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 "ulist.h"
  44 #include "hash.h"
  45
  46 enum task_position {
  47         TASK_EXTENTS,
  48         TASK_FREE_SPACE,
  49         TASK_FS_ROOTS,
  50         TASK_NOTHING, /* have to be the last element */
  51 };
  52
  53 struct task_ctx {
  54         int progress_enabled;
  55         enum task_position tp;
  56
  57         struct task_info *info;
  58 };
  59
  60 static u64 bytes_used = 0;
  61 static u64 total_csum_bytes = 0;
  62 static u64 total_btree_bytes = 0;
  63 static u64 total_fs_tree_bytes = 0;
  64 static u64 total_extent_tree_bytes = 0;
  65 static u64 btree_space_waste = 0;
  66 static u64 data_bytes_allocated = 0;
  67 static u64 data_bytes_referenced = 0;
  68 static int found_old_backref = 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 list_head list;
  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* to_extent_backref(struct list_head *entry)
  97 {
  98         return list_entry(entry, struct extent_backref, list);
  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
 134 static inline struct data_backref* to_data_backref(struct extent_backref *back)
 135 {
 136         return container_of(back, struct data_backref, node);
 137 }
 138
 139 /*
 140  * Much like data_backref, just removed the undetermined members
 141  * and change it to use list_head.
 142  * During extent scan, it is stored in root->orphan_data_extent.
 143  * During fs tree scan, it is then moved to inode_rec->orphan_data_extents.
 144  */
 145 struct orphan_data_extent {
 146         struct list_head list;
 147         u64 root;
 148         u64 objectid;
 149         u64 offset;
 150         u64 disk_bytenr;
 151         u64 disk_len;
 152 };
 153
 154 struct tree_backref {
 155         struct extent_backref node;
 156         union {
 157                 u64 parent;
 158                 u64 root;
 159         };
 160 };
 161
 162 static inline struct tree_backref* to_tree_backref(struct extent_backref *back)
 163 {
 164         return container_of(back, struct tree_backref, node);
 165 }
 166
 167 /* Explicit initialization for extent_record::flag_block_full_backref */
 168 enum { FLAG_UNSET = 2 };
 169
 170 struct extent_record {
 171         struct list_head backrefs;
 172         struct list_head dups;
 173         struct list_head list;
 174         struct cache_extent cache;
 175         struct btrfs_disk_key parent_key;
 176         u64 start;
 177         u64 max_size;
 178         u64 nr;
 179         u64 refs;
 180         u64 extent_item_refs;
 181         u64 generation;
 182         u64 parent_generation;
 183         u64 info_objectid;
 184         u32 num_duplicates;
 185         u8 info_level;
 186         unsigned int flag_block_full_backref:2;
 187         unsigned int found_rec:1;
 188         unsigned int content_checked:1;
 189         unsigned int owner_ref_checked:1;
 190         unsigned int is_root:1;
 191         unsigned int metadata:1;
 192         unsigned int bad_full_backref:1;
 193         unsigned int crossing_stripes:1;
 194         unsigned int wrong_chunk_type:1;
 195 };
 196
 197 static inline struct extent_record* to_extent_record(struct list_head *entry)
 198 {
 199         return container_of(entry, struct extent_record, list);
 200 }
 201
 202 struct inode_backref {
 203         struct list_head list;
 204         unsigned int found_dir_item:1;
 205         unsigned int found_dir_index:1;
 206         unsigned int found_inode_ref:1;
 207         u8 filetype;
 208         u8 ref_type;
 209         int errors;
 210         u64 dir;
 211         u64 index;
 212         u16 namelen;
 213         char name[0];
 214 };
 215
 216 static inline struct inode_backref* to_inode_backref(struct list_head *entry)
 217 {
 218         return list_entry(entry, struct inode_backref, list);
 219 }
 220
 221 struct root_item_record {
 222         struct list_head list;
 223         u64 objectid;
 224         u64 bytenr;
 225         u64 last_snapshot;
 226         u8 level;
 227         u8 drop_level;
 228         int level_size;
 229         struct btrfs_key drop_key;
 230 };
 231
 232 #define REF_ERR_NO_DIR_ITEM             (1 << 0)
 233 #define REF_ERR_NO_DIR_INDEX            (1 << 1)
 234 #define REF_ERR_NO_INODE_REF            (1 << 2)
 235 #define REF_ERR_DUP_DIR_ITEM            (1 << 3)
 236 #define REF_ERR_DUP_DIR_INDEX           (1 << 4)
 237 #define REF_ERR_DUP_INODE_REF           (1 << 5)
 238 #define REF_ERR_INDEX_UNMATCH           (1 << 6)
 239 #define REF_ERR_FILETYPE_UNMATCH        (1 << 7)
 240 #define REF_ERR_NAME_TOO_LONG           (1 << 8) // 100
 241 #define REF_ERR_NO_ROOT_REF             (1 << 9)
 242 #define REF_ERR_NO_ROOT_BACKREF         (1 << 10)
 243 #define REF_ERR_DUP_ROOT_REF            (1 << 11)
 244 #define REF_ERR_DUP_ROOT_BACKREF        (1 << 12)
 245
 246 struct file_extent_hole {
 247         struct rb_node node;
 248         u64 start;
 249         u64 len;
 250 };
 251
 252 struct inode_record {
 253         struct list_head backrefs;
 254         unsigned int checked:1;
 255         unsigned int merging:1;
 256         unsigned int found_inode_item:1;
 257         unsigned int found_dir_item:1;
 258         unsigned int found_file_extent:1;
 259         unsigned int found_csum_item:1;
 260         unsigned int some_csum_missing:1;
 261         unsigned int nodatasum:1;
 262         int errors;
 263
 264         u64 ino;
 265         u32 nlink;
 266         u32 imode;
 267         u64 isize;
 268         u64 nbytes;
 269
 270         u32 found_link;
 271         u64 found_size;
 272         u64 extent_start;
 273         u64 extent_end;
 274         struct rb_root holes;
 275         struct list_head orphan_extents;
 276
 277         u32 refs;
 278 };
 279
 280 #define I_ERR_NO_INODE_ITEM             (1 << 0)
 281 #define I_ERR_NO_ORPHAN_ITEM            (1 << 1)
 282 #define I_ERR_DUP_INODE_ITEM            (1 << 2)
 283 #define I_ERR_DUP_DIR_INDEX             (1 << 3)
 284 #define I_ERR_ODD_DIR_ITEM              (1 << 4)
 285 #define I_ERR_ODD_FILE_EXTENT           (1 << 5)
 286 #define I_ERR_BAD_FILE_EXTENT           (1 << 6)
 287 #define I_ERR_FILE_EXTENT_OVERLAP       (1 << 7)
 288 #define I_ERR_FILE_EXTENT_DISCOUNT      (1 << 8) // 100
 289 #define I_ERR_DIR_ISIZE_WRONG           (1 << 9)
 290 #define I_ERR_FILE_NBYTES_WRONG         (1 << 10) // 400
 291 #define I_ERR_ODD_CSUM_ITEM             (1 << 11)
 292 #define I_ERR_SOME_CSUM_MISSING         (1 << 12)
 293 #define I_ERR_LINK_COUNT_WRONG          (1 << 13)
 294 #define I_ERR_FILE_EXTENT_ORPHAN        (1 << 14)
 295
 296 struct root_backref {
 297         struct list_head list;
 298         unsigned int found_dir_item:1;
 299         unsigned int found_dir_index:1;
 300         unsigned int found_back_ref:1;
 301         unsigned int found_forward_ref:1;
 302         unsigned int reachable:1;
 303         int errors;
 304         u64 ref_root;
 305         u64 dir;
 306         u64 index;
 307         u16 namelen;
 308         char name[0];
 309 };
 310
 311 static inline struct root_backref* to_root_backref(struct list_head *entry)
 312 {
 313         return list_entry(entry, struct root_backref, list);
 314 }
 315
 316 struct root_record {
 317         struct list_head backrefs;
 318         struct cache_extent cache;
 319         unsigned int found_root_item:1;
 320         u64 objectid;
 321         u32 found_ref;
 322 };
 323
 324 struct ptr_node {
 325         struct cache_extent cache;
 326         void *data;
 327 };
 328
 329 struct shared_node {
 330         struct cache_extent cache;
 331         struct cache_tree root_cache;
 332         struct cache_tree inode_cache;
 333         struct inode_record *current;
 334         u32 refs;
 335 };
 336
 337 struct block_info {
 338         u64 start;
 339         u32 size;
 340 };
 341
 342 struct walk_control {
 343         struct cache_tree shared;
 344         struct shared_node *nodes[BTRFS_MAX_LEVEL];
 345         int active_node;
 346         int root_level;
 347 };
 348
 349 struct bad_item {
 350         struct btrfs_key key;
 351         u64 root_id;
 352         struct list_head list;
 353 };
 354
 355 struct extent_entry {
 356         u64 bytenr;
 357         u64 bytes;
 358         int count;
 359         int broken;
 360         struct list_head list;
 361 };
 362
 363 struct root_item_info {
 364         /* level of the root */
 365         u8 level;
 366         /* number of nodes at this level, must be 1 for a root */
 367         int node_count;
 368         u64 bytenr;
 369         u64 gen;
 370         struct cache_extent cache_extent;
 371 };
 372
 373 /*
 374  * Error bit for low memory mode check.
 375  *
 376  * Currently no caller cares about it yet.  Just internal use for error
 377  * classification.
 378  */
 379 #define BACKREF_MISSING         (1 << 0) /* Backref missing in extent tree */
 380 #define BACKREF_MISMATCH        (1 << 1) /* Backref exists but does not match */
 381 #define BYTES_UNALIGNED         (1 << 2) /* Some bytes are not aligned */
 382 #define REFERENCER_MISSING      (1 << 3) /* Referencer not found */
 383 #define REFERENCER_MISMATCH     (1 << 4) /* Referenceer found but does not match */
 384 #define CROSSING_STRIPE_BOUNDARY (1 << 4) /* For kernel scrub workaround */
 385 #define ITEM_SIZE_MISMATCH      (1 << 5) /* Bad item size */
 386 #define UNKNOWN_TYPE            (1 << 6) /* Unknown type */
 387 #define ACCOUNTING_MISMATCH     (1 << 7) /* Used space accounting error */
 388 #define CHUNK_TYPE_MISMATCH     (1 << 8)
 389
 390 static void *print_status_check(void *p)
 391 {
 392         struct task_ctx *priv = p;
 393         const char work_indicator[] = { '.', 'o', 'O', 'o' };
 394         uint32_t count = 0;
 395         static char *task_position_string[] = {
 396                 "checking extents",
 397                 "checking free space cache",
 398                 "checking fs roots",
 399         };
 400
 401         task_period_start(priv->info, 1000 /* 1s */);
 402
 403         if (priv->tp == TASK_NOTHING)
 404                 return NULL;
 405
 406         while (1) {
 407                 printf("%s [%c]\r", task_position_string[priv->tp],
 408                                 work_indicator[count % 4]);
 409                 count++;
 410                 fflush(stdout);
 411                 task_period_wait(priv->info);
 412         }
 413         return NULL;
 414 }
 415
 416 static int print_status_return(void *p)
 417 {
 418         printf("\n");
 419         fflush(stdout);
 420
 421         return 0;
 422 }
 423
 424 static enum btrfs_check_mode parse_check_mode(const char *str)
 425 {
 426         if (strcmp(str, "lowmem") == 0)
 427                 return CHECK_MODE_LOWMEM;
 428         if (strcmp(str, "orig") == 0)
 429                 return CHECK_MODE_ORIGINAL;
 430         if (strcmp(str, "original") == 0)
 431                 return CHECK_MODE_ORIGINAL;
 432
 433         return CHECK_MODE_UNKNOWN;
 434 }
 435
 436 /* Compatible function to allow reuse of old codes */
 437 static u64 first_extent_gap(struct rb_root *holes)
 438 {
 439         struct file_extent_hole *hole;
 440
 441         if (RB_EMPTY_ROOT(holes))
 442                 return (u64)-1;
 443
 444         hole = rb_entry(rb_first(holes), struct file_extent_hole, node);
 445         return hole->start;
 446 }
 447
 448 static int compare_hole(struct rb_node *node1, struct rb_node *node2)
 449 {
 450         struct file_extent_hole *hole1;
 451         struct file_extent_hole *hole2;
 452
 453         hole1 = rb_entry(node1, struct file_extent_hole, node);
 454         hole2 = rb_entry(node2, struct file_extent_hole, node);
 455
 456         if (hole1->start > hole2->start)
 457                 return -1;
 458         if (hole1->start < hole2->start)
 459                 return 1;
 460         /* Now hole1->start == hole2->start */
 461         if (hole1->len >= hole2->len)
 462                 /*
 463                  * Hole 1 will be merge center
 464                  * Same hole will be merged later
 465                  */
 466                 return -1;
 467         /* Hole 2 will be merge center */
 468         return 1;
 469 }
 470
 471 /*
 472  * Add a hole to the record
 473  *
 474  * This will do hole merge for copy_file_extent_holes(),
 475  * which will ensure there won't be continuous holes.
 476  */
 477 static int add_file_extent_hole(struct rb_root *holes,
 478                                 u64 start, u64 len)
 479 {
 480         struct file_extent_hole *hole;
 481         struct file_extent_hole *prev = NULL;
 482         struct file_extent_hole *next = NULL;
 483
 484         hole = malloc(sizeof(*hole));
 485         if (!hole)
 486                 return -ENOMEM;
 487         hole->start = start;
 488         hole->len = len;
 489         /* Since compare will not return 0, no -EEXIST will happen */
 490         rb_insert(holes, &hole->node, compare_hole);
 491
 492         /* simple merge with previous hole */
 493         if (rb_prev(&hole->node))
 494                 prev = rb_entry(rb_prev(&hole->node), struct file_extent_hole,
 495                                 node);
 496         if (prev && prev->start + prev->len >= hole->start) {
 497                 hole->len = hole->start + hole->len - prev->start;
 498                 hole->start = prev->start;
 499                 rb_erase(&prev->node, holes);
 500                 free(prev);
 501                 prev = NULL;
 502         }
 503
 504         /* iterate merge with next holes */
 505         while (1) {
 506                 if (!rb_next(&hole->node))
 507                         break;
 508                 next = rb_entry(rb_next(&hole->node), struct file_extent_hole,
 509                                         node);
 510                 if (hole->start + hole->len >= next->start) {
 511                         if (hole->start + hole->len <= next->start + next->len)
 512                                 hole->len = next->start + next->len -
 513                                             hole->start;
 514                         rb_erase(&next->node, holes);
 515                         free(next);
 516                         next = NULL;
 517                 } else
 518                         break;
 519         }
 520         return 0;
 521 }
 522
 523 static int compare_hole_range(struct rb_node *node, void *data)
 524 {
 525         struct file_extent_hole *hole;
 526         u64 start;
 527
 528         hole = (struct file_extent_hole *)data;
 529         start = hole->start;
 530
 531         hole = rb_entry(node, struct file_extent_hole, node);
 532         if (start < hole->start)
 533                 return -1;
 534         if (start >= hole->start && start < hole->start + hole->len)
 535                 return 0;
 536         return 1;
 537 }
 538
 539 /*
 540  * Delete a hole in the record
 541  *
 542  * This will do the hole split and is much restrict than add.
 543  */
 544 static int del_file_extent_hole(struct rb_root *holes,
 545                                 u64 start, u64 len)
 546 {
 547         struct file_extent_hole *hole;
 548         struct file_extent_hole tmp;
 549         u64 prev_start = 0;
 550         u64 prev_len = 0;
 551         u64 next_start = 0;
 552         u64 next_len = 0;
 553         struct rb_node *node;
 554         int have_prev = 0;
 555         int have_next = 0;
 556         int ret = 0;
 557
 558         tmp.start = start;
 559         tmp.len = len;
 560         node = rb_search(holes, &tmp, compare_hole_range, NULL);
 561         if (!node)
 562                 return -EEXIST;
 563         hole = rb_entry(node, struct file_extent_hole, node);
 564         if (start + len > hole->start + hole->len)
 565                 return -EEXIST;
 566
 567         /*
 568          * Now there will be no overlap, delete the hole and re-add the
 569          * split(s) if they exists.
 570          */
 571         if (start > hole->start) {
 572                 prev_start = hole->start;
 573                 prev_len = start - hole->start;
 574                 have_prev = 1;
 575         }
 576         if (hole->start + hole->len > start + len) {
 577                 next_start = start + len;
 578                 next_len = hole->start + hole->len - start - len;
 579                 have_next = 1;
 580         }
 581         rb_erase(node, holes);
 582         free(hole);
 583         if (have_prev) {
 584                 ret = add_file_extent_hole(holes, prev_start, prev_len);
 585                 if (ret < 0)
 586                         return ret;
 587         }
 588         if (have_next) {
 589                 ret = add_file_extent_hole(holes, next_start, next_len);
 590                 if (ret < 0)
 591                         return ret;
 592         }
 593         return 0;
 594 }
 595
 596 static int copy_file_extent_holes(struct rb_root *dst,
 597                                   struct rb_root *src)
 598 {
 599         struct file_extent_hole *hole;
 600         struct rb_node *node;
 601         int ret = 0;
 602
 603         node = rb_first(src);
 604         while (node) {
 605                 hole = rb_entry(node, struct file_extent_hole, node);
 606                 ret = add_file_extent_hole(dst, hole->start, hole->len);
 607                 if (ret)
 608                         break;
 609                 node = rb_next(node);
 610         }
 611         return ret;
 612 }
 613
 614 static void free_file_extent_holes(struct rb_root *holes)
 615 {
 616         struct rb_node *node;
 617         struct file_extent_hole *hole;
 618
 619         node = rb_first(holes);
 620         while (node) {
 621                 hole = rb_entry(node, struct file_extent_hole, node);
 622                 rb_erase(node, holes);
 623                 free(hole);
 624                 node = rb_first(holes);
 625         }
 626 }
 627
 628 static void reset_cached_block_groups(struct btrfs_fs_info *fs_info);
 629
 630 static void record_root_in_trans(struct btrfs_trans_handle *trans,
 631                                  struct btrfs_root *root)
 632 {
 633         if (root->last_trans != trans->transid) {
 634                 root->track_dirty = 1;
 635                 root->last_trans = trans->transid;
 636                 root->commit_root = root->node;
 637                 extent_buffer_get(root->node);
 638         }
 639 }
 640
 641 static u8 imode_to_type(u32 imode)
 642 {
 643 #define S_SHIFT 12
 644         static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
 645                 [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
 646                 [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
 647                 [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
 648                 [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
 649                 [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
 650                 [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
 651                 [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
 652         };
 653
 654         return btrfs_type_by_mode[(imode & S_IFMT) >> S_SHIFT];
 655 #undef S_SHIFT
 656 }
 657
 658 static int device_record_compare(struct rb_node *node1, struct rb_node *node2)
 659 {
 660         struct device_record *rec1;
 661         struct device_record *rec2;
 662
 663         rec1 = rb_entry(node1, struct device_record, node);
 664         rec2 = rb_entry(node2, struct device_record, node);
 665         if (rec1->devid > rec2->devid)
 666                 return -1;
 667         else if (rec1->devid < rec2->devid)
 668                 return 1;
 669         else
 670                 return 0;
 671 }
 672
 673 static struct inode_record *clone_inode_rec(struct inode_record *orig_rec)
 674 {
 675         struct inode_record *rec;
 676         struct inode_backref *backref;
 677         struct inode_backref *orig;
 678         struct inode_backref *tmp;
 679         struct orphan_data_extent *src_orphan;
 680         struct orphan_data_extent *dst_orphan;
 681         struct rb_node *rb;
 682         size_t size;
 683         int ret;
 684
 685         rec = malloc(sizeof(*rec));
 686         if (!rec)
 687                 return ERR_PTR(-ENOMEM);
 688         memcpy(rec, orig_rec, sizeof(*rec));
 689         rec->refs = 1;
 690         INIT_LIST_HEAD(&rec->backrefs);
 691         INIT_LIST_HEAD(&rec->orphan_extents);
 692         rec->holes = RB_ROOT;
 693
 694         list_for_each_entry(orig, &orig_rec->backrefs, list) {
 695                 size = sizeof(*orig) + orig->namelen + 1;
 696                 backref = malloc(size);
 697                 if (!backref) {
 698                         ret = -ENOMEM;
 699                         goto cleanup;
 700                 }
 701                 memcpy(backref, orig, size);
 702                 list_add_tail(&backref->list, &rec->backrefs);
 703         }
 704         list_for_each_entry(src_orphan, &orig_rec->orphan_extents, list) {
 705                 dst_orphan = malloc(sizeof(*dst_orphan));
 706                 if (!dst_orphan) {
 707                         ret = -ENOMEM;
 708                         goto cleanup;
 709                 }
 710                 memcpy(dst_orphan, src_orphan, sizeof(*src_orphan));
 711                 list_add_tail(&dst_orphan->list, &rec->orphan_extents);
 712         }
 713         ret = copy_file_extent_holes(&rec->holes, &orig_rec->holes);
 714         if (ret < 0)
 715                 goto cleanup_rb;
 716
 717         return rec;
 718
 719 cleanup_rb:
 720         rb = rb_first(&rec->holes);
 721         while (rb) {
 722                 struct file_extent_hole *hole;
 723
 724                 hole = rb_entry(rb, struct file_extent_hole, node);
 725                 rb = rb_next(rb);
 726                 free(hole);
 727         }
 728
 729 cleanup:
 730         if (!list_empty(&rec->backrefs))
 731                 list_for_each_entry_safe(orig, tmp, &rec->backrefs, list) {
 732                         list_del(&orig->list);
 733                         free(orig);
 734                 }
 735
 736         if (!list_empty(&rec->orphan_extents))
 737                 list_for_each_entry_safe(orig, tmp, &rec->orphan_extents, list) {
 738                         list_del(&orig->list);
 739                         free(orig);
 740                 }
 741
 742         free(rec);
 743
 744         return ERR_PTR(ret);
 745 }
 746
 747 static void print_orphan_data_extents(struct list_head *orphan_extents,
 748                                       u64 objectid)
 749 {
 750         struct orphan_data_extent *orphan;
 751
 752         if (list_empty(orphan_extents))
 753                 return;
 754         printf("The following data extent is lost in tree %llu:\n",
 755                objectid);
 756         list_for_each_entry(orphan, orphan_extents, list) {
 757                 printf("\tinode: %llu, offset:%llu, disk_bytenr: %llu, disk_len: %llu\n",
 758                        orphan->objectid, orphan->offset, orphan->disk_bytenr,
 759                        orphan->disk_len);
 760         }
 761 }
 762
 763 static void print_inode_error(struct btrfs_root *root, struct inode_record *rec)
 764 {
 765         u64 root_objectid = root->root_key.objectid;
 766         int errors = rec->errors;
 767
 768         if (!errors)
 769                 return;
 770         /* reloc root errors, we print its corresponding fs root objectid*/
 771         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
 772                 root_objectid = root->root_key.offset;
 773                 fprintf(stderr, "reloc");
 774         }
 775         fprintf(stderr, "root %llu inode %llu errors %x",
 776                 (unsigned long long) root_objectid,
 777                 (unsigned long long) rec->ino, rec->errors);
 778
 779         if (errors & I_ERR_NO_INODE_ITEM)
 780                 fprintf(stderr, ", no inode item");
 781         if (errors & I_ERR_NO_ORPHAN_ITEM)
 782                 fprintf(stderr, ", no orphan item");
 783         if (errors & I_ERR_DUP_INODE_ITEM)
 784                 fprintf(stderr, ", dup inode item");
 785         if (errors & I_ERR_DUP_DIR_INDEX)
 786                 fprintf(stderr, ", dup dir index");
 787         if (errors & I_ERR_ODD_DIR_ITEM)
 788                 fprintf(stderr, ", odd dir item");
 789         if (errors & I_ERR_ODD_FILE_EXTENT)
 790                 fprintf(stderr, ", odd file extent");
 791         if (errors & I_ERR_BAD_FILE_EXTENT)
 792                 fprintf(stderr, ", bad file extent");
 793         if (errors & I_ERR_FILE_EXTENT_OVERLAP)
 794                 fprintf(stderr, ", file extent overlap");
 795         if (errors & I_ERR_FILE_EXTENT_DISCOUNT)
 796                 fprintf(stderr, ", file extent discount");
 797         if (errors & I_ERR_DIR_ISIZE_WRONG)
 798                 fprintf(stderr, ", dir isize wrong");
 799         if (errors & I_ERR_FILE_NBYTES_WRONG)
 800                 fprintf(stderr, ", nbytes wrong");
 801         if (errors & I_ERR_ODD_CSUM_ITEM)
 802                 fprintf(stderr, ", odd csum item");
 803         if (errors & I_ERR_SOME_CSUM_MISSING)
 804                 fprintf(stderr, ", some csum missing");
 805         if (errors & I_ERR_LINK_COUNT_WRONG)
 806                 fprintf(stderr, ", link count wrong");
 807         if (errors & I_ERR_FILE_EXTENT_ORPHAN)
 808                 fprintf(stderr, ", orphan file extent");
 809         fprintf(stderr, "\n");
 810         /* Print the orphan extents if needed */
 811         if (errors & I_ERR_FILE_EXTENT_ORPHAN)
 812                 print_orphan_data_extents(&rec->orphan_extents, root->objectid);
 813
 814         /* Print the holes if needed */
 815         if (errors & I_ERR_FILE_EXTENT_DISCOUNT) {
 816                 struct file_extent_hole *hole;
 817                 struct rb_node *node;
 818                 int found = 0;
 819
 820                 node = rb_first(&rec->holes);
 821                 fprintf(stderr, "Found file extent holes:\n");
 822                 while (node) {
 823                         found = 1;
 824                         hole = rb_entry(node, struct file_extent_hole, node);
 825                         fprintf(stderr, "\tstart: %llu, len: %llu\n",
 826                                 hole->start, hole->len);
 827                         node = rb_next(node);
 828                 }
 829                 if (!found)
 830                         fprintf(stderr, "\tstart: 0, len: %llu\n",
 831                                 round_up(rec->isize, root->sectorsize));
 832         }
 833 }
 834
 835 static void print_ref_error(int errors)
 836 {
 837         if (errors & REF_ERR_NO_DIR_ITEM)
 838                 fprintf(stderr, ", no dir item");
 839         if (errors & REF_ERR_NO_DIR_INDEX)
 840                 fprintf(stderr, ", no dir index");
 841         if (errors & REF_ERR_NO_INODE_REF)
 842                 fprintf(stderr, ", no inode ref");
 843         if (errors & REF_ERR_DUP_DIR_ITEM)
 844                 fprintf(stderr, ", dup dir item");
 845         if (errors & REF_ERR_DUP_DIR_INDEX)
 846                 fprintf(stderr, ", dup dir index");
 847         if (errors & REF_ERR_DUP_INODE_REF)
 848                 fprintf(stderr, ", dup inode ref");
 849         if (errors & REF_ERR_INDEX_UNMATCH)
 850                 fprintf(stderr, ", index mismatch");
 851         if (errors & REF_ERR_FILETYPE_UNMATCH)
 852                 fprintf(stderr, ", filetype mismatch");
 853         if (errors & REF_ERR_NAME_TOO_LONG)
 854                 fprintf(stderr, ", name too long");
 855         if (errors & REF_ERR_NO_ROOT_REF)
 856                 fprintf(stderr, ", no root ref");
 857         if (errors & REF_ERR_NO_ROOT_BACKREF)
 858                 fprintf(stderr, ", no root backref");
 859         if (errors & REF_ERR_DUP_ROOT_REF)
 860                 fprintf(stderr, ", dup root ref");
 861         if (errors & REF_ERR_DUP_ROOT_BACKREF)
 862                 fprintf(stderr, ", dup root backref");
 863         fprintf(stderr, "\n");
 864 }
 865
 866 static struct inode_record *get_inode_rec(struct cache_tree *inode_cache,
 867                                           u64 ino, int mod)
 868 {
 869         struct ptr_node *node;
 870         struct cache_extent *cache;
 871         struct inode_record *rec = NULL;
 872         int ret;
 873
 874         cache = lookup_cache_extent(inode_cache, ino, 1);
 875         if (cache) {
 876                 node = container_of(cache, struct ptr_node, cache);
 877                 rec = node->data;
 878                 if (mod && rec->refs > 1) {
 879                         node->data = clone_inode_rec(rec);
 880                         if (IS_ERR(node->data))
 881                                 return node->data;
 882                         rec->refs--;
 883                         rec = node->data;
 884                 }
 885         } else if (mod) {
 886                 rec = calloc(1, sizeof(*rec));
 887                 if (!rec)
 888                         return ERR_PTR(-ENOMEM);
 889                 rec->ino = ino;
 890                 rec->extent_start = (u64)-1;
 891                 rec->refs = 1;
 892                 INIT_LIST_HEAD(&rec->backrefs);
 893                 INIT_LIST_HEAD(&rec->orphan_extents);
 894                 rec->holes = RB_ROOT;
 895
 896                 node = malloc(sizeof(*node));
 897                 if (!node) {
 898                         free(rec);
 899                         return ERR_PTR(-ENOMEM);
 900                 }
 901                 node->cache.start = ino;
 902                 node->cache.size = 1;
 903                 node->data = rec;
 904
 905                 if (ino == BTRFS_FREE_INO_OBJECTID)
 906                         rec->found_link = 1;
 907
 908                 ret = insert_cache_extent(inode_cache, &node->cache);
 909                 if (ret)
 910                         return ERR_PTR(-EEXIST);
 911         }
 912         return rec;
 913 }
 914
 915 static void free_orphan_data_extents(struct list_head *orphan_extents)
 916 {
 917         struct orphan_data_extent *orphan;
 918
 919         while (!list_empty(orphan_extents)) {
 920                 orphan = list_entry(orphan_extents->next,
 921                                     struct orphan_data_extent, list);
 922                 list_del(&orphan->list);
 923                 free(orphan);
 924         }
 925 }
 926
 927 static void free_inode_rec(struct inode_record *rec)
 928 {
 929         struct inode_backref *backref;
 930
 931         if (--rec->refs > 0)
 932                 return;
 933
 934         while (!list_empty(&rec->backrefs)) {
 935                 backref = to_inode_backref(rec->backrefs.next);
 936                 list_del(&backref->list);
 937                 free(backref);
 938         }
 939         free_orphan_data_extents(&rec->orphan_extents);
 940         free_file_extent_holes(&rec->holes);
 941         free(rec);
 942 }
 943
 944 static int can_free_inode_rec(struct inode_record *rec)
 945 {
 946         if (!rec->errors && rec->checked && rec->found_inode_item &&
 947             rec->nlink == rec->found_link && list_empty(&rec->backrefs))
 948                 return 1;
 949         return 0;
 950 }
 951
 952 static void maybe_free_inode_rec(struct cache_tree *inode_cache,
 953                                  struct inode_record *rec)
 954 {
 955         struct cache_extent *cache;
 956         struct inode_backref *tmp, *backref;
 957         struct ptr_node *node;
 958         u8 filetype;
 959
 960         if (!rec->found_inode_item)
 961                 return;
 962
 963         filetype = imode_to_type(rec->imode);
 964         list_for_each_entry_safe(backref, tmp, &rec->backrefs, list) {
 965                 if (backref->found_dir_item && backref->found_dir_index) {
 966                         if (backref->filetype != filetype)
 967                                 backref->errors |= REF_ERR_FILETYPE_UNMATCH;
 968                         if (!backref->errors && backref->found_inode_ref &&
 969                             rec->nlink == rec->found_link) {
 970                                 list_del(&backref->list);
 971                                 free(backref);
 972                         }
 973                 }
 974         }
 975
 976         if (!rec->checked || rec->merging)
 977                 return;
 978
 979         if (S_ISDIR(rec->imode)) {
 980                 if (rec->found_size != rec->isize)
 981                         rec->errors |= I_ERR_DIR_ISIZE_WRONG;
 982                 if (rec->found_file_extent)
 983                         rec->errors |= I_ERR_ODD_FILE_EXTENT;
 984         } else if (S_ISREG(rec->imode) || S_ISLNK(rec->imode)) {
 985                 if (rec->found_dir_item)
 986                         rec->errors |= I_ERR_ODD_DIR_ITEM;
 987                 if (rec->found_size != rec->nbytes)
 988                         rec->errors |= I_ERR_FILE_NBYTES_WRONG;
 989                 if (rec->nlink > 0 && !no_holes &&
 990                     (rec->extent_end < rec->isize ||
 991                      first_extent_gap(&rec->holes) < rec->isize))
 992                         rec->errors |= I_ERR_FILE_EXTENT_DISCOUNT;
 993         }
 994
 995         if (S_ISREG(rec->imode) || S_ISLNK(rec->imode)) {
 996                 if (rec->found_csum_item && rec->nodatasum)
 997                         rec->errors |= I_ERR_ODD_CSUM_ITEM;
 998                 if (rec->some_csum_missing && !rec->nodatasum)
 999                         rec->errors |= I_ERR_SOME_CSUM_MISSING;
1000         }
1001
1002         BUG_ON(rec->refs != 1);
1003         if (can_free_inode_rec(rec)) {
1004                 cache = lookup_cache_extent(inode_cache, rec->ino, 1);
1005                 node = container_of(cache, struct ptr_node, cache);
1006                 BUG_ON(node->data != rec);
1007                 remove_cache_extent(inode_cache, &node->cache);
1008                 free(node);
1009                 free_inode_rec(rec);
1010         }
1011 }
1012
1013 static int check_orphan_item(struct btrfs_root *root, u64 ino)
1014 {
1015         struct btrfs_path path;
1016         struct btrfs_key key;
1017         int ret;
1018
1019         key.objectid = BTRFS_ORPHAN_OBJECTID;
1020         key.type = BTRFS_ORPHAN_ITEM_KEY;
1021         key.offset = ino;
1022
1023         btrfs_init_path(&path);
1024         ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
1025         btrfs_release_path(&path);
1026         if (ret > 0)
1027                 ret = -ENOENT;
1028         return ret;
1029 }
1030
1031 static int process_inode_item(struct extent_buffer *eb,
1032                               int slot, struct btrfs_key *key,
1033                               struct shared_node *active_node)
1034 {
1035         struct inode_record *rec;
1036         struct btrfs_inode_item *item;
1037
1038         rec = active_node->current;
1039         BUG_ON(rec->ino != key->objectid || rec->refs > 1);
1040         if (rec->found_inode_item) {
1041                 rec->errors |= I_ERR_DUP_INODE_ITEM;
1042                 return 1;
1043         }
1044         item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
1045         rec->nlink = btrfs_inode_nlink(eb, item);
1046         rec->isize = btrfs_inode_size(eb, item);
1047         rec->nbytes = btrfs_inode_nbytes(eb, item);
1048         rec->imode = btrfs_inode_mode(eb, item);
1049         if (btrfs_inode_flags(eb, item) & BTRFS_INODE_NODATASUM)
1050                 rec->nodatasum = 1;
1051         rec->found_inode_item = 1;
1052         if (rec->nlink == 0)
1053                 rec->errors |= I_ERR_NO_ORPHAN_ITEM;
1054         maybe_free_inode_rec(&active_node->inode_cache, rec);
1055         return 0;
1056 }
1057
1058 static struct inode_backref *get_inode_backref(struct inode_record *rec,
1059                                                 const char *name,
1060                                                 int namelen, u64 dir)
1061 {
1062         struct inode_backref *backref;
1063
1064         list_for_each_entry(backref, &rec->backrefs, list) {
1065                 if (rec->ino == BTRFS_MULTIPLE_OBJECTIDS)
1066                         break;
1067                 if (backref->dir != dir || backref->namelen != namelen)
1068                         continue;
1069                 if (memcmp(name, backref->name, namelen))
1070                         continue;
1071                 return backref;
1072         }
1073
1074         backref = malloc(sizeof(*backref) + namelen + 1);
1075         if (!backref)
1076                 return NULL;
1077         memset(backref, 0, sizeof(*backref));
1078         backref->dir = dir;
1079         backref->namelen = namelen;
1080         memcpy(backref->name, name, namelen);
1081         backref->name[namelen] = '\0';
1082         list_add_tail(&backref->list, &rec->backrefs);
1083         return backref;
1084 }
1085
1086 static int add_inode_backref(struct cache_tree *inode_cache,
1087                              u64 ino, u64 dir, u64 index,
1088                              const char *name, int namelen,
1089                              u8 filetype, u8 itemtype, int errors)
1090 {
1091         struct inode_record *rec;
1092         struct inode_backref *backref;
1093
1094         rec = get_inode_rec(inode_cache, ino, 1);
1095         BUG_ON(IS_ERR(rec));
1096         backref = get_inode_backref(rec, name, namelen, dir);
1097         BUG_ON(!backref);
1098         if (errors)
1099                 backref->errors |= errors;
1100         if (itemtype == BTRFS_DIR_INDEX_KEY) {
1101                 if (backref->found_dir_index)
1102                         backref->errors |= REF_ERR_DUP_DIR_INDEX;
1103                 if (backref->found_inode_ref && backref->index != index)
1104                         backref->errors |= REF_ERR_INDEX_UNMATCH;
1105                 if (backref->found_dir_item && backref->filetype != filetype)
1106                         backref->errors |= REF_ERR_FILETYPE_UNMATCH;
1107
1108                 backref->index = index;
1109                 backref->filetype = filetype;
1110                 backref->found_dir_index = 1;
1111         } else if (itemtype == BTRFS_DIR_ITEM_KEY) {
1112                 rec->found_link++;
1113                 if (backref->found_dir_item)
1114                         backref->errors |= REF_ERR_DUP_DIR_ITEM;
1115                 if (backref->found_dir_index && backref->filetype != filetype)
1116                         backref->errors |= REF_ERR_FILETYPE_UNMATCH;
1117
1118                 backref->filetype = filetype;
1119                 backref->found_dir_item = 1;
1120         } else if ((itemtype == BTRFS_INODE_REF_KEY) ||
1121                    (itemtype == BTRFS_INODE_EXTREF_KEY)) {
1122                 if (backref->found_inode_ref)
1123                         backref->errors |= REF_ERR_DUP_INODE_REF;
1124                 if (backref->found_dir_index && backref->index != index)
1125                         backref->errors |= REF_ERR_INDEX_UNMATCH;
1126                 else
1127                         backref->index = index;
1128
1129                 backref->ref_type = itemtype;
1130                 backref->found_inode_ref = 1;
1131         } else {
1132                 BUG_ON(1);
1133         }
1134
1135         maybe_free_inode_rec(inode_cache, rec);
1136         return 0;
1137 }
1138
1139 static int merge_inode_recs(struct inode_record *src, struct inode_record *dst,
1140                             struct cache_tree *dst_cache)
1141 {
1142         struct inode_backref *backref;
1143         u32 dir_count = 0;
1144         int ret = 0;
1145
1146         dst->merging = 1;
1147         list_for_each_entry(backref, &src->backrefs, list) {
1148                 if (backref->found_dir_index) {
1149                         add_inode_backref(dst_cache, dst->ino, backref->dir,
1150                                         backref->index, backref->name,
1151                                         backref->namelen, backref->filetype,
1152                                         BTRFS_DIR_INDEX_KEY, backref->errors);
1153                 }
1154                 if (backref->found_dir_item) {
1155                         dir_count++;
1156                         add_inode_backref(dst_cache, dst->ino,
1157                                         backref->dir, 0, backref->name,
1158                                         backref->namelen, backref->filetype,
1159                                         BTRFS_DIR_ITEM_KEY, backref->errors);
1160                 }
1161                 if (backref->found_inode_ref) {
1162                         add_inode_backref(dst_cache, dst->ino,
1163                                         backref->dir, backref->index,
1164                                         backref->name, backref->namelen, 0,
1165                                         backref->ref_type, backref->errors);
1166                 }
1167         }
1168
1169         if (src->found_dir_item)
1170                 dst->found_dir_item = 1;
1171         if (src->found_file_extent)
1172                 dst->found_file_extent = 1;
1173         if (src->found_csum_item)
1174                 dst->found_csum_item = 1;
1175         if (src->some_csum_missing)
1176                 dst->some_csum_missing = 1;
1177         if (first_extent_gap(&dst->holes) > first_extent_gap(&src->holes)) {
1178                 ret = copy_file_extent_holes(&dst->holes, &src->holes);
1179                 if (ret < 0)
1180                         return ret;
1181         }
1182
1183         BUG_ON(src->found_link < dir_count);
1184         dst->found_link += src->found_link - dir_count;
1185         dst->found_size += src->found_size;
1186         if (src->extent_start != (u64)-1) {
1187                 if (dst->extent_start == (u64)-1) {
1188                         dst->extent_start = src->extent_start;
1189                         dst->extent_end = src->extent_end;
1190                 } else {
1191                         if (dst->extent_end > src->extent_start)
1192                                 dst->errors |= I_ERR_FILE_EXTENT_OVERLAP;
1193                         else if (dst->extent_end < src->extent_start) {
1194                                 ret = add_file_extent_hole(&dst->holes,
1195                                         dst->extent_end,
1196                                         src->extent_start - dst->extent_end);
1197                         }
1198                         if (dst->extent_end < src->extent_end)
1199                                 dst->extent_end = src->extent_end;
1200                 }
1201         }
1202
1203         dst->errors |= src->errors;
1204         if (src->found_inode_item) {
1205                 if (!dst->found_inode_item) {
1206                         dst->nlink = src->nlink;
1207                         dst->isize = src->isize;
1208                         dst->nbytes = src->nbytes;
1209                         dst->imode = src->imode;
1210                         dst->nodatasum = src->nodatasum;
1211                         dst->found_inode_item = 1;
1212                 } else {
1213                         dst->errors |= I_ERR_DUP_INODE_ITEM;
1214                 }
1215         }
1216         dst->merging = 0;
1217
1218         return 0;
1219 }
1220
1221 static int splice_shared_node(struct shared_node *src_node,
1222                               struct shared_node *dst_node)
1223 {
1224         struct cache_extent *cache;
1225         struct ptr_node *node, *ins;
1226         struct cache_tree *src, *dst;
1227         struct inode_record *rec, *conflict;
1228         u64 current_ino = 0;
1229         int splice = 0;
1230         int ret;
1231
1232         if (--src_node->refs == 0)
1233                 splice = 1;
1234         if (src_node->current)
1235                 current_ino = src_node->current->ino;
1236
1237         src = &src_node->root_cache;
1238         dst = &dst_node->root_cache;
1239 again:
1240         cache = search_cache_extent(src, 0);
1241         while (cache) {
1242                 node = container_of(cache, struct ptr_node, cache);
1243                 rec = node->data;
1244                 cache = next_cache_extent(cache);
1245
1246                 if (splice) {
1247                         remove_cache_extent(src, &node->cache);
1248                         ins = node;
1249                 } else {
1250                         ins = malloc(sizeof(*ins));
1251                         BUG_ON(!ins);
1252                         ins->cache.start = node->cache.start;
1253                         ins->cache.size = node->cache.size;
1254                         ins->data = rec;
1255                         rec->refs++;
1256                 }
1257                 ret = insert_cache_extent(dst, &ins->cache);
1258                 if (ret == -EEXIST) {
1259                         conflict = get_inode_rec(dst, rec->ino, 1);
1260                         BUG_ON(IS_ERR(conflict));
1261                         merge_inode_recs(rec, conflict, dst);
1262                         if (rec->checked) {
1263                                 conflict->checked = 1;
1264                                 if (dst_node->current == conflict)
1265                                         dst_node->current = NULL;
1266                         }
1267                         maybe_free_inode_rec(dst, conflict);
1268                         free_inode_rec(rec);
1269                         free(ins);
1270                 } else {
1271                         BUG_ON(ret);
1272                 }
1273         }
1274
1275         if (src == &src_node->root_cache) {
1276                 src = &src_node->inode_cache;
1277                 dst = &dst_node->inode_cache;
1278                 goto again;
1279         }
1280
1281         if (current_ino > 0 && (!dst_node->current ||
1282             current_ino > dst_node->current->ino)) {
1283                 if (dst_node->current) {
1284                         dst_node->current->checked = 1;
1285                         maybe_free_inode_rec(dst, dst_node->current);
1286                 }
1287                 dst_node->current = get_inode_rec(dst, current_ino, 1);
1288                 BUG_ON(IS_ERR(dst_node->current));
1289         }
1290         return 0;
1291 }
1292
1293 static void free_inode_ptr(struct cache_extent *cache)
1294 {
1295         struct ptr_node *node;
1296         struct inode_record *rec;
1297
1298         node = container_of(cache, struct ptr_node, cache);
1299         rec = node->data;
1300         free_inode_rec(rec);
1301         free(node);
1302 }
1303
1304 FREE_EXTENT_CACHE_BASED_TREE(inode_recs, free_inode_ptr);
1305
1306 static struct shared_node *find_shared_node(struct cache_tree *shared,
1307                                             u64 bytenr)
1308 {
1309         struct cache_extent *cache;
1310         struct shared_node *node;
1311
1312         cache = lookup_cache_extent(shared, bytenr, 1);
1313         if (cache) {
1314                 node = container_of(cache, struct shared_node, cache);
1315                 return node;
1316         }
1317         return NULL;
1318 }
1319
1320 static int add_shared_node(struct cache_tree *shared, u64 bytenr, u32 refs)
1321 {
1322         int ret;
1323         struct shared_node *node;
1324
1325         node = calloc(1, sizeof(*node));
1326         if (!node)
1327                 return -ENOMEM;
1328         node->cache.start = bytenr;
1329         node->cache.size = 1;
1330         cache_tree_init(&node->root_cache);
1331         cache_tree_init(&node->inode_cache);
1332         node->refs = refs;
1333
1334         ret = insert_cache_extent(shared, &node->cache);
1335
1336         return ret;
1337 }
1338
1339 static int enter_shared_node(struct btrfs_root *root, u64 bytenr, u32 refs,
1340                              struct walk_control *wc, int level)
1341 {
1342         struct shared_node *node;
1343         struct shared_node *dest;
1344         int ret;
1345
1346         if (level == wc->active_node)
1347                 return 0;
1348
1349         BUG_ON(wc->active_node <= level);
1350         node = find_shared_node(&wc->shared, bytenr);
1351         if (!node) {
1352                 ret = add_shared_node(&wc->shared, bytenr, refs);
1353                 BUG_ON(ret);
1354                 node = find_shared_node(&wc->shared, bytenr);
1355                 wc->nodes[level] = node;
1356                 wc->active_node = level;
1357                 return 0;
1358         }
1359
1360         if (wc->root_level == wc->active_node &&
1361             btrfs_root_refs(&root->root_item) == 0) {
1362                 if (--node->refs == 0) {
1363                         free_inode_recs_tree(&node->root_cache);
1364                         free_inode_recs_tree(&node->inode_cache);
1365                         remove_cache_extent(&wc->shared, &node->cache);
1366                         free(node);
1367                 }
1368                 return 1;
1369         }
1370
1371         dest = wc->nodes[wc->active_node];
1372         splice_shared_node(node, dest);
1373         if (node->refs == 0) {
1374                 remove_cache_extent(&wc->shared, &node->cache);
1375                 free(node);
1376         }
1377         return 1;
1378 }
1379
1380 static int leave_shared_node(struct btrfs_root *root,
1381                              struct walk_control *wc, int level)
1382 {
1383         struct shared_node *node;
1384         struct shared_node *dest;
1385         int i;
1386
1387         if (level == wc->root_level)
1388                 return 0;
1389
1390         for (i = level + 1; i < BTRFS_MAX_LEVEL; i++) {
1391                 if (wc->nodes[i])
1392                         break;
1393         }
1394         BUG_ON(i >= BTRFS_MAX_LEVEL);
1395
1396         node = wc->nodes[wc->active_node];
1397         wc->nodes[wc->active_node] = NULL;
1398         wc->active_node = i;
1399
1400         dest = wc->nodes[wc->active_node];
1401         if (wc->active_node < wc->root_level ||
1402             btrfs_root_refs(&root->root_item) > 0) {
1403                 BUG_ON(node->refs <= 1);
1404                 splice_shared_node(node, dest);
1405         } else {
1406                 BUG_ON(node->refs < 2);
1407                 node->refs--;
1408         }
1409         return 0;
1410 }
1411
1412 /*
1413  * Returns:
1414  * < 0 - on error
1415  * 1   - if the root with id child_root_id is a child of root parent_root_id
1416  * 0   - if the root child_root_id isn't a child of the root parent_root_id but
1417  *       has other root(s) as parent(s)
1418  * 2   - if the root child_root_id doesn't have any parent roots
1419  */
1420 static int is_child_root(struct btrfs_root *root, u64 parent_root_id,
1421                          u64 child_root_id)
1422 {
1423         struct btrfs_path path;
1424         struct btrfs_key key;
1425         struct extent_buffer *leaf;
1426         int has_parent = 0;
1427         int ret;
1428
1429         btrfs_init_path(&path);
1430
1431         key.objectid = parent_root_id;
1432         key.type = BTRFS_ROOT_REF_KEY;
1433         key.offset = child_root_id;
1434         ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, &path,
1435                                 0, 0);
1436         if (ret < 0)
1437                 return ret;
1438         btrfs_release_path(&path);
1439         if (!ret)
1440                 return 1;
1441
1442         key.objectid = child_root_id;
1443         key.type = BTRFS_ROOT_BACKREF_KEY;
1444         key.offset = 0;
1445         ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, &path,
1446                                 0, 0);
1447         if (ret < 0)
1448                 goto out;
1449
1450         while (1) {
1451                 leaf = path.nodes[0];
1452                 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
1453                         ret = btrfs_next_leaf(root->fs_info->tree_root, &path);
1454                         if (ret)
1455                                 break;
1456                         leaf = path.nodes[0];
1457                 }
1458
1459                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
1460                 if (key.objectid != child_root_id ||
1461                     key.type != BTRFS_ROOT_BACKREF_KEY)
1462                         break;
1463
1464                 has_parent = 1;
1465
1466                 if (key.offset == parent_root_id) {
1467                         btrfs_release_path(&path);
1468                         return 1;
1469                 }
1470
1471                 path.slots[0]++;
1472         }
1473 out:
1474         btrfs_release_path(&path);
1475         if (ret < 0)
1476                 return ret;
1477         return has_parent ? 0 : 2;
1478 }
1479
1480 static int process_dir_item(struct btrfs_root *root,
1481                             struct extent_buffer *eb,
1482                             int slot, struct btrfs_key *key,
1483                             struct shared_node *active_node)
1484 {
1485         u32 total;
1486         u32 cur = 0;
1487         u32 len;
1488         u32 name_len;
1489         u32 data_len;
1490         int error;
1491         int nritems = 0;
1492         u8 filetype;
1493         struct btrfs_dir_item *di;
1494         struct inode_record *rec;
1495         struct cache_tree *root_cache;
1496         struct cache_tree *inode_cache;
1497         struct btrfs_key location;
1498         char namebuf[BTRFS_NAME_LEN];
1499
1500         root_cache = &active_node->root_cache;
1501         inode_cache = &active_node->inode_cache;
1502         rec = active_node->current;
1503         rec->found_dir_item = 1;
1504
1505         di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
1506         total = btrfs_item_size_nr(eb, slot);
1507         while (cur < total) {
1508                 nritems++;
1509                 btrfs_dir_item_key_to_cpu(eb, di, &location);
1510                 name_len = btrfs_dir_name_len(eb, di);
1511                 data_len = btrfs_dir_data_len(eb, di);
1512                 filetype = btrfs_dir_type(eb, di);
1513
1514                 rec->found_size += name_len;
1515                 if (name_len <= BTRFS_NAME_LEN) {
1516                         len = name_len;
1517                         error = 0;
1518                 } else {
1519                         len = BTRFS_NAME_LEN;
1520                         error = REF_ERR_NAME_TOO_LONG;
1521                 }
1522                 read_extent_buffer(eb, namebuf, (unsigned long)(di + 1), len);
1523
1524                 if (location.type == BTRFS_INODE_ITEM_KEY) {
1525                         add_inode_backref(inode_cache, location.objectid,
1526                                           key->objectid, key->offset, namebuf,
1527                                           len, filetype, key->type, error);
1528                 } else if (location.type == BTRFS_ROOT_ITEM_KEY) {
1529                         add_inode_backref(root_cache, location.objectid,
1530                                           key->objectid, key->offset,
1531                                           namebuf, len, filetype,
1532                                           key->type, error);
1533                 } else {
1534                         fprintf(stderr, "invalid location in dir item %u\n",
1535                                 location.type);
1536                         add_inode_backref(inode_cache, BTRFS_MULTIPLE_OBJECTIDS,
1537                                           key->objectid, key->offset, namebuf,
1538                                           len, filetype, key->type, error);
1539                 }
1540
1541                 len = sizeof(*di) + name_len + data_len;
1542                 di = (struct btrfs_dir_item *)((char *)di + len);
1543                 cur += len;
1544         }
1545         if (key->type == BTRFS_DIR_INDEX_KEY && nritems > 1)
1546                 rec->errors |= I_ERR_DUP_DIR_INDEX;
1547
1548         return 0;
1549 }
1550
1551 static int process_inode_ref(struct extent_buffer *eb,
1552                              int slot, struct btrfs_key *key,
1553                              struct shared_node *active_node)
1554 {
1555         u32 total;
1556         u32 cur = 0;
1557         u32 len;
1558         u32 name_len;
1559         u64 index;
1560         int error;
1561         struct cache_tree *inode_cache;
1562         struct btrfs_inode_ref *ref;
1563         char namebuf[BTRFS_NAME_LEN];
1564
1565         inode_cache = &active_node->inode_cache;
1566
1567         ref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1568         total = btrfs_item_size_nr(eb, slot);
1569         while (cur < total) {
1570                 name_len = btrfs_inode_ref_name_len(eb, ref);
1571                 index = btrfs_inode_ref_index(eb, ref);
1572                 if (name_len <= BTRFS_NAME_LEN) {
1573                         len = name_len;
1574                         error = 0;
1575                 } else {
1576                         len = BTRFS_NAME_LEN;
1577                         error = REF_ERR_NAME_TOO_LONG;
1578                 }
1579                 read_extent_buffer(eb, namebuf, (unsigned long)(ref + 1), len);
1580                 add_inode_backref(inode_cache, key->objectid, key->offset,
1581                                   index, namebuf, len, 0, key->type, error);
1582
1583                 len = sizeof(*ref) + name_len;
1584                 ref = (struct btrfs_inode_ref *)((char *)ref + len);
1585                 cur += len;
1586         }
1587         return 0;
1588 }
1589
1590 static int process_inode_extref(struct extent_buffer *eb,
1591                                 int slot, struct btrfs_key *key,
1592                                 struct shared_node *active_node)
1593 {
1594         u32 total;
1595         u32 cur = 0;
1596         u32 len;
1597         u32 name_len;
1598         u64 index;
1599         u64 parent;
1600         int error;
1601         struct cache_tree *inode_cache;
1602         struct btrfs_inode_extref *extref;
1603         char namebuf[BTRFS_NAME_LEN];
1604
1605         inode_cache = &active_node->inode_cache;
1606
1607         extref = btrfs_item_ptr(eb, slot, struct btrfs_inode_extref);
1608         total = btrfs_item_size_nr(eb, slot);
1609         while (cur < total) {
1610                 name_len = btrfs_inode_extref_name_len(eb, extref);
1611                 index = btrfs_inode_extref_index(eb, extref);
1612                 parent = btrfs_inode_extref_parent(eb, extref);
1613                 if (name_len <= BTRFS_NAME_LEN) {
1614                         len = name_len;
1615                         error = 0;
1616                 } else {
1617                         len = BTRFS_NAME_LEN;
1618                         error = REF_ERR_NAME_TOO_LONG;
1619                 }
1620                 read_extent_buffer(eb, namebuf,
1621                                    (unsigned long)(extref + 1), len);
1622                 add_inode_backref(inode_cache, key->objectid, parent,
1623                                   index, namebuf, len, 0, key->type, error);
1624
1625                 len = sizeof(*extref) + name_len;
1626                 extref = (struct btrfs_inode_extref *)((char *)extref + len);
1627                 cur += len;
1628         }
1629         return 0;
1630
1631 }
1632
1633 static int count_csum_range(struct btrfs_root *root, u64 start,
1634                             u64 len, u64 *found)
1635 {
1636         struct btrfs_key key;
1637         struct btrfs_path path;
1638         struct extent_buffer *leaf;
1639         int ret;
1640         size_t size;
1641         *found = 0;
1642         u64 csum_end;
1643         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
1644
1645         btrfs_init_path(&path);
1646
1647         key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
1648         key.offset = start;
1649         key.type = BTRFS_EXTENT_CSUM_KEY;
1650
1651         ret = btrfs_search_slot(NULL, root->fs_info->csum_root,
1652                                 &key, &path, 0, 0);
1653         if (ret < 0)
1654                 goto out;
1655         if (ret > 0 && path.slots[0] > 0) {
1656                 leaf = path.nodes[0];
1657                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0] - 1);
1658                 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
1659                     key.type == BTRFS_EXTENT_CSUM_KEY)
1660                         path.slots[0]--;
1661         }
1662
1663         while (len > 0) {
1664                 leaf = path.nodes[0];
1665                 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
1666                         ret = btrfs_next_leaf(root->fs_info->csum_root, &path);
1667                         if (ret > 0)
1668                                 break;
1669                         else if (ret < 0)
1670                                 goto out;
1671                         leaf = path.nodes[0];
1672                 }
1673
1674                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
1675                 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1676                     key.type != BTRFS_EXTENT_CSUM_KEY)
1677                         break;
1678
1679                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
1680                 if (key.offset >= start + len)
1681                         break;
1682
1683                 if (key.offset > start)
1684                         start = key.offset;
1685
1686                 size = btrfs_item_size_nr(leaf, path.slots[0]);
1687                 csum_end = key.offset + (size / csum_size) * root->sectorsize;
1688                 if (csum_end > start) {
1689                         size = min(csum_end - start, len);
1690                         len -= size;
1691                         start += size;
1692                         *found += size;
1693                 }
1694
1695                 path.slots[0]++;
1696         }
1697 out:
1698         btrfs_release_path(&path);
1699         if (ret < 0)
1700                 return ret;
1701         return 0;
1702 }
1703
1704 static int process_file_extent(struct btrfs_root *root,
1705                                 struct extent_buffer *eb,
1706                                 int slot, struct btrfs_key *key,
1707                                 struct shared_node *active_node)
1708 {
1709         struct inode_record *rec;
1710         struct btrfs_file_extent_item *fi;
1711         u64 num_bytes = 0;
1712         u64 disk_bytenr = 0;
1713         u64 extent_offset = 0;
1714         u64 mask = root->sectorsize - 1;
1715         int extent_type;
1716         int ret;
1717
1718         rec = active_node->current;
1719         BUG_ON(rec->ino != key->objectid || rec->refs > 1);
1720         rec->found_file_extent = 1;
1721
1722         if (rec->extent_start == (u64)-1) {
1723                 rec->extent_start = key->offset;
1724                 rec->extent_end = key->offset;
1725         }
1726
1727         if (rec->extent_end > key->offset)
1728                 rec->errors |= I_ERR_FILE_EXTENT_OVERLAP;
1729         else if (rec->extent_end < key->offset) {
1730                 ret = add_file_extent_hole(&rec->holes, rec->extent_end,
1731                                            key->offset - rec->extent_end);
1732                 if (ret < 0)
1733                         return ret;
1734         }
1735
1736         fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
1737         extent_type = btrfs_file_extent_type(eb, fi);
1738
1739         if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1740                 num_bytes = btrfs_file_extent_inline_len(eb, slot, fi);
1741                 if (num_bytes == 0)
1742                         rec->errors |= I_ERR_BAD_FILE_EXTENT;
1743                 rec->found_size += num_bytes;
1744                 num_bytes = (num_bytes + mask) & ~mask;
1745         } else if (extent_type == BTRFS_FILE_EXTENT_REG ||
1746                    extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1747                 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
1748                 disk_bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1749                 extent_offset = btrfs_file_extent_offset(eb, fi);
1750                 if (num_bytes == 0 || (num_bytes & mask))
1751                         rec->errors |= I_ERR_BAD_FILE_EXTENT;
1752                 if (num_bytes + extent_offset >
1753                     btrfs_file_extent_ram_bytes(eb, fi))
1754                         rec->errors |= I_ERR_BAD_FILE_EXTENT;
1755                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC &&
1756                     (btrfs_file_extent_compression(eb, fi) ||
1757                      btrfs_file_extent_encryption(eb, fi) ||
1758                      btrfs_file_extent_other_encoding(eb, fi)))
1759                         rec->errors |= I_ERR_BAD_FILE_EXTENT;
1760                 if (disk_bytenr > 0)
1761                         rec->found_size += num_bytes;
1762         } else {
1763                 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1764         }
1765         rec->extent_end = key->offset + num_bytes;
1766
1767         /*
1768          * The data reloc tree will copy full extents into its inode and then
1769          * copy the corresponding csums.  Because the extent it copied could be
1770          * a preallocated extent that hasn't been written to yet there may be no
1771          * csums to copy, ergo we won't have csums for our file extent.  This is
1772          * ok so just don't bother checking csums if the inode belongs to the
1773          * data reloc tree.
1774          */
1775         if (disk_bytenr > 0 &&
1776             btrfs_header_owner(eb) != BTRFS_DATA_RELOC_TREE_OBJECTID) {
1777                 u64 found;
1778                 if (btrfs_file_extent_compression(eb, fi))
1779                         num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1780                 else
1781                         disk_bytenr += extent_offset;
1782
1783                 ret = count_csum_range(root, disk_bytenr, num_bytes, &found);
1784                 if (ret < 0)
1785                         return ret;
1786                 if (extent_type == BTRFS_FILE_EXTENT_REG) {
1787                         if (found > 0)
1788                                 rec->found_csum_item = 1;
1789                         if (found < num_bytes)
1790                                 rec->some_csum_missing = 1;
1791                 } else if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1792                         if (found > 0)
1793                                 rec->errors |= I_ERR_ODD_CSUM_ITEM;
1794                 }
1795         }
1796         return 0;
1797 }
1798
1799 static int process_one_leaf(struct btrfs_root *root, struct extent_buffer *eb,
1800                             struct walk_control *wc)
1801 {
1802         struct btrfs_key key;
1803         u32 nritems;
1804         int i;
1805         int ret = 0;
1806         struct cache_tree *inode_cache;
1807         struct shared_node *active_node;
1808
1809         if (wc->root_level == wc->active_node &&
1810             btrfs_root_refs(&root->root_item) == 0)
1811                 return 0;
1812
1813         active_node = wc->nodes[wc->active_node];
1814         inode_cache = &active_node->inode_cache;
1815         nritems = btrfs_header_nritems(eb);
1816         for (i = 0; i < nritems; i++) {
1817                 btrfs_item_key_to_cpu(eb, &key, i);
1818
1819                 if (key.objectid == BTRFS_FREE_SPACE_OBJECTID)
1820                         continue;
1821                 if (key.type == BTRFS_ORPHAN_ITEM_KEY)
1822                         continue;
1823
1824                 if (active_node->current == NULL ||
1825                     active_node->current->ino < key.objectid) {
1826                         if (active_node->current) {
1827                                 active_node->current->checked = 1;
1828                                 maybe_free_inode_rec(inode_cache,
1829                                                      active_node->current);
1830                         }
1831                         active_node->current = get_inode_rec(inode_cache,
1832                                                              key.objectid, 1);
1833                         BUG_ON(IS_ERR(active_node->current));
1834                 }
1835                 switch (key.type) {
1836                 case BTRFS_DIR_ITEM_KEY:
1837                 case BTRFS_DIR_INDEX_KEY:
1838                         ret = process_dir_item(root, eb, i, &key, active_node);
1839                         break;
1840                 case BTRFS_INODE_REF_KEY:
1841                         ret = process_inode_ref(eb, i, &key, active_node);
1842                         break;
1843                 case BTRFS_INODE_EXTREF_KEY:
1844                         ret = process_inode_extref(eb, i, &key, active_node);
1845                         break;
1846                 case BTRFS_INODE_ITEM_KEY:
1847                         ret = process_inode_item(eb, i, &key, active_node);
1848                         break;
1849                 case BTRFS_EXTENT_DATA_KEY:
1850                         ret = process_file_extent(root, eb, i, &key,
1851                                                   active_node);
1852                         break;
1853                 default:
1854                         break;
1855                 };
1856         }
1857         return ret;
1858 }
1859
1860 struct node_refs {
1861         u64 bytenr[BTRFS_MAX_LEVEL];
1862         u64 refs[BTRFS_MAX_LEVEL];
1863         int need_check[BTRFS_MAX_LEVEL];
1864 };
1865
1866 static int update_nodes_refs(struct btrfs_root *root, u64 bytenr,
1867                              struct node_refs *nrefs, u64 level);
1868 static int check_inode_item(struct btrfs_root *root, struct btrfs_path *path,
1869                             unsigned int ext_ref);
1870
1871 static int process_one_leaf_v2(struct btrfs_root *root, struct btrfs_path *path,
1872                                struct node_refs *nrefs, int *level, int ext_ref)
1873 {
1874         struct extent_buffer *cur = path->nodes[0];
1875         struct btrfs_key key;
1876         u64 cur_bytenr;
1877         u32 nritems;
1878         u64 first_ino = 0;
1879         int root_level = btrfs_header_level(root->node);
1880         int i;
1881         int ret = 0; /* Final return value */
1882         int err = 0; /* Positive error bitmap */
1883
1884         cur_bytenr = cur->start;
1885
1886         /* skip to first inode item or the first inode number change */
1887         nritems = btrfs_header_nritems(cur);
1888         for (i = 0; i < nritems; i++) {
1889                 btrfs_item_key_to_cpu(cur, &key, i);
1890                 if (i == 0)
1891                         first_ino = key.objectid;
1892                 if (key.type == BTRFS_INODE_ITEM_KEY ||
1893                     (first_ino && first_ino != key.objectid))
1894                         break;
1895         }
1896         if (i == nritems) {
1897                 path->slots[0] = nritems;
1898                 return 0;
1899         }
1900         path->slots[0] = i;
1901
1902 again:
1903         err |= check_inode_item(root, path, ext_ref);
1904
1905         if (err & LAST_ITEM)
1906                 goto out;
1907
1908         /* still have inode items in thie leaf */
1909         if (cur->start == cur_bytenr)
1910                 goto again;
1911
1912         /*
1913          * we have switched to another leaf, above nodes may
1914          * have changed, here walk down the path, if a node
1915          * or leaf is shared, check whether we can skip this
1916          * node or leaf.
1917          */
1918         for (i = root_level; i >= 0; i--) {
1919                 if (path->nodes[i]->start == nrefs->bytenr[i])
1920                         continue;
1921
1922                 ret = update_nodes_refs(root,
1923                                 path->nodes[i]->start,
1924                                 nrefs, i);
1925                 if (ret)
1926                         goto out;
1927
1928                 if (!nrefs->need_check[i]) {
1929                         *level += 1;
1930                         break;
1931                 }
1932         }
1933
1934         for (i = 0; i < *level; i++) {
1935                 free_extent_buffer(path->nodes[i]);
1936                 path->nodes[i] = NULL;
1937         }
1938 out:
1939         err &= ~LAST_ITEM;
1940         /*
1941          * Convert any error bitmap to -EIO, as we should avoid
1942          * mixing positive and negative return value to represent
1943          * error
1944          */
1945         if (err && !ret)
1946                 ret = -EIO;
1947         return ret;
1948 }
1949
1950 static void reada_walk_down(struct btrfs_root *root,
1951                             struct extent_buffer *node, int slot)
1952 {
1953         u64 bytenr;
1954         u64 ptr_gen;
1955         u32 nritems;
1956         u32 blocksize;
1957         int i;
1958         int level;
1959
1960         level = btrfs_header_level(node);
1961         if (level != 1)
1962                 return;
1963
1964         nritems = btrfs_header_nritems(node);
1965         blocksize = root->nodesize;
1966         for (i = slot; i < nritems; i++) {
1967                 bytenr = btrfs_node_blockptr(node, i);
1968                 ptr_gen = btrfs_node_ptr_generation(node, i);
1969                 readahead_tree_block(root, bytenr, blocksize, ptr_gen);
1970         }
1971 }
1972
1973 /*
1974  * Check the child node/leaf by the following condition:
1975  * 1. the first item key of the node/leaf should be the same with the one
1976  *    in parent.
1977  * 2. block in parent node should match the child node/leaf.
1978  * 3. generation of parent node and child's header should be consistent.
1979  *
1980  * Or the child node/leaf pointed by the key in parent is not valid.
1981  *
1982  * We hope to check leaf owner too, but since subvol may share leaves,
1983  * which makes leaf owner check not so strong, key check should be
1984  * sufficient enough for that case.
1985  */
1986 static int check_child_node(struct btrfs_root *root,
1987                             struct extent_buffer *parent, int slot,
1988                             struct extent_buffer *child)
1989 {
1990         struct btrfs_key parent_key;
1991         struct btrfs_key child_key;
1992         int ret = 0;
1993
1994         btrfs_node_key_to_cpu(parent, &parent_key, slot);
1995         if (btrfs_header_level(child) == 0)
1996                 btrfs_item_key_to_cpu(child, &child_key, 0);
1997         else
1998                 btrfs_node_key_to_cpu(child, &child_key, 0);
1999
2000         if (memcmp(&parent_key, &child_key, sizeof(parent_key))) {
2001                 ret = -EINVAL;
2002                 fprintf(stderr,
2003                         "Wrong key of child node/leaf, wanted: (%llu, %u, %llu), have: (%llu, %u, %llu)\n",
2004                         parent_key.objectid, parent_key.type, parent_key.offset,
2005                         child_key.objectid, child_key.type, child_key.offset);
2006         }
2007         if (btrfs_header_bytenr(child) != btrfs_node_blockptr(parent, slot)) {
2008                 ret = -EINVAL;
2009                 fprintf(stderr, "Wrong block of child node/leaf, wanted: %llu, have: %llu\n",
2010                         btrfs_node_blockptr(parent, slot),
2011                         btrfs_header_bytenr(child));
2012         }
2013         if (btrfs_node_ptr_generation(parent, slot) !=
2014             btrfs_header_generation(child)) {
2015                 ret = -EINVAL;
2016                 fprintf(stderr, "Wrong generation of child node/leaf, wanted: %llu, have: %llu\n",
2017                         btrfs_header_generation(child),
2018                         btrfs_node_ptr_generation(parent, slot));
2019         }
2020         return ret;
2021 }
2022
2023 /*
2024  * for a tree node or leaf, if it's shared, indeed we don't need to iterate it
2025  * in every fs or file tree check. Here we find its all root ids, and only check
2026  * it in the fs or file tree which has the smallest root id.
2027  */
2028 static int need_check(struct btrfs_root *root, struct ulist *roots)
2029 {
2030         struct rb_node *node;
2031         struct ulist_node *u;
2032
2033         if (roots->nnodes == 1)
2034                 return 1;
2035
2036         node = rb_first(&roots->root);
2037         u = rb_entry(node, struct ulist_node, rb_node);
2038         /*
2039          * current root id is not smallest, we skip it and let it be checked
2040          * in the fs or file tree who hash the smallest root id.
2041          */
2042         if (root->objectid != u->val)
2043                 return 0;
2044
2045         return 1;
2046 }
2047
2048 /*
2049  * for a tree node or leaf, we record its reference count, so later if we still
2050  * process this node or leaf, don't need to compute its reference count again.
2051  */
2052 static int update_nodes_refs(struct btrfs_root *root, u64 bytenr,
2053                              struct node_refs *nrefs, u64 level)
2054 {
2055         int check, ret;
2056         u64 refs;
2057         struct ulist *roots;
2058
2059         if (nrefs->bytenr[level] != bytenr) {
2060                 ret = btrfs_lookup_extent_info(NULL, root, bytenr,
2061                                        level, 1, &refs, NULL);
2062                 if (ret < 0)
2063                         return ret;
2064
2065                 nrefs->bytenr[level] = bytenr;
2066                 nrefs->refs[level] = refs;
2067                 if (refs > 1) {
2068                         ret = btrfs_find_all_roots(NULL, root->fs_info, bytenr,
2069                                                    0, &roots);
2070                         if (ret)
2071                                 return -EIO;
2072
2073                         check = need_check(root, roots);
2074                         ulist_free(roots);
2075                         nrefs->need_check[level] = check;
2076                 } else {
2077                         nrefs->need_check[level] = 1;
2078                 }
2079         }
2080
2081         return 0;
2082 }
2083
2084 static int walk_down_tree(struct btrfs_root *root, struct btrfs_path *path,
2085                           struct walk_control *wc, int *level,
2086                           struct node_refs *nrefs)
2087 {
2088         enum btrfs_tree_block_status status;
2089         u64 bytenr;
2090         u64 ptr_gen;
2091         struct extent_buffer *next;
2092         struct extent_buffer *cur;
2093         u32 blocksize;
2094         int ret, err = 0;
2095         u64 refs;
2096
2097         WARN_ON(*level < 0);
2098         WARN_ON(*level >= BTRFS_MAX_LEVEL);
2099
2100         if (path->nodes[*level]->start == nrefs->bytenr[*level]) {
2101                 refs = nrefs->refs[*level];
2102                 ret = 0;
2103         } else {
2104                 ret = btrfs_lookup_extent_info(NULL, root,
2105                                        path->nodes[*level]->start,
2106                                        *level, 1, &refs, NULL);
2107                 if (ret < 0) {
2108                         err = ret;
2109                         goto out;
2110                 }
2111                 nrefs->bytenr[*level] = path->nodes[*level]->start;
2112                 nrefs->refs[*level] = refs;
2113         }
2114
2115         if (refs > 1) {
2116                 ret = enter_shared_node(root, path->nodes[*level]->start,
2117                                         refs, wc, *level);
2118                 if (ret > 0) {
2119                         err = ret;
2120                         goto out;
2121                 }
2122         }
2123
2124         while (*level >= 0) {
2125                 WARN_ON(*level < 0);
2126                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2127                 cur = path->nodes[*level];
2128
2129                 if (btrfs_header_level(cur) != *level)
2130                         WARN_ON(1);
2131
2132                 if (path->slots[*level] >= btrfs_header_nritems(cur))
2133                         break;
2134                 if (*level == 0) {
2135                         ret = process_one_leaf(root, cur, wc);
2136                         if (ret < 0)
2137                                 err = ret;
2138                         break;
2139                 }
2140                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2141                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2142                 blocksize = root->nodesize;
2143
2144                 if (bytenr == nrefs->bytenr[*level - 1]) {
2145                         refs = nrefs->refs[*level - 1];
2146                 } else {
2147                         ret = btrfs_lookup_extent_info(NULL, root, bytenr,
2148                                         *level - 1, 1, &refs, NULL);
2149                         if (ret < 0) {
2150                                 refs = 0;
2151                         } else {
2152                                 nrefs->bytenr[*level - 1] = bytenr;
2153                                 nrefs->refs[*level - 1] = refs;
2154                         }
2155                 }
2156
2157                 if (refs > 1) {
2158                         ret = enter_shared_node(root, bytenr, refs,
2159                                                 wc, *level - 1);
2160                         if (ret > 0) {
2161                                 path->slots[*level]++;
2162                                 continue;
2163                         }
2164                 }
2165
2166                 next = btrfs_find_tree_block(root, bytenr, blocksize);
2167                 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
2168                         free_extent_buffer(next);
2169                         reada_walk_down(root, cur, path->slots[*level]);
2170                         next = read_tree_block(root, bytenr, blocksize,
2171                                                ptr_gen);
2172                         if (!extent_buffer_uptodate(next)) {
2173                                 struct btrfs_key node_key;
2174
2175                                 btrfs_node_key_to_cpu(path->nodes[*level],
2176                                                       &node_key,
2177                                                       path->slots[*level]);
2178                                 btrfs_add_corrupt_extent_record(root->fs_info,
2179                                                 &node_key,
2180                                                 path->nodes[*level]->start,
2181                                                 root->nodesize, *level);
2182                                 err = -EIO;
2183                                 goto out;
2184                         }
2185                 }
2186
2187                 ret = check_child_node(root, cur, path->slots[*level], next);
2188                 if (ret) {
2189                         err = ret;
2190                         goto out;
2191                 }
2192
2193                 if (btrfs_is_leaf(next))
2194                         status = btrfs_check_leaf(root, NULL, next);
2195                 else
2196                         status = btrfs_check_node(root, NULL, next);
2197                 if (status != BTRFS_TREE_BLOCK_CLEAN) {
2198                         free_extent_buffer(next);
2199                         err = -EIO;
2200                         goto out;
2201                 }
2202
2203                 *level = *level - 1;
2204                 free_extent_buffer(path->nodes[*level]);
2205                 path->nodes[*level] = next;
2206                 path->slots[*level] = 0;
2207         }
2208 out:
2209         path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2210         return err;
2211 }
2212
2213 static int check_inode_item(struct btrfs_root *root, struct btrfs_path *path,
2214                             unsigned int ext_ref);
2215
2216 static int walk_down_tree_v2(struct btrfs_root *root, struct btrfs_path *path,
2217                              int *level, struct node_refs *nrefs, int ext_ref)
2218 {
2219         enum btrfs_tree_block_status status;
2220         u64 bytenr;
2221         u64 ptr_gen;
2222         struct extent_buffer *next;
2223         struct extent_buffer *cur;
2224         u32 blocksize;
2225         int ret;
2226
2227         WARN_ON(*level < 0);
2228         WARN_ON(*level >= BTRFS_MAX_LEVEL);
2229
2230         ret = update_nodes_refs(root, path->nodes[*level]->start,
2231                                 nrefs, *level);
2232         if (ret < 0)
2233                 return ret;
2234
2235         while (*level >= 0) {
2236                 WARN_ON(*level < 0);
2237                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2238                 cur = path->nodes[*level];
2239
2240                 if (btrfs_header_level(cur) != *level)
2241                         WARN_ON(1);
2242
2243                 if (path->slots[*level] >= btrfs_header_nritems(cur))
2244                         break;
2245                 /* Don't forgot to check leaf/node validation */
2246                 if (*level == 0) {
2247                         ret = btrfs_check_leaf(root, NULL, cur);
2248                         if (ret != BTRFS_TREE_BLOCK_CLEAN) {
2249                                 ret = -EIO;
2250                                 break;
2251                         }
2252                         ret = process_one_leaf_v2(root, path, nrefs,
2253                                                   level, ext_ref);
2254                         break;
2255                 } else {
2256                         ret = btrfs_check_node(root, NULL, cur);
2257                         if (ret != BTRFS_TREE_BLOCK_CLEAN) {
2258                                 ret = -EIO;
2259                                 break;
2260                         }
2261                 }
2262                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2263                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2264                 blocksize = root->nodesize;
2265
2266                 ret = update_nodes_refs(root, bytenr, nrefs, *level - 1);
2267                 if (ret)
2268                         break;
2269                 if (!nrefs->need_check[*level - 1]) {
2270                         path->slots[*level]++;
2271                         continue;
2272                 }
2273
2274                 next = btrfs_find_tree_block(root, bytenr, blocksize);
2275                 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
2276                         free_extent_buffer(next);
2277                         reada_walk_down(root, cur, path->slots[*level]);
2278                         next = read_tree_block(root, bytenr, blocksize,
2279                                                ptr_gen);
2280                         if (!extent_buffer_uptodate(next)) {
2281                                 struct btrfs_key node_key;
2282
2283                                 btrfs_node_key_to_cpu(path->nodes[*level],
2284                                                       &node_key,
2285                                                       path->slots[*level]);
2286                                 btrfs_add_corrupt_extent_record(root->fs_info,
2287                                                 &node_key,
2288                                                 path->nodes[*level]->start,
2289                                                 root->nodesize, *level);
2290                                 ret = -EIO;
2291                                 break;
2292                         }
2293                 }
2294
2295                 ret = check_child_node(root, cur, path->slots[*level], next);
2296                 if (ret < 0)
2297                         break;
2298
2299                 if (btrfs_is_leaf(next))
2300                         status = btrfs_check_leaf(root, NULL, next);
2301                 else
2302                         status = btrfs_check_node(root, NULL, next);
2303                 if (status != BTRFS_TREE_BLOCK_CLEAN) {
2304                         free_extent_buffer(next);
2305                         ret = -EIO;
2306                         break;
2307                 }
2308
2309                 *level = *level - 1;
2310                 free_extent_buffer(path->nodes[*level]);
2311                 path->nodes[*level] = next;
2312                 path->slots[*level] = 0;
2313         }
2314         return ret;
2315 }
2316
2317 static int walk_up_tree(struct btrfs_root *root, struct btrfs_path *path,
2318                         struct walk_control *wc, int *level)
2319 {
2320         int i;
2321         struct extent_buffer *leaf;
2322
2323         for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2324                 leaf = path->nodes[i];
2325                 if (path->slots[i] + 1 < btrfs_header_nritems(leaf)) {
2326                         path->slots[i]++;
2327                         *level = i;
2328                         return 0;
2329                 } else {
2330                         free_extent_buffer(path->nodes[*level]);
2331                         path->nodes[*level] = NULL;
2332                         BUG_ON(*level > wc->active_node);
2333                         if (*level == wc->active_node)
2334                                 leave_shared_node(root, wc, *level);
2335                         *level = i + 1;
2336                 }
2337         }
2338         return 1;
2339 }
2340
2341 static int walk_up_tree_v2(struct btrfs_root *root, struct btrfs_path *path,
2342                            int *level)
2343 {
2344         int i;
2345         struct extent_buffer *leaf;
2346
2347         for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2348                 leaf = path->nodes[i];
2349                 if (path->slots[i] + 1 < btrfs_header_nritems(leaf)) {
2350                         path->slots[i]++;
2351                         *level = i;
2352                         return 0;
2353                 } else {
2354                         free_extent_buffer(path->nodes[*level]);
2355                         path->nodes[*level] = NULL;
2356                         *level = i + 1;
2357                 }
2358         }
2359         return 1;
2360 }
2361
2362 static int check_root_dir(struct inode_record *rec)
2363 {
2364         struct inode_backref *backref;
2365         int ret = -1;
2366
2367         if (!rec->found_inode_item || rec->errors)
2368                 goto out;
2369         if (rec->nlink != 1 || rec->found_link != 0)
2370                 goto out;
2371         if (list_empty(&rec->backrefs))
2372                 goto out;
2373         backref = to_inode_backref(rec->backrefs.next);
2374         if (!backref->found_inode_ref)
2375                 goto out;
2376         if (backref->index != 0 || backref->namelen != 2 ||
2377             memcmp(backref->name, "..", 2))
2378                 goto out;
2379         if (backref->found_dir_index || backref->found_dir_item)
2380                 goto out;
2381         ret = 0;
2382 out:
2383         return ret;
2384 }
2385
2386 static int repair_inode_isize(struct btrfs_trans_handle *trans,
2387                               struct btrfs_root *root, struct btrfs_path *path,
2388                               struct inode_record *rec)
2389 {
2390         struct btrfs_inode_item *ei;
2391         struct btrfs_key key;
2392         int ret;
2393
2394         key.objectid = rec->ino;
2395         key.type = BTRFS_INODE_ITEM_KEY;
2396         key.offset = (u64)-1;
2397
2398         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2399         if (ret < 0)
2400                 goto out;
2401         if (ret) {
2402                 if (!path->slots[0]) {
2403                         ret = -ENOENT;
2404                         goto out;
2405                 }
2406                 path->slots[0]--;
2407                 ret = 0;
2408         }
2409         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2410         if (key.objectid != rec->ino) {
2411                 ret = -ENOENT;
2412                 goto out;
2413         }
2414
2415         ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
2416                             struct btrfs_inode_item);
2417         btrfs_set_inode_size(path->nodes[0], ei, rec->found_size);
2418         btrfs_mark_buffer_dirty(path->nodes[0]);
2419         rec->errors &= ~I_ERR_DIR_ISIZE_WRONG;
2420         printf("reset isize for dir %Lu root %Lu\n", rec->ino,
2421                root->root_key.objectid);
2422 out:
2423         btrfs_release_path(path);
2424         return ret;
2425 }
2426
2427 static int repair_inode_orphan_item(struct btrfs_trans_handle *trans,
2428                                     struct btrfs_root *root,
2429                                     struct btrfs_path *path,
2430                                     struct inode_record *rec)
2431 {
2432         int ret;
2433
2434         ret = btrfs_add_orphan_item(trans, root, path, rec->ino);
2435         btrfs_release_path(path);
2436         if (!ret)
2437                 rec->errors &= ~I_ERR_NO_ORPHAN_ITEM;
2438         return ret;
2439 }
2440
2441 static int repair_inode_nbytes(struct btrfs_trans_handle *trans,
2442                                struct btrfs_root *root,
2443                                struct btrfs_path *path,
2444                                struct inode_record *rec)
2445 {
2446         struct btrfs_inode_item *ei;
2447         struct btrfs_key key;
2448         int ret = 0;
2449
2450         key.objectid = rec->ino;
2451         key.type = BTRFS_INODE_ITEM_KEY;
2452         key.offset = 0;
2453
2454         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2455         if (ret) {
2456                 if (ret > 0)
2457                         ret = -ENOENT;
2458                 goto out;
2459         }
2460
2461         /* Since ret == 0, no need to check anything */
2462         ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
2463                             struct btrfs_inode_item);
2464         btrfs_set_inode_nbytes(path->nodes[0], ei, rec->found_size);
2465         btrfs_mark_buffer_dirty(path->nodes[0]);
2466         rec->errors &= ~I_ERR_FILE_NBYTES_WRONG;
2467         printf("reset nbytes for ino %llu root %llu\n",
2468                rec->ino, root->root_key.objectid);
2469 out:
2470         btrfs_release_path(path);
2471         return ret;
2472 }
2473
2474 static int add_missing_dir_index(struct btrfs_root *root,
2475                                  struct cache_tree *inode_cache,
2476                                  struct inode_record *rec,
2477                                  struct inode_backref *backref)
2478 {
2479         struct btrfs_path path;
2480         struct btrfs_trans_handle *trans;
2481         struct btrfs_dir_item *dir_item;
2482         struct extent_buffer *leaf;
2483         struct btrfs_key key;
2484         struct btrfs_disk_key disk_key;
2485         struct inode_record *dir_rec;
2486         unsigned long name_ptr;
2487         u32 data_size = sizeof(*dir_item) + backref->namelen;
2488         int ret;
2489
2490         trans = btrfs_start_transaction(root, 1);
2491         if (IS_ERR(trans))
2492                 return PTR_ERR(trans);
2493
2494         fprintf(stderr, "repairing missing dir index item for inode %llu\n",
2495                 (unsigned long long)rec->ino);
2496
2497         btrfs_init_path(&path);
2498         key.objectid = backref->dir;
2499         key.type = BTRFS_DIR_INDEX_KEY;
2500         key.offset = backref->index;
2501         ret = btrfs_insert_empty_item(trans, root, &path, &key, data_size);
2502         BUG_ON(ret);
2503
2504         leaf = path.nodes[0];
2505         dir_item = btrfs_item_ptr(leaf, path.slots[0], struct btrfs_dir_item);
2506
2507         disk_key.objectid = cpu_to_le64(rec->ino);
2508         disk_key.type = BTRFS_INODE_ITEM_KEY;
2509         disk_key.offset = 0;
2510
2511         btrfs_set_dir_item_key(leaf, dir_item, &disk_key);
2512         btrfs_set_dir_type(leaf, dir_item, imode_to_type(rec->imode));
2513         btrfs_set_dir_data_len(leaf, dir_item, 0);
2514         btrfs_set_dir_name_len(leaf, dir_item, backref->namelen);
2515         name_ptr = (unsigned long)(dir_item + 1);
2516         write_extent_buffer(leaf, backref->name, name_ptr, backref->namelen);
2517         btrfs_mark_buffer_dirty(leaf);
2518         btrfs_release_path(&path);
2519         btrfs_commit_transaction(trans, root);
2520
2521         backref->found_dir_index = 1;
2522         dir_rec = get_inode_rec(inode_cache, backref->dir, 0);
2523         BUG_ON(IS_ERR(dir_rec));
2524         if (!dir_rec)
2525                 return 0;
2526         dir_rec->found_size += backref->namelen;
2527         if (dir_rec->found_size == dir_rec->isize &&
2528             (dir_rec->errors & I_ERR_DIR_ISIZE_WRONG))
2529                 dir_rec->errors &= ~I_ERR_DIR_ISIZE_WRONG;
2530         if (dir_rec->found_size != dir_rec->isize)
2531                 dir_rec->errors |= I_ERR_DIR_ISIZE_WRONG;
2532
2533         return 0;
2534 }
2535
2536 static int delete_dir_index(struct btrfs_root *root,
2537                             struct cache_tree *inode_cache,
2538                             struct inode_record *rec,
2539                             struct inode_backref *backref)
2540 {
2541         struct btrfs_trans_handle *trans;
2542         struct btrfs_dir_item *di;
2543         struct btrfs_path path;
2544         int ret = 0;
2545
2546         trans = btrfs_start_transaction(root, 1);
2547         if (IS_ERR(trans))
2548                 return PTR_ERR(trans);
2549
2550         fprintf(stderr, "Deleting bad dir index [%llu,%u,%llu] root %llu\n",
2551                 (unsigned long long)backref->dir,
2552                 BTRFS_DIR_INDEX_KEY, (unsigned long long)backref->index,
2553                 (unsigned long long)root->objectid);
2554
2555         btrfs_init_path(&path);
2556         di = btrfs_lookup_dir_index(trans, root, &path, backref->dir,
2557                                     backref->name, backref->namelen,
2558                                     backref->index, -1);
2559         if (IS_ERR(di)) {
2560                 ret = PTR_ERR(di);
2561                 btrfs_release_path(&path);
2562                 btrfs_commit_transaction(trans, root);
2563                 if (ret == -ENOENT)
2564                         return 0;
2565                 return ret;
2566         }
2567
2568         if (!di)
2569                 ret = btrfs_del_item(trans, root, &path);
2570         else
2571                 ret = btrfs_delete_one_dir_name(trans, root, &path, di);
2572         BUG_ON(ret);
2573         btrfs_release_path(&path);
2574         btrfs_commit_transaction(trans, root);
2575         return ret;
2576 }
2577
2578 static int create_inode_item(struct btrfs_root *root,
2579                              struct inode_record *rec,
2580                              struct inode_backref *backref, int root_dir)
2581 {
2582         struct btrfs_trans_handle *trans;
2583         struct btrfs_inode_item inode_item;
2584         time_t now = time(NULL);
2585         int ret;
2586
2587         trans = btrfs_start_transaction(root, 1);
2588         if (IS_ERR(trans)) {
2589                 ret = PTR_ERR(trans);
2590                 return ret;
2591         }
2592
2593         fprintf(stderr, "root %llu inode %llu recreating inode item, this may "
2594                 "be incomplete, please check permissions and content after "
2595                 "the fsck completes.\n", (unsigned long long)root->objectid,
2596                 (unsigned long long)rec->ino);
2597
2598         memset(&inode_item, 0, sizeof(inode_item));
2599         btrfs_set_stack_inode_generation(&inode_item, trans->transid);
2600         if (root_dir)
2601                 btrfs_set_stack_inode_nlink(&inode_item, 1);
2602         else
2603                 btrfs_set_stack_inode_nlink(&inode_item, rec->found_link);
2604         btrfs_set_stack_inode_nbytes(&inode_item, rec->found_size);
2605         if (rec->found_dir_item) {
2606                 if (rec->found_file_extent)
2607                         fprintf(stderr, "root %llu inode %llu has both a dir "
2608                                 "item and extents, unsure if it is a dir or a "
2609                                 "regular file so setting it as a directory\n",
2610                                 (unsigned long long)root->objectid,
2611                                 (unsigned long long)rec->ino);
2612                 btrfs_set_stack_inode_mode(&inode_item, S_IFDIR | 0755);
2613                 btrfs_set_stack_inode_size(&inode_item, rec->found_size);
2614         } else if (!rec->found_dir_item) {
2615                 btrfs_set_stack_inode_size(&inode_item, rec->extent_end);
2616                 btrfs_set_stack_inode_mode(&inode_item, S_IFREG | 0755);
2617         }
2618         btrfs_set_stack_timespec_sec(&inode_item.atime, now);
2619         btrfs_set_stack_timespec_nsec(&inode_item.atime, 0);
2620         btrfs_set_stack_timespec_sec(&inode_item.ctime, now);
2621         btrfs_set_stack_timespec_nsec(&inode_item.ctime, 0);
2622         btrfs_set_stack_timespec_sec(&inode_item.mtime, now);
2623         btrfs_set_stack_timespec_nsec(&inode_item.mtime, 0);
2624         btrfs_set_stack_timespec_sec(&inode_item.otime, 0);
2625         btrfs_set_stack_timespec_nsec(&inode_item.otime, 0);
2626
2627         ret = btrfs_insert_inode(trans, root, rec->ino, &inode_item);
2628         BUG_ON(ret);
2629         btrfs_commit_transaction(trans, root);
2630         return 0;
2631 }
2632
2633 static int repair_inode_backrefs(struct btrfs_root *root,
2634                                  struct inode_record *rec,
2635                                  struct cache_tree *inode_cache,
2636                                  int delete)
2637 {
2638         struct inode_backref *tmp, *backref;
2639         u64 root_dirid = btrfs_root_dirid(&root->root_item);
2640         int ret = 0;
2641         int repaired = 0;
2642
2643         list_for_each_entry_safe(backref, tmp, &rec->backrefs, list) {
2644                 if (!delete && rec->ino == root_dirid) {
2645                         if (!rec->found_inode_item) {
2646                                 ret = create_inode_item(root, rec, backref, 1);
2647                                 if (ret)
2648                                         break;
2649                                 repaired++;
2650                         }
2651                 }
2652
2653                 /* Index 0 for root dir's are special, don't mess with it */
2654                 if (rec->ino == root_dirid && backref->index == 0)
2655                         continue;
2656
2657                 if (delete &&
2658                     ((backref->found_dir_index && !backref->found_inode_ref) ||
2659                      (backref->found_dir_index && backref->found_inode_ref &&
2660                       (backref->errors & REF_ERR_INDEX_UNMATCH)))) {
2661                         ret = delete_dir_index(root, inode_cache, rec, backref);
2662                         if (ret)
2663                                 break;
2664                         repaired++;
2665                         list_del(&backref->list);
2666                         free(backref);
2667                 }
2668
2669                 if (!delete && !backref->found_dir_index &&
2670                     backref->found_dir_item && backref->found_inode_ref) {
2671                         ret = add_missing_dir_index(root, inode_cache, rec,
2672                                                     backref);
2673                         if (ret)
2674                                 break;
2675                         repaired++;
2676                         if (backref->found_dir_item &&
2677                             backref->found_dir_index &&
2678                             backref->found_dir_index) {
2679                                 if (!backref->errors &&
2680                                     backref->found_inode_ref) {
2681                                         list_del(&backref->list);
2682                                         free(backref);
2683                                 }
2684                         }
2685                 }
2686
2687                 if (!delete && (!backref->found_dir_index &&
2688                                 !backref->found_dir_item &&
2689                                 backref->found_inode_ref)) {
2690                         struct btrfs_trans_handle *trans;
2691                         struct btrfs_key location;
2692
2693                         ret = check_dir_conflict(root, backref->name,
2694                                                  backref->namelen,
2695                                                  backref->dir,
2696                                                  backref->index);
2697                         if (ret) {
2698                                 /*
2699                                  * let nlink fixing routine to handle it,
2700                                  * which can do it better.
2701                                  */
2702                                 ret = 0;
2703                                 break;
2704                         }
2705                         location.objectid = rec->ino;
2706                         location.type = BTRFS_INODE_ITEM_KEY;
2707                         location.offset = 0;
2708
2709                         trans = btrfs_start_transaction(root, 1);
2710                         if (IS_ERR(trans)) {
2711                                 ret = PTR_ERR(trans);
2712                                 break;
2713                         }
2714                         fprintf(stderr, "adding missing dir index/item pair "
2715                                 "for inode %llu\n",
2716                                 (unsigned long long)rec->ino);
2717                         ret = btrfs_insert_dir_item(trans, root, backref->name,
2718                                                     backref->namelen,
2719                                                     backref->dir, &location,
2720                                                     imode_to_type(rec->imode),
2721                                                     backref->index);
2722                         BUG_ON(ret);
2723                         btrfs_commit_transaction(trans, root);
2724                         repaired++;
2725                 }
2726
2727                 if (!delete && (backref->found_inode_ref &&
2728                                 backref->found_dir_index &&
2729                                 backref->found_dir_item &&
2730                                 !(backref->errors & REF_ERR_INDEX_UNMATCH) &&
2731                                 !rec->found_inode_item)) {
2732                         ret = create_inode_item(root, rec, backref, 0);
2733                         if (ret)
2734                                 break;
2735                         repaired++;
2736                 }
2737
2738         }
2739         return ret ? ret : repaired;
2740 }
2741
2742 /*
2743  * To determine the file type for nlink/inode_item repair
2744  *
2745  * Return 0 if file type is found and BTRFS_FT_* is stored into type.
2746  * Return -ENOENT if file type is not found.
2747  */
2748 static int find_file_type(struct inode_record *rec, u8 *type)
2749 {
2750         struct inode_backref *backref;
2751
2752         /* For inode item recovered case */
2753         if (rec->found_inode_item) {
2754                 *type = imode_to_type(rec->imode);
2755                 return 0;
2756         }
2757
2758         list_for_each_entry(backref, &rec->backrefs, list) {
2759                 if (backref->found_dir_index || backref->found_dir_item) {
2760                         *type = backref->filetype;
2761                         return 0;
2762                 }
2763         }
2764         return -ENOENT;
2765 }
2766
2767 /*
2768  * To determine the file name for nlink repair
2769  *
2770  * Return 0 if file name is found, set name and namelen.
2771  * Return -ENOENT if file name is not found.
2772  */
2773 static int find_file_name(struct inode_record *rec,
2774                           char *name, int *namelen)
2775 {
2776         struct inode_backref *backref;
2777
2778         list_for_each_entry(backref, &rec->backrefs, list) {
2779                 if (backref->found_dir_index || backref->found_dir_item ||
2780                     backref->found_inode_ref) {
2781                         memcpy(name, backref->name, backref->namelen);
2782                         *namelen = backref->namelen;
2783                         return 0;
2784                 }
2785         }
2786         return -ENOENT;
2787 }
2788
2789 /* Reset the nlink of the inode to the correct one */
2790 static int reset_nlink(struct btrfs_trans_handle *trans,
2791                        struct btrfs_root *root,
2792                        struct btrfs_path *path,
2793                        struct inode_record *rec)
2794 {
2795         struct inode_backref *backref;
2796         struct inode_backref *tmp;
2797         struct btrfs_key key;
2798         struct btrfs_inode_item *inode_item;
2799         int ret = 0;
2800
2801         /* We don't believe this either, reset it and iterate backref */
2802         rec->found_link = 0;
2803
2804         /* Remove all backref including the valid ones */
2805         list_for_each_entry_safe(backref, tmp, &rec->backrefs, list) {
2806                 ret = btrfs_unlink(trans, root, rec->ino, backref->dir,
2807                                    backref->index, backref->name,
2808                                    backref->namelen, 0);
2809                 if (ret < 0)
2810                         goto out;
2811
2812                 /* remove invalid backref, so it won't be added back */
2813                 if (!(backref->found_dir_index &&
2814                       backref->found_dir_item &&
2815                       backref->found_inode_ref)) {
2816                         list_del(&backref->list);
2817                         free(backref);
2818                 } else {
2819                         rec->found_link++;
2820                 }
2821         }
2822
2823         /* Set nlink to 0 */
2824         key.objectid = rec->ino;
2825         key.type = BTRFS_INODE_ITEM_KEY;
2826         key.offset = 0;
2827         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2828         if (ret < 0)
2829                 goto out;
2830         if (ret > 0) {
2831                 ret = -ENOENT;
2832                 goto out;
2833         }
2834         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2835                                     struct btrfs_inode_item);
2836         btrfs_set_inode_nlink(path->nodes[0], inode_item, 0);
2837         btrfs_mark_buffer_dirty(path->nodes[0]);
2838         btrfs_release_path(path);
2839
2840         /*
2841          * Add back valid inode_ref/dir_item/dir_index,
2842          * add_link() will handle the nlink inc, so new nlink must be correct
2843          */
2844         list_for_each_entry(backref, &rec->backrefs, list) {
2845                 ret = btrfs_add_link(trans, root, rec->ino, backref->dir,
2846                                      backref->name, backref->namelen,
2847                                      backref->filetype, &backref->index, 1);
2848                 if (ret < 0)
2849                         goto out;
2850         }
2851 out:
2852         btrfs_release_path(path);
2853         return ret;
2854 }
2855
2856 static int repair_inode_nlinks(struct btrfs_trans_handle *trans,
2857                                struct btrfs_root *root,
2858                                struct btrfs_path *path,
2859                                struct inode_record *rec)
2860 {
2861         char *dir_name = "lost+found";
2862         char namebuf[BTRFS_NAME_LEN] = {0};
2863         u64 lost_found_ino;
2864         u32 mode = 0700;
2865         u8 type = 0;
2866         int namelen = 0;
2867         int name_recovered = 0;
2868         int type_recovered = 0;
2869         int ret = 0;
2870
2871         /*
2872          * Get file name and type first before these invalid inode ref
2873          * are deleted by remove_all_invalid_backref()
2874          */
2875         name_recovered = !find_file_name(rec, namebuf, &namelen);
2876         type_recovered = !find_file_type(rec, &type);
2877
2878         if (!name_recovered) {
2879                 printf("Can't get file name for inode %llu, using '%llu' as fallback\n",
2880                        rec->ino, rec->ino);
2881                 namelen = count_digits(rec->ino);
2882                 sprintf(namebuf, "%llu", rec->ino);
2883                 name_recovered = 1;
2884         }
2885         if (!type_recovered) {
2886                 printf("Can't get file type for inode %llu, using FILE as fallback\n",
2887                        rec->ino);
2888                 type = BTRFS_FT_REG_FILE;
2889                 type_recovered = 1;
2890         }
2891
2892         ret = reset_nlink(trans, root, path, rec);
2893         if (ret < 0) {
2894                 fprintf(stderr,
2895                         "Failed to reset nlink for inode %llu: %s\n",
2896                         rec->ino, strerror(-ret));
2897                 goto out;
2898         }
2899
2900         if (rec->found_link == 0) {
2901                 lost_found_ino = root->highest_inode;
2902                 if (lost_found_ino >= BTRFS_LAST_FREE_OBJECTID) {
2903                         ret = -EOVERFLOW;
2904                         goto out;
2905                 }
2906                 lost_found_ino++;
2907                 ret = btrfs_mkdir(trans, root, dir_name, strlen(dir_name),
2908                                   BTRFS_FIRST_FREE_OBJECTID, &lost_found_ino,
2909                                   mode);
2910                 if (ret < 0) {
2911                         fprintf(stderr, "Failed to create '%s' dir: %s\n",
2912                                 dir_name, strerror(-ret));
2913                         goto out;
2914                 }
2915                 ret = btrfs_add_link(trans, root, rec->ino, lost_found_ino,
2916                                      namebuf, namelen, type, NULL, 1);
2917                 /*
2918                  * Add ".INO" suffix several times to handle case where
2919                  * "FILENAME.INO" is already taken by another file.
2920                  */
2921                 while (ret == -EEXIST) {
2922                         /*
2923                          * Conflicting file name, add ".INO" as suffix * +1 for '.'
2924                          */
2925                         if (namelen + count_digits(rec->ino) + 1 >
2926                             BTRFS_NAME_LEN) {
2927                                 ret = -EFBIG;
2928                                 goto out;
2929                         }
2930                         snprintf(namebuf + namelen, BTRFS_NAME_LEN - namelen,
2931                                  ".%llu", rec->ino);
2932                         namelen += count_digits(rec->ino) + 1;
2933                         ret = btrfs_add_link(trans, root, rec->ino,
2934                                              lost_found_ino, namebuf,
2935                                              namelen, type, NULL, 1);
2936                 }
2937                 if (ret < 0) {
2938                         fprintf(stderr,
2939                                 "Failed to link the inode %llu to %s dir: %s\n",
2940                                 rec->ino, dir_name, strerror(-ret));
2941                         goto out;
2942                 }
2943                 /*
2944                  * Just increase the found_link, don't actually add the
2945                  * backref. This will make things easier and this inode
2946                  * record will be freed after the repair is done.
2947                  * So fsck will not report problem about this inode.
2948                  */
2949                 rec->found_link++;
2950                 printf("Moving file '%.*s' to '%s' dir since it has no valid backref\n",
2951                        namelen, namebuf, dir_name);
2952         }
2953         printf("Fixed the nlink of inode %llu\n", rec->ino);
2954 out:
2955         /*
2956          * Clear the flag anyway, or we will loop forever for the same inode
2957          * as it will not be removed from the bad inode list and the dead loop
2958          * happens.
2959          */
2960         rec->errors &= ~I_ERR_LINK_COUNT_WRONG;
2961         btrfs_release_path(path);
2962         return ret;
2963 }
2964
2965 /*
2966  * Check if there is any normal(reg or prealloc) file extent for given
2967  * ino.
2968  * This is used to determine the file type when neither its dir_index/item or
2969  * inode_item exists.
2970  *
2971  * This will *NOT* report error, if any error happens, just consider it does
2972  * not have any normal file extent.
2973  */
2974 static int find_normal_file_extent(struct btrfs_root *root, u64 ino)
2975 {
2976         struct btrfs_path path;
2977         struct btrfs_key key;
2978         struct btrfs_key found_key;
2979         struct btrfs_file_extent_item *fi;
2980         u8 type;
2981         int ret = 0;
2982
2983         btrfs_init_path(&path);
2984         key.objectid = ino;
2985         key.type = BTRFS_EXTENT_DATA_KEY;
2986         key.offset = 0;
2987
2988         ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
2989         if (ret < 0) {
2990                 ret = 0;
2991                 goto out;
2992         }
2993         if (ret && path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
2994                 ret = btrfs_next_leaf(root, &path);
2995                 if (ret) {
2996                         ret = 0;
2997                         goto out;
2998                 }
2999         }
3000         while (1) {
3001                 btrfs_item_key_to_cpu(path.nodes[0], &found_key,
3002                                       path.slots[0]);
3003                 if (found_key.objectid != ino ||
3004                     found_key.type != BTRFS_EXTENT_DATA_KEY)
3005                         break;
3006                 fi = btrfs_item_ptr(path.nodes[0], path.slots[0],
3007                                     struct btrfs_file_extent_item);
3008                 type = btrfs_file_extent_type(path.nodes[0], fi);
3009                 if (type != BTRFS_FILE_EXTENT_INLINE) {
3010                         ret = 1;
3011                         goto out;
3012                 }
3013         }
3014 out:
3015         btrfs_release_path(&path);
3016         return ret;
3017 }
3018
3019 static u32 btrfs_type_to_imode(u8 type)
3020 {
3021         static u32 imode_by_btrfs_type[] = {
3022                 [BTRFS_FT_REG_FILE]     = S_IFREG,
3023                 [BTRFS_FT_DIR]          = S_IFDIR,
3024                 [BTRFS_FT_CHRDEV]       = S_IFCHR,
3025                 [BTRFS_FT_BLKDEV]       = S_IFBLK,
3026                 [BTRFS_FT_FIFO]         = S_IFIFO,
3027                 [BTRFS_FT_SOCK]         = S_IFSOCK,
3028                 [BTRFS_FT_SYMLINK]      = S_IFLNK,
3029         };
3030
3031         return imode_by_btrfs_type[(type)];
3032 }
3033
3034 static int repair_inode_no_item(struct btrfs_trans_handle *trans,
3035                                 struct btrfs_root *root,
3036                                 struct btrfs_path *path,
3037                                 struct inode_record *rec)
3038 {
3039         u8 filetype;
3040         u32 mode = 0700;
3041         int type_recovered = 0;
3042         int ret = 0;
3043
3044         printf("Trying to rebuild inode:%llu\n", rec->ino);
3045
3046         type_recovered = !find_file_type(rec, &filetype);
3047
3048         /*
3049          * Try to determine inode type if type not found.
3050          *
3051          * For found regular file extent, it must be FILE.
3052          * For found dir_item/index, it must be DIR.
3053          *
3054          * For undetermined one, use FILE as fallback.
3055          *
3056          * TODO:
3057          * 1. If found backref(inode_index/item is already handled) to it,
3058          *    it must be DIR.
3059          *    Need new inode-inode ref structure to allow search for that.
3060          */
3061         if (!type_recovered) {
3062                 if (rec->found_file_extent &&
3063                     find_normal_file_extent(root, rec->ino)) {
3064                         type_recovered = 1;
3065                         filetype = BTRFS_FT_REG_FILE;
3066                 } else if (rec->found_dir_item) {
3067                         type_recovered = 1;
3068                         filetype = BTRFS_FT_DIR;
3069                 } else if (!list_empty(&rec->orphan_extents)) {
3070                         type_recovered = 1;
3071                         filetype = BTRFS_FT_REG_FILE;
3072                 } else{
3073                         printf("Can't determine the filetype for inode %llu, assume it is a normal file\n",
3074                                rec->ino);
3075                         type_recovered = 1;
3076                         filetype = BTRFS_FT_REG_FILE;
3077                 }
3078         }
3079
3080         ret = btrfs_new_inode(trans, root, rec->ino,
3081                               mode | btrfs_type_to_imode(filetype));
3082         if (ret < 0)
3083                 goto out;
3084
3085         /*
3086          * Here inode rebuild is done, we only rebuild the inode item,
3087          * don't repair the nlink(like move to lost+found).
3088          * That is the job of nlink repair.
3089          *
3090          * We just fill the record and return
3091          */
3092         rec->found_dir_item = 1;
3093         rec->imode = mode | btrfs_type_to_imode(filetype);
3094         rec->nlink = 0;
3095         rec->errors &= ~I_ERR_NO_INODE_ITEM;
3096         /* Ensure the inode_nlinks repair function will be called */
3097         rec->errors |= I_ERR_LINK_COUNT_WRONG;
3098 out:
3099         return ret;
3100 }
3101
3102 static int repair_inode_orphan_extent(struct btrfs_trans_handle *trans,
3103                                       struct btrfs_root *root,
3104                                       struct btrfs_path *path,
3105                                       struct inode_record *rec)
3106 {
3107         struct orphan_data_extent *orphan;
3108         struct orphan_data_extent *tmp;
3109         int ret = 0;
3110
3111         list_for_each_entry_safe(orphan, tmp, &rec->orphan_extents, list) {
3112                 /*
3113                  * Check for conflicting file extents
3114                  *
3115                  * Here we don't know whether the extents is compressed or not,
3116                  * so we can only assume it not compressed nor data offset,
3117                  * and use its disk_len as extent length.
3118                  */
3119                 ret = btrfs_get_extent(NULL, root, path, orphan->objectid,
3120                                        orphan->offset, orphan->disk_len, 0);
3121                 btrfs_release_path(path);
3122                 if (ret < 0)
3123                         goto out;
3124                 if (!ret) {
3125                         fprintf(stderr,
3126                                 "orphan extent (%llu, %llu) conflicts, delete the orphan\n",
3127                                 orphan->disk_bytenr, orphan->disk_len);
3128                         ret = btrfs_free_extent(trans,
3129                                         root->fs_info->extent_root,
3130                                         orphan->disk_bytenr, orphan->disk_len,
3131                                         0, root->objectid, orphan->objectid,
3132                                         orphan->offset);
3133                         if (ret < 0)
3134                                 goto out;
3135                 }
3136                 ret = btrfs_insert_file_extent(trans, root, orphan->objectid,
3137                                 orphan->offset, orphan->disk_bytenr,
3138                                 orphan->disk_len, orphan->disk_len);
3139                 if (ret < 0)
3140                         goto out;
3141
3142                 /* Update file size info */
3143                 rec->found_size += orphan->disk_len;
3144                 if (rec->found_size == rec->nbytes)
3145                         rec->errors &= ~I_ERR_FILE_NBYTES_WRONG;
3146
3147                 /* Update the file extent hole info too */
3148                 ret = del_file_extent_hole(&rec->holes, orphan->offset,
3149                                            orphan->disk_len);
3150                 if (ret < 0)
3151                         goto out;
3152                 if (RB_EMPTY_ROOT(&rec->holes))
3153                         rec->errors &= ~I_ERR_FILE_EXTENT_DISCOUNT;
3154
3155                 list_del(&orphan->list);
3156                 free(orphan);
3157         }
3158         rec->errors &= ~I_ERR_FILE_EXTENT_ORPHAN;
3159 out:
3160         return ret;
3161 }
3162
3163 static int repair_inode_discount_extent(struct btrfs_trans_handle *trans,
3164                                         struct btrfs_root *root,
3165                                         struct btrfs_path *path,
3166                                         struct inode_record *rec)
3167 {
3168         struct rb_node *node;
3169         struct file_extent_hole *hole;
3170         int found = 0;
3171         int ret = 0;
3172
3173         node = rb_first(&rec->holes);
3174
3175         while (node) {
3176                 found = 1;
3177                 hole = rb_entry(node, struct file_extent_hole, node);
3178                 ret = btrfs_punch_hole(trans, root, rec->ino,
3179                                        hole->start, hole->len);
3180                 if (ret < 0)
3181                         goto out;
3182                 ret = del_file_extent_hole(&rec->holes, hole->start,
3183                                            hole->len);
3184                 if (ret < 0)
3185                         goto out;
3186                 if (RB_EMPTY_ROOT(&rec->holes))
3187                         rec->errors &= ~I_ERR_FILE_EXTENT_DISCOUNT;
3188                 node = rb_first(&rec->holes);
3189         }
3190         /* special case for a file losing all its file extent */
3191         if (!found) {
3192                 ret = btrfs_punch_hole(trans, root, rec->ino, 0,
3193                                        round_up(rec->isize, root->sectorsize));
3194                 if (ret < 0)
3195                         goto out;
3196         }
3197         printf("Fixed discount file extents for inode: %llu in root: %llu\n",
3198                rec->ino, root->objectid);
3199 out:
3200         return ret;
3201 }
3202
3203 static int try_repair_inode(struct btrfs_root *root, struct inode_record *rec)
3204 {
3205         struct btrfs_trans_handle *trans;
3206         struct btrfs_path path;
3207         int ret = 0;
3208
3209         if (!(rec->errors & (I_ERR_DIR_ISIZE_WRONG |
3210                              I_ERR_NO_ORPHAN_ITEM |
3211                              I_ERR_LINK_COUNT_WRONG |
3212                              I_ERR_NO_INODE_ITEM |
3213                              I_ERR_FILE_EXTENT_ORPHAN |
3214                              I_ERR_FILE_EXTENT_DISCOUNT|
3215                              I_ERR_FILE_NBYTES_WRONG)))
3216                 return rec->errors;
3217
3218         /*
3219          * For nlink repair, it may create a dir and add link, so
3220          * 2 for parent(256)'s dir_index and dir_item
3221          * 2 for lost+found dir's inode_item and inode_ref
3222          * 1 for the new inode_ref of the file
3223          * 2 for lost+found dir's dir_index and dir_item for the file
3224          */
3225         trans = btrfs_start_transaction(root, 7);
3226         if (IS_ERR(trans))
3227                 return PTR_ERR(trans);
3228
3229         btrfs_init_path(&path);
3230         if (rec->errors & I_ERR_NO_INODE_ITEM)
3231                 ret = repair_inode_no_item(trans, root, &path, rec);
3232         if (!ret && rec->errors & I_ERR_FILE_EXTENT_ORPHAN)
3233                 ret = repair_inode_orphan_extent(trans, root, &path, rec);
3234         if (!ret && rec->errors & I_ERR_FILE_EXTENT_DISCOUNT)
3235                 ret = repair_inode_discount_extent(trans, root, &path, rec);
3236         if (!ret && rec->errors & I_ERR_DIR_ISIZE_WRONG)
3237                 ret = repair_inode_isize(trans, root, &path, rec);
3238         if (!ret && rec->errors & I_ERR_NO_ORPHAN_ITEM)
3239                 ret = repair_inode_orphan_item(trans, root, &path, rec);
3240         if (!ret && rec->errors & I_ERR_LINK_COUNT_WRONG)
3241                 ret = repair_inode_nlinks(trans, root, &path, rec);
3242         if (!ret && rec->errors & I_ERR_FILE_NBYTES_WRONG)
3243                 ret = repair_inode_nbytes(trans, root, &path, rec);
3244         btrfs_commit_transaction(trans, root);
3245         btrfs_release_path(&path);
3246         return ret;
3247 }
3248
3249 static int check_inode_recs(struct btrfs_root *root,
3250                             struct cache_tree *inode_cache)
3251 {
3252         struct cache_extent *cache;
3253         struct ptr_node *node;
3254         struct inode_record *rec;
3255         struct inode_backref *backref;
3256         int stage = 0;
3257         int ret = 0;
3258         int err = 0;
3259         u64 error = 0;
3260         u64 root_dirid = btrfs_root_dirid(&root->root_item);
3261
3262         if (btrfs_root_refs(&root->root_item) == 0) {
3263                 if (!cache_tree_empty(inode_cache))
3264                         fprintf(stderr, "warning line %d\n", __LINE__);
3265                 return 0;
3266         }
3267
3268         /*
3269          * We need to record the highest inode number for later 'lost+found'
3270          * dir creation.
3271          * We must select an ino not used/referred by any existing inode, or
3272          * 'lost+found' ino may be a missing ino in a corrupted leaf,
3273          * this may cause 'lost+found' dir has wrong nlinks.
3274          */
3275         cache = last_cache_extent(inode_cache);
3276         if (cache) {
3277                 node = container_of(cache, struct ptr_node, cache);
3278                 rec = node->data;
3279                 if (rec->ino > root->highest_inode)
3280                         root->highest_inode = rec->ino;
3281         }
3282
3283         /*
3284          * We need to repair backrefs first because we could change some of the
3285          * errors in the inode recs.
3286          *
3287          * We also need to go through and delete invalid backrefs first and then
3288          * add the correct ones second.  We do this because we may get EEXIST
3289          * when adding back the correct index because we hadn't yet deleted the
3290          * invalid index.
3291          *
3292          * For example, if we were missing a dir index then the directories
3293          * isize would be wrong, so if we fixed the isize to what we thought it
3294          * would be and then fixed the backref we'd still have a invalid fs, so
3295          * we need to add back the dir index and then check to see if the isize
3296          * is still wrong.
3297          */
3298         while (stage < 3) {
3299                 stage++;
3300                 if (stage == 3 && !err)
3301                         break;
3302
3303                 cache = search_cache_extent(inode_cache, 0);
3304                 while (repair && cache) {
3305                         node = container_of(cache, struct ptr_node, cache);
3306                         rec = node->data;
3307                         cache = next_cache_extent(cache);
3308
3309                         /* Need to free everything up and rescan */
3310                         if (stage == 3) {
3311                                 remove_cache_extent(inode_cache, &node->cache);
3312                                 free(node);
3313                                 free_inode_rec(rec);
3314                                 continue;
3315                         }
3316
3317                         if (list_empty(&rec->backrefs))
3318                                 continue;
3319
3320                         ret = repair_inode_backrefs(root, rec, inode_cache,
3321                                                     stage == 1);
3322                         if (ret < 0) {
3323                                 err = ret;
3324                                 stage = 2;
3325                                 break;
3326                         } if (ret > 0) {
3327                                 err = -EAGAIN;
3328                         }
3329                 }
3330         }
3331         if (err)
3332                 return err;
3333
3334         rec = get_inode_rec(inode_cache, root_dirid, 0);
3335         BUG_ON(IS_ERR(rec));
3336         if (rec) {
3337                 ret = check_root_dir(rec);
3338                 if (ret) {
3339                         fprintf(stderr, "root %llu root dir %llu error\n",
3340                                 (unsigned long long)root->root_key.objectid,
3341                                 (unsigned long long)root_dirid);
3342                         print_inode_error(root, rec);
3343                         error++;
3344                 }
3345         } else {
3346                 if (repair) {
3347                         struct btrfs_trans_handle *trans;
3348
3349                         trans = btrfs_start_transaction(root, 1);
3350                         if (IS_ERR(trans)) {
3351                                 err = PTR_ERR(trans);
3352                                 return err;
3353                         }
3354
3355                         fprintf(stderr,
3356                                 "root %llu missing its root dir, recreating\n",
3357                                 (unsigned long long)root->objectid);
3358
3359                         ret = btrfs_make_root_dir(trans, root, root_dirid);
3360                         BUG_ON(ret);
3361
3362                         btrfs_commit_transaction(trans, root);
3363                         return -EAGAIN;
3364                 }
3365
3366                 fprintf(stderr, "root %llu root dir %llu not found\n",
3367                         (unsigned long long)root->root_key.objectid,
3368                         (unsigned long long)root_dirid);
3369         }
3370
3371         while (1) {
3372                 cache = search_cache_extent(inode_cache, 0);
3373                 if (!cache)
3374                         break;
3375                 node = container_of(cache, struct ptr_node, cache);
3376                 rec = node->data;
3377                 remove_cache_extent(inode_cache, &node->cache);
3378                 free(node);
3379                 if (rec->ino == root_dirid ||
3380                     rec->ino == BTRFS_ORPHAN_OBJECTID) {
3381                         free_inode_rec(rec);
3382                         continue;
3383                 }
3384
3385                 if (rec->errors & I_ERR_NO_ORPHAN_ITEM) {
3386                         ret = check_orphan_item(root, rec->ino);
3387                         if (ret == 0)
3388                                 rec->errors &= ~I_ERR_NO_ORPHAN_ITEM;
3389                         if (can_free_inode_rec(rec)) {
3390                                 free_inode_rec(rec);
3391                                 continue;
3392                         }
3393                 }
3394
3395                 if (!rec->found_inode_item)
3396                         rec->errors |= I_ERR_NO_INODE_ITEM;
3397                 if (rec->found_link != rec->nlink)
3398                         rec->errors |= I_ERR_LINK_COUNT_WRONG;
3399                 if (repair) {
3400                         ret = try_repair_inode(root, rec);
3401                         if (ret == 0 && can_free_inode_rec(rec)) {
3402                                 free_inode_rec(rec);
3403                                 continue;
3404                         }
3405                         ret = 0;
3406                 }
3407
3408                 if (!(repair && ret == 0))
3409                         error++;
3410                 print_inode_error(root, rec);
3411                 list_for_each_entry(backref, &rec->backrefs, list) {
3412                         if (!backref->found_dir_item)
3413                                 backref->errors |= REF_ERR_NO_DIR_ITEM;
3414                         if (!backref->found_dir_index)
3415                                 backref->errors |= REF_ERR_NO_DIR_INDEX;
3416                         if (!backref->found_inode_ref)
3417                                 backref->errors |= REF_ERR_NO_INODE_REF;
3418                         fprintf(stderr, "\tunresolved ref dir %llu index %llu"
3419                                 " namelen %u name %s filetype %d errors %x",
3420                                 (unsigned long long)backref->dir,
3421                                 (unsigned long long)backref->index,
3422                                 backref->namelen, backref->name,
3423                                 backref->filetype, backref->errors);
3424                         print_ref_error(backref->errors);
3425                 }
3426                 free_inode_rec(rec);
3427         }
3428         return (error > 0) ? -1 : 0;
3429 }
3430
3431 static struct root_record *get_root_rec(struct cache_tree *root_cache,
3432                                         u64 objectid)
3433 {
3434         struct cache_extent *cache;
3435         struct root_record *rec = NULL;
3436         int ret;
3437
3438         cache = lookup_cache_extent(root_cache, objectid, 1);
3439         if (cache) {
3440                 rec = container_of(cache, struct root_record, cache);
3441         } else {
3442                 rec = calloc(1, sizeof(*rec));
3443                 if (!rec)
3444                         return ERR_PTR(-ENOMEM);
3445                 rec->objectid = objectid;
3446                 INIT_LIST_HEAD(&rec->backrefs);
3447                 rec->cache.start = objectid;
3448                 rec->cache.size = 1;
3449
3450                 ret = insert_cache_extent(root_cache, &rec->cache);
3451                 if (ret)
3452                         return ERR_PTR(-EEXIST);
3453         }
3454         return rec;
3455 }
3456
3457 static struct root_backref *get_root_backref(struct root_record *rec,
3458                                              u64 ref_root, u64 dir, u64 index,
3459                                              const char *name, int namelen)
3460 {
3461         struct root_backref *backref;
3462
3463         list_for_each_entry(backref, &rec->backrefs, list) {
3464                 if (backref->ref_root != ref_root || backref->dir != dir ||
3465                     backref->namelen != namelen)
3466                         continue;
3467                 if (memcmp(name, backref->name, namelen))
3468                         continue;
3469                 return backref;
3470         }
3471
3472         backref = calloc(1, sizeof(*backref) + namelen + 1);
3473         if (!backref)
3474                 return NULL;
3475         backref->ref_root = ref_root;
3476         backref->dir = dir;
3477         backref->index = index;
3478         backref->namelen = namelen;
3479         memcpy(backref->name, name, namelen);
3480         backref->name[namelen] = '\0';
3481         list_add_tail(&backref->list, &rec->backrefs);
3482         return backref;
3483 }
3484
3485 static void free_root_record(struct cache_extent *cache)
3486 {
3487         struct root_record *rec;
3488         struct root_backref *backref;
3489
3490         rec = container_of(cache, struct root_record, cache);
3491         while (!list_empty(&rec->backrefs)) {
3492                 backref = to_root_backref(rec->backrefs.next);
3493                 list_del(&backref->list);
3494                 free(backref);
3495         }
3496
3497         free(rec);
3498 }
3499
3500 FREE_EXTENT_CACHE_BASED_TREE(root_recs, free_root_record);
3501
3502 static int add_root_backref(struct cache_tree *root_cache,
3503                             u64 root_id, u64 ref_root, u64 dir, u64 index,
3504                             const char *name, int namelen,
3505                             int item_type, int errors)
3506 {
3507         struct root_record *rec;
3508         struct root_backref *backref;
3509
3510         rec = get_root_rec(root_cache, root_id);
3511         BUG_ON(IS_ERR(rec));
3512         backref = get_root_backref(rec, ref_root, dir, index, name, namelen);
3513         BUG_ON(!backref);
3514
3515         backref->errors |= errors;
3516
3517         if (item_type != BTRFS_DIR_ITEM_KEY) {
3518                 if (backref->found_dir_index || backref->found_back_ref ||
3519                     backref->found_forward_ref) {
3520                         if (backref->index != index)
3521                                 backref->errors |= REF_ERR_INDEX_UNMATCH;
3522                 } else {
3523                         backref->index = index;
3524                 }
3525         }
3526
3527         if (item_type == BTRFS_DIR_ITEM_KEY) {
3528                 if (backref->found_forward_ref)
3529                         rec->found_ref++;
3530                 backref->found_dir_item = 1;
3531         } else if (item_type == BTRFS_DIR_INDEX_KEY) {
3532                 backref->found_dir_index = 1;
3533         } else if (item_type == BTRFS_ROOT_REF_KEY) {
3534                 if (backref->found_forward_ref)
3535                         backref->errors |= REF_ERR_DUP_ROOT_REF;
3536                 else if (backref->found_dir_item)
3537                         rec->found_ref++;
3538                 backref->found_forward_ref = 1;
3539         } else if (item_type == BTRFS_ROOT_BACKREF_KEY) {
3540                 if (backref->found_back_ref)
3541                         backref->errors |= REF_ERR_DUP_ROOT_BACKREF;
3542                 backref->found_back_ref = 1;
3543         } else {
3544                 BUG_ON(1);
3545         }
3546
3547         if (backref->found_forward_ref && backref->found_dir_item)
3548                 backref->reachable = 1;
3549         return 0;
3550 }
3551
3552 static int merge_root_recs(struct btrfs_root *root,
3553                            struct cache_tree *src_cache,
3554                            struct cache_tree *dst_cache)
3555 {
3556         struct cache_extent *cache;
3557         struct ptr_node *node;
3558         struct inode_record *rec;
3559         struct inode_backref *backref;
3560         int ret = 0;
3561
3562         if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
3563                 free_inode_recs_tree(src_cache);
3564                 return 0;
3565         }
3566
3567         while (1) {
3568                 cache = search_cache_extent(src_cache, 0);
3569                 if (!cache)
3570                         break;
3571                 node = container_of(cache, struct ptr_node, cache);
3572                 rec = node->data;
3573                 remove_cache_extent(src_cache, &node->cache);
3574                 free(node);
3575
3576                 ret = is_child_root(root, root->objectid, rec->ino);
3577                 if (ret < 0)
3578                         break;
3579                 else if (ret == 0)
3580                         goto skip;
3581
3582                 list_for_each_entry(backref, &rec->backrefs, list) {
3583                         BUG_ON(backref->found_inode_ref);
3584                         if (backref->found_dir_item)
3585                                 add_root_backref(dst_cache, rec->ino,
3586                                         root->root_key.objectid, backref->dir,
3587                                         backref->index, backref->name,
3588                                         backref->namelen, BTRFS_DIR_ITEM_KEY,
3589                                         backref->errors);
3590                         if (backref->found_dir_index)
3591                                 add_root_backref(dst_cache, rec->ino,
3592                                         root->root_key.objectid, backref->dir,
3593                                         backref->index, backref->name,
3594                                         backref->namelen, BTRFS_DIR_INDEX_KEY,
3595                                         backref->errors);
3596                 }
3597 skip:
3598                 free_inode_rec(rec);
3599         }
3600         if (ret < 0)
3601                 return ret;
3602         return 0;
3603 }
3604
3605 static int check_root_refs(struct btrfs_root *root,
3606                            struct cache_tree *root_cache)
3607 {
3608         struct root_record *rec;
3609         struct root_record *ref_root;
3610         struct root_backref *backref;
3611         struct cache_extent *cache;
3612         int loop = 1;
3613         int ret;
3614         int error;
3615         int errors = 0;
3616
3617         rec = get_root_rec(root_cache, BTRFS_FS_TREE_OBJECTID);
3618         BUG_ON(IS_ERR(rec));
3619         rec->found_ref = 1;
3620
3621         /* fixme: this can not detect circular references */
3622         while (loop) {
3623                 loop = 0;
3624                 cache = search_cache_extent(root_cache, 0);
3625                 while (1) {
3626                         if (!cache)
3627                                 break;
3628                         rec = container_of(cache, struct root_record, cache);
3629                         cache = next_cache_extent(cache);
3630
3631                         if (rec->found_ref == 0)
3632                                 continue;
3633
3634                         list_for_each_entry(backref, &rec->backrefs, list) {
3635                                 if (!backref->reachable)
3636                                         continue;
3637
3638                                 ref_root = get_root_rec(root_cache,
3639                                                         backref->ref_root);
3640                                 BUG_ON(IS_ERR(ref_root));
3641                                 if (ref_root->found_ref > 0)
3642                                         continue;
3643
3644                                 backref->reachable = 0;
3645                                 rec->found_ref--;
3646                                 if (rec->found_ref == 0)
3647                                         loop = 1;
3648                         }
3649                 }
3650         }
3651
3652         cache = search_cache_extent(root_cache, 0);
3653         while (1) {
3654                 if (!cache)
3655                         break;
3656                 rec = container_of(cache, struct root_record, cache);
3657                 cache = next_cache_extent(cache);
3658
3659                 if (rec->found_ref == 0 &&
3660                     rec->objectid >= BTRFS_FIRST_FREE_OBJECTID &&
3661                     rec->objectid <= BTRFS_LAST_FREE_OBJECTID) {
3662                         ret = check_orphan_item(root->fs_info->tree_root,
3663                                                 rec->objectid);
3664                         if (ret == 0)
3665                                 continue;
3666
3667                         /*
3668                          * If we don't have a root item then we likely just have
3669                          * a dir item in a snapshot for this root but no actual
3670                          * ref key or anything so it's meaningless.
3671                          */
3672                         if (!rec->found_root_item)
3673                                 continue;
3674                         errors++;
3675                         fprintf(stderr, "fs tree %llu not referenced\n",
3676                                 (unsigned long long)rec->objectid);
3677                 }
3678
3679                 error = 0;
3680                 if (rec->found_ref > 0 && !rec->found_root_item)
3681                         error = 1;
3682                 list_for_each_entry(backref, &rec->backrefs, list) {
3683                         if (!backref->found_dir_item)
3684                                 backref->errors |= REF_ERR_NO_DIR_ITEM;
3685                         if (!backref->found_dir_index)
3686                                 backref->errors |= REF_ERR_NO_DIR_INDEX;
3687                         if (!backref->found_back_ref)
3688                                 backref->errors |= REF_ERR_NO_ROOT_BACKREF;
3689                         if (!backref->found_forward_ref)
3690                                 backref->errors |= REF_ERR_NO_ROOT_REF;
3691                         if (backref->reachable && backref->errors)
3692                                 error = 1;
3693                 }
3694                 if (!error)
3695                         continue;
3696
3697                 errors++;
3698                 fprintf(stderr, "fs tree %llu refs %u %s\n",
3699                         (unsigned long long)rec->objectid, rec->found_ref,
3700                          rec->found_root_item ? "" : "not found");
3701
3702                 list_for_each_entry(backref, &rec->backrefs, list) {
3703                         if (!backref->reachable)
3704                                 continue;
3705                         if (!backref->errors && rec->found_root_item)
3706                                 continue;
3707                         fprintf(stderr, "\tunresolved ref root %llu dir %llu"
3708                                 " index %llu namelen %u name %s errors %x\n",
3709                                 (unsigned long long)backref->ref_root,
3710                                 (unsigned long long)backref->dir,
3711                                 (unsigned long long)backref->index,
3712                                 backref->namelen, backref->name,
3713                                 backref->errors);
3714                         print_ref_error(backref->errors);
3715                 }
3716         }
3717         return errors > 0 ? 1 : 0;
3718 }
3719
3720 static int process_root_ref(struct extent_buffer *eb, int slot,
3721                             struct btrfs_key *key,
3722                             struct cache_tree *root_cache)
3723 {
3724         u64 dirid;
3725         u64 index;
3726         u32 len;
3727         u32 name_len;
3728         struct btrfs_root_ref *ref;
3729         char namebuf[BTRFS_NAME_LEN];
3730         int error;
3731
3732         ref = btrfs_item_ptr(eb, slot, struct btrfs_root_ref);
3733
3734         dirid = btrfs_root_ref_dirid(eb, ref);
3735         index = btrfs_root_ref_sequence(eb, ref);
3736         name_len = btrfs_root_ref_name_len(eb, ref);
3737
3738         if (name_len <= BTRFS_NAME_LEN) {
3739                 len = name_len;
3740                 error = 0;
3741         } else {
3742                 len = BTRFS_NAME_LEN;
3743                 error = REF_ERR_NAME_TOO_LONG;
3744         }
3745         read_extent_buffer(eb, namebuf, (unsigned long)(ref + 1), len);
3746
3747         if (key->type == BTRFS_ROOT_REF_KEY) {
3748                 add_root_backref(root_cache, key->offset, key->objectid, dirid,
3749                                  index, namebuf, len, key->type, error);
3750         } else {
3751                 add_root_backref(root_cache, key->objectid, key->offset, dirid,
3752                                  index, namebuf, len, key->type, error);
3753         }
3754         return 0;
3755 }
3756
3757 static void free_corrupt_block(struct cache_extent *cache)
3758 {
3759         struct btrfs_corrupt_block *corrupt;
3760
3761         corrupt = container_of(cache, struct btrfs_corrupt_block, cache);
3762         free(corrupt);
3763 }
3764
3765 FREE_EXTENT_CACHE_BASED_TREE(corrupt_blocks, free_corrupt_block);
3766
3767 /*
3768  * Repair the btree of the given root.
3769  *
3770  * The fix is to remove the node key in corrupt_blocks cache_tree.
3771  * and rebalance the tree.
3772  * After the fix, the btree should be writeable.
3773  */
3774 static int repair_btree(struct btrfs_root *root,
3775                         struct cache_tree *corrupt_blocks)
3776 {
3777         struct btrfs_trans_handle *trans;
3778         struct btrfs_path path;
3779         struct btrfs_corrupt_block *corrupt;
3780         struct cache_extent *cache;
3781         struct btrfs_key key;
3782         u64 offset;
3783         int level;
3784         int ret = 0;
3785
3786         if (cache_tree_empty(corrupt_blocks))
3787                 return 0;
3788
3789         trans = btrfs_start_transaction(root, 1);
3790         if (IS_ERR(trans)) {
3791                 ret = PTR_ERR(trans);
3792                 fprintf(stderr, "Error starting transaction: %s\n",
3793                         strerror(-ret));
3794                 return ret;
3795         }
3796         btrfs_init_path(&path);
3797         cache = first_cache_extent(corrupt_blocks);
3798         while (cache) {
3799                 corrupt = container_of(cache, struct btrfs_corrupt_block,
3800                                        cache);
3801                 level = corrupt->level;
3802                 path.lowest_level = level;
3803                 key.objectid = corrupt->key.objectid;
3804                 key.type = corrupt->key.type;
3805                 key.offset = corrupt->key.offset;
3806
3807                 /*
3808                  * Here we don't want to do any tree balance, since it may
3809                  * cause a balance with corrupted brother leaf/node,
3810                  * so ins_len set to 0 here.
3811                  * Balance will be done after all corrupt node/leaf is deleted.
3812                  */
3813                 ret = btrfs_search_slot(trans, root, &key, &path, 0, 1);
3814                 if (ret < 0)
3815                         goto out;
3816                 offset = btrfs_node_blockptr(path.nodes[level],
3817                                              path.slots[level]);
3818
3819                 /* Remove the ptr */
3820                 ret = btrfs_del_ptr(trans, root, &path, level,
3821                                     path.slots[level]);
3822                 if (ret < 0)
3823                         goto out;
3824                 /*
3825                  * Remove the corresponding extent
3826                  * return value is not concerned.
3827                  */
3828                 btrfs_release_path(&path);
3829                 ret = btrfs_free_extent(trans, root, offset, root->nodesize,
3830                                         0, root->root_key.objectid,
3831                                         level - 1, 0);
3832                 cache = next_cache_extent(cache);
3833         }
3834
3835         /* Balance the btree using btrfs_search_slot() */
3836         cache = first_cache_extent(corrupt_blocks);
3837         while (cache) {
3838                 corrupt = container_of(cache, struct btrfs_corrupt_block,
3839                                        cache);
3840                 memcpy(&key, &corrupt->key, sizeof(key));
3841                 ret = btrfs_search_slot(trans, root, &key, &path, -1, 1);
3842                 if (ret < 0)
3843                         goto out;
3844                 /* return will always >0 since it won't find the item */
3845                 ret = 0;
3846                 btrfs_release_path(&path);
3847                 cache = next_cache_extent(cache);
3848         }
3849 out:
3850         btrfs_commit_transaction(trans, root);
3851         btrfs_release_path(&path);
3852         return ret;
3853 }
3854
3855 static int check_fs_root(struct btrfs_root *root,
3856                          struct cache_tree *root_cache,
3857                          struct walk_control *wc)
3858 {
3859         int ret = 0;
3860         int err = 0;
3861         int wret;
3862         int level;
3863         struct btrfs_path path;
3864         struct shared_node root_node;
3865         struct root_record *rec;
3866         struct btrfs_root_item *root_item = &root->root_item;
3867         struct cache_tree corrupt_blocks;
3868         struct orphan_data_extent *orphan;
3869         struct orphan_data_extent *tmp;
3870         enum btrfs_tree_block_status status;
3871         struct node_refs nrefs;
3872
3873         /*
3874          * Reuse the corrupt_block cache tree to record corrupted tree block
3875          *
3876          * Unlike the usage in extent tree check, here we do it in a per
3877          * fs/subvol tree base.
3878          */
3879         cache_tree_init(&corrupt_blocks);
3880         root->fs_info->corrupt_blocks = &corrupt_blocks;
3881
3882         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
3883                 rec = get_root_rec(root_cache, root->root_key.objectid);
3884                 BUG_ON(IS_ERR(rec));
3885                 if (btrfs_root_refs(root_item) > 0)
3886                         rec->found_root_item = 1;
3887         }
3888
3889         btrfs_init_path(&path);
3890         memset(&root_node, 0, sizeof(root_node));
3891         cache_tree_init(&root_node.root_cache);
3892         cache_tree_init(&root_node.inode_cache);
3893         memset(&nrefs, 0, sizeof(nrefs));
3894
3895         /* Move the orphan extent record to corresponding inode_record */
3896         list_for_each_entry_safe(orphan, tmp,
3897                                  &root->orphan_data_extents, list) {
3898                 struct inode_record *inode;
3899
3900                 inode = get_inode_rec(&root_node.inode_cache, orphan->objectid,
3901                                       1);
3902                 BUG_ON(IS_ERR(inode));
3903                 inode->errors |= I_ERR_FILE_EXTENT_ORPHAN;
3904                 list_move(&orphan->list, &inode->orphan_extents);
3905         }
3906
3907         level = btrfs_header_level(root->node);
3908         memset(wc->nodes, 0, sizeof(wc->nodes));
3909         wc->nodes[level] = &root_node;
3910         wc->active_node = level;
3911         wc->root_level = level;
3912
3913         /* We may not have checked the root block, lets do that now */
3914         if (btrfs_is_leaf(root->node))
3915                 status = btrfs_check_leaf(root, NULL, root->node);
3916         else
3917                 status = btrfs_check_node(root, NULL, root->node);
3918         if (status != BTRFS_TREE_BLOCK_CLEAN)
3919                 return -EIO;
3920
3921         if (btrfs_root_refs(root_item) > 0 ||
3922             btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
3923                 path.nodes[level] = root->node;
3924                 extent_buffer_get(root->node);
3925                 path.slots[level] = 0;
3926         } else {
3927                 struct btrfs_key key;
3928                 struct btrfs_disk_key found_key;
3929
3930                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
3931                 level = root_item->drop_level;
3932                 path.lowest_level = level;
3933                 if (level > btrfs_header_level(root->node) ||
3934                     level >= BTRFS_MAX_LEVEL) {
3935                         error("ignoring invalid drop level: %u", level);
3936                         goto skip_walking;
3937                 }
3938                 wret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
3939                 if (wret < 0)
3940                         goto skip_walking;
3941                 btrfs_node_key(path.nodes[level], &found_key,
3942                                 path.slots[level]);
3943                 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
3944                                         sizeof(found_key)));
3945         }
3946
3947         while (1) {
3948                 wret = walk_down_tree(root, &path, wc, &level, &nrefs);
3949                 if (wret < 0)
3950                         ret = wret;
3951                 if (wret != 0)
3952                         break;
3953
3954                 wret = walk_up_tree(root, &path, wc, &level);
3955                 if (wret < 0)
3956                         ret = wret;
3957                 if (wret != 0)
3958                         break;
3959         }
3960 skip_walking:
3961         btrfs_release_path(&path);
3962
3963         if (!cache_tree_empty(&corrupt_blocks)) {
3964                 struct cache_extent *cache;
3965                 struct btrfs_corrupt_block *corrupt;
3966
3967                 printf("The following tree block(s) is corrupted in tree %llu:\n",
3968                        root->root_key.objectid);
3969                 cache = first_cache_extent(&corrupt_blocks);
3970                 while (cache) {
3971                         corrupt = container_of(cache,
3972                                                struct btrfs_corrupt_block,
3973                                                cache);
3974                         printf("\ttree block bytenr: %llu, level: %d, node key: (%llu, %u, %llu)\n",
3975                                cache->start, corrupt->level,
3976                                corrupt->key.objectid, corrupt->key.type,
3977                                corrupt->key.offset);
3978                         cache = next_cache_extent(cache);
3979                 }
3980                 if (repair) {
3981                         printf("Try to repair the btree for root %llu\n",
3982                                root->root_key.objectid);
3983                         ret = repair_btree(root, &corrupt_blocks);
3984                         if (ret < 0)
3985                                 fprintf(stderr, "Failed to repair btree: %s\n",
3986                                         strerror(-ret));
3987                         if (!ret)
3988                                 printf("Btree for root %llu is fixed\n",
3989                                        root->root_key.objectid);
3990                 }
3991         }
3992
3993         err = merge_root_recs(root, &root_node.root_cache, root_cache);
3994         if (err < 0)
3995                 ret = err;
3996
3997         if (root_node.current) {
3998                 root_node.current->checked = 1;
3999                 maybe_free_inode_rec(&root_node.inode_cache,
4000                                 root_node.current);
4001         }
4002
4003         err = check_inode_recs(root, &root_node.inode_cache);
4004         if (!ret)
4005                 ret = err;
4006
4007         free_corrupt_blocks_tree(&corrupt_blocks);
4008         root->fs_info->corrupt_blocks = NULL;
4009         free_orphan_data_extents(&root->orphan_data_extents);
4010         return ret;
4011 }
4012
4013 static int fs_root_objectid(u64 objectid)
4014 {
4015         if (objectid == BTRFS_TREE_RELOC_OBJECTID ||
4016             objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
4017                 return 1;
4018         return is_fstree(objectid);
4019 }
4020
4021 static int check_fs_roots(struct btrfs_root *root,
4022                           struct cache_tree *root_cache)
4023 {
4024         struct btrfs_path path;
4025         struct btrfs_key key;
4026         struct walk_control wc;
4027         struct extent_buffer *leaf, *tree_node;
4028         struct btrfs_root *tmp_root;
4029         struct btrfs_root *tree_root = root->fs_info->tree_root;
4030         int ret;
4031         int err = 0;
4032
4033         if (ctx.progress_enabled) {
4034                 ctx.tp = TASK_FS_ROOTS;
4035                 task_start(ctx.info);
4036         }
4037
4038         /*
4039          * Just in case we made any changes to the extent tree that weren't
4040          * reflected into the free space cache yet.
4041          */
4042         if (repair)
4043                 reset_cached_block_groups(root->fs_info);
4044         memset(&wc, 0, sizeof(wc));
4045         cache_tree_init(&wc.shared);
4046         btrfs_init_path(&path);
4047
4048 again:
4049         key.offset = 0;
4050         key.objectid = 0;
4051         key.type = BTRFS_ROOT_ITEM_KEY;
4052         ret = btrfs_search_slot(NULL, tree_root, &key, &path, 0, 0);
4053         if (ret < 0) {
4054                 err = 1;
4055                 goto out;
4056         }
4057         tree_node = tree_root->node;
4058         while (1) {
4059                 if (tree_node != tree_root->node) {
4060                         free_root_recs_tree(root_cache);
4061                         btrfs_release_path(&path);
4062                         goto again;
4063                 }
4064                 leaf = path.nodes[0];
4065                 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
4066                         ret = btrfs_next_leaf(tree_root, &path);
4067                         if (ret) {
4068                                 if (ret < 0)
4069                                         err = 1;
4070                                 break;
4071                         }
4072                         leaf = path.nodes[0];
4073                 }
4074                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
4075                 if (key.type == BTRFS_ROOT_ITEM_KEY &&
4076                     fs_root_objectid(key.objectid)) {
4077                         if (key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
4078                                 tmp_root = btrfs_read_fs_root_no_cache(
4079                                                 root->fs_info, &key);
4080                         } else {
4081                                 key.offset = (u64)-1;
4082                                 tmp_root = btrfs_read_fs_root(
4083                                                 root->fs_info, &key);
4084                         }
4085                         if (IS_ERR(tmp_root)) {
4086                                 err = 1;
4087                                 goto next;
4088                         }
4089                         ret = check_fs_root(tmp_root, root_cache, &wc);
4090                         if (ret == -EAGAIN) {
4091                                 free_root_recs_tree(root_cache);
4092                                 btrfs_release_path(&path);
4093                                 goto again;
4094                         }
4095                         if (ret)
4096                                 err = 1;
4097                         if (key.objectid == BTRFS_TREE_RELOC_OBJECTID)
4098                                 btrfs_free_fs_root(tmp_root);
4099                 } else if (key.type == BTRFS_ROOT_REF_KEY ||
4100                            key.type == BTRFS_ROOT_BACKREF_KEY) {
4101                         process_root_ref(leaf, path.slots[0], &key,
4102                                          root_cache);
4103                 }
4104 next:
4105                 path.slots[0]++;
4106         }
4107 out:
4108         btrfs_release_path(&path);
4109         if (err)
4110                 free_extent_cache_tree(&wc.shared);
4111         if (!cache_tree_empty(&wc.shared))
4112                 fprintf(stderr, "warning line %d\n", __LINE__);
4113
4114         task_stop(ctx.info);
4115
4116         return err;
4117 }
4118
4119 /*
4120  * Find DIR_ITEM/DIR_INDEX for the given key and check it with the specified
4121  * INODE_REF/INODE_EXTREF match.
4122  *
4123  * @root:       the root of the fs/file tree
4124  * @ref_key:    the key of the INODE_REF/INODE_EXTREF
4125  * @key:        the key of the DIR_ITEM/DIR_INDEX
4126  * @index:      the index in the INODE_REF/INODE_EXTREF, be used to
4127  *              distinguish root_dir between normal dir/file
4128  * @name:       the name in the INODE_REF/INODE_EXTREF
4129  * @namelen:    the length of name in the INODE_REF/INODE_EXTREF
4130  * @mode:       the st_mode of INODE_ITEM
4131  *
4132  * Return 0 if no error occurred.
4133  * Return ROOT_DIR_ERROR if found DIR_ITEM/DIR_INDEX for root_dir.
4134  * Return DIR_ITEM_MISSING if couldn't find DIR_ITEM/DIR_INDEX for normal
4135  * dir/file.
4136  * Return DIR_ITEM_MISMATCH if INODE_REF/INODE_EXTREF and DIR_ITEM/DIR_INDEX
4137  * not match for normal dir/file.
4138  */
4139 static int find_dir_item(struct btrfs_root *root, struct btrfs_key *ref_key,
4140                          struct btrfs_key *key, u64 index, char *name,
4141                          u32 namelen, u32 mode)
4142 {
4143         struct btrfs_path path;
4144         struct extent_buffer *node;
4145         struct btrfs_dir_item *di;
4146         struct btrfs_key location;
4147         char namebuf[BTRFS_NAME_LEN] = {0};
4148         u32 total;
4149         u32 cur = 0;
4150         u32 len;
4151         u32 name_len;
4152         u32 data_len;
4153         u8 filetype;
4154         int slot;
4155         int ret;
4156
4157         btrfs_init_path(&path);
4158         ret = btrfs_search_slot(NULL, root, key, &path, 0, 0);
4159         if (ret < 0) {
4160                 ret = DIR_ITEM_MISSING;
4161                 goto out;
4162         }
4163
4164         /* Process root dir and goto out*/
4165         if (index == 0) {
4166                 if (ret == 0) {
4167                         ret = ROOT_DIR_ERROR;
4168                         error(
4169                         "root %llu INODE %s[%llu %llu] ROOT_DIR shouldn't have %s",
4170                                 root->objectid,
4171                                 ref_key->type == BTRFS_INODE_REF_KEY ?
4172                                         "REF" : "EXTREF",
4173                                 ref_key->objectid, ref_key->offset,
4174                                 key->type == BTRFS_DIR_ITEM_KEY ?
4175                                         "DIR_ITEM" : "DIR_INDEX");
4176                 } else {
4177                         ret = 0;
4178                 }
4179
4180                 goto out;
4181         }
4182
4183         /* Process normal file/dir */
4184         if (ret > 0) {
4185                 ret = DIR_ITEM_MISSING;
4186                 error(
4187                 "root %llu INODE %s[%llu %llu] doesn't have related %s[%llu %llu] namelen %u filename %s filetype %d",
4188                         root->objectid,
4189                         ref_key->type == BTRFS_INODE_REF_KEY ? "REF" : "EXTREF",
4190                         ref_key->objectid, ref_key->offset,
4191                         key->type == BTRFS_DIR_ITEM_KEY ?
4192                                 "DIR_ITEM" : "DIR_INDEX",
4193                         key->objectid, key->offset, namelen, name,
4194                         imode_to_type(mode));
4195                 goto out;
4196         }
4197
4198         /* Check whether inode_id/filetype/name match */
4199         node = path.nodes[0];
4200         slot = path.slots[0];
4201         di = btrfs_item_ptr(node, slot, struct btrfs_dir_item);
4202         total = btrfs_item_size_nr(node, slot);
4203         while (cur < total) {
4204                 ret = DIR_ITEM_MISMATCH;
4205                 name_len = btrfs_dir_name_len(node, di);
4206                 data_len = btrfs_dir_data_len(node, di);
4207
4208                 btrfs_dir_item_key_to_cpu(node, di, &location);
4209                 if (location.objectid != ref_key->objectid ||
4210                     location.type !=  BTRFS_INODE_ITEM_KEY ||
4211                     location.offset != 0)
4212                         goto next;
4213
4214                 filetype = btrfs_dir_type(node, di);
4215                 if (imode_to_type(mode) != filetype)
4216                         goto next;
4217
4218                 if (name_len <= BTRFS_NAME_LEN) {
4219                         len = name_len;
4220                 } else {
4221                         len = BTRFS_NAME_LEN;
4222                         warning("root %llu %s[%llu %llu] name too long %u, trimmed",
4223                         root->objectid,
4224                         key->type == BTRFS_DIR_ITEM_KEY ?
4225                         "DIR_ITEM" : "DIR_INDEX",
4226                         key->objectid, key->offset, name_len);
4227                 }
4228                 read_extent_buffer(node, namebuf, (unsigned long)(di + 1), len);
4229                 if (len != namelen || strncmp(namebuf, name, len))
4230                         goto next;
4231
4232                 ret = 0;
4233                 goto out;
4234 next:
4235                 len = sizeof(*di) + name_len + data_len;
4236                 di = (struct btrfs_dir_item *)((char *)di + len);
4237                 cur += len;
4238         }
4239         if (ret == DIR_ITEM_MISMATCH)
4240                 error(
4241                 "root %llu INODE %s[%llu %llu] and %s[%llu %llu] mismatch namelen %u filename %s filetype %d",
4242                         root->objectid,
4243                         ref_key->type == BTRFS_INODE_REF_KEY ? "REF" : "EXTREF",
4244                         ref_key->objectid, ref_key->offset,
4245                         key->type == BTRFS_DIR_ITEM_KEY ?
4246                                 "DIR_ITEM" : "DIR_INDEX",
4247                         key->objectid, key->offset, namelen, name,
4248                         imode_to_type(mode));
4249 out:
4250         btrfs_release_path(&path);
4251         return ret;
4252 }
4253
4254 /*
4255  * Traverse the given INODE_REF and call find_dir_item() to find related
4256  * DIR_ITEM/DIR_INDEX.
4257  *
4258  * @root:       the root of the fs/file tree
4259  * @ref_key:    the key of the INODE_REF
4260  * @refs:       the count of INODE_REF
4261  * @mode:       the st_mode of INODE_ITEM
4262  *
4263  * Return 0 if no error occurred.
4264  */
4265 static int check_inode_ref(struct btrfs_root *root, struct btrfs_key *ref_key,
4266                            struct extent_buffer *node, int slot, u64 *refs,
4267                            int mode)
4268 {
4269         struct btrfs_key key;
4270         struct btrfs_inode_ref *ref;
4271         char namebuf[BTRFS_NAME_LEN] = {0};
4272         u32 total;
4273         u32 cur = 0;
4274         u32 len;
4275         u32 name_len;
4276         u64 index;
4277         int ret, err = 0;
4278
4279         ref = btrfs_item_ptr(node, slot, struct btrfs_inode_ref);
4280         total = btrfs_item_size_nr(node, slot);
4281
4282 next:
4283         /* Update inode ref count */
4284         (*refs)++;
4285
4286         index = btrfs_inode_ref_index(node, ref);
4287         name_len = btrfs_inode_ref_name_len(node, ref);
4288         if (name_len <= BTRFS_NAME_LEN) {
4289                 len = name_len;
4290         } else {
4291                 len = BTRFS_NAME_LEN;
4292                 warning("root %llu INODE_REF[%llu %llu] name too long",
4293                         root->objectid, ref_key->objectid, ref_key->offset);
4294         }
4295
4296         read_extent_buffer(node, namebuf, (unsigned long)(ref + 1), len);
4297
4298         /* Check root dir ref name */
4299         if (index == 0 && strncmp(namebuf, "..", name_len)) {
4300                 error("root %llu INODE_REF[%llu %llu] ROOT_DIR name shouldn't be %s",
4301                       root->objectid, ref_key->objectid, ref_key->offset,
4302                       namebuf);
4303                 err |= ROOT_DIR_ERROR;
4304         }
4305
4306         /* Find related DIR_INDEX */
4307         key.objectid = ref_key->offset;
4308         key.type = BTRFS_DIR_INDEX_KEY;
4309         key.offset = index;
4310         ret = find_dir_item(root, ref_key, &key, index, namebuf, len, mode);
4311         err |= ret;
4312
4313         /* Find related dir_item */
4314         key.objectid = ref_key->offset;
4315         key.type = BTRFS_DIR_ITEM_KEY;
4316         key.offset = btrfs_name_hash(namebuf, len);
4317         ret = find_dir_item(root, ref_key, &key, index, namebuf, len, mode);
4318         err |= ret;
4319
4320         len = sizeof(*ref) + name_len;
4321         ref = (struct btrfs_inode_ref *)((char *)ref + len);
4322         cur += len;
4323         if (cur < total)
4324                 goto next;
4325
4326         return err;
4327 }
4328
4329 /*
4330  * Traverse the given INODE_EXTREF and call find_dir_item() to find related
4331  * DIR_ITEM/DIR_INDEX.
4332  *
4333  * @root:       the root of the fs/file tree
4334  * @ref_key:    the key of the INODE_EXTREF
4335  * @refs:       the count of INODE_EXTREF
4336  * @mode:       the st_mode of INODE_ITEM
4337  *
4338  * Return 0 if no error occurred.
4339  */
4340 static int check_inode_extref(struct btrfs_root *root,
4341                               struct btrfs_key *ref_key,
4342                               struct extent_buffer *node, int slot, u64 *refs,
4343                               int mode)
4344 {
4345         struct btrfs_key key;
4346         struct btrfs_inode_extref *extref;
4347         char namebuf[BTRFS_NAME_LEN] = {0};
4348         u32 total;
4349         u32 cur = 0;
4350         u32 len;
4351         u32 name_len;
4352         u64 index;
4353         u64 parent;
4354         int ret;
4355         int err = 0;
4356
4357         extref = btrfs_item_ptr(node, slot, struct btrfs_inode_extref);
4358         total = btrfs_item_size_nr(node, slot);
4359
4360 next:
4361         /* update inode ref count */
4362         (*refs)++;
4363         name_len = btrfs_inode_extref_name_len(node, extref);
4364         index = btrfs_inode_extref_index(node, extref);
4365         parent = btrfs_inode_extref_parent(node, extref);
4366         if (name_len <= BTRFS_NAME_LEN) {
4367                 len = name_len;
4368         } else {
4369                 len = BTRFS_NAME_LEN;
4370                 warning("root %llu INODE_EXTREF[%llu %llu] name too long",
4371                         root->objectid, ref_key->objectid, ref_key->offset);
4372         }
4373         read_extent_buffer(node, namebuf, (unsigned long)(extref + 1), len);
4374
4375         /* Check root dir ref name */
4376         if (index == 0 && strncmp(namebuf, "..", name_len)) {
4377                 error("root %llu INODE_EXTREF[%llu %llu] ROOT_DIR name shouldn't be %s",
4378                       root->objectid, ref_key->objectid, ref_key->offset,
4379                       namebuf);
4380                 err |= ROOT_DIR_ERROR;
4381         }
4382
4383         /* find related dir_index */
4384         key.objectid = parent;
4385         key.type = BTRFS_DIR_INDEX_KEY;
4386         key.offset = index;
4387         ret = find_dir_item(root, ref_key, &key, index, namebuf, len, mode);
4388         err |= ret;
4389
4390         /* find related dir_item */
4391         key.objectid = parent;
4392         key.type = BTRFS_DIR_ITEM_KEY;
4393         key.offset = btrfs_name_hash(namebuf, len);
4394         ret = find_dir_item(root, ref_key, &key, index, namebuf, len, mode);
4395         err |= ret;
4396
4397         len = sizeof(*extref) + name_len;
4398         extref = (struct btrfs_inode_extref *)((char *)extref + len);
4399         cur += len;
4400
4401         if (cur < total)
4402                 goto next;
4403
4404         return err;
4405 }
4406
4407 /*
4408  * Find INODE_REF/INODE_EXTREF for the given key and check it with the specified
4409  * DIR_ITEM/DIR_INDEX match.
4410  *
4411  * @root:       the root of the fs/file tree
4412  * @key:        the key of the INODE_REF/INODE_EXTREF
4413  * @name:       the name in the INODE_REF/INODE_EXTREF
4414  * @namelen:    the length of name in the INODE_REF/INODE_EXTREF
4415  * @index:      the index in the INODE_REF/INODE_EXTREF, for DIR_ITEM set index
4416  * to (u64)-1
4417  * @ext_ref:    the EXTENDED_IREF feature
4418  *
4419  * Return 0 if no error occurred.
4420  * Return >0 for error bitmap
4421  */
4422 static int find_inode_ref(struct btrfs_root *root, struct btrfs_key *key,
4423                           char *name, int namelen, u64 index,
4424                           unsigned int ext_ref)
4425 {
4426         struct btrfs_path path;
4427         struct btrfs_inode_ref *ref;
4428         struct btrfs_inode_extref *extref;
4429         struct extent_buffer *node;
4430         char ref_namebuf[BTRFS_NAME_LEN] = {0};
4431         u32 total;
4432         u32 cur = 0;
4433         u32 len;
4434         u32 ref_namelen;
4435         u64 ref_index;
4436         u64 parent;
4437         u64 dir_id;
4438         int slot;
4439         int ret;
4440
4441         btrfs_init_path(&path);
4442         ret = btrfs_search_slot(NULL, root, key, &path, 0, 0);
4443         if (ret) {
4444                 ret = INODE_REF_MISSING;
4445                 goto extref;
4446         }
4447
4448         node = path.nodes[0];
4449         slot = path.slots[0];
4450
4451         ref = btrfs_item_ptr(node, slot, struct btrfs_inode_ref);
4452         total = btrfs_item_size_nr(node, slot);
4453
4454         /* Iterate all entry of INODE_REF */
4455         while (cur < total) {
4456                 ret = INODE_REF_MISSING;
4457
4458                 ref_namelen = btrfs_inode_ref_name_len(node, ref);
4459                 ref_index = btrfs_inode_ref_index(node, ref);
4460                 if (index != (u64)-1 && index != ref_index)
4461                         goto next_ref;
4462
4463                 if (ref_namelen <= BTRFS_NAME_LEN) {
4464                         len = ref_namelen;
4465                 } else {
4466                         len = BTRFS_NAME_LEN;
4467                         warning("root %llu INODE %s[%llu %llu] name too long",
4468                                 root->objectid,
4469                                 key->type == BTRFS_INODE_REF_KEY ?
4470                                         "REF" : "EXTREF",
4471                                 key->objectid, key->offset);
4472                 }
4473                 read_extent_buffer(node, ref_namebuf, (unsigned long)(ref + 1),
4474                                    len);
4475
4476                 if (len != namelen || strncmp(ref_namebuf, name, len))
4477                         goto next_ref;
4478
4479                 ret = 0;
4480                 goto out;
4481 next_ref:
4482                 len = sizeof(*ref) + ref_namelen;
4483                 ref = (struct btrfs_inode_ref *)((char *)ref + len);
4484                 cur += len;
4485         }
4486
4487 extref:
4488         /* Skip if not support EXTENDED_IREF feature */
4489         if (!ext_ref)
4490                 goto out;
4491
4492         btrfs_release_path(&path);
4493         btrfs_init_path(&path);
4494
4495         dir_id = key->offset;
4496         key->type = BTRFS_INODE_EXTREF_KEY;
4497         key->offset = btrfs_extref_hash(dir_id, name, namelen);
4498
4499         ret = btrfs_search_slot(NULL, root, key, &path, 0, 0);
4500         if (ret) {
4501                 ret = INODE_REF_MISSING;
4502                 goto out;
4503         }
4504
4505         node = path.nodes[0];
4506         slot = path.slots[0];
4507
4508         extref = btrfs_item_ptr(node, slot, struct btrfs_inode_extref);
4509         cur = 0;
4510         total = btrfs_item_size_nr(node, slot);
4511
4512         /* Iterate all entry of INODE_EXTREF */
4513         while (cur < total) {
4514                 ret = INODE_REF_MISSING;
4515
4516                 ref_namelen = btrfs_inode_extref_name_len(node, extref);
4517                 ref_index = btrfs_inode_extref_index(node, extref);
4518                 parent = btrfs_inode_extref_parent(node, extref);
4519                 if (index != (u64)-1 && index != ref_index)
4520                         goto next_extref;
4521
4522                 if (parent != dir_id)
4523                         goto next_extref;
4524
4525                 if (ref_namelen <= BTRFS_NAME_LEN) {
4526                         len = ref_namelen;
4527                 } else {
4528                         len = BTRFS_NAME_LEN;
4529                         warning("Warning: root %llu INODE %s[%llu %llu] name too long\n",
4530                                 root->objectid,
4531                                 key->type == BTRFS_INODE_REF_KEY ?
4532                                         "REF" : "EXTREF",
4533                                 key->objectid, key->offset);
4534                 }
4535                 read_extent_buffer(node, ref_namebuf,
4536                                    (unsigned long)(extref + 1), len);
4537
4538                 if (len != namelen || strncmp(ref_namebuf, name, len))
4539                         goto next_extref;
4540
4541                 ret = 0;
4542                 goto out;
4543
4544 next_extref:
4545                 len = sizeof(*extref) + ref_namelen;
4546                 extref = (struct btrfs_inode_extref *)((char *)extref + len);
4547                 cur += len;
4548
4549         }
4550 out:
4551         btrfs_release_path(&path);
4552         return ret;
4553 }
4554
4555 /*
4556  * Traverse the given DIR_ITEM/DIR_INDEX and check related INODE_ITEM and
4557  * call find_inode_ref() to check related INODE_REF/INODE_EXTREF.
4558  *
4559  * @root:       the root of the fs/file tree
4560  * @key:        the key of the INODE_REF/INODE_EXTREF
4561  * @size:       the st_size of the INODE_ITEM
4562  * @ext_ref:    the EXTENDED_IREF feature
4563  *
4564  * Return 0 if no error occurred.
4565  */
4566 static int check_dir_item(struct btrfs_root *root, struct btrfs_key *key,
4567                           struct extent_buffer *node, int slot, u64 *size,
4568                           unsigned int ext_ref)
4569 {
4570         struct btrfs_dir_item *di;
4571         struct btrfs_inode_item *ii;
4572         struct btrfs_path path;
4573         struct btrfs_key location;
4574         char namebuf[BTRFS_NAME_LEN] = {0};
4575         u32 total;
4576         u32 cur = 0;
4577         u32 len;
4578         u32 name_len;
4579         u32 data_len;
4580         u8 filetype;
4581         u32 mode;
4582         u64 index;
4583         int ret;
4584         int err = 0;
4585
4586         /*
4587          * For DIR_ITEM set index to (u64)-1, so that find_inode_ref
4588          * ignore index check.
4589          */
4590         index = (key->type == BTRFS_DIR_INDEX_KEY) ? key->offset : (u64)-1;
4591
4592         di = btrfs_item_ptr(node, slot, struct btrfs_dir_item);
4593         total = btrfs_item_size_nr(node, slot);
4594
4595         while (cur < total) {
4596                 data_len = btrfs_dir_data_len(node, di);
4597                 if (data_len)
4598                         error("root %llu %s[%llu %llu] data_len shouldn't be %u",
4599                               root->objectid, key->type == BTRFS_DIR_ITEM_KEY ?
4600                               "DIR_ITEM" : "DIR_INDEX",
4601                               key->objectid, key->offset, data_len);
4602
4603                 name_len = btrfs_dir_name_len(node, di);
4604                 if (name_len <= BTRFS_NAME_LEN) {
4605                         len = name_len;
4606                 } else {
4607                         len = BTRFS_NAME_LEN;
4608                         warning("root %llu %s[%llu %llu] name too long",
4609                                 root->objectid,
4610                                 key->type == BTRFS_DIR_ITEM_KEY ?
4611                                 "DIR_ITEM" : "DIR_INDEX",
4612                                 key->objectid, key->offset);
4613                 }
4614                 (*size) += name_len;
4615
4616                 read_extent_buffer(node, namebuf, (unsigned long)(di + 1), len);
4617                 filetype = btrfs_dir_type(node, di);
4618
4619                 btrfs_init_path(&path);
4620                 btrfs_dir_item_key_to_cpu(node, di, &location);
4621
4622                 /* Ignore related ROOT_ITEM check */
4623                 if (location.type == BTRFS_ROOT_ITEM_KEY)
4624                         goto next;
4625
4626                 /* Check relative INODE_ITEM(existence/filetype) */
4627                 ret = btrfs_search_slot(NULL, root, &location, &path, 0, 0);
4628                 if (ret) {
4629                         err |= INODE_ITEM_MISSING;
4630                         error("root %llu %s[%llu %llu] couldn't find relative INODE_ITEM[%llu] namelen %u filename %s filetype %x",
4631                               root->objectid, key->type == BTRFS_DIR_ITEM_KEY ?
4632                               "DIR_ITEM" : "DIR_INDEX", key->objectid,
4633                               key->offset, location.objectid, name_len,
4634                               namebuf, filetype);
4635                         goto next;
4636                 }
4637
4638                 ii = btrfs_item_ptr(path.nodes[0], path.slots[0],
4639                                     struct btrfs_inode_item);
4640                 mode = btrfs_inode_mode(path.nodes[0], ii);
4641
4642                 if (imode_to_type(mode) != filetype) {
4643                         err |= INODE_ITEM_MISMATCH;
4644                         error("root %llu %s[%llu %llu] relative INODE_ITEM filetype mismatch namelen %u filename %s filetype %d",
4645                               root->objectid, key->type == BTRFS_DIR_ITEM_KEY ?
4646                               "DIR_ITEM" : "DIR_INDEX", key->objectid,
4647                               key->offset, name_len, namebuf, filetype);
4648                 }
4649
4650                 /* Check relative INODE_REF/INODE_EXTREF */
4651                 location.type = BTRFS_INODE_REF_KEY;
4652                 location.offset = key->objectid;
4653                 ret = find_inode_ref(root, &location, namebuf, len,
4654                                        index, ext_ref);
4655                 err |= ret;
4656                 if (ret & INODE_REF_MISSING)
4657                         error("root %llu %s[%llu %llu] relative INODE_REF missing namelen %u filename %s filetype %d",
4658                               root->objectid, key->type == BTRFS_DIR_ITEM_KEY ?
4659                               "DIR_ITEM" : "DIR_INDEX", key->objectid,
4660                               key->offset, name_len, namebuf, filetype);
4661
4662 next:
4663                 btrfs_release_path(&path);
4664                 len = sizeof(*di) + name_len + data_len;
4665                 di = (struct btrfs_dir_item *)((char *)di + len);
4666                 cur += len;
4667
4668                 if (key->type == BTRFS_DIR_INDEX_KEY && cur < total) {
4669                         error("root %llu DIR_INDEX[%llu %llu] should contain only one entry",
4670                               root->objectid, key->objectid, key->offset);
4671                         break;
4672                 }
4673         }
4674
4675         return err;
4676 }
4677
4678 /*
4679  * Check file extent datasum/hole, update the size of the file extents,
4680  * check and update the last offset of the file extent.
4681  *
4682  * @root:       the root of fs/file tree.
4683  * @fkey:       the key of the file extent.
4684  * @nodatasum:  INODE_NODATASUM feature.
4685  * @size:       the sum of all EXTENT_DATA items size for this inode.
4686  * @end:        the offset of the last extent.
4687  *
4688  * Return 0 if no error occurred.
4689  */
4690 static int check_file_extent(struct btrfs_root *root, struct btrfs_key *fkey,
4691                              struct extent_buffer *node, int slot,
4692                              unsigned int nodatasum, u64 *size, u64 *end)
4693 {
4694         struct btrfs_file_extent_item *fi;
4695         u64 disk_bytenr;
4696         u64 disk_num_bytes;
4697         u64 extent_num_bytes;
4698         u64 found;
4699         unsigned int extent_type;
4700         unsigned int is_hole;
4701         int ret;
4702         int err = 0;
4703
4704         fi = btrfs_item_ptr(node, slot, struct btrfs_file_extent_item);
4705
4706         extent_type = btrfs_file_extent_type(node, fi);
4707         /* Skip if file extent is inline */
4708         if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4709                 struct btrfs_item *e = btrfs_item_nr(slot);
4710                 u32 item_inline_len;
4711
4712                 item_inline_len = btrfs_file_extent_inline_item_len(node, e);
4713                 extent_num_bytes = btrfs_file_extent_inline_len(node, slot, fi);
4714                 if (extent_num_bytes == 0 ||
4715                     extent_num_bytes != item_inline_len)
4716                         err |= FILE_EXTENT_ERROR;
4717                 *size += extent_num_bytes;
4718                 return err;
4719         }
4720
4721         /* Check extent type */
4722         if (extent_type != BTRFS_FILE_EXTENT_REG &&
4723                         extent_type != BTRFS_FILE_EXTENT_PREALLOC) {
4724                 err |= FILE_EXTENT_ERROR;
4725                 error("root %llu EXTENT_DATA[%llu %llu] type bad",
4726                       root->objectid, fkey->objectid, fkey->offset);
4727                 return err;
4728         }
4729
4730         /* Check REG_EXTENT/PREALLOC_EXTENT */
4731         disk_bytenr = btrfs_file_extent_disk_bytenr(node, fi);
4732         disk_num_bytes = btrfs_file_extent_disk_num_bytes(node, fi);
4733         extent_num_bytes = btrfs_file_extent_num_bytes(node, fi);
4734         is_hole = (disk_bytenr == 0) && (disk_num_bytes == 0);
4735
4736         /* Check EXTENT_DATA datasum */
4737         ret = count_csum_range(root, disk_bytenr, disk_num_bytes, &found);
4738         if (found > 0 && nodatasum) {
4739                 err |= ODD_CSUM_ITEM;
4740                 error("root %llu EXTENT_DATA[%llu %llu] nodatasum shouldn't have datasum",
4741                       root->objectid, fkey->objectid, fkey->offset);
4742         } else if (extent_type == BTRFS_FILE_EXTENT_REG && !nodatasum &&
4743                    !is_hole &&
4744                    (ret < 0 || found == 0 || found < disk_num_bytes)) {
4745                 err |= CSUM_ITEM_MISSING;
4746                 error("root %llu EXTENT_DATA[%llu %llu] datasum missing",
4747                       root->objectid, fkey->objectid, fkey->offset);
4748         } else if (extent_type == BTRFS_FILE_EXTENT_PREALLOC && found > 0) {
4749                 err |= ODD_CSUM_ITEM;
4750                 error("root %llu EXTENT_DATA[%llu %llu] prealloc shouldn't have datasum",
4751                       root->objectid, fkey->objectid, fkey->offset);
4752         }
4753
4754         /* Check EXTENT_DATA hole */
4755         if (no_holes && is_hole) {
4756                 err |= FILE_EXTENT_ERROR;
4757                 error("root %llu EXTENT_DATA[%llu %llu] shouldn't be hole",
4758                       root->objectid, fkey->objectid, fkey->offset);
4759         } else if (!no_holes && *end != fkey->offset) {
4760                 err |= FILE_EXTENT_ERROR;
4761                 error("root %llu EXTENT_DATA[%llu %llu] interrupt",
4762                       root->objectid, fkey->objectid, fkey->offset);
4763         }
4764
4765         *end += extent_num_bytes;
4766         if (!is_hole)
4767                 *size += extent_num_bytes;
4768
4769         return err;
4770 }
4771
4772 /*
4773  * Check INODE_ITEM and related ITEMs (the same inode number)
4774  * 1. check link count
4775  * 2. check inode ref/extref
4776  * 3. check dir item/index
4777  *
4778  * @ext_ref:    the EXTENDED_IREF feature
4779  *
4780  * Return 0 if no error occurred.
4781  * Return >0 for error or hit the traversal is done(by error bitmap)
4782  */
4783 static int check_inode_item(struct btrfs_root *root, struct btrfs_path *path,
4784                             unsigned int ext_ref)
4785 {
4786         struct extent_buffer *node;
4787         struct btrfs_inode_item *ii;
4788         struct btrfs_key key;
4789         u64 inode_id;
4790         u32 mode;
4791         u64 nlink;
4792         u64 nbytes;
4793         u64 isize;
4794         u64 size = 0;
4795         u64 refs = 0;
4796         u64 extent_end = 0;
4797         u64 extent_size = 0;
4798         unsigned int dir;
4799         unsigned int nodatasum;
4800         int slot;
4801         int ret;
4802         int err = 0;
4803
4804         node = path->nodes[0];
4805         slot = path->slots[0];
4806
4807         btrfs_item_key_to_cpu(node, &key, slot);
4808         inode_id = key.objectid;
4809
4810         if (inode_id == BTRFS_ORPHAN_OBJECTID) {
4811                 ret = btrfs_next_item(root, path);
4812                 if (ret > 0)
4813                         err |= LAST_ITEM;
4814                 return err;
4815         }
4816
4817         ii = btrfs_item_ptr(node, slot, struct btrfs_inode_item);
4818         isize = btrfs_inode_size(node, ii);
4819         nbytes = btrfs_inode_nbytes(node, ii);
4820         mode = btrfs_inode_mode(node, ii);
4821         dir = imode_to_type(mode) == BTRFS_FT_DIR;
4822         nlink = btrfs_inode_nlink(node, ii);
4823         nodatasum = btrfs_inode_flags(node, ii) & BTRFS_INODE_NODATASUM;
4824
4825         while (1) {
4826                 ret = btrfs_next_item(root, path);
4827                 if (ret < 0) {
4828                         /* out will fill 'err' rusing current statistics */
4829                         goto out;
4830                 } else if (ret > 0) {
4831                         err |= LAST_ITEM;
4832                         goto out;
4833                 }
4834
4835                 node = path->nodes[0];
4836                 slot = path->slots[0];
4837                 btrfs_item_key_to_cpu(node, &key, slot);
4838                 if (key.objectid != inode_id)
4839                         goto out;
4840
4841                 switch (key.type) {
4842                 case BTRFS_INODE_REF_KEY:
4843                         ret = check_inode_ref(root, &key, node, slot, &refs,
4844                                               mode);
4845                         err |= ret;
4846                         break;
4847                 case BTRFS_INODE_EXTREF_KEY:
4848                         if (key.type == BTRFS_INODE_EXTREF_KEY && !ext_ref)
4849                                 warning("root %llu EXTREF[%llu %llu] isn't supported",
4850                                         root->objectid, key.objectid,
4851                                         key.offset);
4852                         ret = check_inode_extref(root, &key, node, slot, &refs,
4853                                                  mode);
4854                         err |= ret;
4855                         break;
4856                 case BTRFS_DIR_ITEM_KEY:
4857                 case BTRFS_DIR_INDEX_KEY:
4858                         if (!dir) {
4859                                 warning("root %llu INODE[%llu] mode %u shouldn't have DIR_INDEX[%llu %llu]",
4860                                         root->objectid, inode_id,
4861                                         imode_to_type(mode), key.objectid,
4862                                         key.offset);
4863                         }
4864                         ret = check_dir_item(root, &key, node, slot, &size,
4865                                              ext_ref);
4866                         err |= ret;
4867                         break;
4868                 case BTRFS_EXTENT_DATA_KEY:
4869                         if (dir) {
4870                                 warning("root %llu DIR INODE[%llu] shouldn't EXTENT_DATA[%llu %llu]",
4871                                         root->objectid, inode_id, key.objectid,
4872                                         key.offset);
4873                         }
4874                         ret = check_file_extent(root, &key, node, slot,
4875                                                 nodatasum, &extent_size,
4876                                                 &extent_end);
4877                         err |= ret;
4878                         break;
4879                 case BTRFS_XATTR_ITEM_KEY:
4880                         break;
4881                 default:
4882                         error("ITEM[%llu %u %llu] UNKNOWN TYPE",
4883                               key.objectid, key.type, key.offset);
4884                 }
4885         }
4886
4887 out:
4888         /* verify INODE_ITEM nlink/isize/nbytes */
4889         if (dir) {
4890                 if (nlink != 1) {
4891                         err |= LINK_COUNT_ERROR;
4892                         error("root %llu DIR INODE[%llu] shouldn't have more than one link(%llu)",
4893                               root->objectid, inode_id, nlink);
4894                 }
4895
4896                 /*
4897                  * Just a warning, as dir inode nbytes is just an
4898                  * instructive value.
4899                  */
4900                 if (!IS_ALIGNED(nbytes, root->nodesize)) {
4901                         warning("root %llu DIR INODE[%llu] nbytes should be aligned to %u",
4902                                 root->objectid, inode_id, root->nodesize);
4903                 }
4904
4905                 if (isize != size) {
4906                         err |= ISIZE_ERROR;
4907                         error("root %llu DIR INODE [%llu] size(%llu) not equal to %llu",
4908                               root->objectid, inode_id, isize, size);
4909                 }
4910         } else {
4911                 if (nlink != refs) {
4912                         err |= LINK_COUNT_ERROR;
4913                         error("root %llu INODE[%llu] nlink(%llu) not equal to inode_refs(%llu)",
4914                               root->objectid, inode_id, nlink, refs);
4915                 } else if (!nlink) {
4916                         err |= ORPHAN_ITEM;
4917                 }
4918
4919                 if (!nbytes && !no_holes && extent_end < isize) {
4920                         err |= NBYTES_ERROR;
4921                         error("root %llu INODE[%llu] size (%llu) should have a file extent hole",
4922                               root->objectid, inode_id, isize);
4923                 }
4924
4925                 if (nbytes != extent_size) {
4926                         err |= NBYTES_ERROR;
4927                         error("root %llu INODE[%llu] nbytes(%llu) not equal to extent_size(%llu)",
4928                               root->objectid, inode_id, nbytes, extent_size);
4929                 }
4930         }
4931
4932         return err;
4933 }
4934
4935 static int check_fs_first_inode(struct btrfs_root *root, unsigned int ext_ref)
4936 {
4937         struct btrfs_path *path;
4938         struct btrfs_key key;
4939         int err = 0;
4940         int ret;
4941
4942         path = btrfs_alloc_path();
4943         key.objectid = 256;
4944         key.type = BTRFS_INODE_ITEM_KEY;
4945         key.offset = 0;
4946
4947         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4948         if (ret < 0)
4949                 return ret;
4950         if (ret > 0) {
4951                 ret = 0;
4952                 err |= INODE_ITEM_MISSING;
4953         }
4954
4955         err |= check_inode_item(root, path, ext_ref);
4956         err &= ~LAST_ITEM;
4957         if (err && !ret)
4958                 ret = -EIO;
4959         btrfs_free_path(path);
4960         return ret;
4961 }
4962
4963 /*
4964  * Iterate all item on the tree and call check_inode_item() to check.
4965  *
4966  * @root:       the root of the tree to be checked.
4967  * @ext_ref:    the EXTENDED_IREF feature
4968  *
4969  * Return 0 if no error found.
4970  * Return <0 for error.
4971  */
4972 static int check_fs_root_v2(struct btrfs_root *root, unsigned int ext_ref)
4973 {
4974         struct btrfs_path *path;
4975         struct node_refs nrefs;
4976         struct btrfs_root_item *root_item = &root->root_item;
4977         int ret, wret;
4978         int level;
4979
4980         /*
4981          * We need to manually check the first inode item(256)
4982          * As the following traversal function will only start from
4983          * the first inode item in the leaf, if inode item(256) is missing
4984          * we will just skip it forever.
4985          */
4986         ret = check_fs_first_inode(root, ext_ref);
4987         if (ret < 0)
4988                 return ret;
4989
4990         path = btrfs_alloc_path();
4991         if (!path)
4992                 return -ENOMEM;
4993
4994         memset(&nrefs, 0, sizeof(nrefs));
4995         level = btrfs_header_level(root->node);
4996
4997         if (btrfs_root_refs(root_item) > 0 ||
4998             btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
4999                 path->nodes[level] = root->node;
5000                 path->slots[level] = 0;
5001                 extent_buffer_get(root->node);
5002         } else {
5003                 struct btrfs_key key;
5004
5005                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
5006                 level = root_item->drop_level;
5007                 path->lowest_level = level;
5008                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5009                 if (ret < 0)
5010                         goto out;
5011                 ret = 0;
5012         }
5013
5014         while (1) {
5015                 wret = walk_down_tree_v2(root, path, &level, &nrefs, ext_ref);
5016                 if (wret < 0)
5017                         ret = wret;
5018                 if (wret != 0)
5019                         break;
5020
5021                 wret = walk_up_tree_v2(root, path, &level);
5022                 if (wret < 0)
5023                         ret = wret;
5024                 if (wret != 0)
5025                         break;
5026         }
5027
5028 out:
5029         btrfs_free_path(path);
5030         return ret;
5031 }
5032
5033 /*
5034  * Find the relative ref for root_ref and root_backref.
5035  *
5036  * @root:       the root of the root tree.
5037  * @ref_key:    the key of the root ref.
5038  *
5039  * Return 0 if no error occurred.
5040  */
5041 static int check_root_ref(struct btrfs_root *root, struct btrfs_key *ref_key,
5042                           struct extent_buffer *node, int slot)
5043 {
5044         struct btrfs_path *path;
5045         struct btrfs_key key;
5046         struct btrfs_root_ref *ref;
5047         struct btrfs_root_ref *backref;
5048         char ref_name[BTRFS_NAME_LEN] = {0};
5049         char backref_name[BTRFS_NAME_LEN] = {0};
5050         u64 ref_dirid;
5051         u64 ref_seq;
5052         u32 ref_namelen;
5053         u64 backref_dirid;
5054         u64 backref_seq;
5055         u32 backref_namelen;
5056         u32 len;
5057         int ret;
5058         int err = 0;
5059
5060         ref = btrfs_item_ptr(node, slot, struct btrfs_root_ref);
5061         ref_dirid = btrfs_root_ref_dirid(node, ref);
5062         ref_seq = btrfs_root_ref_sequence(node, ref);
5063         ref_namelen = btrfs_root_ref_name_len(node, ref);
5064
5065         if (ref_namelen <= BTRFS_NAME_LEN) {
5066                 len = ref_namelen;
5067         } else {
5068                 len = BTRFS_NAME_LEN;
5069                 warning("%s[%llu %llu] ref_name too long",
5070                         ref_key->type == BTRFS_ROOT_REF_KEY ?
5071                         "ROOT_REF" : "ROOT_BACKREF", ref_key->objectid,
5072                         ref_key->offset);
5073         }
5074         read_extent_buffer(node, ref_name, (unsigned long)(ref + 1), len);
5075
5076         /* Find relative root_ref */
5077         key.objectid = ref_key->offset;
5078         key.type = BTRFS_ROOT_BACKREF_KEY + BTRFS_ROOT_REF_KEY - ref_key->type;
5079         key.offset = ref_key->objectid;
5080
5081         path = btrfs_alloc_path();
5082         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5083         if (ret) {
5084                 err |= ROOT_REF_MISSING;
5085                 error("%s[%llu %llu] couldn't find relative ref",
5086                       ref_key->type == BTRFS_ROOT_REF_KEY ?
5087                       "ROOT_REF" : "ROOT_BACKREF",
5088                       ref_key->objectid, ref_key->offset);
5089                 goto out;
5090         }
5091
5092         backref = btrfs_item_ptr(path->nodes[0], path->slots[0],
5093                                  struct btrfs_root_ref);
5094         backref_dirid = btrfs_root_ref_dirid(path->nodes[0], backref);
5095         backref_seq = btrfs_root_ref_sequence(path->nodes[0], backref);
5096         backref_namelen = btrfs_root_ref_name_len(path->nodes[0], backref);
5097
5098         if (backref_namelen <= BTRFS_NAME_LEN) {
5099                 len = backref_namelen;
5100         } else {
5101                 len = BTRFS_NAME_LEN;
5102                 warning("%s[%llu %llu] ref_name too long",
5103                         key.type == BTRFS_ROOT_REF_KEY ?
5104                         "ROOT_REF" : "ROOT_BACKREF",
5105                         key.objectid, key.offset);
5106         }
5107         read_extent_buffer(path->nodes[0], backref_name,
5108                            (unsigned long)(backref + 1), len);
5109
5110         if (ref_dirid != backref_dirid || ref_seq != backref_seq ||
5111             ref_namelen != backref_namelen ||
5112             strncmp(ref_name, backref_name, len)) {
5113                 err |= ROOT_REF_MISMATCH;
5114                 error("%s[%llu %llu] mismatch relative ref",
5115                       ref_key->type == BTRFS_ROOT_REF_KEY ?
5116                       "ROOT_REF" : "ROOT_BACKREF",
5117                       ref_key->objectid, ref_key->offset);
5118         }
5119 out:
5120         btrfs_free_path(path);
5121         return err;
5122 }
5123
5124 /*
5125  * Check all fs/file tree in low_memory mode.
5126  *
5127  * 1. for fs tree root item, call check_fs_root_v2()
5128  * 2. for fs tree root ref/backref, call check_root_ref()
5129  *
5130  * Return 0 if no error occurred.
5131  */
5132 static int check_fs_roots_v2(struct btrfs_fs_info *fs_info)
5133 {
5134         struct btrfs_root *tree_root = fs_info->tree_root;
5135         struct btrfs_root *cur_root = NULL;
5136         struct btrfs_path *path;
5137         struct btrfs_key key;
5138         struct extent_buffer *node;
5139         unsigned int ext_ref;
5140         int slot;
5141         int ret;
5142         int err = 0;
5143
5144         ext_ref = btrfs_fs_incompat(fs_info,
5145                                     BTRFS_FEATURE_INCOMPAT_EXTENDED_IREF);
5146
5147         path = btrfs_alloc_path();
5148         if (!path)
5149                 return -ENOMEM;
5150
5151         key.objectid = BTRFS_FS_TREE_OBJECTID;
5152         key.offset = 0;
5153         key.type = BTRFS_ROOT_ITEM_KEY;
5154
5155         ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
5156         if (ret < 0) {
5157                 err = ret;
5158                 goto out;
5159         } else if (ret > 0) {
5160                 err = -ENOENT;
5161                 goto out;
5162         }
5163
5164         while (1) {
5165                 node = path->nodes[0];
5166                 slot = path->slots[0];
5167                 btrfs_item_key_to_cpu(node, &key, slot);
5168                 if (key.objectid > BTRFS_LAST_FREE_OBJECTID)
5169                         goto out;
5170                 if (key.type == BTRFS_ROOT_ITEM_KEY &&
5171                     fs_root_objectid(key.objectid)) {
5172                         if (key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
5173                                 cur_root = btrfs_read_fs_root_no_cache(fs_info,
5174                                                                        &key);
5175                         } else {
5176                                 key.offset = (u64)-1;
5177                                 cur_root = btrfs_read_fs_root(fs_info, &key);
5178                         }
5179
5180                         if (IS_ERR(cur_root)) {
5181                                 error("Fail to read fs/subvol tree: %lld",
5182                                       key.objectid);
5183                                 err = -EIO;
5184                                 goto next;
5185                         }
5186
5187                         ret = check_fs_root_v2(cur_root, ext_ref);
5188                         err |= ret;
5189
5190                         if (key.objectid == BTRFS_TREE_RELOC_OBJECTID)
5191                                 btrfs_free_fs_root(cur_root);
5192                 } else if (key.type == BTRFS_ROOT_REF_KEY ||
5193                                 key.type == BTRFS_ROOT_BACKREF_KEY) {
5194                         ret = check_root_ref(tree_root, &key, node, slot);
5195                         err |= ret;
5196                 }
5197 next:
5198                 ret = btrfs_next_item(tree_root, path);
5199                 if (ret > 0)
5200                         goto out;
5201                 if (ret < 0) {
5202                         err = ret;
5203                         goto out;
5204                 }
5205         }
5206
5207 out:
5208         btrfs_free_path(path);
5209         return err;
5210 }
5211
5212 static int all_backpointers_checked(struct extent_record *rec, int print_errs)
5213 {
5214         struct list_head *cur = rec->backrefs.next;
5215         struct extent_backref *back;
5216         struct tree_backref *tback;
5217         struct data_backref *dback;
5218         u64 found = 0;
5219         int err = 0;
5220
5221         while(cur != &rec->backrefs) {
5222                 back = to_extent_backref(cur);
5223                 cur = cur->next;
5224                 if (!back->found_extent_tree) {
5225                         err = 1;
5226                         if (!print_errs)
5227                                 goto out;
5228                         if (back->is_data) {
5229                                 dback = to_data_backref(back);
5230                                 fprintf(stderr, "Backref %llu %s %llu"
5231                                         " owner %llu offset %llu num_refs %lu"
5232                                         " not found in extent tree\n",
5233                                         (unsigned long long)rec->start,
5234                                         back->full_backref ?
5235                                         "parent" : "root",
5236                                         back->full_backref ?
5237                                         (unsigned long long)dback->parent:
5238                                         (unsigned long long)dback->root,
5239                                         (unsigned long long)dback->owner,
5240                                         (unsigned long long)dback->offset,
5241                                         (unsigned long)dback->num_refs);
5242                         } else {
5243                                 tback = to_tree_backref(back);
5244                                 fprintf(stderr, "Backref %llu parent %llu"
5245                                         " root %llu not found in extent tree\n",
5246                                         (unsigned long long)rec->start,
5247                                         (unsigned long long)tback->parent,
5248                                         (unsigned long long)tback->root);
5249                         }
5250                 }
5251                 if (!back->is_data && !back->found_ref) {
5252                         err = 1;
5253                         if (!print_errs)
5254                                 goto out;
5255                         tback = to_tree_backref(back);
5256                         fprintf(stderr, "Backref %llu %s %llu not referenced back %p\n",
5257                                 (unsigned long long)rec->start,
5258                                 back->full_backref ? "parent" : "root",
5259                                 back->full_backref ?
5260                                 (unsigned long long)tback->parent :
5261                                 (unsigned long long)tback->root, back);
5262                 }
5263                 if (back->is_data) {
5264                         dback = to_data_backref(back);
5265                         if (dback->found_ref != dback->num_refs) {
5266                                 err = 1;
5267                                 if (!print_errs)
5268                                         goto out;
5269                                 fprintf(stderr, "Incorrect local backref count"
5270                                         " on %llu %s %llu owner %llu"
5271                                         " offset %llu found %u wanted %u back %p\n",
5272                                         (unsigned long long)rec->start,
5273                                         back->full_backref ?
5274                                         "parent" : "root",
5275                                         back->full_backref ?
5276                                         (unsigned long long)dback->parent:
5277                                         (unsigned long long)dback->root,
5278                                         (unsigned long long)dback->owner,
5279                                         (unsigned long long)dback->offset,
5280                                         dback->found_ref, dback->num_refs, back);
5281                         }
5282                         if (dback->disk_bytenr != rec->start) {
5283                                 err = 1;
5284                                 if (!print_errs)
5285                                         goto out;
5286                                 fprintf(stderr, "Backref disk bytenr does not"
5287                                         " match extent record, bytenr=%llu, "
5288                                         "ref bytenr=%llu\n",
5289                                         (unsigned long long)rec->start,
5290                                         (unsigned long long)dback->disk_bytenr);
5291                         }
5292
5293                         if (dback->bytes != rec->nr) {
5294                                 err = 1;
5295                                 if (!print_errs)
5296                                         goto out;
5297                                 fprintf(stderr, "Backref bytes do not match "
5298                                         "extent backref, bytenr=%llu, ref "
5299                                         "bytes=%llu, backref bytes=%llu\n",
5300                                         (unsigned long long)rec->start,
5301                                         (unsigned long long)rec->nr,
5302                                         (unsigned long long)dback->bytes);
5303                         }
5304                 }
5305                 if (!back->is_data) {
5306                         found += 1;
5307                 } else {
5308                         dback = to_data_backref(back);
5309                         found += dback->found_ref;
5310                 }
5311         }
5312         if (found != rec->refs) {
5313                 err = 1;
5314                 if (!print_errs)
5315                         goto out;
5316                 fprintf(stderr, "Incorrect global backref count "
5317                         "on %llu found %llu wanted %llu\n",
5318                         (unsigned long long)rec->start,
5319                         (unsigned long long)found,
5320                         (unsigned long long)rec->refs);
5321         }
5322 out:
5323         return err;
5324 }
5325
5326 static int free_all_extent_backrefs(struct extent_record *rec)
5327 {
5328         struct extent_backref *back;
5329         struct list_head *cur;
5330         while (!list_empty(&rec->backrefs)) {
5331                 cur = rec->backrefs.next;
5332                 back = to_extent_backref(cur);
5333                 list_del(cur);
5334                 free(back);
5335         }
5336         return 0;
5337 }
5338
5339 static void free_extent_record_cache(struct btrfs_fs_info *fs_info,
5340                                      struct cache_tree *extent_cache)
5341 {
5342         struct cache_extent *cache;
5343         struct extent_record *rec;
5344
5345         while (1) {
5346                 cache = first_cache_extent(extent_cache);
5347                 if (!cache)
5348                         break;
5349                 rec = container_of(cache, struct extent_record, cache);
5350                 remove_cache_extent(extent_cache, cache);
5351                 free_all_extent_backrefs(rec);
5352                 free(rec);
5353         }
5354 }
5355
5356 static int maybe_free_extent_rec(struct cache_tree *extent_cache,
5357                                  struct extent_record *rec)
5358 {
5359         if (rec->content_checked && rec->owner_ref_checked &&
5360             rec->extent_item_refs == rec->refs && rec->refs > 0 &&
5361             rec->num_duplicates == 0 && !all_backpointers_checked(rec, 0) &&
5362             !rec->bad_full_backref && !rec->crossing_stripes &&
5363             !rec->wrong_chunk_type) {
5364                 remove_cache_extent(extent_cache, &rec->cache);
5365                 free_all_extent_backrefs(rec);
5366                 list_del_init(&rec->list);
5367                 free(rec);
5368         }
5369         return 0;
5370 }
5371
5372 static int check_owner_ref(struct btrfs_root *root,
5373                             struct extent_record *rec,
5374                             struct extent_buffer *buf)
5375 {
5376         struct extent_backref *node;
5377         struct tree_backref *back;
5378         struct btrfs_root *ref_root;
5379         struct btrfs_key key;
5380         struct btrfs_path path;
5381         struct extent_buffer *parent;
5382         int level;
5383         int found = 0;
5384         int ret;
5385
5386         list_for_each_entry(node, &rec->backrefs, list) {
5387                 if (node->is_data)
5388                         continue;
5389                 if (!node->found_ref)
5390                         continue;
5391                 if (node->full_backref)
5392                         continue;
5393                 back = to_tree_backref(node);
5394                 if (btrfs_header_owner(buf) == back->root)
5395                         return 0;
5396         }
5397         BUG_ON(rec->is_root);
5398
5399         /* try to find the block by search corresponding fs tree */
5400         key.objectid = btrfs_header_owner(buf);
5401         key.type = BTRFS_ROOT_ITEM_KEY;
5402         key.offset = (u64)-1;
5403
5404         ref_root = btrfs_read_fs_root(root->fs_info, &key);
5405         if (IS_ERR(ref_root))
5406                 return 1;
5407
5408         level = btrfs_header_level(buf);
5409         if (level == 0)
5410                 btrfs_item_key_to_cpu(buf, &key, 0);
5411         else
5412                 btrfs_node_key_to_cpu(buf, &key, 0);
5413
5414         btrfs_init_path(&path);
5415         path.lowest_level = level + 1;
5416         ret = btrfs_search_slot(NULL, ref_root, &key, &path, 0, 0);
5417         if (ret < 0)
5418                 return 0;
5419
5420         parent = path.nodes[level + 1];
5421         if (parent && buf->start == btrfs_node_blockptr(parent,
5422                                                         path.slots[level + 1]))
5423                 found = 1;
5424
5425         btrfs_release_path(&path);
5426         return found ? 0 : 1;
5427 }
5428
5429 static int is_extent_tree_record(struct extent_record *rec)
5430 {
5431         struct list_head *cur = rec->backrefs.next;
5432         struct extent_backref *node;
5433         struct tree_backref *back;
5434         int is_extent = 0;
5435
5436         while(cur != &rec->backrefs) {
5437                 node = to_extent_backref(cur);
5438                 cur = cur->next;
5439                 if (node->is_data)
5440                         return 0;
5441                 back = to_tree_backref(node);
5442                 if (node->full_backref)
5443                         return 0;
5444                 if (back->root == BTRFS_EXTENT_TREE_OBJECTID)
5445                         is_extent = 1;
5446         }
5447         return is_extent;
5448 }
5449
5450
5451 static int record_bad_block_io(struct btrfs_fs_info *info,
5452                                struct cache_tree *extent_cache,
5453                                u64 start, u64 len)
5454 {
5455         struct extent_record *rec;
5456         struct cache_extent *cache;
5457         struct btrfs_key key;
5458
5459         cache = lookup_cache_extent(extent_cache, start, len);
5460         if (!cache)
5461                 return 0;
5462
5463         rec = container_of(cache, struct extent_record, cache);
5464         if (!is_extent_tree_record(rec))
5465                 return 0;
5466
5467         btrfs_disk_key_to_cpu(&key, &rec->parent_key);
5468         return btrfs_add_corrupt_extent_record(info, &key, start, len, 0);
5469 }
5470
5471 static int swap_values(struct btrfs_root *root, struct btrfs_path *path,
5472                        struct extent_buffer *buf, int slot)
5473 {
5474         if (btrfs_header_level(buf)) {
5475                 struct btrfs_key_ptr ptr1, ptr2;
5476
5477                 read_extent_buffer(buf, &ptr1, btrfs_node_key_ptr_offset(slot),
5478                                    sizeof(struct btrfs_key_ptr));
5479                 read_extent_buffer(buf, &ptr2,
5480                                    btrfs_node_key_ptr_offset(slot + 1),
5481                                    sizeof(struct btrfs_key_ptr));
5482                 write_extent_buffer(buf, &ptr1,
5483                                     btrfs_node_key_ptr_offset(slot + 1),
5484                                     sizeof(struct btrfs_key_ptr));
5485                 write_extent_buffer(buf, &ptr2,
5486                                     btrfs_node_key_ptr_offset(slot),
5487                                     sizeof(struct btrfs_key_ptr));
5488                 if (slot == 0) {
5489                         struct btrfs_disk_key key;
5490                         btrfs_node_key(buf, &key, 0);
5491                         btrfs_fixup_low_keys(root, path, &key,
5492                                              btrfs_header_level(buf) + 1);
5493                 }
5494         } else {
5495                 struct btrfs_item *item1, *item2;
5496                 struct btrfs_key k1, k2;
5497                 char *item1_data, *item2_data;
5498                 u32 item1_offset, item2_offset, item1_size, item2_size;
5499
5500                 item1 = btrfs_item_nr(slot);
5501                 item2 = btrfs_item_nr(slot + 1);
5502                 btrfs_item_key_to_cpu(buf, &k1, slot);
5503                 btrfs_item_key_to_cpu(buf, &k2, slot + 1);
5504                 item1_offset = btrfs_item_offset(buf, item1);
5505                 item2_offset = btrfs_item_offset(buf, item2);
5506                 item1_size = btrfs_item_size(buf, item1);
5507                 item2_size = btrfs_item_size(buf, item2);
5508
5509                 item1_data = malloc(item1_size);
5510                 if (!item1_data)
5511                         return -ENOMEM;
5512                 item2_data = malloc(item2_size);
5513                 if (!item2_data) {
5514                         free(item1_data);
5515                         return -ENOMEM;
5516                 }
5517
5518                 read_extent_buffer(buf, item1_data, item1_offset, item1_size);
5519                 read_extent_buffer(buf, item2_data, item2_offset, item2_size);
5520
5521                 write_extent_buffer(buf, item1_data, item2_offset, item2_size);
5522                 write_extent_buffer(buf, item2_data, item1_offset, item1_size);
5523                 free(item1_data);
5524                 free(item2_data);
5525
5526                 btrfs_set_item_offset(buf, item1, item2_offset);
5527                 btrfs_set_item_offset(buf, item2, item1_offset);
5528                 btrfs_set_item_size(buf, item1, item2_size);
5529                 btrfs_set_item_size(buf, item2, item1_size);
5530
5531                 path->slots[0] = slot;
5532                 btrfs_set_item_key_unsafe(root, path, &k2);
5533                 path->slots[0] = slot + 1;
5534                 btrfs_set_item_key_unsafe(root, path, &k1);
5535         }
5536         return 0;
5537 }
5538
5539 static int fix_key_order(struct btrfs_trans_handle *trans,
5540                          struct btrfs_root *root,
5541                          struct btrfs_path *path)
5542 {
5543         struct extent_buffer *buf;
5544         struct btrfs_key k1, k2;
5545         int i;
5546         int level = path->lowest_level;
5547         int ret = -EIO;
5548
5549         buf = path->nodes[level];
5550         for (i = 0; i < btrfs_header_nritems(buf) - 1; i++) {
5551                 if (level) {
5552                         btrfs_node_key_to_cpu(buf, &k1, i);
5553                         btrfs_node_key_to_cpu(buf, &k2, i + 1);
5554                 } else {
5555                         btrfs_item_key_to_cpu(buf, &k1, i);
5556                         btrfs_item_key_to_cpu(buf, &k2, i + 1);
5557                 }
5558                 if (btrfs_comp_cpu_keys(&k1, &k2) < 0)
5559                         continue;
5560                 ret = swap_values(root, path, buf, i);
5561                 if (ret)
5562                         break;
5563                 btrfs_mark_buffer_dirty(buf);
5564                 i = 0;
5565         }
5566         return ret;
5567 }
5568
5569 static int delete_bogus_item(struct btrfs_trans_handle *trans,
5570                              struct btrfs_root *root,
5571                              struct btrfs_path *path,
5572                              struct extent_buffer *buf, int slot)
5573 {
5574         struct btrfs_key key;
5575         int nritems = btrfs_header_nritems(buf);
5576
5577         btrfs_item_key_to_cpu(buf, &key, slot);
5578
5579         /* These are all the keys we can deal with missing. */
5580         if (key.type != BTRFS_DIR_INDEX_KEY &&
5581             key.type != BTRFS_EXTENT_ITEM_KEY &&
5582             key.type != BTRFS_METADATA_ITEM_KEY &&
5583             key.type != BTRFS_TREE_BLOCK_REF_KEY &&
5584             key.type != BTRFS_EXTENT_DATA_REF_KEY)
5585                 return -1;
5586
5587         printf("Deleting bogus item [%llu,%u,%llu] at slot %d on block %llu\n",
5588                (unsigned long long)key.objectid, key.type,
5589                (unsigned long long)key.offset, slot, buf->start);
5590         memmove_extent_buffer(buf, btrfs_item_nr_offset(slot),
5591                               btrfs_item_nr_offset(slot + 1),
5592                               sizeof(struct btrfs_item) *
5593                               (nritems - slot - 1));
5594         btrfs_set_header_nritems(buf, nritems - 1);
5595         if (slot == 0) {
5596                 struct btrfs_disk_key disk_key;
5597
5598                 btrfs_item_key(buf, &disk_key, 0);
5599                 btrfs_fixup_low_keys(root, path, &disk_key, 1);
5600         }
5601         btrfs_mark_buffer_dirty(buf);
5602         return 0;
5603 }
5604
5605 static int fix_item_offset(struct btrfs_trans_handle *trans,
5606                            struct btrfs_root *root,
5607                            struct btrfs_path *path)
5608 {
5609         struct extent_buffer *buf;
5610         int i;
5611         int ret = 0;
5612
5613         /* We should only get this for leaves */
5614         BUG_ON(path->lowest_level);
5615         buf = path->nodes[0];
5616 again:
5617         for (i = 0; i < btrfs_header_nritems(buf); i++) {
5618                 unsigned int shift = 0, offset;
5619
5620                 if (i == 0 && btrfs_item_end_nr(buf, i) !=
5621                     BTRFS_LEAF_DATA_SIZE(root)) {
5622                         if (btrfs_item_end_nr(buf, i) >
5623                             BTRFS_LEAF_DATA_SIZE(root)) {
5624                                 ret = delete_bogus_item(trans, root, path,
5625                                                         buf, i);
5626                                 if (!ret)
5627                                         goto again;
5628                                 fprintf(stderr, "item is off the end of the "
5629                                         "leaf, can't fix\n");
5630                                 ret = -EIO;
5631                                 break;
5632                         }
5633                         shift = BTRFS_LEAF_DATA_SIZE(root) -
5634                                 btrfs_item_end_nr(buf, i);
5635                 } else if (i > 0 && btrfs_item_end_nr(buf, i) !=
5636                            btrfs_item_offset_nr(buf, i - 1)) {
5637                         if (btrfs_item_end_nr(buf, i) >
5638                             btrfs_item_offset_nr(buf, i - 1)) {
5639                                 ret = delete_bogus_item(trans, root, path,
5640                                                         buf, i);
5641                                 if (!ret)
5642                                         goto again;
5643                                 fprintf(stderr, "items overlap, can't fix\n");
5644                                 ret = -EIO;
5645                                 break;
5646                         }
5647                         shift = btrfs_item_offset_nr(buf, i - 1) -
5648                                 btrfs_item_end_nr(buf, i);
5649                 }
5650                 if (!shift)
5651                         continue;
5652
5653                 printf("Shifting item nr %d by %u bytes in block %llu\n",
5654                        i, shift, (unsigned long long)buf->start);
5655                 offset = btrfs_item_offset_nr(buf, i);
5656                 memmove_extent_buffer(buf,
5657                                       btrfs_leaf_data(buf) + offset + shift,
5658                                       btrfs_leaf_data(buf) + offset,
5659                                       btrfs_item_size_nr(buf, i));
5660                 btrfs_set_item_offset(buf, btrfs_item_nr(i),
5661                                       offset + shift);
5662                 btrfs_mark_buffer_dirty(buf);
5663         }
5664
5665         /*
5666          * We may have moved things, in which case we want to exit so we don't
5667          * write those changes out.  Once we have proper abort functionality in
5668          * progs this can be changed to something nicer.
5669          */
5670         BUG_ON(ret);
5671         return ret;
5672 }
5673
5674 /*
5675  * Attempt to fix basic block failures.  If we can't fix it for whatever reason
5676  * then just return -EIO.
5677  */
5678 static int try_to_fix_bad_block(struct btrfs_root *root,
5679                                 struct extent_buffer *buf,
5680                                 enum btrfs_tree_block_status status)
5681 {
5682         struct btrfs_trans_handle *trans;
5683         struct ulist *roots;
5684         struct ulist_node *node;
5685         struct btrfs_root *search_root;
5686         struct btrfs_path path;
5687         struct ulist_iterator iter;
5688         struct btrfs_key root_key, key;
5689         int ret;
5690
5691         if (status != BTRFS_TREE_BLOCK_BAD_KEY_ORDER &&
5692             status != BTRFS_TREE_BLOCK_INVALID_OFFSETS)
5693                 return -EIO;
5694
5695         ret = btrfs_find_all_roots(NULL, root->fs_info, buf->start, 0, &roots);
5696         if (ret)
5697                 return -EIO;
5698
5699         btrfs_init_path(&path);
5700         ULIST_ITER_INIT(&iter);
5701         while ((node = ulist_next(roots, &iter))) {
5702                 root_key.objectid = node->val;
5703                 root_key.type = BTRFS_ROOT_ITEM_KEY;
5704                 root_key.offset = (u64)-1;
5705
5706                 search_root = btrfs_read_fs_root(root->fs_info, &root_key);
5707                 if (IS_ERR(root)) {
5708                         ret = -EIO;
5709                         break;
5710                 }
5711
5712
5713                 trans = btrfs_start_transaction(search_root, 0);
5714                 if (IS_ERR(trans)) {
5715                         ret = PTR_ERR(trans);
5716                         break;
5717                 }
5718
5719                 path.lowest_level = btrfs_header_level(buf);
5720                 path.skip_check_block = 1;
5721                 if (path.lowest_level)
5722                         btrfs_node_key_to_cpu(buf, &key, 0);
5723                 else
5724                         btrfs_item_key_to_cpu(buf, &key, 0);
5725                 ret = btrfs_search_slot(trans, search_root, &key, &path, 0, 1);
5726                 if (ret) {
5727                         ret = -EIO;
5728                         btrfs_commit_transaction(trans, search_root);
5729                         break;
5730                 }
5731                 if (status == BTRFS_TREE_BLOCK_BAD_KEY_ORDER)
5732                         ret = fix_key_order(trans, search_root, &path);
5733                 else if (status == BTRFS_TREE_BLOCK_INVALID_OFFSETS)
5734                         ret = fix_item_offset(trans, search_root, &path);
5735                 if (ret) {
5736                         btrfs_commit_transaction(trans, search_root);
5737                         break;
5738                 }
5739                 btrfs_release_path(&path);
5740                 btrfs_commit_transaction(trans, search_root);
5741         }
5742         ulist_free(roots);
5743         btrfs_release_path(&path);
5744         return ret;
5745 }
5746
5747 static int check_block(struct btrfs_root *root,
5748                        struct cache_tree *extent_cache,
5749                        struct extent_buffer *buf, u64 flags)
5750 {
5751         struct extent_record *rec;
5752         struct cache_extent *cache;
5753         struct btrfs_key key;
5754         enum btrfs_tree_block_status status;
5755         int ret = 0;
5756         int level;
5757
5758         cache = lookup_cache_extent(extent_cache, buf->start, buf->len);
5759         if (!cache)
5760                 return 1;
5761         rec = container_of(cache, struct extent_record, cache);
5762         rec->generation = btrfs_header_generation(buf);
5763
5764         level = btrfs_header_level(buf);
5765         if (btrfs_header_nritems(buf) > 0) {
5766
5767                 if (level == 0)
5768                         btrfs_item_key_to_cpu(buf, &key, 0);
5769                 else
5770                         btrfs_node_key_to_cpu(buf, &key, 0);
5771
5772                 rec->info_objectid = key.objectid;
5773         }
5774         rec->info_level = level;
5775
5776         if (btrfs_is_leaf(buf))
5777                 status = btrfs_check_leaf(root, &rec->parent_key, buf);
5778         else
5779                 status = btrfs_check_node(root, &rec->parent_key, buf);
5780
5781         if (status != BTRFS_TREE_BLOCK_CLEAN) {
5782                 if (repair)
5783                         status = try_to_fix_bad_block(root, buf, status);
5784                 if (status != BTRFS_TREE_BLOCK_CLEAN) {
5785                         ret = -EIO;
5786                         fprintf(stderr, "bad block %llu\n",
5787                                 (unsigned long long)buf->start);
5788                 } else {
5789                         /*
5790                          * Signal to callers we need to start the scan over
5791                          * again since we'll have cowed blocks.
5792                          */
5793                         ret = -EAGAIN;
5794                 }
5795         } else {
5796                 rec->content_checked = 1;
5797                 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5798                         rec->owner_ref_checked = 1;
5799                 else {
5800                         ret = check_owner_ref(root, rec, buf);
5801                         if (!ret)
5802                                 rec->owner_ref_checked = 1;
5803                 }
5804         }
5805         if (!ret)
5806                 maybe_free_extent_rec(extent_cache, rec);
5807         return ret;
5808 }
5809
5810 static struct tree_backref *find_tree_backref(struct extent_record *rec,
5811                                                 u64 parent, u64 root)
5812 {
5813         struct list_head *cur = rec->backrefs.next;
5814         struct extent_backref *node;
5815         struct tree_backref *back;
5816
5817         while(cur != &rec->backrefs) {
5818                 node = to_extent_backref(cur);
5819                 cur = cur->next;
5820                 if (node->is_data)
5821                         continue;
5822                 back = to_tree_backref(node);
5823                 if (parent > 0) {
5824                         if (!node->full_backref)
5825                                 continue;
5826                         if (parent == back->parent)
5827                                 return back;
5828                 } else {
5829                         if (node->full_backref)
5830                                 continue;
5831                         if (back->root == root)
5832                                 return back;
5833                 }
5834         }
5835         return NULL;
5836 }
5837
5838 static struct tree_backref *alloc_tree_backref(struct extent_record *rec,
5839                                                 u64 parent, u64 root)
5840 {
5841         struct tree_backref *ref = malloc(sizeof(*ref));
5842
5843         if (!ref)
5844                 return NULL;
5845         memset(&ref->node, 0, sizeof(ref->node));
5846         if (parent > 0) {
5847                 ref->parent = parent;
5848                 ref->node.full_backref = 1;
5849         } else {
5850                 ref->root = root;
5851                 ref->node.full_backref = 0;
5852         }
5853         list_add_tail(&ref->node.list, &rec->backrefs);
5854
5855         return ref;
5856 }
5857
5858 static struct data_backref *find_data_backref(struct extent_record *rec,
5859                                                 u64 parent, u64 root,
5860                                                 u64 owner, u64 offset,
5861                                                 int found_ref,
5862                                                 u64 disk_bytenr, u64 bytes)
5863 {
5864         struct list_head *cur = rec->backrefs.next;
5865         struct extent_backref *node;
5866         struct data_backref *back;
5867
5868         while(cur != &rec->backrefs) {
5869                 node = to_extent_backref(cur);
5870                 cur = cur->next;
5871                 if (!node->is_data)
5872                         continue;
5873                 back = to_data_backref(node);
5874                 if (parent > 0) {
5875                         if (!node->full_backref)
5876                                 continue;
5877                         if (parent == back->parent)
5878                                 return back;
5879                 } else {
5880                         if (node->full_backref)
5881                                 continue;
5882                         if (back->root == root && back->owner == owner &&
5883                             back->offset == offset) {
5884                                 if (found_ref && node->found_ref &&
5885                                     (back->bytes != bytes ||
5886                                     back->disk_bytenr != disk_bytenr))
5887                                         continue;
5888                                 return back;
5889                         }
5890                 }
5891         }
5892         return NULL;
5893 }
5894
5895 static struct data_backref *alloc_data_backref(struct extent_record *rec,
5896                                                 u64 parent, u64 root,
5897                                                 u64 owner, u64 offset,
5898                                                 u64 max_size)
5899 {
5900         struct data_backref *ref = malloc(sizeof(*ref));
5901
5902         if (!ref)
5903                 return NULL;
5904         memset(&ref->node, 0, sizeof(ref->node));
5905         ref->node.is_data = 1;
5906
5907         if (parent > 0) {
5908                 ref->parent = parent;
5909                 ref->owner = 0;
5910                 ref->offset = 0;
5911                 ref->node.full_backref = 1;
5912         } else {
5913                 ref->root = root;
5914                 ref->owner = owner;
5915                 ref->offset = offset;
5916                 ref->node.full_backref = 0;
5917         }
5918         ref->bytes = max_size;
5919         ref->found_ref = 0;
5920         ref->num_refs = 0;
5921         list_add_tail(&ref->node.list, &rec->backrefs);
5922         if (max_size > rec->max_size)
5923                 rec->max_size = max_size;
5924         return ref;
5925 }
5926
5927 /* Check if the type of extent matches with its chunk */
5928 static void check_extent_type(struct extent_record *rec)
5929 {
5930         struct btrfs_block_group_cache *bg_cache;
5931
5932         bg_cache = btrfs_lookup_first_block_group(global_info, rec->start);
5933         if (!bg_cache)
5934                 return;
5935
5936         /* data extent, check chunk directly*/
5937         if (!rec->metadata) {
5938                 if (!(bg_cache->flags & BTRFS_BLOCK_GROUP_DATA))
5939                         rec->wrong_chunk_type = 1;
5940                 return;
5941         }
5942
5943         /* metadata extent, check the obvious case first */
5944         if (!(bg_cache->flags & (BTRFS_BLOCK_GROUP_SYSTEM |
5945                                  BTRFS_BLOCK_GROUP_METADATA))) {
5946                 rec->wrong_chunk_type = 1;
5947                 return;
5948         }
5949
5950         /*
5951          * Check SYSTEM extent, as it's also marked as metadata, we can only
5952          * make sure it's a SYSTEM extent by its backref
5953          */
5954         if (!list_empty(&rec->backrefs)) {
5955                 struct extent_backref *node;
5956                 struct tree_backref *tback;
5957                 u64 bg_type;
5958
5959                 node = to_extent_backref(rec->backrefs.next);
5960                 if (node->is_data) {
5961                         /* tree block shouldn't have data backref */
5962                         rec->wrong_chunk_type = 1;
5963                         return;
5964                 }
5965                 tback = container_of(node, struct tree_backref, node);
5966
5967                 if (tback->root == BTRFS_CHUNK_TREE_OBJECTID)
5968                         bg_type = BTRFS_BLOCK_GROUP_SYSTEM;
5969                 else
5970                         bg_type = BTRFS_BLOCK_GROUP_METADATA;
5971                 if (!(bg_cache->flags & bg_type))
5972                         rec->wrong_chunk_type = 1;
5973         }
5974 }
5975
5976 /*
5977  * Allocate a new extent record, fill default values from @tmpl and insert int
5978  * @extent_cache. Caller is supposed to make sure the [start,nr) is not in
5979  * the cache, otherwise it fails.
5980  */
5981 static int add_extent_rec_nolookup(struct cache_tree *extent_cache,
5982                 struct extent_record *tmpl)
5983 {
5984         struct extent_record *rec;
5985         int ret = 0;
5986
5987         rec = malloc(sizeof(*rec));
5988         if (!rec)
5989                 return -ENOMEM;
5990         rec->start = tmpl->start;
5991         rec->max_size = tmpl->max_size;
5992         rec->nr = max(tmpl->nr, tmpl->max_size);
5993         rec->found_rec = tmpl->found_rec;
5994         rec->content_checked = tmpl->content_checked;
5995         rec->owner_ref_checked = tmpl->owner_ref_checked;
5996         rec->num_duplicates = 0;
5997         rec->metadata = tmpl->metadata;
5998         rec->flag_block_full_backref = FLAG_UNSET;
5999         rec->bad_full_backref = 0;
6000         rec->crossing_stripes = 0;
6001         rec->wrong_chunk_type = 0;
6002         rec->is_root = tmpl->is_root;
6003         rec->refs = tmpl->refs;
6004         rec->extent_item_refs = tmpl->extent_item_refs;
6005         rec->parent_generation = tmpl->parent_generation;
6006         INIT_LIST_HEAD(&rec->backrefs);
6007         INIT_LIST_HEAD(&rec->dups);
6008         INIT_LIST_HEAD(&rec->list);
6009         memcpy(&rec->parent_key, &tmpl->parent_key, sizeof(tmpl->parent_key));
6010         rec->cache.start = tmpl->start;
6011         rec->cache.size = tmpl->nr;
6012         ret = insert_cache_extent(extent_cache, &rec->cache);
6013         if (ret) {
6014                 free(rec);
6015                 return ret;
6016         }
6017         bytes_used += rec->nr;
6018
6019         if (tmpl->metadata)
6020                 rec->crossing_stripes = check_crossing_stripes(global_info,
6021                                 rec->start, global_info->tree_root->nodesize);
6022         check_extent_type(rec);
6023         return ret;
6024 }
6025
6026 /*
6027  * Lookup and modify an extent, some values of @tmpl are interpreted verbatim,
6028  * some are hints:
6029  * - refs              - if found, increase refs
6030  * - is_root           - if found, set
6031  * - content_checked   - if found, set
6032  * - owner_ref_checked - if found, set
6033  *
6034  * If not found, create a new one, initialize and insert.
6035  */
6036 static int add_extent_rec(struct cache_tree *extent_cache,
6037                 struct extent_record *tmpl)
6038 {
6039         struct extent_record *rec;
6040         struct cache_extent *cache;
6041         int ret = 0;
6042         int dup = 0;
6043
6044         cache = lookup_cache_extent(extent_cache, tmpl->start, tmpl->nr);
6045         if (cache) {
6046                 rec = container_of(cache, struct extent_record, cache);
6047                 if (tmpl->refs)
6048                         rec->refs++;
6049                 if (rec->nr == 1)
6050                         rec->nr = max(tmpl->nr, tmpl->max_size);
6051
6052                 /*
6053                  * We need to make sure to reset nr to whatever the extent
6054                  * record says was the real size, this way we can compare it to
6055                  * the backrefs.
6056                  */
6057                 if (tmpl->found_rec) {
6058                         if (tmpl->start != rec->start || rec->found_rec) {
6059                                 struct extent_record *tmp;
6060
6061                                 dup = 1;
6062                                 if (list_empty(&rec->list))
6063                                         list_add_tail(&rec->list,
6064                                                       &duplicate_extents);
6065
6066                                 /*
6067                                  * We have to do this song and dance in case we
6068                                  * find an extent record that falls inside of
6069                                  * our current extent record but does not have
6070                                  * the same objectid.
6071                                  */
6072                                 tmp = malloc(sizeof(*tmp));
6073                                 if (!tmp)
6074                                         return -ENOMEM;
6075                                 tmp->start = tmpl->start;
6076                                 tmp->max_size = tmpl->max_size;
6077                                 tmp->nr = tmpl->nr;
6078                                 tmp->found_rec = 1;
6079                                 tmp->metadata = tmpl->metadata;
6080                                 tmp->extent_item_refs = tmpl->extent_item_refs;
6081                                 INIT_LIST_HEAD(&tmp->list);
6082                                 list_add_tail(&tmp->list, &rec->dups);
6083                                 rec->num_duplicates++;
6084                         } else {
6085                                 rec->nr = tmpl->nr;
6086                                 rec->found_rec = 1;
6087                         }
6088                 }
6089
6090                 if (tmpl->extent_item_refs && !dup) {
6091                         if (rec->extent_item_refs) {
6092                                 fprintf(stderr, "block %llu rec "
6093                                         "extent_item_refs %llu, passed %llu\n",
6094                                         (unsigned long long)tmpl->start,
6095                                         (unsigned long long)
6096                                                         rec->extent_item_refs,
6097                                         (unsigned long long)tmpl->extent_item_refs);
6098                         }
6099                         rec->extent_item_refs = tmpl->extent_item_refs;
6100                 }
6101                 if (tmpl->is_root)
6102                         rec->is_root = 1;
6103                 if (tmpl->content_checked)
6104                         rec->content_checked = 1;
6105                 if (tmpl->owner_ref_checked)
6106                         rec->owner_ref_checked = 1;
6107                 memcpy(&rec->parent_key, &tmpl->parent_key,
6108                                 sizeof(tmpl->parent_key));
6109                 if (tmpl->parent_generation)
6110                         rec->parent_generation = tmpl->parent_generation;
6111                 if (rec->max_size < tmpl->max_size)
6112                         rec->max_size = tmpl->max_size;
6113
6114                 /*
6115                  * A metadata extent can't cross stripe_len boundary, otherwise
6116                  * kernel scrub won't be able to handle it.
6117                  * As now stripe_len is fixed to BTRFS_STRIPE_LEN, just check
6118                  * it.
6119                  */
6120                 if (tmpl->metadata)
6121                         rec->crossing_stripes = check_crossing_stripes(
6122                                         global_info, rec->start,
6123                                         global_info->tree_root->nodesize);
6124                 check_extent_type(rec);
6125                 maybe_free_extent_rec(extent_cache, rec);
6126                 return ret;
6127         }
6128
6129         ret = add_extent_rec_nolookup(extent_cache, tmpl);
6130
6131         return ret;
6132 }
6133
6134 static int add_tree_backref(struct cache_tree *extent_cache, u64 bytenr,
6135                             u64 parent, u64 root, int found_ref)
6136 {
6137         struct extent_record *rec;
6138         struct tree_backref *back;
6139         struct cache_extent *cache;
6140         int ret;
6141
6142         cache = lookup_cache_extent(extent_cache, bytenr, 1);
6143         if (!cache) {
6144                 struct extent_record tmpl;
6145
6146                 memset(&tmpl, 0, sizeof(tmpl));
6147                 tmpl.start = bytenr;
6148                 tmpl.nr = 1;
6149                 tmpl.metadata = 1;
6150
6151                 ret = add_extent_rec_nolookup(extent_cache, &tmpl);
6152                 if (ret)
6153                         return ret;
6154
6155                 /* really a bug in cache_extent implement now */
6156                 cache = lookup_cache_extent(extent_cache, bytenr, 1);
6157                 if (!cache)
6158                         return -ENOENT;
6159         }
6160
6161         rec = container_of(cache, struct extent_record, cache);
6162         if (rec->start != bytenr) {
6163                 /*
6164                  * Several cause, from unaligned bytenr to over lapping extents
6165                  */
6166                 return -EEXIST;
6167         }
6168
6169         back = find_tree_backref(rec, parent, root);
6170         if (!back) {
6171                 back = alloc_tree_backref(rec, parent, root);
6172                 if (!back)
6173                         return -ENOMEM;
6174         }
6175
6176         if (found_ref) {
6177                 if (back->node.found_ref) {
6178                         fprintf(stderr, "Extent back ref already exists "
6179                                 "for %llu parent %llu root %llu \n",
6180                                 (unsigned long long)bytenr,
6181                                 (unsigned long long)parent,
6182                                 (unsigned long long)root);
6183                 }
6184                 back->node.found_ref = 1;
6185         } else {
6186                 if (back->node.found_extent_tree) {
6187                         fprintf(stderr, "Extent back ref already exists "
6188                                 "for %llu parent %llu root %llu \n",
6189                                 (unsigned long long)bytenr,
6190                                 (unsigned long long)parent,
6191                                 (unsigned long long)root);
6192                 }
6193                 back->node.found_extent_tree = 1;
6194         }
6195         check_extent_type(rec);
6196         maybe_free_extent_rec(extent_cache, rec);
6197         return 0;
6198 }
6199
6200 static int add_data_backref(struct cache_tree *extent_cache, u64 bytenr,
6201                             u64 parent, u64 root, u64 owner, u64 offset,
6202                             u32 num_refs, int found_ref, u64 max_size)
6203 {
6204         struct extent_record *rec;
6205         struct data_backref *back;
6206         struct cache_extent *cache;
6207         int ret;
6208
6209         cache = lookup_cache_extent(extent_cache, bytenr, 1);
6210         if (!cache) {
6211                 struct extent_record tmpl;
6212
6213                 memset(&tmpl, 0, sizeof(tmpl));
6214                 tmpl.start = bytenr;
6215                 tmpl.nr = 1;
6216                 tmpl.max_size = max_size;
6217
6218                 ret = add_extent_rec_nolookup(extent_cache, &tmpl);
6219                 if (ret)
6220                         return ret;
6221
6222                 cache = lookup_cache_extent(extent_cache, bytenr, 1);
6223                 if (!cache)
6224                         abort();
6225         }
6226
6227         rec = container_of(cache, struct extent_record, cache);
6228         if (rec->max_size < max_size)
6229                 rec->max_size = max_size;
6230
6231         /*
6232          * If found_ref is set then max_size is the real size and must match the
6233          * existing refs.  So if we have already found a ref then we need to
6234          * make sure that this ref matches the existing one, otherwise we need
6235          * to add a new backref so we can notice that the backrefs don't match
6236          * and we need to figure out who is telling the truth.  This is to
6237          * account for that awful fsync bug I introduced where we'd end up with
6238          * a btrfs_file_extent_item that would have its length include multiple
6239          * prealloc extents or point inside of a prealloc extent.
6240          */
6241         back = find_data_backref(rec, parent, root, owner, offset, found_ref,
6242                                  bytenr, max_size);
6243         if (!back) {
6244                 back = alloc_data_backref(rec, parent, root, owner, offset,
6245                                           max_size);
6246                 BUG_ON(!back);
6247         }
6248
6249         if (found_ref) {
6250                 BUG_ON(num_refs != 1);
6251                 if (back->node.found_ref)
6252                         BUG_ON(back->bytes != max_size);
6253                 back->node.found_ref = 1;
6254                 back->found_ref += 1;
6255                 back->bytes = max_size;
6256                 back->disk_bytenr = bytenr;
6257                 rec->refs += 1;
6258                 rec->content_checked = 1;
6259                 rec->owner_ref_checked = 1;
6260         } else {
6261                 if (back->node.found_extent_tree) {
6262                         fprintf(stderr, "Extent back ref already exists "
6263                                 "for %llu parent %llu root %llu "
6264                                 "owner %llu offset %llu num_refs %lu\n",
6265                                 (unsigned long long)bytenr,
6266                                 (unsigned long long)parent,
6267                                 (unsigned long long)root,
6268                                 (unsigned long long)owner,
6269                                 (unsigned long long)offset,
6270                                 (unsigned long)num_refs);
6271                 }
6272                 back->num_refs = num_refs;
6273                 back->node.found_extent_tree = 1;
6274         }
6275         maybe_free_extent_rec(extent_cache, rec);
6276         return 0;
6277 }
6278
6279 static int add_pending(struct cache_tree *pending,
6280                        struct cache_tree *seen, u64 bytenr, u32 size)
6281 {
6282         int ret;
6283         ret = add_cache_extent(seen, bytenr, size);
6284         if (ret)
6285                 return ret;
6286         add_cache_extent(pending, bytenr, size);
6287         return 0;
6288 }
6289
6290 static int pick_next_pending(struct cache_tree *pending,
6291                         struct cache_tree *reada,
6292                         struct cache_tree *nodes,
6293                         u64 last, struct block_info *bits, int bits_nr,
6294                         int *reada_bits)
6295 {
6296         unsigned long node_start = last;
6297         struct cache_extent *cache;
6298         int ret;
6299
6300         cache = search_cache_extent(reada, 0);
6301         if (cache) {
6302                 bits[0].start = cache->start;
6303                 bits[0].size = cache->size;
6304                 *reada_bits = 1;
6305                 return 1;
6306         }
6307         *reada_bits = 0;
6308         if (node_start > 32768)
6309                 node_start -= 32768;
6310
6311         cache = search_cache_extent(nodes, node_start);
6312         if (!cache)
6313                 cache = search_cache_extent(nodes, 0);
6314
6315         if (!cache) {
6316                  cache = search_cache_extent(pending, 0);
6317                  if (!cache)
6318                          return 0;
6319                  ret = 0;
6320                  do {
6321                          bits[ret].start = cache->start;
6322                          bits[ret].size = cache->size;
6323                          cache = next_cache_extent(cache);
6324                          ret++;
6325                  } while (cache && ret < bits_nr);
6326                  return ret;
6327         }
6328
6329         ret = 0;
6330         do {
6331                 bits[ret].start = cache->start;
6332                 bits[ret].size = cache->size;
6333                 cache = next_cache_extent(cache);
6334                 ret++;
6335         } while (cache && ret < bits_nr);
6336
6337         if (bits_nr - ret > 8) {
6338                 u64 lookup = bits[0].start + bits[0].size;
6339                 struct cache_extent *next;
6340                 next = search_cache_extent(pending, lookup);
6341                 while(next) {
6342                         if (next->start - lookup > 32768)
6343                                 break;
6344                         bits[ret].start = next->start;
6345                         bits[ret].size = next->size;
6346                         lookup = next->start + next->size;
6347                         ret++;
6348                         if (ret == bits_nr)
6349                                 break;
6350                         next = next_cache_extent(next);
6351                         if (!next)
6352                                 break;
6353                 }
6354         }
6355         return ret;
6356 }
6357
6358 static void free_chunk_record(struct cache_extent *cache)
6359 {
6360         struct chunk_record *rec;
6361
6362         rec = container_of(cache, struct chunk_record, cache);
6363         list_del_init(&rec->list);
6364         list_del_init(&rec->dextents);
6365         free(rec);
6366 }
6367
6368 void free_chunk_cache_tree(struct cache_tree *chunk_cache)
6369 {
6370         cache_tree_free_extents(chunk_cache, free_chunk_record);
6371 }
6372
6373 static void free_device_record(struct rb_node *node)
6374 {
6375         struct device_record *rec;
6376
6377         rec = container_of(node, struct device_record, node);
6378         free(rec);
6379 }
6380
6381 FREE_RB_BASED_TREE(device_cache, free_device_record);
6382
6383 int insert_block_group_record(struct block_group_tree *tree,
6384                               struct block_group_record *bg_rec)
6385 {
6386         int ret;
6387
6388         ret = insert_cache_extent(&tree->tree, &bg_rec->cache);
6389         if (ret)
6390                 return ret;
6391
6392         list_add_tail(&bg_rec->list, &tree->block_groups);
6393         return 0;
6394 }
6395
6396 static void free_block_group_record(struct cache_extent *cache)
6397 {
6398         struct block_group_record *rec;
6399
6400         rec = container_of(cache, struct block_group_record, cache);
6401         list_del_init(&rec->list);
6402         free(rec);
6403 }
6404
6405 void free_block_group_tree(struct block_group_tree *tree)
6406 {
6407         cache_tree_free_extents(&tree->tree, free_block_group_record);
6408 }
6409
6410 int insert_device_extent_record(struct device_extent_tree *tree,
6411                                 struct device_extent_record *de_rec)
6412 {
6413         int ret;
6414
6415         /*
6416          * Device extent is a bit different from the other extents, because
6417          * the extents which belong to the different devices may have the
6418          * same start and size, so we need use the special extent cache
6419          * search/insert functions.
6420          */
6421         ret = insert_cache_extent2(&tree->tree, &de_rec->cache);
6422         if (ret)
6423                 return ret;
6424
6425         list_add_tail(&de_rec->chunk_list, &tree->no_chunk_orphans);
6426         list_add_tail(&de_rec->device_list, &tree->no_device_orphans);
6427         return 0;
6428 }
6429
6430 static void free_device_extent_record(struct cache_extent *cache)
6431 {
6432         struct device_extent_record *rec;
6433
6434         rec = container_of(cache, struct device_extent_record, cache);
6435         if (!list_empty(&rec->chunk_list))
6436                 list_del_init(&rec->chunk_list);
6437         if (!list_empty(&rec->device_list))
6438                 list_del_init(&rec->device_list);
6439         free(rec);
6440 }
6441
6442 void free_device_extent_tree(struct device_extent_tree *tree)
6443 {
6444         cache_tree_free_extents(&tree->tree, free_device_extent_record);
6445 }
6446
6447 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6448 static int process_extent_ref_v0(struct cache_tree *extent_cache,
6449                                  struct extent_buffer *leaf, int slot)
6450 {
6451         struct btrfs_extent_ref_v0 *ref0;
6452         struct btrfs_key key;
6453         int ret;
6454
6455         btrfs_item_key_to_cpu(leaf, &key, slot);
6456         ref0 = btrfs_item_ptr(leaf, slot, struct btrfs_extent_ref_v0);
6457         if (btrfs_ref_objectid_v0(leaf, ref0) < BTRFS_FIRST_FREE_OBJECTID) {
6458                 ret = add_tree_backref(extent_cache, key.objectid, key.offset,
6459                                 0, 0);
6460         } else {
6461                 ret = add_data_backref(extent_cache, key.objectid, key.offset,
6462                                 0, 0, 0, btrfs_ref_count_v0(leaf, ref0), 0, 0);
6463         }
6464         return ret;
6465 }
6466 #endif
6467
6468 struct chunk_record *btrfs_new_chunk_record(struct extent_buffer *leaf,
6469                                             struct btrfs_key *key,
6470                                             int slot)
6471 {
6472         struct btrfs_chunk *ptr;
6473         struct chunk_record *rec;
6474         int num_stripes, i;
6475
6476         ptr = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6477         num_stripes = btrfs_chunk_num_stripes(leaf, ptr);
6478
6479         rec = calloc(1, btrfs_chunk_record_size(num_stripes));
6480         if (!rec) {
6481                 fprintf(stderr, "memory allocation failed\n");
6482                 exit(-1);
6483         }
6484
6485         INIT_LIST_HEAD(&rec->list);
6486         INIT_LIST_HEAD(&rec->dextents);
6487         rec->bg_rec = NULL;
6488
6489         rec->cache.start = key->offset;
6490         rec->cache.size = btrfs_chunk_length(leaf, ptr);
6491
6492         rec->generation = btrfs_header_generation(leaf);
6493
6494         rec->objectid = key->objectid;
6495         rec->type = key->type;
6496         rec->offset = key->offset;
6497
6498         rec->length = rec->cache.size;
6499         rec->owner = btrfs_chunk_owner(leaf, ptr);
6500         rec->stripe_len = btrfs_chunk_stripe_len(leaf, ptr);
6501         rec->type_flags = btrfs_chunk_type(leaf, ptr);
6502         rec->io_width = btrfs_chunk_io_width(leaf, ptr);
6503         rec->io_align = btrfs_chunk_io_align(leaf, ptr);
6504         rec->sector_size = btrfs_chunk_sector_size(leaf, ptr);
6505         rec->num_stripes = num_stripes;
6506         rec->sub_stripes = btrfs_chunk_sub_stripes(leaf, ptr);
6507
6508         for (i = 0; i < rec->num_stripes; ++i) {
6509                 rec->stripes[i].devid =
6510                         btrfs_stripe_devid_nr(leaf, ptr, i);
6511                 rec->stripes[i].offset =
6512                         btrfs_stripe_offset_nr(leaf, ptr, i);
6513                 read_extent_buffer(leaf, rec->stripes[i].dev_uuid,
6514                                 (unsigned long)btrfs_stripe_dev_uuid_nr(ptr, i),
6515                                 BTRFS_UUID_SIZE);
6516         }
6517
6518         return rec;
6519 }
6520
6521 static int process_chunk_item(struct cache_tree *chunk_cache,
6522                               struct btrfs_key *key, struct extent_buffer *eb,
6523                               int slot)
6524 {
6525         struct chunk_record *rec;
6526         struct btrfs_chunk *chunk;
6527         int ret = 0;
6528
6529         chunk = btrfs_item_ptr(eb, slot, struct btrfs_chunk);
6530         /*
6531          * Do extra check for this chunk item,
6532          *
6533          * It's still possible one can craft a leaf with CHUNK_ITEM, with
6534          * wrong onwer(3) out of chunk tree, to pass both chunk tree check
6535          * and owner<->key_type check.
6536          */
6537         ret = btrfs_check_chunk_valid(global_info->tree_root, eb, chunk, slot,
6538                                       key->offset);
6539         if (ret < 0) {
6540                 error("chunk(%llu, %llu) is not valid, ignore it",
6541                       key->offset, btrfs_chunk_length(eb, chunk));
6542                 return 0;
6543         }
6544         rec = btrfs_new_chunk_record(eb, key, slot);
6545         ret = insert_cache_extent(chunk_cache, &rec->cache);
6546         if (ret) {
6547                 fprintf(stderr, "Chunk[%llu, %llu] existed.\n",
6548                         rec->offset, rec->length);
6549                 free(rec);
6550         }
6551
6552         return ret;
6553 }
6554
6555 static int process_device_item(struct rb_root *dev_cache,
6556                 struct btrfs_key *key, struct extent_buffer *eb, int slot)
6557 {
6558         struct btrfs_dev_item *ptr;
6559         struct device_record *rec;
6560         int ret = 0;
6561
6562         ptr = btrfs_item_ptr(eb,
6563                 slot, struct btrfs_dev_item);
6564
6565         rec = malloc(sizeof(*rec));
6566         if (!rec) {
6567                 fprintf(stderr, "memory allocation failed\n");
6568                 return -ENOMEM;
6569         }
6570
6571         rec->devid = key->offset;
6572         rec->generation = btrfs_header_generation(eb);
6573
6574         rec->objectid = key->objectid;
6575         rec->type = key->type;
6576         rec->offset = key->offset;
6577
6578         rec->devid = btrfs_device_id(eb, ptr);
6579         rec->total_byte = btrfs_device_total_bytes(eb, ptr);
6580         rec->byte_used = btrfs_device_bytes_used(eb, ptr);
6581
6582         ret = rb_insert(dev_cache, &rec->node, device_record_compare);
6583         if (ret) {
6584                 fprintf(stderr, "Device[%llu] existed.\n", rec->devid);
6585                 free(rec);
6586         }
6587
6588         return ret;
6589 }
6590
6591 struct block_group_record *
6592 btrfs_new_block_group_record(struct extent_buffer *leaf, struct btrfs_key *key,
6593                              int slot)
6594 {
6595         struct btrfs_block_group_item *ptr;
6596         struct block_group_record *rec;
6597
6598         rec = calloc(1, sizeof(*rec));
6599         if (!rec) {
6600                 fprintf(stderr, "memory allocation failed\n");
6601                 exit(-1);
6602         }
6603
6604         rec->cache.start = key->objectid;
6605         rec->cache.size = key->offset;
6606
6607         rec->generation = btrfs_header_generation(leaf);
6608
6609         rec->objectid = key->objectid;
6610         rec->type = key->type;
6611         rec->offset = key->offset;
6612
6613         ptr = btrfs_item_ptr(leaf, slot, struct btrfs_block_group_item);
6614         rec->flags = btrfs_disk_block_group_flags(leaf, ptr);
6615
6616         INIT_LIST_HEAD(&rec->list);
6617
6618         return rec;
6619 }
6620
6621 static int process_block_group_item(struct block_group_tree *block_group_cache,
6622                                     struct btrfs_key *key,
6623                                     struct extent_buffer *eb, int slot)
6624 {
6625         struct block_group_record *rec;
6626         int ret = 0;
6627
6628         rec = btrfs_new_block_group_record(eb, key, slot);
6629         ret = insert_block_group_record(block_group_cache, rec);
6630         if (ret) {
6631                 fprintf(stderr, "Block Group[%llu, %llu] existed.\n",
6632                         rec->objectid, rec->offset);
6633                 free(rec);
6634         }
6635
6636         return ret;
6637 }
6638
6639 struct device_extent_record *
6640 btrfs_new_device_extent_record(struct extent_buffer *leaf,
6641                                struct btrfs_key *key, int slot)
6642 {
6643         struct device_extent_record *rec;
6644         struct btrfs_dev_extent *ptr;
6645
6646         rec = calloc(1, sizeof(*rec));
6647         if (!rec) {
6648                 fprintf(stderr, "memory allocation failed\n");
6649                 exit(-1);
6650         }
6651
6652         rec->cache.objectid = key->objectid;
6653         rec->cache.start = key->offset;
6654
6655         rec->generation = btrfs_header_generation(leaf);
6656
6657         rec->objectid = key->objectid;
6658         rec->type = key->type;
6659         rec->offset = key->offset;
6660
6661         ptr = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
6662         rec->chunk_objecteid =
6663                 btrfs_dev_extent_chunk_objectid(leaf, ptr);
6664         rec->chunk_offset =
6665                 btrfs_dev_extent_chunk_offset(leaf, ptr);
6666         rec->length = btrfs_dev_extent_length(leaf, ptr);
6667         rec->cache.size = rec->length;
6668
6669         INIT_LIST_HEAD(&rec->chunk_list);
6670         INIT_LIST_HEAD(&rec->device_list);
6671
6672         return rec;
6673 }
6674
6675 static int
6676 process_device_extent_item(struct device_extent_tree *dev_extent_cache,
6677                            struct btrfs_key *key, struct extent_buffer *eb,
6678                            int slot)
6679 {
6680         struct device_extent_record *rec;
6681         int ret;
6682
6683         rec = btrfs_new_device_extent_record(eb, key, slot);
6684         ret = insert_device_extent_record(dev_extent_cache, rec);
6685         if (ret) {
6686                 fprintf(stderr,
6687                         "Device extent[%llu, %llu, %llu] existed.\n",
6688                         rec->objectid, rec->offset, rec->length);
6689                 free(rec);
6690         }
6691
6692         return ret;
6693 }
6694
6695 static int process_extent_item(struct btrfs_root *root,
6696                                struct cache_tree *extent_cache,
6697                                struct extent_buffer *eb, int slot)
6698 {
6699         struct btrfs_extent_item *ei;
6700         struct btrfs_extent_inline_ref *iref;
6701         struct btrfs_extent_data_ref *dref;
6702         struct btrfs_shared_data_ref *sref;
6703         struct btrfs_key key;
6704         struct extent_record tmpl;
6705         unsigned long end;
6706         unsigned long ptr;
6707         int ret;
6708         int type;
6709         u32 item_size = btrfs_item_size_nr(eb, slot);
6710         u64 refs = 0;
6711         u64 offset;
6712         u64 num_bytes;
6713         int metadata = 0;
6714
6715         btrfs_item_key_to_cpu(eb, &key, slot);
6716
6717         if (key.type == BTRFS_METADATA_ITEM_KEY) {
6718                 metadata = 1;
6719                 num_bytes = root->nodesize;
6720         } else {
6721                 num_bytes = key.offset;
6722         }
6723
6724         if (!IS_ALIGNED(key.objectid, root->sectorsize)) {
6725                 error("ignoring invalid extent, bytenr %llu is not aligned to %u",
6726                       key.objectid, root->sectorsize);
6727                 return -EIO;
6728         }
6729         if (item_size < sizeof(*ei)) {
6730 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6731                 struct btrfs_extent_item_v0 *ei0;
6732                 BUG_ON(item_size != sizeof(*ei0));
6733                 ei0 = btrfs_item_ptr(eb, slot, struct btrfs_extent_item_v0);
6734                 refs = btrfs_extent_refs_v0(eb, ei0);
6735 #else
6736                 BUG();
6737 #endif
6738                 memset(&tmpl, 0, sizeof(tmpl));
6739                 tmpl.start = key.objectid;
6740                 tmpl.nr = num_bytes;
6741                 tmpl.extent_item_refs = refs;
6742                 tmpl.metadata = metadata;
6743                 tmpl.found_rec = 1;
6744                 tmpl.max_size = num_bytes;
6745
6746                 return add_extent_rec(extent_cache, &tmpl);
6747         }
6748
6749         ei = btrfs_item_ptr(eb, slot, struct btrfs_extent_item);
6750         refs = btrfs_extent_refs(eb, ei);
6751         if (btrfs_extent_flags(eb, ei) & BTRFS_EXTENT_FLAG_TREE_BLOCK)
6752                 metadata = 1;
6753         else
6754                 metadata = 0;
6755         if (metadata && num_bytes != root->nodesize) {
6756                 error("ignore invalid metadata extent, length %llu does not equal to %u",
6757                       num_bytes, root->nodesize);
6758                 return -EIO;
6759         }
6760         if (!metadata && !IS_ALIGNED(num_bytes, root->sectorsize)) {
6761                 error("ignore invalid data extent, length %llu is not aligned to %u",
6762                       num_bytes, root->sectorsize);
6763                 return -EIO;
6764         }
6765
6766         memset(&tmpl, 0, sizeof(tmpl));
6767         tmpl.start = key.objectid;
6768         tmpl.nr = num_bytes;
6769         tmpl.extent_item_refs = refs;
6770         tmpl.metadata = metadata;
6771         tmpl.found_rec = 1;
6772         tmpl.max_size = num_bytes;
6773         add_extent_rec(extent_cache, &tmpl);
6774
6775         ptr = (unsigned long)(ei + 1);
6776         if (btrfs_extent_flags(eb, ei) & BTRFS_EXTENT_FLAG_TREE_BLOCK &&
6777             key.type == BTRFS_EXTENT_ITEM_KEY)
6778                 ptr += sizeof(struct btrfs_tree_block_info);
6779
6780         end = (unsigned long)ei + item_size;
6781         while (ptr < end) {
6782                 iref = (struct btrfs_extent_inline_ref *)ptr;
6783                 type = btrfs_extent_inline_ref_type(eb, iref);
6784                 offset = btrfs_extent_inline_ref_offset(eb, iref);
6785                 switch (type) {
6786                 case BTRFS_TREE_BLOCK_REF_KEY:
6787                         ret = add_tree_backref(extent_cache, key.objectid,
6788                                         0, offset, 0);
6789                         if (ret < 0)
6790                                 error("add_tree_backref failed: %s",
6791                                       strerror(-ret));
6792                         break;
6793                 case BTRFS_SHARED_BLOCK_REF_KEY:
6794                         ret = add_tree_backref(extent_cache, key.objectid,
6795                                         offset, 0, 0);
6796                         if (ret < 0)
6797                                 error("add_tree_backref failed: %s",
6798                                       strerror(-ret));
6799                         break;
6800                 case BTRFS_EXTENT_DATA_REF_KEY:
6801                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
6802                         add_data_backref(extent_cache, key.objectid, 0,
6803                                         btrfs_extent_data_ref_root(eb, dref),
6804                                         btrfs_extent_data_ref_objectid(eb,
6805                                                                        dref),
6806                                         btrfs_extent_data_ref_offset(eb, dref),
6807                                         btrfs_extent_data_ref_count(eb, dref),
6808                                         0, num_bytes);
6809                         break;
6810                 case BTRFS_SHARED_DATA_REF_KEY:
6811                         sref = (struct btrfs_shared_data_ref *)(iref + 1);
6812                         add_data_backref(extent_cache, key.objectid, offset,
6813                                         0, 0, 0,
6814                                         btrfs_shared_data_ref_count(eb, sref),
6815                                         0, num_bytes);
6816                         break;
6817                 default:
6818                         fprintf(stderr, "corrupt extent record: key %Lu %u %Lu\n",
6819                                 key.objectid, key.type, num_bytes);
6820                         goto out;
6821                 }
6822                 ptr += btrfs_extent_inline_ref_size(type);
6823         }
6824         WARN_ON(ptr > end);
6825 out:
6826         return 0;
6827 }
6828
6829 static int check_cache_range(struct btrfs_root *root,
6830                              struct btrfs_block_group_cache *cache,
6831                              u64 offset, u64 bytes)
6832 {
6833         struct btrfs_free_space *entry;
6834         u64 *logical;
6835         u64 bytenr;
6836         int stripe_len;
6837         int i, nr, ret;
6838
6839         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
6840                 bytenr = btrfs_sb_offset(i);
6841                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
6842                                        cache->key.objectid, bytenr, 0,
6843                                        &logical, &nr, &stripe_len);
6844                 if (ret)
6845                         return ret;
6846
6847                 while (nr--) {
6848                         if (logical[nr] + stripe_len <= offset)
6849                                 continue;
6850                         if (offset + bytes <= logical[nr])
6851                                 continue;
6852                         if (logical[nr] == offset) {
6853                                 if (stripe_len >= bytes) {
6854                                         free(logical);
6855                                         return 0;
6856                                 }
6857                                 bytes -= stripe_len;
6858                                 offset += stripe_len;
6859                         } else if (logical[nr] < offset) {
6860                                 if (logical[nr] + stripe_len >=
6861                                     offset + bytes) {
6862                                         free(logical);
6863                                         return 0;
6864                                 }
6865                                 bytes = (offset + bytes) -
6866                                         (logical[nr] + stripe_len);
6867                                 offset = logical[nr] + stripe_len;
6868                         } else {
6869                                 /*
6870                                  * Could be tricky, the super may land in the
6871                                  * middle of the area we're checking.  First
6872                                  * check the easiest case, it's at the end.
6873                                  */
6874                                 if (logical[nr] + stripe_len >=
6875                                     bytes + offset) {
6876                                         bytes = logical[nr] - offset;
6877                                         continue;
6878                                 }
6879
6880                                 /* Check the left side */
6881                                 ret = check_cache_range(root, cache,
6882                                                         offset,
6883                                                         logical[nr] - offset);
6884                                 if (ret) {
6885                                         free(logical);
6886                                         return ret;
6887                                 }
6888
6889                                 /* Now we continue with the right side */
6890                                 bytes = (offset + bytes) -
6891                                         (logical[nr] + stripe_len);
6892                                 offset = logical[nr] + stripe_len;
6893                         }
6894                 }
6895
6896                 free(logical);
6897         }
6898
6899         entry = btrfs_find_free_space(cache->free_space_ctl, offset, bytes);
6900         if (!entry) {
6901                 fprintf(stderr, "There is no free space entry for %Lu-%Lu\n",
6902                         offset, offset+bytes);
6903                 return -EINVAL;
6904         }
6905
6906         if (entry->offset != offset) {
6907                 fprintf(stderr, "Wanted offset %Lu, found %Lu\n", offset,
6908                         entry->offset);
6909                 return -EINVAL;
6910         }
6911
6912         if (entry->bytes != bytes) {
6913                 fprintf(stderr, "Wanted bytes %Lu, found %Lu for off %Lu\n",
6914                         bytes, entry->bytes, offset);
6915                 return -EINVAL;
6916         }
6917
6918         unlink_free_space(cache->free_space_ctl, entry);
6919         free(entry);
6920         return 0;
6921 }
6922
6923 static int verify_space_cache(struct btrfs_root *root,
6924                               struct btrfs_block_group_cache *cache)
6925 {
6926         struct btrfs_path path;
6927         struct extent_buffer *leaf;
6928         struct btrfs_key key;
6929         u64 last;
6930         int ret = 0;
6931
6932         root = root->fs_info->extent_root;
6933
6934         last = max_t(u64, cache->key.objectid, BTRFS_SUPER_INFO_OFFSET);
6935
6936         btrfs_init_path(&path);
6937         key.objectid = last;
6938         key.offset = 0;
6939         key.type = BTRFS_EXTENT_ITEM_KEY;
6940         ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
6941         if (ret < 0)
6942                 goto out;
6943         ret = 0;
6944         while (1) {
6945                 if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
6946                         ret = btrfs_next_leaf(root, &path);
6947                         if (ret < 0)
6948                                 goto out;
6949                         if (ret > 0) {
6950                                 ret = 0;
6951                                 break;
6952                         }
6953                 }
6954                 leaf = path.nodes[0];
6955                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
6956                 if (key.objectid >= cache->key.offset + cache->key.objectid)
6957                         break;
6958                 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
6959                     key.type != BTRFS_METADATA_ITEM_KEY) {
6960                         path.slots[0]++;
6961                         continue;
6962                 }
6963
6964                 if (last == key.objectid) {
6965                         if (key.type == BTRFS_EXTENT_ITEM_KEY)
6966                                 last = key.objectid + key.offset;
6967                         else
6968                                 last = key.objectid + root->nodesize;
6969                         path.slots[0]++;
6970                         continue;
6971                 }
6972
6973                 ret = check_cache_range(root, cache, last,
6974                                         key.objectid - last);
6975                 if (ret)
6976                         break;
6977                 if (key.type == BTRFS_EXTENT_ITEM_KEY)
6978                         last = key.objectid + key.offset;
6979                 else
6980                         last = key.objectid + root->nodesize;
6981                 path.slots[0]++;
6982         }
6983
6984         if (last < cache->key.objectid + cache->key.offset)
6985                 ret = check_cache_range(root, cache, last,
6986                                         cache->key.objectid +
6987                                         cache->key.offset - last);
6988
6989 out:
6990         btrfs_release_path(&path);
6991
6992         if (!ret &&
6993             !RB_EMPTY_ROOT(&cache->free_space_ctl->free_space_offset)) {
6994                 fprintf(stderr, "There are still entries left in the space "
6995                         "cache\n");
6996                 ret = -EINVAL;
6997         }
6998
6999         return ret;
7000 }
7001
7002 static int check_space_cache(struct btrfs_root *root)
7003 {
7004         struct btrfs_block_group_cache *cache;
7005         u64 start = BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE;
7006         int ret;
7007         int error = 0;
7008
7009         if (btrfs_super_cache_generation(root->fs_info->super_copy) != -1ULL &&
7010             btrfs_super_generation(root->fs_info->super_copy) !=
7011             btrfs_super_cache_generation(root->fs_info->super_copy)) {
7012                 printf("cache and super generation don't match, space cache "
7013                        "will be invalidated\n");
7014                 return 0;
7015         }
7016
7017         if (ctx.progress_enabled) {
7018                 ctx.tp = TASK_FREE_SPACE;
7019                 task_start(ctx.info);
7020         }
7021
7022         while (1) {
7023                 cache = btrfs_lookup_first_block_group(root->fs_info, start);
7024                 if (!cache)
7025                         break;
7026
7027                 start = cache->key.objectid + cache->key.offset;
7028                 if (!cache->free_space_ctl) {
7029                         if (btrfs_init_free_space_ctl(cache,
7030                                                       root->sectorsize)) {
7031                                 ret = -ENOMEM;
7032                                 break;
7033                         }
7034                 } else {
7035                         btrfs_remove_free_space_cache(cache);
7036                 }
7037
7038                 if (btrfs_fs_compat_ro(root->fs_info,
7039                                        BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE)) {
7040                         ret = exclude_super_stripes(root, cache);
7041                         if (ret) {
7042                                 fprintf(stderr, "could not exclude super stripes: %s\n",
7043                                         strerror(-ret));
7044                                 error++;
7045                                 continue;
7046                         }
7047                         ret = load_free_space_tree(root->fs_info, cache);
7048                         free_excluded_extents(root, cache);
7049                         if (ret < 0) {
7050                                 fprintf(stderr, "could not load free space tree: %s\n",
7051                                         strerror(-ret));
7052                                 error++;
7053                                 continue;
7054                         }
7055                         error += ret;
7056                 } else {
7057                         ret = load_free_space_cache(root->fs_info, cache);
7058                         if (!ret)
7059                                 continue;
7060                 }
7061
7062                 ret = verify_space_cache(root, cache);
7063                 if (ret) {
7064                         fprintf(stderr, "cache appears valid but isn't %Lu\n",
7065                                 cache->key.objectid);
7066                         error++;
7067                 }
7068         }
7069
7070         task_stop(ctx.info);
7071
7072         return error ? -EINVAL : 0;
7073 }
7074
7075 static int check_extent_csums(struct btrfs_root *root, u64 bytenr,
7076                         u64 num_bytes, unsigned long leaf_offset,
7077                         struct extent_buffer *eb) {
7078
7079         u64 offset = 0;
7080         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
7081         char *data;
7082         unsigned long csum_offset;
7083         u32 csum;
7084         u32 csum_expected;
7085         u64 read_len;
7086         u64 data_checked = 0;
7087         u64 tmp;
7088         int ret = 0;
7089         int mirror;
7090         int num_copies;
7091
7092         if (num_bytes % root->sectorsize)
7093                 return -EINVAL;
7094
7095         data = malloc(num_bytes);
7096         if (!data)
7097                 return -ENOMEM;
7098
7099         while (offset < num_bytes) {
7100                 mirror = 0;
7101 again:
7102                 read_len = num_bytes - offset;
7103                 /* read as much space once a time */
7104                 ret = read_extent_data(root, data + offset,
7105                                 bytenr + offset, &read_len, mirror);
7106                 if (ret)
7107                         goto out;
7108                 data_checked = 0;
7109                 /* verify every 4k data's checksum */
7110                 while (data_checked < read_len) {
7111                         csum = ~(u32)0;
7112                         tmp = offset + data_checked;
7113
7114                         csum = btrfs_csum_data(NULL, (char *)data + tmp,
7115                                                csum, root->sectorsize);
7116                         btrfs_csum_final(csum, (u8 *)&csum);
7117
7118                         csum_offset = leaf_offset +
7119                                  tmp / root->sectorsize * csum_size;
7120                         read_extent_buffer(eb, (char *)&csum_expected,
7121                                            csum_offset, csum_size);
7122                         /* try another mirror */
7123                         if (csum != csum_expected) {
7124                                 fprintf(stderr, "mirror %d bytenr %llu csum %u expected csum %u\n",
7125                                                 mirror, bytenr + tmp,
7126                                                 csum, csum_expected);
7127                                 num_copies = btrfs_num_copies(
7128                                                 &root->fs_info->mapping_tree,
7129                                                 bytenr, num_bytes);
7130                                 if (mirror < num_copies - 1) {
7131                                         mirror += 1;
7132                                         goto again;
7133                                 }
7134                         }
7135                         data_checked += root->sectorsize;
7136                 }
7137                 offset += read_len;
7138         }
7139 out:
7140         free(data);
7141         return ret;
7142 }
7143
7144 static int check_extent_exists(struct btrfs_root *root, u64 bytenr,
7145                                u64 num_bytes)
7146 {
7147         struct btrfs_path path;
7148         struct extent_buffer *leaf;
7149         struct btrfs_key key;
7150         int ret;
7151
7152         btrfs_init_path(&path);
7153         key.objectid = bytenr;
7154         key.type = BTRFS_EXTENT_ITEM_KEY;
7155         key.offset = (u64)-1;
7156
7157 again:
7158         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, &path,
7159                                 0, 0);
7160         if (ret < 0) {
7161                 fprintf(stderr, "Error looking up extent record %d\n", ret);
7162                 btrfs_release_path(&path);
7163                 return ret;
7164         } else if (ret) {
7165                 if (path.slots[0] > 0) {
7166                         path.slots[0]--;
7167                 } else {
7168                         ret = btrfs_prev_leaf(root, &path);
7169                         if (ret < 0) {
7170                                 goto out;
7171                         } else if (ret > 0) {
7172                                 ret = 0;
7173                                 goto out;
7174                         }
7175                 }
7176         }
7177
7178         btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
7179
7180         /*
7181          * Block group items come before extent items if they have the same
7182          * bytenr, so walk back one more just in case.  Dear future traveller,
7183          * first congrats on mastering time travel.  Now if it's not too much
7184          * trouble could you go back to 2006 and tell Chris to make the
7185          * BLOCK_GROUP_ITEM_KEY (and BTRFS_*_REF_KEY) lower than the
7186          * EXTENT_ITEM_KEY please?
7187          */
7188         while (key.type > BTRFS_EXTENT_ITEM_KEY) {
7189                 if (path.slots[0] > 0) {
7190                         path.slots[0]--;
7191                 } else {
7192                         ret = btrfs_prev_leaf(root, &path);
7193                         if (ret < 0) {
7194                                 goto out;
7195                         } else if (ret > 0) {
7196                                 ret = 0;
7197                                 goto out;
7198                         }
7199                 }
7200                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
7201         }
7202
7203         while (num_bytes) {
7204                 if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
7205                         ret = btrfs_next_leaf(root, &path);
7206                         if (ret < 0) {
7207                                 fprintf(stderr, "Error going to next leaf "
7208                                         "%d\n", ret);
7209                                 btrfs_release_path(&path);
7210                                 return ret;
7211                         } else if (ret) {
7212                                 break;
7213                         }
7214                 }
7215                 leaf = path.nodes[0];
7216                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
7217                 if (key.type != BTRFS_EXTENT_ITEM_KEY) {
7218                         path.slots[0]++;
7219                         continue;
7220                 }
7221                 if (key.objectid + key.offset < bytenr) {
7222                         path.slots[0]++;
7223                         continue;
7224                 }
7225                 if (key.objectid > bytenr + num_bytes)
7226                         break;
7227
7228                 if (key.objectid == bytenr) {
7229                         if (key.offset >= num_bytes) {
7230                                 num_bytes = 0;
7231                                 break;
7232                         }
7233                         num_bytes -= key.offset;
7234                         bytenr += key.offset;
7235                 } else if (key.objectid < bytenr) {
7236                         if (key.objectid + key.offset >= bytenr + num_bytes) {
7237                                 num_bytes = 0;
7238                                 break;
7239                         }
7240                         num_bytes = (bytenr + num_bytes) -
7241                                 (key.objectid + key.offset);
7242                         bytenr = key.objectid + key.offset;
7243                 } else {
7244                         if (key.objectid + key.offset < bytenr + num_bytes) {
7245                                 u64 new_start = key.objectid + key.offset;
7246                                 u64 new_bytes = bytenr + num_bytes - new_start;
7247
7248                                 /*
7249                                  * Weird case, the extent is in the middle of
7250                                  * our range, we'll have to search one side
7251                                  * and then the other.  Not sure if this happens
7252                                  * in real life, but no harm in coding it up
7253                                  * anyway just in case.
7254                                  */
7255                                 btrfs_release_path(&path);
7256                                 ret = check_extent_exists(root, new_start,
7257                                                           new_bytes);
7258                                 if (ret) {
7259                                         fprintf(stderr, "Right section didn't "
7260                                                 "have a record\n");
7261                                         break;
7262                                 }
7263                                 num_bytes = key.objectid - bytenr;
7264                                 goto again;
7265                         }
7266                         num_bytes = key.objectid - bytenr;
7267                 }
7268                 path.slots[0]++;
7269         }
7270         ret = 0;
7271
7272 out:
7273         if (num_bytes && !ret) {
7274                 fprintf(stderr, "There are no extents for csum range "
7275                         "%Lu-%Lu\n", bytenr, bytenr+num_bytes);
7276                 ret = 1;
7277         }
7278
7279         btrfs_release_path(&path);
7280         return ret;
7281 }
7282
7283 static int check_csums(struct btrfs_root *root)
7284 {
7285         struct btrfs_path path;
7286         struct extent_buffer *leaf;
7287         struct btrfs_key key;
7288         u64 offset = 0, num_bytes = 0;
7289         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
7290         int errors = 0;
7291         int ret;
7292         u64 data_len;
7293         unsigned long leaf_offset;
7294
7295         root = root->fs_info->csum_root;
7296         if (!extent_buffer_uptodate(root->node)) {
7297                 fprintf(stderr, "No valid csum tree found\n");
7298                 return -ENOENT;
7299         }
7300
7301         btrfs_init_path(&path);
7302         key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
7303         key.type = BTRFS_EXTENT_CSUM_KEY;
7304         key.offset = 0;
7305         ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
7306         if (ret < 0) {
7307                 fprintf(stderr, "Error searching csum tree %d\n", ret);
7308                 btrfs_release_path(&path);
7309                 return ret;
7310         }
7311
7312         if (ret > 0 && path.slots[0])
7313                 path.slots[0]--;
7314         ret = 0;
7315
7316         while (1) {
7317                 if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
7318                         ret = btrfs_next_leaf(root, &path);
7319                         if (ret < 0) {
7320                                 fprintf(stderr, "Error going to next leaf "
7321                                         "%d\n", ret);
7322                                 break;
7323                         }
7324                         if (ret)
7325                                 break;
7326                 }
7327                 leaf = path.nodes[0];
7328
7329                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
7330                 if (key.type != BTRFS_EXTENT_CSUM_KEY) {
7331                         path.slots[0]++;
7332                         continue;
7333                 }
7334
7335                 data_len = (btrfs_item_size_nr(leaf, path.slots[0]) /
7336                               csum_size) * root->sectorsize;
7337                 if (!check_data_csum)
7338                         goto skip_csum_check;
7339                 leaf_offset = btrfs_item_ptr_offset(leaf, path.slots[0]);
7340                 ret = check_extent_csums(root, key.offset, data_len,
7341                                          leaf_offset, leaf);
7342                 if (ret)
7343                         break;
7344 skip_csum_check:
7345                 if (!num_bytes) {
7346                         offset = key.offset;
7347                 } else if (key.offset != offset + num_bytes) {
7348                         ret = check_extent_exists(root, offset, num_bytes);
7349                         if (ret) {
7350                                 fprintf(stderr, "Csum exists for %Lu-%Lu but "
7351                                         "there is no extent record\n",
7352                                         offset, offset+num_bytes);
7353                                 errors++;
7354                         }
7355                         offset = key.offset;
7356                         num_bytes = 0;
7357                 }
7358                 num_bytes += data_len;
7359                 path.slots[0]++;
7360         }
7361
7362         btrfs_release_path(&path);
7363         return errors;
7364 }
7365
7366 static int is_dropped_key(struct btrfs_key *key,
7367                           struct btrfs_key *drop_key) {
7368         if (key->objectid < drop_key->objectid)
7369                 return 1;
7370         else if (key->objectid == drop_key->objectid) {
7371                 if (key->type < drop_key->type)
7372                         return 1;
7373                 else if (key->type == drop_key->type) {
7374                         if (key->offset < drop_key->offset)
7375                                 return 1;
7376                 }
7377         }
7378         return 0;
7379 }
7380
7381 /*
7382  * Here are the rules for FULL_BACKREF.
7383  *
7384  * 1) If BTRFS_HEADER_FLAG_RELOC is set then we have FULL_BACKREF set.
7385  * 2) If btrfs_header_owner(buf) no longer points to buf then we have
7386  *      FULL_BACKREF set.
7387  * 3) We cowed the block walking down a reloc tree.  This is impossible to tell
7388  *    if it happened after the relocation occurred since we'll have dropped the
7389  *    reloc root, so it's entirely possible to have FULL_BACKREF set on buf and
7390  *    have no real way to know for sure.
7391  *
7392  * We process the blocks one root at a time, and we start from the lowest root
7393  * objectid and go to the highest.  So we can just lookup the owner backref for
7394  * the record and if we don't find it then we know it doesn't exist and we have
7395  * a FULL BACKREF.
7396  *
7397  * FIXME: if we ever start reclaiming root objectid's then we need to fix this
7398  * assumption and simply indicate that we _think_ that the FULL BACKREF needs to
7399  * be set or not and then we can check later once we've gathered all the refs.
7400  */
7401 static int calc_extent_flag(struct btrfs_root *root,
7402                            struct cache_tree *extent_cache,
7403                            struct extent_buffer *buf,
7404                            struct root_item_record *ri,
7405                            u64 *flags)
7406 {
7407         struct extent_record *rec;
7408         struct cache_extent *cache;
7409         struct tree_backref *tback;
7410         u64 owner = 0;
7411
7412         cache = lookup_cache_extent(extent_cache, buf->start, 1);
7413         /* we have added this extent before */
7414         if (!cache)
7415                 return -ENOENT;
7416
7417         rec = container_of(cache, struct extent_record, cache);
7418
7419         /*
7420          * Except file/reloc tree, we can not have
7421          * FULL BACKREF MODE
7422          */
7423         if (ri->objectid < BTRFS_FIRST_FREE_OBJECTID)
7424                 goto normal;
7425         /*
7426          * root node
7427          */
7428         if (buf->start == ri->bytenr)
7429                 goto normal;
7430
7431         if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))
7432                 goto full_backref;
7433
7434         owner = btrfs_header_owner(buf);
7435         if (owner == ri->objectid)
7436                 goto normal;
7437
7438         tback = find_tree_backref(rec, 0, owner);
7439         if (!tback)
7440                 goto full_backref;
7441 normal:
7442         *flags = 0;
7443         if (rec->flag_block_full_backref != FLAG_UNSET &&
7444             rec->flag_block_full_backref != 0)
7445                 rec->bad_full_backref = 1;
7446         return 0;
7447 full_backref:
7448         *flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7449         if (rec->flag_block_full_backref != FLAG_UNSET &&
7450             rec->flag_block_full_backref != 1)
7451                 rec->bad_full_backref = 1;
7452         return 0;
7453 }
7454
7455 static void report_mismatch_key_root(u8 key_type, u64 rootid)
7456 {
7457         fprintf(stderr, "Invalid key type(");
7458         print_key_type(stderr, 0, key_type);
7459         fprintf(stderr, ") found in root(");
7460         print_objectid(stderr, rootid, 0);
7461         fprintf(stderr, ")\n");
7462 }
7463
7464 /*
7465  * Check if the key is valid with its extent buffer.
7466  *
7467  * This is a early check in case invalid key exists in a extent buffer
7468  * This is not comprehensive yet, but should prevent wrong key/item passed
7469  * further
7470  */
7471 static int check_type_with_root(u64 rootid, u8 key_type)
7472 {
7473         switch (key_type) {
7474         /* Only valid in chunk tree */
7475         case BTRFS_DEV_ITEM_KEY:
7476         case BTRFS_CHUNK_ITEM_KEY:
7477                 if (rootid != BTRFS_CHUNK_TREE_OBJECTID)
7478                         goto err;
7479                 break;
7480         /* valid in csum and log tree */
7481         case BTRFS_CSUM_TREE_OBJECTID:
7482                 if (!(rootid == BTRFS_TREE_LOG_OBJECTID ||
7483                       is_fstree(rootid)))
7484                         goto err;
7485                 break;
7486         case BTRFS_EXTENT_ITEM_KEY:
7487         case BTRFS_METADATA_ITEM_KEY:
7488         case BTRFS_BLOCK_GROUP_ITEM_KEY:
7489                 if (rootid != BTRFS_EXTENT_TREE_OBJECTID)
7490                         goto err;
7491                 break;
7492         case BTRFS_ROOT_ITEM_KEY:
7493                 if (rootid != BTRFS_ROOT_TREE_OBJECTID)
7494                         goto err;
7495                 break;
7496         case BTRFS_DEV_EXTENT_KEY:
7497                 if (rootid != BTRFS_DEV_TREE_OBJECTID)
7498                         goto err;
7499                 break;
7500         }
7501         return 0;
7502 err:
7503         report_mismatch_key_root(key_type, rootid);
7504         return -EINVAL;
7505 }
7506
7507 static int run_next_block(struct btrfs_root *root,
7508                           struct block_info *bits,
7509                           int bits_nr,
7510                           u64 *last,
7511                           struct cache_tree *pending,
7512                           struct cache_tree *seen,
7513                           struct cache_tree *reada,
7514                           struct cache_tree *nodes,
7515                           struct cache_tree *extent_cache,
7516                           struct cache_tree *chunk_cache,
7517                           struct rb_root *dev_cache,
7518                           struct block_group_tree *block_group_cache,
7519                           struct device_extent_tree *dev_extent_cache,
7520                           struct root_item_record *ri)
7521 {
7522         struct extent_buffer *buf;
7523         struct extent_record *rec = NULL;
7524         u64 bytenr;
7525         u32 size;
7526         u64 parent;
7527         u64 owner;
7528         u64 flags;
7529         u64 ptr;
7530         u64 gen = 0;
7531         int ret = 0;
7532         int i;
7533         int nritems;
7534         struct btrfs_key key;
7535         struct cache_extent *cache;
7536         int reada_bits;
7537
7538         nritems = pick_next_pending(pending, reada, nodes, *last, bits,
7539                                     bits_nr, &reada_bits);
7540         if (nritems == 0)
7541                 return 1;
7542
7543         if (!reada_bits) {
7544                 for(i = 0; i < nritems; i++) {
7545                         ret = add_cache_extent(reada, bits[i].start,
7546                                                bits[i].size);
7547                         if (ret == -EEXIST)
7548                                 continue;
7549
7550                         /* fixme, get the parent transid */
7551                         readahead_tree_block(root, bits[i].start,
7552                                              bits[i].size, 0);
7553                 }
7554         }
7555         *last = bits[0].start;
7556         bytenr = bits[0].start;
7557         size = bits[0].size;
7558
7559         cache = lookup_cache_extent(pending, bytenr, size);
7560         if (cache) {
7561                 remove_cache_extent(pending, cache);
7562                 free(cache);
7563         }
7564         cache = lookup_cache_extent(reada, bytenr, size);
7565         if (cache) {
7566                 remove_cache_extent(reada, cache);
7567                 free(cache);
7568         }
7569         cache = lookup_cache_extent(nodes, bytenr, size);
7570         if (cache) {
7571                 remove_cache_extent(nodes, cache);
7572                 free(cache);
7573         }
7574         cache = lookup_cache_extent(extent_cache, bytenr, size);
7575         if (cache) {
7576                 rec = container_of(cache, struct extent_record, cache);
7577                 gen = rec->parent_generation;
7578         }
7579
7580         /* fixme, get the real parent transid */
7581         buf = read_tree_block(root, bytenr, size, gen);
7582         if (!extent_buffer_uptodate(buf)) {
7583                 record_bad_block_io(root->fs_info,
7584                                     extent_cache, bytenr, size);
7585                 goto out;
7586         }
7587
7588         nritems = btrfs_header_nritems(buf);
7589
7590         flags = 0;
7591         if (!init_extent_tree) {
7592                 ret = btrfs_lookup_extent_info(NULL, root, bytenr,
7593                                        btrfs_header_level(buf), 1, NULL,
7594                                        &flags);
7595                 if (ret < 0) {
7596                         ret = calc_extent_flag(root, extent_cache, buf, ri, &flags);
7597                         if (ret < 0) {
7598                                 fprintf(stderr, "Couldn't calc extent flags\n");
7599                                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7600                         }
7601                 }
7602         } else {
7603                 flags = 0;
7604                 ret = calc_extent_flag(root, extent_cache, buf, ri, &flags);
7605                 if (ret < 0) {
7606                         fprintf(stderr, "Couldn't calc extent flags\n");
7607                         flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7608                 }
7609         }
7610
7611         if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
7612                 if (ri != NULL &&
7613                     ri->objectid != BTRFS_TREE_RELOC_OBJECTID &&
7614                     ri->objectid == btrfs_header_owner(buf)) {
7615                         /*
7616                          * Ok we got to this block from it's original owner and
7617                          * we have FULL_BACKREF set.  Relocation can leave
7618                          * converted blocks over so this is altogether possible,
7619                          * however it's not possible if the generation > the
7620                          * last snapshot, so check for this case.
7621                          */
7622                         if (!btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC) &&
7623                             btrfs_header_generation(buf) > ri->last_snapshot) {
7624                                 flags &= ~BTRFS_BLOCK_FLAG_FULL_BACKREF;
7625                                 rec->bad_full_backref = 1;
7626                         }
7627                 }
7628         } else {
7629                 if (ri != NULL &&
7630                     (ri->objectid == BTRFS_TREE_RELOC_OBJECTID ||
7631                      btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) {
7632                         flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7633                         rec->bad_full_backref = 1;
7634                 }
7635         }
7636
7637         if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
7638                 rec->flag_block_full_backref = 1;
7639                 parent = bytenr;
7640                 owner = 0;
7641         } else {
7642                 rec->flag_block_full_backref = 0;
7643                 parent = 0;
7644                 owner = btrfs_header_owner(buf);
7645         }
7646
7647         ret = check_block(root, extent_cache, buf, flags);
7648         if (ret)
7649                 goto out;
7650
7651         if (btrfs_is_leaf(buf)) {
7652                 btree_space_waste += btrfs_leaf_free_space(root, buf);
7653                 for (i = 0; i < nritems; i++) {
7654                         struct btrfs_file_extent_item *fi;
7655                         btrfs_item_key_to_cpu(buf, &key, i);
7656                         /*
7657                          * Check key type against the leaf owner.
7658                          * Could filter quite a lot of early error if
7659                          * owner is correct
7660                          */
7661                         if (check_type_with_root(btrfs_header_owner(buf),
7662                                                  key.type)) {
7663                                 fprintf(stderr, "ignoring invalid key\n");
7664                                 continue;
7665                         }
7666                         if (key.type == BTRFS_EXTENT_ITEM_KEY) {
7667                                 process_extent_item(root, extent_cache, buf,
7668                                                     i);
7669                                 continue;
7670                         }
7671                         if (key.type == BTRFS_METADATA_ITEM_KEY) {
7672                                 process_extent_item(root, extent_cache, buf,
7673                                                     i);
7674                                 continue;
7675                         }
7676                         if (key.type == BTRFS_EXTENT_CSUM_KEY) {
7677                                 total_csum_bytes +=
7678                                         btrfs_item_size_nr(buf, i);
7679                                 continue;
7680                         }
7681                         if (key.type == BTRFS_CHUNK_ITEM_KEY) {
7682                                 process_chunk_item(chunk_cache, &key, buf, i);
7683                                 continue;
7684                         }
7685                         if (key.type == BTRFS_DEV_ITEM_KEY) {
7686                                 process_device_item(dev_cache, &key, buf, i);
7687                                 continue;
7688                         }
7689                         if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7690                                 process_block_group_item(block_group_cache,
7691                                         &key, buf, i);
7692                                 continue;
7693                         }
7694                         if (key.type == BTRFS_DEV_EXTENT_KEY) {
7695                                 process_device_extent_item(dev_extent_cache,
7696                                         &key, buf, i);
7697                                 continue;
7698
7699                         }
7700                         if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
7701 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
7702                                 process_extent_ref_v0(extent_cache, buf, i);
7703 #else
7704                                 BUG();
7705 #endif
7706                                 continue;
7707                         }
7708
7709                         if (key.type == BTRFS_TREE_BLOCK_REF_KEY) {
7710                                 ret = add_tree_backref(extent_cache,
7711                                                 key.objectid, 0, key.offset, 0);
7712                                 if (ret < 0)
7713                                         error("add_tree_backref failed: %s",
7714                                               strerror(-ret));
7715                                 continue;
7716                         }
7717                         if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
7718                                 ret = add_tree_backref(extent_cache,
7719                                                 key.objectid, key.offset, 0, 0);
7720                                 if (ret < 0)
7721                                         error("add_tree_backref failed: %s",
7722                                               strerror(-ret));
7723                                 continue;
7724                         }
7725                         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
7726                                 struct btrfs_extent_data_ref *ref;
7727                                 ref = btrfs_item_ptr(buf, i,
7728                                                 struct btrfs_extent_data_ref);
7729                                 add_data_backref(extent_cache,
7730                                         key.objectid, 0,
7731                                         btrfs_extent_data_ref_root(buf, ref),
7732                                         btrfs_extent_data_ref_objectid(buf,
7733                                                                        ref),
7734                                         btrfs_extent_data_ref_offset(buf, ref),
7735                                         btrfs_extent_data_ref_count(buf, ref),
7736                                         0, root->sectorsize);
7737                                 continue;
7738                         }
7739                         if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
7740                                 struct btrfs_shared_data_ref *ref;
7741                                 ref = btrfs_item_ptr(buf, i,
7742                                                 struct btrfs_shared_data_ref);
7743                                 add_data_backref(extent_cache,
7744                                         key.objectid, key.offset, 0, 0, 0,
7745                                         btrfs_shared_data_ref_count(buf, ref),
7746                                         0, root->sectorsize);
7747                                 continue;
7748                         }
7749                         if (key.type == BTRFS_ORPHAN_ITEM_KEY) {
7750                                 struct bad_item *bad;
7751
7752                                 if (key.objectid == BTRFS_ORPHAN_OBJECTID)
7753                                         continue;
7754                                 if (!owner)
7755                                         continue;
7756                                 bad = malloc(sizeof(struct bad_item));
7757                                 if (!bad)
7758                                         continue;
7759                                 INIT_LIST_HEAD(&bad->list);
7760                                 memcpy(&bad->key, &key,
7761                                        sizeof(struct btrfs_key));
7762                                 bad->root_id = owner;
7763                                 list_add_tail(&bad->list, &delete_items);
7764                                 continue;
7765                         }
7766                         if (key.type != BTRFS_EXTENT_DATA_KEY)
7767                                 continue;
7768                         fi = btrfs_item_ptr(buf, i,
7769                                             struct btrfs_file_extent_item);
7770                         if (btrfs_file_extent_type(buf, fi) ==
7771                             BTRFS_FILE_EXTENT_INLINE)
7772                                 continue;
7773                         if (btrfs_file_extent_disk_bytenr(buf, fi) == 0)
7774                                 continue;
7775
7776                         data_bytes_allocated +=
7777                                 btrfs_file_extent_disk_num_bytes(buf, fi);
7778                         if (data_bytes_allocated < root->sectorsize) {
7779                                 abort();
7780                         }
7781                         data_bytes_referenced +=
7782                                 btrfs_file_extent_num_bytes(buf, fi);
7783                         add_data_backref(extent_cache,
7784                                 btrfs_file_extent_disk_bytenr(buf, fi),
7785                                 parent, owner, key.objectid, key.offset -
7786                                 btrfs_file_extent_offset(buf, fi), 1, 1,
7787                                 btrfs_file_extent_disk_num_bytes(buf, fi));
7788                 }
7789         } else {
7790                 int level;
7791                 struct btrfs_key first_key;
7792
7793                 first_key.objectid = 0;
7794
7795                 if (nritems > 0)
7796                         btrfs_item_key_to_cpu(buf, &first_key, 0);
7797                 level = btrfs_header_level(buf);
7798                 for (i = 0; i < nritems; i++) {
7799                         struct extent_record tmpl;
7800
7801                         ptr = btrfs_node_blockptr(buf, i);
7802                         size = root->nodesize;
7803                         btrfs_node_key_to_cpu(buf, &key, i);
7804                         if (ri != NULL) {
7805                                 if ((level == ri->drop_level)
7806                                     && is_dropped_key(&key, &ri->drop_key)) {
7807                                         continue;
7808                                 }
7809                         }
7810
7811                         memset(&tmpl, 0, sizeof(tmpl));
7812                         btrfs_cpu_key_to_disk(&tmpl.parent_key, &key);
7813                         tmpl.parent_generation = btrfs_node_ptr_generation(buf, i);
7814                         tmpl.start = ptr;
7815                         tmpl.nr = size;
7816                         tmpl.refs = 1;
7817                         tmpl.metadata = 1;
7818                         tmpl.max_size = size;
7819                         ret = add_extent_rec(extent_cache, &tmpl);
7820                         if (ret < 0)
7821                                 goto out;
7822
7823                         ret = add_tree_backref(extent_cache, ptr, parent,
7824                                         owner, 1);
7825                         if (ret < 0) {
7826                                 error("add_tree_backref failed: %s",
7827                                       strerror(-ret));
7828                                 continue;
7829                         }
7830
7831                         if (level > 1) {
7832                                 add_pending(nodes, seen, ptr, size);
7833                         } else {
7834                                 add_pending(pending, seen, ptr, size);
7835                         }
7836                 }
7837                 btree_space_waste += (BTRFS_NODEPTRS_PER_BLOCK(root) -
7838                                       nritems) * sizeof(struct btrfs_key_ptr);
7839         }
7840         total_btree_bytes += buf->len;
7841         if (fs_root_objectid(btrfs_header_owner(buf)))
7842                 total_fs_tree_bytes += buf->len;
7843         if (btrfs_header_owner(buf) == BTRFS_EXTENT_TREE_OBJECTID)
7844                 total_extent_tree_bytes += buf->len;
7845         if (!found_old_backref &&
7846             btrfs_header_owner(buf) == BTRFS_TREE_RELOC_OBJECTID &&
7847             btrfs_header_backref_rev(buf) == BTRFS_MIXED_BACKREF_REV &&
7848             !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))
7849                 found_old_backref = 1;
7850 out:
7851         free_extent_buffer(buf);
7852         return ret;
7853 }
7854
7855 static int add_root_to_pending(struct extent_buffer *buf,
7856                                struct cache_tree *extent_cache,
7857                                struct cache_tree *pending,
7858                                struct cache_tree *seen,
7859                                struct cache_tree *nodes,
7860                                u64 objectid)
7861 {
7862         struct extent_record tmpl;
7863         int ret;
7864
7865         if (btrfs_header_level(buf) > 0)
7866                 add_pending(nodes, seen, buf->start, buf->len);
7867         else
7868                 add_pending(pending, seen, buf->start, buf->len);
7869
7870         memset(&tmpl, 0, sizeof(tmpl));
7871         tmpl.start = buf->start;
7872         tmpl.nr = buf->len;
7873         tmpl.is_root = 1;
7874         tmpl.refs = 1;
7875         tmpl.metadata = 1;
7876         tmpl.max_size = buf->len;
7877         add_extent_rec(extent_cache, &tmpl);
7878
7879         if (objectid == BTRFS_TREE_RELOC_OBJECTID ||
7880             btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
7881                 ret = add_tree_backref(extent_cache, buf->start, buf->start,
7882                                 0, 1);
7883         else
7884                 ret = add_tree_backref(extent_cache, buf->start, 0, objectid,
7885                                 1);
7886         return ret;
7887 }
7888
7889 /* as we fix the tree, we might be deleting blocks that
7890  * we're tracking for repair.  This hook makes sure we
7891  * remove any backrefs for blocks as we are fixing them.
7892  */
7893 static int free_extent_hook(struct btrfs_trans_handle *trans,
7894                             struct btrfs_root *root,
7895                             u64 bytenr, u64 num_bytes, u64 parent,
7896                             u64 root_objectid, u64 owner, u64 offset,
7897                             int refs_to_drop)
7898 {
7899         struct extent_record *rec;
7900         struct cache_extent *cache;
7901         int is_data;
7902         struct cache_tree *extent_cache = root->fs_info->fsck_extent_cache;
7903
7904         is_data = owner >= BTRFS_FIRST_FREE_OBJECTID;
7905         cache = lookup_cache_extent(extent_cache, bytenr, num_bytes);
7906         if (!cache)
7907                 return 0;
7908
7909         rec = container_of(cache, struct extent_record, cache);
7910         if (is_data) {
7911                 struct data_backref *back;
7912                 back = find_data_backref(rec, parent, root_objectid, owner,
7913                                          offset, 1, bytenr, num_bytes);
7914                 if (!back)
7915                         goto out;
7916                 if (back->node.found_ref) {
7917                         back->found_ref -= refs_to_drop;
7918                         if (rec->refs)
7919                                 rec->refs -= refs_to_drop;
7920                 }
7921                 if (back->node.found_extent_tree) {
7922                         back->num_refs -= refs_to_drop;
7923                         if (rec->extent_item_refs)
7924                                 rec->extent_item_refs -= refs_to_drop;
7925                 }
7926                 if (back->found_ref == 0)
7927                         back->node.found_ref = 0;
7928                 if (back->num_refs == 0)
7929                         back->node.found_extent_tree = 0;
7930
7931                 if (!back->node.found_extent_tree && back->node.found_ref) {
7932                         list_del(&back->node.list);
7933                         free(back);
7934                 }
7935         } else {
7936                 struct tree_backref *back;
7937                 back = find_tree_backref(rec, parent, root_objectid);
7938                 if (!back)
7939                         goto out;
7940                 if (back->node.found_ref) {
7941                         if (rec->refs)
7942                                 rec->refs--;
7943                         back->node.found_ref = 0;
7944                 }
7945                 if (back->node.found_extent_tree) {
7946                         if (rec->extent_item_refs)
7947                                 rec->extent_item_refs--;
7948                         back->node.found_extent_tree = 0;
7949                 }
7950                 if (!back->node.found_extent_tree && back->node.found_ref) {
7951                         list_del(&back->node.list);
7952                         free(back);
7953                 }
7954         }
7955         maybe_free_extent_rec(extent_cache, rec);
7956 out:
7957         return 0;
7958 }
7959
7960 static int delete_extent_records(struct btrfs_trans_handle *trans,
7961                                  struct btrfs_root *root,
7962                                  struct btrfs_path *path,
7963                                  u64 bytenr, u64 new_len)
7964 {
7965         struct btrfs_key key;
7966         struct btrfs_key found_key;
7967         struct extent_buffer *leaf;
7968         int ret;
7969         int slot;
7970
7971
7972         key.objectid = bytenr;
7973         key.type = (u8)-1;
7974         key.offset = (u64)-1;
7975
7976         while(1) {
7977                 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
7978                                         &key, path, 0, 1);
7979                 if (ret < 0)
7980                         break;
7981
7982                 if (ret > 0) {
7983                         ret = 0;
7984                         if (path->slots[0] == 0)
7985                                 break;
7986                         path->slots[0]--;
7987                 }
7988                 ret = 0;
7989
7990                 leaf = path->nodes[0];
7991                 slot = path->slots[0];
7992
7993                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7994                 if (found_key.objectid != bytenr)
7995                         break;
7996
7997                 if (found_key.type != BTRFS_EXTENT_ITEM_KEY &&
7998                     found_key.type != BTRFS_METADATA_ITEM_KEY &&
7999                     found_key.type != BTRFS_TREE_BLOCK_REF_KEY &&
8000                     found_key.type != BTRFS_EXTENT_DATA_REF_KEY &&
8001                     found_key.type != BTRFS_EXTENT_REF_V0_KEY &&
8002                     found_key.type != BTRFS_SHARED_BLOCK_REF_KEY &&
8003                     found_key.type != BTRFS_SHARED_DATA_REF_KEY) {
8004                         btrfs_release_path(path);
8005                         if (found_key.type == 0) {
8006                                 if (found_key.offset == 0)
8007                                         break;
8008                                 key.offset = found_key.offset - 1;
8009                                 key.type = found_key.type;
8010                         }
8011                         key.type = found_key.type - 1;
8012                         key.offset = (u64)-1;
8013                         continue;
8014                 }
8015
8016                 fprintf(stderr, "repair deleting extent record: key %Lu %u %Lu\n",
8017                         found_key.objectid, found_key.type, found_key.offset);
8018
8019                 ret = btrfs_del_item(trans, root->fs_info->extent_root, path);
8020                 if (ret)
8021                         break;
8022                 btrfs_release_path(path);
8023
8024                 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
8025                     found_key.type == BTRFS_METADATA_ITEM_KEY) {
8026                         u64 bytes = (found_key.type == BTRFS_EXTENT_ITEM_KEY) ?
8027                                 found_key.offset : root->nodesize;
8028
8029                         ret = btrfs_update_block_group(trans, root, bytenr,
8030                                                        bytes, 0, 0);
8031                         if (ret)
8032                                 break;
8033                 }
8034         }
8035
8036         btrfs_release_path(path);
8037         return ret;
8038 }
8039
8040 /*
8041  * for a single backref, this will allocate a new extent
8042  * and add the backref to it.
8043  */
8044 static int record_extent(struct btrfs_trans_handle *trans,
8045                          struct btrfs_fs_info *info,
8046                          struct btrfs_path *path,
8047                          struct extent_record *rec,
8048                          struct extent_backref *back,
8049                          int allocated, u64 flags)
8050 {
8051         int ret;
8052         struct btrfs_root *extent_root = info->extent_root;
8053         struct extent_buffer *leaf;
8054         struct btrfs_key ins_key;
8055         struct btrfs_extent_item *ei;
8056         struct data_backref *dback;
8057         struct btrfs_tree_block_info *bi;
8058
8059         if (!back->is_data)
8060                 rec->max_size = max_t(u64, rec->max_size,
8061                                     info->extent_root->nodesize);
8062
8063         if (!allocated) {
8064                 u32 item_size = sizeof(*ei);
8065
8066                 if (!back->is_data)
8067                         item_size += sizeof(*bi);
8068
8069                 ins_key.objectid = rec->start;
8070                 ins_key.offset = rec->max_size;
8071                 ins_key.type = BTRFS_EXTENT_ITEM_KEY;
8072
8073                 ret = btrfs_insert_empty_item(trans, extent_root, path,
8074                                         &ins_key, item_size);
8075                 if (ret)
8076                         goto fail;
8077
8078                 leaf = path->nodes[0];
8079                 ei = btrfs_item_ptr(leaf, path->slots[0],
8080                                     struct btrfs_extent_item);
8081
8082                 btrfs_set_extent_refs(leaf, ei, 0);
8083                 btrfs_set_extent_generation(leaf, ei, rec->generation);
8084
8085                 if (back->is_data) {
8086                         btrfs_set_extent_flags(leaf, ei,
8087                                                BTRFS_EXTENT_FLAG_DATA);
8088                 } else {
8089                         struct btrfs_disk_key copy_key;;
8090
8091                         bi = (struct btrfs_tree_block_info *)(ei + 1);
8092                         memset_extent_buffer(leaf, 0, (unsigned long)bi,
8093                                              sizeof(*bi));
8094
8095                         btrfs_set_disk_key_objectid(&copy_key,
8096                                                     rec->info_objectid);
8097                         btrfs_set_disk_key_type(&copy_key, 0);
8098                         btrfs_set_disk_key_offset(&copy_key, 0);
8099
8100                         btrfs_set_tree_block_level(leaf, bi, rec->info_level);
8101                         btrfs_set_tree_block_key(leaf, bi, &copy_key);
8102
8103                         btrfs_set_extent_flags(leaf, ei,
8104                                                BTRFS_EXTENT_FLAG_TREE_BLOCK | flags);
8105                 }
8106
8107                 btrfs_mark_buffer_dirty(leaf);
8108                 ret = btrfs_update_block_group(trans, extent_root, rec->start,
8109                                                rec->max_size, 1, 0);
8110                 if (ret)
8111                         goto fail;
8112                 btrfs_release_path(path);
8113         }
8114
8115         if (back->is_data) {
8116                 u64 parent;
8117                 int i;
8118
8119                 dback = to_data_backref(back);
8120                 if (back->full_backref)
8121                         parent = dback->parent;
8122                 else
8123                         parent = 0;
8124
8125                 for (i = 0; i < dback->found_ref; i++) {
8126                         /* if parent != 0, we're doing a full backref
8127                          * passing BTRFS_FIRST_FREE_OBJECTID as the owner
8128                          * just makes the backref allocator create a data
8129                          * backref
8130                          */
8131                         ret = btrfs_inc_extent_ref(trans, info->extent_root,
8132                                                    rec->start, rec->max_size,
8133                                                    parent,
8134                                                    dback->root,
8135                                                    parent ?
8136                                                    BTRFS_FIRST_FREE_OBJECTID :
8137                                                    dback->owner,
8138                                                    dback->offset);
8139                         if (ret)
8140                                 break;
8141                 }
8142                 fprintf(stderr, "adding new data backref"
8143                                 " on %llu %s %llu owner %llu"
8144                                 " offset %llu found %d\n",
8145                                 (unsigned long long)rec->start,
8146                                 back->full_backref ?
8147                                 "parent" : "root",
8148                                 back->full_backref ?
8149                                 (unsigned long long)parent :
8150                                 (unsigned long long)dback->root,
8151                                 (unsigned long long)dback->owner,
8152                                 (unsigned long long)dback->offset,
8153                                 dback->found_ref);
8154         } else {
8155                 u64 parent;
8156                 struct tree_backref *tback;
8157
8158                 tback = to_tree_backref(back);
8159                 if (back->full_backref)
8160                         parent = tback->parent;
8161                 else
8162                         parent = 0;
8163
8164                 ret = btrfs_inc_extent_ref(trans, info->extent_root,
8165                                            rec->start, rec->max_size,
8166                                            parent, tback->root, 0, 0);
8167                 fprintf(stderr, "adding new tree backref on "
8168                         "start %llu len %llu parent %llu root %llu\n",
8169                         rec->start, rec->max_size, parent, tback->root);
8170         }
8171 fail:
8172         btrfs_release_path(path);
8173         return ret;
8174 }
8175
8176 static struct extent_entry *find_entry(struct list_head *entries,
8177                                        u64 bytenr, u64 bytes)
8178 {
8179         struct extent_entry *entry = NULL;
8180
8181         list_for_each_entry(entry, entries, list) {
8182                 if (entry->bytenr == bytenr && entry->bytes == bytes)
8183                         return entry;
8184         }
8185
8186         return NULL;
8187 }
8188
8189 static struct extent_entry *find_most_right_entry(struct list_head *entries)
8190 {
8191         struct extent_entry *entry, *best = NULL, *prev = NULL;
8192
8193         list_for_each_entry(entry, entries, list) {
8194                 /*
8195                  * If there are as many broken entries as entries then we know
8196                  * not to trust this particular entry.
8197                  */
8198                 if (entry->broken == entry->count)
8199                         continue;
8200
8201                 /*
8202                  * Special case, when there are only two entries and 'best' is
8203                  * the first one
8204                  */
8205                 if (!prev) {
8206                         best = entry;
8207                         prev = entry;
8208                         continue;
8209                 }
8210
8211                 /*
8212                  * If our current entry == best then we can't be sure our best
8213                  * is really the best, so we need to keep searching.
8214                  */
8215                 if (best && best->count == entry->count) {
8216                         prev = entry;
8217                         best = NULL;
8218                         continue;
8219                 }
8220
8221                 /* Prev == entry, not good enough, have to keep searching */
8222                 if (!prev->broken && prev->count == entry->count)
8223                         continue;
8224
8225                 if (!best)
8226                         best = (prev->count > entry->count) ? prev : entry;
8227                 else if (best->count < entry->count)
8228                         best = entry;
8229                 prev = entry;
8230         }
8231
8232         return best;
8233 }
8234
8235 static int repair_ref(struct btrfs_fs_info *info, struct btrfs_path *path,
8236                       struct data_backref *dback, struct extent_entry *entry)
8237 {
8238         struct btrfs_trans_handle *trans;
8239         struct btrfs_root *root;
8240         struct btrfs_file_extent_item *fi;
8241         struct extent_buffer *leaf;
8242         struct btrfs_key key;
8243         u64 bytenr, bytes;
8244         int ret, err;
8245
8246         key.objectid = dback->root;
8247         key.type = BTRFS_ROOT_ITEM_KEY;
8248         key.offset = (u64)-1;
8249         root = btrfs_read_fs_root(info, &key);
8250         if (IS_ERR(root)) {
8251                 fprintf(stderr, "Couldn't find root for our ref\n");
8252                 return -EINVAL;
8253         }
8254
8255         /*
8256          * The backref points to the original offset of the extent if it was
8257          * split, so we need to search down to the offset we have and then walk
8258          * forward until we find the backref we're looking for.
8259          */
8260         key.objectid = dback->owner;
8261         key.type = BTRFS_EXTENT_DATA_KEY;
8262         key.offset = dback->offset;
8263         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8264         if (ret < 0) {
8265                 fprintf(stderr, "Error looking up ref %d\n", ret);
8266                 return ret;
8267         }
8268
8269         while (1) {
8270                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
8271                         ret = btrfs_next_leaf(root, path);
8272                         if (ret) {
8273                                 fprintf(stderr, "Couldn't find our ref, next\n");
8274                                 return -EINVAL;
8275                         }
8276                 }
8277                 leaf = path->nodes[0];
8278                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
8279                 if (key.objectid != dback->owner ||
8280                     key.type != BTRFS_EXTENT_DATA_KEY) {
8281                         fprintf(stderr, "Couldn't find our ref, search\n");
8282                         return -EINVAL;
8283                 }
8284                 fi = btrfs_item_ptr(leaf, path->slots[0],
8285                                     struct btrfs_file_extent_item);
8286                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
8287                 bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
8288
8289                 if (bytenr == dback->disk_bytenr && bytes == dback->bytes)
8290                         break;
8291                 path->slots[0]++;
8292         }
8293
8294         btrfs_release_path(path);
8295
8296         trans = btrfs_start_transaction(root, 1);
8297         if (IS_ERR(trans))
8298                 return PTR_ERR(trans);
8299
8300         /*
8301          * Ok we have the key of the file extent we want to fix, now we can cow
8302          * down to the thing and fix it.
8303          */
8304         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
8305         if (ret < 0) {
8306                 fprintf(stderr, "Error cowing down to ref [%Lu, %u, %Lu]: %d\n",
8307                         key.objectid, key.type, key.offset, ret);
8308                 goto out;
8309         }
8310         if (ret > 0) {
8311                 fprintf(stderr, "Well that's odd, we just found this key "
8312                         "[%Lu, %u, %Lu]\n", key.objectid, key.type,
8313                         key.offset);
8314                 ret = -EINVAL;
8315                 goto out;
8316         }
8317         leaf = path->nodes[0];
8318         fi = btrfs_item_ptr(leaf, path->slots[0],
8319                             struct btrfs_file_extent_item);
8320
8321         if (btrfs_file_extent_compression(leaf, fi) &&
8322             dback->disk_bytenr != entry->bytenr) {
8323                 fprintf(stderr, "Ref doesn't match the record start and is "
8324                         "compressed, please take a btrfs-image of this file "
8325                         "system and send it to a btrfs developer so they can "
8326                         "complete this functionality for bytenr %Lu\n",
8327                         dback->disk_bytenr);
8328                 ret = -EINVAL;
8329                 goto out;
8330         }
8331
8332         if (dback->node.broken && dback->disk_bytenr != entry->bytenr) {
8333                 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
8334         } else if (dback->disk_bytenr > entry->bytenr) {
8335                 u64 off_diff, offset;
8336
8337                 off_diff = dback->disk_bytenr - entry->bytenr;
8338                 offset = btrfs_file_extent_offset(leaf, fi);
8339                 if (dback->disk_bytenr + offset +
8340                     btrfs_file_extent_num_bytes(leaf, fi) >
8341                     entry->bytenr + entry->bytes) {
8342                         fprintf(stderr, "Ref is past the entry end, please "
8343                                 "take a btrfs-image of this file system and "
8344                                 "send it to a btrfs developer, ref %Lu\n",
8345                                 dback->disk_bytenr);
8346                         ret = -EINVAL;
8347                         goto out;
8348                 }
8349                 offset += off_diff;
8350                 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
8351                 btrfs_set_file_extent_offset(leaf, fi, offset);
8352         } else if (dback->disk_bytenr < entry->bytenr) {
8353                 u64 offset;
8354
8355                 offset = btrfs_file_extent_offset(leaf, fi);
8356                 if (dback->disk_bytenr + offset < entry->bytenr) {
8357                         fprintf(stderr, "Ref is before the entry start, please"
8358                                 " take a btrfs-image of this file system and "
8359                                 "send it to a btrfs developer, ref %Lu\n",
8360                                 dback->disk_bytenr);
8361                         ret = -EINVAL;
8362                         goto out;
8363                 }
8364
8365                 offset += dback->disk_bytenr;
8366                 offset -= entry->bytenr;
8367                 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
8368                 btrfs_set_file_extent_offset(leaf, fi, offset);
8369         }
8370
8371         btrfs_set_file_extent_disk_num_bytes(leaf, fi, entry->bytes);
8372
8373         /*
8374          * Chances are if disk_num_bytes were wrong then so is ram_bytes, but
8375          * only do this if we aren't using compression, otherwise it's a
8376          * trickier case.
8377          */
8378         if (!btrfs_file_extent_compression(leaf, fi))
8379                 btrfs_set_file_extent_ram_bytes(leaf, fi, entry->bytes);
8380         else
8381                 printf("ram bytes may be wrong?\n");
8382         btrfs_mark_buffer_dirty(leaf);
8383 out:
8384         err = btrfs_commit_transaction(trans, root);
8385         btrfs_release_path(path);
8386         return ret ? ret : err;
8387 }
8388
8389 static int verify_backrefs(struct btrfs_fs_info *info, struct btrfs_path *path,
8390                            struct extent_record *rec)
8391 {
8392         struct extent_backref *back;
8393         struct data_backref *dback;
8394         struct extent_entry *entry, *best = NULL;
8395         LIST_HEAD(entries);
8396         int nr_entries = 0;
8397         int broken_entries = 0;
8398         int ret = 0;
8399         short mismatch = 0;
8400
8401         /*
8402          * Metadata is easy and the backrefs should always agree on bytenr and
8403          * size, if not we've got bigger issues.
8404          */
8405         if (rec->metadata)
8406                 return 0;
8407
8408         list_for_each_entry(back, &rec->backrefs, list) {
8409                 if (back->full_backref || !back->is_data)
8410                         continue;
8411
8412                 dback = to_data_backref(back);
8413
8414                 /*
8415                  * We only pay attention to backrefs that we found a real
8416                  * backref for.
8417                  */
8418                 if (dback->found_ref == 0)
8419                         continue;
8420
8421                 /*
8422                  * For now we only catch when the bytes don't match, not the
8423                  * bytenr.  We can easily do this at the same time, but I want
8424                  * to have a fs image to test on before we just add repair
8425                  * functionality willy-nilly so we know we won't screw up the
8426                  * repair.
8427                  */
8428
8429                 entry = find_entry(&entries, dback->disk_bytenr,
8430                                    dback->bytes);
8431                 if (!entry) {
8432                         entry = malloc(sizeof(struct extent_entry));
8433                         if (!entry) {
8434                                 ret = -ENOMEM;
8435                                 goto out;
8436                         }
8437                         memset(entry, 0, sizeof(*entry));
8438                         entry->bytenr = dback->disk_bytenr;
8439                         entry->bytes = dback->bytes;
8440                         list_add_tail(&entry->list, &entries);
8441                         nr_entries++;
8442                 }
8443
8444                 /*
8445                  * If we only have on entry we may think the entries agree when
8446                  * in reality they don't so we have to do some extra checking.
8447                  */
8448                 if (dback->disk_bytenr != rec->start ||
8449                     dback->bytes != rec->nr || back->broken)
8450                         mismatch = 1;
8451
8452                 if (back->broken) {
8453                         entry->broken++;
8454                         broken_entries++;
8455                 }
8456
8457                 entry->count++;
8458         }
8459
8460         /* Yay all the backrefs agree, carry on good sir */
8461         if (nr_entries <= 1 && !mismatch)
8462                 goto out;
8463
8464         fprintf(stderr, "attempting to repair backref discrepency for bytenr "
8465                 "%Lu\n", rec->start);
8466
8467         /*
8468          * First we want to see if the backrefs can agree amongst themselves who
8469          * is right, so figure out which one of the entries has the highest
8470          * count.
8471          */
8472         best = find_most_right_entry(&entries);
8473
8474         /*
8475          * Ok so we may have an even split between what the backrefs think, so
8476          * this is where we use the extent ref to see what it thinks.
8477          */
8478         if (!best) {
8479                 entry = find_entry(&entries, rec->start, rec->nr);
8480                 if (!entry && (!broken_entries || !rec->found_rec)) {
8481                         fprintf(stderr, "Backrefs don't agree with each other "
8482                                 "and extent record doesn't agree with anybody,"
8483                                 " so we can't fix bytenr %Lu bytes %Lu\n",
8484                                 rec->start, rec->nr);
8485                         ret = -EINVAL;
8486                         goto out;
8487                 } else if (!entry) {
8488                         /*
8489                          * Ok our backrefs were broken, we'll assume this is the
8490                          * correct value and add an entry for this range.
8491                          */
8492                         entry = malloc(sizeof(struct extent_entry));
8493                         if (!entry) {
8494                                 ret = -ENOMEM;
8495                                 goto out;
8496                         }
8497                         memset(entry, 0, sizeof(*entry));
8498                         entry->bytenr = rec->start;
8499                         entry->bytes = rec->nr;
8500                         list_add_tail(&entry->list, &entries);
8501                         nr_entries++;
8502                 }
8503                 entry->count++;
8504                 best = find_most_right_entry(&entries);
8505                 if (!best) {
8506                         fprintf(stderr, "Backrefs and extent record evenly "
8507                                 "split on who is right, this is going to "
8508                                 "require user input to fix bytenr %Lu bytes "
8509                                 "%Lu\n", rec->start, rec->nr);
8510                         ret = -EINVAL;
8511                         goto out;
8512                 }
8513         }
8514
8515         /*
8516          * I don't think this can happen currently as we'll abort() if we catch
8517          * this case higher up, but in case somebody removes that we still can't
8518          * deal with it properly here yet, so just bail out of that's the case.
8519          */
8520         if (best->bytenr != rec->start) {
8521                 fprintf(stderr, "Extent start and backref starts don't match, "
8522                         "please use btrfs-image on this file system and send "
8523                         "it to a btrfs developer so they can make fsck fix "
8524                         "this particular case.  bytenr is %Lu, bytes is %Lu\n",
8525                         rec->start, rec->nr);
8526                 ret = -EINVAL;
8527                 goto out;
8528         }
8529
8530         /*
8531          * Ok great we all agreed on an extent record, let's go find the real
8532          * references and fix up the ones that don't match.
8533          */
8534         list_for_each_entry(back, &rec->backrefs, list) {
8535                 if (back->full_backref || !back->is_data)
8536                         continue;
8537
8538                 dback = to_data_backref(back);
8539
8540                 /*
8541                  * Still ignoring backrefs that don't have a real ref attached
8542                  * to them.
8543                  */
8544                 if (dback->found_ref == 0)
8545                         continue;
8546
8547                 if (dback->bytes == best->bytes &&
8548                     dback->disk_bytenr == best->bytenr)
8549                         continue;
8550
8551                 ret = repair_ref(info, path, dback, best);
8552                 if (ret)
8553                         goto out;
8554         }
8555
8556         /*
8557          * Ok we messed with the actual refs, which means we need to drop our
8558          * entire cache and go back and rescan.  I know this is a huge pain and
8559          * adds a lot of extra work, but it's the only way to be safe.  Once all
8560          * the backrefs agree we may not need to do anything to the extent
8561          * record itself.
8562          */
8563         ret = -EAGAIN;
8564 out:
8565         while (!list_empty(&entries)) {
8566                 entry = list_entry(entries.next, struct extent_entry, list);
8567                 list_del_init(&entry->list);
8568                 free(entry);
8569         }
8570         return ret;
8571 }
8572
8573 static int process_duplicates(struct btrfs_root *root,
8574                               struct cache_tree *extent_cache,
8575                               struct extent_record *rec)
8576 {
8577         struct extent_record *good, *tmp;
8578         struct cache_extent *cache;
8579         int ret;
8580
8581         /*
8582          * If we found a extent record for this extent then return, or if we
8583          * have more than one duplicate we are likely going to need to delete
8584          * something.
8585          */
8586         if (rec->found_rec || rec->num_duplicates > 1)
8587                 return 0;
8588
8589         /* Shouldn't happen but just in case */
8590         BUG_ON(!rec->num_duplicates);
8591
8592         /*
8593          * So this happens if we end up with a backref that doesn't match the
8594          * actual extent entry.  So either the backref is bad or the extent
8595          * entry is bad.  Either way we want to have the extent_record actually
8596          * reflect what we found in the extent_tree, so we need to take the
8597          * duplicate out and use that as the extent_record since the only way we
8598          * get a duplicate is if we find a real life BTRFS_EXTENT_ITEM_KEY.
8599          */
8600         remove_cache_extent(extent_cache, &rec->cache);
8601
8602         good = to_extent_record(rec->dups.next);
8603         list_del_init(&good->list);
8604         INIT_LIST_HEAD(&good->backrefs);
8605         INIT_LIST_HEAD(&good->dups);
8606         good->cache.start = good->start;
8607         good->cache.size = good->nr;
8608         good->content_checked = 0;
8609         good->owner_ref_checked = 0;
8610         good->num_duplicates = 0;
8611         good->refs = rec->refs;
8612         list_splice_init(&rec->backrefs, &good->backrefs);
8613         while (1) {
8614                 cache = lookup_cache_extent(extent_cache, good->start,
8615                                             good->nr);
8616                 if (!cache)
8617                         break;
8618                 tmp = container_of(cache, struct extent_record, cache);
8619
8620                 /*
8621                  * If we find another overlapping extent and it's found_rec is
8622                  * set then it's a duplicate and we need to try and delete
8623                  * something.
8624                  */
8625                 if (tmp->found_rec || tmp->num_duplicates > 0) {
8626                         if (list_empty(&good->list))
8627                                 list_add_tail(&good->list,
8628                                               &duplicate_extents);
8629                         good->num_duplicates += tmp->num_duplicates + 1;
8630                         list_splice_init(&tmp->dups, &good->dups);
8631                         list_del_init(&tmp->list);
8632                         list_add_tail(&tmp->list, &good->dups);
8633                         remove_cache_extent(extent_cache, &tmp->cache);
8634                         continue;
8635                 }
8636
8637                 /*
8638                  * Ok we have another non extent item backed extent rec, so lets
8639                  * just add it to this extent and carry on like we did above.
8640                  */
8641                 good->refs += tmp->refs;
8642                 list_splice_init(&tmp->backrefs, &good->backrefs);
8643                 remove_cache_extent(extent_cache, &tmp->cache);
8644                 free(tmp);
8645         }
8646         ret = insert_cache_extent(extent_cache, &good->cache);
8647         BUG_ON(ret);
8648         free(rec);
8649         return good->num_duplicates ? 0 : 1;
8650 }
8651
8652 static int delete_duplicate_records(struct btrfs_root *root,
8653                                     struct extent_record *rec)
8654 {
8655         struct btrfs_trans_handle *trans;
8656         LIST_HEAD(delete_list);
8657         struct btrfs_path path;
8658         struct extent_record *tmp, *good, *n;
8659         int nr_del = 0;
8660         int ret = 0, err;
8661         struct btrfs_key key;
8662
8663         btrfs_init_path(&path);
8664
8665         good = rec;
8666         /* Find the record that covers all of the duplicates. */
8667         list_for_each_entry(tmp, &rec->dups, list) {
8668                 if (good->start < tmp->start)
8669                         continue;
8670                 if (good->nr > tmp->nr)
8671                         continue;
8672
8673                 if (tmp->start + tmp->nr < good->start + good->nr) {
8674                         fprintf(stderr, "Ok we have overlapping extents that "
8675                                 "aren't completely covered by each other, this "
8676                                 "is going to require more careful thought.  "
8677                                 "The extents are [%Lu-%Lu] and [%Lu-%Lu]\n",
8678                                 tmp->start, tmp->nr, good->start, good->nr);
8679                         abort();
8680                 }
8681                 good = tmp;
8682         }
8683
8684         if (good != rec)
8685                 list_add_tail(&rec->list, &delete_list);
8686
8687         list_for_each_entry_safe(tmp, n, &rec->dups, list) {
8688                 if (tmp == good)
8689                         continue;
8690                 list_move_tail(&tmp->list, &delete_list);
8691         }
8692
8693         root = root->fs_info->extent_root;
8694         trans = btrfs_start_transaction(root, 1);
8695         if (IS_ERR(trans)) {
8696                 ret = PTR_ERR(trans);
8697                 goto out;
8698         }
8699
8700         list_for_each_entry(tmp, &delete_list, list) {
8701                 if (tmp->found_rec == 0)
8702                         continue;
8703                 key.objectid = tmp->start;
8704                 key.type = BTRFS_EXTENT_ITEM_KEY;
8705                 key.offset = tmp->nr;
8706
8707                 /* Shouldn't happen but just in case */
8708                 if (tmp->metadata) {
8709                         fprintf(stderr, "Well this shouldn't happen, extent "
8710                                 "record overlaps but is metadata? "
8711                                 "[%Lu, %Lu]\n", tmp->start, tmp->nr);
8712                         abort();
8713                 }
8714
8715                 ret = btrfs_search_slot(trans, root, &key, &path, -1, 1);
8716                 if (ret) {
8717                         if (ret > 0)
8718                                 ret = -EINVAL;
8719                         break;
8720                 }
8721                 ret = btrfs_del_item(trans, root, &path);
8722                 if (ret)
8723                         break;
8724                 btrfs_release_path(&path);
8725                 nr_del++;
8726         }
8727         err = btrfs_commit_transaction(trans, root);
8728         if (err && !ret)
8729                 ret = err;
8730 out:
8731         while (!list_empty(&delete_list)) {
8732                 tmp = to_extent_record(delete_list.next);
8733                 list_del_init(&tmp->list);
8734                 if (tmp == rec)
8735                         continue;
8736                 free(tmp);
8737         }
8738
8739         while (!list_empty(&rec->dups)) {
8740                 tmp = to_extent_record(rec->dups.next);
8741                 list_del_init(&tmp->list);
8742                 free(tmp);
8743         }
8744
8745         btrfs_release_path(&path);
8746
8747         if (!ret && !nr_del)
8748                 rec->num_duplicates = 0;
8749
8750         return ret ? ret : nr_del;
8751 }
8752
8753 static int find_possible_backrefs(struct btrfs_fs_info *info,
8754                                   struct btrfs_path *path,
8755                                   struct cache_tree *extent_cache,
8756                                   struct extent_record *rec)
8757 {
8758         struct btrfs_root *root;
8759         struct extent_backref *back;
8760         struct data_backref *dback;
8761         struct cache_extent *cache;
8762         struct btrfs_file_extent_item *fi;
8763         struct btrfs_key key;
8764         u64 bytenr, bytes;
8765         int ret;
8766
8767         list_for_each_entry(back, &rec->backrefs, list) {
8768                 /* Don't care about full backrefs (poor unloved backrefs) */
8769                 if (back->full_backref || !back->is_data)
8770                         continue;
8771
8772                 dback = to_data_backref(back);
8773
8774                 /* We found this one, we don't need to do a lookup */
8775                 if (dback->found_ref)
8776                         continue;
8777
8778                 key.objectid = dback->root;
8779                 key.type = BTRFS_ROOT_ITEM_KEY;
8780                 key.offset = (u64)-1;
8781
8782                 root = btrfs_read_fs_root(info, &key);
8783
8784                 /* No root, definitely a bad ref, skip */
8785                 if (IS_ERR(root) && PTR_ERR(root) == -ENOENT)
8786                         continue;
8787                 /* Other err, exit */
8788                 if (IS_ERR(root))
8789                         return PTR_ERR(root);
8790
8791                 key.objectid = dback->owner;
8792                 key.type = BTRFS_EXTENT_DATA_KEY;
8793                 key.offset = dback->offset;
8794                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8795                 if (ret) {
8796                         btrfs_release_path(path);
8797                         if (ret < 0)
8798                                 return ret;
8799                         /* Didn't find it, we can carry on */
8800                         ret = 0;
8801                         continue;
8802                 }
8803
8804                 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
8805                                     struct btrfs_file_extent_item);
8806                 bytenr = btrfs_file_extent_disk_bytenr(path->nodes[0], fi);
8807                 bytes = btrfs_file_extent_disk_num_bytes(path->nodes[0], fi);
8808                 btrfs_release_path(path);
8809                 cache = lookup_cache_extent(extent_cache, bytenr, 1);
8810                 if (cache) {
8811                         struct extent_record *tmp;
8812                         tmp = container_of(cache, struct extent_record, cache);
8813
8814                         /*
8815                          * If we found an extent record for the bytenr for this
8816                          * particular backref then we can't add it to our
8817                          * current extent record.  We only want to add backrefs
8818                          * that don't have a corresponding extent item in the
8819                          * extent tree since they likely belong to this record
8820                          * and we need to fix it if it doesn't match bytenrs.
8821                          */
8822                         if  (tmp->found_rec)
8823                                 continue;
8824                 }
8825
8826                 dback->found_ref += 1;
8827                 dback->disk_bytenr = bytenr;
8828                 dback->bytes = bytes;
8829
8830                 /*
8831                  * Set this so the verify backref code knows not to trust the
8832                  * values in this backref.
8833                  */
8834                 back->broken = 1;
8835         }
8836
8837         return 0;
8838 }
8839
8840 /*
8841  * Record orphan data ref into corresponding root.
8842  *
8843  * Return 0 if the extent item contains data ref and recorded.
8844  * Return 1 if the extent item contains no useful data ref
8845  *   On that case, it may contains only shared_dataref or metadata backref
8846  *   or the file extent exists(this should be handled by the extent bytenr
8847  *   recovery routine)
8848  * Return <0 if something goes wrong.
8849  */
8850 static int record_orphan_data_extents(struct btrfs_fs_info *fs_info,
8851                                       struct extent_record *rec)
8852 {
8853         struct btrfs_key key;
8854         struct btrfs_root *dest_root;
8855         struct extent_backref *back;
8856         struct data_backref *dback;
8857         struct orphan_data_extent *orphan;
8858         struct btrfs_path path;
8859         int recorded_data_ref = 0;
8860         int ret = 0;
8861
8862         if (rec->metadata)
8863                 return 1;
8864         btrfs_init_path(&path);
8865         list_for_each_entry(back, &rec->backrefs, list) {
8866                 if (back->full_backref || !back->is_data ||
8867                     !back->found_extent_tree)
8868                         continue;
8869                 dback = to_data_backref(back);
8870                 if (dback->found_ref)
8871                         continue;
8872                 key.objectid = dback->root;
8873                 key.type = BTRFS_ROOT_ITEM_KEY;
8874                 key.offset = (u64)-1;
8875
8876                 dest_root = btrfs_read_fs_root(fs_info, &key);
8877
8878                 /* For non-exist root we just skip it */
8879                 if (IS_ERR(dest_root) || !dest_root)
8880                         continue;
8881
8882                 key.objectid = dback->owner;
8883                 key.type = BTRFS_EXTENT_DATA_KEY;
8884                 key.offset = dback->offset;
8885
8886                 ret = btrfs_search_slot(NULL, dest_root, &key, &path, 0, 0);
8887                 btrfs_release_path(&path);
8888                 /*
8889                  * For ret < 0, it's OK since the fs-tree may be corrupted,
8890                  * we need to record it for inode/file extent rebuild.
8891                  * For ret > 0, we record it only for file extent rebuild.
8892                  * For ret == 0, the file extent exists but only bytenr
8893                  * mismatch, let the original bytenr fix routine to handle,
8894                  * don't record it.
8895                  */
8896                 if (ret == 0)
8897                         continue;
8898                 ret = 0;
8899                 orphan = malloc(sizeof(*orphan));
8900                 if (!orphan) {
8901                         ret = -ENOMEM;
8902                         goto out;
8903                 }
8904                 INIT_LIST_HEAD(&orphan->list);
8905                 orphan->root = dback->root;
8906                 orphan->objectid = dback->owner;
8907                 orphan->offset = dback->offset;
8908                 orphan->disk_bytenr = rec->cache.start;
8909                 orphan->disk_len = rec->cache.size;
8910                 list_add(&dest_root->orphan_data_extents, &orphan->list);
8911                 recorded_data_ref = 1;
8912         }
8913 out:
8914         btrfs_release_path(&path);
8915         if (!ret)
8916                 return !recorded_data_ref;
8917         else
8918                 return ret;
8919 }
8920
8921 /*
8922  * when an incorrect extent item is found, this will delete
8923  * all of the existing entries for it and recreate them
8924  * based on what the tree scan found.
8925  */
8926 static int fixup_extent_refs(struct btrfs_fs_info *info,
8927                              struct cache_tree *extent_cache,
8928                              struct extent_record *rec)
8929 {
8930         struct btrfs_trans_handle *trans = NULL;
8931         int ret;
8932         struct btrfs_path path;
8933         struct list_head *cur = rec->backrefs.next;
8934         struct cache_extent *cache;
8935         struct extent_backref *back;
8936         int allocated = 0;
8937         u64 flags = 0;
8938
8939         if (rec->flag_block_full_backref)
8940                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
8941
8942         btrfs_init_path(&path);
8943         if (rec->refs != rec->extent_item_refs && !rec->metadata) {
8944                 /*
8945                  * Sometimes the backrefs themselves are so broken they don't
8946                  * get attached to any meaningful rec, so first go back and
8947                  * check any of our backrefs that we couldn't find and throw
8948                  * them into the list if we find the backref so that
8949                  * verify_backrefs can figure out what to do.
8950                  */
8951                 ret = find_possible_backrefs(info, &path, extent_cache, rec);
8952                 if (ret < 0)
8953                         goto out;
8954         }
8955
8956         /* step one, make sure all of the backrefs agree */
8957         ret = verify_backrefs(info, &path, rec);
8958         if (ret < 0)
8959                 goto out;
8960
8961         trans = btrfs_start_transaction(info->extent_root, 1);
8962         if (IS_ERR(trans)) {
8963                 ret = PTR_ERR(trans);
8964                 goto out;
8965         }
8966
8967         /* step two, delete all the existing records */
8968         ret = delete_extent_records(trans, info->extent_root, &path,
8969                                     rec->start, rec->max_size);
8970
8971         if (ret < 0)
8972                 goto out;
8973
8974         /* was this block corrupt?  If so, don't add references to it */
8975         cache = lookup_cache_extent(info->corrupt_blocks,
8976                                     rec->start, rec->max_size);
8977         if (cache) {
8978                 ret = 0;
8979                 goto out;
8980         }
8981
8982         /* step three, recreate all the refs we did find */
8983         while(cur != &rec->backrefs) {
8984                 back = to_extent_backref(cur);
8985                 cur = cur->next;
8986
8987                 /*
8988                  * if we didn't find any references, don't create a
8989                  * new extent record
8990                  */
8991                 if (!back->found_ref)
8992                         continue;
8993
8994                 rec->bad_full_backref = 0;
8995                 ret = record_extent(trans, info, &path, rec, back, allocated, flags);
8996                 allocated = 1;
8997
8998                 if (ret)
8999                         goto out;
9000         }
9001 out:
9002         if (trans) {
9003                 int err = btrfs_commit_transaction(trans, info->extent_root);
9004                 if (!ret)
9005                         ret = err;
9006         }
9007
9008         btrfs_release_path(&path);
9009         return ret;
9010 }
9011
9012 static int fixup_extent_flags(struct btrfs_fs_info *fs_info,
9013                               struct extent_record *rec)
9014 {
9015         struct btrfs_trans_handle *trans;
9016         struct btrfs_root *root = fs_info->extent_root;
9017         struct btrfs_path path;
9018         struct btrfs_extent_item *ei;
9019         struct btrfs_key key;
9020         u64 flags;
9021         int ret = 0;
9022
9023         key.objectid = rec->start;
9024         if (rec->metadata) {
9025                 key.type = BTRFS_METADATA_ITEM_KEY;
9026                 key.offset = rec->info_level;
9027         } else {
9028                 key.type = BTRFS_EXTENT_ITEM_KEY;
9029                 key.offset = rec->max_size;
9030         }
9031
9032         trans = btrfs_start_transaction(root, 0);
9033         if (IS_ERR(trans))
9034                 return PTR_ERR(trans);
9035
9036         btrfs_init_path(&path);
9037         ret = btrfs_search_slot(trans, root, &key, &path, 0, 1);
9038         if (ret < 0) {
9039                 btrfs_release_path(&path);
9040                 btrfs_commit_transaction(trans, root);
9041                 return ret;
9042         } else if (ret) {
9043                 fprintf(stderr, "Didn't find extent for %llu\n",
9044                         (unsigned long long)rec->start);
9045                 btrfs_release_path(&path);
9046                 btrfs_commit_transaction(trans, root);
9047                 return -ENOENT;
9048         }
9049
9050         ei = btrfs_item_ptr(path.nodes[0], path.slots[0],
9051                             struct btrfs_extent_item);
9052         flags = btrfs_extent_flags(path.nodes[0], ei);
9053         if (rec->flag_block_full_backref) {
9054                 fprintf(stderr, "setting full backref on %llu\n",
9055                         (unsigned long long)key.objectid);
9056                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
9057         } else {
9058                 fprintf(stderr, "clearing full backref on %llu\n",
9059                         (unsigned long long)key.objectid);
9060                 flags &= ~BTRFS_BLOCK_FLAG_FULL_BACKREF;
9061         }
9062         btrfs_set_extent_flags(path.nodes[0], ei, flags);
9063         btrfs_mark_buffer_dirty(path.nodes[0]);
9064         btrfs_release_path(&path);
9065         return btrfs_commit_transaction(trans, root);
9066 }
9067
9068 /* right now we only prune from the extent allocation tree */
9069 static int prune_one_block(struct btrfs_trans_handle *trans,
9070                            struct btrfs_fs_info *info,
9071                            struct btrfs_corrupt_block *corrupt)
9072 {
9073         int ret;
9074         struct btrfs_path path;
9075         struct extent_buffer *eb;
9076         u64 found;
9077         int slot;
9078         int nritems;
9079         int level = corrupt->level + 1;
9080
9081         btrfs_init_path(&path);
9082 again:
9083         /* we want to stop at the parent to our busted block */
9084         path.lowest_level = level;
9085
9086         ret = btrfs_search_slot(trans, info->extent_root,
9087                                 &corrupt->key, &path, -1, 1);
9088
9089         if (ret < 0)
9090                 goto out;
9091
9092         eb = path.nodes[level];
9093         if (!eb) {
9094                 ret = -ENOENT;
9095                 goto out;
9096         }
9097
9098         /*
9099          * hopefully the search gave us the block we want to prune,
9100          * lets try that first
9101          */
9102         slot = path.slots[level];
9103         found =  btrfs_node_blockptr(eb, slot);
9104         if (found == corrupt->cache.start)
9105                 goto del_ptr;
9106
9107         nritems = btrfs_header_nritems(eb);
9108
9109         /* the search failed, lets scan this node and hope we find it */
9110         for (slot = 0; slot < nritems; slot++) {
9111                 found =  btrfs_node_blockptr(eb, slot);
9112                 if (found == corrupt->cache.start)
9113                         goto del_ptr;
9114         }
9115         /*
9116          * we couldn't find the bad block.  TODO, search all the nodes for pointers
9117          * to this block
9118          */
9119         if (eb == info->extent_root->node) {
9120                 ret = -ENOENT;
9121                 goto out;
9122         } else {
9123                 level++;
9124                 btrfs_release_path(&path);
9125                 goto again;
9126         }
9127
9128 del_ptr:
9129         printk("deleting pointer to block %Lu\n", corrupt->cache.start);
9130         ret = btrfs_del_ptr(trans, info->extent_root, &path, level, slot);
9131
9132 out:
9133         btrfs_release_path(&path);
9134         return ret;
9135 }
9136
9137 static int prune_corrupt_blocks(struct btrfs_fs_info *info)
9138 {
9139         struct btrfs_trans_handle *trans = NULL;
9140         struct cache_extent *cache;
9141         struct btrfs_corrupt_block *corrupt;
9142
9143         while (1) {
9144                 cache = search_cache_extent(info->corrupt_blocks, 0);
9145                 if (!cache)
9146                         break;
9147                 if (!trans) {
9148                         trans = btrfs_start_transaction(info->extent_root, 1);
9149                         if (IS_ERR(trans))
9150                                 return PTR_ERR(trans);
9151                 }
9152                 corrupt = container_of(cache, struct btrfs_corrupt_block, cache);
9153                 prune_one_block(trans, info, corrupt);
9154                 remove_cache_extent(info->corrupt_blocks, cache);
9155         }
9156         if (trans)
9157                 return btrfs_commit_transaction(trans, info->extent_root);
9158         return 0;
9159 }
9160
9161 static void reset_cached_block_groups(struct btrfs_fs_info *fs_info)
9162 {
9163         struct btrfs_block_group_cache *cache;
9164         u64 start, end;
9165         int ret;
9166
9167         while (1) {
9168                 ret = find_first_extent_bit(&fs_info->free_space_cache, 0,
9169                                             &start, &end, EXTENT_DIRTY);
9170                 if (ret)
9171                         break;
9172                 clear_extent_dirty(&fs_info->free_space_cache, start, end,
9173                                    GFP_NOFS);
9174         }
9175
9176         start = 0;
9177         while (1) {
9178                 cache = btrfs_lookup_first_block_group(fs_info, start);
9179                 if (!cache)
9180                         break;
9181                 if (cache->cached)
9182                         cache->cached = 0;
9183                 start = cache->key.objectid + cache->key.offset;
9184         }
9185 }
9186
9187 static int check_extent_refs(struct btrfs_root *root,
9188                              struct cache_tree *extent_cache)
9189 {
9190         struct extent_record *rec;
9191         struct cache_extent *cache;
9192         int err = 0;
9193         int ret = 0;
9194         int fixed = 0;
9195         int had_dups = 0;
9196         int recorded = 0;
9197
9198         if (repair) {
9199                 /*
9200                  * if we're doing a repair, we have to make sure
9201                  * we don't allocate from the problem extents.
9202                  * In the worst case, this will be all the
9203                  * extents in the FS
9204                  */
9205                 cache = search_cache_extent(extent_cache, 0);
9206                 while(cache) {
9207                         rec = container_of(cache, struct extent_record, cache);
9208                         set_extent_dirty(root->fs_info->excluded_extents,
9209                                          rec->start,
9210                                          rec->start + rec->max_size - 1,
9211                                          GFP_NOFS);
9212                         cache = next_cache_extent(cache);
9213                 }
9214
9215                 /* pin down all the corrupted blocks too */
9216                 cache = search_cache_extent(root->fs_info->corrupt_blocks, 0);
9217                 while(cache) {
9218                         set_extent_dirty(root->fs_info->excluded_extents,
9219                                          cache->start,
9220                                          cache->start + cache->size - 1,
9221                                          GFP_NOFS);
9222                         cache = next_cache_extent(cache);
9223                 }
9224                 prune_corrupt_blocks(root->fs_info);
9225                 reset_cached_block_groups(root->fs_info);
9226         }
9227
9228         reset_cached_block_groups(root->fs_info);
9229
9230         /*
9231          * We need to delete any duplicate entries we find first otherwise we
9232          * could mess up the extent tree when we have backrefs that actually
9233          * belong to a different extent item and not the weird duplicate one.
9234          */
9235         while (repair && !list_empty(&duplicate_extents)) {
9236                 rec = to_extent_record(duplicate_extents.next);
9237                 list_del_init(&rec->list);
9238
9239                 /* Sometimes we can find a backref before we find an actual
9240                  * extent, so we need to process it a little bit to see if there
9241                  * truly are multiple EXTENT_ITEM_KEY's for the same range, or
9242                  * if this is a backref screwup.  If we need to delete stuff
9243                  * process_duplicates() will return 0, otherwise it will return
9244                  * 1 and we
9245                  */
9246                 if (process_duplicates(root, extent_cache, rec))
9247                         continue;
9248                 ret = delete_duplicate_records(root, rec);
9249                 if (ret < 0)
9250                         return ret;
9251                 /*
9252                  * delete_duplicate_records will return the number of entries
9253                  * deleted, so if it's greater than 0 then we know we actually
9254                  * did something and we need to remove.
9255                  */
9256                 if (ret)
9257                         had_dups = 1;
9258         }
9259
9260         if (had_dups)
9261                 return -EAGAIN;
9262
9263         while(1) {
9264                 int cur_err = 0;
9265
9266                 fixed = 0;
9267                 recorded = 0;
9268                 cache = search_cache_extent(extent_cache, 0);
9269                 if (!cache)
9270                         break;
9271                 rec = container_of(cache, struct extent_record, cache);
9272                 if (rec->num_duplicates) {
9273                         fprintf(stderr, "extent item %llu has multiple extent "
9274                                 "items\n", (unsigned long long)rec->start);
9275                         err = 1;
9276                         cur_err = 1;
9277                 }
9278
9279                 if (rec->refs != rec->extent_item_refs) {
9280                         fprintf(stderr, "ref mismatch on [%llu %llu] ",
9281                                 (unsigned long long)rec->start,
9282                                 (unsigned long long)rec->nr);
9283                         fprintf(stderr, "extent item %llu, found %llu\n",
9284                                 (unsigned long long)rec->extent_item_refs,
9285                                 (unsigned long long)rec->refs);
9286                         ret = record_orphan_data_extents(root->fs_info, rec);
9287                         if (ret < 0)
9288                                 goto repair_abort;
9289                         if (ret == 0) {
9290                                 recorded = 1;
9291                         } else {
9292                                 /*
9293                                  * we can't use the extent to repair file
9294                                  * extent, let the fallback method handle it.
9295                                  */
9296                                 if (!fixed && repair) {
9297                                         ret = fixup_extent_refs(
9298                                                         root->fs_info,
9299                                                         extent_cache, rec);
9300                                         if (ret)
9301                                                 goto repair_abort;
9302                                         fixed = 1;
9303                                 }
9304                         }
9305                         err = 1;
9306                         cur_err = 1;
9307                 }
9308                 if (all_backpointers_checked(rec, 1)) {
9309                         fprintf(stderr, "backpointer mismatch on [%llu %llu]\n",
9310                                 (unsigned long long)rec->start,
9311                                 (unsigned long long)rec->nr);
9312
9313                         if (!fixed && !recorded && repair) {
9314                                 ret = fixup_extent_refs(root->fs_info,
9315                                                         extent_cache, rec);
9316                                 if (ret)
9317                                         goto repair_abort;
9318                                 fixed = 1;
9319                         }
9320                         cur_err = 1;
9321                         err = 1;
9322                 }
9323                 if (!rec->owner_ref_checked) {
9324                         fprintf(stderr, "owner ref check failed [%llu %llu]\n",
9325                                 (unsigned long long)rec->start,
9326                                 (unsigned long long)rec->nr);
9327                         if (!fixed && !recorded && repair) {
9328                                 ret = fixup_extent_refs(root->fs_info,
9329                                                         extent_cache, rec);
9330                                 if (ret)
9331                                         goto repair_abort;
9332                                 fixed = 1;
9333                         }
9334                         err = 1;
9335                         cur_err = 1;
9336                 }
9337                 if (rec->bad_full_backref) {
9338                         fprintf(stderr, "bad full backref, on [%llu]\n",
9339                                 (unsigned long long)rec->start);
9340                         if (repair) {
9341                                 ret = fixup_extent_flags(root->fs_info, rec);
9342                                 if (ret)
9343                                         goto repair_abort;
9344                                 fixed = 1;
9345                         }
9346                         err = 1;
9347                         cur_err = 1;
9348                 }
9349                 /*
9350                  * Although it's not a extent ref's problem, we reuse this
9351                  * routine for error reporting.
9352                  * No repair function yet.
9353                  */
9354                 if (rec->crossing_stripes) {
9355                         fprintf(stderr,
9356                                 "bad metadata [%llu, %llu) crossing stripe boundary\n",
9357                                 rec->start, rec->start + rec->max_size);
9358                         err = 1;
9359                         cur_err = 1;
9360                 }
9361
9362                 if (rec->wrong_chunk_type) {
9363                         fprintf(stderr,
9364                                 "bad extent [%llu, %llu), type mismatch with chunk\n",
9365                                 rec->start, rec->start + rec->max_size);
9366                         err = 1;
9367                         cur_err = 1;
9368                 }
9369
9370                 remove_cache_extent(extent_cache, cache);
9371                 free_all_extent_backrefs(rec);
9372                 if (!init_extent_tree && repair && (!cur_err || fixed))
9373                         clear_extent_dirty(root->fs_info->excluded_extents,
9374                                            rec->start,
9375                                            rec->start + rec->max_size - 1,
9376                                            GFP_NOFS);
9377                 free(rec);
9378         }
9379 repair_abort:
9380         if (repair) {
9381                 if (ret && ret != -EAGAIN) {
9382                         fprintf(stderr, "failed to repair damaged filesystem, aborting\n");
9383                         exit(1);
9384                 } else if (!ret) {
9385                         struct btrfs_trans_handle *trans;
9386
9387                         root = root->fs_info->extent_root;
9388                         trans = btrfs_start_transaction(root, 1);
9389                         if (IS_ERR(trans)) {
9390                                 ret = PTR_ERR(trans);
9391                                 goto repair_abort;
9392                         }
9393
9394                         btrfs_fix_block_accounting(trans, root);
9395                         ret = btrfs_commit_transaction(trans, root);
9396                         if (ret)
9397                                 goto repair_abort;
9398                 }
9399                 if (err)
9400                         fprintf(stderr, "repaired damaged extent references\n");
9401                 return ret;
9402         }
9403         return err;
9404 }
9405
9406 u64 calc_stripe_length(u64 type, u64 length, int num_stripes)
9407 {
9408         u64 stripe_size;
9409
9410         if (type & BTRFS_BLOCK_GROUP_RAID0) {
9411                 stripe_size = length;
9412                 stripe_size /= num_stripes;
9413         } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
9414                 stripe_size = length * 2;
9415                 stripe_size /= num_stripes;
9416         } else if (type & BTRFS_BLOCK_GROUP_RAID5) {
9417                 stripe_size = length;
9418                 stripe_size /= (num_stripes - 1);
9419         } else if (type & BTRFS_BLOCK_GROUP_RAID6) {
9420                 stripe_size = length;
9421                 stripe_size /= (num_stripes - 2);
9422         } else {
9423                 stripe_size = length;
9424         }
9425         return stripe_size;
9426 }
9427
9428 /*
9429  * Check the chunk with its block group/dev list ref:
9430  * Return 0 if all refs seems valid.
9431  * Return 1 if part of refs seems valid, need later check for rebuild ref
9432  * like missing block group and needs to search extent tree to rebuild them.
9433  * Return -1 if essential refs are missing and unable to rebuild.
9434  */
9435 static int check_chunk_refs(struct chunk_record *chunk_rec,
9436                             struct block_group_tree *block_group_cache,
9437                             struct device_extent_tree *dev_extent_cache,
9438                             int silent)
9439 {
9440         struct cache_extent *block_group_item;
9441         struct block_group_record *block_group_rec;
9442         struct cache_extent *dev_extent_item;
9443         struct device_extent_record *dev_extent_rec;
9444         u64 devid;
9445         u64 offset;
9446         u64 length;
9447         int metadump_v2 = 0;
9448         int i;
9449         int ret = 0;
9450
9451         block_group_item = lookup_cache_extent(&block_group_cache->tree,
9452                                                chunk_rec->offset,
9453                                                chunk_rec->length);
9454         if (block_group_item) {
9455                 block_group_rec = container_of(block_group_item,
9456                                                struct block_group_record,
9457                                                cache);
9458                 if (chunk_rec->length != block_group_rec->offset ||
9459                     chunk_rec->offset != block_group_rec->objectid ||
9460                     (!metadump_v2 &&
9461                      chunk_rec->type_flags != block_group_rec->flags)) {
9462                         if (!silent)
9463                                 fprintf(stderr,
9464                                         "Chunk[%llu, %u, %llu]: length(%llu), offset(%llu), type(%llu) mismatch with block group[%llu, %u, %llu]: offset(%llu), objectid(%llu), flags(%llu)\n",
9465                                         chunk_rec->objectid,
9466                                         chunk_rec->type,
9467                                         chunk_rec->offset,
9468                                         chunk_rec->length,
9469                                         chunk_rec->offset,
9470                                         chunk_rec->type_flags,
9471                                         block_group_rec->objectid,
9472                                         block_group_rec->type,
9473                                         block_group_rec->offset,
9474                                         block_group_rec->offset,
9475                                         block_group_rec->objectid,
9476                                         block_group_rec->flags);
9477                         ret = -1;
9478                 } else {
9479                         list_del_init(&block_group_rec->list);
9480                         chunk_rec->bg_rec = block_group_rec;
9481                 }
9482         } else {
9483                 if (!silent)
9484                         fprintf(stderr,
9485                                 "Chunk[%llu, %u, %llu]: length(%llu), offset(%llu), type(%llu) is not found in block group\n",
9486                                 chunk_rec->objectid,
9487                                 chunk_rec->type,
9488                                 chunk_rec->offset,
9489                                 chunk_rec->length,
9490                                 chunk_rec->offset,
9491                                 chunk_rec->type_flags);
9492                 ret = 1;
9493         }
9494
9495         if (metadump_v2)
9496                 return ret;
9497
9498         length = calc_stripe_length(chunk_rec->type_flags, chunk_rec->length,
9499                                     chunk_rec->num_stripes);
9500         for (i = 0; i < chunk_rec->num_stripes; ++i) {
9501                 devid = chunk_rec->stripes[i].devid;
9502                 offset = chunk_rec->stripes[i].offset;
9503                 dev_extent_item = lookup_cache_extent2(&dev_extent_cache->tree,
9504                                                        devid, offset, length);
9505                 if (dev_extent_item) {
9506                         dev_extent_rec = container_of(dev_extent_item,
9507                                                 struct device_extent_record,
9508                                                 cache);
9509                         if (dev_extent_rec->objectid != devid ||
9510                             dev_extent_rec->offset != offset ||
9511                             dev_extent_rec->chunk_offset != chunk_rec->offset ||
9512                             dev_extent_rec->length != length) {
9513                                 if (!silent)
9514                                         fprintf(stderr,
9515                                                 "Chunk[%llu, %u, %llu] stripe[%llu, %llu] dismatch dev extent[%llu, %llu, %llu]\n",
9516                                                 chunk_rec->objectid,
9517                                                 chunk_rec->type,
9518                                                 chunk_rec->offset,
9519                                                 chunk_rec->stripes[i].devid,
9520                                                 chunk_rec->stripes[i].offset,
9521                                                 dev_extent_rec->objectid,
9522                                                 dev_extent_rec->offset,
9523                                                 dev_extent_rec->length);
9524                                 ret = -1;
9525                         } else {
9526                                 list_move(&dev_extent_rec->chunk_list,
9527                                           &chunk_rec->dextents);
9528                         }
9529                 } else {
9530                         if (!silent)
9531                                 fprintf(stderr,
9532                                         "Chunk[%llu, %u, %llu] stripe[%llu, %llu] is not found in dev extent\n",
9533                                         chunk_rec->objectid,
9534                                         chunk_rec->type,
9535                                         chunk_rec->offset,
9536                                         chunk_rec->stripes[i].devid,
9537                                         chunk_rec->stripes[i].offset);
9538                         ret = -1;
9539                 }
9540         }
9541         return ret;
9542 }
9543
9544 /* check btrfs_chunk -> btrfs_dev_extent / btrfs_block_group_item */
9545 int check_chunks(struct cache_tree *chunk_cache,
9546                  struct block_group_tree *block_group_cache,
9547                  struct device_extent_tree *dev_extent_cache,
9548                  struct list_head *good, struct list_head *bad,
9549                  struct list_head *rebuild, int silent)
9550 {
9551         struct cache_extent *chunk_item;
9552         struct chunk_record *chunk_rec;
9553         struct block_group_record *bg_rec;
9554         struct device_extent_record *dext_rec;
9555         int err;
9556         int ret = 0;
9557
9558         chunk_item = first_cache_extent(chunk_cache);
9559         while (chunk_item) {
9560                 chunk_rec = container_of(chunk_item, struct chunk_record,
9561                                          cache);
9562                 err = check_chunk_refs(chunk_rec, block_group_cache,
9563                                        dev_extent_cache, silent);
9564                 if (err < 0)
9565                         ret = err;
9566                 if (err == 0 && good)
9567                         list_add_tail(&chunk_rec->list, good);
9568                 if (err > 0 && rebuild)
9569                         list_add_tail(&chunk_rec->list, rebuild);
9570                 if (err < 0 && bad)
9571                         list_add_tail(&chunk_rec->list, bad);
9572                 chunk_item = next_cache_extent(chunk_item);
9573         }
9574
9575         list_for_each_entry(bg_rec, &block_group_cache->block_groups, list) {
9576                 if (!silent)
9577                         fprintf(stderr,
9578                                 "Block group[%llu, %llu] (flags = %llu) didn't find the relative chunk.\n",
9579                                 bg_rec->objectid,
9580                                 bg_rec->offset,
9581                                 bg_rec->flags);
9582                 if (!ret)
9583                         ret = 1;
9584         }
9585
9586         list_for_each_entry(dext_rec, &dev_extent_cache->no_chunk_orphans,
9587                             chunk_list) {
9588                 if (!silent)
9589                         fprintf(stderr,
9590                                 "Device extent[%llu, %llu, %llu] didn't find the relative chunk.\n",
9591                                 dext_rec->objectid,
9592                                 dext_rec->offset,
9593                                 dext_rec->length);
9594                 if (!ret)
9595                         ret = 1;
9596         }
9597         return ret;
9598 }
9599
9600
9601 static int check_device_used(struct device_record *dev_rec,
9602                              struct device_extent_tree *dext_cache)
9603 {
9604         struct cache_extent *cache;
9605         struct device_extent_record *dev_extent_rec;
9606         u64 total_byte = 0;
9607
9608         cache = search_cache_extent2(&dext_cache->tree, dev_rec->devid, 0);
9609         while (cache) {
9610                 dev_extent_rec = container_of(cache,
9611                                               struct device_extent_record,
9612                                               cache);
9613                 if (dev_extent_rec->objectid != dev_rec->devid)
9614                         break;
9615
9616                 list_del_init(&dev_extent_rec->device_list);
9617                 total_byte += dev_extent_rec->length;
9618                 cache = next_cache_extent(cache);
9619         }
9620
9621         if (total_byte != dev_rec->byte_used) {
9622                 fprintf(stderr,
9623                         "Dev extent's total-byte(%llu) is not equal to byte-used(%llu) in dev[%llu, %u, %llu]\n",
9624                         total_byte, dev_rec->byte_used, dev_rec->objectid,
9625                         dev_rec->type, dev_rec->offset);
9626                 return -1;
9627         } else {
9628                 return 0;
9629         }
9630 }
9631
9632 /* check btrfs_dev_item -> btrfs_dev_extent */
9633 static int check_devices(struct rb_root *dev_cache,
9634                          struct device_extent_tree *dev_extent_cache)
9635 {
9636         struct rb_node *dev_node;
9637         struct device_record *dev_rec;
9638         struct device_extent_record *dext_rec;
9639         int err;
9640         int ret = 0;
9641
9642         dev_node = rb_first(dev_cache);
9643         while (dev_node) {
9644                 dev_rec = container_of(dev_node, struct device_record, node);
9645                 err = check_device_used(dev_rec, dev_extent_cache);
9646                 if (err)
9647                         ret = err;
9648
9649                 dev_node = rb_next(dev_node);
9650         }
9651         list_for_each_entry(dext_rec, &dev_extent_cache->no_device_orphans,
9652                             device_list) {
9653                 fprintf(stderr,
9654                         "Device extent[%llu, %llu, %llu] didn't find its device.\n",
9655                         dext_rec->objectid, dext_rec->offset, dext_rec->length);
9656                 if (!ret)
9657                         ret = 1;
9658         }
9659         return ret;
9660 }
9661
9662 static int add_root_item_to_list(struct list_head *head,
9663                                   u64 objectid, u64 bytenr, u64 last_snapshot,
9664                                   u8 level, u8 drop_level,
9665                                   int level_size, struct btrfs_key *drop_key)
9666 {
9667
9668         struct root_item_record *ri_rec;
9669         ri_rec = malloc(sizeof(*ri_rec));
9670         if (!ri_rec)
9671                 return -ENOMEM;
9672         ri_rec->bytenr = bytenr;
9673         ri_rec->objectid = objectid;
9674         ri_rec->level = level;
9675         ri_rec->level_size = level_size;
9676         ri_rec->drop_level = drop_level;
9677         ri_rec->last_snapshot = last_snapshot;
9678         if (drop_key)
9679                 memcpy(&ri_rec->drop_key, drop_key, sizeof(*drop_key));
9680         list_add_tail(&ri_rec->list, head);
9681
9682         return 0;
9683 }
9684
9685 static void free_root_item_list(struct list_head *list)
9686 {
9687         struct root_item_record *ri_rec;
9688
9689         while (!list_empty(list)) {
9690                 ri_rec = list_first_entry(list, struct root_item_record,
9691                                           list);
9692                 list_del_init(&ri_rec->list);
9693                 free(ri_rec);
9694         }
9695 }
9696
9697 static int deal_root_from_list(struct list_head *list,
9698                                struct btrfs_root *root,
9699                                struct block_info *bits,
9700                                int bits_nr,
9701                                struct cache_tree *pending,
9702                                struct cache_tree *seen,
9703                                struct cache_tree *reada,
9704                                struct cache_tree *nodes,
9705                                struct cache_tree *extent_cache,
9706                                struct cache_tree *chunk_cache,
9707                                struct rb_root *dev_cache,
9708                                struct block_group_tree *block_group_cache,
9709                                struct device_extent_tree *dev_extent_cache)
9710 {
9711         int ret = 0;
9712         u64 last;
9713
9714         while (!list_empty(list)) {
9715                 struct root_item_record *rec;
9716                 struct extent_buffer *buf;
9717                 rec = list_entry(list->next,
9718                                  struct root_item_record, list);
9719                 last = 0;
9720                 buf = read_tree_block(root->fs_info->tree_root,
9721                                       rec->bytenr, rec->level_size, 0);
9722                 if (!extent_buffer_uptodate(buf)) {
9723                         free_extent_buffer(buf);
9724                         ret = -EIO;
9725                         break;
9726                 }
9727                 ret = add_root_to_pending(buf, extent_cache, pending,
9728                                     seen, nodes, rec->objectid);
9729                 if (ret < 0)
9730                         break;
9731                 /*
9732                  * To rebuild extent tree, we need deal with snapshot
9733                  * one by one, otherwise we deal with node firstly which
9734                  * can maximize readahead.
9735                  */
9736                 while (1) {
9737                         ret = run_next_block(root, bits, bits_nr, &last,
9738                                              pending, seen, reada, nodes,
9739                                              extent_cache, chunk_cache,
9740                                              dev_cache, block_group_cache,
9741                                              dev_extent_cache, rec);
9742                         if (ret != 0)
9743                                 break;
9744                 }
9745                 free_extent_buffer(buf);
9746                 list_del(&rec->list);
9747                 free(rec);
9748                 if (ret < 0)
9749                         break;
9750         }
9751         while (ret >= 0) {
9752                 ret = run_next_block(root, bits, bits_nr, &last, pending, seen,
9753                                      reada, nodes, extent_cache, chunk_cache,
9754                                      dev_cache, block_group_cache,
9755                                      dev_extent_cache, NULL);
9756                 if (ret != 0) {
9757                         if (ret > 0)
9758                                 ret = 0;
9759                         break;
9760                 }
9761         }
9762         return ret;
9763 }
9764
9765 static int check_chunks_and_extents(struct btrfs_root *root)
9766 {
9767         struct rb_root dev_cache;
9768         struct cache_tree chunk_cache;
9769         struct block_group_tree block_group_cache;
9770         struct device_extent_tree dev_extent_cache;
9771         struct cache_tree extent_cache;
9772         struct cache_tree seen;
9773         struct cache_tree pending;
9774         struct cache_tree reada;
9775         struct cache_tree nodes;
9776         struct extent_io_tree excluded_extents;
9777         struct cache_tree corrupt_blocks;
9778         struct btrfs_path path;
9779         struct btrfs_key key;
9780         struct btrfs_key found_key;
9781         int ret, err = 0;
9782         struct block_info *bits;
9783         int bits_nr;
9784         struct extent_buffer *leaf;
9785         int slot;
9786         struct btrfs_root_item ri;
9787         struct list_head dropping_trees;
9788         struct list_head normal_trees;
9789         struct btrfs_root *root1;
9790         u64 objectid;
9791         u32 level_size;
9792         u8 level;
9793
9794         dev_cache = RB_ROOT;
9795         cache_tree_init(&chunk_cache);
9796         block_group_tree_init(&block_group_cache);
9797         device_extent_tree_init(&dev_extent_cache);
9798
9799         cache_tree_init(&extent_cache);
9800         cache_tree_init(&seen);
9801         cache_tree_init(&pending);
9802         cache_tree_init(&nodes);
9803         cache_tree_init(&reada);
9804         cache_tree_init(&corrupt_blocks);
9805         extent_io_tree_init(&excluded_extents);
9806         INIT_LIST_HEAD(&dropping_trees);
9807         INIT_LIST_HEAD(&normal_trees);
9808
9809         if (repair) {
9810                 root->fs_info->excluded_extents = &excluded_extents;
9811                 root->fs_info->fsck_extent_cache = &extent_cache;
9812                 root->fs_info->free_extent_hook = free_extent_hook;
9813                 root->fs_info->corrupt_blocks = &corrupt_blocks;
9814         }
9815
9816         bits_nr = 1024;
9817         bits = malloc(bits_nr * sizeof(struct block_info));
9818         if (!bits) {
9819                 perror("malloc");
9820                 exit(1);
9821         }
9822
9823         if (ctx.progress_enabled) {
9824                 ctx.tp = TASK_EXTENTS;
9825                 task_start(ctx.info);
9826         }
9827
9828 again:
9829         root1 = root->fs_info->tree_root;
9830         level = btrfs_header_level(root1->node);
9831         ret = add_root_item_to_list(&normal_trees, root1->root_key.objectid,
9832                                     root1->node->start, 0, level, 0,
9833                                     root1->nodesize, NULL);
9834         if (ret < 0)
9835                 goto out;
9836         root1 = root->fs_info->chunk_root;
9837         level = btrfs_header_level(root1->node);
9838         ret = add_root_item_to_list(&normal_trees, root1->root_key.objectid,
9839                                     root1->node->start, 0, level, 0,
9840                                     root1->nodesize, NULL);
9841         if (ret < 0)
9842                 goto out;
9843         btrfs_init_path(&path);
9844         key.offset = 0;
9845         key.objectid = 0;
9846         key.type = BTRFS_ROOT_ITEM_KEY;
9847         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
9848                                         &key, &path, 0, 0);
9849         if (ret < 0)
9850                 goto out;
9851         while(1) {
9852                 leaf = path.nodes[0];
9853                 slot = path.slots[0];
9854                 if (slot >= btrfs_header_nritems(path.nodes[0])) {
9855                         ret = btrfs_next_leaf(root, &path);
9856                         if (ret != 0)
9857                                 break;
9858                         leaf = path.nodes[0];
9859                         slot = path.slots[0];
9860                 }
9861                 btrfs_item_key_to_cpu(leaf, &found_key, path.slots[0]);
9862                 if (found_key.type == BTRFS_ROOT_ITEM_KEY) {
9863                         unsigned long offset;
9864                         u64 last_snapshot;
9865
9866                         offset = btrfs_item_ptr_offset(leaf, path.slots[0]);
9867                         read_extent_buffer(leaf, &ri, offset, sizeof(ri));
9868                         last_snapshot = btrfs_root_last_snapshot(&ri);
9869                         if (btrfs_disk_key_objectid(&ri.drop_progress) == 0) {
9870                                 level = btrfs_root_level(&ri);
9871                                 level_size = root->nodesize;
9872                                 ret = add_root_item_to_list(&normal_trees,
9873                                                 found_key.objectid,
9874                                                 btrfs_root_bytenr(&ri),
9875                                                 last_snapshot, level,
9876                                                 0, level_size, NULL);
9877                                 if (ret < 0)
9878                                         goto out;
9879                         } else {
9880                                 level = btrfs_root_level(&ri);
9881                                 level_size = root->nodesize;
9882                                 objectid = found_key.objectid;
9883                                 btrfs_disk_key_to_cpu(&found_key,
9884                                                       &ri.drop_progress);
9885                                 ret = add_root_item_to_list(&dropping_trees,
9886                                                 objectid,
9887                                                 btrfs_root_bytenr(&ri),
9888                                                 last_snapshot, level,
9889                                                 ri.drop_level,
9890                                                 level_size, &found_key);
9891                                 if (ret < 0)
9892                                         goto out;
9893                         }
9894                 }
9895                 path.slots[0]++;
9896         }
9897         btrfs_release_path(&path);
9898
9899         /*
9900          * check_block can return -EAGAIN if it fixes something, please keep
9901          * this in mind when dealing with return values from these functions, if
9902          * we get -EAGAIN we want to fall through and restart the loop.
9903          */
9904         ret = deal_root_from_list(&normal_trees, root, bits, bits_nr, &pending,
9905                                   &seen, &reada, &nodes, &extent_cache,
9906                                   &chunk_cache, &dev_cache, &block_group_cache,
9907                                   &dev_extent_cache);
9908         if (ret < 0) {
9909                 if (ret == -EAGAIN)
9910                         goto loop;
9911                 goto out;
9912         }
9913         ret = deal_root_from_list(&dropping_trees, root, bits, bits_nr,
9914                                   &pending, &seen, &reada, &nodes,
9915                                   &extent_cache, &chunk_cache, &dev_cache,
9916                                   &block_group_cache, &dev_extent_cache);
9917         if (ret < 0) {
9918                 if (ret == -EAGAIN)
9919                         goto loop;
9920                 goto out;
9921         }
9922
9923         ret = check_chunks(&chunk_cache, &block_group_cache,
9924                            &dev_extent_cache, NULL, NULL, NULL, 0);
9925         if (ret) {
9926                 if (ret == -EAGAIN)
9927                         goto loop;
9928                 err = ret;
9929         }
9930
9931         ret = check_extent_refs(root, &extent_cache);
9932         if (ret < 0) {
9933                 if (ret == -EAGAIN)
9934                         goto loop;
9935                 goto out;
9936         }
9937
9938         ret = check_devices(&dev_cache, &dev_extent_cache);
9939         if (ret && err)
9940                 ret = err;
9941
9942 out:
9943         task_stop(ctx.info);
9944         if (repair) {
9945                 free_corrupt_blocks_tree(root->fs_info->corrupt_blocks);
9946                 extent_io_tree_cleanup(&excluded_extents);
9947                 root->fs_info->fsck_extent_cache = NULL;
9948                 root->fs_info->free_extent_hook = NULL;
9949                 root->fs_info->corrupt_blocks = NULL;
9950                 root->fs_info->excluded_extents = NULL;
9951         }
9952         free(bits);
9953         free_chunk_cache_tree(&chunk_cache);
9954         free_device_cache_tree(&dev_cache);
9955         free_block_group_tree(&block_group_cache);
9956         free_device_extent_tree(&dev_extent_cache);
9957         free_extent_cache_tree(&seen);
9958         free_extent_cache_tree(&pending);
9959         free_extent_cache_tree(&reada);
9960         free_extent_cache_tree(&nodes);
9961         return ret;
9962 loop:
9963         free_corrupt_blocks_tree(root->fs_info->corrupt_blocks);
9964         free_extent_cache_tree(&seen);
9965         free_extent_cache_tree(&pending);
9966         free_extent_cache_tree(&reada);
9967         free_extent_cache_tree(&nodes);
9968         free_chunk_cache_tree(&chunk_cache);
9969         free_block_group_tree(&block_group_cache);
9970         free_device_cache_tree(&dev_cache);
9971         free_device_extent_tree(&dev_extent_cache);
9972         free_extent_record_cache(root->fs_info, &extent_cache);
9973         free_root_item_list(&normal_trees);
9974         free_root_item_list(&dropping_trees);
9975         extent_io_tree_cleanup(&excluded_extents);
9976         goto again;
9977 }
9978
9979 /*
9980  * Check backrefs of a tree block given by @bytenr or @eb.
9981  *
9982  * @root:       the root containing the @bytenr or @eb
9983  * @eb:         tree block extent buffer, can be NULL
9984  * @bytenr:     bytenr of the tree block to search
9985  * @level:      tree level of the tree block
9986  * @owner:      owner of the tree block
9987  *
9988  * Return >0 for any error found and output error message
9989  * Return 0 for no error found
9990  */
9991 static int check_tree_block_ref(struct btrfs_root *root,
9992                                 struct extent_buffer *eb, u64 bytenr,
9993                                 int level, u64 owner)
9994 {
9995         struct btrfs_key key;
9996         struct btrfs_root *extent_root = root->fs_info->extent_root;
9997         struct btrfs_path path;
9998         struct btrfs_extent_item *ei;
9999         struct btrfs_extent_inline_ref *iref;
10000         struct extent_buffer *leaf;
10001         unsigned long end;
10002         unsigned long ptr;
10003         int slot;
10004         int skinny_level;
10005         int type;
10006         u32 nodesize = root->nodesize;
10007         u32 item_size;
10008         u64 offset;
10009         int found_ref = 0;
10010         int err = 0;
10011         int ret;
10012
10013         btrfs_init_path(&path);
10014         key.objectid = bytenr;
10015         if (btrfs_fs_incompat(root->fs_info,
10016                               BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA))
10017                 key.type = BTRFS_METADATA_ITEM_KEY;
10018         else
10019                 key.type = BTRFS_EXTENT_ITEM_KEY;
10020         key.offset = (u64)-1;
10021
10022         /* Search for the backref in extent tree */
10023         ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
10024         if (ret < 0) {
10025                 err |= BACKREF_MISSING;
10026                 goto out;
10027         }
10028         ret = btrfs_previous_extent_item(extent_root, &path, bytenr);
10029         if (ret) {
10030                 err |= BACKREF_MISSING;
10031                 goto out;
10032         }
10033
10034         leaf = path.nodes[0];
10035         slot = path.slots[0];
10036         btrfs_item_key_to_cpu(leaf, &key, slot);
10037
10038         ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
10039
10040         if (key.type == BTRFS_METADATA_ITEM_KEY) {
10041                 skinny_level = (int)key.offset;
10042                 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
10043         } else {
10044                 struct btrfs_tree_block_info *info;
10045
10046                 info = (struct btrfs_tree_block_info *)(ei + 1);
10047                 skinny_level = btrfs_tree_block_level(leaf, info);
10048                 iref = (struct btrfs_extent_inline_ref *)(info + 1);
10049         }
10050
10051         if (eb) {
10052                 u64 header_gen;
10053                 u64 extent_gen;
10054
10055                 if (!(btrfs_extent_flags(leaf, ei) &
10056                       BTRFS_EXTENT_FLAG_TREE_BLOCK)) {
10057                         error(
10058                 "extent[%llu %u] backref type mismatch, missing bit: %llx",
10059                                 key.objectid, nodesize,
10060                                 BTRFS_EXTENT_FLAG_TREE_BLOCK);
10061                         err = BACKREF_MISMATCH;
10062                 }
10063                 header_gen = btrfs_header_generation(eb);
10064                 extent_gen = btrfs_extent_generation(leaf, ei);
10065                 if (header_gen != extent_gen) {
10066                         error(
10067         "extent[%llu %u] backref generation mismatch, wanted: %llu, have: %llu",
10068                                 key.objectid, nodesize, header_gen,
10069                                 extent_gen);
10070                         err = BACKREF_MISMATCH;
10071                 }
10072                 if (level != skinny_level) {
10073                         error(
10074                         "extent[%llu %u] level mismatch, wanted: %u, have: %u",
10075                                 key.objectid, nodesize, level, skinny_level);
10076                         err = BACKREF_MISMATCH;
10077                 }
10078                 if (!is_fstree(owner) && btrfs_extent_refs(leaf, ei) != 1) {
10079                         error(
10080                         "extent[%llu %u] is referred by other roots than %llu",
10081                                 key.objectid, nodesize, root->objectid);
10082                         err = BACKREF_MISMATCH;
10083                 }
10084         }
10085
10086         /*
10087          * Iterate the extent/metadata item to find the exact backref
10088          */
10089         item_size = btrfs_item_size_nr(leaf, slot);
10090         ptr = (unsigned long)iref;
10091         end = (unsigned long)ei + item_size;
10092         while (ptr < end) {
10093                 iref = (struct btrfs_extent_inline_ref *)ptr;
10094                 type = btrfs_extent_inline_ref_type(leaf, iref);
10095                 offset = btrfs_extent_inline_ref_offset(leaf, iref);
10096
10097                 if (type == BTRFS_TREE_BLOCK_REF_KEY &&
10098                         (offset == root->objectid || offset == owner)) {
10099                         found_ref = 1;
10100                 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
10101                         /* Check if the backref points to valid referencer */
10102                         found_ref = !check_tree_block_ref(root, NULL, offset,
10103                                                           level + 1, owner);
10104                 }
10105
10106                 if (found_ref)
10107                         break;
10108                 ptr += btrfs_extent_inline_ref_size(type);
10109         }
10110
10111         /*
10112          * Inlined extent item doesn't have what we need, check
10113          * TREE_BLOCK_REF_KEY
10114          */
10115         if (!found_ref) {
10116                 btrfs_release_path(&path);
10117                 key.objectid = bytenr;
10118                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
10119                 key.offset = root->objectid;
10120
10121                 ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
10122                 if (!ret)
10123                         found_ref = 1;
10124         }
10125         if (!found_ref)
10126                 err |= BACKREF_MISSING;
10127 out:
10128         btrfs_release_path(&path);
10129         if (eb && (err & BACKREF_MISSING))
10130                 error("extent[%llu %u] backref lost (owner: %llu, level: %u)",
10131                         bytenr, nodesize, owner, level);
10132         return err;
10133 }
10134
10135 /*
10136  * Check EXTENT_DATA item, mainly for its dbackref in extent tree
10137  *
10138  * Return >0 any error found and output error message
10139  * Return 0 for no error found
10140  */
10141 static int check_extent_data_item(struct btrfs_root *root,
10142                                   struct extent_buffer *eb, int slot)
10143 {
10144         struct btrfs_file_extent_item *fi;
10145         struct btrfs_path path;
10146         struct btrfs_root *extent_root = root->fs_info->extent_root;
10147         struct btrfs_key fi_key;
10148         struct btrfs_key dbref_key;
10149         struct extent_buffer *leaf;
10150         struct btrfs_extent_item *ei;
10151         struct btrfs_extent_inline_ref *iref;
10152         struct btrfs_extent_data_ref *dref;
10153         u64 owner;
10154         u64 file_extent_gen;
10155         u64 disk_bytenr;
10156         u64 disk_num_bytes;
10157         u64 extent_num_bytes;
10158         u64 extent_flags;
10159         u64 extent_gen;
10160         u32 item_size;
10161         unsigned long end;
10162         unsigned long ptr;
10163         int type;
10164         u64 ref_root;
10165         int found_dbackref = 0;
10166         int err = 0;
10167         int ret;
10168
10169         btrfs_item_key_to_cpu(eb, &fi_key, slot);
10170         fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
10171         file_extent_gen = btrfs_file_extent_generation(eb, fi);
10172
10173         /* Nothing to check for hole and inline data extents */
10174         if (btrfs_file_extent_type(eb, fi) == BTRFS_FILE_EXTENT_INLINE ||
10175             btrfs_file_extent_disk_bytenr(eb, fi) == 0)
10176                 return 0;
10177
10178         disk_bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
10179         disk_num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
10180         extent_num_bytes = btrfs_file_extent_num_bytes(eb, fi);
10181
10182         /* Check unaligned disk_num_bytes and num_bytes */
10183         if (!IS_ALIGNED(disk_num_bytes, root->sectorsize)) {
10184                 error(
10185 "file extent [%llu, %llu] has unaligned disk num bytes: %llu, should be aligned to %u",
10186                         fi_key.objectid, fi_key.offset, disk_num_bytes,
10187                         root->sectorsize);
10188                 err |= BYTES_UNALIGNED;
10189         } else {
10190                 data_bytes_allocated += disk_num_bytes;
10191         }
10192         if (!IS_ALIGNED(extent_num_bytes, root->sectorsize)) {
10193                 error(
10194 "file extent [%llu, %llu] has unaligned num bytes: %llu, should be aligned to %u",
10195                         fi_key.objectid, fi_key.offset, extent_num_bytes,
10196                         root->sectorsize);
10197                 err |= BYTES_UNALIGNED;
10198         } else {
10199                 data_bytes_referenced += extent_num_bytes;
10200         }
10201         owner = btrfs_header_owner(eb);
10202
10203         /* Check the extent item of the file extent in extent tree */
10204         btrfs_init_path(&path);
10205         dbref_key.objectid = btrfs_file_extent_disk_bytenr(eb, fi);
10206         dbref_key.type = BTRFS_EXTENT_ITEM_KEY;
10207         dbref_key.offset = btrfs_file_extent_disk_num_bytes(eb, fi);
10208
10209         ret = btrfs_search_slot(NULL, extent_root, &dbref_key, &path, 0, 0);
10210         if (ret) {
10211                 err |= BACKREF_MISSING;
10212                 goto error;
10213         }
10214
10215         leaf = path.nodes[0];
10216         slot = path.slots[0];
10217         ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
10218
10219         extent_flags = btrfs_extent_flags(leaf, ei);
10220         extent_gen = btrfs_extent_generation(leaf, ei);
10221
10222         if (!(extent_flags & BTRFS_EXTENT_FLAG_DATA)) {
10223                 error(
10224                     "extent[%llu %llu] backref type mismatch, wanted bit: %llx",
10225                     disk_bytenr, disk_num_bytes,
10226                     BTRFS_EXTENT_FLAG_DATA);
10227                 err |= BACKREF_MISMATCH;
10228         }
10229
10230         if (file_extent_gen < extent_gen) {
10231                 error(
10232 "extent[%llu %llu] backref generation mismatch, wanted: <=%llu, have: %llu",
10233                         disk_bytenr, disk_num_bytes, file_extent_gen,
10234                         extent_gen);
10235                 err |= BACKREF_MISMATCH;
10236         }
10237
10238         /* Check data backref inside that extent item */
10239         item_size = btrfs_item_size_nr(leaf, path.slots[0]);
10240         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
10241         ptr = (unsigned long)iref;
10242         end = (unsigned long)ei + item_size;
10243         while (ptr < end) {
10244                 iref = (struct btrfs_extent_inline_ref *)ptr;
10245                 type = btrfs_extent_inline_ref_type(leaf, iref);
10246                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
10247
10248                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
10249                         ref_root = btrfs_extent_data_ref_root(leaf, dref);
10250                         if (ref_root == owner || ref_root == root->objectid)
10251                                 found_dbackref = 1;
10252                 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
10253                         found_dbackref = !check_tree_block_ref(root, NULL,
10254                                 btrfs_extent_inline_ref_offset(leaf, iref),
10255                                 0, owner);
10256                 }
10257
10258                 if (found_dbackref)
10259                         break;
10260                 ptr += btrfs_extent_inline_ref_size(type);
10261         }
10262
10263         /* Didn't found inlined data backref, try EXTENT_DATA_REF_KEY */
10264         if (!found_dbackref) {
10265                 btrfs_release_path(&path);
10266
10267                 btrfs_init_path(&path);
10268                 dbref_key.objectid = btrfs_file_extent_disk_bytenr(eb, fi);
10269                 dbref_key.type = BTRFS_EXTENT_DATA_REF_KEY;
10270                 dbref_key.offset = hash_extent_data_ref(root->objectid,
10271                                 fi_key.objectid, fi_key.offset);
10272
10273                 ret = btrfs_search_slot(NULL, root->fs_info->extent_root,
10274                                         &dbref_key, &path, 0, 0);
10275                 if (!ret)
10276                         found_dbackref = 1;
10277         }
10278
10279         if (!found_dbackref)
10280                 err |= BACKREF_MISSING;
10281 error:
10282         btrfs_release_path(&path);
10283         if (err & BACKREF_MISSING) {
10284                 error("data extent[%llu %llu] backref lost",
10285                       disk_bytenr, disk_num_bytes);
10286         }
10287         return err;
10288 }
10289
10290 /*
10291  * Get real tree block level for the case like shared block
10292  * Return >= 0 as tree level
10293  * Return <0 for error
10294  */
10295 static int query_tree_block_level(struct btrfs_fs_info *fs_info, u64 bytenr)
10296 {
10297         struct extent_buffer *eb;
10298         struct btrfs_path path;
10299         struct btrfs_key key;
10300         struct btrfs_extent_item *ei;
10301         u64 flags;
10302         u64 transid;
10303         u32 nodesize = btrfs_super_nodesize(fs_info->super_copy);
10304         u8 backref_level;
10305         u8 header_level;
10306         int ret;
10307
10308         /* Search extent tree for extent generation and level */
10309         key.objectid = bytenr;
10310         key.type = BTRFS_METADATA_ITEM_KEY;
10311         key.offset = (u64)-1;
10312
10313         btrfs_init_path(&path);
10314         ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, &path, 0, 0);
10315         if (ret < 0)
10316                 goto release_out;
10317         ret = btrfs_previous_extent_item(fs_info->extent_root, &path, bytenr);
10318         if (ret < 0)
10319                 goto release_out;
10320         if (ret > 0) {
10321                 ret = -ENOENT;
10322                 goto release_out;
10323         }
10324
10325         btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
10326         ei = btrfs_item_ptr(path.nodes[0], path.slots[0],
10327                             struct btrfs_extent_item);
10328         flags = btrfs_extent_flags(path.nodes[0], ei);
10329         if (!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) {
10330                 ret = -ENOENT;
10331                 goto release_out;
10332         }
10333
10334         /* Get transid for later read_tree_block() check */
10335         transid = btrfs_extent_generation(path.nodes[0], ei);
10336
10337         /* Get backref level as one source */
10338         if (key.type == BTRFS_METADATA_ITEM_KEY) {
10339                 backref_level = key.offset;
10340         } else {
10341                 struct btrfs_tree_block_info *info;
10342
10343                 info = (struct btrfs_tree_block_info *)(ei + 1);
10344                 backref_level = btrfs_tree_block_level(path.nodes[0], info);
10345         }
10346         btrfs_release_path(&path);
10347
10348         /* Get level from tree block as an alternative source */
10349         eb = read_tree_block_fs_info(fs_info, bytenr, nodesize, transid);
10350         if (!extent_buffer_uptodate(eb)) {
10351                 free_extent_buffer(eb);
10352                 return -EIO;
10353         }
10354         header_level = btrfs_header_level(eb);
10355         free_extent_buffer(eb);
10356
10357         if (header_level != backref_level)
10358                 return -EIO;
10359         return header_level;
10360
10361 release_out:
10362         btrfs_release_path(&path);
10363         return ret;
10364 }
10365
10366 /*
10367  * Check if a tree block backref is valid (points to a valid tree block)
10368  * if level == -1, level will be resolved
10369  * Return >0 for any error found and print error message
10370  */
10371 static int check_tree_block_backref(struct btrfs_fs_info *fs_info, u64 root_id,
10372                                     u64 bytenr, int level)
10373 {
10374         struct btrfs_root *root;
10375         struct btrfs_key key;
10376         struct btrfs_path path;
10377         struct extent_buffer *eb;
10378         struct extent_buffer *node;
10379         u32 nodesize = btrfs_super_nodesize(fs_info->super_copy);
10380         int err = 0;
10381         int ret;
10382
10383         /* Query level for level == -1 special case */
10384         if (level == -1)
10385                 level = query_tree_block_level(fs_info, bytenr);
10386         if (level < 0) {
10387                 err |= REFERENCER_MISSING;
10388                 goto out;
10389         }
10390
10391         key.objectid = root_id;
10392         key.type = BTRFS_ROOT_ITEM_KEY;
10393         key.offset = (u64)-1;
10394
10395         root = btrfs_read_fs_root(fs_info, &key);
10396         if (IS_ERR(root)) {
10397                 err |= REFERENCER_MISSING;
10398                 goto out;
10399         }
10400
10401         /* Read out the tree block to get item/node key */
10402         eb = read_tree_block(root, bytenr, root->nodesize, 0);
10403         if (!extent_buffer_uptodate(eb)) {
10404                 err |= REFERENCER_MISSING;
10405                 free_extent_buffer(eb);
10406                 goto out;
10407         }
10408
10409         /* Empty tree, no need to check key */
10410         if (!btrfs_header_nritems(eb) && !level) {
10411                 free_extent_buffer(eb);
10412                 goto out;
10413         }
10414
10415         if (level)
10416                 btrfs_node_key_to_cpu(eb, &key, 0);
10417         else
10418                 btrfs_item_key_to_cpu(eb, &key, 0);
10419
10420         free_extent_buffer(eb);
10421
10422         btrfs_init_path(&path);
10423         path.lowest_level = level;
10424         /* Search with the first key, to ensure we can reach it */
10425         ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
10426         if (ret < 0) {
10427                 err |= REFERENCER_MISSING;
10428                 goto release_out;
10429         }
10430
10431         node = path.nodes[level];
10432         if (btrfs_header_bytenr(node) != bytenr) {
10433                 error(
10434         "extent [%llu %d] referencer bytenr mismatch, wanted: %llu, have: %llu",
10435                         bytenr, nodesize, bytenr,
10436                         btrfs_header_bytenr(node));
10437                 err |= REFERENCER_MISMATCH;
10438         }
10439         if (btrfs_header_level(node) != level) {
10440                 error(
10441         "extent [%llu %d] referencer level mismatch, wanted: %d, have: %d",
10442                         bytenr, nodesize, level,
10443                         btrfs_header_level(node));
10444                 err |= REFERENCER_MISMATCH;
10445         }
10446
10447 release_out:
10448         btrfs_release_path(&path);
10449 out:
10450         if (err & REFERENCER_MISSING) {
10451                 if (level < 0)
10452                         error("extent [%llu %d] lost referencer (owner: %llu)",
10453                                 bytenr, nodesize, root_id);
10454                 else
10455                         error(
10456                 "extent [%llu %d] lost referencer (owner: %llu, level: %u)",
10457                                 bytenr, nodesize, root_id, level);
10458         }
10459
10460         return err;
10461 }
10462
10463 /*
10464  * Check referencer for shared block backref
10465  * If level == -1, this function will resolve the level.
10466  */
10467 static int check_shared_block_backref(struct btrfs_fs_info *fs_info,
10468                                      u64 parent, u64 bytenr, int level)
10469 {
10470         struct extent_buffer *eb;
10471         u32 nodesize = btrfs_super_nodesize(fs_info->super_copy);
10472         u32 nr;
10473         int found_parent = 0;
10474         int i;
10475
10476         eb = read_tree_block_fs_info(fs_info, parent, nodesize, 0);
10477         if (!extent_buffer_uptodate(eb))
10478                 goto out;
10479
10480         if (level == -1)
10481                 level = query_tree_block_level(fs_info, bytenr);
10482         if (level < 0)
10483                 goto out;
10484
10485         if (level + 1 != btrfs_header_level(eb))
10486                 goto out;
10487
10488         nr = btrfs_header_nritems(eb);
10489         for (i = 0; i < nr; i++) {
10490                 if (bytenr == btrfs_node_blockptr(eb, i)) {
10491                         found_parent = 1;
10492                         break;
10493                 }
10494         }
10495 out:
10496         free_extent_buffer(eb);
10497         if (!found_parent) {
10498                 error(
10499         "shared extent[%llu %u] lost its parent (parent: %llu, level: %u)",
10500                         bytenr, nodesize, parent, level);
10501                 return REFERENCER_MISSING;
10502         }
10503         return 0;
10504 }
10505
10506 /*
10507  * Check referencer for normal (inlined) data ref
10508  * If len == 0, it will be resolved by searching in extent tree
10509  */
10510 static int check_extent_data_backref(struct btrfs_fs_info *fs_info,
10511                                      u64 root_id, u64 objectid, u64 offset,
10512                                      u64 bytenr, u64 len, u32 count)
10513 {
10514         struct btrfs_root *root;
10515         struct btrfs_root *extent_root = fs_info->extent_root;
10516         struct btrfs_key key;
10517         struct btrfs_path path;
10518         struct extent_buffer *leaf;
10519         struct btrfs_file_extent_item *fi;
10520         u32 found_count = 0;
10521         int slot;
10522         int ret = 0;
10523
10524         if (!len) {
10525                 key.objectid = bytenr;
10526                 key.type = BTRFS_EXTENT_ITEM_KEY;
10527                 key.offset = (u64)-1;
10528
10529                 btrfs_init_path(&path);
10530                 ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
10531                 if (ret < 0)
10532                         goto out;
10533                 ret = btrfs_previous_extent_item(extent_root, &path, bytenr);
10534                 if (ret)
10535                         goto out;
10536                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
10537                 if (key.objectid != bytenr ||
10538                     key.type != BTRFS_EXTENT_ITEM_KEY)
10539                         goto out;
10540                 len = key.offset;
10541                 btrfs_release_path(&path);
10542         }
10543         key.objectid = root_id;
10544         key.type = BTRFS_ROOT_ITEM_KEY;
10545         key.offset = (u64)-1;
10546         btrfs_init_path(&path);
10547
10548         root = btrfs_read_fs_root(fs_info, &key);
10549         if (IS_ERR(root))
10550                 goto out;
10551
10552         key.objectid = objectid;
10553         key.type = BTRFS_EXTENT_DATA_KEY;
10554         /*
10555          * It can be nasty as data backref offset is
10556          * file offset - file extent offset, which is smaller or
10557          * equal to original backref offset.  The only special case is
10558          * overflow.  So we need to special check and do further search.
10559          */
10560         key.offset = offset & (1ULL << 63) ? 0 : offset;
10561
10562         ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
10563         if (ret < 0)
10564                 goto out;
10565
10566         /*
10567          * Search afterwards to get correct one
10568          * NOTE: As we must do a comprehensive check on the data backref to
10569          * make sure the dref count also matches, we must iterate all file
10570          * extents for that inode.
10571          */
10572         while (1) {
10573                 leaf = path.nodes[0];
10574                 slot = path.slots[0];
10575
10576                 btrfs_item_key_to_cpu(leaf, &key, slot);
10577                 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
10578                         break;
10579                 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
10580                 /*
10581                  * Except normal disk bytenr and disk num bytes, we still
10582                  * need to do extra check on dbackref offset as
10583                  * dbackref offset = file_offset - file_extent_offset
10584                  */
10585                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == bytenr &&
10586                     btrfs_file_extent_disk_num_bytes(leaf, fi) == len &&
10587                     (u64)(key.offset - btrfs_file_extent_offset(leaf, fi)) ==
10588                     offset)
10589                         found_count++;
10590
10591                 ret = btrfs_next_item(root, &path);
10592                 if (ret)
10593                         break;
10594         }
10595 out:
10596         btrfs_release_path(&path);
10597         if (found_count != count) {
10598                 error(
10599 "extent[%llu, %llu] referencer count mismatch (root: %llu, owner: %llu, offset: %llu) wanted: %u, have: %u",
10600                         bytenr, len, root_id, objectid, offset, count, found_count);
10601                 return REFERENCER_MISSING;
10602         }
10603         return 0;
10604 }
10605
10606 /*
10607  * Check if the referencer of a shared data backref exists
10608  */
10609 static int check_shared_data_backref(struct btrfs_fs_info *fs_info,
10610                                      u64 parent, u64 bytenr)
10611 {
10612         struct extent_buffer *eb;
10613         struct btrfs_key key;
10614         struct btrfs_file_extent_item *fi;
10615         u32 nodesize = btrfs_super_nodesize(fs_info->super_copy);
10616         u32 nr;
10617         int found_parent = 0;
10618         int i;
10619
10620         eb = read_tree_block_fs_info(fs_info, parent, nodesize, 0);
10621         if (!extent_buffer_uptodate(eb))
10622                 goto out;
10623
10624         nr = btrfs_header_nritems(eb);
10625         for (i = 0; i < nr; i++) {
10626                 btrfs_item_key_to_cpu(eb, &key, i);
10627                 if (key.type != BTRFS_EXTENT_DATA_KEY)
10628                         continue;
10629
10630                 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
10631                 if (btrfs_file_extent_type(eb, fi) == BTRFS_FILE_EXTENT_INLINE)
10632                         continue;
10633
10634                 if (btrfs_file_extent_disk_bytenr(eb, fi) == bytenr) {
10635                         found_parent = 1;
10636                         break;
10637                 }
10638         }
10639
10640 out:
10641         free_extent_buffer(eb);
10642         if (!found_parent) {
10643                 error("shared extent %llu referencer lost (parent: %llu)",
10644                         bytenr, parent);
10645                 return REFERENCER_MISSING;
10646         }
10647         return 0;
10648 }
10649
10650 /*
10651  * This function will check a given extent item, including its backref and
10652  * itself (like crossing stripe boundary and type)
10653  *
10654  * Since we don't use extent_record anymore, introduce new error bit
10655  */
10656 static int check_extent_item(struct btrfs_fs_info *fs_info,
10657                              struct extent_buffer *eb, int slot)
10658 {
10659         struct btrfs_extent_item *ei;
10660         struct btrfs_extent_inline_ref *iref;
10661         struct btrfs_extent_data_ref *dref;
10662         unsigned long end;
10663         unsigned long ptr;
10664         int type;
10665         u32 nodesize = btrfs_super_nodesize(fs_info->super_copy);
10666         u32 item_size = btrfs_item_size_nr(eb, slot);
10667         u64 flags;
10668         u64 offset;
10669         int metadata = 0;
10670         int level;
10671         struct btrfs_key key;
10672         int ret;
10673         int err = 0;
10674
10675         btrfs_item_key_to_cpu(eb, &key, slot);
10676         if (key.type == BTRFS_EXTENT_ITEM_KEY)
10677                 bytes_used += key.offset;
10678         else
10679                 bytes_used += nodesize;
10680
10681         if (item_size < sizeof(*ei)) {
10682                 /*
10683                  * COMPAT_EXTENT_TREE_V0 case, but it's already a super
10684                  * old thing when on disk format is still un-determined.
10685                  * No need to care about it anymore
10686                  */
10687                 error("unsupported COMPAT_EXTENT_TREE_V0 detected");
10688                 return -ENOTTY;
10689         }
10690
10691         ei = btrfs_item_ptr(eb, slot, struct btrfs_extent_item);
10692         flags = btrfs_extent_flags(eb, ei);
10693
10694         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
10695                 metadata = 1;
10696         if (metadata && check_crossing_stripes(global_info, key.objectid,
10697                                                eb->len)) {
10698                 error("bad metadata [%llu, %llu) crossing stripe boundary",
10699                       key.objectid, key.objectid + nodesize);
10700                 err |= CROSSING_STRIPE_BOUNDARY;
10701         }
10702
10703         ptr = (unsigned long)(ei + 1);
10704
10705         if (metadata && key.type == BTRFS_EXTENT_ITEM_KEY) {
10706                 /* Old EXTENT_ITEM metadata */
10707                 struct btrfs_tree_block_info *info;
10708
10709                 info = (struct btrfs_tree_block_info *)ptr;
10710                 level = btrfs_tree_block_level(eb, info);
10711                 ptr += sizeof(struct btrfs_tree_block_info);
10712         } else {
10713                 /* New METADATA_ITEM */
10714                 level = key.offset;
10715         }
10716         end = (unsigned long)ei + item_size;
10717
10718         if (ptr >= end) {
10719                 err |= ITEM_SIZE_MISMATCH;
10720                 goto out;
10721         }
10722
10723         /* Now check every backref in this extent item */
10724 next:
10725         iref = (struct btrfs_extent_inline_ref *)ptr;
10726         type = btrfs_extent_inline_ref_type(eb, iref);
10727         offset = btrfs_extent_inline_ref_offset(eb, iref);
10728         switch (type) {
10729         case BTRFS_TREE_BLOCK_REF_KEY:
10730                 ret = check_tree_block_backref(fs_info, offset, key.objectid,
10731                                                level);
10732                 err |= ret;
10733                 break;
10734         case BTRFS_SHARED_BLOCK_REF_KEY:
10735                 ret = check_shared_block_backref(fs_info, offset, key.objectid,
10736                                                  level);
10737                 err |= ret;
10738                 break;
10739         case BTRFS_EXTENT_DATA_REF_KEY:
10740                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
10741                 ret = check_extent_data_backref(fs_info,
10742                                 btrfs_extent_data_ref_root(eb, dref),
10743                                 btrfs_extent_data_ref_objectid(eb, dref),
10744                                 btrfs_extent_data_ref_offset(eb, dref),
10745                                 key.objectid, key.offset,
10746                                 btrfs_extent_data_ref_count(eb, dref));
10747                 err |= ret;
10748                 break;
10749         case BTRFS_SHARED_DATA_REF_KEY:
10750                 ret = check_shared_data_backref(fs_info, offset, key.objectid);
10751                 err |= ret;
10752                 break;
10753         default:
10754                 error("extent[%llu %d %llu] has unknown ref type: %d",
10755                         key.objectid, key.type, key.offset, type);
10756                 err |= UNKNOWN_TYPE;
10757                 goto out;
10758         }
10759
10760         ptr += btrfs_extent_inline_ref_size(type);
10761         if (ptr < end)
10762                 goto next;
10763
10764 out:
10765         return err;
10766 }
10767
10768 /*
10769  * Check if a dev extent item is referred correctly by its chunk
10770  */
10771 static int check_dev_extent_item(struct btrfs_fs_info *fs_info,
10772                                  struct extent_buffer *eb, int slot)
10773 {
10774         struct btrfs_root *chunk_root = fs_info->chunk_root;
10775         struct btrfs_dev_extent *ptr;
10776         struct btrfs_path path;
10777         struct btrfs_key chunk_key;
10778         struct btrfs_key devext_key;
10779         struct btrfs_chunk *chunk;
10780         struct extent_buffer *l;
10781         int num_stripes;
10782         u64 length;
10783         int i;
10784         int found_chunk = 0;
10785         int ret;
10786
10787         btrfs_item_key_to_cpu(eb, &devext_key, slot);
10788         ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_extent);
10789         length = btrfs_dev_extent_length(eb, ptr);
10790
10791         chunk_key.objectid = btrfs_dev_extent_chunk_objectid(eb, ptr);
10792         chunk_key.type = BTRFS_CHUNK_ITEM_KEY;
10793         chunk_key.offset = btrfs_dev_extent_chunk_offset(eb, ptr);
10794
10795         btrfs_init_path(&path);
10796         ret = btrfs_search_slot(NULL, chunk_root, &chunk_key, &path, 0, 0);
10797         if (ret)
10798                 goto out;
10799
10800         l = path.nodes[0];
10801         chunk = btrfs_item_ptr(l, path.slots[0], struct btrfs_chunk);
10802         if (btrfs_chunk_length(l, chunk) != length)
10803                 goto out;
10804
10805         num_stripes = btrfs_chunk_num_stripes(l, chunk);
10806         for (i = 0; i < num_stripes; i++) {
10807                 u64 devid = btrfs_stripe_devid_nr(l, chunk, i);
10808                 u64 offset = btrfs_stripe_offset_nr(l, chunk, i);
10809
10810                 if (devid == devext_key.objectid &&
10811                     offset == devext_key.offset) {
10812                         found_chunk = 1;
10813                         break;
10814                 }
10815         }
10816 out:
10817         btrfs_release_path(&path);
10818         if (!found_chunk) {
10819                 error(
10820                 "device extent[%llu, %llu, %llu] did not find the related chunk",
10821                         devext_key.objectid, devext_key.offset, length);
10822                 return REFERENCER_MISSING;
10823         }
10824         return 0;
10825 }
10826
10827 /*
10828  * Check if the used space is correct with the dev item
10829  */
10830 static int check_dev_item(struct btrfs_fs_info *fs_info,
10831                           struct extent_buffer *eb, int slot)
10832 {
10833         struct btrfs_root *dev_root = fs_info->dev_root;
10834         struct btrfs_dev_item *dev_item;
10835         struct btrfs_path path;
10836         struct btrfs_key key;
10837         struct btrfs_dev_extent *ptr;
10838         u64 dev_id;
10839         u64 used;
10840         u64 total = 0;
10841         int ret;
10842
10843         dev_item = btrfs_item_ptr(eb, slot, struct btrfs_dev_item);
10844         dev_id = btrfs_device_id(eb, dev_item);
10845         used = btrfs_device_bytes_used(eb, dev_item);
10846
10847         key.objectid = dev_id;
10848         key.type = BTRFS_DEV_EXTENT_KEY;
10849         key.offset = 0;
10850
10851         btrfs_init_path(&path);
10852         ret = btrfs_search_slot(NULL, dev_root, &key, &path, 0, 0);
10853         if (ret < 0) {
10854                 btrfs_item_key_to_cpu(eb, &key, slot);
10855                 error("cannot find any related dev extent for dev[%llu, %u, %llu]",
10856                         key.objectid, key.type, key.offset);
10857                 btrfs_release_path(&path);
10858                 return REFERENCER_MISSING;
10859         }
10860
10861         /* Iterate dev_extents to calculate the used space of a device */
10862         while (1) {
10863                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
10864
10865                 if (key.objectid > dev_id)
10866                         break;
10867                 if (key.type != BTRFS_DEV_EXTENT_KEY || key.objectid != dev_id)
10868                         goto next;
10869
10870                 ptr = btrfs_item_ptr(path.nodes[0], path.slots[0],
10871                                      struct btrfs_dev_extent);
10872                 total += btrfs_dev_extent_length(path.nodes[0], ptr);
10873 next:
10874                 ret = btrfs_next_item(dev_root, &path);
10875                 if (ret)
10876                         break;
10877         }
10878         btrfs_release_path(&path);
10879
10880         if (used != total) {
10881                 btrfs_item_key_to_cpu(eb, &key, slot);
10882                 error(
10883 "Dev extent's total-byte %llu is not equal to bytes-used %llu in dev[%llu, %u, %llu]",
10884                         total, used, BTRFS_ROOT_TREE_OBJECTID,
10885                         BTRFS_DEV_EXTENT_KEY, dev_id);
10886                 return ACCOUNTING_MISMATCH;
10887         }
10888         return 0;
10889 }
10890
10891 /*
10892  * Check a block group item with its referener (chunk) and its used space
10893  * with extent/metadata item
10894  */
10895 static int check_block_group_item(struct btrfs_fs_info *fs_info,
10896                                   struct extent_buffer *eb, int slot)
10897 {
10898         struct btrfs_root *extent_root = fs_info->extent_root;
10899         struct btrfs_root *chunk_root = fs_info->chunk_root;
10900         struct btrfs_block_group_item *bi;
10901         struct btrfs_block_group_item bg_item;
10902         struct btrfs_path path;
10903         struct btrfs_key bg_key;
10904         struct btrfs_key chunk_key;
10905         struct btrfs_key extent_key;
10906         struct btrfs_chunk *chunk;
10907         struct extent_buffer *leaf;
10908         struct btrfs_extent_item *ei;
10909         u32 nodesize = btrfs_super_nodesize(fs_info->super_copy);
10910         u64 flags;
10911         u64 bg_flags;
10912         u64 used;
10913         u64 total = 0;
10914         int ret;
10915         int err = 0;
10916
10917         btrfs_item_key_to_cpu(eb, &bg_key, slot);
10918         bi = btrfs_item_ptr(eb, slot, struct btrfs_block_group_item);
10919         read_extent_buffer(eb, &bg_item, (unsigned long)bi, sizeof(bg_item));
10920         used = btrfs_block_group_used(&bg_item);
10921         bg_flags = btrfs_block_group_flags(&bg_item);
10922
10923         chunk_key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
10924         chunk_key.type = BTRFS_CHUNK_ITEM_KEY;
10925         chunk_key.offset = bg_key.objectid;
10926
10927         btrfs_init_path(&path);
10928         /* Search for the referencer chunk */
10929         ret = btrfs_search_slot(NULL, chunk_root, &chunk_key, &path, 0, 0);
10930         if (ret) {
10931                 error(
10932                 "block group[%llu %llu] did not find the related chunk item",
10933                         bg_key.objectid, bg_key.offset);
10934                 err |= REFERENCER_MISSING;
10935         } else {
10936                 chunk = btrfs_item_ptr(path.nodes[0], path.slots[0],
10937                                         struct btrfs_chunk);
10938                 if (btrfs_chunk_length(path.nodes[0], chunk) !=
10939                                                 bg_key.offset) {
10940                         error(
10941         "block group[%llu %llu] related chunk item length does not match",
10942                                 bg_key.objectid, bg_key.offset);
10943                         err |= REFERENCER_MISMATCH;
10944                 }
10945         }
10946         btrfs_release_path(&path);
10947
10948         /* Search from the block group bytenr */
10949         extent_key.objectid = bg_key.objectid;
10950         extent_key.type = 0;
10951         extent_key.offset = 0;
10952
10953         btrfs_init_path(&path);
10954         ret = btrfs_search_slot(NULL, extent_root, &extent_key, &path, 0, 0);
10955         if (ret < 0)
10956                 goto out;
10957
10958         /* Iterate extent tree to account used space */
10959         while (1) {
10960                 leaf = path.nodes[0];
10961                 btrfs_item_key_to_cpu(leaf, &extent_key, path.slots[0]);
10962                 if (extent_key.objectid >= bg_key.objectid + bg_key.offset)
10963                         break;
10964
10965                 if (extent_key.type != BTRFS_METADATA_ITEM_KEY &&
10966                     extent_key.type != BTRFS_EXTENT_ITEM_KEY)
10967                         goto next;
10968                 if (extent_key.objectid < bg_key.objectid)
10969                         goto next;
10970
10971                 if (extent_key.type == BTRFS_METADATA_ITEM_KEY)
10972                         total += nodesize;
10973                 else
10974                         total += extent_key.offset;
10975
10976                 ei = btrfs_item_ptr(leaf, path.slots[0],
10977                                     struct btrfs_extent_item);
10978                 flags = btrfs_extent_flags(leaf, ei);
10979                 if (flags & BTRFS_EXTENT_FLAG_DATA) {
10980                         if (!(bg_flags & BTRFS_BLOCK_GROUP_DATA)) {
10981                                 error(
10982                         "bad extent[%llu, %llu) type mismatch with chunk",
10983                                         extent_key.objectid,
10984                                         extent_key.objectid + extent_key.offset);
10985                                 err |= CHUNK_TYPE_MISMATCH;
10986                         }
10987                 } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
10988                         if (!(bg_flags & (BTRFS_BLOCK_GROUP_SYSTEM |
10989                                     BTRFS_BLOCK_GROUP_METADATA))) {
10990                                 error(
10991                         "bad extent[%llu, %llu) type mismatch with chunk",
10992                                         extent_key.objectid,
10993                                         extent_key.objectid + nodesize);
10994                                 err |= CHUNK_TYPE_MISMATCH;
10995                         }
10996                 }
10997 next:
10998                 ret = btrfs_next_item(extent_root, &path);
10999                 if (ret)
11000                         break;
11001         }
11002
11003 out:
11004         btrfs_release_path(&path);
11005
11006         if (total != used) {
11007                 error(
11008                 "block group[%llu %llu] used %llu but extent items used %llu",
11009                         bg_key.objectid, bg_key.offset, used, total);
11010                 err |= ACCOUNTING_MISMATCH;
11011         }
11012         return err;
11013 }
11014
11015 /*
11016  * Check a chunk item.
11017  * Including checking all referred dev_extents and block group
11018  */
11019 static int check_chunk_item(struct btrfs_fs_info *fs_info,
11020                             struct extent_buffer *eb, int slot)
11021 {
11022         struct btrfs_root *extent_root = fs_info->extent_root;
11023         struct btrfs_root *dev_root = fs_info->dev_root;
11024         struct btrfs_path path;
11025         struct btrfs_key chunk_key;
11026         struct btrfs_key bg_key;
11027         struct btrfs_key devext_key;
11028         struct btrfs_chunk *chunk;
11029         struct extent_buffer *leaf;
11030         struct btrfs_block_group_item *bi;
11031         struct btrfs_block_group_item bg_item;
11032         struct btrfs_dev_extent *ptr;
11033         u32 sectorsize = btrfs_super_sectorsize(fs_info->super_copy);
11034         u64 length;
11035         u64 chunk_end;
11036         u64 type;
11037         u64 profile;
11038         int num_stripes;
11039         u64 offset;
11040         u64 objectid;
11041         int i;
11042         int ret;
11043         int err = 0;
11044
11045         btrfs_item_key_to_cpu(eb, &chunk_key, slot);
11046         chunk = btrfs_item_ptr(eb, slot, struct btrfs_chunk);
11047         length = btrfs_chunk_length(eb, chunk);
11048         chunk_end = chunk_key.offset + length;
11049         if (!IS_ALIGNED(length, sectorsize)) {
11050                 error("chunk[%llu %llu) not aligned to %u",
11051                         chunk_key.offset, chunk_end, sectorsize);
11052                 err |= BYTES_UNALIGNED;
11053                 goto out;
11054         }
11055
11056         type = btrfs_chunk_type(eb, chunk);
11057         profile = type & BTRFS_BLOCK_GROUP_PROFILE_MASK;
11058         if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
11059                 error("chunk[%llu %llu) has no chunk type",
11060                         chunk_key.offset, chunk_end);
11061                 err |= UNKNOWN_TYPE;
11062         }
11063         if (profile && (profile & (profile - 1))) {
11064                 error("chunk[%llu %llu) multiple profiles detected: %llx",
11065                         chunk_key.offset, chunk_end, profile);
11066                 err |= UNKNOWN_TYPE;
11067         }
11068
11069         bg_key.objectid = chunk_key.offset;
11070         bg_key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
11071         bg_key.offset = length;
11072
11073         btrfs_init_path(&path);
11074         ret = btrfs_search_slot(NULL, extent_root, &bg_key, &path, 0, 0);
11075         if (ret) {
11076                 error(
11077                 "chunk[%llu %llu) did not find the related block group item",
11078                         chunk_key.offset, chunk_end);
11079                 err |= REFERENCER_MISSING;
11080         } else{
11081                 leaf = path.nodes[0];
11082                 bi = btrfs_item_ptr(leaf, path.slots[0],
11083                                     struct btrfs_block_group_item);
11084                 read_extent_buffer(leaf, &bg_item, (unsigned long)bi,
11085                                    sizeof(bg_item));
11086                 if (btrfs_block_group_flags(&bg_item) != type) {
11087                         error(
11088 "chunk[%llu %llu) related block group item flags mismatch, wanted: %llu, have: %llu",
11089                                 chunk_key.offset, chunk_end, type,
11090                                 btrfs_block_group_flags(&bg_item));
11091                         err |= REFERENCER_MISSING;
11092                 }
11093         }
11094
11095         num_stripes = btrfs_chunk_num_stripes(eb, chunk);
11096         for (i = 0; i < num_stripes; i++) {
11097                 btrfs_release_path(&path);
11098                 btrfs_init_path(&path);
11099                 devext_key.objectid = btrfs_stripe_devid_nr(eb, chunk, i);
11100                 devext_key.type = BTRFS_DEV_EXTENT_KEY;
11101                 devext_key.offset = btrfs_stripe_offset_nr(eb, chunk, i);
11102
11103                 ret = btrfs_search_slot(NULL, dev_root, &devext_key, &path,
11104                                         0, 0);
11105                 if (ret)
11106                         goto not_match_dev;
11107
11108                 leaf = path.nodes[0];
11109                 ptr = btrfs_item_ptr(leaf, path.slots[0],
11110                                      struct btrfs_dev_extent);
11111                 objectid = btrfs_dev_extent_chunk_objectid(leaf, ptr);
11112                 offset = btrfs_dev_extent_chunk_offset(leaf, ptr);
11113                 if (objectid != chunk_key.objectid ||
11114                     offset != chunk_key.offset ||
11115                     btrfs_dev_extent_length(leaf, ptr) != length)
11116                         goto not_match_dev;
11117                 continue;
11118 not_match_dev:
11119                 err |= BACKREF_MISSING;
11120                 error(
11121                 "chunk[%llu %llu) stripe %d did not find the related dev extent",
11122                         chunk_key.objectid, chunk_end, i);
11123                 continue;
11124         }
11125         btrfs_release_path(&path);
11126 out:
11127         return err;
11128 }
11129
11130 /*
11131  * Main entry function to check known items and update related accounting info
11132  */
11133 static int check_leaf_items(struct btrfs_root *root, struct extent_buffer *eb)
11134 {
11135         struct btrfs_fs_info *fs_info = root->fs_info;
11136         struct btrfs_key key;
11137         int slot = 0;
11138         int type;
11139         struct btrfs_extent_data_ref *dref;
11140         int ret;
11141         int err = 0;
11142
11143 next:
11144         btrfs_item_key_to_cpu(eb, &key, slot);
11145         type = key.type;
11146
11147         switch (type) {
11148         case BTRFS_EXTENT_DATA_KEY:
11149                 ret = check_extent_data_item(root, eb, slot);
11150                 err |= ret;
11151                 break;
11152         case BTRFS_BLOCK_GROUP_ITEM_KEY:
11153                 ret = check_block_group_item(fs_info, eb, slot);
11154                 err |= ret;
11155                 break;
11156         case BTRFS_DEV_ITEM_KEY:
11157                 ret = check_dev_item(fs_info, eb, slot);
11158                 err |= ret;
11159                 break;
11160         case BTRFS_CHUNK_ITEM_KEY:
11161                 ret = check_chunk_item(fs_info, eb, slot);
11162                 err |= ret;
11163                 break;
11164         case BTRFS_DEV_EXTENT_KEY:
11165                 ret = check_dev_extent_item(fs_info, eb, slot);
11166                 err |= ret;
11167                 break;
11168         case BTRFS_EXTENT_ITEM_KEY:
11169         case BTRFS_METADATA_ITEM_KEY:
11170                 ret = check_extent_item(fs_info, eb, slot);
11171                 err |= ret;
11172                 break;
11173         case BTRFS_EXTENT_CSUM_KEY:
11174                 total_csum_bytes += btrfs_item_size_nr(eb, slot);
11175                 break;
11176         case BTRFS_TREE_BLOCK_REF_KEY:
11177                 ret = check_tree_block_backref(fs_info, key.offset,
11178                                                key.objectid, -1);
11179                 err |= ret;
11180                 break;
11181         case BTRFS_EXTENT_DATA_REF_KEY:
11182                 dref = btrfs_item_ptr(eb, slot, struct btrfs_extent_data_ref);
11183                 ret = check_extent_data_backref(fs_info,
11184                                 btrfs_extent_data_ref_root(eb, dref),
11185                                 btrfs_extent_data_ref_objectid(eb, dref),
11186                                 btrfs_extent_data_ref_offset(eb, dref),
11187                                 key.objectid, 0,
11188                                 btrfs_extent_data_ref_count(eb, dref));
11189                 err |= ret;
11190                 break;
11191         case BTRFS_SHARED_BLOCK_REF_KEY:
11192                 ret = check_shared_block_backref(fs_info, key.offset,
11193                                                  key.objectid, -1);
11194                 err |= ret;
11195                 break;
11196         case BTRFS_SHARED_DATA_REF_KEY:
11197                 ret = check_shared_data_backref(fs_info, key.offset,
11198                                                 key.objectid);
11199                 err |= ret;
11200                 break;
11201         default:
11202                 break;
11203         }
11204
11205         if (++slot < btrfs_header_nritems(eb))
11206                 goto next;
11207
11208         return err;
11209 }
11210
11211 /*
11212  * Helper function for later fs/subvol tree check.  To determine if a tree
11213  * block should be checked.
11214  * This function will ensure only the direct referencer with lowest rootid to
11215  * check a fs/subvolume tree block.
11216  *
11217  * Backref check at extent tree would detect errors like missing subvolume
11218  * tree, so we can do aggressive check to reduce duplicated checks.
11219  */
11220 static int should_check(struct btrfs_root *root, struct extent_buffer *eb)
11221 {
11222         struct btrfs_root *extent_root = root->fs_info->extent_root;
11223         struct btrfs_key key;
11224         struct btrfs_path path;
11225         struct extent_buffer *leaf;
11226         int slot;
11227         struct btrfs_extent_item *ei;
11228         unsigned long ptr;
11229         unsigned long end;
11230         int type;
11231         u32 item_size;
11232         u64 offset;
11233         struct btrfs_extent_inline_ref *iref;
11234         int ret;
11235
11236         btrfs_init_path(&path);
11237         key.objectid = btrfs_header_bytenr(eb);
11238         key.type = BTRFS_METADATA_ITEM_KEY;
11239         key.offset = (u64)-1;
11240
11241         /*
11242          * Any failure in backref resolving means we can't determine
11243          * whom the tree block belongs to.
11244          * So in that case, we need to check that tree block
11245          */
11246         ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
11247         if (ret < 0)
11248                 goto need_check;
11249
11250         ret = btrfs_previous_extent_item(extent_root, &path,
11251                                          btrfs_header_bytenr(eb));
11252         if (ret)
11253                 goto need_check;
11254
11255         leaf = path.nodes[0];
11256         slot = path.slots[0];
11257         btrfs_item_key_to_cpu(leaf, &key, slot);
11258         ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
11259
11260         if (key.type == BTRFS_METADATA_ITEM_KEY) {
11261                 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
11262         } else {
11263                 struct btrfs_tree_block_info *info;
11264
11265                 info = (struct btrfs_tree_block_info *)(ei + 1);
11266                 iref = (struct btrfs_extent_inline_ref *)(info + 1);
11267         }
11268
11269         item_size = btrfs_item_size_nr(leaf, slot);
11270         ptr = (unsigned long)iref;
11271         end = (unsigned long)ei + item_size;
11272         while (ptr < end) {
11273                 iref = (struct btrfs_extent_inline_ref *)ptr;
11274                 type = btrfs_extent_inline_ref_type(leaf, iref);
11275                 offset = btrfs_extent_inline_ref_offset(leaf, iref);
11276
11277                 /*
11278                  * We only check the tree block if current root is
11279                  * the lowest referencer of it.
11280                  */
11281                 if (type == BTRFS_TREE_BLOCK_REF_KEY &&
11282                     offset < root->objectid) {
11283                         btrfs_release_path(&path);
11284                         return 0;
11285                 }
11286
11287                 ptr += btrfs_extent_inline_ref_size(type);
11288         }
11289         /*
11290          * Normally we should also check keyed tree block ref, but that may be
11291          * very time consuming.  Inlined ref should already make us skip a lot
11292          * of refs now.  So skip search keyed tree block ref.
11293          */
11294
11295 need_check:
11296         btrfs_release_path(&path);
11297         return 1;
11298 }
11299
11300 /*
11301  * Traversal function for tree block. We will do:
11302  * 1) Skip shared fs/subvolume tree blocks
11303  * 2) Update related bytes accounting
11304  * 3) Pre-order traversal
11305  */
11306 static int traverse_tree_block(struct btrfs_root *root,
11307                                 struct extent_buffer *node)
11308 {
11309         struct extent_buffer *eb;
11310         struct btrfs_key key;
11311         struct btrfs_key drop_key;
11312         int level;
11313         u64 nr;
11314         int i;
11315         int err = 0;
11316         int ret;
11317
11318         /*
11319          * Skip shared fs/subvolume tree block, in that case they will
11320          * be checked by referencer with lowest rootid
11321          */
11322         if (is_fstree(root->objectid) && !should_check(root, node))
11323                 return 0;
11324
11325         /* Update bytes accounting */
11326         total_btree_bytes += node->len;
11327         if (fs_root_objectid(btrfs_header_owner(node)))
11328                 total_fs_tree_bytes += node->len;
11329         if (btrfs_header_owner(node) == BTRFS_EXTENT_TREE_OBJECTID)
11330                 total_extent_tree_bytes += node->len;
11331         if (!found_old_backref &&
11332             btrfs_header_owner(node) == BTRFS_TREE_RELOC_OBJECTID &&
11333             btrfs_header_backref_rev(node) == BTRFS_MIXED_BACKREF_REV &&
11334             !btrfs_header_flag(node, BTRFS_HEADER_FLAG_RELOC))
11335                 found_old_backref = 1;
11336
11337         /* pre-order tranversal, check itself first */
11338         level = btrfs_header_level(node);
11339         ret = check_tree_block_ref(root, node, btrfs_header_bytenr(node),
11340                                    btrfs_header_level(node),
11341                                    btrfs_header_owner(node));
11342         err |= ret;
11343         if (err)
11344                 error(
11345         "check %s failed root %llu bytenr %llu level %d, force continue check",
11346                         level ? "node":"leaf", root->objectid,
11347                         btrfs_header_bytenr(node), btrfs_header_level(node));
11348
11349         if (!level) {
11350                 btree_space_waste += btrfs_leaf_free_space(root, node);
11351                 ret = check_leaf_items(root, node);
11352                 err |= ret;
11353                 return err;
11354         }
11355
11356         nr = btrfs_header_nritems(node);
11357         btrfs_disk_key_to_cpu(&drop_key, &root->root_item.drop_progress);
11358         btree_space_waste += (BTRFS_NODEPTRS_PER_BLOCK(root) - nr) *
11359                 sizeof(struct btrfs_key_ptr);
11360
11361         /* Then check all its children */
11362         for (i = 0; i < nr; i++) {
11363                 u64 blocknr = btrfs_node_blockptr(node, i);
11364
11365                 btrfs_node_key_to_cpu(node, &key, i);
11366                 if (level == root->root_item.drop_level &&
11367                     is_dropped_key(&key, &drop_key))
11368                         continue;
11369
11370                 /*
11371                  * As a btrfs tree has most 8 levels (0..7), so it's quite safe
11372                  * to call the function itself.
11373                  */
11374                 eb = read_tree_block(root, blocknr, root->nodesize, 0);
11375                 if (extent_buffer_uptodate(eb)) {
11376                         ret = traverse_tree_block(root, eb);
11377                         err |= ret;
11378                 }
11379                 free_extent_buffer(eb);
11380         }
11381
11382         return err;
11383 }
11384
11385 /*
11386  * Low memory usage version check_chunks_and_extents.
11387  */
11388 static int check_chunks_and_extents_v2(struct btrfs_root *root)
11389 {
11390         struct btrfs_path path;
11391         struct btrfs_key key;
11392         struct btrfs_root *root1;
11393         struct btrfs_root *cur_root;
11394         int err = 0;
11395         int ret;
11396
11397         root1 = root->fs_info->chunk_root;
11398         ret = traverse_tree_block(root1, root1->node);
11399         err |= ret;
11400
11401         root1 = root->fs_info->tree_root;
11402         ret = traverse_tree_block(root1, root1->node);
11403         err |= ret;
11404
11405         btrfs_init_path(&path);
11406         key.objectid = BTRFS_EXTENT_TREE_OBJECTID;
11407         key.offset = 0;
11408         key.type = BTRFS_ROOT_ITEM_KEY;
11409
11410         ret = btrfs_search_slot(NULL, root1, &key, &path, 0, 0);
11411         if (ret) {
11412                 error("cannot find extent treet in tree_root");
11413                 goto out;
11414         }
11415
11416         while (1) {
11417                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
11418                 if (key.type != BTRFS_ROOT_ITEM_KEY)
11419                         goto next;
11420                 key.offset = (u64)-1;
11421
11422                 cur_root = btrfs_read_fs_root(root->fs_info, &key);
11423                 if (IS_ERR(cur_root) || !cur_root) {
11424                         error("failed to read tree: %lld", key.objectid);
11425                         goto next;
11426                 }
11427
11428                 ret = traverse_tree_block(cur_root, cur_root->node);
11429                 err |= ret;
11430
11431 next:
11432                 ret = btrfs_next_item(root1, &path);
11433                 if (ret)
11434                         goto out;
11435         }
11436
11437 out:
11438         btrfs_release_path(&path);
11439         return err;
11440 }
11441
11442 static int btrfs_fsck_reinit_root(struct btrfs_trans_handle *trans,
11443                            struct btrfs_root *root, int overwrite)
11444 {
11445         struct extent_buffer *c;
11446         struct extent_buffer *old = root->node;
11447         int level;
11448         int ret;
11449         struct btrfs_disk_key disk_key = {0,0,0};
11450
11451         level = 0;
11452
11453         if (overwrite) {
11454                 c = old;
11455                 extent_buffer_get(c);
11456                 goto init;
11457         }
11458         c = btrfs_alloc_free_block(trans, root,
11459                                    root->nodesize,
11460                                    root->root_key.objectid,
11461                                    &disk_key, level, 0, 0);
11462         if (IS_ERR(c)) {
11463                 c = old;
11464                 extent_buffer_get(c);
11465                 overwrite = 1;
11466         }
11467 init:
11468         memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
11469         btrfs_set_header_level(c, level);
11470         btrfs_set_header_bytenr(c, c->start);
11471         btrfs_set_header_generation(c, trans->transid);
11472         btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
11473         btrfs_set_header_owner(c, root->root_key.objectid);
11474
11475         write_extent_buffer(c, root->fs_info->fsid,
11476                             btrfs_header_fsid(), BTRFS_FSID_SIZE);
11477
11478         write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
11479                             btrfs_header_chunk_tree_uuid(c),
11480                             BTRFS_UUID_SIZE);
11481
11482         btrfs_mark_buffer_dirty(c);
11483         /*
11484          * this case can happen in the following case:
11485          *
11486          * 1.overwrite previous root.
11487          *
11488          * 2.reinit reloc data root, this is because we skip pin
11489          * down reloc data tree before which means we can allocate
11490          * same block bytenr here.
11491          */
11492         if (old->start == c->start) {
11493                 btrfs_set_root_generation(&root->root_item,
11494                                           trans->transid);
11495                 root->root_item.level = btrfs_header_level(root->node);
11496                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
11497                                         &root->root_key, &root->root_item);
11498                 if (ret) {
11499                         free_extent_buffer(c);
11500                         return ret;
11501                 }
11502         }
11503         free_extent_buffer(old);
11504         root->node = c;
11505         add_root_to_dirty_list(root);
11506         return 0;
11507 }
11508
11509 static int pin_down_tree_blocks(struct btrfs_fs_info *fs_info,
11510                                 struct extent_buffer *eb, int tree_root)
11511 {
11512         struct extent_buffer *tmp;
11513         struct btrfs_root_item *ri;
11514         struct btrfs_key key;
11515         u64 bytenr;
11516         u32 nodesize;
11517         int level = btrfs_header_level(eb);
11518         int nritems;
11519         int ret;
11520         int i;
11521
11522         /*
11523          * If we have pinned this block before, don't pin it again.
11524          * This can not only avoid forever loop with broken filesystem
11525          * but also give us some speedups.
11526          */
11527         if (test_range_bit(&fs_info->pinned_extents, eb->start,
11528                            eb->start + eb->len - 1, EXTENT_DIRTY, 0))
11529                 return 0;
11530
11531         btrfs_pin_extent(fs_info, eb->start, eb->len);
11532
11533         nodesize = btrfs_super_nodesize(fs_info->super_copy);
11534         nritems = btrfs_header_nritems(eb);
11535         for (i = 0; i < nritems; i++) {
11536                 if (level == 0) {
11537                         btrfs_item_key_to_cpu(eb, &key, i);
11538                         if (key.type != BTRFS_ROOT_ITEM_KEY)
11539                                 continue;
11540                         /* Skip the extent root and reloc roots */
11541                         if (key.objectid == BTRFS_EXTENT_TREE_OBJECTID ||
11542                             key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
11543                             key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
11544                                 continue;
11545                         ri = btrfs_item_ptr(eb, i, struct btrfs_root_item);
11546                         bytenr = btrfs_disk_root_bytenr(eb, ri);
11547
11548                         /*
11549                          * If at any point we start needing the real root we
11550                          * will have to build a stump root for the root we are
11551                          * in, but for now this doesn't actually use the root so
11552                          * just pass in extent_root.
11553                          */
11554                         tmp = read_tree_block(fs_info->extent_root, bytenr,
11555                                               nodesize, 0);
11556                         if (!extent_buffer_uptodate(tmp)) {
11557                                 fprintf(stderr, "Error reading root block\n");
11558                                 return -EIO;
11559                         }
11560                         ret = pin_down_tree_blocks(fs_info, tmp, 0);
11561                         free_extent_buffer(tmp);
11562                         if (ret)
11563                                 return ret;
11564                 } else {
11565                         bytenr = btrfs_node_blockptr(eb, i);
11566
11567                         /* If we aren't the tree root don't read the block */
11568                         if (level == 1 && !tree_root) {
11569                                 btrfs_pin_extent(fs_info, bytenr, nodesize);
11570                                 continue;
11571                         }
11572
11573                         tmp = read_tree_block(fs_info->extent_root, bytenr,
11574                                               nodesize, 0);
11575                         if (!extent_buffer_uptodate(tmp)) {
11576                                 fprintf(stderr, "Error reading tree block\n");
11577                                 return -EIO;
11578                         }
11579                         ret = pin_down_tree_blocks(fs_info, tmp, tree_root);
11580                         free_extent_buffer(tmp);
11581                         if (ret)
11582                                 return ret;
11583                 }
11584         }
11585
11586         return 0;
11587 }
11588
11589 static int pin_metadata_blocks(struct btrfs_fs_info *fs_info)
11590 {
11591         int ret;
11592
11593         ret = pin_down_tree_blocks(fs_info, fs_info->chunk_root->node, 0);
11594         if (ret)
11595                 return ret;
11596
11597         return pin_down_tree_blocks(fs_info, fs_info->tree_root->node, 1);
11598 }
11599
11600 static int reset_block_groups(struct btrfs_fs_info *fs_info)
11601 {
11602         struct btrfs_block_group_cache *cache;
11603         struct btrfs_path path;
11604         struct extent_buffer *leaf;
11605         struct btrfs_chunk *chunk;
11606         struct btrfs_key key;
11607         int ret;
11608         u64 start;
11609
11610         btrfs_init_path(&path);
11611         key.objectid = 0;
11612         key.type = BTRFS_CHUNK_ITEM_KEY;
11613         key.offset = 0;
11614         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, &path, 0, 0);
11615         if (ret < 0) {
11616                 btrfs_release_path(&path);
11617                 return ret;
11618         }
11619
11620         /*
11621          * We do this in case the block groups were screwed up and had alloc
11622          * bits that aren't actually set on the chunks.  This happens with
11623          * restored images every time and could happen in real life I guess.
11624          */
11625         fs_info->avail_data_alloc_bits = 0;
11626         fs_info->avail_metadata_alloc_bits = 0;
11627         fs_info->avail_system_alloc_bits = 0;
11628
11629         /* First we need to create the in-memory block groups */
11630         while (1) {
11631                 if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
11632                         ret = btrfs_next_leaf(fs_info->chunk_root, &path);
11633                         if (ret < 0) {
11634                                 btrfs_release_path(&path);
11635                                 return ret;
11636                         }
11637                         if (ret) {
11638                                 ret = 0;
11639                                 break;
11640                         }
11641                 }
11642                 leaf = path.nodes[0];
11643                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
11644                 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
11645                         path.slots[0]++;
11646                         continue;
11647                 }
11648
11649                 chunk = btrfs_item_ptr(leaf, path.slots[0], struct btrfs_chunk);
11650                 btrfs_add_block_group(fs_info, 0,
11651                                       btrfs_chunk_type(leaf, chunk),
11652                                       key.objectid, key.offset,
11653                                       btrfs_chunk_length(leaf, chunk));
11654                 set_extent_dirty(&fs_info->free_space_cache, key.offset,
11655                                  key.offset + btrfs_chunk_length(leaf, chunk),
11656                                  GFP_NOFS);
11657                 path.slots[0]++;
11658         }
11659         start = 0;
11660         while (1) {
11661                 cache = btrfs_lookup_first_block_group(fs_info, start);
11662                 if (!cache)
11663                         break;
11664                 cache->cached = 1;
11665                 start = cache->key.objectid + cache->key.offset;
11666         }
11667
11668         btrfs_release_path(&path);
11669         return 0;
11670 }
11671
11672 static int reset_balance(struct btrfs_trans_handle *trans,
11673                          struct btrfs_fs_info *fs_info)
11674 {
11675         struct btrfs_root *root = fs_info->tree_root;
11676         struct btrfs_path path;
11677         struct extent_buffer *leaf;
11678         struct btrfs_key key;
11679         int del_slot, del_nr = 0;
11680         int ret;
11681         int found = 0;
11682
11683         btrfs_init_path(&path);
11684         key.objectid = BTRFS_BALANCE_OBJECTID;
11685         key.type = BTRFS_BALANCE_ITEM_KEY;
11686         key.offset = 0;
11687         ret = btrfs_search_slot(trans, root, &key, &path, -1, 1);
11688         if (ret) {
11689                 if (ret > 0)
11690                         ret = 0;
11691                 if (!ret)
11692                         goto reinit_data_reloc;
11693                 else
11694                         goto out;
11695         }
11696
11697         ret = btrfs_del_item(trans, root, &path);
11698         if (ret)
11699                 goto out;
11700         btrfs_release_path(&path);
11701
11702         key.objectid = BTRFS_TREE_RELOC_OBJECTID;
11703         key.type = BTRFS_ROOT_ITEM_KEY;
11704         key.offset = 0;
11705         ret = btrfs_search_slot(trans, root, &key, &path, -1, 1);
11706         if (ret < 0)
11707                 goto out;
11708         while (1) {
11709                 if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
11710                         if (!found)
11711                                 break;
11712
11713                         if (del_nr) {
11714                                 ret = btrfs_del_items(trans, root, &path,
11715                                                       del_slot, del_nr);
11716                                 del_nr = 0;
11717                                 if (ret)
11718                                         goto out;
11719                         }
11720                         key.offset++;
11721                         btrfs_release_path(&path);
11722
11723                         found = 0;
11724                         ret = btrfs_search_slot(trans, root, &key, &path,
11725                                                 -1, 1);
11726                         if (ret < 0)
11727                                 goto out;
11728                         continue;
11729                 }
11730                 found = 1;
11731                 leaf = path.nodes[0];
11732                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
11733                 if (key.objectid > BTRFS_TREE_RELOC_OBJECTID)
11734                         break;
11735                 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
11736                         path.slots[0]++;
11737                         continue;
11738                 }
11739                 if (!del_nr) {
11740                         del_slot = path.slots[0];
11741                         del_nr = 1;
11742                 } else {
11743                         del_nr++;
11744                 }
11745                 path.slots[0]++;
11746         }
11747
11748         if (del_nr) {
11749                 ret = btrfs_del_items(trans, root, &path, del_slot, del_nr);
11750                 if (ret)
11751                         goto out;
11752         }
11753         btrfs_release_path(&path);
11754
11755 reinit_data_reloc:
11756         key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
11757         key.type = BTRFS_ROOT_ITEM_KEY;
11758         key.offset = (u64)-1;
11759         root = btrfs_read_fs_root(fs_info, &key);
11760         if (IS_ERR(root)) {
11761                 fprintf(stderr, "Error reading data reloc tree\n");
11762                 ret = PTR_ERR(root);
11763                 goto out;
11764         }
11765         record_root_in_trans(trans, root);
11766         ret = btrfs_fsck_reinit_root(trans, root, 0);
11767         if (ret)
11768                 goto out;
11769         ret = btrfs_make_root_dir(trans, root, BTRFS_FIRST_FREE_OBJECTID);
11770 out:
11771         btrfs_release_path(&path);
11772         return ret;
11773 }
11774
11775 static int reinit_extent_tree(struct btrfs_trans_handle *trans,
11776                               struct btrfs_fs_info *fs_info)
11777 {
11778         u64 start = 0;
11779         int ret;
11780
11781         /*
11782          * The only reason we don't do this is because right now we're just
11783          * walking the trees we find and pinning down their bytes, we don't look
11784          * at any of the leaves.  In order to do mixed groups we'd have to check
11785          * the leaves of any fs roots and pin down the bytes for any file
11786          * extents we find.  Not hard but why do it if we don't have to?
11787          */
11788         if (btrfs_fs_incompat(fs_info, BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)) {
11789                 fprintf(stderr, "We don't support re-initing the extent tree "
11790                         "for mixed block groups yet, please notify a btrfs "
11791                         "developer you want to do this so they can add this "
11792                         "functionality.\n");
11793                 return -EINVAL;
11794         }
11795
11796         /*
11797          * first we need to walk all of the trees except the extent tree and pin
11798          * down the bytes that are in use so we don't overwrite any existing
11799          * metadata.
11800          */
11801         ret = pin_metadata_blocks(fs_info);
11802         if (ret) {
11803                 fprintf(stderr, "error pinning down used bytes\n");
11804                 return ret;
11805         }
11806
11807         /*
11808          * Need to drop all the block groups since we're going to recreate all
11809          * of them again.
11810          */
11811         btrfs_free_block_groups(fs_info);
11812         ret = reset_block_groups(fs_info);
11813         if (ret) {
11814                 fprintf(stderr, "error resetting the block groups\n");
11815                 return ret;
11816         }
11817
11818         /* Ok we can allocate now, reinit the extent root */
11819         ret = btrfs_fsck_reinit_root(trans, fs_info->extent_root, 0);
11820         if (ret) {
11821                 fprintf(stderr, "extent root initialization failed\n");
11822                 /*
11823                  * When the transaction code is updated we should end the
11824                  * transaction, but for now progs only knows about commit so
11825                  * just return an error.
11826                  */
11827                 return ret;
11828         }
11829
11830         /*
11831          * Now we have all the in-memory block groups setup so we can make
11832          * allocations properly, and the metadata we care about is safe since we
11833          * pinned all of it above.
11834          */
11835         while (1) {
11836                 struct btrfs_block_group_cache *cache;
11837
11838                 cache = btrfs_lookup_first_block_group(fs_info, start);
11839                 if (!cache)
11840                         break;
11841                 start = cache->key.objectid + cache->key.offset;
11842                 ret = btrfs_insert_item(trans, fs_info->extent_root,
11843                                         &cache->key, &cache->item,
11844                                         sizeof(cache->item));
11845                 if (ret) {
11846                         fprintf(stderr, "Error adding block group\n");
11847                         return ret;
11848                 }
11849                 btrfs_extent_post_op(trans, fs_info->extent_root);
11850         }
11851
11852         ret = reset_balance(trans, fs_info);
11853         if (ret)
11854                 fprintf(stderr, "error resetting the pending balance\n");
11855
11856         return ret;
11857 }
11858
11859 static int recow_extent_buffer(struct btrfs_root *root, struct extent_buffer *eb)
11860 {
11861         struct btrfs_path path;
11862         struct btrfs_trans_handle *trans;
11863         struct btrfs_key key;
11864         int ret;
11865
11866         printf("Recowing metadata block %llu\n", eb->start);
11867         key.objectid = btrfs_header_owner(eb);
11868         key.type = BTRFS_ROOT_ITEM_KEY;
11869         key.offset = (u64)-1;
11870
11871         root = btrfs_read_fs_root(root->fs_info, &key);
11872         if (IS_ERR(root)) {
11873                 fprintf(stderr, "Couldn't find owner root %llu\n",
11874                         key.objectid);
11875                 return PTR_ERR(root);
11876         }
11877
11878         trans = btrfs_start_transaction(root, 1);
11879         if (IS_ERR(trans))
11880                 return PTR_ERR(trans);
11881
11882         btrfs_init_path(&path);
11883         path.lowest_level = btrfs_header_level(eb);
11884         if (path.lowest_level)
11885                 btrfs_node_key_to_cpu(eb, &key, 0);
11886         else
11887                 btrfs_item_key_to_cpu(eb, &key, 0);
11888
11889         ret = btrfs_search_slot(trans, root, &key, &path, 0, 1);
11890         btrfs_commit_transaction(trans, root);
11891         btrfs_release_path(&path);
11892         return ret;
11893 }
11894
11895 static int delete_bad_item(struct btrfs_root *root, struct bad_item *bad)
11896 {
11897         struct btrfs_path path;
11898         struct btrfs_trans_handle *trans;
11899         struct btrfs_key key;
11900         int ret;
11901
11902         printf("Deleting bad item [%llu,%u,%llu]\n", bad->key.objectid,
11903                bad->key.type, bad->key.offset);
11904         key.objectid = bad->root_id;
11905         key.type = BTRFS_ROOT_ITEM_KEY;
11906         key.offset = (u64)-1;
11907
11908         root = btrfs_read_fs_root(root->fs_info, &key);
11909         if (IS_ERR(root)) {
11910                 fprintf(stderr, "Couldn't find owner root %llu\n",
11911                         key.objectid);
11912                 return PTR_ERR(root);
11913         }
11914
11915         trans = btrfs_start_transaction(root, 1);
11916         if (IS_ERR(trans))
11917                 return PTR_ERR(trans);
11918
11919         btrfs_init_path(&path);
11920         ret = btrfs_search_slot(trans, root, &bad->key, &path, -1, 1);
11921         if (ret) {
11922                 if (ret > 0)
11923                         ret = 0;
11924                 goto out;
11925         }
11926         ret = btrfs_del_item(trans, root, &path);
11927 out:
11928         btrfs_commit_transaction(trans, root);
11929         btrfs_release_path(&path);
11930         return ret;
11931 }
11932
11933 static int zero_log_tree(struct btrfs_root *root)
11934 {
11935         struct btrfs_trans_handle *trans;
11936         int ret;
11937
11938         trans = btrfs_start_transaction(root, 1);
11939         if (IS_ERR(trans)) {
11940                 ret = PTR_ERR(trans);
11941                 return ret;
11942         }
11943         btrfs_set_super_log_root(root->fs_info->super_copy, 0);
11944         btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
11945         ret = btrfs_commit_transaction(trans, root);
11946         return ret;
11947 }
11948
11949 static int populate_csum(struct btrfs_trans_handle *trans,
11950                          struct btrfs_root *csum_root, char *buf, u64 start,
11951                          u64 len)
11952 {
11953         u64 offset = 0;
11954         u64 sectorsize;
11955         int ret = 0;
11956
11957         while (offset < len) {
11958                 sectorsize = csum_root->sectorsize;
11959                 ret = read_extent_data(csum_root, buf, start + offset,
11960                                        &sectorsize, 0);
11961                 if (ret)
11962                         break;
11963                 ret = btrfs_csum_file_block(trans, csum_root, start + len,
11964                                             start + offset, buf, sectorsize);
11965                 if (ret)
11966                         break;
11967                 offset += sectorsize;
11968         }
11969         return ret;
11970 }
11971
11972 static int fill_csum_tree_from_one_fs_root(struct btrfs_trans_handle *trans,
11973                                       struct btrfs_root *csum_root,
11974                                       struct btrfs_root *cur_root)
11975 {
11976         struct btrfs_path path;
11977         struct btrfs_key key;
11978         struct extent_buffer *node;
11979         struct btrfs_file_extent_item *fi;
11980         char *buf = NULL;
11981         u64 start = 0;
11982         u64 len = 0;
11983         int slot = 0;
11984         int ret = 0;
11985
11986         buf = malloc(cur_root->fs_info->csum_root->sectorsize);
11987         if (!buf)
11988                 return -ENOMEM;
11989
11990         btrfs_init_path(&path);
11991         key.objectid = 0;
11992         key.offset = 0;
11993         key.type = 0;
11994         ret = btrfs_search_slot(NULL, cur_root, &key, &path, 0, 0);
11995         if (ret < 0)
11996                 goto out;
11997         /* Iterate all regular file extents and fill its csum */
11998         while (1) {
11999                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
12000
12001                 if (key.type != BTRFS_EXTENT_DATA_KEY)
12002                         goto next;
12003                 node = path.nodes[0];
12004                 slot = path.slots[0];
12005                 fi = btrfs_item_ptr(node, slot, struct btrfs_file_extent_item);
12006                 if (btrfs_file_extent_type(node, fi) != BTRFS_FILE_EXTENT_REG)
12007                         goto next;
12008                 start = btrfs_file_extent_disk_bytenr(node, fi);
12009                 len = btrfs_file_extent_disk_num_bytes(node, fi);
12010
12011                 ret = populate_csum(trans, csum_root, buf, start, len);
12012                 if (ret == -EEXIST)
12013                         ret = 0;
12014                 if (ret < 0)
12015                         goto out;
12016 next:
12017                 /*
12018                  * TODO: if next leaf is corrupted, jump to nearest next valid
12019                  * leaf.
12020                  */
12021                 ret = btrfs_next_item(cur_root, &path);
12022                 if (ret < 0)
12023                         goto out;
12024                 if (ret > 0) {
12025                         ret = 0;
12026                         goto out;
12027                 }
12028         }
12029
12030 out:
12031         btrfs_release_path(&path);
12032         free(buf);
12033         return ret;
12034 }
12035
12036 static int fill_csum_tree_from_fs(struct btrfs_trans_handle *trans,
12037                                   struct btrfs_root *csum_root)
12038 {
12039         struct btrfs_fs_info *fs_info = csum_root->fs_info;
12040         struct btrfs_path path;
12041         struct btrfs_root *tree_root = fs_info->tree_root;
12042         struct btrfs_root *cur_root;
12043         struct extent_buffer *node;
12044         struct btrfs_key key;
12045         int slot = 0;
12046         int ret = 0;
12047
12048         btrfs_init_path(&path);
12049         key.objectid = BTRFS_FS_TREE_OBJECTID;
12050         key.offset = 0;
12051         key.type = BTRFS_ROOT_ITEM_KEY;
12052         ret = btrfs_search_slot(NULL, tree_root, &key, &path, 0, 0);
12053         if (ret < 0)
12054                 goto out;
12055         if (ret > 0) {
12056                 ret = -ENOENT;
12057                 goto out;
12058         }
12059
12060         while (1) {
12061                 node = path.nodes[0];
12062                 slot = path.slots[0];
12063                 btrfs_item_key_to_cpu(node, &key, slot);
12064                 if (key.objectid > BTRFS_LAST_FREE_OBJECTID)
12065                         goto out;
12066                 if (key.type != BTRFS_ROOT_ITEM_KEY)
12067                         goto next;
12068                 if (!is_fstree(key.objectid))
12069                         goto next;
12070                 key.offset = (u64)-1;
12071
12072                 cur_root = btrfs_read_fs_root(fs_info, &key);
12073                 if (IS_ERR(cur_root) || !cur_root) {
12074                         fprintf(stderr, "Fail to read fs/subvol tree: %lld\n",
12075                                 key.objectid);
12076                         goto out;
12077                 }
12078                 ret = fill_csum_tree_from_one_fs_root(trans, csum_root,
12079                                 cur_root);
12080                 if (ret < 0)
12081                         goto out;
12082 next:
12083                 ret = btrfs_next_item(tree_root, &path);
12084                 if (ret > 0) {
12085                         ret = 0;
12086                         goto out;
12087                 }
12088                 if (ret < 0)
12089                         goto out;
12090         }
12091
12092 out:
12093         btrfs_release_path(&path);
12094         return ret;
12095 }
12096
12097 static int fill_csum_tree_from_extent(struct btrfs_trans_handle *trans,
12098                                       struct btrfs_root *csum_root)
12099 {
12100         struct btrfs_root *extent_root = csum_root->fs_info->extent_root;
12101         struct btrfs_path path;
12102         struct btrfs_extent_item *ei;
12103         struct extent_buffer *leaf;
12104         char *buf;
12105         struct btrfs_key key;
12106         int ret;
12107
12108         btrfs_init_path(&path);
12109         key.objectid = 0;
12110         key.type = BTRFS_EXTENT_ITEM_KEY;
12111         key.offset = 0;
12112         ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
12113         if (ret < 0) {
12114                 btrfs_release_path(&path);
12115                 return ret;
12116         }
12117
12118         buf = malloc(csum_root->sectorsize);
12119         if (!buf) {
12120                 btrfs_release_path(&path);
12121                 return -ENOMEM;
12122         }
12123
12124         while (1) {
12125                 if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
12126                         ret = btrfs_next_leaf(extent_root, &path);
12127                         if (ret < 0)
12128                                 break;
12129                         if (ret) {
12130                                 ret = 0;
12131                                 break;
12132                         }
12133                 }
12134                 leaf = path.nodes[0];
12135
12136                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
12137                 if (key.type != BTRFS_EXTENT_ITEM_KEY) {
12138                         path.slots[0]++;
12139                         continue;
12140                 }
12141
12142                 ei = btrfs_item_ptr(leaf, path.slots[0],
12143                                     struct btrfs_extent_item);
12144                 if (!(btrfs_extent_flags(leaf, ei) &
12145                       BTRFS_EXTENT_FLAG_DATA)) {
12146                         path.slots[0]++;
12147                         continue;
12148                 }
12149
12150                 ret = populate_csum(trans, csum_root, buf, key.objectid,
12151                                     key.offset);
12152                 if (ret)
12153                         break;
12154                 path.slots[0]++;
12155         }
12156
12157         btrfs_release_path(&path);
12158         free(buf);
12159         return ret;
12160 }
12161
12162 /*
12163  * Recalculate the csum and put it into the csum tree.
12164  *
12165  * Extent tree init will wipe out all the extent info, so in that case, we
12166  * can't depend on extent tree, but use fs tree.  If search_fs_tree is set, we
12167  * will use fs/subvol trees to init the csum tree.
12168  */
12169 static int fill_csum_tree(struct btrfs_trans_handle *trans,
12170                           struct btrfs_root *csum_root,
12171                           int search_fs_tree)
12172 {
12173         if (search_fs_tree)
12174                 return fill_csum_tree_from_fs(trans, csum_root);
12175         else
12176                 return fill_csum_tree_from_extent(trans, csum_root);
12177 }
12178
12179 static void free_roots_info_cache(void)
12180 {
12181         if (!roots_info_cache)
12182                 return;
12183
12184         while (!cache_tree_empty(roots_info_cache)) {
12185                 struct cache_extent *entry;
12186                 struct root_item_info *rii;
12187
12188                 entry = first_cache_extent(roots_info_cache);
12189                 if (!entry)
12190                         break;
12191                 remove_cache_extent(roots_info_cache, entry);
12192                 rii = container_of(entry, struct root_item_info, cache_extent);
12193                 free(rii);
12194         }
12195
12196         free(roots_info_cache);
12197         roots_info_cache = NULL;
12198 }
12199
12200 static int build_roots_info_cache(struct btrfs_fs_info *info)
12201 {
12202         int ret = 0;
12203         struct btrfs_key key;
12204         struct extent_buffer *leaf;
12205         struct btrfs_path path;
12206
12207         if (!roots_info_cache) {
12208                 roots_info_cache = malloc(sizeof(*roots_info_cache));
12209                 if (!roots_info_cache)
12210                         return -ENOMEM;
12211                 cache_tree_init(roots_info_cache);
12212         }
12213
12214         btrfs_init_path(&path);
12215         key.objectid = 0;
12216         key.type = BTRFS_EXTENT_ITEM_KEY;
12217         key.offset = 0;
12218         ret = btrfs_search_slot(NULL, info->extent_root, &key, &path, 0, 0);
12219         if (ret < 0)
12220                 goto out;
12221         leaf = path.nodes[0];
12222
12223         while (1) {
12224                 struct btrfs_key found_key;
12225                 struct btrfs_extent_item *ei;
12226                 struct btrfs_extent_inline_ref *iref;
12227                 int slot = path.slots[0];
12228                 int type;
12229                 u64 flags;
12230                 u64 root_id;
12231                 u8 level;
12232                 struct cache_extent *entry;
12233                 struct root_item_info *rii;
12234
12235                 if (slot >= btrfs_header_nritems(leaf)) {
12236                         ret = btrfs_next_leaf(info->extent_root, &path);
12237                         if (ret < 0) {
12238                                 break;
12239                         } else if (ret) {
12240                                 ret = 0;
12241                                 break;
12242                         }
12243                         leaf = path.nodes[0];
12244                         slot = path.slots[0];
12245                 }
12246
12247                 btrfs_item_key_to_cpu(leaf, &found_key, path.slots[0]);
12248
12249                 if (found_key.type != BTRFS_EXTENT_ITEM_KEY &&
12250                     found_key.type != BTRFS_METADATA_ITEM_KEY)
12251                         goto next;
12252
12253                 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
12254                 flags = btrfs_extent_flags(leaf, ei);
12255
12256                 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
12257                     !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
12258                         goto next;
12259
12260                 if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
12261                         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
12262                         level = found_key.offset;
12263                 } else {
12264                         struct btrfs_tree_block_info *binfo;
12265
12266                         binfo = (struct btrfs_tree_block_info *)(ei + 1);
12267                         iref = (struct btrfs_extent_inline_ref *)(binfo + 1);
12268                         level = btrfs_tree_block_level(leaf, binfo);
12269                 }
12270
12271                 /*
12272                  * For a root extent, it must be of the following type and the
12273                  * first (and only one) iref in the item.
12274                  */
12275                 type = btrfs_extent_inline_ref_type(leaf, iref);
12276                 if (type != BTRFS_TREE_BLOCK_REF_KEY)
12277                         goto next;
12278
12279                 root_id = btrfs_extent_inline_ref_offset(leaf, iref);
12280                 entry = lookup_cache_extent(roots_info_cache, root_id, 1);
12281                 if (!entry) {
12282                         rii = malloc(sizeof(struct root_item_info));
12283                         if (!rii) {
12284                                 ret = -ENOMEM;
12285                                 goto out;
12286                         }
12287                         rii->cache_extent.start = root_id;
12288                         rii->cache_extent.size = 1;
12289                         rii->level = (u8)-1;
12290                         entry = &rii->cache_extent;
12291                         ret = insert_cache_extent(roots_info_cache, entry);
12292                         ASSERT(ret == 0);
12293                 } else {
12294                         rii = container_of(entry, struct root_item_info,
12295                                            cache_extent);
12296                 }
12297
12298                 ASSERT(rii->cache_extent.start == root_id);
12299                 ASSERT(rii->cache_extent.size == 1);
12300
12301                 if (level > rii->level || rii->level == (u8)-1) {
12302                         rii->level = level;
12303                         rii->bytenr = found_key.objectid;
12304                         rii->gen = btrfs_extent_generation(leaf, ei);
12305                         rii->node_count = 1;
12306                 } else if (level == rii->level) {
12307                         rii->node_count++;
12308                 }
12309 next:
12310                 path.slots[0]++;
12311         }
12312
12313 out:
12314         btrfs_release_path(&path);
12315
12316         return ret;
12317 }
12318
12319 static int maybe_repair_root_item(struct btrfs_fs_info *info,
12320                                   struct btrfs_path *path,
12321                                   const struct btrfs_key *root_key,
12322                                   const int read_only_mode)
12323 {
12324         const u64 root_id = root_key->objectid;
12325         struct cache_extent *entry;
12326         struct root_item_info *rii;
12327         struct btrfs_root_item ri;
12328         unsigned long offset;
12329
12330         entry = lookup_cache_extent(roots_info_cache, root_id, 1);
12331         if (!entry) {
12332                 fprintf(stderr,
12333                         "Error: could not find extent items for root %llu\n",
12334                         root_key->objectid);
12335                 return -ENOENT;
12336         }
12337
12338         rii = container_of(entry, struct root_item_info, cache_extent);
12339         ASSERT(rii->cache_extent.start == root_id);
12340         ASSERT(rii->cache_extent.size == 1);
12341
12342         if (rii->node_count != 1) {
12343                 fprintf(stderr,
12344                         "Error: could not find btree root extent for root %llu\n",
12345                         root_id);
12346                 return -ENOENT;
12347         }
12348
12349         offset = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
12350         read_extent_buffer(path->nodes[0], &ri, offset, sizeof(ri));
12351
12352         if (btrfs_root_bytenr(&ri) != rii->bytenr ||
12353             btrfs_root_level(&ri) != rii->level ||
12354             btrfs_root_generation(&ri) != rii->gen) {
12355
12356                 /*
12357                  * If we're in repair mode but our caller told us to not update
12358                  * the root item, i.e. just check if it needs to be updated, don't
12359                  * print this message, since the caller will call us again shortly
12360                  * for the same root item without read only mode (the caller will
12361                  * open a transaction first).
12362                  */
12363                 if (!(read_only_mode && repair))
12364                         fprintf(stderr,
12365                                 "%sroot item for root %llu,"
12366                                 " current bytenr %llu, current gen %llu, current level %u,"
12367                                 " new bytenr %llu, new gen %llu, new level %u\n",
12368                                 (read_only_mode ? "" : "fixing "),
12369                                 root_id,
12370                                 btrfs_root_bytenr(&ri), btrfs_root_generation(&ri),
12371                                 btrfs_root_level(&ri),
12372                                 rii->bytenr, rii->gen, rii->level);
12373
12374                 if (btrfs_root_generation(&ri) > rii->gen) {
12375                         fprintf(stderr,
12376                                 "root %llu has a root item with a more recent gen (%llu) compared to the found root node (%llu)\n",
12377                                 root_id, btrfs_root_generation(&ri), rii->gen);
12378                         return -EINVAL;
12379                 }
12380
12381                 if (!read_only_mode) {
12382                         btrfs_set_root_bytenr(&ri, rii->bytenr);
12383                         btrfs_set_root_level(&ri, rii->level);
12384                         btrfs_set_root_generation(&ri, rii->gen);
12385                         write_extent_buffer(path->nodes[0], &ri,
12386                                             offset, sizeof(ri));
12387                 }
12388
12389                 return 1;
12390         }
12391
12392         return 0;
12393 }
12394
12395 /*
12396  * A regression introduced in the 3.17 kernel (more specifically in 3.17-rc2),
12397  * caused read-only snapshots to be corrupted if they were created at a moment
12398  * when the source subvolume/snapshot had orphan items. The issue was that the
12399  * on-disk root items became incorrect, referring to the pre orphan cleanup root
12400  * node instead of the post orphan cleanup root node.
12401  * So this function, and its callees, just detects and fixes those cases. Even
12402  * though the regression was for read-only snapshots, this function applies to
12403  * any snapshot/subvolume root.
12404  * This must be run before any other repair code - not doing it so, makes other
12405  * repair code delete or modify backrefs in the extent tree for example, which
12406  * will result in an inconsistent fs after repairing the root items.
12407  */
12408 static int repair_root_items(struct btrfs_fs_info *info)
12409 {
12410         struct btrfs_path path;
12411         struct btrfs_key key;
12412         struct extent_buffer *leaf;
12413         struct btrfs_trans_handle *trans = NULL;
12414         int ret = 0;
12415         int bad_roots = 0;
12416         int need_trans = 0;
12417
12418         btrfs_init_path(&path);
12419
12420         ret = build_roots_info_cache(info);
12421         if (ret)
12422                 goto out;
12423
12424         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
12425         key.type = BTRFS_ROOT_ITEM_KEY;
12426         key.offset = 0;
12427
12428 again:
12429         /*
12430          * Avoid opening and committing transactions if a leaf doesn't have
12431          * any root items that need to be fixed, so that we avoid rotating
12432          * backup roots unnecessarily.
12433          */
12434         if (need_trans) {
12435                 trans = btrfs_start_transaction(info->tree_root, 1);
12436                 if (IS_ERR(trans)) {
12437                         ret = PTR_ERR(trans);
12438                         goto out;
12439                 }
12440         }
12441
12442         ret = btrfs_search_slot(trans, info->tree_root, &key, &path,
12443                                 0, trans ? 1 : 0);
12444         if (ret < 0)
12445                 goto out;
12446         leaf = path.nodes[0];
12447
12448         while (1) {
12449                 struct btrfs_key found_key;
12450
12451                 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
12452                         int no_more_keys = find_next_key(&path, &key);
12453
12454                         btrfs_release_path(&path);
12455                         if (trans) {
12456                                 ret = btrfs_commit_transaction(trans,
12457                                                                info->tree_root);
12458                                 trans = NULL;
12459                                 if (ret < 0)
12460                                         goto out;
12461                         }
12462                         need_trans = 0;
12463                         if (no_more_keys)
12464                                 break;
12465                         goto again;
12466                 }
12467
12468                 btrfs_item_key_to_cpu(leaf, &found_key, path.slots[0]);
12469
12470                 if (found_key.type != BTRFS_ROOT_ITEM_KEY)
12471                         goto next;
12472                 if (found_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
12473                         goto next;
12474
12475                 ret = maybe_repair_root_item(info, &path, &found_key,
12476                                              trans ? 0 : 1);
12477                 if (ret < 0)
12478                         goto out;
12479                 if (ret) {
12480                         if (!trans && repair) {
12481                                 need_trans = 1;
12482                                 key = found_key;
12483                                 btrfs_release_path(&path);
12484                                 goto again;
12485                         }
12486                         bad_roots++;
12487                 }
12488 next:
12489                 path.slots[0]++;
12490         }
12491         ret = 0;
12492 out:
12493         free_roots_info_cache();
12494         btrfs_release_path(&path);
12495         if (trans)
12496                 btrfs_commit_transaction(trans, info->tree_root);
12497         if (ret < 0)
12498                 return ret;
12499
12500         return bad_roots;
12501 }
12502
12503 static int clear_free_space_cache(struct btrfs_fs_info *fs_info)
12504 {
12505         struct btrfs_trans_handle *trans;
12506         struct btrfs_block_group_cache *bg_cache;
12507         u64 current = 0;
12508         int ret = 0;
12509
12510         /* Clear all free space cache inodes and its extent data */
12511         while (1) {
12512                 bg_cache = btrfs_lookup_first_block_group(fs_info, current);
12513                 if (!bg_cache)
12514                         break;
12515                 ret = btrfs_clear_free_space_cache(fs_info, bg_cache);
12516                 if (ret < 0)
12517                         return ret;
12518                 current = bg_cache->key.objectid + bg_cache->key.offset;
12519         }
12520
12521         /* Don't forget to set cache_generation to -1 */
12522         trans = btrfs_start_transaction(fs_info->tree_root, 0);
12523         if (IS_ERR(trans)) {
12524                 error("failed to update super block cache generation");
12525                 return PTR_ERR(trans);
12526         }
12527         btrfs_set_super_cache_generation(fs_info->super_copy, (u64)-1);
12528         btrfs_commit_transaction(trans, fs_info->tree_root);
12529
12530         return ret;
12531 }
12532
12533 const char * const cmd_check_usage[] = {
12534         "btrfs check [options] <device>",
12535         "Check structural integrity of a filesystem (unmounted).",
12536         "Check structural integrity of an unmounted filesystem. Verify internal",
12537         "trees' consistency and item connectivity. In the repair mode try to",
12538         "fix the problems found. ",
12539         "WARNING: the repair mode is considered dangerous",
12540         "",
12541         "-s|--super <superblock>     use this superblock copy",
12542         "-b|--backup                 use the first valid backup root copy",
12543         "--repair                    try to repair the filesystem",
12544         "--readonly                  run in read-only mode (default)",
12545         "--init-csum-tree            create a new CRC tree",
12546         "--init-extent-tree          create a new extent tree",
12547         "--mode <MODE>               allows choice of memory/IO trade-offs",
12548         "                            where MODE is one of:",
12549         "                            original - read inodes and extents to memory (requires",
12550         "                                       more memory, does less IO)",
12551         "                            lowmem   - try to use less memory but read blocks again",
12552         "                                       when needed",
12553         "--check-data-csum           verify checksums of data blocks",
12554         "-Q|--qgroup-report          print a report on qgroup consistency",
12555         "-E|--subvol-extents <subvolid>",
12556         "                            print subvolume extents and sharing state",
12557         "-r|--tree-root <bytenr>     use the given bytenr for the tree root",
12558         "--chunk-root <bytenr>       use the given bytenr for the chunk tree root",
12559         "-p|--progress               indicate progress",
12560         "--clear-space-cache v1|v2   clear space cache for v1 or v2",
12561         NULL
12562 };
12563
12564 int cmd_check(int argc, char **argv)
12565 {
12566         struct cache_tree root_cache;
12567         struct btrfs_root *root;
12568         struct btrfs_fs_info *info;
12569         u64 bytenr = 0;
12570         u64 subvolid = 0;
12571         u64 tree_root_bytenr = 0;
12572         u64 chunk_root_bytenr = 0;
12573         char uuidbuf[BTRFS_UUID_UNPARSED_SIZE];
12574         int ret;
12575         int err = 0;
12576         u64 num;
12577         int init_csum_tree = 0;
12578         int readonly = 0;
12579         int clear_space_cache = 0;
12580         int qgroup_report = 0;
12581         int qgroups_repaired = 0;
12582         unsigned ctree_flags = OPEN_CTREE_EXCLUSIVE;
12583
12584         while(1) {
12585                 int c;
12586                 enum { GETOPT_VAL_REPAIR = 257, GETOPT_VAL_INIT_CSUM,
12587                         GETOPT_VAL_INIT_EXTENT, GETOPT_VAL_CHECK_CSUM,
12588                         GETOPT_VAL_READONLY, GETOPT_VAL_CHUNK_TREE,
12589                         GETOPT_VAL_MODE, GETOPT_VAL_CLEAR_SPACE_CACHE };
12590                 static const struct option long_options[] = {
12591                         { "super", required_argument, NULL, 's' },
12592                         { "repair", no_argument, NULL, GETOPT_VAL_REPAIR },
12593                         { "readonly", no_argument, NULL, GETOPT_VAL_READONLY },
12594                         { "init-csum-tree", no_argument, NULL,
12595                                 GETOPT_VAL_INIT_CSUM },
12596                         { "init-extent-tree", no_argument, NULL,
12597                                 GETOPT_VAL_INIT_EXTENT },
12598                         { "check-data-csum", no_argument, NULL,
12599                                 GETOPT_VAL_CHECK_CSUM },
12600                         { "backup", no_argument, NULL, 'b' },
12601                         { "subvol-extents", required_argument, NULL, 'E' },
12602                         { "qgroup-report", no_argument, NULL, 'Q' },
12603                         { "tree-root", required_argument, NULL, 'r' },
12604                         { "chunk-root", required_argument, NULL,
12605                                 GETOPT_VAL_CHUNK_TREE },
12606                         { "progress", no_argument, NULL, 'p' },
12607                         { "mode", required_argument, NULL,
12608                                 GETOPT_VAL_MODE },
12609                         { "clear-space-cache", required_argument, NULL,
12610                                 GETOPT_VAL_CLEAR_SPACE_CACHE},
12611                         { NULL, 0, NULL, 0}
12612                 };
12613
12614                 c = getopt_long(argc, argv, "as:br:p", long_options, NULL);
12615                 if (c < 0)
12616                         break;
12617                 switch(c) {
12618                         case 'a': /* ignored */ break;
12619                         case 'b':
12620                                 ctree_flags |= OPEN_CTREE_BACKUP_ROOT;
12621                                 break;
12622                         case 's':
12623                                 num = arg_strtou64(optarg);
12624                                 if (num >= BTRFS_SUPER_MIRROR_MAX) {
12625                                         error(
12626                                         "super mirror should be less than %d",
12627                                                 BTRFS_SUPER_MIRROR_MAX);
12628                                         exit(1);
12629                                 }
12630                                 bytenr = btrfs_sb_offset(((int)num));
12631                                 printf("using SB copy %llu, bytenr %llu\n", num,
12632                                        (unsigned long long)bytenr);
12633                                 break;
12634                         case 'Q':
12635                                 qgroup_report = 1;
12636                                 break;
12637                         case 'E':
12638                                 subvolid = arg_strtou64(optarg);
12639                                 break;
12640                         case 'r':
12641                                 tree_root_bytenr = arg_strtou64(optarg);
12642                                 break;
12643                         case GETOPT_VAL_CHUNK_TREE:
12644                                 chunk_root_bytenr = arg_strtou64(optarg);
12645                                 break;
12646                         case 'p':
12647                                 ctx.progress_enabled = true;
12648                                 break;
12649                         case '?':
12650                         case 'h':
12651                                 usage(cmd_check_usage);
12652                         case GETOPT_VAL_REPAIR:
12653                                 printf("enabling repair mode\n");
12654                                 repair = 1;
12655                                 ctree_flags |= OPEN_CTREE_WRITES;
12656                                 break;
12657                         case GETOPT_VAL_READONLY:
12658                                 readonly = 1;
12659                                 break;
12660                         case GETOPT_VAL_INIT_CSUM:
12661                                 printf("Creating a new CRC tree\n");
12662                                 init_csum_tree = 1;
12663                                 repair = 1;
12664                                 ctree_flags |= OPEN_CTREE_WRITES;
12665                                 break;
12666                         case GETOPT_VAL_INIT_EXTENT:
12667                                 init_extent_tree = 1;
12668                                 ctree_flags |= (OPEN_CTREE_WRITES |
12669                                                 OPEN_CTREE_NO_BLOCK_GROUPS);
12670                                 repair = 1;
12671                                 break;
12672                         case GETOPT_VAL_CHECK_CSUM:
12673                                 check_data_csum = 1;
12674                                 break;
12675                         case GETOPT_VAL_MODE:
12676                                 check_mode = parse_check_mode(optarg);
12677                                 if (check_mode == CHECK_MODE_UNKNOWN) {
12678                                         error("unknown mode: %s", optarg);
12679                                         exit(1);
12680                                 }
12681                                 break;
12682                         case GETOPT_VAL_CLEAR_SPACE_CACHE:
12683                                 if (strcmp(optarg, "v1") == 0) {
12684                                         clear_space_cache = 1;
12685                                 } else if (strcmp(optarg, "v2") == 0) {
12686                                         clear_space_cache = 2;
12687                                         ctree_flags |= OPEN_CTREE_INVALIDATE_FST;
12688                                 } else {
12689                                         error(
12690                 "invalid argument to --clear-space-cache, must be v1 or v2");
12691                                         exit(1);
12692                                 }
12693                                 ctree_flags |= OPEN_CTREE_WRITES;
12694                                 break;
12695                 }
12696         }
12697
12698         if (check_argc_exact(argc - optind, 1))
12699                 usage(cmd_check_usage);
12700
12701         if (ctx.progress_enabled) {
12702                 ctx.tp = TASK_NOTHING;
12703                 ctx.info = task_init(print_status_check, print_status_return, &ctx);
12704         }
12705
12706         /* This check is the only reason for --readonly to exist */
12707         if (readonly && repair) {
12708                 error("repair options are not compatible with --readonly");
12709                 exit(1);
12710         }
12711
12712         /*
12713          * Not supported yet
12714          */
12715         if (repair && check_mode == CHECK_MODE_LOWMEM) {
12716                 error("low memory mode doesn't support repair yet");
12717                 exit(1);
12718         }
12719
12720         radix_tree_init();
12721         cache_tree_init(&root_cache);
12722
12723         if((ret = check_mounted(argv[optind])) < 0) {
12724                 error("could not check mount status: %s", strerror(-ret));
12725                 err |= !!ret;
12726                 goto err_out;
12727         } else if(ret) {
12728                 error("%s is currently mounted, aborting", argv[optind]);
12729                 ret = -EBUSY;
12730                 err |= !!ret;
12731                 goto err_out;
12732         }
12733
12734         /* only allow partial opening under repair mode */
12735         if (repair)
12736                 ctree_flags |= OPEN_CTREE_PARTIAL;
12737
12738         info = open_ctree_fs_info(argv[optind], bytenr, tree_root_bytenr,
12739                                   chunk_root_bytenr, ctree_flags);
12740         if (!info) {
12741                 error("cannot open file system");
12742                 ret = -EIO;
12743                 err |= !!ret;
12744                 goto err_out;
12745         }
12746
12747         global_info = info;
12748         root = info->fs_root;
12749         if (clear_space_cache == 1) {
12750                 if (btrfs_fs_compat_ro(info,
12751                                 BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE)) {
12752                         error(
12753                 "free space cache v2 detected, use --clear-space-cache v2");
12754                         ret = 1;
12755                         goto close_out;
12756                 }
12757                 printf("Clearing free space cache\n");
12758                 ret = clear_free_space_cache(info);
12759                 if (ret) {
12760                         error("failed to clear free space cache");
12761                         ret = 1;
12762                 } else {
12763                         printf("Free space cache cleared\n");
12764                 }
12765                 goto close_out;
12766         } else if (clear_space_cache == 2) {
12767                 if (!btrfs_fs_compat_ro(info,
12768                                         BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE)) {
12769                         printf("no free space cache v2 to clear\n");
12770                         ret = 0;
12771                         goto close_out;
12772                 }
12773                 printf("Clear free space cache v2\n");
12774                 ret = btrfs_clear_free_space_tree(info);
12775                 if (ret) {
12776                         error("failed to clear free space cache v2: %d", ret);
12777                         ret = 1;
12778                 } else {
12779                         printf("free space cache v2 cleared\n");
12780                 }
12781                 goto close_out;
12782         }
12783
12784         /*
12785          * repair mode will force us to commit transaction which
12786          * will make us fail to load log tree when mounting.
12787          */
12788         if (repair && btrfs_super_log_root(info->super_copy)) {
12789                 ret = ask_user("repair mode will force to clear out log tree, are you sure?");
12790                 if (!ret) {
12791                         ret = 1;
12792                         err |= !!ret;
12793                         goto close_out;
12794                 }
12795                 ret = zero_log_tree(root);
12796                 err |= !!ret;
12797                 if (ret) {
12798                         error("failed to zero log tree: %d", ret);
12799                         goto close_out;
12800                 }
12801         }
12802
12803         uuid_unparse(info->super_copy->fsid, uuidbuf);
12804         if (qgroup_report) {
12805                 printf("Print quota groups for %s\nUUID: %s\n", argv[optind],
12806                        uuidbuf);
12807                 ret = qgroup_verify_all(info);
12808                 err |= !!ret;
12809                 if (ret == 0)
12810                         report_qgroups(1);
12811                 goto close_out;
12812         }
12813         if (subvolid) {
12814                 printf("Print extent state for subvolume %llu on %s\nUUID: %s\n",
12815                        subvolid, argv[optind], uuidbuf);
12816                 ret = print_extent_state(info, subvolid);
12817                 err |= !!ret;
12818                 goto close_out;
12819         }
12820         printf("Checking filesystem on %s\nUUID: %s\n", argv[optind], uuidbuf);
12821
12822         if (!extent_buffer_uptodate(info->tree_root->node) ||
12823             !extent_buffer_uptodate(info->dev_root->node) ||
12824             !extent_buffer_uptodate(info->chunk_root->node)) {
12825                 error("critical roots corrupted, unable to check the filesystem");
12826                 err |= !!ret;
12827                 ret = -EIO;
12828                 goto close_out;
12829         }
12830
12831         if (init_extent_tree || init_csum_tree) {
12832                 struct btrfs_trans_handle *trans;
12833
12834                 trans = btrfs_start_transaction(info->extent_root, 0);
12835                 if (IS_ERR(trans)) {
12836                         error("error starting transaction");
12837                         ret = PTR_ERR(trans);
12838                         err |= !!ret;
12839                         goto close_out;
12840                 }
12841
12842                 if (init_extent_tree) {
12843                         printf("Creating a new extent tree\n");
12844                         ret = reinit_extent_tree(trans, info);
12845                         err |= !!ret;
12846                         if (ret)
12847                                 goto close_out;
12848                 }
12849
12850                 if (init_csum_tree) {
12851                         printf("Reinitialize checksum tree\n");
12852                         ret = btrfs_fsck_reinit_root(trans, info->csum_root, 0);
12853                         if (ret) {
12854                                 error("checksum tree initialization failed: %d",
12855                                                 ret);
12856                                 ret = -EIO;
12857                                 err |= !!ret;
12858                                 goto close_out;
12859                         }
12860
12861                         ret = fill_csum_tree(trans, info->csum_root,
12862                                              init_extent_tree);
12863                         err |= !!ret;
12864                         if (ret) {
12865                                 error("checksum tree refilling failed: %d", ret);
12866                                 return -EIO;
12867                         }
12868                 }
12869                 /*
12870                  * Ok now we commit and run the normal fsck, which will add
12871                  * extent entries for all of the items it finds.
12872                  */
12873                 ret = btrfs_commit_transaction(trans, info->extent_root);
12874                 err |= !!ret;
12875                 if (ret)
12876                         goto close_out;
12877         }
12878         if (!extent_buffer_uptodate(info->extent_root->node)) {
12879                 error("critical: extent_root, unable to check the filesystem");
12880                 ret = -EIO;
12881                 err |= !!ret;
12882                 goto close_out;
12883         }
12884         if (!extent_buffer_uptodate(info->csum_root->node)) {
12885                 error("critical: csum_root, unable to check the filesystem");
12886                 ret = -EIO;
12887                 err |= !!ret;
12888                 goto close_out;
12889         }
12890
12891         if (!ctx.progress_enabled)
12892                 fprintf(stderr, "checking extents\n");
12893         if (check_mode == CHECK_MODE_LOWMEM)
12894                 ret = check_chunks_and_extents_v2(root);
12895         else
12896                 ret = check_chunks_and_extents(root);
12897         err |= !!ret;
12898         if (ret)
12899                 error(
12900                 "errors found in extent allocation tree or chunk allocation");
12901
12902         ret = repair_root_items(info);
12903         err |= !!ret;
12904         if (ret < 0)
12905                 goto close_out;
12906         if (repair) {
12907                 fprintf(stderr, "Fixed %d roots.\n", ret);
12908                 ret = 0;
12909         } else if (ret > 0) {
12910                 fprintf(stderr,
12911                        "Found %d roots with an outdated root item.\n",
12912                        ret);
12913                 fprintf(stderr,
12914                         "Please run a filesystem check with the option --repair to fix them.\n");
12915                 ret = 1;
12916                 err |= !!ret;
12917                 goto close_out;
12918         }
12919
12920         if (!ctx.progress_enabled) {
12921                 if (btrfs_fs_compat_ro(info, BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE))
12922                         fprintf(stderr, "checking free space tree\n");
12923                 else
12924                         fprintf(stderr, "checking free space cache\n");
12925         }
12926         ret = check_space_cache(root);
12927         err |= !!ret;
12928         if (ret)
12929                 goto out;
12930
12931         /*
12932          * We used to have to have these hole extents in between our real
12933          * extents so if we don't have this flag set we need to make sure there
12934          * are no gaps in the file extents for inodes, otherwise we can just
12935          * ignore it when this happens.
12936          */
12937         no_holes = btrfs_fs_incompat(root->fs_info,
12938                                      BTRFS_FEATURE_INCOMPAT_NO_HOLES);
12939         if (!ctx.progress_enabled)
12940                 fprintf(stderr, "checking fs roots\n");
12941         if (check_mode == CHECK_MODE_LOWMEM)
12942                 ret = check_fs_roots_v2(root->fs_info);
12943         else
12944                 ret = check_fs_roots(root, &root_cache);
12945         err |= !!ret;
12946         if (ret)
12947                 goto out;
12948
12949         fprintf(stderr, "checking csums\n");
12950         ret = check_csums(root);
12951         err |= !!ret;
12952         if (ret)
12953                 goto out;
12954
12955         fprintf(stderr, "checking root refs\n");
12956         /* For low memory mode, check_fs_roots_v2 handles root refs */
12957         if (check_mode != CHECK_MODE_LOWMEM) {
12958                 ret = check_root_refs(root, &root_cache);
12959                 err |= !!ret;
12960                 if (ret)
12961                         goto out;
12962         }
12963
12964         while (repair && !list_empty(&root->fs_info->recow_ebs)) {
12965                 struct extent_buffer *eb;
12966
12967                 eb = list_first_entry(&root->fs_info->recow_ebs,
12968                                       struct extent_buffer, recow);
12969                 list_del_init(&eb->recow);
12970                 ret = recow_extent_buffer(root, eb);
12971                 err |= !!ret;
12972                 if (ret)
12973                         break;
12974         }
12975
12976         while (!list_empty(&delete_items)) {
12977                 struct bad_item *bad;
12978
12979                 bad = list_first_entry(&delete_items, struct bad_item, list);
12980                 list_del_init(&bad->list);
12981                 if (repair) {
12982                         ret = delete_bad_item(root, bad);
12983                         err |= !!ret;
12984                 }
12985                 free(bad);
12986         }
12987
12988         if (info->quota_enabled) {
12989                 fprintf(stderr, "checking quota groups\n");
12990                 ret = qgroup_verify_all(info);
12991                 err |= !!ret;
12992                 if (ret)
12993                         goto out;
12994                 report_qgroups(0);
12995                 ret = repair_qgroups(info, &qgroups_repaired);
12996                 err |= !!ret;
12997                 if (err)
12998                         goto out;
12999                 ret = 0;
13000         }
13001
13002         if (!list_empty(&root->fs_info->recow_ebs)) {
13003                 error("transid errors in file system");
13004                 ret = 1;
13005                 err |= !!ret;
13006         }
13007 out:
13008         if (found_old_backref) { /*
13009                  * there was a disk format change when mixed
13010                  * backref was in testing tree. The old format
13011                  * existed about one week.
13012                  */
13013                 printf("\n * Found old mixed backref format. "
13014                        "The old format is not supported! *"
13015                        "\n * Please mount the FS in readonly mode, "
13016                        "backup data and re-format the FS. *\n\n");
13017                 err |= 1;
13018         }
13019         printf("found %llu bytes used err is %d\n",
13020                (unsigned long long)bytes_used, ret);
13021         printf("total csum bytes: %llu\n",(unsigned long long)total_csum_bytes);
13022         printf("total tree bytes: %llu\n",
13023                (unsigned long long)total_btree_bytes);
13024         printf("total fs tree bytes: %llu\n",
13025                (unsigned long long)total_fs_tree_bytes);
13026         printf("total extent tree bytes: %llu\n",
13027                (unsigned long long)total_extent_tree_bytes);
13028         printf("btree space waste bytes: %llu\n",
13029                (unsigned long long)btree_space_waste);
13030         printf("file data blocks allocated: %llu\n referenced %llu\n",
13031                 (unsigned long long)data_bytes_allocated,
13032                 (unsigned long long)data_bytes_referenced);
13033
13034         free_qgroup_counts();
13035         free_root_recs_tree(&root_cache);
13036 close_out:
13037         close_ctree(root);
13038 err_out:
13039         if (ctx.progress_enabled)
13040                 task_deinit(ctx.info);
13041
13042         return err;
13043 }