 1;
 759         else
 760                 return 0;
 761 }
 762
 763 static struct inode_record *clone_inode_rec(struct inode_record *orig_rec)
 764 {
 765         struct inode_record *rec;
 766         struct inode_backref *backref;
 767         struct inode_backref *orig;
 768         struct inode_backref *tmp;
 769         struct orphan_data_extent *src_orphan;
 770         struct orphan_data_extent *dst_orphan;
 771         struct rb_node *rb;
 772         size_t size;
 773         int ret;
 774
 775         rec = malloc(sizeof(*rec));
 776         if (!rec)
 777                 return ERR_PTR(-ENOMEM);
 778         memcpy(rec, orig_rec, sizeof(*rec));
 779         rec->refs = 1;
 780         INIT_LIST_HEAD(&rec->backrefs);
 781         INIT_LIST_HEAD(&rec->orphan_extents);
 782         rec->holes = RB_ROOT;
 783
 784         list_for_each_entry(orig, &orig_rec->backrefs, list) {
 785                 size = sizeof(*orig) + orig->namelen + 1;
 786                 backref = malloc(size);
 787                 if (!backref) {
 788                         ret = -ENOMEM;
 789                         goto cleanup;
 790                 }
 791                 memcpy(backref, orig, size);
 792                 list_add_tail(&backref->list, &rec->backrefs);
 793         }
 794         list_for_each_entry(src_orphan, &orig_rec->orphan_extents, list) {
 795                 dst_orphan = malloc(sizeof(*dst_orphan));
 796                 if (!dst_orphan) {
 797                         ret = -ENOMEM;
 798                         goto cleanup;
 799                 }
 800                 memcpy(dst_orphan, src_orphan, sizeof(*src_orphan));
 801                 list_add_tail(&dst_orphan->list, &rec->orphan_extents);
 802         }
 803         ret = copy_file_extent_holes(&rec->holes, &orig_rec->holes);
 804         if (ret < 0)
 805                 goto cleanup_rb;
 806
 807         return rec;
 808
 809 cleanup_rb:
 810         rb = rb_first(&rec->holes);
 811         while (rb) {
 812                 struct file_extent_hole *hole;
 813
 814                 hole = rb_entry(rb, struct file_extent_hole, node);
 815                 rb = rb_next(rb);
 816                 free(hole);
 817         }
 818
 819 cleanup:
 820         if (!list_empty(&rec->backrefs))
 821                 list_for_each_entry_safe(orig, tmp, &rec->backrefs, list) {
 822                         list_del(&orig->list);
 823                         free(orig);
 824                 }
 825
 826         if (!list_empty(&rec->orphan_extents))
 827                 list_for_each_entry_safe(orig, tmp, &rec->orphan_extents, list) {
 828                         list_del(&orig->list);
 829                         free(orig);
 830                 }
 831
 832         free(rec);
 833
 834         return ERR_PTR(ret);
 835 }
 836
 837 static void print_orphan_data_extents(struct list_head *orphan_extents,
 838                                       u64 objectid)
 839 {
 840         struct orphan_data_extent *orphan;
 841
 842         if (list_empty(orphan_extents))
 843                 return;
 844         printf("The following data extent is lost in tree %llu:\n",
 845                objectid);
 846         list_for_each_entry(orphan, orphan_extents, list) {
 847                 printf("\tinode: %llu, offset:%llu, disk_bytenr: %llu, disk_len: %llu\n",
 848                        orphan->objectid, orphan->offset, orphan->disk_bytenr,
 849                        orphan->disk_len);
 850         }
 851 }
 852
 853 static void print_inode_error(struct btrfs_root *root, struct inode_record *rec)
 854 {
 855         u64 root_objectid = root->root_key.objectid;
 856         int errors = rec->errors;
 857
 858         if (!errors)
 859                 return;
 860         /* reloc root errors, we print its corresponding fs root objectid*/
 861         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
 862                 root_objectid = root->root_key.offset;
 863                 fprintf(stderr, "reloc");
 864         }
 865         fprintf(stderr, "root %llu inode %llu errors %x",
 866                 (unsigned long long) root_objectid,
 867                 (unsigned long long) rec->ino, rec->errors);
 868
 869         if (errors & I_ERR_NO_INODE_ITEM)
 870                 fprintf(stderr, ", no inode item");
 871         if (errors & I_ERR_NO_ORPHAN_ITEM)
 872                 fprintf(stderr, ", no orphan item");
 873         if (errors & I_ERR_DUP_INODE_ITEM)
 874                 fprintf(stderr, ", dup inode item");
 875         if (errors & I_ERR_DUP_DIR_INDEX)
 876                 fprintf(stderr, ", dup dir index");
 877         if (errors & I_ERR_ODD_DIR_ITEM)
 878                 fprintf(stderr, ", odd dir item");
 879         if (errors & I_ERR_ODD_FILE_EXTENT)
 880                 fprintf(stderr, ", odd file extent");
 881         if (errors & I_ERR_BAD_FILE_EXTENT)
 882                 fprintf(stderr, ", bad file extent");
 883         if (errors & I_ERR_FILE_EXTENT_OVERLAP)
 884                 fprintf(stderr, ", file extent overlap");
 885         if (errors & I_ERR_FILE_EXTENT_DISCOUNT)
 886                 fprintf(stderr, ", file extent discount");
 887         if (errors & I_ERR_DIR_ISIZE_WRONG)
 888                 fprintf(stderr, ", dir isize wrong");
 889         if (errors & I_ERR_FILE_NBYTES_WRONG)
 890                 fprintf(stderr, ", nbytes wrong");
 891         if (errors & I_ERR_ODD_CSUM_ITEM)
 892                 fprintf(stderr, ", odd csum item");
 893         if (errors & I_ERR_SOME_CSUM_MISSING)
 894                 fprintf(stderr, ", some csum missing");
 895         if (errors & I_ERR_LINK_COUNT_WRONG)
 896                 fprintf(stderr, ", link count wrong");
 897         if (errors & I_ERR_FILE_EXTENT_ORPHAN)
 898                 fprintf(stderr, ", orphan file extent");
 899         fprintf(stderr, "\n");
 900         /* Print the orphan extents if needed */
 901         if (errors & I_ERR_FILE_EXTENT_ORPHAN)
 902                 print_orphan_data_extents(&rec->orphan_extents, root->objectid);
 903
 904         /* Print the holes if needed */
 905         if (errors & I_ERR_FILE_EXTENT_DISCOUNT) {
 906                 struct file_extent_hole *hole;
 907                 struct rb_node *node;
 908                 int found = 0;
 909
 910                 node = rb_first(&rec->holes);
 911                 fprintf(stderr, "Found file extent holes:\n");
 912                 while (node) {
 913                         found = 1;
 914                         hole = rb_entry(node, struct file_extent_hole, node);
 915                         fprintf(stderr, "\tstart: %llu, len: %llu\n",
 916                                 hole->start, hole->len);
 917                         node = rb_next(node);
 918                 }
 919                 if (!found)
 920                         fprintf(stderr, "\tstart: 0, len: %llu\n",
 921                                 round_up(rec->isize,
 922                                          root->fs_info->sectorsize));
 923         }
 924 }
 925
 926 static void print_ref_error(int errors)
 927 {
 928         if (errors & REF_ERR_NO_DIR_ITEM)
 929                 fprintf(stderr, ", no dir item");
 930         if (errors & REF_ERR_NO_DIR_INDEX)
 931                 fprintf(stderr, ", no dir index");
 932         if (errors & REF_ERR_NO_INODE_REF)
 933                 fprintf(stderr, ", no inode ref");
 934         if (errors & REF_ERR_DUP_DIR_ITEM)
 935                 fprintf(stderr, ", dup dir item");
 936         if (errors & REF_ERR_DUP_DIR_INDEX)
 937                 fprintf(stderr, ", dup dir index");
 938         if (errors & REF_ERR_DUP_INODE_REF)
 939                 fprintf(stderr, ", dup inode ref");
 940         if (errors & REF_ERR_INDEX_UNMATCH)
 941                 fprintf(stderr, ", index mismatch");
 942         if (errors & REF_ERR_FILETYPE_UNMATCH)
 943                 fprintf(stderr, ", filetype mismatch");
 944         if (errors & REF_ERR_NAME_TOO_LONG)
 945                 fprintf(stderr, ", name too long");
 946         if (errors & REF_ERR_NO_ROOT_REF)
 947                 fprintf(stderr, ", no root ref");
 948         if (errors & REF_ERR_NO_ROOT_BACKREF)
 949                 fprintf(stderr, ", no root backref");
 950         if (errors & REF_ERR_DUP_ROOT_REF)
 951                 fprintf(stderr, ", dup root ref");
 952         if (errors & REF_ERR_DUP_ROOT_BACKREF)
 953                 fprintf(stderr, ", dup root backref");
 954         fprintf(stderr, "\n");
 955 }
 956
 957 static struct inode_record *get_inode_rec(struct cache_tree *inode_cache,
 958                                           u64 ino, int mod)
 959 {
 960         struct ptr_node *node;
 961         struct cache_extent *cache;
 962         struct inode_record *rec = NULL;
 963         int ret;
 964
 965         cache = lookup_cache_extent(inode_cache, ino, 1);
 966         if (cache) {
 967                 node = container_of(cache, struct ptr_node, cache);
 968                 rec = node->data;
 969                 if (mod && rec->refs > 1) {
 970                         node->data = clone_inode_rec(rec);
 971                         if (IS_ERR(node->data))
 972                                 return node->data;
 973                         rec->refs--;
 974                         rec = node->data;
 975                 }
 976         } else if (mod) {
 977                 rec = calloc(1, sizeof(*rec));
 978                 if (!rec)
 979                         return ERR_PTR(-ENOMEM);
 980                 rec->ino = ino;
 981                 rec->extent_start = (u64)-1;
 982                 rec->refs = 1;
 983                 INIT_LIST_HEAD(&rec->backrefs);
 984                 INIT_LIST_HEAD(&rec->orphan_extents);
 985                 rec->holes = RB_ROOT;
 986
 987                 node = malloc(sizeof(*node));
 988                 if (!node) {
 989                         free(rec);
 990                         return ERR_PTR(-ENOMEM);
 991                 }
 992                 node->cache.start = ino;
 993                 node->cache.size = 1;
 994                 node->data = rec;
 995
 996                 if (ino == BTRFS_FREE_INO_OBJECTID)
 997                         rec->found_link = 1;
 998
 999                 ret = insert_cache_extent(inode_cache, &node->cache);
1000                 if (ret)
1001                         return ERR_PTR(-EEXIST);
1002         }
1003         return rec;
1004 }
1005
1006 static void free_orphan_data_extents(struct list_head *orphan_extents)
1007 {
1008         struct orphan_data_extent *orphan;
1009
1010         while (!list_empty(orphan_extents)) {
1011                 orphan = list_entry(orphan_extents->next,
1012                                     struct orphan_data_extent, list);
1013                 list_del(&orphan->list);
1014                 free(orphan);
1015         }
1016 }
1017
1018 static void free_inode_rec(struct inode_record *rec)
1019 {
1020         struct inode_backref *backref;
1021
1022         if (--rec->refs > 0)
1023                 return;
1024
1025         while (!list_empty(&rec->backrefs)) {
1026                 backref = to_inode_backref(rec->backrefs.next);
1027                 list_del(&backref->list);
1028                 free(backref);
1029         }
1030         free_orphan_data_extents(&rec->orphan_extents);
1031         free_file_extent_holes(&rec->holes);
1032         free(rec);
1033 }
1034
1035 static int can_free_inode_rec(struct inode_record *rec)
1036 {
1037         if (!rec->errors && rec->checked && rec->found_inode_item &&
1038             rec->nlink == rec->found_link && list_empty(&rec->backrefs))
1039                 return 1;
1040         return 0;
1041 }
1042
1043 static void maybe_free_inode_rec(struct cache_tree *inode_cache,
1044                                  struct inode_record *rec)
1045 {
1046         struct cache_extent *cache;
1047         struct inode_backref *tmp, *backref;
1048         struct ptr_node *node;
1049         u8 filetype;
1050
1051         if (!rec->found_inode_item)
1052                 return;
1053
1054         filetype = imode_to_type(rec->imode);
1055         list_for_each_entry_safe(backref, tmp, &rec->backrefs, list) {
1056                 if (backref->found_dir_item && backref->found_dir_index) {
1057                         if (backref->filetype != filetype)
1058                                 backref->errors |= REF_ERR_FILETYPE_UNMATCH;
1059                         if (!backref->errors && backref->found_inode_ref &&
1060                             rec->nlink == rec->found_link) {
1061                                 list_del(&backref->list);
1062                                 free(backref);
1063                         }
1064                 }
1065         }
1066
1067         if (!rec->checked || rec->merging)
1068                 return;
1069
1070         if (S_ISDIR(rec->imode)) {
1071                 if (rec->found_size != rec->isize)
1072                         rec->errors |= I_ERR_DIR_ISIZE_WRONG;
1073                 if (rec->found_file_extent)
1074                         rec->errors |= I_ERR_ODD_FILE_EXTENT;
1075         } else if (S_ISREG(rec->imode) || S_ISLNK(rec->imode)) {
1076                 if (rec->found_dir_item)
1077                         rec->errors |= I_ERR_ODD_DIR_ITEM;
1078                 if (rec->found_size != rec->nbytes)
1079                         rec->errors |= I_ERR_FILE_NBYTES_WRONG;
1080                 if (rec->nlink > 0 && !no_holes &&
1081                     (rec->extent_end < rec->isize ||
1082                      first_extent_gap(&rec->holes) < rec->isize))
1083                         rec->errors |= I_ERR_FILE_EXTENT_DISCOUNT;
1084         }
1085
1086         if (S_ISREG(rec->imode) || S_ISLNK(rec->imode)) {
1087                 if (rec->found_csum_item && rec->nodatasum)
1088                         rec->errors |= I_ERR_ODD_CSUM_ITEM;
1089                 if (rec->some_csum_missing && !rec->nodatasum)
1090                         rec->errors |= I_ERR_SOME_CSUM_MISSING;
1091         }
1092
1093         BUG_ON(rec->refs != 1);
1094         if (can_free_inode_rec(rec)) {
1095                 cache = lookup_cache_extent(inode_cache, rec->ino, 1);
1096                 node = container_of(cache, struct ptr_node, cache);
1097                 BUG_ON(node->data != rec);
1098                 remove_cache_extent(inode_cache, &node->cache);
1099                 free(node);
1100                 free_inode_rec(rec);
1101         }
1102 }
1103
1104 static int check_orphan_item(struct btrfs_root *root, u64 ino)
1105 {
1106         struct btrfs_path path;
1107         struct btrfs_key key;
1108         int ret;
1109
1110         key.objectid = BTRFS_ORPHAN_OBJECTID;
1111         key.type = BTRFS_ORPHAN_ITEM_KEY;
1112         key.offset = ino;
1113
1114         btrfs_init_path(&path);
1115         ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
1116         btrfs_release_path(&path);
1117         if (ret > 0)
1118                 ret = -ENOENT;
1119         return ret;
1120 }
1121
1122 static int process_inode_item(struct extent_buffer *eb,
1123                               int slot, struct btrfs_key *key,
1124                               struct shared_node *active_node)
1125 {
1126         struct inode_record *rec;
1127         struct btrfs_inode_item *item;
1128
1129         rec = active_node->current;
1130         BUG_ON(rec->ino != key->objectid || rec->refs > 1);
1131         if (rec->found_inode_item) {
1132                 rec->errors |= I_ERR_DUP_INODE_ITEM;
1133                 return 1;
1134         }
1135         item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
1136         rec->nlink = btrfs_inode_nlink(eb, item);
1137         rec->isize = btrfs_inode_size(eb, item);
1138         rec->nbytes = btrfs_inode_nbytes(eb, item);
1139         rec->imode = btrfs_inode_mode(eb, item);
1140         if (btrfs_inode_flags(eb, item) & BTRFS_INODE_NODATASUM)
1141                 rec->nodatasum = 1;
1142         rec->found_inode_item = 1;
1143         if (rec->nlink == 0)
1144                 rec->errors |= I_ERR_NO_ORPHAN_ITEM;
1145         maybe_free_inode_rec(&active_node->inode_cache, rec);
1146         return 0;
1147 }
1148
1149 static struct inode_backref *get_inode_backref(struct inode_record *rec,
1150                                                 const char *name,
1151                                                 int namelen, u64 dir)
1152 {
1153         struct inode_backref *backref;
1154
1155         list_for_each_entry(backref, &rec->backrefs, list) {
1156                 if (rec->ino == BTRFS_MULTIPLE_OBJECTIDS)
1157                         break;
1158                 if (backref->dir != dir || backref->namelen != namelen)
1159                         continue;
1160                 if (memcmp(name, backref->name, namelen))
1161                         continue;
1162                 return backref;
1163         }
1164
1165         backref = malloc(sizeof(*backref) + namelen + 1);
1166         if (!backref)
1167                 return NULL;
1168         memset(backref, 0, sizeof(*backref));
1169         backref->dir = dir;
1170         backref->namelen = namelen;
1171         memcpy(backref->name, name, namelen);
1172         backref->name[namelen] = '\0';
1173         list_add_tail(&backref->list, &rec->backrefs);
1174         return backref;
1175 }
1176
1177 static int add_inode_backref(struct cache_tree *inode_cache,
1178                              u64 ino, u64 dir, u64 index,
1179                              const char *name, int namelen,
1180                              u8 filetype, u8 itemtype, int errors)
1181 {
1182         struct inode_record *rec;
1183         struct inode_backref *backref;
1184
1185         rec = get_inode_rec(inode_cache, ino, 1);
1186         BUG_ON(IS_ERR(rec));
1187         backref = get_inode_backref(rec, name, namelen, dir);
1188         BUG_ON(!backref);
1189         if (errors)
1190                 backref->errors |= errors;
1191         if (itemtype == BTRFS_DIR_INDEX_KEY) {
1192                 if (backref->found_dir_index)
1193                         backref->errors |= REF_ERR_DUP_DIR_INDEX;
1194                 if (backref->found_inode_ref && backref->index != index)
1195                         backref->errors |= REF_ERR_INDEX_UNMATCH;
1196                 if (backref->found_dir_item && backref->filetype != filetype)
1197                         backref->errors |= REF_ERR_FILETYPE_UNMATCH;
1198
1199                 backref->index = index;
1200                 backref->filetype = filetype;
1201                 backref->found_dir_index = 1;
1202         } else if (itemtype == BTRFS_DIR_ITEM_KEY) {
1203                 rec->found_link++;
1204                 if (backref->found_dir_item)
1205                         backref->errors |= REF_ERR_DUP_DIR_ITEM;
1206                 if (backref->found_dir_index && backref->filetype != filetype)
1207                         backref->errors |= REF_ERR_FILETYPE_UNMATCH;
1208
1209                 backref->filetype = filetype;
1210                 backref->found_dir_item = 1;
1211         } else if ((itemtype == BTRFS_INODE_REF_KEY) ||
1212                    (itemtype == BTRFS_INODE_EXTREF_KEY)) {
1213                 if (backref->found_inode_ref)
1214                         backref->errors |= REF_ERR_DUP_INODE_REF;
1215                 if (backref->found_dir_index && backref->index != index)
1216                         backref->errors |= REF_ERR_INDEX_UNMATCH;
1217                 else
1218                         backref->index = index;
1219
1220                 backref->ref_type = itemtype;
1221                 backref->found_inode_ref = 1;
1222         } else {
1223                 BUG_ON(1);
1224         }
1225
1226         maybe_free_inode_rec(inode_cache, rec);
1227         return 0;
1228 }
1229
1230 static int merge_inode_recs(struct inode_record *src, struct inode_record *dst,
1231                             struct cache_tree *dst_cache)
1232 {
1233         struct inode_backref *backref;
1234         u32 dir_count = 0;
1235         int ret = 0;
1236
1237         dst->merging = 1;
1238         list_for_each_entry(backref, &src->backrefs, list) {
1239                 if (backref->found_dir_index) {
1240                         add_inode_backref(dst_cache, dst->ino, backref->dir,
1241                                         backref->index, backref->name,
1242                                         backref->namelen, backref->filetype,
1243                                         BTRFS_DIR_INDEX_KEY, backref->errors);
1244                 }
1245                 if (backref->found_dir_item) {
1246                         dir_count++;
1247                         add_inode_backref(dst_cache, dst->ino,
1248                                         backref->dir, 0, backref->name,
1249                                         backref->namelen, backref->filetype,
1250                                         BTRFS_DIR_ITEM_KEY, backref->errors);
1251                 }
1252                 if (backref->found_inode_ref) {
1253                         add_inode_backref(dst_cache, dst->ino,
1254                                         backref->dir, backref->index,
1255                                         backref->name, backref->namelen, 0,
1256                                         backref->ref_type, backref->errors);
1257                 }
1258         }
1259
1260         if (src->found_dir_item)
1261                 dst->found_dir_item = 1;
1262         if (src->found_file_extent)
1263                 dst->found_file_extent = 1;
1264         if (src->found_csum_item)
1265                 dst->found_csum_item = 1;
1266         if (src->some_csum_missing)
1267                 dst->some_csum_missing = 1;
1268         if (first_extent_gap(&dst->holes) > first_extent_gap(&src->holes)) {
1269                 ret = copy_file_extent_holes(&dst->holes, &src->holes);
1270                 if (ret < 0)
1271                         return ret;
1272         }
1273
1274         BUG_ON(src->found_link < dir_count);
1275         dst->found_link += src->found_link - dir_count;
1276         dst->found_size += src->found_size;
1277         if (src->extent_start != (u64)-1) {
1278                 if (dst->extent_start == (u64)-1) {
1279                         dst->extent_start = src->extent_start;
1280                         dst->extent_end = src->extent_end;
1281                 } else {
1282                         if (dst->extent_end > src->extent_start)
1283                                 dst->errors |= I_ERR_FILE_EXTENT_OVERLAP;
1284                         else if (dst->extent_end < src->extent_start) {
1285                                 ret = add_file_extent_hole(&dst->holes,
1286                                         dst->extent_end,
1287                                         src->extent_start - dst->extent_end);
1288                         }
1289                         if (dst->extent_end < src->extent_end)
1290                                 dst->extent_end = src->extent_end;
1291                 }
1292         }
1293
1294         dst->errors |= src->errors;
1295         if (src->found_inode_item) {
1296                 if (!dst->found_inode_item) {
1297                         dst->nlink = src->nlink;
1298                         dst->isize = src->isize;
1299                         dst->nbytes = src->nbytes;
1300                         dst->imode = src->imode;
1301                         dst->nodatasum = src->nodatasum;
1302                         dst->found_inode_item = 1;
1303                 } else {
1304                         dst->errors |= I_ERR_DUP_INODE_ITEM;
1305                 }
1306         }
1307         dst->merging = 0;
1308
1309         return 0;
1310 }
1311
1312 static int splice_shared_node(struct shared_node *src_node,
1313                               struct shared_node *dst_node)
1314 {
1315         struct cache_extent *cache;
1316         struct ptr_node *node, *ins;
1317         struct cache_tree *src, *dst;
1318         struct inode_record *rec, *conflict;
1319         u64 current_ino = 0;
1320         int splice = 0;
1321         int ret;
1322
1323         if (--src_node->refs == 0)
1324                 splice = 1;
1325         if (src_node->current)
1326                 current_ino = src_node->current->ino;
1327
1328         src = &src_node->root_cache;
1329         dst = &dst_node->root_cache;
1330 again:
1331         cache = search_cache_extent(src, 0);
1332         while (cache) {
1333                 node = container_of(cache, struct ptr_node, cache);
1334                 rec = node->data;
1335                 cache = next_cache_extent(cache);
1336
1337                 if (splice) {
1338                         remove_cache_extent(src, &node->cache);
1339                         ins = node;
1340                 } else {
1341                         ins = malloc(sizeof(*ins));
1342                         BUG_ON(!ins);
1343                         ins->cache.start = node->cache.start;
1344                         ins->cache.size = node->cache.size;
1345                         ins->data = rec;
1346                         rec->refs++;
1347                 }
1348                 ret = insert_cache_extent(dst, &ins->cache);
1349                 if (ret == -EEXIST) {
1350                         conflict = get_inode_rec(dst, rec->ino, 1);
1351                         BUG_ON(IS_ERR(conflict));
1352                         merge_inode_recs(rec, conflict, dst);
1353                         if (rec->checked) {
1354                                 conflict->checked = 1;
1355                                 if (dst_node->current == conflict)
1356                                         dst_node->current = NULL;
1357                         }
1358                         maybe_free_inode_rec(dst, conflict);
1359                         free_inode_rec(rec);
1360                         free(ins);
1361                 } else {
1362                         BUG_ON(ret);
1363                 }
1364         }
1365
1366         if (src == &src_node->root_cache) {
1367                 src = &src_node->inode_cache;
1368                 dst = &dst_node->inode_cache;
1369                 goto again;
1370         }
1371
1372         if (current_ino > 0 && (!dst_node->current ||
1373             current_ino > dst_node->current->ino)) {
1374                 if (dst_node->current) {
1375                         dst_node->current->checked = 1;
1376                         maybe_free_inode_rec(dst, dst_node->current);
1377                 }
1378                 dst_node->current = get_inode_rec(dst, current_ino, 1);
1379                 BUG_ON(IS_ERR(dst_node->current));
1380         }
1381         return 0;
1382 }
1383
1384 static void free_inode_ptr(struct cache_extent *cache)
1385 {
1386         struct ptr_node *node;
1387         struct inode_record *rec;
1388
1389         node = container_of(cache, struct ptr_node, cache);
1390         rec = node->data;
1391         free_inode_rec(rec);
1392         free(node);
1393 }
1394
1395 FREE_EXTENT_CACHE_BASED_TREE(inode_recs, free_inode_ptr);
1396
1397 static struct shared_node *find_shared_node(struct cache_tree *shared,
1398                                             u64 bytenr)
1399 {
1400         struct cache_extent *cache;
1401         struct shared_node *node;
1402
1403         cache = lookup_cache_extent(shared, bytenr, 1);
1404         if (cache) {
1405                 node = container_of(cache, struct shared_node, cache);
1406                 return node;
1407         }
1408         return NULL;
1409 }
1410
1411 static int add_shared_node(struct cache_tree *shared, u64 bytenr, u32 refs)
1412 {
1413         int ret;
1414         struct shared_node *node;
1415
1416         node = calloc(1, sizeof(*node));
1417         if (!node)
1418                 return -ENOMEM;
1419         node->cache.start = bytenr;
1420         node->cache.size = 1;
1421         cache_tree_init(&node->root_cache);
1422         cache_tree_init(&node->inode_cache);
1423         node->refs = refs;
1424
1425         ret = insert_cache_extent(shared, &node->cache);
1426
1427         return ret;
1428 }
1429
1430 static int enter_shared_node(struct btrfs_root *root, u64 bytenr, u32 refs,
1431                              struct walk_control *wc, int level)
1432 {
1433         struct shared_node *node;
1434         struct shared_node *dest;
1435         int ret;
1436
1437         if (level == wc->active_node)
1438                 return 0;
1439
1440         BUG_ON(wc->active_node <= level);
1441         node = find_shared_node(&wc->shared, bytenr);
1442         if (!node) {
1443                 ret = add_shared_node(&wc->shared, bytenr, refs);
1444                 BUG_ON(ret);
1445                 node = find_shared_node(&wc->shared, bytenr);
1446                 wc->nodes[level] = node;
1447                 wc->active_node = level;
1448                 return 0;
1449         }
1450
1451         if (wc->root_level == wc->active_node &&
1452             btrfs_root_refs(&root->root_item) == 0) {
1453                 if (--node->refs == 0) {
1454                         free_inode_recs_tree(&node->root_cache);
1455                         free_inode_recs_tree(&node->inode_cache);
1456                         remove_cache_extent(&wc->shared, &node->cache);
1457                         free(node);
1458                 }
1459                 return 1;
1460         }
1461
1462         dest = wc->nodes[wc->active_node];
1463         splice_shared_node(node, dest);
1464         if (node->refs == 0) {
1465                 remove_cache_extent(&wc->shared, &node->cache);
1466                 free(node);
1467         }
1468         return 1;
1469 }
1470
1471 static int leave_shared_node(struct btrfs_root *root,
1472                              struct walk_control *wc, int level)
1473 {
1474         struct shared_node *node;
1475         struct shared_node *dest;
1476         int i;
1477
1478         if (level == wc->root_level)
1479                 return 0;
1480
1481         for (i = level + 1; i < BTRFS_MAX_LEVEL; i++) {
1482                 if (wc->nodes[i])
1483                         break;
1484         }
1485         BUG_ON(i >= BTRFS_MAX_LEVEL);
1486
1487         node = wc->nodes[wc->active_node];
1488         wc->nodes[wc->active_node] = NULL;
1489         wc->active_node = i;
1490
1491         dest = wc->nodes[wc->active_node];
1492         if (wc->active_node < wc->root_level ||
1493             btrfs_root_refs(&root->root_item) > 0) {
1494                 BUG_ON(node->refs <= 1);
1495                 splice_shared_node(node, dest);
1496         } else {
1497                 BUG_ON(node->refs < 2);
1498                 node->refs--;
1499         }
1500         return 0;
1501 }
1502
1503 /*
1504  * Returns:
1505  * < 0 - on error
1506  * 1   - if the root with id child_root_id is a child of root parent_root_id
1507  * 0   - if the root child_root_id isn't a child of the root parent_root_id but
1508  *       has other root(s) as parent(s)
1509  * 2   - if the root child_root_id doesn't have any parent roots
1510  */
1511 static int is_child_root(struct btrfs_root *root, u64 parent_root_id,
1512                          u64 child_root_id)
1513 {
1514         struct btrfs_path path;
1515         struct btrfs_key key;
1516         struct extent_buffer *leaf;
1517         int has_parent = 0;
1518         int ret;
1519
1520         btrfs_init_path(&path);
1521
1522         key.objectid = parent_root_id;
1523         key.type = BTRFS_ROOT_REF_KEY;
1524         key.offset = child_root_id;
1525         ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, &path,
1526                                 0, 0);
1527         if (ret < 0)
1528                 return ret;
1529         btrfs_release_path(&path);
1530         if (!ret)
1531                 return 1;
1532
1533         key.objectid = child_root_id;
1534         key.type = BTRFS_ROOT_BACKREF_KEY;
1535         key.offset = 0;
1536         ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, &path,
1537                                 0, 0);
1538         if (ret < 0)
1539                 goto out;
1540
1541         while (1) {
1542                 leaf = path.nodes[0];
1543                 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
1544                         ret = btrfs_next_leaf(root->fs_info->tree_root, &path);
1545                         if (ret)
1546                                 break;
1547                         leaf = path.nodes[0];
1548                 }
1549
1550                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
1551                 if (key.objectid != child_root_id ||
1552                     key.type != BTRFS_ROOT_BACKREF_KEY)
1553                         break;
1554
1555                 has_parent = 1;
1556
1557                 if (key.offset == parent_root_id) {
1558                         btrfs_release_path(&path);
1559                         return 1;
1560                 }
1561
1562                 path.slots[0]++;
1563         }
1564 out:
1565         btrfs_release_path(&path);
1566         if (ret < 0)
1567                 return ret;
1568         return has_parent ? 0 : 2;
1569 }
1570
1571 static int process_dir_item(struct extent_buffer *eb,
1572                             int slot, struct btrfs_key *key,
1573                             struct shared_node *active_node)
1574 {
1575         u32 total;
1576         u32 cur = 0;
1577         u32 len;
1578         u32 name_len;
1579         u32 data_len;
1580         int error;
1581         int nritems = 0;
1582         u8 filetype;
1583         struct btrfs_dir_item *di;
1584         struct inode_record *rec;
1585         struct cache_tree *root_cache;
1586         struct cache_tree *inode_cache;
1587         struct btrfs_key location;
1588         char namebuf[BTRFS_NAME_LEN];
1589
1590         root_cache = &active_node->root_cache;
1591         inode_cache = &active_node->inode_cache;
1592         rec = active_node->current;
1593         rec->found_dir_item = 1;
1594
1595         di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
1596         total = btrfs_item_size_nr(eb, slot);
1597         while (cur < total) {
1598                 nritems++;
1599                 btrfs_dir_item_key_to_cpu(eb, di, &location);
1600                 name_len = btrfs_dir_name_len(eb, di);
1601                 data_len = btrfs_dir_data_len(eb, di);
1602                 filetype = btrfs_dir_type(eb, di);
1603
1604                 rec->found_size += name_len;
1605                 if (cur + sizeof(*di) + name_len > total ||
1606                     name_len > BTRFS_NAME_LEN) {
1607                         error = REF_ERR_NAME_TOO_LONG;
1608
1609                         if (cur + sizeof(*di) > total)
1610                                 break;
1611                         len = min_t(u32, total - cur - sizeof(*di),
1612                                     BTRFS_NAME_LEN);
1613                 } else {
1614                         len = name_len;
1615                         error = 0;
1616                 }
1617
1618                 read_extent_buffer(eb, namebuf, (unsigned long)(di + 1), len);
1619
1620                 if (key->type == BTRFS_DIR_ITEM_KEY &&
1621                     key->offset != btrfs_name_hash(namebuf, len)) {
1622                         rec->errors |= I_ERR_ODD_DIR_ITEM;
1623                         error("DIR_ITEM[%llu %llu] name %s namelen %u filetype %u mismatch with its hash, wanted %llu have %llu",
1624                         key->objectid, key->offset, namebuf, len, filetype,
1625                         key->offset, btrfs_name_hash(namebuf, len));
1626                 }
1627
1628                 if (location.type == BTRFS_INODE_ITEM_KEY) {
1629                         add_inode_backref(inode_cache, location.objectid,
1630                                           key->objectid, key->offset, namebuf,
1631                                           len, filetype, key->type, error);
1632                 } else if (location.type == BTRFS_ROOT_ITEM_KEY) {
1633                         add_inode_backref(root_cache, location.objectid,
1634                                           key->objectid, key->offset,
1635                                           namebuf, len, filetype,
1636                                           key->type, error);
1637                 } else {
1638                         fprintf(stderr, "invalid location in dir item %u\n",
1639                                 location.type);
1640                         add_inode_backref(inode_cache, BTRFS_MULTIPLE_OBJECTIDS,
1641                                           key->objectid, key->offset, namebuf,
1642                                           len, filetype, key->type, error);
1643                 }
1644
1645                 len = sizeof(*di) + name_len + data_len;
1646                 di = (struct btrfs_dir_item *)((char *)di + len);
1647                 cur += len;
1648         }
1649         if (key->type == BTRFS_DIR_INDEX_KEY && nritems > 1)
1650                 rec->errors |= I_ERR_DUP_DIR_INDEX;
1651
1652         return 0;
1653 }
1654
1655 static int process_inode_ref(struct extent_buffer *eb,
1656                              int slot, struct btrfs_key *key,
1657                              struct shared_node *active_node)
1658 {
1659         u32 total;
1660         u32 cur = 0;
1661         u32 len;
1662         u32 name_len;
1663         u64 index;
1664         int error;
1665         struct cache_tree *inode_cache;
1666         struct btrfs_inode_ref *ref;
1667         char namebuf[BTRFS_NAME_LEN];
1668
1669         inode_cache = &active_node->inode_cache;
1670
1671         ref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1672         total = btrfs_item_size_nr(eb, slot);
1673         while (cur < total) {
1674                 name_len = btrfs_inode_ref_name_len(eb, ref);
1675                 index = btrfs_inode_ref_index(eb, ref);
1676
1677                 /* inode_ref + namelen should not cross item boundary */
1678                 if (cur + sizeof(*ref) + name_len > total ||
1679                     name_len > BTRFS_NAME_LEN) {
1680                         if (total < cur + sizeof(*ref))
1681                                 break;
1682
1683                         /* Still try to read out the remaining part */
1684                         len = min_t(u32, total - cur - sizeof(*ref),
1685                                     BTRFS_NAME_LEN);
1686                         error = REF_ERR_NAME_TOO_LONG;
1687                 } else {
1688                         len = name_len;
1689                         error = 0;
1690                 }
1691
1692                 read_extent_buffer(eb, namebuf, (unsigned long)(ref + 1), len);
1693                 add_inode_backref(inode_cache, key->objectid, key->offset,
1694                                   index, namebuf, len, 0, key->type, error);
1695
1696                 len = sizeof(*ref) + name_len;
1697                 ref = (struct btrfs_inode_ref *)((char *)ref + len);
1698                 cur += len;
1699         }
1700         return 0;
1701 }
1702
1703 static int process_inode_extref(struct extent_buffer *eb,
1704                                 int slot, struct btrfs_key *key,
1705                                 struct shared_node *active_node)
1706 {
1707         u32 total;
1708         u32 cur = 0;
1709         u32 len;
1710         u32 name_len;
1711         u64 index;
1712         u64 parent;
1713         int error;
1714         struct cache_tree *inode_cache;
1715         struct btrfs_inode_extref *extref;
1716         char namebuf[BTRFS_NAME_LEN];
1717
1718         inode_cache = &active_node->inode_cache;
1719
1720         extref = btrfs_item_ptr(eb, slot, struct btrfs_inode_extref);
1721         total = btrfs_item_size_nr(eb, slot);
1722         while (cur < total) {
1723                 name_len = btrfs_inode_extref_name_len(eb, extref);
1724                 index = btrfs_inode_extref_index(eb, extref);
1725                 parent = btrfs_inode_extref_parent(eb, extref);
1726                 if (name_len <= BTRFS_NAME_LEN) {
1727                         len = name_len;
1728                         error = 0;
1729                 } else {
1730                         len = BTRFS_NAME_LEN;
1731                         error = REF_ERR_NAME_TOO_LONG;
1732                 }
1733                 read_extent_buffer(eb, namebuf,
1734                                    (unsigned long)(extref + 1), len);
1735                 add_inode_backref(inode_cache, key->objectid, parent,
1736                                   index, namebuf, len, 0, key->type, error);
1737
1738                 len = sizeof(*extref) + name_len;
1739                 extref = (struct btrfs_inode_extref *)((char *)extref + len);
1740                 cur += len;
1741         }
1742         return 0;
1743
1744 }
1745
1746 static int count_csum_range(struct btrfs_root *root, u64 start,
1747                             u64 len, u64 *found)
1748 {
1749         struct btrfs_key key;
1750         struct btrfs_path path;
1751         struct extent_buffer *leaf;
1752         int ret;
1753         size_t size;
1754         *found = 0;
1755         u64 csum_end;
1756         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
1757
1758         btrfs_init_path(&path);
1759
1760         key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
1761         key.offset = start;
1762         key.type = BTRFS_EXTENT_CSUM_KEY;
1763
1764         ret = btrfs_search_slot(NULL, root->fs_info->csum_root,
1765                                 &key, &path, 0, 0);
1766         if (ret < 0)
1767                 goto out;
1768         if (ret > 0 && path.slots[0] > 0) {
1769                 leaf = path.nodes[0];
1770                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0] - 1);
1771                 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
1772                     key.type == BTRFS_EXTENT_CSUM_KEY)
1773                         path.slots[0]--;
1774         }
1775
1776         while (len > 0) {
1777                 leaf = path.nodes[0];
1778                 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
1779                         ret = btrfs_next_leaf(root->fs_info->csum_root, &path);
1780                         if (ret > 0)
1781                                 break;
1782                         else if (ret < 0)
1783                                 goto out;
1784                         leaf = path.nodes[0];
1785                 }
1786
1787                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
1788                 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1789                     key.type != BTRFS_EXTENT_CSUM_KEY)
1790                         break;
1791
1792                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
1793                 if (key.offset >= start + len)
1794                         break;
1795
1796                 if (key.offset > start)
1797                         start = key.offset;
1798
1799                 size = btrfs_item_size_nr(leaf, path.slots[0]);
1800                 csum_end = key.offset + (size / csum_size) *
1801                            root->fs_info->sectorsize;
1802                 if (csum_end > start) {
1803                         size = min(csum_end - start, len);
1804                         len -= size;
1805                         start += size;
1806                         *found += size;
1807                 }
1808
1809                 path.slots[0]++;
1810         }
1811 out:
1812         btrfs_release_path(&path);
1813         if (ret < 0)
1814                 return ret;
1815         return 0;
1816 }
1817
1818 static int process_file_extent(struct btrfs_root *root,
1819                                 struct extent_buffer *eb,
1820                                 int slot, struct btrfs_key *key,
1821                                 struct shared_node *active_node)
1822 {
1823         struct inode_record *rec;
1824         struct btrfs_file_extent_item *fi;
1825         u64 num_bytes = 0;
1826         u64 disk_bytenr = 0;
1827         u64 extent_offset = 0;
1828         u64 mask = root->fs_info->sectorsize - 1;
1829         int extent_type;
1830         int ret;
1831
1832         rec = active_node->current;
1833         BUG_ON(rec->ino != key->objectid || rec->refs > 1);
1834         rec->found_file_extent = 1;
1835
1836         if (rec->extent_start == (u64)-1) {
1837                 rec->extent_start = key->offset;
1838                 rec->extent_end = key->offset;
1839         }
1840
1841         if (rec->extent_end > key->offset)
1842                 rec->errors |= I_ERR_FILE_EXTENT_OVERLAP;
1843         else if (rec->extent_end < key->offset) {
1844                 ret = add_file_extent_hole(&rec->holes, rec->extent_end,
1845                                            key->offset - rec->extent_end);
1846                 if (ret < 0)
1847                         return ret;
1848         }
1849
1850         fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
1851         extent_type = btrfs_file_extent_type(eb, fi);
1852
1853         if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1854                 num_bytes = btrfs_file_extent_inline_len(eb, slot, fi);
1855                 if (num_bytes == 0)
1856                         rec->errors |= I_ERR_BAD_FILE_EXTENT;
1857                 rec->found_size += num_bytes;
1858                 num_bytes = (num_bytes + mask) & ~mask;
1859         } else if (extent_type == BTRFS_FILE_EXTENT_REG ||
1860                    extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1861                 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
1862                 disk_bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1863                 extent_offset = btrfs_file_extent_offset(eb, fi);
1864                 if (num_bytes == 0 || (num_bytes & mask))
1865                         rec->errors |= I_ERR_BAD_FILE_EXTENT;
1866                 if (num_bytes + extent_offset >
1867                     btrfs_file_extent_ram_bytes(eb, fi))
1868                         rec->errors |= I_ERR_BAD_FILE_EXTENT;
1869                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC &&
1870                     (btrfs_file_extent_compression(eb, fi) ||
1871                      btrfs_file_extent_encryption(eb, fi) ||
1872                      btrfs_file_extent_other_encoding(eb, fi)))
1873                         rec->errors |= I_ERR_BAD_FILE_EXTENT;
1874                 if (disk_bytenr > 0)
1875                         rec->found_size += num_bytes;
1876         } else {
1877                 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1878         }
1879         rec->extent_end = key->offset + num_bytes;
1880
1881         /*
1882          * The data reloc tree will copy full extents into its inode and then
1883          * copy the corresponding csums.  Because the extent it copied could be
1884          * a preallocated extent that hasn't been written to yet there may be no
1885          * csums to copy, ergo we won't have csums for our file extent.  This is
1886          * ok so just don't bother checking csums if the inode belongs to the
1887          * data reloc tree.
1888          */
1889         if (disk_bytenr > 0 &&
1890             btrfs_header_owner(eb) != BTRFS_DATA_RELOC_TREE_OBJECTID) {
1891                 u64 found;
1892                 if (btrfs_file_extent_compression(eb, fi))
1893                         num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1894                 else
1895                         disk_bytenr += extent_offset;
1896
1897                 ret = count_csum_range(root, disk_bytenr, num_bytes, &found);
1898                 if (ret < 0)
1899                         return ret;
1900                 if (extent_type == BTRFS_FILE_EXTENT_REG) {
1901                         if (found > 0)
1902                                 rec->found_csum_item = 1;
1903                         if (found < num_bytes)
1904                                 rec->some_csum_missing = 1;
1905                 } else if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1906                         if (found > 0)
1907                                 rec->errors |= I_ERR_ODD_CSUM_ITEM;
1908                 }
1909         }
1910         return 0;
1911 }
1912
1913 static int process_one_leaf(struct btrfs_root *root, struct extent_buffer *eb,
1914                             struct walk_control *wc)
1915 {
1916         struct btrfs_key key;
1917         u32 nritems;
1918         int i;
1919         int ret = 0;
1920         struct cache_tree *inode_cache;
1921         struct shared_node *active_node;
1922
1923         if (wc->root_level == wc->active_node &&
1924             btrfs_root_refs(&root->root_item) == 0)
1925                 return 0;
1926
1927         active_node = wc->nodes[wc->active_node];
1928         inode_cache = &active_node->inode_cache;
1929         nritems = btrfs_header_nritems(eb);
1930         for (i = 0; i < nritems; i++) {
1931                 btrfs_item_key_to_cpu(eb, &key, i);
1932
1933                 if (key.objectid == BTRFS_FREE_SPACE_OBJECTID)
1934                         continue;
1935                 if (key.type == BTRFS_ORPHAN_ITEM_KEY)
1936                         continue;
1937
1938                 if (active_node->current == NULL ||
1939                     active_node->current->ino < key.objectid) {
1940                         if (active_node->current) {
1941                                 active_node->current->checked = 1;
1942                                 maybe_free_inode_rec(inode_cache,
1943                                                      active_node->current);
1944                         }
1945                         active_node->current = get_inode_rec(inode_cache,
1946                                                              key.objectid, 1);
1947                         BUG_ON(IS_ERR(active_node->current));
1948                 }
1949                 switch (key.type) {
1950                 case BTRFS_DIR_ITEM_KEY:
1951                 case BTRFS_DIR_INDEX_KEY:
1952                         ret = process_dir_item(eb, i, &key, active_node);
1953                         break;
1954                 case BTRFS_INODE_REF_KEY:
1955                         ret = process_inode_ref(eb, i, &key, active_node);
1956                         break;
1957                 case BTRFS_INODE_EXTREF_KEY:
1958                         ret = process_inode_extref(eb, i, &key, active_node);
1959                         break;
1960                 case BTRFS_INODE_ITEM_KEY:
1961                         ret = process_inode_item(eb, i, &key, active_node);
1962                         break;
1963                 case BTRFS_EXTENT_DATA_KEY:
1964                         ret = process_file_extent(root, eb, i, &key,
1965                                                   active_node);
1966                         break;
1967                 default:
1968                         break;
1969                 };
1970         }
1971         return ret;
1972 }
1973
1974 struct node_refs {
1975         u64 bytenr[BTRFS_MAX_LEVEL];
1976         u64 refs[BTRFS_MAX_LEVEL];
1977         int need_check[BTRFS_MAX_LEVEL];
1978         /* field for checking all trees */
1979         int checked[BTRFS_MAX_LEVEL];
1980         /* the corresponding extent should be marked as full backref or not */
1981         int full_backref[BTRFS_MAX_LEVEL];
1982 };
1983
1984 static int update_nodes_refs(struct btrfs_root *root, u64 bytenr,
1985                              struct extent_buffer *eb, struct node_refs *nrefs,
1986                              u64 level, int check_all);
1987 static int check_inode_item(struct btrfs_root *root, struct btrfs_path *path,
1988                             unsigned int ext_ref);
1989
1990 /*
1991  * Returns >0  Found error, not fatal, should continue
1992  * Returns <0  Fatal error, must exit the whole check
1993  * Returns 0   No errors found
1994  */
1995 static int process_one_leaf_v2(struct btrfs_root *root, struct btrfs_path *path,
1996                                struct node_refs *nrefs, int *level, int ext_ref)
1997 {
1998         struct extent_buffer *cur = path->nodes[0];
1999         struct btrfs_key key;
2000         u64 cur_bytenr;
2001         u32 nritems;
2002         u64 first_ino = 0;
2003         int root_level = btrfs_header_level(root->node);
2004         int i;
2005         int ret = 0; /* Final return value */
2006         int err = 0; /* Positive error bitmap */
2007
2008         cur_bytenr = cur->start;
2009
2010         /* skip to first inode item or the first inode number change */
2011         nritems = btrfs_header_nritems(cur);
2012         for (i = 0; i < nritems; i++) {
2013                 btrfs_item_key_to_cpu(cur, &key, i);
2014                 if (i == 0)
2015                         first_ino = key.objectid;
2016                 if (key.type == BTRFS_INODE_ITEM_KEY ||
2017                     (first_ino && first_ino != key.objectid))
2018                         break;
2019         }
2020         if (i == nritems) {
2021                 path->slots[0] = nritems;
2022                 return 0;
2023         }
2024         path->slots[0] = i;
2025
2026 again:
2027         err |= check_inode_item(root, path, ext_ref);
2028
2029         /* modify cur since check_inode_item may change path */
2030         cur = path->nodes[0];
2031
2032         if (err & LAST_ITEM)
2033                 goto out;
2034
2035         /* still have inode items in thie leaf */
2036         if (cur->start == cur_bytenr)
2037                 goto again;
2038
2039         /*
2040          * we have switched to another leaf, above nodes may
2041          * have changed, here walk down the path, if a node
2042          * or leaf is shared, check whether we can skip this
2043          * node or leaf.
2044          */
2045         for (i = root_level; i >= 0; i--) {
2046                 if (path->nodes[i]->start == nrefs->bytenr[i])
2047                         continue;
2048
2049                 ret = update_nodes_refs(root, path->nodes[i]->start,
2050                                 path->nodes[i], nrefs, i, 0);
2051                 if (ret)
2052                         goto out;
2053
2054                 if (!nrefs->need_check[i]) {
2055                         *level += 1;
2056                         break;
2057                 }
2058         }
2059
2060         for (i = 0; i < *level; i++) {
2061                 free_extent_buffer(path->nodes[i]);
2062                 path->nodes[i] = NULL;
2063         }
2064 out:
2065         err &= ~LAST_ITEM;
2066         if (err && !ret)
2067                 ret = err;
2068         return ret;
2069 }
2070
2071 static void reada_walk_down(struct btrfs_root *root,
2072                             struct extent_buffer *node, int slot)
2073 {
2074         struct btrfs_fs_info *fs_info = root->fs_info;
2075         u64 bytenr;
2076         u64 ptr_gen;
2077         u32 nritems;
2078         int i;
2079         int level;
2080
2081         level = btrfs_header_level(node);
2082         if (level != 1)
2083                 return;
2084
2085         nritems = btrfs_header_nritems(node);
2086         for (i = slot; i < nritems; i++) {
2087                 bytenr = btrfs_node_blockptr(node, i);
2088                 ptr_gen = btrfs_node_ptr_generation(node, i);
2089                 readahead_tree_block(fs_info, bytenr, ptr_gen);
2090         }
2091 }
2092
2093 /*
2094  * Check the child node/leaf by the following condition:
2095  * 1. the first item key of the node/leaf should be the same with the one
2096  *    in parent.
2097  * 2. block in parent node should match the child node/leaf.
2098  * 3. generation of parent node and child's header should be consistent.
2099  *
2100  * Or the child node/leaf pointed by the key in parent is not valid.
2101  *
2102  * We hope to check leaf owner too, but since subvol may share leaves,
2103  * which makes leaf owner check not so strong, key check should be
2104  * sufficient enough for that case.
2105  */
2106 static int check_child_node(struct extent_buffer *parent, int slot,
2107                             struct extent_buffer *child)
2108 {
2109         struct btrfs_key parent_key;
2110         struct btrfs_key child_key;
2111         int ret = 0;
2112
2113         btrfs_node_key_to_cpu(parent, &parent_key, slot);
2114         if (btrfs_header_level(child) == 0)
2115                 btrfs_item_key_to_cpu(child, &child_key, 0);
2116         else
2117                 btrfs_node_key_to_cpu(child, &child_key, 0);
2118
2119         if (memcmp(&parent_key, &child_key, sizeof(parent_key))) {
2120                 ret = -EINVAL;
2121                 fprintf(stderr,
2122                         "Wrong key of child node/leaf, wanted: (%llu, %u, %llu), have: (%llu, %u, %llu)\n",
2123                         parent_key.objectid, parent_key.type, parent_key.offset,
2124                         child_key.objectid, child_key.type, child_key.offset);
2125         }
2126         if (btrfs_header_bytenr(child) != btrfs_node_blockptr(parent, slot)) {
2127                 ret = -EINVAL;
2128                 fprintf(stderr, "Wrong block of child node/leaf, wanted: %llu, have: %llu\n",
2129                         btrfs_node_blockptr(parent, slot),
2130                         btrfs_header_bytenr(child));
2131         }
2132         if (btrfs_node_ptr_generation(parent, slot) !=
2133             btrfs_header_generation(child)) {
2134                 ret = -EINVAL;
2135                 fprintf(stderr, "Wrong generation of child node/leaf, wanted: %llu, have: %llu\n",
2136                         btrfs_header_generation(child),
2137                         btrfs_node_ptr_generation(parent, slot));
2138         }
2139         return ret;
2140 }
2141
2142 /*
2143  * for a tree node or leaf, if it's shared, indeed we don't need to iterate it
2144  * in every fs or file tree check. Here we find its all root ids, and only check
2145  * it in the fs or file tree which has the smallest root id.
2146  */
2147 static int need_check(struct btrfs_root *root, struct ulist *roots)
2148 {
2149         struct rb_node *node;
2150         struct ulist_node *u;
2151
2152         if (roots->nnodes == 1)
2153                 return 1;
2154
2155         node = rb_first(&roots->root);
2156         u = rb_entry(node, struct ulist_node, rb_node);
2157         /*
2158          * current root id is not smallest, we skip it and let it be checked
2159          * in the fs or file tree who hash the smallest root id.
2160          */
2161         if (root->objectid != u->val)
2162                 return 0;
2163
2164         return 1;
2165 }
2166
2167 static int calc_extent_flag_v2(struct btrfs_root *root, struct extent_buffer *eb,
2168                                u64 *flags_ret)
2169 {
2170         struct btrfs_root *extent_root = root->fs_info->extent_root;
2171         struct btrfs_root_item *ri = &root->root_item;
2172         struct btrfs_extent_inline_ref *iref;
2173         struct btrfs_extent_item *ei;
2174         struct btrfs_key key;
2175         struct btrfs_path *path = NULL;
2176         unsigned long ptr;
2177         unsigned long end;
2178         u64 flags;
2179         u64 owner = 0;
2180         u64 offset;
2181         int slot;
2182         int type;
2183         int ret = 0;
2184
2185         /*
2186          * Except file/reloc tree, we can not have FULL BACKREF MODE
2187          */
2188         if (root->objectid < BTRFS_FIRST_FREE_OBJECTID)
2189                 goto normal;
2190
2191         /* root node */
2192         if (eb->start == btrfs_root_bytenr(ri))
2193                 goto normal;
2194
2195         if (btrfs_header_flag(eb, BTRFS_HEADER_FLAG_RELOC))
2196                 goto full_backref;
2197
2198         owner = btrfs_header_owner(eb);
2199         if (owner == root->objectid)
2200                 goto normal;
2201
2202         path = btrfs_alloc_path();
2203         if (!path)
2204                 return -ENOMEM;
2205
2206         key.objectid = btrfs_header_bytenr(eb);
2207         key.type = (u8)-1;
2208         key.offset = (u64)-1;
2209
2210         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2211         if (ret <= 0) {
2212                 ret = -EIO;
2213                 goto out;
2214         }
2215
2216         if (ret > 0) {
2217                 ret = btrfs_previous_extent_item(extent_root, path,
2218                                                  key.objectid);
2219                 if (ret)
2220                         goto full_backref;
2221
2222         }
2223         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2224
2225         eb = path->nodes[0];
2226         slot = path->slots[0];
2227         ei = btrfs_item_ptr(eb, slot, struct btrfs_extent_item);
2228
2229         flags = btrfs_extent_flags(eb, ei);
2230         if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
2231                 goto full_backref;
2232
2233         ptr = (unsigned long)(ei + 1);
2234         end = (unsigned long)ei + btrfs_item_size_nr(eb, slot);
2235
2236         if (key.type == BTRFS_EXTENT_ITEM_KEY)
2237                 ptr += sizeof(struct btrfs_tree_block_info);
2238
2239 next:
2240         /* Reached extent item ends normally */
2241         if (ptr == end)
2242                 goto full_backref;
2243
2244         /* Beyond extent item end, wrong item size */
2245         if (ptr > end) {
2246                 error("extent item at bytenr %llu slot %d has wrong size",
2247                         eb->start, slot);
2248                 goto full_backref;
2249         }
2250
2251         iref = (struct btrfs_extent_inline_ref *)ptr;
2252         offset = btrfs_extent_inline_ref_offset(eb, iref);
2253         type = btrfs_extent_inline_ref_type(eb, iref);
2254
2255         if (type == BTRFS_TREE_BLOCK_REF_KEY && offset == owner)
2256                 goto normal;
2257         ptr += btrfs_extent_inline_ref_size(type);
2258         goto next;
2259
2260 normal:
2261         *flags_ret &= ~BTRFS_BLOCK_FLAG_FULL_BACKREF;
2262         goto out;
2263
2264 full_backref:
2265         *flags_ret |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
2266 out:
2267         btrfs_free_path(path);
2268         return ret;
2269 }
2270
2271 /*
2272  * for a tree node or leaf, we record its reference count, so later if we still
2273  * process this node or leaf, don't need to compute its reference count again.
2274  *
2275  * @bytenr  if @bytenr == (u64)-1, only update nrefs->full_backref[level]
2276  */
2277 static int update_nodes_refs(struct btrfs_root *root, u64 bytenr,
2278                              struct extent_buffer *eb, struct node_refs *nrefs,
2279                              u64 level, int check_all)
2280 {
2281         struct ulist *roots;
2282         u64 refs = 0;
2283         u64 flags = 0;
2284         int root_level = btrfs_header_level(root->node);
2285         int check;
2286         int ret;
2287
2288         if (nrefs->bytenr[level] == bytenr)
2289                 return 0;
2290
2291         if (bytenr != (u64)-1) {
2292                 /* the return value of this function seems a mistake */
2293                 ret = btrfs_lookup_extent_info(NULL, root, bytenr,
2294                                        level, 1, &refs, &flags);
2295                 /* temporary fix */
2296                 if (ret < 0 && !check_all)
2297                         return ret;
2298
2299                 nrefs->bytenr[level] = bytenr;
2300                 nrefs->refs[level] = refs;
2301                 nrefs->full_backref[level] = 0;
2302                 nrefs->checked[level] = 0;
2303
2304                 if (refs > 1) {
2305                         ret = btrfs_find_all_roots(NULL, root->fs_info, bytenr,
2306                                                    0, &roots);
2307                         if (ret)
2308                                 return -EIO;
2309
2310                         check = need_check(root, roots);
2311                         ulist_free(roots);
2312                         nrefs->need_check[level] = check;
2313                 } else {
2314                         if (!check_all) {
2315                                 nrefs->need_check[level] = 1;
2316                         } else {
2317                                 if (level == root_level) {
2318                                         nrefs->need_check[level] = 1;
2319                                 } else {
2320                                         /*
2321                                          * The node refs may have not been
2322                                          * updated if upper needs checking (the
2323                                          * lowest root_objectid) the node can
2324                                          * be checked.
2325                                          */
2326                                         nrefs->need_check[level] =
2327                                                 nrefs->need_check[level + 1];
2328                                 }
2329                         }
2330                 }
2331         }
2332
2333         if (check_all && eb) {
2334                 calc_extent_flag_v2(root, eb, &flags);
2335                 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
2336                         nrefs->full_backref[level] = 1;
2337         }
2338
2339         return 0;
2340 }
2341
2342 /*
2343  * @level           if @level == -1 means extent data item
2344  *                  else normal treeblocl.
2345  */
2346 static int should_check_extent_strictly(struct btrfs_root *root,
2347                                         struct node_refs *nrefs, int level)
2348 {
2349         int root_level = btrfs_header_level(root->node);
2350
2351         if (level > root_level || level < -1)
2352                 return 1;
2353         if (level == root_level)
2354                 return 1;
2355         /*
2356          * if the upper node is marked full backref, it should contain shared
2357          * backref of the parent (except owner == root->objectid).
2358          */
2359         while (++level <= root_level)
2360                 if (nrefs->refs[level] > 1)
2361                         return 0;
2362
2363         return 1;
2364 }
2365
2366 static int walk_down_tree(struct btrfs_root *root, struct btrfs_path *path,
2367                           struct walk_control *wc, int *level,
2368                           struct node_refs *nrefs)
2369 {
2370         enum btrfs_tree_block_status status;
2371         u64 bytenr;
2372         u64 ptr_gen;
2373         struct btrfs_fs_info *fs_info = root->fs_info;
2374         struct extent_buffer *next;
2375         struct extent_buffer *cur;
2376         int ret, err = 0;
2377         u64 refs;
2378
2379         WARN_ON(*level < 0);
2380         WARN_ON(*level >= BTRFS_MAX_LEVEL);
2381
2382         if (path->nodes[*level]->start == nrefs->bytenr[*level]) {
2383                 refs = nrefs->refs[*level];
2384                 ret = 0;
2385         } else {
2386                 ret = btrfs_lookup_extent_info(NULL, root,
2387                                        path->nodes[*level]->start,
2388                                        *level, 1, &refs, NULL);
2389                 if (ret < 0) {
2390                         err = ret;
2391                         goto out;
2392                 }
2393                 nrefs->bytenr[*level] = path->nodes[*level]->start;
2394                 nrefs->refs[*level] = refs;
2395         }
2396
2397         if (refs > 1) {
2398                 ret = enter_shared_node(root, path->nodes[*level]->start,
2399                                         refs, wc, *level);
2400                 if (ret > 0) {
2401                         err = ret;
2402                         goto out;
2403                 }
2404         }
2405
2406         while (*level >= 0) {
2407                 WARN_ON(*level < 0);
2408                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2409                 cur = path->nodes[*level];
2410
2411                 if (btrfs_header_level(cur) != *level)
2412                         WARN_ON(1);
2413
2414                 if (path->slots[*level] >= btrfs_header_nritems(cur))
2415                         break;
2416                 if (*level == 0) {
2417                         ret = process_one_leaf(root, cur, wc);
2418                         if (ret < 0)
2419                                 err = ret;
2420                         break;
2421                 }
2422                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2423                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2424
2425                 if (bytenr == nrefs->bytenr[*level - 1]) {
2426                         refs = nrefs->refs[*level - 1];
2427                 } else {
2428                         ret = btrfs_lookup_extent_info(NULL, root, bytenr,
2429                                         *level - 1, 1, &refs, NULL);
2430                         if (ret < 0) {
2431                                 refs = 0;
2432                         } else {
2433                                 nrefs->bytenr[*level - 1] = bytenr;
2434                                 nrefs->refs[*level - 1] = refs;
2435                         }
2436                 }
2437
2438                 if (refs > 1) {
2439                         ret = enter_shared_node(root, bytenr, refs,
2440                                                 wc, *level - 1);
2441                         if (ret > 0) {
2442                                 path->slots[*level]++;
2443                                 continue;
2444                         }
2445                 }
2446
2447                 next = btrfs_find_tree_block(fs_info, bytenr, fs_info->nodesize);
2448                 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
2449                         free_extent_buffer(next);
2450                         reada_walk_down(root, cur, path->slots[*level]);
2451                         next = read_tree_block(root->fs_info, bytenr, ptr_gen);
2452                         if (!extent_buffer_uptodate(next)) {
2453                                 struct btrfs_key node_key;
2454
2455                                 btrfs_node_key_to_cpu(path->nodes[*level],
2456                                                       &node_key,
2457                                                       path->slots[*level]);
2458                                 btrfs_add_corrupt_extent_record(root->fs_info,
2459                                                 &node_key,
2460                                                 path->nodes[*level]->start,
2461                                                 root->fs_info->nodesize,
2462                                                 *level);
2463                                 err = -EIO;
2464                                 goto out;
2465                         }
2466                 }
2467
2468                 ret = check_child_node(cur, path->slots[*level], next);
2469                 if (ret) {
2470                         free_extent_buffer(next);
2471                         err = ret;
2472                         goto out;
2473                 }
2474
2475                 if (btrfs_is_leaf(next))
2476                         status = btrfs_check_leaf(root, NULL, next);
2477                 else
2478                         status = btrfs_check_node(root, NULL, next);
2479                 if (status != BTRFS_TREE_BLOCK_CLEAN) {
2480                         free_extent_buffer(next);
2481                         err = -EIO;
2482                         goto out;
2483                 }
2484
2485                 *level = *level - 1;
2486                 free_extent_buffer(path->nodes[*level]);
2487                 path->nodes[*level] = next;
2488                 path->slots[*level] = 0;
2489         }
2490 out:
2491         path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2492         return err;
2493 }
2494
2495 static int fs_root_objectid(u64 objectid);
2496
2497 /*
2498  * Update global fs information.
2499  */
2500 static void account_bytes(struct btrfs_root *root, struct btrfs_path *path,
2501                          int level)
2502 {
2503         u32 free_nrs;
2504         struct extent_buffer *eb = path->nodes[level];
2505
2506         total_btree_bytes += eb->len;
2507         if (fs_root_objectid(root->objectid))
2508                 total_fs_tree_bytes += eb->len;
2509         if (btrfs_header_owner(eb) == BTRFS_EXTENT_TREE_OBJECTID)
2510                 total_extent_tree_bytes += eb->len;
2511
2512         if (level == 0) {
2513                 btree_space_waste += btrfs_leaf_free_space(root, eb);
2514         } else {
2515                 free_nrs = (BTRFS_NODEPTRS_PER_BLOCK(root) -
2516                             btrfs_header_nritems(eb));
2517                 btree_space_waste += free_nrs * sizeof(struct btrfs_key_ptr);
2518         }
2519 }
2520
2521 /*
2522  * This function only handles BACKREF_MISSING,
2523  * If corresponding extent item exists, increase the ref, else insert an extent
2524  * item and backref.
2525  *
2526  * Returns error bits after repair.
2527  */
2528 static int repair_tree_block_ref(struct btrfs_trans_handle *trans,
2529                                  struct btrfs_root *root,
2530                                  struct extent_buffer *node,
2531                                  struct node_refs *nrefs, int level, int err)
2532 {
2533         struct btrfs_fs_info *fs_info = root->fs_info;
2534         struct btrfs_root *extent_root = fs_info->extent_root;
2535         struct btrfs_path path;
2536         struct btrfs_extent_item *ei;
2537         struct btrfs_tree_block_info *bi;
2538         struct btrfs_key key;
2539         struct extent_buffer *eb;
2540         u32 size = sizeof(*ei);
2541         u32 node_size = root->fs_info->nodesize;
2542         int insert_extent = 0;
2543         int skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
2544         int root_level = btrfs_header_level(root->node);
2545         int generation;
2546         int ret;
2547         u64 owner;
2548         u64 bytenr;
2549         u64 flags = BTRFS_EXTENT_FLAG_TREE_BLOCK;
2550         u64 parent = 0;
2551
2552         if ((err & BACKREF_MISSING) == 0)
2553                 return err;
2554
2555         WARN_ON(level > BTRFS_MAX_LEVEL);
2556         WARN_ON(level < 0);
2557
2558         btrfs_init_path(&path);
2559         bytenr = btrfs_header_bytenr(node);
2560         owner = btrfs_header_owner(node);
2561         generation = btrfs_header_generation(node);
2562
2563         key.objectid = bytenr;
2564         key.type = (u8)-1;
2565         key.offset = (u64)-1;
2566
2567         /* Search for the extent item */
2568         ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
2569         if (ret <= 0) {
2570                 ret = -EIO;
2571                 goto out;
2572         }
2573
2574         ret = btrfs_previous_extent_item(extent_root, &path, bytenr);
2575         if (ret)
2576                 insert_extent = 1;
2577
2578         /* calculate if the extent item flag is full backref or not */
2579         if (nrefs->full_backref[level] != 0)
2580                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
2581
2582         /* insert an extent item */
2583         if (insert_extent) {
2584                 struct btrfs_disk_key copy_key;
2585
2586                 generation = btrfs_header_generation(node);
2587
2588                 if (level < root_level && nrefs->full_backref[level + 1] &&
2589                     owner != root->objectid) {
2590                         flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
2591                 }
2592
2593                 key.objectid = bytenr;
2594                 if (!skinny_metadata) {
2595                         key.type = BTRFS_EXTENT_ITEM_KEY;
2596                         key.offset = node_size;
2597                         size += sizeof(*bi);
2598                 } else {
2599                         key.type = BTRFS_METADATA_ITEM_KEY;
2600                         key.offset = level;
2601                 }
2602
2603                 btrfs_release_path(&path);
2604                 ret = btrfs_insert_empty_item(trans, extent_root, &path, &key,
2605                                               size);
2606                 if (ret)
2607                         goto out;
2608
2609                 eb = path.nodes[0];
2610                 ei = btrfs_item_ptr(eb, path.slots[0], struct btrfs_extent_item);
2611
2612                 btrfs_set_extent_refs(eb, ei, 0);
2613                 btrfs_set_extent_generation(eb, ei, generation);
2614                 btrfs_set_extent_flags(eb, ei, flags);
2615
2616                 if (!skinny_metadata) {
2617                         bi = (struct btrfs_tree_block_info *)(ei + 1);
2618                         memset_extent_buffer(eb, 0, (unsigned long)bi,
2619                                              sizeof(*bi));
2620                         btrfs_set_disk_key_objectid(&copy_key, root->objectid);
2621                         btrfs_set_disk_key_type(&copy_key, 0);
2622                         btrfs_set_disk_key_offset(&copy_key, 0);
2623
2624                         btrfs_set_tree_block_level(eb, bi, level);
2625                         btrfs_set_tree_block_key(eb, bi, &copy_key);
2626                 }
2627                 btrfs_mark_buffer_dirty(eb);
2628                 printf("Added an extent item [%llu %u]\n", bytenr, node_size);
2629                 btrfs_update_block_group(trans, extent_root, bytenr, node_size,
2630                                          1, 0);
2631
2632                 nrefs->refs[level] = 0;
2633                 nrefs->full_backref[level] =
2634                         flags & BTRFS_BLOCK_FLAG_FULL_BACKREF;
2635                 btrfs_release_path(&path);
2636         }
2637
2638         if (level < root_level && nrefs->full_backref[level + 1] &&
2639             owner != root->objectid)
2640                 parent = nrefs->bytenr[level + 1];
2641
2642         /* increase the ref */
2643         ret = btrfs_inc_extent_ref(trans, extent_root, bytenr, node_size,
2644                         parent, root->objectid, level, 0);
2645
2646         nrefs->refs[level]++;
2647 out:
2648         btrfs_release_path(&path);
2649         if (ret) {
2650                 error(
2651         "failed to repair tree block ref start %llu root %llu due to %s",
2652                       bytenr, root->objectid, strerror(-ret));
2653         } else {
2654                 printf("Added one tree block ref start %llu %s %llu\n",
2655                        bytenr, parent ? "parent" : "root",
2656                        parent ? parent : root->objectid);
2657                 err &= ~BACKREF_MISSING;
2658         }
2659
2660         return err;
2661 }
2662
2663 static int check_inode_item(struct btrfs_root *root, struct btrfs_path *path,
2664                             unsigned int ext_ref);
2665 static int check_tree_block_ref(struct btrfs_root *root,
2666                                 struct extent_buffer *eb, u64 bytenr,
2667                                 int level, u64 owner, struct node_refs *nrefs);
2668 static int check_leaf_items(struct btrfs_trans_handle *trans,
2669                             struct btrfs_root *root, struct btrfs_path *path,
2670                             struct node_refs *nrefs, int account_bytes);
2671
2672 /*
2673  * @trans      just for lowmem repair mode
2674  * @check all  if not 0 then check all tree block backrefs and items
2675  *             0 then just check relationship of items in fs tree(s)
2676  *
2677  * Returns >0  Found error, should continue
2678  * Returns <0  Fatal error, must exit the whole check
2679  * Returns 0   No errors found
2680  */
2681 static int walk_down_tree_v2(struct btrfs_trans_handle *trans,
2682                              struct btrfs_root *root, struct btrfs_path *path,
2683                              int *level, struct node_refs *nrefs, int ext_ref,
2684                              int check_all)
2685
2686 {
2687         enum btrfs_tree_block_status status;
2688         u64 bytenr;
2689         u64 ptr_gen;
2690         struct btrfs_fs_info *fs_info = root->fs_info;
2691         struct extent_buffer *next;
2692         struct extent_buffer *cur;
2693         int ret;
2694         int err = 0;
2695         int check;
2696         int account_file_data = 0;
2697
2698         WARN_ON(*level < 0);
2699         WARN_ON(*level >= BTRFS_MAX_LEVEL);
2700
2701         ret = update_nodes_refs(root, btrfs_header_bytenr(path->nodes[*level]),
2702                                 path->nodes[*level], nrefs, *level, check_all);
2703         if (ret < 0)
2704                 return ret;
2705
2706         while (*level >= 0) {
2707                 WARN_ON(*level < 0);
2708                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2709                 cur = path->nodes[*level];
2710                 bytenr = btrfs_header_bytenr(cur);
2711                 check = nrefs->need_check[*level];
2712
2713                 if (btrfs_header_level(cur) != *level)
2714                         WARN_ON(1);
2715                /*
2716                 * Update bytes accounting and check tree block ref
2717                 * NOTE: Doing accounting and check before checking nritems
2718                 * is necessary because of empty node/leaf.
2719                 */
2720                 if ((check_all && !nrefs->checked[*level]) ||
2721                     (!check_all && nrefs->need_check[*level])) {
2722                         ret = check_tree_block_ref(root, cur,
2723                            btrfs_header_bytenr(cur), btrfs_header_level(cur),
2724                            btrfs_header_owner(cur), nrefs);
2725
2726                         if (repair && ret)
2727                                 ret = repair_tree_block_ref(trans, root,
2728                                     path->nodes[*level], nrefs, *level, ret);
2729                         err |= ret;
2730
2731                         if (check_all && nrefs->need_check[*level] &&
2732                                 nrefs->refs[*level]) {
2733                                 account_bytes(root, path, *level);
2734                                 account_file_data = 1;
2735                         }
2736                         nrefs->checked[*level] = 1;
2737                 }
2738
2739                 if (path->slots[*level] >= btrfs_header_nritems(cur))
2740                         break;
2741
2742                 /* Don't forgot to check leaf/node validation */
2743                 if (*level == 0) {
2744                         /* skip duplicate check */
2745                         if (check || !check_all) {
2746                                 ret = btrfs_check_leaf(root, NULL, cur);
2747                                 if (ret != BTRFS_TREE_BLOCK_CLEAN) {
2748                                         err |= -EIO;
2749                                         break;
2750                                 }
2751                         }
2752
2753                         ret = 0;
2754                         if (!check_all)
2755                                 ret = process_one_leaf_v2(root, path, nrefs,
2756                                                           level, ext_ref);
2757                         else
2758                                 ret = check_leaf_items(trans, root, path,
2759                                                nrefs, account_file_data);
2760                         err |= ret;
2761                         break;
2762                 } else {
2763                         if (check || !check_all) {
2764                                 ret = btrfs_check_node(root, NULL, cur);
2765                                 if (ret != BTRFS_TREE_BLOCK_CLEAN) {
2766                                         err |= -EIO;
2767                                         break;
2768                                 }
2769                         }
2770                 }
2771
2772                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2773                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2774
2775                 ret = update_nodes_refs(root, bytenr, NULL, nrefs, *level - 1,
2776                                         check_all);
2777                 if (ret < 0)
2778                         break;
2779                 /*
2780                  * check all trees in check_chunks_and_extent_v2
2781                  * check shared node once in check_fs_roots
2782                  */
2783                 if (!check_all && !nrefs->need_check[*level - 1]) {
2784                         path->slots[*level]++;
2785                         continue;
2786                 }
2787
2788                 next = btrfs_find_tree_block(fs_info, bytenr, fs_info->nodesize);
2789                 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
2790                         free_extent_buffer(next);
2791                         reada_walk_down(root, cur, path->slots[*level]);
2792                         next = read_tree_block(fs_info, bytenr, ptr_gen);
2793                         if (!extent_buffer_uptodate(next)) {
2794                                 struct btrfs_key node_key;
2795
2796                                 btrfs_node_key_to_cpu(path->nodes[*level],
2797                                                       &node_key,
2798                                                       path->slots[*level]);
2799                                 btrfs_add_corrupt_extent_record(fs_info,
2800                                         &node_key, path->nodes[*level]->start,
2801                                         fs_info->nodesize, *level);
2802                                 err |= -EIO;
2803                                 break;
2804                         }
2805                 }
2806
2807                 ret = check_child_node(cur, path->slots[*level], next);
2808                 err |= ret;
2809                 if (ret < 0)
2810                         break;
2811
2812                 if (btrfs_is_leaf(next))
2813                         status = btrfs_check_leaf(root, NULL, next);
2814                 else
2815                         status = btrfs_check_node(root, NULL, next);
2816                 if (status != BTRFS_TREE_BLOCK_CLEAN) {
2817                         free_extent_buffer(next);
2818                         err |= -EIO;
2819                         break;
2820                 }
2821
2822                 *level = *level - 1;
2823                 free_extent_buffer(path->nodes[*level]);
2824                 path->nodes[*level] = next;
2825                 path->slots[*level] = 0;
2826                 account_file_data = 0;
2827
2828                 update_nodes_refs(root, (u64)-1, next, nrefs, *level, check_all);
2829         }
2830         return err;
2831 }
2832
2833 static int walk_up_tree(struct btrfs_root *root, struct btrfs_path *path,
2834                         struct walk_control *wc, int *level)
2835 {
2836         int i;
2837         struct extent_buffer *leaf;
2838
2839         for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2840                 leaf = path->nodes[i];
2841                 if (path->slots[i] + 1 < btrfs_header_nritems(leaf)) {
2842                         path->slots[i]++;
2843                         *level = i;
2844                         return 0;
2845                 } else {
2846                         free_extent_buffer(path->nodes[*level]);
2847                         path->nodes[*level] = NULL;
2848                         BUG_ON(*level > wc->active_node);
2849                         if (*level == wc->active_node)
2850                                 leave_shared_node(root, wc, *level);
2851                         *level = i + 1;
2852                 }
2853         }
2854         return 1;
2855 }
2856
2857 static int walk_up_tree_v2(struct btrfs_root *root, struct btrfs_path *path,
2858                            int *level)
2859 {
2860         int i;
2861         struct extent_buffer *leaf;
2862
2863         for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2864                 leaf = path->nodes[i];
2865                 if (path->slots[i] + 1 < btrfs_header_nritems(leaf)) {
2866                         path->slots[i]++;
2867                         *level = i;
2868                         return 0;
2869                 } else {
2870                         free_extent_buffer(path->nodes[*level]);
2871                         path->nodes[*level] = NULL;
2872                         *level = i + 1;
2873                 }
2874         }
2875         return 1;
2876 }
2877
2878 static int check_root_dir(struct inode_record *rec)
2879 {
2880         struct inode_backref *backref;
2881         int ret = -1;
2882
2883         if (!rec->found_inode_item || rec->errors)
2884                 goto out;
2885         if (rec->nlink != 1 || rec->found_link != 0)
2886                 goto out;
2887         if (list_empty(&rec->backrefs))
2888                 goto out;
2889         backref = to_inode_backref(rec->backrefs.next);
2890         if (!backref->found_inode_ref)
2891                 goto out;
2892         if (backref->index != 0 || backref->namelen != 2 ||
2893             memcmp(backref->name, "..", 2))
2894                 goto out;
2895         if (backref->found_dir_index || backref->found_dir_item)
2896                 goto out;
2897         ret = 0;
2898 out:
2899         return ret;
2900 }
2901
2902 static int repair_inode_isize(struct btrfs_trans_handle *trans,
2903                               struct btrfs_root *root, struct btrfs_path *path,
2904                               struct inode_record *rec)
2905 {
2906         struct btrfs_inode_item *ei;
2907         struct btrfs_key key;
2908         int ret;
2909
2910         key.objectid = rec->ino;
2911         key.type = BTRFS_INODE_ITEM_KEY;
2912         key.offset = (u64)-1;
2913
2914         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2915         if (ret < 0)
2916                 goto out;
2917         if (ret) {
2918                 if (!path->slots[0]) {
2919                         ret = -ENOENT;
2920                         goto out;
2921                 }
2922                 path->slots[0]--;
2923                 ret = 0;
2924         }
2925         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2926         if (key.objectid != rec->ino) {
2927                 ret = -ENOENT;
2928                 goto out;
2929         }
2930
2931         ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
2932                             struct btrfs_inode_item);
2933         btrfs_set_inode_size(path->nodes[0], ei, rec->found_size);
2934         btrfs_mark_buffer_dirty(path->nodes[0]);
2935         rec->errors &= ~I_ERR_DIR_ISIZE_WRONG;
2936         printf("reset isize for dir %Lu root %Lu\n", rec->ino,
2937                root->root_key.objectid);
2938 out:
2939         btrfs_release_path(path);
2940         return ret;
2941 }
2942
2943 static int repair_inode_orphan_item(struct btrfs_trans_handle *trans,
2944                                     struct btrfs_root *root,
2945                                     struct btrfs_path *path,
2946                                     struct inode_record *rec)
2947 {
2948         int ret;
2949
2950         ret = btrfs_add_orphan_item(trans, root, path, rec->ino);
2951         btrfs_release_path(path);
2952         if (!ret)
2953                 rec->errors &= ~I_ERR_NO_ORPHAN_ITEM;
2954         return ret;
2955 }
2956
2957 static int repair_inode_nbytes(struct btrfs_trans_handle *trans,
2958                                struct btrfs_root *root,
2959                                struct btrfs_path *path,
2960                                struct inode_record *rec)
2961 {
2962         struct btrfs_inode_item *ei;
2963         struct btrfs_key key;
2964         int ret = 0;
2965
2966         key.objectid = rec->ino;
2967         key.type = BTRFS_INODE_ITEM_KEY;
2968         key.offset = 0;
2969
2970         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2971         if (ret) {
2972                 if (ret > 0)
2973                         ret = -ENOENT;
2974                 goto out;
2975         }
2976
2977         /* Since ret == 0, no need to check anything */
2978         ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
2979                             struct btrfs_inode_item);
2980         btrfs_set_inode_nbytes(path->nodes[0], ei, rec->found_size);
2981         btrfs_mark_buffer_dirty(path->nodes[0]);
2982         rec->errors &= ~I_ERR_FILE_NBYTES_WRONG;
2983         printf("reset nbytes for ino %llu root %llu\n",
2984                rec->ino, root->root_key.objectid);
2985 out:
2986         btrfs_release_path(path);
2987         return ret;
2988 }
2989
2990 static int add_missing_dir_index(struct btrfs_root *root,
2991                                  struct cache_tree *inode_cache,
2992                                  struct inode_record *rec,
2993                                  struct inode_backref *backref)
2994 {
2995         struct btrfs_path path;
2996         struct btrfs_trans_handle *trans;
2997         struct btrfs_dir_item *dir_item;
2998         struct extent_buffer *leaf;
2999         struct btrfs_key key;
3000         struct btrfs_disk_key disk_key;
3001         struct inode_record *dir_rec;
3002         unsigned long name_ptr;
3003         u32 data_size = sizeof(*dir_item) + backref->namelen;
3004         int ret;
3005
3006         trans = btrfs_start_transaction(root, 1);
3007         if (IS_ERR(trans))
3008                 return PTR_ERR(trans);
3009
3010         fprintf(stderr, "repairing missing dir index item for inode %llu\n",
3011                 (unsigned long long)rec->ino);
3012
3013         btrfs_init_path(&path);
3014         key.objectid = backref->dir;
3015         key.type = BTRFS_DIR_INDEX_KEY;
3016         key.offset = backref->index;
3017         ret = btrfs_insert_empty_item(trans, root, &path, &key, data_size);
3018         BUG_ON(ret);
3019
3020         leaf = path.nodes[0];
3021         dir_item = btrfs_item_ptr(leaf, path.slots[0], struct btrfs_dir_item);
3022
3023         disk_key.objectid = cpu_to_le64(rec->ino);
3024         disk_key.type = BTRFS_INODE_ITEM_KEY;
3025         disk_key.offset = 0;
3026
3027         btrfs_set_dir_item_key(leaf, dir_item, &disk_key);
3028         btrfs_set_dir_type(leaf, dir_item, imode_to_type(rec->imode));
3029         btrfs_set_dir_data_len(leaf, dir_item, 0);
3030         btrfs_set_dir_name_len(leaf, dir_item, backref->namelen);
3031         name_ptr = (unsigned long)(dir_item + 1);
3032         write_extent_buffer(leaf, backref->name, name_ptr, backref->namelen);
3033         btrfs_mark_buffer_dirty(leaf);
3034         btrfs_release_path(&path);
3035         btrfs_commit_transaction(trans, root);
3036
3037         backref->found_dir_index = 1;
3038         dir_rec = get_inode_rec(inode_cache, backref->dir, 0);
3039         BUG_ON(IS_ERR(dir_rec));
3040         if (!dir_rec)
3041                 return 0;
3042         dir_rec->found_size += backref->namelen;
3043         if (dir_rec->found_size == dir_rec->isize &&
3044             (dir_rec->errors & I_ERR_DIR_ISIZE_WRONG))
3045                 dir_rec->errors &= ~I_ERR_DIR_ISIZE_WRONG;
3046         if (dir_rec->found_size != dir_rec->isize)
3047                 dir_rec->errors |= I_ERR_DIR_ISIZE_WRONG;
3048
3049         return 0;
3050 }
3051
3052 static int delete_dir_index(struct btrfs_root *root,
3053                             struct inode_backref *backref)
3054 {
3055         struct btrfs_trans_handle *trans;
3056         struct btrfs_dir_item *di;
3057         struct btrfs_path path;
3058         int ret = 0;
3059
3060         trans = btrfs_start_transaction(root, 1);
3061         if (IS_ERR(trans))
3062                 return PTR_ERR(trans);
3063
3064         fprintf(stderr, "Deleting bad dir index [%llu,%u,%llu] root %llu\n",
3065                 (unsigned long long)backref->dir,
3066                 BTRFS_DIR_INDEX_KEY, (unsigned long long)backref->index,
3067                 (unsigned long long)root->objectid);
3068
3069         btrfs_init_path(&path);
3070         di = btrfs_lookup_dir_index(trans, root, &path, backref->dir,
3071                                     backref->name, backref->namelen,
3072                                     backref->index, -1);
3073         if (IS_ERR(di)) {
3074                 ret = PTR_ERR(di);
3075                 btrfs_release_path(&path);
3076                 btrfs_commit_transaction(trans, root);
3077                 if (ret == -ENOENT)
3078                         return 0;
3079                 return ret;
3080         }
3081
3082         if (!di)
3083                 ret = btrfs_del_item(trans, root, &path);
3084         else
3085                 ret = btrfs_delete_one_dir_name(trans, root, &path, di);
3086         BUG_ON(ret);
3087         btrfs_release_path(&path);
3088         btrfs_commit_transaction(trans, root);
3089         return ret;
3090 }
3091
3092 static int __create_inode_item(struct btrfs_trans_handle *trans,
3093                                struct btrfs_root *root, u64 ino, u64 size,
3094                                u64 nbytes, u64 nlink, u32 mode)
3095 {
3096         struct btrfs_inode_item ii;
3097         time_t now = time(NULL);
3098         int ret;
3099
3100         btrfs_set_stack_inode_size(&ii, size);
3101         btrfs_set_stack_inode_nbytes(&ii, nbytes);
3102         btrfs_set_stack_inode_nlink(&ii, nlink);
3103         btrfs_set_stack_inode_mode(&ii, mode);
3104         btrfs_set_stack_inode_generation(&ii, trans->transid);
3105         btrfs_set_stack_timespec_nsec(&ii.atime, 0);
3106         btrfs_set_stack_timespec_sec(&ii.ctime, now);
3107         btrfs_set_stack_timespec_nsec(&ii.ctime, 0);
3108         btrfs_set_stack_timespec_sec(&ii.mtime, now);
3109         btrfs_set_stack_timespec_nsec(&ii.mtime, 0);
3110         btrfs_set_stack_timespec_sec(&ii.otime, 0);
3111         btrfs_set_stack_timespec_nsec(&ii.otime, 0);
3112
3113         ret = btrfs_insert_inode(trans, root, ino, &ii);
3114         ASSERT(!ret);
3115
3116         warning("root %llu inode %llu recreating inode item, this may "
3117                 "be incomplete, please check permissions and content after "
3118                 "the fsck completes.\n", (unsigned long long)root->objectid,
3119                 (unsigned long long)ino);
3120
3121         return 0;
3122 }
3123
3124 static int create_inode_item_lowmem(struct btrfs_trans_handle *trans,
3125                                     struct btrfs_root *root, u64 ino,
3126                                     u8 filetype)
3127 {
3128         u32 mode = (filetype == BTRFS_FT_DIR ? S_IFDIR : S_IFREG) | 0755;
3129
3130         return __create_inode_item(trans, root, ino, 0, 0, 0, mode);
3131 }
3132
3133 static int create_inode_item(struct btrfs_root *root,
3134                              struct inode_record *rec, int root_dir)
3135 {
3136         struct btrfs_trans_handle *trans;
3137         u64 nlink = 0;
3138         u32 mode = 0;
3139         u64 size = 0;
3140         int ret;
3141
3142         trans = btrfs_start_transaction(root, 1);
3143         if (IS_ERR(trans)) {
3144                 ret = PTR_ERR(trans);
3145                 return ret;
3146         }
3147
3148         nlink = root_dir ? 1 : rec->found_link;
3149         if (rec->found_dir_item) {
3150                 if (rec->found_file_extent)
3151                         fprintf(stderr, "root %llu inode %llu has both a dir "
3152                                 "item and extents, unsure if it is a dir or a "
3153                                 "regular file so setting it as a directory\n",
3154                                 (unsigned long long)root->objectid,
3155                                 (unsigned long long)rec->ino);
3156                 mode = S_IFDIR | 0755;
3157                 size = rec->found_size;
3158         } else if (!rec->found_dir_item) {
3159                 size = rec->extent_end;
3160                 mode =  S_IFREG | 0755;
3161         }
3162
3163         ret = __create_inode_item(trans, root, rec->ino, size, rec->nbytes,
3164                                   nlink, mode);
3165         btrfs_commit_transaction(trans, root);
3166         return 0;
3167 }
3168
3169 static int repair_inode_backrefs(struct btrfs_root *root,
3170                                  struct inode_record *rec,
3171                                  struct cache_tree *inode_cache,
3172                                  int delete)
3173 {
3174         struct inode_backref *tmp, *backref;
3175         u64 root_dirid = btrfs_root_dirid(&root->root_item);
3176         int ret = 0;
3177         int repaired = 0;
3178
3179         list_for_each_entry_safe(backref, tmp, &rec->backrefs, list) {
3180                 if (!delete && rec->ino == root_dirid) {
3181                         if (!rec->found_inode_item) {
3182                                 ret = create_inode_item(root, rec, 1);
3183                                 if (ret)
3184                                         break;
3185                                 repaired++;
3186                         }
3187                 }
3188
3189                 /* Index 0 for root dir's are special, don't mess with it */
3190                 if (rec->ino == root_dirid && backref->index == 0)
3191                         continue;
3192
3193                 if (delete &&
3194                     ((backref->found_dir_index && !backref->found_inode_ref) ||
3195                      (backref->found_dir_index && backref->found_inode_ref &&
3196                       (backref->errors & REF_ERR_INDEX_UNMATCH)))) {
3197                         ret = delete_dir_index(root, backref);
3198                         if (ret)
3199                                 break;
3200                         repaired++;
3201                         list_del(&backref->list);
3202                         free(backref);
3203                         continue;
3204                 }
3205
3206                 if (!delete && !backref->found_dir_index &&
3207                     backref->found_dir_item && backref->found_inode_ref) {
3208                         ret = add_missing_dir_index(root, inode_cache, rec,
3209                                                     backref);
3210                         if (ret)
3211                                 break;
3212                         repaired++;
3213                         if (backref->found_dir_item &&
3214                             backref->found_dir_index) {
3215                                 if (!backref->errors &&
3216                                     backref->found_inode_ref) {
3217                                         list_del(&backref->list);
3218                                         free(backref);
3219                                         continue;
3220                                 }
3221                         }
3222                 }
3223
3224                 if (!delete && (!backref->found_dir_index &&
3225                                 !backref->found_dir_item &&
3226                                 backref->found_inode_ref)) {
3227                         struct btrfs_trans_handle *trans;
3228                         struct btrfs_key location;
3229
3230                         ret = check_dir_conflict(root, backref->name,
3231                                                  backref->namelen,
3232                                                  backref->dir,
3233                                                  backref->index);
3234                         if (ret) {
3235                                 /*
3236                                  * let nlink fixing routine to handle it,
3237                                  * which can do it better.
3238                                  */
3239                                 ret = 0;
3240                                 break;
3241                         }
3242                         location.objectid = rec->ino;
3243                         location.type = BTRFS_INODE_ITEM_KEY;
3244                         location.offset = 0;
3245
3246                         trans = btrfs_start_transaction(root, 1);
3247                         if (IS_ERR(trans)) {
3248                                 ret = PTR_ERR(trans);
3249                                 break;
3250                         }
3251                         fprintf(stderr, "adding missing dir index/item pair "
3252                                 "for inode %llu\n",
3253                                 (unsigned long long)rec->ino);
3254                         ret = btrfs_insert_dir_item(trans, root, backref->name,
3255                                                     backref->namelen,
3256                                                     backref->dir, &location,
3257                                                     imode_to_type(rec->imode),
3258                                                     backref->index);
3259                         BUG_ON(ret);
3260                         btrfs_commit_transaction(trans, root);
3261                         repaired++;
3262                 }
3263
3264                 if (!delete && (backref->found_inode_ref &&
3265                                 backref->found_dir_index &&
3266                                 backref->found_dir_item &&
3267                                 !(backref->errors & REF_ERR_INDEX_UNMATCH) &&
3268                                 !rec->found_inode_item)) {
3269                         ret = create_inode_item(root, rec, 0);
3270                         if (ret)
3271                                 break;
3272                         repaired++;
3273                 }
3274
3275         }
3276         return ret ? ret : repaired;
3277 }
3278
3279 /*
3280  * To determine the file type for nlink/inode_item repair
3281  *
3282  * Return 0 if file type is found and BTRFS_FT_* is stored into type.
3283  * Return -ENOENT if file type is not found.
3284  */
3285 static int find_file_type(struct inode_record *rec, u8 *type)
3286 {
3287         struct inode_backref *backref;
3288
3289         /* For inode item recovered case */
3290         if (rec->found_inode_item) {
3291                 *type = imode_to_type(rec->imode);
3292                 return 0;
3293         }
3294
3295         list_for_each_entry(backref, &rec->backrefs, list) {
3296                 if (backref->found_dir_index || backref->found_dir_item) {
3297                         *type = backref->filetype;
3298                         return 0;
3299                 }
3300         }
3301         return -ENOENT;
3302 }
3303
3304 /*
3305  * To determine the file name for nlink repair
3306  *
3307  * Return 0 if file name is found, set name and namelen.
3308  * Return -ENOENT if file name is not found.
3309  */
3310 static int find_file_name(struct inode_record *rec,
3311                           char *name, int *namelen)
3312 {
3313         struct inode_backref *backref;
3314
3315         list_for_each_entry(backref, &rec->backrefs, list) {
3316                 if (backref->found_dir_index || backref->found_dir_item ||
3317                     backref->found_inode_ref) {
3318                         memcpy(name, backref->name, backref->namelen);
3319                         *namelen = backref->namelen;
3320                         return 0;
3321                 }
3322         }
3323         return -ENOENT;
3324 }
3325
3326 /* Reset the nlink of the inode to the correct one */
3327 static int reset_nlink(struct btrfs_trans_handle *trans,
3328                        struct btrfs_root *root,
3329                        struct btrfs_path *path,
3330                        struct inode_record *rec)
3331 {
3332         struct inode_backref *backref;
3333         struct inode_backref *tmp;
3334         struct btrfs_key key;
3335         struct btrfs_inode_item *inode_item;
3336         int ret = 0;
3337
3338         /* We don't believe this either, reset it and iterate backref */
3339         rec->found_link = 0;
3340
3341         /* Remove all backref including the valid ones */
3342         list_for_each_entry_safe(backref, tmp, &rec->backrefs, list) {
3343                 ret = btrfs_unlink(trans, root, rec->ino, backref->dir,
3344                                    backref->index, backref->name,
3345                                    backref->namelen, 0);
3346                 if (ret < 0)
3347                         goto out;
3348
3349                 /* remove invalid backref, so it won't be added back */
3350                 if (!(backref->found_dir_index &&
3351                       backref->found_dir_item &&
3352                       backref->found_inode_ref)) {
3353                         list_del(&backref->list);
3354                         free(backref);
3355                 } else {
3356                         rec->found_link++;
3357                 }
3358         }
3359
3360         /* Set nlink to 0 */
3361         key.objectid = rec->ino;
3362         key.type = BTRFS_INODE_ITEM_KEY;
3363         key.offset = 0;
3364         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
3365         if (ret < 0)
3366                 goto out;
3367         if (ret > 0) {
3368                 ret = -ENOENT;
3369                 goto out;
3370         }
3371         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3372                                     struct btrfs_inode_item);
3373         btrfs_set_inode_nlink(path->nodes[0], inode_item, 0);
3374         btrfs_mark_buffer_dirty(path->nodes[0]);
3375         btrfs_release_path(path);
3376
3377         /*
3378          * Add back valid inode_ref/dir_item/dir_index,
3379          * add_link() will handle the nlink inc, so new nlink must be correct
3380          */
3381         list_for_each_entry(backref, &rec->backrefs, list) {
3382                 ret = btrfs_add_link(trans, root, rec->ino, backref->dir,
3383                                      backref->name, backref->namelen,
3384                                      backref->filetype, &backref->index, 1, 0);
3385                 if (ret < 0)
3386                         goto out;
3387         }
3388 out:
3389         btrfs_release_path(path);
3390         return ret;
3391 }
3392
3393 static int get_highest_inode(struct btrfs_trans_handle *trans,
3394                                 struct btrfs_root *root,
3395                                 struct btrfs_path *path,
3396                                 u64 *highest_ino)
3397 {
3398         struct btrfs_key key, found_key;
3399         int ret;
3400
3401         btrfs_init_path(path);
3402         key.objectid = BTRFS_LAST_FREE_OBJECTID;
3403         key.offset = -1;
3404         key.type = BTRFS_INODE_ITEM_KEY;
3405         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3406         if (ret == 1) {
3407                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3408                                 path->slots[0] - 1);
3409                 *highest_ino = found_key.objectid;
3410                 ret = 0;
3411         }
3412         if (*highest_ino >= BTRFS_LAST_FREE_OBJECTID)
3413                 ret = -EOVERFLOW;
3414         btrfs_release_path(path);
3415         return ret;
3416 }
3417
3418 /*
3419  * Link inode to dir 'lost+found'. Increase @ref_count.
3420  *
3421  * Returns 0 means success.
3422  * Returns <0 means failure.
3423  */
3424 static int link_inode_to_lostfound(struct btrfs_trans_handle *trans,
3425                                    struct btrfs_root *root,
3426                                    struct btrfs_path *path,
3427                                    u64 ino, char *namebuf, u32 name_len,
3428                                    u8 filetype, u64 *ref_count)
3429 {
3430         char *dir_name = "lost+found";
3431         u64 lost_found_ino;
3432         int ret;
3433         u32 mode = 0700;
3434
3435         btrfs_release_path(path);
3436         ret = get_highest_inode(trans, root, path, &lost_found_ino);
3437         if (ret < 0)
3438                 goto out;
3439         lost_found_ino++;
3440
3441         ret = btrfs_mkdir(trans, root, dir_name, strlen(dir_name),
3442                           BTRFS_FIRST_FREE_OBJECTID, &lost_found_ino,
3443                           mode);
3444         if (ret < 0) {
3445                 error("failed to create '%s' dir: %s", dir_name, strerror(-ret));
3446                 goto out;
3447         }
3448         ret = btrfs_add_link(trans, root, ino, lost_found_ino,
3449                              namebuf, name_len, filetype, NULL, 1, 0);
3450         /*
3451          * Add ".INO" suffix several times to handle case where
3452          * "FILENAME.INO" is already taken by another file.
3453          */
3454         while (ret == -EEXIST) {
3455                 /*
3456                  * Conflicting file name, add ".INO" as suffix * +1 for '.'
3457                  */
3458                 if (name_len + count_digits(ino) + 1 > BTRFS_NAME_LEN) {
3459                         ret = -EFBIG;
3460                         goto out;
3461                 }
3462                 snprintf(namebuf + name_len, BTRFS_NAME_LEN - name_len,
3463                          ".%llu", ino);
3464                 name_len += count_digits(ino) + 1;
3465                 ret = btrfs_add_link(trans, root, ino, lost_found_ino, namebuf,
3466                                      name_len, filetype, NULL, 1, 0);
3467         }
3468         if (ret < 0) {
3469                 error("failed to link the inode %llu to %s dir: %s",
3470                       ino, dir_name, strerror(-ret));
3471                 goto out;
3472         }
3473
3474         ++*ref_count;
3475         printf("Moving file '%.*s' to '%s' dir since it has no valid backref\n",
3476                name_len, namebuf, dir_name);
3477 out:
3478         btrfs_release_path(path);
3479         if (ret)
3480                 error("failed to move file '%.*s' to '%s' dir", name_len,
3481                                 namebuf, dir_name);
3482         return ret;
3483 }
3484
3485 static int repair_inode_nlinks(struct btrfs_trans_handle *trans,
3486                                struct btrfs_root *root,
3487                                struct btrfs_path *path,
3488                                struct inode_record *rec)
3489 {
3490         char namebuf[BTRFS_NAME_LEN] = {0};
3491         u8 type = 0;
3492         int namelen = 0;
3493         int name_recovered = 0;
3494         int type_recovered = 0;
3495         int ret = 0;
3496
3497         /*
3498          * Get file name and type first before these invalid inode ref
3499          * are deleted by remove_all_invalid_backref()
3500          */
3501         name_recovered = !find_file_name(rec, namebuf, &namelen);
3502         type_recovered = !find_file_type(rec, &type);
3503
3504         if (!name_recovered) {
3505                 printf("Can't get file name for inode %llu, using '%llu' as fallback\n",
3506                        rec->ino, rec->ino);
3507                 namelen = count_digits(rec->ino);
3508                 sprintf(namebuf, "%llu", rec->ino);
3509                 name_recovered = 1;
3510         }
3511         if (!type_recovered) {
3512                 printf("Can't get file type for inode %llu, using FILE as fallback\n",
3513                        rec->ino);
3514                 type = BTRFS_FT_REG_FILE;
3515                 type_recovered = 1;
3516         }
3517
3518         ret = reset_nlink(trans, root, path, rec);
3519         if (ret < 0) {
3520                 fprintf(stderr,
3521                         "Failed to reset nlink for inode %llu: %s\n",
3522                         rec->ino, strerror(-ret));
3523                 goto out;
3524         }
3525
3526         if (rec->found_link == 0) {
3527                 ret = link_inode_to_lostfound(trans, root, path, rec->ino,
3528                                               namebuf, namelen, type,
3529                                               (u64 *)&rec->found_link);
3530                 if (ret)
3531                         goto out;
3532         }
3533         printf("Fixed the nlink of inode %llu\n", rec->ino);
3534 out:
3535         /*
3536          * Clear the flag anyway, or we will loop forever for the same inode
3537          * as it will not be removed from the bad inode list and the dead loop
3538          * happens.
3539          */
3540         rec->errors &= ~I_ERR_LINK_COUNT_WRONG;
3541         btrfs_release_path(path);
3542         return ret;
3543 }
3544
3545 /*
3546  * Check if there is any normal(reg or prealloc) file extent for given
3547  * ino.
3548  * This is used to determine the file type when neither its dir_index/item or
3549  * inode_item exists.
3550  *
3551  * This will *NOT* report error, if any error happens, just consider it does
3552  * not have any normal file extent.
3553  */
3554 static int find_normal_file_extent(struct btrfs_root *root, u64 ino)
3555 {
3556         struct btrfs_path path;
3557         struct btrfs_key key;
3558         struct btrfs_key found_key;
3559         struct btrfs_file_extent_item *fi;
3560         u8 type;
3561         int ret = 0;
3562
3563         btrfs_init_path(&path);
3564         key.objectid = ino;
3565         key.type = BTRFS_EXTENT_DATA_KEY;
3566         key.offset = 0;
3567
3568         ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
3569         if (ret < 0) {
3570                 ret = 0;
3571                 goto out;
3572         }
3573         if (ret && path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
3574                 ret = btrfs_next_leaf(root, &path);
3575                 if (ret) {
3576                         ret = 0;
3577                         goto out;
3578                 }
3579         }
3580         while (1) {
3581                 btrfs_item_key_to_cpu(path.nodes[0], &found_key,
3582                                       path.slots[0]);
3583                 if (found_key.objectid != ino ||
3584                     found_key.type != BTRFS_EXTENT_DATA_KEY)
3585                         break;
3586                 fi = btrfs_item_ptr(path.nodes[0], path.slots[0],
3587                                     struct btrfs_file_extent_item);
3588                 type = btrfs_file_extent_type(path.nodes[0], fi);
3589                 if (type != BTRFS_FILE_EXTENT_INLINE) {
3590                         ret = 1;
3591                         goto out;
3592                 }
3593         }
3594 out:
3595         btrfs_release_path(&path);
3596         return ret;
3597 }
3598
3599 static u32 btrfs_type_to_imode(u8 type)
3600 {
3601         static u32 imode_by_btrfs_type[] = {
3602                 [BTRFS_FT_REG_FILE]     = S_IFREG,
3603                 [BTRFS_FT_DIR]          = S_IFDIR,
3604                 [BTRFS_FT_CHRDEV]       = S_IFCHR,
3605                 [BTRFS_FT_BLKDEV]       = S_IFBLK,
3606                 [BTRFS_FT_FIFO]         = S_IFIFO,
3607                 [BTRFS_FT_SOCK]         = S_IFSOCK,
3608                 [BTRFS_FT_SYMLINK]      = S_IFLNK,
3609         };
3610
3611         return imode_by_btrfs_type[(type)];
3612 }
3613
3614 static int repair_inode_no_item(struct btrfs_trans_handle *trans,
3615                                 struct btrfs_root *root,
3616                                 struct btrfs_path *path,
3617                                 struct inode_record *rec)
3618 {
3619         u8 filetype;
3620         u32 mode = 0700;
3621         int type_recovered = 0;
3622         int ret = 0;
3623
3624         printf("Trying to rebuild inode:%llu\n", rec->ino);
3625
3626         type_recovered = !find_file_type(rec, &filetype);
3627
3628         /*
3629          * Try to determine inode type if type not found.
3630          *
3631          * For found regular file extent, it must be FILE.
3632          * For found dir_item/index, it must be DIR.
3633          *
3634          * For undetermined one, use FILE as fallback.
3635          *
3636          * TODO:
3637          * 1. If found backref(inode_index/item is already handled) to it,
3638          *    it must be DIR.
3639          *    Need new inode-inode ref structure to allow search for that.
3640          */
3641         if (!type_recovered) {
3642                 if (rec->found_file_extent &&
3643                     find_normal_file_extent(root, rec->ino)) {
3644                         type_recovered = 1;
3645                         filetype = BTRFS_FT_REG_FILE;
3646                 } else if (rec->found_dir_item) {
3647                         type_recovered = 1;
3648                         filetype = BTRFS_FT_DIR;
3649                 } else if (!list_empty(&rec->orphan_extents)) {
3650                         type_recovered = 1;
3651                         filetype = BTRFS_FT_REG_FILE;
3652                 } else{
3653                         printf("Can't determine the filetype for inode %llu, assume it is a normal file\n",
3654                                rec->ino);
3655                         type_recovered = 1;
3656                         filetype = BTRFS_FT_REG_FILE;
3657                 }
3658         }
3659
3660         ret = btrfs_new_inode(trans, root, rec->ino,
3661                               mode | btrfs_type_to_imode(filetype));
3662         if (ret < 0)
3663                 goto out;
3664
3665         /*
3666          * Here inode rebuild is done, we only rebuild the inode item,
3667          * don't repair the nlink(like move to lost+found).
3668          * That is the job of nlink repair.
3669          *
3670          * We just fill the record and return
3671          */
3672         rec->found_dir_item = 1;
3673         rec->imode = mode | btrfs_type_to_imode(filetype);
3674         rec->nlink = 0;
3675         rec->errors &= ~I_ERR_NO_INODE_ITEM;
3676         /* Ensure the inode_nlinks repair function will be called */
3677         rec->errors |= I_ERR_LINK_COUNT_WRONG;
3678 out:
3679         return ret;
3680 }
3681
3682 static int repair_inode_orphan_extent(struct btrfs_trans_handle *trans,
3683                                       struct btrfs_root *root,
3684                                       struct btrfs_path *path,
3685                                       struct inode_record *rec)
3686 {
3687         struct orphan_data_extent *orphan;
3688         struct orphan_data_extent *tmp;
3689         int ret = 0;
3690
3691         list_for_each_entry_safe(orphan, tmp, &rec->orphan_extents, list) {
3692                 /*
3693                  * Check for conflicting file extents
3694                  *
3695                  * Here we don't know whether the extents is compressed or not,
3696                  * so we can only assume it not compressed nor data offset,
3697                  * and use its disk_len as extent length.
3698                  */
3699                 ret = btrfs_get_extent(NULL, root, path, orphan->objectid,
3700                                        orphan->offset, orphan->disk_len, 0);
3701                 btrfs_release_path(path);
3702                 if (ret < 0)
3703                         goto out;
3704                 if (!ret) {
3705                         fprintf(stderr,
3706                                 "orphan extent (%llu, %llu) conflicts, delete the orphan\n",
3707                                 orphan->disk_bytenr, orphan->disk_len);
3708                         ret = btrfs_free_extent(trans,
3709                                         root->fs_info->extent_root,
3710                                         orphan->disk_bytenr, orphan->disk_len,
3711                                         0, root->objectid, orphan->objectid,
3712                                         orphan->offset);
3713                         if (ret < 0)
3714                                 goto out;
3715                 }
3716                 ret = btrfs_insert_file_extent(trans, root, orphan->objectid,
3717                                 orphan->offset, orphan->disk_bytenr,
3718                                 orphan->disk_len, orphan->disk_len);
3719                 if (ret < 0)
3720                         goto out;
3721
3722                 /* Update file size info */
3723                 rec->found_size += orphan->disk_len;
3724                 if (rec->found_size == rec->nbytes)
3725                         rec->errors &= ~I_ERR_FILE_NBYTES_WRONG;
3726
3727                 /* Update the file extent hole info too */
3728                 ret = del_file_extent_hole(&rec->holes, orphan->offset,
3729                                            orphan->disk_len);
3730                 if (ret < 0)
3731                         goto out;
3732                 if (RB_EMPTY_ROOT(&rec->holes))
3733                         rec->errors &= ~I_ERR_FILE_EXTENT_DISCOUNT;
3734
3735                 list_del(&orphan->list);
3736                 free(orphan);
3737         }
3738         rec->errors &= ~I_ERR_FILE_EXTENT_ORPHAN;
3739 out:
3740         return ret;
3741 }
3742
3743 static int repair_inode_discount_extent(struct btrfs_trans_handle *trans,
3744                                         struct btrfs_root *root,
3745                                         struct btrfs_path *path,
3746                                         struct inode_record *rec)
3747 {
3748         struct rb_node *node;
3749         struct file_extent_hole *hole;
3750         int found = 0;
3751         int ret = 0;
3752
3753         node = rb_first(&rec->holes);
3754
3755         while (node) {
3756                 found = 1;
3757                 hole = rb_entry(node, struct file_extent_hole, node);
3758                 ret = btrfs_punch_hole(trans, root, rec->ino,
3759                                        hole->start, hole->len);
3760                 if (ret < 0)
3761                         goto out;
3762                 ret = del_file_extent_hole(&rec->holes, hole->start,
3763                                            hole->len);
3764                 if (ret < 0)
3765                         goto out;
3766                 if (RB_EMPTY_ROOT(&rec->holes))
3767                         rec->errors &= ~I_ERR_FILE_EXTENT_DISCOUNT;
3768                 node = rb_first(&rec->holes);
3769         }
3770         /* special case for a file losing all its file extent */
3771         if (!found) {
3772                 ret = btrfs_punch_hole(trans, root, rec->ino, 0,
3773                                        round_up(rec->isize,
3774                                                 root->fs_info->sectorsize));
3775                 if (ret < 0)
3776                         goto out;
3777         }
3778         printf("Fixed discount file extents for inode: %llu in root: %llu\n",
3779                rec->ino, root->objectid);
3780 out:
3781         return ret;
3782 }
3783
3784 static int try_repair_inode(struct btrfs_root *root, struct inode_record *rec)
3785 {
3786         struct btrfs_trans_handle *trans;
3787         struct btrfs_path path;
3788         int ret = 0;
3789
3790         if (!(rec->errors & (I_ERR_DIR_ISIZE_WRONG |
3791                              I_ERR_NO_ORPHAN_ITEM |
3792                              I_ERR_LINK_COUNT_WRONG |
3793                              I_ERR_NO_INODE_ITEM |
3794                              I_ERR_FILE_EXTENT_ORPHAN |
3795                              I_ERR_FILE_EXTENT_DISCOUNT|
3796                              I_ERR_FILE_NBYTES_WRONG)))
3797                 return rec->errors;
3798
3799         /*
3800          * For nlink repair, it may create a dir and add link, so
3801          * 2 for parent(256)'s dir_index and dir_item
3802          * 2 for lost+found dir's inode_item and inode_ref
3803          * 1 for the new inode_ref of the file
3804          * 2 for lost+found dir's dir_index and dir_item for the file
3805          */
3806         trans = btrfs_start_transaction(root, 7);
3807         if (IS_ERR(trans))
3808                 return PTR_ERR(trans);
3809
3810         btrfs_init_path(&path);
3811         if (rec->errors & I_ERR_NO_INODE_ITEM)
3812                 ret = repair_inode_no_item(trans, root, &path, rec);
3813         if (!ret && rec->errors & I_ERR_FILE_EXTENT_ORPHAN)
3814                 ret = repair_inode_orphan_extent(trans, root, &path, rec);
3815         if (!ret && rec->errors & I_ERR_FILE_EXTENT_DISCOUNT)
3816                 ret = repair_inode_discount_extent(trans, root, &path, rec);
3817         if (!ret && rec->errors & I_ERR_DIR_ISIZE_WRONG)
3818                 ret = repair_inode_isize(trans, root, &path, rec);
3819         if (!ret && rec->errors & I_ERR_NO_ORPHAN_ITEM)
3820                 ret = repair_inode_orphan_item(trans, root, &path, rec);
3821         if (!ret && rec->errors & I_ERR_LINK_COUNT_WRONG)
3822                 ret = repair_inode_nlinks(trans, root, &path, rec);
3823         if (!ret && rec->errors & I_ERR_FILE_NBYTES_WRONG)
3824                 ret = repair_inode_nbytes(trans, root, &path, rec);
3825         btrfs_commit_transaction(trans, root);
3826         btrfs_release_path(&path);
3827         return ret;
3828 }
3829
3830 static int check_inode_recs(struct btrfs_root *root,
3831                             struct cache_tree *inode_cache)
3832 {
3833         struct cache_extent *cache;
3834         struct ptr_node *node;
3835         struct inode_record *rec;
3836         struct inode_backref *backref;
3837         int stage = 0;
3838         int ret = 0;
3839         int err = 0;
3840         u64 error = 0;
3841         u64 root_dirid = btrfs_root_dirid(&root->root_item);
3842
3843         if (btrfs_root_refs(&root->root_item) == 0) {
3844                 if (!cache_tree_empty(inode_cache))
3845                         fprintf(stderr, "warning line %d\n", __LINE__);
3846                 return 0;
3847         }
3848
3849         /*
3850          * We need to repair backrefs first because we could change some of the
3851          * errors in the inode recs.
3852          *
3853          * We also need to go through and delete invalid backrefs first and then
3854          * add the correct ones second.  We do this because we may get EEXIST
3855          * when adding back the correct index because we hadn't yet deleted the
3856          * invalid index.
3857          *
3858          * For example, if we were missing a dir index then the directories
3859          * isize would be wrong, so if we fixed the isize to what we thought it
3860          * would be and then fixed the backref we'd still have a invalid fs, so
3861          * we need to add back the dir index and then check to see if the isize
3862          * is still wrong.
3863          */
3864         while (stage < 3) {
3865                 stage++;
3866                 if (stage == 3 && !err)
3867                         break;
3868
3869                 cache = search_cache_extent(inode_cache, 0);
3870                 while (repair && cache) {
3871                         node = container_of(cache, struct ptr_node, cache);
3872                         rec = node->data;
3873                         cache = next_cache_extent(cache);
3874
3875                         /* Need to free everything up and rescan */
3876                         if (stage == 3) {
3877                                 remove_cache_extent(inode_cache, &node->cache);
3878                                 free(node);
3879                                 free_inode_rec(rec);
3880                                 continue;
3881                         }
3882
3883                         if (list_empty(&rec->backrefs))
3884                                 continue;
3885
3886                         ret = repair_inode_backrefs(root, rec, inode_cache,
3887                                                     stage == 1);
3888                         if (ret < 0) {
3889                                 err = ret;
3890                                 stage = 2;
3891                                 break;
3892                         } if (ret > 0) {
3893                                 err = -EAGAIN;
3894                         }
3895                 }
3896         }
3897         if (err)
3898                 return err;
3899
3900         rec = get_inode_rec(inode_cache, root_dirid, 0);
3901         BUG_ON(IS_ERR(rec));
3902         if (rec) {
3903                 ret = check_root_dir(rec);
3904                 if (ret) {
3905                         fprintf(stderr, "root %llu root dir %llu error\n",
3906                                 (unsigned long long)root->root_key.objectid,
3907                                 (unsigned long long)root_dirid);
3908                         print_inode_error(root, rec);
3909                         error++;
3910                 }
3911         } else {
3912                 if (repair) {
3913                         struct btrfs_trans_handle *trans;
3914
3915                         trans = btrfs_start_transaction(root, 1);
3916                         if (IS_ERR(trans)) {
3917                                 err = PTR_ERR(trans);
3918                                 return err;
3919                         }
3920
3921                         fprintf(stderr,
3922                                 "root %llu missing its root dir, recreating\n",
3923                                 (unsigned long long)root->objectid);
3924
3925                         ret = btrfs_make_root_dir(trans, root, root_dirid);
3926                         BUG_ON(ret);
3927
3928                         btrfs_commit_transaction(trans, root);
3929                         return -EAGAIN;
3930                 }
3931
3932                 fprintf(stderr, "root %llu root dir %llu not found\n",
3933                         (unsigned long long)root->root_key.objectid,
3934                         (unsigned long long)root_dirid);
3935         }
3936
3937         while (1) {
3938                 cache = search_cache_extent(inode_cache, 0);
3939                 if (!cache)
3940                         break;
3941                 node = container_of(cache, struct ptr_node, cache);
3942                 rec = node->data;
3943                 remove_cache_extent(inode_cache, &node->cache);
3944                 free(node);
3945                 if (rec->ino == root_dirid ||
3946                     rec->ino == BTRFS_ORPHAN_OBJECTID) {
3947                         free_inode_rec(rec);
3948                         continue;
3949                 }
3950
3951                 if (rec->errors & I_ERR_NO_ORPHAN_ITEM) {
3952                         ret = check_orphan_item(root, rec->ino);
3953                         if (ret == 0)
3954                                 rec->errors &= ~I_ERR_NO_ORPHAN_ITEM;
3955                         if (can_free_inode_rec(rec)) {
3956                                 free_inode_rec(rec);
3957                                 continue;
3958                         }
3959                 }
3960
3961                 if (!rec->found_inode_item)
3962                         rec->errors |= I_ERR_NO_INODE_ITEM;
3963                 if (rec->found_link != rec->nlink)
3964                         rec->errors |= I_ERR_LINK_COUNT_WRONG;
3965                 if (repair) {
3966                         ret = try_repair_inode(root, rec);
3967                         if (ret == 0 && can_free_inode_rec(rec)) {
3968                                 free_inode_rec(rec);
3969                                 continue;
3970                         }
3971                         ret = 0;
3972                 }
3973
3974                 if (!(repair && ret == 0))
3975                         error++;
3976                 print_inode_error(root, rec);
3977                 list_for_each_entry(backref, &rec->backrefs, list) {
3978                         if (!backref->found_dir_item)
3979                                 backref->errors |= REF_ERR_NO_DIR_ITEM;
3980                         if (!backref->found_dir_index)
3981                                 backref->errors |= REF_ERR_NO_DIR_INDEX;
3982                         if (!backref->found_inode_ref)
3983                                 backref->errors |= REF_ERR_NO_INODE_REF;
3984                         fprintf(stderr, "\tunresolved ref dir %llu index %llu"
3985                                 " namelen %u name %s filetype %d errors %x",
3986                                 (unsigned long long)backref->dir,
3987                                 (unsigned long long)backref->index,
3988                                 backref->namelen, backref->name,
3989                                 backref->filetype, backref->errors);
3990                         print_ref_error(backref->errors);
3991                 }
3992                 free_inode_rec(rec);
3993         }
3994         return (error > 0) ? -1 : 0;
3995 }
3996
3997 static struct root_record *get_root_rec(struct cache_tree *root_cache,
3998                                         u64 objectid)
3999 {
4000         struct cache_extent *cache;
4001         struct root_record *rec = NULL;
4002         int ret;
4003
4004         cache = lookup_cache_extent(root_cache, objectid, 1);
4005         if (cache) {
4006                 rec = container_of(cache, struct root_record, cache);
4007         } else {
4008                 rec = calloc(1, sizeof(*rec));
4009                 if (!rec)
4010                         return ERR_PTR(-ENOMEM);
4011                 rec->objectid = objectid;
4012                 INIT_LIST_HEAD(&rec->backrefs);
4013                 rec->cache.start = objectid;
4014                 rec->cache.size = 1;
4015
4016                 ret = insert_cache_extent(root_cache, &rec->cache);
4017                 if (ret)
4018                         return ERR_PTR(-EEXIST);
4019         }
4020         return rec;
4021 }
4022
4023 static struct root_backref *get_root_backref(struct root_record *rec,
4024                                              u64 ref_root, u64 dir, u64 index,
4025                                              const char *name, int namelen)
4026 {
4027         struct root_backref *backref;
4028
4029         list_for_each_entry(backref, &rec->backrefs, list) {
4030                 if (backref->ref_root != ref_root || backref->dir != dir ||
4031                     backref->namelen != namelen)
4032                         continue;
4033                 if (memcmp(name, backref->name, namelen))
4034                         continue;
4035                 return backref;
4036         }
4037
4038         backref = calloc(1, sizeof(*backref) + namelen + 1);
4039         if (!backref)
4040                 return NULL;
4041         backref->ref_root = ref_root;
4042         backref->dir = dir;
4043         backref->index = index;
4044         backref->namelen = namelen;
4045         memcpy(backref->name, name, namelen);
4046         backref->name[namelen] = '\0';
4047         list_add_tail(&backref->list, &rec->backrefs);
4048         return backref;
4049 }
4050
4051 static void free_root_record(struct cache_extent *cache)
4052 {
4053         struct root_record *rec;
4054         struct root_backref *backref;
4055
4056         rec = container_of(cache, struct root_record, cache);
4057         while (!list_empty(&rec->backrefs)) {
4058                 backref = to_root_backref(rec->backrefs.next);
4059                 list_del(&backref->list);
4060                 free(backref);
4061         }
4062
4063         free(rec);
4064 }
4065
4066 FREE_EXTENT_CACHE_BASED_TREE(root_recs, free_root_record);
4067
4068 static int add_root_backref(struct cache_tree *root_cache,
4069                             u64 root_id, u64 ref_root, u64 dir, u64 index,
4070                             const char *name, int namelen,
4071                             int item_type, int errors)
4072 {
4073         struct root_record *rec;
4074         struct root_backref *backref;
4075
4076         rec = get_root_rec(root_cache, root_id);
4077         BUG_ON(IS_ERR(rec));
4078         backref = get_root_backref(rec, ref_root, dir, index, name, namelen);
4079         BUG_ON(!backref);
4080
4081         backref->errors |= errors;
4082
4083         if (item_type != BTRFS_DIR_ITEM_KEY) {
4084                 if (backref->found_dir_index || backref->found_back_ref ||
4085                     backref->found_forward_ref) {
4086                         if (backref->index != index)
4087                                 backref->errors |= REF_ERR_INDEX_UNMATCH;
4088                 } else {
4089                         backref->index = index;
4090                 }
4091         }
4092
4093         if (item_type == BTRFS_DIR_ITEM_KEY) {
4094                 if (backref->found_forward_ref)
4095                         rec->found_ref++;
4096                 backref->found_dir_item = 1;
4097         } else if (item_type == BTRFS_DIR_INDEX_KEY) {
4098                 backref->found_dir_index = 1;
4099         } else if (item_type == BTRFS_ROOT_REF_KEY) {
4100                 if (backref->found_forward_ref)
4101                         backref->errors |= REF_ERR_DUP_ROOT_REF;
4102                 else if (backref->found_dir_item)
4103                         rec->found_ref++;
4104                 backref->found_forward_ref = 1;
4105         } else if (item_type == BTRFS_ROOT_BACKREF_KEY) {
4106                 if (backref->found_back_ref)
4107                         backref->errors |= REF_ERR_DUP_ROOT_BACKREF;
4108                 backref->found_back_ref = 1;
4109         } else {
4110                 BUG_ON(1);
4111         }
4112
4113         if (backref->found_forward_ref && backref->found_dir_item)
4114                 backref->reachable = 1;
4115         return 0;
4116 }
4117
4118 static int merge_root_recs(struct btrfs_root *root,
4119                            struct cache_tree *src_cache,
4120                            struct cache_tree *dst_cache)
4121 {
4122         struct cache_extent *cache;
4123         struct ptr_node *node;
4124         struct inode_record *rec;
4125         struct inode_backref *backref;
4126         int ret = 0;
4127
4128         if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
4129                 free_inode_recs_tree(src_cache);
4130                 return 0;
4131         }
4132
4133         while (1) {
4134                 cache = search_cache_extent(src_cache, 0);
4135                 if (!cache)
4136                         break;
4137                 node = container_of(cache, struct ptr_node, cache);
4138                 rec = node->data;
4139                 remove_cache_extent(src_cache, &node->cache);
4140                 free(node);
4141
4142                 ret = is_child_root(root, root->objectid, rec->ino);
4143                 if (ret < 0)
4144                         break;
4145                 else if (ret == 0)
4146                         goto skip;
4147
4148                 list_for_each_entry(backref, &rec->backrefs, list) {
4149                         BUG_ON(backref->found_inode_ref);
4150                         if (backref->found_dir_item)
4151                                 add_root_backref(dst_cache, rec->ino,
4152                                         root->root_key.objectid, backref->dir,
4153                                         backref->index, backref->name,
4154                                         backref->namelen, BTRFS_DIR_ITEM_KEY,
4155                                         backref->errors);
4156                         if (backref->found_dir_index)
4157                                 add_root_backref(dst_cache, rec->ino,
4158                                         root->root_key.objectid, backref->dir,
4159                                         backref->index, backref->name,
4160                                         backref->namelen, BTRFS_DIR_INDEX_KEY,
4161                                         backref->errors);
4162                 }
4163 skip:
4164                 free_inode_rec(rec);
4165         }
4166         if (ret < 0)
4167                 return ret;
4168         return 0;
4169 }
4170
4171 static int check_root_refs(struct btrfs_root *root,
4172                            struct cache_tree *root_cache)
4173 {
4174         struct root_record *rec;
4175         struct root_record *ref_root;
4176         struct root_backref *backref;
4177         struct cache_extent *cache;
4178         int loop = 1;
4179         int ret;
4180         int error;
4181         int errors = 0;
4182
4183         rec = get_root_rec(root_cache, BTRFS_FS_TREE_OBJECTID);
4184         BUG_ON(IS_ERR(rec));
4185         rec->found_ref = 1;
4186
4187         /* fixme: this can not detect circular references */
4188         while (loop) {
4189                 loop = 0;
4190                 cache = search_cache_extent(root_cache, 0);
4191                 while (1) {
4192                         if (!cache)
4193                                 break;
4194                         rec = container_of(cache, struct root_record, cache);
4195                         cache = next_cache_extent(cache);
4196
4197                         if (rec->found_ref == 0)
4198                                 continue;
4199
4200                         list_for_each_entry(backref, &rec->backrefs, list) {
4201                                 if (!backref->reachable)
4202                                         continue;
4203
4204                                 ref_root = get_root_rec(root_cache,
4205                                                         backref->ref_root);
4206                                 BUG_ON(IS_ERR(ref_root));
4207                                 if (ref_root->found_ref > 0)
4208                                         continue;
4209
4210                                 backref->reachable = 0;
4211                                 rec->found_ref--;
4212                                 if (rec->found_ref == 0)
4213                                         loop = 1;
4214                         }
4215                 }
4216         }
4217
4218         cache = search_cache_extent(root_cache, 0);
4219         while (1) {
4220                 if (!cache)
4221                         break;
4222                 rec = container_of(cache, struct root_record, cache);
4223                 cache = next_cache_extent(cache);
4224
4225                 if (rec->found_ref == 0 &&
4226                     rec->objectid >= BTRFS_FIRST_FREE_OBJECTID &&
4227                     rec->objectid <= BTRFS_LAST_FREE_OBJECTID) {
4228                         ret = check_orphan_item(root->fs_info->tree_root,
4229                                                 rec->objectid);
4230                         if (ret == 0)
4231                                 continue;
4232
4233                         /*
4234                          * If we don't have a root item then we likely just have
4235                          * a dir item in a snapshot for this root but no actual
4236                          * ref key or anything so it's meaningless.
4237                          */
4238                         if (!rec->found_root_item)
4239                                 continue;
4240                         errors++;
4241                         fprintf(stderr, "fs tree %llu not referenced\n",
4242                                 (unsigned long long)rec->objectid);
4243                 }
4244
4245                 error = 0;
4246                 if (rec->found_ref > 0 && !rec->found_root_item)
4247                         error = 1;
4248                 list_for_each_entry(backref, &rec->backrefs, list) {
4249                         if (!backref->found_dir_item)
4250                                 backref->errors |= REF_ERR_NO_DIR_ITEM;
4251                         if (!backref->found_dir_index)
4252                                 backref->errors |= REF_ERR_NO_DIR_INDEX;
4253                         if (!backref->found_back_ref)
4254                                 backref->errors |= REF_ERR_NO_ROOT_BACKREF;
4255                         if (!backref->found_forward_ref)
4256                                 backref->errors |= REF_ERR_NO_ROOT_REF;
4257                         if (backref->reachable && backref->errors)
4258                                 error = 1;
4259                 }
4260                 if (!error)
4261                         continue;
4262
4263                 errors++;
4264                 fprintf(stderr, "fs tree %llu refs %u %s\n",
4265                         (unsigned long long)rec->objectid, rec->found_ref,
4266                          rec->found_root_item ? "" : "not found");
4267
4268                 list_for_each_entry(backref, &rec->backrefs, list) {
4269                         if (!backref->reachable)
4270                                 continue;
4271                         if (!backref->errors && rec->found_root_item)
4272                                 continue;
4273                         fprintf(stderr, "\tunresolved ref root %llu dir %llu"
4274                                 " index %llu namelen %u name %s errors %x\n",
4275                                 (unsigned long long)backref->ref_root,
4276                                 (unsigned long long)backref->dir,
4277                                 (unsigned long long)backref->index,
4278                                 backref->namelen, backref->name,
4279                                 backref->errors);
4280                         print_ref_error(backref->errors);
4281                 }
4282         }
4283         return errors > 0 ? 1 : 0;
4284 }
4285
4286 static int process_root_ref(struct extent_buffer *eb, int slot,
4287                             struct btrfs_key *key,
4288                             struct cache_tree *root_cache)
4289 {
4290         u64 dirid;
4291         u64 index;
4292         u32 len;
4293         u32 name_len;
4294         struct btrfs_root_ref *ref;
4295         char namebuf[BTRFS_NAME_LEN];
4296         int error;
4297
4298         ref = btrfs_item_ptr(eb, slot, struct btrfs_root_ref);
4299
4300         dirid = btrfs_root_ref_dirid(eb, ref);
4301         index = btrfs_root_ref_sequence(eb, ref);
4302         name_len = btrfs_root_ref_name_len(eb, ref);
4303
4304         if (name_len <= BTRFS_NAME_LEN) {
4305                 len = name_len;
4306                 error = 0;
4307         } else {
4308                 len = BTRFS_NAME_LEN;
4309                 error = REF_ERR_NAME_TOO_LONG;
4310         }
4311         read_extent_buffer(eb, namebuf, (unsigned long)(ref + 1), len);
4312
4313         if (key->type == BTRFS_ROOT_REF_KEY) {
4314                 add_root_backref(root_cache, key->offset, key->objectid, dirid,
4315                                  index, namebuf, len, key->type, error);
4316         } else {
4317                 add_root_backref(root_cache, key->objectid, key->offset, dirid,
4318                                  index, namebuf, len, key->type, error);
4319         }
4320         return 0;
4321 }
4322
4323 static void free_corrupt_block(struct cache_extent *cache)
4324 {
4325         struct btrfs_corrupt_block *corrupt;
4326
4327         corrupt = container_of(cache, struct btrfs_corrupt_block, cache);
4328         free(corrupt);
4329 }
4330
4331 FREE_EXTENT_CACHE_BASED_TREE(corrupt_blocks, free_corrupt_block);
4332
4333 /*
4334  * Repair the btree of the given root.
4335  *
4336  * The fix is to remove the node key in corrupt_blocks cache_tree.
4337  * and rebalance the tree.
4338  * After the fix, the btree should be writeable.
4339  */
4340 static int repair_btree(struct btrfs_root *root,
4341                         struct cache_tree *corrupt_blocks)
4342 {
4343         struct btrfs_trans_handle *trans;
4344         struct btrfs_path path;
4345         struct btrfs_corrupt_block *corrupt;
4346         struct cache_extent *cache;
4347         struct btrfs_key key;
4348         u64 offset;
4349         int level;
4350         int ret = 0;
4351
4352         if (cache_tree_empty(corrupt_blocks))
4353                 return 0;
4354
4355         trans = btrfs_start_transaction(root, 1);
4356         if (IS_ERR(trans)) {
4357                 ret = PTR_ERR(trans);
4358                 fprintf(stderr, "Error starting transaction: %s\n",
4359                         strerror(-ret));
4360                 return ret;
4361         }
4362         btrfs_init_path(&path);
4363         cache = first_cache_extent(corrupt_blocks);
4364         while (cache) {
4365                 corrupt = container_of(cache, struct btrfs_corrupt_block,
4366                                        cache);
4367                 level = corrupt->level;
4368                 path.lowest_level = level;
4369                 key.objectid = corrupt->key.objectid;
4370                 key.type = corrupt->key.type;
4371                 key.offset = corrupt->key.offset;
4372
4373                 /*
4374                  * Here we don't want to do any tree balance, since it may
4375                  * cause a balance with corrupted brother leaf/node,
4376                  * so ins_len set to 0 here.
4377                  * Balance will be done after all corrupt node/leaf is deleted.
4378                  */
4379                 ret = btrfs_search_slot(trans, root, &key, &path, 0, 1);
4380                 if (ret < 0)
4381                         goto out;
4382                 offset = btrfs_node_blockptr(path.nodes[level],
4383                                              path.slots[level]);
4384
4385                 /* Remove the ptr */
4386                 ret = btrfs_del_ptr(root, &path, level, path.slots[level]);
4387                 if (ret < 0)
4388                         goto out;
4389                 /*
4390                  * Remove the corresponding extent
4391                  * return value is not concerned.
4392                  */
4393                 btrfs_release_path(&path);
4394                 ret = btrfs_free_extent(trans, root, offset,
4395                                 root->fs_info->nodesize, 0,
4396                                 root->root_key.objectid, level - 1, 0);
4397                 cache = next_cache_extent(cache);
4398         }
4399
4400         /* Balance the btree using btrfs_search_slot() */
4401         cache = first_cache_extent(corrupt_blocks);
4402         while (cache) {
4403                 corrupt = container_of(cache, struct btrfs_corrupt_block,
4404                                        cache);
4405                 memcpy(&key, &corrupt->key, sizeof(key));
4406                 ret = btrfs_search_slot(trans, root, &key, &path, -1, 1);
4407                 if (ret < 0)
4408                         goto out;
4409                 /* return will always >0 since it won't find the item */
4410                 ret = 0;
4411                 btrfs_release_path(&path);
4412                 cache = next_cache_extent(cache);
4413         }
4414 out:
4415         btrfs_commit_transaction(trans, root);
4416         btrfs_release_path(&path);
4417         return ret;
4418 }
4419
4420 static int check_fs_root(struct btrfs_root *root,
4421                          struct cache_tree *root_cache,
4422                          struct walk_control *wc)
4423 {
4424         int ret = 0;
4425         int err = 0;
4426         int wret;
4427         int level;
4428         struct btrfs_path path;
4429         struct shared_node root_node;
4430         struct root_record *rec;
4431         struct btrfs_root_item *root_item = &root->root_item;
4432         struct cache_tree corrupt_blocks;
4433         struct orphan_data_extent *orphan;
4434         struct orphan_data_extent *tmp;
4435         enum btrfs_tree_block_status status;
4436         struct node_refs nrefs;
4437
4438         /*
4439          * Reuse the corrupt_block cache tree to record corrupted tree block
4440          *
4441          * Unlike the usage in extent tree check, here we do it in a per
4442          * fs/subvol tree base.
4443          */
4444         cache_tree_init(&corrupt_blocks);
4445         root->fs_info->corrupt_blocks = &corrupt_blocks;
4446
4447         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
4448                 rec = get_root_rec(root_cache, root->root_key.objectid);
4449                 BUG_ON(IS_ERR(rec));
4450                 if (btrfs_root_refs(root_item) > 0)
4451                         rec->found_root_item = 1;
4452         }
4453
4454         btrfs_init_path(&path);
4455         memset(&root_node, 0, sizeof(root_node));
4456         cache_tree_init(&root_node.root_cache);
4457         cache_tree_init(&root_node.inode_cache);
4458         memset(&nrefs, 0, sizeof(nrefs));
4459
4460         /* Move the orphan extent record to corresponding inode_record */
4461         list_for_each_entry_safe(orphan, tmp,
4462                                  &root->orphan_data_extents, list) {
4463                 struct inode_record *inode;
4464
4465                 inode = get_inode_rec(&root_node.inode_cache, orphan->objectid,
4466                                       1);
4467                 BUG_ON(IS_ERR(inode));
4468                 inode->errors |= I_ERR_FILE_EXTENT_ORPHAN;
4469                 list_move(&orphan->list, &inode->orphan_extents);
4470         }
4471
4472         level = btrfs_header_level(root->node);
4473         memset(wc->nodes, 0, sizeof(wc->nodes));
4474         wc->nodes[level] = &root_node;
4475         wc->active_node = level;
4476         wc->root_level = level;
4477
4478         /* We may not have checked the root block, lets do that now */
4479         if (btrfs_is_leaf(root->node))
4480                 status = btrfs_check_leaf(root, NULL, root->node);
4481         else
4482                 status = btrfs_check_node(root, NULL, root->node);
4483         if (status != BTRFS_TREE_BLOCK_CLEAN)
4484                 return -EIO;
4485
4486         if (btrfs_root_refs(root_item) > 0 ||
4487             btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
4488                 path.nodes[level] = root->node;
4489                 extent_buffer_get(root->node);
4490                 path.slots[level] = 0;
4491         } else {
4492                 struct btrfs_key key;
4493                 struct btrfs_disk_key found_key;
4494
4495                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
4496                 level = root_item->drop_level;
4497                 path.lowest_level = level;
4498                 if (level > btrfs_header_level(root->node) ||
4499                     level >= BTRFS_MAX_LEVEL) {
4500                         error("ignoring invalid drop level: %u", level);
4501                         goto skip_walking;
4502                 }
4503                 wret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
4504                 if (wret < 0)
4505                         goto skip_walking;
4506                 btrfs_node_key(path.nodes[level], &found_key,
4507                                 path.slots[level]);
4508                 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
4509                                         sizeof(found_key)));
4510         }
4511
4512         while (1) {
4513                 wret = walk_down_tree(root, &path, wc, &level, &nrefs);
4514                 if (wret < 0)
4515                         ret = wret;
4516                 if (wret != 0)
4517                         break;
4518
4519                 wret = walk_up_tree(root, &path, wc, &level);
4520                 if (wret < 0)
4521                         ret = wret;
4522                 if (wret != 0)
4523                         break;
4524         }
4525 skip_walking:
4526         btrfs_release_path(&path);
4527
4528         if (!cache_tree_empty(&corrupt_blocks)) {
4529                 struct cache_extent *cache;
4530                 struct btrfs_corrupt_block *corrupt;
4531
4532                 printf("The following tree block(s) is corrupted in tree %llu:\n",
4533                        root->root_key.objectid);
4534                 cache = first_cache_extent(&corrupt_blocks);
4535                 while (cache) {
4536                         corrupt = container_of(cache,
4537                                                struct btrfs_corrupt_block,
4538                                                cache);
4539                         printf("\ttree block bytenr: %llu, level: %d, node key: (%llu, %u, %llu)\n",
4540                                cache->start, corrupt->level,
4541                                corrupt->key.objectid, corrupt->key.type,
4542                                corrupt->key.offset);
4543                         cache = next_cache_extent(cache);
4544                 }
4545                 if (repair) {
4546                         printf("Try to repair the btree for root %llu\n",
4547                                root->root_key.objectid);
4548                         ret = repair_btree(root, &corrupt_blocks);
4549                         if (ret < 0)
4550                                 fprintf(stderr, "Failed to repair btree: %s\n",
4551                                         strerror(-ret));
4552                         if (!ret)
4553                                 printf("Btree for root %llu is fixed\n",
4554                                        root->root_key.objectid);
4555                 }
4556         }
4557
4558         err = merge_root_recs(root, &root_node.root_cache, root_cache);
4559         if (err < 0)
4560                 ret = err;
4561
4562         if (root_node.current) {
4563                 root_node.current->checked = 1;
4564                 maybe_free_inode_rec(&root_node.inode_cache,
4565                                 root_node.current);
4566         }
4567
4568         err = check_inode_recs(root, &root_node.inode_cache);
4569         if (!ret)
4570                 ret = err;
4571
4572         free_corrupt_blocks_tree(&corrupt_blocks);
4573         root->fs_info->corrupt_blocks = NULL;
4574         free_orphan_data_extents(&root->orphan_data_extents);
4575         return ret;
4576 }
4577
4578 static int fs_root_objectid(u64 objectid)
4579 {
4580         if (objectid == BTRFS_TREE_RELOC_OBJECTID ||
4581             objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
4582                 return 1;
4583         return is_fstree(objectid);
4584 }
4585
4586 static int check_fs_roots(struct btrfs_fs_info *fs_info,
4587                           struct cache_tree *root_cache)
4588 {
4589         struct btrfs_path path;
4590         struct btrfs_key key;
4591         struct walk_control wc;
4592         struct extent_buffer *leaf, *tree_node;
4593         struct btrfs_root *tmp_root;
4594         struct btrfs_root *tree_root = fs_info->tree_root;
4595         int ret;
4596         int err = 0;
4597
4598         if (ctx.progress_enabled) {
4599                 ctx.tp = TASK_FS_ROOTS;
4600                 task_start(ctx.info);
4601         }
4602
4603         /*
4604          * Just in case we made any changes to the extent tree that weren't
4605          * reflected into the free space cache yet.
4606          */
4607         if (repair)
4608                 reset_cached_block_groups(fs_info);
4609         memset(&wc, 0, sizeof(wc));
4610         cache_tree_init(&wc.shared);
4611         btrfs_init_path(&path);
4612
4613 again:
4614         key.offset = 0;
4615         key.objectid = 0;
4616         key.type = BTRFS_ROOT_ITEM_KEY;
4617         ret = btrfs_search_slot(NULL, tree_root, &key, &path, 0, 0);
4618         if (ret < 0) {
4619                 err = 1;
4620                 goto out;
4621         }
4622         tree_node = tree_root->node;
4623         while (1) {
4624                 if (tree_node != tree_root->node) {
4625                         free_root_recs_tree(root_cache);
4626                         btrfs_release_path(&path);
4627                         goto again;
4628                 }
4629                 leaf = path.nodes[0];
4630                 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
4631                         ret = btrfs_next_leaf(tree_root, &path);
4632                         if (ret) {
4633                                 if (ret < 0)
4634                                         err = 1;
4635                                 break;
4636                         }
4637                         leaf = path.nodes[0];
4638                 }
4639                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
4640                 if (key.type == BTRFS_ROOT_ITEM_KEY &&
4641                     fs_root_objectid(key.objectid)) {
4642                         if (key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
4643                                 tmp_root = btrfs_read_fs_root_no_cache(
4644                                                 fs_info, &key);
4645                         } else {
4646                                 key.offset = (u64)-1;
4647                                 tmp_root = btrfs_read_fs_root(
4648                                                 fs_info, &key);
4649                         }
4650                         if (IS_ERR(tmp_root)) {
4651                                 err = 1;
4652                                 goto next;
4653                         }
4654                         ret = check_fs_root(tmp_root, root_cache, &wc);
4655                         if (ret == -EAGAIN) {
4656                                 free_root_recs_tree(root_cache);
4657                                 btrfs_release_path(&path);
4658                                 goto again;
4659                         }
4660                         if (ret)
4661                                 err = 1;
4662                         if (key.objectid == BTRFS_TREE_RELOC_OBJECTID)
4663                                 btrfs_free_fs_root(tmp_root);
4664                 } else if (key.type == BTRFS_ROOT_REF_KEY ||
4665                            key.type == BTRFS_ROOT_BACKREF_KEY) {
4666                         process_root_ref(leaf, path.slots[0], &key,
4667                                          root_cache);
4668                 }
4669 next:
4670                 path.slots[0]++;
4671         }
4672 out:
4673         btrfs_release_path(&path);
4674         if (err)
4675                 free_extent_cache_tree(&wc.shared);
4676         if (!cache_tree_empty(&wc.shared))
4677                 fprintf(stderr, "warning line %d\n", __LINE__);
4678
4679         task_stop(ctx.info);
4680
4681         return err;
4682 }
4683
4684 /*
4685  * Find the @index according by @ino and name.
4686  * Notice:time efficiency is O(N)
4687  *
4688  * @root:       the root of the fs/file tree
4689  * @index_ret:  the index as return value
4690  * @namebuf:    the name to match
4691  * @name_len:   the length of name to match
4692  * @file_type:  the file_type of INODE_ITEM to match
4693  *
4694  * Returns 0 if found and *@index_ret will be modified with right value
4695  * Returns< 0 not found and *@index_ret will be (u64)-1
4696  */
4697 static int find_dir_index(struct btrfs_root *root, u64 dirid, u64 location_id,
4698                           u64 *index_ret, char *namebuf, u32 name_len,
4699                           u8 file_type)
4700 {
4701         struct btrfs_path path;
4702         struct extent_buffer *node;
4703         struct btrfs_dir_item *di;
4704         struct btrfs_key key;
4705         struct btrfs_key location;
4706         char name[BTRFS_NAME_LEN] = {0};
4707
4708         u32 total;
4709         u32 cur = 0;
4710         u32 len;
4711         u32 data_len;
4712         u8 filetype;
4713         int slot;
4714         int ret;
4715
4716         ASSERT(index_ret);
4717
4718         /* search from the last index */
4719         key.objectid = dirid;
4720         key.offset = (u64)-1;
4721         key.type = BTRFS_DIR_INDEX_KEY;
4722
4723         btrfs_init_path(&path);
4724         ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
4725         if (ret < 0)
4726                 return ret;
4727
4728 loop:
4729         ret = btrfs_previous_item(root, &path, dirid, BTRFS_DIR_INDEX_KEY);
4730         if (ret) {
4731                 ret = -ENOENT;
4732                 *index_ret = (64)-1;
4733                 goto out;
4734         }
4735         /* Check whether inode_id/filetype/name match */
4736         node = path.nodes[0];
4737         slot = path.slots[0];
4738         di = btrfs_item_ptr(node, slot, struct btrfs_dir_item);
4739         total = btrfs_item_size_nr(node, slot);
4740         while (cur < total) {
4741                 ret = -ENOENT;
4742                 len = btrfs_dir_name_len(node, di);
4743                 data_len = btrfs_dir_data_len(node, di);
4744
4745                 btrfs_dir_item_key_to_cpu(node, di, &location);
4746                 if (location.objectid != location_id ||
4747                     location.type != BTRFS_INODE_ITEM_KEY ||
4748                     location.offset != 0)
4749                         goto next;
4750
4751                 filetype = btrfs_dir_type(node, di);
4752                 if (file_type != filetype)
4753                         goto next;
4754
4755                 if (len > BTRFS_NAME_LEN)
4756                         len = BTRFS_NAME_LEN;
4757
4758                 read_extent_buffer(node, name, (unsigned long)(di + 1), len);
4759                 if (len != name_len || strncmp(namebuf, name, len))
4760                         goto next;
4761
4762                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
4763                 *index_ret = key.offset;
4764                 ret = 0;
4765                 goto out;
4766 next:
4767                 len += sizeof(*di) + data_len;
4768                 di = (struct btrfs_dir_item *)((char *)di + len);
4769                 cur += len;
4770         }
4771         goto loop;
4772
4773 out:
4774         btrfs_release_path(&path);
4775         return ret;
4776 }
4777
4778 /*
4779  * Find DIR_ITEM/DIR_INDEX for the given key and check it with the specified
4780  * INODE_REF/INODE_EXTREF match.
4781  *
4782  * @root:       the root of the fs/file tree
4783  * @key:        the key of the DIR_ITEM/DIR_INDEX, key->offset will be right
4784  *              value while find index
4785  * @location_key: location key of the struct btrfs_dir_item to match
4786  * @name:       the name to match
4787  * @namelen:    the length of name
4788  * @file_type:  the type of file to math
4789  *
4790  * Return 0 if no error occurred.
4791  * Return DIR_ITEM_MISSING/DIR_INDEX_MISSING if couldn't find
4792  * DIR_ITEM/DIR_INDEX
4793  * Return DIR_ITEM_MISMATCH/DIR_INDEX_MISMATCH if INODE_REF/INODE_EXTREF
4794  * and DIR_ITEM/DIR_INDEX mismatch
4795  */
4796 static int find_dir_item(struct btrfs_root *root, struct btrfs_key *key,
4797                          struct btrfs_key *location_key, char *name,
4798                          u32 namelen, u8 file_type)
4799 {
4800         struct btrfs_path path;
4801         struct extent_buffer *node;
4802         struct btrfs_dir_item *di;
4803         struct btrfs_key location;
4804         char namebuf[BTRFS_NAME_LEN] = {0};
4805         u32 total;
4806         u32 cur = 0;
4807         u32 len;
4808         u32 data_len;
4809         u8 filetype;
4810         int slot;
4811         int ret;
4812
4813         /* get the index by traversing all index */
4814         if (key->type == BTRFS_DIR_INDEX_KEY && key->offset == (u64)-1) {
4815                 ret = find_dir_index(root, key->objectid,
4816                                      location_key->objectid, &key->offset,
4817                                      name, namelen, file_type);
4818                 if (ret)
4819                         ret = DIR_INDEX_MISSING;
4820                 return ret;
4821         }
4822
4823         btrfs_init_path(&path);
4824         ret = btrfs_search_slot(NULL, root, key, &path, 0, 0);
4825         if (ret) {
4826                 ret = key->type == BTRFS_DIR_ITEM_KEY ? DIR_ITEM_MISSING :
4827                         DIR_INDEX_MISSING;
4828                 goto out;
4829         }
4830
4831         /* Check whether inode_id/filetype/name match */
4832         node = path.nodes[0];
4833         slot = path.slots[0];
4834         di = btrfs_item_ptr(node, slot, struct btrfs_dir_item);
4835         total = btrfs_item_size_nr(node, slot);
4836         while (cur < total) {
4837                 ret = key->type == BTRFS_DIR_ITEM_KEY ?
4838                         DIR_ITEM_MISMATCH : DIR_INDEX_MISMATCH;
4839
4840                 len = btrfs_dir_name_len(node, di);
4841                 data_len = btrfs_dir_data_len(node, di);
4842
4843                 btrfs_dir_item_key_to_cpu(node, di, &location);
4844                 if (location.objectid != location_key->objectid ||
4845                     location.type != location_key->type ||
4846                     location.offset != location_key->offset)
4847                         goto next;
4848
4849                 filetype = btrfs_dir_type(node, di);
4850                 if (file_type != filetype)
4851                         goto next;
4852
4853                 if (len > BTRFS_NAME_LEN) {
4854                         len = BTRFS_NAME_LEN;
4855                         warning("root %llu %s[%llu %llu] name too long %u, trimmed",
4856                         root->objectid,
4857                         key->type == BTRFS_DIR_ITEM_KEY ?
4858                         "DIR_ITEM" : "DIR_INDEX",
4859                         key->objectid, key->offset, len);
4860                 }
4861                 read_extent_buffer(node, namebuf, (unsigned long)(di + 1),
4862                                    len);
4863                 if (len != namelen || strncmp(namebuf, name, len))
4864                         goto next;
4865
4866                 ret = 0;
4867                 goto out;
4868 next:
4869                 len += sizeof(*di) + data_len;
4870                 di = (struct btrfs_dir_item *)((char *)di + len);
4871                 cur += len;
4872         }
4873
4874 out:
4875         btrfs_release_path(&path);
4876         return ret;
4877 }
4878
4879 /*
4880  * Prints inode ref error message
4881  */
4882 static void print_inode_ref_err(struct btrfs_root *root, struct btrfs_key *key,
4883                                 u64 index, const char *namebuf, int name_len,
4884                                 u8 filetype, int err)
4885 {
4886         if (!err)
4887                 return;
4888
4889         /* root dir error */
4890         if (key->objectid == BTRFS_FIRST_FREE_OBJECTID) {
4891                 error(
4892         "root %llu root dir shouldn't have INODE REF[%llu %llu] name %s",
4893                       root->objectid, key->objectid, key->offset, namebuf);
4894                 return;
4895         }
4896
4897         /* normal error */
4898         if (err & (DIR_ITEM_MISMATCH | DIR_ITEM_MISSING))
4899                 error("root %llu DIR ITEM[%llu %llu] %s name %s filetype %u",
4900                       root->objectid, key->offset,
4901                       btrfs_name_hash(namebuf, name_len),
4902                       err & DIR_ITEM_MISMATCH ? "mismatch" : "missing",
4903                       namebuf, filetype);
4904         if (err & (DIR_INDEX_MISMATCH | DIR_INDEX_MISSING))
4905                 error("root %llu DIR INDEX[%llu %llu] %s name %s filetype %u",
4906                       root->objectid, key->offset, index,
4907                       err & DIR_ITEM_MISMATCH ? "mismatch" : "missing",
4908                       namebuf, filetype);
4909 }
4910
4911 /*
4912  * Insert the missing inode item.
4913  *
4914  * Returns 0 means success.
4915  * Returns <0 means error.
4916  */
4917 static int repair_inode_item_missing(struct btrfs_root *root, u64 ino,
4918                                      u8 filetype)
4919 {
4920         struct btrfs_key key;
4921         struct btrfs_trans_handle *trans;
4922         struct btrfs_path path;
4923         int ret;
4924
4925         key.objectid = ino;
4926         key.type = BTRFS_INODE_ITEM_KEY;
4927         key.offset = 0;
4928
4929         btrfs_init_path(&path);
4930         trans = btrfs_start_transaction(root, 1);
4931         if (IS_ERR(trans)) {
4932                 ret = -EIO;
4933                 goto out;
4934         }
4935
4936         ret = btrfs_search_slot(trans, root, &key, &path, 1, 1);
4937         if (ret < 0 || !ret)
4938                 goto fail;
4939
4940         /* insert inode item */
4941         create_inode_item_lowmem(trans, root, ino, filetype);
4942         ret = 0;
4943 fail:
4944         btrfs_commit_transaction(trans, root);
4945 out:
4946         if (ret)
4947                 error("failed to repair root %llu INODE ITEM[%llu] missing",
4948                       root->objectid, ino);
4949         btrfs_release_path(&path);
4950         return ret;
4951 }
4952
4953 /*
4954  * The ternary means dir item, dir index and relative inode ref.
4955  * The function handles errs: INODE_MISSING, DIR_INDEX_MISSING
4956  * DIR_INDEX_MISMATCH, DIR_ITEM_MISSING, DIR_ITEM_MISMATCH by the follow
4957  * strategy:
4958  * If two of three is missing or mismatched, delete the existing one.
4959  * If one of three is missing or mismatched, add the missing one.
4960  *
4961  * returns 0 means success.
4962  * returns not 0 means on error;
4963  */
4964 int repair_ternary_lowmem(struct btrfs_root *root, u64 dir_ino, u64 ino,
4965                           u64 index, char *name, int name_len, u8 filetype,
4966                           int err)
4967 {
4968         struct btrfs_trans_handle *trans;
4969         int stage = 0;
4970         int ret = 0;
4971
4972         /*
4973          * stage shall be one of following valild values:
4974          *      0: Fine, nothing to do.
4975          *      1: One of three is wrong, so add missing one.
4976          *      2: Two of three is wrong, so delete existed one.
4977          */
4978         if (err & (DIR_INDEX_MISMATCH | DIR_INDEX_MISSING))
4979                 stage++;
4980         if (err & (DIR_ITEM_MISMATCH | DIR_ITEM_MISSING))
4981                 stage++;
4982         if (err & (INODE_REF_MISSING))
4983                 stage++;
4984
4985         /* stage must be smllarer than 3 */
4986         ASSERT(stage < 3);
4987
4988         trans = btrfs_start_transaction(root, 1);
4989         if (stage == 2) {
4990                 ret = btrfs_unlink(trans, root, ino, dir_ino, index, name,
4991                                    name_len, 0);
4992                 goto out;
4993         }
4994         if (stage == 1) {
4995                 ret = btrfs_add_link(trans, root, ino, dir_ino, name, name_len,
4996                                filetype, &index, 1, 1);
4997                 goto out;
4998         }
4999 out:
5000         btrfs_commit_transaction(trans, root);
5001
5002         if (ret)
5003                 error("fail to repair inode %llu name %s filetype %u",
5004                       ino, name, filetype);
5005         else
5006                 printf("%s ref/dir_item of inode %llu name %s filetype %u\n",
5007                        stage == 2 ? "Delete" : "Add",
5008                        ino, name, filetype);
5009
5010         return ret;
5011 }
5012
5013 /*
5014  * Traverse the given INODE_REF and call find_dir_item() to find related
5015  * DIR_ITEM/DIR_INDEX.
5016  *
5017  * @root:       the root of the fs/file tree
5018  * @ref_key:    the key of the INODE_REF
5019  * @path        the path provides node and slot
5020  * @refs:       the count of INODE_REF
5021  * @mode:       the st_mode of INODE_ITEM
5022  * @name_ret:   returns with the first ref's name
5023  * @name_len_ret:    len of the name_ret
5024  *
5025  * Return 0 if no error occurred.
5026  */
5027 static int check_inode_ref(struct btrfs_root *root, struct btrfs_key *ref_key,
5028                            struct btrfs_path *path, char *name_ret,
5029                            u32 *namelen_ret, u64 *refs_ret, int mode)
5030 {
5031         struct btrfs_key key;
5032         struct btrfs_key location;
5033         struct btrfs_inode_ref *ref;
5034         struct extent_buffer *node;
5035         char namebuf[BTRFS_NAME_LEN] = {0};
5036         u32 total;
5037         u32 cur = 0;
5038         u32 len;
5039         u32 name_len;
5040         u64 index;
5041         int ret;
5042         int err = 0;
5043         int tmp_err;
5044         int slot;
5045         int need_research = 0;
5046         u64 refs;
5047
5048 begin:
5049         err = 0;
5050         cur = 0;
5051         refs = *refs_ret;
5052
5053         /* since after repair, path and the dir item may be changed */
5054         if (need_research) {
5055                 need_research = 0;
5056                 btrfs_release_path(path);
5057                 ret = btrfs_search_slot(NULL, root, ref_key, path, 0, 0);
5058                 /* the item was deleted, let path point to the last checked item */
5059                 if (ret > 0) {
5060                         if (path->slots[0] == 0)
5061                                 btrfs_prev_leaf(root, path);
5062                         else
5063                                 path->slots[0]--;
5064                 }
5065                 if (ret)
5066                         goto out;
5067         }
5068
5069         location.objectid = ref_key->objectid;
5070         location.type = BTRFS_INODE_ITEM_KEY;
5071         location.offset = 0;
5072         node = path->nodes[0];
5073         slot = path->slots[0];
5074
5075         memset(namebuf, 0, sizeof(namebuf) / sizeof(*namebuf));
5076         ref = btrfs_item_ptr(node, slot, struct btrfs_inode_ref);
5077         total = btrfs_item_size_nr(node, slot);
5078
5079 next:
5080         /* Update inode ref count */
5081         refs++;
5082         tmp_err = 0;
5083         index = btrfs_inode_ref_index(node, ref);
5084         name_len = btrfs_inode_ref_name_len(node, ref);
5085
5086         if (name_len <= BTRFS_NAME_LEN) {
5087                 len = name_len;
5088         } else {
5089                 len = BTRFS_NAME_LEN;
5090                 warning("root %llu INODE_REF[%llu %llu] name too long",
5091                         root->objectid, ref_key->objectid, ref_key->offset);
5092         }
5093
5094         read_extent_buffer(node, namebuf, (unsigned long)(ref + 1), len);
5095
5096         /* copy the first name found to name_ret */
5097         if (refs == 1 && name_ret) {
5098                 memcpy(name_ret, namebuf, len);
5099                 *namelen_ret = len;
5100         }
5101
5102         /* Check root dir ref */
5103         if (ref_key->objectid == BTRFS_FIRST_FREE_OBJECTID) {
5104                 if (index != 0 || len != strlen("..") ||
5105                     strncmp("..", namebuf, len) ||
5106                     ref_key->offset != BTRFS_FIRST_FREE_OBJECTID) {
5107                         /* set err bits then repair will delete the ref */
5108                         err |= DIR_INDEX_MISSING;
5109                         err |= DIR_ITEM_MISSING;
5110                 }
5111                 goto end;
5112         }
5113
5114         /* Find related DIR_INDEX */
5115         key.objectid = ref_key->offset;
5116         key.type = BTRFS_DIR_INDEX_KEY;
5117         key.offset = index;
5118         tmp_err |= find_dir_item(root, &key, &location, namebuf, len,
5119                             imode_to_type(mode));
5120
5121         /* Find related dir_item */
5122         key.objectid = ref_key->offset;
5123         key.type = BTRFS_DIR_ITEM_KEY;
5124         key.offset = btrfs_name_hash(namebuf, len);
5125         tmp_err |= find_dir_item(root, &key, &location, namebuf, len,
5126                             imode_to_type(mode));
5127 end:
5128         if (tmp_err && repair) {
5129                 ret = repair_ternary_lowmem(root, ref_key->offset,
5130                                             ref_key->objectid, index, namebuf,
5131                                             name_len, imode_to_type(mode),
5132                                             tmp_err);
5133                 if (!ret) {
5134                         need_research = 1;
5135                         goto begin;
5136                 }
5137         }
5138         print_inode_ref_err(root, ref_key, index, namebuf, name_len,
5139                             imode_to_type(mode), tmp_err);
5140         err |= tmp_err;
5141         len = sizeof(*ref) + name_len;
5142         ref = (struct btrfs_inode_ref *)((char *)ref + len);
5143         cur += len;
5144         if (cur < total)
5145                 goto next;
5146
5147 out:
5148         *refs_ret = refs;
5149         return err;
5150 }
5151
5152 /*
5153  * Traverse the given INODE_EXTREF and call find_dir_item() to find related
5154  * DIR_ITEM/DIR_INDEX.
5155  *
5156  * @root:       the root of the fs/file tree
5157  * @ref_key:    the key of the INODE_EXTREF
5158  * @refs:       the count of INODE_EXTREF
5159  * @mode:       the st_mode of INODE_ITEM
5160  *
5161  * Return 0 if no error occurred.
5162  */
5163 static int check_inode_extref(struct btrfs_root *root,
5164                               struct btrfs_key *ref_key,
5165                               struct extent_buffer *node, int slot, u64 *refs,
5166                               int mode)
5167 {
5168         struct btrfs_key key;
5169         struct btrfs_key location;
5170         struct btrfs_inode_extref *extref;
5171         char namebuf[BTRFS_NAME_LEN] = {0};
5172         u32 total;
5173         u32 cur = 0;
5174         u32 len;
5175         u32 name_len;
5176         u64 index;
5177         u64 parent;
5178         int ret;
5179         int err = 0;
5180
5181         location.objectid = ref_key->objectid;
5182         location.type = BTRFS_INODE_ITEM_KEY;
5183         location.offset = 0;
5184
5185         extref = btrfs_item_ptr(node, slot, struct btrfs_inode_extref);
5186         total = btrfs_item_size_nr(node, slot);
5187
5188 next:
5189         /* update inode ref count */
5190         (*refs)++;
5191         name_len = btrfs_inode_extref_name_len(node, extref);
5192         index = btrfs_inode_extref_index(node, extref);
5193         parent = btrfs_inode_extref_parent(node, extref);
5194         if (name_len <= BTRFS_NAME_LEN) {
5195                 len = name_len;
5196         } else {
5197                 len = BTRFS_NAME_LEN;
5198                 warning("root %llu INODE_EXTREF[%llu %llu] name too long",
5199                         root->objectid, ref_key->objectid, ref_key->offset);
5200         }
5201         read_extent_buffer(node, namebuf, (unsigned long)(extref + 1), len);
5202
5203         /* Check root dir ref name */
5204         if (index == 0 && strncmp(namebuf, "..", name_len)) {
5205                 error("root %llu INODE_EXTREF[%llu %llu] ROOT_DIR name shouldn't be %s",
5206                       root->objectid, ref_key->objectid, ref_key->offset,
5207                       namebuf);
5208                 err |= ROOT_DIR_ERROR;
5209         }
5210
5211         /* find related dir_index */
5212         key.objectid = parent;
5213         key.type = BTRFS_DIR_INDEX_KEY;
5214         key.offset = index;
5215         ret = find_dir_item(root, &key, &location, namebuf, len, mode);
5216         err |= ret;
5217
5218         /* find related dir_item */
5219         key.objectid = parent;
5220         key.type = BTRFS_DIR_ITEM_KEY;
5221         key.offset = btrfs_name_hash(namebuf, len);
5222         ret = find_dir_item(root, &key, &location, namebuf, len, mode);
5223         err |= ret;
5224
5225         len = sizeof(*extref) + name_len;
5226         extref = (struct btrfs_inode_extref *)((char *)extref + len);
5227         cur += len;
5228
5229         if (cur < total)
5230                 goto next;
5231
5232         return err;
5233 }
5234
5235 /*
5236  * Find INODE_REF/INODE_EXTREF for the given key and check it with the specified
5237  * DIR_ITEM/DIR_INDEX match.
5238  * Return with @index_ret.
5239  *
5240  * @root:       the root of the fs/file tree
5241  * @key:        the key of the INODE_REF/INODE_EXTREF
5242  * @name:       the name in the INODE_REF/INODE_EXTREF
5243  * @namelen:    the length of name in the INODE_REF/INODE_EXTREF
5244  * @index_ret:  the index in the INODE_REF/INODE_EXTREF,
5245  *              value (64)-1 means do not check index
5246  * @ext_ref:    the EXTENDED_IREF feature
5247  *
5248  * Return 0 if no error occurred.
5249  * Return >0 for error bitmap
5250  */
5251 static int find_inode_ref(struct btrfs_root *root, struct btrfs_key *key,
5252                           char *name, int namelen, u64 *index_ret,
5253                           unsigned int ext_ref)
5254 {
5255         struct btrfs_path path;
5256         struct btrfs_inode_ref *ref;
5257         struct btrfs_inode_extref *extref;
5258         struct extent_buffer *node;
5259         char ref_namebuf[BTRFS_NAME_LEN] = {0};
5260         u32 total;
5261         u32 cur = 0;
5262         u32 len;
5263         u32 ref_namelen;
5264         u64 ref_index;
5265         u64 parent;
5266         u64 dir_id;
5267         int slot;
5268         int ret;
5269
5270         ASSERT(index_ret);
5271
5272         btrfs_init_path(&path);
5273         ret = btrfs_search_slot(NULL, root, key, &path, 0, 0);
5274         if (ret) {
5275                 ret = INODE_REF_MISSING;
5276                 goto extref;
5277         }
5278
5279         node = path.nodes[0];
5280         slot = path.slots[0];
5281
5282         ref = btrfs_item_ptr(node, slot, struct btrfs_inode_ref);
5283         total = btrfs_item_size_nr(node, slot);
5284
5285         /* Iterate all entry of INODE_REF */
5286         while (cur < total) {
5287                 ret = INODE_REF_MISSING;
5288
5289                 ref_namelen = btrfs_inode_ref_name_len(node, ref);
5290                 ref_index = btrfs_inode_ref_index(node, ref);
5291                 if (*index_ret != (u64)-1 && *index_ret != ref_index)
5292                         goto next_ref;
5293
5294                 if (cur + sizeof(*ref) + ref_namelen > total ||
5295                     ref_namelen > BTRFS_NAME_LEN) {
5296                         warning("root %llu INODE %s[%llu %llu] name too long",
5297                                 root->objectid,
5298                                 key->type == BTRFS_INODE_REF_KEY ?
5299                                         "REF" : "EXTREF",
5300                                 key->objectid, key->offset);
5301
5302                         if (cur + sizeof(*ref) > total)
5303                                 break;
5304                         len = min_t(u32, total - cur - sizeof(*ref),
5305                                     BTRFS_NAME_LEN);
5306                 } else {
5307                         len = ref_namelen;
5308                 }
5309
5310                 read_extent_buffer(node, ref_namebuf, (unsigned long)(ref + 1),
5311                                    len);
5312
5313                 if (len != namelen || strncmp(ref_namebuf, name, len))
5314                         goto next_ref;
5315
5316                 *index_ret = ref_index;
5317                 ret = 0;
5318                 goto out;
5319 next_ref:
5320                 len = sizeof(*ref) + ref_namelen;
5321                 ref = (struct btrfs_inode_ref *)((char *)ref + len);
5322                 cur += len;
5323         }
5324
5325 extref:
5326         /* Skip if not support EXTENDED_IREF feature */
5327         if (!ext_ref)
5328                 goto out;
5329
5330         btrfs_release_path(&path);
5331         btrfs_init_path(&path);
5332
5333         dir_id = key->offset;
5334         key->type = BTRFS_INODE_EXTREF_KEY;
5335         key->offset = btrfs_extref_hash(dir_id, name, namelen);
5336
5337         ret = btrfs_search_slot(NULL, root, key, &path, 0, 0);
5338         if (ret) {
5339                 ret = INODE_REF_MISSING;
5340                 goto out;
5341         }
5342
5343         node = path.nodes[0];
5344         slot = path.slots[0];
5345
5346         extref = btrfs_item_ptr(node, slot, struct btrfs_inode_extref);
5347         cur = 0;
5348         total = btrfs_item_size_nr(node, slot);
5349
5350         /* Iterate all entry of INODE_EXTREF */
5351         while (cur < total) {
5352                 ret = INODE_REF_MISSING;
5353
5354                 ref_namelen = btrfs_inode_extref_name_len(node, extref);
5355                 ref_index = btrfs_inode_extref_index(node, extref);
5356                 parent = btrfs_inode_extref_parent(node, extref);
5357                 if (*index_ret != (u64)-1 && *index_ret != ref_index)
5358                         goto next_extref;
5359
5360                 if (parent != dir_id)
5361                         goto next_extref;
5362
5363                 if (ref_namelen <= BTRFS_NAME_LEN) {
5364                         len = ref_namelen;
5365                 } else {
5366                         len = BTRFS_NAME_LEN;
5367                         warning("root %llu INODE %s[%llu %llu] name too long",
5368                                 root->objectid,
5369                                 key->type == BTRFS_INODE_REF_KEY ?
5370                                         "REF" : "EXTREF",
5371                                 key->objectid, key->offset);
5372                 }
5373                 read_extent_buffer(node, ref_namebuf,
5374                                    (unsigned long)(extref + 1), len);
5375
5376                 if (len != namelen || strncmp(ref_namebuf, name, len))
5377                         goto next_extref;
5378
5379                 *index_ret = ref_index;
5380                 ret = 0;
5381                 goto out;
5382
5383 next_extref:
5384                 len = sizeof(*extref) + ref_namelen;
5385                 extref = (struct btrfs_inode_extref *)((char *)extref + len);
5386                 cur += len;
5387
5388         }
5389 out:
5390         btrfs_release_path(&path);
5391         return ret;
5392 }
5393
5394 static void print_dir_item_err(struct btrfs_root *root, struct btrfs_key *key,
5395                                u64 ino, u64 index, const char *namebuf,
5396                                int name_len, u8 filetype, int err)
5397 {
5398         if (err & (DIR_ITEM_MISMATCH | DIR_ITEM_MISSING)) {
5399                 error("root %llu DIR ITEM[%llu %llu] name %s filetype %d %s",
5400                       root->objectid, key->objectid, key->offset, namebuf,
5401                       filetype,
5402                       err & DIR_ITEM_MISMATCH ? "mismath" : "missing");
5403         }
5404
5405         if (err & (DIR_INDEX_MISMATCH | DIR_INDEX_MISSING)) {
5406                 error("root %llu DIR INDEX[%llu %llu] name %s filetype %d %s",
5407                       root->objectid, key->objectid, index, namebuf, filetype,
5408                       err & DIR_ITEM_MISMATCH ? "mismath" : "missing");
5409         }
5410
5411         if (err & (INODE_ITEM_MISSING | INODE_ITEM_MISMATCH)) {
5412                 error(
5413                 "root %llu INODE_ITEM[%llu] index %llu name %s filetype %d %s",
5414                       root->objectid, ino, index, namebuf, filetype,
5415                       err & INODE_ITEM_MISMATCH ? "mismath" : "missing");
5416         }
5417
5418         if (err & INODE_REF_MISSING)
5419                 error(
5420                 "root %llu INODE REF[%llu, %llu] name %s filetype %u missing",
5421                       root->objectid, ino, key->objectid, namebuf, filetype);
5422
5423 }
5424
5425 /*
5426  * Call repair_inode_item_missing and repair_ternary_lowmem to repair
5427  *
5428  * Returns error after repair
5429  */
5430 static int repair_dir_item(struct btrfs_root *root, u64 dirid, u64 ino,
5431                            u64 index, u8 filetype, char *namebuf, u32 name_len,
5432                            int err)
5433 {
5434         int ret;
5435
5436         if (err & INODE_ITEM_MISSING) {
5437                 ret = repair_inode_item_missing(root, ino, filetype);
5438                 if (!ret)
5439                         err &= ~(INODE_ITEM_MISMATCH | INODE_ITEM_MISSING);
5440         }
5441
5442         if (err & ~(INODE_ITEM_MISMATCH | INODE_ITEM_MISSING)) {
5443                 ret = repair_ternary_lowmem(root, dirid, ino, index, namebuf,
5444                                             name_len, filetype, err);
5445                 if (!ret) {
5446                         err &= ~(DIR_INDEX_MISMATCH | DIR_INDEX_MISSING);
5447                         err &= ~(DIR_ITEM_MISMATCH | DIR_ITEM_MISSING);
5448                         err &= ~(INODE_REF_MISSING);
5449                 }
5450         }
5451         return err;
5452 }
5453
5454 static int __count_dir_isize(struct btrfs_root *root, u64 ino, int type,
5455                 u64 *size_ret)
5456 {
5457         struct btrfs_key key;
5458         struct btrfs_path path;
5459         u32 len;
5460         struct btrfs_dir_item *di;
5461         int ret;
5462         int cur = 0;
5463         int total = 0;
5464
5465         ASSERT(size_ret);
5466         *size_ret = 0;
5467
5468         key.objectid = ino;
5469         key.type = type;
5470         key.offset = (u64)-1;
5471
5472         btrfs_init_path(&path);
5473         ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
5474         if (ret < 0) {
5475                 ret = -EIO;
5476                 goto out;
5477         }
5478         /* if found, go to spacial case */
5479         if (ret == 0)
5480                 goto special_case;
5481
5482 loop:
5483         ret = btrfs_previous_item(root, &path, ino, type);
5484
5485         if (ret) {
5486                 ret = 0;
5487                 goto out;
5488         }
5489
5490 special_case:
5491         di = btrfs_item_ptr(path.nodes[0], path.slots[0], struct btrfs_dir_item);
5492         cur = 0;
5493         total = btrfs_item_size_nr(path.nodes[0], path.slots[0]);
5494
5495         while (cur < total) {
5496                 len = btrfs_dir_name_len(path.nodes[0], di);
5497                 if (len > BTRFS_NAME_LEN)
5498                         len = BTRFS_NAME_LEN;
5499                 *size_ret += len;
5500
5501                 len += btrfs_dir_data_len(path.nodes[0], di);
5502                 len += sizeof(*di);
5503                 di = (struct btrfs_dir_item *)((char *)di + len);
5504                 cur += len;
5505         }
5506         goto loop;
5507
5508 out:
5509         btrfs_release_path(&path);
5510         return ret;
5511 }
5512
5513 static int count_dir_isize(struct btrfs_root *root, u64 ino, u64 *size)
5514 {
5515         u64 item_size;
5516         u64 index_size;
5517         int ret;
5518
5519         ASSERT(size);
5520         ret = __count_dir_isize(root, ino, BTRFS_DIR_ITEM_KEY, &item_size);
5521         if (ret)
5522                 goto out;
5523
5524         ret = __count_dir_isize(root, ino, BTRFS_DIR_INDEX_KEY, &index_size);
5525         if (ret)
5526                 goto out;
5527
5528         *size = item_size + index_size;
5529
5530 out:
5531         if (ret)
5532                 error("failed to count root %llu INODE[%llu] root size",
5533                       root->objectid, ino);
5534         return ret;
5535 }
5536
5537 /*
5538  * Traverse the given DIR_ITEM/DIR_INDEX and check related INODE_ITEM and
5539  * call find_inode_ref() to check related INODE_REF/INODE_EXTREF.
5540  *
5541  * @root:       the root of the fs/file tree
5542  * @key:        the key of the INODE_REF/INODE_EXTREF
5543  * @path:       the path
5544  * @size:       the st_size of the INODE_ITEM
5545  * @ext_ref:    the EXTENDED_IREF feature
5546  *
5547  * Return 0 if no error occurred.
5548  * Return DIR_COUNT_AGAIN if the isize of the inode should be recalculated.
5549  */
5550 static int check_dir_item(struct btrfs_root *root, struct btrfs_key *di_key,
5551                           struct btrfs_path *path, u64 *size,
5552                           unsigned int ext_ref)
5553 {
5554         struct btrfs_dir_item *di;
5555         struct btrfs_inode_item *ii;
5556         struct btrfs_key key;
5557         struct btrfs_key location;
5558         struct extent_buffer *node;
5559         int slot;
5560         char namebuf[BTRFS_NAME_LEN] = {0};
5561         u32 total;
5562         u32 cur = 0;
5563         u32 len;
5564         u32 name_len;
5565         u32 data_len;
5566         u8 filetype;
5567         u32 mode = 0;
5568         u64 index;
5569         int ret;
5570         int err;
5571         int tmp_err;
5572         int need_research = 0;
5573
5574         /*
5575          * For DIR_ITEM set index to (u64)-1, so that find_inode_ref
5576          * ignore index check.
5577          */
5578         if (di_key->type == BTRFS_DIR_INDEX_KEY)
5579                 index = di_key->offset;
5580         else
5581                 index = (u64)-1;
5582 begin:
5583         err = 0;
5584         cur = 0;
5585
5586         /* since after repair, path and the dir item may be changed */
5587         if (need_research) {
5588                 need_research = 0;
5589                 err |= DIR_COUNT_AGAIN;
5590                 btrfs_release_path(path);
5591                 ret = btrfs_search_slot(NULL, root, di_key, path, 0, 0);
5592                 /* the item was deleted, let path point the last checked item */
5593                 if (ret > 0) {
5594                         if (path->slots[0] == 0)
5595                                 btrfs_prev_leaf(root, path);
5596                         else
5597                                 path->slots[0]--;
5598                 }
5599                 if (ret)
5600                         goto out;
5601         }
5602
5603         node = path->nodes[0];
5604         slot = path->slots[0];
5605
5606         di = btrfs_item_ptr(node, slot, struct btrfs_dir_item);
5607         total = btrfs_item_size_nr(node, slot);
5608         memset(namebuf, 0, sizeof(namebuf) / sizeof(*namebuf));
5609
5610         while (cur < total) {
5611                 data_len = btrfs_dir_data_len(node, di);
5612                 tmp_err = 0;
5613                 if (data_len)
5614                         error("root %llu %s[%llu %llu] data_len shouldn't be %u",
5615                               root->objectid,
5616               di_key->type == BTRFS_DIR_ITEM_KEY ? "DIR_ITEM" : "DIR_INDEX",
5617                               di_key->objectid, di_key->offset, data_len);
5618
5619                 name_len = btrfs_dir_name_len(node, di);
5620                 if (name_len <= BTRFS_NAME_LEN) {
5621                         len = name_len;
5622                 } else {
5623                         len = BTRFS_NAME_LEN;
5624                         warning("root %llu %s[%llu %llu] name too long",
5625                                 root->objectid,
5626                 di_key->type == BTRFS_DIR_ITEM_KEY ? "DIR_ITEM" : "DIR_INDEX",
5627                                 di_key->objectid, di_key->offset);
5628                 }
5629                 (*size) += name_len;
5630                 read_extent_buffer(node, namebuf, (unsigned long)(di + 1),
5631                                    len);
5632                 filetype = btrfs_dir_type(node, di);
5633
5634                 if (di_key->type == BTRFS_DIR_ITEM_KEY &&
5635                     di_key->offset != btrfs_name_hash(namebuf, len)) {
5636                         err |= -EIO;
5637                         error("root %llu DIR_ITEM[%llu %llu] name %s namelen %u filetype %u mismatch with its hash, wanted %llu have %llu",
5638                         root->objectid, di_key->objectid, di_key->offset,
5639                         namebuf, len, filetype, di_key->offset,
5640                         btrfs_name_hash(namebuf, len));
5641                 }
5642
5643                 btrfs_dir_item_key_to_cpu(node, di, &location);
5644                 /* Ignore related ROOT_ITEM check */
5645                 if (location.type == BTRFS_ROOT_ITEM_KEY)
5646                         goto next;
5647
5648                 btrfs_release_path(path);
5649                 /* Check relative INODE_ITEM(existence/filetype) */
5650                 ret = btrfs_search_slot(NULL, root, &location, path, 0, 0);
5651                 if (ret) {
5652                         tmp_err |= INODE_ITEM_MISSING;
5653                         goto next;
5654                 }
5655
5656                 ii = btrfs_item_ptr(path->nodes[0], path->slots[0],
5657                                     struct btrfs_inode_item);
5658                 mode = btrfs_inode_mode(path->nodes[0], ii);
5659                 if (imode_to_type(mode) != filetype) {
5660                         tmp_err |= INODE_ITEM_MISMATCH;
5661                         goto next;
5662                 }
5663
5664                 /* Check relative INODE_REF/INODE_EXTREF */
5665                 key.objectid = location.objectid;
5666                 key.type = BTRFS_INODE_REF_KEY;
5667                 key.offset = di_key->objectid;
5668                 tmp_err |= find_inode_ref(root, &key, namebuf, len,
5669                                           &index, ext_ref);
5670
5671                 /* check relative INDEX/ITEM */
5672                 key.objectid = di_key->objectid;
5673                 if (key.type == BTRFS_DIR_ITEM_KEY) {
5674                         key.type = BTRFS_DIR_INDEX_KEY;
5675                         key.offset = index;
5676                 } else {
5677                         key.type = BTRFS_DIR_ITEM_KEY;
5678                         key.offset = btrfs_name_hash(namebuf, name_len);
5679                 }
5680
5681                 tmp_err |= find_dir_item(root, &key, &location, namebuf,
5682                                          name_len, filetype);
5683                 /* find_dir_item may find index */
5684                 if (key.type == BTRFS_DIR_INDEX_KEY)
5685                         index = key.offset;
5686 next:
5687
5688                 if (tmp_err && repair) {
5689                         ret = repair_dir_item(root, di_key->objectid,
5690                                               location.objectid, index,
5691                                               imode_to_type(mode), namebuf,
5692                                               name_len, tmp_err);
5693                         if (ret != tmp_err) {
5694                                 need_research = 1;
5695                                 goto begin;
5696                         }
5697                 }
5698                 btrfs_release_path(path);
5699                 print_dir_item_err(root, di_key, location.objectid, index,
5700                                    namebuf, name_len, filetype, tmp_err);
5701                 err |= tmp_err;
5702                 len = sizeof(*di) + name_len + data_len;
5703                 di = (struct btrfs_dir_item *)((char *)di + len);
5704                 cur += len;
5705
5706                 if (di_key->type == BTRFS_DIR_INDEX_KEY && cur < total) {
5707                         error("root %llu DIR_INDEX[%llu %llu] should contain only one entry",
5708                               root->objectid, di_key->objectid,
5709                               di_key->offset);
5710                         break;
5711                 }
5712         }
5713 out:
5714         /* research path */
5715         btrfs_release_path(path);
5716         ret = btrfs_search_slot(NULL, root, di_key, path, 0, 0);
5717         if (ret)
5718                 err |= ret > 0 ? -ENOENT : ret;
5719         return err;
5720 }
5721
5722 /*
5723  * Wrapper function of btrfs_punch_hole.
5724  *
5725  * Returns 0 means success.
5726  * Returns not 0 means error.
5727  */
5728 static int punch_extent_hole(struct btrfs_root *root, u64 ino, u64 start,
5729                              u64 len)
5730 {
5731         struct btrfs_trans_handle *trans;
5732         int ret = 0;
5733
5734         trans = btrfs_start_transaction(root, 1);
5735         if (IS_ERR(trans))
5736                 return PTR_ERR(trans);
5737
5738         ret = btrfs_punch_hole(trans, root, ino, start, len);
5739         if (ret)
5740                 error("failed to add hole [%llu, %llu] in inode [%llu]",
5741                       start, len, ino);
5742         else
5743                 printf("Add a hole [%llu, %llu] in inode [%llu]\n", start, len,
5744                        ino);
5745
5746         btrfs_commit_transaction(trans, root);
5747         return ret;
5748 }
5749
5750 /*
5751  * Check file extent datasum/hole, update the size of the file extents,
5752  * check and update the last offset of the file extent.
5753  *
5754  * @root:       the root of fs/file tree.
5755  * @fkey:       the key of the file extent.
5756  * @nodatasum:  INODE_NODATASUM feature.
5757  * @size:       the sum of all EXTENT_DATA items size for this inode.
5758  * @end:        the offset of the last extent.
5759  *
5760  * Return 0 if no error occurred.
5761  */
5762 static int check_file_extent(struct btrfs_root *root, struct btrfs_key *fkey,
5763                              struct extent_buffer *node, int slot,
5764                              unsigned int nodatasum, u64 *size, u64 *end)
5765 {
5766         struct btrfs_file_extent_item *fi;
5767         u64 disk_bytenr;
5768         u64 disk_num_bytes;
5769         u64 extent_num_bytes;
5770         u64 extent_offset;
5771         u64 csum_found;         /* In byte size, sectorsize aligned */
5772         u64 search_start;       /* Logical range start we search for csum */
5773         u64 search_len;         /* Logical range len we search for csum */
5774         unsigned int extent_type;
5775         unsigned int is_hole;
5776         int compressed = 0;
5777         int ret;
5778         int err = 0;
5779
5780         fi = btrfs_item_ptr(node, slot, struct btrfs_file_extent_item);
5781
5782         /* Check inline extent */
5783         extent_type = btrfs_file_extent_type(node, fi);
5784         if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
5785                 struct btrfs_item *e = btrfs_item_nr(slot);
5786                 u32 item_inline_len;
5787
5788                 item_inline_len = btrfs_file_extent_inline_item_len(node, e);
5789                 extent_num_bytes = btrfs_file_extent_inline_len(node, slot, fi);
5790                 compressed = btrfs_file_extent_compression(node, fi);
5791                 if (extent_num_bytes == 0) {
5792                         error(
5793                 "root %llu EXTENT_DATA[%llu %llu] has empty inline extent",
5794                                 root->objectid, fkey->objectid, fkey->offset);
5795                         err |= FILE_EXTENT_ERROR;
5796                 }
5797                 if (!compressed && extent_num_bytes != item_inline_len) {
5798                         error(
5799                 "root %llu EXTENT_DATA[%llu %llu] wrong inline size, have: %llu, expected: %u",
5800                                 root->objectid, fkey->objectid, fkey->offset,
5801                                 extent_num_bytes, item_inline_len);
5802                         err |= FILE_EXTENT_ERROR;
5803                 }
5804                 *end += extent_num_bytes;
5805                 *size += extent_num_bytes;
5806                 return err;
5807         }
5808
5809         /* Check extent type */
5810         if (extent_type != BTRFS_FILE_EXTENT_REG &&
5811                         extent_type != BTRFS_FILE_EXTENT_PREALLOC) {
5812                 err |= FILE_EXTENT_ERROR;
5813                 error("root %llu EXTENT_DATA[%llu %llu] type bad",
5814                       root->objectid, fkey->objectid, fkey->offset);
5815                 return err;
5816         }
5817
5818         /* Check REG_EXTENT/PREALLOC_EXTENT */
5819         disk_bytenr = btrfs_file_extent_disk_bytenr(node, fi);
5820         disk_num_bytes = btrfs_file_extent_disk_num_bytes(node, fi);
5821         extent_num_bytes = btrfs_file_extent_num_bytes(node, fi);
5822         extent_offset = btrfs_file_extent_offset(node, fi);
5823         compressed = btrfs_file_extent_compression(node, fi);
5824         is_hole = (disk_bytenr == 0) && (disk_num_bytes == 0);
5825
5826         /*
5827          * Check EXTENT_DATA csum
5828          *
5829          * For plain (uncompressed) extent, we should only check the range
5830          * we're referring to, as it's possible that part of prealloc extent
5831          * has been written, and has csum:
5832          *
5833          * |<--- Original large preallocated extent A ---->|
5834          * |<- Prealloc File Extent ->|<- Regular Extent ->|
5835          *      No csum                         Has csum
5836          *
5837          * For compressed extent, we should check the whole range.
5838          */
5839         if (!compressed) {
5840                 search_start = disk_bytenr + extent_offset;
5841                 search_len = extent_num_bytes;
5842         } else {
5843                 search_start = disk_bytenr;
5844                 search_len = disk_num_bytes;
5845         }
5846         ret = count_csum_range(root, search_start, search_len, &csum_found);
5847         if (csum_found > 0 && nodatasum) {
5848                 err |= ODD_CSUM_ITEM;
5849                 error("root %llu EXTENT_DATA[%llu %llu] nodatasum shouldn't have datasum",
5850                       root->objectid, fkey->objectid, fkey->offset);
5851         } else if (extent_type == BTRFS_FILE_EXTENT_REG && !nodatasum &&
5852                    !is_hole && (ret < 0 || csum_found < search_len)) {
5853                 err |= CSUM_ITEM_MISSING;
5854                 error("root %llu EXTENT_DATA[%llu %llu] csum missing, have: %llu, expected: %llu",
5855                       root->objectid, fkey->objectid, fkey->offset,
5856                       csum_found, search_len);
5857         } else if (extent_type == BTRFS_FILE_EXTENT_PREALLOC && csum_found > 0) {
5858                 err |= ODD_CSUM_ITEM;
5859                 error("root %llu EXTENT_DATA[%llu %llu] prealloc shouldn't have csum, but has: %llu",
5860                       root->objectid, fkey->objectid, fkey->offset, csum_found);
5861         }
5862
5863         /* Check EXTENT_DATA hole */
5864         if (!no_holes && *end != fkey->offset) {
5865                 if (repair)
5866                         ret = punch_extent_hole(root, fkey->objectid,
5867                                                 *end, fkey->offset - *end);
5868                 if (!repair || ret) {
5869                         err |= FILE_EXTENT_ERROR;
5870                         error(
5871                 "root %llu EXTENT_DATA[%llu %llu] interrupt, should start at %llu",
5872                         root->objectid, fkey->objectid, fkey->offset, *end);
5873                 }
5874         }
5875
5876         *end += extent_num_bytes;
5877         if (!is_hole)
5878                 *size += extent_num_bytes;
5879
5880         return err;
5881 }
5882
5883 /*
5884  * Set inode item nbytes to @nbytes
5885  *
5886  * Returns  0     on success
5887  * Returns  != 0  on error
5888  */
5889 static int repair_inode_nbytes_lowmem(struct btrfs_root *root,
5890                                       struct btrfs_path *path,
5891                                       u64 ino, u64 nbytes)
5892 {
5893         struct btrfs_trans_handle *trans;
5894         struct btrfs_inode_item *ii;
5895         struct btrfs_key key;
5896         struct btrfs_key research_key;
5897         int err = 0;
5898         int ret;
5899
5900         btrfs_item_key_to_cpu(path->nodes[0], &research_key, path->slots[0]);
5901
5902         key.objectid = ino;
5903         key.type = BTRFS_INODE_ITEM_KEY;
5904         key.offset = 0;
5905
5906         trans = btrfs_start_transaction(root, 1);
5907         if (IS_ERR(trans)) {
5908                 ret = PTR_ERR(trans);
5909                 err |= ret;
5910                 goto out;
5911         }
5912
5913         btrfs_release_path(path);
5914         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
5915         if (ret > 0)
5916                 ret = -ENOENT;
5917         if (ret) {
5918                 err |= ret;
5919                 goto fail;
5920         }
5921
5922         ii = btrfs_item_ptr(path->nodes[0], path->slots[0],
5923                             struct btrfs_inode_item);
5924         btrfs_set_inode_nbytes(path->nodes[0], ii, nbytes);
5925         btrfs_mark_buffer_dirty(path->nodes[0]);
5926 fail:
5927         btrfs_commit_transaction(trans, root);
5928 out:
5929         if (ret)
5930                 error("failed to set nbytes in inode %llu root %llu",
5931                       ino, root->root_key.objectid);
5932         else
5933                 printf("Set nbytes in inode item %llu root %llu\n to %llu", ino,
5934                        root->root_key.objectid, nbytes);
5935
5936         /* research path */
5937         btrfs_release_path(path);
5938         ret = btrfs_search_slot(NULL, root, &research_key, path, 0, 0);
5939         err |= ret;
5940
5941         return err;
5942 }
5943
5944 /*
5945  * Set directory inode isize to @isize.
5946  *
5947  * Returns 0     on success.
5948  * Returns != 0  on error.
5949  */
5950 static int repair_dir_isize_lowmem(struct btrfs_root *root,
5951                                    struct btrfs_path *path,
5952                                    u64 ino, u64 isize)
5953 {
5954         struct btrfs_trans_handle *trans;
5955         struct btrfs_inode_item *ii;
5956         struct btrfs_key key;
5957         struct btrfs_key research_key;
5958         int ret;
5959         int err = 0;
5960
5961         btrfs_item_key_to_cpu(path->nodes[0], &research_key, path->slots[0]);
5962
5963         key.objectid = ino;
5964         key.type = BTRFS_INODE_ITEM_KEY;
5965         key.offset = 0;
5966
5967         trans = btrfs_start_transaction(root, 1);
5968         if (IS_ERR(trans)) {
5969                 ret = PTR_ERR(trans);
5970                 err |= ret;
5971                 goto out;
5972         }
5973
5974         btrfs_release_path(path);
5975         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
5976         if (ret > 0)
5977                 ret = -ENOENT;
5978         if (ret) {
5979                 err |= ret;
5980                 goto fail;
5981         }
5982
5983         ii = btrfs_item_ptr(path->nodes[0], path->slots[0],
5984                             struct btrfs_inode_item);
5985         btrfs_set_inode_size(path->nodes[0], ii, isize);
5986         btrfs_mark_buffer_dirty(path->nodes[0]);
5987 fail:
5988         btrfs_commit_transaction(trans, root);
5989 out:
5990         if (ret)
5991                 error("failed to set isize in inode %llu root %llu",
5992                       ino, root->root_key.objectid);
5993         else
5994                 printf("Set isize in inode %llu root %llu to %llu\n",
5995                        ino, root->root_key.objectid, isize);
5996
5997         btrfs_release_path(path);
5998         ret = btrfs_search_slot(NULL, root, &research_key, path, 0, 0);
5999         err |= ret;
6000
6001         return err;
6002 }
6003
6004 /*
6005  * Wrapper function for btrfs_add_orphan_item().
6006  *
6007  * Returns 0     on success.
6008  * Returns != 0  on error.
6009  */
6010 static int repair_inode_orphan_item_lowmem(struct btrfs_root *root,
6011                                            struct btrfs_path *path, u64 ino)
6012 {
6013         struct btrfs_trans_handle *trans;
6014         struct btrfs_key research_key;
6015         int ret;
6016         int err = 0;
6017
6018         btrfs_item_key_to_cpu(path->nodes[0], &research_key, path->slots[0]);
6019
6020         trans = btrfs_start_transaction(root, 1);
6021         if (IS_ERR(trans)) {
6022                 ret = PTR_ERR(trans);
6023                 err |= ret;
6024                 goto out;
6025         }
6026
6027         btrfs_release_path(path);
6028         ret = btrfs_add_orphan_item(trans, root, path, ino);
6029         err |= ret;
6030         btrfs_commit_transaction(trans, root);
6031 out:
6032         if (ret)
6033                 error("failed to add inode %llu as orphan item root %llu",
6034                       ino, root->root_key.objectid);
6035         else
6036                 printf("Added inode %llu as orphan item root %llu\n",
6037                        ino, root->root_key.objectid);
6038
6039         btrfs_release_path(path);
6040         ret = btrfs_search_slot(NULL, root, &research_key, path, 0, 0);
6041         err |= ret;
6042
6043         return err;
6044 }
6045
6046 /* Set inode_item nlink to @ref_count.
6047  * If @ref_count == 0, move it to "lost+found" and increase @ref_count.
6048  *
6049  * Returns 0 on success
6050  */
6051 static int repair_inode_nlinks_lowmem(struct btrfs_root *root,
6052                                       struct btrfs_path *path, u64 ino,
6053                                       const char *name, u32 namelen,
6054                                       u64 ref_count, u8 filetype, u64 *nlink)
6055 {
6056         struct btrfs_trans_handle *trans;
6057         struct btrfs_inode_item *ii;
6058         struct btrfs_key key;
6059         struct btrfs_key old_key;
6060         char namebuf[BTRFS_NAME_LEN] = {0};
6061         int name_len;
6062         int ret;
6063         int ret2;
6064
6065         /* save the key */
6066         btrfs_item_key_to_cpu(path->nodes[0], &old_key, path->slots[0]);
6067
6068         if (name && namelen) {
6069                 ASSERT(namelen <= BTRFS_NAME_LEN);
6070                 memcpy(namebuf, name, namelen);
6071                 name_len = namelen;
6072         } else {
6073                 sprintf(namebuf, "%llu", ino);
6074                 name_len = count_digits(ino);
6075                 printf("Can't find file name for inode %llu, use %s instead\n",
6076                        ino, namebuf);
6077         }
6078
6079         trans = btrfs_start_transaction(root, 1);
6080         if (IS_ERR(trans)) {
6081                 ret = PTR_ERR(trans);
6082                 goto out;
6083         }
6084
6085         btrfs_release_path(path);
6086         /* if refs is 0, put it into lostfound */
6087         if (ref_count == 0) {
6088                 ret = link_inode_to_lostfound(trans, root, path, ino, namebuf,
6089                                               name_len, filetype, &ref_count);
6090                 if (ret)
6091                         goto fail;
6092         }
6093
6094         /* reset inode_item's nlink to ref_count */
6095         key.objectid = ino;
6096         key.type = BTRFS_INODE_ITEM_KEY;
6097         key.offset = 0;
6098
6099         btrfs_release_path(path);
6100         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
6101         if (ret > 0)
6102                 ret = -ENOENT;
6103         if (ret)
6104                 goto fail;
6105
6106         ii = btrfs_item_ptr(path->nodes[0], path->slots[0],
6107                             struct btrfs_inode_item);
6108         btrfs_set_inode_nlink(path->nodes[0], ii, ref_count);
6109         btrfs_mark_buffer_dirty(path->nodes[0]);
6110
6111         if (nlink)
6112                 *nlink = ref_count;
6113 fail:
6114         btrfs_commit_transaction(trans, root);
6115 out:
6116         if (ret)
6117                 error(
6118         "fail to repair nlink of inode %llu root %llu name %s filetype %u",
6119                        root->objectid, ino, namebuf, filetype);
6120         else
6121                 printf("Fixed nlink of inode %llu root %llu name %s filetype %u\n",
6122                        root->objectid, ino, namebuf, filetype);
6123
6124         /* research */
6125         btrfs_release_path(path);
6126         ret2 = btrfs_search_slot(NULL, root, &old_key, path, 0, 0);
6127         if (ret2 < 0)
6128                 return ret |= ret2;
6129         return ret;
6130 }
6131
6132 /*
6133  * Check INODE_ITEM and related ITEMs (the same inode number)
6134  * 1. check link count
6135  * 2. check inode ref/extref
6136  * 3. check dir item/index
6137  *
6138  * @ext_ref:    the EXTENDED_IREF feature
6139  *
6140  * Return 0 if no error occurred.
6141  * Return >0 for error or hit the traversal is done(by error bitmap)
6142  */
6143 static int check_inode_item(struct btrfs_root *root, struct btrfs_path *path,
6144                             unsigned int ext_ref)
6145 {
6146         struct extent_buffer *node;
6147         struct btrfs_inode_item *ii;
6148         struct btrfs_key key;
6149         struct btrfs_key last_key;
6150         u64 inode_id;
6151         u32 mode;
6152         u64 nlink;
6153         u64 nbytes;
6154         u64 isize;
6155         u64 size = 0;
6156         u64 refs = 0;
6157         u64 extent_end = 0;
6158         u64 extent_size = 0;
6159         unsigned int dir;
6160         unsigned int nodatasum;
6161         int slot;
6162         int ret;
6163         int err = 0;
6164         char namebuf[BTRFS_NAME_LEN] = {0};
6165         u32 name_len = 0;
6166
6167         node = path->nodes[0];
6168         slot = path->slots[0];
6169
6170         btrfs_item_key_to_cpu(node, &key, slot);
6171         inode_id = key.objectid;
6172
6173         if (inode_id == BTRFS_ORPHAN_OBJECTID) {
6174                 ret = btrfs_next_item(root, path);
6175                 if (ret > 0)
6176                         err |= LAST_ITEM;
6177                 return err;
6178         }
6179
6180         ii = btrfs_item_ptr(node, slot, struct btrfs_inode_item);
6181         isize = btrfs_inode_size(node, ii);
6182         nbytes = btrfs_inode_nbytes(node, ii);
6183         mode = btrfs_inode_mode(node, ii);
6184         dir = imode_to_type(mode) == BTRFS_FT_DIR;
6185         nlink = btrfs_inode_nlink(node, ii);
6186         nodatasum = btrfs_inode_flags(node, ii) & BTRFS_INODE_NODATASUM;
6187
6188         while (1) {
6189                 btrfs_item_key_to_cpu(path->nodes[0], &last_key, path->slots[0]);
6190                 ret = btrfs_next_item(root, path);
6191                 if (ret < 0) {
6192                         /* out will fill 'err' rusing current statistics */
6193                         goto out;
6194                 } else if (ret > 0) {
6195                         err |= LAST_ITEM;
6196                         goto out;
6197                 }
6198
6199                 node = path->nodes[0];
6200                 slot = path->slots[0];
6201                 btrfs_item_key_to_cpu(node, &key, slot);
6202                 if (key.objectid != inode_id)
6203                         goto out;
6204
6205                 switch (key.type) {
6206                 case BTRFS_INODE_REF_KEY:
6207                         ret = check_inode_ref(root, &key, path, namebuf,
6208                                               &name_len, &refs, mode);
6209                         err |= ret;
6210                         break;
6211                 case BTRFS_INODE_EXTREF_KEY:
6212                         if (key.type == BTRFS_INODE_EXTREF_KEY && !ext_ref)
6213                                 warning("root %llu EXTREF[%llu %llu] isn't supported",
6214                                         root->objectid, key.objectid,
6215                                         key.offset);
6216                         ret = check_inode_extref(root, &key, node, slot, &refs,
6217                                                  mode);
6218                         err |= ret;
6219                         break;
6220                 case BTRFS_DIR_ITEM_KEY:
6221                 case BTRFS_DIR_INDEX_KEY:
6222                         if (!dir) {
6223                                 warning("root %llu INODE[%llu] mode %u shouldn't have DIR_INDEX[%llu %llu]",
6224                                         root->objectid, inode_id,
6225                                         imode_to_type(mode), key.objectid,
6226                                         key.offset);
6227                         }
6228                         ret = check_dir_item(root, &key, path, &size, ext_ref);
6229                         err |= ret;
6230                         break;
6231                 case BTRFS_EXTENT_DATA_KEY:
6232                         if (dir) {
6233                                 warning("root %llu DIR INODE[%llu] shouldn't EXTENT_DATA[%llu %llu]",
6234                                         root->objectid, inode_id, key.objectid,
6235                                         key.offset);
6236                         }
6237                         ret = check_file_extent(root, &key, node, slot,
6238                                                 nodatasum, &extent_size,
6239                                                 &extent_end);
6240                         err |= ret;
6241                         break;
6242                 case BTRFS_XATTR_ITEM_KEY:
6243                         break;
6244                 default:
6245                         error("ITEM[%llu %u %llu] UNKNOWN TYPE",
6246                               key.objectid, key.type, key.offset);
6247                 }
6248         }
6249
6250 out:
6251         if (err & LAST_ITEM) {
6252                 btrfs_release_path(path);
6253                 ret = btrfs_search_slot(NULL, root, &last_key, path, 0, 0);
6254                 if (ret)
6255                         return err;
6256         }
6257
6258         /* verify INODE_ITEM nlink/isize/nbytes */
6259         if (dir) {
6260                 if (repair && (err & DIR_COUNT_AGAIN)) {
6261                         err &= ~DIR_COUNT_AGAIN;
6262                         count_dir_isize(root, inode_id, &size);
6263                 }
6264
6265                 if ((nlink != 1 || refs != 1) && repair) {
6266                         ret = repair_inode_nlinks_lowmem(root, path, inode_id,
6267                                 namebuf, name_len, refs, imode_to_type(mode),
6268                                 &nlink);
6269                 }
6270
6271                 if (nlink != 1) {
6272                         err |= LINK_COUNT_ERROR;
6273                         error("root %llu DIR INODE[%llu] shouldn't have more than one link(%llu)",
6274                               root->objectid, inode_id, nlink);
6275                 }
6276
6277                 /*
6278                  * Just a warning, as dir inode nbytes is just an
6279                  * instructive value.
6280                  */
6281                 if (!IS_ALIGNED(nbytes, root->fs_info->nodesize)) {
6282                         warning("root %llu DIR INODE[%llu] nbytes should be aligned to %u",
6283                                 root->objectid, inode_id,
6284                                 root->fs_info->nodesize);
6285                 }
6286
6287                 if (isize != size) {
6288                         if (repair)
6289                                 ret = repair_dir_isize_lowmem(root, path,
6290                                                               inode_id, size);
6291                         if (!repair || ret) {
6292                                 err |= ISIZE_ERROR;
6293                                 error(
6294                 "root %llu DIR INODE [%llu] size %llu not equal to %llu",
6295                                       root->objectid, inode_id, isize, size);
6296                         }
6297                 }
6298         } else {
6299                 if (nlink != refs) {
6300                         if (repair)
6301                                 ret = repair_inode_nlinks_lowmem(root, path,
6302                                          inode_id, namebuf, name_len, refs,
6303                                          imode_to_type(mode), &nlink);
6304                         if (!repair || ret) {
6305                                 err |= LINK_COUNT_ERROR;
6306                                 error(
6307                 "root %llu INODE[%llu] nlink(%llu) not equal to inode_refs(%llu)",
6308                                       root->objectid, inode_id, nlink, refs);
6309                         }
6310                 } else if (!nlink) {
6311                         if (repair)
6312                                 ret = repair_inode_orphan_item_lowmem(root,
6313                                                               path, inode_id);
6314                         if (!repair || ret) {
6315                                 err |= ORPHAN_ITEM;
6316                                 error("root %llu INODE[%llu] is orphan item",
6317                                       root->objectid, inode_id);
6318                         }
6319                 }
6320
6321                 if (!nbytes && !no_holes && extent_end < isize) {
6322                         if (repair)
6323                                 ret = punch_extent_hole(root, inode_id,
6324                                                 extent_end, isize - extent_end);
6325                         if (!repair || ret) {
6326                                 err |= NBYTES_ERROR;
6327                                 error(
6328         "root %llu INODE[%llu] size %llu should have a file extent hole",
6329                                       root->objectid, inode_id, isize);
6330                         }
6331                 }
6332
6333                 if (nbytes != extent_size) {
6334                         if (repair)
6335                                 ret = repair_inode_nbytes_lowmem(root, path,
6336                                                          inode_id, extent_size);
6337                         if (!repair || ret) {
6338                                 err |= NBYTES_ERROR;
6339                                 error(
6340         "root %llu INODE[%llu] nbytes %llu not equal to extent_size %llu",
6341                                       root->objectid, inode_id, nbytes,
6342                                       extent_size);
6343                         }
6344                 }
6345         }
6346
6347         if (err & LAST_ITEM)
6348                 btrfs_next_item(root, path);
6349         return err;
6350 }
6351
6352 /*
6353  * Insert the missing inode item and inode ref.
6354  *
6355  * Normal INODE_ITEM_MISSING and INODE_REF_MISSING are handled in backref * dir.
6356  * Root dir should be handled specially because root dir is the root of fs.
6357  *
6358  * returns err (>0 or 0) after repair
6359  */
6360 static int repair_fs_first_inode(struct btrfs_root *root, int err)
6361 {
6362         struct btrfs_trans_handle *trans;
6363         struct btrfs_key key;
6364         struct btrfs_path path;
6365         int filetype = BTRFS_FT_DIR;
6366         int ret = 0;
6367
6368         btrfs_init_path(&path);
6369
6370         if (err & INODE_REF_MISSING) {
6371                 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
6372                 key.type = BTRFS_INODE_REF_KEY;
6373                 key.offset = BTRFS_FIRST_FREE_OBJECTID;
6374
6375                 trans = btrfs_start_transaction(root, 1);
6376                 if (IS_ERR(trans)) {
6377                         ret = PTR_ERR(trans);
6378                         goto out;
6379                 }
6380
6381                 btrfs_release_path(&path);
6382                 ret = btrfs_search_slot(trans, root, &key, &path, 1, 1);
6383                 if (ret)
6384                         goto trans_fail;
6385
6386                 ret = btrfs_insert_inode_ref(trans, root, "..", 2,
6387                                              BTRFS_FIRST_FREE_OBJECTID,
6388                                              BTRFS_FIRST_FREE_OBJECTID, 0);
6389                 if (ret)
6390                         goto trans_fail;
6391
6392                 printf("Add INODE_REF[%llu %llu] name %s\n",
6393                        BTRFS_FIRST_FREE_OBJECTID, BTRFS_FIRST_FREE_OBJECTID,
6394                        "..");
6395                 err &= ~INODE_REF_MISSING;
6396 trans_fail:
6397                 if (ret)
6398                         error("fail to insert first inode's ref");
6399                 btrfs_commit_transaction(trans, root);
6400         }
6401
6402         if (err & INODE_ITEM_MISSING) {
6403                 ret = repair_inode_item_missing(root,
6404                                         BTRFS_FIRST_FREE_OBJECTID, filetype);
6405                 if (ret)
6406                         goto out;
6407                 err &= ~INODE_ITEM_MISSING;
6408         }
6409 out:
6410         if (ret)
6411                 error("fail to repair first inode");
6412         btrfs_release_path(&path);
6413         return err;
6414 }
6415
6416 /*
6417  * check first root dir's inode_item and inode_ref
6418  *
6419  * returns 0 means no error
6420  * returns >0 means error
6421  * returns <0 means fatal error
6422  */
6423 static int check_fs_first_inode(struct btrfs_root *root, unsigned int ext_ref)
6424 {
6425         struct btrfs_path path;
6426         struct btrfs_key key;
6427         struct btrfs_inode_item *ii;
6428         u64 index;
6429         u32 mode;
6430         int err = 0;
6431         int ret;
6432
6433         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
6434         key.type = BTRFS_INODE_ITEM_KEY;
6435         key.offset = 0;
6436
6437         /* For root being dropped, we don't need to check first inode */
6438         if (btrfs_root_refs(&root->root_item) == 0 &&
6439             btrfs_disk_key_objectid(&root->root_item.drop_progress) >=
6440             BTRFS_FIRST_FREE_OBJECTID)
6441                 return 0;
6442
6443         btrfs_init_path(&path);
6444         ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
6445         if (ret < 0)
6446                 goto out;
6447         if (ret > 0) {
6448                 ret = 0;
6449                 err |= INODE_ITEM_MISSING;
6450         } else {
6451                 ii = btrfs_item_ptr(path.nodes[0], path.slots[0],
6452                                     struct btrfs_inode_item);
6453                 mode = btrfs_inode_mode(path.nodes[0], ii);
6454                 if (imode_to_type(mode) != BTRFS_FT_DIR)
6455                         err |= INODE_ITEM_MISMATCH;
6456         }
6457
6458         /* lookup first inode ref */
6459         key.offset = BTRFS_FIRST_FREE_OBJECTID;
6460         key.type = BTRFS_INODE_REF_KEY;
6461         /* special index value */
6462         index = 0;
6463
6464         ret = find_inode_ref(root, &key, "..", strlen(".."), &index, ext_ref);
6465         if (ret < 0)
6466                 goto out;
6467         err |= ret;
6468
6469 out:
6470         btrfs_release_path(&path);
6471
6472         if (err && repair)
6473                 err = repair_fs_first_inode(root, err);
6474
6475         if (err & (INODE_ITEM_MISSING | INODE_ITEM_MISMATCH))
6476                 error("root dir INODE_ITEM is %s",
6477                       err & INODE_ITEM_MISMATCH ? "mismatch" : "missing");
6478         if (err & INODE_REF_MISSING)
6479                 error("root dir INODE_REF is missing");
6480
6481         return ret < 0 ? ret : err;
6482 }
6483
6484 static struct tree_backref *find_tree_backref(struct extent_record *rec,
6485                                                 u64 parent, u64 root)
6486 {
6487         struct rb_node *node;
6488         struct tree_backref *back = NULL;
6489         struct tree_backref match = {
6490                 .node = {
6491                         .is_data = 0,
6492                 },
6493         };
6494
6495         if (parent) {
6496                 match.parent = parent;
6497                 match.node.full_backref = 1;
6498         } else {
6499                 match.root = root;
6500         }
6501
6502         node = rb_search(&rec->backref_tree, &match.node.node,
6503                          (rb_compare_keys)compare_extent_backref, NULL);
6504         if (node)
6505                 back = to_tree_backref(rb_node_to_extent_backref(node));
6506
6507         return back;
6508 }
6509
6510 static struct data_backref *find_data_backref(struct extent_record *rec,
6511                                                 u64 parent, u64 root,
6512                                                 u64 owner, u64 offset,
6513                                                 int found_ref,
6514                                                 u64 disk_bytenr, u64 bytes)
6515 {
6516         struct rb_node *node;
6517         struct data_backref *back = NULL;
6518         struct data_backref match = {
6519                 .node = {
6520                         .is_data = 1,
6521                 },
6522                 .owner = owner,
6523                 .offset = offset,
6524                 .bytes = bytes,
6525                 .found_ref = found_ref,
6526                 .disk_bytenr = disk_bytenr,
6527         };
6528
6529         if (parent) {
6530                 match.parent = parent;
6531                 match.node.full_backref = 1;
6532         } else {
6533                 match.root = root;
6534         }
6535
6536         node = rb_search(&rec->backref_tree, &match.node.node,
6537                          (rb_compare_keys)compare_extent_backref, NULL);
6538         if (node)
6539                 back = to_data_backref(rb_node_to_extent_backref(node));
6540
6541         return back;
6542 }
6543 /*
6544  * This function calls walk_down_tree_v2 and walk_up_tree_v2 to check tree
6545  * blocks and integrity of fs tree items.
6546  *
6547  * @root:         the root of the tree to be checked.
6548  * @ext_ref       feature EXTENDED_IREF is enable or not.
6549  * @account       if NOT 0 means check the tree (including tree)'s treeblocks.
6550  *                otherwise means check fs tree(s) items relationship and
6551  *                @root MUST be a fs tree root.
6552  * Returns 0      represents OK.
6553  * Returns not 0  represents error.
6554  */
6555 static int check_btrfs_root(struct btrfs_trans_handle *trans,
6556                             struct btrfs_root *root, unsigned int ext_ref,
6557                             int check_all)
6558
6559 {
6560         struct btrfs_path path;
6561         struct node_refs nrefs;
6562         struct btrfs_root_item *root_item = &root->root_item;
6563         int ret;
6564         int level;
6565         int err = 0;
6566
6567         memset(&nrefs, 0, sizeof(nrefs));
6568         if (!check_all) {
6569                 /*
6570                  * We need to manually check the first inode item (256)
6571                  * As the following traversal function will only start from
6572                  * the first inode item in the leaf, if inode item (256) is
6573                  * missing we will skip it forever.
6574                  */
6575                 ret = check_fs_first_inode(root, ext_ref);
6576                 if (ret < 0)
6577                         return ret;
6578         }
6579
6580
6581         level = btrfs_header_level(root->node);
6582         btrfs_init_path(&path);
6583
6584         if (btrfs_root_refs(root_item) > 0 ||
6585             btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6586                 path.nodes[level] = root->node;
6587                 path.slots[level] = 0;
6588                 extent_buffer_get(root->node);
6589         } else {
6590                 struct btrfs_key key;
6591
6592                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6593                 level = root_item->drop_level;
6594                 path.lowest_level = level;
6595                 ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
6596                 if (ret < 0)
6597                         goto out;
6598                 ret = 0;
6599         }
6600
6601         while (1) {
6602                 ret = walk_down_tree_v2(trans, root, &path, &level, &nrefs,
6603                                         ext_ref, check_all);
6604
6605                 err |= !!ret;
6606
6607                 /* if ret is negative, walk shall stop */
6608                 if (ret < 0) {
6609                         ret = err;
6610                         break;
6611                 }
6612
6613                 ret = walk_up_tree_v2(root, &path, &level);
6614                 if (ret != 0) {
6615                         /* Normal exit, reset ret to err */
6616                         ret = err;
6617                         break;
6618                 }
6619         }
6620
6621 out:
6622         btrfs_release_path(&path);
6623         return ret;
6624 }
6625
6626 /*
6627  * Iterate all items in the tree and call check_inode_item() to check.
6628  *
6629  * @root:       the root of the tree to be checked.
6630  * @ext_ref:    the EXTENDED_IREF feature
6631  *
6632  * Return 0 if no error found.
6633  * Return <0 for error.
6634  */
6635 static int check_fs_root_v2(struct btrfs_root *root, unsigned int ext_ref)
6636 {
6637         reset_cached_block_groups(root->fs_info);
6638         return check_btrfs_root(NULL, root, ext_ref, 0);
6639 }
6640
6641 /*
6642  * Find the relative ref for root_ref and root_backref.
6643  *
6644  * @root:       the root of the root tree.
6645  * @ref_key:    the key of the root ref.
6646  *
6647  * Return 0 if no error occurred.
6648  */
6649 static int check_root_ref(struct btrfs_root *root, struct btrfs_key *ref_key,
6650                           struct extent_buffer *node, int slot)
6651 {
6652         struct btrfs_path path;
6653         struct btrfs_key key;
6654         struct btrfs_root_ref *ref;
6655         struct btrfs_root_ref *backref;
6656         char ref_name[BTRFS_NAME_LEN] = {0};
6657         char backref_name[BTRFS_NAME_LEN] = {0};
6658         u64 ref_dirid;
6659         u64 ref_seq;
6660         u32 ref_namelen;
6661         u64 backref_dirid;
6662         u64 backref_seq;
6663         u32 backref_namelen;
6664         u32 len;
6665         int ret;
6666         int err = 0;
6667
6668         ref = btrfs_item_ptr(node, slot, struct btrfs_root_ref);
6669         ref_dirid = btrfs_root_ref_dirid(node, ref);
6670         ref_seq = btrfs_root_ref_sequence(node, ref);
6671         ref_namelen = btrfs_root_ref_name_len(node, ref);
6672
6673         if (ref_namelen <= BTRFS_NAME_LEN) {
6674                 len = ref_namelen;
6675         } else {
6676                 len = BTRFS_NAME_LEN;
6677                 warning("%s[%llu %llu] ref_name too long",
6678                         ref_key->type == BTRFS_ROOT_REF_KEY ?
6679                         "ROOT_REF" : "ROOT_BACKREF", ref_key->objectid,
6680                         ref_key->offset);
6681         }
6682         read_extent_buffer(node, ref_name, (unsigned long)(ref + 1), len);
6683
6684         /* Find relative root_ref */
6685         key.objectid = ref_key->offset;
6686         key.type = BTRFS_ROOT_BACKREF_KEY + BTRFS_ROOT_REF_KEY - ref_key->type;
6687         key.offset = ref_key->objectid;
6688
6689         btrfs_init_path(&path);
6690         ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
6691         if (ret) {
6692                 err |= ROOT_REF_MISSING;
6693                 error("%s[%llu %llu] couldn't find relative ref",
6694                       ref_key->type == BTRFS_ROOT_REF_KEY ?
6695                       "ROOT_REF" : "ROOT_BACKREF",
6696                       ref_key->objectid, ref_key->offset);
6697                 goto out;
6698         }
6699
6700         backref = btrfs_item_ptr(path.nodes[0], path.slots[0],
6701                                  struct btrfs_root_ref);
6702         backref_dirid = btrfs_root_ref_dirid(path.nodes[0], backref);
6703         backref_seq = btrfs_root_ref_sequence(path.nodes[0], backref);
6704         backref_namelen = btrfs_root_ref_name_len(path.nodes[0], backref);
6705
6706         if (backref_namelen <= BTRFS_NAME_LEN) {
6707                 len = backref_namelen;
6708         } else {
6709                 len = BTRFS_NAME_LEN;
6710                 warning("%s[%llu %llu] ref_name too long",
6711                         key.type == BTRFS_ROOT_REF_KEY ?
6712                         "ROOT_REF" : "ROOT_BACKREF",
6713                         key.objectid, key.offset);
6714         }
6715         read_extent_buffer(path.nodes[0], backref_name,
6716                            (unsigned long)(backref + 1), len);
6717
6718         if (ref_dirid != backref_dirid || ref_seq != backref_seq ||
6719             ref_namelen != backref_namelen ||
6720             strncmp(ref_name, backref_name, len)) {
6721                 err |= ROOT_REF_MISMATCH;
6722                 error("%s[%llu %llu] mismatch relative ref",
6723                       ref_key->type == BTRFS_ROOT_REF_KEY ?
6724                       "ROOT_REF" : "ROOT_BACKREF",
6725                       ref_key->objectid, ref_key->offset);
6726         }
6727 out:
6728         btrfs_release_path(&path);
6729         return err;
6730 }
6731
6732 /*
6733  * Check all fs/file tree in low_memory mode.
6734  *
6735  * 1. for fs tree root item, call check_fs_root_v2()
6736  * 2. for fs tree root ref/backref, call check_root_ref()
6737  *
6738  * Return 0 if no error occurred.
6739  */
6740 static int check_fs_roots_v2(struct btrfs_fs_info *fs_info)
6741 {
6742         struct btrfs_root *tree_root = fs_info->tree_root;
6743         struct btrfs_root *cur_root = NULL;
6744         struct btrfs_path path;
6745         struct btrfs_key key;
6746         struct extent_buffer *node;
6747         unsigned int ext_ref;
6748         int slot;
6749         int ret;
6750         int err = 0;
6751
6752         ext_ref = btrfs_fs_incompat(fs_info, EXTENDED_IREF);
6753
6754         btrfs_init_path(&path);
6755         key.objectid = BTRFS_FS_TREE_OBJECTID;
6756         key.offset = 0;
6757         key.type = BTRFS_ROOT_ITEM_KEY;
6758
6759         ret = btrfs_search_slot(NULL, tree_root, &key, &path, 0, 0);
6760         if (ret < 0) {
6761                 err = ret;
6762                 goto out;
6763         } else if (ret > 0) {
6764                 err = -ENOENT;
6765                 goto out;
6766         }
6767
6768         while (1) {
6769                 node = path.nodes[0];
6770                 slot = path.slots[0];
6771                 btrfs_item_key_to_cpu(node, &key, slot);
6772                 if (key.objectid > BTRFS_LAST_FREE_OBJECTID)
6773                         goto out;
6774                 if (key.type == BTRFS_ROOT_ITEM_KEY &&
6775                     fs_root_objectid(key.objectid)) {
6776                         if (key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
6777                                 cur_root = btrfs_read_fs_root_no_cache(fs_info,
6778                                                                        &key);
6779                         } else {
6780                                 key.offset = (u64)-1;
6781                                 cur_root = btrfs_read_fs_root(fs_info, &key);
6782                         }
6783
6784                         if (IS_ERR(cur_root)) {
6785                                 error("Fail to read fs/subvol tree: %lld",
6786                                       key.objectid);
6787                                 err = -EIO;
6788                                 goto next;
6789                         }
6790
6791                         ret = check_fs_root_v2(cur_root, ext_ref);
6792                         err |= ret;
6793
6794                         if (key.objectid == BTRFS_TREE_RELOC_OBJECTID)
6795                                 btrfs_free_fs_root(cur_root);
6796                 } else if (key.type == BTRFS_ROOT_REF_KEY ||
6797                                 key.type == BTRFS_ROOT_BACKREF_KEY) {
6798                         ret = check_root_ref(tree_root, &key, node, slot);
6799                         err |= ret;
6800                 }
6801 next:
6802                 ret = btrfs_next_item(tree_root, &path);
6803                 if (ret > 0)
6804                         goto out;
6805                 if (ret < 0) {
6806                         err = ret;
6807                         goto out;
6808                 }
6809         }
6810
6811 out:
6812         btrfs_release_path(&path);
6813         return err;
6814 }
6815
6816 static int do_check_fs_roots(struct btrfs_fs_info *fs_info,
6817                           struct cache_tree *root_cache)
6818 {
6819         int ret;
6820
6821         if (!ctx.progress_enabled)
6822                 fprintf(stderr, "checking fs roots\n");
6823         if (check_mode == CHECK_MODE_LOWMEM)
6824                 ret = check_fs_roots_v2(fs_info);
6825         else
6826                 ret = check_fs_roots(fs_info, root_cache);
6827
6828         return ret;
6829 }
6830
6831 static int all_backpointers_checked(struct extent_record *rec, int print_errs)
6832 {
6833         struct extent_backref *back, *tmp;
6834         struct tree_backref *tback;
6835         struct data_backref *dback;
6836         u64 found = 0;
6837         int err = 0;
6838
6839         rbtree_postorder_for_each_entry_safe(back, tmp,
6840                                              &rec->backref_tree, node) {
6841                 if (!back->found_extent_tree) {
6842                         err = 1;
6843                         if (!print_errs)
6844                                 goto out;
6845                         if (back->is_data) {
6846                                 dback = to_data_backref(back);
6847                                 fprintf(stderr, "Data backref %llu %s %llu"
6848                                         " owner %llu offset %llu num_refs %lu"
6849                                         " not found in extent tree\n",
6850                                         (unsigned long long)rec->start,
6851                                         back->full_backref ?
6852                                         "parent" : "root",
6853                                         back->full_backref ?
6854                                         (unsigned long long)dback->parent:
6855                                         (unsigned long long)dback->root,
6856                                         (unsigned long long)dback->owner,
6857                                         (unsigned long long)dback->offset,
6858                                         (unsigned long)dback->num_refs);
6859                         } else {
6860                                 tback = to_tree_backref(back);
6861                                 fprintf(stderr, "Tree backref %llu parent %llu"
6862                                         " root %llu not found in extent tree\n",
6863                                         (unsigned long long)rec->start,
6864                                         (unsigned long long)tback->parent,
6865                                         (unsigned long long)tback->root);
6866                         }
6867                 }
6868                 if (!back->is_data && !back->found_ref) {
6869                         err = 1;
6870                         if (!print_errs)
6871                                 goto out;
6872                         tback = to_tree_backref(back);
6873                         fprintf(stderr, "Backref %llu %s %llu not referenced back %p\n",
6874                                 (unsigned long long)rec->start,
6875                                 back->full_backref ? "parent" : "root",
6876                                 back->full_backref ?
6877                                 (unsigned long long)tback->parent :
6878                                 (unsigned long long)tback->root, back);
6879                 }
6880                 if (back->is_data) {
6881                         dback = to_data_backref(back);
6882                         if (dback->found_ref != dback->num_refs) {
6883                                 err = 1;
6884                                 if (!print_errs)
6885                                         goto out;
6886                                 fprintf(stderr, "Incorrect local backref count"
6887                                         " on %llu %s %llu owner %llu"
6888                                         " offset %llu found %u wanted %u back %p\n",
6889                                         (unsigned long long)rec->start,
6890                                         back->full_backref ?
6891                                         "parent" : "root",
6892                                         back->full_backref ?
6893                                         (unsigned long long)dback->parent:
6894                                         (unsigned long long)dback->root,
6895                                         (unsigned long long)dback->owner,
6896                                         (unsigned long long)dback->offset,
6897                                         dback->found_ref, dback->num_refs, back);
6898                         }
6899                         if (dback->disk_bytenr != rec->start) {
6900                                 err = 1;
6901                                 if (!print_errs)
6902                                         goto out;
6903                                 fprintf(stderr, "Backref disk bytenr does not"
6904                                         " match extent record, bytenr=%llu, "
6905                                         "ref bytenr=%llu\n",
6906                                         (unsigned long long)rec->start,
6907                                         (unsigned long long)dback->disk_bytenr);
6908                         }
6909
6910                         if (dback->bytes != rec->nr) {
6911                                 err = 1;
6912                                 if (!print_errs)
6913                                         goto out;
6914                                 fprintf(stderr, "Backref bytes do not match "
6915                                         "extent backref, bytenr=%llu, ref "
6916                                         "bytes=%llu, backref bytes=%llu\n",
6917                                         (unsigned long long)rec->start,
6918                                         (unsigned long long)rec->nr,
6919                                         (unsigned long long)dback->bytes);
6920                         }
6921                 }
6922                 if (!back->is_data) {
6923                         found += 1;
6924                 } else {
6925                         dback = to_data_backref(back);
6926                         found += dback->found_ref;
6927                 }
6928         }
6929         if (found != rec->refs) {
6930                 err = 1;
6931                 if (!print_errs)
6932                         goto out;
6933                 fprintf(stderr, "Incorrect global backref count "
6934                         "on %llu found %llu wanted %llu\n",
6935                         (unsigned long long)rec->start,
6936                         (unsigned long long)found,
6937                         (unsigned long long)rec->refs);
6938         }
6939 out:
6940         return err;
6941 }
6942
6943 static void __free_one_backref(struct rb_node *node)
6944 {
6945         struct extent_backref *back = rb_node_to_extent_backref(node);
6946
6947         free(back);
6948 }
6949
6950 static void free_all_extent_backrefs(struct extent_record *rec)
6951 {
6952         rb_free_nodes(&rec->backref_tree, __free_one_backref);
6953 }
6954
6955 static void free_extent_record_cache(struct cache_tree *extent_cache)
6956 {
6957         struct cache_extent *cache;
6958         struct extent_record *rec;
6959
6960         while (1) {
6961                 cache = first_cache_extent(extent_cache);
6962                 if (!cache)
6963                         break;
6964                 rec = container_of(cache, struct extent_record, cache);
6965                 remove_cache_extent(extent_cache, cache);
6966                 free_all_extent_backrefs(rec);
6967                 free(rec);
6968         }
6969 }
6970
6971 static int maybe_free_extent_rec(struct cache_tree *extent_cache,
6972                                  struct extent_record *rec)
6973 {
6974         if (rec->content_checked && rec->owner_ref_checked &&
6975             rec->extent_item_refs == rec->refs && rec->refs > 0 &&
6976             rec->num_duplicates == 0 && !all_backpointers_checked(rec, 0) &&
6977             !rec->bad_full_backref && !rec->crossing_stripes &&
6978             !rec->wrong_chunk_type) {
6979                 remove_cache_extent(extent_cache, &rec->cache);
6980                 free_all_extent_backrefs(rec);
6981                 list_del_init(&rec->list);
6982                 free(rec);
6983         }
6984         return 0;
6985 }
6986
6987 static int check_owner_ref(struct btrfs_root *root,
6988                             struct extent_record *rec,
6989                             struct extent_buffer *buf)
6990 {
6991         struct extent_backref *node, *tmp;
6992         struct tree_backref *back;
6993         struct btrfs_root *ref_root;
6994         struct btrfs_key key;
6995         struct btrfs_path path;
6996         struct extent_buffer *parent;
6997         int level;
6998         int found = 0;
6999         int ret;
7000
7001         rbtree_postorder_for_each_entry_safe(node, tmp,
7002                                              &rec->backref_tree, node) {
7003                 if (node->is_data)
7004                         continue;
7005                 if (!node->found_ref)
7006                         continue;
7007                 if (node->full_backref)
7008                         continue;
7009                 back = to_tree_backref(node);
7010                 if (btrfs_header_owner(buf) == back->root)
7011                         return 0;
7012         }
7013         BUG_ON(rec->is_root);
7014
7015         /* try to find the block by search corresponding fs tree */
7016         key.objectid = btrfs_header_owner(buf);
7017         key.type = BTRFS_ROOT_ITEM_KEY;
7018         key.offset = (u64)-1;
7019
7020         ref_root = btrfs_read_fs_root(root->fs_info, &key);
7021         if (IS_ERR(ref_root))
7022                 return 1;
7023
7024         level = btrfs_header_level(buf);
7025         if (level == 0)
7026                 btrfs_item_key_to_cpu(buf, &key, 0);
7027         else
7028                 btrfs_node_key_to_cpu(buf, &key, 0);
7029
7030         btrfs_init_path(&path);
7031         path.lowest_level = level + 1;
7032         ret = btrfs_search_slot(NULL, ref_root, &key, &path, 0, 0);
7033         if (ret < 0)
7034                 return 0;
7035
7036         parent = path.nodes[level + 1];
7037         if (parent && buf->start == btrfs_node_blockptr(parent,
7038                                                         path.slots[level + 1]))
7039                 found = 1;
7040
7041         btrfs_release_path(&path);
7042         return found ? 0 : 1;
7043 }
7044
7045 static int is_extent_tree_record(struct extent_record *rec)
7046 {
7047         struct extent_backref *node, *tmp;
7048         struct tree_backref *back;
7049         int is_extent = 0;
7050
7051         rbtree_postorder_for_each_entry_safe(node, tmp,
7052                                              &rec->backref_tree, node) {
7053                 if (node->is_data)
7054                         return 0;
7055                 back = to_tree_backref(node);
7056                 if (node->full_backref)
7057                         return 0;
7058                 if (back->root == BTRFS_EXTENT_TREE_OBJECTID)
7059                         is_extent = 1;
7060         }
7061         return is_extent;
7062 }
7063
7064
7065 static int record_bad_block_io(struct btrfs_fs_info *info,
7066                                struct cache_tree *extent_cache,
7067                                u64 start, u64 len)
7068 {
7069         struct extent_record *rec;
7070         struct cache_extent *cache;
7071         struct btrfs_key key;
7072
7073         cache = lookup_cache_extent(extent_cache, start, len);
7074         if (!cache)
7075                 return 0;
7076
7077         rec = container_of(cache, struct extent_record, cache);
7078         if (!is_extent_tree_record(rec))
7079                 return 0;
7080
7081         btrfs_disk_key_to_cpu(&key, &rec->parent_key);
7082         return btrfs_add_corrupt_extent_record(info, &key, start, len, 0);
7083 }
7084
7085 static int swap_values(struct btrfs_root *root, struct btrfs_path *path,
7086                        struct extent_buffer *buf, int slot)
7087 {
7088         if (btrfs_header_level(buf)) {
7089                 struct btrfs_key_ptr ptr1, ptr2;
7090
7091                 read_extent_buffer(buf, &ptr1, btrfs_node_key_ptr_offset(slot),
7092                                    sizeof(struct btrfs_key_ptr));
7093                 read_extent_buffer(buf, &ptr2,
7094                                    btrfs_node_key_ptr_offset(slot + 1),
7095                                    sizeof(struct btrfs_key_ptr));
7096                 write_extent_buffer(buf, &ptr1,
7097                                     btrfs_node_key_ptr_offset(slot + 1),
7098                                     sizeof(struct btrfs_key_ptr));
7099                 write_extent_buffer(buf, &ptr2,
7100                                     btrfs_node_key_ptr_offset(slot),
7101                                     sizeof(struct btrfs_key_ptr));
7102                 if (slot == 0) {
7103                         struct btrfs_disk_key key;
7104                         btrfs_node_key(buf, &key, 0);
7105                         btrfs_fixup_low_keys(root, path, &key,
7106                                              btrfs_header_level(buf) + 1);
7107                 }
7108         } else {
7109                 struct btrfs_item *item1, *item2;
7110                 struct btrfs_key k1, k2;
7111                 char *item1_data, *item2_data;
7112                 u32 item1_offset, item2_offset, item1_size, item2_size;
7113
7114                 item1 = btrfs_item_nr(slot);
7115                 item2 = btrfs_item_nr(slot + 1);
7116                 btrfs_item_key_to_cpu(buf, &k1, slot);
7117                 btrfs_item_key_to_cpu(buf, &k2, slot + 1);
7118                 item1_offset = btrfs_item_offset(buf, item1);
7119                 item2_offset = btrfs_item_offset(buf, item2);
7120                 item1_size = btrfs_item_size(buf, item1);
7121                 item2_size = btrfs_item_size(buf, item2);
7122
7123                 item1_data = malloc(item1_size);
7124                 if (!item1_data)
7125                         return -ENOMEM;
7126                 item2_data = malloc(item2_size);
7127                 if (!item2_data) {
7128                         free(item1_data);
7129                         return -ENOMEM;
7130                 }
7131
7132                 read_extent_buffer(buf, item1_data, item1_offset, item1_size);
7133                 read_extent_buffer(buf, item2_data, item2_offset, item2_size);
7134
7135                 write_extent_buffer(buf, item1_data, item2_offset, item2_size);
7136                 write_extent_buffer(buf, item2_data, item1_offset, item1_size);
7137                 free(item1_data);
7138                 free(item2_data);
7139
7140                 btrfs_set_item_offset(buf, item1, item2_offset);
7141                 btrfs_set_item_offset(buf, item2, item1_offset);
7142                 btrfs_set_item_size(buf, item1, item2_size);
7143                 btrfs_set_item_size(buf, item2, item1_size);
7144
7145                 path->slots[0] = slot;
7146                 btrfs_set_item_key_unsafe(root, path, &k2);
7147                 path->slots[0] = slot + 1;
7148                 btrfs_set_item_key_unsafe(root, path, &k1);
7149         }
7150         return 0;
7151 }
7152
7153 static int fix_key_order(struct btrfs_root *root, struct btrfs_path *path)
7154 {
7155         struct extent_buffer *buf;
7156         struct btrfs_key k1, k2;
7157         int i;
7158         int level = path->lowest_level;
7159         int ret = -EIO;
7160
7161         buf = path->nodes[level];
7162         for (i = 0; i < btrfs_header_nritems(buf) - 1; i++) {
7163                 if (level) {
7164                         btrfs_node_key_to_cpu(buf, &k1, i);
7165                         btrfs_node_key_to_cpu(buf, &k2, i + 1);
7166                 } else {
7167                         btrfs_item_key_to_cpu(buf, &k1, i);
7168                         btrfs_item_key_to_cpu(buf, &k2, i + 1);
7169                 }
7170                 if (btrfs_comp_cpu_keys(&k1, &k2) < 0)
7171                         continue;
7172                 ret = swap_values(root, path, buf, i);
7173                 if (ret)
7174                         break;
7175                 btrfs_mark_buffer_dirty(buf);
7176                 i = 0;
7177         }
7178         return ret;
7179 }
7180
7181 static int delete_bogus_item(struct btrfs_root *root,
7182                              struct btrfs_path *path,
7183                              struct extent_buffer *buf, int slot)
7184 {
7185         struct btrfs_key key;
7186         int nritems = btrfs_header_nritems(buf);
7187
7188         btrfs_item_key_to_cpu(buf, &key, slot);
7189
7190         /* These are all the keys we can deal with missing. */
7191         if (key.type != BTRFS_DIR_INDEX_KEY &&
7192             key.type != BTRFS_EXTENT_ITEM_KEY &&
7193             key.type != BTRFS_METADATA_ITEM_KEY &&
7194             key.type != BTRFS_TREE_BLOCK_REF_KEY &&
7195             key.type != BTRFS_EXTENT_DATA_REF_KEY)
7196                 return -1;
7197
7198         printf("Deleting bogus item [%llu,%u,%llu] at slot %d on block %llu\n",
7199                (unsigned long long)key.objectid, key.type,
7200                (unsigned long long)key.offset, slot, buf->start);
7201         memmove_extent_buffer(buf, btrfs_item_nr_offset(slot),
7202                               btrfs_item_nr_offset(slot + 1),
7203                               sizeof(struct btrfs_item) *
7204                               (nritems - slot - 1));
7205         btrfs_set_header_nritems(buf, nritems - 1);
7206         if (slot == 0) {
7207                 struct btrfs_disk_key disk_key;
7208
7209                 btrfs_item_key(buf, &disk_key, 0);
7210                 btrfs_fixup_low_keys(root, path, &disk_key, 1);
7211         }
7212         btrfs_mark_buffer_dirty(buf);
7213         return 0;
7214 }
7215
7216 static int fix_item_offset(struct btrfs_root *root, struct btrfs_path *path)
7217 {
7218         struct extent_buffer *buf;
7219         int i;
7220         int ret = 0;
7221
7222         /* We should only get this for leaves */
7223         BUG_ON(path->lowest_level);
7224         buf = path->nodes[0];
7225 again:
7226         for (i = 0; i < btrfs_header_nritems(buf); i++) {
7227                 unsigned int shift = 0, offset;
7228
7229                 if (i == 0 && btrfs_item_end_nr(buf, i) !=
7230                     BTRFS_LEAF_DATA_SIZE(root)) {
7231                         if (btrfs_item_end_nr(buf, i) >
7232                             BTRFS_LEAF_DATA_SIZE(root)) {
7233                                 ret = delete_bogus_item(root, path, buf, i);
7234                                 if (!ret)
7235                                         goto again;
7236                                 fprintf(stderr, "item is off the end of the "
7237                                         "leaf, can't fix\n");
7238                                 ret = -EIO;
7239                                 break;
7240                         }
7241                         shift = BTRFS_LEAF_DATA_SIZE(root) -
7242                                 btrfs_item_end_nr(buf, i);
7243                 } else if (i > 0 && btrfs_item_end_nr(buf, i) !=
7244                            btrfs_item_offset_nr(buf, i - 1)) {
7245                         if (btrfs_item_end_nr(buf, i) >
7246                             btrfs_item_offset_nr(buf, i - 1)) {
7247                                 ret = delete_bogus_item(root, path, buf, i);
7248                                 if (!ret)
7249                                         goto again;
7250                                 fprintf(stderr, "items overlap, can't fix\n");
7251                                 ret = -EIO;
7252                                 break;
7253                         }
7254                         shift = btrfs_item_offset_nr(buf, i - 1) -
7255                                 btrfs_item_end_nr(buf, i);
7256                 }
7257                 if (!shift)
7258                         continue;
7259
7260                 printf("Shifting item nr %d by %u bytes in block %llu\n",
7261                        i, shift, (unsigned long long)buf->start);
7262                 offset = btrfs_item_offset_nr(buf, i);
7263                 memmove_extent_buffer(buf,
7264                                       btrfs_leaf_data(buf) + offset + shift,
7265                                       btrfs_leaf_data(buf) + offset,
7266                                       btrfs_item_size_nr(buf, i));
7267                 btrfs_set_item_offset(buf, btrfs_item_nr(i),
7268                                       offset + shift);
7269                 btrfs_mark_buffer_dirty(buf);
7270         }
7271
7272         /*
7273          * We may have moved things, in which case we want to exit so we don't
7274          * write those changes out.  Once we have proper abort functionality in
7275          * progs this can be changed to something nicer.
7276          */
7277         BUG_ON(ret);
7278         return ret;
7279 }
7280
7281 /*
7282  * Attempt to fix basic block failures.  If we can't fix it for whatever reason
7283  * then just return -EIO.
7284  */
7285 static int try_to_fix_bad_block(struct btrfs_root *root,
7286                                 struct extent_buffer *buf,
7287                                 enum btrfs_tree_block_status status)
7288 {
7289         struct btrfs_trans_handle *trans;
7290         struct ulist *roots;
7291         struct ulist_node *node;
7292         struct btrfs_root *search_root;
7293         struct btrfs_path path;
7294         struct ulist_iterator iter;
7295         struct btrfs_key root_key, key;
7296         int ret;
7297
7298         if (status != BTRFS_TREE_BLOCK_BAD_KEY_ORDER &&
7299             status != BTRFS_TREE_BLOCK_INVALID_OFFSETS)
7300                 return -EIO;
7301
7302         ret = btrfs_find_all_roots(NULL, root->fs_info, buf->start, 0, &roots);
7303         if (ret)
7304                 return -EIO;
7305
7306         btrfs_init_path(&path);
7307         ULIST_ITER_INIT(&iter);
7308         while ((node = ulist_next(roots, &iter))) {
7309                 root_key.objectid = node->val;
7310                 root_key.type = BTRFS_ROOT_ITEM_KEY;
7311                 root_key.offset = (u64)-1;
7312
7313                 search_root = btrfs_read_fs_root(root->fs_info, &root_key);
7314                 if (IS_ERR(root)) {
7315                         ret = -EIO;
7316                         break;
7317                 }
7318
7319
7320                 trans = btrfs_start_transaction(search_root, 0);
7321                 if (IS_ERR(trans)) {
7322                         ret = PTR_ERR(trans);
7323                         break;
7324                 }
7325
7326                 path.lowest_level = btrfs_header_level(buf);
7327                 path.skip_check_block = 1;
7328                 if (path.lowest_level)
7329                         btrfs_node_key_to_cpu(buf, &key, 0);
7330                 else
7331                         btrfs_item_key_to_cpu(buf, &key, 0);
7332                 ret = btrfs_search_slot(trans, search_root, &key, &path, 0, 1);
7333                 if (ret) {
7334                         ret = -EIO;
7335                         btrfs_commit_transaction(trans, search_root);
7336                         break;
7337                 }
7338                 if (status == BTRFS_TREE_BLOCK_BAD_KEY_ORDER)
7339                         ret = fix_key_order(search_root, &path);
7340                 else if (status == BTRFS_TREE_BLOCK_INVALID_OFFSETS)
7341                         ret = fix_item_offset(search_root, &path);
7342                 if (ret) {
7343                         btrfs_commit_transaction(trans, search_root);
7344                         break;
7345                 }
7346                 btrfs_release_path(&path);
7347                 btrfs_commit_transaction(trans, search_root);
7348         }
7349         ulist_free(roots);
7350         btrfs_release_path(&path);
7351         return ret;
7352 }
7353
7354 static int check_block(struct btrfs_root *root,
7355                        struct cache_tree *extent_cache,
7356                        struct extent_buffer *buf, u64 flags)
7357 {
7358         struct extent_record *rec;
7359         struct cache_extent *cache;
7360         struct btrfs_key key;
7361         enum btrfs_tree_block_status status;
7362         int ret = 0;
7363         int level;
7364
7365         cache = lookup_cache_extent(extent_cache, buf->start, buf->len);
7366         if (!cache)
7367                 return 1;
7368         rec = container_of(cache, struct extent_record, cache);
7369         rec->generation = btrfs_header_generation(buf);
7370
7371         level = btrfs_header_level(buf);
7372         if (btrfs_header_nritems(buf) > 0) {
7373
7374                 if (level == 0)
7375                         btrfs_item_key_to_cpu(buf, &key, 0);
7376                 else
7377                         btrfs_node_key_to_cpu(buf, &key, 0);
7378
7379                 rec->info_objectid = key.objectid;
7380         }
7381         rec->info_level = level;
7382
7383         if (btrfs_is_leaf(buf))
7384                 status = btrfs_check_leaf(root, &rec->parent_key, buf);
7385         else
7386                 status = btrfs_check_node(root, &rec->parent_key, buf);
7387
7388         if (status != BTRFS_TREE_BLOCK_CLEAN) {
7389                 if (repair)
7390                         status = try_to_fix_bad_block(root, buf, status);
7391                 if (status != BTRFS_TREE_BLOCK_CLEAN) {
7392                         ret = -EIO;
7393                         fprintf(stderr, "bad block %llu\n",
7394                                 (unsigned long long)buf->start);
7395                 } else {
7396                         /*
7397                          * Signal to callers we need to start the scan over
7398                          * again since we'll have cowed blocks.
7399                          */
7400                         ret = -EAGAIN;
7401                 }
7402         } else {
7403                 rec->content_checked = 1;
7404                 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7405                         rec->owner_ref_checked = 1;
7406                 else {
7407                         ret = check_owner_ref(root, rec, buf);
7408                         if (!ret)
7409                                 rec->owner_ref_checked = 1;
7410                 }
7411         }
7412         if (!ret)
7413                 maybe_free_extent_rec(extent_cache, rec);
7414         return ret;
7415 }
7416
7417 #if 0
7418 static struct tree_backref *find_tree_backref(struct extent_record *rec,
7419                                                 u64 parent, u64 root)
7420 {
7421         struct list_head *cur = rec->backrefs.next;
7422         struct extent_backref *node;
7423         struct tree_backref *back;
7424
7425         while(cur != &rec->backrefs) {
7426                 node = to_extent_backref(cur);
7427                 cur = cur->next;
7428                 if (node->is_data)
7429                         continue;
7430                 back = to_tree_backref(node);
7431                 if (parent > 0) {
7432                         if (!node->full_backref)
7433                                 continue;
7434                         if (parent == back->parent)
7435                                 return back;
7436                 } else {
7437                         if (node->full_backref)
7438                                 continue;
7439                         if (back->root == root)
7440                                 return back;
7441                 }
7442         }
7443         return NULL;
7444 }
7445 #endif
7446
7447 static struct tree_backref *alloc_tree_backref(struct extent_record *rec,
7448                                                 u64 parent, u64 root)
7449 {
7450         struct tree_backref *ref = malloc(sizeof(*ref));
7451
7452         if (!ref)
7453                 return NULL;
7454         memset(&ref->node, 0, sizeof(ref->node));
7455         if (parent > 0) {
7456                 ref->parent = parent;
7457                 ref->node.full_backref = 1;
7458         } else {
7459                 ref->root = root;
7460                 ref->node.full_backref = 0;
7461         }
7462
7463         return ref;
7464 }
7465
7466 #if 0
7467 static struct data_backref *find_data_backref(struct extent_record *rec,
7468                                                 u64 parent, u64 root,
7469                                                 u64 owner, u64 offset,
7470                                                 int found_ref,
7471                                                 u64 disk_bytenr, u64 bytes)
7472 {
7473         struct list_head *cur = rec->backrefs.next;
7474         struct extent_backref *node;
7475         struct data_backref *back;
7476
7477         while(cur != &rec->backrefs) {
7478                 node = to_extent_backref(cur);
7479                 cur = cur->next;
7480                 if (!node->is_data)
7481                         continue;
7482                 back = to_data_backref(node);
7483                 if (parent > 0) {
7484                         if (!node->full_backref)
7485                                 continue;
7486                         if (parent == back->parent)
7487                                 return back;
7488                 } else {
7489                         if (node->full_backref)
7490                                 continue;
7491                         if (back->root == root && back->owner == owner &&
7492                             back->offset == offset) {
7493                                 if (found_ref && node->found_ref &&
7494                                     (back->bytes != bytes ||
7495                                     back->disk_bytenr != disk_bytenr))
7496                                         continue;
7497                                 return back;
7498                         }
7499                 }
7500         }
7501         return NULL;
7502 }
7503 #endif
7504
7505 static struct data_backref *alloc_data_backref(struct extent_record *rec,
7506                                                 u64 parent, u64 root,
7507                                                 u64 owner, u64 offset,
7508                                                 u64 max_size)
7509 {
7510         struct data_backref *ref = malloc(sizeof(*ref));
7511
7512         if (!ref)
7513                 return NULL;
7514         memset(&ref->node, 0, sizeof(ref->node));
7515         ref->node.is_data = 1;
7516
7517         if (parent > 0) {
7518                 ref->parent = parent;
7519                 ref->owner = 0;
7520                 ref->offset = 0;
7521                 ref->node.full_backref = 1;
7522         } else {
7523                 ref->root = root;
7524                 ref->owner = owner;
7525                 ref->offset = offset;
7526                 ref->node.full_backref = 0;
7527         }
7528         ref->bytes = max_size;
7529         ref->found_ref = 0;
7530         ref->num_refs = 0;
7531         if (max_size > rec->max_size)
7532                 rec->max_size = max_size;
7533         return ref;
7534 }
7535
7536 /* Check if the type of extent matches with its chunk */
7537 static void check_extent_type(struct extent_record *rec)
7538 {
7539         struct btrfs_block_group_cache *bg_cache;
7540
7541         bg_cache = btrfs_lookup_first_block_group(global_info, rec->start);
7542         if (!bg_cache)
7543                 return;
7544
7545         /* data extent, check chunk directly*/
7546         if (!rec->metadata) {
7547                 if (!(bg_cache->flags & BTRFS_BLOCK_GROUP_DATA))
7548                         rec->wrong_chunk_type = 1;
7549                 return;
7550         }
7551
7552         /* metadata extent, check the obvious case first */
7553         if (!(bg_cache->flags & (BTRFS_BLOCK_GROUP_SYSTEM |
7554                                  BTRFS_BLOCK_GROUP_METADATA))) {
7555                 rec->wrong_chunk_type = 1;
7556                 return;
7557         }
7558
7559         /*
7560          * Check SYSTEM extent, as it's also marked as metadata, we can only
7561          * make sure it's a SYSTEM extent by its backref
7562          */
7563         if (!RB_EMPTY_ROOT(&rec->backref_tree)) {
7564                 struct extent_backref *node;
7565                 struct tree_backref *tback;
7566                 u64 bg_type;
7567
7568                 node = rb_node_to_extent_backref(rb_first(&rec->backref_tree));
7569                 if (node->is_data) {
7570                         /* tree block shouldn't have data backref */
7571                         rec->wrong_chunk_type = 1;
7572                         return;
7573                 }
7574                 tback = container_of(node, struct tree_backref, node);
7575
7576                 if (tback->root == BTRFS_CHUNK_TREE_OBJECTID)
7577                         bg_type = BTRFS_BLOCK_GROUP_SYSTEM;
7578                 else
7579                         bg_type = BTRFS_BLOCK_GROUP_METADATA;
7580                 if (!(bg_cache->flags & bg_type))
7581                         rec->wrong_chunk_type = 1;
7582         }
7583 }
7584
7585 /*
7586  * Allocate a new extent record, fill default values from @tmpl and insert int
7587  * @extent_cache. Caller is supposed to make sure the [start,nr) is not in
7588  * the cache, otherwise it fails.
7589  */
7590 static int add_extent_rec_nolookup(struct cache_tree *extent_cache,
7591                 struct extent_record *tmpl)
7592 {
7593         struct extent_record *rec;
7594         int ret = 0;
7595
7596         BUG_ON(tmpl->max_size == 0);
7597         rec = malloc(sizeof(*rec));
7598         if (!rec)
7599                 return -ENOMEM;
7600         rec->start = tmpl->start;
7601         rec->max_size = tmpl->max_size;
7602         rec->nr = max(tmpl->nr, tmpl->max_size);
7603         rec->found_rec = tmpl->found_rec;
7604         rec->content_checked = tmpl->content_checked;
7605         rec->owner_ref_checked = tmpl->owner_ref_checked;
7606         rec->num_duplicates = 0;
7607         rec->metadata = tmpl->metadata;
7608         rec->flag_block_full_backref = FLAG_UNSET;
7609         rec->bad_full_backref = 0;
7610         rec->crossing_stripes = 0;
7611         rec->wrong_chunk_type = 0;
7612         rec->is_root = tmpl->is_root;
7613         rec->refs = tmpl->refs;
7614         rec->extent_item_refs = tmpl->extent_item_refs;
7615         rec->parent_generation = tmpl->parent_generation;
7616         INIT_LIST_HEAD(&rec->backrefs);
7617         INIT_LIST_HEAD(&rec->dups);
7618         INIT_LIST_HEAD(&rec->list);
7619         rec->backref_tree = RB_ROOT;
7620         memcpy(&rec->parent_key, &tmpl->parent_key, sizeof(tmpl->parent_key));
7621         rec->cache.start = tmpl->start;
7622         rec->cache.size = tmpl->nr;
7623         ret = insert_cache_extent(extent_cache, &rec->cache);
7624         if (ret) {
7625                 free(rec);
7626                 return ret;
7627         }
7628         bytes_used += rec->nr;
7629
7630         if (tmpl->metadata)
7631                 rec->crossing_stripes = check_crossing_stripes(global_info,
7632                                 rec->start, global_info->nodesize);
7633         check_extent_type(rec);
7634         return ret;
7635 }
7636
7637 /*
7638  * Lookup and modify an extent, some values of @tmpl are interpreted verbatim,
7639  * some are hints:
7640  * - refs              - if found, increase refs
7641  * - is_root           - if found, set
7642  * - content_checked   - if found, set
7643  * - owner_ref_checked - if found, set
7644  *
7645  * If not found, create a new one, initialize and insert.
7646  */
7647 static int add_extent_rec(struct cache_tree *extent_cache,
7648                 struct extent_record *tmpl)
7649 {
7650         struct extent_record *rec;
7651         struct cache_extent *cache;
7652         int ret = 0;
7653         int dup = 0;
7654
7655         cache = lookup_cache_extent(extent_cache, tmpl->start, tmpl->nr);
7656         if (cache) {
7657                 rec = container_of(cache, struct extent_record, cache);
7658                 if (tmpl->refs)
7659                         rec->refs++;
7660                 if (rec->nr == 1)
7661                         rec->nr = max(tmpl->nr, tmpl->max_size);
7662
7663                 /*
7664                  * We need to make sure to reset nr to whatever the extent
7665                  * record says was the real size, this way we can compare it to
7666                  * the backrefs.
7667                  */
7668                 if (tmpl->found_rec) {
7669                         if (tmpl->start != rec->start || rec->found_rec) {
7670                                 struct extent_record *tmp;
7671
7672                                 dup = 1;
7673                                 if (list_empty(&rec->list))
7674                                         list_add_tail(&rec->list,
7675                                                       &duplicate_extents);
7676
7677                                 /*
7678                                  * We have to do this song and dance in case we
7679                                  * find an extent record that falls inside of
7680                                  * our current extent record but does not have
7681                                  * the same objectid.
7682                                  */
7683                                 tmp = malloc(sizeof(*tmp));
7684                                 if (!tmp)
7685                                         return -ENOMEM;
7686                                 tmp->start = tmpl->start;
7687                                 tmp->max_size = tmpl->max_size;
7688                                 tmp->nr = tmpl->nr;
7689                                 tmp->found_rec = 1;
7690                                 tmp->metadata = tmpl->metadata;
7691                                 tmp->extent_item_refs = tmpl->extent_item_refs;
7692                                 INIT_LIST_HEAD(&tmp->list);
7693                                 list_add_tail(&tmp->list, &rec->dups);
7694                                 rec->num_duplicates++;
7695                         } else {
7696                                 rec->nr = tmpl->nr;
7697                                 rec->found_rec = 1;
7698                         }
7699                 }
7700
7701                 if (tmpl->extent_item_refs && !dup) {
7702                         if (rec->extent_item_refs) {
7703                                 fprintf(stderr, "block %llu rec "
7704                                         "extent_item_refs %llu, passed %llu\n",
7705                                         (unsigned long long)tmpl->start,
7706                                         (unsigned long long)
7707                                                         rec->extent_item_refs,
7708                                         (unsigned long long)tmpl->extent_item_refs);
7709                         }
7710                         rec->extent_item_refs = tmpl->extent_item_refs;
7711                 }
7712                 if (tmpl->is_root)
7713                         rec->is_root = 1;
7714                 if (tmpl->content_checked)
7715                         rec->content_checked = 1;
7716                 if (tmpl->owner_ref_checked)
7717                         rec->owner_ref_checked = 1;
7718                 memcpy(&rec->parent_key, &tmpl->parent_key,
7719                                 sizeof(tmpl->parent_key));
7720                 if (tmpl->parent_generation)
7721                         rec->parent_generation = tmpl->parent_generation;
7722                 if (rec->max_size < tmpl->max_size)
7723                         rec->max_size = tmpl->max_size;
7724
7725                 /*
7726                  * A metadata extent can't cross stripe_len boundary, otherwise
7727                  * kernel scrub won't be able to handle it.
7728                  * As now stripe_len is fixed to BTRFS_STRIPE_LEN, just check
7729                  * it.
7730                  */
7731                 if (tmpl->metadata)
7732                         rec->crossing_stripes = check_crossing_stripes(
7733                                         global_info, rec->start,
7734                                         global_info->nodesize);
7735                 check_extent_type(rec);
7736                 maybe_free_extent_rec(extent_cache, rec);
7737                 return ret;
7738         }
7739
7740         ret = add_extent_rec_nolookup(extent_cache, tmpl);
7741
7742         return ret;
7743 }
7744
7745 static int add_tree_backref(struct cache_tree *extent_cache, u64 bytenr,
7746                             u64 parent, u64 root, int found_ref)
7747 {
7748         struct extent_record *rec;
7749         struct tree_backref *back;
7750         struct cache_extent *cache;
7751         int ret;
7752         bool insert = false;
7753
7754         cache = lookup_cache_extent(extent_cache, bytenr, 1);
7755         if (!cache) {
7756                 struct extent_record tmpl;
7757
7758                 memset(&tmpl, 0, sizeof(tmpl));
7759                 tmpl.start = bytenr;
7760                 tmpl.nr = 1;
7761                 tmpl.metadata = 1;
7762                 tmpl.max_size = 1;
7763
7764                 ret = add_extent_rec_nolookup(extent_cache, &tmpl);
7765                 if (ret)
7766                         return ret;
7767
7768                 /* really a bug in cache_extent implement now */
7769                 cache = lookup_cache_extent(extent_cache, bytenr, 1);
7770                 if (!cache)
7771                         return -ENOENT;
7772         }
7773
7774         rec = container_of(cache, struct extent_record, cache);
7775         if (rec->start != bytenr) {
7776                 /*
7777                  * Several cause, from unaligned bytenr to over lapping extents
7778                  */
7779                 return -EEXIST;
7780         }
7781
7782         back = find_tree_backref(rec, parent, root);
7783         if (!back) {
7784                 back = alloc_tree_backref(rec, parent, root);
7785                 if (!back)
7786                         return -ENOMEM;
7787                 insert = true;
7788         }
7789
7790         if (found_ref) {
7791                 if (back->node.found_ref) {
7792                         fprintf(stderr, "Extent back ref already exists "
7793                                 "for %llu parent %llu root %llu \n",
7794                                 (unsigned long long)bytenr,
7795                                 (unsigned long long)parent,
7796                                 (unsigned long long)root);
7797                 }
7798                 back->node.found_ref = 1;
7799         } else {
7800                 if (back->node.found_extent_tree) {
7801                         fprintf(stderr, "Extent back ref already exists "
7802                                 "for %llu parent %llu root %llu \n",
7803                                 (unsigned long long)bytenr,
7804                                 (unsigned long long)parent,
7805                                 (unsigned long long)root);
7806                 }
7807                 back->node.found_extent_tree = 1;
7808         }
7809         if (insert)
7810                 WARN_ON(rb_insert(&rec->backref_tree, &back->node.node,
7811                         compare_extent_backref));
7812         check_extent_type(rec);
7813         maybe_free_extent_rec(extent_cache, rec);
7814         return 0;
7815 }
7816
7817 static int add_data_backref(struct cache_tree *extent_cache, u64 bytenr,
7818                             u64 parent, u64 root, u64 owner, u64 offset,
7819                             u32 num_refs, int found_ref, u64 max_size)
7820 {
7821         struct extent_record *rec;
7822         struct data_backref *back;
7823         struct cache_extent *cache;
7824         int ret;
7825         bool insert = false;
7826
7827         cache = lookup_cache_extent(extent_cache, bytenr, 1);
7828         if (!cache) {
7829                 struct extent_record tmpl;
7830
7831                 memset(&tmpl, 0, sizeof(tmpl));
7832                 tmpl.start = bytenr;
7833                 tmpl.nr = 1;
7834                 tmpl.max_size = max_size;
7835
7836                 ret = add_extent_rec_nolookup(extent_cache, &tmpl);
7837                 if (ret)
7838                         return ret;
7839
7840                 cache = lookup_cache_extent(extent_cache, bytenr, 1);
7841                 if (!cache)
7842                         abort();
7843         }
7844
7845         rec = container_of(cache, struct extent_record, cache);
7846         if (rec->max_size < max_size)
7847                 rec->max_size = max_size;
7848
7849         /*
7850          * If found_ref is set then max_size is the real size and must match the
7851          * existing refs.  So if we have already found a ref then we need to
7852          * make sure that this ref matches the existing one, otherwise we need
7853          * to add a new backref so we can notice that the backrefs don't match
7854          * and we need to figure out who is telling the truth.  This is to
7855          * account for that awful fsync bug I introduced where we'd end up with
7856          * a btrfs_file_extent_item that would have its length include multiple
7857          * prealloc extents or point inside of a prealloc extent.
7858          */
7859         back = find_data_backref(rec, parent, root, owner, offset, found_ref,
7860                                  bytenr, max_size);
7861         if (!back) {
7862                 back = alloc_data_backref(rec, parent, root, owner, offset,
7863                                           max_size);
7864                 BUG_ON(!back);
7865                 insert = true;
7866         }
7867
7868         if (found_ref) {
7869                 BUG_ON(num_refs != 1);
7870                 if (back->node.found_ref)
7871                         BUG_ON(back->bytes != max_size);
7872                 back->node.found_ref = 1;
7873                 back->found_ref += 1;
7874                 if (back->bytes != max_size || back->disk_bytenr != bytenr) {
7875                         back->bytes = max_size;
7876                         back->disk_bytenr = bytenr;
7877
7878                         /* Need to reinsert if not already in the tree */
7879                         if (!insert) {
7880                                 rb_erase(&back->node.node, &rec->backref_tree);
7881                                 insert = true;
7882                         }
7883                 }
7884                 rec->refs += 1;
7885                 rec->content_checked = 1;
7886                 rec->owner_ref_checked = 1;
7887         } else {
7888                 if (back->node.found_extent_tree) {
7889                         fprintf(stderr, "Extent back ref already exists "
7890                                 "for %llu parent %llu root %llu "
7891                                 "owner %llu offset %llu num_refs %lu\n",
7892                                 (unsigned long long)bytenr,
7893                                 (unsigned long long)parent,
7894                                 (unsigned long long)root,
7895                                 (unsigned long long)owner,
7896                                 (unsigned long long)offset,
7897                                 (unsigned long)num_refs);
7898                 }
7899                 back->num_refs = num_refs;
7900                 back->node.found_extent_tree = 1;
7901         }
7902         if (insert)
7903                 WARN_ON(rb_insert(&rec->backref_tree, &back->node.node,
7904                         compare_extent_backref));
7905
7906         maybe_free_extent_rec(extent_cache, rec);
7907         return 0;
7908 }
7909
7910 static int add_pending(struct cache_tree *pending,
7911                        struct cache_tree *seen, u64 bytenr, u32 size)
7912 {
7913         int ret;
7914         ret = add_cache_extent(seen, bytenr, size);
7915         if (ret)
7916                 return ret;
7917         add_cache_extent(pending, bytenr, size);
7918         return 0;
7919 }
7920
7921 static int pick_next_pending(struct cache_tree *pending,
7922                         struct cache_tree *reada,
7923                         struct cache_tree *nodes,
7924                         u64 last, struct block_info *bits, int bits_nr,
7925                         int *reada_bits)
7926 {
7927         unsigned long node_start = last;
7928         struct cache_extent *cache;
7929         int ret;
7930
7931         cache = search_cache_extent(reada, 0);
7932         if (cache) {
7933                 bits[0].start = cache->start;
7934                 bits[0].size = cache->size;
7935                 *reada_bits = 1;
7936                 return 1;
7937         }
7938         *reada_bits = 0;
7939         if (node_start > 32768)
7940                 node_start -= 32768;
7941
7942         cache = search_cache_extent(nodes, node_start);
7943         if (!cache)
7944                 cache = search_cache_extent(nodes, 0);
7945
7946         if (!cache) {
7947                  cache = search_cache_extent(pending, 0);
7948                  if (!cache)
7949                          return 0;
7950                  ret = 0;
7951                  do {
7952                          bits[ret].start = cache->start;
7953                          bits[ret].size = cache->size;
7954                          cache = next_cache_extent(cache);
7955                          ret++;
7956                  } while (cache && ret < bits_nr);
7957                  return ret;
7958         }
7959
7960         ret = 0;
7961         do {
7962                 bits[ret].start = cache->start;
7963                 bits[ret].size = cache->size;
7964                 cache = next_cache_extent(cache);
7965                 ret++;
7966         } while (cache && ret < bits_nr);
7967
7968         if (bits_nr - ret > 8) {
7969                 u64 lookup = bits[0].start + bits[0].size;
7970                 struct cache_extent *next;
7971                 next = search_cache_extent(pending, lookup);
7972                 while(next) {
7973                         if (next->start - lookup > 32768)
7974                                 break;
7975                         bits[ret].start = next->start;
7976                         bits[ret].size = next->size;
7977                         lookup = next->start + next->size;
7978                         ret++;
7979                         if (ret == bits_nr)
7980                                 break;
7981                         next = next_cache_extent(next);
7982                         if (!next)
7983                                 break;
7984                 }
7985         }
7986         return ret;
7987 }
7988
7989 static void free_chunk_record(struct cache_extent *cache)
7990 {
7991         struct chunk_record *rec;
7992
7993         rec = container_of(cache, struct chunk_record, cache);
7994         list_del_init(&rec->list);
7995         list_del_init(&rec->dextents);
7996         free(rec);
7997 }
7998
7999 void free_chunk_cache_tree(struct cache_tree *chunk_cache)
8000 {
8001         cache_tree_free_extents(chunk_cache, free_chunk_record);
8002 }
8003
8004 static void free_device_record(struct rb_node *node)
8005 {
8006         struct device_record *rec;
8007
8008         rec = container_of(node, struct device_record, node);
8009         free(rec);
8010 }
8011
8012 FREE_RB_BASED_TREE(device_cache, free_device_record);
8013
8014 int insert_block_group_record(struct block_group_tree *tree,
8015                               struct block_group_record *bg_rec)
8016 {
8017         int ret;
8018
8019         ret = insert_cache_extent(&tree->tree, &bg_rec->cache);
8020         if (ret)
8021                 return ret;
8022
8023         list_add_tail(&bg_rec->list, &tree->block_groups);
8024         return 0;
8025 }
8026
8027 static void free_block_group_record(struct cache_extent *cache)
8028 {
8029         struct block_group_record *rec;
8030
8031         rec = container_of(cache, struct block_group_record, cache);
8032         list_del_init(&rec->list);
8033         free(rec);
8034 }
8035
8036 void free_block_group_tree(struct block_group_tree *tree)
8037 {
8038         cache_tree_free_extents(&tree->tree, free_block_group_record);
8039 }
8040
8041 int insert_device_extent_record(struct device_extent_tree *tree,
8042                                 struct device_extent_record *de_rec)
8043 {
8044         int ret;
8045
8046         /*
8047          * Device extent is a bit different from the other extents, because
8048          * the extents which belong to the different devices may have the
8049          * same start and size, so we need use the special extent cache
8050          * search/insert functions.
8051          */
8052         ret = insert_cache_extent2(&tree->tree, &de_rec->cache);
8053         if (ret)
8054                 return ret;
8055
8056         list_add_tail(&de_rec->chunk_list, &tree->no_chunk_orphans);
8057         list_add_tail(&de_rec->device_list, &tree->no_device_orphans);
8058         return 0;
8059 }
8060
8061 static void free_device_extent_record(struct cache_extent *cache)
8062 {
8063         struct device_extent_record *rec;
8064
8065         rec = container_of(cache, struct device_extent_record, cache);
8066         if (!list_empty(&rec->chunk_list))
8067                 list_del_init(&rec->chunk_list);
8068         if (!list_empty(&rec->device_list))
8069                 list_del_init(&rec->device_list);
8070         free(rec);
8071 }
8072
8073 void free_device_extent_tree(struct device_extent_tree *tree)
8074 {
8075         cache_tree_free_extents(&tree->tree, free_device_extent_record);
8076 }
8077
8078 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
8079 static int process_extent_ref_v0(struct cache_tree *extent_cache,
8080                                  struct extent_buffer *leaf, int slot)
8081 {
8082         struct btrfs_extent_ref_v0 *ref0;
8083         struct btrfs_key key;
8084         int ret;
8085
8086         btrfs_item_key_to_cpu(leaf, &key, slot);
8087         ref0 = btrfs_item_ptr(leaf, slot, struct btrfs_extent_ref_v0);
8088         if (btrfs_ref_objectid_v0(leaf, ref0) < BTRFS_FIRST_FREE_OBJECTID) {
8089                 ret = add_tree_backref(extent_cache, key.objectid, key.offset,
8090                                 0, 0);
8091         } else {
8092                 ret = add_data_backref(extent_cache, key.objectid, key.offset,
8093                                 0, 0, 0, btrfs_ref_count_v0(leaf, ref0), 0, 0);
8094         }
8095         return ret;
8096 }
8097 #endif
8098
8099 struct chunk_record *btrfs_new_chunk_record(struct extent_buffer *leaf,
8100                                             struct btrfs_key *key,
8101                                             int slot)
8102 {
8103         struct btrfs_chunk *ptr;
8104         struct chunk_record *rec;
8105         int num_stripes, i;
8106
8107         ptr = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
8108         num_stripes = btrfs_chunk_num_stripes(leaf, ptr);
8109
8110         rec = calloc(1, btrfs_chunk_record_size(num_stripes));
8111         if (!rec) {
8112                 fprintf(stderr, "memory allocation failed\n");
8113                 exit(-1);
8114         }
8115
8116         INIT_LIST_HEAD(&rec->list);
8117         INIT_LIST_HEAD(&rec->dextents);
8118         rec->bg_rec = NULL;
8119
8120         rec->cache.start = key->offset;
8121         rec->cache.size = btrfs_chunk_length(leaf, ptr);
8122
8123         rec->generation = btrfs_header_generation(leaf);
8124
8125         rec->objectid = key->objectid;
8126         rec->type = key->type;
8127         rec->offset = key->offset;
8128
8129         rec->length = rec->cache.size;
8130         rec->owner = btrfs_chunk_owner(leaf, ptr);
8131         rec->stripe_len = btrfs_chunk_stripe_len(leaf, ptr);
8132         rec->type_flags = btrfs_chunk_type(leaf, ptr);
8133         rec->io_width = btrfs_chunk_io_width(leaf, ptr);
8134         rec->io_align = btrfs_chunk_io_align(leaf, ptr);
8135         rec->sector_size = btrfs_chunk_sector_size(leaf, ptr);
8136         rec->num_stripes = num_stripes;
8137         rec->sub_stripes = btrfs_chunk_sub_stripes(leaf, ptr);
8138
8139         for (i = 0; i < rec->num_stripes; ++i) {
8140                 rec->stripes[i].devid =
8141                         btrfs_stripe_devid_nr(leaf, ptr, i);
8142                 rec->stripes[i].offset =
8143                         btrfs_stripe_offset_nr(leaf, ptr, i);
8144                 read_extent_buffer(leaf, rec->stripes[i].dev_uuid,
8145                                 (unsigned long)btrfs_stripe_dev_uuid_nr(ptr, i),
8146                                 BTRFS_UUID_SIZE);
8147         }
8148
8149         return rec;
8150 }
8151
8152 static int process_chunk_item(struct cache_tree *chunk_cache,
8153                               struct btrfs_key *key, struct extent_buffer *eb,
8154                               int slot)
8155 {
8156         struct chunk_record *rec;
8157         struct btrfs_chunk *chunk;
8158         int ret = 0;
8159
8160         chunk = btrfs_item_ptr(eb, slot, struct btrfs_chunk);
8161         /*
8162          * Do extra check for this chunk item,
8163          *
8164          * It's still possible one can craft a leaf with CHUNK_ITEM, with
8165          * wrong onwer(3) out of chunk tree, to pass both chunk tree check
8166          * and owner<->key_type check.
8167          */
8168         ret = btrfs_check_chunk_valid(global_info, eb, chunk, slot,
8169                                       key->offset);
8170         if (ret < 0) {
8171                 error("chunk(%llu, %llu) is not valid, ignore it",
8172                       key->offset, btrfs_chunk_length(eb, chunk));
8173                 return 0;
8174         }
8175         rec = btrfs_new_chunk_record(eb, key, slot);
8176         ret = insert_cache_extent(chunk_cache, &rec->cache);
8177         if (ret) {
8178                 fprintf(stderr, "Chunk[%llu, %llu] existed.\n",
8179                         rec->offset, rec->length);
8180                 free(rec);
8181         }
8182
8183         return ret;
8184 }
8185
8186 static int process_device_item(struct rb_root *dev_cache,
8187                 struct btrfs_key *key, struct extent_buffer *eb, int slot)
8188 {
8189         struct btrfs_dev_item *ptr;
8190         struct device_record *rec;
8191         int ret = 0;
8192
8193         ptr = btrfs_item_ptr(eb,
8194                 slot, struct btrfs_dev_item);
8195
8196         rec = malloc(sizeof(*rec));
8197         if (!rec) {
8198                 fprintf(stderr, "memory allocation failed\n");
8199                 return -ENOMEM;
8200         }
8201
8202         rec->devid = key->offset;
8203         rec->generation = btrfs_header_generation(eb);
8204
8205         rec->objectid = key->objectid;
8206         rec->type = key->type;
8207         rec->offset = key->offset;
8208
8209         rec->devid = btrfs_device_id(eb, ptr);
8210         rec->total_byte = btrfs_device_total_bytes(eb, ptr);
8211         rec->byte_used = btrfs_device_bytes_used(eb, ptr);
8212
8213         ret = rb_insert(dev_cache, &rec->node, device_record_compare);
8214         if (ret) {
8215                 fprintf(stderr, "Device[%llu] existed.\n", rec->devid);
8216                 free(rec);
8217         }
8218
8219         return ret;
8220 }
8221
8222 struct block_group_record *
8223 btrfs_new_block_group_record(struct extent_buffer *leaf, struct btrfs_key *key,
8224                              int slot)
8225 {
8226         struct btrfs_block_group_item *ptr;
8227         struct block_group_record *rec;
8228
8229         rec = calloc(1, sizeof(*rec));
8230         if (!rec) {
8231                 fprintf(stderr, "memory allocation failed\n");
8232                 exit(-1);
8233         }
8234
8235         rec->cache.start = key->objectid;
8236         rec->cache.size = key->offset;
8237
8238         rec->generation = btrfs_header_generation(leaf);
8239
8240         rec->objectid = key->objectid;
8241         rec->type = key->type;
8242         rec->offset = key->offset;
8243
8244         ptr = btrfs_item_ptr(leaf, slot, struct btrfs_block_group_item);
8245         rec->flags = btrfs_disk_block_group_flags(leaf, ptr);
8246
8247         INIT_LIST_HEAD(&rec->list);
8248
8249         return rec;
8250 }
8251
8252 static int process_block_group_item(struct block_group_tree *block_group_cache,
8253                                     struct btrfs_key *key,
8254                                     struct extent_buffer *eb, int slot)
8255 {
8256         struct block_group_record *rec;
8257         int ret = 0;
8258
8259         rec = btrfs_new_block_group_record(eb, key, slot);
8260         ret = insert_block_group_record(block_group_cache, rec);
8261         if (ret) {
8262                 fprintf(stderr, "Block Group[%llu, %llu] existed.\n",
8263                         rec->objectid, rec->offset);
8264                 free(rec);
8265         }
8266
8267         return ret;
8268 }
8269
8270 struct device_extent_record *
8271 btrfs_new_device_extent_record(struct extent_buffer *leaf,
8272                                struct btrfs_key *key, int slot)
8273 {
8274         struct device_extent_record *rec;
8275         struct btrfs_dev_extent *ptr;
8276
8277         rec = calloc(1, sizeof(*rec));
8278         if (!rec) {
8279                 fprintf(stderr, "memory allocation failed\n");
8280                 exit(-1);
8281         }
8282
8283         rec->cache.objectid = key->objectid;
8284         rec->cache.start = key->offset;
8285
8286         rec->generation = btrfs_header_generation(leaf);
8287
8288         rec->objectid = key->objectid;
8289         rec->type = key->type;
8290         rec->offset = key->offset;
8291
8292         ptr = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
8293         rec->chunk_objecteid =
8294                 btrfs_dev_extent_chunk_objectid(leaf, ptr);
8295         rec->chunk_offset =
8296                 btrfs_dev_extent_chunk_offset(leaf, ptr);
8297         rec->length = btrfs_dev_extent_length(leaf, ptr);
8298         rec->cache.size = rec->length;
8299
8300         INIT_LIST_HEAD(&rec->chunk_list);
8301         INIT_LIST_HEAD(&rec->device_list);
8302
8303         return rec;
8304 }
8305
8306 static int
8307 process_device_extent_item(struct device_extent_tree *dev_extent_cache,
8308                            struct btrfs_key *key, struct extent_buffer *eb,
8309                            int slot)
8310 {
8311         struct device_extent_record *rec;
8312         int ret;
8313
8314         rec = btrfs_new_device_extent_record(eb, key, slot);
8315         ret = insert_device_extent_record(dev_extent_cache, rec);
8316         if (ret) {
8317                 fprintf(stderr,
8318                         "Device extent[%llu, %llu, %llu] existed.\n",
8319                         rec->objectid, rec->offset, rec->length);
8320                 free(rec);
8321         }
8322
8323         return ret;
8324 }
8325
8326 static int process_extent_item(struct btrfs_root *root,
8327                                struct cache_tree *extent_cache,
8328                                struct extent_buffer *eb, int slot)
8329 {
8330         struct btrfs_extent_item *ei;
8331         struct btrfs_extent_inline_ref *iref;
8332         struct btrfs_extent_data_ref *dref;
8333         struct btrfs_shared_data_ref *sref;
8334         struct btrfs_key key;
8335         struct extent_record tmpl;
8336         unsigned long end;
8337         unsigned long ptr;
8338         int ret;
8339         int type;
8340         u32 item_size = btrfs_item_size_nr(eb, slot);
8341         u64 refs = 0;
8342         u64 offset;
8343         u64 num_bytes;
8344         int metadata = 0;
8345
8346         btrfs_item_key_to_cpu(eb, &key, slot);
8347
8348         if (key.type == BTRFS_METADATA_ITEM_KEY) {
8349                 metadata = 1;
8350                 num_bytes = root->fs_info->nodesize;
8351         } else {
8352                 num_bytes = key.offset;
8353         }
8354
8355         if (!IS_ALIGNED(key.objectid, root->fs_info->sectorsize)) {
8356                 error("ignoring invalid extent, bytenr %llu is not aligned to %u",
8357                       key.objectid, root->fs_info->sectorsize);
8358                 return -EIO;
8359         }
8360         if (item_size < sizeof(*ei)) {
8361 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
8362                 struct btrfs_extent_item_v0 *ei0;
8363                 BUG_ON(item_size != sizeof(*ei0));
8364                 ei0 = btrfs_item_ptr(eb, slot, struct btrfs_extent_item_v0);
8365                 refs = btrfs_extent_refs_v0(eb, ei0);
8366 #else
8367                 BUG();
8368 #endif
8369                 memset(&tmpl, 0, sizeof(tmpl));
8370                 tmpl.start = key.objectid;
8371                 tmpl.nr = num_bytes;
8372                 tmpl.extent_item_refs = refs;
8373                 tmpl.metadata = metadata;
8374                 tmpl.found_rec = 1;
8375                 tmpl.max_size = num_bytes;
8376
8377                 return add_extent_rec(extent_cache, &tmpl);
8378         }
8379
8380         ei = btrfs_item_ptr(eb, slot, struct btrfs_extent_item);
8381         refs = btrfs_extent_refs(eb, ei);
8382         if (btrfs_extent_flags(eb, ei) & BTRFS_EXTENT_FLAG_TREE_BLOCK)
8383                 metadata = 1;
8384         else
8385                 metadata = 0;
8386         if (metadata && num_bytes != root->fs_info->nodesize) {
8387                 error("ignore invalid metadata extent, length %llu does not equal to %u",
8388                       num_bytes, root->fs_info->nodesize);
8389                 return -EIO;
8390         }
8391         if (!metadata && !IS_ALIGNED(num_bytes, root->fs_info->sectorsize)) {
8392                 error("ignore invalid data extent, length %llu is not aligned to %u",
8393                       num_bytes, root->fs_info->sectorsize);
8394                 return -EIO;
8395         }
8396
8397         memset(&tmpl, 0, sizeof(tmpl));
8398         tmpl.start = key.objectid;
8399         tmpl.nr = num_bytes;
8400         tmpl.extent_item_refs = refs;
8401         tmpl.metadata = metadata;
8402         tmpl.found_rec = 1;
8403         tmpl.max_size = num_bytes;
8404         add_extent_rec(extent_cache, &tmpl);
8405
8406         ptr = (unsigned long)(ei + 1);
8407         if (btrfs_extent_flags(eb, ei) & BTRFS_EXTENT_FLAG_TREE_BLOCK &&
8408             key.type == BTRFS_EXTENT_ITEM_KEY)
8409                 ptr += sizeof(struct btrfs_tree_block_info);
8410
8411         end = (unsigned long)ei + item_size;
8412         while (ptr < end) {
8413                 iref = (struct btrfs_extent_inline_ref *)ptr;
8414                 type = btrfs_extent_inline_ref_type(eb, iref);
8415                 offset = btrfs_extent_inline_ref_offset(eb, iref);
8416                 switch (type) {
8417                 case BTRFS_TREE_BLOCK_REF_KEY:
8418                         ret = add_tree_backref(extent_cache, key.objectid,
8419                                         0, offset, 0);
8420                         if (ret < 0)
8421                                 error(
8422                         "add_tree_backref failed (extent items tree block): %s",
8423                                       strerror(-ret));
8424                         break;
8425                 case BTRFS_SHARED_BLOCK_REF_KEY:
8426                         ret = add_tree_backref(extent_cache, key.objectid,
8427                                         offset, 0, 0);
8428                         if (ret < 0)
8429                                 error(
8430                         "add_tree_backref failed (extent items shared block): %s",
8431                                       strerror(-ret));
8432                         break;
8433                 case BTRFS_EXTENT_DATA_REF_KEY:
8434                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
8435                         add_data_backref(extent_cache, key.objectid, 0,
8436                                         btrfs_extent_data_ref_root(eb, dref),
8437                                         btrfs_extent_data_ref_objectid(eb,
8438                                                                        dref),
8439                                         btrfs_extent_data_ref_offset(eb, dref),
8440                                         btrfs_extent_data_ref_count(eb, dref),
8441                                         0, num_bytes);
8442                         break;
8443                 case BTRFS_SHARED_DATA_REF_KEY:
8444                         sref = (struct btrfs_shared_data_ref *)(iref + 1);
8445                         add_data_backref(extent_cache, key.objectid, offset,
8446                                         0, 0, 0,
8447                                         btrfs_shared_data_ref_count(eb, sref),
8448                                         0, num_bytes);
8449                         break;
8450                 default:
8451                         fprintf(stderr, "corrupt extent record: key %Lu %u %Lu\n",
8452                                 key.objectid, key.type, num_bytes);
8453                         goto out;
8454                 }
8455                 ptr += btrfs_extent_inline_ref_size(type);
8456         }
8457         WARN_ON(ptr > end);
8458 out:
8459         return 0;
8460 }
8461
8462 static int check_cache_range(struct btrfs_root *root,
8463                              struct btrfs_block_group_cache *cache,
8464                              u64 offset, u64 bytes)
8465 {
8466         struct btrfs_free_space *entry;
8467         u64 *logical;
8468         u64 bytenr;
8469         int stripe_len;
8470         int i, nr, ret;
8471
8472         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
8473                 bytenr = btrfs_sb_offset(i);
8474                 ret = btrfs_rmap_block(root->fs_info,
8475                                        cache->key.objectid, bytenr, 0,
8476                                        &logical, &nr, &stripe_len);
8477                 if (ret)
8478                         return ret;
8479
8480                 while (nr--) {
8481                         if (logical[nr] + stripe_len <= offset)
8482                                 continue;
8483                         if (offset + bytes <= logical[nr])
8484                                 continue;
8485                         if (logical[nr] == offset) {
8486                                 if (stripe_len >= bytes) {
8487                                         free(logical);
8488                                         return 0;
8489                                 }
8490                                 bytes -= stripe_len;
8491                                 offset += stripe_len;
8492                         } else if (logical[nr] < offset) {
8493                                 if (logical[nr] + stripe_len >=
8494                                     offset + bytes) {
8495                                         free(logical);
8496                                         return 0;
8497                                 }
8498                                 bytes = (offset + bytes) -
8499                                         (logical[nr] + stripe_len);
8500                                 offset = logical[nr] + stripe_len;
8501                         } else {
8502                                 /*
8503                                  * Could be tricky, the super may land in the
8504                                  * middle of the area we're checking.  First
8505                                  * check the easiest case, it's at the end.
8506                                  */
8507                                 if (logical[nr] + stripe_len >=
8508                                     bytes + offset) {
8509                                         bytes = logical[nr] - offset;
8510                                         continue;
8511                                 }
8512
8513                                 /* Check the left side */
8514                                 ret = check_cache_range(root, cache,
8515                                                         offset,
8516                                                         logical[nr] - offset);
8517                                 if (ret) {
8518                                         free(logical);
8519                                         return ret;
8520                                 }
8521
8522                                 /* Now we continue with the right side */
8523                                 bytes = (offset + bytes) -
8524                                         (logical[nr] + stripe_len);
8525                                 offset = logical[nr] + stripe_len;
8526                         }
8527                 }
8528
8529                 free(logical);
8530         }
8531
8532         entry = btrfs_find_free_space(cache->free_space_ctl, offset, bytes);
8533         if (!entry) {
8534                 fprintf(stderr, "There is no free space entry for %Lu-%Lu\n",
8535                         offset, offset+bytes);
8536                 return -EINVAL;
8537         }
8538
8539         if (entry->offset != offset) {
8540                 fprintf(stderr, "Wanted offset %Lu, found %Lu\n", offset,
8541                         entry->offset);
8542                 return -EINVAL;
8543         }
8544
8545         if (entry->bytes != bytes) {
8546                 fprintf(stderr, "Wanted bytes %Lu, found %Lu for off %Lu\n",
8547                         bytes, entry->bytes, offset);
8548                 return -EINVAL;
8549         }
8550
8551         unlink_free_space(cache->free_space_ctl, entry);
8552         free(entry);
8553         return 0;
8554 }
8555
8556 static int verify_space_cache(struct btrfs_root *root,
8557                               struct btrfs_block_group_cache *cache)
8558 {
8559         struct btrfs_path path;
8560         struct extent_buffer *leaf;
8561         struct btrfs_key key;
8562         u64 last;
8563         int ret = 0;
8564
8565         root = root->fs_info->extent_root;
8566
8567         last = max_t(u64, cache->key.objectid, BTRFS_SUPER_INFO_OFFSET);
8568
8569         btrfs_init_path(&path);
8570         key.objectid = last;
8571         key.offset = 0;
8572         key.type = BTRFS_EXTENT_ITEM_KEY;
8573         ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
8574         if (ret < 0)
8575                 goto out;
8576         ret = 0;
8577         while (1) {
8578                 if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
8579                         ret = btrfs_next_leaf(root, &path);
8580                         if (ret < 0)
8581                                 goto out;
8582                         if (ret > 0) {
8583                                 ret = 0;
8584                                 break;
8585                         }
8586                 }
8587                 leaf = path.nodes[0];
8588                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
8589                 if (key.objectid >= cache->key.offset + cache->key.objectid)
8590                         break;
8591                 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
8592                     key.type != BTRFS_METADATA_ITEM_KEY) {
8593                         path.slots[0]++;
8594                         continue;
8595                 }
8596
8597                 if (last == key.objectid) {
8598                         if (key.type == BTRFS_EXTENT_ITEM_KEY)
8599                                 last = key.objectid + key.offset;
8600                         else
8601                                 last = key.objectid + root->fs_info->nodesize;
8602                         path.slots[0]++;
8603                         continue;
8604                 }
8605
8606                 ret = check_cache_range(root, cache, last,
8607                                         key.objectid - last);
8608                 if (ret)
8609                         break;
8610                 if (key.type == BTRFS_EXTENT_ITEM_KEY)
8611                         last = key.objectid + key.offset;
8612                 else
8613                         last = key.objectid + root->fs_info->nodesize;
8614                 path.slots[0]++;
8615         }
8616
8617         if (last < cache->key.objectid + cache->key.offset)
8618                 ret = check_cache_range(root, cache, last,
8619                                         cache->key.objectid +
8620                                         cache->key.offset - last);
8621
8622 out:
8623         btrfs_release_path(&path);
8624
8625         if (!ret &&
8626             !RB_EMPTY_ROOT(&cache->free_space_ctl->free_space_offset)) {
8627                 fprintf(stderr, "There are still entries left in the space "
8628                         "cache\n");
8629                 ret = -EINVAL;
8630         }
8631
8632         return ret;
8633 }
8634
8635 static int check_space_cache(struct btrfs_root *root)
8636 {
8637         struct btrfs_block_group_cache *cache;
8638         u64 start = BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE;
8639         int ret;
8640         int error = 0;
8641
8642         if (btrfs_super_cache_generation(root->fs_info->super_copy) != -1ULL &&
8643             btrfs_super_generation(root->fs_info->super_copy) !=
8644             btrfs_super_cache_generation(root->fs_info->super_copy)) {
8645                 printf("cache and super generation don't match, space cache "
8646                        "will be invalidated\n");
8647                 return 0;
8648         }
8649
8650         if (ctx.progress_enabled) {
8651                 ctx.tp = TASK_FREE_SPACE;
8652                 task_start(ctx.info);
8653         }
8654
8655         while (1) {
8656                 cache = btrfs_lookup_first_block_group(root->fs_info, start);
8657                 if (!cache)
8658                         break;
8659
8660                 start = cache->key.objectid + cache->key.offset;
8661                 if (!cache->free_space_ctl) {
8662                         if (btrfs_init_free_space_ctl(cache,
8663                                                 root->fs_info->sectorsize)) {
8664                                 ret = -ENOMEM;
8665                                 break;
8666                         }
8667                 } else {
8668                         btrfs_remove_free_space_cache(cache);
8669                 }
8670
8671                 if (btrfs_fs_compat_ro(root->fs_info, FREE_SPACE_TREE)) {
8672                         ret = exclude_super_stripes(root, cache);
8673                         if (ret) {
8674                                 fprintf(stderr, "could not exclude super stripes: %s\n",
8675                                         strerror(-ret));
8676                                 error++;
8677                                 continue;
8678                         }
8679                         ret = load_free_space_tree(root->fs_info, cache);
8680                         free_excluded_extents(root, cache);
8681                         if (ret < 0) {
8682                                 fprintf(stderr, "could not load free space tree: %s\n",
8683                                         strerror(-ret));
8684                                 error++;
8685                                 continue;
8686                         }
8687                         error += ret;
8688                 } else {
8689                         ret = load_free_space_cache(root->fs_info, cache);
8690                         if (!ret)
8691                                 continue;
8692                 }
8693
8694                 ret = verify_space_cache(root, cache);
8695                 if (ret) {
8696                         fprintf(stderr, "cache appears valid but isn't %Lu\n",
8697                                 cache->key.objectid);
8698                         error++;
8699                 }
8700         }
8701
8702         task_stop(ctx.info);
8703
8704         return error ? -EINVAL : 0;
8705 }
8706
8707 static int check_extent_csums(struct btrfs_root *root, u64 bytenr,
8708                         u64 num_bytes, unsigned long leaf_offset,
8709                         struct extent_buffer *eb) {
8710
8711         struct btrfs_fs_info *fs_info = root->fs_info;
8712         u64 offset = 0;
8713         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
8714         char *data;
8715         unsigned long csum_offset;
8716         u32 csum;
8717         u32 csum_expected;
8718         u64 read_len;
8719         u64 data_checked = 0;
8720         u64 tmp;
8721         int ret = 0;
8722         int mirror;
8723         int num_copies;
8724
8725         if (num_bytes % fs_info->sectorsize)
8726                 return -EINVAL;
8727
8728         data = malloc(num_bytes);
8729         if (!data)
8730                 return -ENOMEM;
8731
8732         while (offset < num_bytes) {
8733                 mirror = 0;
8734 again:
8735                 read_len = num_bytes - offset;
8736                 /* read as much space once a time */
8737                 ret = read_extent_data(fs_info, data + offset,
8738                                 bytenr + offset, &read_len, mirror);
8739                 if (ret)
8740                         goto out;
8741                 data_checked = 0;
8742                 /* verify every 4k data's checksum */
8743                 while (data_checked < read_len) {
8744                         csum = ~(u32)0;
8745                         tmp = offset + data_checked;
8746
8747                         csum = btrfs_csum_data((char *)data + tmp,
8748                                                csum, fs_info->sectorsize);
8749                         btrfs_csum_final(csum, (u8 *)&csum);
8750
8751                         csum_offset = leaf_offset +
8752                                  tmp / fs_info->sectorsize * csum_size;
8753                         read_extent_buffer(eb, (char *)&csum_expected,
8754                                            csum_offset, csum_size);
8755                         /* try another mirror */
8756                         if (csum != csum_expected) {
8757                                 fprintf(stderr, "mirror %d bytenr %llu csum %u expected csum %u\n",
8758                                                 mirror, bytenr + tmp,
8759                                                 csum, csum_expected);
8760                                 num_copies = btrfs_num_copies(root->fs_info,
8761                                                 bytenr, num_bytes);
8762                                 if (mirror < num_copies - 1) {
8763                                         mirror += 1;
8764                                         goto again;
8765                                 }
8766                         }
8767                         data_checked += fs_info->sectorsize;
8768                 }
8769                 offset += read_len;
8770         }
8771 out:
8772         free(data);
8773         return ret;
8774 }
8775
8776 static int check_extent_exists(struct btrfs_root *root, u64 bytenr,
8777                                u64 num_bytes)
8778 {
8779         struct btrfs_path path;
8780         struct extent_buffer *leaf;
8781         struct btrfs_key key;
8782         int ret;
8783
8784         btrfs_init_path(&path);
8785         key.objectid = bytenr;
8786         key.type = BTRFS_EXTENT_ITEM_KEY;
8787         key.offset = (u64)-1;
8788
8789 again:
8790         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, &path,
8791                                 0, 0);
8792         if (ret < 0) {
8793                 fprintf(stderr, "Error looking up extent record %d\n", ret);
8794                 btrfs_release_path(&path);
8795                 return ret;
8796         } else if (ret) {
8797                 if (path.slots[0] > 0) {
8798                         path.slots[0]--;
8799                 } else {
8800                         ret = btrfs_prev_leaf(root, &path);
8801                         if (ret < 0) {
8802                                 goto out;
8803                         } else if (ret > 0) {
8804                                 ret = 0;
8805                                 goto out;
8806                         }
8807                 }
8808         }
8809
8810         btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
8811
8812         /*
8813          * Block group items come before extent items if they have the same
8814          * bytenr, so walk back one more just in case.  Dear future traveller,
8815          * first congrats on mastering time travel.  Now if it's not too much
8816          * trouble could you go back to 2006 and tell Chris to make the
8817          * BLOCK_GROUP_ITEM_KEY (and BTRFS_*_REF_KEY) lower than the
8818          * EXTENT_ITEM_KEY please?
8819          */
8820         while (key.type > BTRFS_EXTENT_ITEM_KEY) {
8821                 if (path.slots[0] > 0) {
8822                         path.slots[0]--;
8823                 } else {
8824                         ret = btrfs_prev_leaf(root, &path);
8825                         if (ret < 0) {
8826                                 goto out;
8827                         } else if (ret > 0) {
8828                                 ret = 0;
8829                                 goto out;
8830                         }
8831                 }
8832                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
8833         }
8834
8835         while (num_bytes) {
8836                 if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
8837                         ret = btrfs_next_leaf(root, &path);
8838                         if (ret < 0) {
8839                                 fprintf(stderr, "Error going to next leaf "
8840                                         "%d\n", ret);
8841                                 btrfs_release_path(&path);
8842                                 return ret;
8843                         } else if (ret) {
8844                                 break;
8845                         }
8846                 }
8847                 leaf = path.nodes[0];
8848                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
8849                 if (key.type != BTRFS_EXTENT_ITEM_KEY) {
8850                         path.slots[0]++;
8851                         continue;
8852                 }
8853                 if (key.objectid + key.offset < bytenr) {
8854                         path.slots[0]++;
8855                         continue;
8856                 }
8857                 if (key.objectid > bytenr + num_bytes)
8858                         break;
8859
8860                 if (key.objectid == bytenr) {
8861                         if (key.offset >= num_bytes) {
8862                                 num_bytes = 0;
8863                                 break;
8864                         }
8865                         num_bytes -= key.offset;
8866                         bytenr += key.offset;
8867                 } else if (key.objectid < bytenr) {
8868                         if (key.objectid + key.offset >= bytenr + num_bytes) {
8869                                 num_bytes = 0;
8870                                 break;
8871                         }
8872                         num_bytes = (bytenr + num_bytes) -
8873                                 (key.objectid + key.offset);
8874                         bytenr = key.objectid + key.offset;
8875                 } else {
8876                         if (key.objectid + key.offset < bytenr + num_bytes) {
8877                                 u64 new_start = key.objectid + key.offset;
8878                                 u64 new_bytes = bytenr + num_bytes - new_start;
8879
8880                                 /*
8881                                  * Weird case, the extent is in the middle of
8882                                  * our range, we'll have to search one side
8883                                  * and then the other.  Not sure if this happens
8884                                  * in real life, but no harm in coding it up
8885                                  * anyway just in case.
8886                                  */
8887                                 btrfs_release_path(&path);
8888                                 ret = check_extent_exists(root, new_start,
8889                                                           new_bytes);
8890                                 if (ret) {
8891                                         fprintf(stderr, "Right section didn't "
8892                                                 "have a record\n");
8893                                         break;
8894                                 }
8895                                 num_bytes = key.objectid - bytenr;
8896                                 goto again;
8897                         }
8898                         num_bytes = key.objectid - bytenr;
8899                 }
8900                 path.slots[0]++;
8901         }
8902         ret = 0;
8903
8904 out:
8905         if (num_bytes && !ret) {
8906                 fprintf(stderr, "There are no extents for csum range "
8907                         "%Lu-%Lu\n", bytenr, bytenr+num_bytes);
8908                 ret = 1;
8909         }
8910
8911         btrfs_release_path(&path);
8912         return ret;
8913 }
8914
8915 static int check_csums(struct btrfs_root *root)
8916 {
8917         struct btrfs_path path;
8918         struct extent_buffer *leaf;
8919         struct btrfs_key key;
8920         u64 offset = 0, num_bytes = 0;
8921         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
8922         int errors = 0;
8923         int ret;
8924         u64 data_len;
8925         unsigned long leaf_offset;
8926
8927         root = root->fs_info->csum_root;
8928         if (!extent_buffer_uptodate(root->node)) {
8929                 fprintf(stderr, "No valid csum tree found\n");
8930                 return -ENOENT;
8931         }
8932
8933         btrfs_init_path(&path);
8934         key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
8935         key.type = BTRFS_EXTENT_CSUM_KEY;
8936         key.offset = 0;
8937         ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
8938         if (ret < 0) {
8939                 fprintf(stderr, "Error searching csum tree %d\n", ret);
8940                 btrfs_release_path(&path);
8941                 return ret;
8942         }
8943
8944         if (ret > 0 && path.slots[0])
8945                 path.slots[0]--;
8946         ret = 0;
8947
8948         while (1) {
8949                 if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
8950                         ret = btrfs_next_leaf(root, &path);
8951                         if (ret < 0) {
8952                                 fprintf(stderr, "Error going to next leaf "
8953                                         "%d\n", ret);
8954                                 break;
8955                         }
8956                         if (ret)
8957                                 break;
8958                 }
8959                 leaf = path.nodes[0];
8960
8961                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
8962                 if (key.type != BTRFS_EXTENT_CSUM_KEY) {
8963                         path.slots[0]++;
8964                         continue;
8965                 }
8966
8967                 data_len = (btrfs_item_size_nr(leaf, path.slots[0]) /
8968                               csum_size) * root->fs_info->sectorsize;
8969                 if (!check_data_csum)
8970                         goto skip_csum_check;
8971                 leaf_offset = btrfs_item_ptr_offset(leaf, path.slots[0]);
8972                 ret = check_extent_csums(root, key.offset, data_len,
8973                                          leaf_offset, leaf);
8974                 if (ret)
8975                         break;
8976 skip_csum_check:
8977                 if (!num_bytes) {
8978                         offset = key.offset;
8979                 } else if (key.offset != offset + num_bytes) {
8980                         ret = check_extent_exists(root, offset, num_bytes);
8981                         if (ret) {
8982                                 fprintf(stderr, "Csum exists for %Lu-%Lu but "
8983                                         "there is no extent record\n",
8984                                         offset, offset+num_bytes);
8985                                 errors++;
8986                         }
8987                         offset = key.offset;
8988                         num_bytes = 0;
8989                 }
8990                 num_bytes += data_len;
8991                 path.slots[0]++;
8992         }
8993
8994         btrfs_release_path(&path);
8995         return errors;
8996 }
8997
8998 static int is_dropped_key(struct btrfs_key *key,
8999                           struct btrfs_key *drop_key) {
9000         if (key->objectid < drop_key->objectid)
9001                 return 1;
9002         else if (key->objectid == drop_key->objectid) {
9003                 if (key->type < drop_key->type)
9004                         return 1;
9005                 else if (key->type == drop_key->type) {
9006                         if (key->offset < drop_key->offset)
9007                                 return 1;
9008                 }
9009         }
9010         return 0;
9011 }
9012
9013 /*
9014  * Here are the rules for FULL_BACKREF.
9015  *
9016  * 1) If BTRFS_HEADER_FLAG_RELOC is set then we have FULL_BACKREF set.
9017  * 2) If btrfs_header_owner(buf) no longer points to buf then we have
9018  *      FULL_BACKREF set.
9019  * 3) We cowed the block walking down a reloc tree.  This is impossible to tell
9020  *    if it happened after the relocation occurred since we'll have dropped the
9021  *    reloc root, so it's entirely possible to have FULL_BACKREF set on buf and
9022  *    have no real way to know for sure.
9023  *
9024  * We process the blocks one root at a time, and we start from the lowest root
9025  * objectid and go to the highest.  So we can just lookup the owner backref for
9026  * the record and if we don't find it then we know it doesn't exist and we have
9027  * a FULL BACKREF.
9028  *
9029  * FIXME: if we ever start reclaiming root objectid's then we need to fix this
9030  * assumption and simply indicate that we _think_ that the FULL BACKREF needs to
9031  * be set or not and then we can check later once we've gathered all the refs.
9032  */
9033 static int calc_extent_flag(struct cache_tree *extent_cache,
9034                            struct extent_buffer *buf,
9035                            struct root_item_record *ri,
9036                            u64 *flags)
9037 {
9038         struct extent_record *rec;
9039         struct cache_extent *cache;
9040         struct tree_backref *tback;
9041         u64 owner = 0;
9042
9043         cache = lookup_cache_extent(extent_cache, buf->start, 1);
9044         /* we have added this extent before */
9045         if (!cache)
9046                 return -ENOENT;
9047
9048         rec = container_of(cache, struct extent_record, cache);
9049
9050         /*
9051          * Except file/reloc tree, we can not have
9052          * FULL BACKREF MODE
9053          */
9054         if (ri->objectid < BTRFS_FIRST_FREE_OBJECTID)
9055                 goto normal;
9056         /*
9057          * root node
9058          */
9059         if (buf->start == ri->bytenr)
9060                 goto normal;
9061
9062         if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))
9063                 goto full_backref;
9064
9065         owner = btrfs_header_owner(buf);
9066         if (owner == ri->objectid)
9067                 goto normal;
9068
9069         tback = find_tree_backref(rec, 0, owner);
9070         if (!tback)
9071                 goto full_backref;
9072 normal:
9073         *flags = 0;
9074         if (rec->flag_block_full_backref != FLAG_UNSET &&
9075             rec->flag_block_full_backref != 0)
9076                 rec->bad_full_backref = 1;
9077         return 0;
9078 full_backref:
9079         *flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
9080         if (rec->flag_block_full_backref != FLAG_UNSET &&
9081             rec->flag_block_full_backref != 1)
9082                 rec->bad_full_backref = 1;
9083         return 0;
9084 }
9085
9086 static void report_mismatch_key_root(u8 key_type, u64 rootid)
9087 {
9088         fprintf(stderr, "Invalid key type(");
9089         print_key_type(stderr, 0, key_type);
9090         fprintf(stderr, ") found in root(");
9091         print_objectid(stderr, rootid, 0);
9092         fprintf(stderr, ")\n");
9093 }
9094
9095 /*
9096  * Check if the key is valid with its extent buffer.
9097  *
9098  * This is a early check in case invalid key exists in a extent buffer
9099  * This is not comprehensive yet, but should prevent wrong key/item passed
9100  * further
9101  */
9102 static int check_type_with_root(u64 rootid, u8 key_type)
9103 {
9104         switch (key_type) {
9105         /* Only valid in chunk tree */
9106         case BTRFS_DEV_ITEM_KEY:
9107         case BTRFS_CHUNK_ITEM_KEY:
9108                 if (rootid != BTRFS_CHUNK_TREE_OBJECTID)
9109                         goto err;
9110                 break;
9111         /* valid in csum and log tree */
9112         case BTRFS_CSUM_TREE_OBJECTID:
9113                 if (!(rootid == BTRFS_TREE_LOG_OBJECTID ||
9114                       is_fstree(rootid)))
9115                         goto err;
9116                 break;
9117         case BTRFS_EXTENT_ITEM_KEY:
9118         case BTRFS_METADATA_ITEM_KEY:
9119         case BTRFS_BLOCK_GROUP_ITEM_KEY:
9120                 if (rootid != BTRFS_EXTENT_TREE_OBJECTID)
9121                         goto err;
9122                 break;
9123         case BTRFS_ROOT_ITEM_KEY:
9124                 if (rootid != BTRFS_ROOT_TREE_OBJECTID)
9125                         goto err;
9126                 break;
9127         case BTRFS_DEV_EXTENT_KEY:
9128                 if (rootid != BTRFS_DEV_TREE_OBJECTID)
9129                         goto err;
9130                 break;
9131         }
9132         return 0;
9133 err:
9134         report_mismatch_key_root(key_type, rootid);
9135         return -EINVAL;
9136 }
9137
9138 static int run_next_block(struct btrfs_root *root,
9139                           struct block_info *bits,
9140                           int bits_nr,
9141                           u64 *last,
9142                           struct cache_tree *pending,
9143                           struct cache_tree *seen,
9144                           struct cache_tree *reada,
9145                           struct cache_tree *nodes,
9146                           struct cache_tree *extent_cache,
9147                           struct cache_tree *chunk_cache,
9148                           struct rb_root *dev_cache,
9149                           struct block_group_tree *block_group_cache,
9150                           struct device_extent_tree *dev_extent_cache,
9151                           struct root_item_record *ri)
9152 {
9153         struct btrfs_fs_info *fs_info = root->fs_info;
9154         struct extent_buffer *buf;
9155         struct extent_record *rec = NULL;
9156         u64 bytenr;
9157         u32 size;
9158         u64 parent;
9159         u64 owner;
9160         u64 flags;
9161         u64 ptr;
9162         u64 gen = 0;
9163         int ret = 0;
9164         int i;
9165         int nritems;
9166         struct btrfs_key key;
9167         struct cache_extent *cache;
9168         int reada_bits;
9169
9170         nritems = pick_next_pending(pending, reada, nodes, *last, bits,
9171                                     bits_nr, &reada_bits);
9172         if (nritems == 0)
9173                 return 1;
9174
9175         if (!reada_bits) {
9176                 for(i = 0; i < nritems; i++) {
9177                         ret = add_cache_extent(reada, bits[i].start,
9178                                                bits[i].size);
9179                         if (ret == -EEXIST)
9180                                 continue;
9181
9182                         /* fixme, get the parent transid */
9183                         readahead_tree_block(fs_info, bits[i].start, 0);
9184                 }
9185         }
9186         *last = bits[0].start;
9187         bytenr = bits[0].start;
9188         size = bits[0].size;
9189
9190         cache = lookup_cache_extent(pending, bytenr, size);
9191         if (cache) {
9192                 remove_cache_extent(pending, cache);
9193                 free(cache);
9194         }
9195         cache = lookup_cache_extent(reada, bytenr, size);
9196         if (cache) {
9197                 remove_cache_extent(reada, cache);
9198                 free(cache);
9199         }
9200         cache = lookup_cache_extent(nodes, bytenr, size);
9201         if (cache) {
9202                 remove_cache_extent(nodes, cache);
9203                 free(cache);
9204         }
9205         cache = lookup_cache_extent(extent_cache, bytenr, size);
9206         if (cache) {
9207                 rec = container_of(cache, struct extent_record, cache);
9208                 gen = rec->parent_generation;
9209         }
9210
9211         /* fixme, get the real parent transid */
9212         buf = read_tree_block(root->fs_info, bytenr, gen);
9213         if (!extent_buffer_uptodate(buf)) {
9214                 record_bad_block_io(root->fs_info,
9215                                     extent_cache, bytenr, size);
9216                 goto out;
9217         }
9218
9219         nritems = btrfs_header_nritems(buf);
9220
9221         flags = 0;
9222         if (!init_extent_tree) {
9223                 ret = btrfs_lookup_extent_info(NULL, root, bytenr,
9224                                        btrfs_header_level(buf), 1, NULL,
9225                                        &flags);
9226                 if (ret < 0) {
9227                         ret = calc_extent_flag(extent_cache, buf, ri, &flags);
9228                         if (ret < 0) {
9229                                 fprintf(stderr, "Couldn't calc extent flags\n");
9230                                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
9231                         }
9232                 }
9233         } else {
9234                 flags = 0;
9235                 ret = calc_extent_flag(extent_cache, buf, ri, &flags);
9236                 if (ret < 0) {
9237                         fprintf(stderr, "Couldn't calc extent flags\n");
9238                         flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
9239                 }
9240         }
9241
9242         if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
9243                 if (ri != NULL &&
9244                     ri->objectid != BTRFS_TREE_RELOC_OBJECTID &&
9245                     ri->objectid == btrfs_header_owner(buf)) {
9246                         /*
9247                          * Ok we got to this block from it's original owner and
9248                          * we have FULL_BACKREF set.  Relocation can leave
9249                          * converted blocks over so this is altogether possible,
9250                          * however it's not possible if the generation > the
9251                          * last snapshot, so check for this case.
9252                          */
9253                         if (!btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC) &&
9254                             btrfs_header_generation(buf) > ri->last_snapshot) {
9255                                 flags &= ~BTRFS_BLOCK_FLAG_FULL_BACKREF;
9256                                 rec->bad_full_backref = 1;
9257                         }
9258                 }
9259         } else {
9260                 if (ri != NULL &&
9261                     (ri->objectid == BTRFS_TREE_RELOC_OBJECTID ||
9262                      btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) {
9263                         flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
9264                         rec->bad_full_backref = 1;
9265                 }
9266         }
9267
9268         if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
9269                 rec->flag_block_full_backref = 1;
9270                 parent = bytenr;
9271                 owner = 0;
9272         } else {
9273                 rec->flag_block_full_backref = 0;
9274                 parent = 0;
9275                 owner = btrfs_header_owner(buf);
9276         }
9277
9278         ret = check_block(root, extent_cache, buf, flags);
9279         if (ret)
9280                 goto out;
9281
9282         if (btrfs_is_leaf(buf)) {
9283                 btree_space_waste += btrfs_leaf_free_space(root, buf);
9284                 for (i = 0; i < nritems; i++) {
9285                         struct btrfs_file_extent_item *fi;
9286                         btrfs_item_key_to_cpu(buf, &key, i);
9287                         /*
9288                          * Check key type against the leaf owner.
9289                          * Could filter quite a lot of early error if
9290                          * owner is correct
9291                          */
9292                         if (check_type_with_root(btrfs_header_owner(buf),
9293                                                  key.type)) {
9294                                 fprintf(stderr, "ignoring invalid key\n");
9295                                 continue;
9296                         }
9297                         if (key.type == BTRFS_EXTENT_ITEM_KEY) {
9298                                 process_extent_item(root, extent_cache, buf,
9299                                                     i);
9300                                 continue;
9301                         }
9302                         if (key.type == BTRFS_METADATA_ITEM_KEY) {
9303                                 process_extent_item(root, extent_cache, buf,
9304                                                     i);
9305                                 continue;
9306                         }
9307                         if (key.type == BTRFS_EXTENT_CSUM_KEY) {
9308                                 total_csum_bytes +=
9309                                         btrfs_item_size_nr(buf, i);
9310                                 continue;
9311                         }
9312                         if (key.type == BTRFS_CHUNK_ITEM_KEY) {
9313                                 process_chunk_item(chunk_cache, &key, buf, i);
9314                                 continue;
9315                         }
9316                         if (key.type == BTRFS_DEV_ITEM_KEY) {
9317                                 process_device_item(dev_cache, &key, buf, i);
9318                                 continue;
9319                         }
9320                         if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9321                                 process_block_group_item(block_group_cache,
9322                                         &key, buf, i);
9323                                 continue;
9324                         }
9325                         if (key.type == BTRFS_DEV_EXTENT_KEY) {
9326                                 process_device_extent_item(dev_extent_cache,
9327                                         &key, buf, i);
9328                                 continue;
9329
9330                         }
9331                         if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
9332 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
9333                                 process_extent_ref_v0(extent_cache, buf, i);
9334 #else
9335                                 BUG();
9336 #endif
9337                                 continue;
9338                         }
9339
9340                         if (key.type == BTRFS_TREE_BLOCK_REF_KEY) {
9341                                 ret = add_tree_backref(extent_cache,
9342                                                 key.objectid, 0, key.offset, 0);
9343                                 if (ret < 0)
9344                                         error(
9345                                 "add_tree_backref failed (leaf tree block): %s",
9346                                               strerror(-ret));
9347                                 continue;
9348                         }
9349                         if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
9350                                 ret = add_tree_backref(extent_cache,
9351                                                 key.objectid, key.offset, 0, 0);
9352                                 if (ret < 0)
9353                                         error(
9354                                 "add_tree_backref failed (leaf shared block): %s",
9355                                               strerror(-ret));
9356                                 continue;
9357                         }
9358                         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
9359                                 struct btrfs_extent_data_ref *ref;
9360                                 ref = btrfs_item_ptr(buf, i,
9361                                                 struct btrfs_extent_data_ref);
9362                                 add_data_backref(extent_cache,
9363                                         key.objectid, 0,
9364                                         btrfs_extent_data_ref_root(buf, ref),
9365                                         btrfs_extent_data_ref_objectid(buf,
9366                                                                        ref),
9367                                         btrfs_extent_data_ref_offset(buf, ref),
9368                                         btrfs_extent_data_ref_count(buf, ref),
9369                                         0, root->fs_info->sectorsize);
9370                                 continue;
9371                         }
9372                         if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
9373                                 struct btrfs_shared_data_ref *ref;
9374                                 ref = btrfs_item_ptr(buf, i,
9375                                                 struct btrfs_shared_data_ref);
9376                                 add_data_backref(extent_cache,
9377                                         key.objectid, key.offset, 0, 0, 0,
9378                                         btrfs_shared_data_ref_count(buf, ref),
9379                                         0, root->fs_info->sectorsize);
9380                                 continue;
9381                         }
9382                         if (key.type == BTRFS_ORPHAN_ITEM_KEY) {
9383                                 struct bad_item *bad;
9384
9385                                 if (key.objectid == BTRFS_ORPHAN_OBJECTID)
9386                                         continue;
9387                                 if (!owner)
9388                                         continue;
9389                                 bad = malloc(sizeof(struct bad_item));
9390                                 if (!bad)
9391                                         continue;
9392                                 INIT_LIST_HEAD(&bad->list);
9393                                 memcpy(&bad->key, &key,
9394                                        sizeof(struct btrfs_key));
9395                                 bad->root_id = owner;
9396                                 list_add_tail(&bad->list, &delete_items);
9397                                 continue;
9398                         }
9399                         if (key.type != BTRFS_EXTENT_DATA_KEY)
9400                                 continue;
9401                         fi = btrfs_item_ptr(buf, i,
9402                                             struct btrfs_file_extent_item);
9403                         if (btrfs_file_extent_type(buf, fi) ==
9404                             BTRFS_FILE_EXTENT_INLINE)
9405                                 continue;
9406                         if (btrfs_file_extent_disk_bytenr(buf, fi) == 0)
9407                                 continue;
9408
9409                         data_bytes_allocated +=
9410                                 btrfs_file_extent_disk_num_bytes(buf, fi);
9411                         if (data_bytes_allocated < root->fs_info->sectorsize) {
9412                                 abort();
9413                         }
9414                         data_bytes_referenced +=
9415                                 btrfs_file_extent_num_bytes(buf, fi);
9416                         add_data_backref(extent_cache,
9417                                 btrfs_file_extent_disk_bytenr(buf, fi),
9418                                 parent, owner, key.objectid, key.offset -
9419                                 btrfs_file_extent_offset(buf, fi), 1, 1,
9420                                 btrfs_file_extent_disk_num_bytes(buf, fi));
9421                 }
9422         } else {
9423                 int level;
9424                 struct btrfs_key first_key;
9425
9426                 first_key.objectid = 0;
9427
9428                 if (nritems > 0)
9429                         btrfs_item_key_to_cpu(buf, &first_key, 0);
9430                 level = btrfs_header_level(buf);
9431                 for (i = 0; i < nritems; i++) {
9432                         struct extent_record tmpl;
9433
9434                         ptr = btrfs_node_blockptr(buf, i);
9435                         size = root->fs_info->nodesize;
9436                         btrfs_node_key_to_cpu(buf, &key, i);
9437                         if (ri != NULL) {
9438                                 if ((level == ri->drop_level)
9439                                     && is_dropped_key(&key, &ri->drop_key)) {
9440                                         continue;
9441                                 }
9442                         }
9443
9444                         memset(&tmpl, 0, sizeof(tmpl));
9445                         btrfs_cpu_key_to_disk(&tmpl.parent_key, &key);
9446                         tmpl.parent_generation = btrfs_node_ptr_generation(buf, i);
9447                         tmpl.start = ptr;
9448                         tmpl.nr = size;
9449                         tmpl.refs = 1;
9450                         tmpl.metadata = 1;
9451                         tmpl.max_size = size;
9452                         ret = add_extent_rec(extent_cache, &tmpl);
9453                         if (ret < 0)
9454                                 goto out;
9455
9456                         ret = add_tree_backref(extent_cache, ptr, parent,
9457                                         owner, 1);
9458                         if (ret < 0) {
9459                                 error(
9460                                 "add_tree_backref failed (non-leaf block): %s",
9461                                       strerror(-ret));
9462                                 continue;
9463                         }
9464
9465                         if (level > 1) {
9466                                 add_pending(nodes, seen, ptr, size);
9467                         } else {
9468                                 add_pending(pending, seen, ptr, size);
9469                         }
9470                 }
9471                 btree_space_waste += (BTRFS_NODEPTRS_PER_BLOCK(root) -
9472                                       nritems) * sizeof(struct btrfs_key_ptr);
9473         }
9474         total_btree_bytes += buf->len;
9475         if (fs_root_objectid(btrfs_header_owner(buf)))
9476                 total_fs_tree_bytes += buf->len;
9477         if (btrfs_header_owner(buf) == BTRFS_EXTENT_TREE_OBJECTID)
9478                 total_extent_tree_bytes += buf->len;
9479 out:
9480         free_extent_buffer(buf);
9481         return ret;
9482 }
9483
9484 static int add_root_to_pending(struct extent_buffer *buf,
9485                                struct cache_tree *extent_cache,
9486                                struct cache_tree *pending,
9487                                struct cache_tree *seen,
9488                                struct cache_tree *nodes,
9489                                u64 objectid)
9490 {
9491         struct extent_record tmpl;
9492         int ret;
9493
9494         if (btrfs_header_level(buf) > 0)
9495                 add_pending(nodes, seen, buf->start, buf->len);
9496         else
9497                 add_pending(pending, seen, buf->start, buf->len);
9498
9499         memset(&tmpl, 0, sizeof(tmpl));
9500         tmpl.start = buf->start;
9501         tmpl.nr = buf->len;
9502         tmpl.is_root = 1;
9503         tmpl.refs = 1;
9504         tmpl.metadata = 1;
9505         tmpl.max_size = buf->len;
9506         add_extent_rec(extent_cache, &tmpl);
9507
9508         if (objectid == BTRFS_TREE_RELOC_OBJECTID ||
9509             btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
9510                 ret = add_tree_backref(extent_cache, buf->start, buf->start,
9511                                 0, 1);
9512         else
9513                 ret = add_tree_backref(extent_cache, buf->start, 0, objectid,
9514                                 1);
9515         return ret;
9516 }
9517
9518 /* as we fix the tree, we might be deleting blocks that
9519  * we're tracking for repair.  This hook makes sure we
9520  * remove any backrefs for blocks as we are fixing them.
9521  */
9522 static int free_extent_hook(struct btrfs_trans_handle *trans,
9523                             struct btrfs_root *root,
9524                             u64 bytenr, u64 num_bytes, u64 parent,
9525                             u64 root_objectid, u64 owner, u64 offset,
9526                             int refs_to_drop)
9527 {
9528         struct extent_record *rec;
9529         struct cache_extent *cache;
9530         int is_data;
9531         struct cache_tree *extent_cache = root->fs_info->fsck_extent_cache;
9532
9533         is_data = owner >= BTRFS_FIRST_FREE_OBJECTID;
9534         cache = lookup_cache_extent(extent_cache, bytenr, num_bytes);
9535         if (!cache)
9536                 return 0;
9537
9538         rec = container_of(cache, struct extent_record, cache);
9539         if (is_data) {
9540                 struct data_backref *back;
9541                 back = find_data_backref(rec, parent, root_objectid, owner,
9542                                          offset, 1, bytenr, num_bytes);
9543                 if (!back)
9544                         goto out;
9545                 if (back->node.found_ref) {
9546                         back->found_ref -= refs_to_drop;
9547                         if (rec->refs)
9548                                 rec->refs -= refs_to_drop;
9549                 }
9550                 if (back->node.found_extent_tree) {
9551                         back->num_refs -= refs_to_drop;
9552                         if (rec->extent_item_refs)
9553                                 rec->extent_item_refs -= refs_to_drop;
9554                 }
9555                 if (back->found_ref == 0)
9556                         back->node.found_ref = 0;
9557                 if (back->num_refs == 0)
9558                         back->node.found_extent_tree = 0;
9559
9560                 if (!back->node.found_extent_tree && back->node.found_ref) {
9561                         rb_erase(&back->node.node, &rec->backref_tree);
9562                         free(back);
9563                 }
9564         } else {
9565                 struct tree_backref *back;
9566                 back = find_tree_backref(rec, parent, root_objectid);
9567                 if (!back)
9568                         goto out;
9569                 if (back->node.found_ref) {
9570                         if (rec->refs)
9571                                 rec->refs--;
9572                         back->node.found_ref = 0;
9573                 }
9574                 if (back->node.found_extent_tree) {
9575                         if (rec->extent_item_refs)
9576                                 rec->extent_item_refs--;
9577                         back->node.found_extent_tree = 0;
9578                 }
9579                 if (!back->node.found_extent_tree && back->node.found_ref) {
9580                         rb_erase(&back->node.node, &rec->backref_tree);
9581                         free(back);
9582                 }
9583         }
9584         maybe_free_extent_rec(extent_cache, rec);
9585 out:
9586         return 0;
9587 }
9588
9589 static int delete_extent_records(struct btrfs_trans_handle *trans,
9590                                  struct btrfs_root *root,
9591                                  struct btrfs_path *path,
9592                                  u64 bytenr)
9593 {
9594         struct btrfs_key key;
9595         struct btrfs_key found_key;
9596         struct extent_buffer *leaf;
9597         int ret;
9598         int slot;
9599
9600
9601         key.objectid = bytenr;
9602         key.type = (u8)-1;
9603         key.offset = (u64)-1;
9604
9605         while(1) {
9606                 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
9607                                         &key, path, 0, 1);
9608                 if (ret < 0)
9609                         break;
9610
9611                 if (ret > 0) {
9612                         ret = 0;
9613                         if (path->slots[0] == 0)
9614                                 break;
9615                         path->slots[0]--;
9616                 }
9617                 ret = 0;
9618
9619                 leaf = path->nodes[0];
9620                 slot = path->slots[0];
9621
9622                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9623                 if (found_key.objectid != bytenr)
9624                         break;
9625
9626                 if (found_key.type != BTRFS_EXTENT_ITEM_KEY &&
9627                     found_key.type != BTRFS_METADATA_ITEM_KEY &&
9628                     found_key.type != BTRFS_TREE_BLOCK_REF_KEY &&
9629                     found_key.type != BTRFS_EXTENT_DATA_REF_KEY &&
9630                     found_key.type != BTRFS_EXTENT_REF_V0_KEY &&
9631                     found_key.type != BTRFS_SHARED_BLOCK_REF_KEY &&
9632                     found_key.type != BTRFS_SHARED_DATA_REF_KEY) {
9633                         btrfs_release_path(path);
9634                         if (found_key.type == 0) {
9635                                 if (found_key.offset == 0)
9636                                         break;
9637                                 key.offset = found_key.offset - 1;
9638                                 key.type = found_key.type;
9639                         }
9640                         key.type = found_key.type - 1;
9641                         key.offset = (u64)-1;
9642                         continue;
9643                 }
9644
9645                 fprintf(stderr, "repair deleting extent record: key %Lu %u %Lu\n",
9646                         found_key.objectid, found_key.type, found_key.offset);
9647
9648                 ret = btrfs_del_item(trans, root->fs_info->extent_root, path);
9649                 if (ret)
9650                         break;
9651                 btrfs_release_path(path);
9652
9653                 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
9654                     found_key.type == BTRFS_METADATA_ITEM_KEY) {
9655                         u64 bytes = (found_key.type == BTRFS_EXTENT_ITEM_KEY) ?
9656                                 found_key.offset : root->fs_info->nodesize;
9657
9658                         ret = btrfs_update_block_group(trans, root, bytenr,
9659                                                        bytes, 0, 0);
9660                         if (ret)
9661                                 break;
9662                 }
9663         }
9664
9665         btrfs_release_path(path);
9666         return ret;
9667 }
9668
9669 /*
9670  * for a single backref, this will allocate a new extent
9671  * and add the backref to it.
9672  */
9673 static int record_extent(struct btrfs_trans_handle *trans,
9674                          struct btrfs_fs_info *info,
9675                          struct btrfs_path *path,
9676                          struct extent_record *rec,
9677                          struct extent_backref *back,
9678                          int allocated, u64 flags)
9679 {
9680         int ret = 0;
9681         struct btrfs_root *extent_root = info->extent_root;
9682         struct extent_buffer *leaf;
9683         struct btrfs_key ins_key;
9684         struct btrfs_extent_item *ei;
9685         struct data_backref *dback;
9686         struct btrfs_tree_block_info *bi;
9687
9688         if (!back->is_data)
9689                 rec->max_size = max_t(u64, rec->max_size,
9690                                     info->nodesize);
9691
9692         if (!allocated) {
9693                 u32 item_size = sizeof(*ei);
9694
9695                 if (!back->is_data)
9696                         item_size += sizeof(*bi);
9697
9698                 ins_key.objectid = rec->start;
9699                 ins_key.offset = rec->max_size;
9700                 ins_key.type = BTRFS_EXTENT_ITEM_KEY;
9701
9702                 ret = btrfs_insert_empty_item(trans, extent_root, path,
9703                                         &ins_key, item_size);
9704                 if (ret)
9705                         goto fail;
9706
9707                 leaf = path->nodes[0];
9708                 ei = btrfs_item_ptr(leaf, path->slots[0],
9709                                     struct btrfs_extent_item);
9710
9711                 btrfs_set_extent_refs(leaf, ei, 0);
9712                 btrfs_set_extent_generation(leaf, ei, rec->generation);
9713
9714                 if (back->is_data) {
9715                         btrfs_set_extent_flags(leaf, ei,
9716                                                BTRFS_EXTENT_FLAG_DATA);
9717                 } else {
9718                         struct btrfs_disk_key copy_key;;
9719
9720                         bi = (struct btrfs_tree_block_info *)(ei + 1);
9721                         memset_extent_buffer(leaf, 0, (unsigned long)bi,
9722                                              sizeof(*bi));
9723
9724                         btrfs_set_disk_key_objectid(&copy_key,
9725                                                     rec->info_objectid);
9726                         btrfs_set_disk_key_type(&copy_key, 0);
9727                         btrfs_set_disk_key_offset(&copy_key, 0);
9728
9729                         btrfs_set_tree_block_level(leaf, bi, rec->info_level);
9730                         btrfs_set_tree_block_key(leaf, bi, &copy_key);
9731
9732                         btrfs_set_extent_flags(leaf, ei,
9733                                                BTRFS_EXTENT_FLAG_TREE_BLOCK | flags);
9734                 }
9735
9736                 btrfs_mark_buffer_dirty(leaf);
9737                 ret = btrfs_update_block_group(trans, extent_root, rec->start,
9738                                                rec->max_size, 1, 0);
9739                 if (ret)
9740                         goto fail;
9741                 btrfs_release_path(path);
9742         }
9743
9744         if (back->is_data) {
9745                 u64 parent;
9746                 int i;
9747
9748                 dback = to_data_backref(back);
9749                 if (back->full_backref)
9750                         parent = dback->parent;
9751                 else
9752                         parent = 0;
9753
9754                 for (i = 0; i < dback->found_ref; i++) {
9755                         /* if parent != 0, we're doing a full backref
9756                          * passing BTRFS_FIRST_FREE_OBJECTID as the owner
9757                          * just makes the backref allocator create a data
9758                          * backref
9759                          */
9760                         ret = btrfs_inc_extent_ref(trans, info->extent_root,
9761                                                    rec->start, rec->max_size,
9762                                                    parent,
9763                                                    dback->root,
9764                                                    parent ?
9765                                                    BTRFS_FIRST_FREE_OBJECTID :
9766                                                    dback->owner,
9767                                                    dback->offset);
9768                         if (ret)
9769                                 break;
9770                 }
9771                 fprintf(stderr, "adding new data backref"
9772                                 " on %llu %s %llu owner %llu"
9773                                 " offset %llu found %d\n",
9774                                 (unsigned long long)rec->start,
9775                                 back->full_backref ?
9776                                 "parent" : "root",
9777                                 back->full_backref ?
9778                                 (unsigned long long)parent :
9779                                 (unsigned long long)dback->root,
9780                                 (unsigned long long)dback->owner,
9781                                 (unsigned long long)dback->offset,
9782                                 dback->found_ref);
9783         } else {
9784                 u64 parent;
9785                 struct tree_backref *tback;
9786
9787                 tback = to_tree_backref(back);
9788                 if (back->full_backref)
9789                         parent = tback->parent;
9790                 else
9791                         parent = 0;
9792
9793                 ret = btrfs_inc_extent_ref(trans, info->extent_root,
9794                                            rec->start, rec->max_size,
9795                                            parent, tback->root, 0, 0);
9796                 fprintf(stderr, "adding new tree backref on "
9797                         "start %llu len %llu parent %llu root %llu\n",
9798                         rec->start, rec->max_size, parent, tback->root);
9799         }
9800 fail:
9801         btrfs_release_path(path);
9802         return ret;
9803 }
9804
9805 static struct extent_entry *find_entry(struct list_head *entries,
9806                                        u64 bytenr, u64 bytes)
9807 {
9808         struct extent_entry *entry = NULL;
9809
9810         list_for_each_entry(entry, entries, list) {
9811                 if (entry->bytenr == bytenr && entry->bytes == bytes)
9812                         return entry;
9813         }
9814
9815         return NULL;
9816 }
9817
9818 static struct extent_entry *find_most_right_entry(struct list_head *entries)
9819 {
9820         struct extent_entry *entry, *best = NULL, *prev = NULL;
9821
9822         list_for_each_entry(entry, entries, list) {
9823                 /*
9824                  * If there are as many broken entries as entries then we know
9825                  * not to trust this particular entry.
9826                  */
9827                 if (entry->broken == entry->count)
9828                         continue;
9829
9830                 /*
9831                  * Special case, when there are only two entries and 'best' is
9832                  * the first one
9833                  */
9834                 if (!prev) {
9835                         best = entry;
9836                         prev = entry;
9837                         continue;
9838                 }
9839
9840                 /*
9841                  * If our current entry == best then we can't be sure our best
9842                  * is really the best, so we need to keep searching.
9843                  */
9844                 if (best && best->count == entry->count) {
9845                         prev = entry;
9846                         best = NULL;
9847                         continue;
9848                 }
9849
9850                 /* Prev == entry, not good enough, have to keep searching */
9851                 if (!prev->broken && prev->count == entry->count)
9852                         continue;
9853
9854                 if (!best)
9855                         best = (prev->count > entry->count) ? prev : entry;
9856                 else if (best->count < entry->count)
9857                         best = entry;
9858                 prev = entry;
9859         }
9860
9861         return best;
9862 }
9863
9864 static int repair_ref(struct btrfs_fs_info *info, struct btrfs_path *path,
9865                       struct data_backref *dback, struct extent_entry *entry)
9866 {
9867         struct btrfs_trans_handle *trans;
9868         struct btrfs_root *root;
9869         struct btrfs_file_extent_item *fi;
9870         struct extent_buffer *leaf;
9871         struct btrfs_key key;
9872         u64 bytenr, bytes;
9873         int ret, err;
9874
9875         key.objectid = dback->root;
9876         key.type = BTRFS_ROOT_ITEM_KEY;
9877         key.offset = (u64)-1;
9878         root = btrfs_read_fs_root(info, &key);
9879         if (IS_ERR(root)) {
9880                 fprintf(stderr, "Couldn't find root for our ref\n");
9881                 return -EINVAL;
9882         }
9883
9884         /*
9885          * The backref points to the original offset of the extent if it was
9886          * split, so we need to search down to the offset we have and then walk
9887          * forward until we find the backref we're looking for.
9888          */
9889         key.objectid = dback->owner;
9890         key.type = BTRFS_EXTENT_DATA_KEY;
9891         key.offset = dback->offset;
9892         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
9893         if (ret < 0) {
9894                 fprintf(stderr, "Error looking up ref %d\n", ret);
9895                 return ret;
9896         }
9897
9898         while (1) {
9899                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
9900                         ret = btrfs_next_leaf(root, path);
9901                         if (ret) {
9902                                 fprintf(stderr, "Couldn't find our ref, next\n");
9903                                 return -EINVAL;
9904                         }
9905                 }
9906                 leaf = path->nodes[0];
9907                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
9908                 if (key.objectid != dback->owner ||
9909                     key.type != BTRFS_EXTENT_DATA_KEY) {
9910                         fprintf(stderr, "Couldn't find our ref, search\n");
9911                         return -EINVAL;
9912                 }
9913                 fi = btrfs_item_ptr(leaf, path->slots[0],
9914                                     struct btrfs_file_extent_item);
9915                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
9916                 bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
9917
9918                 if (bytenr == dback->disk_bytenr && bytes == dback->bytes)
9919                         break;
9920                 path->slots[0]++;
9921         }
9922
9923         btrfs_release_path(path);
9924
9925         trans = btrfs_start_transaction(root, 1);
9926         if (IS_ERR(trans))
9927                 return PTR_ERR(trans);
9928
9929         /*
9930          * Ok we have the key of the file extent we want to fix, now we can cow
9931          * down to the thing and fix it.
9932          */
9933         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
9934         if (ret < 0) {
9935                 fprintf(stderr, "Error cowing down to ref [%Lu, %u, %Lu]: %d\n",
9936                         key.objectid, key.type, key.offset, ret);
9937                 goto out;
9938         }
9939         if (ret > 0) {
9940                 fprintf(stderr, "Well that's odd, we just found this key "
9941                         "[%Lu, %u, %Lu]\n", key.objectid, key.type,
9942                         key.offset);
9943                 ret = -EINVAL;
9944                 goto out;
9945         }
9946         leaf = path->nodes[0];
9947         fi = btrfs_item_ptr(leaf, path->slots[0],
9948                             struct btrfs_file_extent_item);
9949
9950         if (btrfs_file_extent_compression(leaf, fi) &&
9951             dback->disk_bytenr != entry->bytenr) {
9952                 fprintf(stderr, "Ref doesn't match the record start and is "
9953                         "compressed, please take a btrfs-image of this file "
9954                         "system and send it to a btrfs developer so they can "
9955                         "complete this functionality for bytenr %Lu\n",
9956                         dback->disk_bytenr);
9957                 ret = -EINVAL;
9958                 goto out;
9959         }
9960
9961         if (dback->node.broken && dback->disk_bytenr != entry->bytenr) {
9962                 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
9963         } else if (dback->disk_bytenr > entry->bytenr) {
9964                 u64 off_diff, offset;
9965
9966                 off_diff = dback->disk_bytenr - entry->bytenr;
9967                 offset = btrfs_file_extent_offset(leaf, fi);
9968                 if (dback->disk_bytenr + offset +
9969                     btrfs_file_extent_num_bytes(leaf, fi) >
9970                     entry->bytenr + entry->bytes) {
9971                         fprintf(stderr, "Ref is past the entry end, please "
9972                                 "take a btrfs-image of this file system and "
9973                                 "send it to a btrfs developer, ref %Lu\n",
9974                                 dback->disk_bytenr);
9975                         ret = -EINVAL;
9976                         goto out;
9977                 }
9978                 offset += off_diff;
9979                 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
9980                 btrfs_set_file_extent_offset(leaf, fi, offset);
9981         } else if (dback->disk_bytenr < entry->bytenr) {
9982                 u64 offset;
9983
9984                 offset = btrfs_file_extent_offset(leaf, fi);
9985                 if (dback->disk_bytenr + offset < entry->bytenr) {
9986                         fprintf(stderr, "Ref is before the entry start, please"
9987                                 " take a btrfs-image of this file system and "
9988                                 "send it to a btrfs developer, ref %Lu\n",
9989                                 dback->disk_bytenr);
9990                         ret = -EINVAL;
9991                         goto out;
9992                 }
9993
9994                 offset += dback->disk_bytenr;
9995                 offset -= entry->bytenr;
9996                 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
9997                 btrfs_set_file_extent_offset(leaf, fi, offset);
9998         }
9999
10000         btrfs_set_file_extent_disk_num_bytes(leaf, fi, entry->bytes);
10001
10002         /*
10003          * Chances are if disk_num_bytes were wrong then so is ram_bytes, but
10004          * only do this if we aren't using compression, otherwise it's a
10005          * trickier case.
10006          */
10007         if (!btrfs_file_extent_compression(leaf, fi))
10008                 btrfs_set_file_extent_ram_bytes(leaf, fi, entry->bytes);
10009         else
10010                 printf("ram bytes may be wrong?\n");
10011         btrfs_mark_buffer_dirty(leaf);
10012 out:
10013         err = btrfs_commit_transaction(trans, root);
10014         btrfs_release_path(path);
10015         return ret ? ret : err;
10016 }
10017
10018 static int verify_backrefs(struct btrfs_fs_info *info, struct btrfs_path *path,
10019                            struct extent_record *rec)
10020 {
10021         struct extent_backref *back, *tmp;
10022         struct data_backref *dback;
10023         struct extent_entry *entry, *best = NULL;
10024         LIST_HEAD(entries);
10025         int nr_entries = 0;
10026         int broken_entries = 0;
10027         int ret = 0;
10028         short mismatch = 0;
10029
10030         /*
10031          * Metadata is easy and the backrefs should always agree on bytenr and
10032          * size, if not we've got bigger issues.
10033          */
10034         if (rec->metadata)
10035                 return 0;
10036
10037         rbtree_postorder_for_each_entry_safe(back, tmp,
10038                                              &rec->backref_tree, node) {
10039                 if (back->full_backref || !back->is_data)
10040                         continue;
10041
10042                 dback = to_data_backref(back);
10043
10044                 /*
10045                  * We only pay attention to backrefs that we found a real
10046                  * backref for.
10047                  */
10048                 if (dback->found_ref == 0)
10049                         continue;
10050
10051                 /*
10052                  * For now we only catch when the bytes don't match, not the
10053                  * bytenr.  We can easily do this at the same time, but I want
10054                  * to have a fs image to test on before we just add repair
10055                  * functionality willy-nilly so we know we won't screw up the
10056                  * repair.
10057                  */
10058
10059                 entry = find_entry(&entries, dback->disk_bytenr,
10060                                    dback->bytes);
10061                 if (!entry) {
10062                         entry = malloc(sizeof(struct extent_entry));
10063                         if (!entry) {
10064                                 ret = -ENOMEM;
10065                                 goto out;
10066                         }
10067                         memset(entry, 0, sizeof(*entry));
10068                         entry->bytenr = dback->disk_bytenr;
10069                         entry->bytes = dback->bytes;
10070                         list_add_tail(&entry->list, &entries);
10071                         nr_entries++;
10072                 }
10073
10074                 /*
10075                  * If we only have on entry we may think the entries agree when
10076                  * in reality they don't so we have to do some extra checking.
10077                  */
10078                 if (dback->disk_bytenr != rec->start ||
10079                     dback->bytes != rec->nr || back->broken)
10080                         mismatch = 1;
10081
10082                 if (back->broken) {
10083                         entry->broken++;
10084                         broken_entries++;
10085                 }
10086
10087                 entry->count++;
10088         }
10089
10090         /* Yay all the backrefs agree, carry on good sir */
10091         if (nr_entries <= 1 && !mismatch)
10092                 goto out;
10093
10094         fprintf(stderr, "attempting to repair backref discrepency for bytenr "
10095                 "%Lu\n", rec->start);
10096
10097         /*
10098          * First we want to see if the backrefs can agree amongst themselves who
10099          * is right, so figure out which one of the entries has the highest
10100          * count.
10101          */
10102         best = find_most_right_entry(&entries);
10103
10104         /*
10105          * Ok so we may have an even split between what the backrefs think, so
10106          * this is where we use the extent ref to see what it thinks.
10107          */
10108         if (!best) {
10109                 entry = find_entry(&entries, rec->start, rec->nr);
10110                 if (!entry && (!broken_entries || !rec->found_rec)) {
10111                         fprintf(stderr, "Backrefs don't agree with each other "
10112                                 "and extent record doesn't agree with anybody,"
10113                                 " so we can't fix bytenr %Lu bytes %Lu\n",
10114                                 rec->start, rec->nr);
10115                         ret = -EINVAL;
10116                         goto out;
10117                 } else if (!entry) {
10118                         /*
10119                          * Ok our backrefs were broken, we'll assume this is the
10120                          * correct value and add an entry for this range.
10121                          */
10122                         entry = malloc(sizeof(struct extent_entry));
10123                         if (!entry) {
10124                                 ret = -ENOMEM;
10125                                 goto out;
10126                         }
10127                         memset(entry, 0, sizeof(*entry));
10128                         entry->bytenr = rec->start;
10129                         entry->bytes = rec->nr;
10130                         list_add_tail(&entry->list, &entries);
10131                         nr_entries++;
10132                 }
10133                 entry->count++;
10134                 best = find_most_right_entry(&entries);
10135                 if (!best) {
10136                         fprintf(stderr, "Backrefs and extent record evenly "
10137                                 "split on who is right, this is going to "
10138                                 "require user input to fix bytenr %Lu bytes "
10139                                 "%Lu\n", rec->start, rec->nr);
10140                         ret = -EINVAL;
10141                         goto out;
10142                 }
10143         }
10144
10145         /*
10146          * I don't think this can happen currently as we'll abort() if we catch
10147          * this case higher up, but in case somebody removes that we still can't
10148          * deal with it properly here yet, so just bail out of that's the case.
10149          */
10150         if (best->bytenr != rec->start) {
10151                 fprintf(stderr, "Extent start and backref starts don't match, "
10152                         "please use btrfs-image on this file system and send "
10153                         "it to a btrfs developer so they can make fsck fix "
10154                         "this particular case.  bytenr is %Lu, bytes is %Lu\n",
10155                         rec->start, rec->nr);
10156                 ret = -EINVAL;
10157                 goto out;
10158         }
10159
10160         /*
10161          * Ok great we all agreed on an extent record, let's go find the real
10162          * references and fix up the ones that don't match.
10163          */
10164         rbtree_postorder_for_each_entry_safe(back, tmp,
10165                                              &rec->backref_tree, node) {
10166                 if (back->full_backref || !back->is_data)
10167                         continue;
10168
10169                 dback = to_data_backref(back);
10170
10171                 /*
10172                  * Still ignoring backrefs that don't have a real ref attached
10173                  * to them.
10174                  */
10175                 if (dback->found_ref == 0)
10176                         continue;
10177
10178                 if (dback->bytes == best->bytes &&
10179                     dback->disk_bytenr == best->bytenr)
10180                         continue;
10181
10182                 ret = repair_ref(info, path, dback, best);
10183                 if (ret)
10184                         goto out;
10185         }
10186
10187         /*
10188          * Ok we messed with the actual refs, which means we need to drop our
10189          * entire cache and go back and rescan.  I know this is a huge pain and
10190          * adds a lot of extra work, but it's the only way to be safe.  Once all
10191          * the backrefs agree we may not need to do anything to the extent
10192          * record itself.
10193          */
10194         ret = -EAGAIN;
10195 out:
10196         while (!list_empty(&entries)) {
10197                 entry = list_entry(entries.next, struct extent_entry, list);
10198                 list_del_init(&entry->list);
10199                 free(entry);
10200         }
10201         return ret;
10202 }
10203
10204 static int process_duplicates(struct cache_tree *extent_cache,
10205                               struct extent_record *rec)
10206 {
10207         struct extent_record *good, *tmp;
10208         struct cache_extent *cache;
10209         int ret;
10210
10211         /*
10212          * If we found a extent record for this extent then return, or if we
10213          * have more than one duplicate we are likely going to need to delete
10214          * something.
10215          */
10216         if (rec->found_rec || rec->num_duplicates > 1)
10217                 return 0;
10218
10219         /* Shouldn't happen but just in case */
10220         BUG_ON(!rec->num_duplicates);
10221
10222         /*
10223          * So this happens if we end up with a backref that doesn't match the
10224          * actual extent entry.  So either the backref is bad or the extent
10225          * entry is bad.  Either way we want to have the extent_record actually
10226          * reflect what we found in the extent_tree, so we need to take the
10227          * duplicate out and use that as the extent_record since the only way we
10228          * get a duplicate is if we find a real life BTRFS_EXTENT_ITEM_KEY.
10229          */
10230         remove_cache_extent(extent_cache, &rec->cache);
10231
10232         good = to_extent_record(rec->dups.next);
10233         list_del_init(&good->list);
10234         INIT_LIST_HEAD(&good->backrefs);
10235         INIT_LIST_HEAD(&good->dups);
10236         good->cache.start = good->start;
10237         good->cache.size = good->nr;
10238         good->content_checked = 0;
10239         good->owner_ref_checked = 0;
10240         good->num_duplicates = 0;
10241         good->refs = rec->refs;
10242         list_splice_init(&rec->backrefs, &good->backrefs);
10243         while (1) {
10244                 cache = lookup_cache_extent(extent_cache, good->start,
10245                                             good->nr);
10246                 if (!cache)
10247                         break;
10248                 tmp = container_of(cache, struct extent_record, cache);
10249
10250                 /*
10251                  * If we find another overlapping extent and it's found_rec is
10252                  * set then it's a duplicate and we need to try and delete
10253                  * something.
10254                  */
10255                 if (tmp->found_rec || tmp->num_duplicates > 0) {
10256                         if (list_empty(&good->list))
10257                                 list_add_tail(&good->list,
10258                                               &duplicate_extents);
10259                         good->num_duplicates += tmp->num_duplicates + 1;
10260                         list_splice_init(&tmp->dups, &good->dups);
10261                         list_del_init(&tmp->list);
10262                         list_add_tail(&tmp->list, &good->dups);
10263                         remove_cache_extent(extent_cache, &tmp->cache);
10264                         continue;
10265                 }
10266
10267                 /*
10268                  * Ok we have another non extent item backed extent rec, so lets
10269                  * just add it to this extent and carry on like we did above.
10270                  */
10271                 good->refs += tmp->refs;
10272                 list_splice_init(&tmp->backrefs, &good->backrefs);
10273                 remove_cache_extent(extent_cache, &tmp->cache);
10274                 free(tmp);
10275         }
10276         ret = insert_cache_extent(extent_cache, &good->cache);
10277         BUG_ON(ret);
10278         free(rec);
10279         return good->num_duplicates ? 0 : 1;
10280 }
10281
10282 static int delete_duplicate_records(struct btrfs_root *root,
10283                                     struct extent_record *rec)
10284 {
10285         struct btrfs_trans_handle *trans;
10286         LIST_HEAD(delete_list);
10287         struct btrfs_path path;
10288         struct extent_record *tmp, *good, *n;
10289         int nr_del = 0;
10290         int ret = 0, err;
10291         struct btrfs_key key;
10292
10293         btrfs_init_path(&path);
10294
10295         good = rec;
10296         /* Find the record that covers all of the duplicates. */
10297         list_for_each_entry(tmp, &rec->dups, list) {
10298                 if (good->start < tmp->start)
10299                         continue;
10300                 if (good->nr > tmp->nr)
10301                         continue;
10302
10303                 if (tmp->start + tmp->nr < good->start + good->nr) {
10304                         fprintf(stderr, "Ok we have overlapping extents that "
10305                                 "aren't completely covered by each other, this "
10306                                 "is going to require more careful thought.  "
10307                                 "The extents are [%Lu-%Lu] and [%Lu-%Lu]\n",
10308                                 tmp->start, tmp->nr, good->start, good->nr);
10309                         abort();
10310                 }
10311                 good = tmp;
10312         }
10313
10314         if (good != rec)
10315                 list_add_tail(&rec->list, &delete_list);
10316
10317         list_for_each_entry_safe(tmp, n, &rec->dups, list) {
10318                 if (tmp == good)
10319                         continue;
10320                 list_move_tail(&tmp->list, &delete_list);
10321         }
10322
10323         root = root->fs_info->extent_root;
10324         trans = btrfs_start_transaction(root, 1);
10325         if (IS_ERR(trans)) {
10326                 ret = PTR_ERR(trans);
10327                 goto out;
10328         }
10329
10330         list_for_each_entry(tmp, &delete_list, list) {
10331                 if (tmp->found_rec == 0)
10332                         continue;
10333                 key.objectid = tmp->start;
10334                 key.type = BTRFS_EXTENT_ITEM_KEY;
10335                 key.offset = tmp->nr;
10336
10337                 /* Shouldn't happen but just in case */
10338                 if (tmp->metadata) {
10339                         fprintf(stderr, "Well this shouldn't happen, extent "
10340                                 "record overlaps but is metadata? "
10341                                 "[%Lu, %Lu]\n", tmp->start, tmp->nr);
10342                         abort();
10343                 }
10344
10345                 ret = btrfs_search_slot(trans, root, &key, &path, -1, 1);
10346                 if (ret) {
10347                         if (ret > 0)
10348                                 ret = -EINVAL;
10349                         break;
10350                 }
10351                 ret = btrfs_del_item(trans, root, &path);
10352                 if (ret)
10353                         break;
10354                 btrfs_release_path(&path);
10355                 nr_del++;
10356         }
10357         err = btrfs_commit_transaction(trans, root);
10358         if (err && !ret)
10359                 ret = err;
10360 out:
10361         while (!list_empty(&delete_list)) {
10362                 tmp = to_extent_record(delete_list.next);
10363                 list_del_init(&tmp->list);
10364                 if (tmp == rec)
10365                         continue;
10366                 free(tmp);
10367         }
10368
10369         while (!list_empty(&rec->dups)) {
10370                 tmp = to_extent_record(rec->dups.next);
10371                 list_del_init(&tmp->list);
10372                 free(tmp);
10373         }
10374
10375         btrfs_release_path(&path);
10376
10377         if (!ret && !nr_del)
10378                 rec->num_duplicates = 0;
10379
10380         return ret ? ret : nr_del;
10381 }
10382
10383 static int find_possible_backrefs(struct btrfs_fs_info *info,
10384                                   struct btrfs_path *path,
10385                                   struct cache_tree *extent_cache,
10386                                   struct extent_record *rec)
10387 {
10388         struct btrfs_root *root;
10389         struct extent_backref *back, *tmp;
10390         struct data_backref *dback;
10391         struct cache_extent *cache;
10392         struct btrfs_file_extent_item *fi;
10393         struct btrfs_key key;
10394         u64 bytenr, bytes;
10395         int ret;
10396
10397         rbtree_postorder_for_each_entry_safe(back, tmp,
10398                                              &rec->backref_tree, node) {
10399                 /* Don't care about full backrefs (poor unloved backrefs) */
10400                 if (back->full_backref || !back->is_data)
10401                         continue;
10402
10403                 dback = to_data_backref(back);
10404
10405                 /* We found this one, we don't need to do a lookup */
10406                 if (dback->found_ref)
10407                         continue;
10408
10409                 key.objectid = dback->root;
10410                 key.type = BTRFS_ROOT_ITEM_KEY;
10411                 key.offset = (u64)-1;
10412
10413                 root = btrfs_read_fs_root(info, &key);
10414
10415                 /* No root, definitely a bad ref, skip */
10416                 if (IS_ERR(root) && PTR_ERR(root) == -ENOENT)
10417                         continue;
10418                 /* Other err, exit */
10419                 if (IS_ERR(root))
10420                         return PTR_ERR(root);
10421
10422                 key.objectid = dback->owner;
10423                 key.type = BTRFS_EXTENT_DATA_KEY;
10424                 key.offset = dback->offset;
10425                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
10426                 if (ret) {
10427                         btrfs_release_path(path);
10428                         if (ret < 0)
10429                                 return ret;
10430                         /* Didn't find it, we can carry on */
10431                         ret = 0;
10432                         continue;
10433                 }
10434
10435                 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
10436                                     struct btrfs_file_extent_item);
10437                 bytenr = btrfs_file_extent_disk_bytenr(path->nodes[0], fi);
10438                 bytes = btrfs_file_extent_disk_num_bytes(path->nodes[0], fi);
10439                 btrfs_release_path(path);
10440                 cache = lookup_cache_extent(extent_cache, bytenr, 1);
10441                 if (cache) {
10442                         struct extent_record *tmp;
10443                         tmp = container_of(cache, struct extent_record, cache);
10444
10445                         /*
10446                          * If we found an extent record for the bytenr for this
10447                          * particular backref then we can't add it to our
10448                          * current extent record.  We only want to add backrefs
10449                          * that don't have a corresponding extent item in the
10450                          * extent tree since they likely belong to this record
10451                          * and we need to fix it if it doesn't match bytenrs.
10452                          */
10453                         if  (tmp->found_rec)
10454                                 continue;
10455                 }
10456
10457                 dback->found_ref += 1;
10458                 dback->disk_bytenr = bytenr;
10459                 dback->bytes = bytes;
10460
10461                 /*
10462                  * Set this so the verify backref code knows not to trust the
10463                  * values in this backref.
10464                  */
10465                 back->broken = 1;
10466         }
10467
10468         return 0;
10469 }
10470
10471 /*
10472  * Record orphan data ref into corresponding root.
10473  *
10474  * Return 0 if the extent item contains data ref and recorded.
10475  * Return 1 if the extent item contains no useful data ref
10476  *   On that case, it may contains only shared_dataref or metadata backref
10477  *   or the file extent exists(this should be handled by the extent bytenr
10478  *   recovery routine)
10479  * Return <0 if something goes wrong.
10480  */
10481 static int record_orphan_data_extents(struct btrfs_fs_info *fs_info,
10482                                       struct extent_record *rec)
10483 {
10484         struct btrfs_key key;
10485         struct btrfs_root *dest_root;
10486         struct extent_backref *back, *tmp;
10487         struct data_backref *dback;
10488         struct orphan_data_extent *orphan;
10489         struct btrfs_path path;
10490         int recorded_data_ref = 0;
10491         int ret = 0;
10492
10493         if (rec->metadata)
10494                 return 1;
10495         btrfs_init_path(&path);
10496         rbtree_postorder_for_each_entry_safe(back, tmp,
10497                                              &rec->backref_tree, node) {
10498                 if (back->full_backref || !back->is_data ||
10499                     !back->found_extent_tree)
10500                         continue;
10501                 dback = to_data_backref(back);
10502                 if (dback->found_ref)
10503                         continue;
10504                 key.objectid = dback->root;
10505                 key.type = BTRFS_ROOT_ITEM_KEY;
10506                 key.offset = (u64)-1;
10507
10508                 dest_root = btrfs_read_fs_root(fs_info, &key);
10509
10510                 /* For non-exist root we just skip it */
10511                 if (IS_ERR(dest_root) || !dest_root)
10512                         continue;
10513
10514                 key.objectid = dback->owner;
10515                 key.type = BTRFS_EXTENT_DATA_KEY;
10516                 key.offset = dback->offset;
10517
10518                 ret = btrfs_search_slot(NULL, dest_root, &key, &path, 0, 0);
10519                 btrfs_release_path(&path);
10520                 /*
10521                  * For ret < 0, it's OK since the fs-tree may be corrupted,
10522                  * we need to record it for inode/file extent rebuild.
10523                  * For ret > 0, we record it only for file extent rebuild.
10524                  * For ret == 0, the file extent exists but only bytenr
10525                  * mismatch, let the original bytenr fix routine to handle,
10526                  * don't record it.
10527                  */
10528                 if (ret == 0)
10529                         continue;
10530                 ret = 0;
10531                 orphan = malloc(sizeof(*orphan));
10532                 if (!orphan) {
10533                         ret = -ENOMEM;
10534                         goto out;
10535                 }
10536                 INIT_LIST_HEAD(&orphan->list);
10537                 orphan->root = dback->root;
10538                 orphan->objectid = dback->owner;
10539                 orphan->offset = dback->offset;
10540                 orphan->disk_bytenr = rec->cache.start;
10541                 orphan->disk_len = rec->cache.size;
10542                 list_add(&dest_root->orphan_data_extents, &orphan->list);
10543                 recorded_data_ref = 1;
10544         }
10545 out:
10546         btrfs_release_path(&path);
10547         if (!ret)
10548                 return !recorded_data_ref;
10549         else
10550                 return ret;
10551 }
10552
10553 /*
10554  * when an incorrect extent item is found, this will delete
10555  * all of the existing entries for it and recreate them
10556  * based on what the tree scan found.
10557  */
10558 static int fixup_extent_refs(struct btrfs_fs_info *info,
10559                              struct cache_tree *extent_cache,
10560                              struct extent_record *rec)
10561 {
10562         struct btrfs_trans_handle *trans = NULL;
10563         int ret;
10564         struct btrfs_path path;
10565         struct cache_extent *cache;
10566         struct extent_backref *back, *tmp;
10567         int allocated = 0;
10568         u64 flags = 0;
10569
10570         if (rec->flag_block_full_backref)
10571                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
10572
10573         btrfs_init_path(&path);
10574         if (rec->refs != rec->extent_item_refs && !rec->metadata) {
10575                 /*
10576                  * Sometimes the backrefs themselves are so broken they don't
10577                  * get attached to any meaningful rec, so first go back and
10578                  * check any of our backrefs that we couldn't find and throw
10579                  * them into the list if we find the backref so that
10580                  * verify_backrefs can figure out what to do.
10581                  */
10582                 ret = find_possible_backrefs(info, &path, extent_cache, rec);
10583                 if (ret < 0)
10584                         goto out;
10585         }
10586
10587         /* step one, make sure all of the backrefs agree */
10588         ret = verify_backrefs(info, &path, rec);
10589         if (ret < 0)
10590                 goto out;
10591
10592         trans = btrfs_start_transaction(info->extent_root, 1);
10593         if (IS_ERR(trans)) {
10594                 ret = PTR_ERR(trans);
10595                 goto out;
10596         }
10597
10598         /* step two, delete all the existing records */
10599         ret = delete_extent_records(trans, info->extent_root, &path,
10600                                     rec->start);
10601
10602         if (ret < 0)
10603                 goto out;
10604
10605         /* was this block corrupt?  If so, don't add references to it */
10606         cache = lookup_cache_extent(info->corrupt_blocks,
10607                                     rec->start, rec->max_size);
10608         if (cache) {
10609                 ret = 0;
10610                 goto out;
10611         }
10612
10613         /* step three, recreate all the refs we did find */
10614         rbtree_postorder_for_each_entry_safe(back, tmp,
10615                                              &rec->backref_tree, node) {
10616                 /*
10617                  * if we didn't find any references, don't create a
10618                  * new extent record
10619                  */
10620                 if (!back->found_ref)
10621                         continue;
10622
10623                 rec->bad_full_backref = 0;
10624                 ret = record_extent(trans, info, &path, rec, back, allocated, flags);
10625                 allocated = 1;
10626
10627                 if (ret)
10628                         goto out;
10629         }
10630 out:
10631         if (trans) {
10632                 int err = btrfs_commit_transaction(trans, info->extent_root);
10633                 if (!ret)
10634                         ret = err;
10635         }
10636
10637         if (!ret)
10638                 fprintf(stderr, "Repaired extent references for %llu\n",
10639                                 (unsigned long long)rec->start);
10640
10641         btrfs_release_path(&path);
10642         return ret;
10643 }
10644
10645 static int fixup_extent_flags(struct btrfs_fs_info *fs_info,
10646                               struct extent_record *rec)
10647 {
10648         struct btrfs_trans_handle *trans;
10649         struct btrfs_root *root = fs_info->extent_root;
10650         struct btrfs_path path;
10651         struct btrfs_extent_item *ei;
10652         struct btrfs_key key;
10653         u64 flags;
10654         int ret = 0;
10655
10656         key.objectid = rec->start;
10657         if (rec->metadata) {
10658                 key.type = BTRFS_METADATA_ITEM_KEY;
10659                 key.offset = rec->info_level;
10660         } else {
10661                 key.type = BTRFS_EXTENT_ITEM_KEY;
10662                 key.offset = rec->max_size;
10663         }
10664
10665         trans = btrfs_start_transaction(root, 0);
10666         if (IS_ERR(trans))
10667                 return PTR_ERR(trans);
10668
10669         btrfs_init_path(&path);
10670         ret = btrfs_search_slot(trans, root, &key, &path, 0, 1);
10671         if (ret < 0) {
10672                 btrfs_release_path(&path);
10673                 btrfs_commit_transaction(trans, root);
10674                 return ret;
10675         } else if (ret) {
10676                 fprintf(stderr, "Didn't find extent for %llu\n",
10677                         (unsigned long long)rec->start);
10678                 btrfs_release_path(&path);
10679                 btrfs_commit_transaction(trans, root);
10680                 return -ENOENT;
10681         }
10682
10683         ei = btrfs_item_ptr(path.nodes[0], path.slots[0],
10684                             struct btrfs_extent_item);
10685         flags = btrfs_extent_flags(path.nodes[0], ei);
10686         if (rec->flag_block_full_backref) {
10687                 fprintf(stderr, "setting full backref on %llu\n",
10688                         (unsigned long long)key.objectid);
10689                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
10690         } else {
10691                 fprintf(stderr, "clearing full backref on %llu\n",
10692                         (unsigned long long)key.objectid);
10693                 flags &= ~BTRFS_BLOCK_FLAG_FULL_BACKREF;
10694         }
10695         btrfs_set_extent_flags(path.nodes[0], ei, flags);
10696         btrfs_mark_buffer_dirty(path.nodes[0]);
10697         btrfs_release_path(&path);
10698         ret = btrfs_commit_transaction(trans, root);
10699         if (!ret)
10700                 fprintf(stderr, "Repaired extent flags for %llu\n",
10701                                 (unsigned long long)rec->start);
10702
10703         return ret;
10704 }
10705
10706 /* right now we only prune from the extent allocation tree */
10707 static int prune_one_block(struct btrfs_trans_handle *trans,
10708                            struct btrfs_fs_info *info,
10709                            struct btrfs_corrupt_block *corrupt)
10710 {
10711         int ret;
10712         struct btrfs_path path;
10713         struct extent_buffer *eb;
10714         u64 found;
10715         int slot;
10716         int nritems;
10717         int level = corrupt->level + 1;
10718
10719         btrfs_init_path(&path);
10720 again:
10721         /* we want to stop at the parent to our busted block */
10722         path.lowest_level = level;
10723
10724         ret = btrfs_search_slot(trans, info->extent_root,
10725                                 &corrupt->key, &path, -1, 1);
10726
10727         if (ret < 0)
10728                 goto out;
10729
10730         eb = path.nodes[level];
10731         if (!eb) {
10732                 ret = -ENOENT;
10733                 goto out;
10734         }
10735
10736         /*
10737          * hopefully the search gave us the block we want to prune,
10738          * lets try that first
10739          */
10740         slot = path.slots[level];
10741         found =  btrfs_node_blockptr(eb, slot);
10742         if (found == corrupt->cache.start)
10743                 goto del_ptr;
10744
10745         nritems = btrfs_header_nritems(eb);
10746
10747         /* the search failed, lets scan this node and hope we find it */
10748         for (slot = 0; slot < nritems; slot++) {
10749                 found =  btrfs_node_blockptr(eb, slot);
10750                 if (found == corrupt->cache.start)
10751                         goto del_ptr;
10752         }
10753         /*
10754          * we couldn't find the bad block.  TODO, search all the nodes for pointers
10755          * to this block
10756          */
10757         if (eb == info->extent_root->node) {
10758                 ret = -ENOENT;
10759                 goto out;
10760         } else {
10761                 level++;
10762                 btrfs_release_path(&path);
10763                 goto again;
10764         }
10765
10766 del_ptr:
10767         printk("deleting pointer to block %Lu\n", corrupt->cache.start);
10768         ret = btrfs_del_ptr(info->extent_root, &path, level, slot);
10769
10770 out:
10771         btrfs_release_path(&path);
10772         return ret;
10773 }
10774
10775 static int prune_corrupt_blocks(struct btrfs_fs_info *info)
10776 {
10777         struct btrfs_trans_handle *trans = NULL;
10778         struct cache_extent *cache;
10779         struct btrfs_corrupt_block *corrupt;
10780
10781         while (1) {
10782                 cache = search_cache_extent(info->corrupt_blocks, 0);
10783                 if (!cache)
10784                         break;
10785                 if (!trans) {
10786                         trans = btrfs_start_transaction(info->extent_root, 1);
10787                         if (IS_ERR(trans))
10788                                 return PTR_ERR(trans);
10789                 }
10790                 corrupt = container_of(cache, struct btrfs_corrupt_block, cache);
10791                 prune_one_block(trans, info, corrupt);
10792                 remove_cache_extent(info->corrupt_blocks, cache);
10793         }
10794         if (trans)
10795                 return btrfs_commit_transaction(trans, info->extent_root);
10796         return 0;
10797 }
10798
10799 static void reset_cached_block_groups(struct btrfs_fs_info *fs_info)
10800 {
10801         struct btrfs_block_group_cache *cache;
10802         u64 start, end;
10803         int ret;
10804
10805         while (1) {
10806                 ret = find_first_extent_bit(&fs_info->free_space_cache, 0,
10807                                             &start, &end, EXTENT_DIRTY);
10808                 if (ret)
10809                         break;
10810                 clear_extent_dirty(&fs_info->free_space_cache, start, end);
10811         }
10812
10813         start = 0;
10814         while (1) {
10815                 cache = btrfs_lookup_first_block_group(fs_info, start);
10816                 if (!cache)
10817                         break;
10818                 if (cache->cached)
10819                         cache->cached = 0;
10820                 start = cache->key.objectid + cache->key.offset;
10821         }
10822 }
10823
10824 static int check_extent_refs(struct btrfs_root *root,
10825                              struct cache_tree *extent_cache)
10826 {
10827         struct extent_record *rec;
10828         struct cache_extent *cache;
10829         int ret = 0;
10830         int had_dups = 0;
10831         int err = 0;
10832
10833         if (repair) {
10834                 /*
10835                  * if we're doing a repair, we have to make sure
10836                  * we don't allocate from the problem extents.
10837                  * In the worst case, this will be all the
10838                  * extents in the FS
10839                  */
10840                 cache = search_cache_extent(extent_cache, 0);
10841                 while(cache) {
10842                         rec = container_of(cache, struct extent_record, cache);
10843                         set_extent_dirty(root->fs_info->excluded_extents,
10844                                          rec->start,
10845                                          rec->start + rec->max_size - 1);
10846                         cache = next_cache_extent(cache);
10847                 }
10848
10849                 /* pin down all the corrupted blocks too */
10850                 cache = search_cache_extent(root->fs_info->corrupt_blocks, 0);
10851                 while(cache) {
10852                         set_extent_dirty(root->fs_info->excluded_extents,
10853                                          cache->start,
10854                                          cache->start + cache->size - 1);
10855                         cache = next_cache_extent(cache);
10856                 }
10857                 prune_corrupt_blocks(root->fs_info);
10858                 reset_cached_block_groups(root->fs_info);
10859         }
10860
10861         reset_cached_block_groups(root->fs_info);
10862
10863         /*
10864          * We need to delete any duplicate entries we find first otherwise we
10865          * could mess up the extent tree when we have backrefs that actually
10866          * belong to a different extent item and not the weird duplicate one.
10867          */
10868         while (repair && !list_empty(&duplicate_extents)) {
10869                 rec = to_extent_record(duplicate_extents.next);
10870                 list_del_init(&rec->list);
10871
10872                 /* Sometimes we can find a backref before we find an actual
10873                  * extent, so we need to process it a little bit to see if there
10874                  * truly are multiple EXTENT_ITEM_KEY's for the same range, or
10875                  * if this is a backref screwup.  If we need to delete stuff
10876                  * process_duplicates() will return 0, otherwise it will return
10877                  * 1 and we
10878                  */
10879                 if (process_duplicates(extent_cache, rec))
10880                         continue;
10881                 ret = delete_duplicate_records(root, rec);
10882                 if (ret < 0)
10883                         return ret;
10884                 /*
10885                  * delete_duplicate_records will return the number of entries
10886                  * deleted, so if it's greater than 0 then we know we actually
10887                  * did something and we need to remove.
10888                  */
10889                 if (ret)
10890                         had_dups = 1;
10891         }
10892
10893         if (had_dups)
10894                 return -EAGAIN;
10895
10896         while(1) {
10897                 int cur_err = 0;
10898                 int fix = 0;
10899
10900                 cache = search_cache_extent(extent_cache, 0);
10901                 if (!cache)
10902                         break;
10903                 rec = container_of(cache, struct extent_record, cache);
10904                 if (rec->num_duplicates) {
10905                         fprintf(stderr, "extent item %llu has multiple extent "
10906                                 "items\n", (unsigned long long)rec->start);
10907                         cur_err = 1;
10908                 }
10909
10910                 if (rec->refs != rec->extent_item_refs) {
10911                         fprintf(stderr, "ref mismatch on [%llu %llu] ",
10912                                 (unsigned long long)rec->start,
10913                                 (unsigned long long)rec->nr);
10914                         fprintf(stderr, "extent item %llu, found %llu\n",
10915                                 (unsigned long long)rec->extent_item_refs,
10916                                 (unsigned long long)rec->refs);
10917                         ret = record_orphan_data_extents(root->fs_info, rec);
10918                         if (ret < 0)
10919                                 goto repair_abort;
10920                         fix = ret;
10921                         cur_err = 1;
10922                 }
10923                 if (all_backpointers_checked(rec, 1)) {
10924                         fprintf(stderr, "backpointer mismatch on [%llu %llu]\n",
10925                                 (unsigned long long)rec->start,
10926                                 (unsigned long long)rec->nr);
10927                         fix = 1;
10928                         cur_err = 1;
10929                 }
10930                 if (!rec->owner_ref_checked) {
10931                         fprintf(stderr, "owner ref check failed [%llu %llu]\n",
10932                                 (unsigned long long)rec->start,
10933                                 (unsigned long long)rec->nr);
10934                         fix = 1;
10935                         cur_err = 1;
10936                 }
10937
10938                 if (repair && fix) {
10939                         ret = fixup_extent_refs(root->fs_info, extent_cache, rec);
10940                         if (ret)
10941                                 goto repair_abort;
10942                 }
10943
10944
10945                 if (rec->bad_full_backref) {
10946                         fprintf(stderr, "bad full backref, on [%llu]\n",
10947                                 (unsigned long long)rec->start);
10948                         if (repair) {
10949                                 ret = fixup_extent_flags(root->fs_info, rec);
10950                                 if (ret)
10951                                         goto repair_abort;
10952                                 fix = 1;
10953                         }
10954                         cur_err = 1;
10955                 }
10956                 /*
10957                  * Although it's not a extent ref's problem, we reuse this
10958                  * routine for error reporting.
10959                  * No repair function yet.
10960                  */
10961                 if (rec->crossing_stripes) {
10962                         fprintf(stderr,
10963                                 "bad metadata [%llu, %llu) crossing stripe boundary\n",
10964                                 rec->start, rec->start + rec->max_size);
10965                         cur_err = 1;
10966                 }
10967
10968                 if (rec->wrong_chunk_type) {
10969                         fprintf(stderr,
10970                                 "bad extent [%llu, %llu), type mismatch with chunk\n",
10971                                 rec->start, rec->start + rec->max_size);
10972                         cur_err = 1;
10973                 }
10974
10975                 err = cur_err;
10976                 remove_cache_extent(extent_cache, cache);
10977                 free_all_extent_backrefs(rec);
10978                 if (!init_extent_tree && repair && (!cur_err || fix))
10979                         clear_extent_dirty(root->fs_info->excluded_extents,
10980                                            rec->start,
10981                                            rec->start + rec->max_size - 1);
10982                 free(rec);
10983         }
10984 repair_abort:
10985         if (repair) {
10986                 if (ret && ret != -EAGAIN) {
10987                         fprintf(stderr, "failed to repair damaged filesystem, aborting\n");
10988                         exit(1);
10989                 } else if (!ret) {
10990                         struct btrfs_trans_handle *trans;
10991
10992                         root = root->fs_info->extent_root;
10993                         trans = btrfs_start_transaction(root, 1);
10994                         if (IS_ERR(trans)) {
10995                                 ret = PTR_ERR(trans);
10996                                 goto repair_abort;
10997                         }
10998
10999                         ret = btrfs_fix_block_accounting(trans, root);
11000                         if (ret)
11001                                 goto repair_abort;
11002                         ret = btrfs_commit_transaction(trans, root);
11003                         if (ret)
11004                                 goto repair_abort;
11005                 }
11006                 return ret;
11007         }
11008
11009         if (err)
11010                 err = -EIO;
11011         return err;
11012 }
11013
11014 u64 calc_stripe_length(u64 type, u64 length, int num_stripes)
11015 {
11016         u64 stripe_size;
11017
11018         if (type & BTRFS_BLOCK_GROUP_RAID0) {
11019                 stripe_size = length;
11020                 stripe_size /= num_stripes;
11021         } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
11022                 stripe_size = length * 2;
11023                 stripe_size /= num_stripes;
11024         } else if (type & BTRFS_BLOCK_GROUP_RAID5) {
11025                 stripe_size = length;
11026                 stripe_size /= (num_stripes - 1);
11027         } else if (type & BTRFS_BLOCK_GROUP_RAID6) {
11028                 stripe_size = length;
11029                 stripe_size /= (num_stripes - 2);
11030         } else {
11031                 stripe_size = length;
11032         }
11033         return stripe_size;
11034 }
11035
11036 /*
11037  * Check the chunk with its block group/dev list ref:
11038  * Return 0 if all refs seems valid.
11039  * Return 1 if part of refs seems valid, need later check for rebuild ref
11040  * like missing block group and needs to search extent tree to rebuild them.
11041  * Return -1 if essential refs are missing and unable to rebuild.
11042  */
11043 static int check_chunk_refs(struct chunk_record *chunk_rec,
11044                             struct block_group_tree *block_group_cache,
11045                             struct device_extent_tree *dev_extent_cache,
11046                             int silent)
11047 {
11048         struct cache_extent *block_group_item;
11049         struct block_group_record *block_group_rec;
11050         struct cache_extent *dev_extent_item;
11051         struct device_extent_record *dev_extent_rec;
11052         u64 devid;
11053         u64 offset;
11054         u64 length;
11055         int metadump_v2 = 0;
11056         int i;
11057         int ret = 0;
11058
11059         block_group_item = lookup_cache_extent(&block_group_cache->tree,
11060                                                chunk_rec->offset,
11061                                                chunk_rec->length);
11062         if (block_group_item) {
11063                 block_group_rec = container_of(block_group_item,
11064                                                struct block_group_record,
11065                                                cache);
11066                 if (chunk_rec->length != block_group_rec->offset ||
11067                     chunk_rec->offset != block_group_rec->objectid ||
11068                     (!metadump_v2 &&
11069                      chunk_rec->type_flags != block_group_rec->flags)) {
11070                         if (!silent)
11071                                 fprintf(stderr,
11072                                         "Chunk[%llu, %u, %llu]: length(%llu), offset(%llu), type(%llu) mismatch with block group[%llu, %u, %llu]: offset(%llu), objectid(%llu), flags(%llu)\n",
11073                                         chunk_rec->objectid,
11074                                         chunk_rec->type,
11075                                         chunk_rec->offset,
11076                                         chunk_rec->length,
11077                                         chunk_rec->offset,
11078                                         chunk_rec->type_flags,
11079                                         block_group_rec->objectid,
11080                                         block_group_rec->type,
11081                                         block_group_rec->offset,
11082                                         block_group_rec->offset,
11083                                         block_group_rec->objectid,
11084                                         block_group_rec->flags);
11085                         ret = -1;
11086                 } else {
11087                         list_del_init(&block_group_rec->list);
11088                         chunk_rec->bg_rec = block_group_rec;
11089                 }
11090         } else {
11091                 if (!silent)
11092                         fprintf(stderr,
11093                                 "Chunk[%llu, %u, %llu]: length(%llu), offset(%llu), type(%llu) is not found in block group\n",
11094                                 chunk_rec->objectid,
11095                                 chunk_rec->type,
11096                                 chunk_rec->offset,
11097                                 chunk_rec->length,
11098                                 chunk_rec->offset,
11099                                 chunk_rec->type_flags);
11100                 ret = 1;
11101         }
11102
11103         if (metadump_v2)
11104                 return ret;
11105
11106         length = calc_stripe_length(chunk_rec->type_flags, chunk_rec->length,
11107                                     chunk_rec->num_stripes);
11108         for (i = 0; i < chunk_rec->num_stripes; ++i) {
11109                 devid = chunk_rec->stripes[i].devid;
11110                 offset = chunk_rec->stripes[i].offset;
11111                 dev_extent_item = lookup_cache_extent2(&dev_extent_cache->tree,
11112                                                        devid, offset, length);
11113                 if (dev_extent_item) {
11114                         dev_extent_rec = container_of(dev_extent_item,
11115                                                 struct device_extent_record,
11116                                                 cache);
11117                         if (dev_extent_rec->objectid != devid ||
11118                             dev_extent_rec->offset != offset ||
11119                             dev_extent_rec->chunk_offset != chunk_rec->offset ||
11120                             dev_extent_rec->length != length) {
11121                                 if (!silent)
11122                                         fprintf(stderr,
11123                                                 "Chunk[%llu, %u, %llu] stripe[%llu, %llu] dismatch dev extent[%llu, %llu, %llu]\n",
11124                                                 chunk_rec->objectid,
11125                                                 chunk_rec->type,
11126                                                 chunk_rec->offset,
11127                                                 chunk_rec->stripes[i].devid,
11128                                                 chunk_rec->stripes[i].offset,
11129                                                 dev_extent_rec->objectid,
11130                                                 dev_extent_rec->offset,
11131                                                 dev_extent_rec->length);
11132                                 ret = -1;
11133                         } else {
11134                                 list_move(&dev_extent_rec->chunk_list,
11135                                           &chunk_rec->dextents);
11136                         }
11137                 } else {
11138                         if (!silent)
11139                                 fprintf(stderr,
11140                                         "Chunk[%llu, %u, %llu] stripe[%llu, %llu] is not found in dev extent\n",
11141                                         chunk_rec->objectid,
11142                                         chunk_rec->type,
11143                                         chunk_rec->offset,
11144                                         chunk_rec->stripes[i].devid,
11145                                         chunk_rec->stripes[i].offset);
11146                         ret = -1;
11147                 }
11148         }
11149         return ret;
11150 }
11151
11152 /* check btrfs_chunk -> btrfs_dev_extent / btrfs_block_group_item */
11153 int check_chunks(struct cache_tree *chunk_cache,
11154                  struct block_group_tree *block_group_cache,
11155                  struct device_extent_tree *dev_extent_cache,
11156                  struct list_head *good, struct list_head *bad,
11157                  struct list_head *rebuild, int silent)
11158 {
11159         struct cache_extent *chunk_item;
11160         struct chunk_record *chunk_rec;
11161         struct block_group_record *bg_rec;
11162         struct device_extent_record *dext_rec;
11163         int err;
11164         int ret = 0;
11165
11166         chunk_item = first_cache_extent(chunk_cache);
11167         while (chunk_item) {
11168                 chunk_rec = container_of(chunk_item, struct chunk_record,
11169                                          cache);
11170                 err = check_chunk_refs(chunk_rec, block_group_cache,
11171                                        dev_extent_cache, silent);
11172                 if (err < 0)
11173                         ret = err;
11174                 if (err == 0 && good)
11175                         list_add_tail(&chunk_rec->list, good);
11176                 if (err > 0 && rebuild)
11177                         list_add_tail(&chunk_rec->list, rebuild);
11178                 if (err < 0 && bad)
11179                         list_add_tail(&chunk_rec->list, bad);
11180                 chunk_item = next_cache_extent(chunk_item);
11181         }
11182
11183         list_for_each_entry(bg_rec, &block_group_cache->block_groups, list) {
11184                 if (!silent)
11185                         fprintf(stderr,
11186                                 "Block group[%llu, %llu] (flags = %llu) didn't find the relative chunk.\n",
11187                                 bg_rec->objectid,
11188                                 bg_rec->offset,
11189                                 bg_rec->flags);
11190                 if (!ret)
11191                         ret = 1;
11192         }
11193
11194         list_for_each_entry(dext_rec, &dev_extent_cache->no_chunk_orphans,
11195                             chunk_list) {
11196                 if (!silent)
11197                         fprintf(stderr,
11198                                 "Device extent[%llu, %llu, %llu] didn't find the relative chunk.\n",
11199                                 dext_rec->objectid,
11200                                 dext_rec->offset,
11201                                 dext_rec->length);
11202                 if (!ret)
11203                         ret = 1;
11204         }
11205         return ret;
11206 }
11207
11208
11209 static int check_device_used(struct device_record *dev_rec,
11210                              struct device_extent_tree *dext_cache)
11211 {
11212         struct cache_extent *cache;
11213         struct device_extent_record *dev_extent_rec;
11214         u64 total_byte = 0;
11215
11216         cache = search_cache_extent2(&dext_cache->tree, dev_rec->devid, 0);
11217         while (cache) {
11218                 dev_extent_rec = container_of(cache,
11219                                               struct device_extent_record,
11220                                               cache);
11221                 if (dev_extent_rec->objectid != dev_rec->devid)
11222                         break;
11223
11224                 list_del_init(&dev_extent_rec->device_list);
11225                 total_byte += dev_extent_rec->length;
11226                 cache = next_cache_extent(cache);
11227         }
11228
11229         if (total_byte != dev_rec->byte_used) {
11230                 fprintf(stderr,
11231                         "Dev extent's total-byte(%llu) is not equal to byte-used(%llu) in dev[%llu, %u, %llu]\n",
11232                         total_byte, dev_rec->byte_used, dev_rec->objectid,
11233                         dev_rec->type, dev_rec->offset);
11234                 return -1;
11235         } else {
11236                 return 0;
11237         }
11238 }
11239
11240 /*
11241  * Extra (optional) check for dev_item size to report possbile problem on a new
11242  * kernel.
11243  */
11244 static void check_dev_size_alignment(u64 devid, u64 total_bytes, u32 sectorsize)
11245 {
11246         if (!IS_ALIGNED(total_bytes, sectorsize)) {
11247                 warning(
11248 "unaligned total_bytes detected for devid %llu, have %llu should be aligned to %u",
11249                         devid, total_bytes, sectorsize);
11250                 warning(
11251 "this is OK for older kernel, but may cause kernel warning for newer kernels");
11252                 warning("this can be fixed by 'btrfs rescue fix-device-size'");
11253         }
11254 }
11255
11256 /*
11257  * Unlike device size alignment check above, some super total_bytes check
11258  * failure can lead to mount failure for newer kernel.
11259  *
11260  * So this function will return the error for a fatal super total_bytes problem.
11261  */
11262 static bool is_super_size_valid(struct btrfs_fs_info *fs_info)
11263 {
11264         struct btrfs_device *dev;
11265         struct list_head *dev_list = &fs_info->fs_devices->devices;
11266         u64 total_bytes = 0;
11267         u64 super_bytes = btrfs_super_total_bytes(fs_info->super_copy);
11268
11269         list_for_each_entry(dev, dev_list, dev_list)
11270                 total_bytes += dev->total_bytes;
11271
11272         /* Important check, which can cause unmountable fs */
11273         if (super_bytes < total_bytes) {
11274                 error("super total bytes %llu smaller than real device(s) size %llu",
11275                         super_bytes, total_bytes);
11276                 error("mounting this fs may fail for newer kernels");
11277                 error("this can be fixed by 'btrfs rescue fix-device-size'");
11278                 return false;
11279         }
11280
11281         /*
11282          * Optional check, just to make everything aligned and match with each
11283          * other.
11284          *
11285          * For a btrfs-image restored fs, we don't need to check it anyway.
11286          */
11287         if (btrfs_super_flags(fs_info->super_copy) &
11288             (BTRFS_SUPER_FLAG_METADUMP | BTRFS_SUPER_FLAG_METADUMP_V2))
11289                 return true;
11290         if (!IS_ALIGNED(super_bytes, fs_info->sectorsize) ||
11291             !IS_ALIGNED(total_bytes, fs_info->sectorsize) ||
11292             super_bytes != total_bytes) {
11293                 warning("minor unaligned/mismatch device size detected");
11294                 warning(
11295                 "recommended to use 'btrfs rescue fix-device-size' to fix it");
11296         }
11297         return true;
11298 }
11299
11300 /* check btrfs_dev_item -> btrfs_dev_extent */
11301 static int check_devices(struct rb_root *dev_cache,
11302                          struct device_extent_tree *dev_extent_cache)
11303 {
11304         struct rb_node *dev_node;
11305         struct device_record *dev_rec;
11306         struct device_extent_record *dext_rec;
11307         int err;
11308         int ret = 0;
11309
11310         dev_node = rb_first(dev_cache);
11311         while (dev_node) {
11312                 dev_rec = container_of(dev_node, struct device_record, node);
11313                 err = check_device_used(dev_rec, dev_extent_cache);
11314                 if (err)
11315                         ret = err;
11316
11317                 check_dev_size_alignment(dev_rec->devid, dev_rec->total_byte,
11318                                          global_info->sectorsize);
11319                 dev_node = rb_next(dev_node);
11320         }
11321         list_for_each_entry(dext_rec, &dev_extent_cache->no_device_orphans,
11322                             device_list) {
11323                 fprintf(stderr,
11324                         "Device extent[%llu, %llu, %llu] didn't find its device.\n",
11325                         dext_rec->objectid, dext_rec->offset, dext_rec->length);
11326                 if (!ret)
11327                         ret = 1;
11328         }
11329         return ret;
11330 }
11331
11332 static int add_root_item_to_list(struct list_head *head,
11333                                   u64 objectid, u64 bytenr, u64 last_snapshot,
11334                                   u8 level, u8 drop_level,
11335                                   struct btrfs_key *drop_key)
11336 {
11337
11338         struct root_item_record *ri_rec;
11339         ri_rec = malloc(sizeof(*ri_rec));
11340         if (!ri_rec)
11341                 return -ENOMEM;
11342         ri_rec->bytenr = bytenr;
11343         ri_rec->objectid = objectid;
11344         ri_rec->level = level;
11345         ri_rec->drop_level = drop_level;
11346         ri_rec->last_snapshot = last_snapshot;
11347         if (drop_key)
11348                 memcpy(&ri_rec->drop_key, drop_key, sizeof(*drop_key));
11349         list_add_tail(&ri_rec->list, head);
11350
11351         return 0;
11352 }
11353
11354 static void free_root_item_list(struct list_head *list)
11355 {
11356         struct root_item_record *ri_rec;
11357
11358         while (!list_empty(list)) {
11359                 ri_rec = list_first_entry(list, struct root_item_record,
11360                                           list);
11361                 list_del_init(&ri_rec->list);
11362                 free(ri_rec);
11363         }
11364 }
11365
11366 static int deal_root_from_list(struct list_head *list,
11367                                struct btrfs_root *root,
11368                                struct block_info *bits,
11369                                int bits_nr,
11370                                struct cache_tree *pending,
11371                                struct cache_tree *seen,
11372                                struct cache_tree *reada,
11373                                struct cache_tree *nodes,
11374                                struct cache_tree *extent_cache,
11375                                struct cache_tree *chunk_cache,
11376                                struct rb_root *dev_cache,
11377                                struct block_group_tree *block_group_cache,
11378                                struct device_extent_tree *dev_extent_cache)
11379 {
11380         int ret = 0;
11381         u64 last;
11382
11383         while (!list_empty(list)) {
11384                 struct root_item_record *rec;
11385                 struct extent_buffer *buf;
11386                 rec = list_entry(list->next,
11387                                  struct root_item_record, list);
11388                 last = 0;
11389                 buf = read_tree_block(root->fs_info, rec->bytenr, 0);
11390                 if (!extent_buffer_uptodate(buf)) {
11391                         free_extent_buffer(buf);
11392                         ret = -EIO;
11393                         break;
11394                 }
11395                 ret = add_root_to_pending(buf, extent_cache, pending,
11396                                     seen, nodes, rec->objectid);
11397                 if (ret < 0)
11398                         break;
11399                 /*
11400                  * To rebuild extent tree, we need deal with snapshot
11401                  * one by one, otherwise we deal with node firstly which
11402                  * can maximize readahead.
11403                  */
11404                 while (1) {
11405                         ret = run_next_block(root, bits, bits_nr, &last,
11406                                              pending, seen, reada, nodes,
11407                                              extent_cache, chunk_cache,
11408                                              dev_cache, block_group_cache,
11409                                              dev_extent_cache, rec);
11410                         if (ret != 0)
11411                                 break;
11412                 }
11413                 free_extent_buffer(buf);
11414                 list_del(&rec->list);
11415                 free(rec);
11416                 if (ret < 0)
11417                         break;
11418         }
11419         while (ret >= 0) {
11420                 ret = run_next_block(root, bits, bits_nr, &last, pending, seen,
11421                                      reada, nodes, extent_cache, chunk_cache,
11422                                      dev_cache, block_group_cache,
11423                                      dev_extent_cache, NULL);
11424                 if (ret != 0) {
11425                         if (ret > 0)
11426                                 ret = 0;
11427                         break;
11428                 }
11429         }
11430         return ret;
11431 }
11432
11433 static int check_chunks_and_extents(struct btrfs_fs_info *fs_info)
11434 {
11435         struct rb_root dev_cache;
11436         struct cache_tree chunk_cache;
11437         struct block_group_tree block_group_cache;
11438         struct device_extent_tree dev_extent_cache;
11439         struct cache_tree extent_cache;
11440         struct cache_tree seen;
11441         struct cache_tree pending;
11442         struct cache_tree reada;
11443         struct cache_tree nodes;
11444         struct extent_io_tree excluded_extents;
11445         struct cache_tree corrupt_blocks;
11446         struct btrfs_path path;
11447         struct btrfs_key key;
11448         struct btrfs_key found_key;
11449         int ret, err = 0;
11450         struct block_info *bits;
11451         int bits_nr;
11452         struct extent_buffer *leaf;
11453         int slot;
11454         struct btrfs_root_item ri;
11455         struct list_head dropping_trees;
11456         struct list_head normal_trees;
11457         struct btrfs_root *root1;
11458         struct btrfs_root *root;
11459         u64 objectid;
11460         u8 level;
11461
11462         root = fs_info->fs_root;
11463         dev_cache = RB_ROOT;
11464         cache_tree_init(&chunk_cache);
11465         block_group_tree_init(&block_group_cache);
11466         device_extent_tree_init(&dev_extent_cache);
11467
11468         cache_tree_init(&extent_cache);
11469         cache_tree_init(&seen);
11470         cache_tree_init(&pending);
11471         cache_tree_init(&nodes);
11472         cache_tree_init(&reada);
11473         cache_tree_init(&corrupt_blocks);
11474         extent_io_tree_init(&excluded_extents);
11475         INIT_LIST_HEAD(&dropping_trees);
11476         INIT_LIST_HEAD(&normal_trees);
11477
11478         if (repair) {
11479                 fs_info->excluded_extents = &excluded_extents;
11480                 fs_info->fsck_extent_cache = &extent_cache;
11481                 fs_info->free_extent_hook = free_extent_hook;
11482                 fs_info->corrupt_blocks = &corrupt_blocks;
11483         }
11484
11485         bits_nr = 1024;
11486         bits = malloc(bits_nr * sizeof(struct block_info));
11487         if (!bits) {
11488                 perror("malloc");
11489                 exit(1);
11490         }
11491
11492         if (ctx.progress_enabled) {
11493                 ctx.tp = TASK_EXTENTS;
11494                 task_start(ctx.info);
11495         }
11496
11497 again:
11498         root1 = fs_info->tree_root;
11499         level = btrfs_header_level(root1->node);
11500         ret = add_root_item_to_list(&normal_trees, root1->root_key.objectid,
11501                                     root1->node->start, 0, level, 0, NULL);
11502         if (ret < 0)
11503                 goto out;
11504         root1 = fs_info->chunk_root;
11505         level = btrfs_header_level(root1->node);
11506         ret = add_root_item_to_list(&normal_trees, root1->root_key.objectid,
11507                                     root1->node->start, 0, level, 0, NULL);
11508         if (ret < 0)
11509                 goto out;
11510         btrfs_init_path(&path);
11511         key.offset = 0;
11512         key.objectid = 0;
11513         key.type = BTRFS_ROOT_ITEM_KEY;
11514         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, &path, 0, 0);
11515         if (ret < 0)
11516                 goto out;
11517         while(1) {
11518                 leaf = path.nodes[0];
11519                 slot = path.slots[0];
11520                 if (slot >= btrfs_header_nritems(path.nodes[0])) {
11521                         ret = btrfs_next_leaf(root, &path);
11522                         if (ret != 0)
11523                                 break;
11524                         leaf = path.nodes[0];
11525                         slot = path.slots[0];
11526                 }
11527                 btrfs_item_key_to_cpu(leaf, &found_key, path.slots[0]);
11528                 if (found_key.type == BTRFS_ROOT_ITEM_KEY) {
11529                         unsigned long offset;
11530                         u64 last_snapshot;
11531
11532                         offset = btrfs_item_ptr_offset(leaf, path.slots[0]);
11533                         read_extent_buffer(leaf, &ri, offset, sizeof(ri));
11534                         last_snapshot = btrfs_root_last_snapshot(&ri);
11535                         if (btrfs_disk_key_objectid(&ri.drop_progress) == 0) {
11536                                 level = btrfs_root_level(&ri);
11537                                 ret = add_root_item_to_list(&normal_trees,
11538                                                 found_key.objectid,
11539                                                 btrfs_root_bytenr(&ri),
11540                                                 last_snapshot, level,
11541                                                 0, NULL);
11542                                 if (ret < 0)
11543                                         goto out;
11544                         } else {
11545                                 level = btrfs_root_level(&ri);
11546                                 objectid = found_key.objectid;
11547                                 btrfs_disk_key_to_cpu(&found_key,
11548                                                       &ri.drop_progress);
11549                                 ret = add_root_item_to_list(&dropping_trees,
11550                                                 objectid,
11551                                                 btrfs_root_bytenr(&ri),
11552                                                 last_snapshot, level,
11553                                                 ri.drop_level, &found_key);
11554                                 if (ret < 0)
11555                                         goto out;
11556                         }
11557                 }
11558                 path.slots[0]++;
11559         }
11560         btrfs_release_path(&path);
11561
11562         /*
11563          * check_block can return -EAGAIN if it fixes something, please keep
11564          * this in mind when dealing with return values from these functions, if
11565          * we get -EAGAIN we want to fall through and restart the loop.
11566          */
11567         ret = deal_root_from_list(&normal_trees, root, bits, bits_nr, &pending,
11568                                   &seen, &reada, &nodes, &extent_cache,
11569                                   &chunk_cache, &dev_cache, &block_group_cache,
11570                                   &dev_extent_cache);
11571         if (ret < 0) {
11572                 if (ret == -EAGAIN)
11573                         goto loop;
11574                 goto out;
11575         }
11576         ret = deal_root_from_list(&dropping_trees, root, bits, bits_nr,
11577                                   &pending, &seen, &reada, &nodes,
11578                                   &extent_cache, &chunk_cache, &dev_cache,
11579                                   &block_group_cache, &dev_extent_cache);
11580         if (ret < 0) {
11581                 if (ret == -EAGAIN)
11582                         goto loop;
11583                 goto out;
11584         }
11585
11586         ret = check_chunks(&chunk_cache, &block_group_cache,
11587                            &dev_extent_cache, NULL, NULL, NULL, 0);
11588         if (ret) {
11589                 if (ret == -EAGAIN)
11590                         goto loop;
11591                 err = ret;
11592         }
11593
11594         ret = check_extent_refs(root, &extent_cache);
11595         if (ret < 0) {
11596                 if (ret == -EAGAIN)
11597                         goto loop;
11598                 goto out;
11599         }
11600
11601         ret = check_devices(&dev_cache, &dev_extent_cache);
11602         if (ret && err)
11603                 ret = err;
11604
11605 out:
11606         task_stop(ctx.info);
11607         if (repair) {
11608                 free_corrupt_blocks_tree(fs_info->corrupt_blocks);
11609                 extent_io_tree_cleanup(&excluded_extents);
11610                 fs_info->fsck_extent_cache = NULL;
11611                 fs_info->free_extent_hook = NULL;
11612                 fs_info->corrupt_blocks = NULL;
11613                 fs_info->excluded_extents = NULL;
11614         }
11615         free(bits);
11616         free_chunk_cache_tree(&chunk_cache);
11617         free_device_cache_tree(&dev_cache);
11618         free_block_group_tree(&block_group_cache);
11619         free_device_extent_tree(&dev_extent_cache);
11620         free_extent_cache_tree(&seen);
11621         free_extent_cache_tree(&pending);
11622         free_extent_cache_tree(&reada);
11623         free_extent_cache_tree(&nodes);
11624         free_root_item_list(&normal_trees);
11625         free_root_item_list(&dropping_trees);
11626         return ret;
11627 loop:
11628         free_corrupt_blocks_tree(fs_info->corrupt_blocks);
11629         free_extent_cache_tree(&seen);
11630         free_extent_cache_tree(&pending);
11631         free_extent_cache_tree(&reada);
11632         free_extent_cache_tree(&nodes);
11633         free_chunk_cache_tree(&chunk_cache);
11634         free_block_group_tree(&block_group_cache);
11635         free_device_cache_tree(&dev_cache);
11636         free_device_extent_tree(&dev_extent_cache);
11637         free_extent_record_cache(&extent_cache);
11638         free_root_item_list(&normal_trees);
11639         free_root_item_list(&dropping_trees);
11640         extent_io_tree_cleanup(&excluded_extents);
11641         goto again;
11642 }
11643
11644 static int check_extent_inline_ref(struct extent_buffer *eb,
11645                    struct btrfs_key *key, struct btrfs_extent_inline_ref *iref)
11646 {
11647         int ret;
11648         u8 type = btrfs_extent_inline_ref_type(eb, iref);
11649
11650         switch (type) {
11651         case BTRFS_TREE_BLOCK_REF_KEY:
11652         case BTRFS_EXTENT_DATA_REF_KEY:
11653         case BTRFS_SHARED_BLOCK_REF_KEY:
11654         case BTRFS_SHARED_DATA_REF_KEY:
11655                 ret = 0;
11656                 break;
11657         default:
11658                 error("extent[%llu %u %llu] has unknown ref type: %d",
11659                       key->objectid, key->type, key->offset, type);
11660                 ret = UNKNOWN_TYPE;
11661                 break;
11662         }
11663
11664         return ret;
11665 }
11666
11667 /*
11668  * Check backrefs of a tree block given by @bytenr or @eb.
11669  *
11670  * @root:       the root containing the @bytenr or @eb
11671  * @eb:         tree block extent buffer, can be NULL
11672  * @bytenr:     bytenr of the tree block to search
11673  * @level:      tree level of the tree block
11674  * @owner:      owner of the tree block
11675  *
11676  * Return >0 for any error found and output error message
11677  * Return 0 for no error found
11678  */
11679 static int check_tree_block_ref(struct btrfs_root *root,
11680                                 struct extent_buffer *eb, u64 bytenr,
11681                                 int level, u64 owner, struct node_refs *nrefs)
11682 {
11683         struct btrfs_key key;
11684         struct btrfs_root *extent_root = root->fs_info->extent_root;
11685         struct btrfs_path path;
11686         struct btrfs_extent_item *ei;
11687         struct btrfs_extent_inline_ref *iref;
11688         struct extent_buffer *leaf;
11689         unsigned long end;
11690         unsigned long ptr;
11691         int slot;
11692         int skinny_level;
11693         int root_level = btrfs_header_level(root->node);
11694         int type;
11695         u32 nodesize = root->fs_info->nodesize;
11696         u32 item_size;
11697         u64 offset;
11698         int tree_reloc_root = 0;
11699         int found_ref = 0;
11700         int err = 0;
11701         int ret;
11702         int strict = 1;
11703         int parent = 0;
11704
11705         if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID &&
11706             btrfs_header_bytenr(root->node) == bytenr)
11707                 tree_reloc_root = 1;
11708         btrfs_init_path(&path);
11709         key.objectid = bytenr;
11710         if (btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
11711                 key.type = BTRFS_METADATA_ITEM_KEY;
11712         else
11713                 key.type = BTRFS_EXTENT_ITEM_KEY;
11714         key.offset = (u64)-1;
11715
11716         /* Search for the backref in extent tree */
11717         ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
11718         if (ret < 0) {
11719                 err |= BACKREF_MISSING;
11720                 goto out;
11721         }
11722         ret = btrfs_previous_extent_item(extent_root, &path, bytenr);
11723         if (ret) {
11724                 err |= BACKREF_MISSING;
11725                 goto out;
11726         }
11727
11728         leaf = path.nodes[0];
11729         slot = path.slots[0];
11730         btrfs_item_key_to_cpu(leaf, &key, slot);
11731
11732         ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
11733
11734         if (key.type == BTRFS_METADATA_ITEM_KEY) {
11735                 skinny_level = (int)key.offset;
11736                 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
11737         } else {
11738                 struct btrfs_tree_block_info *info;
11739
11740                 info = (struct btrfs_tree_block_info *)(ei + 1);
11741                 skinny_level = btrfs_tree_block_level(leaf, info);
11742                 iref = (struct btrfs_extent_inline_ref *)(info + 1);
11743         }
11744
11745
11746         if (eb) {
11747                 u64 header_gen;
11748                 u64 extent_gen;
11749
11750                 /*
11751                  * Due to the feature of shared tree blocks, if the upper node
11752                  * is a fs root or shared node, the extent of checked node may
11753                  * not be updated until the next CoW.
11754                  */
11755                 if (nrefs)
11756                         strict = should_check_extent_strictly(root, nrefs,
11757                                         level);
11758                 if (!(btrfs_extent_flags(leaf, ei) &
11759                       BTRFS_EXTENT_FLAG_TREE_BLOCK)) {
11760                         error(
11761                 "extent[%llu %u] backref type mismatch, missing bit: %llx",
11762                                 key.objectid, nodesize,
11763                                 BTRFS_EXTENT_FLAG_TREE_BLOCK);
11764                         err = BACKREF_MISMATCH;
11765                 }
11766                 header_gen = btrfs_header_generation(eb);
11767                 extent_gen = btrfs_extent_generation(leaf, ei);
11768                 if (header_gen != extent_gen) {
11769                         error(
11770         "extent[%llu %u] backref generation mismatch, wanted: %llu, have: %llu",
11771                                 key.objectid, nodesize, header_gen,
11772                                 extent_gen);
11773                         err = BACKREF_MISMATCH;
11774                 }
11775                 if (level != skinny_level) {
11776                         error(
11777                         "extent[%llu %u] level mismatch, wanted: %u, have: %u",
11778                                 key.objectid, nodesize, level, skinny_level);
11779                         err = BACKREF_MISMATCH;
11780                 }
11781                 if (!is_fstree(owner) && btrfs_extent_refs(leaf, ei) != 1) {
11782                         error(
11783                         "extent[%llu %u] is referred by other roots than %llu",
11784                                 key.objectid, nodesize, root->objectid);
11785                         err = BACKREF_MISMATCH;
11786                 }
11787         }
11788
11789         /*
11790          * Iterate the extent/metadata item to find the exact backref
11791          */
11792         item_size = btrfs_item_size_nr(leaf, slot);
11793         ptr = (unsigned long)iref;
11794         end = (unsigned long)ei + item_size;
11795
11796         while (ptr < end) {
11797                 iref = (struct btrfs_extent_inline_ref *)ptr;
11798                 type = btrfs_extent_inline_ref_type(leaf, iref);
11799                 offset = btrfs_extent_inline_ref_offset(leaf, iref);
11800
11801                 ret = check_extent_inline_ref(leaf, &key, iref);
11802                 if (ret) {
11803                         err |= ret;
11804                         break;
11805                 }
11806                 if (type == BTRFS_TREE_BLOCK_REF_KEY) {
11807                         if (offset == root->objectid)
11808                                 found_ref = 1;
11809                         if (!strict && owner == offset)
11810                                 found_ref = 1;
11811                 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
11812                         /*
11813                          * Backref of tree reloc root points to itself, no need
11814                          * to check backref any more.
11815                          */
11816                         if (tree_reloc_root) {
11817                                 found_ref = 1;
11818                         } else {
11819                                 /*
11820                                  * Check if the backref points to valid
11821                                  * referencer
11822                                  */
11823                                 found_ref = !check_tree_block_ref( root, NULL,
11824                                                 offset, level + 1, owner,
11825                                                 NULL);
11826                         }
11827                 }
11828
11829                 if (found_ref)
11830                         break;
11831                 ptr += btrfs_extent_inline_ref_size(type);
11832         }
11833
11834         /*
11835          * Inlined extent item doesn't have what we need, check
11836          * TREE_BLOCK_REF_KEY
11837          */
11838         if (!found_ref) {
11839                 btrfs_release_path(&path);
11840                 key.objectid = bytenr;
11841                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
11842                 key.offset = root->objectid;
11843
11844                 ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
11845                 if (!ret)
11846                         found_ref = 1;
11847         }
11848         if (!found_ref)
11849                 err |= BACKREF_MISSING;
11850 out:
11851         btrfs_release_path(&path);
11852         if (nrefs && strict &&
11853             level < root_level && nrefs->full_backref[level + 1])
11854                 parent = nrefs->bytenr[level + 1];
11855         if (eb && (err & BACKREF_MISSING))
11856                 error(
11857         "extent[%llu %u] backref lost (owner: %llu, level: %u) %s %llu",
11858                       bytenr, nodesize, owner, level,
11859                       parent ? "parent" : "root",
11860                       parent ? parent : root->objectid);
11861         return err;
11862 }
11863
11864 /*
11865  * If @err contains BACKREF_MISSING then add extent of the
11866  * file_extent_data_item.
11867  *
11868  * Returns error bits after reapir.
11869  */
11870 static int repair_extent_data_item(struct btrfs_trans_handle *trans,
11871                                    struct btrfs_root *root,
11872                                    struct btrfs_path *pathp,
11873                                    struct node_refs *nrefs,
11874                                    int err)
11875 {
11876         struct btrfs_file_extent_item *fi;
11877         struct btrfs_key fi_key;
11878         struct btrfs_key key;
11879         struct btrfs_extent_item *ei;
11880         struct btrfs_path path;
11881         struct btrfs_root *extent_root = root->fs_info->extent_root;
11882         struct extent_buffer *eb;
11883         u64 size;
11884         u64 disk_bytenr;
11885         u64 num_bytes;
11886         u64 parent;
11887         u64 offset;
11888         u64 extent_offset;
11889         u64 file_offset;
11890         int generation;
11891         int slot;
11892         int ret = 0;
11893
11894         eb = pathp->nodes[0];
11895         slot = pathp->slots[0];
11896         btrfs_item_key_to_cpu(eb, &fi_key, slot);
11897         fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
11898
11899         if (btrfs_file_extent_type(eb, fi) == BTRFS_FILE_EXTENT_INLINE ||
11900             btrfs_file_extent_disk_bytenr(eb, fi) == 0)
11901                 return err;
11902
11903         file_offset = fi_key.offset;
11904         generation = btrfs_file_extent_generation(eb, fi);
11905         disk_bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
11906         num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
11907         extent_offset = btrfs_file_extent_offset(eb, fi);
11908         offset = file_offset - extent_offset;
11909
11910         /* now repair only adds backref */
11911         if ((err & BACKREF_MISSING) == 0)
11912                 return err;
11913
11914         /* search extent item */
11915         key.objectid = disk_bytenr;
11916         key.type = BTRFS_EXTENT_ITEM_KEY;
11917         key.offset = num_bytes;
11918
11919         btrfs_init_path(&path);
11920         ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
11921         if (ret < 0) {
11922                 ret = -EIO;
11923                 goto out;
11924         }
11925
11926         /* insert an extent item */
11927         if (ret > 0) {
11928                 key.objectid = disk_bytenr;
11929                 key.type = BTRFS_EXTENT_ITEM_KEY;
11930                 key.offset = num_bytes;
11931                 size = sizeof(*ei);
11932
11933                 btrfs_release_path(&path);
11934                 ret = btrfs_insert_empty_item(trans, extent_root, &path, &key,
11935                                               size);
11936                 if (ret)
11937                         goto out;
11938                 eb = path.nodes[0];
11939                 ei = btrfs_item_ptr(eb, path.slots[0], struct btrfs_extent_item);
11940
11941                 btrfs_set_extent_refs(eb, ei, 0);
11942                 btrfs_set_extent_generation(eb, ei, generation);
11943                 btrfs_set_extent_flags(eb, ei, BTRFS_EXTENT_FLAG_DATA);
11944
11945                 btrfs_mark_buffer_dirty(eb);
11946                 ret = btrfs_update_block_group(trans, extent_root, disk_bytenr,
11947                                                num_bytes, 1, 0);
11948                 btrfs_release_path(&path);
11949         }
11950
11951         if (nrefs->full_backref[0])
11952                 parent = btrfs_header_bytenr(eb);
11953         else
11954                 parent = 0;
11955
11956         ret = btrfs_inc_extent_ref(trans, root, disk_bytenr, num_bytes, parent,
11957                                    root->objectid,
11958                    parent ? BTRFS_FIRST_FREE_OBJECTID : fi_key.objectid,
11959                                    offset);
11960         if (ret) {
11961                 error(
11962                 "failed to increase extent data backref[%llu %llu] root %llu",
11963                       disk_bytenr, num_bytes, root->objectid);
11964                 goto out;
11965         } else {
11966                 printf("Add one extent data backref [%llu %llu]\n",
11967                        disk_bytenr, num_bytes);
11968         }
11969
11970         err &= ~BACKREF_MISSING;
11971 out:
11972         if (ret)
11973                 error("can't repair root %llu extent data item[%llu %llu]",
11974                       root->objectid, disk_bytenr, num_bytes);
11975         return err;
11976 }
11977
11978 /*
11979  * Check EXTENT_DATA item, mainly for its dbackref in extent tree
11980  *
11981  * Return >0 any error found and output error message
11982  * Return 0 for no error found
11983  */
11984 static int check_extent_data_item(struct btrfs_root *root,
11985                                   struct btrfs_path *pathp,
11986                                   struct node_refs *nrefs,  int account_bytes)
11987 {
11988         struct btrfs_file_extent_item *fi;
11989         struct extent_buffer *eb = pathp->nodes[0];
11990         struct btrfs_path path;
11991         struct btrfs_root *extent_root = root->fs_info->extent_root;
11992         struct btrfs_key fi_key;
11993         struct btrfs_key dbref_key;
11994         struct extent_buffer *leaf;
11995         struct btrfs_extent_item *ei;
11996         struct btrfs_extent_inline_ref *iref;
11997         struct btrfs_extent_data_ref *dref;
11998         u64 owner;
11999         u64 disk_bytenr;
12000         u64 disk_num_bytes;
12001         u64 extent_num_bytes;
12002         u64 extent_flags;
12003         u32 item_size;
12004         unsigned long end;
12005         unsigned long ptr;
12006         int type;
12007         u64 ref_root;
12008         int found_dbackref = 0;
12009         int slot = pathp->slots[0];
12010         int err = 0;
12011         int ret;
12012         int strict;
12013
12014         btrfs_item_key_to_cpu(eb, &fi_key, slot);
12015         fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
12016
12017         /* Nothing to check for hole and inline data extents */
12018         if (btrfs_file_extent_type(eb, fi) == BTRFS_FILE_EXTENT_INLINE ||
12019             btrfs_file_extent_disk_bytenr(eb, fi) == 0)
12020                 return 0;
12021
12022         disk_bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
12023         disk_num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
12024         extent_num_bytes = btrfs_file_extent_num_bytes(eb, fi);
12025
12026         /* Check unaligned disk_num_bytes and num_bytes */
12027         if (!IS_ALIGNED(disk_num_bytes, root->fs_info->sectorsize)) {
12028                 error(
12029 "file extent [%llu, %llu] has unaligned disk num bytes: %llu, should be aligned to %u",
12030                         fi_key.objectid, fi_key.offset, disk_num_bytes,
12031                         root->fs_info->sectorsize);
12032                 err |= BYTES_UNALIGNED;
12033         } else if (account_bytes) {
12034                 data_bytes_allocated += disk_num_bytes;
12035         }
12036         if (!IS_ALIGNED(extent_num_bytes, root->fs_info->sectorsize)) {
12037                 error(
12038 "file extent [%llu, %llu] has unaligned num bytes: %llu, should be aligned to %u",
12039                         fi_key.objectid, fi_key.offset, extent_num_bytes,
12040                         root->fs_info->sectorsize);
12041                 err |= BYTES_UNALIGNED;
12042         } else if (account_bytes) {
12043                 data_bytes_referenced += extent_num_bytes;
12044         }
12045         owner = btrfs_header_owner(eb);
12046
12047         /* Check the extent item of the file extent in extent tree */
12048         btrfs_init_path(&path);
12049         dbref_key.objectid = btrfs_file_extent_disk_bytenr(eb, fi);
12050         dbref_key.type = BTRFS_EXTENT_ITEM_KEY;
12051         dbref_key.offset = btrfs_file_extent_disk_num_bytes(eb, fi);
12052
12053         ret = btrfs_search_slot(NULL, extent_root, &dbref_key, &path, 0, 0);
12054         if (ret)
12055                 goto out;
12056
12057         leaf = path.nodes[0];
12058         slot = path.slots[0];
12059         ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
12060
12061         extent_flags = btrfs_extent_flags(leaf, ei);
12062
12063         if (!(extent_flags & BTRFS_EXTENT_FLAG_DATA)) {
12064                 error(
12065                     "extent[%llu %llu] backref type mismatch, wanted bit: %llx",
12066                     disk_bytenr, disk_num_bytes,
12067                     BTRFS_EXTENT_FLAG_DATA);
12068                 err |= BACKREF_MISMATCH;
12069         }
12070
12071         /* Check data backref inside that extent item */
12072         item_size = btrfs_item_size_nr(leaf, path.slots[0]);
12073         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
12074         ptr = (unsigned long)iref;
12075         end = (unsigned long)ei + item_size;
12076         strict = should_check_extent_strictly(root, nrefs, -1);
12077
12078         while (ptr < end) {
12079                 iref = (struct btrfs_extent_inline_ref *)ptr;
12080                 type = btrfs_extent_inline_ref_type(leaf, iref);
12081                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
12082
12083                 ret = check_extent_inline_ref(leaf, &dbref_key, iref);
12084                 if (ret) {
12085                         err |= ret;
12086                         break;
12087                 }
12088                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
12089                         ref_root = btrfs_extent_data_ref_root(leaf, dref);
12090                         if (ref_root == root->objectid)
12091                                 found_dbackref = 1;
12092                         else if (!strict && owner == ref_root)
12093                                 found_dbackref = 1;
12094                 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
12095                         found_dbackref = !check_tree_block_ref(root, NULL,
12096                                 btrfs_extent_inline_ref_offset(leaf, iref),
12097                                 0, owner, NULL);
12098                 }
12099
12100                 if (found_dbackref)
12101                         break;
12102                 ptr += btrfs_extent_inline_ref_size(type);
12103         }
12104
12105         if (!found_dbackref) {
12106                 btrfs_release_path(&path);
12107
12108                 /* Didn't find inlined data backref, try EXTENT_DATA_REF_KEY */
12109                 dbref_key.objectid = btrfs_file_extent_disk_bytenr(eb, fi);
12110                 dbref_key.type = BTRFS_EXTENT_DATA_REF_KEY;
12111                 dbref_key.offset = hash_extent_data_ref(root->objectid,
12112                                 fi_key.objectid, fi_key.offset);
12113
12114                 ret = btrfs_search_slot(NULL, root->fs_info->extent_root,
12115                                         &dbref_key, &path, 0, 0);
12116                 if (!ret) {
12117                         found_dbackref = 1;
12118                         goto out;
12119                 }
12120
12121                 btrfs_release_path(&path);
12122
12123                 /*
12124                  * Neither inlined nor EXTENT_DATA_REF found, try
12125                  * SHARED_DATA_REF as last chance.
12126                  */
12127                 dbref_key.objectid = disk_bytenr;
12128                 dbref_key.type = BTRFS_SHARED_DATA_REF_KEY;
12129                 dbref_key.offset = eb->start;
12130
12131                 ret = btrfs_search_slot(NULL, root->fs_info->extent_root,
12132                                         &dbref_key, &path, 0, 0);
12133                 if (!ret) {
12134                         found_dbackref = 1;
12135                         goto out;
12136                 }
12137         }
12138
12139 out:
12140         if (!found_dbackref)
12141                 err |= BACKREF_MISSING;
12142         btrfs_release_path(&path);
12143         if (err & BACKREF_MISSING) {
12144                 error("data extent[%llu %llu] backref lost",
12145                       disk_bytenr, disk_num_bytes);
12146         }
12147         return err;
12148 }
12149
12150 /*
12151  * Get real tree block level for the case like shared block
12152  * Return >= 0 as tree level
12153  * Return <0 for error
12154  */
12155 static int query_tree_block_level(struct btrfs_fs_info *fs_info, u64 bytenr)
12156 {
12157         struct extent_buffer *eb;
12158         struct btrfs_path path;
12159         struct btrfs_key key;
12160         struct btrfs_extent_item *ei;
12161         u64 flags;
12162         u64 transid;
12163         u8 backref_level;
12164         u8 header_level;
12165         int ret;
12166
12167         /* Search extent tree for extent generation and level */
12168         key.objectid = bytenr;
12169         key.type = BTRFS_METADATA_ITEM_KEY;
12170         key.offset = (u64)-1;
12171
12172         btrfs_init_path(&path);
12173         ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, &path, 0, 0);
12174         if (ret < 0)
12175                 goto release_out;
12176         ret = btrfs_previous_extent_item(fs_info->extent_root, &path, bytenr);
12177         if (ret < 0)
12178                 goto release_out;
12179         if (ret > 0) {
12180                 ret = -ENOENT;
12181                 goto release_out;
12182         }
12183
12184         btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
12185         ei = btrfs_item_ptr(path.nodes[0], path.slots[0],
12186                             struct btrfs_extent_item);
12187         flags = btrfs_extent_flags(path.nodes[0], ei);
12188         if (!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) {
12189                 ret = -ENOENT;
12190                 goto release_out;
12191         }
12192
12193         /* Get transid for later read_tree_block() check */
12194         transid = btrfs_extent_generation(path.nodes[0], ei);
12195
12196         /* Get backref level as one source */
12197         if (key.type == BTRFS_METADATA_ITEM_KEY) {
12198                 backref_level = key.offset;
12199         } else {
12200                 struct btrfs_tree_block_info *info;
12201
12202                 info = (struct btrfs_tree_block_info *)(ei + 1);
12203                 backref_level = btrfs_tree_block_level(path.nodes[0], info);
12204         }
12205         btrfs_release_path(&path);
12206
12207         /* Get level from tree block as an alternative source */
12208         eb = read_tree_block(fs_info, bytenr, transid);
12209         if (!extent_buffer_uptodate(eb)) {
12210                 free_extent_buffer(eb);
12211                 return -EIO;
12212         }
12213         header_level = btrfs_header_level(eb);
12214         free_extent_buffer(eb);
12215
12216         if (header_level != backref_level)
12217                 return -EIO;
12218         return header_level;
12219
12220 release_out:
12221         btrfs_release_path(&path);
12222         return ret;
12223 }
12224
12225 /*
12226  * Check if a tree block backref is valid (points to a valid tree block)
12227  * if level == -1, level will be resolved
12228  * Return >0 for any error found and print error message
12229  */
12230 static int check_tree_block_backref(struct btrfs_fs_info *fs_info, u64 root_id,
12231                                     u64 bytenr, int level)
12232 {
12233         struct btrfs_root *root;
12234         struct btrfs_key key;
12235         struct btrfs_path path;
12236         struct extent_buffer *eb;
12237         struct extent_buffer *node;
12238         u32 nodesize = btrfs_super_nodesize(fs_info->super_copy);
12239         int err = 0;
12240         int ret;
12241
12242         /* Query level for level == -1 special case */
12243         if (level == -1)
12244                 level = query_tree_block_level(fs_info, bytenr);
12245         if (level < 0) {
12246                 err |= REFERENCER_MISSING;
12247                 goto out;
12248         }
12249
12250         key.objectid = root_id;
12251         key.type = BTRFS_ROOT_ITEM_KEY;
12252         key.offset = (u64)-1;
12253
12254         root = btrfs_read_fs_root(fs_info, &key);
12255         if (IS_ERR(root)) {
12256                 err |= REFERENCER_MISSING;
12257                 goto out;
12258         }
12259
12260         /* Read out the tree block to get item/node key */
12261         eb = read_tree_block(fs_info, bytenr, 0);
12262         if (!extent_buffer_uptodate(eb)) {
12263                 err |= REFERENCER_MISSING;
12264                 free_extent_buffer(eb);
12265                 goto out;
12266         }
12267
12268         /* Empty tree, no need to check key */
12269         if (!btrfs_header_nritems(eb) && !level) {
12270                 free_extent_buffer(eb);
12271                 goto out;
12272         }
12273
12274         if (level)
12275                 btrfs_node_key_to_cpu(eb, &key, 0);
12276         else
12277                 btrfs_item_key_to_cpu(eb, &key, 0);
12278
12279         free_extent_buffer(eb);
12280
12281         btrfs_init_path(&path);
12282         path.lowest_level = level;
12283         /* Search with the first key, to ensure we can reach it */
12284         ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
12285         if (ret < 0) {
12286                 err |= REFERENCER_MISSING;
12287                 goto release_out;
12288         }
12289
12290         node = path.nodes[level];
12291         if (btrfs_header_bytenr(node) != bytenr) {
12292                 error(
12293         "extent [%llu %d] referencer bytenr mismatch, wanted: %llu, have: %llu",
12294                         bytenr, nodesize, bytenr,
12295                         btrfs_header_bytenr(node));
12296                 err |= REFERENCER_MISMATCH;
12297         }
12298         if (btrfs_header_level(node) != level) {
12299                 error(
12300         "extent [%llu %d] referencer level mismatch, wanted: %d, have: %d",
12301                         bytenr, nodesize, level,
12302                         btrfs_header_level(node));
12303                 err |= REFERENCER_MISMATCH;
12304         }
12305
12306 release_out:
12307         btrfs_release_path(&path);
12308 out:
12309         if (err & REFERENCER_MISSING) {
12310                 if (level < 0)
12311                         error("extent [%llu %d] lost referencer (owner: %llu)",
12312                                 bytenr, nodesize, root_id);
12313                 else
12314                         error(
12315                 "extent [%llu %d] lost referencer (owner: %llu, level: %u)",
12316                                 bytenr, nodesize, root_id, level);
12317         }
12318
12319         return err;
12320 }
12321
12322 /*
12323  * Check if tree block @eb is tree reloc root.
12324  * Return 0 if it's not or any problem happens
12325  * Return 1 if it's a tree reloc root
12326  */
12327 static int is_tree_reloc_root(struct btrfs_fs_info *fs_info,
12328                                  struct extent_buffer *eb)
12329 {
12330         struct btrfs_root *tree_reloc_root;
12331         struct btrfs_key key;
12332         u64 bytenr = btrfs_header_bytenr(eb);
12333         u64 owner = btrfs_header_owner(eb);
12334         int ret = 0;
12335
12336         key.objectid = BTRFS_TREE_RELOC_OBJECTID;
12337         key.offset = owner;
12338         key.type = BTRFS_ROOT_ITEM_KEY;
12339
12340         tree_reloc_root = btrfs_read_fs_root_no_cache(fs_info, &key);
12341         if (IS_ERR(tree_reloc_root))
12342                 return 0;
12343
12344         if (bytenr == btrfs_header_bytenr(tree_reloc_root->node))
12345                 ret = 1;
12346         btrfs_free_fs_root(tree_reloc_root);
12347         return ret;
12348 }
12349
12350 /*
12351  * Check referencer for shared block backref
12352  * If level == -1, this function will resolve the level.
12353  */
12354 static int check_shared_block_backref(struct btrfs_fs_info *fs_info,
12355                                      u64 parent, u64 bytenr, int level)
12356 {
12357         struct extent_buffer *eb;
12358         u32 nr;
12359         int found_parent = 0;
12360         int i;
12361
12362         eb = read_tree_block(fs_info, parent, 0);
12363         if (!extent_buffer_uptodate(eb))
12364                 goto out;
12365
12366         if (level == -1)
12367                 level = query_tree_block_level(fs_info, bytenr);
12368         if (level < 0)
12369                 goto out;
12370
12371         /* It's possible it's a tree reloc root */
12372         if (parent == bytenr) {
12373                 if (is_tree_reloc_root(fs_info, eb))
12374                         found_parent = 1;
12375                 goto out;
12376         }
12377
12378         if (level + 1 != btrfs_header_level(eb))
12379                 goto out;
12380
12381         nr = btrfs_header_nritems(eb);
12382         for (i = 0; i < nr; i++) {
12383                 if (bytenr == btrfs_node_blockptr(eb, i)) {
12384                         found_parent = 1;
12385                         break;
12386                 }
12387         }
12388 out:
12389         free_extent_buffer(eb);
12390         if (!found_parent) {
12391                 error(
12392         "shared extent[%llu %u] lost its parent (parent: %llu, level: %u)",
12393                         bytenr, fs_info->nodesize, parent, level);
12394                 return REFERENCER_MISSING;
12395         }
12396         return 0;
12397 }
12398
12399 /*
12400  * Check referencer for normal (inlined) data ref
12401  * If len == 0, it will be resolved by searching in extent tree
12402  */
12403 static int check_extent_data_backref(struct btrfs_fs_info *fs_info,
12404                                      u64 root_id, u64 objectid, u64 offset,
12405                                      u64 bytenr, u64 len, u32 count)
12406 {
12407         struct btrfs_root *root;
12408         struct btrfs_root *extent_root = fs_info->extent_root;
12409         struct btrfs_key key;
12410         struct btrfs_path path;
12411         struct extent_buffer *leaf;
12412         struct btrfs_file_extent_item *fi;
12413         u32 found_count = 0;
12414         int slot;
12415         int ret = 0;
12416
12417         if (!len) {
12418                 key.objectid = bytenr;
12419                 key.type = BTRFS_EXTENT_ITEM_KEY;
12420                 key.offset = (u64)-1;
12421
12422                 btrfs_init_path(&path);
12423                 ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
12424                 if (ret < 0)
12425                         goto out;
12426                 ret = btrfs_previous_extent_item(extent_root, &path, bytenr);
12427                 if (ret)
12428                         goto out;
12429                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
12430                 if (key.objectid != bytenr ||
12431                     key.type != BTRFS_EXTENT_ITEM_KEY)
12432                         goto out;
12433                 len = key.offset;
12434                 btrfs_release_path(&path);
12435         }
12436         key.objectid = root_id;
12437         key.type = BTRFS_ROOT_ITEM_KEY;
12438         key.offset = (u64)-1;
12439         btrfs_init_path(&path);
12440
12441         root = btrfs_read_fs_root(fs_info, &key);
12442         if (IS_ERR(root))
12443                 goto out;
12444
12445         key.objectid = objectid;
12446         key.type = BTRFS_EXTENT_DATA_KEY;
12447         /*
12448          * It can be nasty as data backref offset is
12449          * file offset - file extent offset, which is smaller or
12450          * equal to original backref offset.  The only special case is
12451          * overflow.  So we need to special check and do further search.
12452          */
12453         key.offset = offset & (1ULL << 63) ? 0 : offset;
12454
12455         ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
12456         if (ret < 0)
12457                 goto out;
12458
12459         /*
12460          * Search afterwards to get correct one
12461          * NOTE: As we must do a comprehensive check on the data backref to
12462          * make sure the dref count also matches, we must iterate all file
12463          * extents for that inode.
12464          */
12465         while (1) {
12466                 leaf = path.nodes[0];
12467                 slot = path.slots[0];
12468
12469                 if (slot >= btrfs_header_nritems(leaf) ||
12470                     btrfs_header_owner(leaf) != root_id)
12471                         goto next;
12472                 btrfs_item_key_to_cpu(leaf, &key, slot);
12473                 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
12474                         break;
12475                 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
12476                 /*
12477                  * Except normal disk bytenr and disk num bytes, we still
12478                  * need to do extra check on dbackref offset as
12479                  * dbackref offset = file_offset - file_extent_offset
12480                  */
12481                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == bytenr &&
12482                     btrfs_file_extent_disk_num_bytes(leaf, fi) == len &&
12483                     (u64)(key.offset - btrfs_file_extent_offset(leaf, fi)) ==
12484                     offset)
12485                         found_count++;
12486
12487 next:
12488                 ret = btrfs_next_item(root, &path);
12489                 if (ret)
12490                         break;
12491         }
12492 out:
12493         btrfs_release_path(&path);
12494         if (found_count != count) {
12495                 error(
12496 "extent[%llu, %llu] referencer count mismatch (root: %llu, owner: %llu, offset: %llu) wanted: %u, have: %u",
12497                         bytenr, len, root_id, objectid, offset, count, found_count);
12498                 return REFERENCER_MISSING;
12499         }
12500         return 0;
12501 }
12502
12503 /*
12504  * Check if the referencer of a shared data backref exists
12505  */
12506 static int check_shared_data_backref(struct btrfs_fs_info *fs_info,
12507                                      u64 parent, u64 bytenr)
12508 {
12509         struct extent_buffer *eb;
12510         struct btrfs_key key;
12511         struct btrfs_file_extent_item *fi;
12512         u32 nr;
12513         int found_parent = 0;
12514         int i;
12515
12516         eb = read_tree_block(fs_info, parent, 0);
12517         if (!extent_buffer_uptodate(eb))
12518                 goto out;
12519
12520         nr = btrfs_header_nritems(eb);
12521         for (i = 0; i < nr; i++) {
12522                 btrfs_item_key_to_cpu(eb, &key, i);
12523                 if (key.type != BTRFS_EXTENT_DATA_KEY)
12524                         continue;
12525
12526                 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
12527                 if (btrfs_file_extent_type(eb, fi) == BTRFS_FILE_EXTENT_INLINE)
12528                         continue;
12529
12530                 if (btrfs_file_extent_disk_bytenr(eb, fi) == bytenr) {
12531                         found_parent = 1;
12532                         break;
12533                 }
12534         }
12535
12536 out:
12537         free_extent_buffer(eb);
12538         if (!found_parent) {
12539                 error("shared extent %llu referencer lost (parent: %llu)",
12540                         bytenr, parent);
12541                 return REFERENCER_MISSING;
12542         }
12543         return 0;
12544 }
12545
12546 /*
12547  * Only delete backref if REFERENCER_MISSING now
12548  *
12549  * Returns <0   the extent was deleted
12550  * Returns >0   the backref was deleted but extent still exists, returned value
12551  *               means error after repair
12552  * Returns  0   nothing happened
12553  */
12554 static int repair_extent_item(struct btrfs_trans_handle *trans,
12555                       struct btrfs_root *root, struct btrfs_path *path,
12556                       u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
12557                       u64 owner, u64 offset, int err)
12558 {
12559         struct btrfs_key old_key;
12560         int freed = 0;
12561         int ret;
12562
12563         btrfs_item_key_to_cpu(path->nodes[0], &old_key, path->slots[0]);
12564
12565         if (err & (REFERENCER_MISSING | REFERENCER_MISMATCH)) {
12566                 /* delete the backref */
12567                 ret = btrfs_free_extent(trans, root->fs_info->fs_root, bytenr,
12568                           num_bytes, parent, root_objectid, owner, offset);
12569                 if (!ret) {
12570                         freed = 1;
12571                         err &= ~REFERENCER_MISSING;
12572                         printf("Delete backref in extent [%llu %llu]\n",
12573                                bytenr, num_bytes);
12574                 } else {
12575                         error("fail to delete backref in extent [%llu %llu]",
12576                                bytenr, num_bytes);
12577                 }
12578         }
12579
12580         /* btrfs_free_extent may delete the extent */
12581         btrfs_release_path(path);
12582         ret = btrfs_search_slot(NULL, root, &old_key, path, 0, 0);
12583
12584         if (ret)
12585                 ret = -ENOENT;
12586         else if (freed)
12587                 ret = err;
12588         return ret;
12589 }
12590
12591 /*
12592  * This function will check a given extent item, including its backref and
12593  * itself (like crossing stripe boundary and type)
12594  *
12595  * Since we don't use extent_record anymore, introduce new error bit
12596  */
12597 static int check_extent_item(struct btrfs_trans_handle *trans,
12598                              struct btrfs_fs_info *fs_info,
12599                              struct btrfs_path *path)
12600 {
12601         struct btrfs_extent_item *ei;
12602         struct btrfs_extent_inline_ref *iref;
12603         struct btrfs_extent_data_ref *dref;
12604         struct extent_buffer *eb = path->nodes[0];
12605         unsigned long end;
12606         unsigned long ptr;
12607         int slot = path->slots[0];
12608         int type;
12609         u32 nodesize = btrfs_super_nodesize(fs_info->super_copy);
12610         u32 item_size = btrfs_item_size_nr(eb, slot);
12611         u64 flags;
12612         u64 offset;
12613         u64 parent;
12614         u64 num_bytes;
12615         u64 root_objectid;
12616         u64 owner;
12617         u64 owner_offset;
12618         int metadata = 0;
12619         int level;
12620         struct btrfs_key key;
12621         int ret;
12622         int err = 0;
12623
12624         btrfs_item_key_to_cpu(eb, &key, slot);
12625         if (key.type == BTRFS_EXTENT_ITEM_KEY) {
12626                 bytes_used += key.offset;
12627                 num_bytes = key.offset;
12628         } else {
12629                 bytes_used += nodesize;
12630                 num_bytes = nodesize;
12631         }
12632
12633         if (item_size < sizeof(*ei)) {
12634                 /*
12635                  * COMPAT_EXTENT_TREE_V0 case, but it's already a super
12636                  * old thing when on disk format is still un-determined.
12637                  * No need to care about it anymore
12638                  */
12639                 error("unsupported COMPAT_EXTENT_TREE_V0 detected");
12640                 return -ENOTTY;
12641         }
12642
12643         ei = btrfs_item_ptr(eb, slot, struct btrfs_extent_item);
12644         flags = btrfs_extent_flags(eb, ei);
12645
12646         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
12647                 metadata = 1;
12648         if (metadata && check_crossing_stripes(global_info, key.objectid,
12649                                                eb->len)) {
12650                 error("bad metadata [%llu, %llu) crossing stripe boundary",
12651                       key.objectid, key.objectid + nodesize);
12652                 err |= CROSSING_STRIPE_BOUNDARY;
12653         }
12654
12655         ptr = (unsigned long)(ei + 1);
12656
12657         if (metadata && key.type == BTRFS_EXTENT_ITEM_KEY) {
12658                 /* Old EXTENT_ITEM metadata */
12659                 struct btrfs_tree_block_info *info;
12660
12661                 info = (struct btrfs_tree_block_info *)ptr;
12662                 level = btrfs_tree_block_level(eb, info);
12663                 ptr += sizeof(struct btrfs_tree_block_info);
12664         } else {
12665                 /* New METADATA_ITEM */
12666                 level = key.offset;
12667         }
12668         end = (unsigned long)ei + item_size;
12669
12670 next:
12671         /* Reached extent item end normally */
12672         if (ptr == end)
12673                 goto out;
12674
12675         /* Beyond extent item end, wrong item size */
12676         if (ptr > end) {
12677                 err |= ITEM_SIZE_MISMATCH;
12678                 error("extent item at bytenr %llu slot %d has wrong size",
12679                         eb->start, slot);
12680                 goto out;
12681         }
12682
12683         parent = 0;
12684         root_objectid = 0;
12685         owner = 0;
12686         owner_offset = 0;
12687         /* Now check every backref in this extent item */
12688         iref = (struct btrfs_extent_inline_ref *)ptr;
12689         type = btrfs_extent_inline_ref_type(eb, iref);
12690         offset = btrfs_extent_inline_ref_offset(eb, iref);
12691         switch (type) {
12692         case BTRFS_TREE_BLOCK_REF_KEY:
12693                 root_objectid = offset;
12694                 owner = level;
12695                 ret = check_tree_block_backref(fs_info, offset, key.objectid,
12696                                                level);
12697                 err |= ret;
12698                 break;
12699         case BTRFS_SHARED_BLOCK_REF_KEY:
12700                 parent = offset;
12701                 ret = check_shared_block_backref(fs_info, offset, key.objectid,
12702                                                  level);
12703                 err |= ret;
12704                 break;
12705         case BTRFS_EXTENT_DATA_REF_KEY:
12706                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
12707                 root_objectid = btrfs_extent_data_ref_root(eb, dref);
12708                 owner = btrfs_extent_data_ref_objectid(eb, dref);
12709                 owner_offset = btrfs_extent_data_ref_offset(eb, dref);
12710                 ret = check_extent_data_backref(fs_info, root_objectid, owner,
12711                                         owner_offset, key.objectid, key.offset,
12712                                         btrfs_extent_data_ref_count(eb, dref));
12713                 err |= ret;
12714                 break;
12715         case BTRFS_SHARED_DATA_REF_KEY:
12716                 parent = offset;
12717                 ret = check_shared_data_backref(fs_info, offset, key.objectid);
12718                 err |= ret;
12719                 break;
12720         default:
12721                 error("extent[%llu %d %llu] has unknown ref type: %d",
12722                         key.objectid, key.type, key.offset, type);
12723                 ret = UNKNOWN_TYPE;
12724                 err |= ret;
12725                 goto out;
12726         }
12727
12728         if (err && repair) {
12729                 ret = repair_extent_item(trans, fs_info->extent_root, path,
12730                          key.objectid, num_bytes, parent, root_objectid,
12731                          owner, owner_offset, ret);
12732                 if (ret < 0)
12733                         goto out;
12734                 if (ret) {
12735                         goto next;
12736                         err = ret;
12737                 }
12738         }
12739
12740         ptr += btrfs_extent_inline_ref_size(type);
12741         goto next;
12742
12743 out:
12744         return err;
12745 }
12746
12747 /*
12748  * Check if a dev extent item is referred correctly by its chunk
12749  */
12750 static int check_dev_extent_item(struct btrfs_fs_info *fs_info,
12751                                  struct extent_buffer *eb, int slot)
12752 {
12753         struct btrfs_root *chunk_root = fs_info->chunk_root;
12754         struct btrfs_dev_extent *ptr;
12755         struct btrfs_path path;
12756         struct btrfs_key chunk_key;
12757         struct btrfs_key devext_key;
12758         struct btrfs_chunk *chunk;
12759         struct extent_buffer *l;
12760         int num_stripes;
12761         u64 length;
12762         int i;
12763         int found_chunk = 0;
12764         int ret;
12765
12766         btrfs_item_key_to_cpu(eb, &devext_key, slot);
12767         ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_extent);
12768         length = btrfs_dev_extent_length(eb, ptr);
12769
12770         chunk_key.objectid = btrfs_dev_extent_chunk_objectid(eb, ptr);
12771         chunk_key.type = BTRFS_CHUNK_ITEM_KEY;
12772         chunk_key.offset = btrfs_dev_extent_chunk_offset(eb, ptr);
12773
12774         btrfs_init_path(&path);
12775         ret = btrfs_search_slot(NULL, chunk_root, &chunk_key, &path, 0, 0);
12776         if (ret)
12777                 goto out;
12778
12779         l = path.nodes[0];
12780         chunk = btrfs_item_ptr(l, path.slots[0], struct btrfs_chunk);
12781         ret = btrfs_check_chunk_valid(fs_info, l, chunk, path.slots[0],
12782                                       chunk_key.offset);
12783         if (ret < 0)
12784                 goto out;
12785
12786         if (btrfs_stripe_length(fs_info, l, chunk) != length)
12787                 goto out;
12788
12789         num_stripes = btrfs_chunk_num_stripes(l, chunk);
12790         for (i = 0; i < num_stripes; i++) {
12791                 u64 devid = btrfs_stripe_devid_nr(l, chunk, i);
12792                 u64 offset = btrfs_stripe_offset_nr(l, chunk, i);
12793
12794                 if (devid == devext_key.objectid &&
12795                     offset == devext_key.offset) {
12796                         found_chunk = 1;
12797                         break;
12798                 }
12799         }
12800 out:
12801         btrfs_release_path(&path);
12802         if (!found_chunk) {
12803                 error(
12804                 "device extent[%llu, %llu, %llu] did not find the related chunk",
12805                         devext_key.objectid, devext_key.offset, length);
12806                 return REFERENCER_MISSING;
12807         }
12808         return 0;
12809 }
12810
12811 /*
12812  * Check if the used space is correct with the dev item
12813  */
12814 static int check_dev_item(struct btrfs_fs_info *fs_info,
12815                           struct extent_buffer *eb, int slot)
12816 {
12817         struct btrfs_root *dev_root = fs_info->dev_root;
12818         struct btrfs_dev_item *dev_item;
12819         struct btrfs_path path;
12820         struct btrfs_key key;
12821         struct btrfs_dev_extent *ptr;
12822         u64 total_bytes;
12823         u64 dev_id;
12824         u64 used;
12825         u64 total = 0;
12826         int ret;
12827
12828         dev_item = btrfs_item_ptr(eb, slot, struct btrfs_dev_item);
12829         dev_id = btrfs_device_id(eb, dev_item);
12830         used = btrfs_device_bytes_used(eb, dev_item);
12831         total_bytes = btrfs_device_total_bytes(eb, dev_item);
12832
12833         key.objectid = dev_id;
12834         key.type = BTRFS_DEV_EXTENT_KEY;
12835         key.offset = 0;
12836
12837         btrfs_init_path(&path);
12838         ret = btrfs_search_slot(NULL, dev_root, &key, &path, 0, 0);
12839         if (ret < 0) {
12840                 btrfs_item_key_to_cpu(eb, &key, slot);
12841                 error("cannot find any related dev extent for dev[%llu, %u, %llu]",
12842                         key.objectid, key.type, key.offset);
12843                 btrfs_release_path(&path);
12844                 return REFERENCER_MISSING;
12845         }
12846
12847         /* Iterate dev_extents to calculate the used space of a device */
12848         while (1) {
12849                 if (path.slots[0] >= btrfs_header_nritems(path.nodes[0]))
12850                         goto next;
12851
12852                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
12853                 if (key.objectid > dev_id)
12854                         break;
12855                 if (key.type != BTRFS_DEV_EXTENT_KEY || key.objectid != dev_id)
12856                         goto next;
12857
12858                 ptr = btrfs_item_ptr(path.nodes[0], path.slots[0],
12859                                      struct btrfs_dev_extent);
12860                 total += btrfs_dev_extent_length(path.nodes[0], ptr);
12861 next:
12862                 ret = btrfs_next_item(dev_root, &path);
12863                 if (ret)
12864                         break;
12865         }
12866         btrfs_release_path(&path);
12867
12868         if (used != total) {
12869                 btrfs_item_key_to_cpu(eb, &key, slot);
12870                 error(
12871 "Dev extent's total-byte %llu is not equal to bytes-used %llu in dev[%llu, %u, %llu]",
12872                         total, used, BTRFS_ROOT_TREE_OBJECTID,
12873                         BTRFS_DEV_EXTENT_KEY, dev_id);
12874                 return ACCOUNTING_MISMATCH;
12875         }
12876         check_dev_size_alignment(dev_id, total_bytes, fs_info->sectorsize);
12877
12878         return 0;
12879 }
12880
12881 /*
12882  * Check a block group item with its referener (chunk) and its used space
12883  * with extent/metadata item
12884  */
12885 static int check_block_group_item(struct btrfs_fs_info *fs_info,
12886                                   struct extent_buffer *eb, int slot)
12887 {
12888         struct btrfs_root *extent_root = fs_info->extent_root;
12889         struct btrfs_root *chunk_root = fs_info->chunk_root;
12890         struct btrfs_block_group_item *bi;
12891         struct btrfs_block_group_item bg_item;
12892         struct btrfs_path path;
12893         struct btrfs_key bg_key;
12894         struct btrfs_key chunk_key;
12895         struct btrfs_key extent_key;
12896         struct btrfs_chunk *chunk;
12897         struct extent_buffer *leaf;
12898         struct btrfs_extent_item *ei;
12899         u32 nodesize = btrfs_super_nodesize(fs_info->super_copy);
12900         u64 flags;
12901         u64 bg_flags;
12902         u64 used;
12903         u64 total = 0;
12904         int ret;
12905         int err = 0;
12906
12907         btrfs_item_key_to_cpu(eb, &bg_key, slot);
12908         bi = btrfs_item_ptr(eb, slot, struct btrfs_block_group_item);
12909         read_extent_buffer(eb, &bg_item, (unsigned long)bi, sizeof(bg_item));
12910         used = btrfs_block_group_used(&bg_item);
12911         bg_flags = btrfs_block_group_flags(&bg_item);
12912
12913         chunk_key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
12914         chunk_key.type = BTRFS_CHUNK_ITEM_KEY;
12915         chunk_key.offset = bg_key.objectid;
12916
12917         btrfs_init_path(&path);
12918         /* Search for the referencer chunk */
12919         ret = btrfs_search_slot(NULL, chunk_root, &chunk_key, &path, 0, 0);
12920         if (ret) {
12921                 error(
12922                 "block group[%llu %llu] did not find the related chunk item",
12923                         bg_key.objectid, bg_key.offset);
12924                 err |= REFERENCER_MISSING;
12925         } else {
12926                 chunk = btrfs_item_ptr(path.nodes[0], path.slots[0],
12927                                         struct btrfs_chunk);
12928                 if (btrfs_chunk_length(path.nodes[0], chunk) !=
12929                                                 bg_key.offset) {
12930                         error(
12931         "block group[%llu %llu] related chunk item length does not match",
12932                                 bg_key.objectid, bg_key.offset);
12933                         err |= REFERENCER_MISMATCH;
12934                 }
12935         }
12936         btrfs_release_path(&path);
12937
12938         /* Search from the block group bytenr */
12939         extent_key.objectid = bg_key.objectid;
12940         extent_key.type = 0;
12941         extent_key.offset = 0;
12942
12943         btrfs_init_path(&path);
12944         ret = btrfs_search_slot(NULL, extent_root, &extent_key, &path, 0, 0);
12945         if (ret < 0)
12946                 goto out;
12947
12948         /* Iterate extent tree to account used space */
12949         while (1) {
12950                 leaf = path.nodes[0];
12951
12952                 /* Search slot can point to the last item beyond leaf nritems */
12953                 if (path.slots[0] >= btrfs_header_nritems(leaf))
12954                         goto next;
12955
12956                 btrfs_item_key_to_cpu(leaf, &extent_key, path.slots[0]);
12957                 if (extent_key.objectid >= bg_key.objectid + bg_key.offset)
12958                         break;
12959
12960                 if (extent_key.type != BTRFS_METADATA_ITEM_KEY &&
12961                     extent_key.type != BTRFS_EXTENT_ITEM_KEY)
12962                         goto next;
12963                 if (extent_key.objectid < bg_key.objectid)
12964                         goto next;
12965
12966                 if (extent_key.type == BTRFS_METADATA_ITEM_KEY)
12967                         total += nodesize;
12968                 else
12969                         total += extent_key.offset;
12970
12971                 ei = btrfs_item_ptr(leaf, path.slots[0],
12972                                     struct btrfs_extent_item);
12973                 flags = btrfs_extent_flags(leaf, ei);
12974                 if (flags & BTRFS_EXTENT_FLAG_DATA) {
12975                         if (!(bg_flags & BTRFS_BLOCK_GROUP_DATA)) {
12976                                 error(
12977                         "bad extent[%llu, %llu) type mismatch with chunk",
12978                                         extent_key.objectid,
12979                                         extent_key.objectid + extent_key.offset);
12980                                 err |= CHUNK_TYPE_MISMATCH;
12981                         }
12982                 } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
12983                         if (!(bg_flags & (BTRFS_BLOCK_GROUP_SYSTEM |
12984                                     BTRFS_BLOCK_GROUP_METADATA))) {
12985                                 error(
12986                         "bad extent[%llu, %llu) type mismatch with chunk",
12987                                         extent_key.objectid,
12988                                         extent_key.objectid + nodesize);
12989                                 err |= CHUNK_TYPE_MISMATCH;
12990                         }
12991                 }
12992 next:
12993                 ret = btrfs_next_item(extent_root, &path);
12994                 if (ret)
12995                         break;
12996         }
12997
12998 out:
12999         btrfs_release_path(&path);
13000
13001         if (total != used) {
13002                 error(
13003                 "block group[%llu %llu] used %llu but extent items used %llu",
13004                         bg_key.objectid, bg_key.offset, used, total);
13005                 err |= BG_ACCOUNTING_ERROR;
13006         }
13007         return err;
13008 }
13009
13010 /*
13011  * Add block group item to the extent tree if @err contains REFERENCER_MISSING.
13012  * FIXME: We still need to repair error of dev_item.
13013  *
13014  * Returns error after repair.
13015  */
13016 static int repair_chunk_item(struct btrfs_trans_handle *trans,
13017                              struct btrfs_root *chunk_root,
13018                              struct btrfs_path *path, int err)
13019 {
13020         struct btrfs_chunk *chunk;
13021         struct btrfs_key chunk_key;
13022         struct extent_buffer *eb = path->nodes[0];
13023         u64 length;
13024         int slot = path->slots[0];
13025         u64 type;
13026         int ret = 0;
13027
13028         btrfs_item_key_to_cpu(eb, &chunk_key, slot);
13029         if (chunk_key.type != BTRFS_CHUNK_ITEM_KEY)
13030                 return err;
13031         chunk = btrfs_item_ptr(eb, slot, struct btrfs_chunk);
13032         type = btrfs_chunk_type(path->nodes[0], chunk);
13033         length = btrfs_chunk_length(eb, chunk);
13034
13035         if (err & REFERENCER_MISSING) {
13036                 ret = btrfs_make_block_group(trans, chunk_root->fs_info, 0,
13037                      type, chunk_key.objectid, chunk_key.offset, length);
13038                 if (ret) {
13039                         error("fail to add block group item[%llu %llu]",
13040                               chunk_key.offset, length);
13041                         goto out;
13042                 } else {
13043                         err &= ~REFERENCER_MISSING;
13044                         printf("Added block group item[%llu %llu]\n",
13045                                chunk_key.offset, length);
13046                 }
13047         }
13048
13049 out:
13050         return err;
13051 }
13052
13053 /*
13054  * Check a chunk item.
13055  * Including checking all referred dev_extents and block group
13056  */
13057 static int check_chunk_item(struct btrfs_fs_info *fs_info,
13058                             struct extent_buffer *eb, int slot)
13059 {
13060         struct btrfs_root *extent_root = fs_info->extent_root;
13061         struct btrfs_root *dev_root = fs_info->dev_root;
13062         struct btrfs_path path;
13063         struct btrfs_key chunk_key;
13064         struct btrfs_key bg_key;
13065         struct btrfs_key devext_key;
13066         struct btrfs_chunk *chunk;
13067         struct extent_buffer *leaf;
13068         struct btrfs_block_group_item *bi;
13069         struct btrfs_block_group_item bg_item;
13070         struct btrfs_dev_extent *ptr;
13071         u64 length;
13072         u64 chunk_end;
13073         u64 stripe_len;
13074         u64 type;
13075         int num_stripes;
13076         u64 offset;
13077         u64 objectid;
13078         int i;
13079         int ret;
13080         int err = 0;
13081
13082         btrfs_item_key_to_cpu(eb, &chunk_key, slot);
13083         chunk = btrfs_item_ptr(eb, slot, struct btrfs_chunk);
13084         length = btrfs_chunk_length(eb, chunk);
13085         chunk_end = chunk_key.offset + length;
13086         ret = btrfs_check_chunk_valid(fs_info, eb, chunk, slot,
13087                                       chunk_key.offset);
13088         if (ret < 0) {
13089                 error("chunk[%llu %llu) is invalid", chunk_key.offset,
13090                         chunk_end);
13091                 err |= BYTES_UNALIGNED | UNKNOWN_TYPE;
13092                 goto out;
13093         }
13094         type = btrfs_chunk_type(eb, chunk);
13095
13096         bg_key.objectid = chunk_key.offset;
13097         bg_key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
13098         bg_key.offset = length;
13099
13100         btrfs_init_path(&path);
13101         ret = btrfs_search_slot(NULL, extent_root, &bg_key, &path, 0, 0);
13102         if (ret) {
13103                 error(
13104                 "chunk[%llu %llu) did not find the related block group item",
13105                         chunk_key.offset, chunk_end);
13106                 err |= REFERENCER_MISSING;
13107         } else{
13108                 leaf = path.nodes[0];
13109                 bi = btrfs_item_ptr(leaf, path.slots[0],
13110                                     struct btrfs_block_group_item);
13111                 read_extent_buffer(leaf, &bg_item, (unsigned long)bi,
13112                                    sizeof(bg_item));
13113                 if (btrfs_block_group_flags(&bg_item) != type) {
13114                         error(
13115 "chunk[%llu %llu) related block group item flags mismatch, wanted: %llu, have: %llu",
13116                                 chunk_key.offset, chunk_end, type,
13117                                 btrfs_block_group_flags(&bg_item));
13118                         err |= REFERENCER_MISSING;
13119                 }
13120         }
13121
13122         num_stripes = btrfs_chunk_num_stripes(eb, chunk);
13123         stripe_len = btrfs_stripe_length(fs_info, eb, chunk);
13124         for (i = 0; i < num_stripes; i++) {
13125                 btrfs_release_path(&path);
13126                 btrfs_init_path(&path);
13127                 devext_key.objectid = btrfs_stripe_devid_nr(eb, chunk, i);
13128                 devext_key.type = BTRFS_DEV_EXTENT_KEY;
13129                 devext_key.offset = btrfs_stripe_offset_nr(eb, chunk, i);
13130
13131                 ret = btrfs_search_slot(NULL, dev_root, &devext_key, &path,
13132                                         0, 0);
13133                 if (ret)
13134                         goto not_match_dev;
13135
13136                 leaf = path.nodes[0];
13137                 ptr = btrfs_item_ptr(leaf, path.slots[0],
13138                                      struct btrfs_dev_extent);
13139                 objectid = btrfs_dev_extent_chunk_objectid(leaf, ptr);
13140                 offset = btrfs_dev_extent_chunk_offset(leaf, ptr);
13141                 if (objectid != chunk_key.objectid ||
13142                     offset != chunk_key.offset ||
13143                     btrfs_dev_extent_length(leaf, ptr) != stripe_len)
13144                         goto not_match_dev;
13145                 continue;
13146 not_match_dev:
13147                 err |= BACKREF_MISSING;
13148                 error(
13149                 "chunk[%llu %llu) stripe %d did not find the related dev extent",
13150                         chunk_key.objectid, chunk_end, i);
13151                 continue;
13152         }
13153         btrfs_release_path(&path);
13154 out:
13155         return err;
13156 }
13157
13158 static int delete_extent_tree_item(struct btrfs_trans_handle *trans,
13159                                    struct btrfs_root *root,
13160                                    struct btrfs_path *path)
13161 {
13162         struct btrfs_key key;
13163         int ret = 0;
13164
13165         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
13166         btrfs_release_path(path);
13167         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
13168         if (ret) {
13169                 ret = -ENOENT;
13170                 goto out;
13171         }
13172
13173         ret = btrfs_del_item(trans, root, path);
13174         if (ret)
13175                 goto out;
13176
13177         if (path->slots[0] == 0)
13178                 btrfs_prev_leaf(root, path);
13179         else
13180                 path->slots[0]--;
13181 out:
13182         if (ret)
13183                 error("failed to delete root %llu item[%llu, %u, %llu]",
13184                       root->objectid, key.objectid, key.type, key.offset);
13185         else
13186                 printf("Deleted root %llu item[%llu, %u, %llu]\n",
13187                        root->objectid, key.objectid, key.type, key.offset);
13188         return ret;
13189 }
13190
13191 /*
13192  * Main entry function to check known items and update related accounting info
13193  */
13194 static int check_leaf_items(struct btrfs_trans_handle *trans,
13195                             struct btrfs_root *root, struct btrfs_path *path,
13196                             struct node_refs *nrefs, int account_bytes)
13197 {
13198         struct btrfs_fs_info *fs_info = root->fs_info;
13199         struct btrfs_key key;
13200         struct extent_buffer *eb;
13201         int slot;
13202         int type;
13203         struct btrfs_extent_data_ref *dref;
13204         int ret = 0;
13205         int err = 0;
13206
13207 again:
13208         eb = path->nodes[0];
13209         slot = path->slots[0];
13210         if (slot >= btrfs_header_nritems(eb)) {
13211                 if (slot == 0) {
13212                         error("empty leaf [%llu %u] root %llu", eb->start,
13213                                 root->fs_info->nodesize, root->objectid);
13214                         err |= EIO;
13215                 }
13216                 goto out;
13217         }
13218
13219         btrfs_item_key_to_cpu(eb, &key, slot);
13220         type = key.type;
13221
13222         switch (type) {
13223         case BTRFS_EXTENT_DATA_KEY:
13224                 ret = check_extent_data_item(root, path, nrefs, account_bytes);
13225                 if (repair && ret)
13226                         ret = repair_extent_data_item(trans, root, path, nrefs,
13227                                                       ret);
13228                 err |= ret;
13229                 break;
13230         case BTRFS_BLOCK_GROUP_ITEM_KEY:
13231                 ret = check_block_group_item(fs_info, eb, slot);
13232                 if (repair &&
13233                     ret & REFERENCER_MISSING)
13234                         ret = delete_extent_tree_item(trans, root, path);
13235                 err |= ret;
13236                 break;
13237         case BTRFS_DEV_ITEM_KEY:
13238                 ret = check_dev_item(fs_info, eb, slot);
13239                 err |= ret;
13240                 break;
13241         case BTRFS_CHUNK_ITEM_KEY:
13242                 ret = check_chunk_item(fs_info, eb, slot);
13243                 if (repair && ret)
13244                         ret = repair_chunk_item(trans, root, path, ret);
13245                 err |= ret;
13246                 break;
13247         case BTRFS_DEV_EXTENT_KEY:
13248                 ret = check_dev_extent_item(fs_info, eb, slot);
13249                 err |= ret;
13250                 break;
13251         case BTRFS_EXTENT_ITEM_KEY:
13252         case BTRFS_METADATA_ITEM_KEY:
13253                 ret = check_extent_item(trans, fs_info, path);
13254                 err |= ret;
13255                 break;
13256         case BTRFS_EXTENT_CSUM_KEY:
13257                 total_csum_bytes += btrfs_item_size_nr(eb, slot);
13258                 err |= ret;
13259                 break;
13260         case BTRFS_TREE_BLOCK_REF_KEY:
13261                 ret = check_tree_block_backref(fs_info, key.offset,
13262                                                key.objectid, -1);
13263                 if (repair &&
13264                     ret & (REFERENCER_MISMATCH | REFERENCER_MISSING))
13265                         ret = delete_extent_tree_item(trans, root, path);
13266                 err |= ret;
13267                 break;
13268         case BTRFS_EXTENT_DATA_REF_KEY:
13269                 dref = btrfs_item_ptr(eb, slot, struct btrfs_extent_data_ref);
13270                 ret = check_extent_data_backref(fs_info,
13271                                 btrfs_extent_data_ref_root(eb, dref),
13272                                 btrfs_extent_data_ref_objectid(eb, dref),
13273                                 btrfs_extent_data_ref_offset(eb, dref),
13274                                 key.objectid, 0,
13275                                 btrfs_extent_data_ref_count(eb, dref));
13276                 if (repair &&
13277                     ret & (REFERENCER_MISMATCH | REFERENCER_MISSING))
13278                         ret = delete_extent_tree_item(trans, root, path);
13279                 err |= ret;
13280                 break;
13281         case BTRFS_SHARED_BLOCK_REF_KEY:
13282                 ret = check_shared_block_backref(fs_info, key.offset,
13283                                                  key.objectid, -1);
13284                 if (repair &&
13285                     ret & (REFERENCER_MISMATCH | REFERENCER_MISSING))
13286                         ret = delete_extent_tree_item(trans, root, path);
13287                 err |= ret;
13288                 break;
13289         case BTRFS_SHARED_DATA_REF_KEY:
13290                 ret = check_shared_data_backref(fs_info, key.offset,
13291                                                 key.objectid);
13292                 if (repair &&
13293                     ret & (REFERENCER_MISMATCH | REFERENCER_MISSING))
13294                         ret = delete_extent_tree_item(trans, root, path);
13295                 err |= ret;
13296                 break;
13297         default:
13298                 break;
13299         }
13300
13301         ++path->slots[0];
13302         goto again;
13303 out:
13304         return err;
13305 }
13306
13307 /*
13308  * Low memory usage version check_chunks_and_extents.
13309  */
13310 static int check_chunks_and_extents_v2(struct btrfs_fs_info *fs_info)
13311 {
13312         struct btrfs_trans_handle *trans = NULL;
13313         struct btrfs_path path;
13314         struct btrfs_key old_key;
13315         struct btrfs_key key;
13316         struct btrfs_root *root1;
13317         struct btrfs_root *root;
13318         struct btrfs_root *cur_root;
13319         int err = 0;
13320         int ret;
13321
13322         root = fs_info->fs_root;
13323
13324         if (repair) {
13325                 trans = btrfs_start_transaction(fs_info->extent_root, 1);
13326                 if (IS_ERR(trans)) {
13327                         error("failed to start transaction before check");
13328                         return PTR_ERR(trans);
13329                 }
13330         }
13331
13332         root1 = root->fs_info->chunk_root;
13333         ret = check_btrfs_root(trans, root1, 0, 1);
13334         err |= ret;
13335
13336         root1 = root->fs_info->tree_root;
13337         ret = check_btrfs_root(trans, root1, 0, 1);
13338         err |= ret;
13339
13340         btrfs_init_path(&path);
13341         key.objectid = BTRFS_EXTENT_TREE_OBJECTID;
13342         key.offset = 0;
13343         key.type = BTRFS_ROOT_ITEM_KEY;
13344
13345         ret = btrfs_search_slot(NULL, root1, &key, &path, 0, 0);
13346         if (ret) {
13347                 error("cannot find extent tree in tree_root");
13348                 goto out;
13349         }
13350
13351         while (1) {
13352                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
13353                 if (key.type != BTRFS_ROOT_ITEM_KEY)
13354                         goto next;
13355                 old_key = key;
13356                 key.offset = (u64)-1;
13357
13358                 if (key.objectid == BTRFS_TREE_RELOC_OBJECTID)
13359                         cur_root = btrfs_read_fs_root_no_cache(root->fs_info,
13360                                         &key);
13361                 else
13362                         cur_root = btrfs_read_fs_root(root->fs_info, &key);
13363                 if (IS_ERR(cur_root) || !cur_root) {
13364                         error("failed to read tree: %lld", key.objectid);
13365                         goto next;
13366                 }
13367
13368                 ret = check_btrfs_root(trans, cur_root, 0, 1);
13369                 err |= ret;
13370
13371                 if (key.objectid == BTRFS_TREE_RELOC_OBJECTID)
13372                         btrfs_free_fs_root(cur_root);
13373
13374                 btrfs_release_path(&path);
13375                 ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
13376                                         &old_key, &path, 0, 0);
13377                 if (ret)
13378                         goto out;
13379 next:
13380                 ret = btrfs_next_item(root1, &path);
13381                 if (ret)
13382                         goto out;
13383         }
13384 out:
13385
13386         /* if repair, update block accounting */
13387         if (repair) {
13388                 ret = btrfs_fix_block_accounting(trans, root);
13389                 if (ret)
13390                         err |= ret;
13391                 else
13392                         err &= ~BG_ACCOUNTING_ERROR;
13393         }
13394
13395         if (trans)
13396                 btrfs_commit_transaction(trans, root->fs_info->extent_root);
13397
13398         btrfs_release_path(&path);
13399
13400         return err;
13401 }
13402
13403 static int do_check_chunks_and_extents(struct btrfs_fs_info *fs_info)
13404 {
13405         int ret;
13406
13407         if (!ctx.progress_enabled)
13408                 fprintf(stderr, "checking extents\n");
13409         if (check_mode == CHECK_MODE_LOWMEM)
13410                 ret = check_chunks_and_extents_v2(fs_info);
13411         else
13412                 ret = check_chunks_and_extents(fs_info);
13413
13414         /* Also repair device size related problems */
13415         if (repair && !ret) {
13416                 ret = btrfs_fix_device_and_super_size(fs_info);
13417                 if (ret > 0)
13418                         ret = 0;
13419         }
13420         return ret;
13421 }
13422
13423 static int btrfs_fsck_reinit_root(struct btrfs_trans_handle *trans,
13424                            struct btrfs_root *root, int overwrite)
13425 {
13426         struct extent_buffer *c;
13427         struct extent_buffer *old = root->node;
13428         int level;
13429         int ret;
13430         struct btrfs_disk_key disk_key = {0,0,0};
13431
13432         level = 0;
13433
13434         if (overwrite) {
13435                 c = old;
13436                 extent_buffer_get(c);
13437                 goto init;
13438         }
13439         c = btrfs_alloc_free_block(trans, root,
13440                                    root->fs_info->nodesize,
13441                                    root->root_key.objectid,
13442                                    &disk_key, level, 0, 0);
13443         if (IS_ERR(c)) {
13444                 c = old;
13445                 extent_buffer_get(c);
13446                 overwrite = 1;
13447         }
13448 init:
13449         memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
13450         btrfs_set_header_level(c, level);
13451         btrfs_set_header_bytenr(c, c->start);
13452         btrfs_set_header_generation(c, trans->transid);
13453         btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
13454         btrfs_set_header_owner(c, root->root_key.objectid);
13455
13456         write_extent_buffer(c, root->fs_info->fsid,
13457                             btrfs_header_fsid(), BTRFS_FSID_SIZE);
13458
13459         write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
13460                             btrfs_header_chunk_tree_uuid(c),
13461                             BTRFS_UUID_SIZE);
13462
13463         btrfs_mark_buffer_dirty(c);
13464         /*
13465          * this case can happen in the following case:
13466          *
13467          * 1.overwrite previous root.
13468          *
13469          * 2.reinit reloc data root, this is because we skip pin
13470          * down reloc data tree before which means we can allocate
13471          * same block bytenr here.
13472          */
13473         if (old->start == c->start) {
13474                 btrfs_set_root_generation(&root->root_item,
13475                                           trans->transid);
13476                 root->root_item.level = btrfs_header_level(root->node);
13477                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
13478                                         &root->root_key, &root->root_item);
13479                 if (ret) {
13480                         free_extent_buffer(c);
13481                         return ret;
13482                 }
13483         }
13484         free_extent_buffer(old);
13485         root->node = c;
13486         add_root_to_dirty_list(root);
13487         return 0;
13488 }
13489
13490 static int pin_down_tree_blocks(struct btrfs_fs_info *fs_info,
13491                                 struct extent_buffer *eb, int tree_root)
13492 {
13493         struct extent_buffer *tmp;
13494         struct btrfs_root_item *ri;
13495         struct btrfs_key key;
13496         u64 bytenr;
13497         int level = btrfs_header_level(eb);
13498         int nritems;
13499         int ret;
13500         int i;
13501
13502         /*
13503          * If we have pinned this block before, don't pin it again.
13504          * This can not only avoid forever loop with broken filesystem
13505          * but also give us some speedups.
13506          */
13507         if (test_range_bit(&fs_info->pinned_extents, eb->start,
13508                            eb->start + eb->len - 1, EXTENT_DIRTY, 0))
13509                 return 0;
13510
13511         btrfs_pin_extent(fs_info, eb->start, eb->len);
13512
13513         nritems = btrfs_header_nritems(eb);
13514         for (i = 0; i < nritems; i++) {
13515                 if (level == 0) {
13516                         btrfs_item_key_to_cpu(eb, &key, i);
13517                         if (key.type != BTRFS_ROOT_ITEM_KEY)
13518                                 continue;
13519                         /* Skip the extent root and reloc roots */
13520                         if (key.objectid == BTRFS_EXTENT_TREE_OBJECTID ||
13521                             key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
13522                             key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
13523                                 continue;
13524                         ri = btrfs_item_ptr(eb, i, struct btrfs_root_item);
13525                         bytenr = btrfs_disk_root_bytenr(eb, ri);
13526
13527                         /*
13528                          * If at any point we start needing the real root we
13529                          * will have to build a stump root for the root we are
13530                          * in, but for now this doesn't actually use the root so
13531                          * just pass in extent_root.
13532                          */
13533                         tmp = read_tree_block(fs_info, bytenr, 0);
13534                         if (!extent_buffer_uptodate(tmp)) {
13535                                 fprintf(stderr, "Error reading root block\n");
13536                                 return -EIO;
13537                         }
13538                         ret = pin_down_tree_blocks(fs_info, tmp, 0);
13539                         free_extent_buffer(tmp);
13540                         if (ret)
13541                                 return ret;
13542                 } else {
13543                         bytenr = btrfs_node_blockptr(eb, i);
13544
13545                         /* If we aren't the tree root don't read the block */
13546                         if (level == 1 && !tree_root) {
13547                                 btrfs_pin_extent(fs_info, bytenr,
13548                                                 fs_info->nodesize);
13549                                 continue;
13550                         }
13551
13552                         tmp = read_tree_block(fs_info, bytenr, 0);
13553                         if (!extent_buffer_uptodate(tmp)) {
13554                                 fprintf(stderr, "Error reading tree block\n");
13555                                 return -EIO;
13556                         }
13557                         ret = pin_down_tree_blocks(fs_info, tmp, tree_root);
13558                         free_extent_buffer(tmp);
13559                         if (ret)
13560                                 return ret;
13561                 }
13562         }
13563
13564         return 0;
13565 }
13566
13567 static int pin_metadata_blocks(struct btrfs_fs_info *fs_info)
13568 {
13569         int ret;
13570
13571         ret = pin_down_tree_blocks(fs_info, fs_info->chunk_root->node, 0);
13572         if (ret)
13573                 return ret;
13574
13575         return pin_down_tree_blocks(fs_info, fs_info->tree_root->node, 1);
13576 }
13577
13578 static int reset_block_groups(struct btrfs_fs_info *fs_info)
13579 {
13580         struct btrfs_block_group_cache *cache;
13581         struct btrfs_path path;
13582         struct extent_buffer *leaf;
13583         struct btrfs_chunk *chunk;
13584         struct btrfs_key key;
13585         int ret;
13586         u64 start;
13587
13588         btrfs_init_path(&path);
13589         key.objectid = 0;
13590         key.type = BTRFS_CHUNK_ITEM_KEY;
13591         key.offset = 0;
13592         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, &path, 0, 0);
13593         if (ret < 0) {
13594                 btrfs_release_path(&path);
13595                 return ret;
13596         }
13597
13598         /*
13599          * We do this in case the block groups were screwed up and had alloc
13600          * bits that aren't actually set on the chunks.  This happens with
13601          * restored images every time and could happen in real life I guess.
13602          */
13603         fs_info->avail_data_alloc_bits = 0;
13604         fs_info->avail_metadata_alloc_bits = 0;
13605         fs_info->avail_system_alloc_bits = 0;
13606
13607         /* First we need to create the in-memory block groups */
13608         while (1) {
13609                 if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
13610                         ret = btrfs_next_leaf(fs_info->chunk_root, &path);
13611                         if (ret < 0) {
13612                                 btrfs_release_path(&path);
13613                                 return ret;
13614                         }
13615                         if (ret) {
13616                                 ret = 0;
13617                                 break;
13618                         }
13619                 }
13620                 leaf = path.nodes[0];
13621                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
13622                 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
13623                         path.slots[0]++;
13624                         continue;
13625                 }
13626
13627                 chunk = btrfs_item_ptr(leaf, path.slots[0], struct btrfs_chunk);
13628                 btrfs_add_block_group(fs_info, 0,
13629                                       btrfs_chunk_type(leaf, chunk),
13630                                       key.objectid, key.offset,
13631                                       btrfs_chunk_length(leaf, chunk));
13632                 set_extent_dirty(&fs_info->free_space_cache, key.offset,
13633                                  key.offset + btrfs_chunk_length(leaf, chunk));
13634                 path.slots[0]++;
13635         }
13636         start = 0;
13637         while (1) {
13638                 cache = btrfs_lookup_first_block_group(fs_info, start);
13639                 if (!cache)
13640                         break;
13641                 cache->cached = 1;
13642                 start = cache->key.objectid + cache->key.offset;
13643         }
13644
13645         btrfs_release_path(&path);
13646         return 0;
13647 }
13648
13649 static int reset_balance(struct btrfs_trans_handle *trans,
13650                          struct btrfs_fs_info *fs_info)
13651 {
13652         struct btrfs_root *root = fs_info->tree_root;
13653         struct btrfs_path path;
13654         struct extent_buffer *leaf;
13655         struct btrfs_key key;
13656         int del_slot, del_nr = 0;
13657         int ret;
13658         int found = 0;
13659
13660         btrfs_init_path(&path);
13661         key.objectid = BTRFS_BALANCE_OBJECTID;
13662         key.type = BTRFS_BALANCE_ITEM_KEY;
13663         key.offset = 0;
13664         ret = btrfs_search_slot(trans, root, &key, &path, -1, 1);
13665         if (ret) {
13666                 if (ret > 0)
13667                         ret = 0;
13668                 if (!ret)
13669                         goto reinit_data_reloc;
13670                 else
13671                         goto out;
13672         }
13673
13674         ret = btrfs_del_item(trans, root, &path);
13675         if (ret)
13676                 goto out;
13677         btrfs_release_path(&path);
13678
13679         key.objectid = BTRFS_TREE_RELOC_OBJECTID;
13680         key.type = BTRFS_ROOT_ITEM_KEY;
13681         key.offset = 0;
13682         ret = btrfs_search_slot(trans, root, &key, &path, -1, 1);
13683         if (ret < 0)
13684                 goto out;
13685         while (1) {
13686                 if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
13687                         if (!found)
13688                                 break;
13689
13690                         if (del_nr) {
13691                                 ret = btrfs_del_items(trans, root, &path,
13692                                                       del_slot, del_nr);
13693                                 del_nr = 0;
13694                                 if (ret)
13695                                         goto out;
13696                         }
13697                         key.offset++;
13698                         btrfs_release_path(&path);
13699
13700                         found = 0;
13701                         ret = btrfs_search_slot(trans, root, &key, &path,
13702                                                 -1, 1);
13703                         if (ret < 0)
13704                                 goto out;
13705                         continue;
13706                 }
13707                 found = 1;
13708                 leaf = path.nodes[0];
13709                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
13710                 if (key.objectid > BTRFS_TREE_RELOC_OBJECTID)
13711                         break;
13712                 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
13713                         path.slots[0]++;
13714                         continue;
13715                 }
13716                 if (!del_nr) {
13717                         del_slot = path.slots[0];
13718                         del_nr = 1;
13719                 } else {
13720                         del_nr++;
13721                 }
13722                 path.slots[0]++;
13723         }
13724
13725         if (del_nr) {
13726                 ret = btrfs_del_items(trans, root, &path, del_slot, del_nr);
13727                 if (ret)
13728                         goto out;
13729         }
13730         btrfs_release_path(&path);
13731
13732 reinit_data_reloc:
13733         key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
13734         key.type = BTRFS_ROOT_ITEM_KEY;
13735         key.offset = (u64)-1;
13736         root = btrfs_read_fs_root(fs_info, &key);
13737         if (IS_ERR(root)) {
13738                 fprintf(stderr, "Error reading data reloc tree\n");
13739                 ret = PTR_ERR(root);
13740                 goto out;
13741         }
13742         record_root_in_trans(trans, root);
13743         ret = btrfs_fsck_reinit_root(trans, root, 0);
13744         if (ret)
13745                 goto out;
13746         ret = btrfs_make_root_dir(trans, root, BTRFS_FIRST_FREE_OBJECTID);
13747 out:
13748         btrfs_release_path(&path);
13749         return ret;
13750 }
13751
13752 static int reinit_extent_tree(struct btrfs_trans_handle *trans,
13753                               struct btrfs_fs_info *fs_info)
13754 {
13755         u64 start = 0;
13756         int ret;
13757
13758         /*
13759          * The only reason we don't do this is because right now we're just
13760          * walking the trees we find and pinning down their bytes, we don't look
13761          * at any of the leaves.  In order to do mixed groups we'd have to check
13762          * the leaves of any fs roots and pin down the bytes for any file
13763          * extents we find.  Not hard but why do it if we don't have to?
13764          */
13765         if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
13766                 fprintf(stderr, "We don't support re-initing the extent tree "
13767                         "for mixed block groups yet, please notify a btrfs "
13768                         "developer you want to do this so they can add this "
13769                         "functionality.\n");
13770                 return -EINVAL;
13771         }
13772
13773         /*
13774          * first we need to walk all of the trees except the extent tree and pin
13775          * down the bytes that are in use so we don't overwrite any existing
13776          * metadata.
13777          */
13778         ret = pin_metadata_blocks(fs_info);
13779         if (ret) {
13780                 fprintf(stderr, "error pinning down used bytes\n");
13781                 return ret;
13782         }
13783
13784         /*
13785          * Need to drop all the block groups since we're going to recreate all
13786          * of them again.
13787          */
13788         btrfs_free_block_groups(fs_info);
13789         ret = reset_block_groups(fs_info);
13790         if (ret) {
13791                 fprintf(stderr, "error resetting the block groups\n");
13792                 return ret;
13793         }
13794
13795         /* Ok we can allocate now, reinit the extent root */
13796         ret = btrfs_fsck_reinit_root(trans, fs_info->extent_root, 0);
13797         if (ret) {
13798                 fprintf(stderr, "extent root initialization failed\n");
13799                 /*
13800                  * When the transaction code is updated we should end the
13801                  * transaction, but for now progs only knows about commit so
13802                  * just return an error.
13803                  */
13804                 return ret;
13805         }
13806
13807         /*
13808          * Now we have all the in-memory block groups setup so we can make
13809          * allocations properly, and the metadata we care about is safe since we
13810          * pinned all of it above.
13811          */
13812         while (1) {
13813                 struct btrfs_block_group_cache *cache;
13814
13815                 cache = btrfs_lookup_first_block_group(fs_info, start);
13816                 if (!cache)
13817                         break;
13818                 start = cache->key.objectid + cache->key.offset;
13819                 ret = btrfs_insert_item(trans, fs_info->extent_root,
13820                                         &cache->key, &cache->item,
13821                                         sizeof(cache->item));
13822                 if (ret) {
13823                         fprintf(stderr, "Error adding block group\n");
13824                         return ret;
13825                 }
13826                 btrfs_extent_post_op(trans, fs_info->extent_root);
13827         }
13828
13829         ret = reset_balance(trans, fs_info);
13830         if (ret)
13831                 fprintf(stderr, "error resetting the pending balance\n");
13832
13833         return ret;
13834 }
13835
13836 static int recow_extent_buffer(struct btrfs_root *root, struct extent_buffer *eb)
13837 {
13838         struct btrfs_path path;
13839         struct btrfs_trans_handle *trans;
13840         struct btrfs_key key;
13841         int ret;
13842
13843         printf("Recowing metadata block %llu\n", eb->start);
13844         key.objectid = btrfs_header_owner(eb);
13845         key.type = BTRFS_ROOT_ITEM_KEY;
13846         key.offset = (u64)-1;
13847
13848         root = btrfs_read_fs_root(root->fs_info, &key);
13849         if (IS_ERR(root)) {
13850                 fprintf(stderr, "Couldn't find owner root %llu\n",
13851                         key.objectid);
13852                 return PTR_ERR(root);
13853         }
13854
13855         trans = btrfs_start_transaction(root, 1);
13856         if (IS_ERR(trans))
13857                 return PTR_ERR(trans);
13858
13859         btrfs_init_path(&path);
13860         path.lowest_level = btrfs_header_level(eb);
13861         if (path.lowest_level)
13862                 btrfs_node_key_to_cpu(eb, &key, 0);
13863         else
13864                 btrfs_item_key_to_cpu(eb, &key, 0);
13865
13866         ret = btrfs_search_slot(trans, root, &key, &path, 0, 1);
13867         btrfs_commit_transaction(trans, root);
13868         btrfs_release_path(&path);
13869         return ret;
13870 }
13871
13872 static int delete_bad_item(struct btrfs_root *root, struct bad_item *bad)
13873 {
13874         struct btrfs_path path;
13875         struct btrfs_trans_handle *trans;
13876         struct btrfs_key key;
13877         int ret;
13878
13879         printf("Deleting bad item [%llu,%u,%llu]\n", bad->key.objectid,
13880                bad->key.type, bad->key.offset);
13881         key.objectid = bad->root_id;
13882         key.type = BTRFS_ROOT_ITEM_KEY;
13883         key.offset = (u64)-1;
13884
13885         root = btrfs_read_fs_root(root->fs_info, &key);
13886         if (IS_ERR(root)) {
13887                 fprintf(stderr, "Couldn't find owner root %llu\n",
13888                         key.objectid);
13889                 return PTR_ERR(root);
13890         }
13891
13892         trans = btrfs_start_transaction(root, 1);
13893         if (IS_ERR(trans))
13894                 return PTR_ERR(trans);
13895
13896         btrfs_init_path(&path);
13897         ret = btrfs_search_slot(trans, root, &bad->key, &path, -1, 1);
13898         if (ret) {
13899                 if (ret > 0)
13900                         ret = 0;
13901                 goto out;
13902         }
13903         ret = btrfs_del_item(trans, root, &path);
13904 out:
13905         btrfs_commit_transaction(trans, root);
13906         btrfs_release_path(&path);
13907         return ret;
13908 }
13909
13910 static int zero_log_tree(struct btrfs_root *root)
13911 {
13912         struct btrfs_trans_handle *trans;
13913         int ret;
13914
13915         trans = btrfs_start_transaction(root, 1);
13916         if (IS_ERR(trans)) {
13917                 ret = PTR_ERR(trans);
13918                 return ret;
13919         }
13920         btrfs_set_super_log_root(root->fs_info->super_copy, 0);
13921         btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
13922         ret = btrfs_commit_transaction(trans, root);
13923         return ret;
13924 }
13925
13926 static int populate_csum(struct btrfs_trans_handle *trans,
13927                          struct btrfs_root *csum_root, char *buf, u64 start,
13928                          u64 len)
13929 {
13930         struct btrfs_fs_info *fs_info = csum_root->fs_info;
13931         u64 offset = 0;
13932         u64 sectorsize;
13933         int ret = 0;
13934
13935         while (offset < len) {
13936                 sectorsize = fs_info->sectorsize;
13937                 ret = read_extent_data(fs_info, buf, start + offset,
13938                                        &sectorsize, 0);
13939                 if (ret)
13940                         break;
13941                 ret = btrfs_csum_file_block(trans, csum_root, start + len,
13942                                             start + offset, buf, sectorsize);
13943                 if (ret)
13944                         break;
13945                 offset += sectorsize;
13946         }
13947         return ret;
13948 }
13949
13950 static int fill_csum_tree_from_one_fs_root(struct btrfs_trans_handle *trans,
13951                                       struct btrfs_root *csum_root,
13952                                       struct btrfs_root *cur_root)
13953 {
13954         struct btrfs_path path;
13955         struct btrfs_key key;
13956         struct extent_buffer *node;
13957         struct btrfs_file_extent_item *fi;
13958         char *buf = NULL;
13959         u64 start = 0;
13960         u64 len = 0;
13961         int slot = 0;
13962         int ret = 0;
13963
13964         buf = malloc(cur_root->fs_info->sectorsize);
13965         if (!buf)
13966                 return -ENOMEM;
13967
13968         btrfs_init_path(&path);
13969         key.objectid = 0;
13970         key.offset = 0;
13971         key.type = 0;
13972         ret = btrfs_search_slot(NULL, cur_root, &key, &path, 0, 0);
13973         if (ret < 0)
13974                 goto out;
13975         /* Iterate all regular file extents and fill its csum */
13976         while (1) {
13977                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
13978
13979                 if (key.type != BTRFS_EXTENT_DATA_KEY)
13980                         goto next;
13981                 node = path.nodes[0];
13982                 slot = path.slots[0];
13983                 fi = btrfs_item_ptr(node, slot, struct btrfs_file_extent_item);
13984                 if (btrfs_file_extent_type(node, fi) != BTRFS_FILE_EXTENT_REG)
13985                         goto next;
13986                 start = btrfs_file_extent_disk_bytenr(node, fi);
13987                 len = btrfs_file_extent_disk_num_bytes(node, fi);
13988
13989                 ret = populate_csum(trans, csum_root, buf, start, len);
13990                 if (ret == -EEXIST)
13991                         ret = 0;
13992                 if (ret < 0)
13993                         goto out;
13994 next:
13995                 /*
13996                  * TODO: if next leaf is corrupted, jump to nearest next valid
13997                  * leaf.
13998                  */
13999                 ret = btrfs_next_item(cur_root, &path);
14000                 if (ret < 0)
14001                         goto out;
14002                 if (ret > 0) {
14003                         ret = 0;
14004                         goto out;
14005                 }
14006         }
14007
14008 out:
14009         btrfs_release_path(&path);
14010         free(buf);
14011         return ret;
14012 }
14013
14014 static int fill_csum_tree_from_fs(struct btrfs_trans_handle *trans,
14015                                   struct btrfs_root *csum_root)
14016 {
14017         struct btrfs_fs_info *fs_info = csum_root->fs_info;
14018         struct btrfs_path path;
14019         struct btrfs_root *tree_root = fs_info->tree_root;
14020         struct btrfs_root *cur_root;
14021         struct extent_buffer *node;
14022         struct btrfs_key key;
14023         int slot = 0;
14024         int ret = 0;
14025
14026         btrfs_init_path(&path);
14027         key.objectid = BTRFS_FS_TREE_OBJECTID;
14028         key.offset = 0;
14029         key.type = BTRFS_ROOT_ITEM_KEY;
14030         ret = btrfs_search_slot(NULL, tree_root, &key, &path, 0, 0);
14031         if (ret < 0)
14032                 goto out;
14033         if (ret > 0) {
14034                 ret = -ENOENT;
14035                 goto out;
14036         }
14037
14038         while (1) {
14039                 node = path.nodes[0];
14040                 slot = path.slots[0];
14041                 btrfs_item_key_to_cpu(node, &key, slot);
14042                 if (key.objectid > BTRFS_LAST_FREE_OBJECTID)
14043                         goto out;
14044                 if (key.type != BTRFS_ROOT_ITEM_KEY)
14045                         goto next;
14046                 if (!is_fstree(key.objectid))
14047                         goto next;
14048                 key.offset = (u64)-1;
14049
14050                 cur_root = btrfs_read_fs_root(fs_info, &key);
14051                 if (IS_ERR(cur_root) || !cur_root) {
14052                         fprintf(stderr, "Fail to read fs/subvol tree: %lld\n",
14053                                 key.objectid);
14054                         goto out;
14055                 }
14056                 ret = fill_csum_tree_from_one_fs_root(trans, csum_root,
14057                                 cur_root);
14058                 if (ret < 0)
14059                         goto out;
14060 next:
14061                 ret = btrfs_next_item(tree_root, &path);
14062                 if (ret > 0) {
14063                         ret = 0;
14064                         goto out;
14065                 }
14066                 if (ret < 0)
14067                         goto out;
14068         }
14069
14070 out:
14071         btrfs_release_path(&path);
14072         return ret;
14073 }
14074
14075 static int fill_csum_tree_from_extent(struct btrfs_trans_handle *trans,
14076                                       struct btrfs_root *csum_root)
14077 {
14078         struct btrfs_root *extent_root = csum_root->fs_info->extent_root;
14079         struct btrfs_path path;
14080         struct btrfs_extent_item *ei;
14081         struct extent_buffer *leaf;
14082         char *buf;
14083         struct btrfs_key key;
14084         int ret;
14085
14086         btrfs_init_path(&path);
14087         key.objectid = 0;
14088         key.type = BTRFS_EXTENT_ITEM_KEY;
14089         key.offset = 0;
14090         ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
14091         if (ret < 0) {
14092                 btrfs_release_path(&path);
14093                 return ret;
14094         }
14095
14096         buf = malloc(csum_root->fs_info->sectorsize);
14097         if (!buf) {
14098                 btrfs_release_path(&path);
14099                 return -ENOMEM;
14100         }
14101
14102         while (1) {
14103                 if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
14104                         ret = btrfs_next_leaf(extent_root, &path);
14105                         if (ret < 0)
14106                                 break;
14107                         if (ret) {
14108                                 ret = 0;
14109                                 break;
14110                         }
14111                 }
14112                 leaf = path.nodes[0];
14113
14114                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
14115                 if (key.type != BTRFS_EXTENT_ITEM_KEY) {
14116                         path.slots[0]++;
14117                         continue;
14118                 }
14119
14120                 ei = btrfs_item_ptr(leaf, path.slots[0],
14121                                     struct btrfs_extent_item);
14122                 if (!(btrfs_extent_flags(leaf, ei) &
14123                       BTRFS_EXTENT_FLAG_DATA)) {
14124                         path.slots[0]++;
14125                         continue;
14126                 }
14127
14128                 ret = populate_csum(trans, csum_root, buf, key.objectid,
14129                                     key.offset);
14130                 if (ret)
14131                         break;
14132                 path.slots[0]++;
14133         }
14134
14135         btrfs_release_path(&path);
14136         free(buf);
14137         return ret;
14138 }
14139
14140 /*
14141  * Recalculate the csum and put it into the csum tree.
14142  *
14143  * Extent tree init will wipe out all the extent info, so in that case, we
14144  * can't depend on extent tree, but use fs tree.  If search_fs_tree is set, we
14145  * will use fs/subvol trees to init the csum tree.
14146  */
14147 static int fill_csum_tree(struct btrfs_trans_handle *trans,
14148                           struct btrfs_root *csum_root,
14149                           int search_fs_tree)
14150 {
14151         if (search_fs_tree)
14152                 return fill_csum_tree_from_fs(trans, csum_root);
14153         else
14154                 return fill_csum_tree_from_extent(trans, csum_root);
14155 }
14156
14157 static void free_roots_info_cache(void)
14158 {
14159         if (!roots_info_cache)
14160                 return;
14161
14162         while (!cache_tree_empty(roots_info_cache)) {
14163                 struct cache_extent *entry;
14164                 struct root_item_info *rii;
14165
14166                 entry = first_cache_extent(roots_info_cache);
14167                 if (!entry)
14168                         break;
14169                 remove_cache_extent(roots_info_cache, entry);
14170                 rii = container_of(entry, struct root_item_info, cache_extent);
14171                 free(rii);
14172         }
14173
14174         free(roots_info_cache);
14175         roots_info_cache = NULL;
14176 }
14177
14178 static int build_roots_info_cache(struct btrfs_fs_info *info)
14179 {
14180         int ret = 0;
14181         struct btrfs_key key;
14182         struct extent_buffer *leaf;
14183         struct btrfs_path path;
14184
14185         if (!roots_info_cache) {
14186                 roots_info_cache = malloc(sizeof(*roots_info_cache));
14187                 if (!roots_info_cache)
14188                         return -ENOMEM;
14189                 cache_tree_init(roots_info_cache);
14190         }
14191
14192         btrfs_init_path(&path);
14193         key.objectid = 0;
14194         key.type = BTRFS_EXTENT_ITEM_KEY;
14195         key.offset = 0;
14196         ret = btrfs_search_slot(NULL, info->extent_root, &key, &path, 0, 0);
14197         if (ret < 0)
14198                 goto out;
14199         leaf = path.nodes[0];
14200
14201         while (1) {
14202                 struct btrfs_key found_key;
14203                 struct btrfs_extent_item *ei;
14204                 struct btrfs_extent_inline_ref *iref;
14205                 int slot = path.slots[0];
14206                 int type;
14207                 u64 flags;
14208                 u64 root_id;
14209                 u8 level;
14210                 struct cache_extent *entry;
14211                 struct root_item_info *rii;
14212
14213                 if (slot >= btrfs_header_nritems(leaf)) {
14214                         ret = btrfs_next_leaf(info->extent_root, &path);
14215                         if (ret < 0) {
14216                                 break;
14217                         } else if (ret) {
14218                                 ret = 0;
14219                                 break;
14220                         }
14221                         leaf = path.nodes[0];
14222                         slot = path.slots[0];
14223                 }
14224
14225                 btrfs_item_key_to_cpu(leaf, &found_key, path.slots[0]);
14226
14227                 if (found_key.type != BTRFS_EXTENT_ITEM_KEY &&
14228                     found_key.type != BTRFS_METADATA_ITEM_KEY)
14229                         goto next;
14230
14231                 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
14232                 flags = btrfs_extent_flags(leaf, ei);
14233
14234                 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
14235                     !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
14236                         goto next;
14237
14238                 if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
14239                         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
14240                         level = found_key.offset;
14241                 } else {
14242                         struct btrfs_tree_block_info *binfo;
14243
14244                         binfo = (struct btrfs_tree_block_info *)(ei + 1);
14245                         iref = (struct btrfs_extent_inline_ref *)(binfo + 1);
14246                         level = btrfs_tree_block_level(leaf, binfo);
14247                 }
14248
14249                 /*
14250                  * For a root extent, it must be of the following type and the
14251                  * first (and only one) iref in the item.
14252                  */
14253                 type = btrfs_extent_inline_ref_type(leaf, iref);
14254                 if (type != BTRFS_TREE_BLOCK_REF_KEY)
14255                         goto next;
14256
14257                 root_id = btrfs_extent_inline_ref_offset(leaf, iref);
14258                 entry = lookup_cache_extent(roots_info_cache, root_id, 1);
14259                 if (!entry) {
14260                         rii = malloc(sizeof(struct root_item_info));
14261                         if (!rii) {
14262                                 ret = -ENOMEM;
14263                                 goto out;
14264                         }
14265                         rii->cache_extent.start = root_id;
14266                         rii->cache_extent.size = 1;
14267                         rii->level = (u8)-1;
14268                         entry = &rii->cache_extent;
14269                         ret = insert_cache_extent(roots_info_cache, entry);
14270                         ASSERT(ret == 0);
14271                 } else {
14272                         rii = container_of(entry, struct root_item_info,
14273                                            cache_extent);
14274                 }
14275
14276                 ASSERT(rii->cache_extent.start == root_id);
14277                 ASSERT(rii->cache_extent.size == 1);
14278
14279                 if (level > rii->level || rii->level == (u8)-1) {
14280                         rii->level = level;
14281                         rii->bytenr = found_key.objectid;
14282                         rii->gen = btrfs_extent_generation(leaf, ei);
14283                         rii->node_count = 1;
14284                 } else if (level == rii->level) {
14285                         rii->node_count++;
14286                 }
14287 next:
14288                 path.slots[0]++;
14289         }
14290
14291 out:
14292         btrfs_release_path(&path);
14293
14294         return ret;
14295 }
14296
14297 static int maybe_repair_root_item(struct btrfs_path *path,
14298                                   const struct btrfs_key *root_key,
14299                                   const int read_only_mode)
14300 {
14301         const u64 root_id = root_key->objectid;
14302         struct cache_extent *entry;
14303         struct root_item_info *rii;
14304         struct btrfs_root_item ri;
14305         unsigned long offset;
14306
14307         entry = lookup_cache_extent(roots_info_cache, root_id, 1);
14308         if (!entry) {
14309                 fprintf(stderr,
14310                         "Error: could not find extent items for root %llu\n",
14311                         root_key->objectid);
14312                 return -ENOENT;
14313         }
14314
14315         rii = container_of(entry, struct root_item_info, cache_extent);
14316         ASSERT(rii->cache_extent.start == root_id);
14317         ASSERT(rii->cache_extent.size == 1);
14318
14319         if (rii->node_count != 1) {
14320                 fprintf(stderr,
14321                         "Error: could not find btree root extent for root %llu\n",
14322                         root_id);
14323                 return -ENOENT;
14324         }
14325
14326         offset = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
14327         read_extent_buffer(path->nodes[0], &ri, offset, sizeof(ri));
14328
14329         if (btrfs_root_bytenr(&ri) != rii->bytenr ||
14330             btrfs_root_level(&ri) != rii->level ||
14331             btrfs_root_generation(&ri) != rii->gen) {
14332
14333                 /*
14334                  * If we're in repair mode but our caller told us to not update
14335                  * the root item, i.e. just check if it needs to be updated, don't
14336                  * print this message, since the caller will call us again shortly
14337                  * for the same root item without read only mode (the caller will
14338                  * open a transaction first).
14339                  */
14340                 if (!(read_only_mode && repair))
14341                         fprintf(stderr,
14342                                 "%sroot item for root %llu,"
14343                                 " current bytenr %llu, current gen %llu, current level %u,"
14344                                 " new bytenr %llu, new gen %llu, new level %u\n",
14345                                 (read_only_mode ? "" : "fixing "),
14346                                 root_id,
14347                                 btrfs_root_bytenr(&ri), btrfs_root_generation(&ri),
14348                                 btrfs_root_level(&ri),
14349                                 rii->bytenr, rii->gen, rii->level);
14350
14351                 if (btrfs_root_generation(&ri) > rii->gen) {
14352                         fprintf(stderr,
14353                                 "root %llu has a root item with a more recent gen (%llu) compared to the found root node (%llu)\n",
14354                                 root_id, btrfs_root_generation(&ri), rii->gen);
14355                         return -EINVAL;
14356                 }
14357
14358                 if (!read_only_mode) {
14359                         btrfs_set_root_bytenr(&ri, rii->bytenr);
14360                         btrfs_set_root_level(&ri, rii->level);
14361                         btrfs_set_root_generation(&ri, rii->gen);
14362                         write_extent_buffer(path->nodes[0], &ri,
14363                                             offset, sizeof(ri));
14364                 }
14365
14366                 return 1;
14367         }
14368
14369         return 0;
14370 }
14371
14372 /*
14373  * A regression introduced in the 3.17 kernel (more specifically in 3.17-rc2),
14374  * caused read-only snapshots to be corrupted if they were created at a moment
14375  * when the source subvolume/snapshot had orphan items. The issue was that the
14376  * on-disk root items became incorrect, referring to the pre orphan cleanup root
14377  * node instead of the post orphan cleanup root node.
14378  * So this function, and its callees, just detects and fixes those cases. Even
14379  * though the regression was for read-only snapshots, this function applies to
14380  * any snapshot/subvolume root.
14381  * This must be run before any other repair code - not doing it so, makes other
14382  * repair code delete or modify backrefs in the extent tree for example, which
14383  * will result in an inconsistent fs after repairing the root items.
14384  */
14385 static int repair_root_items(struct btrfs_fs_info *info)
14386 {
14387         struct btrfs_path path;
14388         struct btrfs_key key;
14389         struct extent_buffer *leaf;
14390         struct btrfs_trans_handle *trans = NULL;
14391         int ret = 0;
14392         int bad_roots = 0;
14393         int need_trans = 0;
14394
14395         btrfs_init_path(&path);
14396
14397         ret = build_roots_info_cache(info);
14398         if (ret)
14399                 goto out;
14400
14401         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
14402         key.type = BTRFS_ROOT_ITEM_KEY;
14403         key.offset = 0;
14404
14405 again:
14406         /*
14407          * Avoid opening and committing transactions if a leaf doesn't have
14408          * any root items that need to be fixed, so that we avoid rotating
14409          * backup roots unnecessarily.
14410          */
14411         if (need_trans) {
14412                 trans = btrfs_start_transaction(info->tree_root, 1);
14413                 if (IS_ERR(trans)) {
14414                         ret = PTR_ERR(trans);
14415                         goto out;
14416                 }
14417         }
14418
14419         ret = btrfs_search_slot(trans, info->tree_root, &key, &path,
14420                                 0, trans ? 1 : 0);
14421         if (ret < 0)
14422                 goto out;
14423         leaf = path.nodes[0];
14424
14425         while (1) {
14426                 struct btrfs_key found_key;
14427
14428                 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
14429                         int no_more_keys = find_next_key(&path, &key);
14430
14431                         btrfs_release_path(&path);
14432                         if (trans) {
14433                                 ret = btrfs_commit_transaction(trans,
14434                                                                info->tree_root);
14435                                 trans = NULL;
14436                                 if (ret < 0)
14437                                         goto out;
14438                         }
14439                         need_trans = 0;
14440                         if (no_more_keys)
14441                                 break;
14442                         goto again;
14443                 }
14444
14445                 btrfs_item_key_to_cpu(leaf, &found_key, path.slots[0]);
14446
14447                 if (found_key.type != BTRFS_ROOT_ITEM_KEY)
14448                         goto next;
14449                 if (found_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
14450                         goto next;
14451
14452                 ret = maybe_repair_root_item(&path, &found_key, trans ? 0 : 1);
14453                 if (ret < 0)
14454                         goto out;
14455                 if (ret) {
14456                         if (!trans && repair) {
14457                                 need_trans = 1;
14458                                 key = found_key;
14459                                 btrfs_release_path(&path);
14460                                 goto again;
14461                         }
14462                         bad_roots++;
14463                 }
14464 next:
14465                 path.slots[0]++;
14466         }
14467         ret = 0;
14468 out:
14469         free_roots_info_cache();
14470         btrfs_release_path(&path);
14471         if (trans)
14472                 btrfs_commit_transaction(trans, info->tree_root);
14473         if (ret < 0)
14474                 return ret;
14475
14476         return bad_roots;
14477 }
14478
14479 static int clear_free_space_cache(struct btrfs_fs_info *fs_info)
14480 {
14481         struct btrfs_trans_handle *trans;
14482         struct btrfs_block_group_cache *bg_cache;
14483         u64 current = 0;
14484         int ret = 0;
14485
14486         /* Clear all free space cache inodes and its extent data */
14487         while (1) {
14488                 bg_cache = btrfs_lookup_first_block_group(fs_info, current);
14489                 if (!bg_cache)
14490                         break;
14491                 ret = btrfs_clear_free_space_cache(fs_info, bg_cache);
14492                 if (ret < 0)
14493                         return ret;
14494                 current = bg_cache->key.objectid + bg_cache->key.offset;
14495         }
14496
14497         /* Don't forget to set cache_generation to -1 */
14498         trans = btrfs_start_transaction(fs_info->tree_root, 0);
14499         if (IS_ERR(trans)) {
14500                 error("failed to update super block cache generation");
14501                 return PTR_ERR(trans);
14502         }
14503         btrfs_set_super_cache_generation(fs_info->super_copy, (u64)-1);
14504         btrfs_commit_transaction(trans, fs_info->tree_root);
14505
14506         return ret;
14507 }
14508
14509 static int do_clear_free_space_cache(struct btrfs_fs_info *fs_info,
14510                 int clear_version)
14511 {
14512         int ret = 0;
14513
14514         if (clear_version == 1) {
14515                 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
14516                         error(
14517                 "free space cache v2 detected, use --clear-space-cache v2");
14518                         ret = 1;
14519                         goto close_out;
14520                 }
14521                 printf("Clearing free space cache\n");
14522                 ret = clear_free_space_cache(fs_info);
14523                 if (ret) {
14524                         error("failed to clear free space cache");
14525                         ret = 1;
14526                 } else {
14527                         printf("Free space cache cleared\n");
14528                 }
14529         } else if (clear_version == 2) {
14530                 if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
14531                         printf("no free space cache v2 to clear\n");
14532                         ret = 0;
14533                         goto close_out;
14534                 }
14535                 printf("Clear free space cache v2\n");
14536                 ret = btrfs_clear_free_space_tree(fs_info);
14537                 if (ret) {
14538                         error("failed to clear free space cache v2: %d", ret);
14539                         ret = 1;
14540                 } else {
14541                         printf("free space cache v2 cleared\n");
14542                 }
14543         }
14544 close_out:
14545         return ret;
14546 }
14547
14548 const char * const cmd_check_usage[] = {
14549         "btrfs check [options] <device>",
14550         "Check structural integrity of a filesystem (unmounted).",
14551         "Check structural integrity of an unmounted filesystem. Verify internal",
14552         "trees' consistency and item connectivity. In the repair mode try to",
14553         "fix the problems found. ",
14554         "WARNING: the repair mode is considered dangerous",
14555         "",
14556         "-s|--super <superblock>     use this superblock copy",
14557         "-b|--backup                 use the first valid backup root copy",
14558         "--force                     skip mount checks, repair is not possible",
14559         "--repair                    try to repair the filesystem",
14560         "--readonly                  run in read-only mode (default)",
14561         "--init-csum-tree            create a new CRC tree",
14562         "--init-extent-tree          create a new extent tree",
14563         "--mode <MODE>               allows choice of memory/IO trade-offs",
14564         "                            where MODE is one of:",
14565         "                            original - read inodes and extents to memory (requires",
14566         "                                       more memory, does less IO)",
14567         "                            lowmem   - try to use less memory but read blocks again",
14568         "                                       when needed",
14569         "--check-data-csum           verify checksums of data blocks",
14570         "-Q|--qgroup-report          print a report on qgroup consistency",
14571         "-E|--subvol-extents <subvolid>",
14572         "                            print subvolume extents and sharing state",
14573         "-r|--tree-root <bytenr>     use the given bytenr for the tree root",
14574         "--chunk-root <bytenr>       use the given bytenr for the chunk tree root",
14575         "-p|--progress               indicate progress",
14576         "--clear-space-cache v1|v2   clear space cache for v1 or v2",
14577         NULL
14578 };
14579
14580 int cmd_check(int argc, char **argv)
14581 {
14582         struct cache_tree root_cache;
14583         struct btrfs_root *root;
14584         struct btrfs_fs_info *info;
14585         u64 bytenr = 0;
14586         u64 subvolid = 0;
14587         u64 tree_root_bytenr = 0;
14588         u64 chunk_root_bytenr = 0;
14589         char uuidbuf[BTRFS_UUID_UNPARSED_SIZE];
14590         int ret = 0;
14591         int err = 0;
14592         u64 num;
14593         int init_csum_tree = 0;
14594         int readonly = 0;
14595         int clear_space_cache = 0;
14596         int qgroup_report = 0;
14597         int qgroups_repaired = 0;
14598         unsigned ctree_flags = OPEN_CTREE_EXCLUSIVE;
14599         int force = 0;
14600
14601         while(1) {
14602                 int c;
14603                 enum { GETOPT_VAL_REPAIR = 257, GETOPT_VAL_INIT_CSUM,
14604                         GETOPT_VAL_INIT_EXTENT, GETOPT_VAL_CHECK_CSUM,
14605                         GETOPT_VAL_READONLY, GETOPT_VAL_CHUNK_TREE,
14606                         GETOPT_VAL_MODE, GETOPT_VAL_CLEAR_SPACE_CACHE,
14607                         GETOPT_VAL_FORCE };
14608                 static const struct option long_options[] = {
14609                         { "super", required_argument, NULL, 's' },
14610                         { "repair", no_argument, NULL, GETOPT_VAL_REPAIR },
14611                         { "readonly", no_argument, NULL, GETOPT_VAL_READONLY },
14612                         { "init-csum-tree", no_argument, NULL,
14613                                 GETOPT_VAL_INIT_CSUM },
14614                         { "init-extent-tree", no_argument, NULL,
14615                                 GETOPT_VAL_INIT_EXTENT },
14616                         { "check-data-csum", no_argument, NULL,
14617                                 GETOPT_VAL_CHECK_CSUM },
14618                         { "backup", no_argument, NULL, 'b' },
14619                         { "subvol-extents", required_argument, NULL, 'E' },
14620                         { "qgroup-report", no_argument, NULL, 'Q' },
14621                         { "tree-root", required_argument, NULL, 'r' },
14622                         { "chunk-root", required_argument, NULL,
14623                                 GETOPT_VAL_CHUNK_TREE },
14624                         { "progress", no_argument, NULL, 'p' },
14625                         { "mode", required_argument, NULL,
14626                                 GETOPT_VAL_MODE },
14627                         { "clear-space-cache", required_argument, NULL,
14628                                 GETOPT_VAL_CLEAR_SPACE_CACHE},
14629                         { "force", no_argument, NULL, GETOPT_VAL_FORCE },
14630                         { NULL, 0, NULL, 0}
14631                 };
14632
14633                 c = getopt_long(argc, argv, "as:br:pEQ", long_options, NULL);
14634                 if (c < 0)
14635                         break;
14636                 switch(c) {
14637                         case 'a': /* ignored */ break;
14638                         case 'b':
14639                                 ctree_flags |= OPEN_CTREE_BACKUP_ROOT;
14640                                 break;
14641                         case 's':
14642                                 num = arg_strtou64(optarg);
14643                                 if (num >= BTRFS_SUPER_MIRROR_MAX) {
14644                                         error(
14645                                         "super mirror should be less than %d",
14646                                                 BTRFS_SUPER_MIRROR_MAX);
14647                                         exit(1);
14648                                 }
14649                                 bytenr = btrfs_sb_offset(((int)num));
14650                                 printf("using SB copy %llu, bytenr %llu\n", num,
14651                                        (unsigned long long)bytenr);
14652                                 break;
14653                         case 'Q':
14654                                 qgroup_report = 1;
14655                                 break;
14656                         case 'E':
14657                                 subvolid = arg_strtou64(optarg);
14658                                 break;
14659                         case 'r':
14660                                 tree_root_bytenr = arg_strtou64(optarg);
14661                                 break;
14662                         case GETOPT_VAL_CHUNK_TREE:
14663                                 chunk_root_bytenr = arg_strtou64(optarg);
14664                                 break;
14665                         case 'p':
14666                                 ctx.progress_enabled = true;
14667                                 break;
14668                         case '?':
14669                         case 'h':
14670                                 usage(cmd_check_usage);
14671                         case GETOPT_VAL_REPAIR:
14672                                 printf("enabling repair mode\n");
14673                                 repair = 1;
14674                                 ctree_flags |= OPEN_CTREE_WRITES;
14675                                 break;
14676                         case GETOPT_VAL_READONLY:
14677                                 readonly = 1;
14678                                 break;
14679                         case GETOPT_VAL_INIT_CSUM:
14680                                 printf("Creating a new CRC tree\n");
14681                                 init_csum_tree = 1;
14682                                 repair = 1;
14683                                 ctree_flags |= OPEN_CTREE_WRITES;
14684                                 break;
14685                         case GETOPT_VAL_INIT_EXTENT:
14686                                 init_extent_tree = 1;
14687                                 ctree_flags |= (OPEN_CTREE_WRITES |
14688                                                 OPEN_CTREE_NO_BLOCK_GROUPS);
14689                                 repair = 1;
14690                                 break;
14691                         case GETOPT_VAL_CHECK_CSUM:
14692                                 check_data_csum = 1;
14693                                 break;
14694                         case GETOPT_VAL_MODE:
14695                                 check_mode = parse_check_mode(optarg);
14696                                 if (check_mode == CHECK_MODE_UNKNOWN) {
14697                                         error("unknown mode: %s", optarg);
14698                                         exit(1);
14699                                 }
14700                                 break;
14701                         case GETOPT_VAL_CLEAR_SPACE_CACHE:
14702                                 if (strcmp(optarg, "v1") == 0) {
14703                                         clear_space_cache = 1;
14704                                 } else if (strcmp(optarg, "v2") == 0) {
14705                                         clear_space_cache = 2;
14706                                         ctree_flags |= OPEN_CTREE_INVALIDATE_FST;
14707                                 } else {
14708                                         error(
14709                 "invalid argument to --clear-space-cache, must be v1 or v2");
14710                                         exit(1);
14711                                 }
14712                                 ctree_flags |= OPEN_CTREE_WRITES;
14713                                 break;
14714                         case GETOPT_VAL_FORCE:
14715                                 force = 1;
14716                                 break;
14717                 }
14718         }
14719
14720         if (check_argc_exact(argc - optind, 1))
14721                 usage(cmd_check_usage);
14722
14723         if (ctx.progress_enabled) {
14724                 ctx.tp = TASK_NOTHING;
14725                 ctx.info = task_init(print_status_check, print_status_return, &ctx);
14726         }
14727
14728         /* This check is the only reason for --readonly to exist */
14729         if (readonly && repair) {
14730                 error("repair options are not compatible with --readonly");
14731                 exit(1);
14732         }
14733
14734         /*
14735          * experimental and dangerous
14736          */
14737         if (repair && check_mode == CHECK_MODE_LOWMEM)
14738                 warning("low-memory mode repair support is only partial");
14739
14740         radix_tree_init();
14741         cache_tree_init(&root_cache);
14742
14743         ret = check_mounted(argv[optind]);
14744         if (!force) {
14745                 if (ret < 0) {
14746                         error("could not check mount status: %s",
14747                                         strerror(-ret));
14748                         err |= !!ret;
14749                         goto err_out;
14750                 } else if (ret) {
14751                         error(
14752 "%s is currently mounted, use --force if you really intend to check the filesystem",
14753                                 argv[optind]);
14754                         ret = -EBUSY;
14755                         err |= !!ret;
14756                         goto err_out;
14757                 }
14758         } else {
14759                 if (repair) {
14760                         error("repair and --force is not yet supported");
14761                         ret = 1;
14762                         err |= !!ret;
14763                         goto err_out;
14764                 }
14765                 if (ret < 0) {
14766                         warning(
14767 "cannot check mount status of %s, the filesystem could be mounted, continuing because of --force",
14768                                 argv[optind]);
14769                 } else if (ret) {
14770                         warning(
14771                         "filesystem mounted, continuing because of --force");
14772                 }
14773                 /* A block device is mounted in exclusive mode by kernel */
14774                 ctree_flags &= ~OPEN_CTREE_EXCLUSIVE;
14775         }
14776
14777         /* only allow partial opening under repair mode */
14778         if (repair)
14779                 ctree_flags |= OPEN_CTREE_PARTIAL;
14780
14781         info = open_ctree_fs_info(argv[optind], bytenr, tree_root_bytenr,
14782                                   chunk_root_bytenr, ctree_flags);
14783         if (!info) {
14784                 error("cannot open file system");
14785                 ret = -EIO;
14786                 err |= !!ret;
14787                 goto err_out;
14788         }
14789
14790         global_info = info;
14791         root = info->fs_root;
14792         uuid_unparse(info->super_copy->fsid, uuidbuf);
14793
14794         printf("Checking filesystem on %s\nUUID: %s\n", argv[optind], uuidbuf);
14795
14796         /*
14797          * Check the bare minimum before starting anything else that could rely
14798          * on it, namely the tree roots, any local consistency checks
14799          */
14800         if (!extent_buffer_uptodate(info->tree_root->node) ||
14801             !extent_buffer_uptodate(info->dev_root->node) ||
14802             !extent_buffer_uptodate(info->chunk_root->node)) {
14803                 error("critical roots corrupted, unable to check the filesystem");
14804                 err |= !!ret;
14805                 ret = -EIO;
14806                 goto close_out;
14807         }
14808
14809         if (clear_space_cache) {
14810                 ret = do_clear_free_space_cache(info, clear_space_cache);
14811                 err |= !!ret;
14812                 goto close_out;
14813         }
14814
14815         /*
14816          * repair mode will force us to commit transaction which
14817          * will make us fail to load log tree when mounting.
14818          */
14819         if (repair && btrfs_super_log_root(info->super_copy)) {
14820                 ret = ask_user("repair mode will force to clear out log tree, are you sure?");
14821                 if (!ret) {
14822                         ret = 1;
14823                         err |= !!ret;
14824                         goto close_out;
14825                 }
14826                 ret = zero_log_tree(root);
14827                 err |= !!ret;
14828                 if (ret) {
14829                         error("failed to zero log tree: %d", ret);
14830                         goto close_out;
14831                 }
14832         }
14833
14834         if (qgroup_report) {
14835                 printf("Print quota groups for %s\nUUID: %s\n", argv[optind],
14836                        uuidbuf);
14837                 ret = qgroup_verify_all(info);
14838                 err |= !!ret;
14839                 if (ret == 0)
14840                         report_qgroups(1);
14841                 goto close_out;
14842         }
14843         if (subvolid) {
14844                 printf("Print extent state for subvolume %llu on %s\nUUID: %s\n",
14845                        subvolid, argv[optind], uuidbuf);
14846                 ret = print_extent_state(info, subvolid);
14847                 err |= !!ret;
14848                 goto close_out;
14849         }
14850
14851         if (init_extent_tree || init_csum_tree) {
14852                 struct btrfs_trans_handle *trans;
14853
14854                 trans = btrfs_start_transaction(info->extent_root, 0);
14855                 if (IS_ERR(trans)) {
14856                         error("error starting transaction");
14857                         ret = PTR_ERR(trans);
14858                         err |= !!ret;
14859                         goto close_out;
14860                 }
14861
14862                 if (init_extent_tree) {
14863                         printf("Creating a new extent tree\n");
14864                         ret = reinit_extent_tree(trans, info);
14865                         err |= !!ret;
14866                         if (ret)
14867                                 goto close_out;
14868                 }
14869
14870                 if (init_csum_tree) {
14871                         printf("Reinitialize checksum tree\n");
14872                         ret = btrfs_fsck_reinit_root(trans, info->csum_root, 0);
14873                         if (ret) {
14874                                 error("checksum tree initialization failed: %d",
14875                                                 ret);
14876                                 ret = -EIO;
14877                                 err |= !!ret;
14878                                 goto close_out;
14879                         }
14880
14881                         ret = fill_csum_tree(trans, info->csum_root,
14882                                              init_extent_tree);
14883                         err |= !!ret;
14884                         if (ret) {
14885                                 error("checksum tree refilling failed: %d", ret);
14886                                 return -EIO;
14887                         }
14888                 }
14889                 /*
14890                  * Ok now we commit and run the normal fsck, which will add
14891                  * extent entries for all of the items it finds.
14892                  */
14893                 ret = btrfs_commit_transaction(trans, info->extent_root);
14894                 err |= !!ret;
14895                 if (ret)
14896                         goto close_out;
14897         }
14898         if (!extent_buffer_uptodate(info->extent_root->node)) {
14899                 error("critical: extent_root, unable to check the filesystem");
14900                 ret = -EIO;
14901                 err |= !!ret;
14902                 goto close_out;
14903         }
14904         if (!extent_buffer_uptodate(info->csum_root->node)) {
14905                 error("critical: csum_root, unable to check the filesystem");
14906                 ret = -EIO;
14907                 err |= !!ret;
14908                 goto close_out;
14909         }
14910
14911         if (!init_extent_tree) {
14912                 ret = repair_root_items(info);
14913                 if (ret < 0) {
14914                         err = !!ret;
14915                         error("failed to repair root items: %s", strerror(-ret));
14916                         goto close_out;
14917                 }
14918                 if (repair) {
14919                         fprintf(stderr, "Fixed %d roots.\n", ret);
14920                         ret = 0;
14921                 } else if (ret > 0) {
14922                         fprintf(stderr,
14923                                 "Found %d roots with an outdated root item.\n",
14924                                 ret);
14925                         fprintf(stderr,
14926         "Please run a filesystem check with the option --repair to fix them.\n");
14927                         ret = 1;
14928                         err |= ret;
14929                         goto close_out;
14930                 }
14931         }
14932
14933         ret = do_check_chunks_and_extents(info);
14934         err |= !!ret;
14935         if (ret)
14936                 error(
14937                 "errors found in extent allocation tree or chunk allocation");
14938
14939         /* Only re-check super size after we checked and repaired the fs */
14940         err |= !is_super_size_valid(info);
14941
14942         if (!ctx.progress_enabled) {
14943                 if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE))
14944                         fprintf(stderr, "checking free space tree\n");
14945                 else
14946                         fprintf(stderr, "checking free space cache\n");
14947         }
14948         ret = check_space_cache(root);
14949         err |= !!ret;
14950         if (ret) {
14951                 if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE))
14952                         error("errors found in free space tree");
14953                 else
14954                         error("errors found in free space cache");
14955                 goto out;
14956         }
14957
14958         /*
14959          * We used to have to have these hole extents in between our real
14960          * extents so if we don't have this flag set we need to make sure there
14961          * are no gaps in the file extents for inodes, otherwise we can just
14962          * ignore it when this happens.
14963          */
14964         no_holes = btrfs_fs_incompat(root->fs_info, NO_HOLES);
14965         ret = do_check_fs_roots(info, &root_cache);
14966         err |= !!ret;
14967         if (ret) {
14968                 error("errors found in fs roots");
14969                 goto out;
14970         }
14971
14972         fprintf(stderr, "checking csums\n");
14973         ret = check_csums(root);
14974         err |= !!ret;
14975         if (ret) {
14976                 error("errors found in csum tree");
14977                 goto out;
14978         }
14979
14980         fprintf(stderr, "checking root refs\n");
14981         /* For low memory mode, check_fs_roots_v2 handles root refs */
14982         if (check_mode != CHECK_MODE_LOWMEM) {
14983                 ret = check_root_refs(root, &root_cache);
14984                 err |= !!ret;
14985                 if (ret) {
14986                         error("errors found in root refs");
14987                         goto out;
14988                 }
14989         }
14990
14991         while (repair && !list_empty(&root->fs_info->recow_ebs)) {
14992                 struct extent_buffer *eb;
14993
14994                 eb = list_first_entry(&root->fs_info->recow_ebs,
14995                                       struct extent_buffer, recow);
14996                 list_del_init(&eb->recow);
14997                 ret = recow_extent_buffer(root, eb);
14998                 err |= !!ret;
14999                 if (ret) {
15000                         error("fails to fix transid errors");
15001                         break;
15002                 }
15003         }
15004
15005         while (!list_empty(&delete_items)) {
15006                 struct bad_item *bad;
15007
15008                 bad = list_first_entry(&delete_items, struct bad_item, list);
15009                 list_del_init(&bad->list);
15010                 if (repair) {
15011                         ret = delete_bad_item(root, bad);
15012                         err |= !!ret;
15013                 }
15014                 free(bad);
15015         }
15016
15017         if (info->quota_enabled) {
15018                 fprintf(stderr, "checking quota groups\n");
15019                 ret = qgroup_verify_all(info);
15020                 err |= !!ret;
15021                 if (ret) {
15022                         error("failed to check quota groups");
15023                         goto out;
15024                 }
15025                 report_qgroups(0);
15026                 ret = repair_qgroups(info, &qgroups_repaired);
15027                 err |= !!ret;
15028                 if (err) {
15029                         error("failed to repair quota groups");
15030                         goto out;
15031                 }
15032                 ret = 0;
15033         }
15034
15035         if (!list_empty(&root->fs_info->recow_ebs)) {
15036                 error("transid errors in file system");
15037                 ret = 1;
15038                 err |= !!ret;
15039         }
15040 out:
15041         printf("found %llu bytes used, ",
15042                (unsigned long long)bytes_used);
15043         if (err)
15044                 printf("error(s) found\n");
15045         else
15046                 printf("no error found\n");
15047         printf("total csum bytes: %llu\n",(unsigned long long)total_csum_bytes);
15048         printf("total tree bytes: %llu\n",
15049                (unsigned long long)total_btree_bytes);
15050         printf("total fs tree bytes: %llu\n",
15051                (unsigned long long)total_fs_tree_bytes);
15052         printf("total extent tree bytes: %llu\n",
15053                (unsigned long long)total_extent_tree_bytes);
15054         printf("btree space waste bytes: %llu\n",
15055                (unsigned long long)btree_space_waste);
15056         printf("file data blocks allocated: %llu\n referenced %llu\n",
15057                 (unsigned long long)data_bytes_allocated,
15058                 (unsigned long long)data_bytes_referenced);
15059
15060         free_qgroup_counts();
15061         free_root_recs_tree(&root_cache);
15062 close_out:
15063         close_ctree(root);
15064 err_out:
15065         if (ctx.progress_enabled)
15066                 task_deinit(ctx.info);
15067
15068         return err;
15069 }